]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
58 | #include <linux/mmu_context.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/slab.h> | |
6c271ce2 | 61 | #include <linux/kthread.h> |
2b188cc1 | 62 | #include <linux/blkdev.h> |
edafccee | 63 | #include <linux/bvec.h> |
2b188cc1 JA |
64 | #include <linux/net.h> |
65 | #include <net/sock.h> | |
66 | #include <net/af_unix.h> | |
6b06314c | 67 | #include <net/scm.h> |
2b188cc1 JA |
68 | #include <linux/anon_inodes.h> |
69 | #include <linux/sched/mm.h> | |
70 | #include <linux/uaccess.h> | |
71 | #include <linux/nospec.h> | |
edafccee JA |
72 | #include <linux/sizes.h> |
73 | #include <linux/hugetlb.h> | |
aa4c3967 | 74 | #include <linux/highmem.h> |
15b71abe JA |
75 | #include <linux/namei.h> |
76 | #include <linux/fsnotify.h> | |
4840e418 | 77 | #include <linux/fadvise.h> |
3e4827b0 | 78 | #include <linux/eventpoll.h> |
ff002b30 | 79 | #include <linux/fs_struct.h> |
7d67af2c | 80 | #include <linux/splice.h> |
b41e9852 | 81 | #include <linux/task_work.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
101 | |
102 | struct io_uring { | |
103 | u32 head ____cacheline_aligned_in_smp; | |
104 | u32 tail ____cacheline_aligned_in_smp; | |
105 | }; | |
106 | ||
1e84b97b | 107 | /* |
75b28aff HV |
108 | * This data is shared with the application through the mmap at offsets |
109 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
110 | * |
111 | * The offsets to the member fields are published through struct | |
112 | * io_sqring_offsets when calling io_uring_setup. | |
113 | */ | |
75b28aff | 114 | struct io_rings { |
1e84b97b SB |
115 | /* |
116 | * Head and tail offsets into the ring; the offsets need to be | |
117 | * masked to get valid indices. | |
118 | * | |
75b28aff HV |
119 | * The kernel controls head of the sq ring and the tail of the cq ring, |
120 | * and the application controls tail of the sq ring and the head of the | |
121 | * cq ring. | |
1e84b97b | 122 | */ |
75b28aff | 123 | struct io_uring sq, cq; |
1e84b97b | 124 | /* |
75b28aff | 125 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
126 | * ring_entries - 1) |
127 | */ | |
75b28aff HV |
128 | u32 sq_ring_mask, cq_ring_mask; |
129 | /* Ring sizes (constant, power of 2) */ | |
130 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
131 | /* |
132 | * Number of invalid entries dropped by the kernel due to | |
133 | * invalid index stored in array | |
134 | * | |
135 | * Written by the kernel, shouldn't be modified by the | |
136 | * application (i.e. get number of "new events" by comparing to | |
137 | * cached value). | |
138 | * | |
139 | * After a new SQ head value was read by the application this | |
140 | * counter includes all submissions that were dropped reaching | |
141 | * the new SQ head (and possibly more). | |
142 | */ | |
75b28aff | 143 | u32 sq_dropped; |
1e84b97b SB |
144 | /* |
145 | * Runtime flags | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application. | |
149 | * | |
150 | * The application needs a full memory barrier before checking | |
151 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
152 | */ | |
75b28aff | 153 | u32 sq_flags; |
1e84b97b SB |
154 | /* |
155 | * Number of completion events lost because the queue was full; | |
156 | * this should be avoided by the application by making sure | |
0b4295b5 | 157 | * there are not more requests pending than there is space in |
1e84b97b SB |
158 | * the completion queue. |
159 | * | |
160 | * Written by the kernel, shouldn't be modified by the | |
161 | * application (i.e. get number of "new events" by comparing to | |
162 | * cached value). | |
163 | * | |
164 | * As completion events come in out of order this counter is not | |
165 | * ordered with any other data. | |
166 | */ | |
75b28aff | 167 | u32 cq_overflow; |
1e84b97b SB |
168 | /* |
169 | * Ring buffer of completion events. | |
170 | * | |
171 | * The kernel writes completion events fresh every time they are | |
172 | * produced, so the application is allowed to modify pending | |
173 | * entries. | |
174 | */ | |
75b28aff | 175 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
176 | }; |
177 | ||
edafccee JA |
178 | struct io_mapped_ubuf { |
179 | u64 ubuf; | |
180 | size_t len; | |
181 | struct bio_vec *bvec; | |
182 | unsigned int nr_bvecs; | |
183 | }; | |
184 | ||
65e19f54 JA |
185 | struct fixed_file_table { |
186 | struct file **files; | |
31b51510 JA |
187 | }; |
188 | ||
05589553 XW |
189 | struct fixed_file_ref_node { |
190 | struct percpu_ref refs; | |
191 | struct list_head node; | |
192 | struct list_head file_list; | |
193 | struct fixed_file_data *file_data; | |
194 | struct work_struct work; | |
195 | }; | |
196 | ||
05f3fb3c JA |
197 | struct fixed_file_data { |
198 | struct fixed_file_table *table; | |
199 | struct io_ring_ctx *ctx; | |
200 | ||
05589553 | 201 | struct percpu_ref *cur_refs; |
05f3fb3c | 202 | struct percpu_ref refs; |
05f3fb3c | 203 | struct completion done; |
05589553 XW |
204 | struct list_head ref_list; |
205 | spinlock_t lock; | |
05f3fb3c JA |
206 | }; |
207 | ||
5a2e745d JA |
208 | struct io_buffer { |
209 | struct list_head list; | |
210 | __u64 addr; | |
211 | __s32 len; | |
212 | __u16 bid; | |
213 | }; | |
214 | ||
2b188cc1 JA |
215 | struct io_ring_ctx { |
216 | struct { | |
217 | struct percpu_ref refs; | |
218 | } ____cacheline_aligned_in_smp; | |
219 | ||
220 | struct { | |
221 | unsigned int flags; | |
e1d85334 RD |
222 | unsigned int compat: 1; |
223 | unsigned int account_mem: 1; | |
224 | unsigned int cq_overflow_flushed: 1; | |
225 | unsigned int drain_next: 1; | |
226 | unsigned int eventfd_async: 1; | |
2b188cc1 | 227 | |
75b28aff HV |
228 | /* |
229 | * Ring buffer of indices into array of io_uring_sqe, which is | |
230 | * mmapped by the application using the IORING_OFF_SQES offset. | |
231 | * | |
232 | * This indirection could e.g. be used to assign fixed | |
233 | * io_uring_sqe entries to operations and only submit them to | |
234 | * the queue when needed. | |
235 | * | |
236 | * The kernel modifies neither the indices array nor the entries | |
237 | * array. | |
238 | */ | |
239 | u32 *sq_array; | |
2b188cc1 JA |
240 | unsigned cached_sq_head; |
241 | unsigned sq_entries; | |
242 | unsigned sq_mask; | |
6c271ce2 | 243 | unsigned sq_thread_idle; |
498ccd9e | 244 | unsigned cached_sq_dropped; |
206aefde | 245 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 246 | unsigned long sq_check_overflow; |
de0617e4 JA |
247 | |
248 | struct list_head defer_list; | |
5262f567 | 249 | struct list_head timeout_list; |
1d7bb1d5 | 250 | struct list_head cq_overflow_list; |
fcb323cc JA |
251 | |
252 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 253 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
254 | } ____cacheline_aligned_in_smp; |
255 | ||
206aefde JA |
256 | struct io_rings *rings; |
257 | ||
2b188cc1 | 258 | /* IO offload */ |
561fb04a | 259 | struct io_wq *io_wq; |
6c271ce2 | 260 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 261 | struct mm_struct *sqo_mm; |
6c271ce2 | 262 | wait_queue_head_t sqo_wait; |
75b28aff | 263 | |
6b06314c JA |
264 | /* |
265 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
266 | * readers must ensure that ->refs is alive as long as the file* is | |
267 | * used. Only updated through io_uring_register(2). | |
268 | */ | |
05f3fb3c | 269 | struct fixed_file_data *file_data; |
6b06314c | 270 | unsigned nr_user_files; |
b14cca0c PB |
271 | int ring_fd; |
272 | struct file *ring_file; | |
6b06314c | 273 | |
edafccee JA |
274 | /* if used, fixed mapped user buffers */ |
275 | unsigned nr_user_bufs; | |
276 | struct io_mapped_ubuf *user_bufs; | |
277 | ||
2b188cc1 JA |
278 | struct user_struct *user; |
279 | ||
0b8c0ec7 | 280 | const struct cred *creds; |
181e448d | 281 | |
206aefde JA |
282 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
283 | struct completion *completions; | |
284 | ||
0ddf92e8 JA |
285 | /* if all else fails... */ |
286 | struct io_kiocb *fallback_req; | |
287 | ||
206aefde JA |
288 | #if defined(CONFIG_UNIX) |
289 | struct socket *ring_sock; | |
290 | #endif | |
291 | ||
5a2e745d JA |
292 | struct idr io_buffer_idr; |
293 | ||
071698e1 JA |
294 | struct idr personality_idr; |
295 | ||
206aefde JA |
296 | struct { |
297 | unsigned cached_cq_tail; | |
298 | unsigned cq_entries; | |
299 | unsigned cq_mask; | |
300 | atomic_t cq_timeouts; | |
ad3eb2c8 | 301 | unsigned long cq_check_overflow; |
206aefde JA |
302 | struct wait_queue_head cq_wait; |
303 | struct fasync_struct *cq_fasync; | |
304 | struct eventfd_ctx *cq_ev_fd; | |
305 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
306 | |
307 | struct { | |
308 | struct mutex uring_lock; | |
309 | wait_queue_head_t wait; | |
310 | } ____cacheline_aligned_in_smp; | |
311 | ||
312 | struct { | |
313 | spinlock_t completion_lock; | |
e94f141b | 314 | |
def596e9 JA |
315 | /* |
316 | * ->poll_list is protected by the ctx->uring_lock for | |
317 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
318 | * For SQPOLL, only the single threaded io_sq_thread() will | |
319 | * manipulate the list, hence no extra locking is needed there. | |
320 | */ | |
321 | struct list_head poll_list; | |
78076bb6 JA |
322 | struct hlist_head *cancel_hash; |
323 | unsigned cancel_hash_bits; | |
e94f141b | 324 | bool poll_multi_file; |
31b51510 | 325 | |
fcb323cc JA |
326 | spinlock_t inflight_lock; |
327 | struct list_head inflight_list; | |
2b188cc1 | 328 | } ____cacheline_aligned_in_smp; |
85faa7b8 JA |
329 | |
330 | struct work_struct exit_work; | |
2b188cc1 JA |
331 | }; |
332 | ||
09bb8394 JA |
333 | /* |
334 | * First field must be the file pointer in all the | |
335 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
336 | */ | |
221c5eb2 JA |
337 | struct io_poll_iocb { |
338 | struct file *file; | |
0969e783 JA |
339 | union { |
340 | struct wait_queue_head *head; | |
341 | u64 addr; | |
342 | }; | |
221c5eb2 | 343 | __poll_t events; |
8c838788 | 344 | bool done; |
221c5eb2 | 345 | bool canceled; |
392edb45 | 346 | struct wait_queue_entry wait; |
221c5eb2 JA |
347 | }; |
348 | ||
b5dba59e JA |
349 | struct io_close { |
350 | struct file *file; | |
351 | struct file *put_file; | |
352 | int fd; | |
353 | }; | |
354 | ||
ad8a48ac JA |
355 | struct io_timeout_data { |
356 | struct io_kiocb *req; | |
357 | struct hrtimer timer; | |
358 | struct timespec64 ts; | |
359 | enum hrtimer_mode mode; | |
cc42e0ac | 360 | u32 seq_offset; |
ad8a48ac JA |
361 | }; |
362 | ||
8ed8d3c3 JA |
363 | struct io_accept { |
364 | struct file *file; | |
365 | struct sockaddr __user *addr; | |
366 | int __user *addr_len; | |
367 | int flags; | |
09952e3e | 368 | unsigned long nofile; |
8ed8d3c3 JA |
369 | }; |
370 | ||
371 | struct io_sync { | |
372 | struct file *file; | |
373 | loff_t len; | |
374 | loff_t off; | |
375 | int flags; | |
d63d1b5e | 376 | int mode; |
8ed8d3c3 JA |
377 | }; |
378 | ||
fbf23849 JA |
379 | struct io_cancel { |
380 | struct file *file; | |
381 | u64 addr; | |
382 | }; | |
383 | ||
b29472ee JA |
384 | struct io_timeout { |
385 | struct file *file; | |
386 | u64 addr; | |
387 | int flags; | |
26a61679 | 388 | unsigned count; |
b29472ee JA |
389 | }; |
390 | ||
9adbd45d JA |
391 | struct io_rw { |
392 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
393 | struct kiocb kiocb; | |
394 | u64 addr; | |
395 | u64 len; | |
396 | }; | |
397 | ||
3fbb51c1 JA |
398 | struct io_connect { |
399 | struct file *file; | |
400 | struct sockaddr __user *addr; | |
401 | int addr_len; | |
402 | }; | |
403 | ||
e47293fd JA |
404 | struct io_sr_msg { |
405 | struct file *file; | |
fddaface JA |
406 | union { |
407 | struct user_msghdr __user *msg; | |
408 | void __user *buf; | |
409 | }; | |
e47293fd | 410 | int msg_flags; |
bcda7baa | 411 | int bgid; |
fddaface | 412 | size_t len; |
bcda7baa | 413 | struct io_buffer *kbuf; |
e47293fd JA |
414 | }; |
415 | ||
15b71abe JA |
416 | struct io_open { |
417 | struct file *file; | |
418 | int dfd; | |
eddc7ef5 | 419 | union { |
eddc7ef5 JA |
420 | unsigned mask; |
421 | }; | |
15b71abe | 422 | struct filename *filename; |
eddc7ef5 | 423 | struct statx __user *buffer; |
c12cedf2 | 424 | struct open_how how; |
4022e7af | 425 | unsigned long nofile; |
15b71abe JA |
426 | }; |
427 | ||
05f3fb3c JA |
428 | struct io_files_update { |
429 | struct file *file; | |
430 | u64 arg; | |
431 | u32 nr_args; | |
432 | u32 offset; | |
433 | }; | |
434 | ||
4840e418 JA |
435 | struct io_fadvise { |
436 | struct file *file; | |
437 | u64 offset; | |
438 | u32 len; | |
439 | u32 advice; | |
440 | }; | |
441 | ||
c1ca757b JA |
442 | struct io_madvise { |
443 | struct file *file; | |
444 | u64 addr; | |
445 | u32 len; | |
446 | u32 advice; | |
447 | }; | |
448 | ||
3e4827b0 JA |
449 | struct io_epoll { |
450 | struct file *file; | |
451 | int epfd; | |
452 | int op; | |
453 | int fd; | |
454 | struct epoll_event event; | |
e47293fd JA |
455 | }; |
456 | ||
7d67af2c PB |
457 | struct io_splice { |
458 | struct file *file_out; | |
459 | struct file *file_in; | |
460 | loff_t off_out; | |
461 | loff_t off_in; | |
462 | u64 len; | |
463 | unsigned int flags; | |
464 | }; | |
465 | ||
ddf0322d JA |
466 | struct io_provide_buf { |
467 | struct file *file; | |
468 | __u64 addr; | |
469 | __s32 len; | |
470 | __u32 bgid; | |
471 | __u16 nbufs; | |
472 | __u16 bid; | |
473 | }; | |
474 | ||
f499a021 JA |
475 | struct io_async_connect { |
476 | struct sockaddr_storage address; | |
477 | }; | |
478 | ||
03b1230c JA |
479 | struct io_async_msghdr { |
480 | struct iovec fast_iov[UIO_FASTIOV]; | |
481 | struct iovec *iov; | |
482 | struct sockaddr __user *uaddr; | |
483 | struct msghdr msg; | |
b537916c | 484 | struct sockaddr_storage addr; |
03b1230c JA |
485 | }; |
486 | ||
f67676d1 JA |
487 | struct io_async_rw { |
488 | struct iovec fast_iov[UIO_FASTIOV]; | |
489 | struct iovec *iov; | |
490 | ssize_t nr_segs; | |
491 | ssize_t size; | |
492 | }; | |
493 | ||
1a6b74fc | 494 | struct io_async_ctx { |
f67676d1 JA |
495 | union { |
496 | struct io_async_rw rw; | |
03b1230c | 497 | struct io_async_msghdr msg; |
f499a021 | 498 | struct io_async_connect connect; |
2d28390a | 499 | struct io_timeout_data timeout; |
f67676d1 | 500 | }; |
1a6b74fc JA |
501 | }; |
502 | ||
6b47ee6e PB |
503 | enum { |
504 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
505 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
506 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
507 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
508 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 509 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 510 | |
dea3b49c | 511 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
512 | REQ_F_LINK_NEXT_BIT, |
513 | REQ_F_FAIL_LINK_BIT, | |
514 | REQ_F_INFLIGHT_BIT, | |
515 | REQ_F_CUR_POS_BIT, | |
516 | REQ_F_NOWAIT_BIT, | |
517 | REQ_F_IOPOLL_COMPLETED_BIT, | |
518 | REQ_F_LINK_TIMEOUT_BIT, | |
519 | REQ_F_TIMEOUT_BIT, | |
520 | REQ_F_ISREG_BIT, | |
521 | REQ_F_MUST_PUNT_BIT, | |
522 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
523 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 524 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 525 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 526 | REQ_F_POLLED_BIT, |
bcda7baa | 527 | REQ_F_BUFFER_SELECTED_BIT, |
84557871 JA |
528 | |
529 | /* not a real bit, just to check we're not overflowing the space */ | |
530 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
531 | }; |
532 | ||
533 | enum { | |
534 | /* ctx owns file */ | |
535 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
536 | /* drain existing IO first */ | |
537 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
538 | /* linked sqes */ | |
539 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
540 | /* doesn't sever on completion < 0 */ | |
541 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
542 | /* IOSQE_ASYNC */ | |
543 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
544 | /* IOSQE_BUFFER_SELECT */ |
545 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 546 | |
dea3b49c PB |
547 | /* head of a link */ |
548 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
549 | /* already grabbed next link */ |
550 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
551 | /* fail rest of links */ | |
552 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
553 | /* on inflight list */ | |
554 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
555 | /* read/write uses file position */ | |
556 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
557 | /* must not punt to workers */ | |
558 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
559 | /* polled IO has completed */ | |
560 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
561 | /* has linked timeout */ | |
562 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
563 | /* timeout request */ | |
564 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
565 | /* regular file */ | |
566 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
567 | /* must be punted even for NONBLOCK */ | |
568 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
569 | /* no timeout sequence */ | |
570 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
571 | /* completion under lock */ | |
572 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
573 | /* needs cleanup */ |
574 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
575 | /* in overflow list */ |
576 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
577 | /* already went through poll handler */ |
578 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
579 | /* buffer already selected */ |
580 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
d7718a9d JA |
581 | }; |
582 | ||
583 | struct async_poll { | |
584 | struct io_poll_iocb poll; | |
585 | struct io_wq_work work; | |
6b47ee6e PB |
586 | }; |
587 | ||
09bb8394 JA |
588 | /* |
589 | * NOTE! Each of the iocb union members has the file pointer | |
590 | * as the first entry in their struct definition. So you can | |
591 | * access the file pointer through any of the sub-structs, | |
592 | * or directly as just 'ki_filp' in this struct. | |
593 | */ | |
2b188cc1 | 594 | struct io_kiocb { |
221c5eb2 | 595 | union { |
09bb8394 | 596 | struct file *file; |
9adbd45d | 597 | struct io_rw rw; |
221c5eb2 | 598 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
599 | struct io_accept accept; |
600 | struct io_sync sync; | |
fbf23849 | 601 | struct io_cancel cancel; |
b29472ee | 602 | struct io_timeout timeout; |
3fbb51c1 | 603 | struct io_connect connect; |
e47293fd | 604 | struct io_sr_msg sr_msg; |
15b71abe | 605 | struct io_open open; |
b5dba59e | 606 | struct io_close close; |
05f3fb3c | 607 | struct io_files_update files_update; |
4840e418 | 608 | struct io_fadvise fadvise; |
c1ca757b | 609 | struct io_madvise madvise; |
3e4827b0 | 610 | struct io_epoll epoll; |
7d67af2c | 611 | struct io_splice splice; |
ddf0322d | 612 | struct io_provide_buf pbuf; |
221c5eb2 | 613 | }; |
2b188cc1 | 614 | |
1a6b74fc | 615 | struct io_async_ctx *io; |
c398ecb3 | 616 | int cflags; |
cf6fd4bd | 617 | bool needs_fixed_file; |
d625c6ee | 618 | u8 opcode; |
2b188cc1 JA |
619 | |
620 | struct io_ring_ctx *ctx; | |
d7718a9d | 621 | struct list_head list; |
2b188cc1 | 622 | unsigned int flags; |
c16361c1 | 623 | refcount_t refs; |
3537b6a7 JA |
624 | struct task_struct *task; |
625 | unsigned long fsize; | |
2b188cc1 | 626 | u64 user_data; |
9e645e11 | 627 | u32 result; |
de0617e4 | 628 | u32 sequence; |
2b188cc1 | 629 | |
d7718a9d JA |
630 | struct list_head link_list; |
631 | ||
fcb323cc JA |
632 | struct list_head inflight_entry; |
633 | ||
05589553 XW |
634 | struct percpu_ref *fixed_file_refs; |
635 | ||
b41e9852 JA |
636 | union { |
637 | /* | |
638 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
639 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
640 | * async armed poll handlers for regular commands. The latter | |
641 | * restore the work, if needed. | |
b41e9852 JA |
642 | */ |
643 | struct { | |
b41e9852 | 644 | struct callback_head task_work; |
d7718a9d JA |
645 | struct hlist_node hash_node; |
646 | struct async_poll *apoll; | |
b41e9852 JA |
647 | }; |
648 | struct io_wq_work work; | |
649 | }; | |
2b188cc1 JA |
650 | }; |
651 | ||
652 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 653 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 654 | |
9a56a232 JA |
655 | struct io_submit_state { |
656 | struct blk_plug plug; | |
657 | ||
2579f913 JA |
658 | /* |
659 | * io_kiocb alloc cache | |
660 | */ | |
661 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 662 | unsigned int free_reqs; |
2579f913 | 663 | |
9a56a232 JA |
664 | /* |
665 | * File reference cache | |
666 | */ | |
667 | struct file *file; | |
668 | unsigned int fd; | |
669 | unsigned int has_refs; | |
670 | unsigned int used_refs; | |
671 | unsigned int ios_left; | |
672 | }; | |
673 | ||
d3656344 JA |
674 | struct io_op_def { |
675 | /* needs req->io allocated for deferral/async */ | |
676 | unsigned async_ctx : 1; | |
677 | /* needs current->mm setup, does mm access */ | |
678 | unsigned needs_mm : 1; | |
679 | /* needs req->file assigned */ | |
680 | unsigned needs_file : 1; | |
681 | /* needs req->file assigned IFF fd is >= 0 */ | |
682 | unsigned fd_non_neg : 1; | |
683 | /* hash wq insertion if file is a regular file */ | |
684 | unsigned hash_reg_file : 1; | |
685 | /* unbound wq insertion if file is a non-regular file */ | |
686 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
687 | /* opcode is not supported by this kernel */ |
688 | unsigned not_supported : 1; | |
f86cd20c JA |
689 | /* needs file table */ |
690 | unsigned file_table : 1; | |
ff002b30 JA |
691 | /* needs ->fs */ |
692 | unsigned needs_fs : 1; | |
8a72758c JA |
693 | /* set if opcode supports polled "wait" */ |
694 | unsigned pollin : 1; | |
695 | unsigned pollout : 1; | |
bcda7baa JA |
696 | /* op supports buffer selection */ |
697 | unsigned buffer_select : 1; | |
d3656344 JA |
698 | }; |
699 | ||
700 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
701 | [IORING_OP_NOP] = {}, |
702 | [IORING_OP_READV] = { | |
d3656344 JA |
703 | .async_ctx = 1, |
704 | .needs_mm = 1, | |
705 | .needs_file = 1, | |
706 | .unbound_nonreg_file = 1, | |
8a72758c | 707 | .pollin = 1, |
4d954c25 | 708 | .buffer_select = 1, |
d3656344 | 709 | }, |
0463b6c5 | 710 | [IORING_OP_WRITEV] = { |
d3656344 JA |
711 | .async_ctx = 1, |
712 | .needs_mm = 1, | |
713 | .needs_file = 1, | |
714 | .hash_reg_file = 1, | |
715 | .unbound_nonreg_file = 1, | |
8a72758c | 716 | .pollout = 1, |
d3656344 | 717 | }, |
0463b6c5 | 718 | [IORING_OP_FSYNC] = { |
d3656344 JA |
719 | .needs_file = 1, |
720 | }, | |
0463b6c5 | 721 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
722 | .needs_file = 1, |
723 | .unbound_nonreg_file = 1, | |
8a72758c | 724 | .pollin = 1, |
d3656344 | 725 | }, |
0463b6c5 | 726 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
727 | .needs_file = 1, |
728 | .hash_reg_file = 1, | |
729 | .unbound_nonreg_file = 1, | |
8a72758c | 730 | .pollout = 1, |
d3656344 | 731 | }, |
0463b6c5 | 732 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
733 | .needs_file = 1, |
734 | .unbound_nonreg_file = 1, | |
735 | }, | |
0463b6c5 PB |
736 | [IORING_OP_POLL_REMOVE] = {}, |
737 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
738 | .needs_file = 1, |
739 | }, | |
0463b6c5 | 740 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
741 | .async_ctx = 1, |
742 | .needs_mm = 1, | |
743 | .needs_file = 1, | |
744 | .unbound_nonreg_file = 1, | |
ff002b30 | 745 | .needs_fs = 1, |
8a72758c | 746 | .pollout = 1, |
d3656344 | 747 | }, |
0463b6c5 | 748 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
749 | .async_ctx = 1, |
750 | .needs_mm = 1, | |
751 | .needs_file = 1, | |
752 | .unbound_nonreg_file = 1, | |
ff002b30 | 753 | .needs_fs = 1, |
8a72758c | 754 | .pollin = 1, |
52de1fe1 | 755 | .buffer_select = 1, |
d3656344 | 756 | }, |
0463b6c5 | 757 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
758 | .async_ctx = 1, |
759 | .needs_mm = 1, | |
760 | }, | |
0463b6c5 PB |
761 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
762 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
763 | .needs_mm = 1, |
764 | .needs_file = 1, | |
765 | .unbound_nonreg_file = 1, | |
f86cd20c | 766 | .file_table = 1, |
8a72758c | 767 | .pollin = 1, |
d3656344 | 768 | }, |
0463b6c5 PB |
769 | [IORING_OP_ASYNC_CANCEL] = {}, |
770 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
771 | .async_ctx = 1, |
772 | .needs_mm = 1, | |
773 | }, | |
0463b6c5 | 774 | [IORING_OP_CONNECT] = { |
d3656344 JA |
775 | .async_ctx = 1, |
776 | .needs_mm = 1, | |
777 | .needs_file = 1, | |
778 | .unbound_nonreg_file = 1, | |
8a72758c | 779 | .pollout = 1, |
d3656344 | 780 | }, |
0463b6c5 | 781 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
782 | .needs_file = 1, |
783 | }, | |
0463b6c5 | 784 | [IORING_OP_OPENAT] = { |
d3656344 JA |
785 | .needs_file = 1, |
786 | .fd_non_neg = 1, | |
f86cd20c | 787 | .file_table = 1, |
ff002b30 | 788 | .needs_fs = 1, |
d3656344 | 789 | }, |
0463b6c5 | 790 | [IORING_OP_CLOSE] = { |
d3656344 | 791 | .needs_file = 1, |
f86cd20c | 792 | .file_table = 1, |
d3656344 | 793 | }, |
0463b6c5 | 794 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 795 | .needs_mm = 1, |
f86cd20c | 796 | .file_table = 1, |
d3656344 | 797 | }, |
0463b6c5 | 798 | [IORING_OP_STATX] = { |
d3656344 JA |
799 | .needs_mm = 1, |
800 | .needs_file = 1, | |
801 | .fd_non_neg = 1, | |
ff002b30 | 802 | .needs_fs = 1, |
d3656344 | 803 | }, |
0463b6c5 | 804 | [IORING_OP_READ] = { |
3a6820f2 JA |
805 | .needs_mm = 1, |
806 | .needs_file = 1, | |
807 | .unbound_nonreg_file = 1, | |
8a72758c | 808 | .pollin = 1, |
bcda7baa | 809 | .buffer_select = 1, |
3a6820f2 | 810 | }, |
0463b6c5 | 811 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
812 | .needs_mm = 1, |
813 | .needs_file = 1, | |
814 | .unbound_nonreg_file = 1, | |
8a72758c | 815 | .pollout = 1, |
3a6820f2 | 816 | }, |
0463b6c5 | 817 | [IORING_OP_FADVISE] = { |
4840e418 JA |
818 | .needs_file = 1, |
819 | }, | |
0463b6c5 | 820 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
821 | .needs_mm = 1, |
822 | }, | |
0463b6c5 | 823 | [IORING_OP_SEND] = { |
fddaface JA |
824 | .needs_mm = 1, |
825 | .needs_file = 1, | |
826 | .unbound_nonreg_file = 1, | |
8a72758c | 827 | .pollout = 1, |
fddaface | 828 | }, |
0463b6c5 | 829 | [IORING_OP_RECV] = { |
fddaface JA |
830 | .needs_mm = 1, |
831 | .needs_file = 1, | |
832 | .unbound_nonreg_file = 1, | |
8a72758c | 833 | .pollin = 1, |
bcda7baa | 834 | .buffer_select = 1, |
fddaface | 835 | }, |
0463b6c5 | 836 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
837 | .needs_file = 1, |
838 | .fd_non_neg = 1, | |
f86cd20c | 839 | .file_table = 1, |
ff002b30 | 840 | .needs_fs = 1, |
cebdb986 | 841 | }, |
3e4827b0 JA |
842 | [IORING_OP_EPOLL_CTL] = { |
843 | .unbound_nonreg_file = 1, | |
844 | .file_table = 1, | |
845 | }, | |
7d67af2c PB |
846 | [IORING_OP_SPLICE] = { |
847 | .needs_file = 1, | |
848 | .hash_reg_file = 1, | |
849 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
850 | }, |
851 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 852 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
853 | }; |
854 | ||
561fb04a | 855 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 856 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 857 | static void io_put_req(struct io_kiocb *req); |
978db57e | 858 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
859 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
860 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
861 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
862 | struct io_uring_files_update *ip, | |
863 | unsigned nr_args); | |
f86cd20c | 864 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 865 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
866 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
867 | int fd, struct file **out_file, bool fixed); | |
868 | static void __io_queue_sqe(struct io_kiocb *req, | |
869 | const struct io_uring_sqe *sqe); | |
de0617e4 | 870 | |
2b188cc1 JA |
871 | static struct kmem_cache *req_cachep; |
872 | ||
873 | static const struct file_operations io_uring_fops; | |
874 | ||
875 | struct sock *io_uring_get_socket(struct file *file) | |
876 | { | |
877 | #if defined(CONFIG_UNIX) | |
878 | if (file->f_op == &io_uring_fops) { | |
879 | struct io_ring_ctx *ctx = file->private_data; | |
880 | ||
881 | return ctx->ring_sock->sk; | |
882 | } | |
883 | #endif | |
884 | return NULL; | |
885 | } | |
886 | EXPORT_SYMBOL(io_uring_get_socket); | |
887 | ||
888 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
889 | { | |
890 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
891 | ||
206aefde | 892 | complete(&ctx->completions[0]); |
2b188cc1 JA |
893 | } |
894 | ||
895 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
896 | { | |
897 | struct io_ring_ctx *ctx; | |
78076bb6 | 898 | int hash_bits; |
2b188cc1 JA |
899 | |
900 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
901 | if (!ctx) | |
902 | return NULL; | |
903 | ||
0ddf92e8 JA |
904 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
905 | if (!ctx->fallback_req) | |
906 | goto err; | |
907 | ||
206aefde JA |
908 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
909 | if (!ctx->completions) | |
910 | goto err; | |
911 | ||
78076bb6 JA |
912 | /* |
913 | * Use 5 bits less than the max cq entries, that should give us around | |
914 | * 32 entries per hash list if totally full and uniformly spread. | |
915 | */ | |
916 | hash_bits = ilog2(p->cq_entries); | |
917 | hash_bits -= 5; | |
918 | if (hash_bits <= 0) | |
919 | hash_bits = 1; | |
920 | ctx->cancel_hash_bits = hash_bits; | |
921 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
922 | GFP_KERNEL); | |
923 | if (!ctx->cancel_hash) | |
924 | goto err; | |
925 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
926 | ||
21482896 | 927 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
928 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
929 | goto err; | |
2b188cc1 JA |
930 | |
931 | ctx->flags = p->flags; | |
932 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 933 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
934 | init_completion(&ctx->completions[0]); |
935 | init_completion(&ctx->completions[1]); | |
5a2e745d | 936 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 937 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
938 | mutex_init(&ctx->uring_lock); |
939 | init_waitqueue_head(&ctx->wait); | |
940 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 941 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 942 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 943 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
944 | init_waitqueue_head(&ctx->inflight_wait); |
945 | spin_lock_init(&ctx->inflight_lock); | |
946 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 947 | return ctx; |
206aefde | 948 | err: |
0ddf92e8 JA |
949 | if (ctx->fallback_req) |
950 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 951 | kfree(ctx->completions); |
78076bb6 | 952 | kfree(ctx->cancel_hash); |
206aefde JA |
953 | kfree(ctx); |
954 | return NULL; | |
2b188cc1 JA |
955 | } |
956 | ||
9d858b21 | 957 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 958 | { |
a197f664 JL |
959 | struct io_ring_ctx *ctx = req->ctx; |
960 | ||
498ccd9e JA |
961 | return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped |
962 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
963 | } |
964 | ||
9d858b21 | 965 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 966 | { |
87987898 | 967 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 968 | return __req_need_defer(req); |
de0617e4 | 969 | |
9d858b21 | 970 | return false; |
de0617e4 JA |
971 | } |
972 | ||
7adf4eaf | 973 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
974 | { |
975 | struct io_kiocb *req; | |
976 | ||
7adf4eaf | 977 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 978 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
979 | list_del_init(&req->list); |
980 | return req; | |
981 | } | |
982 | ||
983 | return NULL; | |
984 | } | |
985 | ||
5262f567 JA |
986 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
987 | { | |
7adf4eaf JA |
988 | struct io_kiocb *req; |
989 | ||
990 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
991 | if (req) { |
992 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
993 | return NULL; | |
fb4b3d3f | 994 | if (!__req_need_defer(req)) { |
93bd25bb JA |
995 | list_del_init(&req->list); |
996 | return req; | |
997 | } | |
7adf4eaf JA |
998 | } |
999 | ||
1000 | return NULL; | |
5262f567 JA |
1001 | } |
1002 | ||
de0617e4 | 1003 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1004 | { |
75b28aff | 1005 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1006 | |
07910158 PB |
1007 | /* order cqe stores with ring update */ |
1008 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1009 | |
07910158 PB |
1010 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1011 | wake_up_interruptible(&ctx->cq_wait); | |
1012 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1013 | } |
1014 | } | |
1015 | ||
cccf0ee8 JA |
1016 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1017 | const struct io_op_def *def) | |
1018 | { | |
1019 | if (!req->work.mm && def->needs_mm) { | |
1020 | mmgrab(current->mm); | |
1021 | req->work.mm = current->mm; | |
2b188cc1 | 1022 | } |
cccf0ee8 JA |
1023 | if (!req->work.creds) |
1024 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1025 | if (!req->work.fs && def->needs_fs) { |
1026 | spin_lock(¤t->fs->lock); | |
1027 | if (!current->fs->in_exec) { | |
1028 | req->work.fs = current->fs; | |
1029 | req->work.fs->users++; | |
1030 | } else { | |
1031 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1032 | } | |
1033 | spin_unlock(¤t->fs->lock); | |
1034 | } | |
6ab23144 JA |
1035 | if (!req->work.task_pid) |
1036 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1037 | } |
1038 | ||
cccf0ee8 | 1039 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1040 | { |
cccf0ee8 JA |
1041 | if (req->work.mm) { |
1042 | mmdrop(req->work.mm); | |
1043 | req->work.mm = NULL; | |
1044 | } | |
1045 | if (req->work.creds) { | |
1046 | put_cred(req->work.creds); | |
1047 | req->work.creds = NULL; | |
1048 | } | |
ff002b30 JA |
1049 | if (req->work.fs) { |
1050 | struct fs_struct *fs = req->work.fs; | |
1051 | ||
1052 | spin_lock(&req->work.fs->lock); | |
1053 | if (--fs->users) | |
1054 | fs = NULL; | |
1055 | spin_unlock(&req->work.fs->lock); | |
1056 | if (fs) | |
1057 | free_fs_struct(fs); | |
1058 | } | |
561fb04a JA |
1059 | } |
1060 | ||
8766dd51 | 1061 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1062 | struct io_kiocb **link) |
18d9be1a | 1063 | { |
d3656344 | 1064 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1065 | |
d3656344 JA |
1066 | if (req->flags & REQ_F_ISREG) { |
1067 | if (def->hash_reg_file) | |
8766dd51 | 1068 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1069 | } else { |
1070 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1071 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1072 | } |
cccf0ee8 JA |
1073 | |
1074 | io_req_work_grab_env(req, def); | |
54a91f3b | 1075 | |
94ae5e77 | 1076 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1077 | } |
1078 | ||
a197f664 | 1079 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1080 | { |
a197f664 | 1081 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1082 | struct io_kiocb *link; |
94ae5e77 | 1083 | |
8766dd51 | 1084 | io_prep_async_work(req, &link); |
561fb04a | 1085 | |
8766dd51 PB |
1086 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1087 | &req->work, req->flags); | |
1088 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1089 | |
1090 | if (link) | |
1091 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1092 | } |
1093 | ||
5262f567 JA |
1094 | static void io_kill_timeout(struct io_kiocb *req) |
1095 | { | |
1096 | int ret; | |
1097 | ||
2d28390a | 1098 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1099 | if (ret != -1) { |
1100 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1101 | list_del_init(&req->list); |
f0e20b89 | 1102 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1103 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1104 | io_put_req(req); |
5262f567 JA |
1105 | } |
1106 | } | |
1107 | ||
1108 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1109 | { | |
1110 | struct io_kiocb *req, *tmp; | |
1111 | ||
1112 | spin_lock_irq(&ctx->completion_lock); | |
1113 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1114 | io_kill_timeout(req); | |
1115 | spin_unlock_irq(&ctx->completion_lock); | |
1116 | } | |
1117 | ||
de0617e4 JA |
1118 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1119 | { | |
1120 | struct io_kiocb *req; | |
1121 | ||
5262f567 JA |
1122 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1123 | io_kill_timeout(req); | |
1124 | ||
de0617e4 JA |
1125 | __io_commit_cqring(ctx); |
1126 | ||
87987898 | 1127 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1128 | io_queue_async_work(req); |
de0617e4 JA |
1129 | } |
1130 | ||
2b188cc1 JA |
1131 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1132 | { | |
75b28aff | 1133 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1134 | unsigned tail; |
1135 | ||
1136 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1137 | /* |
1138 | * writes to the cq entry need to come after reading head; the | |
1139 | * control dependency is enough as we're using WRITE_ONCE to | |
1140 | * fill the cq entry | |
1141 | */ | |
75b28aff | 1142 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1143 | return NULL; |
1144 | ||
1145 | ctx->cached_cq_tail++; | |
75b28aff | 1146 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1147 | } |
1148 | ||
f2842ab5 JA |
1149 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1150 | { | |
f0b493e6 JA |
1151 | if (!ctx->cq_ev_fd) |
1152 | return false; | |
f2842ab5 JA |
1153 | if (!ctx->eventfd_async) |
1154 | return true; | |
b41e9852 | 1155 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1156 | } |
1157 | ||
b41e9852 | 1158 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1159 | { |
1160 | if (waitqueue_active(&ctx->wait)) | |
1161 | wake_up(&ctx->wait); | |
1162 | if (waitqueue_active(&ctx->sqo_wait)) | |
1163 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1164 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1165 | eventfd_signal(ctx->cq_ev_fd, 1); |
1166 | } | |
1167 | ||
c4a2ed72 JA |
1168 | /* Returns true if there are no backlogged entries after the flush */ |
1169 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1170 | { |
1171 | struct io_rings *rings = ctx->rings; | |
1172 | struct io_uring_cqe *cqe; | |
1173 | struct io_kiocb *req; | |
1174 | unsigned long flags; | |
1175 | LIST_HEAD(list); | |
1176 | ||
1177 | if (!force) { | |
1178 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1179 | return true; |
1d7bb1d5 JA |
1180 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1181 | rings->cq_ring_entries)) | |
c4a2ed72 | 1182 | return false; |
1d7bb1d5 JA |
1183 | } |
1184 | ||
1185 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1186 | ||
1187 | /* if force is set, the ring is going away. always drop after that */ | |
1188 | if (force) | |
69b3e546 | 1189 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1190 | |
c4a2ed72 | 1191 | cqe = NULL; |
1d7bb1d5 JA |
1192 | while (!list_empty(&ctx->cq_overflow_list)) { |
1193 | cqe = io_get_cqring(ctx); | |
1194 | if (!cqe && !force) | |
1195 | break; | |
1196 | ||
1197 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1198 | list); | |
1199 | list_move(&req->list, &list); | |
2ca10259 | 1200 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1201 | if (cqe) { |
1202 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1203 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1204 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1205 | } else { |
1206 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1207 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1208 | } | |
1209 | } | |
1210 | ||
1211 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1212 | if (cqe) { |
1213 | clear_bit(0, &ctx->sq_check_overflow); | |
1214 | clear_bit(0, &ctx->cq_check_overflow); | |
1215 | } | |
1d7bb1d5 JA |
1216 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1217 | io_cqring_ev_posted(ctx); | |
1218 | ||
1219 | while (!list_empty(&list)) { | |
1220 | req = list_first_entry(&list, struct io_kiocb, list); | |
1221 | list_del(&req->list); | |
ec9c02ad | 1222 | io_put_req(req); |
1d7bb1d5 | 1223 | } |
c4a2ed72 JA |
1224 | |
1225 | return cqe != NULL; | |
1d7bb1d5 JA |
1226 | } |
1227 | ||
bcda7baa | 1228 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1229 | { |
78e19bbe | 1230 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1231 | struct io_uring_cqe *cqe; |
1232 | ||
78e19bbe | 1233 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1234 | |
2b188cc1 JA |
1235 | /* |
1236 | * If we can't get a cq entry, userspace overflowed the | |
1237 | * submission (by quite a lot). Increment the overflow count in | |
1238 | * the ring. | |
1239 | */ | |
1240 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1241 | if (likely(cqe)) { |
78e19bbe | 1242 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1243 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1244 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1245 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1246 | WRITE_ONCE(ctx->rings->cq_overflow, |
1247 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1248 | } else { |
ad3eb2c8 JA |
1249 | if (list_empty(&ctx->cq_overflow_list)) { |
1250 | set_bit(0, &ctx->sq_check_overflow); | |
1251 | set_bit(0, &ctx->cq_check_overflow); | |
1252 | } | |
2ca10259 | 1253 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1254 | refcount_inc(&req->refs); |
1255 | req->result = res; | |
bcda7baa | 1256 | req->cflags = cflags; |
1d7bb1d5 | 1257 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1258 | } |
1259 | } | |
1260 | ||
bcda7baa JA |
1261 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1262 | { | |
1263 | __io_cqring_fill_event(req, res, 0); | |
1264 | } | |
1265 | ||
1266 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1267 | { |
78e19bbe | 1268 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1269 | unsigned long flags; |
1270 | ||
1271 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1272 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1273 | io_commit_cqring(ctx); |
1274 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1275 | ||
8c838788 | 1276 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1277 | } |
1278 | ||
bcda7baa JA |
1279 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1280 | { | |
1281 | __io_cqring_add_event(req, res, 0); | |
1282 | } | |
1283 | ||
0ddf92e8 JA |
1284 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1285 | { | |
1286 | return req == (struct io_kiocb *) | |
1287 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1288 | } | |
1289 | ||
1290 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1291 | { | |
1292 | struct io_kiocb *req; | |
1293 | ||
1294 | req = ctx->fallback_req; | |
1295 | if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req)) | |
1296 | return req; | |
1297 | ||
1298 | return NULL; | |
1299 | } | |
1300 | ||
0553b8bd PB |
1301 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1302 | struct io_submit_state *state) | |
2b188cc1 | 1303 | { |
fd6fab2c | 1304 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1305 | struct io_kiocb *req; |
1306 | ||
2579f913 | 1307 | if (!state) { |
fd6fab2c | 1308 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1309 | if (unlikely(!req)) |
0ddf92e8 | 1310 | goto fallback; |
2579f913 JA |
1311 | } else if (!state->free_reqs) { |
1312 | size_t sz; | |
1313 | int ret; | |
1314 | ||
1315 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1316 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1317 | ||
1318 | /* | |
1319 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1320 | * retry single alloc to be on the safe side. | |
1321 | */ | |
1322 | if (unlikely(ret <= 0)) { | |
1323 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1324 | if (!state->reqs[0]) | |
0ddf92e8 | 1325 | goto fallback; |
fd6fab2c JA |
1326 | ret = 1; |
1327 | } | |
2579f913 | 1328 | state->free_reqs = ret - 1; |
6c8a3134 | 1329 | req = state->reqs[ret - 1]; |
2579f913 | 1330 | } else { |
2579f913 | 1331 | state->free_reqs--; |
6c8a3134 | 1332 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1333 | } |
1334 | ||
2579f913 | 1335 | return req; |
0ddf92e8 | 1336 | fallback: |
0553b8bd | 1337 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1338 | } |
1339 | ||
8da11c19 PB |
1340 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1341 | bool fixed) | |
1342 | { | |
1343 | if (fixed) | |
05589553 | 1344 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1345 | else |
1346 | fput(file); | |
1347 | } | |
1348 | ||
c6ca97b3 | 1349 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1350 | { |
929a3af9 PB |
1351 | if (req->flags & REQ_F_NEED_CLEANUP) |
1352 | io_cleanup_req(req); | |
1353 | ||
96fd84d8 | 1354 | kfree(req->io); |
8da11c19 PB |
1355 | if (req->file) |
1356 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1357 | if (req->task) |
1358 | put_task_struct(req->task); | |
cccf0ee8 JA |
1359 | |
1360 | io_req_work_drop_env(req); | |
def596e9 JA |
1361 | } |
1362 | ||
9e645e11 | 1363 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1364 | { |
c6ca97b3 | 1365 | __io_req_aux_free(req); |
fcb323cc | 1366 | |
fcb323cc | 1367 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1368 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1369 | unsigned long flags; |
1370 | ||
1371 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1372 | list_del(&req->inflight_entry); | |
1373 | if (waitqueue_active(&ctx->inflight_wait)) | |
1374 | wake_up(&ctx->inflight_wait); | |
1375 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1376 | } | |
2b85edfc PB |
1377 | |
1378 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1379 | if (likely(!io_is_fallback_req(req))) |
1380 | kmem_cache_free(req_cachep, req); | |
1381 | else | |
1382 | clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req); | |
e65ef56d JA |
1383 | } |
1384 | ||
c6ca97b3 JA |
1385 | struct req_batch { |
1386 | void *reqs[IO_IOPOLL_BATCH]; | |
1387 | int to_free; | |
1388 | int need_iter; | |
1389 | }; | |
1390 | ||
1391 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1392 | { | |
1393 | if (!rb->to_free) | |
1394 | return; | |
1395 | if (rb->need_iter) { | |
1396 | int i, inflight = 0; | |
1397 | unsigned long flags; | |
1398 | ||
1399 | for (i = 0; i < rb->to_free; i++) { | |
1400 | struct io_kiocb *req = rb->reqs[i]; | |
1401 | ||
10fef4be | 1402 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1403 | req->file = NULL; |
05589553 | 1404 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1405 | } |
c6ca97b3 JA |
1406 | if (req->flags & REQ_F_INFLIGHT) |
1407 | inflight++; | |
c6ca97b3 JA |
1408 | __io_req_aux_free(req); |
1409 | } | |
1410 | if (!inflight) | |
1411 | goto do_free; | |
1412 | ||
1413 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1414 | for (i = 0; i < rb->to_free; i++) { | |
1415 | struct io_kiocb *req = rb->reqs[i]; | |
1416 | ||
10fef4be | 1417 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1418 | list_del(&req->inflight_entry); |
1419 | if (!--inflight) | |
1420 | break; | |
1421 | } | |
1422 | } | |
1423 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1424 | ||
1425 | if (waitqueue_active(&ctx->inflight_wait)) | |
1426 | wake_up(&ctx->inflight_wait); | |
1427 | } | |
1428 | do_free: | |
1429 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1430 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1431 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1432 | } |
1433 | ||
a197f664 | 1434 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1435 | { |
a197f664 | 1436 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1437 | int ret; |
1438 | ||
2d28390a | 1439 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1440 | if (ret != -1) { |
78e19bbe | 1441 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1442 | io_commit_cqring(ctx); |
dea3b49c | 1443 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1444 | io_put_req(req); |
2665abfd JA |
1445 | return true; |
1446 | } | |
1447 | ||
1448 | return false; | |
e65ef56d JA |
1449 | } |
1450 | ||
ba816ad6 | 1451 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1452 | { |
2665abfd | 1453 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1454 | bool wake_ev = false; |
9e645e11 | 1455 | |
4d7dd462 JA |
1456 | /* Already got next link */ |
1457 | if (req->flags & REQ_F_LINK_NEXT) | |
1458 | return; | |
1459 | ||
9e645e11 JA |
1460 | /* |
1461 | * The list should never be empty when we are called here. But could | |
1462 | * potentially happen if the chain is messed up, check to be on the | |
1463 | * safe side. | |
1464 | */ | |
4493233e PB |
1465 | while (!list_empty(&req->link_list)) { |
1466 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1467 | struct io_kiocb, link_list); | |
94ae5e77 | 1468 | |
4493233e PB |
1469 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1470 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1471 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1472 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1473 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1474 | continue; | |
1475 | } | |
9e645e11 | 1476 | |
4493233e PB |
1477 | list_del_init(&req->link_list); |
1478 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1479 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1480 | *nxtptr = nxt; |
94ae5e77 | 1481 | break; |
9e645e11 | 1482 | } |
2665abfd | 1483 | |
4d7dd462 | 1484 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1485 | if (wake_ev) |
1486 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1487 | } |
1488 | ||
1489 | /* | |
dea3b49c | 1490 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1491 | */ |
1492 | static void io_fail_links(struct io_kiocb *req) | |
1493 | { | |
2665abfd | 1494 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1495 | unsigned long flags; |
1496 | ||
1497 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1498 | |
1499 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1500 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1501 | struct io_kiocb, link_list); | |
9e645e11 | 1502 | |
4493233e | 1503 | list_del_init(&link->link_list); |
c826bd7a | 1504 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1505 | |
1506 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1507 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1508 | io_link_cancel_timeout(link); |
2665abfd | 1509 | } else { |
78e19bbe | 1510 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1511 | __io_double_put_req(link); |
2665abfd | 1512 | } |
5d960724 | 1513 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1514 | } |
2665abfd JA |
1515 | |
1516 | io_commit_cqring(ctx); | |
1517 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1518 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1519 | } |
1520 | ||
4d7dd462 | 1521 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1522 | { |
dea3b49c | 1523 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1524 | return; |
2665abfd | 1525 | |
9e645e11 JA |
1526 | /* |
1527 | * If LINK is set, we have dependent requests in this chain. If we | |
1528 | * didn't fail this request, queue the first one up, moving any other | |
1529 | * dependencies to the next request. In case of failure, fail the rest | |
1530 | * of the chain. | |
1531 | */ | |
2665abfd JA |
1532 | if (req->flags & REQ_F_FAIL_LINK) { |
1533 | io_fail_links(req); | |
7c9e7f0f JA |
1534 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1535 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1536 | struct io_ring_ctx *ctx = req->ctx; |
1537 | unsigned long flags; | |
1538 | ||
1539 | /* | |
1540 | * If this is a timeout link, we could be racing with the | |
1541 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1542 | * protect against that. |
2665abfd JA |
1543 | */ |
1544 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1545 | io_req_link_next(req, nxt); | |
1546 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1547 | } else { | |
1548 | io_req_link_next(req, nxt); | |
9e645e11 | 1549 | } |
4d7dd462 | 1550 | } |
9e645e11 | 1551 | |
c69f8dbe JL |
1552 | static void io_free_req(struct io_kiocb *req) |
1553 | { | |
944e58bf PB |
1554 | struct io_kiocb *nxt = NULL; |
1555 | ||
1556 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1557 | __io_free_req(req); |
944e58bf PB |
1558 | |
1559 | if (nxt) | |
1560 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1561 | } |
1562 | ||
7a743e22 PB |
1563 | static void io_link_work_cb(struct io_wq_work **workptr) |
1564 | { | |
18a542ff PB |
1565 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1566 | struct io_kiocb *link; | |
7a743e22 | 1567 | |
18a542ff | 1568 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1569 | io_queue_linked_timeout(link); |
1570 | io_wq_submit_work(workptr); | |
1571 | } | |
1572 | ||
1573 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1574 | { | |
1575 | struct io_kiocb *link; | |
8766dd51 PB |
1576 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1577 | ||
1578 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1579 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1580 | |
1581 | *workptr = &nxt->work; | |
1582 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1583 | if (link) |
7a743e22 | 1584 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1585 | } |
1586 | ||
ba816ad6 JA |
1587 | /* |
1588 | * Drop reference to request, return next in chain (if there is one) if this | |
1589 | * was the last reference to this request. | |
1590 | */ | |
f9bd67f6 | 1591 | __attribute__((nonnull)) |
ec9c02ad | 1592 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1593 | { |
2a44f467 JA |
1594 | if (refcount_dec_and_test(&req->refs)) { |
1595 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1596 | __io_free_req(req); |
2a44f467 | 1597 | } |
2b188cc1 JA |
1598 | } |
1599 | ||
e65ef56d JA |
1600 | static void io_put_req(struct io_kiocb *req) |
1601 | { | |
1602 | if (refcount_dec_and_test(&req->refs)) | |
1603 | io_free_req(req); | |
2b188cc1 JA |
1604 | } |
1605 | ||
e9fd9396 PB |
1606 | static void io_steal_work(struct io_kiocb *req, |
1607 | struct io_wq_work **workptr) | |
7a743e22 PB |
1608 | { |
1609 | /* | |
1610 | * It's in an io-wq worker, so there always should be at least | |
1611 | * one reference, which will be dropped in io_put_work() just | |
1612 | * after the current handler returns. | |
1613 | * | |
1614 | * It also means, that if the counter dropped to 1, then there is | |
1615 | * no asynchronous users left, so it's safe to steal the next work. | |
1616 | */ | |
7a743e22 PB |
1617 | if (refcount_read(&req->refs) == 1) { |
1618 | struct io_kiocb *nxt = NULL; | |
1619 | ||
1620 | io_req_find_next(req, &nxt); | |
1621 | if (nxt) | |
1622 | io_wq_assign_next(workptr, nxt); | |
1623 | } | |
1624 | } | |
1625 | ||
978db57e JA |
1626 | /* |
1627 | * Must only be used if we don't need to care about links, usually from | |
1628 | * within the completion handling itself. | |
1629 | */ | |
1630 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1631 | { |
1632 | /* drop both submit and complete references */ | |
1633 | if (refcount_sub_and_test(2, &req->refs)) | |
1634 | __io_free_req(req); | |
1635 | } | |
1636 | ||
978db57e JA |
1637 | static void io_double_put_req(struct io_kiocb *req) |
1638 | { | |
1639 | /* drop both submit and complete references */ | |
1640 | if (refcount_sub_and_test(2, &req->refs)) | |
1641 | io_free_req(req); | |
1642 | } | |
1643 | ||
1d7bb1d5 | 1644 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1645 | { |
84f97dc2 JA |
1646 | struct io_rings *rings = ctx->rings; |
1647 | ||
ad3eb2c8 JA |
1648 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1649 | /* | |
1650 | * noflush == true is from the waitqueue handler, just ensure | |
1651 | * we wake up the task, and the next invocation will flush the | |
1652 | * entries. We cannot safely to it from here. | |
1653 | */ | |
1654 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1655 | return -1U; | |
1d7bb1d5 | 1656 | |
ad3eb2c8 JA |
1657 | io_cqring_overflow_flush(ctx, false); |
1658 | } | |
1d7bb1d5 | 1659 | |
a3a0e43f JA |
1660 | /* See comment at the top of this file */ |
1661 | smp_rmb(); | |
ad3eb2c8 | 1662 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1663 | } |
1664 | ||
fb5ccc98 PB |
1665 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1666 | { | |
1667 | struct io_rings *rings = ctx->rings; | |
1668 | ||
1669 | /* make sure SQ entry isn't read before tail */ | |
1670 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1671 | } | |
1672 | ||
8237e045 | 1673 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1674 | { |
dea3b49c | 1675 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1676 | return false; |
e94f141b | 1677 | |
c6ca97b3 JA |
1678 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1679 | rb->need_iter++; | |
1680 | ||
1681 | rb->reqs[rb->to_free++] = req; | |
1682 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1683 | io_free_req_many(req->ctx, rb); | |
1684 | return true; | |
e94f141b JA |
1685 | } |
1686 | ||
bcda7baa JA |
1687 | static int io_put_kbuf(struct io_kiocb *req) |
1688 | { | |
4d954c25 | 1689 | struct io_buffer *kbuf; |
bcda7baa JA |
1690 | int cflags; |
1691 | ||
4d954c25 | 1692 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1693 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1694 | cflags |= IORING_CQE_F_BUFFER; | |
1695 | req->rw.addr = 0; | |
1696 | kfree(kbuf); | |
1697 | return cflags; | |
1698 | } | |
1699 | ||
def596e9 JA |
1700 | /* |
1701 | * Find and free completed poll iocbs | |
1702 | */ | |
1703 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1704 | struct list_head *done) | |
1705 | { | |
8237e045 | 1706 | struct req_batch rb; |
def596e9 | 1707 | struct io_kiocb *req; |
def596e9 | 1708 | |
c6ca97b3 | 1709 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1710 | while (!list_empty(done)) { |
bcda7baa JA |
1711 | int cflags = 0; |
1712 | ||
def596e9 JA |
1713 | req = list_first_entry(done, struct io_kiocb, list); |
1714 | list_del(&req->list); | |
1715 | ||
bcda7baa JA |
1716 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1717 | cflags = io_put_kbuf(req); | |
1718 | ||
1719 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1720 | (*nr_events)++; |
1721 | ||
8237e045 JA |
1722 | if (refcount_dec_and_test(&req->refs) && |
1723 | !io_req_multi_free(&rb, req)) | |
1724 | io_free_req(req); | |
def596e9 | 1725 | } |
def596e9 | 1726 | |
09bb8394 | 1727 | io_commit_cqring(ctx); |
32b2244a XW |
1728 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1729 | io_cqring_ev_posted(ctx); | |
8237e045 | 1730 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1731 | } |
1732 | ||
581f9810 BM |
1733 | static void io_iopoll_queue(struct list_head *again) |
1734 | { | |
1735 | struct io_kiocb *req; | |
1736 | ||
1737 | do { | |
1738 | req = list_first_entry(again, struct io_kiocb, list); | |
1739 | list_del(&req->list); | |
1740 | refcount_inc(&req->refs); | |
1741 | io_queue_async_work(req); | |
1742 | } while (!list_empty(again)); | |
1743 | } | |
1744 | ||
def596e9 JA |
1745 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1746 | long min) | |
1747 | { | |
1748 | struct io_kiocb *req, *tmp; | |
1749 | LIST_HEAD(done); | |
581f9810 | 1750 | LIST_HEAD(again); |
def596e9 JA |
1751 | bool spin; |
1752 | int ret; | |
1753 | ||
1754 | /* | |
1755 | * Only spin for completions if we don't have multiple devices hanging | |
1756 | * off our complete list, and we're under the requested amount. | |
1757 | */ | |
1758 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1759 | ||
1760 | ret = 0; | |
1761 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1762 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1763 | |
1764 | /* | |
581f9810 BM |
1765 | * Move completed and retryable entries to our local lists. |
1766 | * If we find a request that requires polling, break out | |
1767 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1768 | */ |
1769 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1770 | list_move_tail(&req->list, &done); | |
1771 | continue; | |
1772 | } | |
1773 | if (!list_empty(&done)) | |
1774 | break; | |
1775 | ||
581f9810 BM |
1776 | if (req->result == -EAGAIN) { |
1777 | list_move_tail(&req->list, &again); | |
1778 | continue; | |
1779 | } | |
1780 | if (!list_empty(&again)) | |
1781 | break; | |
1782 | ||
def596e9 JA |
1783 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1784 | if (ret < 0) | |
1785 | break; | |
1786 | ||
1787 | if (ret && spin) | |
1788 | spin = false; | |
1789 | ret = 0; | |
1790 | } | |
1791 | ||
1792 | if (!list_empty(&done)) | |
1793 | io_iopoll_complete(ctx, nr_events, &done); | |
1794 | ||
581f9810 BM |
1795 | if (!list_empty(&again)) |
1796 | io_iopoll_queue(&again); | |
1797 | ||
def596e9 JA |
1798 | return ret; |
1799 | } | |
1800 | ||
1801 | /* | |
d195a66e | 1802 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1803 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1804 | * as a non-spinning completion check. | |
1805 | */ | |
1806 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1807 | long min) | |
1808 | { | |
08f5439f | 1809 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1810 | int ret; |
1811 | ||
1812 | ret = io_do_iopoll(ctx, nr_events, min); | |
1813 | if (ret < 0) | |
1814 | return ret; | |
1815 | if (!min || *nr_events >= min) | |
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | return 1; | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * We can't just wait for polled events to come to us, we have to actively | |
1824 | * find and complete them. | |
1825 | */ | |
1826 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1827 | { | |
1828 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1829 | return; | |
1830 | ||
1831 | mutex_lock(&ctx->uring_lock); | |
1832 | while (!list_empty(&ctx->poll_list)) { | |
1833 | unsigned int nr_events = 0; | |
1834 | ||
1835 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1836 | |
1837 | /* | |
1838 | * Ensure we allow local-to-the-cpu processing to take place, | |
1839 | * in this case we need to ensure that we reap all events. | |
1840 | */ | |
1841 | cond_resched(); | |
def596e9 JA |
1842 | } |
1843 | mutex_unlock(&ctx->uring_lock); | |
1844 | } | |
1845 | ||
c7849be9 XW |
1846 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1847 | long min) | |
def596e9 | 1848 | { |
2b2ed975 | 1849 | int iters = 0, ret = 0; |
500f9fba | 1850 | |
c7849be9 XW |
1851 | /* |
1852 | * We disallow the app entering submit/complete with polling, but we | |
1853 | * still need to lock the ring to prevent racing with polled issue | |
1854 | * that got punted to a workqueue. | |
1855 | */ | |
1856 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1857 | do { |
1858 | int tmin = 0; | |
1859 | ||
a3a0e43f JA |
1860 | /* |
1861 | * Don't enter poll loop if we already have events pending. | |
1862 | * If we do, we can potentially be spinning for commands that | |
1863 | * already triggered a CQE (eg in error). | |
1864 | */ | |
1d7bb1d5 | 1865 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1866 | break; |
1867 | ||
500f9fba JA |
1868 | /* |
1869 | * If a submit got punted to a workqueue, we can have the | |
1870 | * application entering polling for a command before it gets | |
1871 | * issued. That app will hold the uring_lock for the duration | |
1872 | * of the poll right here, so we need to take a breather every | |
1873 | * now and then to ensure that the issue has a chance to add | |
1874 | * the poll to the issued list. Otherwise we can spin here | |
1875 | * forever, while the workqueue is stuck trying to acquire the | |
1876 | * very same mutex. | |
1877 | */ | |
1878 | if (!(++iters & 7)) { | |
1879 | mutex_unlock(&ctx->uring_lock); | |
1880 | mutex_lock(&ctx->uring_lock); | |
1881 | } | |
1882 | ||
def596e9 JA |
1883 | if (*nr_events < min) |
1884 | tmin = min - *nr_events; | |
1885 | ||
1886 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1887 | if (ret <= 0) | |
1888 | break; | |
1889 | ret = 0; | |
1890 | } while (min && !*nr_events && !need_resched()); | |
1891 | ||
500f9fba | 1892 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1893 | return ret; |
1894 | } | |
1895 | ||
491381ce | 1896 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1897 | { |
491381ce JA |
1898 | /* |
1899 | * Tell lockdep we inherited freeze protection from submission | |
1900 | * thread. | |
1901 | */ | |
1902 | if (req->flags & REQ_F_ISREG) { | |
1903 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1904 | |
491381ce | 1905 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1906 | } |
491381ce | 1907 | file_end_write(req->file); |
2b188cc1 JA |
1908 | } |
1909 | ||
4e88d6e7 JA |
1910 | static inline void req_set_fail_links(struct io_kiocb *req) |
1911 | { | |
1912 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1913 | req->flags |= REQ_F_FAIL_LINK; | |
1914 | } | |
1915 | ||
ba816ad6 | 1916 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1917 | { |
9adbd45d | 1918 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1919 | int cflags = 0; |
2b188cc1 | 1920 | |
491381ce JA |
1921 | if (kiocb->ki_flags & IOCB_WRITE) |
1922 | kiocb_end_write(req); | |
2b188cc1 | 1923 | |
4e88d6e7 JA |
1924 | if (res != req->result) |
1925 | req_set_fail_links(req); | |
bcda7baa JA |
1926 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1927 | cflags = io_put_kbuf(req); | |
1928 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1929 | } |
1930 | ||
1931 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1932 | { | |
9adbd45d | 1933 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1934 | |
1935 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1936 | io_put_req(req); |
2b188cc1 JA |
1937 | } |
1938 | ||
def596e9 JA |
1939 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1940 | { | |
9adbd45d | 1941 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1942 | |
491381ce JA |
1943 | if (kiocb->ki_flags & IOCB_WRITE) |
1944 | kiocb_end_write(req); | |
def596e9 | 1945 | |
4e88d6e7 JA |
1946 | if (res != req->result) |
1947 | req_set_fail_links(req); | |
9e645e11 | 1948 | req->result = res; |
def596e9 JA |
1949 | if (res != -EAGAIN) |
1950 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1951 | } | |
1952 | ||
1953 | /* | |
1954 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1955 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1956 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1957 | * accessing the kiocb cookie. | |
1958 | */ | |
1959 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1960 | { | |
1961 | struct io_ring_ctx *ctx = req->ctx; | |
1962 | ||
1963 | /* | |
1964 | * Track whether we have multiple files in our lists. This will impact | |
1965 | * how we do polling eventually, not spinning if we're on potentially | |
1966 | * different devices. | |
1967 | */ | |
1968 | if (list_empty(&ctx->poll_list)) { | |
1969 | ctx->poll_multi_file = false; | |
1970 | } else if (!ctx->poll_multi_file) { | |
1971 | struct io_kiocb *list_req; | |
1972 | ||
1973 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1974 | list); | |
9adbd45d | 1975 | if (list_req->file != req->file) |
def596e9 JA |
1976 | ctx->poll_multi_file = true; |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * For fast devices, IO may have already completed. If it has, add | |
1981 | * it to the front so we find it first. | |
1982 | */ | |
1983 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1984 | list_add(&req->list, &ctx->poll_list); | |
1985 | else | |
1986 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1987 | |
1988 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1989 | wq_has_sleeper(&ctx->sqo_wait)) | |
1990 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1991 | } |
1992 | ||
3d6770fb | 1993 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1994 | { |
3d6770fb | 1995 | if (state->file) { |
9a56a232 JA |
1996 | int diff = state->has_refs - state->used_refs; |
1997 | ||
1998 | if (diff) | |
1999 | fput_many(state->file, diff); | |
2000 | state->file = NULL; | |
2001 | } | |
2002 | } | |
2003 | ||
2004 | /* | |
2005 | * Get as many references to a file as we have IOs left in this submission, | |
2006 | * assuming most submissions are for one file, or at least that each file | |
2007 | * has more than one submission. | |
2008 | */ | |
8da11c19 | 2009 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2010 | { |
2011 | if (!state) | |
2012 | return fget(fd); | |
2013 | ||
2014 | if (state->file) { | |
2015 | if (state->fd == fd) { | |
2016 | state->used_refs++; | |
2017 | state->ios_left--; | |
2018 | return state->file; | |
2019 | } | |
3d6770fb | 2020 | io_file_put(state); |
9a56a232 JA |
2021 | } |
2022 | state->file = fget_many(fd, state->ios_left); | |
2023 | if (!state->file) | |
2024 | return NULL; | |
2025 | ||
2026 | state->fd = fd; | |
2027 | state->has_refs = state->ios_left; | |
2028 | state->used_refs = 1; | |
2029 | state->ios_left--; | |
2030 | return state->file; | |
2031 | } | |
2032 | ||
2b188cc1 JA |
2033 | /* |
2034 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2035 | * any file. For now, just ensure that anything potentially problematic is done | |
2036 | * inline. | |
2037 | */ | |
2038 | static bool io_file_supports_async(struct file *file) | |
2039 | { | |
2040 | umode_t mode = file_inode(file)->i_mode; | |
2041 | ||
10d59345 | 2042 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2043 | return true; |
2044 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2045 | return true; | |
2046 | ||
2047 | return false; | |
2048 | } | |
2049 | ||
3529d8c2 JA |
2050 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2051 | bool force_nonblock) | |
2b188cc1 | 2052 | { |
def596e9 | 2053 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2054 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2055 | unsigned ioprio; |
2056 | int ret; | |
2b188cc1 | 2057 | |
491381ce JA |
2058 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2059 | req->flags |= REQ_F_ISREG; | |
2060 | ||
2b188cc1 | 2061 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2062 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2063 | req->flags |= REQ_F_CUR_POS; | |
2064 | kiocb->ki_pos = req->file->f_pos; | |
2065 | } | |
2b188cc1 | 2066 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2067 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2068 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2069 | if (unlikely(ret)) | |
2070 | return ret; | |
2b188cc1 JA |
2071 | |
2072 | ioprio = READ_ONCE(sqe->ioprio); | |
2073 | if (ioprio) { | |
2074 | ret = ioprio_check_cap(ioprio); | |
2075 | if (ret) | |
09bb8394 | 2076 | return ret; |
2b188cc1 JA |
2077 | |
2078 | kiocb->ki_ioprio = ioprio; | |
2079 | } else | |
2080 | kiocb->ki_ioprio = get_current_ioprio(); | |
2081 | ||
8449eeda | 2082 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2083 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2084 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2085 | req->flags |= REQ_F_NOWAIT; |
2086 | ||
2087 | if (force_nonblock) | |
2b188cc1 | 2088 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2089 | |
def596e9 | 2090 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2091 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2092 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2093 | return -EOPNOTSUPP; |
2b188cc1 | 2094 | |
def596e9 JA |
2095 | kiocb->ki_flags |= IOCB_HIPRI; |
2096 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2097 | req->result = 0; |
def596e9 | 2098 | } else { |
09bb8394 JA |
2099 | if (kiocb->ki_flags & IOCB_HIPRI) |
2100 | return -EINVAL; | |
def596e9 JA |
2101 | kiocb->ki_complete = io_complete_rw; |
2102 | } | |
9adbd45d | 2103 | |
3529d8c2 JA |
2104 | req->rw.addr = READ_ONCE(sqe->addr); |
2105 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2106 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2107 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2108 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2109 | return 0; |
2b188cc1 JA |
2110 | } |
2111 | ||
2112 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2113 | { | |
2114 | switch (ret) { | |
2115 | case -EIOCBQUEUED: | |
2116 | break; | |
2117 | case -ERESTARTSYS: | |
2118 | case -ERESTARTNOINTR: | |
2119 | case -ERESTARTNOHAND: | |
2120 | case -ERESTART_RESTARTBLOCK: | |
2121 | /* | |
2122 | * We can't just restart the syscall, since previously | |
2123 | * submitted sqes may already be in progress. Just fail this | |
2124 | * IO with EINTR. | |
2125 | */ | |
2126 | ret = -EINTR; | |
2127 | /* fall through */ | |
2128 | default: | |
2129 | kiocb->ki_complete(kiocb, ret, 0); | |
2130 | } | |
2131 | } | |
2132 | ||
014db007 | 2133 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2134 | { |
ba04291e JA |
2135 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2136 | ||
2137 | if (req->flags & REQ_F_CUR_POS) | |
2138 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2139 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2140 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2141 | else |
2142 | io_rw_done(kiocb, ret); | |
2143 | } | |
2144 | ||
9adbd45d | 2145 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2146 | struct iov_iter *iter) |
edafccee | 2147 | { |
9adbd45d JA |
2148 | struct io_ring_ctx *ctx = req->ctx; |
2149 | size_t len = req->rw.len; | |
edafccee JA |
2150 | struct io_mapped_ubuf *imu; |
2151 | unsigned index, buf_index; | |
2152 | size_t offset; | |
2153 | u64 buf_addr; | |
2154 | ||
2155 | /* attempt to use fixed buffers without having provided iovecs */ | |
2156 | if (unlikely(!ctx->user_bufs)) | |
2157 | return -EFAULT; | |
2158 | ||
9adbd45d | 2159 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2160 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2161 | return -EFAULT; | |
2162 | ||
2163 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2164 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2165 | buf_addr = req->rw.addr; |
edafccee JA |
2166 | |
2167 | /* overflow */ | |
2168 | if (buf_addr + len < buf_addr) | |
2169 | return -EFAULT; | |
2170 | /* not inside the mapped region */ | |
2171 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2172 | return -EFAULT; | |
2173 | ||
2174 | /* | |
2175 | * May not be a start of buffer, set size appropriately | |
2176 | * and advance us to the beginning. | |
2177 | */ | |
2178 | offset = buf_addr - imu->ubuf; | |
2179 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2180 | |
2181 | if (offset) { | |
2182 | /* | |
2183 | * Don't use iov_iter_advance() here, as it's really slow for | |
2184 | * using the latter parts of a big fixed buffer - it iterates | |
2185 | * over each segment manually. We can cheat a bit here, because | |
2186 | * we know that: | |
2187 | * | |
2188 | * 1) it's a BVEC iter, we set it up | |
2189 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2190 | * first and last bvec | |
2191 | * | |
2192 | * So just find our index, and adjust the iterator afterwards. | |
2193 | * If the offset is within the first bvec (or the whole first | |
2194 | * bvec, just use iov_iter_advance(). This makes it easier | |
2195 | * since we can just skip the first segment, which may not | |
2196 | * be PAGE_SIZE aligned. | |
2197 | */ | |
2198 | const struct bio_vec *bvec = imu->bvec; | |
2199 | ||
2200 | if (offset <= bvec->bv_len) { | |
2201 | iov_iter_advance(iter, offset); | |
2202 | } else { | |
2203 | unsigned long seg_skip; | |
2204 | ||
2205 | /* skip first vec */ | |
2206 | offset -= bvec->bv_len; | |
2207 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2208 | ||
2209 | iter->bvec = bvec + seg_skip; | |
2210 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2211 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2212 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2213 | } |
2214 | } | |
2215 | ||
5e559561 | 2216 | return len; |
edafccee JA |
2217 | } |
2218 | ||
bcda7baa JA |
2219 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2220 | { | |
2221 | if (needs_lock) | |
2222 | mutex_unlock(&ctx->uring_lock); | |
2223 | } | |
2224 | ||
2225 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2226 | { | |
2227 | /* | |
2228 | * "Normal" inline submissions always hold the uring_lock, since we | |
2229 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2230 | * The only exception is when we've detached the request and issue it | |
2231 | * from an async worker thread, grab the lock for that case. | |
2232 | */ | |
2233 | if (needs_lock) | |
2234 | mutex_lock(&ctx->uring_lock); | |
2235 | } | |
2236 | ||
2237 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2238 | int bgid, struct io_buffer *kbuf, | |
2239 | bool needs_lock) | |
2240 | { | |
2241 | struct io_buffer *head; | |
2242 | ||
2243 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2244 | return kbuf; | |
2245 | ||
2246 | io_ring_submit_lock(req->ctx, needs_lock); | |
2247 | ||
2248 | lockdep_assert_held(&req->ctx->uring_lock); | |
2249 | ||
2250 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2251 | if (head) { | |
2252 | if (!list_empty(&head->list)) { | |
2253 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2254 | list); | |
2255 | list_del(&kbuf->list); | |
2256 | } else { | |
2257 | kbuf = head; | |
2258 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2259 | } | |
2260 | if (*len > kbuf->len) | |
2261 | *len = kbuf->len; | |
2262 | } else { | |
2263 | kbuf = ERR_PTR(-ENOBUFS); | |
2264 | } | |
2265 | ||
2266 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2267 | ||
2268 | return kbuf; | |
2269 | } | |
2270 | ||
4d954c25 JA |
2271 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2272 | bool needs_lock) | |
2273 | { | |
2274 | struct io_buffer *kbuf; | |
2275 | int bgid; | |
2276 | ||
2277 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2278 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2279 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2280 | if (IS_ERR(kbuf)) | |
2281 | return kbuf; | |
2282 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2283 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2284 | return u64_to_user_ptr(kbuf->addr); | |
2285 | } | |
2286 | ||
2287 | #ifdef CONFIG_COMPAT | |
2288 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2289 | bool needs_lock) | |
2290 | { | |
2291 | struct compat_iovec __user *uiov; | |
2292 | compat_ssize_t clen; | |
2293 | void __user *buf; | |
2294 | ssize_t len; | |
2295 | ||
2296 | uiov = u64_to_user_ptr(req->rw.addr); | |
2297 | if (!access_ok(uiov, sizeof(*uiov))) | |
2298 | return -EFAULT; | |
2299 | if (__get_user(clen, &uiov->iov_len)) | |
2300 | return -EFAULT; | |
2301 | if (clen < 0) | |
2302 | return -EINVAL; | |
2303 | ||
2304 | len = clen; | |
2305 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2306 | if (IS_ERR(buf)) | |
2307 | return PTR_ERR(buf); | |
2308 | iov[0].iov_base = buf; | |
2309 | iov[0].iov_len = (compat_size_t) len; | |
2310 | return 0; | |
2311 | } | |
2312 | #endif | |
2313 | ||
2314 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2315 | bool needs_lock) | |
2316 | { | |
2317 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2318 | void __user *buf; | |
2319 | ssize_t len; | |
2320 | ||
2321 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2322 | return -EFAULT; | |
2323 | ||
2324 | len = iov[0].iov_len; | |
2325 | if (len < 0) | |
2326 | return -EINVAL; | |
2327 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2328 | if (IS_ERR(buf)) | |
2329 | return PTR_ERR(buf); | |
2330 | iov[0].iov_base = buf; | |
2331 | iov[0].iov_len = len; | |
2332 | return 0; | |
2333 | } | |
2334 | ||
2335 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2336 | bool needs_lock) | |
2337 | { | |
2338 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2339 | return 0; | |
2340 | if (!req->rw.len) | |
2341 | return 0; | |
2342 | else if (req->rw.len > 1) | |
2343 | return -EINVAL; | |
2344 | ||
2345 | #ifdef CONFIG_COMPAT | |
2346 | if (req->ctx->compat) | |
2347 | return io_compat_import(req, iov, needs_lock); | |
2348 | #endif | |
2349 | ||
2350 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2351 | } | |
2352 | ||
cf6fd4bd | 2353 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2354 | struct iovec **iovec, struct iov_iter *iter, |
2355 | bool needs_lock) | |
2b188cc1 | 2356 | { |
9adbd45d JA |
2357 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2358 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2359 | ssize_t ret; |
edafccee JA |
2360 | u8 opcode; |
2361 | ||
d625c6ee | 2362 | opcode = req->opcode; |
7d009165 | 2363 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2364 | *iovec = NULL; |
9adbd45d | 2365 | return io_import_fixed(req, rw, iter); |
edafccee | 2366 | } |
2b188cc1 | 2367 | |
bcda7baa JA |
2368 | /* buffer index only valid with fixed read/write, or buffer select */ |
2369 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2370 | return -EINVAL; |
2371 | ||
3a6820f2 | 2372 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2373 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2374 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2375 | if (IS_ERR(buf)) { | |
bcda7baa | 2376 | *iovec = NULL; |
4d954c25 | 2377 | return PTR_ERR(buf); |
bcda7baa | 2378 | } |
3f9d6441 | 2379 | req->rw.len = sqe_len; |
bcda7baa JA |
2380 | } |
2381 | ||
3a6820f2 JA |
2382 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2383 | *iovec = NULL; | |
3a901598 | 2384 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2385 | } |
2386 | ||
f67676d1 JA |
2387 | if (req->io) { |
2388 | struct io_async_rw *iorw = &req->io->rw; | |
2389 | ||
2390 | *iovec = iorw->iov; | |
2391 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2392 | if (iorw->iov == iorw->fast_iov) | |
2393 | *iovec = NULL; | |
2394 | return iorw->size; | |
2395 | } | |
2396 | ||
4d954c25 JA |
2397 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2398 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2399 | if (!ret) { |
2400 | ret = (*iovec)->iov_len; | |
2401 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2402 | } | |
4d954c25 JA |
2403 | *iovec = NULL; |
2404 | return ret; | |
2405 | } | |
2406 | ||
2b188cc1 | 2407 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2408 | if (req->ctx->compat) |
2b188cc1 JA |
2409 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2410 | iovec, iter); | |
2411 | #endif | |
2412 | ||
2413 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2414 | } | |
2415 | ||
31b51510 | 2416 | /* |
32960613 JA |
2417 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2418 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2419 | */ |
32960613 JA |
2420 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2421 | struct iov_iter *iter) | |
2422 | { | |
2423 | ssize_t ret = 0; | |
2424 | ||
2425 | /* | |
2426 | * Don't support polled IO through this interface, and we can't | |
2427 | * support non-blocking either. For the latter, this just causes | |
2428 | * the kiocb to be handled from an async context. | |
2429 | */ | |
2430 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2431 | return -EOPNOTSUPP; | |
2432 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2433 | return -EAGAIN; | |
2434 | ||
2435 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2436 | struct iovec iovec; |
32960613 JA |
2437 | ssize_t nr; |
2438 | ||
311ae9e1 PB |
2439 | if (!iov_iter_is_bvec(iter)) { |
2440 | iovec = iov_iter_iovec(iter); | |
2441 | } else { | |
2442 | /* fixed buffers import bvec */ | |
2443 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2444 | + iter->iov_offset; | |
2445 | iovec.iov_len = min(iter->count, | |
2446 | iter->bvec->bv_len - iter->iov_offset); | |
2447 | } | |
2448 | ||
32960613 JA |
2449 | if (rw == READ) { |
2450 | nr = file->f_op->read(file, iovec.iov_base, | |
2451 | iovec.iov_len, &kiocb->ki_pos); | |
2452 | } else { | |
2453 | nr = file->f_op->write(file, iovec.iov_base, | |
2454 | iovec.iov_len, &kiocb->ki_pos); | |
2455 | } | |
2456 | ||
311ae9e1 PB |
2457 | if (iov_iter_is_bvec(iter)) |
2458 | kunmap(iter->bvec->bv_page); | |
2459 | ||
32960613 JA |
2460 | if (nr < 0) { |
2461 | if (!ret) | |
2462 | ret = nr; | |
2463 | break; | |
2464 | } | |
2465 | ret += nr; | |
2466 | if (nr != iovec.iov_len) | |
2467 | break; | |
2468 | iov_iter_advance(iter, nr); | |
2469 | } | |
2470 | ||
2471 | return ret; | |
2472 | } | |
2473 | ||
b7bb4f7d | 2474 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2475 | struct iovec *iovec, struct iovec *fast_iov, |
2476 | struct iov_iter *iter) | |
2477 | { | |
2478 | req->io->rw.nr_segs = iter->nr_segs; | |
2479 | req->io->rw.size = io_size; | |
2480 | req->io->rw.iov = iovec; | |
2481 | if (!req->io->rw.iov) { | |
2482 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2483 | if (req->io->rw.iov != fast_iov) |
2484 | memcpy(req->io->rw.iov, fast_iov, | |
2485 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2486 | } else { |
2487 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2488 | } |
2489 | } | |
2490 | ||
3d9932a8 XW |
2491 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2492 | { | |
2493 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2494 | return req->io == NULL; | |
2495 | } | |
2496 | ||
b7bb4f7d | 2497 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2498 | { |
d3656344 JA |
2499 | if (!io_op_defs[req->opcode].async_ctx) |
2500 | return 0; | |
3d9932a8 XW |
2501 | |
2502 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2503 | } |
2504 | ||
b7bb4f7d JA |
2505 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2506 | struct iovec *iovec, struct iovec *fast_iov, | |
2507 | struct iov_iter *iter) | |
2508 | { | |
980ad263 | 2509 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2510 | return 0; |
5d204bcf | 2511 | if (!req->io) { |
3d9932a8 | 2512 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2513 | return -ENOMEM; |
b7bb4f7d | 2514 | |
5d204bcf JA |
2515 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2516 | } | |
b7bb4f7d | 2517 | return 0; |
f67676d1 JA |
2518 | } |
2519 | ||
3529d8c2 JA |
2520 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2521 | bool force_nonblock) | |
f67676d1 | 2522 | { |
3529d8c2 JA |
2523 | struct io_async_ctx *io; |
2524 | struct iov_iter iter; | |
f67676d1 JA |
2525 | ssize_t ret; |
2526 | ||
3529d8c2 JA |
2527 | ret = io_prep_rw(req, sqe, force_nonblock); |
2528 | if (ret) | |
2529 | return ret; | |
f67676d1 | 2530 | |
3529d8c2 JA |
2531 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2532 | return -EBADF; | |
f67676d1 | 2533 | |
5f798bea PB |
2534 | /* either don't need iovec imported or already have it */ |
2535 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2536 | return 0; |
2537 | ||
2538 | io = req->io; | |
2539 | io->rw.iov = io->rw.fast_iov; | |
2540 | req->io = NULL; | |
bcda7baa | 2541 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2542 | req->io = io; |
2543 | if (ret < 0) | |
2544 | return ret; | |
2545 | ||
2546 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2547 | return 0; | |
f67676d1 JA |
2548 | } |
2549 | ||
014db007 | 2550 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2551 | { |
2552 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2553 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2554 | struct iov_iter iter; |
31b51510 | 2555 | size_t iov_count; |
f67676d1 | 2556 | ssize_t io_size, ret; |
2b188cc1 | 2557 | |
bcda7baa | 2558 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2559 | if (ret < 0) |
2560 | return ret; | |
2b188cc1 | 2561 | |
fd6c2e4c JA |
2562 | /* Ensure we clear previously set non-block flag */ |
2563 | if (!force_nonblock) | |
29de5f6a | 2564 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2565 | |
797f3f53 | 2566 | req->result = 0; |
f67676d1 | 2567 | io_size = ret; |
dea3b49c | 2568 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2569 | req->result = io_size; |
2570 | ||
2571 | /* | |
2572 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2573 | * we know to async punt it even if it was opened O_NONBLOCK | |
2574 | */ | |
29de5f6a | 2575 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2576 | goto copy_iov; |
9e645e11 | 2577 | |
31b51510 | 2578 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2579 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2580 | if (!ret) { |
2581 | ssize_t ret2; | |
2582 | ||
9adbd45d JA |
2583 | if (req->file->f_op->read_iter) |
2584 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2585 | else |
9adbd45d | 2586 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2587 | |
9d93a3f5 | 2588 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2589 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2590 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2591 | } else { |
2592 | copy_iov: | |
b7bb4f7d | 2593 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2594 | inline_vecs, &iter); |
2595 | if (ret) | |
2596 | goto out_free; | |
29de5f6a JA |
2597 | /* any defer here is final, must blocking retry */ |
2598 | if (!(req->flags & REQ_F_NOWAIT)) | |
2599 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2600 | return -EAGAIN; |
2601 | } | |
2b188cc1 | 2602 | } |
f67676d1 | 2603 | out_free: |
1e95081c | 2604 | kfree(iovec); |
99bc4c38 | 2605 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2606 | return ret; |
2607 | } | |
2608 | ||
3529d8c2 JA |
2609 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2610 | bool force_nonblock) | |
f67676d1 | 2611 | { |
3529d8c2 JA |
2612 | struct io_async_ctx *io; |
2613 | struct iov_iter iter; | |
f67676d1 JA |
2614 | ssize_t ret; |
2615 | ||
3529d8c2 JA |
2616 | ret = io_prep_rw(req, sqe, force_nonblock); |
2617 | if (ret) | |
2618 | return ret; | |
f67676d1 | 2619 | |
3529d8c2 JA |
2620 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2621 | return -EBADF; | |
f67676d1 | 2622 | |
4ed734b0 JA |
2623 | req->fsize = rlimit(RLIMIT_FSIZE); |
2624 | ||
5f798bea PB |
2625 | /* either don't need iovec imported or already have it */ |
2626 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2627 | return 0; |
2628 | ||
2629 | io = req->io; | |
2630 | io->rw.iov = io->rw.fast_iov; | |
2631 | req->io = NULL; | |
bcda7baa | 2632 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2633 | req->io = io; |
2634 | if (ret < 0) | |
2635 | return ret; | |
2636 | ||
2637 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2638 | return 0; | |
f67676d1 JA |
2639 | } |
2640 | ||
014db007 | 2641 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2642 | { |
2643 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2644 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2645 | struct iov_iter iter; |
31b51510 | 2646 | size_t iov_count; |
f67676d1 | 2647 | ssize_t ret, io_size; |
2b188cc1 | 2648 | |
bcda7baa | 2649 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2650 | if (ret < 0) |
2651 | return ret; | |
2b188cc1 | 2652 | |
fd6c2e4c JA |
2653 | /* Ensure we clear previously set non-block flag */ |
2654 | if (!force_nonblock) | |
9adbd45d | 2655 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2656 | |
797f3f53 | 2657 | req->result = 0; |
f67676d1 | 2658 | io_size = ret; |
dea3b49c | 2659 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2660 | req->result = io_size; |
9e645e11 | 2661 | |
f67676d1 JA |
2662 | /* |
2663 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2664 | * we know to async punt it even if it was opened O_NONBLOCK | |
2665 | */ | |
29de5f6a | 2666 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2667 | goto copy_iov; |
31b51510 | 2668 | |
10d59345 JA |
2669 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2670 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2671 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2672 | goto copy_iov; |
31b51510 | 2673 | |
f67676d1 | 2674 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2675 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2676 | if (!ret) { |
9bf7933f RP |
2677 | ssize_t ret2; |
2678 | ||
2b188cc1 JA |
2679 | /* |
2680 | * Open-code file_start_write here to grab freeze protection, | |
2681 | * which will be released by another thread in | |
2682 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2683 | * released so that it doesn't complain about the held lock when | |
2684 | * we return to userspace. | |
2685 | */ | |
491381ce | 2686 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2687 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2688 | SB_FREEZE_WRITE, true); |
9adbd45d | 2689 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2690 | SB_FREEZE_WRITE); |
2691 | } | |
2692 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2693 | |
4ed734b0 JA |
2694 | if (!force_nonblock) |
2695 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2696 | ||
9adbd45d JA |
2697 | if (req->file->f_op->write_iter) |
2698 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2699 | else |
9adbd45d | 2700 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2701 | |
2702 | if (!force_nonblock) | |
2703 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2704 | ||
faac996c | 2705 | /* |
bff6035d | 2706 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2707 | * retry them without IOCB_NOWAIT. |
2708 | */ | |
2709 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2710 | ret2 = -EAGAIN; | |
f67676d1 | 2711 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2712 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2713 | } else { |
2714 | copy_iov: | |
b7bb4f7d | 2715 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2716 | inline_vecs, &iter); |
2717 | if (ret) | |
2718 | goto out_free; | |
29de5f6a JA |
2719 | /* any defer here is final, must blocking retry */ |
2720 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2721 | return -EAGAIN; |
2722 | } | |
2b188cc1 | 2723 | } |
31b51510 | 2724 | out_free: |
99bc4c38 | 2725 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2726 | kfree(iovec); |
2b188cc1 JA |
2727 | return ret; |
2728 | } | |
2729 | ||
7d67af2c PB |
2730 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2731 | { | |
2732 | struct io_splice* sp = &req->splice; | |
2733 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2734 | int ret; | |
2735 | ||
2736 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2737 | return 0; | |
2738 | ||
2739 | sp->file_in = NULL; | |
2740 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2741 | sp->off_out = READ_ONCE(sqe->off); | |
2742 | sp->len = READ_ONCE(sqe->len); | |
2743 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2744 | ||
2745 | if (unlikely(sp->flags & ~valid_flags)) | |
2746 | return -EINVAL; | |
2747 | ||
2748 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2749 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2750 | if (ret) | |
2751 | return ret; | |
2752 | req->flags |= REQ_F_NEED_CLEANUP; | |
2753 | ||
2754 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2755 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2756 | ||
2757 | return 0; | |
2758 | } | |
2759 | ||
2760 | static bool io_splice_punt(struct file *file) | |
2761 | { | |
2762 | if (get_pipe_info(file)) | |
2763 | return false; | |
2764 | if (!io_file_supports_async(file)) | |
2765 | return true; | |
88357580 | 2766 | return !(file->f_flags & O_NONBLOCK); |
7d67af2c PB |
2767 | } |
2768 | ||
014db007 | 2769 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2770 | { |
2771 | struct io_splice *sp = &req->splice; | |
2772 | struct file *in = sp->file_in; | |
2773 | struct file *out = sp->file_out; | |
2774 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2775 | loff_t *poff_in, *poff_out; | |
2776 | long ret; | |
2777 | ||
2778 | if (force_nonblock) { | |
2779 | if (io_splice_punt(in) || io_splice_punt(out)) | |
2780 | return -EAGAIN; | |
2781 | flags |= SPLICE_F_NONBLOCK; | |
2782 | } | |
2783 | ||
2784 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2785 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2786 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2787 | if (force_nonblock && ret == -EAGAIN) | |
2788 | return -EAGAIN; | |
2789 | ||
2790 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2791 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2792 | ||
2793 | io_cqring_add_event(req, ret); | |
2794 | if (ret != sp->len) | |
2795 | req_set_fail_links(req); | |
014db007 | 2796 | io_put_req(req); |
7d67af2c PB |
2797 | return 0; |
2798 | } | |
2799 | ||
2b188cc1 JA |
2800 | /* |
2801 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2802 | */ | |
78e19bbe | 2803 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2804 | { |
2805 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2806 | |
def596e9 JA |
2807 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2808 | return -EINVAL; | |
2809 | ||
78e19bbe | 2810 | io_cqring_add_event(req, 0); |
e65ef56d | 2811 | io_put_req(req); |
2b188cc1 JA |
2812 | return 0; |
2813 | } | |
2814 | ||
3529d8c2 | 2815 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2816 | { |
6b06314c | 2817 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2818 | |
09bb8394 JA |
2819 | if (!req->file) |
2820 | return -EBADF; | |
c992fe29 | 2821 | |
6b06314c | 2822 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2823 | return -EINVAL; |
edafccee | 2824 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2825 | return -EINVAL; |
2826 | ||
8ed8d3c3 JA |
2827 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2828 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2829 | return -EINVAL; | |
2830 | ||
2831 | req->sync.off = READ_ONCE(sqe->off); | |
2832 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2833 | return 0; |
2834 | } | |
2835 | ||
8ed8d3c3 JA |
2836 | static bool io_req_cancelled(struct io_kiocb *req) |
2837 | { | |
2838 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2839 | req_set_fail_links(req); | |
2840 | io_cqring_add_event(req, -ECANCELED); | |
2841 | io_put_req(req); | |
2842 | return true; | |
2843 | } | |
2844 | ||
2845 | return false; | |
2846 | } | |
2847 | ||
014db007 | 2848 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2849 | { |
8ed8d3c3 | 2850 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2851 | int ret; |
2852 | ||
9adbd45d | 2853 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2854 | end > 0 ? end : LLONG_MAX, |
2855 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2856 | if (ret < 0) | |
2857 | req_set_fail_links(req); | |
2858 | io_cqring_add_event(req, ret); | |
014db007 | 2859 | io_put_req(req); |
8ed8d3c3 JA |
2860 | } |
2861 | ||
5ea62161 | 2862 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2863 | { |
5ea62161 | 2864 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2865 | |
2866 | if (io_req_cancelled(req)) | |
2867 | return; | |
014db007 | 2868 | __io_fsync(req); |
e9fd9396 | 2869 | io_steal_work(req, workptr); |
8ed8d3c3 | 2870 | } |
c992fe29 | 2871 | |
014db007 | 2872 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2873 | { |
c992fe29 | 2874 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2875 | if (force_nonblock) { |
8ed8d3c3 | 2876 | req->work.func = io_fsync_finish; |
c992fe29 | 2877 | return -EAGAIN; |
8ed8d3c3 | 2878 | } |
014db007 | 2879 | __io_fsync(req); |
c992fe29 CH |
2880 | return 0; |
2881 | } | |
2882 | ||
014db007 | 2883 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2884 | { |
8ed8d3c3 JA |
2885 | int ret; |
2886 | ||
4ed734b0 | 2887 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2888 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2889 | req->sync.len); | |
4ed734b0 | 2890 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2891 | if (ret < 0) |
2892 | req_set_fail_links(req); | |
2893 | io_cqring_add_event(req, ret); | |
014db007 | 2894 | io_put_req(req); |
5ea62161 PB |
2895 | } |
2896 | ||
2897 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2898 | { | |
2899 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2900 | |
594506fe PB |
2901 | if (io_req_cancelled(req)) |
2902 | return; | |
014db007 | 2903 | __io_fallocate(req); |
e9fd9396 | 2904 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2905 | } |
2906 | ||
d63d1b5e JA |
2907 | static int io_fallocate_prep(struct io_kiocb *req, |
2908 | const struct io_uring_sqe *sqe) | |
2909 | { | |
2910 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2911 | return -EINVAL; | |
2912 | ||
2913 | req->sync.off = READ_ONCE(sqe->off); | |
2914 | req->sync.len = READ_ONCE(sqe->addr); | |
2915 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2916 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2917 | return 0; |
2918 | } | |
2919 | ||
014db007 | 2920 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2921 | { |
d63d1b5e | 2922 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2923 | if (force_nonblock) { |
d63d1b5e | 2924 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2925 | return -EAGAIN; |
8ed8d3c3 | 2926 | } |
5d17b4a4 | 2927 | |
014db007 | 2928 | __io_fallocate(req); |
5d17b4a4 JA |
2929 | return 0; |
2930 | } | |
2931 | ||
15b71abe | 2932 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2933 | { |
f8748881 | 2934 | const char __user *fname; |
15b71abe | 2935 | int ret; |
b7bb4f7d | 2936 | |
15b71abe JA |
2937 | if (sqe->ioprio || sqe->buf_index) |
2938 | return -EINVAL; | |
9c280f90 | 2939 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2940 | return -EBADF; |
0bdbdd08 PB |
2941 | if (req->flags & REQ_F_NEED_CLEANUP) |
2942 | return 0; | |
03b1230c | 2943 | |
15b71abe | 2944 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2945 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2946 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2947 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2948 | if (force_o_largefile()) |
2949 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2950 | |
f8748881 | 2951 | req->open.filename = getname(fname); |
15b71abe JA |
2952 | if (IS_ERR(req->open.filename)) { |
2953 | ret = PTR_ERR(req->open.filename); | |
2954 | req->open.filename = NULL; | |
2955 | return ret; | |
2956 | } | |
3529d8c2 | 2957 | |
4022e7af | 2958 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2959 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2960 | return 0; |
03b1230c JA |
2961 | } |
2962 | ||
cebdb986 | 2963 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2964 | { |
cebdb986 JA |
2965 | struct open_how __user *how; |
2966 | const char __user *fname; | |
2967 | size_t len; | |
0fa03c62 JA |
2968 | int ret; |
2969 | ||
cebdb986 | 2970 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2971 | return -EINVAL; |
9c280f90 | 2972 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2973 | return -EBADF; |
0bdbdd08 PB |
2974 | if (req->flags & REQ_F_NEED_CLEANUP) |
2975 | return 0; | |
0fa03c62 | 2976 | |
cebdb986 JA |
2977 | req->open.dfd = READ_ONCE(sqe->fd); |
2978 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2979 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2980 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2981 | |
cebdb986 JA |
2982 | if (len < OPEN_HOW_SIZE_VER0) |
2983 | return -EINVAL; | |
3529d8c2 | 2984 | |
cebdb986 JA |
2985 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2986 | len); | |
2987 | if (ret) | |
2988 | return ret; | |
3529d8c2 | 2989 | |
cebdb986 JA |
2990 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2991 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2992 | |
cebdb986 JA |
2993 | req->open.filename = getname(fname); |
2994 | if (IS_ERR(req->open.filename)) { | |
2995 | ret = PTR_ERR(req->open.filename); | |
2996 | req->open.filename = NULL; | |
2997 | return ret; | |
2998 | } | |
2999 | ||
4022e7af | 3000 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3001 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
3002 | return 0; |
3003 | } | |
3004 | ||
014db007 | 3005 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3006 | { |
3007 | struct open_flags op; | |
15b71abe JA |
3008 | struct file *file; |
3009 | int ret; | |
3010 | ||
f86cd20c | 3011 | if (force_nonblock) |
15b71abe | 3012 | return -EAGAIN; |
15b71abe | 3013 | |
cebdb986 | 3014 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3015 | if (ret) |
3016 | goto err; | |
3017 | ||
4022e7af | 3018 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3019 | if (ret < 0) |
3020 | goto err; | |
3021 | ||
3022 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3023 | if (IS_ERR(file)) { | |
3024 | put_unused_fd(ret); | |
3025 | ret = PTR_ERR(file); | |
3026 | } else { | |
3027 | fsnotify_open(file); | |
3028 | fd_install(ret, file); | |
3029 | } | |
3030 | err: | |
3031 | putname(req->open.filename); | |
8fef80bf | 3032 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3033 | if (ret < 0) |
3034 | req_set_fail_links(req); | |
3035 | io_cqring_add_event(req, ret); | |
014db007 | 3036 | io_put_req(req); |
15b71abe JA |
3037 | return 0; |
3038 | } | |
3039 | ||
014db007 | 3040 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3041 | { |
3042 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3043 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3044 | } |
3045 | ||
067524e9 JA |
3046 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3047 | const struct io_uring_sqe *sqe) | |
3048 | { | |
3049 | struct io_provide_buf *p = &req->pbuf; | |
3050 | u64 tmp; | |
3051 | ||
3052 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3053 | return -EINVAL; | |
3054 | ||
3055 | tmp = READ_ONCE(sqe->fd); | |
3056 | if (!tmp || tmp > USHRT_MAX) | |
3057 | return -EINVAL; | |
3058 | ||
3059 | memset(p, 0, sizeof(*p)); | |
3060 | p->nbufs = tmp; | |
3061 | p->bgid = READ_ONCE(sqe->buf_group); | |
3062 | return 0; | |
3063 | } | |
3064 | ||
3065 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3066 | int bgid, unsigned nbufs) | |
3067 | { | |
3068 | unsigned i = 0; | |
3069 | ||
3070 | /* shouldn't happen */ | |
3071 | if (!nbufs) | |
3072 | return 0; | |
3073 | ||
3074 | /* the head kbuf is the list itself */ | |
3075 | while (!list_empty(&buf->list)) { | |
3076 | struct io_buffer *nxt; | |
3077 | ||
3078 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3079 | list_del(&nxt->list); | |
3080 | kfree(nxt); | |
3081 | if (++i == nbufs) | |
3082 | return i; | |
3083 | } | |
3084 | i++; | |
3085 | kfree(buf); | |
3086 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3087 | ||
3088 | return i; | |
3089 | } | |
3090 | ||
3091 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3092 | { | |
3093 | struct io_provide_buf *p = &req->pbuf; | |
3094 | struct io_ring_ctx *ctx = req->ctx; | |
3095 | struct io_buffer *head; | |
3096 | int ret = 0; | |
3097 | ||
3098 | io_ring_submit_lock(ctx, !force_nonblock); | |
3099 | ||
3100 | lockdep_assert_held(&ctx->uring_lock); | |
3101 | ||
3102 | ret = -ENOENT; | |
3103 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3104 | if (head) | |
3105 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3106 | ||
3107 | io_ring_submit_lock(ctx, !force_nonblock); | |
3108 | if (ret < 0) | |
3109 | req_set_fail_links(req); | |
3110 | io_cqring_add_event(req, ret); | |
3111 | io_put_req(req); | |
3112 | return 0; | |
3113 | } | |
3114 | ||
ddf0322d JA |
3115 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3116 | const struct io_uring_sqe *sqe) | |
3117 | { | |
3118 | struct io_provide_buf *p = &req->pbuf; | |
3119 | u64 tmp; | |
3120 | ||
3121 | if (sqe->ioprio || sqe->rw_flags) | |
3122 | return -EINVAL; | |
3123 | ||
3124 | tmp = READ_ONCE(sqe->fd); | |
3125 | if (!tmp || tmp > USHRT_MAX) | |
3126 | return -E2BIG; | |
3127 | p->nbufs = tmp; | |
3128 | p->addr = READ_ONCE(sqe->addr); | |
3129 | p->len = READ_ONCE(sqe->len); | |
3130 | ||
3131 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3132 | return -EFAULT; | |
3133 | ||
3134 | p->bgid = READ_ONCE(sqe->buf_group); | |
3135 | tmp = READ_ONCE(sqe->off); | |
3136 | if (tmp > USHRT_MAX) | |
3137 | return -E2BIG; | |
3138 | p->bid = tmp; | |
3139 | return 0; | |
3140 | } | |
3141 | ||
3142 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3143 | { | |
3144 | struct io_buffer *buf; | |
3145 | u64 addr = pbuf->addr; | |
3146 | int i, bid = pbuf->bid; | |
3147 | ||
3148 | for (i = 0; i < pbuf->nbufs; i++) { | |
3149 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3150 | if (!buf) | |
3151 | break; | |
3152 | ||
3153 | buf->addr = addr; | |
3154 | buf->len = pbuf->len; | |
3155 | buf->bid = bid; | |
3156 | addr += pbuf->len; | |
3157 | bid++; | |
3158 | if (!*head) { | |
3159 | INIT_LIST_HEAD(&buf->list); | |
3160 | *head = buf; | |
3161 | } else { | |
3162 | list_add_tail(&buf->list, &(*head)->list); | |
3163 | } | |
3164 | } | |
3165 | ||
3166 | return i ? i : -ENOMEM; | |
3167 | } | |
3168 | ||
ddf0322d JA |
3169 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3170 | { | |
3171 | struct io_provide_buf *p = &req->pbuf; | |
3172 | struct io_ring_ctx *ctx = req->ctx; | |
3173 | struct io_buffer *head, *list; | |
3174 | int ret = 0; | |
3175 | ||
3176 | io_ring_submit_lock(ctx, !force_nonblock); | |
3177 | ||
3178 | lockdep_assert_held(&ctx->uring_lock); | |
3179 | ||
3180 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3181 | ||
3182 | ret = io_add_buffers(p, &head); | |
3183 | if (ret < 0) | |
3184 | goto out; | |
3185 | ||
3186 | if (!list) { | |
3187 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3188 | GFP_KERNEL); | |
3189 | if (ret < 0) { | |
067524e9 | 3190 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3191 | goto out; |
3192 | } | |
3193 | } | |
3194 | out: | |
3195 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3196 | if (ret < 0) | |
3197 | req_set_fail_links(req); | |
3198 | io_cqring_add_event(req, ret); | |
3199 | io_put_req(req); | |
3200 | return 0; | |
cebdb986 JA |
3201 | } |
3202 | ||
3e4827b0 JA |
3203 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3204 | const struct io_uring_sqe *sqe) | |
3205 | { | |
3206 | #if defined(CONFIG_EPOLL) | |
3207 | if (sqe->ioprio || sqe->buf_index) | |
3208 | return -EINVAL; | |
3209 | ||
3210 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3211 | req->epoll.op = READ_ONCE(sqe->len); | |
3212 | req->epoll.fd = READ_ONCE(sqe->off); | |
3213 | ||
3214 | if (ep_op_has_event(req->epoll.op)) { | |
3215 | struct epoll_event __user *ev; | |
3216 | ||
3217 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3218 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3219 | return -EFAULT; | |
3220 | } | |
3221 | ||
3222 | return 0; | |
3223 | #else | |
3224 | return -EOPNOTSUPP; | |
3225 | #endif | |
3226 | } | |
3227 | ||
014db007 | 3228 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3229 | { |
3230 | #if defined(CONFIG_EPOLL) | |
3231 | struct io_epoll *ie = &req->epoll; | |
3232 | int ret; | |
3233 | ||
3234 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3235 | if (force_nonblock && ret == -EAGAIN) | |
3236 | return -EAGAIN; | |
3237 | ||
3238 | if (ret < 0) | |
3239 | req_set_fail_links(req); | |
3240 | io_cqring_add_event(req, ret); | |
014db007 | 3241 | io_put_req(req); |
3e4827b0 JA |
3242 | return 0; |
3243 | #else | |
3244 | return -EOPNOTSUPP; | |
3245 | #endif | |
3246 | } | |
3247 | ||
c1ca757b JA |
3248 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3249 | { | |
3250 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3251 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3252 | return -EINVAL; | |
3253 | ||
3254 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3255 | req->madvise.len = READ_ONCE(sqe->len); | |
3256 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3257 | return 0; | |
3258 | #else | |
3259 | return -EOPNOTSUPP; | |
3260 | #endif | |
3261 | } | |
3262 | ||
014db007 | 3263 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3264 | { |
3265 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3266 | struct io_madvise *ma = &req->madvise; | |
3267 | int ret; | |
3268 | ||
3269 | if (force_nonblock) | |
3270 | return -EAGAIN; | |
3271 | ||
3272 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3273 | if (ret < 0) | |
3274 | req_set_fail_links(req); | |
3275 | io_cqring_add_event(req, ret); | |
014db007 | 3276 | io_put_req(req); |
c1ca757b JA |
3277 | return 0; |
3278 | #else | |
3279 | return -EOPNOTSUPP; | |
3280 | #endif | |
3281 | } | |
3282 | ||
4840e418 JA |
3283 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3284 | { | |
3285 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3286 | return -EINVAL; | |
3287 | ||
3288 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3289 | req->fadvise.len = READ_ONCE(sqe->len); | |
3290 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3291 | return 0; | |
3292 | } | |
3293 | ||
014db007 | 3294 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3295 | { |
3296 | struct io_fadvise *fa = &req->fadvise; | |
3297 | int ret; | |
3298 | ||
3e69426d JA |
3299 | if (force_nonblock) { |
3300 | switch (fa->advice) { | |
3301 | case POSIX_FADV_NORMAL: | |
3302 | case POSIX_FADV_RANDOM: | |
3303 | case POSIX_FADV_SEQUENTIAL: | |
3304 | break; | |
3305 | default: | |
3306 | return -EAGAIN; | |
3307 | } | |
3308 | } | |
4840e418 JA |
3309 | |
3310 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3311 | if (ret < 0) | |
3312 | req_set_fail_links(req); | |
3313 | io_cqring_add_event(req, ret); | |
014db007 | 3314 | io_put_req(req); |
4840e418 JA |
3315 | return 0; |
3316 | } | |
3317 | ||
eddc7ef5 JA |
3318 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3319 | { | |
f8748881 | 3320 | const char __user *fname; |
eddc7ef5 JA |
3321 | unsigned lookup_flags; |
3322 | int ret; | |
3323 | ||
3324 | if (sqe->ioprio || sqe->buf_index) | |
3325 | return -EINVAL; | |
9c280f90 | 3326 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3327 | return -EBADF; |
0bdbdd08 PB |
3328 | if (req->flags & REQ_F_NEED_CLEANUP) |
3329 | return 0; | |
eddc7ef5 JA |
3330 | |
3331 | req->open.dfd = READ_ONCE(sqe->fd); | |
3332 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3333 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3334 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3335 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3336 | |
c12cedf2 | 3337 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3338 | return -EINVAL; |
3339 | ||
f8748881 | 3340 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3341 | if (IS_ERR(req->open.filename)) { |
3342 | ret = PTR_ERR(req->open.filename); | |
3343 | req->open.filename = NULL; | |
3344 | return ret; | |
3345 | } | |
3346 | ||
8fef80bf | 3347 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3348 | return 0; |
3349 | } | |
3350 | ||
014db007 | 3351 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3352 | { |
3353 | struct io_open *ctx = &req->open; | |
3354 | unsigned lookup_flags; | |
3355 | struct path path; | |
3356 | struct kstat stat; | |
3357 | int ret; | |
3358 | ||
3359 | if (force_nonblock) | |
3360 | return -EAGAIN; | |
3361 | ||
c12cedf2 | 3362 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3363 | return -EINVAL; |
3364 | ||
3365 | retry: | |
3366 | /* filename_lookup() drops it, keep a reference */ | |
3367 | ctx->filename->refcnt++; | |
3368 | ||
3369 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3370 | NULL); | |
3371 | if (ret) | |
3372 | goto err; | |
3373 | ||
c12cedf2 | 3374 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3375 | path_put(&path); |
3376 | if (retry_estale(ret, lookup_flags)) { | |
3377 | lookup_flags |= LOOKUP_REVAL; | |
3378 | goto retry; | |
3379 | } | |
3380 | if (!ret) | |
3381 | ret = cp_statx(&stat, ctx->buffer); | |
3382 | err: | |
3383 | putname(ctx->filename); | |
8fef80bf | 3384 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3385 | if (ret < 0) |
3386 | req_set_fail_links(req); | |
3387 | io_cqring_add_event(req, ret); | |
014db007 | 3388 | io_put_req(req); |
eddc7ef5 JA |
3389 | return 0; |
3390 | } | |
3391 | ||
b5dba59e JA |
3392 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3393 | { | |
3394 | /* | |
3395 | * If we queue this for async, it must not be cancellable. That would | |
3396 | * leave the 'file' in an undeterminate state. | |
3397 | */ | |
3398 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3399 | ||
3400 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3401 | sqe->rw_flags || sqe->buf_index) | |
3402 | return -EINVAL; | |
9c280f90 | 3403 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3404 | return -EBADF; |
b5dba59e JA |
3405 | |
3406 | req->close.fd = READ_ONCE(sqe->fd); | |
3407 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3408 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3409 | return -EBADF; |
3410 | ||
3411 | return 0; | |
3412 | } | |
3413 | ||
a93b3331 | 3414 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3415 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3416 | { |
3417 | int ret; | |
3418 | ||
3419 | ret = filp_close(req->close.put_file, req->work.files); | |
3420 | if (ret < 0) | |
3421 | req_set_fail_links(req); | |
3422 | io_cqring_add_event(req, ret); | |
3423 | fput(req->close.put_file); | |
014db007 | 3424 | io_put_req(req); |
a93b3331 PB |
3425 | } |
3426 | ||
b5dba59e JA |
3427 | static void io_close_finish(struct io_wq_work **workptr) |
3428 | { | |
3429 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3430 | |
7fbeb95d | 3431 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3432 | __io_close_finish(req); |
e9fd9396 | 3433 | io_steal_work(req, workptr); |
b5dba59e JA |
3434 | } |
3435 | ||
014db007 | 3436 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3437 | { |
3438 | int ret; | |
3439 | ||
3440 | req->close.put_file = NULL; | |
3441 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3442 | if (ret < 0) | |
3443 | return ret; | |
3444 | ||
3445 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3446 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3447 | /* submission ref will be dropped, take it for async */ |
3448 | refcount_inc(&req->refs); | |
3449 | ||
a2100672 PB |
3450 | req->work.func = io_close_finish; |
3451 | /* | |
3452 | * Do manual async queue here to avoid grabbing files - we don't | |
3453 | * need the files, and it'll cause io_close_finish() to close | |
3454 | * the file again and cause a double CQE entry for this request | |
3455 | */ | |
3456 | io_queue_async_work(req); | |
3457 | return 0; | |
3458 | } | |
b5dba59e JA |
3459 | |
3460 | /* | |
3461 | * No ->flush(), safely close from here and just punt the | |
3462 | * fput() to async context. | |
3463 | */ | |
014db007 | 3464 | __io_close_finish(req); |
1a417f4e | 3465 | return 0; |
b5dba59e JA |
3466 | } |
3467 | ||
3529d8c2 | 3468 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3469 | { |
3470 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3471 | |
3472 | if (!req->file) | |
3473 | return -EBADF; | |
5d17b4a4 JA |
3474 | |
3475 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3476 | return -EINVAL; | |
3477 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3478 | return -EINVAL; | |
3479 | ||
8ed8d3c3 JA |
3480 | req->sync.off = READ_ONCE(sqe->off); |
3481 | req->sync.len = READ_ONCE(sqe->len); | |
3482 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3483 | return 0; |
3484 | } | |
3485 | ||
014db007 | 3486 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3487 | { |
8ed8d3c3 JA |
3488 | int ret; |
3489 | ||
9adbd45d | 3490 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3491 | req->sync.flags); |
3492 | if (ret < 0) | |
3493 | req_set_fail_links(req); | |
3494 | io_cqring_add_event(req, ret); | |
014db007 | 3495 | io_put_req(req); |
5ea62161 PB |
3496 | } |
3497 | ||
3498 | ||
3499 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3500 | { | |
3501 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3502 | |
3503 | if (io_req_cancelled(req)) | |
3504 | return; | |
014db007 | 3505 | __io_sync_file_range(req); |
594506fe | 3506 | io_put_req(req); /* put submission ref */ |
5d17b4a4 JA |
3507 | } |
3508 | ||
014db007 | 3509 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3510 | { |
5d17b4a4 | 3511 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3512 | if (force_nonblock) { |
8ed8d3c3 | 3513 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3514 | return -EAGAIN; |
8ed8d3c3 | 3515 | } |
5d17b4a4 | 3516 | |
014db007 | 3517 | __io_sync_file_range(req); |
5d17b4a4 JA |
3518 | return 0; |
3519 | } | |
3520 | ||
469956e8 | 3521 | #if defined(CONFIG_NET) |
02d27d89 PB |
3522 | static int io_setup_async_msg(struct io_kiocb *req, |
3523 | struct io_async_msghdr *kmsg) | |
3524 | { | |
3525 | if (req->io) | |
3526 | return -EAGAIN; | |
3527 | if (io_alloc_async_ctx(req)) { | |
3528 | if (kmsg->iov != kmsg->fast_iov) | |
3529 | kfree(kmsg->iov); | |
3530 | return -ENOMEM; | |
3531 | } | |
3532 | req->flags |= REQ_F_NEED_CLEANUP; | |
3533 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3534 | return -EAGAIN; | |
3535 | } | |
3536 | ||
3529d8c2 | 3537 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3538 | { |
e47293fd | 3539 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3540 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3541 | int ret; |
03b1230c | 3542 | |
e47293fd JA |
3543 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3544 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3545 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3546 | |
d8768362 JA |
3547 | #ifdef CONFIG_COMPAT |
3548 | if (req->ctx->compat) | |
3549 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3550 | #endif | |
3551 | ||
fddaface | 3552 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3553 | return 0; |
5f798bea PB |
3554 | /* iovec is already imported */ |
3555 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3556 | return 0; | |
3529d8c2 | 3557 | |
d9688565 | 3558 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3559 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3560 | &io->msg.iov); |
99bc4c38 PB |
3561 | if (!ret) |
3562 | req->flags |= REQ_F_NEED_CLEANUP; | |
3563 | return ret; | |
03b1230c JA |
3564 | } |
3565 | ||
014db007 | 3566 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3567 | { |
0b416c3e | 3568 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3569 | struct socket *sock; |
3570 | int ret; | |
3571 | ||
3572 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3573 | return -EINVAL; | |
3574 | ||
3575 | sock = sock_from_file(req->file, &ret); | |
3576 | if (sock) { | |
b7bb4f7d | 3577 | struct io_async_ctx io; |
0fa03c62 JA |
3578 | unsigned flags; |
3579 | ||
03b1230c | 3580 | if (req->io) { |
0b416c3e | 3581 | kmsg = &req->io->msg; |
b537916c | 3582 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3583 | /* if iov is set, it's allocated already */ |
3584 | if (!kmsg->iov) | |
3585 | kmsg->iov = kmsg->fast_iov; | |
3586 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3587 | } else { |
3529d8c2 JA |
3588 | struct io_sr_msg *sr = &req->sr_msg; |
3589 | ||
0b416c3e | 3590 | kmsg = &io.msg; |
b537916c | 3591 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3592 | |
3593 | io.msg.iov = io.msg.fast_iov; | |
3594 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3595 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3596 | if (ret) |
3529d8c2 | 3597 | return ret; |
03b1230c | 3598 | } |
0fa03c62 | 3599 | |
e47293fd JA |
3600 | flags = req->sr_msg.msg_flags; |
3601 | if (flags & MSG_DONTWAIT) | |
3602 | req->flags |= REQ_F_NOWAIT; | |
3603 | else if (force_nonblock) | |
3604 | flags |= MSG_DONTWAIT; | |
3605 | ||
0b416c3e | 3606 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3607 | if (force_nonblock && ret == -EAGAIN) |
3608 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3609 | if (ret == -ERESTARTSYS) |
3610 | ret = -EINTR; | |
0fa03c62 JA |
3611 | } |
3612 | ||
1e95081c | 3613 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3614 | kfree(kmsg->iov); |
99bc4c38 | 3615 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3616 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3617 | if (ret < 0) |
3618 | req_set_fail_links(req); | |
014db007 | 3619 | io_put_req(req); |
5d17b4a4 | 3620 | return 0; |
03b1230c | 3621 | } |
aa1fa28f | 3622 | |
014db007 | 3623 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3624 | { |
fddaface JA |
3625 | struct socket *sock; |
3626 | int ret; | |
3627 | ||
3628 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3629 | return -EINVAL; | |
3630 | ||
3631 | sock = sock_from_file(req->file, &ret); | |
3632 | if (sock) { | |
3633 | struct io_sr_msg *sr = &req->sr_msg; | |
3634 | struct msghdr msg; | |
3635 | struct iovec iov; | |
3636 | unsigned flags; | |
3637 | ||
3638 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3639 | &msg.msg_iter); | |
3640 | if (ret) | |
3641 | return ret; | |
3642 | ||
3643 | msg.msg_name = NULL; | |
3644 | msg.msg_control = NULL; | |
3645 | msg.msg_controllen = 0; | |
3646 | msg.msg_namelen = 0; | |
3647 | ||
3648 | flags = req->sr_msg.msg_flags; | |
3649 | if (flags & MSG_DONTWAIT) | |
3650 | req->flags |= REQ_F_NOWAIT; | |
3651 | else if (force_nonblock) | |
3652 | flags |= MSG_DONTWAIT; | |
3653 | ||
0b7b21e4 JA |
3654 | msg.msg_flags = flags; |
3655 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3656 | if (force_nonblock && ret == -EAGAIN) |
3657 | return -EAGAIN; | |
3658 | if (ret == -ERESTARTSYS) | |
3659 | ret = -EINTR; | |
3660 | } | |
3661 | ||
3662 | io_cqring_add_event(req, ret); | |
3663 | if (ret < 0) | |
3664 | req_set_fail_links(req); | |
014db007 | 3665 | io_put_req(req); |
fddaface | 3666 | return 0; |
fddaface JA |
3667 | } |
3668 | ||
52de1fe1 JA |
3669 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3670 | { | |
3671 | struct io_sr_msg *sr = &req->sr_msg; | |
3672 | struct iovec __user *uiov; | |
3673 | size_t iov_len; | |
3674 | int ret; | |
3675 | ||
3676 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3677 | &uiov, &iov_len); | |
3678 | if (ret) | |
3679 | return ret; | |
3680 | ||
3681 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3682 | if (iov_len > 1) | |
3683 | return -EINVAL; | |
3684 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3685 | return -EFAULT; | |
3686 | sr->len = io->msg.iov[0].iov_len; | |
3687 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3688 | sr->len); | |
3689 | io->msg.iov = NULL; | |
3690 | } else { | |
3691 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3692 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3693 | if (ret > 0) | |
3694 | ret = 0; | |
3695 | } | |
3696 | ||
3697 | return ret; | |
3698 | } | |
3699 | ||
3700 | #ifdef CONFIG_COMPAT | |
3701 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3702 | struct io_async_ctx *io) | |
3703 | { | |
3704 | struct compat_msghdr __user *msg_compat; | |
3705 | struct io_sr_msg *sr = &req->sr_msg; | |
3706 | struct compat_iovec __user *uiov; | |
3707 | compat_uptr_t ptr; | |
3708 | compat_size_t len; | |
3709 | int ret; | |
3710 | ||
3711 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3712 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3713 | &ptr, &len); | |
3714 | if (ret) | |
3715 | return ret; | |
3716 | ||
3717 | uiov = compat_ptr(ptr); | |
3718 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3719 | compat_ssize_t clen; | |
3720 | ||
3721 | if (len > 1) | |
3722 | return -EINVAL; | |
3723 | if (!access_ok(uiov, sizeof(*uiov))) | |
3724 | return -EFAULT; | |
3725 | if (__get_user(clen, &uiov->iov_len)) | |
3726 | return -EFAULT; | |
3727 | if (clen < 0) | |
3728 | return -EINVAL; | |
3729 | sr->len = io->msg.iov[0].iov_len; | |
3730 | io->msg.iov = NULL; | |
3731 | } else { | |
3732 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3733 | &io->msg.iov, | |
3734 | &io->msg.msg.msg_iter); | |
3735 | if (ret < 0) | |
3736 | return ret; | |
3737 | } | |
3738 | ||
3739 | return 0; | |
3740 | } | |
3741 | #endif | |
3742 | ||
3743 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3744 | { | |
3745 | io->msg.iov = io->msg.fast_iov; | |
3746 | ||
3747 | #ifdef CONFIG_COMPAT | |
3748 | if (req->ctx->compat) | |
3749 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3750 | #endif |
52de1fe1 JA |
3751 | |
3752 | return __io_recvmsg_copy_hdr(req, io); | |
3753 | } | |
3754 | ||
bcda7baa JA |
3755 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3756 | int *cflags, bool needs_lock) | |
3757 | { | |
3758 | struct io_sr_msg *sr = &req->sr_msg; | |
3759 | struct io_buffer *kbuf; | |
3760 | ||
3761 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3762 | return NULL; | |
3763 | ||
3764 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3765 | if (IS_ERR(kbuf)) | |
3766 | return kbuf; | |
3767 | ||
3768 | sr->kbuf = kbuf; | |
3769 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3770 | ||
3771 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3772 | *cflags |= IORING_CQE_F_BUFFER; | |
3773 | return kbuf; | |
fddaface JA |
3774 | } |
3775 | ||
3529d8c2 JA |
3776 | static int io_recvmsg_prep(struct io_kiocb *req, |
3777 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3778 | { |
e47293fd | 3779 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3780 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3781 | int ret; |
3529d8c2 JA |
3782 | |
3783 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3784 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3785 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3786 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3787 | |
d8768362 JA |
3788 | #ifdef CONFIG_COMPAT |
3789 | if (req->ctx->compat) | |
3790 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3791 | #endif | |
3792 | ||
fddaface | 3793 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3794 | return 0; |
5f798bea PB |
3795 | /* iovec is already imported */ |
3796 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3797 | return 0; | |
03b1230c | 3798 | |
52de1fe1 | 3799 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3800 | if (!ret) |
3801 | req->flags |= REQ_F_NEED_CLEANUP; | |
3802 | return ret; | |
aa1fa28f JA |
3803 | } |
3804 | ||
014db007 | 3805 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3806 | { |
0b416c3e | 3807 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3808 | struct socket *sock; |
52de1fe1 | 3809 | int ret, cflags = 0; |
03b1230c JA |
3810 | |
3811 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3812 | return -EINVAL; | |
3813 | ||
3814 | sock = sock_from_file(req->file, &ret); | |
3815 | if (sock) { | |
52de1fe1 | 3816 | struct io_buffer *kbuf; |
b7bb4f7d | 3817 | struct io_async_ctx io; |
03b1230c JA |
3818 | unsigned flags; |
3819 | ||
03b1230c | 3820 | if (req->io) { |
0b416c3e | 3821 | kmsg = &req->io->msg; |
b537916c | 3822 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3823 | /* if iov is set, it's allocated already */ |
3824 | if (!kmsg->iov) | |
3825 | kmsg->iov = kmsg->fast_iov; | |
3826 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3827 | } else { |
0b416c3e | 3828 | kmsg = &io.msg; |
b537916c | 3829 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3830 | |
52de1fe1 | 3831 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3832 | if (ret) |
3529d8c2 | 3833 | return ret; |
03b1230c JA |
3834 | } |
3835 | ||
52de1fe1 JA |
3836 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3837 | if (IS_ERR(kbuf)) { | |
3838 | return PTR_ERR(kbuf); | |
3839 | } else if (kbuf) { | |
3840 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3841 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3842 | 1, req->sr_msg.len); | |
3843 | } | |
3844 | ||
e47293fd JA |
3845 | flags = req->sr_msg.msg_flags; |
3846 | if (flags & MSG_DONTWAIT) | |
3847 | req->flags |= REQ_F_NOWAIT; | |
3848 | else if (force_nonblock) | |
3849 | flags |= MSG_DONTWAIT; | |
3850 | ||
3851 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3852 | kmsg->uaddr, flags); | |
02d27d89 PB |
3853 | if (force_nonblock && ret == -EAGAIN) |
3854 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3855 | if (ret == -ERESTARTSYS) |
3856 | ret = -EINTR; | |
3857 | } | |
3858 | ||
1e95081c | 3859 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3860 | kfree(kmsg->iov); |
99bc4c38 | 3861 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3862 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3863 | if (ret < 0) |
3864 | req_set_fail_links(req); | |
014db007 | 3865 | io_put_req(req); |
03b1230c | 3866 | return 0; |
0fa03c62 | 3867 | } |
5d17b4a4 | 3868 | |
014db007 | 3869 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3870 | { |
bcda7baa | 3871 | struct io_buffer *kbuf = NULL; |
fddaface | 3872 | struct socket *sock; |
bcda7baa | 3873 | int ret, cflags = 0; |
fddaface JA |
3874 | |
3875 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3876 | return -EINVAL; | |
3877 | ||
3878 | sock = sock_from_file(req->file, &ret); | |
3879 | if (sock) { | |
3880 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3881 | void __user *buf = sr->buf; |
fddaface JA |
3882 | struct msghdr msg; |
3883 | struct iovec iov; | |
3884 | unsigned flags; | |
3885 | ||
bcda7baa JA |
3886 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3887 | if (IS_ERR(kbuf)) | |
3888 | return PTR_ERR(kbuf); | |
3889 | else if (kbuf) | |
3890 | buf = u64_to_user_ptr(kbuf->addr); | |
3891 | ||
3892 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3893 | &msg.msg_iter); |
bcda7baa JA |
3894 | if (ret) { |
3895 | kfree(kbuf); | |
fddaface | 3896 | return ret; |
bcda7baa | 3897 | } |
fddaface | 3898 | |
bcda7baa | 3899 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3900 | msg.msg_name = NULL; |
3901 | msg.msg_control = NULL; | |
3902 | msg.msg_controllen = 0; | |
3903 | msg.msg_namelen = 0; | |
3904 | msg.msg_iocb = NULL; | |
3905 | msg.msg_flags = 0; | |
3906 | ||
3907 | flags = req->sr_msg.msg_flags; | |
3908 | if (flags & MSG_DONTWAIT) | |
3909 | req->flags |= REQ_F_NOWAIT; | |
3910 | else if (force_nonblock) | |
3911 | flags |= MSG_DONTWAIT; | |
3912 | ||
0b7b21e4 | 3913 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3914 | if (force_nonblock && ret == -EAGAIN) |
3915 | return -EAGAIN; | |
3916 | if (ret == -ERESTARTSYS) | |
3917 | ret = -EINTR; | |
3918 | } | |
3919 | ||
bcda7baa JA |
3920 | kfree(kbuf); |
3921 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3922 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3923 | if (ret < 0) |
3924 | req_set_fail_links(req); | |
014db007 | 3925 | io_put_req(req); |
fddaface | 3926 | return 0; |
fddaface JA |
3927 | } |
3928 | ||
3529d8c2 | 3929 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3930 | { |
8ed8d3c3 JA |
3931 | struct io_accept *accept = &req->accept; |
3932 | ||
17f2fe35 JA |
3933 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3934 | return -EINVAL; | |
8042d6ce | 3935 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3936 | return -EINVAL; |
3937 | ||
d55e5f5b JA |
3938 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3939 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3940 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3941 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3942 | return 0; |
8ed8d3c3 | 3943 | } |
17f2fe35 | 3944 | |
014db007 | 3945 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3946 | { |
3947 | struct io_accept *accept = &req->accept; | |
3948 | unsigned file_flags; | |
3949 | int ret; | |
3950 | ||
3951 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3952 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3953 | accept->addr_len, accept->flags, |
3954 | accept->nofile); | |
8ed8d3c3 | 3955 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3956 | return -EAGAIN; |
8e3cca12 JA |
3957 | if (ret == -ERESTARTSYS) |
3958 | ret = -EINTR; | |
4e88d6e7 JA |
3959 | if (ret < 0) |
3960 | req_set_fail_links(req); | |
78e19bbe | 3961 | io_cqring_add_event(req, ret); |
014db007 | 3962 | io_put_req(req); |
17f2fe35 | 3963 | return 0; |
8ed8d3c3 JA |
3964 | } |
3965 | ||
3966 | static void io_accept_finish(struct io_wq_work **workptr) | |
3967 | { | |
3968 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3969 | |
3970 | if (io_req_cancelled(req)) | |
3971 | return; | |
014db007 | 3972 | __io_accept(req, false); |
e9fd9396 | 3973 | io_steal_work(req, workptr); |
8ed8d3c3 | 3974 | } |
8ed8d3c3 | 3975 | |
014db007 | 3976 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3977 | { |
8ed8d3c3 JA |
3978 | int ret; |
3979 | ||
014db007 | 3980 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3981 | if (ret == -EAGAIN && force_nonblock) { |
3982 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3983 | return -EAGAIN; |
3984 | } | |
3985 | return 0; | |
0fa03c62 | 3986 | } |
5d17b4a4 | 3987 | |
3529d8c2 | 3988 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3989 | { |
3529d8c2 JA |
3990 | struct io_connect *conn = &req->connect; |
3991 | struct io_async_ctx *io = req->io; | |
f499a021 | 3992 | |
3fbb51c1 JA |
3993 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3994 | return -EINVAL; | |
3995 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3996 | return -EINVAL; | |
3997 | ||
3529d8c2 JA |
3998 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3999 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4000 | ||
4001 | if (!io) | |
4002 | return 0; | |
4003 | ||
4004 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4005 | &io->connect.address); |
f499a021 JA |
4006 | } |
4007 | ||
014db007 | 4008 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4009 | { |
f499a021 | 4010 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4011 | unsigned file_flags; |
3fbb51c1 | 4012 | int ret; |
f8e85cf2 | 4013 | |
f499a021 JA |
4014 | if (req->io) { |
4015 | io = req->io; | |
4016 | } else { | |
3529d8c2 JA |
4017 | ret = move_addr_to_kernel(req->connect.addr, |
4018 | req->connect.addr_len, | |
4019 | &__io.connect.address); | |
f499a021 JA |
4020 | if (ret) |
4021 | goto out; | |
4022 | io = &__io; | |
4023 | } | |
4024 | ||
3fbb51c1 JA |
4025 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4026 | ||
4027 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4028 | req->connect.addr_len, file_flags); | |
87f80d62 | 4029 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4030 | if (req->io) |
4031 | return -EAGAIN; | |
4032 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4033 | ret = -ENOMEM; |
4034 | goto out; | |
4035 | } | |
b7bb4f7d | 4036 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4037 | return -EAGAIN; |
f499a021 | 4038 | } |
f8e85cf2 JA |
4039 | if (ret == -ERESTARTSYS) |
4040 | ret = -EINTR; | |
f499a021 | 4041 | out: |
4e88d6e7 JA |
4042 | if (ret < 0) |
4043 | req_set_fail_links(req); | |
f8e85cf2 | 4044 | io_cqring_add_event(req, ret); |
014db007 | 4045 | io_put_req(req); |
f8e85cf2 | 4046 | return 0; |
469956e8 Y |
4047 | } |
4048 | #else /* !CONFIG_NET */ | |
4049 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4050 | { | |
f8e85cf2 | 4051 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4052 | } |
4053 | ||
469956e8 Y |
4054 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4055 | { | |
4056 | return -EOPNOTSUPP; | |
4057 | } | |
4058 | ||
4059 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4060 | { | |
4061 | return -EOPNOTSUPP; | |
4062 | } | |
4063 | ||
4064 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4065 | const struct io_uring_sqe *sqe) | |
4066 | { | |
4067 | return -EOPNOTSUPP; | |
4068 | } | |
4069 | ||
4070 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4071 | { | |
4072 | return -EOPNOTSUPP; | |
4073 | } | |
4074 | ||
4075 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4076 | { | |
4077 | return -EOPNOTSUPP; | |
4078 | } | |
4079 | ||
4080 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4081 | { | |
4082 | return -EOPNOTSUPP; | |
4083 | } | |
4084 | ||
4085 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4086 | { | |
4087 | return -EOPNOTSUPP; | |
4088 | } | |
4089 | ||
4090 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4091 | { | |
4092 | return -EOPNOTSUPP; | |
4093 | } | |
4094 | ||
4095 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4096 | { | |
f8e85cf2 | 4097 | return -EOPNOTSUPP; |
f8e85cf2 | 4098 | } |
469956e8 | 4099 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4100 | |
d7718a9d JA |
4101 | struct io_poll_table { |
4102 | struct poll_table_struct pt; | |
4103 | struct io_kiocb *req; | |
4104 | int error; | |
4105 | }; | |
4106 | ||
4107 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4108 | struct wait_queue_head *head) | |
4109 | { | |
4110 | if (unlikely(poll->head)) { | |
4111 | pt->error = -EINVAL; | |
4112 | return; | |
4113 | } | |
4114 | ||
4115 | pt->error = 0; | |
4116 | poll->head = head; | |
4117 | add_wait_queue(head, &poll->wait); | |
4118 | } | |
4119 | ||
4120 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4121 | struct poll_table_struct *p) | |
4122 | { | |
4123 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4124 | ||
4125 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4126 | } | |
4127 | ||
4128 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4129 | __poll_t mask, task_work_func_t func) | |
4130 | { | |
4131 | struct task_struct *tsk; | |
aa96bf8a | 4132 | int ret; |
d7718a9d JA |
4133 | |
4134 | /* for instances that support it check for an event match first: */ | |
4135 | if (mask && !(mask & poll->events)) | |
4136 | return 0; | |
4137 | ||
4138 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4139 | ||
4140 | list_del_init(&poll->wait.entry); | |
4141 | ||
4142 | tsk = req->task; | |
4143 | req->result = mask; | |
4144 | init_task_work(&req->task_work, func); | |
4145 | /* | |
aa96bf8a JA |
4146 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4147 | * threads to ensure the work gets run, we can't always rely on exit | |
4148 | * cancelation taking care of this. | |
d7718a9d | 4149 | */ |
aa96bf8a JA |
4150 | ret = task_work_add(tsk, &req->task_work, true); |
4151 | if (unlikely(ret)) { | |
4152 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4153 | task_work_add(tsk, &req->task_work, true); | |
4154 | } | |
d7718a9d JA |
4155 | wake_up_process(tsk); |
4156 | return 1; | |
4157 | } | |
4158 | ||
4159 | static void io_async_task_func(struct callback_head *cb) | |
4160 | { | |
4161 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4162 | struct async_poll *apoll = req->apoll; | |
4163 | struct io_ring_ctx *ctx = req->ctx; | |
4164 | ||
4165 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4166 | ||
4167 | WARN_ON_ONCE(!list_empty(&req->apoll->poll.wait.entry)); | |
4168 | ||
4169 | if (hash_hashed(&req->hash_node)) { | |
4170 | spin_lock_irq(&ctx->completion_lock); | |
4171 | hash_del(&req->hash_node); | |
4172 | spin_unlock_irq(&ctx->completion_lock); | |
4173 | } | |
4174 | ||
4175 | /* restore ->work in case we need to retry again */ | |
4176 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4177 | ||
4178 | __set_current_state(TASK_RUNNING); | |
4179 | mutex_lock(&ctx->uring_lock); | |
4180 | __io_queue_sqe(req, NULL); | |
4181 | mutex_unlock(&ctx->uring_lock); | |
4182 | ||
4183 | kfree(apoll); | |
4184 | } | |
4185 | ||
4186 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4187 | void *key) | |
4188 | { | |
4189 | struct io_kiocb *req = wait->private; | |
4190 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4191 | ||
4192 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4193 | key_to_poll(key)); | |
4194 | ||
4195 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4196 | } | |
4197 | ||
4198 | static void io_poll_req_insert(struct io_kiocb *req) | |
4199 | { | |
4200 | struct io_ring_ctx *ctx = req->ctx; | |
4201 | struct hlist_head *list; | |
4202 | ||
4203 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4204 | hlist_add_head(&req->hash_node, list); | |
4205 | } | |
4206 | ||
4207 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4208 | struct io_poll_iocb *poll, | |
4209 | struct io_poll_table *ipt, __poll_t mask, | |
4210 | wait_queue_func_t wake_func) | |
4211 | __acquires(&ctx->completion_lock) | |
4212 | { | |
4213 | struct io_ring_ctx *ctx = req->ctx; | |
4214 | bool cancel = false; | |
4215 | ||
4216 | poll->file = req->file; | |
4217 | poll->head = NULL; | |
4218 | poll->done = poll->canceled = false; | |
4219 | poll->events = mask; | |
4220 | ||
4221 | ipt->pt._key = mask; | |
4222 | ipt->req = req; | |
4223 | ipt->error = -EINVAL; | |
4224 | ||
4225 | INIT_LIST_HEAD(&poll->wait.entry); | |
4226 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4227 | poll->wait.private = req; | |
4228 | ||
4229 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4230 | ||
4231 | spin_lock_irq(&ctx->completion_lock); | |
4232 | if (likely(poll->head)) { | |
4233 | spin_lock(&poll->head->lock); | |
4234 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4235 | if (ipt->error) | |
4236 | cancel = true; | |
4237 | ipt->error = 0; | |
4238 | mask = 0; | |
4239 | } | |
4240 | if (mask || ipt->error) | |
4241 | list_del_init(&poll->wait.entry); | |
4242 | else if (cancel) | |
4243 | WRITE_ONCE(poll->canceled, true); | |
4244 | else if (!poll->done) /* actually waiting for an event */ | |
4245 | io_poll_req_insert(req); | |
4246 | spin_unlock(&poll->head->lock); | |
4247 | } | |
4248 | ||
4249 | return mask; | |
4250 | } | |
4251 | ||
4252 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4253 | { | |
4254 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4255 | struct io_ring_ctx *ctx = req->ctx; | |
4256 | struct async_poll *apoll; | |
4257 | struct io_poll_table ipt; | |
4258 | __poll_t mask, ret; | |
4259 | ||
4260 | if (!req->file || !file_can_poll(req->file)) | |
4261 | return false; | |
4262 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4263 | return false; | |
4264 | if (!def->pollin && !def->pollout) | |
4265 | return false; | |
4266 | ||
4267 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4268 | if (unlikely(!apoll)) | |
4269 | return false; | |
4270 | ||
4271 | req->flags |= REQ_F_POLLED; | |
4272 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4273 | ||
3537b6a7 | 4274 | get_task_struct(current); |
d7718a9d JA |
4275 | req->task = current; |
4276 | req->apoll = apoll; | |
4277 | INIT_HLIST_NODE(&req->hash_node); | |
4278 | ||
8755d97a | 4279 | mask = 0; |
d7718a9d | 4280 | if (def->pollin) |
8755d97a | 4281 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4282 | if (def->pollout) |
4283 | mask |= POLLOUT | POLLWRNORM; | |
4284 | mask |= POLLERR | POLLPRI; | |
4285 | ||
4286 | ipt.pt._qproc = io_async_queue_proc; | |
4287 | ||
4288 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4289 | io_async_wake); | |
4290 | if (ret) { | |
4291 | ipt.error = 0; | |
4292 | apoll->poll.done = true; | |
4293 | spin_unlock_irq(&ctx->completion_lock); | |
4294 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4295 | kfree(apoll); | |
4296 | return false; | |
4297 | } | |
4298 | spin_unlock_irq(&ctx->completion_lock); | |
4299 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4300 | apoll->poll.events); | |
4301 | return true; | |
4302 | } | |
4303 | ||
4304 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4305 | struct io_poll_iocb *poll) | |
221c5eb2 | 4306 | { |
b41e9852 | 4307 | bool do_complete = false; |
221c5eb2 JA |
4308 | |
4309 | spin_lock(&poll->head->lock); | |
4310 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4311 | if (!list_empty(&poll->wait.entry)) { |
4312 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4313 | do_complete = true; |
221c5eb2 JA |
4314 | } |
4315 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4316 | return do_complete; |
4317 | } | |
4318 | ||
4319 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4320 | { | |
b1f573bd | 4321 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4322 | bool do_complete; |
4323 | ||
4324 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4325 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4326 | } else { | |
b1f573bd | 4327 | apoll = req->apoll; |
d7718a9d JA |
4328 | /* non-poll requests have submit ref still */ |
4329 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4330 | if (do_complete) | |
4331 | io_put_req(req); | |
4332 | } | |
4333 | ||
78076bb6 | 4334 | hash_del(&req->hash_node); |
d7718a9d | 4335 | |
b1f573bd XW |
4336 | if (apoll) { |
4337 | /* | |
4338 | * restore ->work because we need to call io_req_work_drop_env. | |
4339 | */ | |
4340 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4341 | kfree(apoll); | |
4342 | } | |
4343 | ||
b41e9852 JA |
4344 | if (do_complete) { |
4345 | io_cqring_fill_event(req, -ECANCELED); | |
4346 | io_commit_cqring(req->ctx); | |
4347 | req->flags |= REQ_F_COMP_LOCKED; | |
4348 | io_put_req(req); | |
4349 | } | |
4350 | ||
4351 | return do_complete; | |
221c5eb2 JA |
4352 | } |
4353 | ||
4354 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4355 | { | |
78076bb6 | 4356 | struct hlist_node *tmp; |
221c5eb2 | 4357 | struct io_kiocb *req; |
78076bb6 | 4358 | int i; |
221c5eb2 JA |
4359 | |
4360 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4361 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4362 | struct hlist_head *list; | |
4363 | ||
4364 | list = &ctx->cancel_hash[i]; | |
4365 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
4366 | io_poll_remove_one(req); | |
221c5eb2 JA |
4367 | } |
4368 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 JA |
4369 | |
4370 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4371 | } |
4372 | ||
47f46768 JA |
4373 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4374 | { | |
78076bb6 | 4375 | struct hlist_head *list; |
47f46768 JA |
4376 | struct io_kiocb *req; |
4377 | ||
78076bb6 JA |
4378 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4379 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4380 | if (sqe_addr != req->user_data) |
4381 | continue; | |
4382 | if (io_poll_remove_one(req)) | |
eac406c6 | 4383 | return 0; |
b41e9852 | 4384 | return -EALREADY; |
47f46768 JA |
4385 | } |
4386 | ||
4387 | return -ENOENT; | |
4388 | } | |
4389 | ||
3529d8c2 JA |
4390 | static int io_poll_remove_prep(struct io_kiocb *req, |
4391 | const struct io_uring_sqe *sqe) | |
0969e783 | 4392 | { |
0969e783 JA |
4393 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4394 | return -EINVAL; | |
4395 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4396 | sqe->poll_events) | |
4397 | return -EINVAL; | |
4398 | ||
4399 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4400 | return 0; |
4401 | } | |
4402 | ||
221c5eb2 JA |
4403 | /* |
4404 | * Find a running poll command that matches one specified in sqe->addr, | |
4405 | * and remove it if found. | |
4406 | */ | |
fc4df999 | 4407 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4408 | { |
4409 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4410 | u64 addr; |
47f46768 | 4411 | int ret; |
221c5eb2 | 4412 | |
0969e783 | 4413 | addr = req->poll.addr; |
221c5eb2 | 4414 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4415 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4416 | spin_unlock_irq(&ctx->completion_lock); |
4417 | ||
78e19bbe | 4418 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4419 | if (ret < 0) |
4420 | req_set_fail_links(req); | |
e65ef56d | 4421 | io_put_req(req); |
221c5eb2 JA |
4422 | return 0; |
4423 | } | |
4424 | ||
b0dd8a41 | 4425 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4426 | { |
a197f664 JL |
4427 | struct io_ring_ctx *ctx = req->ctx; |
4428 | ||
8c838788 | 4429 | req->poll.done = true; |
b0a20349 | 4430 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4431 | io_commit_cqring(ctx); |
221c5eb2 JA |
4432 | } |
4433 | ||
b41e9852 | 4434 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4435 | { |
221c5eb2 | 4436 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4437 | struct io_poll_iocb *poll = &req->poll; |
4438 | ||
4439 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4440 | struct poll_table_struct pt = { ._key = poll->events }; | |
4441 | ||
4442 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4443 | } | |
221c5eb2 | 4444 | |
221c5eb2 | 4445 | spin_lock_irq(&ctx->completion_lock); |
a6ba632d JA |
4446 | if (!req->result && !READ_ONCE(poll->canceled)) { |
4447 | add_wait_queue(poll->head, &poll->wait); | |
4448 | spin_unlock_irq(&ctx->completion_lock); | |
4449 | return; | |
4450 | } | |
78076bb6 | 4451 | hash_del(&req->hash_node); |
b41e9852 JA |
4452 | io_poll_complete(req, req->result, 0); |
4453 | req->flags |= REQ_F_COMP_LOCKED; | |
4454 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4455 | spin_unlock_irq(&ctx->completion_lock); |
4456 | ||
8c838788 | 4457 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4458 | } |
4459 | ||
b41e9852 | 4460 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4461 | { |
b41e9852 JA |
4462 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4463 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4464 | |
b41e9852 | 4465 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4466 | if (nxt) { |
4467 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4468 | |
d7718a9d | 4469 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4470 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4471 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4472 | } |
f0b493e6 JA |
4473 | } |
4474 | ||
221c5eb2 JA |
4475 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4476 | void *key) | |
4477 | { | |
c2f2eb7d JA |
4478 | struct io_kiocb *req = wait->private; |
4479 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4480 | |
d7718a9d | 4481 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4482 | } |
4483 | ||
221c5eb2 JA |
4484 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4485 | struct poll_table_struct *p) | |
4486 | { | |
4487 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4488 | ||
d7718a9d | 4489 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4490 | } |
4491 | ||
3529d8c2 | 4492 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4493 | { |
4494 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4495 | u16 events; |
221c5eb2 JA |
4496 | |
4497 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4498 | return -EINVAL; | |
4499 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4500 | return -EINVAL; | |
09bb8394 JA |
4501 | if (!poll->file) |
4502 | return -EBADF; | |
221c5eb2 | 4503 | |
221c5eb2 JA |
4504 | events = READ_ONCE(sqe->poll_events); |
4505 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4506 | |
3537b6a7 | 4507 | get_task_struct(current); |
b41e9852 | 4508 | req->task = current; |
0969e783 JA |
4509 | return 0; |
4510 | } | |
4511 | ||
014db007 | 4512 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4513 | { |
4514 | struct io_poll_iocb *poll = &req->poll; | |
4515 | struct io_ring_ctx *ctx = req->ctx; | |
4516 | struct io_poll_table ipt; | |
0969e783 | 4517 | __poll_t mask; |
0969e783 | 4518 | |
78076bb6 | 4519 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4520 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4521 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4522 | |
d7718a9d JA |
4523 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4524 | io_poll_wake); | |
221c5eb2 | 4525 | |
8c838788 | 4526 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4527 | ipt.error = 0; |
b0dd8a41 | 4528 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4529 | } |
221c5eb2 JA |
4530 | spin_unlock_irq(&ctx->completion_lock); |
4531 | ||
8c838788 JA |
4532 | if (mask) { |
4533 | io_cqring_ev_posted(ctx); | |
014db007 | 4534 | io_put_req(req); |
221c5eb2 | 4535 | } |
8c838788 | 4536 | return ipt.error; |
221c5eb2 JA |
4537 | } |
4538 | ||
5262f567 JA |
4539 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4540 | { | |
ad8a48ac JA |
4541 | struct io_timeout_data *data = container_of(timer, |
4542 | struct io_timeout_data, timer); | |
4543 | struct io_kiocb *req = data->req; | |
4544 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4545 | unsigned long flags; |
4546 | ||
5262f567 JA |
4547 | atomic_inc(&ctx->cq_timeouts); |
4548 | ||
4549 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4550 | /* |
11365043 JA |
4551 | * We could be racing with timeout deletion. If the list is empty, |
4552 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4553 | */ |
842f9612 | 4554 | if (!list_empty(&req->list)) { |
11365043 | 4555 | struct io_kiocb *prev; |
5262f567 | 4556 | |
11365043 JA |
4557 | /* |
4558 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4559 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4560 | * pointer will be increased, otherwise other timeout reqs may |
4561 | * return in advance without waiting for enough wait_nr. | |
4562 | */ | |
4563 | prev = req; | |
4564 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4565 | prev->sequence++; | |
11365043 | 4566 | list_del_init(&req->list); |
11365043 | 4567 | } |
5262f567 | 4568 | |
78e19bbe | 4569 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4570 | io_commit_cqring(ctx); |
4571 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4572 | ||
4573 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4574 | req_set_fail_links(req); |
5262f567 JA |
4575 | io_put_req(req); |
4576 | return HRTIMER_NORESTART; | |
4577 | } | |
4578 | ||
47f46768 JA |
4579 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4580 | { | |
4581 | struct io_kiocb *req; | |
4582 | int ret = -ENOENT; | |
4583 | ||
4584 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4585 | if (user_data == req->user_data) { | |
4586 | list_del_init(&req->list); | |
4587 | ret = 0; | |
4588 | break; | |
4589 | } | |
4590 | } | |
4591 | ||
4592 | if (ret == -ENOENT) | |
4593 | return ret; | |
4594 | ||
2d28390a | 4595 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4596 | if (ret == -1) |
4597 | return -EALREADY; | |
4598 | ||
4e88d6e7 | 4599 | req_set_fail_links(req); |
47f46768 JA |
4600 | io_cqring_fill_event(req, -ECANCELED); |
4601 | io_put_req(req); | |
4602 | return 0; | |
4603 | } | |
4604 | ||
3529d8c2 JA |
4605 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4606 | const struct io_uring_sqe *sqe) | |
b29472ee | 4607 | { |
b29472ee JA |
4608 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4609 | return -EINVAL; | |
4610 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4611 | return -EINVAL; | |
4612 | ||
4613 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4614 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4615 | if (req->timeout.flags) | |
4616 | return -EINVAL; | |
4617 | ||
b29472ee JA |
4618 | return 0; |
4619 | } | |
4620 | ||
11365043 JA |
4621 | /* |
4622 | * Remove or update an existing timeout command | |
4623 | */ | |
fc4df999 | 4624 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4625 | { |
4626 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4627 | int ret; |
11365043 | 4628 | |
11365043 | 4629 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4630 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4631 | |
47f46768 | 4632 | io_cqring_fill_event(req, ret); |
11365043 JA |
4633 | io_commit_cqring(ctx); |
4634 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4635 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4636 | if (ret < 0) |
4637 | req_set_fail_links(req); | |
ec9c02ad | 4638 | io_put_req(req); |
11365043 | 4639 | return 0; |
5262f567 JA |
4640 | } |
4641 | ||
3529d8c2 | 4642 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4643 | bool is_timeout_link) |
5262f567 | 4644 | { |
ad8a48ac | 4645 | struct io_timeout_data *data; |
a41525ab | 4646 | unsigned flags; |
5262f567 | 4647 | |
ad8a48ac | 4648 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4649 | return -EINVAL; |
ad8a48ac | 4650 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4651 | return -EINVAL; |
2d28390a JA |
4652 | if (sqe->off && is_timeout_link) |
4653 | return -EINVAL; | |
a41525ab JA |
4654 | flags = READ_ONCE(sqe->timeout_flags); |
4655 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4656 | return -EINVAL; |
bdf20073 | 4657 | |
26a61679 JA |
4658 | req->timeout.count = READ_ONCE(sqe->off); |
4659 | ||
3529d8c2 | 4660 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4661 | return -ENOMEM; |
4662 | ||
4663 | data = &req->io->timeout; | |
ad8a48ac | 4664 | data->req = req; |
ad8a48ac JA |
4665 | req->flags |= REQ_F_TIMEOUT; |
4666 | ||
4667 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4668 | return -EFAULT; |
4669 | ||
11365043 | 4670 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4671 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4672 | else |
ad8a48ac | 4673 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4674 | |
ad8a48ac JA |
4675 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4676 | return 0; | |
4677 | } | |
4678 | ||
fc4df999 | 4679 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac JA |
4680 | { |
4681 | unsigned count; | |
4682 | struct io_ring_ctx *ctx = req->ctx; | |
4683 | struct io_timeout_data *data; | |
4684 | struct list_head *entry; | |
4685 | unsigned span = 0; | |
ad8a48ac | 4686 | |
2d28390a | 4687 | data = &req->io->timeout; |
93bd25bb | 4688 | |
5262f567 JA |
4689 | /* |
4690 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4691 | * timeout event to be satisfied. If it isn't set, then this is |
4692 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4693 | */ |
26a61679 | 4694 | count = req->timeout.count; |
93bd25bb JA |
4695 | if (!count) { |
4696 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4697 | spin_lock_irq(&ctx->completion_lock); | |
4698 | entry = ctx->timeout_list.prev; | |
4699 | goto add; | |
4700 | } | |
5262f567 JA |
4701 | |
4702 | req->sequence = ctx->cached_sq_head + count - 1; | |
2d28390a | 4703 | data->seq_offset = count; |
5262f567 JA |
4704 | |
4705 | /* | |
4706 | * Insertion sort, ensuring the first entry in the list is always | |
4707 | * the one we need first. | |
4708 | */ | |
5262f567 JA |
4709 | spin_lock_irq(&ctx->completion_lock); |
4710 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4711 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5da0fb1a | 4712 | unsigned nxt_sq_head; |
4713 | long long tmp, tmp_nxt; | |
2d28390a | 4714 | u32 nxt_offset = nxt->io->timeout.seq_offset; |
5262f567 | 4715 | |
93bd25bb JA |
4716 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4717 | continue; | |
4718 | ||
5da0fb1a | 4719 | /* |
4720 | * Since cached_sq_head + count - 1 can overflow, use type long | |
4721 | * long to store it. | |
4722 | */ | |
4723 | tmp = (long long)ctx->cached_sq_head + count - 1; | |
cc42e0ac PB |
4724 | nxt_sq_head = nxt->sequence - nxt_offset + 1; |
4725 | tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1; | |
5da0fb1a | 4726 | |
4727 | /* | |
4728 | * cached_sq_head may overflow, and it will never overflow twice | |
4729 | * once there is some timeout req still be valid. | |
4730 | */ | |
4731 | if (ctx->cached_sq_head < nxt_sq_head) | |
8b07a65a | 4732 | tmp += UINT_MAX; |
5da0fb1a | 4733 | |
a1f58ba4 | 4734 | if (tmp > tmp_nxt) |
5262f567 | 4735 | break; |
a1f58ba4 | 4736 | |
4737 | /* | |
4738 | * Sequence of reqs after the insert one and itself should | |
4739 | * be adjusted because each timeout req consumes a slot. | |
4740 | */ | |
4741 | span++; | |
4742 | nxt->sequence++; | |
5262f567 | 4743 | } |
a1f58ba4 | 4744 | req->sequence -= span; |
93bd25bb | 4745 | add: |
5262f567 | 4746 | list_add(&req->list, entry); |
ad8a48ac JA |
4747 | data->timer.function = io_timeout_fn; |
4748 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4749 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4750 | return 0; |
4751 | } | |
5262f567 | 4752 | |
62755e35 JA |
4753 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4754 | { | |
4755 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4756 | ||
4757 | return req->user_data == (unsigned long) data; | |
4758 | } | |
4759 | ||
e977d6d3 | 4760 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4761 | { |
62755e35 | 4762 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4763 | int ret = 0; |
4764 | ||
62755e35 JA |
4765 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4766 | switch (cancel_ret) { | |
4767 | case IO_WQ_CANCEL_OK: | |
4768 | ret = 0; | |
4769 | break; | |
4770 | case IO_WQ_CANCEL_RUNNING: | |
4771 | ret = -EALREADY; | |
4772 | break; | |
4773 | case IO_WQ_CANCEL_NOTFOUND: | |
4774 | ret = -ENOENT; | |
4775 | break; | |
4776 | } | |
4777 | ||
e977d6d3 JA |
4778 | return ret; |
4779 | } | |
4780 | ||
47f46768 JA |
4781 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4782 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4783 | int success_ret) |
47f46768 JA |
4784 | { |
4785 | unsigned long flags; | |
4786 | int ret; | |
4787 | ||
4788 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4789 | if (ret != -ENOENT) { | |
4790 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4791 | goto done; | |
4792 | } | |
4793 | ||
4794 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4795 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4796 | if (ret != -ENOENT) | |
4797 | goto done; | |
4798 | ret = io_poll_cancel(ctx, sqe_addr); | |
4799 | done: | |
b0dd8a41 JA |
4800 | if (!ret) |
4801 | ret = success_ret; | |
47f46768 JA |
4802 | io_cqring_fill_event(req, ret); |
4803 | io_commit_cqring(ctx); | |
4804 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4805 | io_cqring_ev_posted(ctx); | |
4806 | ||
4e88d6e7 JA |
4807 | if (ret < 0) |
4808 | req_set_fail_links(req); | |
014db007 | 4809 | io_put_req(req); |
47f46768 JA |
4810 | } |
4811 | ||
3529d8c2 JA |
4812 | static int io_async_cancel_prep(struct io_kiocb *req, |
4813 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4814 | { |
fbf23849 | 4815 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4816 | return -EINVAL; |
4817 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4818 | sqe->cancel_flags) | |
4819 | return -EINVAL; | |
4820 | ||
fbf23849 JA |
4821 | req->cancel.addr = READ_ONCE(sqe->addr); |
4822 | return 0; | |
4823 | } | |
4824 | ||
014db007 | 4825 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4826 | { |
4827 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4828 | |
014db007 | 4829 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4830 | return 0; |
4831 | } | |
4832 | ||
05f3fb3c JA |
4833 | static int io_files_update_prep(struct io_kiocb *req, |
4834 | const struct io_uring_sqe *sqe) | |
4835 | { | |
4836 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4837 | return -EINVAL; | |
4838 | ||
4839 | req->files_update.offset = READ_ONCE(sqe->off); | |
4840 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4841 | if (!req->files_update.nr_args) | |
4842 | return -EINVAL; | |
4843 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4844 | return 0; | |
4845 | } | |
4846 | ||
4847 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4848 | { |
4849 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4850 | struct io_uring_files_update up; |
4851 | int ret; | |
fbf23849 | 4852 | |
f86cd20c | 4853 | if (force_nonblock) |
05f3fb3c | 4854 | return -EAGAIN; |
05f3fb3c JA |
4855 | |
4856 | up.offset = req->files_update.offset; | |
4857 | up.fds = req->files_update.arg; | |
4858 | ||
4859 | mutex_lock(&ctx->uring_lock); | |
4860 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4861 | mutex_unlock(&ctx->uring_lock); | |
4862 | ||
4863 | if (ret < 0) | |
4864 | req_set_fail_links(req); | |
4865 | io_cqring_add_event(req, ret); | |
4866 | io_put_req(req); | |
5262f567 JA |
4867 | return 0; |
4868 | } | |
4869 | ||
3529d8c2 JA |
4870 | static int io_req_defer_prep(struct io_kiocb *req, |
4871 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4872 | { |
e781573e | 4873 | ssize_t ret = 0; |
f67676d1 | 4874 | |
f1d96a8f PB |
4875 | if (!sqe) |
4876 | return 0; | |
4877 | ||
f86cd20c JA |
4878 | if (io_op_defs[req->opcode].file_table) { |
4879 | ret = io_grab_files(req); | |
4880 | if (unlikely(ret)) | |
4881 | return ret; | |
4882 | } | |
4883 | ||
cccf0ee8 JA |
4884 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4885 | ||
d625c6ee | 4886 | switch (req->opcode) { |
e781573e JA |
4887 | case IORING_OP_NOP: |
4888 | break; | |
f67676d1 JA |
4889 | case IORING_OP_READV: |
4890 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4891 | case IORING_OP_READ: |
3529d8c2 | 4892 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4893 | break; |
4894 | case IORING_OP_WRITEV: | |
4895 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4896 | case IORING_OP_WRITE: |
3529d8c2 | 4897 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4898 | break; |
0969e783 | 4899 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4900 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4901 | break; |
4902 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4903 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4904 | break; |
8ed8d3c3 | 4905 | case IORING_OP_FSYNC: |
3529d8c2 | 4906 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4907 | break; |
4908 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4909 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4910 | break; |
03b1230c | 4911 | case IORING_OP_SENDMSG: |
fddaface | 4912 | case IORING_OP_SEND: |
3529d8c2 | 4913 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4914 | break; |
4915 | case IORING_OP_RECVMSG: | |
fddaface | 4916 | case IORING_OP_RECV: |
3529d8c2 | 4917 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4918 | break; |
f499a021 | 4919 | case IORING_OP_CONNECT: |
3529d8c2 | 4920 | ret = io_connect_prep(req, sqe); |
f499a021 | 4921 | break; |
2d28390a | 4922 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4923 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4924 | break; |
b29472ee | 4925 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4926 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4927 | break; |
fbf23849 | 4928 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4929 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4930 | break; |
2d28390a | 4931 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4932 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4933 | break; |
8ed8d3c3 | 4934 | case IORING_OP_ACCEPT: |
3529d8c2 | 4935 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4936 | break; |
d63d1b5e JA |
4937 | case IORING_OP_FALLOCATE: |
4938 | ret = io_fallocate_prep(req, sqe); | |
4939 | break; | |
15b71abe JA |
4940 | case IORING_OP_OPENAT: |
4941 | ret = io_openat_prep(req, sqe); | |
4942 | break; | |
b5dba59e JA |
4943 | case IORING_OP_CLOSE: |
4944 | ret = io_close_prep(req, sqe); | |
4945 | break; | |
05f3fb3c JA |
4946 | case IORING_OP_FILES_UPDATE: |
4947 | ret = io_files_update_prep(req, sqe); | |
4948 | break; | |
eddc7ef5 JA |
4949 | case IORING_OP_STATX: |
4950 | ret = io_statx_prep(req, sqe); | |
4951 | break; | |
4840e418 JA |
4952 | case IORING_OP_FADVISE: |
4953 | ret = io_fadvise_prep(req, sqe); | |
4954 | break; | |
c1ca757b JA |
4955 | case IORING_OP_MADVISE: |
4956 | ret = io_madvise_prep(req, sqe); | |
4957 | break; | |
cebdb986 JA |
4958 | case IORING_OP_OPENAT2: |
4959 | ret = io_openat2_prep(req, sqe); | |
4960 | break; | |
3e4827b0 JA |
4961 | case IORING_OP_EPOLL_CTL: |
4962 | ret = io_epoll_ctl_prep(req, sqe); | |
4963 | break; | |
7d67af2c PB |
4964 | case IORING_OP_SPLICE: |
4965 | ret = io_splice_prep(req, sqe); | |
4966 | break; | |
ddf0322d JA |
4967 | case IORING_OP_PROVIDE_BUFFERS: |
4968 | ret = io_provide_buffers_prep(req, sqe); | |
4969 | break; | |
067524e9 JA |
4970 | case IORING_OP_REMOVE_BUFFERS: |
4971 | ret = io_remove_buffers_prep(req, sqe); | |
4972 | break; | |
f67676d1 | 4973 | default: |
e781573e JA |
4974 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4975 | req->opcode); | |
4976 | ret = -EINVAL; | |
b7bb4f7d | 4977 | break; |
f67676d1 JA |
4978 | } |
4979 | ||
b7bb4f7d | 4980 | return ret; |
f67676d1 JA |
4981 | } |
4982 | ||
3529d8c2 | 4983 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 4984 | { |
a197f664 | 4985 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 4986 | int ret; |
de0617e4 | 4987 | |
9d858b21 BL |
4988 | /* Still need defer if there is pending req in defer list. */ |
4989 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
4990 | return 0; |
4991 | ||
3529d8c2 | 4992 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
4993 | return -EAGAIN; |
4994 | ||
3529d8c2 | 4995 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 4996 | if (ret < 0) |
2d28390a | 4997 | return ret; |
2d28390a | 4998 | |
de0617e4 | 4999 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5000 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5001 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5002 | return 0; |
5003 | } | |
5004 | ||
915967f6 | 5005 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5006 | list_add_tail(&req->list, &ctx->defer_list); |
5007 | spin_unlock_irq(&ctx->completion_lock); | |
5008 | return -EIOCBQUEUED; | |
5009 | } | |
5010 | ||
99bc4c38 PB |
5011 | static void io_cleanup_req(struct io_kiocb *req) |
5012 | { | |
5013 | struct io_async_ctx *io = req->io; | |
5014 | ||
5015 | switch (req->opcode) { | |
5016 | case IORING_OP_READV: | |
5017 | case IORING_OP_READ_FIXED: | |
5018 | case IORING_OP_READ: | |
bcda7baa JA |
5019 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5020 | kfree((void *)(unsigned long)req->rw.addr); | |
5021 | /* fallthrough */ | |
99bc4c38 PB |
5022 | case IORING_OP_WRITEV: |
5023 | case IORING_OP_WRITE_FIXED: | |
5024 | case IORING_OP_WRITE: | |
5025 | if (io->rw.iov != io->rw.fast_iov) | |
5026 | kfree(io->rw.iov); | |
5027 | break; | |
99bc4c38 | 5028 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5029 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5030 | kfree(req->sr_msg.kbuf); | |
5031 | /* fallthrough */ | |
5032 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5033 | if (io->msg.iov != io->msg.fast_iov) |
5034 | kfree(io->msg.iov); | |
5035 | break; | |
bcda7baa JA |
5036 | case IORING_OP_RECV: |
5037 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5038 | kfree(req->sr_msg.kbuf); | |
5039 | break; | |
8fef80bf PB |
5040 | case IORING_OP_OPENAT: |
5041 | case IORING_OP_OPENAT2: | |
5042 | case IORING_OP_STATX: | |
5043 | putname(req->open.filename); | |
5044 | break; | |
7d67af2c PB |
5045 | case IORING_OP_SPLICE: |
5046 | io_put_file(req, req->splice.file_in, | |
5047 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5048 | break; | |
99bc4c38 PB |
5049 | } |
5050 | ||
5051 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5052 | } | |
5053 | ||
3529d8c2 | 5054 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5055 | bool force_nonblock) |
2b188cc1 | 5056 | { |
a197f664 | 5057 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5058 | int ret; |
2b188cc1 | 5059 | |
d625c6ee | 5060 | switch (req->opcode) { |
2b188cc1 | 5061 | case IORING_OP_NOP: |
78e19bbe | 5062 | ret = io_nop(req); |
2b188cc1 JA |
5063 | break; |
5064 | case IORING_OP_READV: | |
edafccee | 5065 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5066 | case IORING_OP_READ: |
3529d8c2 JA |
5067 | if (sqe) { |
5068 | ret = io_read_prep(req, sqe, force_nonblock); | |
5069 | if (ret < 0) | |
5070 | break; | |
5071 | } | |
014db007 | 5072 | ret = io_read(req, force_nonblock); |
edafccee | 5073 | break; |
3529d8c2 | 5074 | case IORING_OP_WRITEV: |
edafccee | 5075 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5076 | case IORING_OP_WRITE: |
3529d8c2 JA |
5077 | if (sqe) { |
5078 | ret = io_write_prep(req, sqe, force_nonblock); | |
5079 | if (ret < 0) | |
5080 | break; | |
5081 | } | |
014db007 | 5082 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5083 | break; |
c992fe29 | 5084 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5085 | if (sqe) { |
5086 | ret = io_prep_fsync(req, sqe); | |
5087 | if (ret < 0) | |
5088 | break; | |
5089 | } | |
014db007 | 5090 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5091 | break; |
221c5eb2 | 5092 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5093 | if (sqe) { |
5094 | ret = io_poll_add_prep(req, sqe); | |
5095 | if (ret) | |
5096 | break; | |
5097 | } | |
014db007 | 5098 | ret = io_poll_add(req); |
221c5eb2 JA |
5099 | break; |
5100 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5101 | if (sqe) { |
5102 | ret = io_poll_remove_prep(req, sqe); | |
5103 | if (ret < 0) | |
5104 | break; | |
5105 | } | |
fc4df999 | 5106 | ret = io_poll_remove(req); |
221c5eb2 | 5107 | break; |
5d17b4a4 | 5108 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5109 | if (sqe) { |
5110 | ret = io_prep_sfr(req, sqe); | |
5111 | if (ret < 0) | |
5112 | break; | |
5113 | } | |
014db007 | 5114 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5115 | break; |
0fa03c62 | 5116 | case IORING_OP_SENDMSG: |
fddaface | 5117 | case IORING_OP_SEND: |
3529d8c2 JA |
5118 | if (sqe) { |
5119 | ret = io_sendmsg_prep(req, sqe); | |
5120 | if (ret < 0) | |
5121 | break; | |
5122 | } | |
fddaface | 5123 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5124 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5125 | else |
014db007 | 5126 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5127 | break; |
aa1fa28f | 5128 | case IORING_OP_RECVMSG: |
fddaface | 5129 | case IORING_OP_RECV: |
3529d8c2 JA |
5130 | if (sqe) { |
5131 | ret = io_recvmsg_prep(req, sqe); | |
5132 | if (ret) | |
5133 | break; | |
5134 | } | |
fddaface | 5135 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5136 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5137 | else |
014db007 | 5138 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5139 | break; |
5262f567 | 5140 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5141 | if (sqe) { |
5142 | ret = io_timeout_prep(req, sqe, false); | |
5143 | if (ret) | |
5144 | break; | |
5145 | } | |
fc4df999 | 5146 | ret = io_timeout(req); |
5262f567 | 5147 | break; |
11365043 | 5148 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5149 | if (sqe) { |
5150 | ret = io_timeout_remove_prep(req, sqe); | |
5151 | if (ret) | |
5152 | break; | |
5153 | } | |
fc4df999 | 5154 | ret = io_timeout_remove(req); |
11365043 | 5155 | break; |
17f2fe35 | 5156 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5157 | if (sqe) { |
5158 | ret = io_accept_prep(req, sqe); | |
5159 | if (ret) | |
5160 | break; | |
5161 | } | |
014db007 | 5162 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5163 | break; |
f8e85cf2 | 5164 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5165 | if (sqe) { |
5166 | ret = io_connect_prep(req, sqe); | |
5167 | if (ret) | |
5168 | break; | |
5169 | } | |
014db007 | 5170 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5171 | break; |
62755e35 | 5172 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5173 | if (sqe) { |
5174 | ret = io_async_cancel_prep(req, sqe); | |
5175 | if (ret) | |
5176 | break; | |
5177 | } | |
014db007 | 5178 | ret = io_async_cancel(req); |
62755e35 | 5179 | break; |
d63d1b5e JA |
5180 | case IORING_OP_FALLOCATE: |
5181 | if (sqe) { | |
5182 | ret = io_fallocate_prep(req, sqe); | |
5183 | if (ret) | |
5184 | break; | |
5185 | } | |
014db007 | 5186 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5187 | break; |
15b71abe JA |
5188 | case IORING_OP_OPENAT: |
5189 | if (sqe) { | |
5190 | ret = io_openat_prep(req, sqe); | |
5191 | if (ret) | |
5192 | break; | |
5193 | } | |
014db007 | 5194 | ret = io_openat(req, force_nonblock); |
15b71abe | 5195 | break; |
b5dba59e JA |
5196 | case IORING_OP_CLOSE: |
5197 | if (sqe) { | |
5198 | ret = io_close_prep(req, sqe); | |
5199 | if (ret) | |
5200 | break; | |
5201 | } | |
014db007 | 5202 | ret = io_close(req, force_nonblock); |
b5dba59e | 5203 | break; |
05f3fb3c JA |
5204 | case IORING_OP_FILES_UPDATE: |
5205 | if (sqe) { | |
5206 | ret = io_files_update_prep(req, sqe); | |
5207 | if (ret) | |
5208 | break; | |
5209 | } | |
5210 | ret = io_files_update(req, force_nonblock); | |
5211 | break; | |
eddc7ef5 JA |
5212 | case IORING_OP_STATX: |
5213 | if (sqe) { | |
5214 | ret = io_statx_prep(req, sqe); | |
5215 | if (ret) | |
5216 | break; | |
5217 | } | |
014db007 | 5218 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5219 | break; |
4840e418 JA |
5220 | case IORING_OP_FADVISE: |
5221 | if (sqe) { | |
5222 | ret = io_fadvise_prep(req, sqe); | |
5223 | if (ret) | |
5224 | break; | |
5225 | } | |
014db007 | 5226 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5227 | break; |
c1ca757b JA |
5228 | case IORING_OP_MADVISE: |
5229 | if (sqe) { | |
5230 | ret = io_madvise_prep(req, sqe); | |
5231 | if (ret) | |
5232 | break; | |
5233 | } | |
014db007 | 5234 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5235 | break; |
cebdb986 JA |
5236 | case IORING_OP_OPENAT2: |
5237 | if (sqe) { | |
5238 | ret = io_openat2_prep(req, sqe); | |
5239 | if (ret) | |
5240 | break; | |
5241 | } | |
014db007 | 5242 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5243 | break; |
3e4827b0 JA |
5244 | case IORING_OP_EPOLL_CTL: |
5245 | if (sqe) { | |
5246 | ret = io_epoll_ctl_prep(req, sqe); | |
5247 | if (ret) | |
5248 | break; | |
5249 | } | |
014db007 | 5250 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5251 | break; |
7d67af2c PB |
5252 | case IORING_OP_SPLICE: |
5253 | if (sqe) { | |
5254 | ret = io_splice_prep(req, sqe); | |
5255 | if (ret < 0) | |
5256 | break; | |
5257 | } | |
014db007 | 5258 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5259 | break; |
ddf0322d JA |
5260 | case IORING_OP_PROVIDE_BUFFERS: |
5261 | if (sqe) { | |
5262 | ret = io_provide_buffers_prep(req, sqe); | |
5263 | if (ret) | |
5264 | break; | |
5265 | } | |
5266 | ret = io_provide_buffers(req, force_nonblock); | |
5267 | break; | |
067524e9 JA |
5268 | case IORING_OP_REMOVE_BUFFERS: |
5269 | if (sqe) { | |
5270 | ret = io_remove_buffers_prep(req, sqe); | |
5271 | if (ret) | |
5272 | break; | |
5273 | } | |
5274 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5275 | break; |
2b188cc1 JA |
5276 | default: |
5277 | ret = -EINVAL; | |
5278 | break; | |
5279 | } | |
5280 | ||
def596e9 JA |
5281 | if (ret) |
5282 | return ret; | |
5283 | ||
5284 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5285 | const bool in_async = io_wq_current_is_worker(); |
5286 | ||
9e645e11 | 5287 | if (req->result == -EAGAIN) |
def596e9 JA |
5288 | return -EAGAIN; |
5289 | ||
11ba820b JA |
5290 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5291 | if (in_async) | |
5292 | mutex_lock(&ctx->uring_lock); | |
5293 | ||
def596e9 | 5294 | io_iopoll_req_issued(req); |
11ba820b JA |
5295 | |
5296 | if (in_async) | |
5297 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5298 | } |
5299 | ||
5300 | return 0; | |
2b188cc1 JA |
5301 | } |
5302 | ||
561fb04a | 5303 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5304 | { |
561fb04a | 5305 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5306 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5307 | int ret = 0; |
2b188cc1 | 5308 | |
0c9d5ccd JA |
5309 | /* if NO_CANCEL is set, we must still run the work */ |
5310 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5311 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5312 | ret = -ECANCELED; |
0c9d5ccd | 5313 | } |
31b51510 | 5314 | |
561fb04a | 5315 | if (!ret) { |
561fb04a | 5316 | do { |
014db007 | 5317 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5318 | /* |
5319 | * We can get EAGAIN for polled IO even though we're | |
5320 | * forcing a sync submission from here, since we can't | |
5321 | * wait for request slots on the block side. | |
5322 | */ | |
5323 | if (ret != -EAGAIN) | |
5324 | break; | |
5325 | cond_resched(); | |
5326 | } while (1); | |
5327 | } | |
31b51510 | 5328 | |
561fb04a | 5329 | if (ret) { |
4e88d6e7 | 5330 | req_set_fail_links(req); |
78e19bbe | 5331 | io_cqring_add_event(req, ret); |
817869d2 | 5332 | io_put_req(req); |
edafccee | 5333 | } |
2b188cc1 | 5334 | |
e9fd9396 | 5335 | io_steal_work(req, workptr); |
2b188cc1 JA |
5336 | } |
5337 | ||
15b71abe | 5338 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 5339 | { |
d3656344 | 5340 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 5341 | return 0; |
0b5faf6b | 5342 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
5343 | return 0; |
5344 | return 1; | |
09bb8394 JA |
5345 | } |
5346 | ||
65e19f54 JA |
5347 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5348 | int index) | |
5349 | { | |
5350 | struct fixed_file_table *table; | |
5351 | ||
05f3fb3c JA |
5352 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5353 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5354 | } |
5355 | ||
8da11c19 PB |
5356 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5357 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5358 | { |
a197f664 | 5359 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5360 | struct file *file; |
09bb8394 | 5361 | |
8da11c19 | 5362 | if (fixed) { |
05f3fb3c | 5363 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5364 | (unsigned) fd >= ctx->nr_user_files)) |
5365 | return -EBADF; | |
b7620121 | 5366 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5367 | file = io_file_from_index(ctx, fd); |
5368 | if (!file) | |
08a45173 | 5369 | return -EBADF; |
05589553 XW |
5370 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5371 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5372 | } else { |
c826bd7a | 5373 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5374 | file = __io_file_get(state, fd); |
5375 | if (unlikely(!file)) | |
09bb8394 JA |
5376 | return -EBADF; |
5377 | } | |
5378 | ||
8da11c19 | 5379 | *out_file = file; |
09bb8394 JA |
5380 | return 0; |
5381 | } | |
5382 | ||
8da11c19 | 5383 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
9c280f90 | 5384 | int fd, unsigned int flags) |
8da11c19 | 5385 | { |
8da11c19 PB |
5386 | bool fixed; |
5387 | ||
8da11c19 PB |
5388 | if (!io_req_needs_file(req, fd)) |
5389 | return 0; | |
5390 | ||
5391 | fixed = (flags & IOSQE_FIXED_FILE); | |
5392 | if (unlikely(!fixed && req->needs_fixed_file)) | |
5393 | return -EBADF; | |
5394 | ||
5395 | return io_file_get(state, req, fd, &req->file, fixed); | |
5396 | } | |
5397 | ||
a197f664 | 5398 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5399 | { |
5400 | int ret = -EBADF; | |
a197f664 | 5401 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5402 | |
f86cd20c JA |
5403 | if (req->work.files) |
5404 | return 0; | |
b14cca0c | 5405 | if (!ctx->ring_file) |
b5dba59e JA |
5406 | return -EBADF; |
5407 | ||
fcb323cc JA |
5408 | rcu_read_lock(); |
5409 | spin_lock_irq(&ctx->inflight_lock); | |
5410 | /* | |
5411 | * We use the f_ops->flush() handler to ensure that we can flush | |
5412 | * out work accessing these files if the fd is closed. Check if | |
5413 | * the fd has changed since we started down this path, and disallow | |
5414 | * this operation if it has. | |
5415 | */ | |
b14cca0c | 5416 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5417 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5418 | req->flags |= REQ_F_INFLIGHT; | |
5419 | req->work.files = current->files; | |
5420 | ret = 0; | |
5421 | } | |
5422 | spin_unlock_irq(&ctx->inflight_lock); | |
5423 | rcu_read_unlock(); | |
5424 | ||
5425 | return ret; | |
5426 | } | |
5427 | ||
2665abfd | 5428 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5429 | { |
ad8a48ac JA |
5430 | struct io_timeout_data *data = container_of(timer, |
5431 | struct io_timeout_data, timer); | |
5432 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5433 | struct io_ring_ctx *ctx = req->ctx; |
5434 | struct io_kiocb *prev = NULL; | |
5435 | unsigned long flags; | |
2665abfd JA |
5436 | |
5437 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5438 | ||
5439 | /* | |
5440 | * We don't expect the list to be empty, that will only happen if we | |
5441 | * race with the completion of the linked work. | |
5442 | */ | |
4493233e PB |
5443 | if (!list_empty(&req->link_list)) { |
5444 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5445 | link_list); | |
5d960724 | 5446 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5447 | list_del_init(&req->link_list); |
5d960724 JA |
5448 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5449 | } else | |
76a46e06 | 5450 | prev = NULL; |
2665abfd JA |
5451 | } |
5452 | ||
5453 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5454 | ||
5455 | if (prev) { | |
4e88d6e7 | 5456 | req_set_fail_links(prev); |
014db007 | 5457 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5458 | io_put_req(prev); |
47f46768 JA |
5459 | } else { |
5460 | io_cqring_add_event(req, -ETIME); | |
5461 | io_put_req(req); | |
2665abfd | 5462 | } |
2665abfd JA |
5463 | return HRTIMER_NORESTART; |
5464 | } | |
5465 | ||
ad8a48ac | 5466 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5467 | { |
76a46e06 | 5468 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5469 | |
76a46e06 JA |
5470 | /* |
5471 | * If the list is now empty, then our linked request finished before | |
5472 | * we got a chance to setup the timer | |
5473 | */ | |
5474 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5475 | if (!list_empty(&req->link_list)) { |
2d28390a | 5476 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5477 | |
ad8a48ac JA |
5478 | data->timer.function = io_link_timeout_fn; |
5479 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5480 | data->mode); | |
2665abfd | 5481 | } |
76a46e06 | 5482 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5483 | |
2665abfd | 5484 | /* drop submission reference */ |
76a46e06 JA |
5485 | io_put_req(req); |
5486 | } | |
2665abfd | 5487 | |
ad8a48ac | 5488 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5489 | { |
5490 | struct io_kiocb *nxt; | |
5491 | ||
dea3b49c | 5492 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5493 | return NULL; |
d7718a9d JA |
5494 | /* for polled retry, if flag is set, we already went through here */ |
5495 | if (req->flags & REQ_F_POLLED) | |
5496 | return NULL; | |
2665abfd | 5497 | |
4493233e PB |
5498 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5499 | link_list); | |
d625c6ee | 5500 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5501 | return NULL; |
2665abfd | 5502 | |
76a46e06 | 5503 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5504 | return nxt; |
2665abfd JA |
5505 | } |
5506 | ||
3529d8c2 | 5507 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5508 | { |
4a0a7a18 | 5509 | struct io_kiocb *linked_timeout; |
4bc4494e | 5510 | struct io_kiocb *nxt; |
193155c8 | 5511 | const struct cred *old_creds = NULL; |
e0c5c576 | 5512 | int ret; |
2b188cc1 | 5513 | |
4a0a7a18 JA |
5514 | again: |
5515 | linked_timeout = io_prep_linked_timeout(req); | |
5516 | ||
193155c8 JA |
5517 | if (req->work.creds && req->work.creds != current_cred()) { |
5518 | if (old_creds) | |
5519 | revert_creds(old_creds); | |
5520 | if (old_creds == req->work.creds) | |
5521 | old_creds = NULL; /* restored original creds */ | |
5522 | else | |
5523 | old_creds = override_creds(req->work.creds); | |
5524 | } | |
5525 | ||
014db007 | 5526 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5527 | |
5528 | /* | |
5529 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5530 | * doesn't support non-blocking read/write attempts | |
5531 | */ | |
5532 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5533 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5534 | if (io_arm_poll_handler(req)) { |
5535 | if (linked_timeout) | |
5536 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5537 | goto exit; |
d7718a9d | 5538 | } |
86a761f8 | 5539 | punt: |
f86cd20c | 5540 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5541 | ret = io_grab_files(req); |
5542 | if (ret) | |
5543 | goto err; | |
2b188cc1 | 5544 | } |
bbad27b2 PB |
5545 | |
5546 | /* | |
5547 | * Queued up for async execution, worker will release | |
5548 | * submit reference when the iocb is actually submitted. | |
5549 | */ | |
5550 | io_queue_async_work(req); | |
4bc4494e | 5551 | goto exit; |
2b188cc1 | 5552 | } |
e65ef56d | 5553 | |
fcb323cc | 5554 | err: |
4bc4494e | 5555 | nxt = NULL; |
76a46e06 | 5556 | /* drop submission reference */ |
2a44f467 | 5557 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5558 | |
f9bd67f6 | 5559 | if (linked_timeout) { |
76a46e06 | 5560 | if (!ret) |
f9bd67f6 | 5561 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5562 | else |
f9bd67f6 | 5563 | io_put_req(linked_timeout); |
76a46e06 JA |
5564 | } |
5565 | ||
e65ef56d | 5566 | /* and drop final reference, if we failed */ |
9e645e11 | 5567 | if (ret) { |
78e19bbe | 5568 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5569 | req_set_fail_links(req); |
e65ef56d | 5570 | io_put_req(req); |
9e645e11 | 5571 | } |
4a0a7a18 JA |
5572 | if (nxt) { |
5573 | req = nxt; | |
86a761f8 PB |
5574 | |
5575 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5576 | goto punt; | |
4a0a7a18 JA |
5577 | goto again; |
5578 | } | |
4bc4494e | 5579 | exit: |
193155c8 JA |
5580 | if (old_creds) |
5581 | revert_creds(old_creds); | |
2b188cc1 JA |
5582 | } |
5583 | ||
3529d8c2 | 5584 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5585 | { |
5586 | int ret; | |
5587 | ||
3529d8c2 | 5588 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5589 | if (ret) { |
5590 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5591 | fail_req: |
78e19bbe | 5592 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5593 | req_set_fail_links(req); |
78e19bbe | 5594 | io_double_put_req(req); |
4fe2c963 | 5595 | } |
2550878f | 5596 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5597 | ret = io_req_defer_prep(req, sqe); |
5598 | if (unlikely(ret < 0)) | |
5599 | goto fail_req; | |
ce35a47a JA |
5600 | /* |
5601 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5602 | * to async execution. | |
5603 | */ | |
5604 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5605 | io_queue_async_work(req); | |
5606 | } else { | |
3529d8c2 | 5607 | __io_queue_sqe(req, sqe); |
ce35a47a | 5608 | } |
4fe2c963 JL |
5609 | } |
5610 | ||
1b4a51b6 | 5611 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5612 | { |
94ae5e77 | 5613 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5614 | io_cqring_add_event(req, -ECANCELED); |
5615 | io_double_put_req(req); | |
5616 | } else | |
3529d8c2 | 5617 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5618 | } |
5619 | ||
1d4240cc | 5620 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3529d8c2 | 5621 | struct io_submit_state *state, struct io_kiocb **link) |
9e645e11 | 5622 | { |
a197f664 | 5623 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5624 | int ret; |
9e645e11 | 5625 | |
9e645e11 JA |
5626 | /* |
5627 | * If we already have a head request, queue this one for async | |
5628 | * submittal once the head completes. If we don't have a head but | |
5629 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5630 | * submitted sync once the chain is complete. If none of those | |
5631 | * conditions are true (normal request), then just queue it. | |
5632 | */ | |
5633 | if (*link) { | |
9d76377f | 5634 | struct io_kiocb *head = *link; |
4e88d6e7 | 5635 | |
8cdf2193 PB |
5636 | /* |
5637 | * Taking sequential execution of a link, draining both sides | |
5638 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5639 | * requests in the link. So, it drains the head and the | |
5640 | * next after the link request. The last one is done via | |
5641 | * drain_next flag to persist the effect across calls. | |
5642 | */ | |
ef4ff581 | 5643 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5644 | head->flags |= REQ_F_IO_DRAIN; |
5645 | ctx->drain_next = 1; | |
5646 | } | |
1d4240cc PB |
5647 | if (io_alloc_async_ctx(req)) |
5648 | return -EAGAIN; | |
9e645e11 | 5649 | |
3529d8c2 | 5650 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5651 | if (ret) { |
4e88d6e7 | 5652 | /* fail even hard links since we don't submit */ |
9d76377f | 5653 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5654 | return ret; |
2d28390a | 5655 | } |
9d76377f PB |
5656 | trace_io_uring_link(ctx, req, head); |
5657 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5658 | |
5659 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5660 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5661 | io_queue_link_head(head); |
5662 | *link = NULL; | |
5663 | } | |
9e645e11 | 5664 | } else { |
711be031 PB |
5665 | if (unlikely(ctx->drain_next)) { |
5666 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5667 | ctx->drain_next = 0; |
711be031 | 5668 | } |
ef4ff581 | 5669 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5670 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5671 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5672 | |
1d4240cc PB |
5673 | if (io_alloc_async_ctx(req)) |
5674 | return -EAGAIN; | |
5675 | ||
711be031 PB |
5676 | ret = io_req_defer_prep(req, sqe); |
5677 | if (ret) | |
5678 | req->flags |= REQ_F_FAIL_LINK; | |
5679 | *link = req; | |
5680 | } else { | |
5681 | io_queue_sqe(req, sqe); | |
5682 | } | |
9e645e11 | 5683 | } |
2e6e1fde | 5684 | |
1d4240cc | 5685 | return 0; |
9e645e11 JA |
5686 | } |
5687 | ||
9a56a232 JA |
5688 | /* |
5689 | * Batched submission is done, ensure local IO is flushed out. | |
5690 | */ | |
5691 | static void io_submit_state_end(struct io_submit_state *state) | |
5692 | { | |
5693 | blk_finish_plug(&state->plug); | |
3d6770fb | 5694 | io_file_put(state); |
2579f913 | 5695 | if (state->free_reqs) |
6c8a3134 | 5696 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5697 | } |
5698 | ||
5699 | /* | |
5700 | * Start submission side cache. | |
5701 | */ | |
5702 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5703 | unsigned int max_ios) |
9a56a232 JA |
5704 | { |
5705 | blk_start_plug(&state->plug); | |
2579f913 | 5706 | state->free_reqs = 0; |
9a56a232 JA |
5707 | state->file = NULL; |
5708 | state->ios_left = max_ios; | |
5709 | } | |
5710 | ||
2b188cc1 JA |
5711 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5712 | { | |
75b28aff | 5713 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5714 | |
caf582c6 PB |
5715 | /* |
5716 | * Ensure any loads from the SQEs are done at this point, | |
5717 | * since once we write the new head, the application could | |
5718 | * write new data to them. | |
5719 | */ | |
5720 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5721 | } |
5722 | ||
2b188cc1 | 5723 | /* |
3529d8c2 | 5724 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5725 | * that is mapped by userspace. This means that care needs to be taken to |
5726 | * ensure that reads are stable, as we cannot rely on userspace always | |
5727 | * being a good citizen. If members of the sqe are validated and then later | |
5728 | * used, it's important that those reads are done through READ_ONCE() to | |
5729 | * prevent a re-load down the line. | |
5730 | */ | |
709b302f | 5731 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5732 | { |
75b28aff | 5733 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5734 | unsigned head; |
5735 | ||
5736 | /* | |
5737 | * The cached sq head (or cq tail) serves two purposes: | |
5738 | * | |
5739 | * 1) allows us to batch the cost of updating the user visible | |
5740 | * head updates. | |
5741 | * 2) allows the kernel side to track the head on its own, even | |
5742 | * though the application is the one updating it. | |
5743 | */ | |
ee7d46d9 | 5744 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5745 | if (likely(head < ctx->sq_entries)) |
5746 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5747 | |
5748 | /* drop invalid entries */ | |
498ccd9e | 5749 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5750 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5751 | return NULL; |
5752 | } | |
5753 | ||
5754 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5755 | { | |
5756 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5757 | } |
5758 | ||
ef4ff581 PB |
5759 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5760 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5761 | IOSQE_BUFFER_SELECT) | |
5762 | ||
5763 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5764 | const struct io_uring_sqe *sqe, | |
5765 | struct io_submit_state *state, bool async) | |
0553b8bd | 5766 | { |
ef4ff581 PB |
5767 | unsigned int sqe_flags; |
5768 | int id, fd; | |
5769 | ||
0553b8bd PB |
5770 | /* |
5771 | * All io need record the previous position, if LINK vs DARIN, | |
5772 | * it can be used to mark the position of the first IO in the | |
5773 | * link list. | |
5774 | */ | |
5775 | req->sequence = ctx->cached_sq_head; | |
5776 | req->opcode = READ_ONCE(sqe->opcode); | |
5777 | req->user_data = READ_ONCE(sqe->user_data); | |
5778 | req->io = NULL; | |
5779 | req->file = NULL; | |
5780 | req->ctx = ctx; | |
5781 | req->flags = 0; | |
5782 | /* one is dropped after submission, the other at completion */ | |
5783 | refcount_set(&req->refs, 2); | |
5784 | req->task = NULL; | |
5785 | req->result = 0; | |
ef4ff581 | 5786 | req->needs_fixed_file = async; |
0553b8bd | 5787 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5788 | |
5789 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5790 | return -EINVAL; | |
5791 | ||
5792 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5793 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5794 | return -EFAULT; | |
5795 | use_mm(ctx->sqo_mm); | |
5796 | } | |
5797 | ||
5798 | sqe_flags = READ_ONCE(sqe->flags); | |
5799 | /* enforce forwards compatibility on users */ | |
5800 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5801 | return -EINVAL; | |
5802 | ||
5803 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5804 | !io_op_defs[req->opcode].buffer_select) | |
5805 | return -EOPNOTSUPP; | |
5806 | ||
5807 | id = READ_ONCE(sqe->personality); | |
5808 | if (id) { | |
5809 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5810 | if (unlikely(!req->work.creds)) | |
5811 | return -EINVAL; | |
5812 | get_cred(req->work.creds); | |
5813 | } | |
5814 | ||
5815 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5816 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5817 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5818 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5819 | ||
5820 | fd = READ_ONCE(sqe->fd); | |
5821 | return io_req_set_file(state, req, fd, sqe_flags); | |
0553b8bd PB |
5822 | } |
5823 | ||
fb5ccc98 | 5824 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5825 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5826 | { |
5827 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5828 | struct io_kiocb *link = NULL; |
9e645e11 | 5829 | int i, submitted = 0; |
6c271ce2 | 5830 | |
c4a2ed72 | 5831 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5832 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5833 | if (!list_empty(&ctx->cq_overflow_list) && | |
5834 | !io_cqring_overflow_flush(ctx, false)) | |
5835 | return -EBUSY; | |
5836 | } | |
6c271ce2 | 5837 | |
ee7d46d9 PB |
5838 | /* make sure SQ entry isn't read before tail */ |
5839 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5840 | |
2b85edfc PB |
5841 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5842 | return -EAGAIN; | |
6c271ce2 JA |
5843 | |
5844 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5845 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5846 | statep = &state; |
5847 | } | |
5848 | ||
b14cca0c PB |
5849 | ctx->ring_fd = ring_fd; |
5850 | ctx->ring_file = ring_file; | |
5851 | ||
6c271ce2 | 5852 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5853 | const struct io_uring_sqe *sqe; |
196be95c | 5854 | struct io_kiocb *req; |
1cb1edb2 | 5855 | int err; |
fb5ccc98 | 5856 | |
b1e50e54 PB |
5857 | sqe = io_get_sqe(ctx); |
5858 | if (unlikely(!sqe)) { | |
5859 | io_consume_sqe(ctx); | |
5860 | break; | |
5861 | } | |
0553b8bd | 5862 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5863 | if (unlikely(!req)) { |
5864 | if (!submitted) | |
5865 | submitted = -EAGAIN; | |
fb5ccc98 | 5866 | break; |
196be95c | 5867 | } |
fb5ccc98 | 5868 | |
ef4ff581 | 5869 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5870 | io_consume_sqe(ctx); |
d3656344 JA |
5871 | /* will complete beyond this point, count as submitted */ |
5872 | submitted++; | |
5873 | ||
ef4ff581 | 5874 | if (unlikely(err)) { |
1cb1edb2 PB |
5875 | fail_req: |
5876 | io_cqring_add_event(req, err); | |
d3656344 | 5877 | io_double_put_req(req); |
196be95c PB |
5878 | break; |
5879 | } | |
fb5ccc98 | 5880 | |
354420f7 JA |
5881 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5882 | true, async); | |
1d4240cc PB |
5883 | err = io_submit_sqe(req, sqe, statep, &link); |
5884 | if (err) | |
5885 | goto fail_req; | |
6c271ce2 JA |
5886 | } |
5887 | ||
9466f437 PB |
5888 | if (unlikely(submitted != nr)) { |
5889 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5890 | ||
5891 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5892 | } | |
9e645e11 | 5893 | if (link) |
1b4a51b6 | 5894 | io_queue_link_head(link); |
6c271ce2 JA |
5895 | if (statep) |
5896 | io_submit_state_end(&state); | |
5897 | ||
ae9428ca PB |
5898 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5899 | io_commit_sqring(ctx); | |
5900 | ||
6c271ce2 JA |
5901 | return submitted; |
5902 | } | |
5903 | ||
bf9c2f1c PB |
5904 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5905 | { | |
5906 | struct mm_struct *mm = current->mm; | |
5907 | ||
5908 | if (mm) { | |
5909 | unuse_mm(mm); | |
5910 | mmput(mm); | |
5911 | } | |
5912 | } | |
5913 | ||
6c271ce2 JA |
5914 | static int io_sq_thread(void *data) |
5915 | { | |
6c271ce2 | 5916 | struct io_ring_ctx *ctx = data; |
181e448d | 5917 | const struct cred *old_cred; |
6c271ce2 JA |
5918 | mm_segment_t old_fs; |
5919 | DEFINE_WAIT(wait); | |
6c271ce2 | 5920 | unsigned long timeout; |
bdcd3eab | 5921 | int ret = 0; |
6c271ce2 | 5922 | |
206aefde | 5923 | complete(&ctx->completions[1]); |
a4c0b3de | 5924 | |
6c271ce2 JA |
5925 | old_fs = get_fs(); |
5926 | set_fs(USER_DS); | |
181e448d | 5927 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5928 | |
bdcd3eab | 5929 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5930 | while (!kthread_should_park()) { |
fb5ccc98 | 5931 | unsigned int to_submit; |
6c271ce2 | 5932 | |
bdcd3eab | 5933 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5934 | unsigned nr_events = 0; |
5935 | ||
bdcd3eab XW |
5936 | mutex_lock(&ctx->uring_lock); |
5937 | if (!list_empty(&ctx->poll_list)) | |
5938 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5939 | else | |
6c271ce2 | 5940 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5941 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5942 | } |
5943 | ||
fb5ccc98 | 5944 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5945 | |
5946 | /* | |
5947 | * If submit got -EBUSY, flag us as needing the application | |
5948 | * to enter the kernel to reap and flush events. | |
5949 | */ | |
5950 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5951 | /* |
5952 | * Drop cur_mm before scheduling, we can't hold it for | |
5953 | * long periods (or over schedule()). Do this before | |
5954 | * adding ourselves to the waitqueue, as the unuse/drop | |
5955 | * may sleep. | |
5956 | */ | |
bf9c2f1c | 5957 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 5958 | |
6c271ce2 JA |
5959 | /* |
5960 | * We're polling. If we're within the defined idle | |
5961 | * period, then let us spin without work before going | |
c1edbf5f JA |
5962 | * to sleep. The exception is if we got EBUSY doing |
5963 | * more IO, we should wait for the application to | |
5964 | * reap events and wake us up. | |
6c271ce2 | 5965 | */ |
bdcd3eab | 5966 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5967 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5968 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5969 | if (current->task_works) |
5970 | task_work_run(); | |
9831a90c | 5971 | cond_resched(); |
6c271ce2 JA |
5972 | continue; |
5973 | } | |
5974 | ||
6c271ce2 JA |
5975 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5976 | TASK_INTERRUPTIBLE); | |
5977 | ||
bdcd3eab XW |
5978 | /* |
5979 | * While doing polled IO, before going to sleep, we need | |
5980 | * to check if there are new reqs added to poll_list, it | |
5981 | * is because reqs may have been punted to io worker and | |
5982 | * will be added to poll_list later, hence check the | |
5983 | * poll_list again. | |
5984 | */ | |
5985 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5986 | !list_empty_careful(&ctx->poll_list)) { | |
5987 | finish_wait(&ctx->sqo_wait, &wait); | |
5988 | continue; | |
5989 | } | |
5990 | ||
6c271ce2 | 5991 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 5992 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
5993 | /* make sure to read SQ tail after writing flags */ |
5994 | smp_mb(); | |
6c271ce2 | 5995 | |
fb5ccc98 | 5996 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 5997 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 5998 | if (kthread_should_park()) { |
6c271ce2 JA |
5999 | finish_wait(&ctx->sqo_wait, &wait); |
6000 | break; | |
6001 | } | |
b41e9852 JA |
6002 | if (current->task_works) { |
6003 | task_work_run(); | |
10bea96d | 6004 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6005 | continue; |
6006 | } | |
6c271ce2 JA |
6007 | if (signal_pending(current)) |
6008 | flush_signals(current); | |
6009 | schedule(); | |
6010 | finish_wait(&ctx->sqo_wait, &wait); | |
6011 | ||
75b28aff | 6012 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6013 | continue; |
6014 | } | |
6015 | finish_wait(&ctx->sqo_wait, &wait); | |
6016 | ||
75b28aff | 6017 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6018 | } |
6019 | ||
8a4955ff | 6020 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6021 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6022 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6023 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6024 | } |
6025 | ||
b41e9852 JA |
6026 | if (current->task_works) |
6027 | task_work_run(); | |
6028 | ||
6c271ce2 | 6029 | set_fs(old_fs); |
bf9c2f1c | 6030 | io_sq_thread_drop_mm(ctx); |
181e448d | 6031 | revert_creds(old_cred); |
06058632 | 6032 | |
2bbcd6d3 | 6033 | kthread_parkme(); |
06058632 | 6034 | |
6c271ce2 JA |
6035 | return 0; |
6036 | } | |
6037 | ||
bda52162 JA |
6038 | struct io_wait_queue { |
6039 | struct wait_queue_entry wq; | |
6040 | struct io_ring_ctx *ctx; | |
6041 | unsigned to_wait; | |
6042 | unsigned nr_timeouts; | |
6043 | }; | |
6044 | ||
1d7bb1d5 | 6045 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6046 | { |
6047 | struct io_ring_ctx *ctx = iowq->ctx; | |
6048 | ||
6049 | /* | |
d195a66e | 6050 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6051 | * started waiting. For timeouts, we always want to return to userspace, |
6052 | * regardless of event count. | |
6053 | */ | |
1d7bb1d5 | 6054 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6055 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6056 | } | |
6057 | ||
6058 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6059 | int wake_flags, void *key) | |
6060 | { | |
6061 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6062 | wq); | |
6063 | ||
1d7bb1d5 JA |
6064 | /* use noflush == true, as we can't safely rely on locking context */ |
6065 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6066 | return -1; |
6067 | ||
6068 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6069 | } | |
6070 | ||
2b188cc1 JA |
6071 | /* |
6072 | * Wait until events become available, if we don't already have some. The | |
6073 | * application must reap them itself, as they reside on the shared cq ring. | |
6074 | */ | |
6075 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6076 | const sigset_t __user *sig, size_t sigsz) | |
6077 | { | |
bda52162 JA |
6078 | struct io_wait_queue iowq = { |
6079 | .wq = { | |
6080 | .private = current, | |
6081 | .func = io_wake_function, | |
6082 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6083 | }, | |
6084 | .ctx = ctx, | |
6085 | .to_wait = min_events, | |
6086 | }; | |
75b28aff | 6087 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6088 | int ret = 0; |
2b188cc1 | 6089 | |
b41e9852 JA |
6090 | do { |
6091 | if (io_cqring_events(ctx, false) >= min_events) | |
6092 | return 0; | |
6093 | if (!current->task_works) | |
6094 | break; | |
6095 | task_work_run(); | |
6096 | } while (1); | |
2b188cc1 JA |
6097 | |
6098 | if (sig) { | |
9e75ad5d AB |
6099 | #ifdef CONFIG_COMPAT |
6100 | if (in_compat_syscall()) | |
6101 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6102 | sigsz); |
9e75ad5d AB |
6103 | else |
6104 | #endif | |
b772434b | 6105 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6106 | |
2b188cc1 JA |
6107 | if (ret) |
6108 | return ret; | |
6109 | } | |
6110 | ||
bda52162 | 6111 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6112 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6113 | do { |
6114 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6115 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6116 | if (current->task_works) |
6117 | task_work_run(); | |
1d7bb1d5 | 6118 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6119 | break; |
6120 | schedule(); | |
6121 | if (signal_pending(current)) { | |
e9ffa5c2 | 6122 | ret = -EINTR; |
bda52162 JA |
6123 | break; |
6124 | } | |
6125 | } while (1); | |
6126 | finish_wait(&ctx->wait, &iowq.wq); | |
6127 | ||
e9ffa5c2 | 6128 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6129 | |
75b28aff | 6130 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6131 | } |
6132 | ||
6b06314c JA |
6133 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6134 | { | |
6135 | #if defined(CONFIG_UNIX) | |
6136 | if (ctx->ring_sock) { | |
6137 | struct sock *sock = ctx->ring_sock->sk; | |
6138 | struct sk_buff *skb; | |
6139 | ||
6140 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6141 | kfree_skb(skb); | |
6142 | } | |
6143 | #else | |
6144 | int i; | |
6145 | ||
65e19f54 JA |
6146 | for (i = 0; i < ctx->nr_user_files; i++) { |
6147 | struct file *file; | |
6148 | ||
6149 | file = io_file_from_index(ctx, i); | |
6150 | if (file) | |
6151 | fput(file); | |
6152 | } | |
6b06314c JA |
6153 | #endif |
6154 | } | |
6155 | ||
05f3fb3c JA |
6156 | static void io_file_ref_kill(struct percpu_ref *ref) |
6157 | { | |
6158 | struct fixed_file_data *data; | |
6159 | ||
6160 | data = container_of(ref, struct fixed_file_data, refs); | |
6161 | complete(&data->done); | |
6162 | } | |
6163 | ||
6b06314c JA |
6164 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6165 | { | |
05f3fb3c | 6166 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6167 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6168 | unsigned nr_tables, i; |
05589553 | 6169 | unsigned long flags; |
65e19f54 | 6170 | |
05f3fb3c | 6171 | if (!data) |
6b06314c JA |
6172 | return -ENXIO; |
6173 | ||
05589553 XW |
6174 | spin_lock_irqsave(&data->lock, flags); |
6175 | if (!list_empty(&data->ref_list)) | |
6176 | ref_node = list_first_entry(&data->ref_list, | |
6177 | struct fixed_file_ref_node, node); | |
6178 | spin_unlock_irqrestore(&data->lock, flags); | |
6179 | if (ref_node) | |
6180 | percpu_ref_kill(&ref_node->refs); | |
6181 | ||
6182 | percpu_ref_kill(&data->refs); | |
6183 | ||
6184 | /* wait for all refs nodes to complete */ | |
2faf852d | 6185 | wait_for_completion(&data->done); |
05f3fb3c | 6186 | |
6b06314c | 6187 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6188 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6189 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6190 | kfree(data->table[i].files); |
6191 | kfree(data->table); | |
05589553 XW |
6192 | percpu_ref_exit(&data->refs); |
6193 | kfree(data); | |
05f3fb3c | 6194 | ctx->file_data = NULL; |
6b06314c JA |
6195 | ctx->nr_user_files = 0; |
6196 | return 0; | |
6197 | } | |
6198 | ||
6c271ce2 JA |
6199 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6200 | { | |
6201 | if (ctx->sqo_thread) { | |
206aefde | 6202 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6203 | /* |
6204 | * The park is a bit of a work-around, without it we get | |
6205 | * warning spews on shutdown with SQPOLL set and affinity | |
6206 | * set to a single CPU. | |
6207 | */ | |
06058632 | 6208 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6209 | kthread_stop(ctx->sqo_thread); |
6210 | ctx->sqo_thread = NULL; | |
6211 | } | |
6212 | } | |
6213 | ||
6b06314c JA |
6214 | static void io_finish_async(struct io_ring_ctx *ctx) |
6215 | { | |
6c271ce2 JA |
6216 | io_sq_thread_stop(ctx); |
6217 | ||
561fb04a JA |
6218 | if (ctx->io_wq) { |
6219 | io_wq_destroy(ctx->io_wq); | |
6220 | ctx->io_wq = NULL; | |
6b06314c JA |
6221 | } |
6222 | } | |
6223 | ||
6224 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6225 | /* |
6226 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6227 | * the io_uring can be safely unregistered on process exit, even if we have | |
6228 | * loops in the file referencing. | |
6229 | */ | |
6230 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6231 | { | |
6232 | struct sock *sk = ctx->ring_sock->sk; | |
6233 | struct scm_fp_list *fpl; | |
6234 | struct sk_buff *skb; | |
08a45173 | 6235 | int i, nr_files; |
6b06314c | 6236 | |
6b06314c JA |
6237 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6238 | if (!fpl) | |
6239 | return -ENOMEM; | |
6240 | ||
6241 | skb = alloc_skb(0, GFP_KERNEL); | |
6242 | if (!skb) { | |
6243 | kfree(fpl); | |
6244 | return -ENOMEM; | |
6245 | } | |
6246 | ||
6247 | skb->sk = sk; | |
6b06314c | 6248 | |
08a45173 | 6249 | nr_files = 0; |
6b06314c JA |
6250 | fpl->user = get_uid(ctx->user); |
6251 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6252 | struct file *file = io_file_from_index(ctx, i + offset); |
6253 | ||
6254 | if (!file) | |
08a45173 | 6255 | continue; |
65e19f54 | 6256 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6257 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6258 | nr_files++; | |
6b06314c JA |
6259 | } |
6260 | ||
08a45173 JA |
6261 | if (nr_files) { |
6262 | fpl->max = SCM_MAX_FD; | |
6263 | fpl->count = nr_files; | |
6264 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6265 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6266 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6267 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6268 | |
08a45173 JA |
6269 | for (i = 0; i < nr_files; i++) |
6270 | fput(fpl->fp[i]); | |
6271 | } else { | |
6272 | kfree_skb(skb); | |
6273 | kfree(fpl); | |
6274 | } | |
6b06314c JA |
6275 | |
6276 | return 0; | |
6277 | } | |
6278 | ||
6279 | /* | |
6280 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6281 | * causes regular reference counting to break down. We rely on the UNIX | |
6282 | * garbage collection to take care of this problem for us. | |
6283 | */ | |
6284 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6285 | { | |
6286 | unsigned left, total; | |
6287 | int ret = 0; | |
6288 | ||
6289 | total = 0; | |
6290 | left = ctx->nr_user_files; | |
6291 | while (left) { | |
6292 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6293 | |
6294 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6295 | if (ret) | |
6296 | break; | |
6297 | left -= this_files; | |
6298 | total += this_files; | |
6299 | } | |
6300 | ||
6301 | if (!ret) | |
6302 | return 0; | |
6303 | ||
6304 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6305 | struct file *file = io_file_from_index(ctx, total); |
6306 | ||
6307 | if (file) | |
6308 | fput(file); | |
6b06314c JA |
6309 | total++; |
6310 | } | |
6311 | ||
6312 | return ret; | |
6313 | } | |
6314 | #else | |
6315 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6316 | { | |
6317 | return 0; | |
6318 | } | |
6319 | #endif | |
6320 | ||
65e19f54 JA |
6321 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6322 | unsigned nr_files) | |
6323 | { | |
6324 | int i; | |
6325 | ||
6326 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6327 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6328 | unsigned this_files; |
6329 | ||
6330 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6331 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6332 | GFP_KERNEL); | |
6333 | if (!table->files) | |
6334 | break; | |
6335 | nr_files -= this_files; | |
6336 | } | |
6337 | ||
6338 | if (i == nr_tables) | |
6339 | return 0; | |
6340 | ||
6341 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6342 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6343 | kfree(table->files); |
6344 | } | |
6345 | return 1; | |
6346 | } | |
6347 | ||
05f3fb3c JA |
6348 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6349 | { | |
6350 | #if defined(CONFIG_UNIX) | |
6351 | struct sock *sock = ctx->ring_sock->sk; | |
6352 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6353 | struct sk_buff *skb; | |
6354 | int i; | |
6355 | ||
6356 | __skb_queue_head_init(&list); | |
6357 | ||
6358 | /* | |
6359 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6360 | * remove this entry and rearrange the file array. | |
6361 | */ | |
6362 | skb = skb_dequeue(head); | |
6363 | while (skb) { | |
6364 | struct scm_fp_list *fp; | |
6365 | ||
6366 | fp = UNIXCB(skb).fp; | |
6367 | for (i = 0; i < fp->count; i++) { | |
6368 | int left; | |
6369 | ||
6370 | if (fp->fp[i] != file) | |
6371 | continue; | |
6372 | ||
6373 | unix_notinflight(fp->user, fp->fp[i]); | |
6374 | left = fp->count - 1 - i; | |
6375 | if (left) { | |
6376 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6377 | left * sizeof(struct file *)); | |
6378 | } | |
6379 | fp->count--; | |
6380 | if (!fp->count) { | |
6381 | kfree_skb(skb); | |
6382 | skb = NULL; | |
6383 | } else { | |
6384 | __skb_queue_tail(&list, skb); | |
6385 | } | |
6386 | fput(file); | |
6387 | file = NULL; | |
6388 | break; | |
6389 | } | |
6390 | ||
6391 | if (!file) | |
6392 | break; | |
6393 | ||
6394 | __skb_queue_tail(&list, skb); | |
6395 | ||
6396 | skb = skb_dequeue(head); | |
6397 | } | |
6398 | ||
6399 | if (skb_peek(&list)) { | |
6400 | spin_lock_irq(&head->lock); | |
6401 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6402 | __skb_queue_tail(head, skb); | |
6403 | spin_unlock_irq(&head->lock); | |
6404 | } | |
6405 | #else | |
6406 | fput(file); | |
6407 | #endif | |
6408 | } | |
6409 | ||
6410 | struct io_file_put { | |
05589553 | 6411 | struct list_head list; |
05f3fb3c | 6412 | struct file *file; |
05f3fb3c JA |
6413 | }; |
6414 | ||
05589553 | 6415 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6416 | { |
05589553 XW |
6417 | struct fixed_file_ref_node *ref_node; |
6418 | struct fixed_file_data *file_data; | |
6419 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6420 | struct io_file_put *pfile, *tmp; |
05589553 | 6421 | unsigned long flags; |
65e19f54 | 6422 | |
05589553 XW |
6423 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6424 | file_data = ref_node->file_data; | |
6425 | ctx = file_data->ctx; | |
6426 | ||
6427 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6428 | list_del_init(&pfile->list); | |
6429 | io_ring_file_put(ctx, pfile->file); | |
6430 | kfree(pfile); | |
65e19f54 | 6431 | } |
05589553 XW |
6432 | |
6433 | spin_lock_irqsave(&file_data->lock, flags); | |
6434 | list_del_init(&ref_node->node); | |
6435 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6436 | ||
6437 | percpu_ref_exit(&ref_node->refs); | |
6438 | kfree(ref_node); | |
6439 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6440 | } |
65e19f54 | 6441 | |
05589553 | 6442 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6443 | { |
05589553 | 6444 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6445 | |
05589553 XW |
6446 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6447 | ||
6448 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6449 | } |
65e19f54 | 6450 | |
05589553 XW |
6451 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6452 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6453 | { |
05589553 | 6454 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6455 | |
05589553 XW |
6456 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6457 | if (!ref_node) | |
6458 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6459 | |
05589553 XW |
6460 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6461 | 0, GFP_KERNEL)) { | |
6462 | kfree(ref_node); | |
6463 | return ERR_PTR(-ENOMEM); | |
6464 | } | |
6465 | INIT_LIST_HEAD(&ref_node->node); | |
6466 | INIT_LIST_HEAD(&ref_node->file_list); | |
6467 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6468 | ref_node->file_data = ctx->file_data; | |
6469 | return ref_node; | |
6470 | ||
6471 | } | |
6472 | ||
6473 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6474 | { | |
6475 | percpu_ref_exit(&ref_node->refs); | |
6476 | kfree(ref_node); | |
65e19f54 JA |
6477 | } |
6478 | ||
6b06314c JA |
6479 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6480 | unsigned nr_args) | |
6481 | { | |
6482 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6483 | unsigned nr_tables; |
05f3fb3c | 6484 | struct file *file; |
6b06314c JA |
6485 | int fd, ret = 0; |
6486 | unsigned i; | |
05589553 XW |
6487 | struct fixed_file_ref_node *ref_node; |
6488 | unsigned long flags; | |
6b06314c | 6489 | |
05f3fb3c | 6490 | if (ctx->file_data) |
6b06314c JA |
6491 | return -EBUSY; |
6492 | if (!nr_args) | |
6493 | return -EINVAL; | |
6494 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6495 | return -EMFILE; | |
6496 | ||
05f3fb3c JA |
6497 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6498 | if (!ctx->file_data) | |
6499 | return -ENOMEM; | |
6500 | ctx->file_data->ctx = ctx; | |
6501 | init_completion(&ctx->file_data->done); | |
05589553 | 6502 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6503 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6504 | |
65e19f54 | 6505 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6506 | ctx->file_data->table = kcalloc(nr_tables, |
6507 | sizeof(struct fixed_file_table), | |
65e19f54 | 6508 | GFP_KERNEL); |
05f3fb3c JA |
6509 | if (!ctx->file_data->table) { |
6510 | kfree(ctx->file_data); | |
6511 | ctx->file_data = NULL; | |
6b06314c | 6512 | return -ENOMEM; |
05f3fb3c JA |
6513 | } |
6514 | ||
05589553 | 6515 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6516 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6517 | kfree(ctx->file_data->table); | |
6518 | kfree(ctx->file_data); | |
6519 | ctx->file_data = NULL; | |
6b06314c | 6520 | return -ENOMEM; |
05f3fb3c | 6521 | } |
6b06314c | 6522 | |
65e19f54 | 6523 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6524 | percpu_ref_exit(&ctx->file_data->refs); |
6525 | kfree(ctx->file_data->table); | |
6526 | kfree(ctx->file_data); | |
6527 | ctx->file_data = NULL; | |
65e19f54 JA |
6528 | return -ENOMEM; |
6529 | } | |
6530 | ||
08a45173 | 6531 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6532 | struct fixed_file_table *table; |
6533 | unsigned index; | |
6534 | ||
6b06314c JA |
6535 | ret = -EFAULT; |
6536 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6537 | break; | |
08a45173 JA |
6538 | /* allow sparse sets */ |
6539 | if (fd == -1) { | |
6540 | ret = 0; | |
6541 | continue; | |
6542 | } | |
6b06314c | 6543 | |
05f3fb3c | 6544 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6545 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6546 | file = fget(fd); |
6b06314c JA |
6547 | |
6548 | ret = -EBADF; | |
05f3fb3c | 6549 | if (!file) |
6b06314c | 6550 | break; |
05f3fb3c | 6551 | |
6b06314c JA |
6552 | /* |
6553 | * Don't allow io_uring instances to be registered. If UNIX | |
6554 | * isn't enabled, then this causes a reference cycle and this | |
6555 | * instance can never get freed. If UNIX is enabled we'll | |
6556 | * handle it just fine, but there's still no point in allowing | |
6557 | * a ring fd as it doesn't support regular read/write anyway. | |
6558 | */ | |
05f3fb3c JA |
6559 | if (file->f_op == &io_uring_fops) { |
6560 | fput(file); | |
6b06314c JA |
6561 | break; |
6562 | } | |
6b06314c | 6563 | ret = 0; |
05f3fb3c | 6564 | table->files[index] = file; |
6b06314c JA |
6565 | } |
6566 | ||
6567 | if (ret) { | |
65e19f54 | 6568 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6569 | file = io_file_from_index(ctx, i); |
6570 | if (file) | |
6571 | fput(file); | |
6572 | } | |
6573 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6574 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6575 | |
05f3fb3c JA |
6576 | kfree(ctx->file_data->table); |
6577 | kfree(ctx->file_data); | |
6578 | ctx->file_data = NULL; | |
6b06314c JA |
6579 | ctx->nr_user_files = 0; |
6580 | return ret; | |
6581 | } | |
6582 | ||
6583 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6584 | if (ret) { |
6b06314c | 6585 | io_sqe_files_unregister(ctx); |
05589553 XW |
6586 | return ret; |
6587 | } | |
6b06314c | 6588 | |
05589553 XW |
6589 | ref_node = alloc_fixed_file_ref_node(ctx); |
6590 | if (IS_ERR(ref_node)) { | |
6591 | io_sqe_files_unregister(ctx); | |
6592 | return PTR_ERR(ref_node); | |
6593 | } | |
6594 | ||
6595 | ctx->file_data->cur_refs = &ref_node->refs; | |
6596 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6597 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6598 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6599 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6600 | return ret; |
6601 | } | |
6602 | ||
c3a31e60 JA |
6603 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6604 | int index) | |
6605 | { | |
6606 | #if defined(CONFIG_UNIX) | |
6607 | struct sock *sock = ctx->ring_sock->sk; | |
6608 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6609 | struct sk_buff *skb; | |
6610 | ||
6611 | /* | |
6612 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6613 | * file set. If there's no room, fall back to allocating a new skb | |
6614 | * and filling it in. | |
6615 | */ | |
6616 | spin_lock_irq(&head->lock); | |
6617 | skb = skb_peek(head); | |
6618 | if (skb) { | |
6619 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6620 | ||
6621 | if (fpl->count < SCM_MAX_FD) { | |
6622 | __skb_unlink(skb, head); | |
6623 | spin_unlock_irq(&head->lock); | |
6624 | fpl->fp[fpl->count] = get_file(file); | |
6625 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6626 | fpl->count++; | |
6627 | spin_lock_irq(&head->lock); | |
6628 | __skb_queue_head(head, skb); | |
6629 | } else { | |
6630 | skb = NULL; | |
6631 | } | |
6632 | } | |
6633 | spin_unlock_irq(&head->lock); | |
6634 | ||
6635 | if (skb) { | |
6636 | fput(file); | |
6637 | return 0; | |
6638 | } | |
6639 | ||
6640 | return __io_sqe_files_scm(ctx, 1, index); | |
6641 | #else | |
6642 | return 0; | |
6643 | #endif | |
6644 | } | |
6645 | ||
a5318d3c | 6646 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6647 | struct file *file) |
05f3fb3c | 6648 | { |
a5318d3c | 6649 | struct io_file_put *pfile; |
05589553 XW |
6650 | struct percpu_ref *refs = data->cur_refs; |
6651 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6652 | |
05f3fb3c | 6653 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6654 | if (!pfile) |
6655 | return -ENOMEM; | |
05f3fb3c | 6656 | |
05589553 | 6657 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6658 | pfile->file = file; |
05589553 XW |
6659 | list_add(&pfile->list, &ref_node->file_list); |
6660 | ||
a5318d3c | 6661 | return 0; |
05f3fb3c JA |
6662 | } |
6663 | ||
6664 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6665 | struct io_uring_files_update *up, | |
6666 | unsigned nr_args) | |
6667 | { | |
6668 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6669 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6670 | struct file *file; |
c3a31e60 JA |
6671 | __s32 __user *fds; |
6672 | int fd, i, err; | |
6673 | __u32 done; | |
05589553 XW |
6674 | unsigned long flags; |
6675 | bool needs_switch = false; | |
c3a31e60 | 6676 | |
05f3fb3c | 6677 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6678 | return -EOVERFLOW; |
6679 | if (done > ctx->nr_user_files) | |
6680 | return -EINVAL; | |
6681 | ||
05589553 XW |
6682 | ref_node = alloc_fixed_file_ref_node(ctx); |
6683 | if (IS_ERR(ref_node)) | |
6684 | return PTR_ERR(ref_node); | |
6685 | ||
c3a31e60 | 6686 | done = 0; |
05f3fb3c | 6687 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6688 | while (nr_args) { |
65e19f54 JA |
6689 | struct fixed_file_table *table; |
6690 | unsigned index; | |
6691 | ||
c3a31e60 JA |
6692 | err = 0; |
6693 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6694 | err = -EFAULT; | |
6695 | break; | |
6696 | } | |
05f3fb3c JA |
6697 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6698 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6699 | index = i & IORING_FILE_TABLE_MASK; |
6700 | if (table->files[index]) { | |
05f3fb3c | 6701 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6702 | err = io_queue_file_removal(data, file); |
6703 | if (err) | |
6704 | break; | |
65e19f54 | 6705 | table->files[index] = NULL; |
05589553 | 6706 | needs_switch = true; |
c3a31e60 JA |
6707 | } |
6708 | if (fd != -1) { | |
c3a31e60 JA |
6709 | file = fget(fd); |
6710 | if (!file) { | |
6711 | err = -EBADF; | |
6712 | break; | |
6713 | } | |
6714 | /* | |
6715 | * Don't allow io_uring instances to be registered. If | |
6716 | * UNIX isn't enabled, then this causes a reference | |
6717 | * cycle and this instance can never get freed. If UNIX | |
6718 | * is enabled we'll handle it just fine, but there's | |
6719 | * still no point in allowing a ring fd as it doesn't | |
6720 | * support regular read/write anyway. | |
6721 | */ | |
6722 | if (file->f_op == &io_uring_fops) { | |
6723 | fput(file); | |
6724 | err = -EBADF; | |
6725 | break; | |
6726 | } | |
65e19f54 | 6727 | table->files[index] = file; |
c3a31e60 JA |
6728 | err = io_sqe_file_register(ctx, file, i); |
6729 | if (err) | |
6730 | break; | |
6731 | } | |
6732 | nr_args--; | |
6733 | done++; | |
05f3fb3c JA |
6734 | up->offset++; |
6735 | } | |
6736 | ||
05589553 XW |
6737 | if (needs_switch) { |
6738 | percpu_ref_kill(data->cur_refs); | |
6739 | spin_lock_irqsave(&data->lock, flags); | |
6740 | list_add(&ref_node->node, &data->ref_list); | |
6741 | data->cur_refs = &ref_node->refs; | |
6742 | spin_unlock_irqrestore(&data->lock, flags); | |
6743 | percpu_ref_get(&ctx->file_data->refs); | |
6744 | } else | |
6745 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6746 | |
6747 | return done ? done : err; | |
6748 | } | |
05589553 | 6749 | |
05f3fb3c JA |
6750 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6751 | unsigned nr_args) | |
6752 | { | |
6753 | struct io_uring_files_update up; | |
6754 | ||
6755 | if (!ctx->file_data) | |
6756 | return -ENXIO; | |
6757 | if (!nr_args) | |
6758 | return -EINVAL; | |
6759 | if (copy_from_user(&up, arg, sizeof(up))) | |
6760 | return -EFAULT; | |
6761 | if (up.resv) | |
6762 | return -EINVAL; | |
6763 | ||
6764 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6765 | } | |
c3a31e60 | 6766 | |
e9fd9396 | 6767 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6768 | { |
6769 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6770 | ||
e9fd9396 | 6771 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6772 | io_put_req(req); |
6773 | } | |
6774 | ||
24369c2e PB |
6775 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6776 | struct io_uring_params *p) | |
6777 | { | |
6778 | struct io_wq_data data; | |
6779 | struct fd f; | |
6780 | struct io_ring_ctx *ctx_attach; | |
6781 | unsigned int concurrency; | |
6782 | int ret = 0; | |
6783 | ||
6784 | data.user = ctx->user; | |
e9fd9396 | 6785 | data.free_work = io_free_work; |
24369c2e PB |
6786 | |
6787 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6788 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6789 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6790 | ||
6791 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6792 | if (IS_ERR(ctx->io_wq)) { | |
6793 | ret = PTR_ERR(ctx->io_wq); | |
6794 | ctx->io_wq = NULL; | |
6795 | } | |
6796 | return ret; | |
6797 | } | |
6798 | ||
6799 | f = fdget(p->wq_fd); | |
6800 | if (!f.file) | |
6801 | return -EBADF; | |
6802 | ||
6803 | if (f.file->f_op != &io_uring_fops) { | |
6804 | ret = -EINVAL; | |
6805 | goto out_fput; | |
6806 | } | |
6807 | ||
6808 | ctx_attach = f.file->private_data; | |
6809 | /* @io_wq is protected by holding the fd */ | |
6810 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6811 | ret = -EINVAL; | |
6812 | goto out_fput; | |
6813 | } | |
6814 | ||
6815 | ctx->io_wq = ctx_attach->io_wq; | |
6816 | out_fput: | |
6817 | fdput(f); | |
6818 | return ret; | |
6819 | } | |
6820 | ||
6c271ce2 JA |
6821 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6822 | struct io_uring_params *p) | |
2b188cc1 JA |
6823 | { |
6824 | int ret; | |
6825 | ||
6c271ce2 | 6826 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6827 | mmgrab(current->mm); |
6828 | ctx->sqo_mm = current->mm; | |
6829 | ||
6c271ce2 | 6830 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6831 | ret = -EPERM; |
6832 | if (!capable(CAP_SYS_ADMIN)) | |
6833 | goto err; | |
6834 | ||
917257da JA |
6835 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6836 | if (!ctx->sq_thread_idle) | |
6837 | ctx->sq_thread_idle = HZ; | |
6838 | ||
6c271ce2 | 6839 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6840 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6841 | |
917257da | 6842 | ret = -EINVAL; |
44a9bd18 JA |
6843 | if (cpu >= nr_cpu_ids) |
6844 | goto err; | |
7889f44d | 6845 | if (!cpu_online(cpu)) |
917257da JA |
6846 | goto err; |
6847 | ||
6c271ce2 JA |
6848 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6849 | ctx, cpu, | |
6850 | "io_uring-sq"); | |
6851 | } else { | |
6852 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6853 | "io_uring-sq"); | |
6854 | } | |
6855 | if (IS_ERR(ctx->sqo_thread)) { | |
6856 | ret = PTR_ERR(ctx->sqo_thread); | |
6857 | ctx->sqo_thread = NULL; | |
6858 | goto err; | |
6859 | } | |
6860 | wake_up_process(ctx->sqo_thread); | |
6861 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6862 | /* Can't have SQ_AFF without SQPOLL */ | |
6863 | ret = -EINVAL; | |
6864 | goto err; | |
6865 | } | |
6866 | ||
24369c2e PB |
6867 | ret = io_init_wq_offload(ctx, p); |
6868 | if (ret) | |
2b188cc1 | 6869 | goto err; |
2b188cc1 JA |
6870 | |
6871 | return 0; | |
6872 | err: | |
54a91f3b | 6873 | io_finish_async(ctx); |
2b188cc1 JA |
6874 | mmdrop(ctx->sqo_mm); |
6875 | ctx->sqo_mm = NULL; | |
6876 | return ret; | |
6877 | } | |
6878 | ||
6879 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6880 | { | |
6881 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6882 | } | |
6883 | ||
6884 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6885 | { | |
6886 | unsigned long page_limit, cur_pages, new_pages; | |
6887 | ||
6888 | /* Don't allow more pages than we can safely lock */ | |
6889 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6890 | ||
6891 | do { | |
6892 | cur_pages = atomic_long_read(&user->locked_vm); | |
6893 | new_pages = cur_pages + nr_pages; | |
6894 | if (new_pages > page_limit) | |
6895 | return -ENOMEM; | |
6896 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6897 | new_pages) != cur_pages); | |
6898 | ||
6899 | return 0; | |
6900 | } | |
6901 | ||
6902 | static void io_mem_free(void *ptr) | |
6903 | { | |
52e04ef4 MR |
6904 | struct page *page; |
6905 | ||
6906 | if (!ptr) | |
6907 | return; | |
2b188cc1 | 6908 | |
52e04ef4 | 6909 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6910 | if (put_page_testzero(page)) |
6911 | free_compound_page(page); | |
6912 | } | |
6913 | ||
6914 | static void *io_mem_alloc(size_t size) | |
6915 | { | |
6916 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6917 | __GFP_NORETRY; | |
6918 | ||
6919 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6920 | } | |
6921 | ||
75b28aff HV |
6922 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6923 | size_t *sq_offset) | |
6924 | { | |
6925 | struct io_rings *rings; | |
6926 | size_t off, sq_array_size; | |
6927 | ||
6928 | off = struct_size(rings, cqes, cq_entries); | |
6929 | if (off == SIZE_MAX) | |
6930 | return SIZE_MAX; | |
6931 | ||
6932 | #ifdef CONFIG_SMP | |
6933 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6934 | if (off == 0) | |
6935 | return SIZE_MAX; | |
6936 | #endif | |
6937 | ||
6938 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6939 | if (sq_array_size == SIZE_MAX) | |
6940 | return SIZE_MAX; | |
6941 | ||
6942 | if (check_add_overflow(off, sq_array_size, &off)) | |
6943 | return SIZE_MAX; | |
6944 | ||
6945 | if (sq_offset) | |
6946 | *sq_offset = off; | |
6947 | ||
6948 | return off; | |
6949 | } | |
6950 | ||
2b188cc1 JA |
6951 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6952 | { | |
75b28aff | 6953 | size_t pages; |
2b188cc1 | 6954 | |
75b28aff HV |
6955 | pages = (size_t)1 << get_order( |
6956 | rings_size(sq_entries, cq_entries, NULL)); | |
6957 | pages += (size_t)1 << get_order( | |
6958 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6959 | |
75b28aff | 6960 | return pages; |
2b188cc1 JA |
6961 | } |
6962 | ||
edafccee JA |
6963 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6964 | { | |
6965 | int i, j; | |
6966 | ||
6967 | if (!ctx->user_bufs) | |
6968 | return -ENXIO; | |
6969 | ||
6970 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6971 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6972 | ||
6973 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6974 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6975 | |
6976 | if (ctx->account_mem) | |
6977 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6978 | kvfree(imu->bvec); |
edafccee JA |
6979 | imu->nr_bvecs = 0; |
6980 | } | |
6981 | ||
6982 | kfree(ctx->user_bufs); | |
6983 | ctx->user_bufs = NULL; | |
6984 | ctx->nr_user_bufs = 0; | |
6985 | return 0; | |
6986 | } | |
6987 | ||
6988 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6989 | void __user *arg, unsigned index) | |
6990 | { | |
6991 | struct iovec __user *src; | |
6992 | ||
6993 | #ifdef CONFIG_COMPAT | |
6994 | if (ctx->compat) { | |
6995 | struct compat_iovec __user *ciovs; | |
6996 | struct compat_iovec ciov; | |
6997 | ||
6998 | ciovs = (struct compat_iovec __user *) arg; | |
6999 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7000 | return -EFAULT; | |
7001 | ||
d55e5f5b | 7002 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7003 | dst->iov_len = ciov.iov_len; |
7004 | return 0; | |
7005 | } | |
7006 | #endif | |
7007 | src = (struct iovec __user *) arg; | |
7008 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7009 | return -EFAULT; | |
7010 | return 0; | |
7011 | } | |
7012 | ||
7013 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7014 | unsigned nr_args) | |
7015 | { | |
7016 | struct vm_area_struct **vmas = NULL; | |
7017 | struct page **pages = NULL; | |
7018 | int i, j, got_pages = 0; | |
7019 | int ret = -EINVAL; | |
7020 | ||
7021 | if (ctx->user_bufs) | |
7022 | return -EBUSY; | |
7023 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7024 | return -EINVAL; | |
7025 | ||
7026 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7027 | GFP_KERNEL); | |
7028 | if (!ctx->user_bufs) | |
7029 | return -ENOMEM; | |
7030 | ||
7031 | for (i = 0; i < nr_args; i++) { | |
7032 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7033 | unsigned long off, start, end, ubuf; | |
7034 | int pret, nr_pages; | |
7035 | struct iovec iov; | |
7036 | size_t size; | |
7037 | ||
7038 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7039 | if (ret) | |
a278682d | 7040 | goto err; |
edafccee JA |
7041 | |
7042 | /* | |
7043 | * Don't impose further limits on the size and buffer | |
7044 | * constraints here, we'll -EINVAL later when IO is | |
7045 | * submitted if they are wrong. | |
7046 | */ | |
7047 | ret = -EFAULT; | |
7048 | if (!iov.iov_base || !iov.iov_len) | |
7049 | goto err; | |
7050 | ||
7051 | /* arbitrary limit, but we need something */ | |
7052 | if (iov.iov_len > SZ_1G) | |
7053 | goto err; | |
7054 | ||
7055 | ubuf = (unsigned long) iov.iov_base; | |
7056 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7057 | start = ubuf >> PAGE_SHIFT; | |
7058 | nr_pages = end - start; | |
7059 | ||
7060 | if (ctx->account_mem) { | |
7061 | ret = io_account_mem(ctx->user, nr_pages); | |
7062 | if (ret) | |
7063 | goto err; | |
7064 | } | |
7065 | ||
7066 | ret = 0; | |
7067 | if (!pages || nr_pages > got_pages) { | |
7068 | kfree(vmas); | |
7069 | kfree(pages); | |
d4ef6475 | 7070 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7071 | GFP_KERNEL); |
d4ef6475 | 7072 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7073 | sizeof(struct vm_area_struct *), |
7074 | GFP_KERNEL); | |
7075 | if (!pages || !vmas) { | |
7076 | ret = -ENOMEM; | |
7077 | if (ctx->account_mem) | |
7078 | io_unaccount_mem(ctx->user, nr_pages); | |
7079 | goto err; | |
7080 | } | |
7081 | got_pages = nr_pages; | |
7082 | } | |
7083 | ||
d4ef6475 | 7084 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7085 | GFP_KERNEL); |
7086 | ret = -ENOMEM; | |
7087 | if (!imu->bvec) { | |
7088 | if (ctx->account_mem) | |
7089 | io_unaccount_mem(ctx->user, nr_pages); | |
7090 | goto err; | |
7091 | } | |
7092 | ||
7093 | ret = 0; | |
7094 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7095 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7096 | FOLL_WRITE | FOLL_LONGTERM, |
7097 | pages, vmas); | |
edafccee JA |
7098 | if (pret == nr_pages) { |
7099 | /* don't support file backed memory */ | |
7100 | for (j = 0; j < nr_pages; j++) { | |
7101 | struct vm_area_struct *vma = vmas[j]; | |
7102 | ||
7103 | if (vma->vm_file && | |
7104 | !is_file_hugepages(vma->vm_file)) { | |
7105 | ret = -EOPNOTSUPP; | |
7106 | break; | |
7107 | } | |
7108 | } | |
7109 | } else { | |
7110 | ret = pret < 0 ? pret : -EFAULT; | |
7111 | } | |
7112 | up_read(¤t->mm->mmap_sem); | |
7113 | if (ret) { | |
7114 | /* | |
7115 | * if we did partial map, or found file backed vmas, | |
7116 | * release any pages we did get | |
7117 | */ | |
27c4d3a3 | 7118 | if (pret > 0) |
f1f6a7dd | 7119 | unpin_user_pages(pages, pret); |
edafccee JA |
7120 | if (ctx->account_mem) |
7121 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7122 | kvfree(imu->bvec); |
edafccee JA |
7123 | goto err; |
7124 | } | |
7125 | ||
7126 | off = ubuf & ~PAGE_MASK; | |
7127 | size = iov.iov_len; | |
7128 | for (j = 0; j < nr_pages; j++) { | |
7129 | size_t vec_len; | |
7130 | ||
7131 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7132 | imu->bvec[j].bv_page = pages[j]; | |
7133 | imu->bvec[j].bv_len = vec_len; | |
7134 | imu->bvec[j].bv_offset = off; | |
7135 | off = 0; | |
7136 | size -= vec_len; | |
7137 | } | |
7138 | /* store original address for later verification */ | |
7139 | imu->ubuf = ubuf; | |
7140 | imu->len = iov.iov_len; | |
7141 | imu->nr_bvecs = nr_pages; | |
7142 | ||
7143 | ctx->nr_user_bufs++; | |
7144 | } | |
d4ef6475 MR |
7145 | kvfree(pages); |
7146 | kvfree(vmas); | |
edafccee JA |
7147 | return 0; |
7148 | err: | |
d4ef6475 MR |
7149 | kvfree(pages); |
7150 | kvfree(vmas); | |
edafccee JA |
7151 | io_sqe_buffer_unregister(ctx); |
7152 | return ret; | |
7153 | } | |
7154 | ||
9b402849 JA |
7155 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7156 | { | |
7157 | __s32 __user *fds = arg; | |
7158 | int fd; | |
7159 | ||
7160 | if (ctx->cq_ev_fd) | |
7161 | return -EBUSY; | |
7162 | ||
7163 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7164 | return -EFAULT; | |
7165 | ||
7166 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7167 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7168 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7169 | ctx->cq_ev_fd = NULL; | |
7170 | return ret; | |
7171 | } | |
7172 | ||
7173 | return 0; | |
7174 | } | |
7175 | ||
7176 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7177 | { | |
7178 | if (ctx->cq_ev_fd) { | |
7179 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7180 | ctx->cq_ev_fd = NULL; | |
7181 | return 0; | |
7182 | } | |
7183 | ||
7184 | return -ENXIO; | |
7185 | } | |
7186 | ||
5a2e745d JA |
7187 | static int __io_destroy_buffers(int id, void *p, void *data) |
7188 | { | |
7189 | struct io_ring_ctx *ctx = data; | |
7190 | struct io_buffer *buf = p; | |
7191 | ||
067524e9 | 7192 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7193 | return 0; |
7194 | } | |
7195 | ||
7196 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7197 | { | |
7198 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7199 | idr_destroy(&ctx->io_buffer_idr); | |
7200 | } | |
7201 | ||
2b188cc1 JA |
7202 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7203 | { | |
6b06314c | 7204 | io_finish_async(ctx); |
2b188cc1 JA |
7205 | if (ctx->sqo_mm) |
7206 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7207 | |
7208 | io_iopoll_reap_events(ctx); | |
edafccee | 7209 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7210 | io_sqe_files_unregister(ctx); |
9b402849 | 7211 | io_eventfd_unregister(ctx); |
5a2e745d | 7212 | io_destroy_buffers(ctx); |
41726c9a | 7213 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7214 | |
2b188cc1 | 7215 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7216 | if (ctx->ring_sock) { |
7217 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7218 | sock_release(ctx->ring_sock); |
355e8d26 | 7219 | } |
2b188cc1 JA |
7220 | #endif |
7221 | ||
75b28aff | 7222 | io_mem_free(ctx->rings); |
2b188cc1 | 7223 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7224 | |
7225 | percpu_ref_exit(&ctx->refs); | |
7226 | if (ctx->account_mem) | |
7227 | io_unaccount_mem(ctx->user, | |
7228 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7229 | free_uid(ctx->user); | |
181e448d | 7230 | put_cred(ctx->creds); |
206aefde | 7231 | kfree(ctx->completions); |
78076bb6 | 7232 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7233 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7234 | kfree(ctx); |
7235 | } | |
7236 | ||
7237 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7238 | { | |
7239 | struct io_ring_ctx *ctx = file->private_data; | |
7240 | __poll_t mask = 0; | |
7241 | ||
7242 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7243 | /* |
7244 | * synchronizes with barrier from wq_has_sleeper call in | |
7245 | * io_commit_cqring | |
7246 | */ | |
2b188cc1 | 7247 | smp_rmb(); |
75b28aff HV |
7248 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7249 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7250 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7251 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7252 | mask |= EPOLLIN | EPOLLRDNORM; |
7253 | ||
7254 | return mask; | |
7255 | } | |
7256 | ||
7257 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7258 | { | |
7259 | struct io_ring_ctx *ctx = file->private_data; | |
7260 | ||
7261 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7262 | } | |
7263 | ||
071698e1 JA |
7264 | static int io_remove_personalities(int id, void *p, void *data) |
7265 | { | |
7266 | struct io_ring_ctx *ctx = data; | |
7267 | const struct cred *cred; | |
7268 | ||
7269 | cred = idr_remove(&ctx->personality_idr, id); | |
7270 | if (cred) | |
7271 | put_cred(cred); | |
7272 | return 0; | |
7273 | } | |
7274 | ||
85faa7b8 JA |
7275 | static void io_ring_exit_work(struct work_struct *work) |
7276 | { | |
7277 | struct io_ring_ctx *ctx; | |
7278 | ||
7279 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7280 | if (ctx->rings) | |
7281 | io_cqring_overflow_flush(ctx, true); | |
7282 | ||
7283 | wait_for_completion(&ctx->completions[0]); | |
7284 | io_ring_ctx_free(ctx); | |
7285 | } | |
7286 | ||
2b188cc1 JA |
7287 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7288 | { | |
7289 | mutex_lock(&ctx->uring_lock); | |
7290 | percpu_ref_kill(&ctx->refs); | |
7291 | mutex_unlock(&ctx->uring_lock); | |
7292 | ||
df069d80 JA |
7293 | /* |
7294 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7295 | * work after we've killed the ctx ref above. This is important to do | |
7296 | * before we cancel existing commands, as the thread could otherwise | |
7297 | * be queueing new work post that. If that's work we need to cancel, | |
7298 | * it could cause shutdown to hang. | |
7299 | */ | |
7300 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
7301 | cpu_relax(); | |
7302 | ||
5262f567 | 7303 | io_kill_timeouts(ctx); |
221c5eb2 | 7304 | io_poll_remove_all(ctx); |
561fb04a JA |
7305 | |
7306 | if (ctx->io_wq) | |
7307 | io_wq_cancel_all(ctx->io_wq); | |
7308 | ||
def596e9 | 7309 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7310 | /* if we failed setting up the ctx, we might not have any rings */ |
7311 | if (ctx->rings) | |
7312 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7313 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7314 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7315 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7316 | } |
7317 | ||
7318 | static int io_uring_release(struct inode *inode, struct file *file) | |
7319 | { | |
7320 | struct io_ring_ctx *ctx = file->private_data; | |
7321 | ||
7322 | file->private_data = NULL; | |
7323 | io_ring_ctx_wait_and_kill(ctx); | |
7324 | return 0; | |
7325 | } | |
7326 | ||
fcb323cc JA |
7327 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7328 | struct files_struct *files) | |
7329 | { | |
7330 | struct io_kiocb *req; | |
7331 | DEFINE_WAIT(wait); | |
7332 | ||
7333 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 7334 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
7335 | |
7336 | spin_lock_irq(&ctx->inflight_lock); | |
7337 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7338 | if (req->work.files != files) |
7339 | continue; | |
7340 | /* req is being completed, ignore */ | |
7341 | if (!refcount_inc_not_zero(&req->refs)) | |
7342 | continue; | |
7343 | cancel_req = req; | |
7344 | break; | |
fcb323cc | 7345 | } |
768134d4 | 7346 | if (cancel_req) |
fcb323cc | 7347 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7348 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7349 | spin_unlock_irq(&ctx->inflight_lock); |
7350 | ||
768134d4 JA |
7351 | /* We need to keep going until we don't find a matching req */ |
7352 | if (!cancel_req) | |
fcb323cc | 7353 | break; |
2f6d9b9d | 7354 | |
2ca10259 JA |
7355 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7356 | spin_lock_irq(&ctx->completion_lock); | |
7357 | list_del(&cancel_req->list); | |
7358 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7359 | if (list_empty(&ctx->cq_overflow_list)) { | |
7360 | clear_bit(0, &ctx->sq_check_overflow); | |
7361 | clear_bit(0, &ctx->cq_check_overflow); | |
7362 | } | |
7363 | spin_unlock_irq(&ctx->completion_lock); | |
7364 | ||
7365 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7366 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7367 | ||
7368 | /* | |
7369 | * Put inflight ref and overflow ref. If that's | |
7370 | * all we had, then we're done with this request. | |
7371 | */ | |
7372 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7373 | io_put_req(cancel_req); | |
7374 | continue; | |
7375 | } | |
7376 | } | |
7377 | ||
2f6d9b9d BL |
7378 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7379 | io_put_req(cancel_req); | |
fcb323cc JA |
7380 | schedule(); |
7381 | } | |
768134d4 | 7382 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7383 | } |
7384 | ||
7385 | static int io_uring_flush(struct file *file, void *data) | |
7386 | { | |
7387 | struct io_ring_ctx *ctx = file->private_data; | |
7388 | ||
7389 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7390 | |
7391 | /* | |
7392 | * If the task is going away, cancel work it may have pending | |
7393 | */ | |
7394 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7395 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7396 | ||
fcb323cc JA |
7397 | return 0; |
7398 | } | |
7399 | ||
6c5c240e RP |
7400 | static void *io_uring_validate_mmap_request(struct file *file, |
7401 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7402 | { |
2b188cc1 | 7403 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7404 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7405 | struct page *page; |
7406 | void *ptr; | |
7407 | ||
7408 | switch (offset) { | |
7409 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7410 | case IORING_OFF_CQ_RING: |
7411 | ptr = ctx->rings; | |
2b188cc1 JA |
7412 | break; |
7413 | case IORING_OFF_SQES: | |
7414 | ptr = ctx->sq_sqes; | |
7415 | break; | |
2b188cc1 | 7416 | default: |
6c5c240e | 7417 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7418 | } |
7419 | ||
7420 | page = virt_to_head_page(ptr); | |
a50b854e | 7421 | if (sz > page_size(page)) |
6c5c240e RP |
7422 | return ERR_PTR(-EINVAL); |
7423 | ||
7424 | return ptr; | |
7425 | } | |
7426 | ||
7427 | #ifdef CONFIG_MMU | |
7428 | ||
7429 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7430 | { | |
7431 | size_t sz = vma->vm_end - vma->vm_start; | |
7432 | unsigned long pfn; | |
7433 | void *ptr; | |
7434 | ||
7435 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7436 | if (IS_ERR(ptr)) | |
7437 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7438 | |
7439 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7440 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7441 | } | |
7442 | ||
6c5c240e RP |
7443 | #else /* !CONFIG_MMU */ |
7444 | ||
7445 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7446 | { | |
7447 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7448 | } | |
7449 | ||
7450 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7451 | { | |
7452 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7453 | } | |
7454 | ||
7455 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7456 | unsigned long addr, unsigned long len, | |
7457 | unsigned long pgoff, unsigned long flags) | |
7458 | { | |
7459 | void *ptr; | |
7460 | ||
7461 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7462 | if (IS_ERR(ptr)) | |
7463 | return PTR_ERR(ptr); | |
7464 | ||
7465 | return (unsigned long) ptr; | |
7466 | } | |
7467 | ||
7468 | #endif /* !CONFIG_MMU */ | |
7469 | ||
2b188cc1 JA |
7470 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7471 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7472 | size_t, sigsz) | |
7473 | { | |
7474 | struct io_ring_ctx *ctx; | |
7475 | long ret = -EBADF; | |
7476 | int submitted = 0; | |
7477 | struct fd f; | |
7478 | ||
b41e9852 JA |
7479 | if (current->task_works) |
7480 | task_work_run(); | |
7481 | ||
6c271ce2 | 7482 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7483 | return -EINVAL; |
7484 | ||
7485 | f = fdget(fd); | |
7486 | if (!f.file) | |
7487 | return -EBADF; | |
7488 | ||
7489 | ret = -EOPNOTSUPP; | |
7490 | if (f.file->f_op != &io_uring_fops) | |
7491 | goto out_fput; | |
7492 | ||
7493 | ret = -ENXIO; | |
7494 | ctx = f.file->private_data; | |
7495 | if (!percpu_ref_tryget(&ctx->refs)) | |
7496 | goto out_fput; | |
7497 | ||
6c271ce2 JA |
7498 | /* |
7499 | * For SQ polling, the thread will do all submissions and completions. | |
7500 | * Just return the requested submit count, and wake the thread if | |
7501 | * we were asked to. | |
7502 | */ | |
b2a9eada | 7503 | ret = 0; |
6c271ce2 | 7504 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7505 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7506 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7507 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7508 | wake_up(&ctx->sqo_wait); | |
7509 | submitted = to_submit; | |
b2a9eada | 7510 | } else if (to_submit) { |
2b188cc1 | 7511 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7512 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7513 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7514 | |
7515 | if (submitted != to_submit) | |
7516 | goto out; | |
2b188cc1 JA |
7517 | } |
7518 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7519 | unsigned nr_events = 0; |
7520 | ||
2b188cc1 JA |
7521 | min_complete = min(min_complete, ctx->cq_entries); |
7522 | ||
32b2244a XW |
7523 | /* |
7524 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7525 | * space applications don't need to do io completion events | |
7526 | * polling again, they can rely on io_sq_thread to do polling | |
7527 | * work, which can reduce cpu usage and uring_lock contention. | |
7528 | */ | |
7529 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7530 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7531 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7532 | } else { |
7533 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7534 | } | |
2b188cc1 JA |
7535 | } |
7536 | ||
7c504e65 | 7537 | out: |
6805b32e | 7538 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7539 | out_fput: |
7540 | fdput(f); | |
7541 | return submitted ? submitted : ret; | |
7542 | } | |
7543 | ||
bebdb65e | 7544 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7545 | static int io_uring_show_cred(int id, void *p, void *data) |
7546 | { | |
7547 | const struct cred *cred = p; | |
7548 | struct seq_file *m = data; | |
7549 | struct user_namespace *uns = seq_user_ns(m); | |
7550 | struct group_info *gi; | |
7551 | kernel_cap_t cap; | |
7552 | unsigned __capi; | |
7553 | int g; | |
7554 | ||
7555 | seq_printf(m, "%5d\n", id); | |
7556 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7557 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7558 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7559 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7560 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7561 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7562 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7563 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7564 | seq_puts(m, "\n\tGroups:\t"); | |
7565 | gi = cred->group_info; | |
7566 | for (g = 0; g < gi->ngroups; g++) { | |
7567 | seq_put_decimal_ull(m, g ? " " : "", | |
7568 | from_kgid_munged(uns, gi->gid[g])); | |
7569 | } | |
7570 | seq_puts(m, "\n\tCapEff:\t"); | |
7571 | cap = cred->cap_effective; | |
7572 | CAP_FOR_EACH_U32(__capi) | |
7573 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7574 | seq_putc(m, '\n'); | |
7575 | return 0; | |
7576 | } | |
7577 | ||
7578 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7579 | { | |
7580 | int i; | |
7581 | ||
7582 | mutex_lock(&ctx->uring_lock); | |
7583 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7584 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7585 | struct fixed_file_table *table; | |
7586 | struct file *f; | |
7587 | ||
7588 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7589 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7590 | if (f) | |
7591 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7592 | else | |
7593 | seq_printf(m, "%5u: <none>\n", i); | |
7594 | } | |
7595 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7596 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7597 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7598 | ||
7599 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7600 | (unsigned int) buf->len); | |
7601 | } | |
7602 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7603 | seq_printf(m, "Personalities:\n"); | |
7604 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7605 | } | |
d7718a9d JA |
7606 | seq_printf(m, "PollList:\n"); |
7607 | spin_lock_irq(&ctx->completion_lock); | |
7608 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7609 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7610 | struct io_kiocb *req; | |
7611 | ||
7612 | hlist_for_each_entry(req, list, hash_node) | |
7613 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7614 | req->task->task_works != NULL); | |
7615 | } | |
7616 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7617 | mutex_unlock(&ctx->uring_lock); |
7618 | } | |
7619 | ||
7620 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7621 | { | |
7622 | struct io_ring_ctx *ctx = f->private_data; | |
7623 | ||
7624 | if (percpu_ref_tryget(&ctx->refs)) { | |
7625 | __io_uring_show_fdinfo(ctx, m); | |
7626 | percpu_ref_put(&ctx->refs); | |
7627 | } | |
7628 | } | |
bebdb65e | 7629 | #endif |
87ce955b | 7630 | |
2b188cc1 JA |
7631 | static const struct file_operations io_uring_fops = { |
7632 | .release = io_uring_release, | |
fcb323cc | 7633 | .flush = io_uring_flush, |
2b188cc1 | 7634 | .mmap = io_uring_mmap, |
6c5c240e RP |
7635 | #ifndef CONFIG_MMU |
7636 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7637 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7638 | #endif | |
2b188cc1 JA |
7639 | .poll = io_uring_poll, |
7640 | .fasync = io_uring_fasync, | |
bebdb65e | 7641 | #ifdef CONFIG_PROC_FS |
87ce955b | 7642 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7643 | #endif |
2b188cc1 JA |
7644 | }; |
7645 | ||
7646 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7647 | struct io_uring_params *p) | |
7648 | { | |
75b28aff HV |
7649 | struct io_rings *rings; |
7650 | size_t size, sq_array_offset; | |
2b188cc1 | 7651 | |
75b28aff HV |
7652 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7653 | if (size == SIZE_MAX) | |
7654 | return -EOVERFLOW; | |
7655 | ||
7656 | rings = io_mem_alloc(size); | |
7657 | if (!rings) | |
2b188cc1 JA |
7658 | return -ENOMEM; |
7659 | ||
75b28aff HV |
7660 | ctx->rings = rings; |
7661 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7662 | rings->sq_ring_mask = p->sq_entries - 1; | |
7663 | rings->cq_ring_mask = p->cq_entries - 1; | |
7664 | rings->sq_ring_entries = p->sq_entries; | |
7665 | rings->cq_ring_entries = p->cq_entries; | |
7666 | ctx->sq_mask = rings->sq_ring_mask; | |
7667 | ctx->cq_mask = rings->cq_ring_mask; | |
7668 | ctx->sq_entries = rings->sq_ring_entries; | |
7669 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7670 | |
7671 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7672 | if (size == SIZE_MAX) { |
7673 | io_mem_free(ctx->rings); | |
7674 | ctx->rings = NULL; | |
2b188cc1 | 7675 | return -EOVERFLOW; |
eb065d30 | 7676 | } |
2b188cc1 JA |
7677 | |
7678 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7679 | if (!ctx->sq_sqes) { |
7680 | io_mem_free(ctx->rings); | |
7681 | ctx->rings = NULL; | |
2b188cc1 | 7682 | return -ENOMEM; |
eb065d30 | 7683 | } |
2b188cc1 | 7684 | |
2b188cc1 JA |
7685 | return 0; |
7686 | } | |
7687 | ||
7688 | /* | |
7689 | * Allocate an anonymous fd, this is what constitutes the application | |
7690 | * visible backing of an io_uring instance. The application mmaps this | |
7691 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7692 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7693 | */ | |
7694 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7695 | { | |
7696 | struct file *file; | |
7697 | int ret; | |
7698 | ||
7699 | #if defined(CONFIG_UNIX) | |
7700 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7701 | &ctx->ring_sock); | |
7702 | if (ret) | |
7703 | return ret; | |
7704 | #endif | |
7705 | ||
7706 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7707 | if (ret < 0) | |
7708 | goto err; | |
7709 | ||
7710 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7711 | O_RDWR | O_CLOEXEC); | |
7712 | if (IS_ERR(file)) { | |
7713 | put_unused_fd(ret); | |
7714 | ret = PTR_ERR(file); | |
7715 | goto err; | |
7716 | } | |
7717 | ||
7718 | #if defined(CONFIG_UNIX) | |
7719 | ctx->ring_sock->file = file; | |
7720 | #endif | |
7721 | fd_install(ret, file); | |
7722 | return ret; | |
7723 | err: | |
7724 | #if defined(CONFIG_UNIX) | |
7725 | sock_release(ctx->ring_sock); | |
7726 | ctx->ring_sock = NULL; | |
7727 | #endif | |
7728 | return ret; | |
7729 | } | |
7730 | ||
7731 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7732 | { | |
7733 | struct user_struct *user = NULL; | |
7734 | struct io_ring_ctx *ctx; | |
7735 | bool account_mem; | |
7736 | int ret; | |
7737 | ||
8110c1a6 | 7738 | if (!entries) |
2b188cc1 | 7739 | return -EINVAL; |
8110c1a6 JA |
7740 | if (entries > IORING_MAX_ENTRIES) { |
7741 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7742 | return -EINVAL; | |
7743 | entries = IORING_MAX_ENTRIES; | |
7744 | } | |
2b188cc1 JA |
7745 | |
7746 | /* | |
7747 | * Use twice as many entries for the CQ ring. It's possible for the | |
7748 | * application to drive a higher depth than the size of the SQ ring, | |
7749 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7750 | * some flexibility in overcommitting a bit. If the application has |
7751 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7752 | * of CQ ring entries manually. | |
2b188cc1 JA |
7753 | */ |
7754 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7755 | if (p->flags & IORING_SETUP_CQSIZE) { |
7756 | /* | |
7757 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7758 | * to a power-of-two, if it isn't already. We do NOT impose | |
7759 | * any cq vs sq ring sizing. | |
7760 | */ | |
8110c1a6 | 7761 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7762 | return -EINVAL; |
8110c1a6 JA |
7763 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7764 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7765 | return -EINVAL; | |
7766 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7767 | } | |
33a107f0 JA |
7768 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7769 | } else { | |
7770 | p->cq_entries = 2 * p->sq_entries; | |
7771 | } | |
2b188cc1 JA |
7772 | |
7773 | user = get_uid(current_user()); | |
7774 | account_mem = !capable(CAP_IPC_LOCK); | |
7775 | ||
7776 | if (account_mem) { | |
7777 | ret = io_account_mem(user, | |
7778 | ring_pages(p->sq_entries, p->cq_entries)); | |
7779 | if (ret) { | |
7780 | free_uid(user); | |
7781 | return ret; | |
7782 | } | |
7783 | } | |
7784 | ||
7785 | ctx = io_ring_ctx_alloc(p); | |
7786 | if (!ctx) { | |
7787 | if (account_mem) | |
7788 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7789 | p->cq_entries)); | |
7790 | free_uid(user); | |
7791 | return -ENOMEM; | |
7792 | } | |
7793 | ctx->compat = in_compat_syscall(); | |
7794 | ctx->account_mem = account_mem; | |
7795 | ctx->user = user; | |
0b8c0ec7 | 7796 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7797 | |
7798 | ret = io_allocate_scq_urings(ctx, p); | |
7799 | if (ret) | |
7800 | goto err; | |
7801 | ||
6c271ce2 | 7802 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7803 | if (ret) |
7804 | goto err; | |
7805 | ||
2b188cc1 | 7806 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7807 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7808 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7809 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7810 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7811 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7812 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7813 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7814 | |
7815 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7816 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7817 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7818 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7819 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7820 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7821 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7822 | |
044c1ab3 JA |
7823 | /* |
7824 | * Install ring fd as the very last thing, so we don't risk someone | |
7825 | * having closed it before we finish setup | |
7826 | */ | |
7827 | ret = io_uring_get_fd(ctx); | |
7828 | if (ret < 0) | |
7829 | goto err; | |
7830 | ||
da8c9690 | 7831 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7832 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7833 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7834 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7835 | return ret; |
7836 | err: | |
7837 | io_ring_ctx_wait_and_kill(ctx); | |
7838 | return ret; | |
7839 | } | |
7840 | ||
7841 | /* | |
7842 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7843 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7844 | * params structure passed in. | |
7845 | */ | |
7846 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7847 | { | |
7848 | struct io_uring_params p; | |
7849 | long ret; | |
7850 | int i; | |
7851 | ||
7852 | if (copy_from_user(&p, params, sizeof(p))) | |
7853 | return -EFAULT; | |
7854 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7855 | if (p.resv[i]) | |
7856 | return -EINVAL; | |
7857 | } | |
7858 | ||
6c271ce2 | 7859 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7860 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7861 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7862 | return -EINVAL; |
7863 | ||
7864 | ret = io_uring_create(entries, &p); | |
7865 | if (ret < 0) | |
7866 | return ret; | |
7867 | ||
7868 | if (copy_to_user(params, &p, sizeof(p))) | |
7869 | return -EFAULT; | |
7870 | ||
7871 | return ret; | |
7872 | } | |
7873 | ||
7874 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7875 | struct io_uring_params __user *, params) | |
7876 | { | |
7877 | return io_uring_setup(entries, params); | |
7878 | } | |
7879 | ||
66f4af93 JA |
7880 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7881 | { | |
7882 | struct io_uring_probe *p; | |
7883 | size_t size; | |
7884 | int i, ret; | |
7885 | ||
7886 | size = struct_size(p, ops, nr_args); | |
7887 | if (size == SIZE_MAX) | |
7888 | return -EOVERFLOW; | |
7889 | p = kzalloc(size, GFP_KERNEL); | |
7890 | if (!p) | |
7891 | return -ENOMEM; | |
7892 | ||
7893 | ret = -EFAULT; | |
7894 | if (copy_from_user(p, arg, size)) | |
7895 | goto out; | |
7896 | ret = -EINVAL; | |
7897 | if (memchr_inv(p, 0, size)) | |
7898 | goto out; | |
7899 | ||
7900 | p->last_op = IORING_OP_LAST - 1; | |
7901 | if (nr_args > IORING_OP_LAST) | |
7902 | nr_args = IORING_OP_LAST; | |
7903 | ||
7904 | for (i = 0; i < nr_args; i++) { | |
7905 | p->ops[i].op = i; | |
7906 | if (!io_op_defs[i].not_supported) | |
7907 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7908 | } | |
7909 | p->ops_len = i; | |
7910 | ||
7911 | ret = 0; | |
7912 | if (copy_to_user(arg, p, size)) | |
7913 | ret = -EFAULT; | |
7914 | out: | |
7915 | kfree(p); | |
7916 | return ret; | |
7917 | } | |
7918 | ||
071698e1 JA |
7919 | static int io_register_personality(struct io_ring_ctx *ctx) |
7920 | { | |
7921 | const struct cred *creds = get_current_cred(); | |
7922 | int id; | |
7923 | ||
7924 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7925 | USHRT_MAX, GFP_KERNEL); | |
7926 | if (id < 0) | |
7927 | put_cred(creds); | |
7928 | return id; | |
7929 | } | |
7930 | ||
7931 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7932 | { | |
7933 | const struct cred *old_creds; | |
7934 | ||
7935 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7936 | if (old_creds) { | |
7937 | put_cred(old_creds); | |
7938 | return 0; | |
7939 | } | |
7940 | ||
7941 | return -EINVAL; | |
7942 | } | |
7943 | ||
7944 | static bool io_register_op_must_quiesce(int op) | |
7945 | { | |
7946 | switch (op) { | |
7947 | case IORING_UNREGISTER_FILES: | |
7948 | case IORING_REGISTER_FILES_UPDATE: | |
7949 | case IORING_REGISTER_PROBE: | |
7950 | case IORING_REGISTER_PERSONALITY: | |
7951 | case IORING_UNREGISTER_PERSONALITY: | |
7952 | return false; | |
7953 | default: | |
7954 | return true; | |
7955 | } | |
7956 | } | |
7957 | ||
edafccee JA |
7958 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7959 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7960 | __releases(ctx->uring_lock) |
7961 | __acquires(ctx->uring_lock) | |
edafccee JA |
7962 | { |
7963 | int ret; | |
7964 | ||
35fa71a0 JA |
7965 | /* |
7966 | * We're inside the ring mutex, if the ref is already dying, then | |
7967 | * someone else killed the ctx or is already going through | |
7968 | * io_uring_register(). | |
7969 | */ | |
7970 | if (percpu_ref_is_dying(&ctx->refs)) | |
7971 | return -ENXIO; | |
7972 | ||
071698e1 | 7973 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7974 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7975 | |
05f3fb3c JA |
7976 | /* |
7977 | * Drop uring mutex before waiting for references to exit. If | |
7978 | * another thread is currently inside io_uring_enter() it might | |
7979 | * need to grab the uring_lock to make progress. If we hold it | |
7980 | * here across the drain wait, then we can deadlock. It's safe | |
7981 | * to drop the mutex here, since no new references will come in | |
7982 | * after we've killed the percpu ref. | |
7983 | */ | |
7984 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7985 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7986 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7987 | if (ret) { |
7988 | percpu_ref_resurrect(&ctx->refs); | |
7989 | ret = -EINTR; | |
7990 | goto out; | |
7991 | } | |
05f3fb3c | 7992 | } |
edafccee JA |
7993 | |
7994 | switch (opcode) { | |
7995 | case IORING_REGISTER_BUFFERS: | |
7996 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
7997 | break; | |
7998 | case IORING_UNREGISTER_BUFFERS: | |
7999 | ret = -EINVAL; | |
8000 | if (arg || nr_args) | |
8001 | break; | |
8002 | ret = io_sqe_buffer_unregister(ctx); | |
8003 | break; | |
6b06314c JA |
8004 | case IORING_REGISTER_FILES: |
8005 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8006 | break; | |
8007 | case IORING_UNREGISTER_FILES: | |
8008 | ret = -EINVAL; | |
8009 | if (arg || nr_args) | |
8010 | break; | |
8011 | ret = io_sqe_files_unregister(ctx); | |
8012 | break; | |
c3a31e60 JA |
8013 | case IORING_REGISTER_FILES_UPDATE: |
8014 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8015 | break; | |
9b402849 | 8016 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8017 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8018 | ret = -EINVAL; |
8019 | if (nr_args != 1) | |
8020 | break; | |
8021 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8022 | if (ret) |
8023 | break; | |
8024 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8025 | ctx->eventfd_async = 1; | |
8026 | else | |
8027 | ctx->eventfd_async = 0; | |
9b402849 JA |
8028 | break; |
8029 | case IORING_UNREGISTER_EVENTFD: | |
8030 | ret = -EINVAL; | |
8031 | if (arg || nr_args) | |
8032 | break; | |
8033 | ret = io_eventfd_unregister(ctx); | |
8034 | break; | |
66f4af93 JA |
8035 | case IORING_REGISTER_PROBE: |
8036 | ret = -EINVAL; | |
8037 | if (!arg || nr_args > 256) | |
8038 | break; | |
8039 | ret = io_probe(ctx, arg, nr_args); | |
8040 | break; | |
071698e1 JA |
8041 | case IORING_REGISTER_PERSONALITY: |
8042 | ret = -EINVAL; | |
8043 | if (arg || nr_args) | |
8044 | break; | |
8045 | ret = io_register_personality(ctx); | |
8046 | break; | |
8047 | case IORING_UNREGISTER_PERSONALITY: | |
8048 | ret = -EINVAL; | |
8049 | if (arg) | |
8050 | break; | |
8051 | ret = io_unregister_personality(ctx, nr_args); | |
8052 | break; | |
edafccee JA |
8053 | default: |
8054 | ret = -EINVAL; | |
8055 | break; | |
8056 | } | |
8057 | ||
071698e1 | 8058 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8059 | /* bring the ctx back to life */ |
05f3fb3c | 8060 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8061 | out: |
8062 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8063 | } |
edafccee JA |
8064 | return ret; |
8065 | } | |
8066 | ||
8067 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8068 | void __user *, arg, unsigned int, nr_args) | |
8069 | { | |
8070 | struct io_ring_ctx *ctx; | |
8071 | long ret = -EBADF; | |
8072 | struct fd f; | |
8073 | ||
8074 | f = fdget(fd); | |
8075 | if (!f.file) | |
8076 | return -EBADF; | |
8077 | ||
8078 | ret = -EOPNOTSUPP; | |
8079 | if (f.file->f_op != &io_uring_fops) | |
8080 | goto out_fput; | |
8081 | ||
8082 | ctx = f.file->private_data; | |
8083 | ||
8084 | mutex_lock(&ctx->uring_lock); | |
8085 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8086 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8087 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8088 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8089 | out_fput: |
8090 | fdput(f); | |
8091 | return ret; | |
8092 | } | |
8093 | ||
2b188cc1 JA |
8094 | static int __init io_uring_init(void) |
8095 | { | |
d7f62e82 SM |
8096 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8097 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8098 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8099 | } while (0) | |
8100 | ||
8101 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8102 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8103 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8104 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8105 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8106 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8107 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8108 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8109 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8110 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8111 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8112 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8113 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8114 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8115 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8116 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8117 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8118 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8119 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8120 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8121 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8122 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8123 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8124 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8125 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8126 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8127 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8128 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8129 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8130 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8131 | |
d3656344 | 8132 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8133 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8134 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8135 | return 0; | |
8136 | }; | |
8137 | __initcall(io_uring_init); |