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> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
bcf5a063 | 81 | #include <linux/pagemap.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 | 144 | /* |
0d9b5b3a | 145 | * Runtime SQ flags |
1e84b97b SB |
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; |
0d9b5b3a SG |
154 | /* |
155 | * Runtime CQ flags | |
156 | * | |
157 | * Written by the application, shouldn't be modified by the | |
158 | * kernel. | |
159 | */ | |
160 | u32 cq_flags; | |
1e84b97b SB |
161 | /* |
162 | * Number of completion events lost because the queue was full; | |
163 | * this should be avoided by the application by making sure | |
0b4295b5 | 164 | * there are not more requests pending than there is space in |
1e84b97b SB |
165 | * the completion queue. |
166 | * | |
167 | * Written by the kernel, shouldn't be modified by the | |
168 | * application (i.e. get number of "new events" by comparing to | |
169 | * cached value). | |
170 | * | |
171 | * As completion events come in out of order this counter is not | |
172 | * ordered with any other data. | |
173 | */ | |
75b28aff | 174 | u32 cq_overflow; |
1e84b97b SB |
175 | /* |
176 | * Ring buffer of completion events. | |
177 | * | |
178 | * The kernel writes completion events fresh every time they are | |
179 | * produced, so the application is allowed to modify pending | |
180 | * entries. | |
181 | */ | |
75b28aff | 182 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
183 | }; |
184 | ||
edafccee JA |
185 | struct io_mapped_ubuf { |
186 | u64 ubuf; | |
187 | size_t len; | |
188 | struct bio_vec *bvec; | |
189 | unsigned int nr_bvecs; | |
190 | }; | |
191 | ||
65e19f54 JA |
192 | struct fixed_file_table { |
193 | struct file **files; | |
31b51510 JA |
194 | }; |
195 | ||
05589553 XW |
196 | struct fixed_file_ref_node { |
197 | struct percpu_ref refs; | |
198 | struct list_head node; | |
199 | struct list_head file_list; | |
200 | struct fixed_file_data *file_data; | |
4a38aed2 | 201 | struct llist_node llist; |
05589553 XW |
202 | }; |
203 | ||
05f3fb3c JA |
204 | struct fixed_file_data { |
205 | struct fixed_file_table *table; | |
206 | struct io_ring_ctx *ctx; | |
207 | ||
05589553 | 208 | struct percpu_ref *cur_refs; |
05f3fb3c | 209 | struct percpu_ref refs; |
05f3fb3c | 210 | struct completion done; |
05589553 XW |
211 | struct list_head ref_list; |
212 | spinlock_t lock; | |
05f3fb3c JA |
213 | }; |
214 | ||
5a2e745d JA |
215 | struct io_buffer { |
216 | struct list_head list; | |
217 | __u64 addr; | |
218 | __s32 len; | |
219 | __u16 bid; | |
220 | }; | |
221 | ||
2b188cc1 JA |
222 | struct io_ring_ctx { |
223 | struct { | |
224 | struct percpu_ref refs; | |
225 | } ____cacheline_aligned_in_smp; | |
226 | ||
227 | struct { | |
228 | unsigned int flags; | |
e1d85334 | 229 | unsigned int compat: 1; |
aad5d8da | 230 | unsigned int limit_mem: 1; |
e1d85334 RD |
231 | unsigned int cq_overflow_flushed: 1; |
232 | unsigned int drain_next: 1; | |
233 | unsigned int eventfd_async: 1; | |
2b188cc1 | 234 | |
75b28aff HV |
235 | /* |
236 | * Ring buffer of indices into array of io_uring_sqe, which is | |
237 | * mmapped by the application using the IORING_OFF_SQES offset. | |
238 | * | |
239 | * This indirection could e.g. be used to assign fixed | |
240 | * io_uring_sqe entries to operations and only submit them to | |
241 | * the queue when needed. | |
242 | * | |
243 | * The kernel modifies neither the indices array nor the entries | |
244 | * array. | |
245 | */ | |
246 | u32 *sq_array; | |
2b188cc1 JA |
247 | unsigned cached_sq_head; |
248 | unsigned sq_entries; | |
249 | unsigned sq_mask; | |
6c271ce2 | 250 | unsigned sq_thread_idle; |
498ccd9e | 251 | unsigned cached_sq_dropped; |
206aefde | 252 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 253 | unsigned long sq_check_overflow; |
de0617e4 JA |
254 | |
255 | struct list_head defer_list; | |
5262f567 | 256 | struct list_head timeout_list; |
1d7bb1d5 | 257 | struct list_head cq_overflow_list; |
fcb323cc JA |
258 | |
259 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 260 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
261 | } ____cacheline_aligned_in_smp; |
262 | ||
206aefde JA |
263 | struct io_rings *rings; |
264 | ||
2b188cc1 | 265 | /* IO offload */ |
561fb04a | 266 | struct io_wq *io_wq; |
6c271ce2 | 267 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 268 | struct mm_struct *sqo_mm; |
6c271ce2 | 269 | wait_queue_head_t sqo_wait; |
75b28aff | 270 | |
6b06314c JA |
271 | /* |
272 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
273 | * readers must ensure that ->refs is alive as long as the file* is | |
274 | * used. Only updated through io_uring_register(2). | |
275 | */ | |
05f3fb3c | 276 | struct fixed_file_data *file_data; |
6b06314c | 277 | unsigned nr_user_files; |
b14cca0c PB |
278 | int ring_fd; |
279 | struct file *ring_file; | |
6b06314c | 280 | |
edafccee JA |
281 | /* if used, fixed mapped user buffers */ |
282 | unsigned nr_user_bufs; | |
283 | struct io_mapped_ubuf *user_bufs; | |
284 | ||
2b188cc1 JA |
285 | struct user_struct *user; |
286 | ||
0b8c0ec7 | 287 | const struct cred *creds; |
181e448d | 288 | |
0f158b4c JA |
289 | struct completion ref_comp; |
290 | struct completion sq_thread_comp; | |
206aefde | 291 | |
0ddf92e8 JA |
292 | /* if all else fails... */ |
293 | struct io_kiocb *fallback_req; | |
294 | ||
206aefde JA |
295 | #if defined(CONFIG_UNIX) |
296 | struct socket *ring_sock; | |
297 | #endif | |
298 | ||
5a2e745d JA |
299 | struct idr io_buffer_idr; |
300 | ||
071698e1 JA |
301 | struct idr personality_idr; |
302 | ||
206aefde JA |
303 | struct { |
304 | unsigned cached_cq_tail; | |
305 | unsigned cq_entries; | |
306 | unsigned cq_mask; | |
307 | atomic_t cq_timeouts; | |
ad3eb2c8 | 308 | unsigned long cq_check_overflow; |
206aefde JA |
309 | struct wait_queue_head cq_wait; |
310 | struct fasync_struct *cq_fasync; | |
311 | struct eventfd_ctx *cq_ev_fd; | |
312 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
313 | |
314 | struct { | |
315 | struct mutex uring_lock; | |
316 | wait_queue_head_t wait; | |
317 | } ____cacheline_aligned_in_smp; | |
318 | ||
319 | struct { | |
320 | spinlock_t completion_lock; | |
e94f141b | 321 | |
def596e9 JA |
322 | /* |
323 | * ->poll_list is protected by the ctx->uring_lock for | |
324 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
325 | * For SQPOLL, only the single threaded io_sq_thread() will | |
326 | * manipulate the list, hence no extra locking is needed there. | |
327 | */ | |
328 | struct list_head poll_list; | |
78076bb6 JA |
329 | struct hlist_head *cancel_hash; |
330 | unsigned cancel_hash_bits; | |
e94f141b | 331 | bool poll_multi_file; |
31b51510 | 332 | |
fcb323cc JA |
333 | spinlock_t inflight_lock; |
334 | struct list_head inflight_list; | |
2b188cc1 | 335 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 336 | |
4a38aed2 JA |
337 | struct delayed_work file_put_work; |
338 | struct llist_head file_put_llist; | |
339 | ||
85faa7b8 | 340 | struct work_struct exit_work; |
2b188cc1 JA |
341 | }; |
342 | ||
09bb8394 JA |
343 | /* |
344 | * First field must be the file pointer in all the | |
345 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
346 | */ | |
221c5eb2 JA |
347 | struct io_poll_iocb { |
348 | struct file *file; | |
0969e783 JA |
349 | union { |
350 | struct wait_queue_head *head; | |
351 | u64 addr; | |
352 | }; | |
221c5eb2 | 353 | __poll_t events; |
8c838788 | 354 | bool done; |
221c5eb2 | 355 | bool canceled; |
392edb45 | 356 | struct wait_queue_entry wait; |
221c5eb2 JA |
357 | }; |
358 | ||
b5dba59e JA |
359 | struct io_close { |
360 | struct file *file; | |
361 | struct file *put_file; | |
362 | int fd; | |
363 | }; | |
364 | ||
ad8a48ac JA |
365 | struct io_timeout_data { |
366 | struct io_kiocb *req; | |
367 | struct hrtimer timer; | |
368 | struct timespec64 ts; | |
369 | enum hrtimer_mode mode; | |
370 | }; | |
371 | ||
8ed8d3c3 JA |
372 | struct io_accept { |
373 | struct file *file; | |
374 | struct sockaddr __user *addr; | |
375 | int __user *addr_len; | |
376 | int flags; | |
09952e3e | 377 | unsigned long nofile; |
8ed8d3c3 JA |
378 | }; |
379 | ||
380 | struct io_sync { | |
381 | struct file *file; | |
382 | loff_t len; | |
383 | loff_t off; | |
384 | int flags; | |
d63d1b5e | 385 | int mode; |
8ed8d3c3 JA |
386 | }; |
387 | ||
fbf23849 JA |
388 | struct io_cancel { |
389 | struct file *file; | |
390 | u64 addr; | |
391 | }; | |
392 | ||
b29472ee JA |
393 | struct io_timeout { |
394 | struct file *file; | |
395 | u64 addr; | |
396 | int flags; | |
bfe68a22 PB |
397 | u32 off; |
398 | u32 target_seq; | |
b29472ee JA |
399 | }; |
400 | ||
9adbd45d JA |
401 | struct io_rw { |
402 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
403 | struct kiocb kiocb; | |
404 | u64 addr; | |
405 | u64 len; | |
406 | }; | |
407 | ||
3fbb51c1 JA |
408 | struct io_connect { |
409 | struct file *file; | |
410 | struct sockaddr __user *addr; | |
411 | int addr_len; | |
412 | }; | |
413 | ||
e47293fd JA |
414 | struct io_sr_msg { |
415 | struct file *file; | |
fddaface JA |
416 | union { |
417 | struct user_msghdr __user *msg; | |
418 | void __user *buf; | |
419 | }; | |
e47293fd | 420 | int msg_flags; |
bcda7baa | 421 | int bgid; |
fddaface | 422 | size_t len; |
bcda7baa | 423 | struct io_buffer *kbuf; |
e47293fd JA |
424 | }; |
425 | ||
15b71abe JA |
426 | struct io_open { |
427 | struct file *file; | |
428 | int dfd; | |
15b71abe | 429 | struct filename *filename; |
c12cedf2 | 430 | struct open_how how; |
4022e7af | 431 | unsigned long nofile; |
15b71abe JA |
432 | }; |
433 | ||
05f3fb3c JA |
434 | struct io_files_update { |
435 | struct file *file; | |
436 | u64 arg; | |
437 | u32 nr_args; | |
438 | u32 offset; | |
439 | }; | |
440 | ||
4840e418 JA |
441 | struct io_fadvise { |
442 | struct file *file; | |
443 | u64 offset; | |
444 | u32 len; | |
445 | u32 advice; | |
446 | }; | |
447 | ||
c1ca757b JA |
448 | struct io_madvise { |
449 | struct file *file; | |
450 | u64 addr; | |
451 | u32 len; | |
452 | u32 advice; | |
453 | }; | |
454 | ||
3e4827b0 JA |
455 | struct io_epoll { |
456 | struct file *file; | |
457 | int epfd; | |
458 | int op; | |
459 | int fd; | |
460 | struct epoll_event event; | |
e47293fd JA |
461 | }; |
462 | ||
7d67af2c PB |
463 | struct io_splice { |
464 | struct file *file_out; | |
465 | struct file *file_in; | |
466 | loff_t off_out; | |
467 | loff_t off_in; | |
468 | u64 len; | |
469 | unsigned int flags; | |
470 | }; | |
471 | ||
ddf0322d JA |
472 | struct io_provide_buf { |
473 | struct file *file; | |
474 | __u64 addr; | |
475 | __s32 len; | |
476 | __u32 bgid; | |
477 | __u16 nbufs; | |
478 | __u16 bid; | |
479 | }; | |
480 | ||
1d9e1288 BM |
481 | struct io_statx { |
482 | struct file *file; | |
483 | int dfd; | |
484 | unsigned int mask; | |
485 | unsigned int flags; | |
e62753e4 | 486 | const char __user *filename; |
1d9e1288 BM |
487 | struct statx __user *buffer; |
488 | }; | |
489 | ||
f499a021 JA |
490 | struct io_async_connect { |
491 | struct sockaddr_storage address; | |
492 | }; | |
493 | ||
03b1230c JA |
494 | struct io_async_msghdr { |
495 | struct iovec fast_iov[UIO_FASTIOV]; | |
496 | struct iovec *iov; | |
497 | struct sockaddr __user *uaddr; | |
498 | struct msghdr msg; | |
b537916c | 499 | struct sockaddr_storage addr; |
03b1230c JA |
500 | }; |
501 | ||
f67676d1 JA |
502 | struct io_async_rw { |
503 | struct iovec fast_iov[UIO_FASTIOV]; | |
504 | struct iovec *iov; | |
505 | ssize_t nr_segs; | |
506 | ssize_t size; | |
bcf5a063 JA |
507 | struct wait_page_queue wpq; |
508 | struct callback_head task_work; | |
f67676d1 JA |
509 | }; |
510 | ||
1a6b74fc | 511 | struct io_async_ctx { |
f67676d1 JA |
512 | union { |
513 | struct io_async_rw rw; | |
03b1230c | 514 | struct io_async_msghdr msg; |
f499a021 | 515 | struct io_async_connect connect; |
2d28390a | 516 | struct io_timeout_data timeout; |
f67676d1 | 517 | }; |
1a6b74fc JA |
518 | }; |
519 | ||
6b47ee6e PB |
520 | enum { |
521 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
522 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
523 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
524 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
525 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 526 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 527 | |
dea3b49c | 528 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
529 | REQ_F_FAIL_LINK_BIT, |
530 | REQ_F_INFLIGHT_BIT, | |
531 | REQ_F_CUR_POS_BIT, | |
532 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 533 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 534 | REQ_F_ISREG_BIT, |
6b47ee6e | 535 | REQ_F_COMP_LOCKED_BIT, |
99bc4c38 | 536 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 537 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 538 | REQ_F_POLLED_BIT, |
bcda7baa | 539 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 540 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 541 | REQ_F_WORK_INITIALIZED_BIT, |
4dd2824d | 542 | REQ_F_TASK_PINNED_BIT, |
84557871 JA |
543 | |
544 | /* not a real bit, just to check we're not overflowing the space */ | |
545 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
546 | }; |
547 | ||
548 | enum { | |
549 | /* ctx owns file */ | |
550 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
551 | /* drain existing IO first */ | |
552 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
553 | /* linked sqes */ | |
554 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
555 | /* doesn't sever on completion < 0 */ | |
556 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
557 | /* IOSQE_ASYNC */ | |
558 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
559 | /* IOSQE_BUFFER_SELECT */ |
560 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 561 | |
dea3b49c PB |
562 | /* head of a link */ |
563 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
564 | /* fail rest of links */ |
565 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
566 | /* on inflight list */ | |
567 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
568 | /* read/write uses file position */ | |
569 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
570 | /* must not punt to workers */ | |
571 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
6b47ee6e PB |
572 | /* has linked timeout */ |
573 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
6b47ee6e PB |
574 | /* regular file */ |
575 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
6b47ee6e PB |
576 | /* completion under lock */ |
577 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
578 | /* needs cleanup */ |
579 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
580 | /* in overflow list */ |
581 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
582 | /* already went through poll handler */ |
583 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
584 | /* buffer already selected */ |
585 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
586 | /* doesn't need file table for this request */ |
587 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
588 | /* io_wq_work is initialized */ |
589 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
4dd2824d PB |
590 | /* req->task is refcounted */ |
591 | REQ_F_TASK_PINNED = BIT(REQ_F_TASK_PINNED_BIT), | |
d7718a9d JA |
592 | }; |
593 | ||
594 | struct async_poll { | |
595 | struct io_poll_iocb poll; | |
596 | struct io_wq_work work; | |
6b47ee6e PB |
597 | }; |
598 | ||
09bb8394 JA |
599 | /* |
600 | * NOTE! Each of the iocb union members has the file pointer | |
601 | * as the first entry in their struct definition. So you can | |
602 | * access the file pointer through any of the sub-structs, | |
603 | * or directly as just 'ki_filp' in this struct. | |
604 | */ | |
2b188cc1 | 605 | struct io_kiocb { |
221c5eb2 | 606 | union { |
09bb8394 | 607 | struct file *file; |
9adbd45d | 608 | struct io_rw rw; |
221c5eb2 | 609 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
610 | struct io_accept accept; |
611 | struct io_sync sync; | |
fbf23849 | 612 | struct io_cancel cancel; |
b29472ee | 613 | struct io_timeout timeout; |
3fbb51c1 | 614 | struct io_connect connect; |
e47293fd | 615 | struct io_sr_msg sr_msg; |
15b71abe | 616 | struct io_open open; |
b5dba59e | 617 | struct io_close close; |
05f3fb3c | 618 | struct io_files_update files_update; |
4840e418 | 619 | struct io_fadvise fadvise; |
c1ca757b | 620 | struct io_madvise madvise; |
3e4827b0 | 621 | struct io_epoll epoll; |
7d67af2c | 622 | struct io_splice splice; |
ddf0322d | 623 | struct io_provide_buf pbuf; |
1d9e1288 | 624 | struct io_statx statx; |
221c5eb2 | 625 | }; |
2b188cc1 | 626 | |
1a6b74fc | 627 | struct io_async_ctx *io; |
c398ecb3 | 628 | int cflags; |
d625c6ee | 629 | u8 opcode; |
65a6543d XW |
630 | /* polled IO has completed */ |
631 | u8 iopoll_completed; | |
2b188cc1 | 632 | |
4f4eeba8 BM |
633 | u16 buf_index; |
634 | ||
2b188cc1 | 635 | struct io_ring_ctx *ctx; |
d7718a9d | 636 | struct list_head list; |
2b188cc1 | 637 | unsigned int flags; |
c16361c1 | 638 | refcount_t refs; |
3537b6a7 JA |
639 | struct task_struct *task; |
640 | unsigned long fsize; | |
2b188cc1 | 641 | u64 user_data; |
9e645e11 | 642 | u32 result; |
de0617e4 | 643 | u32 sequence; |
2b188cc1 | 644 | |
d7718a9d JA |
645 | struct list_head link_list; |
646 | ||
fcb323cc JA |
647 | struct list_head inflight_entry; |
648 | ||
05589553 XW |
649 | struct percpu_ref *fixed_file_refs; |
650 | ||
b41e9852 JA |
651 | union { |
652 | /* | |
653 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
654 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
655 | * async armed poll handlers for regular commands. The latter | |
656 | * restore the work, if needed. | |
b41e9852 JA |
657 | */ |
658 | struct { | |
d7718a9d JA |
659 | struct hlist_node hash_node; |
660 | struct async_poll *apoll; | |
b41e9852 JA |
661 | }; |
662 | struct io_wq_work work; | |
663 | }; | |
8ef77766 | 664 | struct callback_head task_work; |
2b188cc1 JA |
665 | }; |
666 | ||
def596e9 | 667 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 668 | |
013538bd JA |
669 | struct io_comp_state { |
670 | unsigned int nr; | |
671 | struct list_head list; | |
672 | struct io_ring_ctx *ctx; | |
673 | }; | |
674 | ||
9a56a232 JA |
675 | struct io_submit_state { |
676 | struct blk_plug plug; | |
677 | ||
2579f913 JA |
678 | /* |
679 | * io_kiocb alloc cache | |
680 | */ | |
681 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 682 | unsigned int free_reqs; |
2579f913 | 683 | |
013538bd JA |
684 | /* |
685 | * Batch completion logic | |
686 | */ | |
687 | struct io_comp_state comp; | |
688 | ||
9a56a232 JA |
689 | /* |
690 | * File reference cache | |
691 | */ | |
692 | struct file *file; | |
693 | unsigned int fd; | |
694 | unsigned int has_refs; | |
695 | unsigned int used_refs; | |
696 | unsigned int ios_left; | |
697 | }; | |
698 | ||
d3656344 JA |
699 | struct io_op_def { |
700 | /* needs req->io allocated for deferral/async */ | |
701 | unsigned async_ctx : 1; | |
702 | /* needs current->mm setup, does mm access */ | |
703 | unsigned needs_mm : 1; | |
704 | /* needs req->file assigned */ | |
705 | unsigned needs_file : 1; | |
fd2206e4 JA |
706 | /* don't fail if file grab fails */ |
707 | unsigned needs_file_no_error : 1; | |
d3656344 JA |
708 | /* hash wq insertion if file is a regular file */ |
709 | unsigned hash_reg_file : 1; | |
710 | /* unbound wq insertion if file is a non-regular file */ | |
711 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
712 | /* opcode is not supported by this kernel */ |
713 | unsigned not_supported : 1; | |
f86cd20c JA |
714 | /* needs file table */ |
715 | unsigned file_table : 1; | |
ff002b30 JA |
716 | /* needs ->fs */ |
717 | unsigned needs_fs : 1; | |
8a72758c JA |
718 | /* set if opcode supports polled "wait" */ |
719 | unsigned pollin : 1; | |
720 | unsigned pollout : 1; | |
bcda7baa JA |
721 | /* op supports buffer selection */ |
722 | unsigned buffer_select : 1; | |
d3656344 JA |
723 | }; |
724 | ||
725 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
726 | [IORING_OP_NOP] = {}, |
727 | [IORING_OP_READV] = { | |
d3656344 JA |
728 | .async_ctx = 1, |
729 | .needs_mm = 1, | |
730 | .needs_file = 1, | |
731 | .unbound_nonreg_file = 1, | |
8a72758c | 732 | .pollin = 1, |
4d954c25 | 733 | .buffer_select = 1, |
d3656344 | 734 | }, |
0463b6c5 | 735 | [IORING_OP_WRITEV] = { |
d3656344 JA |
736 | .async_ctx = 1, |
737 | .needs_mm = 1, | |
738 | .needs_file = 1, | |
739 | .hash_reg_file = 1, | |
740 | .unbound_nonreg_file = 1, | |
8a72758c | 741 | .pollout = 1, |
d3656344 | 742 | }, |
0463b6c5 | 743 | [IORING_OP_FSYNC] = { |
d3656344 JA |
744 | .needs_file = 1, |
745 | }, | |
0463b6c5 | 746 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
747 | .needs_file = 1, |
748 | .unbound_nonreg_file = 1, | |
8a72758c | 749 | .pollin = 1, |
d3656344 | 750 | }, |
0463b6c5 | 751 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
752 | .needs_file = 1, |
753 | .hash_reg_file = 1, | |
754 | .unbound_nonreg_file = 1, | |
8a72758c | 755 | .pollout = 1, |
d3656344 | 756 | }, |
0463b6c5 | 757 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
758 | .needs_file = 1, |
759 | .unbound_nonreg_file = 1, | |
760 | }, | |
0463b6c5 PB |
761 | [IORING_OP_POLL_REMOVE] = {}, |
762 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
763 | .needs_file = 1, |
764 | }, | |
0463b6c5 | 765 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
766 | .async_ctx = 1, |
767 | .needs_mm = 1, | |
768 | .needs_file = 1, | |
769 | .unbound_nonreg_file = 1, | |
ff002b30 | 770 | .needs_fs = 1, |
8a72758c | 771 | .pollout = 1, |
d3656344 | 772 | }, |
0463b6c5 | 773 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
774 | .async_ctx = 1, |
775 | .needs_mm = 1, | |
776 | .needs_file = 1, | |
777 | .unbound_nonreg_file = 1, | |
ff002b30 | 778 | .needs_fs = 1, |
8a72758c | 779 | .pollin = 1, |
52de1fe1 | 780 | .buffer_select = 1, |
d3656344 | 781 | }, |
0463b6c5 | 782 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
783 | .async_ctx = 1, |
784 | .needs_mm = 1, | |
785 | }, | |
0463b6c5 PB |
786 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
787 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
788 | .needs_mm = 1, |
789 | .needs_file = 1, | |
790 | .unbound_nonreg_file = 1, | |
f86cd20c | 791 | .file_table = 1, |
8a72758c | 792 | .pollin = 1, |
d3656344 | 793 | }, |
0463b6c5 PB |
794 | [IORING_OP_ASYNC_CANCEL] = {}, |
795 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
796 | .async_ctx = 1, |
797 | .needs_mm = 1, | |
798 | }, | |
0463b6c5 | 799 | [IORING_OP_CONNECT] = { |
d3656344 JA |
800 | .async_ctx = 1, |
801 | .needs_mm = 1, | |
802 | .needs_file = 1, | |
803 | .unbound_nonreg_file = 1, | |
8a72758c | 804 | .pollout = 1, |
d3656344 | 805 | }, |
0463b6c5 | 806 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
807 | .needs_file = 1, |
808 | }, | |
0463b6c5 | 809 | [IORING_OP_OPENAT] = { |
f86cd20c | 810 | .file_table = 1, |
ff002b30 | 811 | .needs_fs = 1, |
d3656344 | 812 | }, |
0463b6c5 | 813 | [IORING_OP_CLOSE] = { |
fd2206e4 JA |
814 | .needs_file = 1, |
815 | .needs_file_no_error = 1, | |
f86cd20c | 816 | .file_table = 1, |
d3656344 | 817 | }, |
0463b6c5 | 818 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 819 | .needs_mm = 1, |
f86cd20c | 820 | .file_table = 1, |
d3656344 | 821 | }, |
0463b6c5 | 822 | [IORING_OP_STATX] = { |
d3656344 | 823 | .needs_mm = 1, |
ff002b30 | 824 | .needs_fs = 1, |
5b0bbee4 | 825 | .file_table = 1, |
d3656344 | 826 | }, |
0463b6c5 | 827 | [IORING_OP_READ] = { |
3a6820f2 JA |
828 | .needs_mm = 1, |
829 | .needs_file = 1, | |
830 | .unbound_nonreg_file = 1, | |
8a72758c | 831 | .pollin = 1, |
bcda7baa | 832 | .buffer_select = 1, |
3a6820f2 | 833 | }, |
0463b6c5 | 834 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
835 | .needs_mm = 1, |
836 | .needs_file = 1, | |
837 | .unbound_nonreg_file = 1, | |
8a72758c | 838 | .pollout = 1, |
3a6820f2 | 839 | }, |
0463b6c5 | 840 | [IORING_OP_FADVISE] = { |
4840e418 JA |
841 | .needs_file = 1, |
842 | }, | |
0463b6c5 | 843 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
844 | .needs_mm = 1, |
845 | }, | |
0463b6c5 | 846 | [IORING_OP_SEND] = { |
fddaface JA |
847 | .needs_mm = 1, |
848 | .needs_file = 1, | |
849 | .unbound_nonreg_file = 1, | |
8a72758c | 850 | .pollout = 1, |
fddaface | 851 | }, |
0463b6c5 | 852 | [IORING_OP_RECV] = { |
fddaface JA |
853 | .needs_mm = 1, |
854 | .needs_file = 1, | |
855 | .unbound_nonreg_file = 1, | |
8a72758c | 856 | .pollin = 1, |
bcda7baa | 857 | .buffer_select = 1, |
fddaface | 858 | }, |
0463b6c5 | 859 | [IORING_OP_OPENAT2] = { |
f86cd20c | 860 | .file_table = 1, |
ff002b30 | 861 | .needs_fs = 1, |
cebdb986 | 862 | }, |
3e4827b0 JA |
863 | [IORING_OP_EPOLL_CTL] = { |
864 | .unbound_nonreg_file = 1, | |
865 | .file_table = 1, | |
866 | }, | |
7d67af2c PB |
867 | [IORING_OP_SPLICE] = { |
868 | .needs_file = 1, | |
869 | .hash_reg_file = 1, | |
870 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
871 | }, |
872 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 873 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
874 | [IORING_OP_TEE] = { |
875 | .needs_file = 1, | |
876 | .hash_reg_file = 1, | |
877 | .unbound_nonreg_file = 1, | |
878 | }, | |
d3656344 JA |
879 | }; |
880 | ||
2e0464d4 BM |
881 | enum io_mem_account { |
882 | ACCT_LOCKED, | |
883 | ACCT_PINNED, | |
884 | }; | |
885 | ||
f3a6fa22 | 886 | static bool io_rw_reissue(struct io_kiocb *req, long res); |
78e19bbe | 887 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 888 | static void io_put_req(struct io_kiocb *req); |
c40f6379 | 889 | static void io_double_put_req(struct io_kiocb *req); |
978db57e | 890 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
891 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
892 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
893 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
894 | struct io_uring_files_update *ip, | |
895 | unsigned nr_args); | |
f86cd20c | 896 | static int io_grab_files(struct io_kiocb *req); |
2237d765 JA |
897 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
898 | struct io_comp_state *cs); | |
99bc4c38 | 899 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
900 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
901 | int fd, struct file **out_file, bool fixed); | |
902 | static void __io_queue_sqe(struct io_kiocb *req, | |
f13fad7b JA |
903 | const struct io_uring_sqe *sqe, |
904 | struct io_comp_state *cs); | |
de0617e4 | 905 | |
b63534c4 JA |
906 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
907 | struct iovec **iovec, struct iov_iter *iter, | |
908 | bool needs_lock); | |
909 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, | |
910 | struct iovec *iovec, struct iovec *fast_iov, | |
911 | struct iov_iter *iter); | |
912 | ||
2b188cc1 JA |
913 | static struct kmem_cache *req_cachep; |
914 | ||
915 | static const struct file_operations io_uring_fops; | |
916 | ||
917 | struct sock *io_uring_get_socket(struct file *file) | |
918 | { | |
919 | #if defined(CONFIG_UNIX) | |
920 | if (file->f_op == &io_uring_fops) { | |
921 | struct io_ring_ctx *ctx = file->private_data; | |
922 | ||
923 | return ctx->ring_sock->sk; | |
924 | } | |
925 | #endif | |
926 | return NULL; | |
927 | } | |
928 | EXPORT_SYMBOL(io_uring_get_socket); | |
929 | ||
4dd2824d PB |
930 | static void io_get_req_task(struct io_kiocb *req) |
931 | { | |
932 | if (req->flags & REQ_F_TASK_PINNED) | |
933 | return; | |
934 | get_task_struct(req->task); | |
935 | req->flags |= REQ_F_TASK_PINNED; | |
936 | } | |
937 | ||
938 | /* not idempotent -- it doesn't clear REQ_F_TASK_PINNED */ | |
939 | static void __io_put_req_task(struct io_kiocb *req) | |
940 | { | |
941 | if (req->flags & REQ_F_TASK_PINNED) | |
942 | put_task_struct(req->task); | |
943 | } | |
944 | ||
c40f6379 JA |
945 | static void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
946 | { | |
947 | struct mm_struct *mm = current->mm; | |
948 | ||
949 | if (mm) { | |
950 | kthread_unuse_mm(mm); | |
951 | mmput(mm); | |
952 | } | |
953 | } | |
954 | ||
955 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
956 | { | |
957 | if (!current->mm) { | |
8eb06d7e | 958 | if (unlikely(!ctx->sqo_mm || !mmget_not_zero(ctx->sqo_mm))) |
c40f6379 JA |
959 | return -EFAULT; |
960 | kthread_use_mm(ctx->sqo_mm); | |
961 | } | |
962 | ||
963 | return 0; | |
964 | } | |
965 | ||
966 | static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx, | |
967 | struct io_kiocb *req) | |
968 | { | |
969 | if (!io_op_defs[req->opcode].needs_mm) | |
970 | return 0; | |
971 | return __io_sq_thread_acquire_mm(ctx); | |
972 | } | |
973 | ||
974 | static inline void req_set_fail_links(struct io_kiocb *req) | |
975 | { | |
976 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
977 | req->flags |= REQ_F_FAIL_LINK; | |
978 | } | |
979 | ||
4a38aed2 JA |
980 | static void io_file_put_work(struct work_struct *work); |
981 | ||
7cdaf587 XW |
982 | /* |
983 | * Note: must call io_req_init_async() for the first time you | |
984 | * touch any members of io_wq_work. | |
985 | */ | |
986 | static inline void io_req_init_async(struct io_kiocb *req) | |
987 | { | |
988 | if (req->flags & REQ_F_WORK_INITIALIZED) | |
989 | return; | |
990 | ||
991 | memset(&req->work, 0, sizeof(req->work)); | |
992 | req->flags |= REQ_F_WORK_INITIALIZED; | |
993 | } | |
994 | ||
0cdaf760 PB |
995 | static inline bool io_async_submit(struct io_ring_ctx *ctx) |
996 | { | |
997 | return ctx->flags & IORING_SETUP_SQPOLL; | |
998 | } | |
999 | ||
2b188cc1 JA |
1000 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1001 | { | |
1002 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1003 | ||
0f158b4c | 1004 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1005 | } |
1006 | ||
8eb7e2d0 PB |
1007 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1008 | { | |
1009 | return !req->timeout.off; | |
1010 | } | |
1011 | ||
2b188cc1 JA |
1012 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1013 | { | |
1014 | struct io_ring_ctx *ctx; | |
78076bb6 | 1015 | int hash_bits; |
2b188cc1 JA |
1016 | |
1017 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1018 | if (!ctx) | |
1019 | return NULL; | |
1020 | ||
0ddf92e8 JA |
1021 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
1022 | if (!ctx->fallback_req) | |
1023 | goto err; | |
1024 | ||
78076bb6 JA |
1025 | /* |
1026 | * Use 5 bits less than the max cq entries, that should give us around | |
1027 | * 32 entries per hash list if totally full and uniformly spread. | |
1028 | */ | |
1029 | hash_bits = ilog2(p->cq_entries); | |
1030 | hash_bits -= 5; | |
1031 | if (hash_bits <= 0) | |
1032 | hash_bits = 1; | |
1033 | ctx->cancel_hash_bits = hash_bits; | |
1034 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1035 | GFP_KERNEL); | |
1036 | if (!ctx->cancel_hash) | |
1037 | goto err; | |
1038 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1039 | ||
21482896 | 1040 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1041 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1042 | goto err; | |
2b188cc1 JA |
1043 | |
1044 | ctx->flags = p->flags; | |
583863ed | 1045 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 | 1046 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1047 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1048 | init_completion(&ctx->ref_comp); |
1049 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1050 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1051 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1052 | mutex_init(&ctx->uring_lock); |
1053 | init_waitqueue_head(&ctx->wait); | |
1054 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 1055 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 1056 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1057 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1058 | init_waitqueue_head(&ctx->inflight_wait); |
1059 | spin_lock_init(&ctx->inflight_lock); | |
1060 | INIT_LIST_HEAD(&ctx->inflight_list); | |
4a38aed2 JA |
1061 | INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work); |
1062 | init_llist_head(&ctx->file_put_llist); | |
2b188cc1 | 1063 | return ctx; |
206aefde | 1064 | err: |
0ddf92e8 JA |
1065 | if (ctx->fallback_req) |
1066 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
78076bb6 | 1067 | kfree(ctx->cancel_hash); |
206aefde JA |
1068 | kfree(ctx); |
1069 | return NULL; | |
2b188cc1 JA |
1070 | } |
1071 | ||
9d858b21 | 1072 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 1073 | { |
a197f664 JL |
1074 | struct io_ring_ctx *ctx = req->ctx; |
1075 | ||
31af27c7 PB |
1076 | return req->sequence != ctx->cached_cq_tail |
1077 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
1078 | } |
1079 | ||
9d858b21 | 1080 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 1081 | { |
87987898 | 1082 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 1083 | return __req_need_defer(req); |
de0617e4 | 1084 | |
9d858b21 | 1085 | return false; |
de0617e4 JA |
1086 | } |
1087 | ||
de0617e4 | 1088 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1089 | { |
75b28aff | 1090 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1091 | |
07910158 PB |
1092 | /* order cqe stores with ring update */ |
1093 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1094 | |
07910158 PB |
1095 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1096 | wake_up_interruptible(&ctx->cq_wait); | |
1097 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1098 | } |
1099 | } | |
1100 | ||
cbdcb435 | 1101 | static void io_req_work_grab_env(struct io_kiocb *req) |
cccf0ee8 | 1102 | { |
351fd535 PB |
1103 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1104 | ||
edcdfcc1 PB |
1105 | io_req_init_async(req); |
1106 | ||
cccf0ee8 JA |
1107 | if (!req->work.mm && def->needs_mm) { |
1108 | mmgrab(current->mm); | |
1109 | req->work.mm = current->mm; | |
2b188cc1 | 1110 | } |
cccf0ee8 JA |
1111 | if (!req->work.creds) |
1112 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1113 | if (!req->work.fs && def->needs_fs) { |
1114 | spin_lock(¤t->fs->lock); | |
1115 | if (!current->fs->in_exec) { | |
1116 | req->work.fs = current->fs; | |
1117 | req->work.fs->users++; | |
1118 | } else { | |
1119 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1120 | } | |
1121 | spin_unlock(¤t->fs->lock); | |
1122 | } | |
2b188cc1 JA |
1123 | } |
1124 | ||
cccf0ee8 | 1125 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1126 | { |
7cdaf587 XW |
1127 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1128 | return; | |
1129 | ||
cccf0ee8 JA |
1130 | if (req->work.mm) { |
1131 | mmdrop(req->work.mm); | |
1132 | req->work.mm = NULL; | |
1133 | } | |
1134 | if (req->work.creds) { | |
1135 | put_cred(req->work.creds); | |
1136 | req->work.creds = NULL; | |
1137 | } | |
ff002b30 JA |
1138 | if (req->work.fs) { |
1139 | struct fs_struct *fs = req->work.fs; | |
1140 | ||
1141 | spin_lock(&req->work.fs->lock); | |
1142 | if (--fs->users) | |
1143 | fs = NULL; | |
1144 | spin_unlock(&req->work.fs->lock); | |
1145 | if (fs) | |
1146 | free_fs_struct(fs); | |
1147 | } | |
561fb04a JA |
1148 | } |
1149 | ||
cbdcb435 | 1150 | static void io_prep_async_work(struct io_kiocb *req) |
18d9be1a | 1151 | { |
d3656344 | 1152 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1153 | |
d3656344 JA |
1154 | if (req->flags & REQ_F_ISREG) { |
1155 | if (def->hash_reg_file) | |
8766dd51 | 1156 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1157 | } else { |
1158 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1159 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1160 | } |
cccf0ee8 | 1161 | |
351fd535 | 1162 | io_req_work_grab_env(req); |
561fb04a JA |
1163 | } |
1164 | ||
cbdcb435 | 1165 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1166 | { |
cbdcb435 | 1167 | struct io_kiocb *cur; |
94ae5e77 | 1168 | |
cbdcb435 PB |
1169 | io_prep_async_work(req); |
1170 | if (req->flags & REQ_F_LINK_HEAD) | |
1171 | list_for_each_entry(cur, &req->link_list, link_list) | |
1172 | io_prep_async_work(cur); | |
1173 | } | |
1174 | ||
1175 | static void __io_queue_async_work(struct io_kiocb *req) | |
1176 | { | |
1177 | struct io_ring_ctx *ctx = req->ctx; | |
1178 | struct io_kiocb *link = io_prep_linked_timeout(req); | |
561fb04a | 1179 | |
8766dd51 PB |
1180 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1181 | &req->work, req->flags); | |
1182 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1183 | |
1184 | if (link) | |
1185 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1186 | } |
1187 | ||
cbdcb435 PB |
1188 | static void io_queue_async_work(struct io_kiocb *req) |
1189 | { | |
1190 | /* init ->work of the whole link before punting */ | |
1191 | io_prep_async_link(req); | |
1192 | __io_queue_async_work(req); | |
1193 | } | |
1194 | ||
5262f567 JA |
1195 | static void io_kill_timeout(struct io_kiocb *req) |
1196 | { | |
1197 | int ret; | |
1198 | ||
2d28390a | 1199 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1200 | if (ret != -1) { |
1201 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1202 | list_del_init(&req->list); |
f0e20b89 | 1203 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1204 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1205 | io_put_req(req); |
5262f567 JA |
1206 | } |
1207 | } | |
1208 | ||
1209 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1210 | { | |
1211 | struct io_kiocb *req, *tmp; | |
1212 | ||
1213 | spin_lock_irq(&ctx->completion_lock); | |
1214 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1215 | io_kill_timeout(req); | |
1216 | spin_unlock_irq(&ctx->completion_lock); | |
1217 | } | |
1218 | ||
04518945 | 1219 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1220 | { |
04518945 PB |
1221 | do { |
1222 | struct io_kiocb *req = list_first_entry(&ctx->defer_list, | |
1223 | struct io_kiocb, list); | |
de0617e4 | 1224 | |
04518945 PB |
1225 | if (req_need_defer(req)) |
1226 | break; | |
1227 | list_del_init(&req->list); | |
cbdcb435 PB |
1228 | /* punt-init is done before queueing for defer */ |
1229 | __io_queue_async_work(req); | |
04518945 PB |
1230 | } while (!list_empty(&ctx->defer_list)); |
1231 | } | |
1232 | ||
360428f8 | 1233 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1234 | { |
360428f8 PB |
1235 | while (!list_empty(&ctx->timeout_list)) { |
1236 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, | |
1237 | struct io_kiocb, list); | |
de0617e4 | 1238 | |
8eb7e2d0 | 1239 | if (io_is_timeout_noseq(req)) |
360428f8 | 1240 | break; |
bfe68a22 PB |
1241 | if (req->timeout.target_seq != ctx->cached_cq_tail |
1242 | - atomic_read(&ctx->cq_timeouts)) | |
360428f8 | 1243 | break; |
bfe68a22 | 1244 | |
360428f8 | 1245 | list_del_init(&req->list); |
5262f567 | 1246 | io_kill_timeout(req); |
360428f8 PB |
1247 | } |
1248 | } | |
5262f567 | 1249 | |
360428f8 PB |
1250 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1251 | { | |
1252 | io_flush_timeouts(ctx); | |
de0617e4 JA |
1253 | __io_commit_cqring(ctx); |
1254 | ||
04518945 PB |
1255 | if (unlikely(!list_empty(&ctx->defer_list))) |
1256 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1257 | } |
1258 | ||
2b188cc1 JA |
1259 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1260 | { | |
75b28aff | 1261 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1262 | unsigned tail; |
1263 | ||
1264 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1265 | /* |
1266 | * writes to the cq entry need to come after reading head; the | |
1267 | * control dependency is enough as we're using WRITE_ONCE to | |
1268 | * fill the cq entry | |
1269 | */ | |
75b28aff | 1270 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1271 | return NULL; |
1272 | ||
1273 | ctx->cached_cq_tail++; | |
75b28aff | 1274 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1275 | } |
1276 | ||
f2842ab5 JA |
1277 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1278 | { | |
f0b493e6 JA |
1279 | if (!ctx->cq_ev_fd) |
1280 | return false; | |
7e55a19c SG |
1281 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1282 | return false; | |
f2842ab5 JA |
1283 | if (!ctx->eventfd_async) |
1284 | return true; | |
b41e9852 | 1285 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1286 | } |
1287 | ||
b41e9852 | 1288 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1289 | { |
1290 | if (waitqueue_active(&ctx->wait)) | |
1291 | wake_up(&ctx->wait); | |
1292 | if (waitqueue_active(&ctx->sqo_wait)) | |
1293 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1294 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1295 | eventfd_signal(ctx->cq_ev_fd, 1); |
1296 | } | |
1297 | ||
c4a2ed72 JA |
1298 | /* Returns true if there are no backlogged entries after the flush */ |
1299 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1300 | { |
1301 | struct io_rings *rings = ctx->rings; | |
1302 | struct io_uring_cqe *cqe; | |
1303 | struct io_kiocb *req; | |
1304 | unsigned long flags; | |
1305 | LIST_HEAD(list); | |
1306 | ||
1307 | if (!force) { | |
1308 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1309 | return true; |
1d7bb1d5 JA |
1310 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1311 | rings->cq_ring_entries)) | |
c4a2ed72 | 1312 | return false; |
1d7bb1d5 JA |
1313 | } |
1314 | ||
1315 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1316 | ||
1317 | /* if force is set, the ring is going away. always drop after that */ | |
1318 | if (force) | |
69b3e546 | 1319 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1320 | |
c4a2ed72 | 1321 | cqe = NULL; |
1d7bb1d5 JA |
1322 | while (!list_empty(&ctx->cq_overflow_list)) { |
1323 | cqe = io_get_cqring(ctx); | |
1324 | if (!cqe && !force) | |
1325 | break; | |
1326 | ||
1327 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1328 | list); | |
1329 | list_move(&req->list, &list); | |
2ca10259 | 1330 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1331 | if (cqe) { |
1332 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1333 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1334 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1335 | } else { |
1336 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1337 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1342 | if (cqe) { |
1343 | clear_bit(0, &ctx->sq_check_overflow); | |
1344 | clear_bit(0, &ctx->cq_check_overflow); | |
1345 | } | |
1d7bb1d5 JA |
1346 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1347 | io_cqring_ev_posted(ctx); | |
1348 | ||
1349 | while (!list_empty(&list)) { | |
1350 | req = list_first_entry(&list, struct io_kiocb, list); | |
1351 | list_del(&req->list); | |
ec9c02ad | 1352 | io_put_req(req); |
1d7bb1d5 | 1353 | } |
c4a2ed72 JA |
1354 | |
1355 | return cqe != NULL; | |
1d7bb1d5 JA |
1356 | } |
1357 | ||
bcda7baa | 1358 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1359 | { |
78e19bbe | 1360 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1361 | struct io_uring_cqe *cqe; |
1362 | ||
78e19bbe | 1363 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1364 | |
2b188cc1 JA |
1365 | /* |
1366 | * If we can't get a cq entry, userspace overflowed the | |
1367 | * submission (by quite a lot). Increment the overflow count in | |
1368 | * the ring. | |
1369 | */ | |
1370 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1371 | if (likely(cqe)) { |
78e19bbe | 1372 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1373 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1374 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1375 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1376 | WRITE_ONCE(ctx->rings->cq_overflow, |
1377 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1378 | } else { |
ad3eb2c8 JA |
1379 | if (list_empty(&ctx->cq_overflow_list)) { |
1380 | set_bit(0, &ctx->sq_check_overflow); | |
1381 | set_bit(0, &ctx->cq_check_overflow); | |
1382 | } | |
2ca10259 | 1383 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1384 | refcount_inc(&req->refs); |
1385 | req->result = res; | |
bcda7baa | 1386 | req->cflags = cflags; |
1d7bb1d5 | 1387 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1388 | } |
1389 | } | |
1390 | ||
bcda7baa JA |
1391 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1392 | { | |
1393 | __io_cqring_fill_event(req, res, 0); | |
1394 | } | |
1395 | ||
e1e16097 | 1396 | static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1397 | { |
78e19bbe | 1398 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1399 | unsigned long flags; |
1400 | ||
1401 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1402 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1403 | io_commit_cqring(ctx); |
1404 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1405 | ||
8c838788 | 1406 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1407 | } |
1408 | ||
229a7b63 | 1409 | static void io_submit_flush_completions(struct io_comp_state *cs) |
bcda7baa | 1410 | { |
229a7b63 JA |
1411 | struct io_ring_ctx *ctx = cs->ctx; |
1412 | ||
1413 | spin_lock_irq(&ctx->completion_lock); | |
1414 | while (!list_empty(&cs->list)) { | |
1415 | struct io_kiocb *req; | |
1416 | ||
1417 | req = list_first_entry(&cs->list, struct io_kiocb, list); | |
1418 | list_del(&req->list); | |
8b3656af | 1419 | __io_cqring_fill_event(req, req->result, req->cflags); |
229a7b63 JA |
1420 | if (!(req->flags & REQ_F_LINK_HEAD)) { |
1421 | req->flags |= REQ_F_COMP_LOCKED; | |
1422 | io_put_req(req); | |
1423 | } else { | |
1424 | spin_unlock_irq(&ctx->completion_lock); | |
1425 | io_put_req(req); | |
1426 | spin_lock_irq(&ctx->completion_lock); | |
1427 | } | |
1428 | } | |
1429 | io_commit_cqring(ctx); | |
1430 | spin_unlock_irq(&ctx->completion_lock); | |
1431 | ||
1432 | io_cqring_ev_posted(ctx); | |
1433 | cs->nr = 0; | |
1434 | } | |
1435 | ||
1436 | static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, | |
1437 | struct io_comp_state *cs) | |
1438 | { | |
1439 | if (!cs) { | |
1440 | io_cqring_add_event(req, res, cflags); | |
1441 | io_put_req(req); | |
1442 | } else { | |
1443 | req->result = res; | |
8b3656af | 1444 | req->cflags = cflags; |
229a7b63 JA |
1445 | list_add_tail(&req->list, &cs->list); |
1446 | if (++cs->nr >= 32) | |
1447 | io_submit_flush_completions(cs); | |
1448 | } | |
e1e16097 JA |
1449 | } |
1450 | ||
1451 | static void io_req_complete(struct io_kiocb *req, long res) | |
1452 | { | |
229a7b63 | 1453 | __io_req_complete(req, res, 0, NULL); |
bcda7baa JA |
1454 | } |
1455 | ||
0ddf92e8 JA |
1456 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1457 | { | |
1458 | return req == (struct io_kiocb *) | |
1459 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1460 | } | |
1461 | ||
1462 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1463 | { | |
1464 | struct io_kiocb *req; | |
1465 | ||
1466 | req = ctx->fallback_req; | |
dd461af6 | 1467 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1468 | return req; |
1469 | ||
1470 | return NULL; | |
1471 | } | |
1472 | ||
0553b8bd PB |
1473 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1474 | struct io_submit_state *state) | |
2b188cc1 | 1475 | { |
fd6fab2c | 1476 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1477 | struct io_kiocb *req; |
1478 | ||
f6b6c7d6 | 1479 | if (!state->free_reqs) { |
2579f913 JA |
1480 | size_t sz; |
1481 | int ret; | |
1482 | ||
1483 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1484 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1485 | ||
1486 | /* | |
1487 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1488 | * retry single alloc to be on the safe side. | |
1489 | */ | |
1490 | if (unlikely(ret <= 0)) { | |
1491 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1492 | if (!state->reqs[0]) | |
0ddf92e8 | 1493 | goto fallback; |
fd6fab2c JA |
1494 | ret = 1; |
1495 | } | |
2579f913 | 1496 | state->free_reqs = ret - 1; |
6c8a3134 | 1497 | req = state->reqs[ret - 1]; |
2579f913 | 1498 | } else { |
2579f913 | 1499 | state->free_reqs--; |
6c8a3134 | 1500 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1501 | } |
1502 | ||
2579f913 | 1503 | return req; |
0ddf92e8 | 1504 | fallback: |
0553b8bd | 1505 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1506 | } |
1507 | ||
8da11c19 PB |
1508 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1509 | bool fixed) | |
1510 | { | |
1511 | if (fixed) | |
05589553 | 1512 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1513 | else |
1514 | fput(file); | |
1515 | } | |
1516 | ||
e6543a81 | 1517 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1518 | { |
929a3af9 PB |
1519 | if (req->flags & REQ_F_NEED_CLEANUP) |
1520 | io_cleanup_req(req); | |
1521 | ||
96fd84d8 | 1522 | kfree(req->io); |
8da11c19 PB |
1523 | if (req->file) |
1524 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
4dd2824d | 1525 | __io_put_req_task(req); |
cccf0ee8 | 1526 | io_req_work_drop_env(req); |
fcb323cc | 1527 | |
fcb323cc | 1528 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1529 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1530 | unsigned long flags; |
1531 | ||
1532 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1533 | list_del(&req->inflight_entry); | |
1534 | if (waitqueue_active(&ctx->inflight_wait)) | |
1535 | wake_up(&ctx->inflight_wait); | |
1536 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1537 | } | |
e6543a81 | 1538 | } |
2b85edfc | 1539 | |
e6543a81 PB |
1540 | static void __io_free_req(struct io_kiocb *req) |
1541 | { | |
ecfc5177 PB |
1542 | struct io_ring_ctx *ctx; |
1543 | ||
e6543a81 | 1544 | io_dismantle_req(req); |
ecfc5177 | 1545 | ctx = req->ctx; |
b1e50e54 PB |
1546 | if (likely(!io_is_fallback_req(req))) |
1547 | kmem_cache_free(req_cachep, req); | |
1548 | else | |
ecfc5177 PB |
1549 | clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req); |
1550 | percpu_ref_put(&ctx->refs); | |
e65ef56d JA |
1551 | } |
1552 | ||
a197f664 | 1553 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1554 | { |
a197f664 | 1555 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1556 | int ret; |
1557 | ||
2d28390a | 1558 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1559 | if (ret != -1) { |
78e19bbe | 1560 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1561 | io_commit_cqring(ctx); |
dea3b49c | 1562 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1563 | io_put_req(req); |
2665abfd JA |
1564 | return true; |
1565 | } | |
1566 | ||
1567 | return false; | |
e65ef56d JA |
1568 | } |
1569 | ||
ab0b6451 | 1570 | static bool __io_kill_linked_timeout(struct io_kiocb *req) |
9e645e11 | 1571 | { |
7c86ffee | 1572 | struct io_kiocb *link; |
ab0b6451 | 1573 | bool wake_ev; |
7c86ffee PB |
1574 | |
1575 | if (list_empty(&req->link_list)) | |
ab0b6451 | 1576 | return false; |
7c86ffee PB |
1577 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1578 | if (link->opcode != IORING_OP_LINK_TIMEOUT) | |
ab0b6451 | 1579 | return false; |
7c86ffee PB |
1580 | |
1581 | list_del_init(&link->link_list); | |
1582 | wake_ev = io_link_cancel_timeout(link); | |
1583 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
ab0b6451 JA |
1584 | return wake_ev; |
1585 | } | |
1586 | ||
1587 | static void io_kill_linked_timeout(struct io_kiocb *req) | |
1588 | { | |
1589 | struct io_ring_ctx *ctx = req->ctx; | |
1590 | bool wake_ev; | |
1591 | ||
1592 | if (!(req->flags & REQ_F_COMP_LOCKED)) { | |
1593 | unsigned long flags; | |
1594 | ||
1595 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1596 | wake_ev = __io_kill_linked_timeout(req); | |
7c86ffee | 1597 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 JA |
1598 | } else { |
1599 | wake_ev = __io_kill_linked_timeout(req); | |
1600 | } | |
1601 | ||
7c86ffee PB |
1602 | if (wake_ev) |
1603 | io_cqring_ev_posted(ctx); | |
1604 | } | |
1605 | ||
9b5f7bd9 | 1606 | static struct io_kiocb *io_req_link_next(struct io_kiocb *req) |
7c86ffee PB |
1607 | { |
1608 | struct io_kiocb *nxt; | |
9e645e11 JA |
1609 | |
1610 | /* | |
1611 | * The list should never be empty when we are called here. But could | |
1612 | * potentially happen if the chain is messed up, check to be on the | |
1613 | * safe side. | |
1614 | */ | |
7c86ffee | 1615 | if (unlikely(list_empty(&req->link_list))) |
9b5f7bd9 | 1616 | return NULL; |
2665abfd | 1617 | |
7c86ffee PB |
1618 | nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
1619 | list_del_init(&req->link_list); | |
1620 | if (!list_empty(&nxt->link_list)) | |
1621 | nxt->flags |= REQ_F_LINK_HEAD; | |
9b5f7bd9 | 1622 | return nxt; |
9e645e11 JA |
1623 | } |
1624 | ||
1625 | /* | |
dea3b49c | 1626 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 | 1627 | */ |
7c86ffee | 1628 | static void __io_fail_links(struct io_kiocb *req) |
9e645e11 | 1629 | { |
2665abfd | 1630 | struct io_ring_ctx *ctx = req->ctx; |
9e645e11 JA |
1631 | |
1632 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1633 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1634 | struct io_kiocb, link_list); | |
9e645e11 | 1635 | |
4493233e | 1636 | list_del_init(&link->link_list); |
c826bd7a | 1637 | trace_io_uring_fail_link(req, link); |
2665abfd | 1638 | |
7c86ffee PB |
1639 | io_cqring_fill_event(link, -ECANCELED); |
1640 | __io_double_put_req(link); | |
5d960724 | 1641 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1642 | } |
2665abfd JA |
1643 | |
1644 | io_commit_cqring(ctx); | |
7c86ffee PB |
1645 | io_cqring_ev_posted(ctx); |
1646 | } | |
1647 | ||
1648 | static void io_fail_links(struct io_kiocb *req) | |
1649 | { | |
1650 | struct io_ring_ctx *ctx = req->ctx; | |
1651 | ||
1652 | if (!(req->flags & REQ_F_COMP_LOCKED)) { | |
1653 | unsigned long flags; | |
1654 | ||
1655 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1656 | __io_fail_links(req); | |
1657 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1658 | } else { | |
1659 | __io_fail_links(req); | |
1660 | } | |
1661 | ||
2665abfd | 1662 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1663 | } |
1664 | ||
3fa5e0f3 | 1665 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
9e645e11 | 1666 | { |
9b0d911a | 1667 | req->flags &= ~REQ_F_LINK_HEAD; |
7c86ffee PB |
1668 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1669 | io_kill_linked_timeout(req); | |
1670 | ||
9e645e11 JA |
1671 | /* |
1672 | * If LINK is set, we have dependent requests in this chain. If we | |
1673 | * didn't fail this request, queue the first one up, moving any other | |
1674 | * dependencies to the next request. In case of failure, fail the rest | |
1675 | * of the chain. | |
1676 | */ | |
9b5f7bd9 PB |
1677 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) |
1678 | return io_req_link_next(req); | |
1679 | io_fail_links(req); | |
1680 | return NULL; | |
4d7dd462 | 1681 | } |
9e645e11 | 1682 | |
3fa5e0f3 PB |
1683 | static struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
1684 | { | |
1685 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) | |
1686 | return NULL; | |
1687 | return __io_req_find_next(req); | |
1688 | } | |
1689 | ||
c2c4c83c JA |
1690 | static int io_req_task_work_add(struct io_kiocb *req, struct callback_head *cb) |
1691 | { | |
1692 | struct task_struct *tsk = req->task; | |
1693 | struct io_ring_ctx *ctx = req->ctx; | |
1694 | int ret, notify = TWA_RESUME; | |
1695 | ||
1696 | /* | |
1697 | * SQPOLL kernel thread doesn't need notification, just a wakeup. | |
1698 | * If we're not using an eventfd, then TWA_RESUME is always fine, | |
1699 | * as we won't have dependencies between request completions for | |
1700 | * other kernel wait conditions. | |
1701 | */ | |
1702 | if (ctx->flags & IORING_SETUP_SQPOLL) | |
1703 | notify = 0; | |
1704 | else if (ctx->cq_ev_fd) | |
1705 | notify = TWA_SIGNAL; | |
1706 | ||
1707 | ret = task_work_add(tsk, cb, notify); | |
1708 | if (!ret) | |
1709 | wake_up_process(tsk); | |
1710 | return ret; | |
1711 | } | |
1712 | ||
c40f6379 JA |
1713 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1714 | { | |
1715 | struct io_ring_ctx *ctx = req->ctx; | |
1716 | ||
1717 | spin_lock_irq(&ctx->completion_lock); | |
1718 | io_cqring_fill_event(req, error); | |
1719 | io_commit_cqring(ctx); | |
1720 | spin_unlock_irq(&ctx->completion_lock); | |
1721 | ||
1722 | io_cqring_ev_posted(ctx); | |
1723 | req_set_fail_links(req); | |
1724 | io_double_put_req(req); | |
1725 | } | |
1726 | ||
1727 | static void io_req_task_cancel(struct callback_head *cb) | |
1728 | { | |
1729 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1730 | ||
1731 | __io_req_task_cancel(req, -ECANCELED); | |
1732 | } | |
1733 | ||
1734 | static void __io_req_task_submit(struct io_kiocb *req) | |
1735 | { | |
1736 | struct io_ring_ctx *ctx = req->ctx; | |
1737 | ||
c40f6379 JA |
1738 | if (!__io_sq_thread_acquire_mm(ctx)) { |
1739 | mutex_lock(&ctx->uring_lock); | |
1740 | __io_queue_sqe(req, NULL, NULL); | |
1741 | mutex_unlock(&ctx->uring_lock); | |
1742 | } else { | |
1743 | __io_req_task_cancel(req, -EFAULT); | |
1744 | } | |
1745 | } | |
1746 | ||
1747 | static void io_req_task_submit(struct callback_head *cb) | |
1748 | { | |
1749 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1750 | ||
1751 | __io_req_task_submit(req); | |
1752 | } | |
1753 | ||
1754 | static void io_req_task_queue(struct io_kiocb *req) | |
1755 | { | |
c40f6379 JA |
1756 | int ret; |
1757 | ||
1758 | init_task_work(&req->task_work, io_req_task_submit); | |
1759 | ||
c2c4c83c | 1760 | ret = io_req_task_work_add(req, &req->task_work); |
c40f6379 | 1761 | if (unlikely(ret)) { |
c2c4c83c JA |
1762 | struct task_struct *tsk; |
1763 | ||
c40f6379 JA |
1764 | init_task_work(&req->task_work, io_req_task_cancel); |
1765 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
1766 | task_work_add(tsk, &req->task_work, 0); |
1767 | wake_up_process(tsk); | |
c40f6379 | 1768 | } |
c40f6379 JA |
1769 | } |
1770 | ||
c3524383 | 1771 | static void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1772 | { |
9b5f7bd9 | 1773 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf | 1774 | |
906a8c3f PB |
1775 | if (nxt) |
1776 | io_req_task_queue(nxt); | |
c69f8dbe JL |
1777 | } |
1778 | ||
c3524383 PB |
1779 | static void io_free_req(struct io_kiocb *req) |
1780 | { | |
1781 | io_queue_next(req); | |
1782 | __io_free_req(req); | |
1783 | } | |
1784 | ||
2d6500d4 PB |
1785 | struct req_batch { |
1786 | void *reqs[IO_IOPOLL_BATCH]; | |
1787 | int to_free; | |
1788 | }; | |
1789 | ||
1790 | static void __io_req_free_batch_flush(struct io_ring_ctx *ctx, | |
1791 | struct req_batch *rb) | |
1792 | { | |
1793 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1794 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
1795 | rb->to_free = 0; | |
1796 | } | |
1797 | ||
1798 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, | |
1799 | struct req_batch *rb) | |
1800 | { | |
1801 | if (rb->to_free) | |
1802 | __io_req_free_batch_flush(ctx, rb); | |
1803 | } | |
1804 | ||
1805 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) | |
1806 | { | |
1807 | if (unlikely(io_is_fallback_req(req))) { | |
1808 | io_free_req(req); | |
1809 | return; | |
1810 | } | |
1811 | if (req->flags & REQ_F_LINK_HEAD) | |
1812 | io_queue_next(req); | |
1813 | ||
1814 | io_dismantle_req(req); | |
1815 | rb->reqs[rb->to_free++] = req; | |
1816 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1817 | __io_req_free_batch_flush(req->ctx, rb); | |
1818 | } | |
1819 | ||
ba816ad6 JA |
1820 | /* |
1821 | * Drop reference to request, return next in chain (if there is one) if this | |
1822 | * was the last reference to this request. | |
1823 | */ | |
9b5f7bd9 | 1824 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 1825 | { |
9b5f7bd9 PB |
1826 | struct io_kiocb *nxt = NULL; |
1827 | ||
2a44f467 | 1828 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 1829 | nxt = io_req_find_next(req); |
4d7dd462 | 1830 | __io_free_req(req); |
2a44f467 | 1831 | } |
9b5f7bd9 | 1832 | return nxt; |
2b188cc1 JA |
1833 | } |
1834 | ||
e65ef56d JA |
1835 | static void io_put_req(struct io_kiocb *req) |
1836 | { | |
1837 | if (refcount_dec_and_test(&req->refs)) | |
1838 | io_free_req(req); | |
2b188cc1 JA |
1839 | } |
1840 | ||
f4db7182 | 1841 | static struct io_wq_work *io_steal_work(struct io_kiocb *req) |
7a743e22 | 1842 | { |
6df1db6b | 1843 | struct io_kiocb *nxt; |
f4db7182 | 1844 | |
7a743e22 | 1845 | /* |
f4db7182 PB |
1846 | * A ref is owned by io-wq in which context we're. So, if that's the |
1847 | * last one, it's safe to steal next work. False negatives are Ok, | |
1848 | * it just will be re-punted async in io_put_work() | |
7a743e22 | 1849 | */ |
f4db7182 PB |
1850 | if (refcount_read(&req->refs) != 1) |
1851 | return NULL; | |
7a743e22 | 1852 | |
9b5f7bd9 | 1853 | nxt = io_req_find_next(req); |
6df1db6b | 1854 | return nxt ? &nxt->work : NULL; |
7a743e22 PB |
1855 | } |
1856 | ||
978db57e JA |
1857 | /* |
1858 | * Must only be used if we don't need to care about links, usually from | |
1859 | * within the completion handling itself. | |
1860 | */ | |
1861 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1862 | { |
1863 | /* drop both submit and complete references */ | |
1864 | if (refcount_sub_and_test(2, &req->refs)) | |
1865 | __io_free_req(req); | |
1866 | } | |
1867 | ||
978db57e JA |
1868 | static void io_double_put_req(struct io_kiocb *req) |
1869 | { | |
1870 | /* drop both submit and complete references */ | |
1871 | if (refcount_sub_and_test(2, &req->refs)) | |
1872 | io_free_req(req); | |
1873 | } | |
1874 | ||
1d7bb1d5 | 1875 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1876 | { |
84f97dc2 JA |
1877 | struct io_rings *rings = ctx->rings; |
1878 | ||
ad3eb2c8 JA |
1879 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1880 | /* | |
1881 | * noflush == true is from the waitqueue handler, just ensure | |
1882 | * we wake up the task, and the next invocation will flush the | |
1883 | * entries. We cannot safely to it from here. | |
1884 | */ | |
1885 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1886 | return -1U; | |
1d7bb1d5 | 1887 | |
ad3eb2c8 JA |
1888 | io_cqring_overflow_flush(ctx, false); |
1889 | } | |
1d7bb1d5 | 1890 | |
a3a0e43f JA |
1891 | /* See comment at the top of this file */ |
1892 | smp_rmb(); | |
ad3eb2c8 | 1893 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1894 | } |
1895 | ||
fb5ccc98 PB |
1896 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1897 | { | |
1898 | struct io_rings *rings = ctx->rings; | |
1899 | ||
1900 | /* make sure SQ entry isn't read before tail */ | |
1901 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1902 | } | |
1903 | ||
bcda7baa JA |
1904 | static int io_put_kbuf(struct io_kiocb *req) |
1905 | { | |
4d954c25 | 1906 | struct io_buffer *kbuf; |
bcda7baa JA |
1907 | int cflags; |
1908 | ||
4d954c25 | 1909 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1910 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1911 | cflags |= IORING_CQE_F_BUFFER; | |
1912 | req->rw.addr = 0; | |
1913 | kfree(kbuf); | |
1914 | return cflags; | |
1915 | } | |
1916 | ||
4c6e277c JA |
1917 | static inline bool io_run_task_work(void) |
1918 | { | |
1919 | if (current->task_works) { | |
1920 | __set_current_state(TASK_RUNNING); | |
1921 | task_work_run(); | |
1922 | return true; | |
1923 | } | |
1924 | ||
1925 | return false; | |
1926 | } | |
1927 | ||
bbde017a XW |
1928 | static void io_iopoll_queue(struct list_head *again) |
1929 | { | |
1930 | struct io_kiocb *req; | |
1931 | ||
1932 | do { | |
1933 | req = list_first_entry(again, struct io_kiocb, list); | |
1934 | list_del(&req->list); | |
cf2f5425 | 1935 | if (!io_rw_reissue(req, -EAGAIN)) |
2237d765 | 1936 | io_complete_rw_common(&req->rw.kiocb, -EAGAIN, NULL); |
bbde017a XW |
1937 | } while (!list_empty(again)); |
1938 | } | |
1939 | ||
def596e9 JA |
1940 | /* |
1941 | * Find and free completed poll iocbs | |
1942 | */ | |
1943 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1944 | struct list_head *done) | |
1945 | { | |
8237e045 | 1946 | struct req_batch rb; |
def596e9 | 1947 | struct io_kiocb *req; |
bbde017a XW |
1948 | LIST_HEAD(again); |
1949 | ||
1950 | /* order with ->result store in io_complete_rw_iopoll() */ | |
1951 | smp_rmb(); | |
def596e9 | 1952 | |
2757a23e | 1953 | rb.to_free = 0; |
def596e9 | 1954 | while (!list_empty(done)) { |
bcda7baa JA |
1955 | int cflags = 0; |
1956 | ||
def596e9 | 1957 | req = list_first_entry(done, struct io_kiocb, list); |
bbde017a XW |
1958 | if (READ_ONCE(req->result) == -EAGAIN) { |
1959 | req->iopoll_completed = 0; | |
1960 | list_move_tail(&req->list, &again); | |
1961 | continue; | |
1962 | } | |
def596e9 JA |
1963 | list_del(&req->list); |
1964 | ||
bcda7baa JA |
1965 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1966 | cflags = io_put_kbuf(req); | |
1967 | ||
1968 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1969 | (*nr_events)++; |
1970 | ||
c3524383 | 1971 | if (refcount_dec_and_test(&req->refs)) |
2d6500d4 | 1972 | io_req_free_batch(&rb, req); |
def596e9 | 1973 | } |
def596e9 | 1974 | |
09bb8394 | 1975 | io_commit_cqring(ctx); |
32b2244a XW |
1976 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1977 | io_cqring_ev_posted(ctx); | |
2d6500d4 | 1978 | io_req_free_batch_finish(ctx, &rb); |
581f9810 | 1979 | |
bbde017a XW |
1980 | if (!list_empty(&again)) |
1981 | io_iopoll_queue(&again); | |
581f9810 BM |
1982 | } |
1983 | ||
def596e9 JA |
1984 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1985 | long min) | |
1986 | { | |
1987 | struct io_kiocb *req, *tmp; | |
1988 | LIST_HEAD(done); | |
1989 | bool spin; | |
1990 | int ret; | |
1991 | ||
1992 | /* | |
1993 | * Only spin for completions if we don't have multiple devices hanging | |
1994 | * off our complete list, and we're under the requested amount. | |
1995 | */ | |
1996 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1997 | ||
1998 | ret = 0; | |
1999 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 2000 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2001 | |
2002 | /* | |
581f9810 BM |
2003 | * Move completed and retryable entries to our local lists. |
2004 | * If we find a request that requires polling, break out | |
2005 | * and complete those lists first, if we have entries there. | |
def596e9 | 2006 | */ |
65a6543d | 2007 | if (READ_ONCE(req->iopoll_completed)) { |
def596e9 JA |
2008 | list_move_tail(&req->list, &done); |
2009 | continue; | |
2010 | } | |
2011 | if (!list_empty(&done)) | |
2012 | break; | |
2013 | ||
2014 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2015 | if (ret < 0) | |
2016 | break; | |
2017 | ||
3aadc23e PB |
2018 | /* iopoll may have completed current req */ |
2019 | if (READ_ONCE(req->iopoll_completed)) | |
2020 | list_move_tail(&req->list, &done); | |
2021 | ||
def596e9 JA |
2022 | if (ret && spin) |
2023 | spin = false; | |
2024 | ret = 0; | |
2025 | } | |
2026 | ||
2027 | if (!list_empty(&done)) | |
2028 | io_iopoll_complete(ctx, nr_events, &done); | |
2029 | ||
2030 | return ret; | |
2031 | } | |
2032 | ||
2033 | /* | |
d195a66e | 2034 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2035 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2036 | * as a non-spinning completion check. | |
2037 | */ | |
2038 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2039 | long min) | |
2040 | { | |
08f5439f | 2041 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
2042 | int ret; |
2043 | ||
2044 | ret = io_do_iopoll(ctx, nr_events, min); | |
2045 | if (ret < 0) | |
2046 | return ret; | |
eba0a4dd | 2047 | if (*nr_events >= min) |
def596e9 JA |
2048 | return 0; |
2049 | } | |
2050 | ||
2051 | return 1; | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * We can't just wait for polled events to come to us, we have to actively | |
2056 | * find and complete them. | |
2057 | */ | |
2058 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
2059 | { | |
2060 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2061 | return; | |
2062 | ||
2063 | mutex_lock(&ctx->uring_lock); | |
2064 | while (!list_empty(&ctx->poll_list)) { | |
2065 | unsigned int nr_events = 0; | |
2066 | ||
2067 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
2068 | |
2069 | /* | |
2070 | * Ensure we allow local-to-the-cpu processing to take place, | |
2071 | * in this case we need to ensure that we reap all events. | |
2072 | */ | |
2073 | cond_resched(); | |
def596e9 JA |
2074 | } |
2075 | mutex_unlock(&ctx->uring_lock); | |
2076 | } | |
2077 | ||
c7849be9 XW |
2078 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
2079 | long min) | |
def596e9 | 2080 | { |
2b2ed975 | 2081 | int iters = 0, ret = 0; |
500f9fba | 2082 | |
c7849be9 XW |
2083 | /* |
2084 | * We disallow the app entering submit/complete with polling, but we | |
2085 | * still need to lock the ring to prevent racing with polled issue | |
2086 | * that got punted to a workqueue. | |
2087 | */ | |
2088 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2089 | do { |
a3a0e43f JA |
2090 | /* |
2091 | * Don't enter poll loop if we already have events pending. | |
2092 | * If we do, we can potentially be spinning for commands that | |
2093 | * already triggered a CQE (eg in error). | |
2094 | */ | |
1d7bb1d5 | 2095 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
2096 | break; |
2097 | ||
500f9fba JA |
2098 | /* |
2099 | * If a submit got punted to a workqueue, we can have the | |
2100 | * application entering polling for a command before it gets | |
2101 | * issued. That app will hold the uring_lock for the duration | |
2102 | * of the poll right here, so we need to take a breather every | |
2103 | * now and then to ensure that the issue has a chance to add | |
2104 | * the poll to the issued list. Otherwise we can spin here | |
2105 | * forever, while the workqueue is stuck trying to acquire the | |
2106 | * very same mutex. | |
2107 | */ | |
2108 | if (!(++iters & 7)) { | |
2109 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2110 | io_run_task_work(); |
500f9fba JA |
2111 | mutex_lock(&ctx->uring_lock); |
2112 | } | |
2113 | ||
eba0a4dd | 2114 | ret = io_iopoll_getevents(ctx, nr_events, min); |
def596e9 JA |
2115 | if (ret <= 0) |
2116 | break; | |
2117 | ret = 0; | |
2118 | } while (min && !*nr_events && !need_resched()); | |
2119 | ||
500f9fba | 2120 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2121 | return ret; |
2122 | } | |
2123 | ||
491381ce | 2124 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2125 | { |
491381ce JA |
2126 | /* |
2127 | * Tell lockdep we inherited freeze protection from submission | |
2128 | * thread. | |
2129 | */ | |
2130 | if (req->flags & REQ_F_ISREG) { | |
2131 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2132 | |
491381ce | 2133 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2134 | } |
491381ce | 2135 | file_end_write(req->file); |
2b188cc1 JA |
2136 | } |
2137 | ||
a1d7c393 JA |
2138 | static void io_complete_rw_common(struct kiocb *kiocb, long res, |
2139 | struct io_comp_state *cs) | |
2b188cc1 | 2140 | { |
9adbd45d | 2141 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 2142 | int cflags = 0; |
2b188cc1 | 2143 | |
491381ce JA |
2144 | if (kiocb->ki_flags & IOCB_WRITE) |
2145 | kiocb_end_write(req); | |
2b188cc1 | 2146 | |
4e88d6e7 JA |
2147 | if (res != req->result) |
2148 | req_set_fail_links(req); | |
bcda7baa JA |
2149 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2150 | cflags = io_put_kbuf(req); | |
a1d7c393 | 2151 | __io_req_complete(req, res, cflags, cs); |
ba816ad6 JA |
2152 | } |
2153 | ||
b63534c4 JA |
2154 | #ifdef CONFIG_BLOCK |
2155 | static bool io_resubmit_prep(struct io_kiocb *req, int error) | |
2156 | { | |
2157 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
2158 | ssize_t ret = -ECANCELED; | |
2159 | struct iov_iter iter; | |
2160 | int rw; | |
2161 | ||
2162 | if (error) { | |
2163 | ret = error; | |
2164 | goto end_req; | |
2165 | } | |
2166 | ||
2167 | switch (req->opcode) { | |
2168 | case IORING_OP_READV: | |
2169 | case IORING_OP_READ_FIXED: | |
2170 | case IORING_OP_READ: | |
2171 | rw = READ; | |
2172 | break; | |
2173 | case IORING_OP_WRITEV: | |
2174 | case IORING_OP_WRITE_FIXED: | |
2175 | case IORING_OP_WRITE: | |
2176 | rw = WRITE; | |
2177 | break; | |
2178 | default: | |
2179 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2180 | req->opcode); | |
2181 | goto end_req; | |
2182 | } | |
2183 | ||
2184 | ret = io_import_iovec(rw, req, &iovec, &iter, false); | |
2185 | if (ret < 0) | |
2186 | goto end_req; | |
2187 | ret = io_setup_async_rw(req, ret, iovec, inline_vecs, &iter); | |
2188 | if (!ret) | |
2189 | return true; | |
2190 | kfree(iovec); | |
2191 | end_req: | |
b63534c4 | 2192 | req_set_fail_links(req); |
e1e16097 | 2193 | io_req_complete(req, ret); |
b63534c4 JA |
2194 | return false; |
2195 | } | |
2196 | ||
2197 | static void io_rw_resubmit(struct callback_head *cb) | |
2198 | { | |
2199 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2200 | struct io_ring_ctx *ctx = req->ctx; | |
2201 | int err; | |
2202 | ||
b63534c4 JA |
2203 | err = io_sq_thread_acquire_mm(ctx, req); |
2204 | ||
2205 | if (io_resubmit_prep(req, err)) { | |
2206 | refcount_inc(&req->refs); | |
2207 | io_queue_async_work(req); | |
2208 | } | |
2209 | } | |
2210 | #endif | |
2211 | ||
2212 | static bool io_rw_reissue(struct io_kiocb *req, long res) | |
2213 | { | |
2214 | #ifdef CONFIG_BLOCK | |
b63534c4 JA |
2215 | int ret; |
2216 | ||
2217 | if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker()) | |
2218 | return false; | |
2219 | ||
b63534c4 | 2220 | init_task_work(&req->task_work, io_rw_resubmit); |
c2c4c83c JA |
2221 | ret = io_req_task_work_add(req, &req->task_work); |
2222 | if (!ret) | |
b63534c4 JA |
2223 | return true; |
2224 | #endif | |
2225 | return false; | |
2226 | } | |
2227 | ||
a1d7c393 JA |
2228 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
2229 | struct io_comp_state *cs) | |
2230 | { | |
2231 | if (!io_rw_reissue(req, res)) | |
2232 | io_complete_rw_common(&req->rw.kiocb, res, cs); | |
2233 | } | |
2234 | ||
ba816ad6 JA |
2235 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) |
2236 | { | |
9adbd45d | 2237 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2238 | |
a1d7c393 | 2239 | __io_complete_rw(req, res, res2, NULL); |
2b188cc1 JA |
2240 | } |
2241 | ||
def596e9 JA |
2242 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2243 | { | |
9adbd45d | 2244 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2245 | |
491381ce JA |
2246 | if (kiocb->ki_flags & IOCB_WRITE) |
2247 | kiocb_end_write(req); | |
def596e9 | 2248 | |
2d7d6792 | 2249 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2250 | req_set_fail_links(req); |
bbde017a XW |
2251 | |
2252 | WRITE_ONCE(req->result, res); | |
2253 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2254 | smp_wmb(); |
2255 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2256 | } |
2257 | ||
2258 | /* | |
2259 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2260 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2261 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2262 | * accessing the kiocb cookie. | |
2263 | */ | |
2264 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
2265 | { | |
2266 | struct io_ring_ctx *ctx = req->ctx; | |
2267 | ||
2268 | /* | |
2269 | * Track whether we have multiple files in our lists. This will impact | |
2270 | * how we do polling eventually, not spinning if we're on potentially | |
2271 | * different devices. | |
2272 | */ | |
2273 | if (list_empty(&ctx->poll_list)) { | |
2274 | ctx->poll_multi_file = false; | |
2275 | } else if (!ctx->poll_multi_file) { | |
2276 | struct io_kiocb *list_req; | |
2277 | ||
2278 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
2279 | list); | |
9adbd45d | 2280 | if (list_req->file != req->file) |
def596e9 JA |
2281 | ctx->poll_multi_file = true; |
2282 | } | |
2283 | ||
2284 | /* | |
2285 | * For fast devices, IO may have already completed. If it has, add | |
2286 | * it to the front so we find it first. | |
2287 | */ | |
65a6543d | 2288 | if (READ_ONCE(req->iopoll_completed)) |
def596e9 JA |
2289 | list_add(&req->list, &ctx->poll_list); |
2290 | else | |
2291 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
2292 | |
2293 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
2294 | wq_has_sleeper(&ctx->sqo_wait)) | |
2295 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
2296 | } |
2297 | ||
9f13c35b | 2298 | static void __io_state_file_put(struct io_submit_state *state) |
9a56a232 | 2299 | { |
9f13c35b | 2300 | int diff = state->has_refs - state->used_refs; |
9a56a232 | 2301 | |
9f13c35b PB |
2302 | if (diff) |
2303 | fput_many(state->file, diff); | |
2304 | state->file = NULL; | |
2305 | } | |
2306 | ||
2307 | static inline void io_state_file_put(struct io_submit_state *state) | |
2308 | { | |
2309 | if (state->file) | |
2310 | __io_state_file_put(state); | |
9a56a232 JA |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * Get as many references to a file as we have IOs left in this submission, | |
2315 | * assuming most submissions are for one file, or at least that each file | |
2316 | * has more than one submission. | |
2317 | */ | |
8da11c19 | 2318 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2319 | { |
2320 | if (!state) | |
2321 | return fget(fd); | |
2322 | ||
2323 | if (state->file) { | |
2324 | if (state->fd == fd) { | |
2325 | state->used_refs++; | |
2326 | state->ios_left--; | |
2327 | return state->file; | |
2328 | } | |
9f13c35b | 2329 | __io_state_file_put(state); |
9a56a232 JA |
2330 | } |
2331 | state->file = fget_many(fd, state->ios_left); | |
2332 | if (!state->file) | |
2333 | return NULL; | |
2334 | ||
2335 | state->fd = fd; | |
2336 | state->has_refs = state->ios_left; | |
2337 | state->used_refs = 1; | |
2338 | state->ios_left--; | |
2339 | return state->file; | |
2340 | } | |
2341 | ||
4503b767 JA |
2342 | static bool io_bdev_nowait(struct block_device *bdev) |
2343 | { | |
2344 | #ifdef CONFIG_BLOCK | |
2345 | return !bdev || queue_is_mq(bdev_get_queue(bdev)); | |
2346 | #else | |
2347 | return true; | |
2348 | #endif | |
2349 | } | |
2350 | ||
2b188cc1 JA |
2351 | /* |
2352 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2353 | * any file. For now, just ensure that anything potentially problematic is done | |
2354 | * inline. | |
2355 | */ | |
af197f50 | 2356 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2357 | { |
2358 | umode_t mode = file_inode(file)->i_mode; | |
2359 | ||
4503b767 JA |
2360 | if (S_ISBLK(mode)) { |
2361 | if (io_bdev_nowait(file->f_inode->i_bdev)) | |
2362 | return true; | |
2363 | return false; | |
2364 | } | |
2365 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2366 | return true; |
4503b767 JA |
2367 | if (S_ISREG(mode)) { |
2368 | if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
2369 | file->f_op != &io_uring_fops) | |
2370 | return true; | |
2371 | return false; | |
2372 | } | |
2b188cc1 | 2373 | |
c5b85625 JA |
2374 | /* any ->read/write should understand O_NONBLOCK */ |
2375 | if (file->f_flags & O_NONBLOCK) | |
2376 | return true; | |
2377 | ||
af197f50 JA |
2378 | if (!(file->f_mode & FMODE_NOWAIT)) |
2379 | return false; | |
2380 | ||
2381 | if (rw == READ) | |
2382 | return file->f_op->read_iter != NULL; | |
2383 | ||
2384 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2385 | } |
2386 | ||
3529d8c2 JA |
2387 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2388 | bool force_nonblock) | |
2b188cc1 | 2389 | { |
def596e9 | 2390 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2391 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2392 | unsigned ioprio; |
2393 | int ret; | |
2b188cc1 | 2394 | |
491381ce JA |
2395 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2396 | req->flags |= REQ_F_ISREG; | |
2397 | ||
2b188cc1 | 2398 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2399 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2400 | req->flags |= REQ_F_CUR_POS; | |
2401 | kiocb->ki_pos = req->file->f_pos; | |
2402 | } | |
2b188cc1 | 2403 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2404 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2405 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2406 | if (unlikely(ret)) | |
2407 | return ret; | |
2b188cc1 JA |
2408 | |
2409 | ioprio = READ_ONCE(sqe->ioprio); | |
2410 | if (ioprio) { | |
2411 | ret = ioprio_check_cap(ioprio); | |
2412 | if (ret) | |
09bb8394 | 2413 | return ret; |
2b188cc1 JA |
2414 | |
2415 | kiocb->ki_ioprio = ioprio; | |
2416 | } else | |
2417 | kiocb->ki_ioprio = get_current_ioprio(); | |
2418 | ||
8449eeda | 2419 | /* don't allow async punt if RWF_NOWAIT was requested */ |
c5b85625 | 2420 | if (kiocb->ki_flags & IOCB_NOWAIT) |
8449eeda SB |
2421 | req->flags |= REQ_F_NOWAIT; |
2422 | ||
b63534c4 JA |
2423 | if (kiocb->ki_flags & IOCB_DIRECT) |
2424 | io_get_req_task(req); | |
2425 | ||
8449eeda | 2426 | if (force_nonblock) |
2b188cc1 | 2427 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2428 | |
def596e9 | 2429 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2430 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2431 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2432 | return -EOPNOTSUPP; |
2b188cc1 | 2433 | |
def596e9 JA |
2434 | kiocb->ki_flags |= IOCB_HIPRI; |
2435 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2436 | req->iopoll_completed = 0; |
f3a6fa22 | 2437 | io_get_req_task(req); |
def596e9 | 2438 | } else { |
09bb8394 JA |
2439 | if (kiocb->ki_flags & IOCB_HIPRI) |
2440 | return -EINVAL; | |
def596e9 JA |
2441 | kiocb->ki_complete = io_complete_rw; |
2442 | } | |
9adbd45d | 2443 | |
3529d8c2 JA |
2444 | req->rw.addr = READ_ONCE(sqe->addr); |
2445 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2446 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2447 | return 0; |
2b188cc1 JA |
2448 | } |
2449 | ||
2450 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2451 | { | |
2452 | switch (ret) { | |
2453 | case -EIOCBQUEUED: | |
2454 | break; | |
2455 | case -ERESTARTSYS: | |
2456 | case -ERESTARTNOINTR: | |
2457 | case -ERESTARTNOHAND: | |
2458 | case -ERESTART_RESTARTBLOCK: | |
2459 | /* | |
2460 | * We can't just restart the syscall, since previously | |
2461 | * submitted sqes may already be in progress. Just fail this | |
2462 | * IO with EINTR. | |
2463 | */ | |
2464 | ret = -EINTR; | |
2465 | /* fall through */ | |
2466 | default: | |
2467 | kiocb->ki_complete(kiocb, ret, 0); | |
2468 | } | |
2469 | } | |
2470 | ||
a1d7c393 JA |
2471 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
2472 | struct io_comp_state *cs) | |
ba816ad6 | 2473 | { |
ba04291e JA |
2474 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2475 | ||
2476 | if (req->flags & REQ_F_CUR_POS) | |
2477 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2478 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
a1d7c393 | 2479 | __io_complete_rw(req, ret, 0, cs); |
ba816ad6 JA |
2480 | else |
2481 | io_rw_done(kiocb, ret); | |
2482 | } | |
2483 | ||
9adbd45d | 2484 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2485 | struct iov_iter *iter) |
edafccee | 2486 | { |
9adbd45d JA |
2487 | struct io_ring_ctx *ctx = req->ctx; |
2488 | size_t len = req->rw.len; | |
edafccee | 2489 | struct io_mapped_ubuf *imu; |
4f4eeba8 | 2490 | u16 index, buf_index; |
edafccee JA |
2491 | size_t offset; |
2492 | u64 buf_addr; | |
2493 | ||
2494 | /* attempt to use fixed buffers without having provided iovecs */ | |
2495 | if (unlikely(!ctx->user_bufs)) | |
2496 | return -EFAULT; | |
2497 | ||
4f4eeba8 | 2498 | buf_index = req->buf_index; |
edafccee JA |
2499 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2500 | return -EFAULT; | |
2501 | ||
2502 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2503 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2504 | buf_addr = req->rw.addr; |
edafccee JA |
2505 | |
2506 | /* overflow */ | |
2507 | if (buf_addr + len < buf_addr) | |
2508 | return -EFAULT; | |
2509 | /* not inside the mapped region */ | |
2510 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2511 | return -EFAULT; | |
2512 | ||
2513 | /* | |
2514 | * May not be a start of buffer, set size appropriately | |
2515 | * and advance us to the beginning. | |
2516 | */ | |
2517 | offset = buf_addr - imu->ubuf; | |
2518 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2519 | |
2520 | if (offset) { | |
2521 | /* | |
2522 | * Don't use iov_iter_advance() here, as it's really slow for | |
2523 | * using the latter parts of a big fixed buffer - it iterates | |
2524 | * over each segment manually. We can cheat a bit here, because | |
2525 | * we know that: | |
2526 | * | |
2527 | * 1) it's a BVEC iter, we set it up | |
2528 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2529 | * first and last bvec | |
2530 | * | |
2531 | * So just find our index, and adjust the iterator afterwards. | |
2532 | * If the offset is within the first bvec (or the whole first | |
2533 | * bvec, just use iov_iter_advance(). This makes it easier | |
2534 | * since we can just skip the first segment, which may not | |
2535 | * be PAGE_SIZE aligned. | |
2536 | */ | |
2537 | const struct bio_vec *bvec = imu->bvec; | |
2538 | ||
2539 | if (offset <= bvec->bv_len) { | |
2540 | iov_iter_advance(iter, offset); | |
2541 | } else { | |
2542 | unsigned long seg_skip; | |
2543 | ||
2544 | /* skip first vec */ | |
2545 | offset -= bvec->bv_len; | |
2546 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2547 | ||
2548 | iter->bvec = bvec + seg_skip; | |
2549 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2550 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2551 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2552 | } |
2553 | } | |
2554 | ||
5e559561 | 2555 | return len; |
edafccee JA |
2556 | } |
2557 | ||
bcda7baa JA |
2558 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2559 | { | |
2560 | if (needs_lock) | |
2561 | mutex_unlock(&ctx->uring_lock); | |
2562 | } | |
2563 | ||
2564 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2565 | { | |
2566 | /* | |
2567 | * "Normal" inline submissions always hold the uring_lock, since we | |
2568 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2569 | * The only exception is when we've detached the request and issue it | |
2570 | * from an async worker thread, grab the lock for that case. | |
2571 | */ | |
2572 | if (needs_lock) | |
2573 | mutex_lock(&ctx->uring_lock); | |
2574 | } | |
2575 | ||
2576 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2577 | int bgid, struct io_buffer *kbuf, | |
2578 | bool needs_lock) | |
2579 | { | |
2580 | struct io_buffer *head; | |
2581 | ||
2582 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2583 | return kbuf; | |
2584 | ||
2585 | io_ring_submit_lock(req->ctx, needs_lock); | |
2586 | ||
2587 | lockdep_assert_held(&req->ctx->uring_lock); | |
2588 | ||
2589 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2590 | if (head) { | |
2591 | if (!list_empty(&head->list)) { | |
2592 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2593 | list); | |
2594 | list_del(&kbuf->list); | |
2595 | } else { | |
2596 | kbuf = head; | |
2597 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2598 | } | |
2599 | if (*len > kbuf->len) | |
2600 | *len = kbuf->len; | |
2601 | } else { | |
2602 | kbuf = ERR_PTR(-ENOBUFS); | |
2603 | } | |
2604 | ||
2605 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2606 | ||
2607 | return kbuf; | |
2608 | } | |
2609 | ||
4d954c25 JA |
2610 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2611 | bool needs_lock) | |
2612 | { | |
2613 | struct io_buffer *kbuf; | |
4f4eeba8 | 2614 | u16 bgid; |
4d954c25 JA |
2615 | |
2616 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2617 | bgid = req->buf_index; |
4d954c25 JA |
2618 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2619 | if (IS_ERR(kbuf)) | |
2620 | return kbuf; | |
2621 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2622 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2623 | return u64_to_user_ptr(kbuf->addr); | |
2624 | } | |
2625 | ||
2626 | #ifdef CONFIG_COMPAT | |
2627 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2628 | bool needs_lock) | |
2629 | { | |
2630 | struct compat_iovec __user *uiov; | |
2631 | compat_ssize_t clen; | |
2632 | void __user *buf; | |
2633 | ssize_t len; | |
2634 | ||
2635 | uiov = u64_to_user_ptr(req->rw.addr); | |
2636 | if (!access_ok(uiov, sizeof(*uiov))) | |
2637 | return -EFAULT; | |
2638 | if (__get_user(clen, &uiov->iov_len)) | |
2639 | return -EFAULT; | |
2640 | if (clen < 0) | |
2641 | return -EINVAL; | |
2642 | ||
2643 | len = clen; | |
2644 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2645 | if (IS_ERR(buf)) | |
2646 | return PTR_ERR(buf); | |
2647 | iov[0].iov_base = buf; | |
2648 | iov[0].iov_len = (compat_size_t) len; | |
2649 | return 0; | |
2650 | } | |
2651 | #endif | |
2652 | ||
2653 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2654 | bool needs_lock) | |
2655 | { | |
2656 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2657 | void __user *buf; | |
2658 | ssize_t len; | |
2659 | ||
2660 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2661 | return -EFAULT; | |
2662 | ||
2663 | len = iov[0].iov_len; | |
2664 | if (len < 0) | |
2665 | return -EINVAL; | |
2666 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2667 | if (IS_ERR(buf)) | |
2668 | return PTR_ERR(buf); | |
2669 | iov[0].iov_base = buf; | |
2670 | iov[0].iov_len = len; | |
2671 | return 0; | |
2672 | } | |
2673 | ||
2674 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2675 | bool needs_lock) | |
2676 | { | |
dddb3e26 JA |
2677 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2678 | struct io_buffer *kbuf; | |
2679 | ||
2680 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2681 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2682 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2683 | return 0; |
dddb3e26 | 2684 | } |
4d954c25 JA |
2685 | if (!req->rw.len) |
2686 | return 0; | |
2687 | else if (req->rw.len > 1) | |
2688 | return -EINVAL; | |
2689 | ||
2690 | #ifdef CONFIG_COMPAT | |
2691 | if (req->ctx->compat) | |
2692 | return io_compat_import(req, iov, needs_lock); | |
2693 | #endif | |
2694 | ||
2695 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2696 | } | |
2697 | ||
cf6fd4bd | 2698 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2699 | struct iovec **iovec, struct iov_iter *iter, |
2700 | bool needs_lock) | |
2b188cc1 | 2701 | { |
9adbd45d JA |
2702 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2703 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2704 | ssize_t ret; |
edafccee JA |
2705 | u8 opcode; |
2706 | ||
d625c6ee | 2707 | opcode = req->opcode; |
7d009165 | 2708 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2709 | *iovec = NULL; |
9adbd45d | 2710 | return io_import_fixed(req, rw, iter); |
edafccee | 2711 | } |
2b188cc1 | 2712 | |
bcda7baa | 2713 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2714 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2715 | return -EINVAL; |
2716 | ||
3a6820f2 | 2717 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2718 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2719 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2720 | if (IS_ERR(buf)) { | |
bcda7baa | 2721 | *iovec = NULL; |
4d954c25 | 2722 | return PTR_ERR(buf); |
bcda7baa | 2723 | } |
3f9d6441 | 2724 | req->rw.len = sqe_len; |
bcda7baa JA |
2725 | } |
2726 | ||
3a6820f2 JA |
2727 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2728 | *iovec = NULL; | |
3a901598 | 2729 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2730 | } |
2731 | ||
f67676d1 JA |
2732 | if (req->io) { |
2733 | struct io_async_rw *iorw = &req->io->rw; | |
2734 | ||
2735 | *iovec = iorw->iov; | |
2736 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2737 | if (iorw->iov == iorw->fast_iov) | |
2738 | *iovec = NULL; | |
2739 | return iorw->size; | |
2740 | } | |
2741 | ||
4d954c25 JA |
2742 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2743 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2744 | if (!ret) { |
2745 | ret = (*iovec)->iov_len; | |
2746 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2747 | } | |
4d954c25 JA |
2748 | *iovec = NULL; |
2749 | return ret; | |
2750 | } | |
2751 | ||
2b188cc1 | 2752 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2753 | if (req->ctx->compat) |
2b188cc1 JA |
2754 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2755 | iovec, iter); | |
2756 | #endif | |
2757 | ||
2758 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2759 | } | |
2760 | ||
31b51510 | 2761 | /* |
32960613 JA |
2762 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2763 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2764 | */ |
32960613 JA |
2765 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2766 | struct iov_iter *iter) | |
2767 | { | |
2768 | ssize_t ret = 0; | |
2769 | ||
2770 | /* | |
2771 | * Don't support polled IO through this interface, and we can't | |
2772 | * support non-blocking either. For the latter, this just causes | |
2773 | * the kiocb to be handled from an async context. | |
2774 | */ | |
2775 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2776 | return -EOPNOTSUPP; | |
2777 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2778 | return -EAGAIN; | |
2779 | ||
2780 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2781 | struct iovec iovec; |
32960613 JA |
2782 | ssize_t nr; |
2783 | ||
311ae9e1 PB |
2784 | if (!iov_iter_is_bvec(iter)) { |
2785 | iovec = iov_iter_iovec(iter); | |
2786 | } else { | |
2787 | /* fixed buffers import bvec */ | |
2788 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2789 | + iter->iov_offset; | |
2790 | iovec.iov_len = min(iter->count, | |
2791 | iter->bvec->bv_len - iter->iov_offset); | |
2792 | } | |
2793 | ||
32960613 JA |
2794 | if (rw == READ) { |
2795 | nr = file->f_op->read(file, iovec.iov_base, | |
2796 | iovec.iov_len, &kiocb->ki_pos); | |
2797 | } else { | |
2798 | nr = file->f_op->write(file, iovec.iov_base, | |
2799 | iovec.iov_len, &kiocb->ki_pos); | |
2800 | } | |
2801 | ||
311ae9e1 PB |
2802 | if (iov_iter_is_bvec(iter)) |
2803 | kunmap(iter->bvec->bv_page); | |
2804 | ||
32960613 JA |
2805 | if (nr < 0) { |
2806 | if (!ret) | |
2807 | ret = nr; | |
2808 | break; | |
2809 | } | |
2810 | ret += nr; | |
2811 | if (nr != iovec.iov_len) | |
2812 | break; | |
2813 | iov_iter_advance(iter, nr); | |
2814 | } | |
2815 | ||
2816 | return ret; | |
2817 | } | |
2818 | ||
b7bb4f7d | 2819 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2820 | struct iovec *iovec, struct iovec *fast_iov, |
2821 | struct iov_iter *iter) | |
2822 | { | |
2823 | req->io->rw.nr_segs = iter->nr_segs; | |
2824 | req->io->rw.size = io_size; | |
2825 | req->io->rw.iov = iovec; | |
2826 | if (!req->io->rw.iov) { | |
2827 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2828 | if (req->io->rw.iov != fast_iov) |
2829 | memcpy(req->io->rw.iov, fast_iov, | |
2830 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2831 | } else { |
2832 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2833 | } |
2834 | } | |
2835 | ||
3d9932a8 XW |
2836 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2837 | { | |
2838 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2839 | return req->io == NULL; | |
2840 | } | |
2841 | ||
b7bb4f7d | 2842 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2843 | { |
d3656344 JA |
2844 | if (!io_op_defs[req->opcode].async_ctx) |
2845 | return 0; | |
3d9932a8 XW |
2846 | |
2847 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2848 | } |
2849 | ||
b7bb4f7d JA |
2850 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2851 | struct iovec *iovec, struct iovec *fast_iov, | |
2852 | struct iov_iter *iter) | |
2853 | { | |
980ad263 | 2854 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2855 | return 0; |
5d204bcf | 2856 | if (!req->io) { |
3d9932a8 | 2857 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2858 | return -ENOMEM; |
b7bb4f7d | 2859 | |
5d204bcf JA |
2860 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2861 | } | |
b7bb4f7d | 2862 | return 0; |
f67676d1 JA |
2863 | } |
2864 | ||
3529d8c2 JA |
2865 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2866 | bool force_nonblock) | |
f67676d1 | 2867 | { |
3529d8c2 JA |
2868 | struct io_async_ctx *io; |
2869 | struct iov_iter iter; | |
f67676d1 JA |
2870 | ssize_t ret; |
2871 | ||
3529d8c2 JA |
2872 | ret = io_prep_rw(req, sqe, force_nonblock); |
2873 | if (ret) | |
2874 | return ret; | |
f67676d1 | 2875 | |
3529d8c2 JA |
2876 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2877 | return -EBADF; | |
f67676d1 | 2878 | |
5f798bea PB |
2879 | /* either don't need iovec imported or already have it */ |
2880 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2881 | return 0; |
2882 | ||
2883 | io = req->io; | |
2884 | io->rw.iov = io->rw.fast_iov; | |
2885 | req->io = NULL; | |
bcda7baa | 2886 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2887 | req->io = io; |
2888 | if (ret < 0) | |
2889 | return ret; | |
2890 | ||
2891 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2892 | return 0; | |
f67676d1 JA |
2893 | } |
2894 | ||
bcf5a063 JA |
2895 | static void io_async_buf_cancel(struct callback_head *cb) |
2896 | { | |
2897 | struct io_async_rw *rw; | |
2898 | struct io_kiocb *req; | |
2899 | ||
2900 | rw = container_of(cb, struct io_async_rw, task_work); | |
2901 | req = rw->wpq.wait.private; | |
c40f6379 | 2902 | __io_req_task_cancel(req, -ECANCELED); |
bcf5a063 JA |
2903 | } |
2904 | ||
2905 | static void io_async_buf_retry(struct callback_head *cb) | |
2906 | { | |
2907 | struct io_async_rw *rw; | |
bcf5a063 JA |
2908 | struct io_kiocb *req; |
2909 | ||
2910 | rw = container_of(cb, struct io_async_rw, task_work); | |
2911 | req = rw->wpq.wait.private; | |
bcf5a063 | 2912 | |
c40f6379 | 2913 | __io_req_task_submit(req); |
bcf5a063 JA |
2914 | } |
2915 | ||
2916 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, | |
2917 | int sync, void *arg) | |
2918 | { | |
2919 | struct wait_page_queue *wpq; | |
2920 | struct io_kiocb *req = wait->private; | |
2921 | struct io_async_rw *rw = &req->io->rw; | |
2922 | struct wait_page_key *key = arg; | |
bcf5a063 JA |
2923 | int ret; |
2924 | ||
2925 | wpq = container_of(wait, struct wait_page_queue, wait); | |
2926 | ||
2927 | ret = wake_page_match(wpq, key); | |
2928 | if (ret != 1) | |
2929 | return ret; | |
2930 | ||
2931 | list_del_init(&wait->entry); | |
2932 | ||
2933 | init_task_work(&rw->task_work, io_async_buf_retry); | |
2934 | /* submit ref gets dropped, acquire a new one */ | |
2935 | refcount_inc(&req->refs); | |
c2c4c83c | 2936 | ret = io_req_task_work_add(req, &rw->task_work); |
bcf5a063 | 2937 | if (unlikely(ret)) { |
c2c4c83c JA |
2938 | struct task_struct *tsk; |
2939 | ||
bcf5a063 JA |
2940 | /* queue just for cancelation */ |
2941 | init_task_work(&rw->task_work, io_async_buf_cancel); | |
2942 | tsk = io_wq_get_task(req->ctx->io_wq); | |
c2c4c83c JA |
2943 | task_work_add(tsk, &rw->task_work, 0); |
2944 | wake_up_process(tsk); | |
bcf5a063 | 2945 | } |
bcf5a063 JA |
2946 | return 1; |
2947 | } | |
2948 | ||
2949 | static bool io_rw_should_retry(struct io_kiocb *req) | |
2950 | { | |
2951 | struct kiocb *kiocb = &req->rw.kiocb; | |
2952 | int ret; | |
2953 | ||
2954 | /* never retry for NOWAIT, we just complete with -EAGAIN */ | |
2955 | if (req->flags & REQ_F_NOWAIT) | |
2956 | return false; | |
2957 | ||
2958 | /* already tried, or we're doing O_DIRECT */ | |
2959 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_WAITQ)) | |
2960 | return false; | |
2961 | /* | |
2962 | * just use poll if we can, and don't attempt if the fs doesn't | |
2963 | * support callback based unlocks | |
2964 | */ | |
2965 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
2966 | return false; | |
2967 | ||
2968 | /* | |
2969 | * If request type doesn't require req->io to defer in general, | |
2970 | * we need to allocate it here | |
2971 | */ | |
2972 | if (!req->io && __io_alloc_async_ctx(req)) | |
2973 | return false; | |
2974 | ||
2975 | ret = kiocb_wait_page_queue_init(kiocb, &req->io->rw.wpq, | |
2976 | io_async_buf_func, req); | |
2977 | if (!ret) { | |
2978 | io_get_req_task(req); | |
2979 | return true; | |
2980 | } | |
2981 | ||
2982 | return false; | |
2983 | } | |
2984 | ||
2985 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
2986 | { | |
2987 | if (req->file->f_op->read_iter) | |
2988 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2989 | return loop_rw_iter(READ, req->file, &req->rw.kiocb, iter); | |
2990 | } | |
2991 | ||
a1d7c393 JA |
2992 | static int io_read(struct io_kiocb *req, bool force_nonblock, |
2993 | struct io_comp_state *cs) | |
2b188cc1 JA |
2994 | { |
2995 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2996 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2997 | struct iov_iter iter; |
31b51510 | 2998 | size_t iov_count; |
f67676d1 | 2999 | ssize_t io_size, ret; |
2b188cc1 | 3000 | |
bcda7baa | 3001 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
3002 | if (ret < 0) |
3003 | return ret; | |
2b188cc1 | 3004 | |
fd6c2e4c JA |
3005 | /* Ensure we clear previously set non-block flag */ |
3006 | if (!force_nonblock) | |
29de5f6a | 3007 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3008 | |
f67676d1 | 3009 | io_size = ret; |
6795c5ab | 3010 | req->result = io_size; |
f67676d1 | 3011 | |
24c74678 | 3012 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3013 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 3014 | goto copy_iov; |
9e645e11 | 3015 | |
31b51510 | 3016 | iov_count = iov_iter_count(&iter); |
9adbd45d | 3017 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 3018 | if (!ret) { |
b63534c4 | 3019 | unsigned long nr_segs = iter.nr_segs; |
4503b767 | 3020 | ssize_t ret2 = 0; |
2b188cc1 | 3021 | |
bcf5a063 | 3022 | ret2 = io_iter_do_read(req, &iter); |
32960613 | 3023 | |
9d93a3f5 | 3024 | /* Catch -EAGAIN return for forced non-blocking submission */ |
4503b767 | 3025 | if (!force_nonblock || (ret2 != -EAGAIN && ret2 != -EIO)) { |
a1d7c393 | 3026 | kiocb_done(kiocb, ret2, cs); |
f67676d1 | 3027 | } else { |
b63534c4 JA |
3028 | iter.count = iov_count; |
3029 | iter.nr_segs = nr_segs; | |
f67676d1 | 3030 | copy_iov: |
b7bb4f7d | 3031 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
3032 | inline_vecs, &iter); |
3033 | if (ret) | |
3034 | goto out_free; | |
bcf5a063 JA |
3035 | /* if we can retry, do so with the callbacks armed */ |
3036 | if (io_rw_should_retry(req)) { | |
3037 | ret2 = io_iter_do_read(req, &iter); | |
3038 | if (ret2 == -EIOCBQUEUED) { | |
3039 | goto out_free; | |
3040 | } else if (ret2 != -EAGAIN) { | |
a1d7c393 | 3041 | kiocb_done(kiocb, ret2, cs); |
bcf5a063 JA |
3042 | goto out_free; |
3043 | } | |
3044 | } | |
3045 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
f67676d1 JA |
3046 | return -EAGAIN; |
3047 | } | |
2b188cc1 | 3048 | } |
f67676d1 | 3049 | out_free: |
6f2cc166 XW |
3050 | if (!(req->flags & REQ_F_NEED_CLEANUP)) |
3051 | kfree(iovec); | |
2b188cc1 JA |
3052 | return ret; |
3053 | } | |
3054 | ||
3529d8c2 JA |
3055 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3056 | bool force_nonblock) | |
f67676d1 | 3057 | { |
3529d8c2 JA |
3058 | struct io_async_ctx *io; |
3059 | struct iov_iter iter; | |
f67676d1 JA |
3060 | ssize_t ret; |
3061 | ||
3529d8c2 JA |
3062 | ret = io_prep_rw(req, sqe, force_nonblock); |
3063 | if (ret) | |
3064 | return ret; | |
f67676d1 | 3065 | |
3529d8c2 JA |
3066 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3067 | return -EBADF; | |
f67676d1 | 3068 | |
4ed734b0 JA |
3069 | req->fsize = rlimit(RLIMIT_FSIZE); |
3070 | ||
5f798bea PB |
3071 | /* either don't need iovec imported or already have it */ |
3072 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
3073 | return 0; |
3074 | ||
3075 | io = req->io; | |
3076 | io->rw.iov = io->rw.fast_iov; | |
3077 | req->io = NULL; | |
bcda7baa | 3078 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
3079 | req->io = io; |
3080 | if (ret < 0) | |
3081 | return ret; | |
3082 | ||
3083 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
3084 | return 0; | |
f67676d1 JA |
3085 | } |
3086 | ||
a1d7c393 JA |
3087 | static int io_write(struct io_kiocb *req, bool force_nonblock, |
3088 | struct io_comp_state *cs) | |
2b188cc1 JA |
3089 | { |
3090 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3091 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 3092 | struct iov_iter iter; |
31b51510 | 3093 | size_t iov_count; |
f67676d1 | 3094 | ssize_t ret, io_size; |
2b188cc1 | 3095 | |
bcda7baa | 3096 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
3097 | if (ret < 0) |
3098 | return ret; | |
2b188cc1 | 3099 | |
fd6c2e4c JA |
3100 | /* Ensure we clear previously set non-block flag */ |
3101 | if (!force_nonblock) | |
9adbd45d | 3102 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 3103 | |
f67676d1 | 3104 | io_size = ret; |
6795c5ab | 3105 | req->result = io_size; |
9e645e11 | 3106 | |
24c74678 | 3107 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3108 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3109 | goto copy_iov; |
31b51510 | 3110 | |
10d59345 JA |
3111 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3112 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3113 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3114 | goto copy_iov; |
31b51510 | 3115 | |
f67676d1 | 3116 | iov_count = iov_iter_count(&iter); |
9adbd45d | 3117 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 3118 | if (!ret) { |
b63534c4 | 3119 | unsigned long nr_segs = iter.nr_segs; |
9bf7933f RP |
3120 | ssize_t ret2; |
3121 | ||
2b188cc1 JA |
3122 | /* |
3123 | * Open-code file_start_write here to grab freeze protection, | |
3124 | * which will be released by another thread in | |
3125 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3126 | * released so that it doesn't complain about the held lock when | |
3127 | * we return to userspace. | |
3128 | */ | |
491381ce | 3129 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 3130 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 3131 | SB_FREEZE_WRITE, true); |
9adbd45d | 3132 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
3133 | SB_FREEZE_WRITE); |
3134 | } | |
3135 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 3136 | |
4ed734b0 JA |
3137 | if (!force_nonblock) |
3138 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
3139 | ||
9adbd45d JA |
3140 | if (req->file->f_op->write_iter) |
3141 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 3142 | else |
9adbd45d | 3143 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
3144 | |
3145 | if (!force_nonblock) | |
3146 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
3147 | ||
faac996c | 3148 | /* |
bff6035d | 3149 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
3150 | * retry them without IOCB_NOWAIT. |
3151 | */ | |
3152 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3153 | ret2 = -EAGAIN; | |
f67676d1 | 3154 | if (!force_nonblock || ret2 != -EAGAIN) { |
a1d7c393 | 3155 | kiocb_done(kiocb, ret2, cs); |
f67676d1 | 3156 | } else { |
b63534c4 JA |
3157 | iter.count = iov_count; |
3158 | iter.nr_segs = nr_segs; | |
f67676d1 | 3159 | copy_iov: |
b7bb4f7d | 3160 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
3161 | inline_vecs, &iter); |
3162 | if (ret) | |
3163 | goto out_free; | |
3164 | return -EAGAIN; | |
3165 | } | |
2b188cc1 | 3166 | } |
31b51510 | 3167 | out_free: |
6f2cc166 XW |
3168 | if (!(req->flags & REQ_F_NEED_CLEANUP)) |
3169 | kfree(iovec); | |
2b188cc1 JA |
3170 | return ret; |
3171 | } | |
3172 | ||
f2a8d5c7 PB |
3173 | static int __io_splice_prep(struct io_kiocb *req, |
3174 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3175 | { |
3176 | struct io_splice* sp = &req->splice; | |
3177 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
3178 | int ret; | |
3179 | ||
3180 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3181 | return 0; | |
3232dd02 PB |
3182 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3183 | return -EINVAL; | |
7d67af2c PB |
3184 | |
3185 | sp->file_in = NULL; | |
7d67af2c PB |
3186 | sp->len = READ_ONCE(sqe->len); |
3187 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3188 | ||
3189 | if (unlikely(sp->flags & ~valid_flags)) | |
3190 | return -EINVAL; | |
3191 | ||
3192 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
3193 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3194 | if (ret) | |
3195 | return ret; | |
3196 | req->flags |= REQ_F_NEED_CLEANUP; | |
3197 | ||
7cdaf587 XW |
3198 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3199 | /* | |
3200 | * Splice operation will be punted aync, and here need to | |
3201 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3202 | */ | |
3203 | io_req_init_async(req); | |
7d67af2c | 3204 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3205 | } |
7d67af2c PB |
3206 | |
3207 | return 0; | |
3208 | } | |
3209 | ||
f2a8d5c7 PB |
3210 | static int io_tee_prep(struct io_kiocb *req, |
3211 | const struct io_uring_sqe *sqe) | |
3212 | { | |
3213 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3214 | return -EINVAL; | |
3215 | return __io_splice_prep(req, sqe); | |
3216 | } | |
3217 | ||
3218 | static int io_tee(struct io_kiocb *req, bool force_nonblock) | |
3219 | { | |
3220 | struct io_splice *sp = &req->splice; | |
3221 | struct file *in = sp->file_in; | |
3222 | struct file *out = sp->file_out; | |
3223 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3224 | long ret = 0; | |
3225 | ||
3226 | if (force_nonblock) | |
3227 | return -EAGAIN; | |
3228 | if (sp->len) | |
3229 | ret = do_tee(in, out, sp->len, flags); | |
3230 | ||
3231 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3232 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3233 | ||
f2a8d5c7 PB |
3234 | if (ret != sp->len) |
3235 | req_set_fail_links(req); | |
e1e16097 | 3236 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3237 | return 0; |
3238 | } | |
3239 | ||
3240 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3241 | { | |
3242 | struct io_splice* sp = &req->splice; | |
3243 | ||
3244 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3245 | sp->off_out = READ_ONCE(sqe->off); | |
3246 | return __io_splice_prep(req, sqe); | |
3247 | } | |
3248 | ||
014db007 | 3249 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
3250 | { |
3251 | struct io_splice *sp = &req->splice; | |
3252 | struct file *in = sp->file_in; | |
3253 | struct file *out = sp->file_out; | |
3254 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3255 | loff_t *poff_in, *poff_out; | |
c9687426 | 3256 | long ret = 0; |
7d67af2c | 3257 | |
2fb3e822 PB |
3258 | if (force_nonblock) |
3259 | return -EAGAIN; | |
7d67af2c PB |
3260 | |
3261 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3262 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3263 | |
948a7749 | 3264 | if (sp->len) |
c9687426 | 3265 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3266 | |
3267 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3268 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3269 | ||
7d67af2c PB |
3270 | if (ret != sp->len) |
3271 | req_set_fail_links(req); | |
e1e16097 | 3272 | io_req_complete(req, ret); |
7d67af2c PB |
3273 | return 0; |
3274 | } | |
3275 | ||
2b188cc1 JA |
3276 | /* |
3277 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3278 | */ | |
229a7b63 | 3279 | static int io_nop(struct io_kiocb *req, struct io_comp_state *cs) |
2b188cc1 JA |
3280 | { |
3281 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3282 | |
def596e9 JA |
3283 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3284 | return -EINVAL; | |
3285 | ||
229a7b63 | 3286 | __io_req_complete(req, 0, 0, cs); |
2b188cc1 JA |
3287 | return 0; |
3288 | } | |
3289 | ||
3529d8c2 | 3290 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3291 | { |
6b06314c | 3292 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3293 | |
09bb8394 JA |
3294 | if (!req->file) |
3295 | return -EBADF; | |
c992fe29 | 3296 | |
6b06314c | 3297 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3298 | return -EINVAL; |
edafccee | 3299 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3300 | return -EINVAL; |
3301 | ||
8ed8d3c3 JA |
3302 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3303 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3304 | return -EINVAL; | |
3305 | ||
3306 | req->sync.off = READ_ONCE(sqe->off); | |
3307 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3308 | return 0; |
3309 | } | |
3310 | ||
ac45abc0 | 3311 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3312 | { |
8ed8d3c3 | 3313 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3314 | int ret; |
3315 | ||
ac45abc0 PB |
3316 | /* fsync always requires a blocking context */ |
3317 | if (force_nonblock) | |
3318 | return -EAGAIN; | |
3319 | ||
9adbd45d | 3320 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3321 | end > 0 ? end : LLONG_MAX, |
3322 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3323 | if (ret < 0) | |
3324 | req_set_fail_links(req); | |
e1e16097 | 3325 | io_req_complete(req, ret); |
c992fe29 CH |
3326 | return 0; |
3327 | } | |
3328 | ||
d63d1b5e JA |
3329 | static int io_fallocate_prep(struct io_kiocb *req, |
3330 | const struct io_uring_sqe *sqe) | |
3331 | { | |
3332 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3333 | return -EINVAL; | |
3232dd02 PB |
3334 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3335 | return -EINVAL; | |
d63d1b5e JA |
3336 | |
3337 | req->sync.off = READ_ONCE(sqe->off); | |
3338 | req->sync.len = READ_ONCE(sqe->addr); | |
3339 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 3340 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
3341 | return 0; |
3342 | } | |
3343 | ||
014db007 | 3344 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3345 | { |
ac45abc0 PB |
3346 | int ret; |
3347 | ||
d63d1b5e | 3348 | /* fallocate always requiring blocking context */ |
ac45abc0 | 3349 | if (force_nonblock) |
5d17b4a4 JA |
3350 | return -EAGAIN; |
3351 | ||
ac45abc0 PB |
3352 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
3353 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, | |
3354 | req->sync.len); | |
3355 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
3356 | if (ret < 0) | |
3357 | req_set_fail_links(req); | |
e1e16097 | 3358 | io_req_complete(req, ret); |
5d17b4a4 JA |
3359 | return 0; |
3360 | } | |
3361 | ||
ec65fea5 | 3362 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3363 | { |
f8748881 | 3364 | const char __user *fname; |
15b71abe | 3365 | int ret; |
b7bb4f7d | 3366 | |
3232dd02 | 3367 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
15b71abe | 3368 | return -EINVAL; |
ec65fea5 | 3369 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3370 | return -EINVAL; |
ec65fea5 | 3371 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3372 | return -EBADF; |
03b1230c | 3373 | |
ec65fea5 PB |
3374 | /* open.how should be already initialised */ |
3375 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3376 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3377 | |
25e72d10 PB |
3378 | req->open.dfd = READ_ONCE(sqe->fd); |
3379 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3380 | req->open.filename = getname(fname); |
15b71abe JA |
3381 | if (IS_ERR(req->open.filename)) { |
3382 | ret = PTR_ERR(req->open.filename); | |
3383 | req->open.filename = NULL; | |
3384 | return ret; | |
3385 | } | |
4022e7af | 3386 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3387 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3388 | return 0; |
03b1230c JA |
3389 | } |
3390 | ||
ec65fea5 PB |
3391 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3392 | { | |
3393 | u64 flags, mode; | |
3394 | ||
3395 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3396 | return 0; | |
3397 | mode = READ_ONCE(sqe->len); | |
3398 | flags = READ_ONCE(sqe->open_flags); | |
3399 | req->open.how = build_open_how(flags, mode); | |
3400 | return __io_openat_prep(req, sqe); | |
3401 | } | |
3402 | ||
cebdb986 | 3403 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3404 | { |
cebdb986 | 3405 | struct open_how __user *how; |
cebdb986 | 3406 | size_t len; |
0fa03c62 JA |
3407 | int ret; |
3408 | ||
0bdbdd08 PB |
3409 | if (req->flags & REQ_F_NEED_CLEANUP) |
3410 | return 0; | |
cebdb986 JA |
3411 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3412 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3413 | if (len < OPEN_HOW_SIZE_VER0) |
3414 | return -EINVAL; | |
3529d8c2 | 3415 | |
cebdb986 JA |
3416 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3417 | len); | |
3418 | if (ret) | |
3419 | return ret; | |
3529d8c2 | 3420 | |
ec65fea5 | 3421 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3422 | } |
3423 | ||
014db007 | 3424 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3425 | { |
3426 | struct open_flags op; | |
15b71abe JA |
3427 | struct file *file; |
3428 | int ret; | |
3429 | ||
f86cd20c | 3430 | if (force_nonblock) |
15b71abe | 3431 | return -EAGAIN; |
15b71abe | 3432 | |
cebdb986 | 3433 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3434 | if (ret) |
3435 | goto err; | |
3436 | ||
4022e7af | 3437 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3438 | if (ret < 0) |
3439 | goto err; | |
3440 | ||
3441 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3442 | if (IS_ERR(file)) { | |
3443 | put_unused_fd(ret); | |
3444 | ret = PTR_ERR(file); | |
3445 | } else { | |
3446 | fsnotify_open(file); | |
3447 | fd_install(ret, file); | |
3448 | } | |
3449 | err: | |
3450 | putname(req->open.filename); | |
8fef80bf | 3451 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3452 | if (ret < 0) |
3453 | req_set_fail_links(req); | |
e1e16097 | 3454 | io_req_complete(req, ret); |
15b71abe JA |
3455 | return 0; |
3456 | } | |
3457 | ||
014db007 | 3458 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 | 3459 | { |
014db007 | 3460 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3461 | } |
3462 | ||
067524e9 JA |
3463 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3464 | const struct io_uring_sqe *sqe) | |
3465 | { | |
3466 | struct io_provide_buf *p = &req->pbuf; | |
3467 | u64 tmp; | |
3468 | ||
3469 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3470 | return -EINVAL; | |
3471 | ||
3472 | tmp = READ_ONCE(sqe->fd); | |
3473 | if (!tmp || tmp > USHRT_MAX) | |
3474 | return -EINVAL; | |
3475 | ||
3476 | memset(p, 0, sizeof(*p)); | |
3477 | p->nbufs = tmp; | |
3478 | p->bgid = READ_ONCE(sqe->buf_group); | |
3479 | return 0; | |
3480 | } | |
3481 | ||
3482 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3483 | int bgid, unsigned nbufs) | |
3484 | { | |
3485 | unsigned i = 0; | |
3486 | ||
3487 | /* shouldn't happen */ | |
3488 | if (!nbufs) | |
3489 | return 0; | |
3490 | ||
3491 | /* the head kbuf is the list itself */ | |
3492 | while (!list_empty(&buf->list)) { | |
3493 | struct io_buffer *nxt; | |
3494 | ||
3495 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3496 | list_del(&nxt->list); | |
3497 | kfree(nxt); | |
3498 | if (++i == nbufs) | |
3499 | return i; | |
3500 | } | |
3501 | i++; | |
3502 | kfree(buf); | |
3503 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3504 | ||
3505 | return i; | |
3506 | } | |
3507 | ||
229a7b63 JA |
3508 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock, |
3509 | struct io_comp_state *cs) | |
067524e9 JA |
3510 | { |
3511 | struct io_provide_buf *p = &req->pbuf; | |
3512 | struct io_ring_ctx *ctx = req->ctx; | |
3513 | struct io_buffer *head; | |
3514 | int ret = 0; | |
3515 | ||
3516 | io_ring_submit_lock(ctx, !force_nonblock); | |
3517 | ||
3518 | lockdep_assert_held(&ctx->uring_lock); | |
3519 | ||
3520 | ret = -ENOENT; | |
3521 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3522 | if (head) | |
3523 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3524 | ||
3525 | io_ring_submit_lock(ctx, !force_nonblock); | |
3526 | if (ret < 0) | |
3527 | req_set_fail_links(req); | |
229a7b63 | 3528 | __io_req_complete(req, ret, 0, cs); |
067524e9 JA |
3529 | return 0; |
3530 | } | |
3531 | ||
ddf0322d JA |
3532 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3533 | const struct io_uring_sqe *sqe) | |
3534 | { | |
3535 | struct io_provide_buf *p = &req->pbuf; | |
3536 | u64 tmp; | |
3537 | ||
3538 | if (sqe->ioprio || sqe->rw_flags) | |
3539 | return -EINVAL; | |
3540 | ||
3541 | tmp = READ_ONCE(sqe->fd); | |
3542 | if (!tmp || tmp > USHRT_MAX) | |
3543 | return -E2BIG; | |
3544 | p->nbufs = tmp; | |
3545 | p->addr = READ_ONCE(sqe->addr); | |
3546 | p->len = READ_ONCE(sqe->len); | |
3547 | ||
efe68c1c | 3548 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3549 | return -EFAULT; |
3550 | ||
3551 | p->bgid = READ_ONCE(sqe->buf_group); | |
3552 | tmp = READ_ONCE(sqe->off); | |
3553 | if (tmp > USHRT_MAX) | |
3554 | return -E2BIG; | |
3555 | p->bid = tmp; | |
3556 | return 0; | |
3557 | } | |
3558 | ||
3559 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3560 | { | |
3561 | struct io_buffer *buf; | |
3562 | u64 addr = pbuf->addr; | |
3563 | int i, bid = pbuf->bid; | |
3564 | ||
3565 | for (i = 0; i < pbuf->nbufs; i++) { | |
3566 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3567 | if (!buf) | |
3568 | break; | |
3569 | ||
3570 | buf->addr = addr; | |
3571 | buf->len = pbuf->len; | |
3572 | buf->bid = bid; | |
3573 | addr += pbuf->len; | |
3574 | bid++; | |
3575 | if (!*head) { | |
3576 | INIT_LIST_HEAD(&buf->list); | |
3577 | *head = buf; | |
3578 | } else { | |
3579 | list_add_tail(&buf->list, &(*head)->list); | |
3580 | } | |
3581 | } | |
3582 | ||
3583 | return i ? i : -ENOMEM; | |
3584 | } | |
3585 | ||
229a7b63 JA |
3586 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock, |
3587 | struct io_comp_state *cs) | |
ddf0322d JA |
3588 | { |
3589 | struct io_provide_buf *p = &req->pbuf; | |
3590 | struct io_ring_ctx *ctx = req->ctx; | |
3591 | struct io_buffer *head, *list; | |
3592 | int ret = 0; | |
3593 | ||
3594 | io_ring_submit_lock(ctx, !force_nonblock); | |
3595 | ||
3596 | lockdep_assert_held(&ctx->uring_lock); | |
3597 | ||
3598 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3599 | ||
3600 | ret = io_add_buffers(p, &head); | |
3601 | if (ret < 0) | |
3602 | goto out; | |
3603 | ||
3604 | if (!list) { | |
3605 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3606 | GFP_KERNEL); | |
3607 | if (ret < 0) { | |
067524e9 | 3608 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3609 | goto out; |
3610 | } | |
3611 | } | |
3612 | out: | |
3613 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3614 | if (ret < 0) | |
3615 | req_set_fail_links(req); | |
229a7b63 | 3616 | __io_req_complete(req, ret, 0, cs); |
ddf0322d | 3617 | return 0; |
cebdb986 JA |
3618 | } |
3619 | ||
3e4827b0 JA |
3620 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3621 | const struct io_uring_sqe *sqe) | |
3622 | { | |
3623 | #if defined(CONFIG_EPOLL) | |
3624 | if (sqe->ioprio || sqe->buf_index) | |
3625 | return -EINVAL; | |
3232dd02 PB |
3626 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3627 | return -EINVAL; | |
3e4827b0 JA |
3628 | |
3629 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3630 | req->epoll.op = READ_ONCE(sqe->len); | |
3631 | req->epoll.fd = READ_ONCE(sqe->off); | |
3632 | ||
3633 | if (ep_op_has_event(req->epoll.op)) { | |
3634 | struct epoll_event __user *ev; | |
3635 | ||
3636 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3637 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3638 | return -EFAULT; | |
3639 | } | |
3640 | ||
3641 | return 0; | |
3642 | #else | |
3643 | return -EOPNOTSUPP; | |
3644 | #endif | |
3645 | } | |
3646 | ||
229a7b63 JA |
3647 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock, |
3648 | struct io_comp_state *cs) | |
3e4827b0 JA |
3649 | { |
3650 | #if defined(CONFIG_EPOLL) | |
3651 | struct io_epoll *ie = &req->epoll; | |
3652 | int ret; | |
3653 | ||
3654 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3655 | if (force_nonblock && ret == -EAGAIN) | |
3656 | return -EAGAIN; | |
3657 | ||
3658 | if (ret < 0) | |
3659 | req_set_fail_links(req); | |
229a7b63 | 3660 | __io_req_complete(req, ret, 0, cs); |
3e4827b0 JA |
3661 | return 0; |
3662 | #else | |
3663 | return -EOPNOTSUPP; | |
3664 | #endif | |
3665 | } | |
3666 | ||
c1ca757b JA |
3667 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3668 | { | |
3669 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3670 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3671 | return -EINVAL; | |
3232dd02 PB |
3672 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3673 | return -EINVAL; | |
c1ca757b JA |
3674 | |
3675 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3676 | req->madvise.len = READ_ONCE(sqe->len); | |
3677 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3678 | return 0; | |
3679 | #else | |
3680 | return -EOPNOTSUPP; | |
3681 | #endif | |
3682 | } | |
3683 | ||
014db007 | 3684 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3685 | { |
3686 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3687 | struct io_madvise *ma = &req->madvise; | |
3688 | int ret; | |
3689 | ||
3690 | if (force_nonblock) | |
3691 | return -EAGAIN; | |
3692 | ||
3693 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3694 | if (ret < 0) | |
3695 | req_set_fail_links(req); | |
e1e16097 | 3696 | io_req_complete(req, ret); |
c1ca757b JA |
3697 | return 0; |
3698 | #else | |
3699 | return -EOPNOTSUPP; | |
3700 | #endif | |
3701 | } | |
3702 | ||
4840e418 JA |
3703 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3704 | { | |
3705 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3706 | return -EINVAL; | |
3232dd02 PB |
3707 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3708 | return -EINVAL; | |
4840e418 JA |
3709 | |
3710 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3711 | req->fadvise.len = READ_ONCE(sqe->len); | |
3712 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3713 | return 0; | |
3714 | } | |
3715 | ||
014db007 | 3716 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3717 | { |
3718 | struct io_fadvise *fa = &req->fadvise; | |
3719 | int ret; | |
3720 | ||
3e69426d JA |
3721 | if (force_nonblock) { |
3722 | switch (fa->advice) { | |
3723 | case POSIX_FADV_NORMAL: | |
3724 | case POSIX_FADV_RANDOM: | |
3725 | case POSIX_FADV_SEQUENTIAL: | |
3726 | break; | |
3727 | default: | |
3728 | return -EAGAIN; | |
3729 | } | |
3730 | } | |
4840e418 JA |
3731 | |
3732 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3733 | if (ret < 0) | |
3734 | req_set_fail_links(req); | |
e1e16097 | 3735 | io_req_complete(req, ret); |
4840e418 JA |
3736 | return 0; |
3737 | } | |
3738 | ||
eddc7ef5 JA |
3739 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3740 | { | |
3232dd02 PB |
3741 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3742 | return -EINVAL; | |
eddc7ef5 JA |
3743 | if (sqe->ioprio || sqe->buf_index) |
3744 | return -EINVAL; | |
9c280f90 | 3745 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3746 | return -EBADF; |
eddc7ef5 | 3747 | |
1d9e1288 BM |
3748 | req->statx.dfd = READ_ONCE(sqe->fd); |
3749 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 3750 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
3751 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3752 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
3753 | |
3754 | return 0; | |
3755 | } | |
3756 | ||
014db007 | 3757 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 | 3758 | { |
1d9e1288 | 3759 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
3760 | int ret; |
3761 | ||
5b0bbee4 JA |
3762 | if (force_nonblock) { |
3763 | /* only need file table for an actual valid fd */ | |
3764 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3765 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3766 | return -EAGAIN; |
5b0bbee4 | 3767 | } |
eddc7ef5 | 3768 | |
e62753e4 BM |
3769 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
3770 | ctx->buffer); | |
eddc7ef5 | 3771 | |
eddc7ef5 JA |
3772 | if (ret < 0) |
3773 | req_set_fail_links(req); | |
e1e16097 | 3774 | io_req_complete(req, ret); |
eddc7ef5 JA |
3775 | return 0; |
3776 | } | |
3777 | ||
b5dba59e JA |
3778 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3779 | { | |
3780 | /* | |
3781 | * If we queue this for async, it must not be cancellable. That would | |
7cdaf587 XW |
3782 | * leave the 'file' in an undeterminate state, and here need to modify |
3783 | * io_wq_work.flags, so initialize io_wq_work firstly. | |
b5dba59e | 3784 | */ |
7cdaf587 | 3785 | io_req_init_async(req); |
b5dba59e JA |
3786 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; |
3787 | ||
3232dd02 PB |
3788 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3789 | return -EINVAL; | |
b5dba59e JA |
3790 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
3791 | sqe->rw_flags || sqe->buf_index) | |
3792 | return -EINVAL; | |
9c280f90 | 3793 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3794 | return -EBADF; |
b5dba59e JA |
3795 | |
3796 | req->close.fd = READ_ONCE(sqe->fd); | |
fd2206e4 JA |
3797 | if ((req->file && req->file->f_op == &io_uring_fops) || |
3798 | req->close.fd == req->ctx->ring_fd) | |
3799 | return -EBADF; | |
b5dba59e | 3800 | |
3af73b28 | 3801 | req->close.put_file = NULL; |
b5dba59e | 3802 | return 0; |
b5dba59e JA |
3803 | } |
3804 | ||
229a7b63 JA |
3805 | static int io_close(struct io_kiocb *req, bool force_nonblock, |
3806 | struct io_comp_state *cs) | |
b5dba59e | 3807 | { |
3af73b28 | 3808 | struct io_close *close = &req->close; |
b5dba59e JA |
3809 | int ret; |
3810 | ||
3af73b28 PB |
3811 | /* might be already done during nonblock submission */ |
3812 | if (!close->put_file) { | |
3813 | ret = __close_fd_get_file(close->fd, &close->put_file); | |
3814 | if (ret < 0) | |
3815 | return (ret == -ENOENT) ? -EBADF : ret; | |
3816 | } | |
b5dba59e JA |
3817 | |
3818 | /* if the file has a flush method, be safe and punt to async */ | |
3af73b28 | 3819 | if (close->put_file->f_op->flush && force_nonblock) { |
24c74678 PB |
3820 | /* was never set, but play safe */ |
3821 | req->flags &= ~REQ_F_NOWAIT; | |
0bf0eefd | 3822 | /* avoid grabbing files - we don't need the files */ |
24c74678 | 3823 | req->flags |= REQ_F_NO_FILE_TABLE; |
0bf0eefd | 3824 | return -EAGAIN; |
a2100672 | 3825 | } |
b5dba59e | 3826 | |
3af73b28 PB |
3827 | /* No ->flush() or already async, safely close from here */ |
3828 | ret = filp_close(close->put_file, req->work.files); | |
3829 | if (ret < 0) | |
3830 | req_set_fail_links(req); | |
3af73b28 PB |
3831 | fput(close->put_file); |
3832 | close->put_file = NULL; | |
229a7b63 | 3833 | __io_req_complete(req, ret, 0, cs); |
1a417f4e | 3834 | return 0; |
b5dba59e JA |
3835 | } |
3836 | ||
3529d8c2 | 3837 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3838 | { |
3839 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3840 | |
3841 | if (!req->file) | |
3842 | return -EBADF; | |
5d17b4a4 JA |
3843 | |
3844 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3845 | return -EINVAL; | |
3846 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3847 | return -EINVAL; | |
3848 | ||
8ed8d3c3 JA |
3849 | req->sync.off = READ_ONCE(sqe->off); |
3850 | req->sync.len = READ_ONCE(sqe->len); | |
3851 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3852 | return 0; |
3853 | } | |
3854 | ||
ac45abc0 | 3855 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3856 | { |
8ed8d3c3 JA |
3857 | int ret; |
3858 | ||
ac45abc0 PB |
3859 | /* sync_file_range always requires a blocking context */ |
3860 | if (force_nonblock) | |
3861 | return -EAGAIN; | |
3862 | ||
9adbd45d | 3863 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3864 | req->sync.flags); |
3865 | if (ret < 0) | |
3866 | req_set_fail_links(req); | |
e1e16097 | 3867 | io_req_complete(req, ret); |
5d17b4a4 JA |
3868 | return 0; |
3869 | } | |
3870 | ||
469956e8 | 3871 | #if defined(CONFIG_NET) |
02d27d89 PB |
3872 | static int io_setup_async_msg(struct io_kiocb *req, |
3873 | struct io_async_msghdr *kmsg) | |
3874 | { | |
3875 | if (req->io) | |
3876 | return -EAGAIN; | |
3877 | if (io_alloc_async_ctx(req)) { | |
3878 | if (kmsg->iov != kmsg->fast_iov) | |
3879 | kfree(kmsg->iov); | |
3880 | return -ENOMEM; | |
3881 | } | |
3882 | req->flags |= REQ_F_NEED_CLEANUP; | |
3883 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3884 | return -EAGAIN; | |
3885 | } | |
3886 | ||
3529d8c2 | 3887 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3888 | { |
e47293fd | 3889 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3890 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3891 | int ret; |
03b1230c | 3892 | |
d2b6f48b PB |
3893 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3894 | return -EINVAL; | |
3895 | ||
e47293fd JA |
3896 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3897 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3898 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3899 | |
d8768362 JA |
3900 | #ifdef CONFIG_COMPAT |
3901 | if (req->ctx->compat) | |
3902 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3903 | #endif | |
3904 | ||
fddaface | 3905 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3906 | return 0; |
5f798bea PB |
3907 | /* iovec is already imported */ |
3908 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3909 | return 0; | |
3529d8c2 | 3910 | |
d9688565 | 3911 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3912 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3913 | &io->msg.iov); |
99bc4c38 PB |
3914 | if (!ret) |
3915 | req->flags |= REQ_F_NEED_CLEANUP; | |
3916 | return ret; | |
03b1230c JA |
3917 | } |
3918 | ||
229a7b63 JA |
3919 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
3920 | struct io_comp_state *cs) | |
aa1fa28f | 3921 | { |
0b416c3e | 3922 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3923 | struct socket *sock; |
3924 | int ret; | |
3925 | ||
0fa03c62 JA |
3926 | sock = sock_from_file(req->file, &ret); |
3927 | if (sock) { | |
b7bb4f7d | 3928 | struct io_async_ctx io; |
0fa03c62 JA |
3929 | unsigned flags; |
3930 | ||
03b1230c | 3931 | if (req->io) { |
0b416c3e | 3932 | kmsg = &req->io->msg; |
b537916c | 3933 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3934 | /* if iov is set, it's allocated already */ |
3935 | if (!kmsg->iov) | |
3936 | kmsg->iov = kmsg->fast_iov; | |
3937 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3938 | } else { |
3529d8c2 JA |
3939 | struct io_sr_msg *sr = &req->sr_msg; |
3940 | ||
0b416c3e | 3941 | kmsg = &io.msg; |
b537916c | 3942 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3943 | |
3944 | io.msg.iov = io.msg.fast_iov; | |
3945 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3946 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3947 | if (ret) |
3529d8c2 | 3948 | return ret; |
03b1230c | 3949 | } |
0fa03c62 | 3950 | |
e47293fd JA |
3951 | flags = req->sr_msg.msg_flags; |
3952 | if (flags & MSG_DONTWAIT) | |
3953 | req->flags |= REQ_F_NOWAIT; | |
3954 | else if (force_nonblock) | |
3955 | flags |= MSG_DONTWAIT; | |
3956 | ||
0b416c3e | 3957 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3958 | if (force_nonblock && ret == -EAGAIN) |
3959 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3960 | if (ret == -ERESTARTSYS) |
3961 | ret = -EINTR; | |
0fa03c62 JA |
3962 | } |
3963 | ||
1e95081c | 3964 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3965 | kfree(kmsg->iov); |
99bc4c38 | 3966 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
3967 | if (ret < 0) |
3968 | req_set_fail_links(req); | |
229a7b63 | 3969 | __io_req_complete(req, ret, 0, cs); |
5d17b4a4 | 3970 | return 0; |
03b1230c | 3971 | } |
aa1fa28f | 3972 | |
229a7b63 JA |
3973 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
3974 | struct io_comp_state *cs) | |
fddaface | 3975 | { |
fddaface JA |
3976 | struct socket *sock; |
3977 | int ret; | |
3978 | ||
fddaface JA |
3979 | sock = sock_from_file(req->file, &ret); |
3980 | if (sock) { | |
3981 | struct io_sr_msg *sr = &req->sr_msg; | |
3982 | struct msghdr msg; | |
3983 | struct iovec iov; | |
3984 | unsigned flags; | |
3985 | ||
3986 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3987 | &msg.msg_iter); | |
3988 | if (ret) | |
3989 | return ret; | |
3990 | ||
3991 | msg.msg_name = NULL; | |
3992 | msg.msg_control = NULL; | |
3993 | msg.msg_controllen = 0; | |
3994 | msg.msg_namelen = 0; | |
3995 | ||
3996 | flags = req->sr_msg.msg_flags; | |
3997 | if (flags & MSG_DONTWAIT) | |
3998 | req->flags |= REQ_F_NOWAIT; | |
3999 | else if (force_nonblock) | |
4000 | flags |= MSG_DONTWAIT; | |
4001 | ||
0b7b21e4 JA |
4002 | msg.msg_flags = flags; |
4003 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
4004 | if (force_nonblock && ret == -EAGAIN) |
4005 | return -EAGAIN; | |
4006 | if (ret == -ERESTARTSYS) | |
4007 | ret = -EINTR; | |
4008 | } | |
4009 | ||
fddaface JA |
4010 | if (ret < 0) |
4011 | req_set_fail_links(req); | |
229a7b63 | 4012 | __io_req_complete(req, ret, 0, cs); |
fddaface | 4013 | return 0; |
fddaface JA |
4014 | } |
4015 | ||
52de1fe1 JA |
4016 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
4017 | { | |
4018 | struct io_sr_msg *sr = &req->sr_msg; | |
4019 | struct iovec __user *uiov; | |
4020 | size_t iov_len; | |
4021 | int ret; | |
4022 | ||
4023 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
4024 | &uiov, &iov_len); | |
4025 | if (ret) | |
4026 | return ret; | |
4027 | ||
4028 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4029 | if (iov_len > 1) | |
4030 | return -EINVAL; | |
4031 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
4032 | return -EFAULT; | |
4033 | sr->len = io->msg.iov[0].iov_len; | |
4034 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
4035 | sr->len); | |
4036 | io->msg.iov = NULL; | |
4037 | } else { | |
4038 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
4039 | &io->msg.iov, &io->msg.msg.msg_iter); | |
4040 | if (ret > 0) | |
4041 | ret = 0; | |
4042 | } | |
4043 | ||
4044 | return ret; | |
4045 | } | |
4046 | ||
4047 | #ifdef CONFIG_COMPAT | |
4048 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
4049 | struct io_async_ctx *io) | |
4050 | { | |
4051 | struct compat_msghdr __user *msg_compat; | |
4052 | struct io_sr_msg *sr = &req->sr_msg; | |
4053 | struct compat_iovec __user *uiov; | |
4054 | compat_uptr_t ptr; | |
4055 | compat_size_t len; | |
4056 | int ret; | |
4057 | ||
4058 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
4059 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
4060 | &ptr, &len); | |
4061 | if (ret) | |
4062 | return ret; | |
4063 | ||
4064 | uiov = compat_ptr(ptr); | |
4065 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4066 | compat_ssize_t clen; | |
4067 | ||
4068 | if (len > 1) | |
4069 | return -EINVAL; | |
4070 | if (!access_ok(uiov, sizeof(*uiov))) | |
4071 | return -EFAULT; | |
4072 | if (__get_user(clen, &uiov->iov_len)) | |
4073 | return -EFAULT; | |
4074 | if (clen < 0) | |
4075 | return -EINVAL; | |
4076 | sr->len = io->msg.iov[0].iov_len; | |
4077 | io->msg.iov = NULL; | |
4078 | } else { | |
4079 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
4080 | &io->msg.iov, | |
4081 | &io->msg.msg.msg_iter); | |
4082 | if (ret < 0) | |
4083 | return ret; | |
4084 | } | |
4085 | ||
4086 | return 0; | |
4087 | } | |
4088 | #endif | |
4089 | ||
4090 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
4091 | { | |
4092 | io->msg.iov = io->msg.fast_iov; | |
4093 | ||
4094 | #ifdef CONFIG_COMPAT | |
4095 | if (req->ctx->compat) | |
4096 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 4097 | #endif |
52de1fe1 JA |
4098 | |
4099 | return __io_recvmsg_copy_hdr(req, io); | |
4100 | } | |
4101 | ||
bcda7baa JA |
4102 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
4103 | int *cflags, bool needs_lock) | |
4104 | { | |
4105 | struct io_sr_msg *sr = &req->sr_msg; | |
4106 | struct io_buffer *kbuf; | |
4107 | ||
4108 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
4109 | return NULL; | |
4110 | ||
4111 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
4112 | if (IS_ERR(kbuf)) | |
4113 | return kbuf; | |
4114 | ||
4115 | sr->kbuf = kbuf; | |
4116 | req->flags |= REQ_F_BUFFER_SELECTED; | |
4117 | ||
4118 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
4119 | *cflags |= IORING_CQE_F_BUFFER; | |
4120 | return kbuf; | |
fddaface JA |
4121 | } |
4122 | ||
3529d8c2 JA |
4123 | static int io_recvmsg_prep(struct io_kiocb *req, |
4124 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4125 | { |
e47293fd | 4126 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 4127 | struct io_async_ctx *io = req->io; |
99bc4c38 | 4128 | int ret; |
3529d8c2 | 4129 | |
d2b6f48b PB |
4130 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4131 | return -EINVAL; | |
4132 | ||
3529d8c2 JA |
4133 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
4134 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 4135 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4136 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4137 | |
d8768362 JA |
4138 | #ifdef CONFIG_COMPAT |
4139 | if (req->ctx->compat) | |
4140 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4141 | #endif | |
4142 | ||
fddaface | 4143 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 4144 | return 0; |
5f798bea PB |
4145 | /* iovec is already imported */ |
4146 | if (req->flags & REQ_F_NEED_CLEANUP) | |
4147 | return 0; | |
03b1230c | 4148 | |
52de1fe1 | 4149 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
4150 | if (!ret) |
4151 | req->flags |= REQ_F_NEED_CLEANUP; | |
4152 | return ret; | |
aa1fa28f JA |
4153 | } |
4154 | ||
229a7b63 JA |
4155 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4156 | struct io_comp_state *cs) | |
aa1fa28f | 4157 | { |
0b416c3e | 4158 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 4159 | struct socket *sock; |
52de1fe1 | 4160 | int ret, cflags = 0; |
03b1230c | 4161 | |
03b1230c JA |
4162 | sock = sock_from_file(req->file, &ret); |
4163 | if (sock) { | |
52de1fe1 | 4164 | struct io_buffer *kbuf; |
b7bb4f7d | 4165 | struct io_async_ctx io; |
03b1230c JA |
4166 | unsigned flags; |
4167 | ||
03b1230c | 4168 | if (req->io) { |
0b416c3e | 4169 | kmsg = &req->io->msg; |
b537916c | 4170 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
4171 | /* if iov is set, it's allocated already */ |
4172 | if (!kmsg->iov) | |
4173 | kmsg->iov = kmsg->fast_iov; | |
4174 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 4175 | } else { |
0b416c3e | 4176 | kmsg = &io.msg; |
b537916c | 4177 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 4178 | |
52de1fe1 | 4179 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 4180 | if (ret) |
3529d8c2 | 4181 | return ret; |
03b1230c JA |
4182 | } |
4183 | ||
52de1fe1 JA |
4184 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
4185 | if (IS_ERR(kbuf)) { | |
4186 | return PTR_ERR(kbuf); | |
4187 | } else if (kbuf) { | |
4188 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
4189 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
4190 | 1, req->sr_msg.len); | |
4191 | } | |
4192 | ||
e47293fd JA |
4193 | flags = req->sr_msg.msg_flags; |
4194 | if (flags & MSG_DONTWAIT) | |
4195 | req->flags |= REQ_F_NOWAIT; | |
4196 | else if (force_nonblock) | |
4197 | flags |= MSG_DONTWAIT; | |
4198 | ||
4199 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
4200 | kmsg->uaddr, flags); | |
02d27d89 PB |
4201 | if (force_nonblock && ret == -EAGAIN) |
4202 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
4203 | if (ret == -ERESTARTSYS) |
4204 | ret = -EINTR; | |
4205 | } | |
4206 | ||
1e95081c | 4207 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 4208 | kfree(kmsg->iov); |
99bc4c38 | 4209 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4210 | if (ret < 0) |
4211 | req_set_fail_links(req); | |
229a7b63 | 4212 | __io_req_complete(req, ret, cflags, cs); |
03b1230c | 4213 | return 0; |
0fa03c62 | 4214 | } |
5d17b4a4 | 4215 | |
229a7b63 JA |
4216 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4217 | struct io_comp_state *cs) | |
fddaface | 4218 | { |
bcda7baa | 4219 | struct io_buffer *kbuf = NULL; |
fddaface | 4220 | struct socket *sock; |
bcda7baa | 4221 | int ret, cflags = 0; |
fddaface | 4222 | |
fddaface JA |
4223 | sock = sock_from_file(req->file, &ret); |
4224 | if (sock) { | |
4225 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 4226 | void __user *buf = sr->buf; |
fddaface JA |
4227 | struct msghdr msg; |
4228 | struct iovec iov; | |
4229 | unsigned flags; | |
4230 | ||
bcda7baa JA |
4231 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
4232 | if (IS_ERR(kbuf)) | |
4233 | return PTR_ERR(kbuf); | |
4234 | else if (kbuf) | |
4235 | buf = u64_to_user_ptr(kbuf->addr); | |
4236 | ||
4237 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 4238 | &msg.msg_iter); |
bcda7baa JA |
4239 | if (ret) { |
4240 | kfree(kbuf); | |
fddaface | 4241 | return ret; |
bcda7baa | 4242 | } |
fddaface | 4243 | |
bcda7baa | 4244 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
4245 | msg.msg_name = NULL; |
4246 | msg.msg_control = NULL; | |
4247 | msg.msg_controllen = 0; | |
4248 | msg.msg_namelen = 0; | |
4249 | msg.msg_iocb = NULL; | |
4250 | msg.msg_flags = 0; | |
4251 | ||
4252 | flags = req->sr_msg.msg_flags; | |
4253 | if (flags & MSG_DONTWAIT) | |
4254 | req->flags |= REQ_F_NOWAIT; | |
4255 | else if (force_nonblock) | |
4256 | flags |= MSG_DONTWAIT; | |
4257 | ||
0b7b21e4 | 4258 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
4259 | if (force_nonblock && ret == -EAGAIN) |
4260 | return -EAGAIN; | |
4261 | if (ret == -ERESTARTSYS) | |
4262 | ret = -EINTR; | |
4263 | } | |
4264 | ||
bcda7baa JA |
4265 | kfree(kbuf); |
4266 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
fddaface JA |
4267 | if (ret < 0) |
4268 | req_set_fail_links(req); | |
229a7b63 | 4269 | __io_req_complete(req, ret, cflags, cs); |
fddaface | 4270 | return 0; |
fddaface JA |
4271 | } |
4272 | ||
3529d8c2 | 4273 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4274 | { |
8ed8d3c3 JA |
4275 | struct io_accept *accept = &req->accept; |
4276 | ||
17f2fe35 JA |
4277 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4278 | return -EINVAL; | |
8042d6ce | 4279 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4280 | return -EINVAL; |
4281 | ||
d55e5f5b JA |
4282 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4283 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4284 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4285 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4286 | return 0; |
8ed8d3c3 | 4287 | } |
17f2fe35 | 4288 | |
229a7b63 JA |
4289 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4290 | struct io_comp_state *cs) | |
8ed8d3c3 JA |
4291 | { |
4292 | struct io_accept *accept = &req->accept; | |
ac45abc0 | 4293 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4294 | int ret; |
4295 | ||
e697deed JX |
4296 | if (req->file->f_flags & O_NONBLOCK) |
4297 | req->flags |= REQ_F_NOWAIT; | |
4298 | ||
8ed8d3c3 | 4299 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4300 | accept->addr_len, accept->flags, |
4301 | accept->nofile); | |
8ed8d3c3 | 4302 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4303 | return -EAGAIN; |
ac45abc0 PB |
4304 | if (ret < 0) { |
4305 | if (ret == -ERESTARTSYS) | |
4306 | ret = -EINTR; | |
4e88d6e7 | 4307 | req_set_fail_links(req); |
ac45abc0 | 4308 | } |
229a7b63 | 4309 | __io_req_complete(req, ret, 0, cs); |
17f2fe35 | 4310 | return 0; |
8ed8d3c3 JA |
4311 | } |
4312 | ||
3529d8c2 | 4313 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4314 | { |
3529d8c2 JA |
4315 | struct io_connect *conn = &req->connect; |
4316 | struct io_async_ctx *io = req->io; | |
f499a021 | 4317 | |
3fbb51c1 JA |
4318 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4319 | return -EINVAL; | |
4320 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4321 | return -EINVAL; | |
4322 | ||
3529d8c2 JA |
4323 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4324 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4325 | ||
4326 | if (!io) | |
4327 | return 0; | |
4328 | ||
4329 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4330 | &io->connect.address); |
f499a021 JA |
4331 | } |
4332 | ||
229a7b63 JA |
4333 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4334 | struct io_comp_state *cs) | |
f8e85cf2 | 4335 | { |
f499a021 | 4336 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4337 | unsigned file_flags; |
3fbb51c1 | 4338 | int ret; |
f8e85cf2 | 4339 | |
f499a021 JA |
4340 | if (req->io) { |
4341 | io = req->io; | |
4342 | } else { | |
3529d8c2 JA |
4343 | ret = move_addr_to_kernel(req->connect.addr, |
4344 | req->connect.addr_len, | |
4345 | &__io.connect.address); | |
f499a021 JA |
4346 | if (ret) |
4347 | goto out; | |
4348 | io = &__io; | |
4349 | } | |
4350 | ||
3fbb51c1 JA |
4351 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4352 | ||
4353 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4354 | req->connect.addr_len, file_flags); | |
87f80d62 | 4355 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4356 | if (req->io) |
4357 | return -EAGAIN; | |
4358 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4359 | ret = -ENOMEM; |
4360 | goto out; | |
4361 | } | |
b7bb4f7d | 4362 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4363 | return -EAGAIN; |
f499a021 | 4364 | } |
f8e85cf2 JA |
4365 | if (ret == -ERESTARTSYS) |
4366 | ret = -EINTR; | |
f499a021 | 4367 | out: |
4e88d6e7 JA |
4368 | if (ret < 0) |
4369 | req_set_fail_links(req); | |
229a7b63 | 4370 | __io_req_complete(req, ret, 0, cs); |
f8e85cf2 | 4371 | return 0; |
469956e8 Y |
4372 | } |
4373 | #else /* !CONFIG_NET */ | |
4374 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4375 | { | |
f8e85cf2 | 4376 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4377 | } |
4378 | ||
1e16c2f9 RD |
4379 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock, |
4380 | struct io_comp_state *cs) | |
469956e8 Y |
4381 | { |
4382 | return -EOPNOTSUPP; | |
4383 | } | |
4384 | ||
1e16c2f9 RD |
4385 | static int io_send(struct io_kiocb *req, bool force_nonblock, |
4386 | struct io_comp_state *cs) | |
469956e8 Y |
4387 | { |
4388 | return -EOPNOTSUPP; | |
4389 | } | |
4390 | ||
4391 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4392 | const struct io_uring_sqe *sqe) | |
4393 | { | |
4394 | return -EOPNOTSUPP; | |
4395 | } | |
4396 | ||
1e16c2f9 RD |
4397 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock, |
4398 | struct io_comp_state *cs) | |
469956e8 Y |
4399 | { |
4400 | return -EOPNOTSUPP; | |
4401 | } | |
4402 | ||
1e16c2f9 RD |
4403 | static int io_recv(struct io_kiocb *req, bool force_nonblock, |
4404 | struct io_comp_state *cs) | |
469956e8 Y |
4405 | { |
4406 | return -EOPNOTSUPP; | |
4407 | } | |
4408 | ||
4409 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4410 | { | |
4411 | return -EOPNOTSUPP; | |
4412 | } | |
4413 | ||
1e16c2f9 RD |
4414 | static int io_accept(struct io_kiocb *req, bool force_nonblock, |
4415 | struct io_comp_state *cs) | |
469956e8 Y |
4416 | { |
4417 | return -EOPNOTSUPP; | |
4418 | } | |
4419 | ||
4420 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4421 | { | |
4422 | return -EOPNOTSUPP; | |
4423 | } | |
4424 | ||
1e16c2f9 RD |
4425 | static int io_connect(struct io_kiocb *req, bool force_nonblock, |
4426 | struct io_comp_state *cs) | |
469956e8 | 4427 | { |
f8e85cf2 | 4428 | return -EOPNOTSUPP; |
f8e85cf2 | 4429 | } |
469956e8 | 4430 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4431 | |
d7718a9d JA |
4432 | struct io_poll_table { |
4433 | struct poll_table_struct pt; | |
4434 | struct io_kiocb *req; | |
4435 | int error; | |
4436 | }; | |
4437 | ||
d7718a9d JA |
4438 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4439 | __poll_t mask, task_work_func_t func) | |
4440 | { | |
aa96bf8a | 4441 | int ret; |
d7718a9d JA |
4442 | |
4443 | /* for instances that support it check for an event match first: */ | |
4444 | if (mask && !(mask & poll->events)) | |
4445 | return 0; | |
4446 | ||
4447 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4448 | ||
4449 | list_del_init(&poll->wait.entry); | |
4450 | ||
d7718a9d JA |
4451 | req->result = mask; |
4452 | init_task_work(&req->task_work, func); | |
4453 | /* | |
e3aabf95 JA |
4454 | * If this fails, then the task is exiting. When a task exits, the |
4455 | * work gets canceled, so just cancel this request as well instead | |
4456 | * of executing it. We can't safely execute it anyway, as we may not | |
4457 | * have the needed state needed for it anyway. | |
d7718a9d | 4458 | */ |
b7db41c9 | 4459 | ret = io_req_task_work_add(req, &req->task_work); |
aa96bf8a | 4460 | if (unlikely(ret)) { |
c2c4c83c JA |
4461 | struct task_struct *tsk; |
4462 | ||
e3aabf95 | 4463 | WRITE_ONCE(poll->canceled, true); |
aa96bf8a | 4464 | tsk = io_wq_get_task(req->ctx->io_wq); |
ce593a6c JA |
4465 | task_work_add(tsk, &req->task_work, 0); |
4466 | wake_up_process(tsk); | |
aa96bf8a | 4467 | } |
d7718a9d JA |
4468 | return 1; |
4469 | } | |
4470 | ||
74ce6ce4 JA |
4471 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4472 | __acquires(&req->ctx->completion_lock) | |
4473 | { | |
4474 | struct io_ring_ctx *ctx = req->ctx; | |
4475 | ||
4476 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4477 | struct poll_table_struct pt = { ._key = poll->events }; | |
4478 | ||
4479 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4480 | } | |
4481 | ||
4482 | spin_lock_irq(&ctx->completion_lock); | |
4483 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4484 | add_wait_queue(poll->head, &poll->wait); | |
4485 | return true; | |
4486 | } | |
4487 | ||
4488 | return false; | |
4489 | } | |
4490 | ||
18bceab1 JA |
4491 | static void io_poll_remove_double(struct io_kiocb *req) |
4492 | { | |
4493 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4494 | ||
4495 | lockdep_assert_held(&req->ctx->completion_lock); | |
4496 | ||
4497 | if (poll && poll->head) { | |
4498 | struct wait_queue_head *head = poll->head; | |
4499 | ||
4500 | spin_lock(&head->lock); | |
4501 | list_del_init(&poll->wait.entry); | |
4502 | if (poll->wait.private) | |
4503 | refcount_dec(&req->refs); | |
4504 | poll->head = NULL; | |
4505 | spin_unlock(&head->lock); | |
4506 | } | |
4507 | } | |
4508 | ||
4509 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4510 | { | |
4511 | struct io_ring_ctx *ctx = req->ctx; | |
4512 | ||
4513 | io_poll_remove_double(req); | |
4514 | req->poll.done = true; | |
4515 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4516 | io_commit_cqring(ctx); | |
4517 | } | |
4518 | ||
4519 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) | |
4520 | { | |
4521 | struct io_ring_ctx *ctx = req->ctx; | |
4522 | ||
4523 | if (io_poll_rewait(req, &req->poll)) { | |
4524 | spin_unlock_irq(&ctx->completion_lock); | |
4525 | return; | |
4526 | } | |
4527 | ||
4528 | hash_del(&req->hash_node); | |
4529 | io_poll_complete(req, req->result, 0); | |
4530 | req->flags |= REQ_F_COMP_LOCKED; | |
9b5f7bd9 | 4531 | *nxt = io_put_req_find_next(req); |
18bceab1 JA |
4532 | spin_unlock_irq(&ctx->completion_lock); |
4533 | ||
4534 | io_cqring_ev_posted(ctx); | |
4535 | } | |
4536 | ||
4537 | static void io_poll_task_func(struct callback_head *cb) | |
4538 | { | |
4539 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4540 | struct io_kiocb *nxt = NULL; | |
4541 | ||
4542 | io_poll_task_handler(req, &nxt); | |
ea1164e5 PB |
4543 | if (nxt) |
4544 | __io_req_task_submit(nxt); | |
18bceab1 JA |
4545 | } |
4546 | ||
4547 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4548 | int sync, void *key) | |
4549 | { | |
4550 | struct io_kiocb *req = wait->private; | |
4551 | struct io_poll_iocb *poll = (struct io_poll_iocb *) req->io; | |
4552 | __poll_t mask = key_to_poll(key); | |
4553 | ||
4554 | /* for instances that support it check for an event match first: */ | |
4555 | if (mask && !(mask & poll->events)) | |
4556 | return 0; | |
4557 | ||
4558 | if (req->poll.head) { | |
4559 | bool done; | |
4560 | ||
4561 | spin_lock(&req->poll.head->lock); | |
4562 | done = list_empty(&req->poll.wait.entry); | |
4563 | if (!done) | |
4564 | list_del_init(&req->poll.wait.entry); | |
4565 | spin_unlock(&req->poll.head->lock); | |
4566 | if (!done) | |
4567 | __io_async_wake(req, poll, mask, io_poll_task_func); | |
4568 | } | |
4569 | refcount_dec(&req->refs); | |
4570 | return 1; | |
4571 | } | |
4572 | ||
4573 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4574 | wait_queue_func_t wake_func) | |
4575 | { | |
4576 | poll->head = NULL; | |
4577 | poll->done = false; | |
4578 | poll->canceled = false; | |
4579 | poll->events = events; | |
4580 | INIT_LIST_HEAD(&poll->wait.entry); | |
4581 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4582 | } | |
4583 | ||
4584 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4585 | struct wait_queue_head *head) | |
4586 | { | |
4587 | struct io_kiocb *req = pt->req; | |
4588 | ||
4589 | /* | |
4590 | * If poll->head is already set, it's because the file being polled | |
4591 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4592 | * for write). Setup a separate io_poll_iocb if this happens. | |
4593 | */ | |
4594 | if (unlikely(poll->head)) { | |
4595 | /* already have a 2nd entry, fail a third attempt */ | |
4596 | if (req->io) { | |
4597 | pt->error = -EINVAL; | |
4598 | return; | |
4599 | } | |
4600 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4601 | if (!poll) { | |
4602 | pt->error = -ENOMEM; | |
4603 | return; | |
4604 | } | |
4605 | io_init_poll_iocb(poll, req->poll.events, io_poll_double_wake); | |
4606 | refcount_inc(&req->refs); | |
4607 | poll->wait.private = req; | |
4608 | req->io = (void *) poll; | |
4609 | } | |
4610 | ||
4611 | pt->error = 0; | |
4612 | poll->head = head; | |
a31eb4a2 JX |
4613 | |
4614 | if (poll->events & EPOLLEXCLUSIVE) | |
4615 | add_wait_queue_exclusive(head, &poll->wait); | |
4616 | else | |
4617 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4618 | } |
4619 | ||
4620 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4621 | struct poll_table_struct *p) | |
4622 | { | |
4623 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4624 | ||
4625 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4626 | } | |
4627 | ||
d7718a9d JA |
4628 | static void io_async_task_func(struct callback_head *cb) |
4629 | { | |
4630 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4631 | struct async_poll *apoll = req->apoll; | |
4632 | struct io_ring_ctx *ctx = req->ctx; | |
4633 | ||
4634 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4635 | ||
74ce6ce4 | 4636 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4637 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4638 | return; |
d7718a9d JA |
4639 | } |
4640 | ||
31067255 | 4641 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 4642 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 4643 | hash_del(&req->hash_node); |
2bae047e | 4644 | |
74ce6ce4 JA |
4645 | spin_unlock_irq(&ctx->completion_lock); |
4646 | ||
44575a67 | 4647 | /* restore ->work in case we need to retry again */ |
405a5d2b XW |
4648 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4649 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
31067255 | 4650 | kfree(apoll); |
44575a67 | 4651 | |
0be0b0e3 PB |
4652 | if (!READ_ONCE(apoll->poll.canceled)) |
4653 | __io_req_task_submit(req); | |
4654 | else | |
4655 | __io_req_task_cancel(req, -ECANCELED); | |
d7718a9d JA |
4656 | } |
4657 | ||
4658 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4659 | void *key) | |
4660 | { | |
4661 | struct io_kiocb *req = wait->private; | |
4662 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4663 | ||
4664 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4665 | key_to_poll(key)); | |
4666 | ||
4667 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4668 | } | |
4669 | ||
4670 | static void io_poll_req_insert(struct io_kiocb *req) | |
4671 | { | |
4672 | struct io_ring_ctx *ctx = req->ctx; | |
4673 | struct hlist_head *list; | |
4674 | ||
4675 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4676 | hlist_add_head(&req->hash_node, list); | |
4677 | } | |
4678 | ||
4679 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4680 | struct io_poll_iocb *poll, | |
4681 | struct io_poll_table *ipt, __poll_t mask, | |
4682 | wait_queue_func_t wake_func) | |
4683 | __acquires(&ctx->completion_lock) | |
4684 | { | |
4685 | struct io_ring_ctx *ctx = req->ctx; | |
4686 | bool cancel = false; | |
4687 | ||
18bceab1 | 4688 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 4689 | poll->file = req->file; |
18bceab1 | 4690 | poll->wait.private = req; |
d7718a9d JA |
4691 | |
4692 | ipt->pt._key = mask; | |
4693 | ipt->req = req; | |
4694 | ipt->error = -EINVAL; | |
4695 | ||
d7718a9d JA |
4696 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
4697 | ||
4698 | spin_lock_irq(&ctx->completion_lock); | |
4699 | if (likely(poll->head)) { | |
4700 | spin_lock(&poll->head->lock); | |
4701 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4702 | if (ipt->error) | |
4703 | cancel = true; | |
4704 | ipt->error = 0; | |
4705 | mask = 0; | |
4706 | } | |
4707 | if (mask || ipt->error) | |
4708 | list_del_init(&poll->wait.entry); | |
4709 | else if (cancel) | |
4710 | WRITE_ONCE(poll->canceled, true); | |
4711 | else if (!poll->done) /* actually waiting for an event */ | |
4712 | io_poll_req_insert(req); | |
4713 | spin_unlock(&poll->head->lock); | |
4714 | } | |
4715 | ||
4716 | return mask; | |
4717 | } | |
4718 | ||
4719 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4720 | { | |
4721 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4722 | struct io_ring_ctx *ctx = req->ctx; | |
4723 | struct async_poll *apoll; | |
4724 | struct io_poll_table ipt; | |
4725 | __poll_t mask, ret; | |
18bceab1 | 4726 | bool had_io; |
d7718a9d JA |
4727 | |
4728 | if (!req->file || !file_can_poll(req->file)) | |
4729 | return false; | |
24c74678 | 4730 | if (req->flags & REQ_F_POLLED) |
d7718a9d JA |
4731 | return false; |
4732 | if (!def->pollin && !def->pollout) | |
4733 | return false; | |
4734 | ||
4735 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4736 | if (unlikely(!apoll)) | |
4737 | return false; | |
4738 | ||
4739 | req->flags |= REQ_F_POLLED; | |
405a5d2b XW |
4740 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4741 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
18bceab1 | 4742 | had_io = req->io != NULL; |
d7718a9d | 4743 | |
4dd2824d | 4744 | io_get_req_task(req); |
d7718a9d JA |
4745 | req->apoll = apoll; |
4746 | INIT_HLIST_NODE(&req->hash_node); | |
4747 | ||
8755d97a | 4748 | mask = 0; |
d7718a9d | 4749 | if (def->pollin) |
8755d97a | 4750 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4751 | if (def->pollout) |
4752 | mask |= POLLOUT | POLLWRNORM; | |
4753 | mask |= POLLERR | POLLPRI; | |
4754 | ||
4755 | ipt.pt._qproc = io_async_queue_proc; | |
4756 | ||
4757 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4758 | io_async_wake); | |
4759 | if (ret) { | |
4760 | ipt.error = 0; | |
18bceab1 JA |
4761 | /* only remove double add if we did it here */ |
4762 | if (!had_io) | |
4763 | io_poll_remove_double(req); | |
d7718a9d | 4764 | spin_unlock_irq(&ctx->completion_lock); |
405a5d2b XW |
4765 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4766 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
d7718a9d JA |
4767 | kfree(apoll); |
4768 | return false; | |
4769 | } | |
4770 | spin_unlock_irq(&ctx->completion_lock); | |
4771 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4772 | apoll->poll.events); | |
4773 | return true; | |
4774 | } | |
4775 | ||
4776 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4777 | struct io_poll_iocb *poll) | |
221c5eb2 | 4778 | { |
b41e9852 | 4779 | bool do_complete = false; |
221c5eb2 JA |
4780 | |
4781 | spin_lock(&poll->head->lock); | |
4782 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4783 | if (!list_empty(&poll->wait.entry)) { |
4784 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4785 | do_complete = true; |
221c5eb2 JA |
4786 | } |
4787 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 4788 | hash_del(&req->hash_node); |
d7718a9d JA |
4789 | return do_complete; |
4790 | } | |
4791 | ||
4792 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4793 | { | |
4794 | bool do_complete; | |
4795 | ||
4796 | if (req->opcode == IORING_OP_POLL_ADD) { | |
18bceab1 | 4797 | io_poll_remove_double(req); |
d7718a9d JA |
4798 | do_complete = __io_poll_remove_one(req, &req->poll); |
4799 | } else { | |
3bfa5bcb JA |
4800 | struct async_poll *apoll = req->apoll; |
4801 | ||
d7718a9d | 4802 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
4803 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
4804 | if (do_complete) { | |
d7718a9d | 4805 | io_put_req(req); |
3bfa5bcb JA |
4806 | /* |
4807 | * restore ->work because we will call | |
4808 | * io_req_work_drop_env below when dropping the | |
4809 | * final reference. | |
4810 | */ | |
405a5d2b XW |
4811 | if (req->flags & REQ_F_WORK_INITIALIZED) |
4812 | memcpy(&req->work, &apoll->work, | |
4813 | sizeof(req->work)); | |
3bfa5bcb JA |
4814 | kfree(apoll); |
4815 | } | |
b1f573bd XW |
4816 | } |
4817 | ||
b41e9852 JA |
4818 | if (do_complete) { |
4819 | io_cqring_fill_event(req, -ECANCELED); | |
4820 | io_commit_cqring(req->ctx); | |
4821 | req->flags |= REQ_F_COMP_LOCKED; | |
4822 | io_put_req(req); | |
4823 | } | |
4824 | ||
4825 | return do_complete; | |
221c5eb2 JA |
4826 | } |
4827 | ||
4828 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4829 | { | |
78076bb6 | 4830 | struct hlist_node *tmp; |
221c5eb2 | 4831 | struct io_kiocb *req; |
8e2e1faf | 4832 | int posted = 0, i; |
221c5eb2 JA |
4833 | |
4834 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4835 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4836 | struct hlist_head *list; | |
4837 | ||
4838 | list = &ctx->cancel_hash[i]; | |
4839 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4840 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4841 | } |
4842 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4843 | |
8e2e1faf JA |
4844 | if (posted) |
4845 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4846 | } |
4847 | ||
47f46768 JA |
4848 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4849 | { | |
78076bb6 | 4850 | struct hlist_head *list; |
47f46768 JA |
4851 | struct io_kiocb *req; |
4852 | ||
78076bb6 JA |
4853 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4854 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4855 | if (sqe_addr != req->user_data) |
4856 | continue; | |
4857 | if (io_poll_remove_one(req)) | |
eac406c6 | 4858 | return 0; |
b41e9852 | 4859 | return -EALREADY; |
47f46768 JA |
4860 | } |
4861 | ||
4862 | return -ENOENT; | |
4863 | } | |
4864 | ||
3529d8c2 JA |
4865 | static int io_poll_remove_prep(struct io_kiocb *req, |
4866 | const struct io_uring_sqe *sqe) | |
0969e783 | 4867 | { |
0969e783 JA |
4868 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4869 | return -EINVAL; | |
4870 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4871 | sqe->poll_events) | |
4872 | return -EINVAL; | |
4873 | ||
4874 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4875 | return 0; |
4876 | } | |
4877 | ||
221c5eb2 JA |
4878 | /* |
4879 | * Find a running poll command that matches one specified in sqe->addr, | |
4880 | * and remove it if found. | |
4881 | */ | |
fc4df999 | 4882 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4883 | { |
4884 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4885 | u64 addr; |
47f46768 | 4886 | int ret; |
221c5eb2 | 4887 | |
0969e783 | 4888 | addr = req->poll.addr; |
221c5eb2 | 4889 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4890 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4891 | spin_unlock_irq(&ctx->completion_lock); |
4892 | ||
4e88d6e7 JA |
4893 | if (ret < 0) |
4894 | req_set_fail_links(req); | |
e1e16097 | 4895 | io_req_complete(req, ret); |
221c5eb2 JA |
4896 | return 0; |
4897 | } | |
4898 | ||
221c5eb2 JA |
4899 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4900 | void *key) | |
4901 | { | |
c2f2eb7d JA |
4902 | struct io_kiocb *req = wait->private; |
4903 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4904 | |
d7718a9d | 4905 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4906 | } |
4907 | ||
221c5eb2 JA |
4908 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4909 | struct poll_table_struct *p) | |
4910 | { | |
4911 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4912 | ||
d7718a9d | 4913 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4914 | } |
4915 | ||
3529d8c2 | 4916 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4917 | { |
4918 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 4919 | u32 events; |
221c5eb2 JA |
4920 | |
4921 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4922 | return -EINVAL; | |
4923 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4924 | return -EINVAL; | |
09bb8394 JA |
4925 | if (!poll->file) |
4926 | return -EBADF; | |
221c5eb2 | 4927 | |
5769a351 JX |
4928 | events = READ_ONCE(sqe->poll32_events); |
4929 | #ifdef __BIG_ENDIAN | |
4930 | events = swahw32(events); | |
4931 | #endif | |
a31eb4a2 JX |
4932 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
4933 | (events & EPOLLEXCLUSIVE); | |
b41e9852 | 4934 | |
4dd2824d | 4935 | io_get_req_task(req); |
0969e783 JA |
4936 | return 0; |
4937 | } | |
4938 | ||
014db007 | 4939 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4940 | { |
4941 | struct io_poll_iocb *poll = &req->poll; | |
4942 | struct io_ring_ctx *ctx = req->ctx; | |
4943 | struct io_poll_table ipt; | |
0969e783 | 4944 | __poll_t mask; |
0969e783 | 4945 | |
78076bb6 | 4946 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4947 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4948 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4949 | |
d7718a9d JA |
4950 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4951 | io_poll_wake); | |
221c5eb2 | 4952 | |
8c838788 | 4953 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4954 | ipt.error = 0; |
b0dd8a41 | 4955 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4956 | } |
221c5eb2 JA |
4957 | spin_unlock_irq(&ctx->completion_lock); |
4958 | ||
8c838788 JA |
4959 | if (mask) { |
4960 | io_cqring_ev_posted(ctx); | |
014db007 | 4961 | io_put_req(req); |
221c5eb2 | 4962 | } |
8c838788 | 4963 | return ipt.error; |
221c5eb2 JA |
4964 | } |
4965 | ||
5262f567 JA |
4966 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4967 | { | |
ad8a48ac JA |
4968 | struct io_timeout_data *data = container_of(timer, |
4969 | struct io_timeout_data, timer); | |
4970 | struct io_kiocb *req = data->req; | |
4971 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4972 | unsigned long flags; |
4973 | ||
5262f567 JA |
4974 | atomic_inc(&ctx->cq_timeouts); |
4975 | ||
4976 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4977 | /* |
11365043 JA |
4978 | * We could be racing with timeout deletion. If the list is empty, |
4979 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4980 | */ |
bfe68a22 | 4981 | if (!list_empty(&req->list)) |
11365043 | 4982 | list_del_init(&req->list); |
5262f567 | 4983 | |
78e19bbe | 4984 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4985 | io_commit_cqring(ctx); |
4986 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4987 | ||
4988 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4989 | req_set_fail_links(req); |
5262f567 JA |
4990 | io_put_req(req); |
4991 | return HRTIMER_NORESTART; | |
4992 | } | |
4993 | ||
47f46768 JA |
4994 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4995 | { | |
4996 | struct io_kiocb *req; | |
4997 | int ret = -ENOENT; | |
4998 | ||
4999 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
5000 | if (user_data == req->user_data) { | |
5001 | list_del_init(&req->list); | |
5002 | ret = 0; | |
5003 | break; | |
5004 | } | |
5005 | } | |
5006 | ||
5007 | if (ret == -ENOENT) | |
5008 | return ret; | |
5009 | ||
2d28390a | 5010 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
5011 | if (ret == -1) |
5012 | return -EALREADY; | |
5013 | ||
4e88d6e7 | 5014 | req_set_fail_links(req); |
47f46768 JA |
5015 | io_cqring_fill_event(req, -ECANCELED); |
5016 | io_put_req(req); | |
5017 | return 0; | |
5018 | } | |
5019 | ||
3529d8c2 JA |
5020 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5021 | const struct io_uring_sqe *sqe) | |
b29472ee | 5022 | { |
b29472ee JA |
5023 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5024 | return -EINVAL; | |
5025 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
5026 | return -EINVAL; | |
5027 | ||
5028 | req->timeout.addr = READ_ONCE(sqe->addr); | |
5029 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
5030 | if (req->timeout.flags) | |
5031 | return -EINVAL; | |
5032 | ||
b29472ee JA |
5033 | return 0; |
5034 | } | |
5035 | ||
11365043 JA |
5036 | /* |
5037 | * Remove or update an existing timeout command | |
5038 | */ | |
fc4df999 | 5039 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
5040 | { |
5041 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5042 | int ret; |
11365043 | 5043 | |
11365043 | 5044 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 5045 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 5046 | |
47f46768 | 5047 | io_cqring_fill_event(req, ret); |
11365043 JA |
5048 | io_commit_cqring(ctx); |
5049 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5050 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5051 | if (ret < 0) |
5052 | req_set_fail_links(req); | |
ec9c02ad | 5053 | io_put_req(req); |
11365043 | 5054 | return 0; |
5262f567 JA |
5055 | } |
5056 | ||
3529d8c2 | 5057 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5058 | bool is_timeout_link) |
5262f567 | 5059 | { |
ad8a48ac | 5060 | struct io_timeout_data *data; |
a41525ab | 5061 | unsigned flags; |
56080b02 | 5062 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5063 | |
ad8a48ac | 5064 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5065 | return -EINVAL; |
ad8a48ac | 5066 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5067 | return -EINVAL; |
56080b02 | 5068 | if (off && is_timeout_link) |
2d28390a | 5069 | return -EINVAL; |
a41525ab JA |
5070 | flags = READ_ONCE(sqe->timeout_flags); |
5071 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5072 | return -EINVAL; |
bdf20073 | 5073 | |
bfe68a22 | 5074 | req->timeout.off = off; |
26a61679 | 5075 | |
3529d8c2 | 5076 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
5077 | return -ENOMEM; |
5078 | ||
5079 | data = &req->io->timeout; | |
ad8a48ac | 5080 | data->req = req; |
ad8a48ac JA |
5081 | |
5082 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5083 | return -EFAULT; |
5084 | ||
11365043 | 5085 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 5086 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 5087 | else |
ad8a48ac | 5088 | data->mode = HRTIMER_MODE_REL; |
11365043 | 5089 | |
ad8a48ac JA |
5090 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5091 | return 0; | |
5092 | } | |
5093 | ||
fc4df999 | 5094 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 5095 | { |
ad8a48ac | 5096 | struct io_ring_ctx *ctx = req->ctx; |
bfe68a22 | 5097 | struct io_timeout_data *data = &req->io->timeout; |
ad8a48ac | 5098 | struct list_head *entry; |
bfe68a22 | 5099 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5100 | |
733f5c95 | 5101 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5102 | |
5262f567 JA |
5103 | /* |
5104 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5105 | * timeout event to be satisfied. If it isn't set, then this is |
5106 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5107 | */ |
8eb7e2d0 | 5108 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5109 | entry = ctx->timeout_list.prev; |
5110 | goto add; | |
5111 | } | |
5262f567 | 5112 | |
bfe68a22 PB |
5113 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5114 | req->timeout.target_seq = tail + off; | |
5262f567 JA |
5115 | |
5116 | /* | |
5117 | * Insertion sort, ensuring the first entry in the list is always | |
5118 | * the one we need first. | |
5119 | */ | |
5262f567 JA |
5120 | list_for_each_prev(entry, &ctx->timeout_list) { |
5121 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5262f567 | 5122 | |
8eb7e2d0 | 5123 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5124 | continue; |
bfe68a22 PB |
5125 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5126 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5127 | break; |
5128 | } | |
93bd25bb | 5129 | add: |
5262f567 | 5130 | list_add(&req->list, entry); |
ad8a48ac JA |
5131 | data->timer.function = io_timeout_fn; |
5132 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5133 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5134 | return 0; |
5135 | } | |
5262f567 | 5136 | |
62755e35 JA |
5137 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5138 | { | |
5139 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5140 | ||
5141 | return req->user_data == (unsigned long) data; | |
5142 | } | |
5143 | ||
e977d6d3 | 5144 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5145 | { |
62755e35 | 5146 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5147 | int ret = 0; |
5148 | ||
4f26bda1 | 5149 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5150 | switch (cancel_ret) { |
5151 | case IO_WQ_CANCEL_OK: | |
5152 | ret = 0; | |
5153 | break; | |
5154 | case IO_WQ_CANCEL_RUNNING: | |
5155 | ret = -EALREADY; | |
5156 | break; | |
5157 | case IO_WQ_CANCEL_NOTFOUND: | |
5158 | ret = -ENOENT; | |
5159 | break; | |
5160 | } | |
5161 | ||
e977d6d3 JA |
5162 | return ret; |
5163 | } | |
5164 | ||
47f46768 JA |
5165 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5166 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5167 | int success_ret) |
47f46768 JA |
5168 | { |
5169 | unsigned long flags; | |
5170 | int ret; | |
5171 | ||
5172 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5173 | if (ret != -ENOENT) { | |
5174 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5175 | goto done; | |
5176 | } | |
5177 | ||
5178 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5179 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5180 | if (ret != -ENOENT) | |
5181 | goto done; | |
5182 | ret = io_poll_cancel(ctx, sqe_addr); | |
5183 | done: | |
b0dd8a41 JA |
5184 | if (!ret) |
5185 | ret = success_ret; | |
47f46768 JA |
5186 | io_cqring_fill_event(req, ret); |
5187 | io_commit_cqring(ctx); | |
5188 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5189 | io_cqring_ev_posted(ctx); | |
5190 | ||
4e88d6e7 JA |
5191 | if (ret < 0) |
5192 | req_set_fail_links(req); | |
014db007 | 5193 | io_put_req(req); |
47f46768 JA |
5194 | } |
5195 | ||
3529d8c2 JA |
5196 | static int io_async_cancel_prep(struct io_kiocb *req, |
5197 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5198 | { |
fbf23849 | 5199 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
5200 | return -EINVAL; |
5201 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
5202 | sqe->cancel_flags) | |
5203 | return -EINVAL; | |
5204 | ||
fbf23849 JA |
5205 | req->cancel.addr = READ_ONCE(sqe->addr); |
5206 | return 0; | |
5207 | } | |
5208 | ||
014db007 | 5209 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
5210 | { |
5211 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5212 | |
014db007 | 5213 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5214 | return 0; |
5215 | } | |
5216 | ||
05f3fb3c JA |
5217 | static int io_files_update_prep(struct io_kiocb *req, |
5218 | const struct io_uring_sqe *sqe) | |
5219 | { | |
5220 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
5221 | return -EINVAL; | |
5222 | ||
5223 | req->files_update.offset = READ_ONCE(sqe->off); | |
5224 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
5225 | if (!req->files_update.nr_args) | |
5226 | return -EINVAL; | |
5227 | req->files_update.arg = READ_ONCE(sqe->addr); | |
5228 | return 0; | |
5229 | } | |
5230 | ||
229a7b63 JA |
5231 | static int io_files_update(struct io_kiocb *req, bool force_nonblock, |
5232 | struct io_comp_state *cs) | |
fbf23849 JA |
5233 | { |
5234 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
5235 | struct io_uring_files_update up; |
5236 | int ret; | |
fbf23849 | 5237 | |
f86cd20c | 5238 | if (force_nonblock) |
05f3fb3c | 5239 | return -EAGAIN; |
05f3fb3c JA |
5240 | |
5241 | up.offset = req->files_update.offset; | |
5242 | up.fds = req->files_update.arg; | |
5243 | ||
5244 | mutex_lock(&ctx->uring_lock); | |
5245 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
5246 | mutex_unlock(&ctx->uring_lock); | |
5247 | ||
5248 | if (ret < 0) | |
5249 | req_set_fail_links(req); | |
229a7b63 | 5250 | __io_req_complete(req, ret, 0, cs); |
5262f567 JA |
5251 | return 0; |
5252 | } | |
5253 | ||
3529d8c2 | 5254 | static int io_req_defer_prep(struct io_kiocb *req, |
debb85f4 | 5255 | const struct io_uring_sqe *sqe) |
f67676d1 | 5256 | { |
e781573e | 5257 | ssize_t ret = 0; |
f67676d1 | 5258 | |
f1d96a8f PB |
5259 | if (!sqe) |
5260 | return 0; | |
5261 | ||
710c2bfb | 5262 | if (io_op_defs[req->opcode].file_table) { |
c40f6379 | 5263 | io_req_init_async(req); |
710c2bfb PB |
5264 | ret = io_grab_files(req); |
5265 | if (unlikely(ret)) | |
5266 | return ret; | |
5267 | } | |
7cdaf587 | 5268 | |
d625c6ee | 5269 | switch (req->opcode) { |
e781573e JA |
5270 | case IORING_OP_NOP: |
5271 | break; | |
f67676d1 JA |
5272 | case IORING_OP_READV: |
5273 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5274 | case IORING_OP_READ: |
3529d8c2 | 5275 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
5276 | break; |
5277 | case IORING_OP_WRITEV: | |
5278 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5279 | case IORING_OP_WRITE: |
3529d8c2 | 5280 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 5281 | break; |
0969e783 | 5282 | case IORING_OP_POLL_ADD: |
3529d8c2 | 5283 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
5284 | break; |
5285 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 5286 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 5287 | break; |
8ed8d3c3 | 5288 | case IORING_OP_FSYNC: |
3529d8c2 | 5289 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
5290 | break; |
5291 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 5292 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 5293 | break; |
03b1230c | 5294 | case IORING_OP_SENDMSG: |
fddaface | 5295 | case IORING_OP_SEND: |
3529d8c2 | 5296 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
5297 | break; |
5298 | case IORING_OP_RECVMSG: | |
fddaface | 5299 | case IORING_OP_RECV: |
3529d8c2 | 5300 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 5301 | break; |
f499a021 | 5302 | case IORING_OP_CONNECT: |
3529d8c2 | 5303 | ret = io_connect_prep(req, sqe); |
f499a021 | 5304 | break; |
2d28390a | 5305 | case IORING_OP_TIMEOUT: |
3529d8c2 | 5306 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 5307 | break; |
b29472ee | 5308 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 5309 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 5310 | break; |
fbf23849 | 5311 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 5312 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 5313 | break; |
2d28390a | 5314 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 5315 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 5316 | break; |
8ed8d3c3 | 5317 | case IORING_OP_ACCEPT: |
3529d8c2 | 5318 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 5319 | break; |
d63d1b5e JA |
5320 | case IORING_OP_FALLOCATE: |
5321 | ret = io_fallocate_prep(req, sqe); | |
5322 | break; | |
15b71abe JA |
5323 | case IORING_OP_OPENAT: |
5324 | ret = io_openat_prep(req, sqe); | |
5325 | break; | |
b5dba59e JA |
5326 | case IORING_OP_CLOSE: |
5327 | ret = io_close_prep(req, sqe); | |
5328 | break; | |
05f3fb3c JA |
5329 | case IORING_OP_FILES_UPDATE: |
5330 | ret = io_files_update_prep(req, sqe); | |
5331 | break; | |
eddc7ef5 JA |
5332 | case IORING_OP_STATX: |
5333 | ret = io_statx_prep(req, sqe); | |
5334 | break; | |
4840e418 JA |
5335 | case IORING_OP_FADVISE: |
5336 | ret = io_fadvise_prep(req, sqe); | |
5337 | break; | |
c1ca757b JA |
5338 | case IORING_OP_MADVISE: |
5339 | ret = io_madvise_prep(req, sqe); | |
5340 | break; | |
cebdb986 JA |
5341 | case IORING_OP_OPENAT2: |
5342 | ret = io_openat2_prep(req, sqe); | |
5343 | break; | |
3e4827b0 JA |
5344 | case IORING_OP_EPOLL_CTL: |
5345 | ret = io_epoll_ctl_prep(req, sqe); | |
5346 | break; | |
7d67af2c PB |
5347 | case IORING_OP_SPLICE: |
5348 | ret = io_splice_prep(req, sqe); | |
5349 | break; | |
ddf0322d JA |
5350 | case IORING_OP_PROVIDE_BUFFERS: |
5351 | ret = io_provide_buffers_prep(req, sqe); | |
5352 | break; | |
067524e9 JA |
5353 | case IORING_OP_REMOVE_BUFFERS: |
5354 | ret = io_remove_buffers_prep(req, sqe); | |
5355 | break; | |
f2a8d5c7 PB |
5356 | case IORING_OP_TEE: |
5357 | ret = io_tee_prep(req, sqe); | |
5358 | break; | |
f67676d1 | 5359 | default: |
e781573e JA |
5360 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5361 | req->opcode); | |
5362 | ret = -EINVAL; | |
b7bb4f7d | 5363 | break; |
f67676d1 JA |
5364 | } |
5365 | ||
b7bb4f7d | 5366 | return ret; |
f67676d1 JA |
5367 | } |
5368 | ||
3529d8c2 | 5369 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5370 | { |
a197f664 | 5371 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5372 | int ret; |
de0617e4 | 5373 | |
9d858b21 | 5374 | /* Still need defer if there is pending req in defer list. */ |
4ee36314 | 5375 | if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) |
de0617e4 JA |
5376 | return 0; |
5377 | ||
650b5481 PB |
5378 | if (!req->io) { |
5379 | if (io_alloc_async_ctx(req)) | |
5380 | return -EAGAIN; | |
debb85f4 | 5381 | ret = io_req_defer_prep(req, sqe); |
650b5481 PB |
5382 | if (ret < 0) |
5383 | return ret; | |
5384 | } | |
cbdcb435 | 5385 | io_prep_async_link(req); |
2d28390a | 5386 | |
de0617e4 | 5387 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5388 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5389 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5390 | return 0; |
5391 | } | |
5392 | ||
915967f6 | 5393 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5394 | list_add_tail(&req->list, &ctx->defer_list); |
5395 | spin_unlock_irq(&ctx->completion_lock); | |
5396 | return -EIOCBQUEUED; | |
5397 | } | |
5398 | ||
99bc4c38 PB |
5399 | static void io_cleanup_req(struct io_kiocb *req) |
5400 | { | |
5401 | struct io_async_ctx *io = req->io; | |
5402 | ||
5403 | switch (req->opcode) { | |
5404 | case IORING_OP_READV: | |
5405 | case IORING_OP_READ_FIXED: | |
5406 | case IORING_OP_READ: | |
bcda7baa JA |
5407 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5408 | kfree((void *)(unsigned long)req->rw.addr); | |
5409 | /* fallthrough */ | |
99bc4c38 PB |
5410 | case IORING_OP_WRITEV: |
5411 | case IORING_OP_WRITE_FIXED: | |
5412 | case IORING_OP_WRITE: | |
5413 | if (io->rw.iov != io->rw.fast_iov) | |
5414 | kfree(io->rw.iov); | |
5415 | break; | |
99bc4c38 | 5416 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5417 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5418 | kfree(req->sr_msg.kbuf); | |
5419 | /* fallthrough */ | |
5420 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5421 | if (io->msg.iov != io->msg.fast_iov) |
5422 | kfree(io->msg.iov); | |
5423 | break; | |
bcda7baa JA |
5424 | case IORING_OP_RECV: |
5425 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5426 | kfree(req->sr_msg.kbuf); | |
5427 | break; | |
8fef80bf PB |
5428 | case IORING_OP_OPENAT: |
5429 | case IORING_OP_OPENAT2: | |
8fef80bf | 5430 | break; |
7d67af2c | 5431 | case IORING_OP_SPLICE: |
f2a8d5c7 | 5432 | case IORING_OP_TEE: |
7d67af2c PB |
5433 | io_put_file(req, req->splice.file_in, |
5434 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5435 | break; | |
99bc4c38 PB |
5436 | } |
5437 | ||
5438 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5439 | } | |
5440 | ||
3529d8c2 | 5441 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 5442 | bool force_nonblock, struct io_comp_state *cs) |
2b188cc1 | 5443 | { |
a197f664 | 5444 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5445 | int ret; |
2b188cc1 | 5446 | |
d625c6ee | 5447 | switch (req->opcode) { |
2b188cc1 | 5448 | case IORING_OP_NOP: |
229a7b63 | 5449 | ret = io_nop(req, cs); |
2b188cc1 JA |
5450 | break; |
5451 | case IORING_OP_READV: | |
edafccee | 5452 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5453 | case IORING_OP_READ: |
3529d8c2 JA |
5454 | if (sqe) { |
5455 | ret = io_read_prep(req, sqe, force_nonblock); | |
5456 | if (ret < 0) | |
5457 | break; | |
5458 | } | |
a1d7c393 | 5459 | ret = io_read(req, force_nonblock, cs); |
edafccee | 5460 | break; |
3529d8c2 | 5461 | case IORING_OP_WRITEV: |
edafccee | 5462 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5463 | case IORING_OP_WRITE: |
3529d8c2 JA |
5464 | if (sqe) { |
5465 | ret = io_write_prep(req, sqe, force_nonblock); | |
5466 | if (ret < 0) | |
5467 | break; | |
5468 | } | |
a1d7c393 | 5469 | ret = io_write(req, force_nonblock, cs); |
2b188cc1 | 5470 | break; |
c992fe29 | 5471 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5472 | if (sqe) { |
5473 | ret = io_prep_fsync(req, sqe); | |
5474 | if (ret < 0) | |
5475 | break; | |
5476 | } | |
014db007 | 5477 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5478 | break; |
221c5eb2 | 5479 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5480 | if (sqe) { |
5481 | ret = io_poll_add_prep(req, sqe); | |
5482 | if (ret) | |
5483 | break; | |
5484 | } | |
014db007 | 5485 | ret = io_poll_add(req); |
221c5eb2 JA |
5486 | break; |
5487 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5488 | if (sqe) { |
5489 | ret = io_poll_remove_prep(req, sqe); | |
5490 | if (ret < 0) | |
5491 | break; | |
5492 | } | |
fc4df999 | 5493 | ret = io_poll_remove(req); |
221c5eb2 | 5494 | break; |
5d17b4a4 | 5495 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5496 | if (sqe) { |
5497 | ret = io_prep_sfr(req, sqe); | |
5498 | if (ret < 0) | |
5499 | break; | |
5500 | } | |
014db007 | 5501 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5502 | break; |
0fa03c62 | 5503 | case IORING_OP_SENDMSG: |
fddaface | 5504 | case IORING_OP_SEND: |
3529d8c2 JA |
5505 | if (sqe) { |
5506 | ret = io_sendmsg_prep(req, sqe); | |
5507 | if (ret < 0) | |
5508 | break; | |
5509 | } | |
fddaface | 5510 | if (req->opcode == IORING_OP_SENDMSG) |
229a7b63 | 5511 | ret = io_sendmsg(req, force_nonblock, cs); |
fddaface | 5512 | else |
229a7b63 | 5513 | ret = io_send(req, force_nonblock, cs); |
0fa03c62 | 5514 | break; |
aa1fa28f | 5515 | case IORING_OP_RECVMSG: |
fddaface | 5516 | case IORING_OP_RECV: |
3529d8c2 JA |
5517 | if (sqe) { |
5518 | ret = io_recvmsg_prep(req, sqe); | |
5519 | if (ret) | |
5520 | break; | |
5521 | } | |
fddaface | 5522 | if (req->opcode == IORING_OP_RECVMSG) |
229a7b63 | 5523 | ret = io_recvmsg(req, force_nonblock, cs); |
fddaface | 5524 | else |
229a7b63 | 5525 | ret = io_recv(req, force_nonblock, cs); |
aa1fa28f | 5526 | break; |
5262f567 | 5527 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5528 | if (sqe) { |
5529 | ret = io_timeout_prep(req, sqe, false); | |
5530 | if (ret) | |
5531 | break; | |
5532 | } | |
fc4df999 | 5533 | ret = io_timeout(req); |
5262f567 | 5534 | break; |
11365043 | 5535 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5536 | if (sqe) { |
5537 | ret = io_timeout_remove_prep(req, sqe); | |
5538 | if (ret) | |
5539 | break; | |
5540 | } | |
fc4df999 | 5541 | ret = io_timeout_remove(req); |
11365043 | 5542 | break; |
17f2fe35 | 5543 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5544 | if (sqe) { |
5545 | ret = io_accept_prep(req, sqe); | |
5546 | if (ret) | |
5547 | break; | |
5548 | } | |
229a7b63 | 5549 | ret = io_accept(req, force_nonblock, cs); |
17f2fe35 | 5550 | break; |
f8e85cf2 | 5551 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5552 | if (sqe) { |
5553 | ret = io_connect_prep(req, sqe); | |
5554 | if (ret) | |
5555 | break; | |
5556 | } | |
229a7b63 | 5557 | ret = io_connect(req, force_nonblock, cs); |
f8e85cf2 | 5558 | break; |
62755e35 | 5559 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5560 | if (sqe) { |
5561 | ret = io_async_cancel_prep(req, sqe); | |
5562 | if (ret) | |
5563 | break; | |
5564 | } | |
014db007 | 5565 | ret = io_async_cancel(req); |
62755e35 | 5566 | break; |
d63d1b5e JA |
5567 | case IORING_OP_FALLOCATE: |
5568 | if (sqe) { | |
5569 | ret = io_fallocate_prep(req, sqe); | |
5570 | if (ret) | |
5571 | break; | |
5572 | } | |
014db007 | 5573 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5574 | break; |
15b71abe JA |
5575 | case IORING_OP_OPENAT: |
5576 | if (sqe) { | |
5577 | ret = io_openat_prep(req, sqe); | |
5578 | if (ret) | |
5579 | break; | |
5580 | } | |
014db007 | 5581 | ret = io_openat(req, force_nonblock); |
15b71abe | 5582 | break; |
b5dba59e JA |
5583 | case IORING_OP_CLOSE: |
5584 | if (sqe) { | |
5585 | ret = io_close_prep(req, sqe); | |
5586 | if (ret) | |
5587 | break; | |
5588 | } | |
229a7b63 | 5589 | ret = io_close(req, force_nonblock, cs); |
b5dba59e | 5590 | break; |
05f3fb3c JA |
5591 | case IORING_OP_FILES_UPDATE: |
5592 | if (sqe) { | |
5593 | ret = io_files_update_prep(req, sqe); | |
5594 | if (ret) | |
5595 | break; | |
5596 | } | |
229a7b63 | 5597 | ret = io_files_update(req, force_nonblock, cs); |
05f3fb3c | 5598 | break; |
eddc7ef5 JA |
5599 | case IORING_OP_STATX: |
5600 | if (sqe) { | |
5601 | ret = io_statx_prep(req, sqe); | |
5602 | if (ret) | |
5603 | break; | |
5604 | } | |
014db007 | 5605 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5606 | break; |
4840e418 JA |
5607 | case IORING_OP_FADVISE: |
5608 | if (sqe) { | |
5609 | ret = io_fadvise_prep(req, sqe); | |
5610 | if (ret) | |
5611 | break; | |
5612 | } | |
014db007 | 5613 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5614 | break; |
c1ca757b JA |
5615 | case IORING_OP_MADVISE: |
5616 | if (sqe) { | |
5617 | ret = io_madvise_prep(req, sqe); | |
5618 | if (ret) | |
5619 | break; | |
5620 | } | |
014db007 | 5621 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5622 | break; |
cebdb986 JA |
5623 | case IORING_OP_OPENAT2: |
5624 | if (sqe) { | |
5625 | ret = io_openat2_prep(req, sqe); | |
5626 | if (ret) | |
5627 | break; | |
5628 | } | |
014db007 | 5629 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5630 | break; |
3e4827b0 JA |
5631 | case IORING_OP_EPOLL_CTL: |
5632 | if (sqe) { | |
5633 | ret = io_epoll_ctl_prep(req, sqe); | |
5634 | if (ret) | |
5635 | break; | |
5636 | } | |
229a7b63 | 5637 | ret = io_epoll_ctl(req, force_nonblock, cs); |
3e4827b0 | 5638 | break; |
7d67af2c PB |
5639 | case IORING_OP_SPLICE: |
5640 | if (sqe) { | |
5641 | ret = io_splice_prep(req, sqe); | |
5642 | if (ret < 0) | |
5643 | break; | |
5644 | } | |
014db007 | 5645 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5646 | break; |
ddf0322d JA |
5647 | case IORING_OP_PROVIDE_BUFFERS: |
5648 | if (sqe) { | |
5649 | ret = io_provide_buffers_prep(req, sqe); | |
5650 | if (ret) | |
5651 | break; | |
5652 | } | |
229a7b63 | 5653 | ret = io_provide_buffers(req, force_nonblock, cs); |
ddf0322d | 5654 | break; |
067524e9 JA |
5655 | case IORING_OP_REMOVE_BUFFERS: |
5656 | if (sqe) { | |
5657 | ret = io_remove_buffers_prep(req, sqe); | |
5658 | if (ret) | |
5659 | break; | |
5660 | } | |
229a7b63 | 5661 | ret = io_remove_buffers(req, force_nonblock, cs); |
3e4827b0 | 5662 | break; |
f2a8d5c7 PB |
5663 | case IORING_OP_TEE: |
5664 | if (sqe) { | |
5665 | ret = io_tee_prep(req, sqe); | |
5666 | if (ret < 0) | |
5667 | break; | |
5668 | } | |
5669 | ret = io_tee(req, force_nonblock); | |
5670 | break; | |
2b188cc1 JA |
5671 | default: |
5672 | ret = -EINVAL; | |
5673 | break; | |
5674 | } | |
5675 | ||
def596e9 JA |
5676 | if (ret) |
5677 | return ret; | |
5678 | ||
b532576e JA |
5679 | /* If the op doesn't have a file, we're not polling for it */ |
5680 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
5681 | const bool in_async = io_wq_current_is_worker(); |
5682 | ||
11ba820b JA |
5683 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5684 | if (in_async) | |
5685 | mutex_lock(&ctx->uring_lock); | |
5686 | ||
def596e9 | 5687 | io_iopoll_req_issued(req); |
11ba820b JA |
5688 | |
5689 | if (in_async) | |
5690 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5691 | } |
5692 | ||
5693 | return 0; | |
2b188cc1 JA |
5694 | } |
5695 | ||
f4db7182 | 5696 | static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
5697 | { |
5698 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 5699 | struct io_kiocb *timeout; |
561fb04a | 5700 | int ret = 0; |
2b188cc1 | 5701 | |
6df1db6b PB |
5702 | timeout = io_prep_linked_timeout(req); |
5703 | if (timeout) | |
5704 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 5705 | |
0c9d5ccd JA |
5706 | /* if NO_CANCEL is set, we must still run the work */ |
5707 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5708 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5709 | ret = -ECANCELED; |
0c9d5ccd | 5710 | } |
31b51510 | 5711 | |
561fb04a | 5712 | if (!ret) { |
561fb04a | 5713 | do { |
f13fad7b | 5714 | ret = io_issue_sqe(req, NULL, false, NULL); |
561fb04a JA |
5715 | /* |
5716 | * We can get EAGAIN for polled IO even though we're | |
5717 | * forcing a sync submission from here, since we can't | |
5718 | * wait for request slots on the block side. | |
5719 | */ | |
5720 | if (ret != -EAGAIN) | |
5721 | break; | |
5722 | cond_resched(); | |
5723 | } while (1); | |
5724 | } | |
31b51510 | 5725 | |
561fb04a | 5726 | if (ret) { |
4e88d6e7 | 5727 | req_set_fail_links(req); |
e1e16097 | 5728 | io_req_complete(req, ret); |
edafccee | 5729 | } |
2b188cc1 | 5730 | |
f4db7182 | 5731 | return io_steal_work(req); |
2b188cc1 JA |
5732 | } |
5733 | ||
65e19f54 JA |
5734 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5735 | int index) | |
5736 | { | |
5737 | struct fixed_file_table *table; | |
5738 | ||
05f3fb3c | 5739 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 5740 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
5741 | } |
5742 | ||
8da11c19 PB |
5743 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5744 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5745 | { |
a197f664 | 5746 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5747 | struct file *file; |
09bb8394 | 5748 | |
8da11c19 | 5749 | if (fixed) { |
05f3fb3c | 5750 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5751 | (unsigned) fd >= ctx->nr_user_files)) |
5752 | return -EBADF; | |
b7620121 | 5753 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 5754 | file = io_file_from_index(ctx, fd); |
fd2206e4 JA |
5755 | if (file) { |
5756 | req->fixed_file_refs = ctx->file_data->cur_refs; | |
5757 | percpu_ref_get(req->fixed_file_refs); | |
5758 | } | |
09bb8394 | 5759 | } else { |
c826bd7a | 5760 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 5761 | file = __io_file_get(state, fd); |
09bb8394 JA |
5762 | } |
5763 | ||
fd2206e4 JA |
5764 | if (file || io_op_defs[req->opcode].needs_file_no_error) { |
5765 | *out_file = file; | |
5766 | return 0; | |
5767 | } | |
5768 | return -EBADF; | |
09bb8394 JA |
5769 | } |
5770 | ||
8da11c19 | 5771 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
63ff8223 | 5772 | int fd) |
8da11c19 | 5773 | { |
8da11c19 PB |
5774 | bool fixed; |
5775 | ||
63ff8223 | 5776 | fixed = (req->flags & REQ_F_FIXED_FILE) != 0; |
0cdaf760 | 5777 | if (unlikely(!fixed && io_async_submit(req->ctx))) |
8da11c19 PB |
5778 | return -EBADF; |
5779 | ||
5780 | return io_file_get(state, req, fd, &req->file, fixed); | |
5781 | } | |
5782 | ||
a197f664 | 5783 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5784 | { |
5785 | int ret = -EBADF; | |
a197f664 | 5786 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5787 | |
5b0bbee4 | 5788 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5789 | return 0; |
b14cca0c | 5790 | if (!ctx->ring_file) |
b5dba59e JA |
5791 | return -EBADF; |
5792 | ||
fcb323cc JA |
5793 | rcu_read_lock(); |
5794 | spin_lock_irq(&ctx->inflight_lock); | |
5795 | /* | |
5796 | * We use the f_ops->flush() handler to ensure that we can flush | |
5797 | * out work accessing these files if the fd is closed. Check if | |
5798 | * the fd has changed since we started down this path, and disallow | |
5799 | * this operation if it has. | |
5800 | */ | |
b14cca0c | 5801 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5802 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5803 | req->flags |= REQ_F_INFLIGHT; | |
5804 | req->work.files = current->files; | |
5805 | ret = 0; | |
5806 | } | |
5807 | spin_unlock_irq(&ctx->inflight_lock); | |
5808 | rcu_read_unlock(); | |
5809 | ||
5810 | return ret; | |
5811 | } | |
5812 | ||
2665abfd | 5813 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5814 | { |
ad8a48ac JA |
5815 | struct io_timeout_data *data = container_of(timer, |
5816 | struct io_timeout_data, timer); | |
5817 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5818 | struct io_ring_ctx *ctx = req->ctx; |
5819 | struct io_kiocb *prev = NULL; | |
5820 | unsigned long flags; | |
2665abfd JA |
5821 | |
5822 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5823 | ||
5824 | /* | |
5825 | * We don't expect the list to be empty, that will only happen if we | |
5826 | * race with the completion of the linked work. | |
5827 | */ | |
4493233e PB |
5828 | if (!list_empty(&req->link_list)) { |
5829 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5830 | link_list); | |
5d960724 | 5831 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5832 | list_del_init(&req->link_list); |
5d960724 JA |
5833 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5834 | } else | |
76a46e06 | 5835 | prev = NULL; |
2665abfd JA |
5836 | } |
5837 | ||
5838 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5839 | ||
5840 | if (prev) { | |
4e88d6e7 | 5841 | req_set_fail_links(prev); |
014db007 | 5842 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5843 | io_put_req(prev); |
47f46768 | 5844 | } else { |
e1e16097 | 5845 | io_req_complete(req, -ETIME); |
2665abfd | 5846 | } |
2665abfd JA |
5847 | return HRTIMER_NORESTART; |
5848 | } | |
5849 | ||
ad8a48ac | 5850 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5851 | { |
76a46e06 | 5852 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5853 | |
76a46e06 JA |
5854 | /* |
5855 | * If the list is now empty, then our linked request finished before | |
5856 | * we got a chance to setup the timer | |
5857 | */ | |
5858 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5859 | if (!list_empty(&req->link_list)) { |
2d28390a | 5860 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5861 | |
ad8a48ac JA |
5862 | data->timer.function = io_link_timeout_fn; |
5863 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5864 | data->mode); | |
2665abfd | 5865 | } |
76a46e06 | 5866 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5867 | |
2665abfd | 5868 | /* drop submission reference */ |
76a46e06 JA |
5869 | io_put_req(req); |
5870 | } | |
2665abfd | 5871 | |
ad8a48ac | 5872 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5873 | { |
5874 | struct io_kiocb *nxt; | |
5875 | ||
dea3b49c | 5876 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5877 | return NULL; |
6df1db6b | 5878 | if (req->flags & REQ_F_LINK_TIMEOUT) |
d7718a9d | 5879 | return NULL; |
2665abfd | 5880 | |
4493233e PB |
5881 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5882 | link_list); | |
d625c6ee | 5883 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5884 | return NULL; |
2665abfd | 5885 | |
76a46e06 | 5886 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5887 | return nxt; |
2665abfd JA |
5888 | } |
5889 | ||
f13fad7b JA |
5890 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5891 | struct io_comp_state *cs) | |
2b188cc1 | 5892 | { |
4a0a7a18 | 5893 | struct io_kiocb *linked_timeout; |
4bc4494e | 5894 | struct io_kiocb *nxt; |
193155c8 | 5895 | const struct cred *old_creds = NULL; |
e0c5c576 | 5896 | int ret; |
2b188cc1 | 5897 | |
4a0a7a18 JA |
5898 | again: |
5899 | linked_timeout = io_prep_linked_timeout(req); | |
5900 | ||
7cdaf587 XW |
5901 | if ((req->flags & REQ_F_WORK_INITIALIZED) && req->work.creds && |
5902 | req->work.creds != current_cred()) { | |
193155c8 JA |
5903 | if (old_creds) |
5904 | revert_creds(old_creds); | |
5905 | if (old_creds == req->work.creds) | |
5906 | old_creds = NULL; /* restored original creds */ | |
5907 | else | |
5908 | old_creds = override_creds(req->work.creds); | |
5909 | } | |
5910 | ||
f13fad7b | 5911 | ret = io_issue_sqe(req, sqe, true, cs); |
491381ce JA |
5912 | |
5913 | /* | |
5914 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5915 | * doesn't support non-blocking read/write attempts | |
5916 | */ | |
24c74678 | 5917 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
d7718a9d JA |
5918 | if (io_arm_poll_handler(req)) { |
5919 | if (linked_timeout) | |
5920 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5921 | goto exit; |
d7718a9d | 5922 | } |
86a761f8 | 5923 | punt: |
7cdaf587 XW |
5924 | io_req_init_async(req); |
5925 | ||
f86cd20c | 5926 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5927 | ret = io_grab_files(req); |
5928 | if (ret) | |
5929 | goto err; | |
2b188cc1 | 5930 | } |
bbad27b2 PB |
5931 | |
5932 | /* | |
5933 | * Queued up for async execution, worker will release | |
5934 | * submit reference when the iocb is actually submitted. | |
5935 | */ | |
5936 | io_queue_async_work(req); | |
4bc4494e | 5937 | goto exit; |
2b188cc1 | 5938 | } |
e65ef56d | 5939 | |
652532ad | 5940 | if (unlikely(ret)) { |
fcb323cc | 5941 | err: |
652532ad PB |
5942 | /* un-prep timeout, so it'll be killed as any other linked */ |
5943 | req->flags &= ~REQ_F_LINK_TIMEOUT; | |
4e88d6e7 | 5944 | req_set_fail_links(req); |
652532ad | 5945 | io_put_req(req); |
e1e16097 | 5946 | io_req_complete(req, ret); |
652532ad | 5947 | goto exit; |
9e645e11 | 5948 | } |
652532ad PB |
5949 | |
5950 | /* drop submission reference */ | |
5951 | nxt = io_put_req_find_next(req); | |
5952 | if (linked_timeout) | |
5953 | io_queue_linked_timeout(linked_timeout); | |
5954 | ||
4a0a7a18 JA |
5955 | if (nxt) { |
5956 | req = nxt; | |
86a761f8 PB |
5957 | |
5958 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5959 | goto punt; | |
4a0a7a18 JA |
5960 | goto again; |
5961 | } | |
4bc4494e | 5962 | exit: |
193155c8 JA |
5963 | if (old_creds) |
5964 | revert_creds(old_creds); | |
2b188cc1 JA |
5965 | } |
5966 | ||
f13fad7b JA |
5967 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5968 | struct io_comp_state *cs) | |
4fe2c963 JL |
5969 | { |
5970 | int ret; | |
5971 | ||
3529d8c2 | 5972 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5973 | if (ret) { |
5974 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5975 | fail_req: |
4e88d6e7 | 5976 | req_set_fail_links(req); |
e1e16097 JA |
5977 | io_put_req(req); |
5978 | io_req_complete(req, ret); | |
4fe2c963 | 5979 | } |
2550878f | 5980 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
bd2ab18a PB |
5981 | if (!req->io) { |
5982 | ret = -EAGAIN; | |
5983 | if (io_alloc_async_ctx(req)) | |
5984 | goto fail_req; | |
debb85f4 | 5985 | ret = io_req_defer_prep(req, sqe); |
bd2ab18a PB |
5986 | if (unlikely(ret < 0)) |
5987 | goto fail_req; | |
5988 | } | |
5989 | ||
ce35a47a JA |
5990 | /* |
5991 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5992 | * to async execution. | |
5993 | */ | |
5994 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5995 | io_queue_async_work(req); | |
5996 | } else { | |
f13fad7b | 5997 | __io_queue_sqe(req, sqe, cs); |
ce35a47a | 5998 | } |
4fe2c963 JL |
5999 | } |
6000 | ||
f13fad7b JA |
6001 | static inline void io_queue_link_head(struct io_kiocb *req, |
6002 | struct io_comp_state *cs) | |
4fe2c963 | 6003 | { |
94ae5e77 | 6004 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6005 | io_put_req(req); |
6006 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6007 | } else |
f13fad7b | 6008 | io_queue_sqe(req, NULL, cs); |
4fe2c963 JL |
6009 | } |
6010 | ||
1d4240cc | 6011 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
f13fad7b | 6012 | struct io_kiocb **link, struct io_comp_state *cs) |
9e645e11 | 6013 | { |
a197f664 | 6014 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6015 | int ret; |
9e645e11 | 6016 | |
9e645e11 JA |
6017 | /* |
6018 | * If we already have a head request, queue this one for async | |
6019 | * submittal once the head completes. If we don't have a head but | |
6020 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6021 | * submitted sync once the chain is complete. If none of those | |
6022 | * conditions are true (normal request), then just queue it. | |
6023 | */ | |
6024 | if (*link) { | |
9d76377f | 6025 | struct io_kiocb *head = *link; |
4e88d6e7 | 6026 | |
8cdf2193 PB |
6027 | /* |
6028 | * Taking sequential execution of a link, draining both sides | |
6029 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6030 | * requests in the link. So, it drains the head and the | |
6031 | * next after the link request. The last one is done via | |
6032 | * drain_next flag to persist the effect across calls. | |
6033 | */ | |
ef4ff581 | 6034 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6035 | head->flags |= REQ_F_IO_DRAIN; |
6036 | ctx->drain_next = 1; | |
6037 | } | |
1d4240cc PB |
6038 | if (io_alloc_async_ctx(req)) |
6039 | return -EAGAIN; | |
9e645e11 | 6040 | |
debb85f4 | 6041 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 6042 | if (ret) { |
4e88d6e7 | 6043 | /* fail even hard links since we don't submit */ |
9d76377f | 6044 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6045 | return ret; |
2d28390a | 6046 | } |
9d76377f | 6047 | trace_io_uring_link(ctx, req, head); |
c40f6379 | 6048 | io_get_req_task(req); |
9d76377f | 6049 | list_add_tail(&req->link_list, &head->link_list); |
32fe525b PB |
6050 | |
6051 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6052 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
f13fad7b | 6053 | io_queue_link_head(head, cs); |
32fe525b PB |
6054 | *link = NULL; |
6055 | } | |
9e645e11 | 6056 | } else { |
711be031 PB |
6057 | if (unlikely(ctx->drain_next)) { |
6058 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6059 | ctx->drain_next = 0; |
711be031 | 6060 | } |
ef4ff581 | 6061 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 6062 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 6063 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 6064 | |
1d4240cc PB |
6065 | if (io_alloc_async_ctx(req)) |
6066 | return -EAGAIN; | |
6067 | ||
debb85f4 | 6068 | ret = io_req_defer_prep(req, sqe); |
711be031 PB |
6069 | if (ret) |
6070 | req->flags |= REQ_F_FAIL_LINK; | |
6071 | *link = req; | |
6072 | } else { | |
f13fad7b | 6073 | io_queue_sqe(req, sqe, cs); |
711be031 | 6074 | } |
9e645e11 | 6075 | } |
2e6e1fde | 6076 | |
1d4240cc | 6077 | return 0; |
9e645e11 JA |
6078 | } |
6079 | ||
9a56a232 JA |
6080 | /* |
6081 | * Batched submission is done, ensure local IO is flushed out. | |
6082 | */ | |
6083 | static void io_submit_state_end(struct io_submit_state *state) | |
6084 | { | |
f13fad7b JA |
6085 | if (!list_empty(&state->comp.list)) |
6086 | io_submit_flush_completions(&state->comp); | |
9a56a232 | 6087 | blk_finish_plug(&state->plug); |
9f13c35b | 6088 | io_state_file_put(state); |
2579f913 | 6089 | if (state->free_reqs) |
6c8a3134 | 6090 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
6091 | } |
6092 | ||
6093 | /* | |
6094 | * Start submission side cache. | |
6095 | */ | |
6096 | static void io_submit_state_start(struct io_submit_state *state, | |
013538bd | 6097 | struct io_ring_ctx *ctx, unsigned int max_ios) |
9a56a232 JA |
6098 | { |
6099 | blk_start_plug(&state->plug); | |
b63534c4 JA |
6100 | #ifdef CONFIG_BLOCK |
6101 | state->plug.nowait = true; | |
6102 | #endif | |
013538bd JA |
6103 | state->comp.nr = 0; |
6104 | INIT_LIST_HEAD(&state->comp.list); | |
6105 | state->comp.ctx = ctx; | |
2579f913 | 6106 | state->free_reqs = 0; |
9a56a232 JA |
6107 | state->file = NULL; |
6108 | state->ios_left = max_ios; | |
6109 | } | |
6110 | ||
2b188cc1 JA |
6111 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6112 | { | |
75b28aff | 6113 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6114 | |
caf582c6 PB |
6115 | /* |
6116 | * Ensure any loads from the SQEs are done at this point, | |
6117 | * since once we write the new head, the application could | |
6118 | * write new data to them. | |
6119 | */ | |
6120 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6121 | } |
6122 | ||
2b188cc1 | 6123 | /* |
3529d8c2 | 6124 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6125 | * that is mapped by userspace. This means that care needs to be taken to |
6126 | * ensure that reads are stable, as we cannot rely on userspace always | |
6127 | * being a good citizen. If members of the sqe are validated and then later | |
6128 | * used, it's important that those reads are done through READ_ONCE() to | |
6129 | * prevent a re-load down the line. | |
6130 | */ | |
709b302f | 6131 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6132 | { |
75b28aff | 6133 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6134 | unsigned head; |
6135 | ||
6136 | /* | |
6137 | * The cached sq head (or cq tail) serves two purposes: | |
6138 | * | |
6139 | * 1) allows us to batch the cost of updating the user visible | |
6140 | * head updates. | |
6141 | * 2) allows the kernel side to track the head on its own, even | |
6142 | * though the application is the one updating it. | |
6143 | */ | |
ee7d46d9 | 6144 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
6145 | if (likely(head < ctx->sq_entries)) |
6146 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6147 | |
6148 | /* drop invalid entries */ | |
498ccd9e | 6149 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6150 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6151 | return NULL; |
6152 | } | |
6153 | ||
6154 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
6155 | { | |
6156 | ctx->cached_sq_head++; | |
2b188cc1 JA |
6157 | } |
6158 | ||
ef4ff581 PB |
6159 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6160 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6161 | IOSQE_BUFFER_SELECT) | |
6162 | ||
6163 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6164 | const struct io_uring_sqe *sqe, | |
0cdaf760 | 6165 | struct io_submit_state *state) |
0553b8bd | 6166 | { |
ef4ff581 | 6167 | unsigned int sqe_flags; |
63ff8223 | 6168 | int id; |
ef4ff581 | 6169 | |
0553b8bd PB |
6170 | /* |
6171 | * All io need record the previous position, if LINK vs DARIN, | |
6172 | * it can be used to mark the position of the first IO in the | |
6173 | * link list. | |
6174 | */ | |
31af27c7 | 6175 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
6176 | req->opcode = READ_ONCE(sqe->opcode); |
6177 | req->user_data = READ_ONCE(sqe->user_data); | |
6178 | req->io = NULL; | |
6179 | req->file = NULL; | |
6180 | req->ctx = ctx; | |
6181 | req->flags = 0; | |
6182 | /* one is dropped after submission, the other at completion */ | |
6183 | refcount_set(&req->refs, 2); | |
4dd2824d | 6184 | req->task = current; |
0553b8bd | 6185 | req->result = 0; |
ef4ff581 PB |
6186 | |
6187 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6188 | return -EINVAL; | |
6189 | ||
9d8426a0 JA |
6190 | if (unlikely(io_sq_thread_acquire_mm(ctx, req))) |
6191 | return -EFAULT; | |
ef4ff581 PB |
6192 | |
6193 | sqe_flags = READ_ONCE(sqe->flags); | |
6194 | /* enforce forwards compatibility on users */ | |
6195 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6196 | return -EINVAL; | |
6197 | ||
6198 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6199 | !io_op_defs[req->opcode].buffer_select) | |
6200 | return -EOPNOTSUPP; | |
6201 | ||
6202 | id = READ_ONCE(sqe->personality); | |
6203 | if (id) { | |
7cdaf587 | 6204 | io_req_init_async(req); |
ef4ff581 PB |
6205 | req->work.creds = idr_find(&ctx->personality_idr, id); |
6206 | if (unlikely(!req->work.creds)) | |
6207 | return -EINVAL; | |
6208 | get_cred(req->work.creds); | |
6209 | } | |
6210 | ||
6211 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6212 | req->flags |= sqe_flags; |
ef4ff581 | 6213 | |
63ff8223 JA |
6214 | if (!io_op_defs[req->opcode].needs_file) |
6215 | return 0; | |
6216 | ||
6217 | return io_req_set_file(state, req, READ_ONCE(sqe->fd)); | |
0553b8bd PB |
6218 | } |
6219 | ||
fb5ccc98 | 6220 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
0cdaf760 | 6221 | struct file *ring_file, int ring_fd) |
6c271ce2 | 6222 | { |
ac8691c4 | 6223 | struct io_submit_state state; |
9e645e11 | 6224 | struct io_kiocb *link = NULL; |
9e645e11 | 6225 | int i, submitted = 0; |
6c271ce2 | 6226 | |
c4a2ed72 | 6227 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
6228 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6229 | if (!list_empty(&ctx->cq_overflow_list) && | |
6230 | !io_cqring_overflow_flush(ctx, false)) | |
6231 | return -EBUSY; | |
6232 | } | |
6c271ce2 | 6233 | |
ee7d46d9 PB |
6234 | /* make sure SQ entry isn't read before tail */ |
6235 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6236 | |
2b85edfc PB |
6237 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6238 | return -EAGAIN; | |
6c271ce2 | 6239 | |
013538bd | 6240 | io_submit_state_start(&state, ctx, nr); |
6c271ce2 | 6241 | |
b14cca0c PB |
6242 | ctx->ring_fd = ring_fd; |
6243 | ctx->ring_file = ring_file; | |
6244 | ||
6c271ce2 | 6245 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6246 | const struct io_uring_sqe *sqe; |
196be95c | 6247 | struct io_kiocb *req; |
1cb1edb2 | 6248 | int err; |
fb5ccc98 | 6249 | |
b1e50e54 PB |
6250 | sqe = io_get_sqe(ctx); |
6251 | if (unlikely(!sqe)) { | |
6252 | io_consume_sqe(ctx); | |
6253 | break; | |
6254 | } | |
ac8691c4 | 6255 | req = io_alloc_req(ctx, &state); |
196be95c PB |
6256 | if (unlikely(!req)) { |
6257 | if (!submitted) | |
6258 | submitted = -EAGAIN; | |
fb5ccc98 | 6259 | break; |
196be95c | 6260 | } |
fb5ccc98 | 6261 | |
ac8691c4 | 6262 | err = io_init_req(ctx, req, sqe, &state); |
709b302f | 6263 | io_consume_sqe(ctx); |
d3656344 JA |
6264 | /* will complete beyond this point, count as submitted */ |
6265 | submitted++; | |
6266 | ||
ef4ff581 | 6267 | if (unlikely(err)) { |
1cb1edb2 | 6268 | fail_req: |
e1e16097 JA |
6269 | io_put_req(req); |
6270 | io_req_complete(req, err); | |
196be95c PB |
6271 | break; |
6272 | } | |
fb5ccc98 | 6273 | |
354420f7 | 6274 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
0cdaf760 | 6275 | true, io_async_submit(ctx)); |
f13fad7b | 6276 | err = io_submit_sqe(req, sqe, &link, &state.comp); |
1d4240cc PB |
6277 | if (err) |
6278 | goto fail_req; | |
6c271ce2 JA |
6279 | } |
6280 | ||
9466f437 PB |
6281 | if (unlikely(submitted != nr)) { |
6282 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
6283 | ||
6284 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
6285 | } | |
9e645e11 | 6286 | if (link) |
f13fad7b | 6287 | io_queue_link_head(link, &state.comp); |
ac8691c4 | 6288 | io_submit_state_end(&state); |
6c271ce2 | 6289 | |
ae9428ca PB |
6290 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6291 | io_commit_sqring(ctx); | |
6292 | ||
6c271ce2 JA |
6293 | return submitted; |
6294 | } | |
6295 | ||
6296 | static int io_sq_thread(void *data) | |
6297 | { | |
6c271ce2 | 6298 | struct io_ring_ctx *ctx = data; |
181e448d | 6299 | const struct cred *old_cred; |
6c271ce2 | 6300 | DEFINE_WAIT(wait); |
6c271ce2 | 6301 | unsigned long timeout; |
bdcd3eab | 6302 | int ret = 0; |
6c271ce2 | 6303 | |
0f158b4c | 6304 | complete(&ctx->sq_thread_comp); |
a4c0b3de | 6305 | |
181e448d | 6306 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 6307 | |
bdcd3eab | 6308 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 6309 | while (!kthread_should_park()) { |
fb5ccc98 | 6310 | unsigned int to_submit; |
6c271ce2 | 6311 | |
bdcd3eab | 6312 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
6313 | unsigned nr_events = 0; |
6314 | ||
bdcd3eab XW |
6315 | mutex_lock(&ctx->uring_lock); |
6316 | if (!list_empty(&ctx->poll_list)) | |
6317 | io_iopoll_getevents(ctx, &nr_events, 0); | |
6318 | else | |
6c271ce2 | 6319 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 6320 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
6321 | } |
6322 | ||
fb5ccc98 | 6323 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
6324 | |
6325 | /* | |
6326 | * If submit got -EBUSY, flag us as needing the application | |
6327 | * to enter the kernel to reap and flush events. | |
6328 | */ | |
b772f07a | 6329 | if (!to_submit || ret == -EBUSY || need_resched()) { |
7143b5ac SG |
6330 | /* |
6331 | * Drop cur_mm before scheduling, we can't hold it for | |
6332 | * long periods (or over schedule()). Do this before | |
6333 | * adding ourselves to the waitqueue, as the unuse/drop | |
6334 | * may sleep. | |
6335 | */ | |
bf9c2f1c | 6336 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 6337 | |
6c271ce2 JA |
6338 | /* |
6339 | * We're polling. If we're within the defined idle | |
6340 | * period, then let us spin without work before going | |
c1edbf5f JA |
6341 | * to sleep. The exception is if we got EBUSY doing |
6342 | * more IO, we should wait for the application to | |
6343 | * reap events and wake us up. | |
6c271ce2 | 6344 | */ |
b772f07a | 6345 | if (!list_empty(&ctx->poll_list) || need_resched() || |
df069d80 JA |
6346 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6347 | !percpu_ref_is_dying(&ctx->refs))) { | |
4c6e277c | 6348 | io_run_task_work(); |
9831a90c | 6349 | cond_resched(); |
6c271ce2 JA |
6350 | continue; |
6351 | } | |
6352 | ||
6c271ce2 JA |
6353 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6354 | TASK_INTERRUPTIBLE); | |
6355 | ||
bdcd3eab XW |
6356 | /* |
6357 | * While doing polled IO, before going to sleep, we need | |
6358 | * to check if there are new reqs added to poll_list, it | |
6359 | * is because reqs may have been punted to io worker and | |
6360 | * will be added to poll_list later, hence check the | |
6361 | * poll_list again. | |
6362 | */ | |
6363 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6364 | !list_empty_careful(&ctx->poll_list)) { | |
6365 | finish_wait(&ctx->sqo_wait, &wait); | |
6366 | continue; | |
6367 | } | |
6368 | ||
6c271ce2 | 6369 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6370 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6371 | /* make sure to read SQ tail after writing flags */ |
6372 | smp_mb(); | |
6c271ce2 | 6373 | |
fb5ccc98 | 6374 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6375 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6376 | if (kthread_should_park()) { |
6c271ce2 JA |
6377 | finish_wait(&ctx->sqo_wait, &wait); |
6378 | break; | |
6379 | } | |
4c6e277c | 6380 | if (io_run_task_work()) { |
10bea96d | 6381 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6382 | continue; |
6383 | } | |
6c271ce2 JA |
6384 | if (signal_pending(current)) |
6385 | flush_signals(current); | |
6386 | schedule(); | |
6387 | finish_wait(&ctx->sqo_wait, &wait); | |
6388 | ||
75b28aff | 6389 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
d4ae271d | 6390 | ret = 0; |
6c271ce2 JA |
6391 | continue; |
6392 | } | |
6393 | finish_wait(&ctx->sqo_wait, &wait); | |
6394 | ||
75b28aff | 6395 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6396 | } |
6397 | ||
8a4955ff | 6398 | mutex_lock(&ctx->uring_lock); |
6b668c9b XW |
6399 | if (likely(!percpu_ref_is_dying(&ctx->refs))) |
6400 | ret = io_submit_sqes(ctx, to_submit, NULL, -1); | |
8a4955ff | 6401 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6402 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6403 | } |
6404 | ||
4c6e277c | 6405 | io_run_task_work(); |
b41e9852 | 6406 | |
bf9c2f1c | 6407 | io_sq_thread_drop_mm(ctx); |
181e448d | 6408 | revert_creds(old_cred); |
06058632 | 6409 | |
2bbcd6d3 | 6410 | kthread_parkme(); |
06058632 | 6411 | |
6c271ce2 JA |
6412 | return 0; |
6413 | } | |
6414 | ||
bda52162 JA |
6415 | struct io_wait_queue { |
6416 | struct wait_queue_entry wq; | |
6417 | struct io_ring_ctx *ctx; | |
6418 | unsigned to_wait; | |
6419 | unsigned nr_timeouts; | |
6420 | }; | |
6421 | ||
1d7bb1d5 | 6422 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6423 | { |
6424 | struct io_ring_ctx *ctx = iowq->ctx; | |
6425 | ||
6426 | /* | |
d195a66e | 6427 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6428 | * started waiting. For timeouts, we always want to return to userspace, |
6429 | * regardless of event count. | |
6430 | */ | |
1d7bb1d5 | 6431 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6432 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6433 | } | |
6434 | ||
6435 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6436 | int wake_flags, void *key) | |
6437 | { | |
6438 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6439 | wq); | |
6440 | ||
1d7bb1d5 JA |
6441 | /* use noflush == true, as we can't safely rely on locking context */ |
6442 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6443 | return -1; |
6444 | ||
6445 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6446 | } | |
6447 | ||
2b188cc1 JA |
6448 | /* |
6449 | * Wait until events become available, if we don't already have some. The | |
6450 | * application must reap them itself, as they reside on the shared cq ring. | |
6451 | */ | |
6452 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6453 | const sigset_t __user *sig, size_t sigsz) | |
6454 | { | |
bda52162 JA |
6455 | struct io_wait_queue iowq = { |
6456 | .wq = { | |
6457 | .private = current, | |
6458 | .func = io_wake_function, | |
6459 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6460 | }, | |
6461 | .ctx = ctx, | |
6462 | .to_wait = min_events, | |
6463 | }; | |
75b28aff | 6464 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6465 | int ret = 0; |
2b188cc1 | 6466 | |
b41e9852 JA |
6467 | do { |
6468 | if (io_cqring_events(ctx, false) >= min_events) | |
6469 | return 0; | |
4c6e277c | 6470 | if (!io_run_task_work()) |
b41e9852 | 6471 | break; |
b41e9852 | 6472 | } while (1); |
2b188cc1 JA |
6473 | |
6474 | if (sig) { | |
9e75ad5d AB |
6475 | #ifdef CONFIG_COMPAT |
6476 | if (in_compat_syscall()) | |
6477 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6478 | sigsz); |
9e75ad5d AB |
6479 | else |
6480 | #endif | |
b772434b | 6481 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6482 | |
2b188cc1 JA |
6483 | if (ret) |
6484 | return ret; | |
6485 | } | |
6486 | ||
bda52162 | 6487 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6488 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6489 | do { |
6490 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6491 | TASK_INTERRUPTIBLE); | |
ce593a6c | 6492 | /* make sure we run task_work before checking for signals */ |
4c6e277c JA |
6493 | if (io_run_task_work()) |
6494 | continue; | |
bda52162 | 6495 | if (signal_pending(current)) { |
b7db41c9 JA |
6496 | if (current->jobctl & JOBCTL_TASK_WORK) { |
6497 | spin_lock_irq(¤t->sighand->siglock); | |
6498 | current->jobctl &= ~JOBCTL_TASK_WORK; | |
6499 | recalc_sigpending(); | |
6500 | spin_unlock_irq(¤t->sighand->siglock); | |
6501 | continue; | |
6502 | } | |
e9ffa5c2 | 6503 | ret = -EINTR; |
bda52162 JA |
6504 | break; |
6505 | } | |
ce593a6c JA |
6506 | if (io_should_wake(&iowq, false)) |
6507 | break; | |
6508 | schedule(); | |
bda52162 JA |
6509 | } while (1); |
6510 | finish_wait(&ctx->wait, &iowq.wq); | |
6511 | ||
e9ffa5c2 | 6512 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6513 | |
75b28aff | 6514 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6515 | } |
6516 | ||
6b06314c JA |
6517 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6518 | { | |
6519 | #if defined(CONFIG_UNIX) | |
6520 | if (ctx->ring_sock) { | |
6521 | struct sock *sock = ctx->ring_sock->sk; | |
6522 | struct sk_buff *skb; | |
6523 | ||
6524 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6525 | kfree_skb(skb); | |
6526 | } | |
6527 | #else | |
6528 | int i; | |
6529 | ||
65e19f54 JA |
6530 | for (i = 0; i < ctx->nr_user_files; i++) { |
6531 | struct file *file; | |
6532 | ||
6533 | file = io_file_from_index(ctx, i); | |
6534 | if (file) | |
6535 | fput(file); | |
6536 | } | |
6b06314c JA |
6537 | #endif |
6538 | } | |
6539 | ||
05f3fb3c JA |
6540 | static void io_file_ref_kill(struct percpu_ref *ref) |
6541 | { | |
6542 | struct fixed_file_data *data; | |
6543 | ||
6544 | data = container_of(ref, struct fixed_file_data, refs); | |
6545 | complete(&data->done); | |
6546 | } | |
6547 | ||
6b06314c JA |
6548 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6549 | { | |
05f3fb3c | 6550 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6551 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 JA |
6552 | unsigned nr_tables, i; |
6553 | ||
05f3fb3c | 6554 | if (!data) |
6b06314c JA |
6555 | return -ENXIO; |
6556 | ||
6a4d07cd | 6557 | spin_lock(&data->lock); |
05589553 XW |
6558 | if (!list_empty(&data->ref_list)) |
6559 | ref_node = list_first_entry(&data->ref_list, | |
6560 | struct fixed_file_ref_node, node); | |
6a4d07cd | 6561 | spin_unlock(&data->lock); |
05589553 XW |
6562 | if (ref_node) |
6563 | percpu_ref_kill(&ref_node->refs); | |
6564 | ||
6565 | percpu_ref_kill(&data->refs); | |
6566 | ||
6567 | /* wait for all refs nodes to complete */ | |
4a38aed2 | 6568 | flush_delayed_work(&ctx->file_put_work); |
2faf852d | 6569 | wait_for_completion(&data->done); |
05f3fb3c | 6570 | |
6b06314c | 6571 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6572 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6573 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6574 | kfree(data->table[i].files); |
6575 | kfree(data->table); | |
05589553 XW |
6576 | percpu_ref_exit(&data->refs); |
6577 | kfree(data); | |
05f3fb3c | 6578 | ctx->file_data = NULL; |
6b06314c JA |
6579 | ctx->nr_user_files = 0; |
6580 | return 0; | |
6581 | } | |
6582 | ||
6c271ce2 JA |
6583 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6584 | { | |
6585 | if (ctx->sqo_thread) { | |
0f158b4c | 6586 | wait_for_completion(&ctx->sq_thread_comp); |
2bbcd6d3 RP |
6587 | /* |
6588 | * The park is a bit of a work-around, without it we get | |
6589 | * warning spews on shutdown with SQPOLL set and affinity | |
6590 | * set to a single CPU. | |
6591 | */ | |
06058632 | 6592 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6593 | kthread_stop(ctx->sqo_thread); |
6594 | ctx->sqo_thread = NULL; | |
6595 | } | |
6596 | } | |
6597 | ||
6b06314c JA |
6598 | static void io_finish_async(struct io_ring_ctx *ctx) |
6599 | { | |
6c271ce2 JA |
6600 | io_sq_thread_stop(ctx); |
6601 | ||
561fb04a JA |
6602 | if (ctx->io_wq) { |
6603 | io_wq_destroy(ctx->io_wq); | |
6604 | ctx->io_wq = NULL; | |
6b06314c JA |
6605 | } |
6606 | } | |
6607 | ||
6608 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6609 | /* |
6610 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6611 | * the io_uring can be safely unregistered on process exit, even if we have | |
6612 | * loops in the file referencing. | |
6613 | */ | |
6614 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6615 | { | |
6616 | struct sock *sk = ctx->ring_sock->sk; | |
6617 | struct scm_fp_list *fpl; | |
6618 | struct sk_buff *skb; | |
08a45173 | 6619 | int i, nr_files; |
6b06314c | 6620 | |
6b06314c JA |
6621 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6622 | if (!fpl) | |
6623 | return -ENOMEM; | |
6624 | ||
6625 | skb = alloc_skb(0, GFP_KERNEL); | |
6626 | if (!skb) { | |
6627 | kfree(fpl); | |
6628 | return -ENOMEM; | |
6629 | } | |
6630 | ||
6631 | skb->sk = sk; | |
6b06314c | 6632 | |
08a45173 | 6633 | nr_files = 0; |
6b06314c JA |
6634 | fpl->user = get_uid(ctx->user); |
6635 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6636 | struct file *file = io_file_from_index(ctx, i + offset); |
6637 | ||
6638 | if (!file) | |
08a45173 | 6639 | continue; |
65e19f54 | 6640 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6641 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6642 | nr_files++; | |
6b06314c JA |
6643 | } |
6644 | ||
08a45173 JA |
6645 | if (nr_files) { |
6646 | fpl->max = SCM_MAX_FD; | |
6647 | fpl->count = nr_files; | |
6648 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6649 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6650 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6651 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6652 | |
08a45173 JA |
6653 | for (i = 0; i < nr_files; i++) |
6654 | fput(fpl->fp[i]); | |
6655 | } else { | |
6656 | kfree_skb(skb); | |
6657 | kfree(fpl); | |
6658 | } | |
6b06314c JA |
6659 | |
6660 | return 0; | |
6661 | } | |
6662 | ||
6663 | /* | |
6664 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6665 | * causes regular reference counting to break down. We rely on the UNIX | |
6666 | * garbage collection to take care of this problem for us. | |
6667 | */ | |
6668 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6669 | { | |
6670 | unsigned left, total; | |
6671 | int ret = 0; | |
6672 | ||
6673 | total = 0; | |
6674 | left = ctx->nr_user_files; | |
6675 | while (left) { | |
6676 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6677 | |
6678 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6679 | if (ret) | |
6680 | break; | |
6681 | left -= this_files; | |
6682 | total += this_files; | |
6683 | } | |
6684 | ||
6685 | if (!ret) | |
6686 | return 0; | |
6687 | ||
6688 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6689 | struct file *file = io_file_from_index(ctx, total); |
6690 | ||
6691 | if (file) | |
6692 | fput(file); | |
6b06314c JA |
6693 | total++; |
6694 | } | |
6695 | ||
6696 | return ret; | |
6697 | } | |
6698 | #else | |
6699 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6700 | { | |
6701 | return 0; | |
6702 | } | |
6703 | #endif | |
6704 | ||
65e19f54 JA |
6705 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6706 | unsigned nr_files) | |
6707 | { | |
6708 | int i; | |
6709 | ||
6710 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6711 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6712 | unsigned this_files; |
6713 | ||
6714 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6715 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6716 | GFP_KERNEL); | |
6717 | if (!table->files) | |
6718 | break; | |
6719 | nr_files -= this_files; | |
6720 | } | |
6721 | ||
6722 | if (i == nr_tables) | |
6723 | return 0; | |
6724 | ||
6725 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6726 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6727 | kfree(table->files); |
6728 | } | |
6729 | return 1; | |
6730 | } | |
6731 | ||
05f3fb3c JA |
6732 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6733 | { | |
6734 | #if defined(CONFIG_UNIX) | |
6735 | struct sock *sock = ctx->ring_sock->sk; | |
6736 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6737 | struct sk_buff *skb; | |
6738 | int i; | |
6739 | ||
6740 | __skb_queue_head_init(&list); | |
6741 | ||
6742 | /* | |
6743 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6744 | * remove this entry and rearrange the file array. | |
6745 | */ | |
6746 | skb = skb_dequeue(head); | |
6747 | while (skb) { | |
6748 | struct scm_fp_list *fp; | |
6749 | ||
6750 | fp = UNIXCB(skb).fp; | |
6751 | for (i = 0; i < fp->count; i++) { | |
6752 | int left; | |
6753 | ||
6754 | if (fp->fp[i] != file) | |
6755 | continue; | |
6756 | ||
6757 | unix_notinflight(fp->user, fp->fp[i]); | |
6758 | left = fp->count - 1 - i; | |
6759 | if (left) { | |
6760 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6761 | left * sizeof(struct file *)); | |
6762 | } | |
6763 | fp->count--; | |
6764 | if (!fp->count) { | |
6765 | kfree_skb(skb); | |
6766 | skb = NULL; | |
6767 | } else { | |
6768 | __skb_queue_tail(&list, skb); | |
6769 | } | |
6770 | fput(file); | |
6771 | file = NULL; | |
6772 | break; | |
6773 | } | |
6774 | ||
6775 | if (!file) | |
6776 | break; | |
6777 | ||
6778 | __skb_queue_tail(&list, skb); | |
6779 | ||
6780 | skb = skb_dequeue(head); | |
6781 | } | |
6782 | ||
6783 | if (skb_peek(&list)) { | |
6784 | spin_lock_irq(&head->lock); | |
6785 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6786 | __skb_queue_tail(head, skb); | |
6787 | spin_unlock_irq(&head->lock); | |
6788 | } | |
6789 | #else | |
6790 | fput(file); | |
6791 | #endif | |
6792 | } | |
6793 | ||
6794 | struct io_file_put { | |
05589553 | 6795 | struct list_head list; |
05f3fb3c | 6796 | struct file *file; |
05f3fb3c JA |
6797 | }; |
6798 | ||
4a38aed2 | 6799 | static void __io_file_put_work(struct fixed_file_ref_node *ref_node) |
65e19f54 | 6800 | { |
4a38aed2 JA |
6801 | struct fixed_file_data *file_data = ref_node->file_data; |
6802 | struct io_ring_ctx *ctx = file_data->ctx; | |
05f3fb3c | 6803 | struct io_file_put *pfile, *tmp; |
05589553 XW |
6804 | |
6805 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6a4d07cd | 6806 | list_del(&pfile->list); |
05589553 XW |
6807 | io_ring_file_put(ctx, pfile->file); |
6808 | kfree(pfile); | |
65e19f54 | 6809 | } |
05589553 | 6810 | |
6a4d07cd JA |
6811 | spin_lock(&file_data->lock); |
6812 | list_del(&ref_node->node); | |
6813 | spin_unlock(&file_data->lock); | |
05589553 XW |
6814 | |
6815 | percpu_ref_exit(&ref_node->refs); | |
6816 | kfree(ref_node); | |
6817 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6818 | } |
65e19f54 | 6819 | |
4a38aed2 JA |
6820 | static void io_file_put_work(struct work_struct *work) |
6821 | { | |
6822 | struct io_ring_ctx *ctx; | |
6823 | struct llist_node *node; | |
6824 | ||
6825 | ctx = container_of(work, struct io_ring_ctx, file_put_work.work); | |
6826 | node = llist_del_all(&ctx->file_put_llist); | |
6827 | ||
6828 | while (node) { | |
6829 | struct fixed_file_ref_node *ref_node; | |
6830 | struct llist_node *next = node->next; | |
6831 | ||
6832 | ref_node = llist_entry(node, struct fixed_file_ref_node, llist); | |
6833 | __io_file_put_work(ref_node); | |
6834 | node = next; | |
6835 | } | |
6836 | } | |
6837 | ||
05589553 | 6838 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6839 | { |
05589553 | 6840 | struct fixed_file_ref_node *ref_node; |
4a38aed2 JA |
6841 | struct io_ring_ctx *ctx; |
6842 | bool first_add; | |
6843 | int delay = HZ; | |
65e19f54 | 6844 | |
05589553 | 6845 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
4a38aed2 | 6846 | ctx = ref_node->file_data->ctx; |
05589553 | 6847 | |
4a38aed2 JA |
6848 | if (percpu_ref_is_dying(&ctx->file_data->refs)) |
6849 | delay = 0; | |
05589553 | 6850 | |
4a38aed2 JA |
6851 | first_add = llist_add(&ref_node->llist, &ctx->file_put_llist); |
6852 | if (!delay) | |
6853 | mod_delayed_work(system_wq, &ctx->file_put_work, 0); | |
6854 | else if (first_add) | |
6855 | queue_delayed_work(system_wq, &ctx->file_put_work, delay); | |
05f3fb3c | 6856 | } |
65e19f54 | 6857 | |
05589553 XW |
6858 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6859 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6860 | { |
05589553 | 6861 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6862 | |
05589553 XW |
6863 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6864 | if (!ref_node) | |
6865 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6866 | |
05589553 XW |
6867 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6868 | 0, GFP_KERNEL)) { | |
6869 | kfree(ref_node); | |
6870 | return ERR_PTR(-ENOMEM); | |
6871 | } | |
6872 | INIT_LIST_HEAD(&ref_node->node); | |
6873 | INIT_LIST_HEAD(&ref_node->file_list); | |
05589553 XW |
6874 | ref_node->file_data = ctx->file_data; |
6875 | return ref_node; | |
05589553 XW |
6876 | } |
6877 | ||
6878 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6879 | { | |
6880 | percpu_ref_exit(&ref_node->refs); | |
6881 | kfree(ref_node); | |
65e19f54 JA |
6882 | } |
6883 | ||
6b06314c JA |
6884 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6885 | unsigned nr_args) | |
6886 | { | |
6887 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6888 | unsigned nr_tables; |
05f3fb3c | 6889 | struct file *file; |
6b06314c JA |
6890 | int fd, ret = 0; |
6891 | unsigned i; | |
05589553 | 6892 | struct fixed_file_ref_node *ref_node; |
6b06314c | 6893 | |
05f3fb3c | 6894 | if (ctx->file_data) |
6b06314c JA |
6895 | return -EBUSY; |
6896 | if (!nr_args) | |
6897 | return -EINVAL; | |
6898 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6899 | return -EMFILE; | |
6900 | ||
05f3fb3c JA |
6901 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6902 | if (!ctx->file_data) | |
6903 | return -ENOMEM; | |
6904 | ctx->file_data->ctx = ctx; | |
6905 | init_completion(&ctx->file_data->done); | |
05589553 | 6906 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6907 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6908 | |
65e19f54 | 6909 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6910 | ctx->file_data->table = kcalloc(nr_tables, |
6911 | sizeof(struct fixed_file_table), | |
65e19f54 | 6912 | GFP_KERNEL); |
05f3fb3c JA |
6913 | if (!ctx->file_data->table) { |
6914 | kfree(ctx->file_data); | |
6915 | ctx->file_data = NULL; | |
6b06314c | 6916 | return -ENOMEM; |
05f3fb3c JA |
6917 | } |
6918 | ||
05589553 | 6919 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6920 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6921 | kfree(ctx->file_data->table); | |
6922 | kfree(ctx->file_data); | |
6923 | ctx->file_data = NULL; | |
6b06314c | 6924 | return -ENOMEM; |
05f3fb3c | 6925 | } |
6b06314c | 6926 | |
65e19f54 | 6927 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6928 | percpu_ref_exit(&ctx->file_data->refs); |
6929 | kfree(ctx->file_data->table); | |
6930 | kfree(ctx->file_data); | |
6931 | ctx->file_data = NULL; | |
65e19f54 JA |
6932 | return -ENOMEM; |
6933 | } | |
6934 | ||
08a45173 | 6935 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6936 | struct fixed_file_table *table; |
6937 | unsigned index; | |
6938 | ||
6b06314c JA |
6939 | ret = -EFAULT; |
6940 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6941 | break; | |
08a45173 JA |
6942 | /* allow sparse sets */ |
6943 | if (fd == -1) { | |
6944 | ret = 0; | |
6945 | continue; | |
6946 | } | |
6b06314c | 6947 | |
05f3fb3c | 6948 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6949 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6950 | file = fget(fd); |
6b06314c JA |
6951 | |
6952 | ret = -EBADF; | |
05f3fb3c | 6953 | if (!file) |
6b06314c | 6954 | break; |
05f3fb3c | 6955 | |
6b06314c JA |
6956 | /* |
6957 | * Don't allow io_uring instances to be registered. If UNIX | |
6958 | * isn't enabled, then this causes a reference cycle and this | |
6959 | * instance can never get freed. If UNIX is enabled we'll | |
6960 | * handle it just fine, but there's still no point in allowing | |
6961 | * a ring fd as it doesn't support regular read/write anyway. | |
6962 | */ | |
05f3fb3c JA |
6963 | if (file->f_op == &io_uring_fops) { |
6964 | fput(file); | |
6b06314c JA |
6965 | break; |
6966 | } | |
6b06314c | 6967 | ret = 0; |
05f3fb3c | 6968 | table->files[index] = file; |
6b06314c JA |
6969 | } |
6970 | ||
6971 | if (ret) { | |
65e19f54 | 6972 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6973 | file = io_file_from_index(ctx, i); |
6974 | if (file) | |
6975 | fput(file); | |
6976 | } | |
6977 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6978 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6979 | |
05f3fb3c JA |
6980 | kfree(ctx->file_data->table); |
6981 | kfree(ctx->file_data); | |
6982 | ctx->file_data = NULL; | |
6b06314c JA |
6983 | ctx->nr_user_files = 0; |
6984 | return ret; | |
6985 | } | |
6986 | ||
6987 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6988 | if (ret) { |
6b06314c | 6989 | io_sqe_files_unregister(ctx); |
05589553 XW |
6990 | return ret; |
6991 | } | |
6b06314c | 6992 | |
05589553 XW |
6993 | ref_node = alloc_fixed_file_ref_node(ctx); |
6994 | if (IS_ERR(ref_node)) { | |
6995 | io_sqe_files_unregister(ctx); | |
6996 | return PTR_ERR(ref_node); | |
6997 | } | |
6998 | ||
6999 | ctx->file_data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7000 | spin_lock(&ctx->file_data->lock); |
05589553 | 7001 | list_add(&ref_node->node, &ctx->file_data->ref_list); |
6a4d07cd | 7002 | spin_unlock(&ctx->file_data->lock); |
05589553 | 7003 | percpu_ref_get(&ctx->file_data->refs); |
6b06314c JA |
7004 | return ret; |
7005 | } | |
7006 | ||
c3a31e60 JA |
7007 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7008 | int index) | |
7009 | { | |
7010 | #if defined(CONFIG_UNIX) | |
7011 | struct sock *sock = ctx->ring_sock->sk; | |
7012 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7013 | struct sk_buff *skb; | |
7014 | ||
7015 | /* | |
7016 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7017 | * file set. If there's no room, fall back to allocating a new skb | |
7018 | * and filling it in. | |
7019 | */ | |
7020 | spin_lock_irq(&head->lock); | |
7021 | skb = skb_peek(head); | |
7022 | if (skb) { | |
7023 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7024 | ||
7025 | if (fpl->count < SCM_MAX_FD) { | |
7026 | __skb_unlink(skb, head); | |
7027 | spin_unlock_irq(&head->lock); | |
7028 | fpl->fp[fpl->count] = get_file(file); | |
7029 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7030 | fpl->count++; | |
7031 | spin_lock_irq(&head->lock); | |
7032 | __skb_queue_head(head, skb); | |
7033 | } else { | |
7034 | skb = NULL; | |
7035 | } | |
7036 | } | |
7037 | spin_unlock_irq(&head->lock); | |
7038 | ||
7039 | if (skb) { | |
7040 | fput(file); | |
7041 | return 0; | |
7042 | } | |
7043 | ||
7044 | return __io_sqe_files_scm(ctx, 1, index); | |
7045 | #else | |
7046 | return 0; | |
7047 | #endif | |
7048 | } | |
7049 | ||
a5318d3c | 7050 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 7051 | struct file *file) |
05f3fb3c | 7052 | { |
a5318d3c | 7053 | struct io_file_put *pfile; |
05589553 XW |
7054 | struct percpu_ref *refs = data->cur_refs; |
7055 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 7056 | |
05f3fb3c | 7057 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
7058 | if (!pfile) |
7059 | return -ENOMEM; | |
05f3fb3c | 7060 | |
05589553 | 7061 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 7062 | pfile->file = file; |
05589553 XW |
7063 | list_add(&pfile->list, &ref_node->file_list); |
7064 | ||
a5318d3c | 7065 | return 0; |
05f3fb3c JA |
7066 | } |
7067 | ||
7068 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
7069 | struct io_uring_files_update *up, | |
7070 | unsigned nr_args) | |
7071 | { | |
7072 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 7073 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 7074 | struct file *file; |
c3a31e60 JA |
7075 | __s32 __user *fds; |
7076 | int fd, i, err; | |
7077 | __u32 done; | |
05589553 | 7078 | bool needs_switch = false; |
c3a31e60 | 7079 | |
05f3fb3c | 7080 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7081 | return -EOVERFLOW; |
7082 | if (done > ctx->nr_user_files) | |
7083 | return -EINVAL; | |
7084 | ||
05589553 XW |
7085 | ref_node = alloc_fixed_file_ref_node(ctx); |
7086 | if (IS_ERR(ref_node)) | |
7087 | return PTR_ERR(ref_node); | |
7088 | ||
c3a31e60 | 7089 | done = 0; |
05f3fb3c | 7090 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 7091 | while (nr_args) { |
65e19f54 JA |
7092 | struct fixed_file_table *table; |
7093 | unsigned index; | |
7094 | ||
c3a31e60 JA |
7095 | err = 0; |
7096 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7097 | err = -EFAULT; | |
7098 | break; | |
7099 | } | |
05f3fb3c JA |
7100 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
7101 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
7102 | index = i & IORING_FILE_TABLE_MASK; |
7103 | if (table->files[index]) { | |
05f3fb3c | 7104 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
7105 | err = io_queue_file_removal(data, file); |
7106 | if (err) | |
7107 | break; | |
65e19f54 | 7108 | table->files[index] = NULL; |
05589553 | 7109 | needs_switch = true; |
c3a31e60 JA |
7110 | } |
7111 | if (fd != -1) { | |
c3a31e60 JA |
7112 | file = fget(fd); |
7113 | if (!file) { | |
7114 | err = -EBADF; | |
7115 | break; | |
7116 | } | |
7117 | /* | |
7118 | * Don't allow io_uring instances to be registered. If | |
7119 | * UNIX isn't enabled, then this causes a reference | |
7120 | * cycle and this instance can never get freed. If UNIX | |
7121 | * is enabled we'll handle it just fine, but there's | |
7122 | * still no point in allowing a ring fd as it doesn't | |
7123 | * support regular read/write anyway. | |
7124 | */ | |
7125 | if (file->f_op == &io_uring_fops) { | |
7126 | fput(file); | |
7127 | err = -EBADF; | |
7128 | break; | |
7129 | } | |
65e19f54 | 7130 | table->files[index] = file; |
c3a31e60 JA |
7131 | err = io_sqe_file_register(ctx, file, i); |
7132 | if (err) | |
7133 | break; | |
7134 | } | |
7135 | nr_args--; | |
7136 | done++; | |
05f3fb3c JA |
7137 | up->offset++; |
7138 | } | |
7139 | ||
05589553 XW |
7140 | if (needs_switch) { |
7141 | percpu_ref_kill(data->cur_refs); | |
6a4d07cd | 7142 | spin_lock(&data->lock); |
05589553 XW |
7143 | list_add(&ref_node->node, &data->ref_list); |
7144 | data->cur_refs = &ref_node->refs; | |
6a4d07cd | 7145 | spin_unlock(&data->lock); |
05589553 XW |
7146 | percpu_ref_get(&ctx->file_data->refs); |
7147 | } else | |
7148 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
7149 | |
7150 | return done ? done : err; | |
7151 | } | |
05589553 | 7152 | |
05f3fb3c JA |
7153 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7154 | unsigned nr_args) | |
7155 | { | |
7156 | struct io_uring_files_update up; | |
7157 | ||
7158 | if (!ctx->file_data) | |
7159 | return -ENXIO; | |
7160 | if (!nr_args) | |
7161 | return -EINVAL; | |
7162 | if (copy_from_user(&up, arg, sizeof(up))) | |
7163 | return -EFAULT; | |
7164 | if (up.resv) | |
7165 | return -EINVAL; | |
7166 | ||
7167 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7168 | } | |
c3a31e60 | 7169 | |
e9fd9396 | 7170 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
7171 | { |
7172 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7173 | ||
e9fd9396 | 7174 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
7175 | io_put_req(req); |
7176 | } | |
7177 | ||
24369c2e PB |
7178 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
7179 | struct io_uring_params *p) | |
7180 | { | |
7181 | struct io_wq_data data; | |
7182 | struct fd f; | |
7183 | struct io_ring_ctx *ctx_attach; | |
7184 | unsigned int concurrency; | |
7185 | int ret = 0; | |
7186 | ||
7187 | data.user = ctx->user; | |
e9fd9396 | 7188 | data.free_work = io_free_work; |
f5fa38c5 | 7189 | data.do_work = io_wq_submit_work; |
24369c2e PB |
7190 | |
7191 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
7192 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
7193 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
7194 | ||
7195 | ctx->io_wq = io_wq_create(concurrency, &data); | |
7196 | if (IS_ERR(ctx->io_wq)) { | |
7197 | ret = PTR_ERR(ctx->io_wq); | |
7198 | ctx->io_wq = NULL; | |
7199 | } | |
7200 | return ret; | |
7201 | } | |
7202 | ||
7203 | f = fdget(p->wq_fd); | |
7204 | if (!f.file) | |
7205 | return -EBADF; | |
7206 | ||
7207 | if (f.file->f_op != &io_uring_fops) { | |
7208 | ret = -EINVAL; | |
7209 | goto out_fput; | |
7210 | } | |
7211 | ||
7212 | ctx_attach = f.file->private_data; | |
7213 | /* @io_wq is protected by holding the fd */ | |
7214 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
7215 | ret = -EINVAL; | |
7216 | goto out_fput; | |
7217 | } | |
7218 | ||
7219 | ctx->io_wq = ctx_attach->io_wq; | |
7220 | out_fput: | |
7221 | fdput(f); | |
7222 | return ret; | |
7223 | } | |
7224 | ||
6c271ce2 JA |
7225 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
7226 | struct io_uring_params *p) | |
2b188cc1 JA |
7227 | { |
7228 | int ret; | |
7229 | ||
6c271ce2 | 7230 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
8eb06d7e PB |
7231 | mmgrab(current->mm); |
7232 | ctx->sqo_mm = current->mm; | |
7233 | ||
3ec482d1 JA |
7234 | ret = -EPERM; |
7235 | if (!capable(CAP_SYS_ADMIN)) | |
7236 | goto err; | |
7237 | ||
917257da JA |
7238 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7239 | if (!ctx->sq_thread_idle) | |
7240 | ctx->sq_thread_idle = HZ; | |
7241 | ||
6c271ce2 | 7242 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7243 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7244 | |
917257da | 7245 | ret = -EINVAL; |
44a9bd18 JA |
7246 | if (cpu >= nr_cpu_ids) |
7247 | goto err; | |
7889f44d | 7248 | if (!cpu_online(cpu)) |
917257da JA |
7249 | goto err; |
7250 | ||
6c271ce2 JA |
7251 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
7252 | ctx, cpu, | |
7253 | "io_uring-sq"); | |
7254 | } else { | |
7255 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
7256 | "io_uring-sq"); | |
7257 | } | |
7258 | if (IS_ERR(ctx->sqo_thread)) { | |
7259 | ret = PTR_ERR(ctx->sqo_thread); | |
7260 | ctx->sqo_thread = NULL; | |
7261 | goto err; | |
7262 | } | |
7263 | wake_up_process(ctx->sqo_thread); | |
7264 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
7265 | /* Can't have SQ_AFF without SQPOLL */ | |
7266 | ret = -EINVAL; | |
7267 | goto err; | |
7268 | } | |
7269 | ||
24369c2e PB |
7270 | ret = io_init_wq_offload(ctx, p); |
7271 | if (ret) | |
2b188cc1 | 7272 | goto err; |
2b188cc1 JA |
7273 | |
7274 | return 0; | |
7275 | err: | |
54a91f3b | 7276 | io_finish_async(ctx); |
8eb06d7e PB |
7277 | if (ctx->sqo_mm) { |
7278 | mmdrop(ctx->sqo_mm); | |
7279 | ctx->sqo_mm = NULL; | |
7280 | } | |
2b188cc1 JA |
7281 | return ret; |
7282 | } | |
7283 | ||
a087e2b5 BM |
7284 | static inline void __io_unaccount_mem(struct user_struct *user, |
7285 | unsigned long nr_pages) | |
2b188cc1 JA |
7286 | { |
7287 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7288 | } | |
7289 | ||
a087e2b5 BM |
7290 | static inline int __io_account_mem(struct user_struct *user, |
7291 | unsigned long nr_pages) | |
2b188cc1 JA |
7292 | { |
7293 | unsigned long page_limit, cur_pages, new_pages; | |
7294 | ||
7295 | /* Don't allow more pages than we can safely lock */ | |
7296 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7297 | ||
7298 | do { | |
7299 | cur_pages = atomic_long_read(&user->locked_vm); | |
7300 | new_pages = cur_pages + nr_pages; | |
7301 | if (new_pages > page_limit) | |
7302 | return -ENOMEM; | |
7303 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7304 | new_pages) != cur_pages); | |
7305 | ||
7306 | return 0; | |
7307 | } | |
7308 | ||
2e0464d4 BM |
7309 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7310 | enum io_mem_account acct) | |
a087e2b5 | 7311 | { |
aad5d8da | 7312 | if (ctx->limit_mem) |
a087e2b5 | 7313 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7314 | |
2e0464d4 BM |
7315 | if (ctx->sqo_mm) { |
7316 | if (acct == ACCT_LOCKED) | |
7317 | ctx->sqo_mm->locked_vm -= nr_pages; | |
7318 | else if (acct == ACCT_PINNED) | |
7319 | atomic64_sub(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7320 | } | |
a087e2b5 BM |
7321 | } |
7322 | ||
2e0464d4 BM |
7323 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages, |
7324 | enum io_mem_account acct) | |
a087e2b5 | 7325 | { |
30975825 BM |
7326 | int ret; |
7327 | ||
7328 | if (ctx->limit_mem) { | |
7329 | ret = __io_account_mem(ctx->user, nr_pages); | |
7330 | if (ret) | |
7331 | return ret; | |
7332 | } | |
7333 | ||
2e0464d4 BM |
7334 | if (ctx->sqo_mm) { |
7335 | if (acct == ACCT_LOCKED) | |
7336 | ctx->sqo_mm->locked_vm += nr_pages; | |
7337 | else if (acct == ACCT_PINNED) | |
7338 | atomic64_add(nr_pages, &ctx->sqo_mm->pinned_vm); | |
7339 | } | |
a087e2b5 BM |
7340 | |
7341 | return 0; | |
7342 | } | |
7343 | ||
2b188cc1 JA |
7344 | static void io_mem_free(void *ptr) |
7345 | { | |
52e04ef4 MR |
7346 | struct page *page; |
7347 | ||
7348 | if (!ptr) | |
7349 | return; | |
2b188cc1 | 7350 | |
52e04ef4 | 7351 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7352 | if (put_page_testzero(page)) |
7353 | free_compound_page(page); | |
7354 | } | |
7355 | ||
7356 | static void *io_mem_alloc(size_t size) | |
7357 | { | |
7358 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
7359 | __GFP_NORETRY; | |
7360 | ||
7361 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7362 | } | |
7363 | ||
75b28aff HV |
7364 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7365 | size_t *sq_offset) | |
7366 | { | |
7367 | struct io_rings *rings; | |
7368 | size_t off, sq_array_size; | |
7369 | ||
7370 | off = struct_size(rings, cqes, cq_entries); | |
7371 | if (off == SIZE_MAX) | |
7372 | return SIZE_MAX; | |
7373 | ||
7374 | #ifdef CONFIG_SMP | |
7375 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7376 | if (off == 0) | |
7377 | return SIZE_MAX; | |
7378 | #endif | |
7379 | ||
7380 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
7381 | if (sq_array_size == SIZE_MAX) | |
7382 | return SIZE_MAX; | |
7383 | ||
7384 | if (check_add_overflow(off, sq_array_size, &off)) | |
7385 | return SIZE_MAX; | |
7386 | ||
7387 | if (sq_offset) | |
7388 | *sq_offset = off; | |
7389 | ||
7390 | return off; | |
7391 | } | |
7392 | ||
2b188cc1 JA |
7393 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
7394 | { | |
75b28aff | 7395 | size_t pages; |
2b188cc1 | 7396 | |
75b28aff HV |
7397 | pages = (size_t)1 << get_order( |
7398 | rings_size(sq_entries, cq_entries, NULL)); | |
7399 | pages += (size_t)1 << get_order( | |
7400 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7401 | |
75b28aff | 7402 | return pages; |
2b188cc1 JA |
7403 | } |
7404 | ||
edafccee JA |
7405 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7406 | { | |
7407 | int i, j; | |
7408 | ||
7409 | if (!ctx->user_bufs) | |
7410 | return -ENXIO; | |
7411 | ||
7412 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7413 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7414 | ||
7415 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7416 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 7417 | |
2e0464d4 | 7418 | io_unaccount_mem(ctx, imu->nr_bvecs, ACCT_PINNED); |
d4ef6475 | 7419 | kvfree(imu->bvec); |
edafccee JA |
7420 | imu->nr_bvecs = 0; |
7421 | } | |
7422 | ||
7423 | kfree(ctx->user_bufs); | |
7424 | ctx->user_bufs = NULL; | |
7425 | ctx->nr_user_bufs = 0; | |
7426 | return 0; | |
7427 | } | |
7428 | ||
7429 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7430 | void __user *arg, unsigned index) | |
7431 | { | |
7432 | struct iovec __user *src; | |
7433 | ||
7434 | #ifdef CONFIG_COMPAT | |
7435 | if (ctx->compat) { | |
7436 | struct compat_iovec __user *ciovs; | |
7437 | struct compat_iovec ciov; | |
7438 | ||
7439 | ciovs = (struct compat_iovec __user *) arg; | |
7440 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7441 | return -EFAULT; | |
7442 | ||
d55e5f5b | 7443 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7444 | dst->iov_len = ciov.iov_len; |
7445 | return 0; | |
7446 | } | |
7447 | #endif | |
7448 | src = (struct iovec __user *) arg; | |
7449 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7450 | return -EFAULT; | |
7451 | return 0; | |
7452 | } | |
7453 | ||
7454 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7455 | unsigned nr_args) | |
7456 | { | |
7457 | struct vm_area_struct **vmas = NULL; | |
7458 | struct page **pages = NULL; | |
7459 | int i, j, got_pages = 0; | |
7460 | int ret = -EINVAL; | |
7461 | ||
7462 | if (ctx->user_bufs) | |
7463 | return -EBUSY; | |
7464 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7465 | return -EINVAL; | |
7466 | ||
7467 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7468 | GFP_KERNEL); | |
7469 | if (!ctx->user_bufs) | |
7470 | return -ENOMEM; | |
7471 | ||
7472 | for (i = 0; i < nr_args; i++) { | |
7473 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7474 | unsigned long off, start, end, ubuf; | |
7475 | int pret, nr_pages; | |
7476 | struct iovec iov; | |
7477 | size_t size; | |
7478 | ||
7479 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7480 | if (ret) | |
a278682d | 7481 | goto err; |
edafccee JA |
7482 | |
7483 | /* | |
7484 | * Don't impose further limits on the size and buffer | |
7485 | * constraints here, we'll -EINVAL later when IO is | |
7486 | * submitted if they are wrong. | |
7487 | */ | |
7488 | ret = -EFAULT; | |
7489 | if (!iov.iov_base || !iov.iov_len) | |
7490 | goto err; | |
7491 | ||
7492 | /* arbitrary limit, but we need something */ | |
7493 | if (iov.iov_len > SZ_1G) | |
7494 | goto err; | |
7495 | ||
7496 | ubuf = (unsigned long) iov.iov_base; | |
7497 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7498 | start = ubuf >> PAGE_SHIFT; | |
7499 | nr_pages = end - start; | |
7500 | ||
2e0464d4 | 7501 | ret = io_account_mem(ctx, nr_pages, ACCT_PINNED); |
a087e2b5 BM |
7502 | if (ret) |
7503 | goto err; | |
edafccee JA |
7504 | |
7505 | ret = 0; | |
7506 | if (!pages || nr_pages > got_pages) { | |
a8c73c1a DE |
7507 | kvfree(vmas); |
7508 | kvfree(pages); | |
d4ef6475 | 7509 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7510 | GFP_KERNEL); |
d4ef6475 | 7511 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7512 | sizeof(struct vm_area_struct *), |
7513 | GFP_KERNEL); | |
7514 | if (!pages || !vmas) { | |
7515 | ret = -ENOMEM; | |
2e0464d4 | 7516 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7517 | goto err; |
7518 | } | |
7519 | got_pages = nr_pages; | |
7520 | } | |
7521 | ||
d4ef6475 | 7522 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7523 | GFP_KERNEL); |
7524 | ret = -ENOMEM; | |
7525 | if (!imu->bvec) { | |
2e0464d4 | 7526 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
edafccee JA |
7527 | goto err; |
7528 | } | |
7529 | ||
7530 | ret = 0; | |
d8ed45c5 | 7531 | mmap_read_lock(current->mm); |
2113b05d | 7532 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7533 | FOLL_WRITE | FOLL_LONGTERM, |
7534 | pages, vmas); | |
edafccee JA |
7535 | if (pret == nr_pages) { |
7536 | /* don't support file backed memory */ | |
7537 | for (j = 0; j < nr_pages; j++) { | |
7538 | struct vm_area_struct *vma = vmas[j]; | |
7539 | ||
7540 | if (vma->vm_file && | |
7541 | !is_file_hugepages(vma->vm_file)) { | |
7542 | ret = -EOPNOTSUPP; | |
7543 | break; | |
7544 | } | |
7545 | } | |
7546 | } else { | |
7547 | ret = pret < 0 ? pret : -EFAULT; | |
7548 | } | |
d8ed45c5 | 7549 | mmap_read_unlock(current->mm); |
edafccee JA |
7550 | if (ret) { |
7551 | /* | |
7552 | * if we did partial map, or found file backed vmas, | |
7553 | * release any pages we did get | |
7554 | */ | |
27c4d3a3 | 7555 | if (pret > 0) |
f1f6a7dd | 7556 | unpin_user_pages(pages, pret); |
2e0464d4 | 7557 | io_unaccount_mem(ctx, nr_pages, ACCT_PINNED); |
d4ef6475 | 7558 | kvfree(imu->bvec); |
edafccee JA |
7559 | goto err; |
7560 | } | |
7561 | ||
7562 | off = ubuf & ~PAGE_MASK; | |
7563 | size = iov.iov_len; | |
7564 | for (j = 0; j < nr_pages; j++) { | |
7565 | size_t vec_len; | |
7566 | ||
7567 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7568 | imu->bvec[j].bv_page = pages[j]; | |
7569 | imu->bvec[j].bv_len = vec_len; | |
7570 | imu->bvec[j].bv_offset = off; | |
7571 | off = 0; | |
7572 | size -= vec_len; | |
7573 | } | |
7574 | /* store original address for later verification */ | |
7575 | imu->ubuf = ubuf; | |
7576 | imu->len = iov.iov_len; | |
7577 | imu->nr_bvecs = nr_pages; | |
7578 | ||
7579 | ctx->nr_user_bufs++; | |
7580 | } | |
d4ef6475 MR |
7581 | kvfree(pages); |
7582 | kvfree(vmas); | |
edafccee JA |
7583 | return 0; |
7584 | err: | |
d4ef6475 MR |
7585 | kvfree(pages); |
7586 | kvfree(vmas); | |
edafccee JA |
7587 | io_sqe_buffer_unregister(ctx); |
7588 | return ret; | |
7589 | } | |
7590 | ||
9b402849 JA |
7591 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7592 | { | |
7593 | __s32 __user *fds = arg; | |
7594 | int fd; | |
7595 | ||
7596 | if (ctx->cq_ev_fd) | |
7597 | return -EBUSY; | |
7598 | ||
7599 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7600 | return -EFAULT; | |
7601 | ||
7602 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7603 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7604 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7605 | ctx->cq_ev_fd = NULL; | |
7606 | return ret; | |
7607 | } | |
7608 | ||
7609 | return 0; | |
7610 | } | |
7611 | ||
7612 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7613 | { | |
7614 | if (ctx->cq_ev_fd) { | |
7615 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7616 | ctx->cq_ev_fd = NULL; | |
7617 | return 0; | |
7618 | } | |
7619 | ||
7620 | return -ENXIO; | |
7621 | } | |
7622 | ||
5a2e745d JA |
7623 | static int __io_destroy_buffers(int id, void *p, void *data) |
7624 | { | |
7625 | struct io_ring_ctx *ctx = data; | |
7626 | struct io_buffer *buf = p; | |
7627 | ||
067524e9 | 7628 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7629 | return 0; |
7630 | } | |
7631 | ||
7632 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7633 | { | |
7634 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7635 | idr_destroy(&ctx->io_buffer_idr); | |
7636 | } | |
7637 | ||
2b188cc1 JA |
7638 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7639 | { | |
6b06314c | 7640 | io_finish_async(ctx); |
30975825 | 7641 | if (ctx->sqo_mm) { |
2b188cc1 | 7642 | mmdrop(ctx->sqo_mm); |
30975825 BM |
7643 | ctx->sqo_mm = NULL; |
7644 | } | |
def596e9 JA |
7645 | |
7646 | io_iopoll_reap_events(ctx); | |
edafccee | 7647 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7648 | io_sqe_files_unregister(ctx); |
9b402849 | 7649 | io_eventfd_unregister(ctx); |
5a2e745d | 7650 | io_destroy_buffers(ctx); |
41726c9a | 7651 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7652 | |
2b188cc1 | 7653 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7654 | if (ctx->ring_sock) { |
7655 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7656 | sock_release(ctx->ring_sock); |
355e8d26 | 7657 | } |
2b188cc1 JA |
7658 | #endif |
7659 | ||
75b28aff | 7660 | io_mem_free(ctx->rings); |
2b188cc1 | 7661 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7662 | |
7663 | percpu_ref_exit(&ctx->refs); | |
2e0464d4 BM |
7664 | io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries), |
7665 | ACCT_LOCKED); | |
2b188cc1 | 7666 | free_uid(ctx->user); |
181e448d | 7667 | put_cred(ctx->creds); |
78076bb6 | 7668 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7669 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7670 | kfree(ctx); |
7671 | } | |
7672 | ||
7673 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7674 | { | |
7675 | struct io_ring_ctx *ctx = file->private_data; | |
7676 | __poll_t mask = 0; | |
7677 | ||
7678 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7679 | /* |
7680 | * synchronizes with barrier from wq_has_sleeper call in | |
7681 | * io_commit_cqring | |
7682 | */ | |
2b188cc1 | 7683 | smp_rmb(); |
75b28aff HV |
7684 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7685 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7686 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7687 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7688 | mask |= EPOLLIN | EPOLLRDNORM; |
7689 | ||
7690 | return mask; | |
7691 | } | |
7692 | ||
7693 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7694 | { | |
7695 | struct io_ring_ctx *ctx = file->private_data; | |
7696 | ||
7697 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7698 | } | |
7699 | ||
071698e1 JA |
7700 | static int io_remove_personalities(int id, void *p, void *data) |
7701 | { | |
7702 | struct io_ring_ctx *ctx = data; | |
7703 | const struct cred *cred; | |
7704 | ||
7705 | cred = idr_remove(&ctx->personality_idr, id); | |
7706 | if (cred) | |
7707 | put_cred(cred); | |
7708 | return 0; | |
7709 | } | |
7710 | ||
85faa7b8 JA |
7711 | static void io_ring_exit_work(struct work_struct *work) |
7712 | { | |
7713 | struct io_ring_ctx *ctx; | |
7714 | ||
7715 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7716 | if (ctx->rings) | |
7717 | io_cqring_overflow_flush(ctx, true); | |
7718 | ||
56952e91 JA |
7719 | /* |
7720 | * If we're doing polled IO and end up having requests being | |
7721 | * submitted async (out-of-line), then completions can come in while | |
7722 | * we're waiting for refs to drop. We need to reap these manually, | |
7723 | * as nobody else will be looking for them. | |
7724 | */ | |
7725 | while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)) { | |
7726 | io_iopoll_reap_events(ctx); | |
7727 | if (ctx->rings) | |
7728 | io_cqring_overflow_flush(ctx, true); | |
7729 | } | |
85faa7b8 JA |
7730 | io_ring_ctx_free(ctx); |
7731 | } | |
7732 | ||
2b188cc1 JA |
7733 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7734 | { | |
7735 | mutex_lock(&ctx->uring_lock); | |
7736 | percpu_ref_kill(&ctx->refs); | |
7737 | mutex_unlock(&ctx->uring_lock); | |
7738 | ||
5262f567 | 7739 | io_kill_timeouts(ctx); |
221c5eb2 | 7740 | io_poll_remove_all(ctx); |
561fb04a JA |
7741 | |
7742 | if (ctx->io_wq) | |
7743 | io_wq_cancel_all(ctx->io_wq); | |
7744 | ||
def596e9 | 7745 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7746 | /* if we failed setting up the ctx, we might not have any rings */ |
7747 | if (ctx->rings) | |
7748 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7749 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7750 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7751 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7752 | } |
7753 | ||
7754 | static int io_uring_release(struct inode *inode, struct file *file) | |
7755 | { | |
7756 | struct io_ring_ctx *ctx = file->private_data; | |
7757 | ||
7758 | file->private_data = NULL; | |
7759 | io_ring_ctx_wait_and_kill(ctx); | |
7760 | return 0; | |
7761 | } | |
7762 | ||
67c4d9e6 PB |
7763 | static bool io_wq_files_match(struct io_wq_work *work, void *data) |
7764 | { | |
7765 | struct files_struct *files = data; | |
7766 | ||
7767 | return work->files == files; | |
7768 | } | |
7769 | ||
fcb323cc JA |
7770 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7771 | struct files_struct *files) | |
7772 | { | |
67c4d9e6 PB |
7773 | if (list_empty_careful(&ctx->inflight_list)) |
7774 | return; | |
7775 | ||
7776 | /* cancel all at once, should be faster than doing it one by one*/ | |
7777 | io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true); | |
7778 | ||
fcb323cc | 7779 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 XW |
7780 | struct io_kiocb *cancel_req = NULL, *req; |
7781 | DEFINE_WAIT(wait); | |
fcb323cc JA |
7782 | |
7783 | spin_lock_irq(&ctx->inflight_lock); | |
7784 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7785 | if (req->work.files != files) |
7786 | continue; | |
7787 | /* req is being completed, ignore */ | |
7788 | if (!refcount_inc_not_zero(&req->refs)) | |
7789 | continue; | |
7790 | cancel_req = req; | |
7791 | break; | |
fcb323cc | 7792 | } |
768134d4 | 7793 | if (cancel_req) |
fcb323cc | 7794 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7795 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7796 | spin_unlock_irq(&ctx->inflight_lock); |
7797 | ||
768134d4 JA |
7798 | /* We need to keep going until we don't find a matching req */ |
7799 | if (!cancel_req) | |
fcb323cc | 7800 | break; |
2f6d9b9d | 7801 | |
2ca10259 JA |
7802 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7803 | spin_lock_irq(&ctx->completion_lock); | |
7804 | list_del(&cancel_req->list); | |
7805 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7806 | if (list_empty(&ctx->cq_overflow_list)) { | |
7807 | clear_bit(0, &ctx->sq_check_overflow); | |
7808 | clear_bit(0, &ctx->cq_check_overflow); | |
7809 | } | |
7810 | spin_unlock_irq(&ctx->completion_lock); | |
7811 | ||
7812 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7813 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7814 | ||
7815 | /* | |
7816 | * Put inflight ref and overflow ref. If that's | |
7817 | * all we had, then we're done with this request. | |
7818 | */ | |
7819 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
4518a3cc | 7820 | io_free_req(cancel_req); |
d8f1b971 | 7821 | finish_wait(&ctx->inflight_wait, &wait); |
2ca10259 JA |
7822 | continue; |
7823 | } | |
7b53d598 PB |
7824 | } else { |
7825 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); | |
7826 | io_put_req(cancel_req); | |
2ca10259 JA |
7827 | } |
7828 | ||
fcb323cc | 7829 | schedule(); |
d8f1b971 | 7830 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7831 | } |
7832 | } | |
7833 | ||
801dd57b | 7834 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
44e728b8 | 7835 | { |
801dd57b PB |
7836 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
7837 | struct task_struct *task = data; | |
44e728b8 | 7838 | |
801dd57b | 7839 | return req->task == task; |
44e728b8 PB |
7840 | } |
7841 | ||
fcb323cc JA |
7842 | static int io_uring_flush(struct file *file, void *data) |
7843 | { | |
7844 | struct io_ring_ctx *ctx = file->private_data; | |
7845 | ||
7846 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7847 | |
7848 | /* | |
7849 | * If the task is going away, cancel work it may have pending | |
7850 | */ | |
801dd57b PB |
7851 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
7852 | io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, current, true); | |
6ab23144 | 7853 | |
fcb323cc JA |
7854 | return 0; |
7855 | } | |
7856 | ||
6c5c240e RP |
7857 | static void *io_uring_validate_mmap_request(struct file *file, |
7858 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7859 | { |
2b188cc1 | 7860 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7861 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7862 | struct page *page; |
7863 | void *ptr; | |
7864 | ||
7865 | switch (offset) { | |
7866 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7867 | case IORING_OFF_CQ_RING: |
7868 | ptr = ctx->rings; | |
2b188cc1 JA |
7869 | break; |
7870 | case IORING_OFF_SQES: | |
7871 | ptr = ctx->sq_sqes; | |
7872 | break; | |
2b188cc1 | 7873 | default: |
6c5c240e | 7874 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7875 | } |
7876 | ||
7877 | page = virt_to_head_page(ptr); | |
a50b854e | 7878 | if (sz > page_size(page)) |
6c5c240e RP |
7879 | return ERR_PTR(-EINVAL); |
7880 | ||
7881 | return ptr; | |
7882 | } | |
7883 | ||
7884 | #ifdef CONFIG_MMU | |
7885 | ||
7886 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7887 | { | |
7888 | size_t sz = vma->vm_end - vma->vm_start; | |
7889 | unsigned long pfn; | |
7890 | void *ptr; | |
7891 | ||
7892 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7893 | if (IS_ERR(ptr)) | |
7894 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7895 | |
7896 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7897 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7898 | } | |
7899 | ||
6c5c240e RP |
7900 | #else /* !CONFIG_MMU */ |
7901 | ||
7902 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7903 | { | |
7904 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7905 | } | |
7906 | ||
7907 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7908 | { | |
7909 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7910 | } | |
7911 | ||
7912 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7913 | unsigned long addr, unsigned long len, | |
7914 | unsigned long pgoff, unsigned long flags) | |
7915 | { | |
7916 | void *ptr; | |
7917 | ||
7918 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7919 | if (IS_ERR(ptr)) | |
7920 | return PTR_ERR(ptr); | |
7921 | ||
7922 | return (unsigned long) ptr; | |
7923 | } | |
7924 | ||
7925 | #endif /* !CONFIG_MMU */ | |
7926 | ||
2b188cc1 JA |
7927 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7928 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7929 | size_t, sigsz) | |
7930 | { | |
7931 | struct io_ring_ctx *ctx; | |
7932 | long ret = -EBADF; | |
7933 | int submitted = 0; | |
7934 | struct fd f; | |
7935 | ||
4c6e277c | 7936 | io_run_task_work(); |
b41e9852 | 7937 | |
6c271ce2 | 7938 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7939 | return -EINVAL; |
7940 | ||
7941 | f = fdget(fd); | |
7942 | if (!f.file) | |
7943 | return -EBADF; | |
7944 | ||
7945 | ret = -EOPNOTSUPP; | |
7946 | if (f.file->f_op != &io_uring_fops) | |
7947 | goto out_fput; | |
7948 | ||
7949 | ret = -ENXIO; | |
7950 | ctx = f.file->private_data; | |
7951 | if (!percpu_ref_tryget(&ctx->refs)) | |
7952 | goto out_fput; | |
7953 | ||
6c271ce2 JA |
7954 | /* |
7955 | * For SQ polling, the thread will do all submissions and completions. | |
7956 | * Just return the requested submit count, and wake the thread if | |
7957 | * we were asked to. | |
7958 | */ | |
b2a9eada | 7959 | ret = 0; |
6c271ce2 | 7960 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7961 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7962 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7963 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7964 | wake_up(&ctx->sqo_wait); | |
7965 | submitted = to_submit; | |
b2a9eada | 7966 | } else if (to_submit) { |
2b188cc1 | 7967 | mutex_lock(&ctx->uring_lock); |
0cdaf760 | 7968 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd); |
2b188cc1 | 7969 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7970 | |
7971 | if (submitted != to_submit) | |
7972 | goto out; | |
2b188cc1 JA |
7973 | } |
7974 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7975 | unsigned nr_events = 0; |
7976 | ||
2b188cc1 JA |
7977 | min_complete = min(min_complete, ctx->cq_entries); |
7978 | ||
32b2244a XW |
7979 | /* |
7980 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7981 | * space applications don't need to do io completion events | |
7982 | * polling again, they can rely on io_sq_thread to do polling | |
7983 | * work, which can reduce cpu usage and uring_lock contention. | |
7984 | */ | |
7985 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7986 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7987 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7988 | } else { |
7989 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7990 | } | |
2b188cc1 JA |
7991 | } |
7992 | ||
7c504e65 | 7993 | out: |
6805b32e | 7994 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7995 | out_fput: |
7996 | fdput(f); | |
7997 | return submitted ? submitted : ret; | |
7998 | } | |
7999 | ||
bebdb65e | 8000 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
8001 | static int io_uring_show_cred(int id, void *p, void *data) |
8002 | { | |
8003 | const struct cred *cred = p; | |
8004 | struct seq_file *m = data; | |
8005 | struct user_namespace *uns = seq_user_ns(m); | |
8006 | struct group_info *gi; | |
8007 | kernel_cap_t cap; | |
8008 | unsigned __capi; | |
8009 | int g; | |
8010 | ||
8011 | seq_printf(m, "%5d\n", id); | |
8012 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
8013 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
8014 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
8015 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
8016 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
8017 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
8018 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
8019 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
8020 | seq_puts(m, "\n\tGroups:\t"); | |
8021 | gi = cred->group_info; | |
8022 | for (g = 0; g < gi->ngroups; g++) { | |
8023 | seq_put_decimal_ull(m, g ? " " : "", | |
8024 | from_kgid_munged(uns, gi->gid[g])); | |
8025 | } | |
8026 | seq_puts(m, "\n\tCapEff:\t"); | |
8027 | cap = cred->cap_effective; | |
8028 | CAP_FOR_EACH_U32(__capi) | |
8029 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
8030 | seq_putc(m, '\n'); | |
8031 | return 0; | |
8032 | } | |
8033 | ||
8034 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
8035 | { | |
8036 | int i; | |
8037 | ||
8038 | mutex_lock(&ctx->uring_lock); | |
8039 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
8040 | for (i = 0; i < ctx->nr_user_files; i++) { | |
8041 | struct fixed_file_table *table; | |
8042 | struct file *f; | |
8043 | ||
8044 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
8045 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
8046 | if (f) | |
8047 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
8048 | else | |
8049 | seq_printf(m, "%5u: <none>\n", i); | |
8050 | } | |
8051 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
8052 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8053 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
8054 | ||
8055 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
8056 | (unsigned int) buf->len); | |
8057 | } | |
8058 | if (!idr_is_empty(&ctx->personality_idr)) { | |
8059 | seq_printf(m, "Personalities:\n"); | |
8060 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
8061 | } | |
d7718a9d JA |
8062 | seq_printf(m, "PollList:\n"); |
8063 | spin_lock_irq(&ctx->completion_lock); | |
8064 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
8065 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
8066 | struct io_kiocb *req; | |
8067 | ||
8068 | hlist_for_each_entry(req, list, hash_node) | |
8069 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
8070 | req->task->task_works != NULL); | |
8071 | } | |
8072 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
8073 | mutex_unlock(&ctx->uring_lock); |
8074 | } | |
8075 | ||
8076 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
8077 | { | |
8078 | struct io_ring_ctx *ctx = f->private_data; | |
8079 | ||
8080 | if (percpu_ref_tryget(&ctx->refs)) { | |
8081 | __io_uring_show_fdinfo(ctx, m); | |
8082 | percpu_ref_put(&ctx->refs); | |
8083 | } | |
8084 | } | |
bebdb65e | 8085 | #endif |
87ce955b | 8086 | |
2b188cc1 JA |
8087 | static const struct file_operations io_uring_fops = { |
8088 | .release = io_uring_release, | |
fcb323cc | 8089 | .flush = io_uring_flush, |
2b188cc1 | 8090 | .mmap = io_uring_mmap, |
6c5c240e RP |
8091 | #ifndef CONFIG_MMU |
8092 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
8093 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
8094 | #endif | |
2b188cc1 JA |
8095 | .poll = io_uring_poll, |
8096 | .fasync = io_uring_fasync, | |
bebdb65e | 8097 | #ifdef CONFIG_PROC_FS |
87ce955b | 8098 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 8099 | #endif |
2b188cc1 JA |
8100 | }; |
8101 | ||
8102 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
8103 | struct io_uring_params *p) | |
8104 | { | |
75b28aff HV |
8105 | struct io_rings *rings; |
8106 | size_t size, sq_array_offset; | |
2b188cc1 | 8107 | |
75b28aff HV |
8108 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
8109 | if (size == SIZE_MAX) | |
8110 | return -EOVERFLOW; | |
8111 | ||
8112 | rings = io_mem_alloc(size); | |
8113 | if (!rings) | |
2b188cc1 JA |
8114 | return -ENOMEM; |
8115 | ||
75b28aff HV |
8116 | ctx->rings = rings; |
8117 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
8118 | rings->sq_ring_mask = p->sq_entries - 1; | |
8119 | rings->cq_ring_mask = p->cq_entries - 1; | |
8120 | rings->sq_ring_entries = p->sq_entries; | |
8121 | rings->cq_ring_entries = p->cq_entries; | |
8122 | ctx->sq_mask = rings->sq_ring_mask; | |
8123 | ctx->cq_mask = rings->cq_ring_mask; | |
8124 | ctx->sq_entries = rings->sq_ring_entries; | |
8125 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
8126 | |
8127 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
8128 | if (size == SIZE_MAX) { |
8129 | io_mem_free(ctx->rings); | |
8130 | ctx->rings = NULL; | |
2b188cc1 | 8131 | return -EOVERFLOW; |
eb065d30 | 8132 | } |
2b188cc1 JA |
8133 | |
8134 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
8135 | if (!ctx->sq_sqes) { |
8136 | io_mem_free(ctx->rings); | |
8137 | ctx->rings = NULL; | |
2b188cc1 | 8138 | return -ENOMEM; |
eb065d30 | 8139 | } |
2b188cc1 | 8140 | |
2b188cc1 JA |
8141 | return 0; |
8142 | } | |
8143 | ||
8144 | /* | |
8145 | * Allocate an anonymous fd, this is what constitutes the application | |
8146 | * visible backing of an io_uring instance. The application mmaps this | |
8147 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
8148 | * we have to tie this fd to a socket for file garbage collection purposes. | |
8149 | */ | |
8150 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
8151 | { | |
8152 | struct file *file; | |
8153 | int ret; | |
8154 | ||
8155 | #if defined(CONFIG_UNIX) | |
8156 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
8157 | &ctx->ring_sock); | |
8158 | if (ret) | |
8159 | return ret; | |
8160 | #endif | |
8161 | ||
8162 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
8163 | if (ret < 0) | |
8164 | goto err; | |
8165 | ||
8166 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
8167 | O_RDWR | O_CLOEXEC); | |
8168 | if (IS_ERR(file)) { | |
8169 | put_unused_fd(ret); | |
8170 | ret = PTR_ERR(file); | |
8171 | goto err; | |
8172 | } | |
8173 | ||
8174 | #if defined(CONFIG_UNIX) | |
8175 | ctx->ring_sock->file = file; | |
8176 | #endif | |
8177 | fd_install(ret, file); | |
8178 | return ret; | |
8179 | err: | |
8180 | #if defined(CONFIG_UNIX) | |
8181 | sock_release(ctx->ring_sock); | |
8182 | ctx->ring_sock = NULL; | |
8183 | #endif | |
8184 | return ret; | |
8185 | } | |
8186 | ||
7f13657d XW |
8187 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
8188 | struct io_uring_params __user *params) | |
2b188cc1 JA |
8189 | { |
8190 | struct user_struct *user = NULL; | |
8191 | struct io_ring_ctx *ctx; | |
aad5d8da | 8192 | bool limit_mem; |
2b188cc1 JA |
8193 | int ret; |
8194 | ||
8110c1a6 | 8195 | if (!entries) |
2b188cc1 | 8196 | return -EINVAL; |
8110c1a6 JA |
8197 | if (entries > IORING_MAX_ENTRIES) { |
8198 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8199 | return -EINVAL; | |
8200 | entries = IORING_MAX_ENTRIES; | |
8201 | } | |
2b188cc1 JA |
8202 | |
8203 | /* | |
8204 | * Use twice as many entries for the CQ ring. It's possible for the | |
8205 | * application to drive a higher depth than the size of the SQ ring, | |
8206 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
8207 | * some flexibility in overcommitting a bit. If the application has |
8208 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
8209 | * of CQ ring entries manually. | |
2b188cc1 JA |
8210 | */ |
8211 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
8212 | if (p->flags & IORING_SETUP_CQSIZE) { |
8213 | /* | |
8214 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
8215 | * to a power-of-two, if it isn't already. We do NOT impose | |
8216 | * any cq vs sq ring sizing. | |
8217 | */ | |
8110c1a6 | 8218 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 8219 | return -EINVAL; |
8110c1a6 JA |
8220 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
8221 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
8222 | return -EINVAL; | |
8223 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
8224 | } | |
33a107f0 JA |
8225 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
8226 | } else { | |
8227 | p->cq_entries = 2 * p->sq_entries; | |
8228 | } | |
2b188cc1 JA |
8229 | |
8230 | user = get_uid(current_user()); | |
aad5d8da | 8231 | limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 8232 | |
aad5d8da | 8233 | if (limit_mem) { |
a087e2b5 | 8234 | ret = __io_account_mem(user, |
2b188cc1 JA |
8235 | ring_pages(p->sq_entries, p->cq_entries)); |
8236 | if (ret) { | |
8237 | free_uid(user); | |
8238 | return ret; | |
8239 | } | |
8240 | } | |
8241 | ||
8242 | ctx = io_ring_ctx_alloc(p); | |
8243 | if (!ctx) { | |
aad5d8da | 8244 | if (limit_mem) |
a087e2b5 | 8245 | __io_unaccount_mem(user, ring_pages(p->sq_entries, |
2b188cc1 JA |
8246 | p->cq_entries)); |
8247 | free_uid(user); | |
8248 | return -ENOMEM; | |
8249 | } | |
8250 | ctx->compat = in_compat_syscall(); | |
2b188cc1 | 8251 | ctx->user = user; |
0b8c0ec7 | 8252 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
8253 | |
8254 | ret = io_allocate_scq_urings(ctx, p); | |
8255 | if (ret) | |
8256 | goto err; | |
8257 | ||
6c271ce2 | 8258 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
8259 | if (ret) |
8260 | goto err; | |
8261 | ||
2b188cc1 | 8262 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
8263 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
8264 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
8265 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
8266 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
8267 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
8268 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
8269 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
8270 | |
8271 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
8272 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
8273 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
8274 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
8275 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
8276 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
8277 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 8278 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 8279 | |
7f13657d XW |
8280 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
8281 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 JX |
8282 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
8283 | IORING_FEAT_POLL_32BITS; | |
7f13657d XW |
8284 | |
8285 | if (copy_to_user(params, p, sizeof(*p))) { | |
8286 | ret = -EFAULT; | |
8287 | goto err; | |
8288 | } | |
044c1ab3 JA |
8289 | /* |
8290 | * Install ring fd as the very last thing, so we don't risk someone | |
8291 | * having closed it before we finish setup | |
8292 | */ | |
8293 | ret = io_uring_get_fd(ctx); | |
8294 | if (ret < 0) | |
8295 | goto err; | |
8296 | ||
c826bd7a | 8297 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2e0464d4 BM |
8298 | io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries), |
8299 | ACCT_LOCKED); | |
30975825 | 8300 | ctx->limit_mem = limit_mem; |
2b188cc1 JA |
8301 | return ret; |
8302 | err: | |
8303 | io_ring_ctx_wait_and_kill(ctx); | |
8304 | return ret; | |
8305 | } | |
8306 | ||
8307 | /* | |
8308 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
8309 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
8310 | * params structure passed in. | |
8311 | */ | |
8312 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
8313 | { | |
8314 | struct io_uring_params p; | |
2b188cc1 JA |
8315 | int i; |
8316 | ||
8317 | if (copy_from_user(&p, params, sizeof(p))) | |
8318 | return -EFAULT; | |
8319 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
8320 | if (p.resv[i]) | |
8321 | return -EINVAL; | |
8322 | } | |
8323 | ||
6c271ce2 | 8324 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 8325 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 8326 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
8327 | return -EINVAL; |
8328 | ||
7f13657d | 8329 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
8330 | } |
8331 | ||
8332 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
8333 | struct io_uring_params __user *, params) | |
8334 | { | |
8335 | return io_uring_setup(entries, params); | |
8336 | } | |
8337 | ||
66f4af93 JA |
8338 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
8339 | { | |
8340 | struct io_uring_probe *p; | |
8341 | size_t size; | |
8342 | int i, ret; | |
8343 | ||
8344 | size = struct_size(p, ops, nr_args); | |
8345 | if (size == SIZE_MAX) | |
8346 | return -EOVERFLOW; | |
8347 | p = kzalloc(size, GFP_KERNEL); | |
8348 | if (!p) | |
8349 | return -ENOMEM; | |
8350 | ||
8351 | ret = -EFAULT; | |
8352 | if (copy_from_user(p, arg, size)) | |
8353 | goto out; | |
8354 | ret = -EINVAL; | |
8355 | if (memchr_inv(p, 0, size)) | |
8356 | goto out; | |
8357 | ||
8358 | p->last_op = IORING_OP_LAST - 1; | |
8359 | if (nr_args > IORING_OP_LAST) | |
8360 | nr_args = IORING_OP_LAST; | |
8361 | ||
8362 | for (i = 0; i < nr_args; i++) { | |
8363 | p->ops[i].op = i; | |
8364 | if (!io_op_defs[i].not_supported) | |
8365 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
8366 | } | |
8367 | p->ops_len = i; | |
8368 | ||
8369 | ret = 0; | |
8370 | if (copy_to_user(arg, p, size)) | |
8371 | ret = -EFAULT; | |
8372 | out: | |
8373 | kfree(p); | |
8374 | return ret; | |
8375 | } | |
8376 | ||
071698e1 JA |
8377 | static int io_register_personality(struct io_ring_ctx *ctx) |
8378 | { | |
8379 | const struct cred *creds = get_current_cred(); | |
8380 | int id; | |
8381 | ||
8382 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
8383 | USHRT_MAX, GFP_KERNEL); | |
8384 | if (id < 0) | |
8385 | put_cred(creds); | |
8386 | return id; | |
8387 | } | |
8388 | ||
8389 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
8390 | { | |
8391 | const struct cred *old_creds; | |
8392 | ||
8393 | old_creds = idr_remove(&ctx->personality_idr, id); | |
8394 | if (old_creds) { | |
8395 | put_cred(old_creds); | |
8396 | return 0; | |
8397 | } | |
8398 | ||
8399 | return -EINVAL; | |
8400 | } | |
8401 | ||
8402 | static bool io_register_op_must_quiesce(int op) | |
8403 | { | |
8404 | switch (op) { | |
8405 | case IORING_UNREGISTER_FILES: | |
8406 | case IORING_REGISTER_FILES_UPDATE: | |
8407 | case IORING_REGISTER_PROBE: | |
8408 | case IORING_REGISTER_PERSONALITY: | |
8409 | case IORING_UNREGISTER_PERSONALITY: | |
8410 | return false; | |
8411 | default: | |
8412 | return true; | |
8413 | } | |
8414 | } | |
8415 | ||
edafccee JA |
8416 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8417 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8418 | __releases(ctx->uring_lock) |
8419 | __acquires(ctx->uring_lock) | |
edafccee JA |
8420 | { |
8421 | int ret; | |
8422 | ||
35fa71a0 JA |
8423 | /* |
8424 | * We're inside the ring mutex, if the ref is already dying, then | |
8425 | * someone else killed the ctx or is already going through | |
8426 | * io_uring_register(). | |
8427 | */ | |
8428 | if (percpu_ref_is_dying(&ctx->refs)) | |
8429 | return -ENXIO; | |
8430 | ||
071698e1 | 8431 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8432 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8433 | |
05f3fb3c JA |
8434 | /* |
8435 | * Drop uring mutex before waiting for references to exit. If | |
8436 | * another thread is currently inside io_uring_enter() it might | |
8437 | * need to grab the uring_lock to make progress. If we hold it | |
8438 | * here across the drain wait, then we can deadlock. It's safe | |
8439 | * to drop the mutex here, since no new references will come in | |
8440 | * after we've killed the percpu ref. | |
8441 | */ | |
8442 | mutex_unlock(&ctx->uring_lock); | |
0f158b4c | 8443 | ret = wait_for_completion_interruptible(&ctx->ref_comp); |
05f3fb3c | 8444 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8445 | if (ret) { |
8446 | percpu_ref_resurrect(&ctx->refs); | |
8447 | ret = -EINTR; | |
8448 | goto out; | |
8449 | } | |
05f3fb3c | 8450 | } |
edafccee JA |
8451 | |
8452 | switch (opcode) { | |
8453 | case IORING_REGISTER_BUFFERS: | |
8454 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8455 | break; | |
8456 | case IORING_UNREGISTER_BUFFERS: | |
8457 | ret = -EINVAL; | |
8458 | if (arg || nr_args) | |
8459 | break; | |
8460 | ret = io_sqe_buffer_unregister(ctx); | |
8461 | break; | |
6b06314c JA |
8462 | case IORING_REGISTER_FILES: |
8463 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8464 | break; | |
8465 | case IORING_UNREGISTER_FILES: | |
8466 | ret = -EINVAL; | |
8467 | if (arg || nr_args) | |
8468 | break; | |
8469 | ret = io_sqe_files_unregister(ctx); | |
8470 | break; | |
c3a31e60 JA |
8471 | case IORING_REGISTER_FILES_UPDATE: |
8472 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8473 | break; | |
9b402849 | 8474 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8475 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8476 | ret = -EINVAL; |
8477 | if (nr_args != 1) | |
8478 | break; | |
8479 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8480 | if (ret) |
8481 | break; | |
8482 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8483 | ctx->eventfd_async = 1; | |
8484 | else | |
8485 | ctx->eventfd_async = 0; | |
9b402849 JA |
8486 | break; |
8487 | case IORING_UNREGISTER_EVENTFD: | |
8488 | ret = -EINVAL; | |
8489 | if (arg || nr_args) | |
8490 | break; | |
8491 | ret = io_eventfd_unregister(ctx); | |
8492 | break; | |
66f4af93 JA |
8493 | case IORING_REGISTER_PROBE: |
8494 | ret = -EINVAL; | |
8495 | if (!arg || nr_args > 256) | |
8496 | break; | |
8497 | ret = io_probe(ctx, arg, nr_args); | |
8498 | break; | |
071698e1 JA |
8499 | case IORING_REGISTER_PERSONALITY: |
8500 | ret = -EINVAL; | |
8501 | if (arg || nr_args) | |
8502 | break; | |
8503 | ret = io_register_personality(ctx); | |
8504 | break; | |
8505 | case IORING_UNREGISTER_PERSONALITY: | |
8506 | ret = -EINVAL; | |
8507 | if (arg) | |
8508 | break; | |
8509 | ret = io_unregister_personality(ctx, nr_args); | |
8510 | break; | |
edafccee JA |
8511 | default: |
8512 | ret = -EINVAL; | |
8513 | break; | |
8514 | } | |
8515 | ||
071698e1 | 8516 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8517 | /* bring the ctx back to life */ |
05f3fb3c | 8518 | percpu_ref_reinit(&ctx->refs); |
c150368b | 8519 | out: |
0f158b4c | 8520 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 8521 | } |
edafccee JA |
8522 | return ret; |
8523 | } | |
8524 | ||
8525 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8526 | void __user *, arg, unsigned int, nr_args) | |
8527 | { | |
8528 | struct io_ring_ctx *ctx; | |
8529 | long ret = -EBADF; | |
8530 | struct fd f; | |
8531 | ||
8532 | f = fdget(fd); | |
8533 | if (!f.file) | |
8534 | return -EBADF; | |
8535 | ||
8536 | ret = -EOPNOTSUPP; | |
8537 | if (f.file->f_op != &io_uring_fops) | |
8538 | goto out_fput; | |
8539 | ||
8540 | ctx = f.file->private_data; | |
8541 | ||
8542 | mutex_lock(&ctx->uring_lock); | |
8543 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8544 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8545 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8546 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8547 | out_fput: |
8548 | fdput(f); | |
8549 | return ret; | |
8550 | } | |
8551 | ||
2b188cc1 JA |
8552 | static int __init io_uring_init(void) |
8553 | { | |
d7f62e82 SM |
8554 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8555 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8556 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8557 | } while (0) | |
8558 | ||
8559 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8560 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8561 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8562 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8563 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8564 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8565 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8566 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8567 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8568 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8569 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8570 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8571 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8572 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8573 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8574 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
8575 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
8576 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
8577 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
8578 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8579 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8580 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8581 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8582 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8583 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8584 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8585 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8586 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8587 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8588 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8589 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8590 | |
d3656344 | 8591 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8592 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8593 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8594 | return 0; | |
8595 | }; | |
8596 | __initcall(io_uring_init); |