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; | |
4a38aed2 | 194 | struct llist_node llist; |
05589553 XW |
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 | |
0f158b4c JA |
282 | struct completion ref_comp; |
283 | struct completion sq_thread_comp; | |
206aefde | 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 | 329 | |
4a38aed2 JA |
330 | struct delayed_work file_put_work; |
331 | struct llist_head file_put_llist; | |
332 | ||
85faa7b8 | 333 | struct work_struct exit_work; |
2b188cc1 JA |
334 | }; |
335 | ||
09bb8394 JA |
336 | /* |
337 | * First field must be the file pointer in all the | |
338 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
339 | */ | |
221c5eb2 JA |
340 | struct io_poll_iocb { |
341 | struct file *file; | |
0969e783 JA |
342 | union { |
343 | struct wait_queue_head *head; | |
344 | u64 addr; | |
345 | }; | |
221c5eb2 | 346 | __poll_t events; |
8c838788 | 347 | bool done; |
221c5eb2 | 348 | bool canceled; |
392edb45 | 349 | struct wait_queue_entry wait; |
221c5eb2 JA |
350 | }; |
351 | ||
b5dba59e JA |
352 | struct io_close { |
353 | struct file *file; | |
354 | struct file *put_file; | |
355 | int fd; | |
356 | }; | |
357 | ||
ad8a48ac JA |
358 | struct io_timeout_data { |
359 | struct io_kiocb *req; | |
360 | struct hrtimer timer; | |
361 | struct timespec64 ts; | |
362 | enum hrtimer_mode mode; | |
363 | }; | |
364 | ||
8ed8d3c3 JA |
365 | struct io_accept { |
366 | struct file *file; | |
367 | struct sockaddr __user *addr; | |
368 | int __user *addr_len; | |
369 | int flags; | |
09952e3e | 370 | unsigned long nofile; |
8ed8d3c3 JA |
371 | }; |
372 | ||
373 | struct io_sync { | |
374 | struct file *file; | |
375 | loff_t len; | |
376 | loff_t off; | |
377 | int flags; | |
d63d1b5e | 378 | int mode; |
8ed8d3c3 JA |
379 | }; |
380 | ||
fbf23849 JA |
381 | struct io_cancel { |
382 | struct file *file; | |
383 | u64 addr; | |
384 | }; | |
385 | ||
b29472ee JA |
386 | struct io_timeout { |
387 | struct file *file; | |
388 | u64 addr; | |
389 | int flags; | |
b55ce732 | 390 | u32 count; |
b29472ee JA |
391 | }; |
392 | ||
9adbd45d JA |
393 | struct io_rw { |
394 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
395 | struct kiocb kiocb; | |
396 | u64 addr; | |
397 | u64 len; | |
398 | }; | |
399 | ||
3fbb51c1 JA |
400 | struct io_connect { |
401 | struct file *file; | |
402 | struct sockaddr __user *addr; | |
403 | int addr_len; | |
404 | }; | |
405 | ||
e47293fd JA |
406 | struct io_sr_msg { |
407 | struct file *file; | |
fddaface JA |
408 | union { |
409 | struct user_msghdr __user *msg; | |
410 | void __user *buf; | |
411 | }; | |
e47293fd | 412 | int msg_flags; |
bcda7baa | 413 | int bgid; |
fddaface | 414 | size_t len; |
bcda7baa | 415 | struct io_buffer *kbuf; |
e47293fd JA |
416 | }; |
417 | ||
15b71abe JA |
418 | struct io_open { |
419 | struct file *file; | |
420 | int dfd; | |
eddc7ef5 | 421 | union { |
eddc7ef5 JA |
422 | unsigned mask; |
423 | }; | |
15b71abe | 424 | struct filename *filename; |
eddc7ef5 | 425 | struct statx __user *buffer; |
c12cedf2 | 426 | struct open_how how; |
4022e7af | 427 | unsigned long nofile; |
15b71abe JA |
428 | }; |
429 | ||
05f3fb3c JA |
430 | struct io_files_update { |
431 | struct file *file; | |
432 | u64 arg; | |
433 | u32 nr_args; | |
434 | u32 offset; | |
435 | }; | |
436 | ||
4840e418 JA |
437 | struct io_fadvise { |
438 | struct file *file; | |
439 | u64 offset; | |
440 | u32 len; | |
441 | u32 advice; | |
442 | }; | |
443 | ||
c1ca757b JA |
444 | struct io_madvise { |
445 | struct file *file; | |
446 | u64 addr; | |
447 | u32 len; | |
448 | u32 advice; | |
449 | }; | |
450 | ||
3e4827b0 JA |
451 | struct io_epoll { |
452 | struct file *file; | |
453 | int epfd; | |
454 | int op; | |
455 | int fd; | |
456 | struct epoll_event event; | |
e47293fd JA |
457 | }; |
458 | ||
7d67af2c PB |
459 | struct io_splice { |
460 | struct file *file_out; | |
461 | struct file *file_in; | |
462 | loff_t off_out; | |
463 | loff_t off_in; | |
464 | u64 len; | |
465 | unsigned int flags; | |
466 | }; | |
467 | ||
ddf0322d JA |
468 | struct io_provide_buf { |
469 | struct file *file; | |
470 | __u64 addr; | |
471 | __s32 len; | |
472 | __u32 bgid; | |
473 | __u16 nbufs; | |
474 | __u16 bid; | |
475 | }; | |
476 | ||
f499a021 JA |
477 | struct io_async_connect { |
478 | struct sockaddr_storage address; | |
479 | }; | |
480 | ||
03b1230c JA |
481 | struct io_async_msghdr { |
482 | struct iovec fast_iov[UIO_FASTIOV]; | |
483 | struct iovec *iov; | |
484 | struct sockaddr __user *uaddr; | |
485 | struct msghdr msg; | |
b537916c | 486 | struct sockaddr_storage addr; |
03b1230c JA |
487 | }; |
488 | ||
f67676d1 JA |
489 | struct io_async_rw { |
490 | struct iovec fast_iov[UIO_FASTIOV]; | |
491 | struct iovec *iov; | |
492 | ssize_t nr_segs; | |
493 | ssize_t size; | |
494 | }; | |
495 | ||
1a6b74fc | 496 | struct io_async_ctx { |
f67676d1 JA |
497 | union { |
498 | struct io_async_rw rw; | |
03b1230c | 499 | struct io_async_msghdr msg; |
f499a021 | 500 | struct io_async_connect connect; |
2d28390a | 501 | struct io_timeout_data timeout; |
f67676d1 | 502 | }; |
1a6b74fc JA |
503 | }; |
504 | ||
6b47ee6e PB |
505 | enum { |
506 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
507 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
508 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
509 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
510 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 511 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 512 | |
dea3b49c | 513 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
514 | REQ_F_LINK_NEXT_BIT, |
515 | REQ_F_FAIL_LINK_BIT, | |
516 | REQ_F_INFLIGHT_BIT, | |
517 | REQ_F_CUR_POS_BIT, | |
518 | REQ_F_NOWAIT_BIT, | |
519 | REQ_F_IOPOLL_COMPLETED_BIT, | |
520 | REQ_F_LINK_TIMEOUT_BIT, | |
521 | REQ_F_TIMEOUT_BIT, | |
522 | REQ_F_ISREG_BIT, | |
523 | REQ_F_MUST_PUNT_BIT, | |
524 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
525 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 526 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 527 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 528 | REQ_F_POLLED_BIT, |
bcda7baa | 529 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 530 | REQ_F_NO_FILE_TABLE_BIT, |
84557871 JA |
531 | |
532 | /* not a real bit, just to check we're not overflowing the space */ | |
533 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
534 | }; |
535 | ||
536 | enum { | |
537 | /* ctx owns file */ | |
538 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
539 | /* drain existing IO first */ | |
540 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
541 | /* linked sqes */ | |
542 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
543 | /* doesn't sever on completion < 0 */ | |
544 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
545 | /* IOSQE_ASYNC */ | |
546 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
547 | /* IOSQE_BUFFER_SELECT */ |
548 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 549 | |
dea3b49c PB |
550 | /* head of a link */ |
551 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
552 | /* already grabbed next link */ |
553 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
554 | /* fail rest of links */ | |
555 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
556 | /* on inflight list */ | |
557 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
558 | /* read/write uses file position */ | |
559 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
560 | /* must not punt to workers */ | |
561 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
562 | /* polled IO has completed */ | |
563 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
564 | /* has linked timeout */ | |
565 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
566 | /* timeout request */ | |
567 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
568 | /* regular file */ | |
569 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
570 | /* must be punted even for NONBLOCK */ | |
571 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
572 | /* no timeout sequence */ | |
573 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
574 | /* completion under lock */ | |
575 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
576 | /* needs cleanup */ |
577 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
578 | /* in overflow list */ |
579 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
580 | /* already went through poll handler */ |
581 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
582 | /* buffer already selected */ |
583 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
584 | /* doesn't need file table for this request */ |
585 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d7718a9d JA |
586 | }; |
587 | ||
588 | struct async_poll { | |
589 | struct io_poll_iocb poll; | |
590 | struct io_wq_work work; | |
6b47ee6e PB |
591 | }; |
592 | ||
09bb8394 JA |
593 | /* |
594 | * NOTE! Each of the iocb union members has the file pointer | |
595 | * as the first entry in their struct definition. So you can | |
596 | * access the file pointer through any of the sub-structs, | |
597 | * or directly as just 'ki_filp' in this struct. | |
598 | */ | |
2b188cc1 | 599 | struct io_kiocb { |
221c5eb2 | 600 | union { |
09bb8394 | 601 | struct file *file; |
9adbd45d | 602 | struct io_rw rw; |
221c5eb2 | 603 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
604 | struct io_accept accept; |
605 | struct io_sync sync; | |
fbf23849 | 606 | struct io_cancel cancel; |
b29472ee | 607 | struct io_timeout timeout; |
3fbb51c1 | 608 | struct io_connect connect; |
e47293fd | 609 | struct io_sr_msg sr_msg; |
15b71abe | 610 | struct io_open open; |
b5dba59e | 611 | struct io_close close; |
05f3fb3c | 612 | struct io_files_update files_update; |
4840e418 | 613 | struct io_fadvise fadvise; |
c1ca757b | 614 | struct io_madvise madvise; |
3e4827b0 | 615 | struct io_epoll epoll; |
7d67af2c | 616 | struct io_splice splice; |
ddf0322d | 617 | struct io_provide_buf pbuf; |
221c5eb2 | 618 | }; |
2b188cc1 | 619 | |
1a6b74fc | 620 | struct io_async_ctx *io; |
c398ecb3 | 621 | int cflags; |
cf6fd4bd | 622 | bool needs_fixed_file; |
d625c6ee | 623 | u8 opcode; |
2b188cc1 JA |
624 | |
625 | struct io_ring_ctx *ctx; | |
d7718a9d | 626 | struct list_head list; |
2b188cc1 | 627 | unsigned int flags; |
c16361c1 | 628 | refcount_t refs; |
3537b6a7 JA |
629 | struct task_struct *task; |
630 | unsigned long fsize; | |
2b188cc1 | 631 | u64 user_data; |
9e645e11 | 632 | u32 result; |
de0617e4 | 633 | u32 sequence; |
2b188cc1 | 634 | |
d7718a9d JA |
635 | struct list_head link_list; |
636 | ||
fcb323cc JA |
637 | struct list_head inflight_entry; |
638 | ||
05589553 XW |
639 | struct percpu_ref *fixed_file_refs; |
640 | ||
b41e9852 JA |
641 | union { |
642 | /* | |
643 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
644 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
645 | * async armed poll handlers for regular commands. The latter | |
646 | * restore the work, if needed. | |
b41e9852 JA |
647 | */ |
648 | struct { | |
b41e9852 | 649 | struct callback_head task_work; |
d7718a9d JA |
650 | struct hlist_node hash_node; |
651 | struct async_poll *apoll; | |
b41e9852 JA |
652 | }; |
653 | struct io_wq_work work; | |
654 | }; | |
2b188cc1 JA |
655 | }; |
656 | ||
657 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 658 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 659 | |
9a56a232 JA |
660 | struct io_submit_state { |
661 | struct blk_plug plug; | |
662 | ||
2579f913 JA |
663 | /* |
664 | * io_kiocb alloc cache | |
665 | */ | |
666 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 667 | unsigned int free_reqs; |
2579f913 | 668 | |
9a56a232 JA |
669 | /* |
670 | * File reference cache | |
671 | */ | |
672 | struct file *file; | |
673 | unsigned int fd; | |
674 | unsigned int has_refs; | |
675 | unsigned int used_refs; | |
676 | unsigned int ios_left; | |
677 | }; | |
678 | ||
d3656344 JA |
679 | struct io_op_def { |
680 | /* needs req->io allocated for deferral/async */ | |
681 | unsigned async_ctx : 1; | |
682 | /* needs current->mm setup, does mm access */ | |
683 | unsigned needs_mm : 1; | |
684 | /* needs req->file assigned */ | |
685 | unsigned needs_file : 1; | |
d3656344 JA |
686 | /* hash wq insertion if file is a regular file */ |
687 | unsigned hash_reg_file : 1; | |
688 | /* unbound wq insertion if file is a non-regular file */ | |
689 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
690 | /* opcode is not supported by this kernel */ |
691 | unsigned not_supported : 1; | |
f86cd20c JA |
692 | /* needs file table */ |
693 | unsigned file_table : 1; | |
ff002b30 JA |
694 | /* needs ->fs */ |
695 | unsigned needs_fs : 1; | |
8a72758c JA |
696 | /* set if opcode supports polled "wait" */ |
697 | unsigned pollin : 1; | |
698 | unsigned pollout : 1; | |
bcda7baa JA |
699 | /* op supports buffer selection */ |
700 | unsigned buffer_select : 1; | |
d3656344 JA |
701 | }; |
702 | ||
703 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
704 | [IORING_OP_NOP] = {}, |
705 | [IORING_OP_READV] = { | |
d3656344 JA |
706 | .async_ctx = 1, |
707 | .needs_mm = 1, | |
708 | .needs_file = 1, | |
709 | .unbound_nonreg_file = 1, | |
8a72758c | 710 | .pollin = 1, |
4d954c25 | 711 | .buffer_select = 1, |
d3656344 | 712 | }, |
0463b6c5 | 713 | [IORING_OP_WRITEV] = { |
d3656344 JA |
714 | .async_ctx = 1, |
715 | .needs_mm = 1, | |
716 | .needs_file = 1, | |
717 | .hash_reg_file = 1, | |
718 | .unbound_nonreg_file = 1, | |
8a72758c | 719 | .pollout = 1, |
d3656344 | 720 | }, |
0463b6c5 | 721 | [IORING_OP_FSYNC] = { |
d3656344 JA |
722 | .needs_file = 1, |
723 | }, | |
0463b6c5 | 724 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
725 | .needs_file = 1, |
726 | .unbound_nonreg_file = 1, | |
8a72758c | 727 | .pollin = 1, |
d3656344 | 728 | }, |
0463b6c5 | 729 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
730 | .needs_file = 1, |
731 | .hash_reg_file = 1, | |
732 | .unbound_nonreg_file = 1, | |
8a72758c | 733 | .pollout = 1, |
d3656344 | 734 | }, |
0463b6c5 | 735 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
736 | .needs_file = 1, |
737 | .unbound_nonreg_file = 1, | |
738 | }, | |
0463b6c5 PB |
739 | [IORING_OP_POLL_REMOVE] = {}, |
740 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
741 | .needs_file = 1, |
742 | }, | |
0463b6c5 | 743 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
744 | .async_ctx = 1, |
745 | .needs_mm = 1, | |
746 | .needs_file = 1, | |
747 | .unbound_nonreg_file = 1, | |
ff002b30 | 748 | .needs_fs = 1, |
8a72758c | 749 | .pollout = 1, |
d3656344 | 750 | }, |
0463b6c5 | 751 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
752 | .async_ctx = 1, |
753 | .needs_mm = 1, | |
754 | .needs_file = 1, | |
755 | .unbound_nonreg_file = 1, | |
ff002b30 | 756 | .needs_fs = 1, |
8a72758c | 757 | .pollin = 1, |
52de1fe1 | 758 | .buffer_select = 1, |
d3656344 | 759 | }, |
0463b6c5 | 760 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
761 | .async_ctx = 1, |
762 | .needs_mm = 1, | |
763 | }, | |
0463b6c5 PB |
764 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
765 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
766 | .needs_mm = 1, |
767 | .needs_file = 1, | |
768 | .unbound_nonreg_file = 1, | |
f86cd20c | 769 | .file_table = 1, |
8a72758c | 770 | .pollin = 1, |
d3656344 | 771 | }, |
0463b6c5 PB |
772 | [IORING_OP_ASYNC_CANCEL] = {}, |
773 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
774 | .async_ctx = 1, |
775 | .needs_mm = 1, | |
776 | }, | |
0463b6c5 | 777 | [IORING_OP_CONNECT] = { |
d3656344 JA |
778 | .async_ctx = 1, |
779 | .needs_mm = 1, | |
780 | .needs_file = 1, | |
781 | .unbound_nonreg_file = 1, | |
8a72758c | 782 | .pollout = 1, |
d3656344 | 783 | }, |
0463b6c5 | 784 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
785 | .needs_file = 1, |
786 | }, | |
0463b6c5 | 787 | [IORING_OP_OPENAT] = { |
f86cd20c | 788 | .file_table = 1, |
ff002b30 | 789 | .needs_fs = 1, |
d3656344 | 790 | }, |
0463b6c5 | 791 | [IORING_OP_CLOSE] = { |
f86cd20c | 792 | .file_table = 1, |
d3656344 | 793 | }, |
0463b6c5 | 794 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 795 | .needs_mm = 1, |
f86cd20c | 796 | .file_table = 1, |
d3656344 | 797 | }, |
0463b6c5 | 798 | [IORING_OP_STATX] = { |
d3656344 | 799 | .needs_mm = 1, |
ff002b30 | 800 | .needs_fs = 1, |
5b0bbee4 | 801 | .file_table = 1, |
d3656344 | 802 | }, |
0463b6c5 | 803 | [IORING_OP_READ] = { |
3a6820f2 JA |
804 | .needs_mm = 1, |
805 | .needs_file = 1, | |
806 | .unbound_nonreg_file = 1, | |
8a72758c | 807 | .pollin = 1, |
bcda7baa | 808 | .buffer_select = 1, |
3a6820f2 | 809 | }, |
0463b6c5 | 810 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
811 | .needs_mm = 1, |
812 | .needs_file = 1, | |
813 | .unbound_nonreg_file = 1, | |
8a72758c | 814 | .pollout = 1, |
3a6820f2 | 815 | }, |
0463b6c5 | 816 | [IORING_OP_FADVISE] = { |
4840e418 JA |
817 | .needs_file = 1, |
818 | }, | |
0463b6c5 | 819 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
820 | .needs_mm = 1, |
821 | }, | |
0463b6c5 | 822 | [IORING_OP_SEND] = { |
fddaface JA |
823 | .needs_mm = 1, |
824 | .needs_file = 1, | |
825 | .unbound_nonreg_file = 1, | |
8a72758c | 826 | .pollout = 1, |
fddaface | 827 | }, |
0463b6c5 | 828 | [IORING_OP_RECV] = { |
fddaface JA |
829 | .needs_mm = 1, |
830 | .needs_file = 1, | |
831 | .unbound_nonreg_file = 1, | |
8a72758c | 832 | .pollin = 1, |
bcda7baa | 833 | .buffer_select = 1, |
fddaface | 834 | }, |
0463b6c5 | 835 | [IORING_OP_OPENAT2] = { |
f86cd20c | 836 | .file_table = 1, |
ff002b30 | 837 | .needs_fs = 1, |
cebdb986 | 838 | }, |
3e4827b0 JA |
839 | [IORING_OP_EPOLL_CTL] = { |
840 | .unbound_nonreg_file = 1, | |
841 | .file_table = 1, | |
842 | }, | |
7d67af2c PB |
843 | [IORING_OP_SPLICE] = { |
844 | .needs_file = 1, | |
845 | .hash_reg_file = 1, | |
846 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
847 | }, |
848 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 849 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
850 | }; |
851 | ||
561fb04a | 852 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 853 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 854 | static void io_put_req(struct io_kiocb *req); |
978db57e | 855 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
856 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
857 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
858 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
859 | struct io_uring_files_update *ip, | |
860 | unsigned nr_args); | |
f86cd20c | 861 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 862 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
863 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
864 | int fd, struct file **out_file, bool fixed); | |
865 | static void __io_queue_sqe(struct io_kiocb *req, | |
866 | const struct io_uring_sqe *sqe); | |
de0617e4 | 867 | |
2b188cc1 JA |
868 | static struct kmem_cache *req_cachep; |
869 | ||
870 | static const struct file_operations io_uring_fops; | |
871 | ||
872 | struct sock *io_uring_get_socket(struct file *file) | |
873 | { | |
874 | #if defined(CONFIG_UNIX) | |
875 | if (file->f_op == &io_uring_fops) { | |
876 | struct io_ring_ctx *ctx = file->private_data; | |
877 | ||
878 | return ctx->ring_sock->sk; | |
879 | } | |
880 | #endif | |
881 | return NULL; | |
882 | } | |
883 | EXPORT_SYMBOL(io_uring_get_socket); | |
884 | ||
4a38aed2 JA |
885 | static void io_file_put_work(struct work_struct *work); |
886 | ||
2b188cc1 JA |
887 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
888 | { | |
889 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
890 | ||
0f158b4c | 891 | complete(&ctx->ref_comp); |
2b188cc1 JA |
892 | } |
893 | ||
894 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
895 | { | |
896 | struct io_ring_ctx *ctx; | |
78076bb6 | 897 | int hash_bits; |
2b188cc1 JA |
898 | |
899 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
900 | if (!ctx) | |
901 | return NULL; | |
902 | ||
0ddf92e8 JA |
903 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
904 | if (!ctx->fallback_req) | |
905 | goto err; | |
906 | ||
78076bb6 JA |
907 | /* |
908 | * Use 5 bits less than the max cq entries, that should give us around | |
909 | * 32 entries per hash list if totally full and uniformly spread. | |
910 | */ | |
911 | hash_bits = ilog2(p->cq_entries); | |
912 | hash_bits -= 5; | |
913 | if (hash_bits <= 0) | |
914 | hash_bits = 1; | |
915 | ctx->cancel_hash_bits = hash_bits; | |
916 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
917 | GFP_KERNEL); | |
918 | if (!ctx->cancel_hash) | |
919 | goto err; | |
920 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
921 | ||
21482896 | 922 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
923 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
924 | goto err; | |
2b188cc1 JA |
925 | |
926 | ctx->flags = p->flags; | |
927 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 928 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
929 | init_completion(&ctx->ref_comp); |
930 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 931 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 932 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
933 | mutex_init(&ctx->uring_lock); |
934 | init_waitqueue_head(&ctx->wait); | |
935 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 936 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 937 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 938 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
939 | init_waitqueue_head(&ctx->inflight_wait); |
940 | spin_lock_init(&ctx->inflight_lock); | |
941 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
942 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
943 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 944 | return ctx; |
206aefde | 945 | err: |
0ddf92e8 JA |
946 | if (ctx->fallback_req) |
947 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 948 | kfree(ctx->cancel_hash); |
206aefde JA |
949 | kfree(ctx); |
950 | return NULL; | |
2b188cc1 JA |
951 | } |
952 | ||
9d858b21 | 953 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 954 | { |
a197f664 JL |
955 | struct io_ring_ctx *ctx = req->ctx; |
956 | ||
31af27c7 PB |
957 | return req->sequence != ctx->cached_cq_tail |
958 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
959 | } |
960 | ||
9d858b21 | 961 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 962 | { |
87987898 | 963 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 964 | return __req_need_defer(req); |
de0617e4 | 965 | |
9d858b21 | 966 | return false; |
de0617e4 JA |
967 | } |
968 | ||
7adf4eaf | 969 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
970 | { |
971 | struct io_kiocb *req; | |
972 | ||
7adf4eaf | 973 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 974 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
975 | list_del_init(&req->list); |
976 | return req; | |
977 | } | |
978 | ||
979 | return NULL; | |
980 | } | |
981 | ||
5262f567 JA |
982 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
983 | { | |
7adf4eaf JA |
984 | struct io_kiocb *req; |
985 | ||
986 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
987 | if (req) { |
988 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
989 | return NULL; | |
fb4b3d3f | 990 | if (!__req_need_defer(req)) { |
93bd25bb JA |
991 | list_del_init(&req->list); |
992 | return req; | |
993 | } | |
7adf4eaf JA |
994 | } |
995 | ||
996 | return NULL; | |
5262f567 JA |
997 | } |
998 | ||
de0617e4 | 999 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1000 | { |
75b28aff | 1001 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1002 | |
07910158 PB |
1003 | /* order cqe stores with ring update */ |
1004 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1005 | |
07910158 PB |
1006 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1007 | wake_up_interruptible(&ctx->cq_wait); | |
1008 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1009 | } |
1010 | } | |
1011 | ||
cccf0ee8 JA |
1012 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1013 | const struct io_op_def *def) | |
1014 | { | |
1015 | if (!req->work.mm && def->needs_mm) { | |
1016 | mmgrab(current->mm); | |
1017 | req->work.mm = current->mm; | |
2b188cc1 | 1018 | } |
cccf0ee8 JA |
1019 | if (!req->work.creds) |
1020 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1021 | if (!req->work.fs && def->needs_fs) { |
1022 | spin_lock(¤t->fs->lock); | |
1023 | if (!current->fs->in_exec) { | |
1024 | req->work.fs = current->fs; | |
1025 | req->work.fs->users++; | |
1026 | } else { | |
1027 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1028 | } | |
1029 | spin_unlock(¤t->fs->lock); | |
1030 | } | |
6ab23144 JA |
1031 | if (!req->work.task_pid) |
1032 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1033 | } |
1034 | ||
cccf0ee8 | 1035 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1036 | { |
cccf0ee8 JA |
1037 | if (req->work.mm) { |
1038 | mmdrop(req->work.mm); | |
1039 | req->work.mm = NULL; | |
1040 | } | |
1041 | if (req->work.creds) { | |
1042 | put_cred(req->work.creds); | |
1043 | req->work.creds = NULL; | |
1044 | } | |
ff002b30 JA |
1045 | if (req->work.fs) { |
1046 | struct fs_struct *fs = req->work.fs; | |
1047 | ||
1048 | spin_lock(&req->work.fs->lock); | |
1049 | if (--fs->users) | |
1050 | fs = NULL; | |
1051 | spin_unlock(&req->work.fs->lock); | |
1052 | if (fs) | |
1053 | free_fs_struct(fs); | |
1054 | } | |
561fb04a JA |
1055 | } |
1056 | ||
8766dd51 | 1057 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1058 | struct io_kiocb **link) |
18d9be1a | 1059 | { |
d3656344 | 1060 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1061 | |
d3656344 JA |
1062 | if (req->flags & REQ_F_ISREG) { |
1063 | if (def->hash_reg_file) | |
8766dd51 | 1064 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1065 | } else { |
1066 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1067 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1068 | } |
cccf0ee8 JA |
1069 | |
1070 | io_req_work_grab_env(req, def); | |
54a91f3b | 1071 | |
94ae5e77 | 1072 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1073 | } |
1074 | ||
a197f664 | 1075 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1076 | { |
a197f664 | 1077 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1078 | struct io_kiocb *link; |
94ae5e77 | 1079 | |
8766dd51 | 1080 | io_prep_async_work(req, &link); |
561fb04a | 1081 | |
8766dd51 PB |
1082 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1083 | &req->work, req->flags); | |
1084 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1085 | |
1086 | if (link) | |
1087 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1088 | } |
1089 | ||
5262f567 JA |
1090 | static void io_kill_timeout(struct io_kiocb *req) |
1091 | { | |
1092 | int ret; | |
1093 | ||
2d28390a | 1094 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1095 | if (ret != -1) { |
1096 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1097 | list_del_init(&req->list); |
f0e20b89 | 1098 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1099 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1100 | io_put_req(req); |
5262f567 JA |
1101 | } |
1102 | } | |
1103 | ||
1104 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1105 | { | |
1106 | struct io_kiocb *req, *tmp; | |
1107 | ||
1108 | spin_lock_irq(&ctx->completion_lock); | |
1109 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1110 | io_kill_timeout(req); | |
1111 | spin_unlock_irq(&ctx->completion_lock); | |
1112 | } | |
1113 | ||
de0617e4 JA |
1114 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1115 | { | |
1116 | struct io_kiocb *req; | |
1117 | ||
5262f567 JA |
1118 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1119 | io_kill_timeout(req); | |
1120 | ||
de0617e4 JA |
1121 | __io_commit_cqring(ctx); |
1122 | ||
87987898 | 1123 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1124 | io_queue_async_work(req); |
de0617e4 JA |
1125 | } |
1126 | ||
2b188cc1 JA |
1127 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1128 | { | |
75b28aff | 1129 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1130 | unsigned tail; |
1131 | ||
1132 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1133 | /* |
1134 | * writes to the cq entry need to come after reading head; the | |
1135 | * control dependency is enough as we're using WRITE_ONCE to | |
1136 | * fill the cq entry | |
1137 | */ | |
75b28aff | 1138 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1139 | return NULL; |
1140 | ||
1141 | ctx->cached_cq_tail++; | |
75b28aff | 1142 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1143 | } |
1144 | ||
f2842ab5 JA |
1145 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1146 | { | |
f0b493e6 JA |
1147 | if (!ctx->cq_ev_fd) |
1148 | return false; | |
f2842ab5 JA |
1149 | if (!ctx->eventfd_async) |
1150 | return true; | |
b41e9852 | 1151 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1152 | } |
1153 | ||
b41e9852 | 1154 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1155 | { |
1156 | if (waitqueue_active(&ctx->wait)) | |
1157 | wake_up(&ctx->wait); | |
1158 | if (waitqueue_active(&ctx->sqo_wait)) | |
1159 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1160 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1161 | eventfd_signal(ctx->cq_ev_fd, 1); |
1162 | } | |
1163 | ||
c4a2ed72 JA |
1164 | /* Returns true if there are no backlogged entries after the flush */ |
1165 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1166 | { |
1167 | struct io_rings *rings = ctx->rings; | |
1168 | struct io_uring_cqe *cqe; | |
1169 | struct io_kiocb *req; | |
1170 | unsigned long flags; | |
1171 | LIST_HEAD(list); | |
1172 | ||
1173 | if (!force) { | |
1174 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1175 | return true; |
1d7bb1d5 JA |
1176 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1177 | rings->cq_ring_entries)) | |
c4a2ed72 | 1178 | return false; |
1d7bb1d5 JA |
1179 | } |
1180 | ||
1181 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1182 | ||
1183 | /* if force is set, the ring is going away. always drop after that */ | |
1184 | if (force) | |
69b3e546 | 1185 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1186 | |
c4a2ed72 | 1187 | cqe = NULL; |
1d7bb1d5 JA |
1188 | while (!list_empty(&ctx->cq_overflow_list)) { |
1189 | cqe = io_get_cqring(ctx); | |
1190 | if (!cqe && !force) | |
1191 | break; | |
1192 | ||
1193 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1194 | list); | |
1195 | list_move(&req->list, &list); | |
2ca10259 | 1196 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1197 | if (cqe) { |
1198 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1199 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1200 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1201 | } else { |
1202 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1203 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1208 | if (cqe) { |
1209 | clear_bit(0, &ctx->sq_check_overflow); | |
1210 | clear_bit(0, &ctx->cq_check_overflow); | |
1211 | } | |
1d7bb1d5 JA |
1212 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1213 | io_cqring_ev_posted(ctx); | |
1214 | ||
1215 | while (!list_empty(&list)) { | |
1216 | req = list_first_entry(&list, struct io_kiocb, list); | |
1217 | list_del(&req->list); | |
ec9c02ad | 1218 | io_put_req(req); |
1d7bb1d5 | 1219 | } |
c4a2ed72 JA |
1220 | |
1221 | return cqe != NULL; | |
1d7bb1d5 JA |
1222 | } |
1223 | ||
bcda7baa | 1224 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1225 | { |
78e19bbe | 1226 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1227 | struct io_uring_cqe *cqe; |
1228 | ||
78e19bbe | 1229 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1230 | |
2b188cc1 JA |
1231 | /* |
1232 | * If we can't get a cq entry, userspace overflowed the | |
1233 | * submission (by quite a lot). Increment the overflow count in | |
1234 | * the ring. | |
1235 | */ | |
1236 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1237 | if (likely(cqe)) { |
78e19bbe | 1238 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1239 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1240 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1241 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1242 | WRITE_ONCE(ctx->rings->cq_overflow, |
1243 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1244 | } else { |
ad3eb2c8 JA |
1245 | if (list_empty(&ctx->cq_overflow_list)) { |
1246 | set_bit(0, &ctx->sq_check_overflow); | |
1247 | set_bit(0, &ctx->cq_check_overflow); | |
1248 | } | |
2ca10259 | 1249 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1250 | refcount_inc(&req->refs); |
1251 | req->result = res; | |
bcda7baa | 1252 | req->cflags = cflags; |
1d7bb1d5 | 1253 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1254 | } |
1255 | } | |
1256 | ||
bcda7baa JA |
1257 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1258 | { | |
1259 | __io_cqring_fill_event(req, res, 0); | |
1260 | } | |
1261 | ||
1262 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1263 | { |
78e19bbe | 1264 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1265 | unsigned long flags; |
1266 | ||
1267 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1268 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1269 | io_commit_cqring(ctx); |
1270 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1271 | ||
8c838788 | 1272 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1273 | } |
1274 | ||
bcda7baa JA |
1275 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1276 | { | |
1277 | __io_cqring_add_event(req, res, 0); | |
1278 | } | |
1279 | ||
0ddf92e8 JA |
1280 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1281 | { | |
1282 | return req == (struct io_kiocb *) | |
1283 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1284 | } | |
1285 | ||
1286 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1287 | { | |
1288 | struct io_kiocb *req; | |
1289 | ||
1290 | req = ctx->fallback_req; | |
dd461af6 | 1291 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1292 | return req; |
1293 | ||
1294 | return NULL; | |
1295 | } | |
1296 | ||
0553b8bd PB |
1297 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1298 | struct io_submit_state *state) | |
2b188cc1 | 1299 | { |
fd6fab2c | 1300 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1301 | struct io_kiocb *req; |
1302 | ||
2579f913 | 1303 | if (!state) { |
fd6fab2c | 1304 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1305 | if (unlikely(!req)) |
0ddf92e8 | 1306 | goto fallback; |
2579f913 JA |
1307 | } else if (!state->free_reqs) { |
1308 | size_t sz; | |
1309 | int ret; | |
1310 | ||
1311 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1312 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1313 | ||
1314 | /* | |
1315 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1316 | * retry single alloc to be on the safe side. | |
1317 | */ | |
1318 | if (unlikely(ret <= 0)) { | |
1319 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1320 | if (!state->reqs[0]) | |
0ddf92e8 | 1321 | goto fallback; |
fd6fab2c JA |
1322 | ret = 1; |
1323 | } | |
2579f913 | 1324 | state->free_reqs = ret - 1; |
6c8a3134 | 1325 | req = state->reqs[ret - 1]; |
2579f913 | 1326 | } else { |
2579f913 | 1327 | state->free_reqs--; |
6c8a3134 | 1328 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1329 | } |
1330 | ||
2579f913 | 1331 | return req; |
0ddf92e8 | 1332 | fallback: |
0553b8bd | 1333 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1334 | } |
1335 | ||
8da11c19 PB |
1336 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1337 | bool fixed) | |
1338 | { | |
1339 | if (fixed) | |
05589553 | 1340 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1341 | else |
1342 | fput(file); | |
1343 | } | |
1344 | ||
c6ca97b3 | 1345 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1346 | { |
929a3af9 PB |
1347 | if (req->flags & REQ_F_NEED_CLEANUP) |
1348 | io_cleanup_req(req); | |
1349 | ||
96fd84d8 | 1350 | kfree(req->io); |
8da11c19 PB |
1351 | if (req->file) |
1352 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1353 | if (req->task) |
1354 | put_task_struct(req->task); | |
cccf0ee8 JA |
1355 | |
1356 | io_req_work_drop_env(req); | |
def596e9 JA |
1357 | } |
1358 | ||
9e645e11 | 1359 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1360 | { |
c6ca97b3 | 1361 | __io_req_aux_free(req); |
fcb323cc | 1362 | |
fcb323cc | 1363 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1364 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1365 | unsigned long flags; |
1366 | ||
1367 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1368 | list_del(&req->inflight_entry); | |
1369 | if (waitqueue_active(&ctx->inflight_wait)) | |
1370 | wake_up(&ctx->inflight_wait); | |
1371 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1372 | } | |
2b85edfc PB |
1373 | |
1374 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1375 | if (likely(!io_is_fallback_req(req))) |
1376 | kmem_cache_free(req_cachep, req); | |
1377 | else | |
dd461af6 | 1378 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1379 | } |
1380 | ||
c6ca97b3 JA |
1381 | struct req_batch { |
1382 | void *reqs[IO_IOPOLL_BATCH]; | |
1383 | int to_free; | |
1384 | int need_iter; | |
1385 | }; | |
1386 | ||
1387 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1388 | { | |
1389 | if (!rb->to_free) | |
1390 | return; | |
1391 | if (rb->need_iter) { | |
1392 | int i, inflight = 0; | |
1393 | unsigned long flags; | |
1394 | ||
1395 | for (i = 0; i < rb->to_free; i++) { | |
1396 | struct io_kiocb *req = rb->reqs[i]; | |
1397 | ||
10fef4be | 1398 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1399 | req->file = NULL; |
05589553 | 1400 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1401 | } |
c6ca97b3 JA |
1402 | if (req->flags & REQ_F_INFLIGHT) |
1403 | inflight++; | |
c6ca97b3 JA |
1404 | __io_req_aux_free(req); |
1405 | } | |
1406 | if (!inflight) | |
1407 | goto do_free; | |
1408 | ||
1409 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1410 | for (i = 0; i < rb->to_free; i++) { | |
1411 | struct io_kiocb *req = rb->reqs[i]; | |
1412 | ||
10fef4be | 1413 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1414 | list_del(&req->inflight_entry); |
1415 | if (!--inflight) | |
1416 | break; | |
1417 | } | |
1418 | } | |
1419 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1420 | ||
1421 | if (waitqueue_active(&ctx->inflight_wait)) | |
1422 | wake_up(&ctx->inflight_wait); | |
1423 | } | |
1424 | do_free: | |
1425 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1426 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1427 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1428 | } |
1429 | ||
a197f664 | 1430 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1431 | { |
a197f664 | 1432 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1433 | int ret; |
1434 | ||
2d28390a | 1435 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1436 | if (ret != -1) { |
78e19bbe | 1437 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1438 | io_commit_cqring(ctx); |
dea3b49c | 1439 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1440 | io_put_req(req); |
2665abfd JA |
1441 | return true; |
1442 | } | |
1443 | ||
1444 | return false; | |
e65ef56d JA |
1445 | } |
1446 | ||
ba816ad6 | 1447 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1448 | { |
2665abfd | 1449 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1450 | bool wake_ev = false; |
9e645e11 | 1451 | |
4d7dd462 JA |
1452 | /* Already got next link */ |
1453 | if (req->flags & REQ_F_LINK_NEXT) | |
1454 | return; | |
1455 | ||
9e645e11 JA |
1456 | /* |
1457 | * The list should never be empty when we are called here. But could | |
1458 | * potentially happen if the chain is messed up, check to be on the | |
1459 | * safe side. | |
1460 | */ | |
4493233e PB |
1461 | while (!list_empty(&req->link_list)) { |
1462 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1463 | struct io_kiocb, link_list); | |
94ae5e77 | 1464 | |
4493233e PB |
1465 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1466 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1467 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1468 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1469 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1470 | continue; | |
1471 | } | |
9e645e11 | 1472 | |
4493233e PB |
1473 | list_del_init(&req->link_list); |
1474 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1475 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1476 | *nxtptr = nxt; |
94ae5e77 | 1477 | break; |
9e645e11 | 1478 | } |
2665abfd | 1479 | |
4d7dd462 | 1480 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1481 | if (wake_ev) |
1482 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1483 | } |
1484 | ||
1485 | /* | |
dea3b49c | 1486 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1487 | */ |
1488 | static void io_fail_links(struct io_kiocb *req) | |
1489 | { | |
2665abfd | 1490 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1491 | unsigned long flags; |
1492 | ||
1493 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1494 | |
1495 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1496 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1497 | struct io_kiocb, link_list); | |
9e645e11 | 1498 | |
4493233e | 1499 | list_del_init(&link->link_list); |
c826bd7a | 1500 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1501 | |
1502 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1503 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1504 | io_link_cancel_timeout(link); |
2665abfd | 1505 | } else { |
78e19bbe | 1506 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1507 | __io_double_put_req(link); |
2665abfd | 1508 | } |
5d960724 | 1509 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1510 | } |
2665abfd JA |
1511 | |
1512 | io_commit_cqring(ctx); | |
1513 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1514 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1515 | } |
1516 | ||
4d7dd462 | 1517 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1518 | { |
dea3b49c | 1519 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1520 | return; |
2665abfd | 1521 | |
9e645e11 JA |
1522 | /* |
1523 | * If LINK is set, we have dependent requests in this chain. If we | |
1524 | * didn't fail this request, queue the first one up, moving any other | |
1525 | * dependencies to the next request. In case of failure, fail the rest | |
1526 | * of the chain. | |
1527 | */ | |
2665abfd JA |
1528 | if (req->flags & REQ_F_FAIL_LINK) { |
1529 | io_fail_links(req); | |
7c9e7f0f JA |
1530 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1531 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1532 | struct io_ring_ctx *ctx = req->ctx; |
1533 | unsigned long flags; | |
1534 | ||
1535 | /* | |
1536 | * If this is a timeout link, we could be racing with the | |
1537 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1538 | * protect against that. |
2665abfd JA |
1539 | */ |
1540 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1541 | io_req_link_next(req, nxt); | |
1542 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1543 | } else { | |
1544 | io_req_link_next(req, nxt); | |
9e645e11 | 1545 | } |
4d7dd462 | 1546 | } |
9e645e11 | 1547 | |
c69f8dbe JL |
1548 | static void io_free_req(struct io_kiocb *req) |
1549 | { | |
944e58bf PB |
1550 | struct io_kiocb *nxt = NULL; |
1551 | ||
1552 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1553 | __io_free_req(req); |
944e58bf PB |
1554 | |
1555 | if (nxt) | |
1556 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1557 | } |
1558 | ||
7a743e22 PB |
1559 | static void io_link_work_cb(struct io_wq_work **workptr) |
1560 | { | |
18a542ff PB |
1561 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1562 | struct io_kiocb *link; | |
7a743e22 | 1563 | |
18a542ff | 1564 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1565 | io_queue_linked_timeout(link); |
1566 | io_wq_submit_work(workptr); | |
1567 | } | |
1568 | ||
1569 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1570 | { | |
1571 | struct io_kiocb *link; | |
8766dd51 PB |
1572 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1573 | ||
1574 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1575 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1576 | |
1577 | *workptr = &nxt->work; | |
1578 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1579 | if (link) |
7a743e22 | 1580 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1581 | } |
1582 | ||
ba816ad6 JA |
1583 | /* |
1584 | * Drop reference to request, return next in chain (if there is one) if this | |
1585 | * was the last reference to this request. | |
1586 | */ | |
f9bd67f6 | 1587 | __attribute__((nonnull)) |
ec9c02ad | 1588 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1589 | { |
2a44f467 JA |
1590 | if (refcount_dec_and_test(&req->refs)) { |
1591 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1592 | __io_free_req(req); |
2a44f467 | 1593 | } |
2b188cc1 JA |
1594 | } |
1595 | ||
e65ef56d JA |
1596 | static void io_put_req(struct io_kiocb *req) |
1597 | { | |
1598 | if (refcount_dec_and_test(&req->refs)) | |
1599 | io_free_req(req); | |
2b188cc1 JA |
1600 | } |
1601 | ||
e9fd9396 PB |
1602 | static void io_steal_work(struct io_kiocb *req, |
1603 | struct io_wq_work **workptr) | |
7a743e22 PB |
1604 | { |
1605 | /* | |
1606 | * It's in an io-wq worker, so there always should be at least | |
1607 | * one reference, which will be dropped in io_put_work() just | |
1608 | * after the current handler returns. | |
1609 | * | |
1610 | * It also means, that if the counter dropped to 1, then there is | |
1611 | * no asynchronous users left, so it's safe to steal the next work. | |
1612 | */ | |
7a743e22 PB |
1613 | if (refcount_read(&req->refs) == 1) { |
1614 | struct io_kiocb *nxt = NULL; | |
1615 | ||
1616 | io_req_find_next(req, &nxt); | |
1617 | if (nxt) | |
1618 | io_wq_assign_next(workptr, nxt); | |
1619 | } | |
1620 | } | |
1621 | ||
978db57e JA |
1622 | /* |
1623 | * Must only be used if we don't need to care about links, usually from | |
1624 | * within the completion handling itself. | |
1625 | */ | |
1626 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1627 | { |
1628 | /* drop both submit and complete references */ | |
1629 | if (refcount_sub_and_test(2, &req->refs)) | |
1630 | __io_free_req(req); | |
1631 | } | |
1632 | ||
978db57e JA |
1633 | static void io_double_put_req(struct io_kiocb *req) |
1634 | { | |
1635 | /* drop both submit and complete references */ | |
1636 | if (refcount_sub_and_test(2, &req->refs)) | |
1637 | io_free_req(req); | |
1638 | } | |
1639 | ||
1d7bb1d5 | 1640 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1641 | { |
84f97dc2 JA |
1642 | struct io_rings *rings = ctx->rings; |
1643 | ||
ad3eb2c8 JA |
1644 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1645 | /* | |
1646 | * noflush == true is from the waitqueue handler, just ensure | |
1647 | * we wake up the task, and the next invocation will flush the | |
1648 | * entries. We cannot safely to it from here. | |
1649 | */ | |
1650 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1651 | return -1U; | |
1d7bb1d5 | 1652 | |
ad3eb2c8 JA |
1653 | io_cqring_overflow_flush(ctx, false); |
1654 | } | |
1d7bb1d5 | 1655 | |
a3a0e43f JA |
1656 | /* See comment at the top of this file */ |
1657 | smp_rmb(); | |
ad3eb2c8 | 1658 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1659 | } |
1660 | ||
fb5ccc98 PB |
1661 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1662 | { | |
1663 | struct io_rings *rings = ctx->rings; | |
1664 | ||
1665 | /* make sure SQ entry isn't read before tail */ | |
1666 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1667 | } | |
1668 | ||
8237e045 | 1669 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1670 | { |
dea3b49c | 1671 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1672 | return false; |
e94f141b | 1673 | |
c6ca97b3 JA |
1674 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1675 | rb->need_iter++; | |
1676 | ||
1677 | rb->reqs[rb->to_free++] = req; | |
1678 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1679 | io_free_req_many(req->ctx, rb); | |
1680 | return true; | |
e94f141b JA |
1681 | } |
1682 | ||
bcda7baa JA |
1683 | static int io_put_kbuf(struct io_kiocb *req) |
1684 | { | |
4d954c25 | 1685 | struct io_buffer *kbuf; |
bcda7baa JA |
1686 | int cflags; |
1687 | ||
4d954c25 | 1688 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1689 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1690 | cflags |= IORING_CQE_F_BUFFER; | |
1691 | req->rw.addr = 0; | |
1692 | kfree(kbuf); | |
1693 | return cflags; | |
1694 | } | |
1695 | ||
def596e9 JA |
1696 | /* |
1697 | * Find and free completed poll iocbs | |
1698 | */ | |
1699 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1700 | struct list_head *done) | |
1701 | { | |
8237e045 | 1702 | struct req_batch rb; |
def596e9 | 1703 | struct io_kiocb *req; |
def596e9 | 1704 | |
c6ca97b3 | 1705 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1706 | while (!list_empty(done)) { |
bcda7baa JA |
1707 | int cflags = 0; |
1708 | ||
def596e9 JA |
1709 | req = list_first_entry(done, struct io_kiocb, list); |
1710 | list_del(&req->list); | |
1711 | ||
bcda7baa JA |
1712 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1713 | cflags = io_put_kbuf(req); | |
1714 | ||
1715 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1716 | (*nr_events)++; |
1717 | ||
8237e045 JA |
1718 | if (refcount_dec_and_test(&req->refs) && |
1719 | !io_req_multi_free(&rb, req)) | |
1720 | io_free_req(req); | |
def596e9 | 1721 | } |
def596e9 | 1722 | |
09bb8394 | 1723 | io_commit_cqring(ctx); |
32b2244a XW |
1724 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1725 | io_cqring_ev_posted(ctx); | |
8237e045 | 1726 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1727 | } |
1728 | ||
581f9810 BM |
1729 | static void io_iopoll_queue(struct list_head *again) |
1730 | { | |
1731 | struct io_kiocb *req; | |
1732 | ||
1733 | do { | |
1734 | req = list_first_entry(again, struct io_kiocb, list); | |
1735 | list_del(&req->list); | |
1736 | refcount_inc(&req->refs); | |
1737 | io_queue_async_work(req); | |
1738 | } while (!list_empty(again)); | |
1739 | } | |
1740 | ||
def596e9 JA |
1741 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1742 | long min) | |
1743 | { | |
1744 | struct io_kiocb *req, *tmp; | |
1745 | LIST_HEAD(done); | |
581f9810 | 1746 | LIST_HEAD(again); |
def596e9 JA |
1747 | bool spin; |
1748 | int ret; | |
1749 | ||
1750 | /* | |
1751 | * Only spin for completions if we don't have multiple devices hanging | |
1752 | * off our complete list, and we're under the requested amount. | |
1753 | */ | |
1754 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1755 | ||
1756 | ret = 0; | |
1757 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1758 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1759 | |
1760 | /* | |
581f9810 BM |
1761 | * Move completed and retryable entries to our local lists. |
1762 | * If we find a request that requires polling, break out | |
1763 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1764 | */ |
1765 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1766 | list_move_tail(&req->list, &done); | |
1767 | continue; | |
1768 | } | |
1769 | if (!list_empty(&done)) | |
1770 | break; | |
1771 | ||
581f9810 BM |
1772 | if (req->result == -EAGAIN) { |
1773 | list_move_tail(&req->list, &again); | |
1774 | continue; | |
1775 | } | |
1776 | if (!list_empty(&again)) | |
1777 | break; | |
1778 | ||
def596e9 JA |
1779 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1780 | if (ret < 0) | |
1781 | break; | |
1782 | ||
1783 | if (ret && spin) | |
1784 | spin = false; | |
1785 | ret = 0; | |
1786 | } | |
1787 | ||
1788 | if (!list_empty(&done)) | |
1789 | io_iopoll_complete(ctx, nr_events, &done); | |
1790 | ||
581f9810 BM |
1791 | if (!list_empty(&again)) |
1792 | io_iopoll_queue(&again); | |
1793 | ||
def596e9 JA |
1794 | return ret; |
1795 | } | |
1796 | ||
1797 | /* | |
d195a66e | 1798 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1799 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1800 | * as a non-spinning completion check. | |
1801 | */ | |
1802 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1803 | long min) | |
1804 | { | |
08f5439f | 1805 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1806 | int ret; |
1807 | ||
1808 | ret = io_do_iopoll(ctx, nr_events, min); | |
1809 | if (ret < 0) | |
1810 | return ret; | |
1811 | if (!min || *nr_events >= min) | |
1812 | return 0; | |
1813 | } | |
1814 | ||
1815 | return 1; | |
1816 | } | |
1817 | ||
1818 | /* | |
1819 | * We can't just wait for polled events to come to us, we have to actively | |
1820 | * find and complete them. | |
1821 | */ | |
1822 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1823 | { | |
1824 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1825 | return; | |
1826 | ||
1827 | mutex_lock(&ctx->uring_lock); | |
1828 | while (!list_empty(&ctx->poll_list)) { | |
1829 | unsigned int nr_events = 0; | |
1830 | ||
1831 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1832 | |
1833 | /* | |
1834 | * Ensure we allow local-to-the-cpu processing to take place, | |
1835 | * in this case we need to ensure that we reap all events. | |
1836 | */ | |
1837 | cond_resched(); | |
def596e9 JA |
1838 | } |
1839 | mutex_unlock(&ctx->uring_lock); | |
1840 | } | |
1841 | ||
c7849be9 XW |
1842 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1843 | long min) | |
def596e9 | 1844 | { |
2b2ed975 | 1845 | int iters = 0, ret = 0; |
500f9fba | 1846 | |
c7849be9 XW |
1847 | /* |
1848 | * We disallow the app entering submit/complete with polling, but we | |
1849 | * still need to lock the ring to prevent racing with polled issue | |
1850 | * that got punted to a workqueue. | |
1851 | */ | |
1852 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1853 | do { |
1854 | int tmin = 0; | |
1855 | ||
a3a0e43f JA |
1856 | /* |
1857 | * Don't enter poll loop if we already have events pending. | |
1858 | * If we do, we can potentially be spinning for commands that | |
1859 | * already triggered a CQE (eg in error). | |
1860 | */ | |
1d7bb1d5 | 1861 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1862 | break; |
1863 | ||
500f9fba JA |
1864 | /* |
1865 | * If a submit got punted to a workqueue, we can have the | |
1866 | * application entering polling for a command before it gets | |
1867 | * issued. That app will hold the uring_lock for the duration | |
1868 | * of the poll right here, so we need to take a breather every | |
1869 | * now and then to ensure that the issue has a chance to add | |
1870 | * the poll to the issued list. Otherwise we can spin here | |
1871 | * forever, while the workqueue is stuck trying to acquire the | |
1872 | * very same mutex. | |
1873 | */ | |
1874 | if (!(++iters & 7)) { | |
1875 | mutex_unlock(&ctx->uring_lock); | |
1876 | mutex_lock(&ctx->uring_lock); | |
1877 | } | |
1878 | ||
def596e9 JA |
1879 | if (*nr_events < min) |
1880 | tmin = min - *nr_events; | |
1881 | ||
1882 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1883 | if (ret <= 0) | |
1884 | break; | |
1885 | ret = 0; | |
1886 | } while (min && !*nr_events && !need_resched()); | |
1887 | ||
500f9fba | 1888 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1889 | return ret; |
1890 | } | |
1891 | ||
491381ce | 1892 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1893 | { |
491381ce JA |
1894 | /* |
1895 | * Tell lockdep we inherited freeze protection from submission | |
1896 | * thread. | |
1897 | */ | |
1898 | if (req->flags & REQ_F_ISREG) { | |
1899 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1900 | |
491381ce | 1901 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1902 | } |
491381ce | 1903 | file_end_write(req->file); |
2b188cc1 JA |
1904 | } |
1905 | ||
4e88d6e7 JA |
1906 | static inline void req_set_fail_links(struct io_kiocb *req) |
1907 | { | |
1908 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1909 | req->flags |= REQ_F_FAIL_LINK; | |
1910 | } | |
1911 | ||
ba816ad6 | 1912 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1913 | { |
9adbd45d | 1914 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1915 | int cflags = 0; |
2b188cc1 | 1916 | |
491381ce JA |
1917 | if (kiocb->ki_flags & IOCB_WRITE) |
1918 | kiocb_end_write(req); | |
2b188cc1 | 1919 | |
4e88d6e7 JA |
1920 | if (res != req->result) |
1921 | req_set_fail_links(req); | |
bcda7baa JA |
1922 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1923 | cflags = io_put_kbuf(req); | |
1924 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1925 | } |
1926 | ||
1927 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1928 | { | |
9adbd45d | 1929 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1930 | |
1931 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1932 | io_put_req(req); |
2b188cc1 JA |
1933 | } |
1934 | ||
def596e9 JA |
1935 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1936 | { | |
9adbd45d | 1937 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1938 | |
491381ce JA |
1939 | if (kiocb->ki_flags & IOCB_WRITE) |
1940 | kiocb_end_write(req); | |
def596e9 | 1941 | |
4e88d6e7 JA |
1942 | if (res != req->result) |
1943 | req_set_fail_links(req); | |
9e645e11 | 1944 | req->result = res; |
def596e9 JA |
1945 | if (res != -EAGAIN) |
1946 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1951 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1952 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1953 | * accessing the kiocb cookie. | |
1954 | */ | |
1955 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1956 | { | |
1957 | struct io_ring_ctx *ctx = req->ctx; | |
1958 | ||
1959 | /* | |
1960 | * Track whether we have multiple files in our lists. This will impact | |
1961 | * how we do polling eventually, not spinning if we're on potentially | |
1962 | * different devices. | |
1963 | */ | |
1964 | if (list_empty(&ctx->poll_list)) { | |
1965 | ctx->poll_multi_file = false; | |
1966 | } else if (!ctx->poll_multi_file) { | |
1967 | struct io_kiocb *list_req; | |
1968 | ||
1969 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1970 | list); | |
9adbd45d | 1971 | if (list_req->file != req->file) |
def596e9 JA |
1972 | ctx->poll_multi_file = true; |
1973 | } | |
1974 | ||
1975 | /* | |
1976 | * For fast devices, IO may have already completed. If it has, add | |
1977 | * it to the front so we find it first. | |
1978 | */ | |
1979 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1980 | list_add(&req->list, &ctx->poll_list); | |
1981 | else | |
1982 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1983 | |
1984 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1985 | wq_has_sleeper(&ctx->sqo_wait)) | |
1986 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1987 | } |
1988 | ||
3d6770fb | 1989 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1990 | { |
3d6770fb | 1991 | if (state->file) { |
9a56a232 JA |
1992 | int diff = state->has_refs - state->used_refs; |
1993 | ||
1994 | if (diff) | |
1995 | fput_many(state->file, diff); | |
1996 | state->file = NULL; | |
1997 | } | |
1998 | } | |
1999 | ||
2000 | /* | |
2001 | * Get as many references to a file as we have IOs left in this submission, | |
2002 | * assuming most submissions are for one file, or at least that each file | |
2003 | * has more than one submission. | |
2004 | */ | |
8da11c19 | 2005 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2006 | { |
2007 | if (!state) | |
2008 | return fget(fd); | |
2009 | ||
2010 | if (state->file) { | |
2011 | if (state->fd == fd) { | |
2012 | state->used_refs++; | |
2013 | state->ios_left--; | |
2014 | return state->file; | |
2015 | } | |
3d6770fb | 2016 | io_file_put(state); |
9a56a232 JA |
2017 | } |
2018 | state->file = fget_many(fd, state->ios_left); | |
2019 | if (!state->file) | |
2020 | return NULL; | |
2021 | ||
2022 | state->fd = fd; | |
2023 | state->has_refs = state->ios_left; | |
2024 | state->used_refs = 1; | |
2025 | state->ios_left--; | |
2026 | return state->file; | |
2027 | } | |
2028 | ||
2b188cc1 JA |
2029 | /* |
2030 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2031 | * any file. For now, just ensure that anything potentially problematic is done | |
2032 | * inline. | |
2033 | */ | |
af197f50 | 2034 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2035 | { |
2036 | umode_t mode = file_inode(file)->i_mode; | |
2037 | ||
10d59345 | 2038 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2039 | return true; |
2040 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2041 | return true; | |
2042 | ||
af197f50 JA |
2043 | if (!(file->f_mode & FMODE_NOWAIT)) |
2044 | return false; | |
2045 | ||
2046 | if (rw == READ) | |
2047 | return file->f_op->read_iter != NULL; | |
2048 | ||
2049 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2050 | } |
2051 | ||
3529d8c2 JA |
2052 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2053 | bool force_nonblock) | |
2b188cc1 | 2054 | { |
def596e9 | 2055 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2056 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2057 | unsigned ioprio; |
2058 | int ret; | |
2b188cc1 | 2059 | |
491381ce JA |
2060 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2061 | req->flags |= REQ_F_ISREG; | |
2062 | ||
2b188cc1 | 2063 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2064 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2065 | req->flags |= REQ_F_CUR_POS; | |
2066 | kiocb->ki_pos = req->file->f_pos; | |
2067 | } | |
2b188cc1 | 2068 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2069 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2070 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2071 | if (unlikely(ret)) | |
2072 | return ret; | |
2b188cc1 JA |
2073 | |
2074 | ioprio = READ_ONCE(sqe->ioprio); | |
2075 | if (ioprio) { | |
2076 | ret = ioprio_check_cap(ioprio); | |
2077 | if (ret) | |
09bb8394 | 2078 | return ret; |
2b188cc1 JA |
2079 | |
2080 | kiocb->ki_ioprio = ioprio; | |
2081 | } else | |
2082 | kiocb->ki_ioprio = get_current_ioprio(); | |
2083 | ||
8449eeda | 2084 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2085 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2086 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2087 | req->flags |= REQ_F_NOWAIT; |
2088 | ||
2089 | if (force_nonblock) | |
2b188cc1 | 2090 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2091 | |
def596e9 | 2092 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2093 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2094 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2095 | return -EOPNOTSUPP; |
2b188cc1 | 2096 | |
def596e9 JA |
2097 | kiocb->ki_flags |= IOCB_HIPRI; |
2098 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2099 | req->result = 0; |
def596e9 | 2100 | } else { |
09bb8394 JA |
2101 | if (kiocb->ki_flags & IOCB_HIPRI) |
2102 | return -EINVAL; | |
def596e9 JA |
2103 | kiocb->ki_complete = io_complete_rw; |
2104 | } | |
9adbd45d | 2105 | |
3529d8c2 JA |
2106 | req->rw.addr = READ_ONCE(sqe->addr); |
2107 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2108 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2109 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2110 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2111 | return 0; |
2b188cc1 JA |
2112 | } |
2113 | ||
2114 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2115 | { | |
2116 | switch (ret) { | |
2117 | case -EIOCBQUEUED: | |
2118 | break; | |
2119 | case -ERESTARTSYS: | |
2120 | case -ERESTARTNOINTR: | |
2121 | case -ERESTARTNOHAND: | |
2122 | case -ERESTART_RESTARTBLOCK: | |
2123 | /* | |
2124 | * We can't just restart the syscall, since previously | |
2125 | * submitted sqes may already be in progress. Just fail this | |
2126 | * IO with EINTR. | |
2127 | */ | |
2128 | ret = -EINTR; | |
2129 | /* fall through */ | |
2130 | default: | |
2131 | kiocb->ki_complete(kiocb, ret, 0); | |
2132 | } | |
2133 | } | |
2134 | ||
014db007 | 2135 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2136 | { |
ba04291e JA |
2137 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2138 | ||
2139 | if (req->flags & REQ_F_CUR_POS) | |
2140 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2141 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2142 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2143 | else |
2144 | io_rw_done(kiocb, ret); | |
2145 | } | |
2146 | ||
9adbd45d | 2147 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2148 | struct iov_iter *iter) |
edafccee | 2149 | { |
9adbd45d JA |
2150 | struct io_ring_ctx *ctx = req->ctx; |
2151 | size_t len = req->rw.len; | |
edafccee JA |
2152 | struct io_mapped_ubuf *imu; |
2153 | unsigned index, buf_index; | |
2154 | size_t offset; | |
2155 | u64 buf_addr; | |
2156 | ||
2157 | /* attempt to use fixed buffers without having provided iovecs */ | |
2158 | if (unlikely(!ctx->user_bufs)) | |
2159 | return -EFAULT; | |
2160 | ||
9adbd45d | 2161 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2162 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2163 | return -EFAULT; | |
2164 | ||
2165 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2166 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2167 | buf_addr = req->rw.addr; |
edafccee JA |
2168 | |
2169 | /* overflow */ | |
2170 | if (buf_addr + len < buf_addr) | |
2171 | return -EFAULT; | |
2172 | /* not inside the mapped region */ | |
2173 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2174 | return -EFAULT; | |
2175 | ||
2176 | /* | |
2177 | * May not be a start of buffer, set size appropriately | |
2178 | * and advance us to the beginning. | |
2179 | */ | |
2180 | offset = buf_addr - imu->ubuf; | |
2181 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2182 | |
2183 | if (offset) { | |
2184 | /* | |
2185 | * Don't use iov_iter_advance() here, as it's really slow for | |
2186 | * using the latter parts of a big fixed buffer - it iterates | |
2187 | * over each segment manually. We can cheat a bit here, because | |
2188 | * we know that: | |
2189 | * | |
2190 | * 1) it's a BVEC iter, we set it up | |
2191 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2192 | * first and last bvec | |
2193 | * | |
2194 | * So just find our index, and adjust the iterator afterwards. | |
2195 | * If the offset is within the first bvec (or the whole first | |
2196 | * bvec, just use iov_iter_advance(). This makes it easier | |
2197 | * since we can just skip the first segment, which may not | |
2198 | * be PAGE_SIZE aligned. | |
2199 | */ | |
2200 | const struct bio_vec *bvec = imu->bvec; | |
2201 | ||
2202 | if (offset <= bvec->bv_len) { | |
2203 | iov_iter_advance(iter, offset); | |
2204 | } else { | |
2205 | unsigned long seg_skip; | |
2206 | ||
2207 | /* skip first vec */ | |
2208 | offset -= bvec->bv_len; | |
2209 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2210 | ||
2211 | iter->bvec = bvec + seg_skip; | |
2212 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2213 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2214 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2215 | } |
2216 | } | |
2217 | ||
5e559561 | 2218 | return len; |
edafccee JA |
2219 | } |
2220 | ||
bcda7baa JA |
2221 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2222 | { | |
2223 | if (needs_lock) | |
2224 | mutex_unlock(&ctx->uring_lock); | |
2225 | } | |
2226 | ||
2227 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2228 | { | |
2229 | /* | |
2230 | * "Normal" inline submissions always hold the uring_lock, since we | |
2231 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2232 | * The only exception is when we've detached the request and issue it | |
2233 | * from an async worker thread, grab the lock for that case. | |
2234 | */ | |
2235 | if (needs_lock) | |
2236 | mutex_lock(&ctx->uring_lock); | |
2237 | } | |
2238 | ||
2239 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2240 | int bgid, struct io_buffer *kbuf, | |
2241 | bool needs_lock) | |
2242 | { | |
2243 | struct io_buffer *head; | |
2244 | ||
2245 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2246 | return kbuf; | |
2247 | ||
2248 | io_ring_submit_lock(req->ctx, needs_lock); | |
2249 | ||
2250 | lockdep_assert_held(&req->ctx->uring_lock); | |
2251 | ||
2252 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2253 | if (head) { | |
2254 | if (!list_empty(&head->list)) { | |
2255 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2256 | list); | |
2257 | list_del(&kbuf->list); | |
2258 | } else { | |
2259 | kbuf = head; | |
2260 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2261 | } | |
2262 | if (*len > kbuf->len) | |
2263 | *len = kbuf->len; | |
2264 | } else { | |
2265 | kbuf = ERR_PTR(-ENOBUFS); | |
2266 | } | |
2267 | ||
2268 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2269 | ||
2270 | return kbuf; | |
2271 | } | |
2272 | ||
4d954c25 JA |
2273 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2274 | bool needs_lock) | |
2275 | { | |
2276 | struct io_buffer *kbuf; | |
2277 | int bgid; | |
2278 | ||
2279 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2280 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2281 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2282 | if (IS_ERR(kbuf)) | |
2283 | return kbuf; | |
2284 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2285 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2286 | return u64_to_user_ptr(kbuf->addr); | |
2287 | } | |
2288 | ||
2289 | #ifdef CONFIG_COMPAT | |
2290 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2291 | bool needs_lock) | |
2292 | { | |
2293 | struct compat_iovec __user *uiov; | |
2294 | compat_ssize_t clen; | |
2295 | void __user *buf; | |
2296 | ssize_t len; | |
2297 | ||
2298 | uiov = u64_to_user_ptr(req->rw.addr); | |
2299 | if (!access_ok(uiov, sizeof(*uiov))) | |
2300 | return -EFAULT; | |
2301 | if (__get_user(clen, &uiov->iov_len)) | |
2302 | return -EFAULT; | |
2303 | if (clen < 0) | |
2304 | return -EINVAL; | |
2305 | ||
2306 | len = clen; | |
2307 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2308 | if (IS_ERR(buf)) | |
2309 | return PTR_ERR(buf); | |
2310 | iov[0].iov_base = buf; | |
2311 | iov[0].iov_len = (compat_size_t) len; | |
2312 | return 0; | |
2313 | } | |
2314 | #endif | |
2315 | ||
2316 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2317 | bool needs_lock) | |
2318 | { | |
2319 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2320 | void __user *buf; | |
2321 | ssize_t len; | |
2322 | ||
2323 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2324 | return -EFAULT; | |
2325 | ||
2326 | len = iov[0].iov_len; | |
2327 | if (len < 0) | |
2328 | return -EINVAL; | |
2329 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2330 | if (IS_ERR(buf)) | |
2331 | return PTR_ERR(buf); | |
2332 | iov[0].iov_base = buf; | |
2333 | iov[0].iov_len = len; | |
2334 | return 0; | |
2335 | } | |
2336 | ||
2337 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2338 | bool needs_lock) | |
2339 | { | |
2340 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2341 | return 0; | |
2342 | if (!req->rw.len) | |
2343 | return 0; | |
2344 | else if (req->rw.len > 1) | |
2345 | return -EINVAL; | |
2346 | ||
2347 | #ifdef CONFIG_COMPAT | |
2348 | if (req->ctx->compat) | |
2349 | return io_compat_import(req, iov, needs_lock); | |
2350 | #endif | |
2351 | ||
2352 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2353 | } | |
2354 | ||
cf6fd4bd | 2355 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2356 | struct iovec **iovec, struct iov_iter *iter, |
2357 | bool needs_lock) | |
2b188cc1 | 2358 | { |
9adbd45d JA |
2359 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2360 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2361 | ssize_t ret; |
edafccee JA |
2362 | u8 opcode; |
2363 | ||
d625c6ee | 2364 | opcode = req->opcode; |
7d009165 | 2365 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2366 | *iovec = NULL; |
9adbd45d | 2367 | return io_import_fixed(req, rw, iter); |
edafccee | 2368 | } |
2b188cc1 | 2369 | |
bcda7baa JA |
2370 | /* buffer index only valid with fixed read/write, or buffer select */ |
2371 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2372 | return -EINVAL; |
2373 | ||
3a6820f2 | 2374 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2375 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2376 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2377 | if (IS_ERR(buf)) { | |
bcda7baa | 2378 | *iovec = NULL; |
4d954c25 | 2379 | return PTR_ERR(buf); |
bcda7baa | 2380 | } |
3f9d6441 | 2381 | req->rw.len = sqe_len; |
bcda7baa JA |
2382 | } |
2383 | ||
3a6820f2 JA |
2384 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2385 | *iovec = NULL; | |
3a901598 | 2386 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2387 | } |
2388 | ||
f67676d1 JA |
2389 | if (req->io) { |
2390 | struct io_async_rw *iorw = &req->io->rw; | |
2391 | ||
2392 | *iovec = iorw->iov; | |
2393 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2394 | if (iorw->iov == iorw->fast_iov) | |
2395 | *iovec = NULL; | |
2396 | return iorw->size; | |
2397 | } | |
2398 | ||
4d954c25 JA |
2399 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2400 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2401 | if (!ret) { |
2402 | ret = (*iovec)->iov_len; | |
2403 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2404 | } | |
4d954c25 JA |
2405 | *iovec = NULL; |
2406 | return ret; | |
2407 | } | |
2408 | ||
2b188cc1 | 2409 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2410 | if (req->ctx->compat) |
2b188cc1 JA |
2411 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2412 | iovec, iter); | |
2413 | #endif | |
2414 | ||
2415 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2416 | } | |
2417 | ||
31b51510 | 2418 | /* |
32960613 JA |
2419 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2420 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2421 | */ |
32960613 JA |
2422 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2423 | struct iov_iter *iter) | |
2424 | { | |
2425 | ssize_t ret = 0; | |
2426 | ||
2427 | /* | |
2428 | * Don't support polled IO through this interface, and we can't | |
2429 | * support non-blocking either. For the latter, this just causes | |
2430 | * the kiocb to be handled from an async context. | |
2431 | */ | |
2432 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2433 | return -EOPNOTSUPP; | |
2434 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2435 | return -EAGAIN; | |
2436 | ||
2437 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2438 | struct iovec iovec; |
32960613 JA |
2439 | ssize_t nr; |
2440 | ||
311ae9e1 PB |
2441 | if (!iov_iter_is_bvec(iter)) { |
2442 | iovec = iov_iter_iovec(iter); | |
2443 | } else { | |
2444 | /* fixed buffers import bvec */ | |
2445 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2446 | + iter->iov_offset; | |
2447 | iovec.iov_len = min(iter->count, | |
2448 | iter->bvec->bv_len - iter->iov_offset); | |
2449 | } | |
2450 | ||
32960613 JA |
2451 | if (rw == READ) { |
2452 | nr = file->f_op->read(file, iovec.iov_base, | |
2453 | iovec.iov_len, &kiocb->ki_pos); | |
2454 | } else { | |
2455 | nr = file->f_op->write(file, iovec.iov_base, | |
2456 | iovec.iov_len, &kiocb->ki_pos); | |
2457 | } | |
2458 | ||
311ae9e1 PB |
2459 | if (iov_iter_is_bvec(iter)) |
2460 | kunmap(iter->bvec->bv_page); | |
2461 | ||
32960613 JA |
2462 | if (nr < 0) { |
2463 | if (!ret) | |
2464 | ret = nr; | |
2465 | break; | |
2466 | } | |
2467 | ret += nr; | |
2468 | if (nr != iovec.iov_len) | |
2469 | break; | |
2470 | iov_iter_advance(iter, nr); | |
2471 | } | |
2472 | ||
2473 | return ret; | |
2474 | } | |
2475 | ||
b7bb4f7d | 2476 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2477 | struct iovec *iovec, struct iovec *fast_iov, |
2478 | struct iov_iter *iter) | |
2479 | { | |
2480 | req->io->rw.nr_segs = iter->nr_segs; | |
2481 | req->io->rw.size = io_size; | |
2482 | req->io->rw.iov = iovec; | |
2483 | if (!req->io->rw.iov) { | |
2484 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2485 | if (req->io->rw.iov != fast_iov) |
2486 | memcpy(req->io->rw.iov, fast_iov, | |
2487 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2488 | } else { |
2489 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2490 | } |
2491 | } | |
2492 | ||
3d9932a8 XW |
2493 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2494 | { | |
2495 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2496 | return req->io == NULL; | |
2497 | } | |
2498 | ||
b7bb4f7d | 2499 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2500 | { |
d3656344 JA |
2501 | if (!io_op_defs[req->opcode].async_ctx) |
2502 | return 0; | |
3d9932a8 XW |
2503 | |
2504 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2505 | } |
2506 | ||
b7bb4f7d JA |
2507 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2508 | struct iovec *iovec, struct iovec *fast_iov, | |
2509 | struct iov_iter *iter) | |
2510 | { | |
980ad263 | 2511 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2512 | return 0; |
5d204bcf | 2513 | if (!req->io) { |
3d9932a8 | 2514 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2515 | return -ENOMEM; |
b7bb4f7d | 2516 | |
5d204bcf JA |
2517 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2518 | } | |
b7bb4f7d | 2519 | return 0; |
f67676d1 JA |
2520 | } |
2521 | ||
3529d8c2 JA |
2522 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2523 | bool force_nonblock) | |
f67676d1 | 2524 | { |
3529d8c2 JA |
2525 | struct io_async_ctx *io; |
2526 | struct iov_iter iter; | |
f67676d1 JA |
2527 | ssize_t ret; |
2528 | ||
3529d8c2 JA |
2529 | ret = io_prep_rw(req, sqe, force_nonblock); |
2530 | if (ret) | |
2531 | return ret; | |
f67676d1 | 2532 | |
3529d8c2 JA |
2533 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2534 | return -EBADF; | |
f67676d1 | 2535 | |
5f798bea PB |
2536 | /* either don't need iovec imported or already have it */ |
2537 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2538 | return 0; |
2539 | ||
2540 | io = req->io; | |
2541 | io->rw.iov = io->rw.fast_iov; | |
2542 | req->io = NULL; | |
bcda7baa | 2543 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2544 | req->io = io; |
2545 | if (ret < 0) | |
2546 | return ret; | |
2547 | ||
2548 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2549 | return 0; | |
f67676d1 JA |
2550 | } |
2551 | ||
014db007 | 2552 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2553 | { |
2554 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2555 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2556 | struct iov_iter iter; |
31b51510 | 2557 | size_t iov_count; |
f67676d1 | 2558 | ssize_t io_size, ret; |
2b188cc1 | 2559 | |
bcda7baa | 2560 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2561 | if (ret < 0) |
2562 | return ret; | |
2b188cc1 | 2563 | |
fd6c2e4c JA |
2564 | /* Ensure we clear previously set non-block flag */ |
2565 | if (!force_nonblock) | |
29de5f6a | 2566 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2567 | |
797f3f53 | 2568 | req->result = 0; |
f67676d1 | 2569 | io_size = ret; |
dea3b49c | 2570 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2571 | req->result = io_size; |
2572 | ||
2573 | /* | |
2574 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2575 | * we know to async punt it even if it was opened O_NONBLOCK | |
2576 | */ | |
af197f50 | 2577 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2578 | goto copy_iov; |
9e645e11 | 2579 | |
31b51510 | 2580 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2581 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2582 | if (!ret) { |
2583 | ssize_t ret2; | |
2584 | ||
9adbd45d JA |
2585 | if (req->file->f_op->read_iter) |
2586 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2587 | else |
9adbd45d | 2588 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2589 | |
9d93a3f5 | 2590 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2591 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2592 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2593 | } else { |
2594 | copy_iov: | |
b7bb4f7d | 2595 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2596 | inline_vecs, &iter); |
2597 | if (ret) | |
2598 | goto out_free; | |
29de5f6a | 2599 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2600 | if (!(req->flags & REQ_F_NOWAIT) && |
2601 | !file_can_poll(req->file)) | |
29de5f6a | 2602 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2603 | return -EAGAIN; |
2604 | } | |
2b188cc1 | 2605 | } |
f67676d1 | 2606 | out_free: |
1e95081c | 2607 | kfree(iovec); |
99bc4c38 | 2608 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2609 | return ret; |
2610 | } | |
2611 | ||
3529d8c2 JA |
2612 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2613 | bool force_nonblock) | |
f67676d1 | 2614 | { |
3529d8c2 JA |
2615 | struct io_async_ctx *io; |
2616 | struct iov_iter iter; | |
f67676d1 JA |
2617 | ssize_t ret; |
2618 | ||
3529d8c2 JA |
2619 | ret = io_prep_rw(req, sqe, force_nonblock); |
2620 | if (ret) | |
2621 | return ret; | |
f67676d1 | 2622 | |
3529d8c2 JA |
2623 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2624 | return -EBADF; | |
f67676d1 | 2625 | |
4ed734b0 JA |
2626 | req->fsize = rlimit(RLIMIT_FSIZE); |
2627 | ||
5f798bea PB |
2628 | /* either don't need iovec imported or already have it */ |
2629 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2630 | return 0; |
2631 | ||
2632 | io = req->io; | |
2633 | io->rw.iov = io->rw.fast_iov; | |
2634 | req->io = NULL; | |
bcda7baa | 2635 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2636 | req->io = io; |
2637 | if (ret < 0) | |
2638 | return ret; | |
2639 | ||
2640 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2641 | return 0; | |
f67676d1 JA |
2642 | } |
2643 | ||
014db007 | 2644 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2645 | { |
2646 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2647 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2648 | struct iov_iter iter; |
31b51510 | 2649 | size_t iov_count; |
f67676d1 | 2650 | ssize_t ret, io_size; |
2b188cc1 | 2651 | |
bcda7baa | 2652 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2653 | if (ret < 0) |
2654 | return ret; | |
2b188cc1 | 2655 | |
fd6c2e4c JA |
2656 | /* Ensure we clear previously set non-block flag */ |
2657 | if (!force_nonblock) | |
9adbd45d | 2658 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2659 | |
797f3f53 | 2660 | req->result = 0; |
f67676d1 | 2661 | io_size = ret; |
dea3b49c | 2662 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2663 | req->result = io_size; |
9e645e11 | 2664 | |
f67676d1 JA |
2665 | /* |
2666 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2667 | * we know to async punt it even if it was opened O_NONBLOCK | |
2668 | */ | |
af197f50 | 2669 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2670 | goto copy_iov; |
31b51510 | 2671 | |
10d59345 JA |
2672 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2673 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2674 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2675 | goto copy_iov; |
31b51510 | 2676 | |
f67676d1 | 2677 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2678 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2679 | if (!ret) { |
9bf7933f RP |
2680 | ssize_t ret2; |
2681 | ||
2b188cc1 JA |
2682 | /* |
2683 | * Open-code file_start_write here to grab freeze protection, | |
2684 | * which will be released by another thread in | |
2685 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2686 | * released so that it doesn't complain about the held lock when | |
2687 | * we return to userspace. | |
2688 | */ | |
491381ce | 2689 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2690 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2691 | SB_FREEZE_WRITE, true); |
9adbd45d | 2692 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2693 | SB_FREEZE_WRITE); |
2694 | } | |
2695 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2696 | |
4ed734b0 JA |
2697 | if (!force_nonblock) |
2698 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2699 | ||
9adbd45d JA |
2700 | if (req->file->f_op->write_iter) |
2701 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2702 | else |
9adbd45d | 2703 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2704 | |
2705 | if (!force_nonblock) | |
2706 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2707 | ||
faac996c | 2708 | /* |
bff6035d | 2709 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2710 | * retry them without IOCB_NOWAIT. |
2711 | */ | |
2712 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2713 | ret2 = -EAGAIN; | |
f67676d1 | 2714 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2715 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2716 | } else { |
2717 | copy_iov: | |
b7bb4f7d | 2718 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2719 | inline_vecs, &iter); |
2720 | if (ret) | |
2721 | goto out_free; | |
29de5f6a | 2722 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2723 | if (!file_can_poll(req->file)) |
2724 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2725 | return -EAGAIN; |
2726 | } | |
2b188cc1 | 2727 | } |
31b51510 | 2728 | out_free: |
99bc4c38 | 2729 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2730 | kfree(iovec); |
2b188cc1 JA |
2731 | return ret; |
2732 | } | |
2733 | ||
7d67af2c PB |
2734 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2735 | { | |
2736 | struct io_splice* sp = &req->splice; | |
2737 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2738 | int ret; | |
2739 | ||
2740 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2741 | return 0; | |
2742 | ||
2743 | sp->file_in = NULL; | |
2744 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2745 | sp->off_out = READ_ONCE(sqe->off); | |
2746 | sp->len = READ_ONCE(sqe->len); | |
2747 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2748 | ||
2749 | if (unlikely(sp->flags & ~valid_flags)) | |
2750 | return -EINVAL; | |
2751 | ||
2752 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2753 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2754 | if (ret) | |
2755 | return ret; | |
2756 | req->flags |= REQ_F_NEED_CLEANUP; | |
2757 | ||
2758 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2759 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2760 | ||
2761 | return 0; | |
2762 | } | |
2763 | ||
014db007 | 2764 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2765 | { |
2766 | struct io_splice *sp = &req->splice; | |
2767 | struct file *in = sp->file_in; | |
2768 | struct file *out = sp->file_out; | |
2769 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2770 | loff_t *poff_in, *poff_out; | |
2771 | long ret; | |
2772 | ||
2fb3e822 PB |
2773 | if (force_nonblock) |
2774 | return -EAGAIN; | |
7d67af2c PB |
2775 | |
2776 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2777 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2778 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2779 | if (force_nonblock && ret == -EAGAIN) | |
2780 | return -EAGAIN; | |
2781 | ||
2782 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2783 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2784 | ||
2785 | io_cqring_add_event(req, ret); | |
2786 | if (ret != sp->len) | |
2787 | req_set_fail_links(req); | |
014db007 | 2788 | io_put_req(req); |
7d67af2c PB |
2789 | return 0; |
2790 | } | |
2791 | ||
2b188cc1 JA |
2792 | /* |
2793 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2794 | */ | |
78e19bbe | 2795 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2796 | { |
2797 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2798 | |
def596e9 JA |
2799 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2800 | return -EINVAL; | |
2801 | ||
78e19bbe | 2802 | io_cqring_add_event(req, 0); |
e65ef56d | 2803 | io_put_req(req); |
2b188cc1 JA |
2804 | return 0; |
2805 | } | |
2806 | ||
3529d8c2 | 2807 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2808 | { |
6b06314c | 2809 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2810 | |
09bb8394 JA |
2811 | if (!req->file) |
2812 | return -EBADF; | |
c992fe29 | 2813 | |
6b06314c | 2814 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2815 | return -EINVAL; |
edafccee | 2816 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2817 | return -EINVAL; |
2818 | ||
8ed8d3c3 JA |
2819 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2820 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2821 | return -EINVAL; | |
2822 | ||
2823 | req->sync.off = READ_ONCE(sqe->off); | |
2824 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2825 | return 0; |
2826 | } | |
2827 | ||
8ed8d3c3 JA |
2828 | static bool io_req_cancelled(struct io_kiocb *req) |
2829 | { | |
2830 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2831 | req_set_fail_links(req); | |
2832 | io_cqring_add_event(req, -ECANCELED); | |
2833 | io_put_req(req); | |
2834 | return true; | |
2835 | } | |
2836 | ||
2837 | return false; | |
2838 | } | |
2839 | ||
014db007 | 2840 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2841 | { |
8ed8d3c3 | 2842 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2843 | int ret; |
2844 | ||
9adbd45d | 2845 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2846 | end > 0 ? end : LLONG_MAX, |
2847 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2848 | if (ret < 0) | |
2849 | req_set_fail_links(req); | |
2850 | io_cqring_add_event(req, ret); | |
014db007 | 2851 | io_put_req(req); |
8ed8d3c3 JA |
2852 | } |
2853 | ||
5ea62161 | 2854 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2855 | { |
5ea62161 | 2856 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2857 | |
2858 | if (io_req_cancelled(req)) | |
2859 | return; | |
014db007 | 2860 | __io_fsync(req); |
e9fd9396 | 2861 | io_steal_work(req, workptr); |
8ed8d3c3 | 2862 | } |
c992fe29 | 2863 | |
014db007 | 2864 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2865 | { |
c992fe29 | 2866 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2867 | if (force_nonblock) { |
8ed8d3c3 | 2868 | req->work.func = io_fsync_finish; |
c992fe29 | 2869 | return -EAGAIN; |
8ed8d3c3 | 2870 | } |
014db007 | 2871 | __io_fsync(req); |
c992fe29 CH |
2872 | return 0; |
2873 | } | |
2874 | ||
014db007 | 2875 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2876 | { |
8ed8d3c3 JA |
2877 | int ret; |
2878 | ||
4ed734b0 | 2879 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2880 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2881 | req->sync.len); | |
4ed734b0 | 2882 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2883 | if (ret < 0) |
2884 | req_set_fail_links(req); | |
2885 | io_cqring_add_event(req, ret); | |
014db007 | 2886 | io_put_req(req); |
5ea62161 PB |
2887 | } |
2888 | ||
2889 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2890 | { | |
2891 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2892 | |
594506fe PB |
2893 | if (io_req_cancelled(req)) |
2894 | return; | |
014db007 | 2895 | __io_fallocate(req); |
e9fd9396 | 2896 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2897 | } |
2898 | ||
d63d1b5e JA |
2899 | static int io_fallocate_prep(struct io_kiocb *req, |
2900 | const struct io_uring_sqe *sqe) | |
2901 | { | |
2902 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2903 | return -EINVAL; | |
2904 | ||
2905 | req->sync.off = READ_ONCE(sqe->off); | |
2906 | req->sync.len = READ_ONCE(sqe->addr); | |
2907 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2908 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2909 | return 0; |
2910 | } | |
2911 | ||
014db007 | 2912 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2913 | { |
d63d1b5e | 2914 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2915 | if (force_nonblock) { |
d63d1b5e | 2916 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2917 | return -EAGAIN; |
8ed8d3c3 | 2918 | } |
5d17b4a4 | 2919 | |
014db007 | 2920 | __io_fallocate(req); |
5d17b4a4 JA |
2921 | return 0; |
2922 | } | |
2923 | ||
15b71abe | 2924 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2925 | { |
f8748881 | 2926 | const char __user *fname; |
15b71abe | 2927 | int ret; |
b7bb4f7d | 2928 | |
15b71abe JA |
2929 | if (sqe->ioprio || sqe->buf_index) |
2930 | return -EINVAL; | |
9c280f90 | 2931 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2932 | return -EBADF; |
0bdbdd08 PB |
2933 | if (req->flags & REQ_F_NEED_CLEANUP) |
2934 | return 0; | |
03b1230c | 2935 | |
15b71abe | 2936 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2937 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2938 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2939 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2940 | if (force_o_largefile()) |
2941 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2942 | |
f8748881 | 2943 | req->open.filename = getname(fname); |
15b71abe JA |
2944 | if (IS_ERR(req->open.filename)) { |
2945 | ret = PTR_ERR(req->open.filename); | |
2946 | req->open.filename = NULL; | |
2947 | return ret; | |
2948 | } | |
3529d8c2 | 2949 | |
4022e7af | 2950 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2951 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2952 | return 0; |
03b1230c JA |
2953 | } |
2954 | ||
cebdb986 | 2955 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2956 | { |
cebdb986 JA |
2957 | struct open_how __user *how; |
2958 | const char __user *fname; | |
2959 | size_t len; | |
0fa03c62 JA |
2960 | int ret; |
2961 | ||
cebdb986 | 2962 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2963 | return -EINVAL; |
9c280f90 | 2964 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2965 | return -EBADF; |
0bdbdd08 PB |
2966 | if (req->flags & REQ_F_NEED_CLEANUP) |
2967 | return 0; | |
0fa03c62 | 2968 | |
cebdb986 JA |
2969 | req->open.dfd = READ_ONCE(sqe->fd); |
2970 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2971 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2972 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2973 | |
cebdb986 JA |
2974 | if (len < OPEN_HOW_SIZE_VER0) |
2975 | return -EINVAL; | |
3529d8c2 | 2976 | |
cebdb986 JA |
2977 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2978 | len); | |
2979 | if (ret) | |
2980 | return ret; | |
3529d8c2 | 2981 | |
cebdb986 JA |
2982 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2983 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2984 | |
cebdb986 JA |
2985 | req->open.filename = getname(fname); |
2986 | if (IS_ERR(req->open.filename)) { | |
2987 | ret = PTR_ERR(req->open.filename); | |
2988 | req->open.filename = NULL; | |
2989 | return ret; | |
2990 | } | |
2991 | ||
4022e7af | 2992 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2993 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2994 | return 0; |
2995 | } | |
2996 | ||
014db007 | 2997 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2998 | { |
2999 | struct open_flags op; | |
15b71abe JA |
3000 | struct file *file; |
3001 | int ret; | |
3002 | ||
f86cd20c | 3003 | if (force_nonblock) |
15b71abe | 3004 | return -EAGAIN; |
15b71abe | 3005 | |
cebdb986 | 3006 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3007 | if (ret) |
3008 | goto err; | |
3009 | ||
4022e7af | 3010 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3011 | if (ret < 0) |
3012 | goto err; | |
3013 | ||
3014 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3015 | if (IS_ERR(file)) { | |
3016 | put_unused_fd(ret); | |
3017 | ret = PTR_ERR(file); | |
3018 | } else { | |
3019 | fsnotify_open(file); | |
3020 | fd_install(ret, file); | |
3021 | } | |
3022 | err: | |
3023 | putname(req->open.filename); | |
8fef80bf | 3024 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3025 | if (ret < 0) |
3026 | req_set_fail_links(req); | |
3027 | io_cqring_add_event(req, ret); | |
014db007 | 3028 | io_put_req(req); |
15b71abe JA |
3029 | return 0; |
3030 | } | |
3031 | ||
014db007 | 3032 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3033 | { |
3034 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3035 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3036 | } |
3037 | ||
067524e9 JA |
3038 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3039 | const struct io_uring_sqe *sqe) | |
3040 | { | |
3041 | struct io_provide_buf *p = &req->pbuf; | |
3042 | u64 tmp; | |
3043 | ||
3044 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3045 | return -EINVAL; | |
3046 | ||
3047 | tmp = READ_ONCE(sqe->fd); | |
3048 | if (!tmp || tmp > USHRT_MAX) | |
3049 | return -EINVAL; | |
3050 | ||
3051 | memset(p, 0, sizeof(*p)); | |
3052 | p->nbufs = tmp; | |
3053 | p->bgid = READ_ONCE(sqe->buf_group); | |
3054 | return 0; | |
3055 | } | |
3056 | ||
3057 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3058 | int bgid, unsigned nbufs) | |
3059 | { | |
3060 | unsigned i = 0; | |
3061 | ||
3062 | /* shouldn't happen */ | |
3063 | if (!nbufs) | |
3064 | return 0; | |
3065 | ||
3066 | /* the head kbuf is the list itself */ | |
3067 | while (!list_empty(&buf->list)) { | |
3068 | struct io_buffer *nxt; | |
3069 | ||
3070 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3071 | list_del(&nxt->list); | |
3072 | kfree(nxt); | |
3073 | if (++i == nbufs) | |
3074 | return i; | |
3075 | } | |
3076 | i++; | |
3077 | kfree(buf); | |
3078 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3079 | ||
3080 | return i; | |
3081 | } | |
3082 | ||
3083 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3084 | { | |
3085 | struct io_provide_buf *p = &req->pbuf; | |
3086 | struct io_ring_ctx *ctx = req->ctx; | |
3087 | struct io_buffer *head; | |
3088 | int ret = 0; | |
3089 | ||
3090 | io_ring_submit_lock(ctx, !force_nonblock); | |
3091 | ||
3092 | lockdep_assert_held(&ctx->uring_lock); | |
3093 | ||
3094 | ret = -ENOENT; | |
3095 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3096 | if (head) | |
3097 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3098 | ||
3099 | io_ring_submit_lock(ctx, !force_nonblock); | |
3100 | if (ret < 0) | |
3101 | req_set_fail_links(req); | |
3102 | io_cqring_add_event(req, ret); | |
3103 | io_put_req(req); | |
3104 | return 0; | |
3105 | } | |
3106 | ||
ddf0322d JA |
3107 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3108 | const struct io_uring_sqe *sqe) | |
3109 | { | |
3110 | struct io_provide_buf *p = &req->pbuf; | |
3111 | u64 tmp; | |
3112 | ||
3113 | if (sqe->ioprio || sqe->rw_flags) | |
3114 | return -EINVAL; | |
3115 | ||
3116 | tmp = READ_ONCE(sqe->fd); | |
3117 | if (!tmp || tmp > USHRT_MAX) | |
3118 | return -E2BIG; | |
3119 | p->nbufs = tmp; | |
3120 | p->addr = READ_ONCE(sqe->addr); | |
3121 | p->len = READ_ONCE(sqe->len); | |
3122 | ||
3123 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3124 | return -EFAULT; | |
3125 | ||
3126 | p->bgid = READ_ONCE(sqe->buf_group); | |
3127 | tmp = READ_ONCE(sqe->off); | |
3128 | if (tmp > USHRT_MAX) | |
3129 | return -E2BIG; | |
3130 | p->bid = tmp; | |
3131 | return 0; | |
3132 | } | |
3133 | ||
3134 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3135 | { | |
3136 | struct io_buffer *buf; | |
3137 | u64 addr = pbuf->addr; | |
3138 | int i, bid = pbuf->bid; | |
3139 | ||
3140 | for (i = 0; i < pbuf->nbufs; i++) { | |
3141 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3142 | if (!buf) | |
3143 | break; | |
3144 | ||
3145 | buf->addr = addr; | |
3146 | buf->len = pbuf->len; | |
3147 | buf->bid = bid; | |
3148 | addr += pbuf->len; | |
3149 | bid++; | |
3150 | if (!*head) { | |
3151 | INIT_LIST_HEAD(&buf->list); | |
3152 | *head = buf; | |
3153 | } else { | |
3154 | list_add_tail(&buf->list, &(*head)->list); | |
3155 | } | |
3156 | } | |
3157 | ||
3158 | return i ? i : -ENOMEM; | |
3159 | } | |
3160 | ||
ddf0322d JA |
3161 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3162 | { | |
3163 | struct io_provide_buf *p = &req->pbuf; | |
3164 | struct io_ring_ctx *ctx = req->ctx; | |
3165 | struct io_buffer *head, *list; | |
3166 | int ret = 0; | |
3167 | ||
3168 | io_ring_submit_lock(ctx, !force_nonblock); | |
3169 | ||
3170 | lockdep_assert_held(&ctx->uring_lock); | |
3171 | ||
3172 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3173 | ||
3174 | ret = io_add_buffers(p, &head); | |
3175 | if (ret < 0) | |
3176 | goto out; | |
3177 | ||
3178 | if (!list) { | |
3179 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3180 | GFP_KERNEL); | |
3181 | if (ret < 0) { | |
067524e9 | 3182 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3183 | goto out; |
3184 | } | |
3185 | } | |
3186 | out: | |
3187 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3188 | if (ret < 0) | |
3189 | req_set_fail_links(req); | |
3190 | io_cqring_add_event(req, ret); | |
3191 | io_put_req(req); | |
3192 | return 0; | |
cebdb986 JA |
3193 | } |
3194 | ||
3e4827b0 JA |
3195 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3196 | const struct io_uring_sqe *sqe) | |
3197 | { | |
3198 | #if defined(CONFIG_EPOLL) | |
3199 | if (sqe->ioprio || sqe->buf_index) | |
3200 | return -EINVAL; | |
3201 | ||
3202 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3203 | req->epoll.op = READ_ONCE(sqe->len); | |
3204 | req->epoll.fd = READ_ONCE(sqe->off); | |
3205 | ||
3206 | if (ep_op_has_event(req->epoll.op)) { | |
3207 | struct epoll_event __user *ev; | |
3208 | ||
3209 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3210 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3211 | return -EFAULT; | |
3212 | } | |
3213 | ||
3214 | return 0; | |
3215 | #else | |
3216 | return -EOPNOTSUPP; | |
3217 | #endif | |
3218 | } | |
3219 | ||
014db007 | 3220 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3221 | { |
3222 | #if defined(CONFIG_EPOLL) | |
3223 | struct io_epoll *ie = &req->epoll; | |
3224 | int ret; | |
3225 | ||
3226 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3227 | if (force_nonblock && ret == -EAGAIN) | |
3228 | return -EAGAIN; | |
3229 | ||
3230 | if (ret < 0) | |
3231 | req_set_fail_links(req); | |
3232 | io_cqring_add_event(req, ret); | |
014db007 | 3233 | io_put_req(req); |
3e4827b0 JA |
3234 | return 0; |
3235 | #else | |
3236 | return -EOPNOTSUPP; | |
3237 | #endif | |
3238 | } | |
3239 | ||
c1ca757b JA |
3240 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3241 | { | |
3242 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3243 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3244 | return -EINVAL; | |
3245 | ||
3246 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3247 | req->madvise.len = READ_ONCE(sqe->len); | |
3248 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3249 | return 0; | |
3250 | #else | |
3251 | return -EOPNOTSUPP; | |
3252 | #endif | |
3253 | } | |
3254 | ||
014db007 | 3255 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3256 | { |
3257 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3258 | struct io_madvise *ma = &req->madvise; | |
3259 | int ret; | |
3260 | ||
3261 | if (force_nonblock) | |
3262 | return -EAGAIN; | |
3263 | ||
3264 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3265 | if (ret < 0) | |
3266 | req_set_fail_links(req); | |
3267 | io_cqring_add_event(req, ret); | |
014db007 | 3268 | io_put_req(req); |
c1ca757b JA |
3269 | return 0; |
3270 | #else | |
3271 | return -EOPNOTSUPP; | |
3272 | #endif | |
3273 | } | |
3274 | ||
4840e418 JA |
3275 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3276 | { | |
3277 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3278 | return -EINVAL; | |
3279 | ||
3280 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3281 | req->fadvise.len = READ_ONCE(sqe->len); | |
3282 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3283 | return 0; | |
3284 | } | |
3285 | ||
014db007 | 3286 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3287 | { |
3288 | struct io_fadvise *fa = &req->fadvise; | |
3289 | int ret; | |
3290 | ||
3e69426d JA |
3291 | if (force_nonblock) { |
3292 | switch (fa->advice) { | |
3293 | case POSIX_FADV_NORMAL: | |
3294 | case POSIX_FADV_RANDOM: | |
3295 | case POSIX_FADV_SEQUENTIAL: | |
3296 | break; | |
3297 | default: | |
3298 | return -EAGAIN; | |
3299 | } | |
3300 | } | |
4840e418 JA |
3301 | |
3302 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3303 | if (ret < 0) | |
3304 | req_set_fail_links(req); | |
3305 | io_cqring_add_event(req, ret); | |
014db007 | 3306 | io_put_req(req); |
4840e418 JA |
3307 | return 0; |
3308 | } | |
3309 | ||
eddc7ef5 JA |
3310 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3311 | { | |
f8748881 | 3312 | const char __user *fname; |
eddc7ef5 JA |
3313 | unsigned lookup_flags; |
3314 | int ret; | |
3315 | ||
3316 | if (sqe->ioprio || sqe->buf_index) | |
3317 | return -EINVAL; | |
9c280f90 | 3318 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3319 | return -EBADF; |
0bdbdd08 PB |
3320 | if (req->flags & REQ_F_NEED_CLEANUP) |
3321 | return 0; | |
eddc7ef5 JA |
3322 | |
3323 | req->open.dfd = READ_ONCE(sqe->fd); | |
3324 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3325 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3326 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3327 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3328 | |
c12cedf2 | 3329 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3330 | return -EINVAL; |
3331 | ||
f8748881 | 3332 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3333 | if (IS_ERR(req->open.filename)) { |
3334 | ret = PTR_ERR(req->open.filename); | |
3335 | req->open.filename = NULL; | |
3336 | return ret; | |
3337 | } | |
3338 | ||
8fef80bf | 3339 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3340 | return 0; |
3341 | } | |
3342 | ||
014db007 | 3343 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3344 | { |
3345 | struct io_open *ctx = &req->open; | |
3346 | unsigned lookup_flags; | |
3347 | struct path path; | |
3348 | struct kstat stat; | |
3349 | int ret; | |
3350 | ||
5b0bbee4 JA |
3351 | if (force_nonblock) { |
3352 | /* only need file table for an actual valid fd */ | |
3353 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3354 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3355 | return -EAGAIN; |
5b0bbee4 | 3356 | } |
eddc7ef5 | 3357 | |
c12cedf2 | 3358 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3359 | return -EINVAL; |
3360 | ||
3361 | retry: | |
3362 | /* filename_lookup() drops it, keep a reference */ | |
3363 | ctx->filename->refcnt++; | |
3364 | ||
3365 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3366 | NULL); | |
3367 | if (ret) | |
3368 | goto err; | |
3369 | ||
c12cedf2 | 3370 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3371 | path_put(&path); |
3372 | if (retry_estale(ret, lookup_flags)) { | |
3373 | lookup_flags |= LOOKUP_REVAL; | |
3374 | goto retry; | |
3375 | } | |
3376 | if (!ret) | |
3377 | ret = cp_statx(&stat, ctx->buffer); | |
3378 | err: | |
3379 | putname(ctx->filename); | |
8fef80bf | 3380 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3381 | if (ret < 0) |
3382 | req_set_fail_links(req); | |
3383 | io_cqring_add_event(req, ret); | |
014db007 | 3384 | io_put_req(req); |
eddc7ef5 JA |
3385 | return 0; |
3386 | } | |
3387 | ||
b5dba59e JA |
3388 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3389 | { | |
3390 | /* | |
3391 | * If we queue this for async, it must not be cancellable. That would | |
3392 | * leave the 'file' in an undeterminate state. | |
3393 | */ | |
3394 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3395 | ||
3396 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3397 | sqe->rw_flags || sqe->buf_index) | |
3398 | return -EINVAL; | |
9c280f90 | 3399 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3400 | return -EBADF; |
b5dba59e JA |
3401 | |
3402 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e JA |
3403 | return 0; |
3404 | } | |
3405 | ||
a93b3331 | 3406 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3407 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3408 | { |
3409 | int ret; | |
3410 | ||
3411 | ret = filp_close(req->close.put_file, req->work.files); | |
3412 | if (ret < 0) | |
3413 | req_set_fail_links(req); | |
3414 | io_cqring_add_event(req, ret); | |
3415 | fput(req->close.put_file); | |
014db007 | 3416 | io_put_req(req); |
a93b3331 PB |
3417 | } |
3418 | ||
b5dba59e JA |
3419 | static void io_close_finish(struct io_wq_work **workptr) |
3420 | { | |
3421 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3422 | |
7fbeb95d | 3423 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3424 | __io_close_finish(req); |
e9fd9396 | 3425 | io_steal_work(req, workptr); |
b5dba59e JA |
3426 | } |
3427 | ||
014db007 | 3428 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3429 | { |
3430 | int ret; | |
3431 | ||
3432 | req->close.put_file = NULL; | |
3433 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
904fbcb1 JA |
3434 | if (ret < 0) { |
3435 | if (ret == -ENOENT) | |
3436 | ret = -EBADF; | |
b5dba59e | 3437 | return ret; |
904fbcb1 | 3438 | } |
b5dba59e JA |
3439 | |
3440 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3441 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3442 | /* submission ref will be dropped, take it for async */ |
3443 | refcount_inc(&req->refs); | |
3444 | ||
a2100672 PB |
3445 | req->work.func = io_close_finish; |
3446 | /* | |
3447 | * Do manual async queue here to avoid grabbing files - we don't | |
3448 | * need the files, and it'll cause io_close_finish() to close | |
3449 | * the file again and cause a double CQE entry for this request | |
3450 | */ | |
3451 | io_queue_async_work(req); | |
3452 | return 0; | |
3453 | } | |
b5dba59e JA |
3454 | |
3455 | /* | |
3456 | * No ->flush(), safely close from here and just punt the | |
3457 | * fput() to async context. | |
3458 | */ | |
014db007 | 3459 | __io_close_finish(req); |
1a417f4e | 3460 | return 0; |
b5dba59e JA |
3461 | } |
3462 | ||
3529d8c2 | 3463 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3464 | { |
3465 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3466 | |
3467 | if (!req->file) | |
3468 | return -EBADF; | |
5d17b4a4 JA |
3469 | |
3470 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3471 | return -EINVAL; | |
3472 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3473 | return -EINVAL; | |
3474 | ||
8ed8d3c3 JA |
3475 | req->sync.off = READ_ONCE(sqe->off); |
3476 | req->sync.len = READ_ONCE(sqe->len); | |
3477 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3478 | return 0; |
3479 | } | |
3480 | ||
014db007 | 3481 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3482 | { |
8ed8d3c3 JA |
3483 | int ret; |
3484 | ||
9adbd45d | 3485 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3486 | req->sync.flags); |
3487 | if (ret < 0) | |
3488 | req_set_fail_links(req); | |
3489 | io_cqring_add_event(req, ret); | |
014db007 | 3490 | io_put_req(req); |
5ea62161 PB |
3491 | } |
3492 | ||
3493 | ||
3494 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3495 | { | |
3496 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3497 | |
3498 | if (io_req_cancelled(req)) | |
3499 | return; | |
014db007 | 3500 | __io_sync_file_range(req); |
7759a0bf | 3501 | io_steal_work(req, workptr); |
5d17b4a4 JA |
3502 | } |
3503 | ||
014db007 | 3504 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3505 | { |
5d17b4a4 | 3506 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3507 | if (force_nonblock) { |
8ed8d3c3 | 3508 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3509 | return -EAGAIN; |
8ed8d3c3 | 3510 | } |
5d17b4a4 | 3511 | |
014db007 | 3512 | __io_sync_file_range(req); |
5d17b4a4 JA |
3513 | return 0; |
3514 | } | |
3515 | ||
469956e8 | 3516 | #if defined(CONFIG_NET) |
02d27d89 PB |
3517 | static int io_setup_async_msg(struct io_kiocb *req, |
3518 | struct io_async_msghdr *kmsg) | |
3519 | { | |
3520 | if (req->io) | |
3521 | return -EAGAIN; | |
3522 | if (io_alloc_async_ctx(req)) { | |
3523 | if (kmsg->iov != kmsg->fast_iov) | |
3524 | kfree(kmsg->iov); | |
3525 | return -ENOMEM; | |
3526 | } | |
3527 | req->flags |= REQ_F_NEED_CLEANUP; | |
3528 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3529 | return -EAGAIN; | |
3530 | } | |
3531 | ||
3529d8c2 | 3532 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3533 | { |
e47293fd | 3534 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3535 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3536 | int ret; |
03b1230c | 3537 | |
e47293fd JA |
3538 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3539 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3540 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3541 | |
d8768362 JA |
3542 | #ifdef CONFIG_COMPAT |
3543 | if (req->ctx->compat) | |
3544 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3545 | #endif | |
3546 | ||
fddaface | 3547 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3548 | return 0; |
5f798bea PB |
3549 | /* iovec is already imported */ |
3550 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3551 | return 0; | |
3529d8c2 | 3552 | |
d9688565 | 3553 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3554 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3555 | &io->msg.iov); |
99bc4c38 PB |
3556 | if (!ret) |
3557 | req->flags |= REQ_F_NEED_CLEANUP; | |
3558 | return ret; | |
03b1230c JA |
3559 | } |
3560 | ||
014db007 | 3561 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3562 | { |
0b416c3e | 3563 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3564 | struct socket *sock; |
3565 | int ret; | |
3566 | ||
3567 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3568 | return -EINVAL; | |
3569 | ||
3570 | sock = sock_from_file(req->file, &ret); | |
3571 | if (sock) { | |
b7bb4f7d | 3572 | struct io_async_ctx io; |
0fa03c62 JA |
3573 | unsigned flags; |
3574 | ||
03b1230c | 3575 | if (req->io) { |
0b416c3e | 3576 | kmsg = &req->io->msg; |
b537916c | 3577 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3578 | /* if iov is set, it's allocated already */ |
3579 | if (!kmsg->iov) | |
3580 | kmsg->iov = kmsg->fast_iov; | |
3581 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3582 | } else { |
3529d8c2 JA |
3583 | struct io_sr_msg *sr = &req->sr_msg; |
3584 | ||
0b416c3e | 3585 | kmsg = &io.msg; |
b537916c | 3586 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3587 | |
3588 | io.msg.iov = io.msg.fast_iov; | |
3589 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3590 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3591 | if (ret) |
3529d8c2 | 3592 | return ret; |
03b1230c | 3593 | } |
0fa03c62 | 3594 | |
e47293fd JA |
3595 | flags = req->sr_msg.msg_flags; |
3596 | if (flags & MSG_DONTWAIT) | |
3597 | req->flags |= REQ_F_NOWAIT; | |
3598 | else if (force_nonblock) | |
3599 | flags |= MSG_DONTWAIT; | |
3600 | ||
0b416c3e | 3601 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3602 | if (force_nonblock && ret == -EAGAIN) |
3603 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3604 | if (ret == -ERESTARTSYS) |
3605 | ret = -EINTR; | |
0fa03c62 JA |
3606 | } |
3607 | ||
1e95081c | 3608 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3609 | kfree(kmsg->iov); |
99bc4c38 | 3610 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3611 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3612 | if (ret < 0) |
3613 | req_set_fail_links(req); | |
014db007 | 3614 | io_put_req(req); |
5d17b4a4 | 3615 | return 0; |
03b1230c | 3616 | } |
aa1fa28f | 3617 | |
014db007 | 3618 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3619 | { |
fddaface JA |
3620 | struct socket *sock; |
3621 | int ret; | |
3622 | ||
3623 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3624 | return -EINVAL; | |
3625 | ||
3626 | sock = sock_from_file(req->file, &ret); | |
3627 | if (sock) { | |
3628 | struct io_sr_msg *sr = &req->sr_msg; | |
3629 | struct msghdr msg; | |
3630 | struct iovec iov; | |
3631 | unsigned flags; | |
3632 | ||
3633 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3634 | &msg.msg_iter); | |
3635 | if (ret) | |
3636 | return ret; | |
3637 | ||
3638 | msg.msg_name = NULL; | |
3639 | msg.msg_control = NULL; | |
3640 | msg.msg_controllen = 0; | |
3641 | msg.msg_namelen = 0; | |
3642 | ||
3643 | flags = req->sr_msg.msg_flags; | |
3644 | if (flags & MSG_DONTWAIT) | |
3645 | req->flags |= REQ_F_NOWAIT; | |
3646 | else if (force_nonblock) | |
3647 | flags |= MSG_DONTWAIT; | |
3648 | ||
0b7b21e4 JA |
3649 | msg.msg_flags = flags; |
3650 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3651 | if (force_nonblock && ret == -EAGAIN) |
3652 | return -EAGAIN; | |
3653 | if (ret == -ERESTARTSYS) | |
3654 | ret = -EINTR; | |
3655 | } | |
3656 | ||
3657 | io_cqring_add_event(req, ret); | |
3658 | if (ret < 0) | |
3659 | req_set_fail_links(req); | |
014db007 | 3660 | io_put_req(req); |
fddaface | 3661 | return 0; |
fddaface JA |
3662 | } |
3663 | ||
52de1fe1 JA |
3664 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3665 | { | |
3666 | struct io_sr_msg *sr = &req->sr_msg; | |
3667 | struct iovec __user *uiov; | |
3668 | size_t iov_len; | |
3669 | int ret; | |
3670 | ||
3671 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3672 | &uiov, &iov_len); | |
3673 | if (ret) | |
3674 | return ret; | |
3675 | ||
3676 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3677 | if (iov_len > 1) | |
3678 | return -EINVAL; | |
3679 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3680 | return -EFAULT; | |
3681 | sr->len = io->msg.iov[0].iov_len; | |
3682 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3683 | sr->len); | |
3684 | io->msg.iov = NULL; | |
3685 | } else { | |
3686 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3687 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3688 | if (ret > 0) | |
3689 | ret = 0; | |
3690 | } | |
3691 | ||
3692 | return ret; | |
3693 | } | |
3694 | ||
3695 | #ifdef CONFIG_COMPAT | |
3696 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3697 | struct io_async_ctx *io) | |
3698 | { | |
3699 | struct compat_msghdr __user *msg_compat; | |
3700 | struct io_sr_msg *sr = &req->sr_msg; | |
3701 | struct compat_iovec __user *uiov; | |
3702 | compat_uptr_t ptr; | |
3703 | compat_size_t len; | |
3704 | int ret; | |
3705 | ||
3706 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3707 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3708 | &ptr, &len); | |
3709 | if (ret) | |
3710 | return ret; | |
3711 | ||
3712 | uiov = compat_ptr(ptr); | |
3713 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3714 | compat_ssize_t clen; | |
3715 | ||
3716 | if (len > 1) | |
3717 | return -EINVAL; | |
3718 | if (!access_ok(uiov, sizeof(*uiov))) | |
3719 | return -EFAULT; | |
3720 | if (__get_user(clen, &uiov->iov_len)) | |
3721 | return -EFAULT; | |
3722 | if (clen < 0) | |
3723 | return -EINVAL; | |
3724 | sr->len = io->msg.iov[0].iov_len; | |
3725 | io->msg.iov = NULL; | |
3726 | } else { | |
3727 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3728 | &io->msg.iov, | |
3729 | &io->msg.msg.msg_iter); | |
3730 | if (ret < 0) | |
3731 | return ret; | |
3732 | } | |
3733 | ||
3734 | return 0; | |
3735 | } | |
3736 | #endif | |
3737 | ||
3738 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3739 | { | |
3740 | io->msg.iov = io->msg.fast_iov; | |
3741 | ||
3742 | #ifdef CONFIG_COMPAT | |
3743 | if (req->ctx->compat) | |
3744 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3745 | #endif |
52de1fe1 JA |
3746 | |
3747 | return __io_recvmsg_copy_hdr(req, io); | |
3748 | } | |
3749 | ||
bcda7baa JA |
3750 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3751 | int *cflags, bool needs_lock) | |
3752 | { | |
3753 | struct io_sr_msg *sr = &req->sr_msg; | |
3754 | struct io_buffer *kbuf; | |
3755 | ||
3756 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3757 | return NULL; | |
3758 | ||
3759 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3760 | if (IS_ERR(kbuf)) | |
3761 | return kbuf; | |
3762 | ||
3763 | sr->kbuf = kbuf; | |
3764 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3765 | ||
3766 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3767 | *cflags |= IORING_CQE_F_BUFFER; | |
3768 | return kbuf; | |
fddaface JA |
3769 | } |
3770 | ||
3529d8c2 JA |
3771 | static int io_recvmsg_prep(struct io_kiocb *req, |
3772 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3773 | { |
e47293fd | 3774 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3775 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3776 | int ret; |
3529d8c2 JA |
3777 | |
3778 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3779 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3780 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3781 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3782 | |
d8768362 JA |
3783 | #ifdef CONFIG_COMPAT |
3784 | if (req->ctx->compat) | |
3785 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3786 | #endif | |
3787 | ||
fddaface | 3788 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3789 | return 0; |
5f798bea PB |
3790 | /* iovec is already imported */ |
3791 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3792 | return 0; | |
03b1230c | 3793 | |
52de1fe1 | 3794 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3795 | if (!ret) |
3796 | req->flags |= REQ_F_NEED_CLEANUP; | |
3797 | return ret; | |
aa1fa28f JA |
3798 | } |
3799 | ||
014db007 | 3800 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3801 | { |
0b416c3e | 3802 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3803 | struct socket *sock; |
52de1fe1 | 3804 | int ret, cflags = 0; |
03b1230c JA |
3805 | |
3806 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3807 | return -EINVAL; | |
3808 | ||
3809 | sock = sock_from_file(req->file, &ret); | |
3810 | if (sock) { | |
52de1fe1 | 3811 | struct io_buffer *kbuf; |
b7bb4f7d | 3812 | struct io_async_ctx io; |
03b1230c JA |
3813 | unsigned flags; |
3814 | ||
03b1230c | 3815 | if (req->io) { |
0b416c3e | 3816 | kmsg = &req->io->msg; |
b537916c | 3817 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3818 | /* if iov is set, it's allocated already */ |
3819 | if (!kmsg->iov) | |
3820 | kmsg->iov = kmsg->fast_iov; | |
3821 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3822 | } else { |
0b416c3e | 3823 | kmsg = &io.msg; |
b537916c | 3824 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3825 | |
52de1fe1 | 3826 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3827 | if (ret) |
3529d8c2 | 3828 | return ret; |
03b1230c JA |
3829 | } |
3830 | ||
52de1fe1 JA |
3831 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3832 | if (IS_ERR(kbuf)) { | |
3833 | return PTR_ERR(kbuf); | |
3834 | } else if (kbuf) { | |
3835 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3836 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3837 | 1, req->sr_msg.len); | |
3838 | } | |
3839 | ||
e47293fd JA |
3840 | flags = req->sr_msg.msg_flags; |
3841 | if (flags & MSG_DONTWAIT) | |
3842 | req->flags |= REQ_F_NOWAIT; | |
3843 | else if (force_nonblock) | |
3844 | flags |= MSG_DONTWAIT; | |
3845 | ||
3846 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3847 | kmsg->uaddr, flags); | |
02d27d89 PB |
3848 | if (force_nonblock && ret == -EAGAIN) |
3849 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3850 | if (ret == -ERESTARTSYS) |
3851 | ret = -EINTR; | |
3852 | } | |
3853 | ||
1e95081c | 3854 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3855 | kfree(kmsg->iov); |
99bc4c38 | 3856 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3857 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3858 | if (ret < 0) |
3859 | req_set_fail_links(req); | |
014db007 | 3860 | io_put_req(req); |
03b1230c | 3861 | return 0; |
0fa03c62 | 3862 | } |
5d17b4a4 | 3863 | |
014db007 | 3864 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3865 | { |
bcda7baa | 3866 | struct io_buffer *kbuf = NULL; |
fddaface | 3867 | struct socket *sock; |
bcda7baa | 3868 | int ret, cflags = 0; |
fddaface JA |
3869 | |
3870 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3871 | return -EINVAL; | |
3872 | ||
3873 | sock = sock_from_file(req->file, &ret); | |
3874 | if (sock) { | |
3875 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3876 | void __user *buf = sr->buf; |
fddaface JA |
3877 | struct msghdr msg; |
3878 | struct iovec iov; | |
3879 | unsigned flags; | |
3880 | ||
bcda7baa JA |
3881 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3882 | if (IS_ERR(kbuf)) | |
3883 | return PTR_ERR(kbuf); | |
3884 | else if (kbuf) | |
3885 | buf = u64_to_user_ptr(kbuf->addr); | |
3886 | ||
3887 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3888 | &msg.msg_iter); |
bcda7baa JA |
3889 | if (ret) { |
3890 | kfree(kbuf); | |
fddaface | 3891 | return ret; |
bcda7baa | 3892 | } |
fddaface | 3893 | |
bcda7baa | 3894 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3895 | msg.msg_name = NULL; |
3896 | msg.msg_control = NULL; | |
3897 | msg.msg_controllen = 0; | |
3898 | msg.msg_namelen = 0; | |
3899 | msg.msg_iocb = NULL; | |
3900 | msg.msg_flags = 0; | |
3901 | ||
3902 | flags = req->sr_msg.msg_flags; | |
3903 | if (flags & MSG_DONTWAIT) | |
3904 | req->flags |= REQ_F_NOWAIT; | |
3905 | else if (force_nonblock) | |
3906 | flags |= MSG_DONTWAIT; | |
3907 | ||
0b7b21e4 | 3908 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3909 | if (force_nonblock && ret == -EAGAIN) |
3910 | return -EAGAIN; | |
3911 | if (ret == -ERESTARTSYS) | |
3912 | ret = -EINTR; | |
3913 | } | |
3914 | ||
bcda7baa JA |
3915 | kfree(kbuf); |
3916 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3917 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3918 | if (ret < 0) |
3919 | req_set_fail_links(req); | |
014db007 | 3920 | io_put_req(req); |
fddaface | 3921 | return 0; |
fddaface JA |
3922 | } |
3923 | ||
3529d8c2 | 3924 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3925 | { |
8ed8d3c3 JA |
3926 | struct io_accept *accept = &req->accept; |
3927 | ||
17f2fe35 JA |
3928 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3929 | return -EINVAL; | |
8042d6ce | 3930 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3931 | return -EINVAL; |
3932 | ||
d55e5f5b JA |
3933 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3934 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3935 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3936 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3937 | return 0; |
8ed8d3c3 | 3938 | } |
17f2fe35 | 3939 | |
014db007 | 3940 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3941 | { |
3942 | struct io_accept *accept = &req->accept; | |
3943 | unsigned file_flags; | |
3944 | int ret; | |
3945 | ||
3946 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3947 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3948 | accept->addr_len, accept->flags, |
3949 | accept->nofile); | |
8ed8d3c3 | 3950 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3951 | return -EAGAIN; |
8e3cca12 JA |
3952 | if (ret == -ERESTARTSYS) |
3953 | ret = -EINTR; | |
4e88d6e7 JA |
3954 | if (ret < 0) |
3955 | req_set_fail_links(req); | |
78e19bbe | 3956 | io_cqring_add_event(req, ret); |
014db007 | 3957 | io_put_req(req); |
17f2fe35 | 3958 | return 0; |
8ed8d3c3 JA |
3959 | } |
3960 | ||
3961 | static void io_accept_finish(struct io_wq_work **workptr) | |
3962 | { | |
3963 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3964 | |
3965 | if (io_req_cancelled(req)) | |
3966 | return; | |
014db007 | 3967 | __io_accept(req, false); |
e9fd9396 | 3968 | io_steal_work(req, workptr); |
8ed8d3c3 | 3969 | } |
8ed8d3c3 | 3970 | |
014db007 | 3971 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3972 | { |
8ed8d3c3 JA |
3973 | int ret; |
3974 | ||
014db007 | 3975 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3976 | if (ret == -EAGAIN && force_nonblock) { |
3977 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3978 | return -EAGAIN; |
3979 | } | |
3980 | return 0; | |
0fa03c62 | 3981 | } |
5d17b4a4 | 3982 | |
3529d8c2 | 3983 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3984 | { |
3529d8c2 JA |
3985 | struct io_connect *conn = &req->connect; |
3986 | struct io_async_ctx *io = req->io; | |
f499a021 | 3987 | |
3fbb51c1 JA |
3988 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3989 | return -EINVAL; | |
3990 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3991 | return -EINVAL; | |
3992 | ||
3529d8c2 JA |
3993 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3994 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3995 | ||
3996 | if (!io) | |
3997 | return 0; | |
3998 | ||
3999 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4000 | &io->connect.address); |
f499a021 JA |
4001 | } |
4002 | ||
014db007 | 4003 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4004 | { |
f499a021 | 4005 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4006 | unsigned file_flags; |
3fbb51c1 | 4007 | int ret; |
f8e85cf2 | 4008 | |
f499a021 JA |
4009 | if (req->io) { |
4010 | io = req->io; | |
4011 | } else { | |
3529d8c2 JA |
4012 | ret = move_addr_to_kernel(req->connect.addr, |
4013 | req->connect.addr_len, | |
4014 | &__io.connect.address); | |
f499a021 JA |
4015 | if (ret) |
4016 | goto out; | |
4017 | io = &__io; | |
4018 | } | |
4019 | ||
3fbb51c1 JA |
4020 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4021 | ||
4022 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4023 | req->connect.addr_len, file_flags); | |
87f80d62 | 4024 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4025 | if (req->io) |
4026 | return -EAGAIN; | |
4027 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4028 | ret = -ENOMEM; |
4029 | goto out; | |
4030 | } | |
b7bb4f7d | 4031 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4032 | return -EAGAIN; |
f499a021 | 4033 | } |
f8e85cf2 JA |
4034 | if (ret == -ERESTARTSYS) |
4035 | ret = -EINTR; | |
f499a021 | 4036 | out: |
4e88d6e7 JA |
4037 | if (ret < 0) |
4038 | req_set_fail_links(req); | |
f8e85cf2 | 4039 | io_cqring_add_event(req, ret); |
014db007 | 4040 | io_put_req(req); |
f8e85cf2 | 4041 | return 0; |
469956e8 Y |
4042 | } |
4043 | #else /* !CONFIG_NET */ | |
4044 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4045 | { | |
f8e85cf2 | 4046 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4047 | } |
4048 | ||
469956e8 Y |
4049 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4050 | { | |
4051 | return -EOPNOTSUPP; | |
4052 | } | |
4053 | ||
4054 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4055 | { | |
4056 | return -EOPNOTSUPP; | |
4057 | } | |
4058 | ||
4059 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4060 | const struct io_uring_sqe *sqe) | |
4061 | { | |
4062 | return -EOPNOTSUPP; | |
4063 | } | |
4064 | ||
4065 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4066 | { | |
4067 | return -EOPNOTSUPP; | |
4068 | } | |
4069 | ||
4070 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4071 | { | |
4072 | return -EOPNOTSUPP; | |
4073 | } | |
4074 | ||
4075 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4076 | { | |
4077 | return -EOPNOTSUPP; | |
4078 | } | |
4079 | ||
4080 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4081 | { | |
4082 | return -EOPNOTSUPP; | |
4083 | } | |
4084 | ||
4085 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4086 | { | |
4087 | return -EOPNOTSUPP; | |
4088 | } | |
4089 | ||
4090 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4091 | { | |
f8e85cf2 | 4092 | return -EOPNOTSUPP; |
f8e85cf2 | 4093 | } |
469956e8 | 4094 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4095 | |
d7718a9d JA |
4096 | struct io_poll_table { |
4097 | struct poll_table_struct pt; | |
4098 | struct io_kiocb *req; | |
4099 | int error; | |
4100 | }; | |
4101 | ||
d7718a9d JA |
4102 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4103 | __poll_t mask, task_work_func_t func) | |
4104 | { | |
4105 | struct task_struct *tsk; | |
aa96bf8a | 4106 | int ret; |
d7718a9d JA |
4107 | |
4108 | /* for instances that support it check for an event match first: */ | |
4109 | if (mask && !(mask & poll->events)) | |
4110 | return 0; | |
4111 | ||
4112 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4113 | ||
4114 | list_del_init(&poll->wait.entry); | |
4115 | ||
4116 | tsk = req->task; | |
4117 | req->result = mask; | |
4118 | init_task_work(&req->task_work, func); | |
4119 | /* | |
aa96bf8a JA |
4120 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4121 | * threads to ensure the work gets run, we can't always rely on exit | |
4122 | * cancelation taking care of this. | |
d7718a9d | 4123 | */ |
aa96bf8a JA |
4124 | ret = task_work_add(tsk, &req->task_work, true); |
4125 | if (unlikely(ret)) { | |
4126 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4127 | task_work_add(tsk, &req->task_work, true); | |
4128 | } | |
d7718a9d JA |
4129 | wake_up_process(tsk); |
4130 | return 1; | |
4131 | } | |
4132 | ||
74ce6ce4 JA |
4133 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4134 | __acquires(&req->ctx->completion_lock) | |
4135 | { | |
4136 | struct io_ring_ctx *ctx = req->ctx; | |
4137 | ||
4138 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4139 | struct poll_table_struct pt = { ._key = poll->events }; | |
4140 | ||
4141 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4142 | } | |
4143 | ||
4144 | spin_lock_irq(&ctx->completion_lock); | |
4145 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4146 | add_wait_queue(poll->head, &poll->wait); | |
4147 | return true; | |
4148 | } | |
4149 | ||
4150 | return false; | |
4151 | } | |
4152 | ||
18bceab1 JA |
4153 | static void io_poll_remove_double(struct io_kiocb *req) |
4154 | { | |
4155 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4156 | ||
4157 | lockdep_assert_held(&req->ctx->completion_lock); | |
4158 | ||
4159 | if (poll && poll->head) { | |
4160 | struct wait_queue_head *head = poll->head; | |
4161 | ||
4162 | spin_lock(&head->lock); | |
4163 | list_del_init(&poll->wait.entry); | |
4164 | if (poll->wait.private) | |
4165 | refcount_dec(&req->refs); | |
4166 | poll->head = NULL; | |
4167 | spin_unlock(&head->lock); | |
4168 | } | |
4169 | } | |
4170 | ||
4171 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4172 | { | |
4173 | struct io_ring_ctx *ctx = req->ctx; | |
4174 | ||
4175 | io_poll_remove_double(req); | |
4176 | req->poll.done = true; | |
4177 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4178 | io_commit_cqring(ctx); | |
4179 | } | |
4180 | ||
4181 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4182 | { | |
4183 | struct io_ring_ctx *ctx = req->ctx; | |
4184 | ||
4185 | if (io_poll_rewait(req, &req->poll)) { | |
4186 | spin_unlock_irq(&ctx->completion_lock); | |
4187 | return; | |
4188 | } | |
4189 | ||
4190 | hash_del(&req->hash_node); | |
4191 | io_poll_complete(req, req->result, 0); | |
4192 | req->flags |= REQ_F_COMP_LOCKED; | |
4193 | io_put_req_find_next(req, nxt); | |
4194 | spin_unlock_irq(&ctx->completion_lock); | |
4195 | ||
4196 | io_cqring_ev_posted(ctx); | |
4197 | } | |
4198 | ||
4199 | static void io_poll_task_func(struct callback_head *cb) | |
4200 | { | |
4201 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4202 | struct io_kiocb *nxt = NULL; | |
4203 | ||
4204 | io_poll_task_handler(req, &nxt); | |
4205 | if (nxt) { | |
4206 | struct io_ring_ctx *ctx = nxt->ctx; | |
4207 | ||
4208 | mutex_lock(&ctx->uring_lock); | |
4209 | __io_queue_sqe(nxt, NULL); | |
4210 | mutex_unlock(&ctx->uring_lock); | |
4211 | } | |
4212 | } | |
4213 | ||
4214 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4215 | int sync, void *key) | |
4216 | { | |
4217 | struct io_kiocb *req = wait->private; | |
4218 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4219 | __poll_t mask = key_to_poll(key); | |
4220 | ||
4221 | /* for instances that support it check for an event match first: */ | |
4222 | if (mask && !(mask & poll->events)) | |
4223 | return 0; | |
4224 | ||
4225 | if (req->poll.head) { | |
4226 | bool done; | |
4227 | ||
4228 | spin_lock(&req->poll.head->lock); | |
4229 | done = list_empty(&req->poll.wait.entry); | |
4230 | if (!done) | |
4231 | list_del_init(&req->poll.wait.entry); | |
4232 | spin_unlock(&req->poll.head->lock); | |
4233 | if (!done) | |
4234 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4235 | } | |
4236 | refcount_dec(&req->refs); | |
4237 | return 1; | |
4238 | } | |
4239 | ||
4240 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4241 | wait_queue_func_t wake_func) | |
4242 | { | |
4243 | poll->head = NULL; | |
4244 | poll->done = false; | |
4245 | poll->canceled = false; | |
4246 | poll->events = events; | |
4247 | INIT_LIST_HEAD(&poll->wait.entry); | |
4248 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4249 | } | |
4250 | ||
4251 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4252 | struct wait_queue_head *head) | |
4253 | { | |
4254 | struct io_kiocb *req = pt->req; | |
4255 | ||
4256 | /* | |
4257 | * If poll->head is already set, it's because the file being polled | |
4258 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4259 | * for write). Setup a separate io_poll_iocb if this happens. | |
4260 | */ | |
4261 | if (unlikely(poll->head)) { | |
4262 | /* already have a 2nd entry, fail a third attempt */ | |
4263 | if (req->io) { | |
4264 | pt->error = -EINVAL; | |
4265 | return; | |
4266 | } | |
4267 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4268 | if (!poll) { | |
4269 | pt->error = -ENOMEM; | |
4270 | return; | |
4271 | } | |
4272 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4273 | refcount_inc(&req->refs); | |
4274 | poll->wait.private = req; | |
4275 | req->io = (void *) poll; | |
4276 | } | |
4277 | ||
4278 | pt->error = 0; | |
4279 | poll->head = head; | |
4280 | add_wait_queue(head, &poll->wait); | |
4281 | } | |
4282 | ||
4283 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4284 | struct poll_table_struct *p) | |
4285 | { | |
4286 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4287 | ||
4288 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4289 | } | |
4290 | ||
d7718a9d JA |
4291 | static void io_async_task_func(struct callback_head *cb) |
4292 | { | |
4293 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4294 | struct async_poll *apoll = req->apoll; | |
4295 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4296 | bool canceled; |
d7718a9d JA |
4297 | |
4298 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4299 | ||
74ce6ce4 | 4300 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4301 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4302 | return; |
d7718a9d JA |
4303 | } |
4304 | ||
74ce6ce4 JA |
4305 | if (hash_hashed(&req->hash_node)) |
4306 | hash_del(&req->hash_node); | |
4307 | ||
2bae047e JA |
4308 | canceled = READ_ONCE(apoll->poll.canceled); |
4309 | if (canceled) { | |
4310 | io_cqring_fill_event(req, -ECANCELED); | |
4311 | io_commit_cqring(ctx); | |
4312 | } | |
4313 | ||
74ce6ce4 JA |
4314 | spin_unlock_irq(&ctx->completion_lock); |
4315 | ||
44575a67 XW |
4316 | /* restore ->work in case we need to retry again */ |
4317 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4318 | ||
2bae047e JA |
4319 | if (canceled) { |
4320 | kfree(apoll); | |
4321 | io_cqring_ev_posted(ctx); | |
4322 | req_set_fail_links(req); | |
44575a67 | 4323 | io_double_put_req(req); |
2bae047e JA |
4324 | return; |
4325 | } | |
4326 | ||
d7718a9d JA |
4327 | __set_current_state(TASK_RUNNING); |
4328 | mutex_lock(&ctx->uring_lock); | |
4329 | __io_queue_sqe(req, NULL); | |
4330 | mutex_unlock(&ctx->uring_lock); | |
4331 | ||
4332 | kfree(apoll); | |
4333 | } | |
4334 | ||
4335 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4336 | void *key) | |
4337 | { | |
4338 | struct io_kiocb *req = wait->private; | |
4339 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4340 | ||
4341 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4342 | key_to_poll(key)); | |
4343 | ||
4344 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4345 | } | |
4346 | ||
4347 | static void io_poll_req_insert(struct io_kiocb *req) | |
4348 | { | |
4349 | struct io_ring_ctx *ctx = req->ctx; | |
4350 | struct hlist_head *list; | |
4351 | ||
4352 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4353 | hlist_add_head(&req->hash_node, list); | |
4354 | } | |
4355 | ||
4356 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4357 | struct io_poll_iocb *poll, | |
4358 | struct io_poll_table *ipt, __poll_t mask, | |
4359 | wait_queue_func_t wake_func) | |
4360 | __acquires(&ctx->completion_lock) | |
4361 | { | |
4362 | struct io_ring_ctx *ctx = req->ctx; | |
4363 | bool cancel = false; | |
4364 | ||
4365 | poll->file = req->file; | |
18bceab1 JA |
4366 | io_init_poll_iocb(poll, mask, wake_func); |
4367 | poll->wait.private = req; | |
d7718a9d JA |
4368 | |
4369 | ipt->pt._key = mask; | |
4370 | ipt->req = req; | |
4371 | ipt->error = -EINVAL; | |
4372 | ||
d7718a9d JA |
4373 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4374 | ||
4375 | spin_lock_irq(&ctx->completion_lock); | |
4376 | if (likely(poll->head)) { | |
4377 | spin_lock(&poll->head->lock); | |
4378 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4379 | if (ipt->error) | |
4380 | cancel = true; | |
4381 | ipt->error = 0; | |
4382 | mask = 0; | |
4383 | } | |
4384 | if (mask || ipt->error) | |
4385 | list_del_init(&poll->wait.entry); | |
4386 | else if (cancel) | |
4387 | WRITE_ONCE(poll->canceled, true); | |
4388 | else if (!poll->done) /* actually waiting for an event */ | |
4389 | io_poll_req_insert(req); | |
4390 | spin_unlock(&poll->head->lock); | |
4391 | } | |
4392 | ||
4393 | return mask; | |
4394 | } | |
4395 | ||
4396 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4397 | { | |
4398 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4399 | struct io_ring_ctx *ctx = req->ctx; | |
4400 | struct async_poll *apoll; | |
4401 | struct io_poll_table ipt; | |
4402 | __poll_t mask, ret; | |
18bceab1 | 4403 | bool had_io; |
d7718a9d JA |
4404 | |
4405 | if (!req->file || !file_can_poll(req->file)) | |
4406 | return false; | |
4407 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4408 | return false; | |
4409 | if (!def->pollin && !def->pollout) | |
4410 | return false; | |
4411 | ||
4412 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4413 | if (unlikely(!apoll)) | |
4414 | return false; | |
4415 | ||
4416 | req->flags |= REQ_F_POLLED; | |
4417 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
18bceab1 | 4418 | had_io = req->io != NULL; |
d7718a9d | 4419 | |
3537b6a7 | 4420 | get_task_struct(current); |
d7718a9d JA |
4421 | req->task = current; |
4422 | req->apoll = apoll; | |
4423 | INIT_HLIST_NODE(&req->hash_node); | |
4424 | ||
8755d97a | 4425 | mask = 0; |
d7718a9d | 4426 | if (def->pollin) |
8755d97a | 4427 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4428 | if (def->pollout) |
4429 | mask |= POLLOUT | POLLWRNORM; | |
4430 | mask |= POLLERR | POLLPRI; | |
4431 | ||
4432 | ipt.pt._qproc = io_async_queue_proc; | |
4433 | ||
4434 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4435 | io_async_wake); | |
4436 | if (ret) { | |
4437 | ipt.error = 0; | |
18bceab1 JA |
4438 | /* only remove double add if we did it here */ |
4439 | if (!had_io) | |
4440 | io_poll_remove_double(req); | |
d7718a9d JA |
4441 | spin_unlock_irq(&ctx->completion_lock); |
4442 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4443 | kfree(apoll); | |
4444 | return false; | |
4445 | } | |
4446 | spin_unlock_irq(&ctx->completion_lock); | |
4447 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4448 | apoll->poll.events); | |
4449 | return true; | |
4450 | } | |
4451 | ||
4452 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4453 | struct io_poll_iocb *poll) | |
221c5eb2 | 4454 | { |
b41e9852 | 4455 | bool do_complete = false; |
221c5eb2 JA |
4456 | |
4457 | spin_lock(&poll->head->lock); | |
4458 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4459 | if (!list_empty(&poll->wait.entry)) { |
4460 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4461 | do_complete = true; |
221c5eb2 JA |
4462 | } |
4463 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4464 | return do_complete; |
4465 | } | |
4466 | ||
4467 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4468 | { | |
b1f573bd | 4469 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4470 | bool do_complete; |
4471 | ||
4472 | if (req->opcode == IORING_OP_POLL_ADD) { | |
18bceab1 | 4473 | io_poll_remove_double(req); |
d7718a9d JA |
4474 | do_complete = __io_poll_remove_one(req, &req->poll); |
4475 | } else { | |
b1f573bd | 4476 | apoll = req->apoll; |
d7718a9d JA |
4477 | /* non-poll requests have submit ref still */ |
4478 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4479 | if (do_complete) | |
4480 | io_put_req(req); | |
4481 | } | |
4482 | ||
78076bb6 | 4483 | hash_del(&req->hash_node); |
d7718a9d | 4484 | |
44575a67 | 4485 | if (do_complete && apoll) { |
b1f573bd XW |
4486 | /* |
4487 | * restore ->work because we need to call io_req_work_drop_env. | |
4488 | */ | |
4489 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4490 | kfree(apoll); | |
4491 | } | |
4492 | ||
b41e9852 JA |
4493 | if (do_complete) { |
4494 | io_cqring_fill_event(req, -ECANCELED); | |
4495 | io_commit_cqring(req->ctx); | |
4496 | req->flags |= REQ_F_COMP_LOCKED; | |
4497 | io_put_req(req); | |
4498 | } | |
4499 | ||
4500 | return do_complete; | |
221c5eb2 JA |
4501 | } |
4502 | ||
4503 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4504 | { | |
78076bb6 | 4505 | struct hlist_node *tmp; |
221c5eb2 | 4506 | struct io_kiocb *req; |
8e2e1faf | 4507 | int posted = 0, i; |
221c5eb2 JA |
4508 | |
4509 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4510 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4511 | struct hlist_head *list; | |
4512 | ||
4513 | list = &ctx->cancel_hash[i]; | |
4514 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4515 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4516 | } |
4517 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4518 | |
8e2e1faf JA |
4519 | if (posted) |
4520 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4521 | } |
4522 | ||
47f46768 JA |
4523 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4524 | { | |
78076bb6 | 4525 | struct hlist_head *list; |
47f46768 JA |
4526 | struct io_kiocb *req; |
4527 | ||
78076bb6 JA |
4528 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4529 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4530 | if (sqe_addr != req->user_data) |
4531 | continue; | |
4532 | if (io_poll_remove_one(req)) | |
eac406c6 | 4533 | return 0; |
b41e9852 | 4534 | return -EALREADY; |
47f46768 JA |
4535 | } |
4536 | ||
4537 | return -ENOENT; | |
4538 | } | |
4539 | ||
3529d8c2 JA |
4540 | static int io_poll_remove_prep(struct io_kiocb *req, |
4541 | const struct io_uring_sqe *sqe) | |
0969e783 | 4542 | { |
0969e783 JA |
4543 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4544 | return -EINVAL; | |
4545 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4546 | sqe->poll_events) | |
4547 | return -EINVAL; | |
4548 | ||
4549 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4550 | return 0; |
4551 | } | |
4552 | ||
221c5eb2 JA |
4553 | /* |
4554 | * Find a running poll command that matches one specified in sqe->addr, | |
4555 | * and remove it if found. | |
4556 | */ | |
fc4df999 | 4557 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4558 | { |
4559 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4560 | u64 addr; |
47f46768 | 4561 | int ret; |
221c5eb2 | 4562 | |
0969e783 | 4563 | addr = req->poll.addr; |
221c5eb2 | 4564 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4565 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4566 | spin_unlock_irq(&ctx->completion_lock); |
4567 | ||
78e19bbe | 4568 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4569 | if (ret < 0) |
4570 | req_set_fail_links(req); | |
e65ef56d | 4571 | io_put_req(req); |
221c5eb2 JA |
4572 | return 0; |
4573 | } | |
4574 | ||
221c5eb2 JA |
4575 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4576 | void *key) | |
4577 | { | |
c2f2eb7d JA |
4578 | struct io_kiocb *req = wait->private; |
4579 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4580 | |
d7718a9d | 4581 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4582 | } |
4583 | ||
221c5eb2 JA |
4584 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4585 | struct poll_table_struct *p) | |
4586 | { | |
4587 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4588 | ||
d7718a9d | 4589 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4590 | } |
4591 | ||
3529d8c2 | 4592 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4593 | { |
4594 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4595 | u16 events; |
221c5eb2 JA |
4596 | |
4597 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4598 | return -EINVAL; | |
4599 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4600 | return -EINVAL; | |
09bb8394 JA |
4601 | if (!poll->file) |
4602 | return -EBADF; | |
221c5eb2 | 4603 | |
221c5eb2 JA |
4604 | events = READ_ONCE(sqe->poll_events); |
4605 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4606 | |
3537b6a7 | 4607 | get_task_struct(current); |
b41e9852 | 4608 | req->task = current; |
0969e783 JA |
4609 | return 0; |
4610 | } | |
4611 | ||
014db007 | 4612 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4613 | { |
4614 | struct io_poll_iocb *poll = &req->poll; | |
4615 | struct io_ring_ctx *ctx = req->ctx; | |
4616 | struct io_poll_table ipt; | |
0969e783 | 4617 | __poll_t mask; |
0969e783 | 4618 | |
78076bb6 | 4619 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4620 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4621 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4622 | |
d7718a9d JA |
4623 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4624 | io_poll_wake); | |
221c5eb2 | 4625 | |
8c838788 | 4626 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4627 | ipt.error = 0; |
b0dd8a41 | 4628 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4629 | } |
221c5eb2 JA |
4630 | spin_unlock_irq(&ctx->completion_lock); |
4631 | ||
8c838788 JA |
4632 | if (mask) { |
4633 | io_cqring_ev_posted(ctx); | |
014db007 | 4634 | io_put_req(req); |
221c5eb2 | 4635 | } |
8c838788 | 4636 | return ipt.error; |
221c5eb2 JA |
4637 | } |
4638 | ||
5262f567 JA |
4639 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4640 | { | |
ad8a48ac JA |
4641 | struct io_timeout_data *data = container_of(timer, |
4642 | struct io_timeout_data, timer); | |
4643 | struct io_kiocb *req = data->req; | |
4644 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4645 | unsigned long flags; |
4646 | ||
5262f567 JA |
4647 | atomic_inc(&ctx->cq_timeouts); |
4648 | ||
4649 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4650 | /* |
11365043 JA |
4651 | * We could be racing with timeout deletion. If the list is empty, |
4652 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4653 | */ |
842f9612 | 4654 | if (!list_empty(&req->list)) { |
11365043 | 4655 | struct io_kiocb *prev; |
5262f567 | 4656 | |
11365043 JA |
4657 | /* |
4658 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4659 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4660 | * pointer will be increased, otherwise other timeout reqs may |
4661 | * return in advance without waiting for enough wait_nr. | |
4662 | */ | |
4663 | prev = req; | |
4664 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4665 | prev->sequence++; | |
11365043 | 4666 | list_del_init(&req->list); |
11365043 | 4667 | } |
5262f567 | 4668 | |
78e19bbe | 4669 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4670 | io_commit_cqring(ctx); |
4671 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4672 | ||
4673 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4674 | req_set_fail_links(req); |
5262f567 JA |
4675 | io_put_req(req); |
4676 | return HRTIMER_NORESTART; | |
4677 | } | |
4678 | ||
47f46768 JA |
4679 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4680 | { | |
4681 | struct io_kiocb *req; | |
4682 | int ret = -ENOENT; | |
4683 | ||
4684 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4685 | if (user_data == req->user_data) { | |
4686 | list_del_init(&req->list); | |
4687 | ret = 0; | |
4688 | break; | |
4689 | } | |
4690 | } | |
4691 | ||
4692 | if (ret == -ENOENT) | |
4693 | return ret; | |
4694 | ||
2d28390a | 4695 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4696 | if (ret == -1) |
4697 | return -EALREADY; | |
4698 | ||
4e88d6e7 | 4699 | req_set_fail_links(req); |
47f46768 JA |
4700 | io_cqring_fill_event(req, -ECANCELED); |
4701 | io_put_req(req); | |
4702 | return 0; | |
4703 | } | |
4704 | ||
3529d8c2 JA |
4705 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4706 | const struct io_uring_sqe *sqe) | |
b29472ee | 4707 | { |
b29472ee JA |
4708 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4709 | return -EINVAL; | |
4710 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4711 | return -EINVAL; | |
4712 | ||
4713 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4714 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4715 | if (req->timeout.flags) | |
4716 | return -EINVAL; | |
4717 | ||
b29472ee JA |
4718 | return 0; |
4719 | } | |
4720 | ||
11365043 JA |
4721 | /* |
4722 | * Remove or update an existing timeout command | |
4723 | */ | |
fc4df999 | 4724 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4725 | { |
4726 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4727 | int ret; |
11365043 | 4728 | |
11365043 | 4729 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4730 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4731 | |
47f46768 | 4732 | io_cqring_fill_event(req, ret); |
11365043 JA |
4733 | io_commit_cqring(ctx); |
4734 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4735 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4736 | if (ret < 0) |
4737 | req_set_fail_links(req); | |
ec9c02ad | 4738 | io_put_req(req); |
11365043 | 4739 | return 0; |
5262f567 JA |
4740 | } |
4741 | ||
3529d8c2 | 4742 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4743 | bool is_timeout_link) |
5262f567 | 4744 | { |
ad8a48ac | 4745 | struct io_timeout_data *data; |
a41525ab | 4746 | unsigned flags; |
5262f567 | 4747 | |
ad8a48ac | 4748 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4749 | return -EINVAL; |
ad8a48ac | 4750 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4751 | return -EINVAL; |
2d28390a JA |
4752 | if (sqe->off && is_timeout_link) |
4753 | return -EINVAL; | |
a41525ab JA |
4754 | flags = READ_ONCE(sqe->timeout_flags); |
4755 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4756 | return -EINVAL; |
bdf20073 | 4757 | |
26a61679 JA |
4758 | req->timeout.count = READ_ONCE(sqe->off); |
4759 | ||
3529d8c2 | 4760 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4761 | return -ENOMEM; |
4762 | ||
4763 | data = &req->io->timeout; | |
ad8a48ac | 4764 | data->req = req; |
ad8a48ac JA |
4765 | req->flags |= REQ_F_TIMEOUT; |
4766 | ||
4767 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4768 | return -EFAULT; |
4769 | ||
11365043 | 4770 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4771 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4772 | else |
ad8a48ac | 4773 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4774 | |
ad8a48ac JA |
4775 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4776 | return 0; | |
4777 | } | |
4778 | ||
fc4df999 | 4779 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4780 | { |
ad8a48ac JA |
4781 | struct io_ring_ctx *ctx = req->ctx; |
4782 | struct io_timeout_data *data; | |
4783 | struct list_head *entry; | |
4784 | unsigned span = 0; | |
b55ce732 | 4785 | u32 count = req->timeout.count; |
22cad158 | 4786 | u32 seq = req->sequence; |
ad8a48ac | 4787 | |
2d28390a | 4788 | data = &req->io->timeout; |
93bd25bb | 4789 | |
5262f567 JA |
4790 | /* |
4791 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4792 | * timeout event to be satisfied. If it isn't set, then this is |
4793 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4794 | */ |
93bd25bb JA |
4795 | if (!count) { |
4796 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4797 | spin_lock_irq(&ctx->completion_lock); | |
4798 | entry = ctx->timeout_list.prev; | |
4799 | goto add; | |
4800 | } | |
5262f567 | 4801 | |
22cad158 | 4802 | req->sequence = seq + count; |
5262f567 JA |
4803 | |
4804 | /* | |
4805 | * Insertion sort, ensuring the first entry in the list is always | |
4806 | * the one we need first. | |
4807 | */ | |
5262f567 JA |
4808 | spin_lock_irq(&ctx->completion_lock); |
4809 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4810 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4811 | unsigned nxt_seq; |
5da0fb1a | 4812 | long long tmp, tmp_nxt; |
b55ce732 | 4813 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4814 | |
93bd25bb JA |
4815 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4816 | continue; | |
4817 | ||
5da0fb1a | 4818 | /* |
22cad158 | 4819 | * Since seq + count can overflow, use type long |
5da0fb1a | 4820 | * long to store it. |
4821 | */ | |
22cad158 PB |
4822 | tmp = (long long)seq + count; |
4823 | nxt_seq = nxt->sequence - nxt_offset; | |
4824 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4825 | |
4826 | /* | |
4827 | * cached_sq_head may overflow, and it will never overflow twice | |
4828 | * once there is some timeout req still be valid. | |
4829 | */ | |
22cad158 | 4830 | if (seq < nxt_seq) |
8b07a65a | 4831 | tmp += UINT_MAX; |
5da0fb1a | 4832 | |
a1f58ba4 | 4833 | if (tmp > tmp_nxt) |
5262f567 | 4834 | break; |
a1f58ba4 | 4835 | |
4836 | /* | |
4837 | * Sequence of reqs after the insert one and itself should | |
4838 | * be adjusted because each timeout req consumes a slot. | |
4839 | */ | |
4840 | span++; | |
4841 | nxt->sequence++; | |
5262f567 | 4842 | } |
a1f58ba4 | 4843 | req->sequence -= span; |
93bd25bb | 4844 | add: |
5262f567 | 4845 | list_add(&req->list, entry); |
ad8a48ac JA |
4846 | data->timer.function = io_timeout_fn; |
4847 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4848 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4849 | return 0; |
4850 | } | |
5262f567 | 4851 | |
62755e35 JA |
4852 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4853 | { | |
4854 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4855 | ||
4856 | return req->user_data == (unsigned long) data; | |
4857 | } | |
4858 | ||
e977d6d3 | 4859 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4860 | { |
62755e35 | 4861 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4862 | int ret = 0; |
4863 | ||
62755e35 JA |
4864 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4865 | switch (cancel_ret) { | |
4866 | case IO_WQ_CANCEL_OK: | |
4867 | ret = 0; | |
4868 | break; | |
4869 | case IO_WQ_CANCEL_RUNNING: | |
4870 | ret = -EALREADY; | |
4871 | break; | |
4872 | case IO_WQ_CANCEL_NOTFOUND: | |
4873 | ret = -ENOENT; | |
4874 | break; | |
4875 | } | |
4876 | ||
e977d6d3 JA |
4877 | return ret; |
4878 | } | |
4879 | ||
47f46768 JA |
4880 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4881 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4882 | int success_ret) |
47f46768 JA |
4883 | { |
4884 | unsigned long flags; | |
4885 | int ret; | |
4886 | ||
4887 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4888 | if (ret != -ENOENT) { | |
4889 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4890 | goto done; | |
4891 | } | |
4892 | ||
4893 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4894 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4895 | if (ret != -ENOENT) | |
4896 | goto done; | |
4897 | ret = io_poll_cancel(ctx, sqe_addr); | |
4898 | done: | |
b0dd8a41 JA |
4899 | if (!ret) |
4900 | ret = success_ret; | |
47f46768 JA |
4901 | io_cqring_fill_event(req, ret); |
4902 | io_commit_cqring(ctx); | |
4903 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4904 | io_cqring_ev_posted(ctx); | |
4905 | ||
4e88d6e7 JA |
4906 | if (ret < 0) |
4907 | req_set_fail_links(req); | |
014db007 | 4908 | io_put_req(req); |
47f46768 JA |
4909 | } |
4910 | ||
3529d8c2 JA |
4911 | static int io_async_cancel_prep(struct io_kiocb *req, |
4912 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4913 | { |
fbf23849 | 4914 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4915 | return -EINVAL; |
4916 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4917 | sqe->cancel_flags) | |
4918 | return -EINVAL; | |
4919 | ||
fbf23849 JA |
4920 | req->cancel.addr = READ_ONCE(sqe->addr); |
4921 | return 0; | |
4922 | } | |
4923 | ||
014db007 | 4924 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4925 | { |
4926 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4927 | |
014db007 | 4928 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4929 | return 0; |
4930 | } | |
4931 | ||
05f3fb3c JA |
4932 | static int io_files_update_prep(struct io_kiocb *req, |
4933 | const struct io_uring_sqe *sqe) | |
4934 | { | |
4935 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4936 | return -EINVAL; | |
4937 | ||
4938 | req->files_update.offset = READ_ONCE(sqe->off); | |
4939 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4940 | if (!req->files_update.nr_args) | |
4941 | return -EINVAL; | |
4942 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4943 | return 0; | |
4944 | } | |
4945 | ||
4946 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4947 | { |
4948 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4949 | struct io_uring_files_update up; |
4950 | int ret; | |
fbf23849 | 4951 | |
f86cd20c | 4952 | if (force_nonblock) |
05f3fb3c | 4953 | return -EAGAIN; |
05f3fb3c JA |
4954 | |
4955 | up.offset = req->files_update.offset; | |
4956 | up.fds = req->files_update.arg; | |
4957 | ||
4958 | mutex_lock(&ctx->uring_lock); | |
4959 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4960 | mutex_unlock(&ctx->uring_lock); | |
4961 | ||
4962 | if (ret < 0) | |
4963 | req_set_fail_links(req); | |
4964 | io_cqring_add_event(req, ret); | |
4965 | io_put_req(req); | |
5262f567 JA |
4966 | return 0; |
4967 | } | |
4968 | ||
3529d8c2 JA |
4969 | static int io_req_defer_prep(struct io_kiocb *req, |
4970 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4971 | { |
e781573e | 4972 | ssize_t ret = 0; |
f67676d1 | 4973 | |
f1d96a8f PB |
4974 | if (!sqe) |
4975 | return 0; | |
4976 | ||
f86cd20c JA |
4977 | if (io_op_defs[req->opcode].file_table) { |
4978 | ret = io_grab_files(req); | |
4979 | if (unlikely(ret)) | |
4980 | return ret; | |
4981 | } | |
4982 | ||
cccf0ee8 JA |
4983 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4984 | ||
d625c6ee | 4985 | switch (req->opcode) { |
e781573e JA |
4986 | case IORING_OP_NOP: |
4987 | break; | |
f67676d1 JA |
4988 | case IORING_OP_READV: |
4989 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4990 | case IORING_OP_READ: |
3529d8c2 | 4991 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4992 | break; |
4993 | case IORING_OP_WRITEV: | |
4994 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4995 | case IORING_OP_WRITE: |
3529d8c2 | 4996 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4997 | break; |
0969e783 | 4998 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4999 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5000 | break; |
5001 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5002 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5003 | break; |
8ed8d3c3 | 5004 | case IORING_OP_FSYNC: |
3529d8c2 | 5005 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5006 | break; |
5007 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5008 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5009 | break; |
03b1230c | 5010 | case IORING_OP_SENDMSG: |
fddaface | 5011 | case IORING_OP_SEND: |
3529d8c2 | 5012 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5013 | break; |
5014 | case IORING_OP_RECVMSG: | |
fddaface | 5015 | case IORING_OP_RECV: |
3529d8c2 | 5016 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5017 | break; |
f499a021 | 5018 | case IORING_OP_CONNECT: |
3529d8c2 | 5019 | ret = io_connect_prep(req, sqe); |
f499a021 | 5020 | break; |
2d28390a | 5021 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5022 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5023 | break; |
b29472ee | 5024 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5025 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5026 | break; |
fbf23849 | 5027 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5028 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5029 | break; |
2d28390a | 5030 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5031 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5032 | break; |
8ed8d3c3 | 5033 | case IORING_OP_ACCEPT: |
3529d8c2 | 5034 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5035 | break; |
d63d1b5e JA |
5036 | case IORING_OP_FALLOCATE: |
5037 | ret = io_fallocate_prep(req, sqe); | |
5038 | break; | |
15b71abe JA |
5039 | case IORING_OP_OPENAT: |
5040 | ret = io_openat_prep(req, sqe); | |
5041 | break; | |
b5dba59e JA |
5042 | case IORING_OP_CLOSE: |
5043 | ret = io_close_prep(req, sqe); | |
5044 | break; | |
05f3fb3c JA |
5045 | case IORING_OP_FILES_UPDATE: |
5046 | ret = io_files_update_prep(req, sqe); | |
5047 | break; | |
eddc7ef5 JA |
5048 | case IORING_OP_STATX: |
5049 | ret = io_statx_prep(req, sqe); | |
5050 | break; | |
4840e418 JA |
5051 | case IORING_OP_FADVISE: |
5052 | ret = io_fadvise_prep(req, sqe); | |
5053 | break; | |
c1ca757b JA |
5054 | case IORING_OP_MADVISE: |
5055 | ret = io_madvise_prep(req, sqe); | |
5056 | break; | |
cebdb986 JA |
5057 | case IORING_OP_OPENAT2: |
5058 | ret = io_openat2_prep(req, sqe); | |
5059 | break; | |
3e4827b0 JA |
5060 | case IORING_OP_EPOLL_CTL: |
5061 | ret = io_epoll_ctl_prep(req, sqe); | |
5062 | break; | |
7d67af2c PB |
5063 | case IORING_OP_SPLICE: |
5064 | ret = io_splice_prep(req, sqe); | |
5065 | break; | |
ddf0322d JA |
5066 | case IORING_OP_PROVIDE_BUFFERS: |
5067 | ret = io_provide_buffers_prep(req, sqe); | |
5068 | break; | |
067524e9 JA |
5069 | case IORING_OP_REMOVE_BUFFERS: |
5070 | ret = io_remove_buffers_prep(req, sqe); | |
5071 | break; | |
f67676d1 | 5072 | default: |
e781573e JA |
5073 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5074 | req->opcode); | |
5075 | ret = -EINVAL; | |
b7bb4f7d | 5076 | break; |
f67676d1 JA |
5077 | } |
5078 | ||
b7bb4f7d | 5079 | return ret; |
f67676d1 JA |
5080 | } |
5081 | ||
3529d8c2 | 5082 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5083 | { |
a197f664 | 5084 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5085 | int ret; |
de0617e4 | 5086 | |
9d858b21 | 5087 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5088 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5089 | return 0; |
5090 | ||
3529d8c2 | 5091 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5092 | return -EAGAIN; |
5093 | ||
3529d8c2 | 5094 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5095 | if (ret < 0) |
2d28390a | 5096 | return ret; |
2d28390a | 5097 | |
de0617e4 | 5098 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5099 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5100 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5101 | return 0; |
5102 | } | |
5103 | ||
915967f6 | 5104 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5105 | list_add_tail(&req->list, &ctx->defer_list); |
5106 | spin_unlock_irq(&ctx->completion_lock); | |
5107 | return -EIOCBQUEUED; | |
5108 | } | |
5109 | ||
99bc4c38 PB |
5110 | static void io_cleanup_req(struct io_kiocb *req) |
5111 | { | |
5112 | struct io_async_ctx *io = req->io; | |
5113 | ||
5114 | switch (req->opcode) { | |
5115 | case IORING_OP_READV: | |
5116 | case IORING_OP_READ_FIXED: | |
5117 | case IORING_OP_READ: | |
bcda7baa JA |
5118 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5119 | kfree((void *)(unsigned long)req->rw.addr); | |
5120 | /* fallthrough */ | |
99bc4c38 PB |
5121 | case IORING_OP_WRITEV: |
5122 | case IORING_OP_WRITE_FIXED: | |
5123 | case IORING_OP_WRITE: | |
5124 | if (io->rw.iov != io->rw.fast_iov) | |
5125 | kfree(io->rw.iov); | |
5126 | break; | |
99bc4c38 | 5127 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5128 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5129 | kfree(req->sr_msg.kbuf); | |
5130 | /* fallthrough */ | |
5131 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5132 | if (io->msg.iov != io->msg.fast_iov) |
5133 | kfree(io->msg.iov); | |
5134 | break; | |
bcda7baa JA |
5135 | case IORING_OP_RECV: |
5136 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5137 | kfree(req->sr_msg.kbuf); | |
5138 | break; | |
8fef80bf PB |
5139 | case IORING_OP_OPENAT: |
5140 | case IORING_OP_OPENAT2: | |
5141 | case IORING_OP_STATX: | |
5142 | putname(req->open.filename); | |
5143 | break; | |
7d67af2c PB |
5144 | case IORING_OP_SPLICE: |
5145 | io_put_file(req, req->splice.file_in, | |
5146 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5147 | break; | |
99bc4c38 PB |
5148 | } |
5149 | ||
5150 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5151 | } | |
5152 | ||
3529d8c2 | 5153 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5154 | bool force_nonblock) |
2b188cc1 | 5155 | { |
a197f664 | 5156 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5157 | int ret; |
2b188cc1 | 5158 | |
d625c6ee | 5159 | switch (req->opcode) { |
2b188cc1 | 5160 | case IORING_OP_NOP: |
78e19bbe | 5161 | ret = io_nop(req); |
2b188cc1 JA |
5162 | break; |
5163 | case IORING_OP_READV: | |
edafccee | 5164 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5165 | case IORING_OP_READ: |
3529d8c2 JA |
5166 | if (sqe) { |
5167 | ret = io_read_prep(req, sqe, force_nonblock); | |
5168 | if (ret < 0) | |
5169 | break; | |
5170 | } | |
014db007 | 5171 | ret = io_read(req, force_nonblock); |
edafccee | 5172 | break; |
3529d8c2 | 5173 | case IORING_OP_WRITEV: |
edafccee | 5174 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5175 | case IORING_OP_WRITE: |
3529d8c2 JA |
5176 | if (sqe) { |
5177 | ret = io_write_prep(req, sqe, force_nonblock); | |
5178 | if (ret < 0) | |
5179 | break; | |
5180 | } | |
014db007 | 5181 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5182 | break; |
c992fe29 | 5183 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5184 | if (sqe) { |
5185 | ret = io_prep_fsync(req, sqe); | |
5186 | if (ret < 0) | |
5187 | break; | |
5188 | } | |
014db007 | 5189 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5190 | break; |
221c5eb2 | 5191 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5192 | if (sqe) { |
5193 | ret = io_poll_add_prep(req, sqe); | |
5194 | if (ret) | |
5195 | break; | |
5196 | } | |
014db007 | 5197 | ret = io_poll_add(req); |
221c5eb2 JA |
5198 | break; |
5199 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5200 | if (sqe) { |
5201 | ret = io_poll_remove_prep(req, sqe); | |
5202 | if (ret < 0) | |
5203 | break; | |
5204 | } | |
fc4df999 | 5205 | ret = io_poll_remove(req); |
221c5eb2 | 5206 | break; |
5d17b4a4 | 5207 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5208 | if (sqe) { |
5209 | ret = io_prep_sfr(req, sqe); | |
5210 | if (ret < 0) | |
5211 | break; | |
5212 | } | |
014db007 | 5213 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5214 | break; |
0fa03c62 | 5215 | case IORING_OP_SENDMSG: |
fddaface | 5216 | case IORING_OP_SEND: |
3529d8c2 JA |
5217 | if (sqe) { |
5218 | ret = io_sendmsg_prep(req, sqe); | |
5219 | if (ret < 0) | |
5220 | break; | |
5221 | } | |
fddaface | 5222 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5223 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5224 | else |
014db007 | 5225 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5226 | break; |
aa1fa28f | 5227 | case IORING_OP_RECVMSG: |
fddaface | 5228 | case IORING_OP_RECV: |
3529d8c2 JA |
5229 | if (sqe) { |
5230 | ret = io_recvmsg_prep(req, sqe); | |
5231 | if (ret) | |
5232 | break; | |
5233 | } | |
fddaface | 5234 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5235 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5236 | else |
014db007 | 5237 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5238 | break; |
5262f567 | 5239 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5240 | if (sqe) { |
5241 | ret = io_timeout_prep(req, sqe, false); | |
5242 | if (ret) | |
5243 | break; | |
5244 | } | |
fc4df999 | 5245 | ret = io_timeout(req); |
5262f567 | 5246 | break; |
11365043 | 5247 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5248 | if (sqe) { |
5249 | ret = io_timeout_remove_prep(req, sqe); | |
5250 | if (ret) | |
5251 | break; | |
5252 | } | |
fc4df999 | 5253 | ret = io_timeout_remove(req); |
11365043 | 5254 | break; |
17f2fe35 | 5255 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5256 | if (sqe) { |
5257 | ret = io_accept_prep(req, sqe); | |
5258 | if (ret) | |
5259 | break; | |
5260 | } | |
014db007 | 5261 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5262 | break; |
f8e85cf2 | 5263 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5264 | if (sqe) { |
5265 | ret = io_connect_prep(req, sqe); | |
5266 | if (ret) | |
5267 | break; | |
5268 | } | |
014db007 | 5269 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5270 | break; |
62755e35 | 5271 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5272 | if (sqe) { |
5273 | ret = io_async_cancel_prep(req, sqe); | |
5274 | if (ret) | |
5275 | break; | |
5276 | } | |
014db007 | 5277 | ret = io_async_cancel(req); |
62755e35 | 5278 | break; |
d63d1b5e JA |
5279 | case IORING_OP_FALLOCATE: |
5280 | if (sqe) { | |
5281 | ret = io_fallocate_prep(req, sqe); | |
5282 | if (ret) | |
5283 | break; | |
5284 | } | |
014db007 | 5285 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5286 | break; |
15b71abe JA |
5287 | case IORING_OP_OPENAT: |
5288 | if (sqe) { | |
5289 | ret = io_openat_prep(req, sqe); | |
5290 | if (ret) | |
5291 | break; | |
5292 | } | |
014db007 | 5293 | ret = io_openat(req, force_nonblock); |
15b71abe | 5294 | break; |
b5dba59e JA |
5295 | case IORING_OP_CLOSE: |
5296 | if (sqe) { | |
5297 | ret = io_close_prep(req, sqe); | |
5298 | if (ret) | |
5299 | break; | |
5300 | } | |
014db007 | 5301 | ret = io_close(req, force_nonblock); |
b5dba59e | 5302 | break; |
05f3fb3c JA |
5303 | case IORING_OP_FILES_UPDATE: |
5304 | if (sqe) { | |
5305 | ret = io_files_update_prep(req, sqe); | |
5306 | if (ret) | |
5307 | break; | |
5308 | } | |
5309 | ret = io_files_update(req, force_nonblock); | |
5310 | break; | |
eddc7ef5 JA |
5311 | case IORING_OP_STATX: |
5312 | if (sqe) { | |
5313 | ret = io_statx_prep(req, sqe); | |
5314 | if (ret) | |
5315 | break; | |
5316 | } | |
014db007 | 5317 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5318 | break; |
4840e418 JA |
5319 | case IORING_OP_FADVISE: |
5320 | if (sqe) { | |
5321 | ret = io_fadvise_prep(req, sqe); | |
5322 | if (ret) | |
5323 | break; | |
5324 | } | |
014db007 | 5325 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5326 | break; |
c1ca757b JA |
5327 | case IORING_OP_MADVISE: |
5328 | if (sqe) { | |
5329 | ret = io_madvise_prep(req, sqe); | |
5330 | if (ret) | |
5331 | break; | |
5332 | } | |
014db007 | 5333 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5334 | break; |
cebdb986 JA |
5335 | case IORING_OP_OPENAT2: |
5336 | if (sqe) { | |
5337 | ret = io_openat2_prep(req, sqe); | |
5338 | if (ret) | |
5339 | break; | |
5340 | } | |
014db007 | 5341 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5342 | break; |
3e4827b0 JA |
5343 | case IORING_OP_EPOLL_CTL: |
5344 | if (sqe) { | |
5345 | ret = io_epoll_ctl_prep(req, sqe); | |
5346 | if (ret) | |
5347 | break; | |
5348 | } | |
014db007 | 5349 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5350 | break; |
7d67af2c PB |
5351 | case IORING_OP_SPLICE: |
5352 | if (sqe) { | |
5353 | ret = io_splice_prep(req, sqe); | |
5354 | if (ret < 0) | |
5355 | break; | |
5356 | } | |
014db007 | 5357 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5358 | break; |
ddf0322d JA |
5359 | case IORING_OP_PROVIDE_BUFFERS: |
5360 | if (sqe) { | |
5361 | ret = io_provide_buffers_prep(req, sqe); | |
5362 | if (ret) | |
5363 | break; | |
5364 | } | |
5365 | ret = io_provide_buffers(req, force_nonblock); | |
5366 | break; | |
067524e9 JA |
5367 | case IORING_OP_REMOVE_BUFFERS: |
5368 | if (sqe) { | |
5369 | ret = io_remove_buffers_prep(req, sqe); | |
5370 | if (ret) | |
5371 | break; | |
5372 | } | |
5373 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5374 | break; |
2b188cc1 JA |
5375 | default: |
5376 | ret = -EINVAL; | |
5377 | break; | |
5378 | } | |
5379 | ||
def596e9 JA |
5380 | if (ret) |
5381 | return ret; | |
5382 | ||
5383 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5384 | const bool in_async = io_wq_current_is_worker(); |
5385 | ||
9e645e11 | 5386 | if (req->result == -EAGAIN) |
def596e9 JA |
5387 | return -EAGAIN; |
5388 | ||
11ba820b JA |
5389 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5390 | if (in_async) | |
5391 | mutex_lock(&ctx->uring_lock); | |
5392 | ||
def596e9 | 5393 | io_iopoll_req_issued(req); |
11ba820b JA |
5394 | |
5395 | if (in_async) | |
5396 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5397 | } |
5398 | ||
5399 | return 0; | |
2b188cc1 JA |
5400 | } |
5401 | ||
561fb04a | 5402 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5403 | { |
561fb04a | 5404 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5405 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5406 | int ret = 0; |
2b188cc1 | 5407 | |
0c9d5ccd JA |
5408 | /* if NO_CANCEL is set, we must still run the work */ |
5409 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5410 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5411 | ret = -ECANCELED; |
0c9d5ccd | 5412 | } |
31b51510 | 5413 | |
561fb04a | 5414 | if (!ret) { |
561fb04a | 5415 | do { |
014db007 | 5416 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5417 | /* |
5418 | * We can get EAGAIN for polled IO even though we're | |
5419 | * forcing a sync submission from here, since we can't | |
5420 | * wait for request slots on the block side. | |
5421 | */ | |
5422 | if (ret != -EAGAIN) | |
5423 | break; | |
5424 | cond_resched(); | |
5425 | } while (1); | |
5426 | } | |
31b51510 | 5427 | |
561fb04a | 5428 | if (ret) { |
4e88d6e7 | 5429 | req_set_fail_links(req); |
78e19bbe | 5430 | io_cqring_add_event(req, ret); |
817869d2 | 5431 | io_put_req(req); |
edafccee | 5432 | } |
2b188cc1 | 5433 | |
e9fd9396 | 5434 | io_steal_work(req, workptr); |
2b188cc1 JA |
5435 | } |
5436 | ||
65e19f54 JA |
5437 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5438 | int index) | |
5439 | { | |
5440 | struct fixed_file_table *table; | |
5441 | ||
05f3fb3c | 5442 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5443 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5444 | } |
5445 | ||
8da11c19 PB |
5446 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5447 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5448 | { |
a197f664 | 5449 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5450 | struct file *file; |
09bb8394 | 5451 | |
8da11c19 | 5452 | if (fixed) { |
05f3fb3c | 5453 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5454 | (unsigned) fd >= ctx->nr_user_files)) |
5455 | return -EBADF; | |
b7620121 | 5456 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5457 | file = io_file_from_index(ctx, fd); |
5458 | if (!file) | |
08a45173 | 5459 | return -EBADF; |
05589553 XW |
5460 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5461 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5462 | } else { |
c826bd7a | 5463 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5464 | file = __io_file_get(state, fd); |
5465 | if (unlikely(!file)) | |
09bb8394 JA |
5466 | return -EBADF; |
5467 | } | |
5468 | ||
8da11c19 | 5469 | *out_file = file; |
09bb8394 JA |
5470 | return 0; |
5471 | } | |
5472 | ||
8da11c19 | 5473 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5474 | int fd) |
8da11c19 | 5475 | { |
8da11c19 PB |
5476 | bool fixed; |
5477 | ||
63ff8223 | 5478 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
8da11c19 PB |
5479 | if (unlikely(!fixed && req->needs_fixed_file)) |
5480 | return -EBADF; | |
5481 | ||
5482 | return io_file_get(state, req, fd, &req->file, fixed); | |
5483 | } | |
5484 | ||
a197f664 | 5485 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5486 | { |
5487 | int ret = -EBADF; | |
a197f664 | 5488 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5489 | |
5b0bbee4 | 5490 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5491 | return 0; |
b14cca0c | 5492 | if (!ctx->ring_file) |
b5dba59e JA |
5493 | return -EBADF; |
5494 | ||
fcb323cc JA |
5495 | rcu_read_lock(); |
5496 | spin_lock_irq(&ctx->inflight_lock); | |
5497 | /* | |
5498 | * We use the f_ops->flush() handler to ensure that we can flush | |
5499 | * out work accessing these files if the fd is closed. Check if | |
5500 | * the fd has changed since we started down this path, and disallow | |
5501 | * this operation if it has. | |
5502 | */ | |
b14cca0c | 5503 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5504 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5505 | req->flags |= REQ_F_INFLIGHT; | |
5506 | req->work.files = current->files; | |
5507 | ret = 0; | |
5508 | } | |
5509 | spin_unlock_irq(&ctx->inflight_lock); | |
5510 | rcu_read_unlock(); | |
5511 | ||
5512 | return ret; | |
5513 | } | |
5514 | ||
2665abfd | 5515 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5516 | { |
ad8a48ac JA |
5517 | struct io_timeout_data *data = container_of(timer, |
5518 | struct io_timeout_data, timer); | |
5519 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5520 | struct io_ring_ctx *ctx = req->ctx; |
5521 | struct io_kiocb *prev = NULL; | |
5522 | unsigned long flags; | |
2665abfd JA |
5523 | |
5524 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5525 | ||
5526 | /* | |
5527 | * We don't expect the list to be empty, that will only happen if we | |
5528 | * race with the completion of the linked work. | |
5529 | */ | |
4493233e PB |
5530 | if (!list_empty(&req->link_list)) { |
5531 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5532 | link_list); | |
5d960724 | 5533 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5534 | list_del_init(&req->link_list); |
5d960724 JA |
5535 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5536 | } else | |
76a46e06 | 5537 | prev = NULL; |
2665abfd JA |
5538 | } |
5539 | ||
5540 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5541 | ||
5542 | if (prev) { | |
4e88d6e7 | 5543 | req_set_fail_links(prev); |
014db007 | 5544 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5545 | io_put_req(prev); |
47f46768 JA |
5546 | } else { |
5547 | io_cqring_add_event(req, -ETIME); | |
5548 | io_put_req(req); | |
2665abfd | 5549 | } |
2665abfd JA |
5550 | return HRTIMER_NORESTART; |
5551 | } | |
5552 | ||
ad8a48ac | 5553 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5554 | { |
76a46e06 | 5555 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5556 | |
76a46e06 JA |
5557 | /* |
5558 | * If the list is now empty, then our linked request finished before | |
5559 | * we got a chance to setup the timer | |
5560 | */ | |
5561 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5562 | if (!list_empty(&req->link_list)) { |
2d28390a | 5563 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5564 | |
ad8a48ac JA |
5565 | data->timer.function = io_link_timeout_fn; |
5566 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5567 | data->mode); | |
2665abfd | 5568 | } |
76a46e06 | 5569 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5570 | |
2665abfd | 5571 | /* drop submission reference */ |
76a46e06 JA |
5572 | io_put_req(req); |
5573 | } | |
2665abfd | 5574 | |
ad8a48ac | 5575 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5576 | { |
5577 | struct io_kiocb *nxt; | |
5578 | ||
dea3b49c | 5579 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5580 | return NULL; |
d7718a9d JA |
5581 | /* for polled retry, if flag is set, we already went through here */ |
5582 | if (req->flags & REQ_F_POLLED) | |
5583 | return NULL; | |
2665abfd | 5584 | |
4493233e PB |
5585 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5586 | link_list); | |
d625c6ee | 5587 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5588 | return NULL; |
2665abfd | 5589 | |
76a46e06 | 5590 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5591 | return nxt; |
2665abfd JA |
5592 | } |
5593 | ||
3529d8c2 | 5594 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5595 | { |
4a0a7a18 | 5596 | struct io_kiocb *linked_timeout; |
4bc4494e | 5597 | struct io_kiocb *nxt; |
193155c8 | 5598 | const struct cred *old_creds = NULL; |
e0c5c576 | 5599 | int ret; |
2b188cc1 | 5600 | |
4a0a7a18 JA |
5601 | again: |
5602 | linked_timeout = io_prep_linked_timeout(req); | |
5603 | ||
193155c8 JA |
5604 | if (req->work.creds && req->work.creds != current_cred()) { |
5605 | if (old_creds) | |
5606 | revert_creds(old_creds); | |
5607 | if (old_creds == req->work.creds) | |
5608 | old_creds = NULL; /* restored original creds */ | |
5609 | else | |
5610 | old_creds = override_creds(req->work.creds); | |
5611 | } | |
5612 | ||
014db007 | 5613 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5614 | |
5615 | /* | |
5616 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5617 | * doesn't support non-blocking read/write attempts | |
5618 | */ | |
5619 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5620 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5621 | if (io_arm_poll_handler(req)) { |
5622 | if (linked_timeout) | |
5623 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5624 | goto exit; |
d7718a9d | 5625 | } |
86a761f8 | 5626 | punt: |
f86cd20c | 5627 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5628 | ret = io_grab_files(req); |
5629 | if (ret) | |
5630 | goto err; | |
2b188cc1 | 5631 | } |
bbad27b2 PB |
5632 | |
5633 | /* | |
5634 | * Queued up for async execution, worker will release | |
5635 | * submit reference when the iocb is actually submitted. | |
5636 | */ | |
5637 | io_queue_async_work(req); | |
4bc4494e | 5638 | goto exit; |
2b188cc1 | 5639 | } |
e65ef56d | 5640 | |
fcb323cc | 5641 | err: |
4bc4494e | 5642 | nxt = NULL; |
76a46e06 | 5643 | /* drop submission reference */ |
2a44f467 | 5644 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5645 | |
f9bd67f6 | 5646 | if (linked_timeout) { |
76a46e06 | 5647 | if (!ret) |
f9bd67f6 | 5648 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5649 | else |
f9bd67f6 | 5650 | io_put_req(linked_timeout); |
76a46e06 JA |
5651 | } |
5652 | ||
e65ef56d | 5653 | /* and drop final reference, if we failed */ |
9e645e11 | 5654 | if (ret) { |
78e19bbe | 5655 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5656 | req_set_fail_links(req); |
e65ef56d | 5657 | io_put_req(req); |
9e645e11 | 5658 | } |
4a0a7a18 JA |
5659 | if (nxt) { |
5660 | req = nxt; | |
86a761f8 PB |
5661 | |
5662 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5663 | goto punt; | |
4a0a7a18 JA |
5664 | goto again; |
5665 | } | |
4bc4494e | 5666 | exit: |
193155c8 JA |
5667 | if (old_creds) |
5668 | revert_creds(old_creds); | |
2b188cc1 JA |
5669 | } |
5670 | ||
3529d8c2 | 5671 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5672 | { |
5673 | int ret; | |
5674 | ||
3529d8c2 | 5675 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5676 | if (ret) { |
5677 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5678 | fail_req: |
78e19bbe | 5679 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5680 | req_set_fail_links(req); |
78e19bbe | 5681 | io_double_put_req(req); |
4fe2c963 | 5682 | } |
2550878f | 5683 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5684 | ret = io_req_defer_prep(req, sqe); |
5685 | if (unlikely(ret < 0)) | |
5686 | goto fail_req; | |
ce35a47a JA |
5687 | /* |
5688 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5689 | * to async execution. | |
5690 | */ | |
5691 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5692 | io_queue_async_work(req); | |
5693 | } else { | |
3529d8c2 | 5694 | __io_queue_sqe(req, sqe); |
ce35a47a | 5695 | } |
4fe2c963 JL |
5696 | } |
5697 | ||
1b4a51b6 | 5698 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5699 | { |
94ae5e77 | 5700 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5701 | io_cqring_add_event(req, -ECANCELED); |
5702 | io_double_put_req(req); | |
5703 | } else | |
3529d8c2 | 5704 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5705 | } |
5706 | ||
1d4240cc | 5707 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
7d01bd74 | 5708 | struct io_kiocb **link) |
9e645e11 | 5709 | { |
a197f664 | 5710 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5711 | int ret; |
9e645e11 | 5712 | |
9e645e11 JA |
5713 | /* |
5714 | * If we already have a head request, queue this one for async | |
5715 | * submittal once the head completes. If we don't have a head but | |
5716 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5717 | * submitted sync once the chain is complete. If none of those | |
5718 | * conditions are true (normal request), then just queue it. | |
5719 | */ | |
5720 | if (*link) { | |
9d76377f | 5721 | struct io_kiocb *head = *link; |
4e88d6e7 | 5722 | |
8cdf2193 PB |
5723 | /* |
5724 | * Taking sequential execution of a link, draining both sides | |
5725 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5726 | * requests in the link. So, it drains the head and the | |
5727 | * next after the link request. The last one is done via | |
5728 | * drain_next flag to persist the effect across calls. | |
5729 | */ | |
ef4ff581 | 5730 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5731 | head->flags |= REQ_F_IO_DRAIN; |
5732 | ctx->drain_next = 1; | |
5733 | } | |
1d4240cc PB |
5734 | if (io_alloc_async_ctx(req)) |
5735 | return -EAGAIN; | |
9e645e11 | 5736 | |
3529d8c2 | 5737 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5738 | if (ret) { |
4e88d6e7 | 5739 | /* fail even hard links since we don't submit */ |
9d76377f | 5740 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5741 | return ret; |
2d28390a | 5742 | } |
9d76377f PB |
5743 | trace_io_uring_link(ctx, req, head); |
5744 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5745 | |
5746 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5747 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5748 | io_queue_link_head(head); |
5749 | *link = NULL; | |
5750 | } | |
9e645e11 | 5751 | } else { |
711be031 PB |
5752 | if (unlikely(ctx->drain_next)) { |
5753 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5754 | ctx->drain_next = 0; |
711be031 | 5755 | } |
ef4ff581 | 5756 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5757 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5758 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5759 | |
1d4240cc PB |
5760 | if (io_alloc_async_ctx(req)) |
5761 | return -EAGAIN; | |
5762 | ||
711be031 PB |
5763 | ret = io_req_defer_prep(req, sqe); |
5764 | if (ret) | |
5765 | req->flags |= REQ_F_FAIL_LINK; | |
5766 | *link = req; | |
5767 | } else { | |
5768 | io_queue_sqe(req, sqe); | |
5769 | } | |
9e645e11 | 5770 | } |
2e6e1fde | 5771 | |
1d4240cc | 5772 | return 0; |
9e645e11 JA |
5773 | } |
5774 | ||
9a56a232 JA |
5775 | /* |
5776 | * Batched submission is done, ensure local IO is flushed out. | |
5777 | */ | |
5778 | static void io_submit_state_end(struct io_submit_state *state) | |
5779 | { | |
5780 | blk_finish_plug(&state->plug); | |
3d6770fb | 5781 | io_file_put(state); |
2579f913 | 5782 | if (state->free_reqs) |
6c8a3134 | 5783 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5784 | } |
5785 | ||
5786 | /* | |
5787 | * Start submission side cache. | |
5788 | */ | |
5789 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5790 | unsigned int max_ios) |
9a56a232 JA |
5791 | { |
5792 | blk_start_plug(&state->plug); | |
2579f913 | 5793 | state->free_reqs = 0; |
9a56a232 JA |
5794 | state->file = NULL; |
5795 | state->ios_left = max_ios; | |
5796 | } | |
5797 | ||
2b188cc1 JA |
5798 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5799 | { | |
75b28aff | 5800 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5801 | |
caf582c6 PB |
5802 | /* |
5803 | * Ensure any loads from the SQEs are done at this point, | |
5804 | * since once we write the new head, the application could | |
5805 | * write new data to them. | |
5806 | */ | |
5807 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5808 | } |
5809 | ||
2b188cc1 | 5810 | /* |
3529d8c2 | 5811 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5812 | * that is mapped by userspace. This means that care needs to be taken to |
5813 | * ensure that reads are stable, as we cannot rely on userspace always | |
5814 | * being a good citizen. If members of the sqe are validated and then later | |
5815 | * used, it's important that those reads are done through READ_ONCE() to | |
5816 | * prevent a re-load down the line. | |
5817 | */ | |
709b302f | 5818 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5819 | { |
75b28aff | 5820 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5821 | unsigned head; |
5822 | ||
5823 | /* | |
5824 | * The cached sq head (or cq tail) serves two purposes: | |
5825 | * | |
5826 | * 1) allows us to batch the cost of updating the user visible | |
5827 | * head updates. | |
5828 | * 2) allows the kernel side to track the head on its own, even | |
5829 | * though the application is the one updating it. | |
5830 | */ | |
ee7d46d9 | 5831 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5832 | if (likely(head < ctx->sq_entries)) |
5833 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5834 | |
5835 | /* drop invalid entries */ | |
498ccd9e | 5836 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5837 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5838 | return NULL; |
5839 | } | |
5840 | ||
5841 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5842 | { | |
5843 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5844 | } |
5845 | ||
ef4ff581 PB |
5846 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5847 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5848 | IOSQE_BUFFER_SELECT) | |
5849 | ||
5850 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5851 | const struct io_uring_sqe *sqe, | |
5852 | struct io_submit_state *state, bool async) | |
0553b8bd | 5853 | { |
ef4ff581 | 5854 | unsigned int sqe_flags; |
63ff8223 | 5855 | int id; |
ef4ff581 | 5856 | |
0553b8bd PB |
5857 | /* |
5858 | * All io need record the previous position, if LINK vs DARIN, | |
5859 | * it can be used to mark the position of the first IO in the | |
5860 | * link list. | |
5861 | */ | |
31af27c7 | 5862 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5863 | req->opcode = READ_ONCE(sqe->opcode); |
5864 | req->user_data = READ_ONCE(sqe->user_data); | |
5865 | req->io = NULL; | |
5866 | req->file = NULL; | |
5867 | req->ctx = ctx; | |
5868 | req->flags = 0; | |
5869 | /* one is dropped after submission, the other at completion */ | |
5870 | refcount_set(&req->refs, 2); | |
5871 | req->task = NULL; | |
5872 | req->result = 0; | |
ef4ff581 | 5873 | req->needs_fixed_file = async; |
0553b8bd | 5874 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5875 | |
5876 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5877 | return -EINVAL; | |
5878 | ||
5879 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5880 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5881 | return -EFAULT; | |
5882 | use_mm(ctx->sqo_mm); | |
5883 | } | |
5884 | ||
5885 | sqe_flags = READ_ONCE(sqe->flags); | |
5886 | /* enforce forwards compatibility on users */ | |
5887 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5888 | return -EINVAL; | |
5889 | ||
5890 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5891 | !io_op_defs[req->opcode].buffer_select) | |
5892 | return -EOPNOTSUPP; | |
5893 | ||
5894 | id = READ_ONCE(sqe->personality); | |
5895 | if (id) { | |
5896 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5897 | if (unlikely(!req->work.creds)) | |
5898 | return -EINVAL; | |
5899 | get_cred(req->work.creds); | |
5900 | } | |
5901 | ||
5902 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5903 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5904 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5905 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5906 | ||
63ff8223 JA |
5907 | if (!io_op_defs[req->opcode].needs_file) |
5908 | return 0; | |
5909 | ||
5910 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
5911 | } |
5912 | ||
fb5ccc98 | 5913 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5914 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5915 | { |
5916 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5917 | struct io_kiocb *link = NULL; |
9e645e11 | 5918 | int i, submitted = 0; |
6c271ce2 | 5919 | |
c4a2ed72 | 5920 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5921 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5922 | if (!list_empty(&ctx->cq_overflow_list) && | |
5923 | !io_cqring_overflow_flush(ctx, false)) | |
5924 | return -EBUSY; | |
5925 | } | |
6c271ce2 | 5926 | |
ee7d46d9 PB |
5927 | /* make sure SQ entry isn't read before tail */ |
5928 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5929 | |
2b85edfc PB |
5930 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5931 | return -EAGAIN; | |
6c271ce2 JA |
5932 | |
5933 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5934 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5935 | statep = &state; |
5936 | } | |
5937 | ||
b14cca0c PB |
5938 | ctx->ring_fd = ring_fd; |
5939 | ctx->ring_file = ring_file; | |
5940 | ||
6c271ce2 | 5941 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5942 | const struct io_uring_sqe *sqe; |
196be95c | 5943 | struct io_kiocb *req; |
1cb1edb2 | 5944 | int err; |
fb5ccc98 | 5945 | |
b1e50e54 PB |
5946 | sqe = io_get_sqe(ctx); |
5947 | if (unlikely(!sqe)) { | |
5948 | io_consume_sqe(ctx); | |
5949 | break; | |
5950 | } | |
0553b8bd | 5951 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5952 | if (unlikely(!req)) { |
5953 | if (!submitted) | |
5954 | submitted = -EAGAIN; | |
fb5ccc98 | 5955 | break; |
196be95c | 5956 | } |
fb5ccc98 | 5957 | |
ef4ff581 | 5958 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5959 | io_consume_sqe(ctx); |
d3656344 JA |
5960 | /* will complete beyond this point, count as submitted */ |
5961 | submitted++; | |
5962 | ||
ef4ff581 | 5963 | if (unlikely(err)) { |
1cb1edb2 PB |
5964 | fail_req: |
5965 | io_cqring_add_event(req, err); | |
d3656344 | 5966 | io_double_put_req(req); |
196be95c PB |
5967 | break; |
5968 | } | |
fb5ccc98 | 5969 | |
354420f7 JA |
5970 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5971 | true, async); | |
7d01bd74 | 5972 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
5973 | if (err) |
5974 | goto fail_req; | |
6c271ce2 JA |
5975 | } |
5976 | ||
9466f437 PB |
5977 | if (unlikely(submitted != nr)) { |
5978 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5979 | ||
5980 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5981 | } | |
9e645e11 | 5982 | if (link) |
1b4a51b6 | 5983 | io_queue_link_head(link); |
6c271ce2 JA |
5984 | if (statep) |
5985 | io_submit_state_end(&state); | |
5986 | ||
ae9428ca PB |
5987 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5988 | io_commit_sqring(ctx); | |
5989 | ||
6c271ce2 JA |
5990 | return submitted; |
5991 | } | |
5992 | ||
bf9c2f1c PB |
5993 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5994 | { | |
5995 | struct mm_struct *mm = current->mm; | |
5996 | ||
5997 | if (mm) { | |
5998 | unuse_mm(mm); | |
5999 | mmput(mm); | |
6000 | } | |
6001 | } | |
6002 | ||
6c271ce2 JA |
6003 | static int io_sq_thread(void *data) |
6004 | { | |
6c271ce2 | 6005 | struct io_ring_ctx *ctx = data; |
181e448d | 6006 | const struct cred *old_cred; |
6c271ce2 JA |
6007 | mm_segment_t old_fs; |
6008 | DEFINE_WAIT(wait); | |
6c271ce2 | 6009 | unsigned long timeout; |
bdcd3eab | 6010 | int ret = 0; |
6c271ce2 | 6011 | |
0f158b4c | 6012 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6013 | |
6c271ce2 JA |
6014 | old_fs = get_fs(); |
6015 | set_fs(USER_DS); | |
181e448d | 6016 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6017 | |
bdcd3eab | 6018 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6019 | while (!kthread_should_park()) { |
fb5ccc98 | 6020 | unsigned int to_submit; |
6c271ce2 | 6021 | |
bdcd3eab | 6022 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
6023 | unsigned nr_events = 0; |
6024 | ||
bdcd3eab XW |
6025 | mutex_lock(&ctx->uring_lock); |
6026 | if (!list_empty(&ctx->poll_list)) | |
6027 | io_iopoll_getevents(ctx, &nr_events, 0); | |
6028 | else | |
6c271ce2 | 6029 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6030 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6031 | } |
6032 | ||
fb5ccc98 | 6033 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6034 | |
6035 | /* | |
6036 | * If submit got -EBUSY, flag us as needing the application | |
6037 | * to enter the kernel to reap and flush events. | |
6038 | */ | |
6039 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
6040 | /* |
6041 | * Drop cur_mm before scheduling, we can't hold it for | |
6042 | * long periods (or over schedule()). Do this before | |
6043 | * adding ourselves to the waitqueue, as the unuse/drop | |
6044 | * may sleep. | |
6045 | */ | |
bf9c2f1c | 6046 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 6047 | |
6c271ce2 JA |
6048 | /* |
6049 | * We're polling. If we're within the defined idle | |
6050 | * period, then let us spin without work before going | |
c1edbf5f JA |
6051 | * to sleep. The exception is if we got EBUSY doing |
6052 | * more IO, we should wait for the application to | |
6053 | * reap events and wake us up. | |
6c271ce2 | 6054 | */ |
bdcd3eab | 6055 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
6056 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6057 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
6058 | if (current->task_works) |
6059 | task_work_run(); | |
9831a90c | 6060 | cond_resched(); |
6c271ce2 JA |
6061 | continue; |
6062 | } | |
6063 | ||
6c271ce2 JA |
6064 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6065 | TASK_INTERRUPTIBLE); | |
6066 | ||
bdcd3eab XW |
6067 | /* |
6068 | * While doing polled IO, before going to sleep, we need | |
6069 | * to check if there are new reqs added to poll_list, it | |
6070 | * is because reqs may have been punted to io worker and | |
6071 | * will be added to poll_list later, hence check the | |
6072 | * poll_list again. | |
6073 | */ | |
6074 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6075 | !list_empty_careful(&ctx->poll_list)) { | |
6076 | finish_wait(&ctx->sqo_wait, &wait); | |
6077 | continue; | |
6078 | } | |
6079 | ||
6c271ce2 | 6080 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6081 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6082 | /* make sure to read SQ tail after writing flags */ |
6083 | smp_mb(); | |
6c271ce2 | 6084 | |
fb5ccc98 | 6085 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6086 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6087 | if (kthread_should_park()) { |
6c271ce2 JA |
6088 | finish_wait(&ctx->sqo_wait, &wait); |
6089 | break; | |
6090 | } | |
b41e9852 JA |
6091 | if (current->task_works) { |
6092 | task_work_run(); | |
10bea96d | 6093 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6094 | continue; |
6095 | } | |
6c271ce2 JA |
6096 | if (signal_pending(current)) |
6097 | flush_signals(current); | |
6098 | schedule(); | |
6099 | finish_wait(&ctx->sqo_wait, &wait); | |
6100 | ||
75b28aff | 6101 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6102 | continue; |
6103 | } | |
6104 | finish_wait(&ctx->sqo_wait, &wait); | |
6105 | ||
75b28aff | 6106 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6107 | } |
6108 | ||
8a4955ff | 6109 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6110 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6111 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6112 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6113 | } |
6114 | ||
b41e9852 JA |
6115 | if (current->task_works) |
6116 | task_work_run(); | |
6117 | ||
6c271ce2 | 6118 | set_fs(old_fs); |
bf9c2f1c | 6119 | io_sq_thread_drop_mm(ctx); |
181e448d | 6120 | revert_creds(old_cred); |
06058632 | 6121 | |
2bbcd6d3 | 6122 | kthread_parkme(); |
06058632 | 6123 | |
6c271ce2 JA |
6124 | return 0; |
6125 | } | |
6126 | ||
bda52162 JA |
6127 | struct io_wait_queue { |
6128 | struct wait_queue_entry wq; | |
6129 | struct io_ring_ctx *ctx; | |
6130 | unsigned to_wait; | |
6131 | unsigned nr_timeouts; | |
6132 | }; | |
6133 | ||
1d7bb1d5 | 6134 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6135 | { |
6136 | struct io_ring_ctx *ctx = iowq->ctx; | |
6137 | ||
6138 | /* | |
d195a66e | 6139 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6140 | * started waiting. For timeouts, we always want to return to userspace, |
6141 | * regardless of event count. | |
6142 | */ | |
1d7bb1d5 | 6143 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6144 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6145 | } | |
6146 | ||
6147 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6148 | int wake_flags, void *key) | |
6149 | { | |
6150 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6151 | wq); | |
6152 | ||
1d7bb1d5 JA |
6153 | /* use noflush == true, as we can't safely rely on locking context */ |
6154 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6155 | return -1; |
6156 | ||
6157 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6158 | } | |
6159 | ||
2b188cc1 JA |
6160 | /* |
6161 | * Wait until events become available, if we don't already have some. The | |
6162 | * application must reap them itself, as they reside on the shared cq ring. | |
6163 | */ | |
6164 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6165 | const sigset_t __user *sig, size_t sigsz) | |
6166 | { | |
bda52162 JA |
6167 | struct io_wait_queue iowq = { |
6168 | .wq = { | |
6169 | .private = current, | |
6170 | .func = io_wake_function, | |
6171 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6172 | }, | |
6173 | .ctx = ctx, | |
6174 | .to_wait = min_events, | |
6175 | }; | |
75b28aff | 6176 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6177 | int ret = 0; |
2b188cc1 | 6178 | |
b41e9852 JA |
6179 | do { |
6180 | if (io_cqring_events(ctx, false) >= min_events) | |
6181 | return 0; | |
6182 | if (!current->task_works) | |
6183 | break; | |
6184 | task_work_run(); | |
6185 | } while (1); | |
2b188cc1 JA |
6186 | |
6187 | if (sig) { | |
9e75ad5d AB |
6188 | #ifdef CONFIG_COMPAT |
6189 | if (in_compat_syscall()) | |
6190 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6191 | sigsz); |
9e75ad5d AB |
6192 | else |
6193 | #endif | |
b772434b | 6194 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6195 | |
2b188cc1 JA |
6196 | if (ret) |
6197 | return ret; | |
6198 | } | |
6199 | ||
bda52162 | 6200 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6201 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6202 | do { |
6203 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6204 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6205 | if (current->task_works) |
6206 | task_work_run(); | |
1d7bb1d5 | 6207 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6208 | break; |
6209 | schedule(); | |
6210 | if (signal_pending(current)) { | |
e9ffa5c2 | 6211 | ret = -EINTR; |
bda52162 JA |
6212 | break; |
6213 | } | |
6214 | } while (1); | |
6215 | finish_wait(&ctx->wait, &iowq.wq); | |
6216 | ||
e9ffa5c2 | 6217 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6218 | |
75b28aff | 6219 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6220 | } |
6221 | ||
6b06314c JA |
6222 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6223 | { | |
6224 | #if defined(CONFIG_UNIX) | |
6225 | if (ctx->ring_sock) { | |
6226 | struct sock *sock = ctx->ring_sock->sk; | |
6227 | struct sk_buff *skb; | |
6228 | ||
6229 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6230 | kfree_skb(skb); | |
6231 | } | |
6232 | #else | |
6233 | int i; | |
6234 | ||
65e19f54 JA |
6235 | for (i = 0; i < ctx->nr_user_files; i++) { |
6236 | struct file *file; | |
6237 | ||
6238 | file = io_file_from_index(ctx, i); | |
6239 | if (file) | |
6240 | fput(file); | |
6241 | } | |
6b06314c JA |
6242 | #endif |
6243 | } | |
6244 | ||
05f3fb3c JA |
6245 | static void io_file_ref_kill(struct percpu_ref *ref) |
6246 | { | |
6247 | struct fixed_file_data *data; | |
6248 | ||
6249 | data = container_of(ref, struct fixed_file_data, refs); | |
6250 | complete(&data->done); | |
6251 | } | |
6252 | ||
6b06314c JA |
6253 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6254 | { | |
05f3fb3c | 6255 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6256 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6257 | unsigned nr_tables, i; |
05589553 | 6258 | unsigned long flags; |
65e19f54 | 6259 | |
05f3fb3c | 6260 | if (!data) |
6b06314c JA |
6261 | return -ENXIO; |
6262 | ||
05589553 XW |
6263 | spin_lock_irqsave(&data->lock, flags); |
6264 | if (!list_empty(&data->ref_list)) | |
6265 | ref_node = list_first_entry(&data->ref_list, | |
6266 | struct fixed_file_ref_node, node); | |
6267 | spin_unlock_irqrestore(&data->lock, flags); | |
6268 | if (ref_node) | |
6269 | percpu_ref_kill(&ref_node->refs); | |
6270 | ||
6271 | percpu_ref_kill(&data->refs); | |
6272 | ||
6273 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6274 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6275 | wait_for_completion(&data->done); |
05f3fb3c | 6276 | |
6b06314c | 6277 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6278 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6279 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6280 | kfree(data->table[i].files); |
6281 | kfree(data->table); | |
05589553 XW |
6282 | percpu_ref_exit(&data->refs); |
6283 | kfree(data); | |
05f3fb3c | 6284 | ctx->file_data = NULL; |
6b06314c JA |
6285 | ctx->nr_user_files = 0; |
6286 | return 0; | |
6287 | } | |
6288 | ||
6c271ce2 JA |
6289 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6290 | { | |
6291 | if (ctx->sqo_thread) { | |
0f158b4c | 6292 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6293 | /* |
6294 | * The park is a bit of a work-around, without it we get | |
6295 | * warning spews on shutdown with SQPOLL set and affinity | |
6296 | * set to a single CPU. | |
6297 | */ | |
06058632 | 6298 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6299 | kthread_stop(ctx->sqo_thread); |
6300 | ctx->sqo_thread = NULL; | |
6301 | } | |
6302 | } | |
6303 | ||
6b06314c JA |
6304 | static void io_finish_async(struct io_ring_ctx *ctx) |
6305 | { | |
6c271ce2 JA |
6306 | io_sq_thread_stop(ctx); |
6307 | ||
561fb04a JA |
6308 | if (ctx->io_wq) { |
6309 | io_wq_destroy(ctx->io_wq); | |
6310 | ctx->io_wq = NULL; | |
6b06314c JA |
6311 | } |
6312 | } | |
6313 | ||
6314 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6315 | /* |
6316 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6317 | * the io_uring can be safely unregistered on process exit, even if we have | |
6318 | * loops in the file referencing. | |
6319 | */ | |
6320 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6321 | { | |
6322 | struct sock *sk = ctx->ring_sock->sk; | |
6323 | struct scm_fp_list *fpl; | |
6324 | struct sk_buff *skb; | |
08a45173 | 6325 | int i, nr_files; |
6b06314c | 6326 | |
6b06314c JA |
6327 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6328 | if (!fpl) | |
6329 | return -ENOMEM; | |
6330 | ||
6331 | skb = alloc_skb(0, GFP_KERNEL); | |
6332 | if (!skb) { | |
6333 | kfree(fpl); | |
6334 | return -ENOMEM; | |
6335 | } | |
6336 | ||
6337 | skb->sk = sk; | |
6b06314c | 6338 | |
08a45173 | 6339 | nr_files = 0; |
6b06314c JA |
6340 | fpl->user = get_uid(ctx->user); |
6341 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6342 | struct file *file = io_file_from_index(ctx, i + offset); |
6343 | ||
6344 | if (!file) | |
08a45173 | 6345 | continue; |
65e19f54 | 6346 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6347 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6348 | nr_files++; | |
6b06314c JA |
6349 | } |
6350 | ||
08a45173 JA |
6351 | if (nr_files) { |
6352 | fpl->max = SCM_MAX_FD; | |
6353 | fpl->count = nr_files; | |
6354 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6355 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6356 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6357 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6358 | |
08a45173 JA |
6359 | for (i = 0; i < nr_files; i++) |
6360 | fput(fpl->fp[i]); | |
6361 | } else { | |
6362 | kfree_skb(skb); | |
6363 | kfree(fpl); | |
6364 | } | |
6b06314c JA |
6365 | |
6366 | return 0; | |
6367 | } | |
6368 | ||
6369 | /* | |
6370 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6371 | * causes regular reference counting to break down. We rely on the UNIX | |
6372 | * garbage collection to take care of this problem for us. | |
6373 | */ | |
6374 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6375 | { | |
6376 | unsigned left, total; | |
6377 | int ret = 0; | |
6378 | ||
6379 | total = 0; | |
6380 | left = ctx->nr_user_files; | |
6381 | while (left) { | |
6382 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6383 | |
6384 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6385 | if (ret) | |
6386 | break; | |
6387 | left -= this_files; | |
6388 | total += this_files; | |
6389 | } | |
6390 | ||
6391 | if (!ret) | |
6392 | return 0; | |
6393 | ||
6394 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6395 | struct file *file = io_file_from_index(ctx, total); |
6396 | ||
6397 | if (file) | |
6398 | fput(file); | |
6b06314c JA |
6399 | total++; |
6400 | } | |
6401 | ||
6402 | return ret; | |
6403 | } | |
6404 | #else | |
6405 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6406 | { | |
6407 | return 0; | |
6408 | } | |
6409 | #endif | |
6410 | ||
65e19f54 JA |
6411 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6412 | unsigned nr_files) | |
6413 | { | |
6414 | int i; | |
6415 | ||
6416 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6417 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6418 | unsigned this_files; |
6419 | ||
6420 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6421 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6422 | GFP_KERNEL); | |
6423 | if (!table->files) | |
6424 | break; | |
6425 | nr_files -= this_files; | |
6426 | } | |
6427 | ||
6428 | if (i == nr_tables) | |
6429 | return 0; | |
6430 | ||
6431 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6432 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6433 | kfree(table->files); |
6434 | } | |
6435 | return 1; | |
6436 | } | |
6437 | ||
05f3fb3c JA |
6438 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6439 | { | |
6440 | #if defined(CONFIG_UNIX) | |
6441 | struct sock *sock = ctx->ring_sock->sk; | |
6442 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6443 | struct sk_buff *skb; | |
6444 | int i; | |
6445 | ||
6446 | __skb_queue_head_init(&list); | |
6447 | ||
6448 | /* | |
6449 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6450 | * remove this entry and rearrange the file array. | |
6451 | */ | |
6452 | skb = skb_dequeue(head); | |
6453 | while (skb) { | |
6454 | struct scm_fp_list *fp; | |
6455 | ||
6456 | fp = UNIXCB(skb).fp; | |
6457 | for (i = 0; i < fp->count; i++) { | |
6458 | int left; | |
6459 | ||
6460 | if (fp->fp[i] != file) | |
6461 | continue; | |
6462 | ||
6463 | unix_notinflight(fp->user, fp->fp[i]); | |
6464 | left = fp->count - 1 - i; | |
6465 | if (left) { | |
6466 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6467 | left * sizeof(struct file *)); | |
6468 | } | |
6469 | fp->count--; | |
6470 | if (!fp->count) { | |
6471 | kfree_skb(skb); | |
6472 | skb = NULL; | |
6473 | } else { | |
6474 | __skb_queue_tail(&list, skb); | |
6475 | } | |
6476 | fput(file); | |
6477 | file = NULL; | |
6478 | break; | |
6479 | } | |
6480 | ||
6481 | if (!file) | |
6482 | break; | |
6483 | ||
6484 | __skb_queue_tail(&list, skb); | |
6485 | ||
6486 | skb = skb_dequeue(head); | |
6487 | } | |
6488 | ||
6489 | if (skb_peek(&list)) { | |
6490 | spin_lock_irq(&head->lock); | |
6491 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6492 | __skb_queue_tail(head, skb); | |
6493 | spin_unlock_irq(&head->lock); | |
6494 | } | |
6495 | #else | |
6496 | fput(file); | |
6497 | #endif | |
6498 | } | |
6499 | ||
6500 | struct io_file_put { | |
05589553 | 6501 | struct list_head list; |
05f3fb3c | 6502 | struct file *file; |
05f3fb3c JA |
6503 | }; |
6504 | ||
4a38aed2 | 6505 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 6506 | { |
4a38aed2 JA |
6507 | struct fixed_file_data *file_data = ref_node->file_data; |
6508 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 6509 | struct io_file_put *pfile, *tmp; |
05589553 | 6510 | unsigned long flags; |
65e19f54 | 6511 | |
05589553 XW |
6512 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { |
6513 | list_del_init(&pfile->list); | |
6514 | io_ring_file_put(ctx, pfile->file); | |
6515 | kfree(pfile); | |
65e19f54 | 6516 | } |
05589553 XW |
6517 | |
6518 | spin_lock_irqsave(&file_data->lock, flags); | |
6519 | list_del_init(&ref_node->node); | |
6520 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6521 | ||
6522 | percpu_ref_exit(&ref_node->refs); | |
6523 | kfree(ref_node); | |
6524 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6525 | } |
65e19f54 | 6526 | |
4a38aed2 JA |
6527 | static void io_file_put_work(struct work_struct *work) |
6528 | { | |
6529 | struct io_ring_ctx *ctx; | |
6530 | struct llist_node *node; | |
6531 | ||
6532 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
6533 | node = llist_del_all(&ctx->file_put_llist); | |
6534 | ||
6535 | while (node) { | |
6536 | struct fixed_file_ref_node *ref_node; | |
6537 | struct llist_node *next = node->next; | |
6538 | ||
6539 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
6540 | __io_file_put_work(ref_node); | |
6541 | node = next; | |
6542 | } | |
6543 | } | |
6544 | ||
05589553 | 6545 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6546 | { |
05589553 | 6547 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
6548 | struct io_ring_ctx *ctx; |
6549 | bool first_add; | |
6550 | int delay = HZ; | |
65e19f54 | 6551 | |
05589553 | 6552 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 6553 | ctx = ref_node->file_data->ctx; |
05589553 | 6554 | |
4a38aed2 JA |
6555 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
6556 | delay = 0; | |
6557 | ||
6558 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); | |
6559 | if (!delay) | |
6560 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
6561 | else if (first_add) | |
6562 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 6563 | } |
65e19f54 | 6564 | |
05589553 XW |
6565 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6566 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6567 | { |
05589553 | 6568 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6569 | |
05589553 XW |
6570 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6571 | if (!ref_node) | |
6572 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6573 | |
05589553 XW |
6574 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6575 | 0, GFP_KERNEL)) { | |
6576 | kfree(ref_node); | |
6577 | return ERR_PTR(-ENOMEM); | |
6578 | } | |
6579 | INIT_LIST_HEAD(&ref_node->node); | |
6580 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
6581 | ref_node->file_data = ctx->file_data; |
6582 | return ref_node; | |
05589553 XW |
6583 | } |
6584 | ||
6585 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6586 | { | |
6587 | percpu_ref_exit(&ref_node->refs); | |
6588 | kfree(ref_node); | |
65e19f54 JA |
6589 | } |
6590 | ||
6b06314c JA |
6591 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6592 | unsigned nr_args) | |
6593 | { | |
6594 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6595 | unsigned nr_tables; |
05f3fb3c | 6596 | struct file *file; |
6b06314c JA |
6597 | int fd, ret = 0; |
6598 | unsigned i; | |
05589553 XW |
6599 | struct fixed_file_ref_node *ref_node; |
6600 | unsigned long flags; | |
6b06314c | 6601 | |
05f3fb3c | 6602 | if (ctx->file_data) |
6b06314c JA |
6603 | return -EBUSY; |
6604 | if (!nr_args) | |
6605 | return -EINVAL; | |
6606 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6607 | return -EMFILE; | |
6608 | ||
05f3fb3c JA |
6609 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6610 | if (!ctx->file_data) | |
6611 | return -ENOMEM; | |
6612 | ctx->file_data->ctx = ctx; | |
6613 | init_completion(&ctx->file_data->done); | |
05589553 | 6614 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6615 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6616 | |
65e19f54 | 6617 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6618 | ctx->file_data->table = kcalloc(nr_tables, |
6619 | sizeof(struct fixed_file_table), | |
65e19f54 | 6620 | GFP_KERNEL); |
05f3fb3c JA |
6621 | if (!ctx->file_data->table) { |
6622 | kfree(ctx->file_data); | |
6623 | ctx->file_data = NULL; | |
6b06314c | 6624 | return -ENOMEM; |
05f3fb3c JA |
6625 | } |
6626 | ||
05589553 | 6627 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6628 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6629 | kfree(ctx->file_data->table); | |
6630 | kfree(ctx->file_data); | |
6631 | ctx->file_data = NULL; | |
6b06314c | 6632 | return -ENOMEM; |
05f3fb3c | 6633 | } |
6b06314c | 6634 | |
65e19f54 | 6635 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6636 | percpu_ref_exit(&ctx->file_data->refs); |
6637 | kfree(ctx->file_data->table); | |
6638 | kfree(ctx->file_data); | |
6639 | ctx->file_data = NULL; | |
65e19f54 JA |
6640 | return -ENOMEM; |
6641 | } | |
6642 | ||
08a45173 | 6643 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6644 | struct fixed_file_table *table; |
6645 | unsigned index; | |
6646 | ||
6b06314c JA |
6647 | ret = -EFAULT; |
6648 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6649 | break; | |
08a45173 JA |
6650 | /* allow sparse sets */ |
6651 | if (fd == -1) { | |
6652 | ret = 0; | |
6653 | continue; | |
6654 | } | |
6b06314c | 6655 | |
05f3fb3c | 6656 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6657 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6658 | file = fget(fd); |
6b06314c JA |
6659 | |
6660 | ret = -EBADF; | |
05f3fb3c | 6661 | if (!file) |
6b06314c | 6662 | break; |
05f3fb3c | 6663 | |
6b06314c JA |
6664 | /* |
6665 | * Don't allow io_uring instances to be registered. If UNIX | |
6666 | * isn't enabled, then this causes a reference cycle and this | |
6667 | * instance can never get freed. If UNIX is enabled we'll | |
6668 | * handle it just fine, but there's still no point in allowing | |
6669 | * a ring fd as it doesn't support regular read/write anyway. | |
6670 | */ | |
05f3fb3c JA |
6671 | if (file->f_op == &io_uring_fops) { |
6672 | fput(file); | |
6b06314c JA |
6673 | break; |
6674 | } | |
6b06314c | 6675 | ret = 0; |
05f3fb3c | 6676 | table->files[index] = file; |
6b06314c JA |
6677 | } |
6678 | ||
6679 | if (ret) { | |
65e19f54 | 6680 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6681 | file = io_file_from_index(ctx, i); |
6682 | if (file) | |
6683 | fput(file); | |
6684 | } | |
6685 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6686 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6687 | |
05f3fb3c JA |
6688 | kfree(ctx->file_data->table); |
6689 | kfree(ctx->file_data); | |
6690 | ctx->file_data = NULL; | |
6b06314c JA |
6691 | ctx->nr_user_files = 0; |
6692 | return ret; | |
6693 | } | |
6694 | ||
6695 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6696 | if (ret) { |
6b06314c | 6697 | io_sqe_files_unregister(ctx); |
05589553 XW |
6698 | return ret; |
6699 | } | |
6b06314c | 6700 | |
05589553 XW |
6701 | ref_node = alloc_fixed_file_ref_node(ctx); |
6702 | if (IS_ERR(ref_node)) { | |
6703 | io_sqe_files_unregister(ctx); | |
6704 | return PTR_ERR(ref_node); | |
6705 | } | |
6706 | ||
6707 | ctx->file_data->cur_refs = &ref_node->refs; | |
6708 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6709 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6710 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6711 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6712 | return ret; |
6713 | } | |
6714 | ||
c3a31e60 JA |
6715 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6716 | int index) | |
6717 | { | |
6718 | #if defined(CONFIG_UNIX) | |
6719 | struct sock *sock = ctx->ring_sock->sk; | |
6720 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6721 | struct sk_buff *skb; | |
6722 | ||
6723 | /* | |
6724 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6725 | * file set. If there's no room, fall back to allocating a new skb | |
6726 | * and filling it in. | |
6727 | */ | |
6728 | spin_lock_irq(&head->lock); | |
6729 | skb = skb_peek(head); | |
6730 | if (skb) { | |
6731 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6732 | ||
6733 | if (fpl->count < SCM_MAX_FD) { | |
6734 | __skb_unlink(skb, head); | |
6735 | spin_unlock_irq(&head->lock); | |
6736 | fpl->fp[fpl->count] = get_file(file); | |
6737 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6738 | fpl->count++; | |
6739 | spin_lock_irq(&head->lock); | |
6740 | __skb_queue_head(head, skb); | |
6741 | } else { | |
6742 | skb = NULL; | |
6743 | } | |
6744 | } | |
6745 | spin_unlock_irq(&head->lock); | |
6746 | ||
6747 | if (skb) { | |
6748 | fput(file); | |
6749 | return 0; | |
6750 | } | |
6751 | ||
6752 | return __io_sqe_files_scm(ctx, 1, index); | |
6753 | #else | |
6754 | return 0; | |
6755 | #endif | |
6756 | } | |
6757 | ||
a5318d3c | 6758 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6759 | struct file *file) |
05f3fb3c | 6760 | { |
a5318d3c | 6761 | struct io_file_put *pfile; |
05589553 XW |
6762 | struct percpu_ref *refs = data->cur_refs; |
6763 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6764 | |
05f3fb3c | 6765 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6766 | if (!pfile) |
6767 | return -ENOMEM; | |
05f3fb3c | 6768 | |
05589553 | 6769 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6770 | pfile->file = file; |
05589553 XW |
6771 | list_add(&pfile->list, &ref_node->file_list); |
6772 | ||
a5318d3c | 6773 | return 0; |
05f3fb3c JA |
6774 | } |
6775 | ||
6776 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6777 | struct io_uring_files_update *up, | |
6778 | unsigned nr_args) | |
6779 | { | |
6780 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6781 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6782 | struct file *file; |
c3a31e60 JA |
6783 | __s32 __user *fds; |
6784 | int fd, i, err; | |
6785 | __u32 done; | |
05589553 XW |
6786 | unsigned long flags; |
6787 | bool needs_switch = false; | |
c3a31e60 | 6788 | |
05f3fb3c | 6789 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6790 | return -EOVERFLOW; |
6791 | if (done > ctx->nr_user_files) | |
6792 | return -EINVAL; | |
6793 | ||
05589553 XW |
6794 | ref_node = alloc_fixed_file_ref_node(ctx); |
6795 | if (IS_ERR(ref_node)) | |
6796 | return PTR_ERR(ref_node); | |
6797 | ||
c3a31e60 | 6798 | done = 0; |
05f3fb3c | 6799 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6800 | while (nr_args) { |
65e19f54 JA |
6801 | struct fixed_file_table *table; |
6802 | unsigned index; | |
6803 | ||
c3a31e60 JA |
6804 | err = 0; |
6805 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6806 | err = -EFAULT; | |
6807 | break; | |
6808 | } | |
05f3fb3c JA |
6809 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6810 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6811 | index = i & IORING_FILE_TABLE_MASK; |
6812 | if (table->files[index]) { | |
05f3fb3c | 6813 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6814 | err = io_queue_file_removal(data, file); |
6815 | if (err) | |
6816 | break; | |
65e19f54 | 6817 | table->files[index] = NULL; |
05589553 | 6818 | needs_switch = true; |
c3a31e60 JA |
6819 | } |
6820 | if (fd != -1) { | |
c3a31e60 JA |
6821 | file = fget(fd); |
6822 | if (!file) { | |
6823 | err = -EBADF; | |
6824 | break; | |
6825 | } | |
6826 | /* | |
6827 | * Don't allow io_uring instances to be registered. If | |
6828 | * UNIX isn't enabled, then this causes a reference | |
6829 | * cycle and this instance can never get freed. If UNIX | |
6830 | * is enabled we'll handle it just fine, but there's | |
6831 | * still no point in allowing a ring fd as it doesn't | |
6832 | * support regular read/write anyway. | |
6833 | */ | |
6834 | if (file->f_op == &io_uring_fops) { | |
6835 | fput(file); | |
6836 | err = -EBADF; | |
6837 | break; | |
6838 | } | |
65e19f54 | 6839 | table->files[index] = file; |
c3a31e60 JA |
6840 | err = io_sqe_file_register(ctx, file, i); |
6841 | if (err) | |
6842 | break; | |
6843 | } | |
6844 | nr_args--; | |
6845 | done++; | |
05f3fb3c JA |
6846 | up->offset++; |
6847 | } | |
6848 | ||
05589553 XW |
6849 | if (needs_switch) { |
6850 | percpu_ref_kill(data->cur_refs); | |
6851 | spin_lock_irqsave(&data->lock, flags); | |
6852 | list_add(&ref_node->node, &data->ref_list); | |
6853 | data->cur_refs = &ref_node->refs; | |
6854 | spin_unlock_irqrestore(&data->lock, flags); | |
6855 | percpu_ref_get(&ctx->file_data->refs); | |
6856 | } else | |
6857 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6858 | |
6859 | return done ? done : err; | |
6860 | } | |
05589553 | 6861 | |
05f3fb3c JA |
6862 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6863 | unsigned nr_args) | |
6864 | { | |
6865 | struct io_uring_files_update up; | |
6866 | ||
6867 | if (!ctx->file_data) | |
6868 | return -ENXIO; | |
6869 | if (!nr_args) | |
6870 | return -EINVAL; | |
6871 | if (copy_from_user(&up, arg, sizeof(up))) | |
6872 | return -EFAULT; | |
6873 | if (up.resv) | |
6874 | return -EINVAL; | |
6875 | ||
6876 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6877 | } | |
c3a31e60 | 6878 | |
e9fd9396 | 6879 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6880 | { |
6881 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6882 | ||
e9fd9396 | 6883 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6884 | io_put_req(req); |
6885 | } | |
6886 | ||
24369c2e PB |
6887 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6888 | struct io_uring_params *p) | |
6889 | { | |
6890 | struct io_wq_data data; | |
6891 | struct fd f; | |
6892 | struct io_ring_ctx *ctx_attach; | |
6893 | unsigned int concurrency; | |
6894 | int ret = 0; | |
6895 | ||
6896 | data.user = ctx->user; | |
e9fd9396 | 6897 | data.free_work = io_free_work; |
24369c2e PB |
6898 | |
6899 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6900 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6901 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6902 | ||
6903 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6904 | if (IS_ERR(ctx->io_wq)) { | |
6905 | ret = PTR_ERR(ctx->io_wq); | |
6906 | ctx->io_wq = NULL; | |
6907 | } | |
6908 | return ret; | |
6909 | } | |
6910 | ||
6911 | f = fdget(p->wq_fd); | |
6912 | if (!f.file) | |
6913 | return -EBADF; | |
6914 | ||
6915 | if (f.file->f_op != &io_uring_fops) { | |
6916 | ret = -EINVAL; | |
6917 | goto out_fput; | |
6918 | } | |
6919 | ||
6920 | ctx_attach = f.file->private_data; | |
6921 | /* @io_wq is protected by holding the fd */ | |
6922 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6923 | ret = -EINVAL; | |
6924 | goto out_fput; | |
6925 | } | |
6926 | ||
6927 | ctx->io_wq = ctx_attach->io_wq; | |
6928 | out_fput: | |
6929 | fdput(f); | |
6930 | return ret; | |
6931 | } | |
6932 | ||
6c271ce2 JA |
6933 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6934 | struct io_uring_params *p) | |
2b188cc1 JA |
6935 | { |
6936 | int ret; | |
6937 | ||
6c271ce2 | 6938 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6939 | mmgrab(current->mm); |
6940 | ctx->sqo_mm = current->mm; | |
6941 | ||
6c271ce2 | 6942 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6943 | ret = -EPERM; |
6944 | if (!capable(CAP_SYS_ADMIN)) | |
6945 | goto err; | |
6946 | ||
917257da JA |
6947 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6948 | if (!ctx->sq_thread_idle) | |
6949 | ctx->sq_thread_idle = HZ; | |
6950 | ||
6c271ce2 | 6951 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6952 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6953 | |
917257da | 6954 | ret = -EINVAL; |
44a9bd18 JA |
6955 | if (cpu >= nr_cpu_ids) |
6956 | goto err; | |
7889f44d | 6957 | if (!cpu_online(cpu)) |
917257da JA |
6958 | goto err; |
6959 | ||
6c271ce2 JA |
6960 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6961 | ctx, cpu, | |
6962 | "io_uring-sq"); | |
6963 | } else { | |
6964 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6965 | "io_uring-sq"); | |
6966 | } | |
6967 | if (IS_ERR(ctx->sqo_thread)) { | |
6968 | ret = PTR_ERR(ctx->sqo_thread); | |
6969 | ctx->sqo_thread = NULL; | |
6970 | goto err; | |
6971 | } | |
6972 | wake_up_process(ctx->sqo_thread); | |
6973 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6974 | /* Can't have SQ_AFF without SQPOLL */ | |
6975 | ret = -EINVAL; | |
6976 | goto err; | |
6977 | } | |
6978 | ||
24369c2e PB |
6979 | ret = io_init_wq_offload(ctx, p); |
6980 | if (ret) | |
2b188cc1 | 6981 | goto err; |
2b188cc1 JA |
6982 | |
6983 | return 0; | |
6984 | err: | |
54a91f3b | 6985 | io_finish_async(ctx); |
2b188cc1 JA |
6986 | mmdrop(ctx->sqo_mm); |
6987 | ctx->sqo_mm = NULL; | |
6988 | return ret; | |
6989 | } | |
6990 | ||
6991 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6992 | { | |
6993 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6994 | } | |
6995 | ||
6996 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6997 | { | |
6998 | unsigned long page_limit, cur_pages, new_pages; | |
6999 | ||
7000 | /* Don't allow more pages than we can safely lock */ | |
7001 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7002 | ||
7003 | do { | |
7004 | cur_pages = atomic_long_read(&user->locked_vm); | |
7005 | new_pages = cur_pages + nr_pages; | |
7006 | if (new_pages > page_limit) | |
7007 | return -ENOMEM; | |
7008 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7009 | new_pages) != cur_pages); | |
7010 | ||
7011 | return 0; | |
7012 | } | |
7013 | ||
7014 | static void io_mem_free(void *ptr) | |
7015 | { | |
52e04ef4 MR |
7016 | struct page *page; |
7017 | ||
7018 | if (!ptr) | |
7019 | return; | |
2b188cc1 | 7020 | |
52e04ef4 | 7021 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7022 | if (put_page_testzero(page)) |
7023 | free_compound_page(page); | |
7024 | } | |
7025 | ||
7026 | static void *io_mem_alloc(size_t size) | |
7027 | { | |
7028 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7029 | __GFP_NORETRY; | |
7030 | ||
7031 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7032 | } | |
7033 | ||
75b28aff HV |
7034 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7035 | size_t *sq_offset) | |
7036 | { | |
7037 | struct io_rings *rings; | |
7038 | size_t off, sq_array_size; | |
7039 | ||
7040 | off = struct_size(rings, cqes, cq_entries); | |
7041 | if (off == SIZE_MAX) | |
7042 | return SIZE_MAX; | |
7043 | ||
7044 | #ifdef CONFIG_SMP | |
7045 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7046 | if (off == 0) | |
7047 | return SIZE_MAX; | |
7048 | #endif | |
7049 | ||
7050 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
7051 | if (sq_array_size == SIZE_MAX) | |
7052 | return SIZE_MAX; | |
7053 | ||
7054 | if (check_add_overflow(off, sq_array_size, &off)) | |
7055 | return SIZE_MAX; | |
7056 | ||
7057 | if (sq_offset) | |
7058 | *sq_offset = off; | |
7059 | ||
7060 | return off; | |
7061 | } | |
7062 | ||
2b188cc1 JA |
7063 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7064 | { | |
75b28aff | 7065 | size_t pages; |
2b188cc1 | 7066 | |
75b28aff HV |
7067 | pages = (size_t)1 << get_order( |
7068 | rings_size(sq_entries, cq_entries, NULL)); | |
7069 | pages += (size_t)1 << get_order( | |
7070 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7071 | |
75b28aff | 7072 | return pages; |
2b188cc1 JA |
7073 | } |
7074 | ||
edafccee JA |
7075 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7076 | { | |
7077 | int i, j; | |
7078 | ||
7079 | if (!ctx->user_bufs) | |
7080 | return -ENXIO; | |
7081 | ||
7082 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7083 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7084 | ||
7085 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7086 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
7087 | |
7088 | if (ctx->account_mem) | |
7089 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 7090 | kvfree(imu->bvec); |
edafccee JA |
7091 | imu->nr_bvecs = 0; |
7092 | } | |
7093 | ||
7094 | kfree(ctx->user_bufs); | |
7095 | ctx->user_bufs = NULL; | |
7096 | ctx->nr_user_bufs = 0; | |
7097 | return 0; | |
7098 | } | |
7099 | ||
7100 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7101 | void __user *arg, unsigned index) | |
7102 | { | |
7103 | struct iovec __user *src; | |
7104 | ||
7105 | #ifdef CONFIG_COMPAT | |
7106 | if (ctx->compat) { | |
7107 | struct compat_iovec __user *ciovs; | |
7108 | struct compat_iovec ciov; | |
7109 | ||
7110 | ciovs = (struct compat_iovec __user *) arg; | |
7111 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7112 | return -EFAULT; | |
7113 | ||
d55e5f5b | 7114 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7115 | dst->iov_len = ciov.iov_len; |
7116 | return 0; | |
7117 | } | |
7118 | #endif | |
7119 | src = (struct iovec __user *) arg; | |
7120 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7121 | return -EFAULT; | |
7122 | return 0; | |
7123 | } | |
7124 | ||
7125 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7126 | unsigned nr_args) | |
7127 | { | |
7128 | struct vm_area_struct **vmas = NULL; | |
7129 | struct page **pages = NULL; | |
7130 | int i, j, got_pages = 0; | |
7131 | int ret = -EINVAL; | |
7132 | ||
7133 | if (ctx->user_bufs) | |
7134 | return -EBUSY; | |
7135 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7136 | return -EINVAL; | |
7137 | ||
7138 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7139 | GFP_KERNEL); | |
7140 | if (!ctx->user_bufs) | |
7141 | return -ENOMEM; | |
7142 | ||
7143 | for (i = 0; i < nr_args; i++) { | |
7144 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7145 | unsigned long off, start, end, ubuf; | |
7146 | int pret, nr_pages; | |
7147 | struct iovec iov; | |
7148 | size_t size; | |
7149 | ||
7150 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7151 | if (ret) | |
a278682d | 7152 | goto err; |
edafccee JA |
7153 | |
7154 | /* | |
7155 | * Don't impose further limits on the size and buffer | |
7156 | * constraints here, we'll -EINVAL later when IO is | |
7157 | * submitted if they are wrong. | |
7158 | */ | |
7159 | ret = -EFAULT; | |
7160 | if (!iov.iov_base || !iov.iov_len) | |
7161 | goto err; | |
7162 | ||
7163 | /* arbitrary limit, but we need something */ | |
7164 | if (iov.iov_len > SZ_1G) | |
7165 | goto err; | |
7166 | ||
7167 | ubuf = (unsigned long) iov.iov_base; | |
7168 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7169 | start = ubuf >> PAGE_SHIFT; | |
7170 | nr_pages = end - start; | |
7171 | ||
7172 | if (ctx->account_mem) { | |
7173 | ret = io_account_mem(ctx->user, nr_pages); | |
7174 | if (ret) | |
7175 | goto err; | |
7176 | } | |
7177 | ||
7178 | ret = 0; | |
7179 | if (!pages || nr_pages > got_pages) { | |
7180 | kfree(vmas); | |
7181 | kfree(pages); | |
d4ef6475 | 7182 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7183 | GFP_KERNEL); |
d4ef6475 | 7184 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7185 | sizeof(struct vm_area_struct *), |
7186 | GFP_KERNEL); | |
7187 | if (!pages || !vmas) { | |
7188 | ret = -ENOMEM; | |
7189 | if (ctx->account_mem) | |
7190 | io_unaccount_mem(ctx->user, nr_pages); | |
7191 | goto err; | |
7192 | } | |
7193 | got_pages = nr_pages; | |
7194 | } | |
7195 | ||
d4ef6475 | 7196 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7197 | GFP_KERNEL); |
7198 | ret = -ENOMEM; | |
7199 | if (!imu->bvec) { | |
7200 | if (ctx->account_mem) | |
7201 | io_unaccount_mem(ctx->user, nr_pages); | |
7202 | goto err; | |
7203 | } | |
7204 | ||
7205 | ret = 0; | |
7206 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7207 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7208 | FOLL_WRITE | FOLL_LONGTERM, |
7209 | pages, vmas); | |
edafccee JA |
7210 | if (pret == nr_pages) { |
7211 | /* don't support file backed memory */ | |
7212 | for (j = 0; j < nr_pages; j++) { | |
7213 | struct vm_area_struct *vma = vmas[j]; | |
7214 | ||
7215 | if (vma->vm_file && | |
7216 | !is_file_hugepages(vma->vm_file)) { | |
7217 | ret = -EOPNOTSUPP; | |
7218 | break; | |
7219 | } | |
7220 | } | |
7221 | } else { | |
7222 | ret = pret < 0 ? pret : -EFAULT; | |
7223 | } | |
7224 | up_read(¤t->mm->mmap_sem); | |
7225 | if (ret) { | |
7226 | /* | |
7227 | * if we did partial map, or found file backed vmas, | |
7228 | * release any pages we did get | |
7229 | */ | |
27c4d3a3 | 7230 | if (pret > 0) |
f1f6a7dd | 7231 | unpin_user_pages(pages, pret); |
edafccee JA |
7232 | if (ctx->account_mem) |
7233 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7234 | kvfree(imu->bvec); |
edafccee JA |
7235 | goto err; |
7236 | } | |
7237 | ||
7238 | off = ubuf & ~PAGE_MASK; | |
7239 | size = iov.iov_len; | |
7240 | for (j = 0; j < nr_pages; j++) { | |
7241 | size_t vec_len; | |
7242 | ||
7243 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7244 | imu->bvec[j].bv_page = pages[j]; | |
7245 | imu->bvec[j].bv_len = vec_len; | |
7246 | imu->bvec[j].bv_offset = off; | |
7247 | off = 0; | |
7248 | size -= vec_len; | |
7249 | } | |
7250 | /* store original address for later verification */ | |
7251 | imu->ubuf = ubuf; | |
7252 | imu->len = iov.iov_len; | |
7253 | imu->nr_bvecs = nr_pages; | |
7254 | ||
7255 | ctx->nr_user_bufs++; | |
7256 | } | |
d4ef6475 MR |
7257 | kvfree(pages); |
7258 | kvfree(vmas); | |
edafccee JA |
7259 | return 0; |
7260 | err: | |
d4ef6475 MR |
7261 | kvfree(pages); |
7262 | kvfree(vmas); | |
edafccee JA |
7263 | io_sqe_buffer_unregister(ctx); |
7264 | return ret; | |
7265 | } | |
7266 | ||
9b402849 JA |
7267 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7268 | { | |
7269 | __s32 __user *fds = arg; | |
7270 | int fd; | |
7271 | ||
7272 | if (ctx->cq_ev_fd) | |
7273 | return -EBUSY; | |
7274 | ||
7275 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7276 | return -EFAULT; | |
7277 | ||
7278 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7279 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7280 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7281 | ctx->cq_ev_fd = NULL; | |
7282 | return ret; | |
7283 | } | |
7284 | ||
7285 | return 0; | |
7286 | } | |
7287 | ||
7288 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7289 | { | |
7290 | if (ctx->cq_ev_fd) { | |
7291 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7292 | ctx->cq_ev_fd = NULL; | |
7293 | return 0; | |
7294 | } | |
7295 | ||
7296 | return -ENXIO; | |
7297 | } | |
7298 | ||
5a2e745d JA |
7299 | static int __io_destroy_buffers(int id, void *p, void *data) |
7300 | { | |
7301 | struct io_ring_ctx *ctx = data; | |
7302 | struct io_buffer *buf = p; | |
7303 | ||
067524e9 | 7304 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7305 | return 0; |
7306 | } | |
7307 | ||
7308 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7309 | { | |
7310 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7311 | idr_destroy(&ctx->io_buffer_idr); | |
7312 | } | |
7313 | ||
2b188cc1 JA |
7314 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7315 | { | |
6b06314c | 7316 | io_finish_async(ctx); |
2b188cc1 JA |
7317 | if (ctx->sqo_mm) |
7318 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7319 | |
7320 | io_iopoll_reap_events(ctx); | |
edafccee | 7321 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7322 | io_sqe_files_unregister(ctx); |
9b402849 | 7323 | io_eventfd_unregister(ctx); |
5a2e745d | 7324 | io_destroy_buffers(ctx); |
41726c9a | 7325 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7326 | |
2b188cc1 | 7327 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7328 | if (ctx->ring_sock) { |
7329 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7330 | sock_release(ctx->ring_sock); |
355e8d26 | 7331 | } |
2b188cc1 JA |
7332 | #endif |
7333 | ||
75b28aff | 7334 | io_mem_free(ctx->rings); |
2b188cc1 | 7335 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7336 | |
7337 | percpu_ref_exit(&ctx->refs); | |
7338 | if (ctx->account_mem) | |
7339 | io_unaccount_mem(ctx->user, | |
7340 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7341 | free_uid(ctx->user); | |
181e448d | 7342 | put_cred(ctx->creds); |
78076bb6 | 7343 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7344 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7345 | kfree(ctx); |
7346 | } | |
7347 | ||
7348 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7349 | { | |
7350 | struct io_ring_ctx *ctx = file->private_data; | |
7351 | __poll_t mask = 0; | |
7352 | ||
7353 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7354 | /* |
7355 | * synchronizes with barrier from wq_has_sleeper call in | |
7356 | * io_commit_cqring | |
7357 | */ | |
2b188cc1 | 7358 | smp_rmb(); |
75b28aff HV |
7359 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7360 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7361 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7362 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7363 | mask |= EPOLLIN | EPOLLRDNORM; |
7364 | ||
7365 | return mask; | |
7366 | } | |
7367 | ||
7368 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7369 | { | |
7370 | struct io_ring_ctx *ctx = file->private_data; | |
7371 | ||
7372 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7373 | } | |
7374 | ||
071698e1 JA |
7375 | static int io_remove_personalities(int id, void *p, void *data) |
7376 | { | |
7377 | struct io_ring_ctx *ctx = data; | |
7378 | const struct cred *cred; | |
7379 | ||
7380 | cred = idr_remove(&ctx->personality_idr, id); | |
7381 | if (cred) | |
7382 | put_cred(cred); | |
7383 | return 0; | |
7384 | } | |
7385 | ||
85faa7b8 JA |
7386 | static void io_ring_exit_work(struct work_struct *work) |
7387 | { | |
7388 | struct io_ring_ctx *ctx; | |
7389 | ||
7390 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7391 | if (ctx->rings) | |
7392 | io_cqring_overflow_flush(ctx, true); | |
7393 | ||
0f158b4c | 7394 | wait_for_completion(&ctx->ref_comp); |
85faa7b8 JA |
7395 | io_ring_ctx_free(ctx); |
7396 | } | |
7397 | ||
2b188cc1 JA |
7398 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7399 | { | |
7400 | mutex_lock(&ctx->uring_lock); | |
7401 | percpu_ref_kill(&ctx->refs); | |
7402 | mutex_unlock(&ctx->uring_lock); | |
7403 | ||
df069d80 JA |
7404 | /* |
7405 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7406 | * work after we've killed the ctx ref above. This is important to do | |
7407 | * before we cancel existing commands, as the thread could otherwise | |
7408 | * be queueing new work post that. If that's work we need to cancel, | |
7409 | * it could cause shutdown to hang. | |
7410 | */ | |
7411 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
3fd44c86 | 7412 | cond_resched(); |
df069d80 | 7413 | |
5262f567 | 7414 | io_kill_timeouts(ctx); |
221c5eb2 | 7415 | io_poll_remove_all(ctx); |
561fb04a JA |
7416 | |
7417 | if (ctx->io_wq) | |
7418 | io_wq_cancel_all(ctx->io_wq); | |
7419 | ||
def596e9 | 7420 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7421 | /* if we failed setting up the ctx, we might not have any rings */ |
7422 | if (ctx->rings) | |
7423 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7424 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7425 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7426 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7427 | } |
7428 | ||
7429 | static int io_uring_release(struct inode *inode, struct file *file) | |
7430 | { | |
7431 | struct io_ring_ctx *ctx = file->private_data; | |
7432 | ||
7433 | file->private_data = NULL; | |
7434 | io_ring_ctx_wait_and_kill(ctx); | |
7435 | return 0; | |
7436 | } | |
7437 | ||
fcb323cc JA |
7438 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7439 | struct files_struct *files) | |
7440 | { | |
fcb323cc | 7441 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7442 | struct io_kiocb *cancel_req = NULL, *req; |
7443 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7444 | |
7445 | spin_lock_irq(&ctx->inflight_lock); | |
7446 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7447 | if (req->work.files != files) |
7448 | continue; | |
7449 | /* req is being completed, ignore */ | |
7450 | if (!refcount_inc_not_zero(&req->refs)) | |
7451 | continue; | |
7452 | cancel_req = req; | |
7453 | break; | |
fcb323cc | 7454 | } |
768134d4 | 7455 | if (cancel_req) |
fcb323cc | 7456 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7457 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7458 | spin_unlock_irq(&ctx->inflight_lock); |
7459 | ||
768134d4 JA |
7460 | /* We need to keep going until we don't find a matching req */ |
7461 | if (!cancel_req) | |
fcb323cc | 7462 | break; |
2f6d9b9d | 7463 | |
2ca10259 JA |
7464 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7465 | spin_lock_irq(&ctx->completion_lock); | |
7466 | list_del(&cancel_req->list); | |
7467 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7468 | if (list_empty(&ctx->cq_overflow_list)) { | |
7469 | clear_bit(0, &ctx->sq_check_overflow); | |
7470 | clear_bit(0, &ctx->cq_check_overflow); | |
7471 | } | |
7472 | spin_unlock_irq(&ctx->completion_lock); | |
7473 | ||
7474 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7475 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7476 | ||
7477 | /* | |
7478 | * Put inflight ref and overflow ref. If that's | |
7479 | * all we had, then we're done with this request. | |
7480 | */ | |
7481 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7482 | io_put_req(cancel_req); | |
d8f1b971 | 7483 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7484 | continue; |
7485 | } | |
7486 | } | |
7487 | ||
2f6d9b9d BL |
7488 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7489 | io_put_req(cancel_req); | |
fcb323cc | 7490 | schedule(); |
d8f1b971 | 7491 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7492 | } |
7493 | } | |
7494 | ||
7495 | static int io_uring_flush(struct file *file, void *data) | |
7496 | { | |
7497 | struct io_ring_ctx *ctx = file->private_data; | |
7498 | ||
7499 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7500 | |
7501 | /* | |
7502 | * If the task is going away, cancel work it may have pending | |
7503 | */ | |
7504 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7505 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7506 | ||
fcb323cc JA |
7507 | return 0; |
7508 | } | |
7509 | ||
6c5c240e RP |
7510 | static void *io_uring_validate_mmap_request(struct file *file, |
7511 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7512 | { |
2b188cc1 | 7513 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7514 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7515 | struct page *page; |
7516 | void *ptr; | |
7517 | ||
7518 | switch (offset) { | |
7519 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7520 | case IORING_OFF_CQ_RING: |
7521 | ptr = ctx->rings; | |
2b188cc1 JA |
7522 | break; |
7523 | case IORING_OFF_SQES: | |
7524 | ptr = ctx->sq_sqes; | |
7525 | break; | |
2b188cc1 | 7526 | default: |
6c5c240e | 7527 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7528 | } |
7529 | ||
7530 | page = virt_to_head_page(ptr); | |
a50b854e | 7531 | if (sz > page_size(page)) |
6c5c240e RP |
7532 | return ERR_PTR(-EINVAL); |
7533 | ||
7534 | return ptr; | |
7535 | } | |
7536 | ||
7537 | #ifdef CONFIG_MMU | |
7538 | ||
7539 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7540 | { | |
7541 | size_t sz = vma->vm_end - vma->vm_start; | |
7542 | unsigned long pfn; | |
7543 | void *ptr; | |
7544 | ||
7545 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7546 | if (IS_ERR(ptr)) | |
7547 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7548 | |
7549 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7550 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7551 | } | |
7552 | ||
6c5c240e RP |
7553 | #else /* !CONFIG_MMU */ |
7554 | ||
7555 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7556 | { | |
7557 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7558 | } | |
7559 | ||
7560 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7561 | { | |
7562 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7563 | } | |
7564 | ||
7565 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7566 | unsigned long addr, unsigned long len, | |
7567 | unsigned long pgoff, unsigned long flags) | |
7568 | { | |
7569 | void *ptr; | |
7570 | ||
7571 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7572 | if (IS_ERR(ptr)) | |
7573 | return PTR_ERR(ptr); | |
7574 | ||
7575 | return (unsigned long) ptr; | |
7576 | } | |
7577 | ||
7578 | #endif /* !CONFIG_MMU */ | |
7579 | ||
2b188cc1 JA |
7580 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7581 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7582 | size_t, sigsz) | |
7583 | { | |
7584 | struct io_ring_ctx *ctx; | |
7585 | long ret = -EBADF; | |
7586 | int submitted = 0; | |
7587 | struct fd f; | |
7588 | ||
b41e9852 JA |
7589 | if (current->task_works) |
7590 | task_work_run(); | |
7591 | ||
6c271ce2 | 7592 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7593 | return -EINVAL; |
7594 | ||
7595 | f = fdget(fd); | |
7596 | if (!f.file) | |
7597 | return -EBADF; | |
7598 | ||
7599 | ret = -EOPNOTSUPP; | |
7600 | if (f.file->f_op != &io_uring_fops) | |
7601 | goto out_fput; | |
7602 | ||
7603 | ret = -ENXIO; | |
7604 | ctx = f.file->private_data; | |
7605 | if (!percpu_ref_tryget(&ctx->refs)) | |
7606 | goto out_fput; | |
7607 | ||
6c271ce2 JA |
7608 | /* |
7609 | * For SQ polling, the thread will do all submissions and completions. | |
7610 | * Just return the requested submit count, and wake the thread if | |
7611 | * we were asked to. | |
7612 | */ | |
b2a9eada | 7613 | ret = 0; |
6c271ce2 | 7614 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7615 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7616 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7617 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7618 | wake_up(&ctx->sqo_wait); | |
7619 | submitted = to_submit; | |
b2a9eada | 7620 | } else if (to_submit) { |
2b188cc1 | 7621 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7622 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7623 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7624 | |
7625 | if (submitted != to_submit) | |
7626 | goto out; | |
2b188cc1 JA |
7627 | } |
7628 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7629 | unsigned nr_events = 0; |
7630 | ||
2b188cc1 JA |
7631 | min_complete = min(min_complete, ctx->cq_entries); |
7632 | ||
32b2244a XW |
7633 | /* |
7634 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7635 | * space applications don't need to do io completion events | |
7636 | * polling again, they can rely on io_sq_thread to do polling | |
7637 | * work, which can reduce cpu usage and uring_lock contention. | |
7638 | */ | |
7639 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7640 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7641 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7642 | } else { |
7643 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7644 | } | |
2b188cc1 JA |
7645 | } |
7646 | ||
7c504e65 | 7647 | out: |
6805b32e | 7648 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7649 | out_fput: |
7650 | fdput(f); | |
7651 | return submitted ? submitted : ret; | |
7652 | } | |
7653 | ||
bebdb65e | 7654 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7655 | static int io_uring_show_cred(int id, void *p, void *data) |
7656 | { | |
7657 | const struct cred *cred = p; | |
7658 | struct seq_file *m = data; | |
7659 | struct user_namespace *uns = seq_user_ns(m); | |
7660 | struct group_info *gi; | |
7661 | kernel_cap_t cap; | |
7662 | unsigned __capi; | |
7663 | int g; | |
7664 | ||
7665 | seq_printf(m, "%5d\n", id); | |
7666 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7667 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7668 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7669 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7670 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7671 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7672 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7673 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7674 | seq_puts(m, "\n\tGroups:\t"); | |
7675 | gi = cred->group_info; | |
7676 | for (g = 0; g < gi->ngroups; g++) { | |
7677 | seq_put_decimal_ull(m, g ? " " : "", | |
7678 | from_kgid_munged(uns, gi->gid[g])); | |
7679 | } | |
7680 | seq_puts(m, "\n\tCapEff:\t"); | |
7681 | cap = cred->cap_effective; | |
7682 | CAP_FOR_EACH_U32(__capi) | |
7683 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7684 | seq_putc(m, '\n'); | |
7685 | return 0; | |
7686 | } | |
7687 | ||
7688 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7689 | { | |
7690 | int i; | |
7691 | ||
7692 | mutex_lock(&ctx->uring_lock); | |
7693 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7694 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7695 | struct fixed_file_table *table; | |
7696 | struct file *f; | |
7697 | ||
7698 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7699 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7700 | if (f) | |
7701 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7702 | else | |
7703 | seq_printf(m, "%5u: <none>\n", i); | |
7704 | } | |
7705 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7706 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7707 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7708 | ||
7709 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7710 | (unsigned int) buf->len); | |
7711 | } | |
7712 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7713 | seq_printf(m, "Personalities:\n"); | |
7714 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7715 | } | |
d7718a9d JA |
7716 | seq_printf(m, "PollList:\n"); |
7717 | spin_lock_irq(&ctx->completion_lock); | |
7718 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7719 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7720 | struct io_kiocb *req; | |
7721 | ||
7722 | hlist_for_each_entry(req, list, hash_node) | |
7723 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7724 | req->task->task_works != NULL); | |
7725 | } | |
7726 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7727 | mutex_unlock(&ctx->uring_lock); |
7728 | } | |
7729 | ||
7730 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7731 | { | |
7732 | struct io_ring_ctx *ctx = f->private_data; | |
7733 | ||
7734 | if (percpu_ref_tryget(&ctx->refs)) { | |
7735 | __io_uring_show_fdinfo(ctx, m); | |
7736 | percpu_ref_put(&ctx->refs); | |
7737 | } | |
7738 | } | |
bebdb65e | 7739 | #endif |
87ce955b | 7740 | |
2b188cc1 JA |
7741 | static const struct file_operations io_uring_fops = { |
7742 | .release = io_uring_release, | |
fcb323cc | 7743 | .flush = io_uring_flush, |
2b188cc1 | 7744 | .mmap = io_uring_mmap, |
6c5c240e RP |
7745 | #ifndef CONFIG_MMU |
7746 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7747 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7748 | #endif | |
2b188cc1 JA |
7749 | .poll = io_uring_poll, |
7750 | .fasync = io_uring_fasync, | |
bebdb65e | 7751 | #ifdef CONFIG_PROC_FS |
87ce955b | 7752 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7753 | #endif |
2b188cc1 JA |
7754 | }; |
7755 | ||
7756 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7757 | struct io_uring_params *p) | |
7758 | { | |
75b28aff HV |
7759 | struct io_rings *rings; |
7760 | size_t size, sq_array_offset; | |
2b188cc1 | 7761 | |
75b28aff HV |
7762 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7763 | if (size == SIZE_MAX) | |
7764 | return -EOVERFLOW; | |
7765 | ||
7766 | rings = io_mem_alloc(size); | |
7767 | if (!rings) | |
2b188cc1 JA |
7768 | return -ENOMEM; |
7769 | ||
75b28aff HV |
7770 | ctx->rings = rings; |
7771 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7772 | rings->sq_ring_mask = p->sq_entries - 1; | |
7773 | rings->cq_ring_mask = p->cq_entries - 1; | |
7774 | rings->sq_ring_entries = p->sq_entries; | |
7775 | rings->cq_ring_entries = p->cq_entries; | |
7776 | ctx->sq_mask = rings->sq_ring_mask; | |
7777 | ctx->cq_mask = rings->cq_ring_mask; | |
7778 | ctx->sq_entries = rings->sq_ring_entries; | |
7779 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7780 | |
7781 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7782 | if (size == SIZE_MAX) { |
7783 | io_mem_free(ctx->rings); | |
7784 | ctx->rings = NULL; | |
2b188cc1 | 7785 | return -EOVERFLOW; |
eb065d30 | 7786 | } |
2b188cc1 JA |
7787 | |
7788 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7789 | if (!ctx->sq_sqes) { |
7790 | io_mem_free(ctx->rings); | |
7791 | ctx->rings = NULL; | |
2b188cc1 | 7792 | return -ENOMEM; |
eb065d30 | 7793 | } |
2b188cc1 | 7794 | |
2b188cc1 JA |
7795 | return 0; |
7796 | } | |
7797 | ||
7798 | /* | |
7799 | * Allocate an anonymous fd, this is what constitutes the application | |
7800 | * visible backing of an io_uring instance. The application mmaps this | |
7801 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7802 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7803 | */ | |
7804 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7805 | { | |
7806 | struct file *file; | |
7807 | int ret; | |
7808 | ||
7809 | #if defined(CONFIG_UNIX) | |
7810 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7811 | &ctx->ring_sock); | |
7812 | if (ret) | |
7813 | return ret; | |
7814 | #endif | |
7815 | ||
7816 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7817 | if (ret < 0) | |
7818 | goto err; | |
7819 | ||
7820 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7821 | O_RDWR | O_CLOEXEC); | |
7822 | if (IS_ERR(file)) { | |
7823 | put_unused_fd(ret); | |
7824 | ret = PTR_ERR(file); | |
7825 | goto err; | |
7826 | } | |
7827 | ||
7828 | #if defined(CONFIG_UNIX) | |
7829 | ctx->ring_sock->file = file; | |
7830 | #endif | |
7831 | fd_install(ret, file); | |
7832 | return ret; | |
7833 | err: | |
7834 | #if defined(CONFIG_UNIX) | |
7835 | sock_release(ctx->ring_sock); | |
7836 | ctx->ring_sock = NULL; | |
7837 | #endif | |
7838 | return ret; | |
7839 | } | |
7840 | ||
7f13657d XW |
7841 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
7842 | struct io_uring_params __user *params) | |
2b188cc1 JA |
7843 | { |
7844 | struct user_struct *user = NULL; | |
7845 | struct io_ring_ctx *ctx; | |
7846 | bool account_mem; | |
7847 | int ret; | |
7848 | ||
8110c1a6 | 7849 | if (!entries) |
2b188cc1 | 7850 | return -EINVAL; |
8110c1a6 JA |
7851 | if (entries > IORING_MAX_ENTRIES) { |
7852 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7853 | return -EINVAL; | |
7854 | entries = IORING_MAX_ENTRIES; | |
7855 | } | |
2b188cc1 JA |
7856 | |
7857 | /* | |
7858 | * Use twice as many entries for the CQ ring. It's possible for the | |
7859 | * application to drive a higher depth than the size of the SQ ring, | |
7860 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7861 | * some flexibility in overcommitting a bit. If the application has |
7862 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7863 | * of CQ ring entries manually. | |
2b188cc1 JA |
7864 | */ |
7865 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7866 | if (p->flags & IORING_SETUP_CQSIZE) { |
7867 | /* | |
7868 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7869 | * to a power-of-two, if it isn't already. We do NOT impose | |
7870 | * any cq vs sq ring sizing. | |
7871 | */ | |
8110c1a6 | 7872 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7873 | return -EINVAL; |
8110c1a6 JA |
7874 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7875 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7876 | return -EINVAL; | |
7877 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7878 | } | |
33a107f0 JA |
7879 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7880 | } else { | |
7881 | p->cq_entries = 2 * p->sq_entries; | |
7882 | } | |
2b188cc1 JA |
7883 | |
7884 | user = get_uid(current_user()); | |
7885 | account_mem = !capable(CAP_IPC_LOCK); | |
7886 | ||
7887 | if (account_mem) { | |
7888 | ret = io_account_mem(user, | |
7889 | ring_pages(p->sq_entries, p->cq_entries)); | |
7890 | if (ret) { | |
7891 | free_uid(user); | |
7892 | return ret; | |
7893 | } | |
7894 | } | |
7895 | ||
7896 | ctx = io_ring_ctx_alloc(p); | |
7897 | if (!ctx) { | |
7898 | if (account_mem) | |
7899 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7900 | p->cq_entries)); | |
7901 | free_uid(user); | |
7902 | return -ENOMEM; | |
7903 | } | |
7904 | ctx->compat = in_compat_syscall(); | |
7905 | ctx->account_mem = account_mem; | |
7906 | ctx->user = user; | |
0b8c0ec7 | 7907 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7908 | |
7909 | ret = io_allocate_scq_urings(ctx, p); | |
7910 | if (ret) | |
7911 | goto err; | |
7912 | ||
6c271ce2 | 7913 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7914 | if (ret) |
7915 | goto err; | |
7916 | ||
2b188cc1 | 7917 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7918 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7919 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7920 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7921 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7922 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7923 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7924 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7925 | |
7926 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7927 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7928 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7929 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7930 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7931 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7932 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7933 | |
7f13657d XW |
7934 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
7935 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
7936 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; | |
7937 | ||
7938 | if (copy_to_user(params, p, sizeof(*p))) { | |
7939 | ret = -EFAULT; | |
7940 | goto err; | |
7941 | } | |
044c1ab3 JA |
7942 | /* |
7943 | * Install ring fd as the very last thing, so we don't risk someone | |
7944 | * having closed it before we finish setup | |
7945 | */ | |
7946 | ret = io_uring_get_fd(ctx); | |
7947 | if (ret < 0) | |
7948 | goto err; | |
7949 | ||
c826bd7a | 7950 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7951 | return ret; |
7952 | err: | |
7953 | io_ring_ctx_wait_and_kill(ctx); | |
7954 | return ret; | |
7955 | } | |
7956 | ||
7957 | /* | |
7958 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7959 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7960 | * params structure passed in. | |
7961 | */ | |
7962 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7963 | { | |
7964 | struct io_uring_params p; | |
2b188cc1 JA |
7965 | int i; |
7966 | ||
7967 | if (copy_from_user(&p, params, sizeof(p))) | |
7968 | return -EFAULT; | |
7969 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7970 | if (p.resv[i]) | |
7971 | return -EINVAL; | |
7972 | } | |
7973 | ||
6c271ce2 | 7974 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7975 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7976 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7977 | return -EINVAL; |
7978 | ||
7f13657d | 7979 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
7980 | } |
7981 | ||
7982 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7983 | struct io_uring_params __user *, params) | |
7984 | { | |
7985 | return io_uring_setup(entries, params); | |
7986 | } | |
7987 | ||
66f4af93 JA |
7988 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7989 | { | |
7990 | struct io_uring_probe *p; | |
7991 | size_t size; | |
7992 | int i, ret; | |
7993 | ||
7994 | size = struct_size(p, ops, nr_args); | |
7995 | if (size == SIZE_MAX) | |
7996 | return -EOVERFLOW; | |
7997 | p = kzalloc(size, GFP_KERNEL); | |
7998 | if (!p) | |
7999 | return -ENOMEM; | |
8000 | ||
8001 | ret = -EFAULT; | |
8002 | if (copy_from_user(p, arg, size)) | |
8003 | goto out; | |
8004 | ret = -EINVAL; | |
8005 | if (memchr_inv(p, 0, size)) | |
8006 | goto out; | |
8007 | ||
8008 | p->last_op = IORING_OP_LAST - 1; | |
8009 | if (nr_args > IORING_OP_LAST) | |
8010 | nr_args = IORING_OP_LAST; | |
8011 | ||
8012 | for (i = 0; i < nr_args; i++) { | |
8013 | p->ops[i].op = i; | |
8014 | if (!io_op_defs[i].not_supported) | |
8015 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8016 | } | |
8017 | p->ops_len = i; | |
8018 | ||
8019 | ret = 0; | |
8020 | if (copy_to_user(arg, p, size)) | |
8021 | ret = -EFAULT; | |
8022 | out: | |
8023 | kfree(p); | |
8024 | return ret; | |
8025 | } | |
8026 | ||
071698e1 JA |
8027 | static int io_register_personality(struct io_ring_ctx *ctx) |
8028 | { | |
8029 | const struct cred *creds = get_current_cred(); | |
8030 | int id; | |
8031 | ||
8032 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8033 | USHRT_MAX, GFP_KERNEL); | |
8034 | if (id < 0) | |
8035 | put_cred(creds); | |
8036 | return id; | |
8037 | } | |
8038 | ||
8039 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8040 | { | |
8041 | const struct cred *old_creds; | |
8042 | ||
8043 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8044 | if (old_creds) { | |
8045 | put_cred(old_creds); | |
8046 | return 0; | |
8047 | } | |
8048 | ||
8049 | return -EINVAL; | |
8050 | } | |
8051 | ||
8052 | static bool io_register_op_must_quiesce(int op) | |
8053 | { | |
8054 | switch (op) { | |
8055 | case IORING_UNREGISTER_FILES: | |
8056 | case IORING_REGISTER_FILES_UPDATE: | |
8057 | case IORING_REGISTER_PROBE: | |
8058 | case IORING_REGISTER_PERSONALITY: | |
8059 | case IORING_UNREGISTER_PERSONALITY: | |
8060 | return false; | |
8061 | default: | |
8062 | return true; | |
8063 | } | |
8064 | } | |
8065 | ||
edafccee JA |
8066 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8067 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8068 | __releases(ctx->uring_lock) |
8069 | __acquires(ctx->uring_lock) | |
edafccee JA |
8070 | { |
8071 | int ret; | |
8072 | ||
35fa71a0 JA |
8073 | /* |
8074 | * We're inside the ring mutex, if the ref is already dying, then | |
8075 | * someone else killed the ctx or is already going through | |
8076 | * io_uring_register(). | |
8077 | */ | |
8078 | if (percpu_ref_is_dying(&ctx->refs)) | |
8079 | return -ENXIO; | |
8080 | ||
071698e1 | 8081 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8082 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8083 | |
05f3fb3c JA |
8084 | /* |
8085 | * Drop uring mutex before waiting for references to exit. If | |
8086 | * another thread is currently inside io_uring_enter() it might | |
8087 | * need to grab the uring_lock to make progress. If we hold it | |
8088 | * here across the drain wait, then we can deadlock. It's safe | |
8089 | * to drop the mutex here, since no new references will come in | |
8090 | * after we've killed the percpu ref. | |
8091 | */ | |
8092 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8093 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8094 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8095 | if (ret) { |
8096 | percpu_ref_resurrect(&ctx->refs); | |
8097 | ret = -EINTR; | |
8098 | goto out; | |
8099 | } | |
05f3fb3c | 8100 | } |
edafccee JA |
8101 | |
8102 | switch (opcode) { | |
8103 | case IORING_REGISTER_BUFFERS: | |
8104 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8105 | break; | |
8106 | case IORING_UNREGISTER_BUFFERS: | |
8107 | ret = -EINVAL; | |
8108 | if (arg || nr_args) | |
8109 | break; | |
8110 | ret = io_sqe_buffer_unregister(ctx); | |
8111 | break; | |
6b06314c JA |
8112 | case IORING_REGISTER_FILES: |
8113 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8114 | break; | |
8115 | case IORING_UNREGISTER_FILES: | |
8116 | ret = -EINVAL; | |
8117 | if (arg || nr_args) | |
8118 | break; | |
8119 | ret = io_sqe_files_unregister(ctx); | |
8120 | break; | |
c3a31e60 JA |
8121 | case IORING_REGISTER_FILES_UPDATE: |
8122 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8123 | break; | |
9b402849 | 8124 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8125 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8126 | ret = -EINVAL; |
8127 | if (nr_args != 1) | |
8128 | break; | |
8129 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8130 | if (ret) |
8131 | break; | |
8132 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8133 | ctx->eventfd_async = 1; | |
8134 | else | |
8135 | ctx->eventfd_async = 0; | |
9b402849 JA |
8136 | break; |
8137 | case IORING_UNREGISTER_EVENTFD: | |
8138 | ret = -EINVAL; | |
8139 | if (arg || nr_args) | |
8140 | break; | |
8141 | ret = io_eventfd_unregister(ctx); | |
8142 | break; | |
66f4af93 JA |
8143 | case IORING_REGISTER_PROBE: |
8144 | ret = -EINVAL; | |
8145 | if (!arg || nr_args > 256) | |
8146 | break; | |
8147 | ret = io_probe(ctx, arg, nr_args); | |
8148 | break; | |
071698e1 JA |
8149 | case IORING_REGISTER_PERSONALITY: |
8150 | ret = -EINVAL; | |
8151 | if (arg || nr_args) | |
8152 | break; | |
8153 | ret = io_register_personality(ctx); | |
8154 | break; | |
8155 | case IORING_UNREGISTER_PERSONALITY: | |
8156 | ret = -EINVAL; | |
8157 | if (arg) | |
8158 | break; | |
8159 | ret = io_unregister_personality(ctx, nr_args); | |
8160 | break; | |
edafccee JA |
8161 | default: |
8162 | ret = -EINVAL; | |
8163 | break; | |
8164 | } | |
8165 | ||
071698e1 | 8166 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8167 | /* bring the ctx back to life */ |
05f3fb3c | 8168 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8169 | out: |
0f158b4c | 8170 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8171 | } |
edafccee JA |
8172 | return ret; |
8173 | } | |
8174 | ||
8175 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8176 | void __user *, arg, unsigned int, nr_args) | |
8177 | { | |
8178 | struct io_ring_ctx *ctx; | |
8179 | long ret = -EBADF; | |
8180 | struct fd f; | |
8181 | ||
8182 | f = fdget(fd); | |
8183 | if (!f.file) | |
8184 | return -EBADF; | |
8185 | ||
8186 | ret = -EOPNOTSUPP; | |
8187 | if (f.file->f_op != &io_uring_fops) | |
8188 | goto out_fput; | |
8189 | ||
8190 | ctx = f.file->private_data; | |
8191 | ||
8192 | mutex_lock(&ctx->uring_lock); | |
8193 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8194 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8195 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8196 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8197 | out_fput: |
8198 | fdput(f); | |
8199 | return ret; | |
8200 | } | |
8201 | ||
2b188cc1 JA |
8202 | static int __init io_uring_init(void) |
8203 | { | |
d7f62e82 SM |
8204 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8205 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8206 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8207 | } while (0) | |
8208 | ||
8209 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8210 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8211 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8212 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8213 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8214 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8215 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8216 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8217 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8218 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8219 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8220 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8221 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8222 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8223 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8224 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8225 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8226 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8227 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8228 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8229 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8230 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8231 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8232 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8233 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8234 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8235 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8236 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8237 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8238 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8239 | |
d3656344 | 8240 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8241 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8242 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8243 | return 0; | |
8244 | }; | |
8245 | __initcall(io_uring_init); |