Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
58 | #include <linux/mmu_context.h> | |
59 | #include <linux/percpu.h> | |
60 | #include <linux/slab.h> | |
6c271ce2 | 61 | #include <linux/kthread.h> |
2b188cc1 | 62 | #include <linux/blkdev.h> |
edafccee | 63 | #include <linux/bvec.h> |
2b188cc1 JA |
64 | #include <linux/net.h> |
65 | #include <net/sock.h> | |
66 | #include <net/af_unix.h> | |
6b06314c | 67 | #include <net/scm.h> |
2b188cc1 JA |
68 | #include <linux/anon_inodes.h> |
69 | #include <linux/sched/mm.h> | |
70 | #include <linux/uaccess.h> | |
71 | #include <linux/nospec.h> | |
edafccee JA |
72 | #include <linux/sizes.h> |
73 | #include <linux/hugetlb.h> | |
aa4c3967 | 74 | #include <linux/highmem.h> |
15b71abe JA |
75 | #include <linux/namei.h> |
76 | #include <linux/fsnotify.h> | |
4840e418 | 77 | #include <linux/fadvise.h> |
3e4827b0 | 78 | #include <linux/eventpoll.h> |
ff002b30 | 79 | #include <linux/fs_struct.h> |
7d67af2c | 80 | #include <linux/splice.h> |
b41e9852 | 81 | #include <linux/task_work.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
101 | |
102 | struct io_uring { | |
103 | u32 head ____cacheline_aligned_in_smp; | |
104 | u32 tail ____cacheline_aligned_in_smp; | |
105 | }; | |
106 | ||
1e84b97b | 107 | /* |
75b28aff HV |
108 | * This data is shared with the application through the mmap at offsets |
109 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
110 | * |
111 | * The offsets to the member fields are published through struct | |
112 | * io_sqring_offsets when calling io_uring_setup. | |
113 | */ | |
75b28aff | 114 | struct io_rings { |
1e84b97b SB |
115 | /* |
116 | * Head and tail offsets into the ring; the offsets need to be | |
117 | * masked to get valid indices. | |
118 | * | |
75b28aff HV |
119 | * The kernel controls head of the sq ring and the tail of the cq ring, |
120 | * and the application controls tail of the sq ring and the head of the | |
121 | * cq ring. | |
1e84b97b | 122 | */ |
75b28aff | 123 | struct io_uring sq, cq; |
1e84b97b | 124 | /* |
75b28aff | 125 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
126 | * ring_entries - 1) |
127 | */ | |
75b28aff HV |
128 | u32 sq_ring_mask, cq_ring_mask; |
129 | /* Ring sizes (constant, power of 2) */ | |
130 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
131 | /* |
132 | * Number of invalid entries dropped by the kernel due to | |
133 | * invalid index stored in array | |
134 | * | |
135 | * Written by the kernel, shouldn't be modified by the | |
136 | * application (i.e. get number of "new events" by comparing to | |
137 | * cached value). | |
138 | * | |
139 | * After a new SQ head value was read by the application this | |
140 | * counter includes all submissions that were dropped reaching | |
141 | * the new SQ head (and possibly more). | |
142 | */ | |
75b28aff | 143 | u32 sq_dropped; |
1e84b97b SB |
144 | /* |
145 | * Runtime flags | |
146 | * | |
147 | * Written by the kernel, shouldn't be modified by the | |
148 | * application. | |
149 | * | |
150 | * The application needs a full memory barrier before checking | |
151 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
152 | */ | |
75b28aff | 153 | u32 sq_flags; |
1e84b97b SB |
154 | /* |
155 | * Number of completion events lost because the queue was full; | |
156 | * this should be avoided by the application by making sure | |
0b4295b5 | 157 | * there are not more requests pending than there is space in |
1e84b97b SB |
158 | * the completion queue. |
159 | * | |
160 | * Written by the kernel, shouldn't be modified by the | |
161 | * application (i.e. get number of "new events" by comparing to | |
162 | * cached value). | |
163 | * | |
164 | * As completion events come in out of order this counter is not | |
165 | * ordered with any other data. | |
166 | */ | |
75b28aff | 167 | u32 cq_overflow; |
1e84b97b SB |
168 | /* |
169 | * Ring buffer of completion events. | |
170 | * | |
171 | * The kernel writes completion events fresh every time they are | |
172 | * produced, so the application is allowed to modify pending | |
173 | * entries. | |
174 | */ | |
75b28aff | 175 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
176 | }; |
177 | ||
edafccee JA |
178 | struct io_mapped_ubuf { |
179 | u64 ubuf; | |
180 | size_t len; | |
181 | struct bio_vec *bvec; | |
182 | unsigned int nr_bvecs; | |
183 | }; | |
184 | ||
65e19f54 JA |
185 | struct fixed_file_table { |
186 | struct file **files; | |
31b51510 JA |
187 | }; |
188 | ||
05589553 XW |
189 | struct fixed_file_ref_node { |
190 | struct percpu_ref refs; | |
191 | struct list_head node; | |
192 | struct list_head file_list; | |
193 | struct fixed_file_data *file_data; | |
194 | struct work_struct work; | |
195 | }; | |
196 | ||
05f3fb3c JA |
197 | struct fixed_file_data { |
198 | struct fixed_file_table *table; | |
199 | struct io_ring_ctx *ctx; | |
200 | ||
05589553 | 201 | struct percpu_ref *cur_refs; |
05f3fb3c | 202 | struct percpu_ref refs; |
05f3fb3c | 203 | struct completion done; |
05589553 XW |
204 | struct list_head ref_list; |
205 | spinlock_t lock; | |
05f3fb3c JA |
206 | }; |
207 | ||
5a2e745d JA |
208 | struct io_buffer { |
209 | struct list_head list; | |
210 | __u64 addr; | |
211 | __s32 len; | |
212 | __u16 bid; | |
213 | }; | |
214 | ||
2b188cc1 JA |
215 | struct io_ring_ctx { |
216 | struct { | |
217 | struct percpu_ref refs; | |
218 | } ____cacheline_aligned_in_smp; | |
219 | ||
220 | struct { | |
221 | unsigned int flags; | |
e1d85334 RD |
222 | unsigned int compat: 1; |
223 | unsigned int account_mem: 1; | |
224 | unsigned int cq_overflow_flushed: 1; | |
225 | unsigned int drain_next: 1; | |
226 | unsigned int eventfd_async: 1; | |
2b188cc1 | 227 | |
75b28aff HV |
228 | /* |
229 | * Ring buffer of indices into array of io_uring_sqe, which is | |
230 | * mmapped by the application using the IORING_OFF_SQES offset. | |
231 | * | |
232 | * This indirection could e.g. be used to assign fixed | |
233 | * io_uring_sqe entries to operations and only submit them to | |
234 | * the queue when needed. | |
235 | * | |
236 | * The kernel modifies neither the indices array nor the entries | |
237 | * array. | |
238 | */ | |
239 | u32 *sq_array; | |
2b188cc1 JA |
240 | unsigned cached_sq_head; |
241 | unsigned sq_entries; | |
242 | unsigned sq_mask; | |
6c271ce2 | 243 | unsigned sq_thread_idle; |
498ccd9e | 244 | unsigned cached_sq_dropped; |
206aefde | 245 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 246 | unsigned long sq_check_overflow; |
de0617e4 JA |
247 | |
248 | struct list_head defer_list; | |
5262f567 | 249 | struct list_head timeout_list; |
1d7bb1d5 | 250 | struct list_head cq_overflow_list; |
fcb323cc JA |
251 | |
252 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 253 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
254 | } ____cacheline_aligned_in_smp; |
255 | ||
206aefde JA |
256 | struct io_rings *rings; |
257 | ||
2b188cc1 | 258 | /* IO offload */ |
561fb04a | 259 | struct io_wq *io_wq; |
6c271ce2 | 260 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 261 | struct mm_struct *sqo_mm; |
6c271ce2 | 262 | wait_queue_head_t sqo_wait; |
75b28aff | 263 | |
6b06314c JA |
264 | /* |
265 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
266 | * readers must ensure that ->refs is alive as long as the file* is | |
267 | * used. Only updated through io_uring_register(2). | |
268 | */ | |
05f3fb3c | 269 | struct fixed_file_data *file_data; |
6b06314c | 270 | unsigned nr_user_files; |
b14cca0c PB |
271 | int ring_fd; |
272 | struct file *ring_file; | |
6b06314c | 273 | |
edafccee JA |
274 | /* if used, fixed mapped user buffers */ |
275 | unsigned nr_user_bufs; | |
276 | struct io_mapped_ubuf *user_bufs; | |
277 | ||
2b188cc1 JA |
278 | struct user_struct *user; |
279 | ||
0b8c0ec7 | 280 | const struct cred *creds; |
181e448d | 281 | |
206aefde JA |
282 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
283 | struct completion *completions; | |
284 | ||
0ddf92e8 JA |
285 | /* if all else fails... */ |
286 | struct io_kiocb *fallback_req; | |
287 | ||
206aefde JA |
288 | #if defined(CONFIG_UNIX) |
289 | struct socket *ring_sock; | |
290 | #endif | |
291 | ||
5a2e745d JA |
292 | struct idr io_buffer_idr; |
293 | ||
071698e1 JA |
294 | struct idr personality_idr; |
295 | ||
206aefde JA |
296 | struct { |
297 | unsigned cached_cq_tail; | |
298 | unsigned cq_entries; | |
299 | unsigned cq_mask; | |
300 | atomic_t cq_timeouts; | |
ad3eb2c8 | 301 | unsigned long cq_check_overflow; |
206aefde JA |
302 | struct wait_queue_head cq_wait; |
303 | struct fasync_struct *cq_fasync; | |
304 | struct eventfd_ctx *cq_ev_fd; | |
305 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
306 | |
307 | struct { | |
308 | struct mutex uring_lock; | |
309 | wait_queue_head_t wait; | |
310 | } ____cacheline_aligned_in_smp; | |
311 | ||
312 | struct { | |
313 | spinlock_t completion_lock; | |
e94f141b | 314 | |
def596e9 JA |
315 | /* |
316 | * ->poll_list is protected by the ctx->uring_lock for | |
317 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
318 | * For SQPOLL, only the single threaded io_sq_thread() will | |
319 | * manipulate the list, hence no extra locking is needed there. | |
320 | */ | |
321 | struct list_head poll_list; | |
78076bb6 JA |
322 | struct hlist_head *cancel_hash; |
323 | unsigned cancel_hash_bits; | |
e94f141b | 324 | bool poll_multi_file; |
31b51510 | 325 | |
fcb323cc JA |
326 | spinlock_t inflight_lock; |
327 | struct list_head inflight_list; | |
2b188cc1 | 328 | } ____cacheline_aligned_in_smp; |
85faa7b8 JA |
329 | |
330 | struct work_struct exit_work; | |
2b188cc1 JA |
331 | }; |
332 | ||
09bb8394 JA |
333 | /* |
334 | * First field must be the file pointer in all the | |
335 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
336 | */ | |
221c5eb2 JA |
337 | struct io_poll_iocb { |
338 | struct file *file; | |
0969e783 JA |
339 | union { |
340 | struct wait_queue_head *head; | |
341 | u64 addr; | |
342 | }; | |
221c5eb2 | 343 | __poll_t events; |
8c838788 | 344 | bool done; |
221c5eb2 | 345 | bool canceled; |
392edb45 | 346 | struct wait_queue_entry wait; |
221c5eb2 JA |
347 | }; |
348 | ||
b5dba59e JA |
349 | struct io_close { |
350 | struct file *file; | |
351 | struct file *put_file; | |
352 | int fd; | |
353 | }; | |
354 | ||
ad8a48ac JA |
355 | struct io_timeout_data { |
356 | struct io_kiocb *req; | |
357 | struct hrtimer timer; | |
358 | struct timespec64 ts; | |
359 | enum hrtimer_mode mode; | |
360 | }; | |
361 | ||
8ed8d3c3 JA |
362 | struct io_accept { |
363 | struct file *file; | |
364 | struct sockaddr __user *addr; | |
365 | int __user *addr_len; | |
366 | int flags; | |
09952e3e | 367 | unsigned long nofile; |
8ed8d3c3 JA |
368 | }; |
369 | ||
370 | struct io_sync { | |
371 | struct file *file; | |
372 | loff_t len; | |
373 | loff_t off; | |
374 | int flags; | |
d63d1b5e | 375 | int mode; |
8ed8d3c3 JA |
376 | }; |
377 | ||
fbf23849 JA |
378 | struct io_cancel { |
379 | struct file *file; | |
380 | u64 addr; | |
381 | }; | |
382 | ||
b29472ee JA |
383 | struct io_timeout { |
384 | struct file *file; | |
385 | u64 addr; | |
386 | int flags; | |
b55ce732 | 387 | u32 count; |
b29472ee JA |
388 | }; |
389 | ||
9adbd45d JA |
390 | struct io_rw { |
391 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
392 | struct kiocb kiocb; | |
393 | u64 addr; | |
394 | u64 len; | |
395 | }; | |
396 | ||
3fbb51c1 JA |
397 | struct io_connect { |
398 | struct file *file; | |
399 | struct sockaddr __user *addr; | |
400 | int addr_len; | |
401 | }; | |
402 | ||
e47293fd JA |
403 | struct io_sr_msg { |
404 | struct file *file; | |
fddaface JA |
405 | union { |
406 | struct user_msghdr __user *msg; | |
407 | void __user *buf; | |
408 | }; | |
e47293fd | 409 | int msg_flags; |
bcda7baa | 410 | int bgid; |
fddaface | 411 | size_t len; |
bcda7baa | 412 | struct io_buffer *kbuf; |
e47293fd JA |
413 | }; |
414 | ||
15b71abe JA |
415 | struct io_open { |
416 | struct file *file; | |
417 | int dfd; | |
eddc7ef5 | 418 | union { |
eddc7ef5 JA |
419 | unsigned mask; |
420 | }; | |
15b71abe | 421 | struct filename *filename; |
eddc7ef5 | 422 | struct statx __user *buffer; |
c12cedf2 | 423 | struct open_how how; |
4022e7af | 424 | unsigned long nofile; |
15b71abe JA |
425 | }; |
426 | ||
05f3fb3c JA |
427 | struct io_files_update { |
428 | struct file *file; | |
429 | u64 arg; | |
430 | u32 nr_args; | |
431 | u32 offset; | |
432 | }; | |
433 | ||
4840e418 JA |
434 | struct io_fadvise { |
435 | struct file *file; | |
436 | u64 offset; | |
437 | u32 len; | |
438 | u32 advice; | |
439 | }; | |
440 | ||
c1ca757b JA |
441 | struct io_madvise { |
442 | struct file *file; | |
443 | u64 addr; | |
444 | u32 len; | |
445 | u32 advice; | |
446 | }; | |
447 | ||
3e4827b0 JA |
448 | struct io_epoll { |
449 | struct file *file; | |
450 | int epfd; | |
451 | int op; | |
452 | int fd; | |
453 | struct epoll_event event; | |
e47293fd JA |
454 | }; |
455 | ||
7d67af2c PB |
456 | struct io_splice { |
457 | struct file *file_out; | |
458 | struct file *file_in; | |
459 | loff_t off_out; | |
460 | loff_t off_in; | |
461 | u64 len; | |
462 | unsigned int flags; | |
463 | }; | |
464 | ||
ddf0322d JA |
465 | struct io_provide_buf { |
466 | struct file *file; | |
467 | __u64 addr; | |
468 | __s32 len; | |
469 | __u32 bgid; | |
470 | __u16 nbufs; | |
471 | __u16 bid; | |
472 | }; | |
473 | ||
f499a021 JA |
474 | struct io_async_connect { |
475 | struct sockaddr_storage address; | |
476 | }; | |
477 | ||
03b1230c JA |
478 | struct io_async_msghdr { |
479 | struct iovec fast_iov[UIO_FASTIOV]; | |
480 | struct iovec *iov; | |
481 | struct sockaddr __user *uaddr; | |
482 | struct msghdr msg; | |
b537916c | 483 | struct sockaddr_storage addr; |
03b1230c JA |
484 | }; |
485 | ||
f67676d1 JA |
486 | struct io_async_rw { |
487 | struct iovec fast_iov[UIO_FASTIOV]; | |
488 | struct iovec *iov; | |
489 | ssize_t nr_segs; | |
490 | ssize_t size; | |
491 | }; | |
492 | ||
1a6b74fc | 493 | struct io_async_ctx { |
f67676d1 JA |
494 | union { |
495 | struct io_async_rw rw; | |
03b1230c | 496 | struct io_async_msghdr msg; |
f499a021 | 497 | struct io_async_connect connect; |
2d28390a | 498 | struct io_timeout_data timeout; |
f67676d1 | 499 | }; |
1a6b74fc JA |
500 | }; |
501 | ||
6b47ee6e PB |
502 | enum { |
503 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
504 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
505 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
506 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
507 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 508 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 509 | |
dea3b49c | 510 | REQ_F_LINK_HEAD_BIT, |
6b47ee6e PB |
511 | REQ_F_LINK_NEXT_BIT, |
512 | REQ_F_FAIL_LINK_BIT, | |
513 | REQ_F_INFLIGHT_BIT, | |
514 | REQ_F_CUR_POS_BIT, | |
515 | REQ_F_NOWAIT_BIT, | |
516 | REQ_F_IOPOLL_COMPLETED_BIT, | |
517 | REQ_F_LINK_TIMEOUT_BIT, | |
518 | REQ_F_TIMEOUT_BIT, | |
519 | REQ_F_ISREG_BIT, | |
520 | REQ_F_MUST_PUNT_BIT, | |
521 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
522 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 523 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 524 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 525 | REQ_F_POLLED_BIT, |
bcda7baa | 526 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 527 | REQ_F_NO_FILE_TABLE_BIT, |
84557871 JA |
528 | |
529 | /* not a real bit, just to check we're not overflowing the space */ | |
530 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
531 | }; |
532 | ||
533 | enum { | |
534 | /* ctx owns file */ | |
535 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
536 | /* drain existing IO first */ | |
537 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
538 | /* linked sqes */ | |
539 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
540 | /* doesn't sever on completion < 0 */ | |
541 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
542 | /* IOSQE_ASYNC */ | |
543 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
544 | /* IOSQE_BUFFER_SELECT */ |
545 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 546 | |
dea3b49c PB |
547 | /* head of a link */ |
548 | REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT), | |
6b47ee6e PB |
549 | /* already grabbed next link */ |
550 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
551 | /* fail rest of links */ | |
552 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
553 | /* on inflight list */ | |
554 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
555 | /* read/write uses file position */ | |
556 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
557 | /* must not punt to workers */ | |
558 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
559 | /* polled IO has completed */ | |
560 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
561 | /* has linked timeout */ | |
562 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
563 | /* timeout request */ | |
564 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
565 | /* regular file */ | |
566 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
567 | /* must be punted even for NONBLOCK */ | |
568 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
569 | /* no timeout sequence */ | |
570 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
571 | /* completion under lock */ | |
572 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
573 | /* needs cleanup */ |
574 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
575 | /* in overflow list */ |
576 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
577 | /* already went through poll handler */ |
578 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
579 | /* buffer already selected */ |
580 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
581 | /* doesn't need file table for this request */ |
582 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
d7718a9d JA |
583 | }; |
584 | ||
585 | struct async_poll { | |
586 | struct io_poll_iocb poll; | |
587 | struct io_wq_work work; | |
6b47ee6e PB |
588 | }; |
589 | ||
09bb8394 JA |
590 | /* |
591 | * NOTE! Each of the iocb union members has the file pointer | |
592 | * as the first entry in their struct definition. So you can | |
593 | * access the file pointer through any of the sub-structs, | |
594 | * or directly as just 'ki_filp' in this struct. | |
595 | */ | |
2b188cc1 | 596 | struct io_kiocb { |
221c5eb2 | 597 | union { |
09bb8394 | 598 | struct file *file; |
9adbd45d | 599 | struct io_rw rw; |
221c5eb2 | 600 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
601 | struct io_accept accept; |
602 | struct io_sync sync; | |
fbf23849 | 603 | struct io_cancel cancel; |
b29472ee | 604 | struct io_timeout timeout; |
3fbb51c1 | 605 | struct io_connect connect; |
e47293fd | 606 | struct io_sr_msg sr_msg; |
15b71abe | 607 | struct io_open open; |
b5dba59e | 608 | struct io_close close; |
05f3fb3c | 609 | struct io_files_update files_update; |
4840e418 | 610 | struct io_fadvise fadvise; |
c1ca757b | 611 | struct io_madvise madvise; |
3e4827b0 | 612 | struct io_epoll epoll; |
7d67af2c | 613 | struct io_splice splice; |
ddf0322d | 614 | struct io_provide_buf pbuf; |
221c5eb2 | 615 | }; |
2b188cc1 | 616 | |
1a6b74fc | 617 | struct io_async_ctx *io; |
c398ecb3 | 618 | int cflags; |
cf6fd4bd | 619 | bool needs_fixed_file; |
d625c6ee | 620 | u8 opcode; |
2b188cc1 JA |
621 | |
622 | struct io_ring_ctx *ctx; | |
d7718a9d | 623 | struct list_head list; |
2b188cc1 | 624 | unsigned int flags; |
c16361c1 | 625 | refcount_t refs; |
3537b6a7 JA |
626 | struct task_struct *task; |
627 | unsigned long fsize; | |
2b188cc1 | 628 | u64 user_data; |
9e645e11 | 629 | u32 result; |
de0617e4 | 630 | u32 sequence; |
2b188cc1 | 631 | |
d7718a9d JA |
632 | struct list_head link_list; |
633 | ||
fcb323cc JA |
634 | struct list_head inflight_entry; |
635 | ||
05589553 XW |
636 | struct percpu_ref *fixed_file_refs; |
637 | ||
b41e9852 JA |
638 | union { |
639 | /* | |
640 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
641 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
642 | * async armed poll handlers for regular commands. The latter | |
643 | * restore the work, if needed. | |
b41e9852 JA |
644 | */ |
645 | struct { | |
b41e9852 | 646 | struct callback_head task_work; |
d7718a9d JA |
647 | struct hlist_node hash_node; |
648 | struct async_poll *apoll; | |
b41e9852 JA |
649 | }; |
650 | struct io_wq_work work; | |
651 | }; | |
2b188cc1 JA |
652 | }; |
653 | ||
654 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 655 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 656 | |
9a56a232 JA |
657 | struct io_submit_state { |
658 | struct blk_plug plug; | |
659 | ||
2579f913 JA |
660 | /* |
661 | * io_kiocb alloc cache | |
662 | */ | |
663 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 664 | unsigned int free_reqs; |
2579f913 | 665 | |
9a56a232 JA |
666 | /* |
667 | * File reference cache | |
668 | */ | |
669 | struct file *file; | |
670 | unsigned int fd; | |
671 | unsigned int has_refs; | |
672 | unsigned int used_refs; | |
673 | unsigned int ios_left; | |
674 | }; | |
675 | ||
d3656344 JA |
676 | struct io_op_def { |
677 | /* needs req->io allocated for deferral/async */ | |
678 | unsigned async_ctx : 1; | |
679 | /* needs current->mm setup, does mm access */ | |
680 | unsigned needs_mm : 1; | |
681 | /* needs req->file assigned */ | |
682 | unsigned needs_file : 1; | |
683 | /* needs req->file assigned IFF fd is >= 0 */ | |
684 | unsigned fd_non_neg : 1; | |
685 | /* hash wq insertion if file is a regular file */ | |
686 | unsigned hash_reg_file : 1; | |
687 | /* unbound wq insertion if file is a non-regular file */ | |
688 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
689 | /* opcode is not supported by this kernel */ |
690 | unsigned not_supported : 1; | |
f86cd20c JA |
691 | /* needs file table */ |
692 | unsigned file_table : 1; | |
ff002b30 JA |
693 | /* needs ->fs */ |
694 | unsigned needs_fs : 1; | |
8a72758c JA |
695 | /* set if opcode supports polled "wait" */ |
696 | unsigned pollin : 1; | |
697 | unsigned pollout : 1; | |
bcda7baa JA |
698 | /* op supports buffer selection */ |
699 | unsigned buffer_select : 1; | |
d3656344 JA |
700 | }; |
701 | ||
702 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
703 | [IORING_OP_NOP] = {}, |
704 | [IORING_OP_READV] = { | |
d3656344 JA |
705 | .async_ctx = 1, |
706 | .needs_mm = 1, | |
707 | .needs_file = 1, | |
708 | .unbound_nonreg_file = 1, | |
8a72758c | 709 | .pollin = 1, |
4d954c25 | 710 | .buffer_select = 1, |
d3656344 | 711 | }, |
0463b6c5 | 712 | [IORING_OP_WRITEV] = { |
d3656344 JA |
713 | .async_ctx = 1, |
714 | .needs_mm = 1, | |
715 | .needs_file = 1, | |
716 | .hash_reg_file = 1, | |
717 | .unbound_nonreg_file = 1, | |
8a72758c | 718 | .pollout = 1, |
d3656344 | 719 | }, |
0463b6c5 | 720 | [IORING_OP_FSYNC] = { |
d3656344 JA |
721 | .needs_file = 1, |
722 | }, | |
0463b6c5 | 723 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
724 | .needs_file = 1, |
725 | .unbound_nonreg_file = 1, | |
8a72758c | 726 | .pollin = 1, |
d3656344 | 727 | }, |
0463b6c5 | 728 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
729 | .needs_file = 1, |
730 | .hash_reg_file = 1, | |
731 | .unbound_nonreg_file = 1, | |
8a72758c | 732 | .pollout = 1, |
d3656344 | 733 | }, |
0463b6c5 | 734 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
735 | .needs_file = 1, |
736 | .unbound_nonreg_file = 1, | |
737 | }, | |
0463b6c5 PB |
738 | [IORING_OP_POLL_REMOVE] = {}, |
739 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
740 | .needs_file = 1, |
741 | }, | |
0463b6c5 | 742 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
743 | .async_ctx = 1, |
744 | .needs_mm = 1, | |
745 | .needs_file = 1, | |
746 | .unbound_nonreg_file = 1, | |
ff002b30 | 747 | .needs_fs = 1, |
8a72758c | 748 | .pollout = 1, |
d3656344 | 749 | }, |
0463b6c5 | 750 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
751 | .async_ctx = 1, |
752 | .needs_mm = 1, | |
753 | .needs_file = 1, | |
754 | .unbound_nonreg_file = 1, | |
ff002b30 | 755 | .needs_fs = 1, |
8a72758c | 756 | .pollin = 1, |
52de1fe1 | 757 | .buffer_select = 1, |
d3656344 | 758 | }, |
0463b6c5 | 759 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
760 | .async_ctx = 1, |
761 | .needs_mm = 1, | |
762 | }, | |
0463b6c5 PB |
763 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
764 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
765 | .needs_mm = 1, |
766 | .needs_file = 1, | |
767 | .unbound_nonreg_file = 1, | |
f86cd20c | 768 | .file_table = 1, |
8a72758c | 769 | .pollin = 1, |
d3656344 | 770 | }, |
0463b6c5 PB |
771 | [IORING_OP_ASYNC_CANCEL] = {}, |
772 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
773 | .async_ctx = 1, |
774 | .needs_mm = 1, | |
775 | }, | |
0463b6c5 | 776 | [IORING_OP_CONNECT] = { |
d3656344 JA |
777 | .async_ctx = 1, |
778 | .needs_mm = 1, | |
779 | .needs_file = 1, | |
780 | .unbound_nonreg_file = 1, | |
8a72758c | 781 | .pollout = 1, |
d3656344 | 782 | }, |
0463b6c5 | 783 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
784 | .needs_file = 1, |
785 | }, | |
0463b6c5 | 786 | [IORING_OP_OPENAT] = { |
d3656344 JA |
787 | .needs_file = 1, |
788 | .fd_non_neg = 1, | |
f86cd20c | 789 | .file_table = 1, |
ff002b30 | 790 | .needs_fs = 1, |
d3656344 | 791 | }, |
0463b6c5 | 792 | [IORING_OP_CLOSE] = { |
d3656344 | 793 | .needs_file = 1, |
f86cd20c | 794 | .file_table = 1, |
d3656344 | 795 | }, |
0463b6c5 | 796 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 797 | .needs_mm = 1, |
f86cd20c | 798 | .file_table = 1, |
d3656344 | 799 | }, |
0463b6c5 | 800 | [IORING_OP_STATX] = { |
d3656344 JA |
801 | .needs_mm = 1, |
802 | .needs_file = 1, | |
803 | .fd_non_neg = 1, | |
ff002b30 | 804 | .needs_fs = 1, |
5b0bbee4 | 805 | .file_table = 1, |
d3656344 | 806 | }, |
0463b6c5 | 807 | [IORING_OP_READ] = { |
3a6820f2 JA |
808 | .needs_mm = 1, |
809 | .needs_file = 1, | |
810 | .unbound_nonreg_file = 1, | |
8a72758c | 811 | .pollin = 1, |
bcda7baa | 812 | .buffer_select = 1, |
3a6820f2 | 813 | }, |
0463b6c5 | 814 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
815 | .needs_mm = 1, |
816 | .needs_file = 1, | |
817 | .unbound_nonreg_file = 1, | |
8a72758c | 818 | .pollout = 1, |
3a6820f2 | 819 | }, |
0463b6c5 | 820 | [IORING_OP_FADVISE] = { |
4840e418 JA |
821 | .needs_file = 1, |
822 | }, | |
0463b6c5 | 823 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
824 | .needs_mm = 1, |
825 | }, | |
0463b6c5 | 826 | [IORING_OP_SEND] = { |
fddaface JA |
827 | .needs_mm = 1, |
828 | .needs_file = 1, | |
829 | .unbound_nonreg_file = 1, | |
8a72758c | 830 | .pollout = 1, |
fddaface | 831 | }, |
0463b6c5 | 832 | [IORING_OP_RECV] = { |
fddaface JA |
833 | .needs_mm = 1, |
834 | .needs_file = 1, | |
835 | .unbound_nonreg_file = 1, | |
8a72758c | 836 | .pollin = 1, |
bcda7baa | 837 | .buffer_select = 1, |
fddaface | 838 | }, |
0463b6c5 | 839 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
840 | .needs_file = 1, |
841 | .fd_non_neg = 1, | |
f86cd20c | 842 | .file_table = 1, |
ff002b30 | 843 | .needs_fs = 1, |
cebdb986 | 844 | }, |
3e4827b0 JA |
845 | [IORING_OP_EPOLL_CTL] = { |
846 | .unbound_nonreg_file = 1, | |
847 | .file_table = 1, | |
848 | }, | |
7d67af2c PB |
849 | [IORING_OP_SPLICE] = { |
850 | .needs_file = 1, | |
851 | .hash_reg_file = 1, | |
852 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
853 | }, |
854 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 855 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
856 | }; |
857 | ||
561fb04a | 858 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 859 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 860 | static void io_put_req(struct io_kiocb *req); |
978db57e | 861 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
862 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
863 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
864 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
865 | struct io_uring_files_update *ip, | |
866 | unsigned nr_args); | |
f86cd20c | 867 | static int io_grab_files(struct io_kiocb *req); |
99bc4c38 | 868 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
869 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
870 | int fd, struct file **out_file, bool fixed); | |
871 | static void __io_queue_sqe(struct io_kiocb *req, | |
872 | const struct io_uring_sqe *sqe); | |
de0617e4 | 873 | |
2b188cc1 JA |
874 | static struct kmem_cache *req_cachep; |
875 | ||
876 | static const struct file_operations io_uring_fops; | |
877 | ||
878 | struct sock *io_uring_get_socket(struct file *file) | |
879 | { | |
880 | #if defined(CONFIG_UNIX) | |
881 | if (file->f_op == &io_uring_fops) { | |
882 | struct io_ring_ctx *ctx = file->private_data; | |
883 | ||
884 | return ctx->ring_sock->sk; | |
885 | } | |
886 | #endif | |
887 | return NULL; | |
888 | } | |
889 | EXPORT_SYMBOL(io_uring_get_socket); | |
890 | ||
891 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
892 | { | |
893 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
894 | ||
206aefde | 895 | complete(&ctx->completions[0]); |
2b188cc1 JA |
896 | } |
897 | ||
898 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
899 | { | |
900 | struct io_ring_ctx *ctx; | |
78076bb6 | 901 | int hash_bits; |
2b188cc1 JA |
902 | |
903 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
904 | if (!ctx) | |
905 | return NULL; | |
906 | ||
0ddf92e8 JA |
907 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
908 | if (!ctx->fallback_req) | |
909 | goto err; | |
910 | ||
206aefde JA |
911 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
912 | if (!ctx->completions) | |
913 | goto err; | |
914 | ||
78076bb6 JA |
915 | /* |
916 | * Use 5 bits less than the max cq entries, that should give us around | |
917 | * 32 entries per hash list if totally full and uniformly spread. | |
918 | */ | |
919 | hash_bits = ilog2(p->cq_entries); | |
920 | hash_bits -= 5; | |
921 | if (hash_bits <= 0) | |
922 | hash_bits = 1; | |
923 | ctx->cancel_hash_bits = hash_bits; | |
924 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
925 | GFP_KERNEL); | |
926 | if (!ctx->cancel_hash) | |
927 | goto err; | |
928 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
929 | ||
21482896 | 930 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
931 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
932 | goto err; | |
2b188cc1 JA |
933 | |
934 | ctx->flags = p->flags; | |
935 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 936 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
937 | init_completion(&ctx->completions[0]); |
938 | init_completion(&ctx->completions[1]); | |
5a2e745d | 939 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 940 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
941 | mutex_init(&ctx->uring_lock); |
942 | init_waitqueue_head(&ctx->wait); | |
943 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 944 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 945 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 946 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
947 | init_waitqueue_head(&ctx->inflight_wait); |
948 | spin_lock_init(&ctx->inflight_lock); | |
949 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 950 | return ctx; |
206aefde | 951 | err: |
0ddf92e8 JA |
952 | if (ctx->fallback_req) |
953 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 954 | kfree(ctx->completions); |
78076bb6 | 955 | kfree(ctx->cancel_hash); |
206aefde JA |
956 | kfree(ctx); |
957 | return NULL; | |
2b188cc1 JA |
958 | } |
959 | ||
9d858b21 | 960 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 961 | { |
a197f664 JL |
962 | struct io_ring_ctx *ctx = req->ctx; |
963 | ||
31af27c7 PB |
964 | return req->sequence != ctx->cached_cq_tail |
965 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
966 | } |
967 | ||
9d858b21 | 968 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 969 | { |
87987898 | 970 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 971 | return __req_need_defer(req); |
de0617e4 | 972 | |
9d858b21 | 973 | return false; |
de0617e4 JA |
974 | } |
975 | ||
7adf4eaf | 976 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
977 | { |
978 | struct io_kiocb *req; | |
979 | ||
7adf4eaf | 980 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 981 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
982 | list_del_init(&req->list); |
983 | return req; | |
984 | } | |
985 | ||
986 | return NULL; | |
987 | } | |
988 | ||
5262f567 JA |
989 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
990 | { | |
7adf4eaf JA |
991 | struct io_kiocb *req; |
992 | ||
993 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
994 | if (req) { |
995 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
996 | return NULL; | |
fb4b3d3f | 997 | if (!__req_need_defer(req)) { |
93bd25bb JA |
998 | list_del_init(&req->list); |
999 | return req; | |
1000 | } | |
7adf4eaf JA |
1001 | } |
1002 | ||
1003 | return NULL; | |
5262f567 JA |
1004 | } |
1005 | ||
de0617e4 | 1006 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1007 | { |
75b28aff | 1008 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 1009 | |
07910158 PB |
1010 | /* order cqe stores with ring update */ |
1011 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 1012 | |
07910158 PB |
1013 | if (wq_has_sleeper(&ctx->cq_wait)) { |
1014 | wake_up_interruptible(&ctx->cq_wait); | |
1015 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1016 | } |
1017 | } | |
1018 | ||
cccf0ee8 JA |
1019 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1020 | const struct io_op_def *def) | |
1021 | { | |
1022 | if (!req->work.mm && def->needs_mm) { | |
1023 | mmgrab(current->mm); | |
1024 | req->work.mm = current->mm; | |
2b188cc1 | 1025 | } |
cccf0ee8 JA |
1026 | if (!req->work.creds) |
1027 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1028 | if (!req->work.fs && def->needs_fs) { |
1029 | spin_lock(¤t->fs->lock); | |
1030 | if (!current->fs->in_exec) { | |
1031 | req->work.fs = current->fs; | |
1032 | req->work.fs->users++; | |
1033 | } else { | |
1034 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1035 | } | |
1036 | spin_unlock(¤t->fs->lock); | |
1037 | } | |
6ab23144 JA |
1038 | if (!req->work.task_pid) |
1039 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1040 | } |
1041 | ||
cccf0ee8 | 1042 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1043 | { |
cccf0ee8 JA |
1044 | if (req->work.mm) { |
1045 | mmdrop(req->work.mm); | |
1046 | req->work.mm = NULL; | |
1047 | } | |
1048 | if (req->work.creds) { | |
1049 | put_cred(req->work.creds); | |
1050 | req->work.creds = NULL; | |
1051 | } | |
ff002b30 JA |
1052 | if (req->work.fs) { |
1053 | struct fs_struct *fs = req->work.fs; | |
1054 | ||
1055 | spin_lock(&req->work.fs->lock); | |
1056 | if (--fs->users) | |
1057 | fs = NULL; | |
1058 | spin_unlock(&req->work.fs->lock); | |
1059 | if (fs) | |
1060 | free_fs_struct(fs); | |
1061 | } | |
561fb04a JA |
1062 | } |
1063 | ||
8766dd51 | 1064 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1065 | struct io_kiocb **link) |
18d9be1a | 1066 | { |
d3656344 | 1067 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1068 | |
d3656344 JA |
1069 | if (req->flags & REQ_F_ISREG) { |
1070 | if (def->hash_reg_file) | |
8766dd51 | 1071 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1072 | } else { |
1073 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1074 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1075 | } |
cccf0ee8 JA |
1076 | |
1077 | io_req_work_grab_env(req, def); | |
54a91f3b | 1078 | |
94ae5e77 | 1079 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1080 | } |
1081 | ||
a197f664 | 1082 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1083 | { |
a197f664 | 1084 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1085 | struct io_kiocb *link; |
94ae5e77 | 1086 | |
8766dd51 | 1087 | io_prep_async_work(req, &link); |
561fb04a | 1088 | |
8766dd51 PB |
1089 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1090 | &req->work, req->flags); | |
1091 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1092 | |
1093 | if (link) | |
1094 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1095 | } |
1096 | ||
5262f567 JA |
1097 | static void io_kill_timeout(struct io_kiocb *req) |
1098 | { | |
1099 | int ret; | |
1100 | ||
2d28390a | 1101 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1102 | if (ret != -1) { |
1103 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1104 | list_del_init(&req->list); |
f0e20b89 | 1105 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1106 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1107 | io_put_req(req); |
5262f567 JA |
1108 | } |
1109 | } | |
1110 | ||
1111 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1112 | { | |
1113 | struct io_kiocb *req, *tmp; | |
1114 | ||
1115 | spin_lock_irq(&ctx->completion_lock); | |
1116 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1117 | io_kill_timeout(req); | |
1118 | spin_unlock_irq(&ctx->completion_lock); | |
1119 | } | |
1120 | ||
de0617e4 JA |
1121 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1122 | { | |
1123 | struct io_kiocb *req; | |
1124 | ||
5262f567 JA |
1125 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1126 | io_kill_timeout(req); | |
1127 | ||
de0617e4 JA |
1128 | __io_commit_cqring(ctx); |
1129 | ||
87987898 | 1130 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1131 | io_queue_async_work(req); |
de0617e4 JA |
1132 | } |
1133 | ||
2b188cc1 JA |
1134 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1135 | { | |
75b28aff | 1136 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1137 | unsigned tail; |
1138 | ||
1139 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1140 | /* |
1141 | * writes to the cq entry need to come after reading head; the | |
1142 | * control dependency is enough as we're using WRITE_ONCE to | |
1143 | * fill the cq entry | |
1144 | */ | |
75b28aff | 1145 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1146 | return NULL; |
1147 | ||
1148 | ctx->cached_cq_tail++; | |
75b28aff | 1149 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1150 | } |
1151 | ||
f2842ab5 JA |
1152 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1153 | { | |
f0b493e6 JA |
1154 | if (!ctx->cq_ev_fd) |
1155 | return false; | |
f2842ab5 JA |
1156 | if (!ctx->eventfd_async) |
1157 | return true; | |
b41e9852 | 1158 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1159 | } |
1160 | ||
b41e9852 | 1161 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1162 | { |
1163 | if (waitqueue_active(&ctx->wait)) | |
1164 | wake_up(&ctx->wait); | |
1165 | if (waitqueue_active(&ctx->sqo_wait)) | |
1166 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1167 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1168 | eventfd_signal(ctx->cq_ev_fd, 1); |
1169 | } | |
1170 | ||
c4a2ed72 JA |
1171 | /* Returns true if there are no backlogged entries after the flush */ |
1172 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1173 | { |
1174 | struct io_rings *rings = ctx->rings; | |
1175 | struct io_uring_cqe *cqe; | |
1176 | struct io_kiocb *req; | |
1177 | unsigned long flags; | |
1178 | LIST_HEAD(list); | |
1179 | ||
1180 | if (!force) { | |
1181 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1182 | return true; |
1d7bb1d5 JA |
1183 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1184 | rings->cq_ring_entries)) | |
c4a2ed72 | 1185 | return false; |
1d7bb1d5 JA |
1186 | } |
1187 | ||
1188 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1189 | ||
1190 | /* if force is set, the ring is going away. always drop after that */ | |
1191 | if (force) | |
69b3e546 | 1192 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1193 | |
c4a2ed72 | 1194 | cqe = NULL; |
1d7bb1d5 JA |
1195 | while (!list_empty(&ctx->cq_overflow_list)) { |
1196 | cqe = io_get_cqring(ctx); | |
1197 | if (!cqe && !force) | |
1198 | break; | |
1199 | ||
1200 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1201 | list); | |
1202 | list_move(&req->list, &list); | |
2ca10259 | 1203 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1204 | if (cqe) { |
1205 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1206 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1207 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1208 | } else { |
1209 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1210 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1215 | if (cqe) { |
1216 | clear_bit(0, &ctx->sq_check_overflow); | |
1217 | clear_bit(0, &ctx->cq_check_overflow); | |
1218 | } | |
1d7bb1d5 JA |
1219 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1220 | io_cqring_ev_posted(ctx); | |
1221 | ||
1222 | while (!list_empty(&list)) { | |
1223 | req = list_first_entry(&list, struct io_kiocb, list); | |
1224 | list_del(&req->list); | |
ec9c02ad | 1225 | io_put_req(req); |
1d7bb1d5 | 1226 | } |
c4a2ed72 JA |
1227 | |
1228 | return cqe != NULL; | |
1d7bb1d5 JA |
1229 | } |
1230 | ||
bcda7baa | 1231 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1232 | { |
78e19bbe | 1233 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1234 | struct io_uring_cqe *cqe; |
1235 | ||
78e19bbe | 1236 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1237 | |
2b188cc1 JA |
1238 | /* |
1239 | * If we can't get a cq entry, userspace overflowed the | |
1240 | * submission (by quite a lot). Increment the overflow count in | |
1241 | * the ring. | |
1242 | */ | |
1243 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1244 | if (likely(cqe)) { |
78e19bbe | 1245 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1246 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1247 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1248 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1249 | WRITE_ONCE(ctx->rings->cq_overflow, |
1250 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1251 | } else { |
ad3eb2c8 JA |
1252 | if (list_empty(&ctx->cq_overflow_list)) { |
1253 | set_bit(0, &ctx->sq_check_overflow); | |
1254 | set_bit(0, &ctx->cq_check_overflow); | |
1255 | } | |
2ca10259 | 1256 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1257 | refcount_inc(&req->refs); |
1258 | req->result = res; | |
bcda7baa | 1259 | req->cflags = cflags; |
1d7bb1d5 | 1260 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1261 | } |
1262 | } | |
1263 | ||
bcda7baa JA |
1264 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1265 | { | |
1266 | __io_cqring_fill_event(req, res, 0); | |
1267 | } | |
1268 | ||
1269 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1270 | { |
78e19bbe | 1271 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1272 | unsigned long flags; |
1273 | ||
1274 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1275 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1276 | io_commit_cqring(ctx); |
1277 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1278 | ||
8c838788 | 1279 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1280 | } |
1281 | ||
bcda7baa JA |
1282 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1283 | { | |
1284 | __io_cqring_add_event(req, res, 0); | |
1285 | } | |
1286 | ||
0ddf92e8 JA |
1287 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1288 | { | |
1289 | return req == (struct io_kiocb *) | |
1290 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1291 | } | |
1292 | ||
1293 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1294 | { | |
1295 | struct io_kiocb *req; | |
1296 | ||
1297 | req = ctx->fallback_req; | |
dd461af6 | 1298 | if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req)) |
0ddf92e8 JA |
1299 | return req; |
1300 | ||
1301 | return NULL; | |
1302 | } | |
1303 | ||
0553b8bd PB |
1304 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx, |
1305 | struct io_submit_state *state) | |
2b188cc1 | 1306 | { |
fd6fab2c | 1307 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1308 | struct io_kiocb *req; |
1309 | ||
2579f913 | 1310 | if (!state) { |
fd6fab2c | 1311 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1312 | if (unlikely(!req)) |
0ddf92e8 | 1313 | goto fallback; |
2579f913 JA |
1314 | } else if (!state->free_reqs) { |
1315 | size_t sz; | |
1316 | int ret; | |
1317 | ||
1318 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1319 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1320 | ||
1321 | /* | |
1322 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1323 | * retry single alloc to be on the safe side. | |
1324 | */ | |
1325 | if (unlikely(ret <= 0)) { | |
1326 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1327 | if (!state->reqs[0]) | |
0ddf92e8 | 1328 | goto fallback; |
fd6fab2c JA |
1329 | ret = 1; |
1330 | } | |
2579f913 | 1331 | state->free_reqs = ret - 1; |
6c8a3134 | 1332 | req = state->reqs[ret - 1]; |
2579f913 | 1333 | } else { |
2579f913 | 1334 | state->free_reqs--; |
6c8a3134 | 1335 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1336 | } |
1337 | ||
2579f913 | 1338 | return req; |
0ddf92e8 | 1339 | fallback: |
0553b8bd | 1340 | return io_get_fallback_req(ctx); |
2b188cc1 JA |
1341 | } |
1342 | ||
8da11c19 PB |
1343 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1344 | bool fixed) | |
1345 | { | |
1346 | if (fixed) | |
05589553 | 1347 | percpu_ref_put(req->fixed_file_refs); |
8da11c19 PB |
1348 | else |
1349 | fput(file); | |
1350 | } | |
1351 | ||
c6ca97b3 | 1352 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1353 | { |
929a3af9 PB |
1354 | if (req->flags & REQ_F_NEED_CLEANUP) |
1355 | io_cleanup_req(req); | |
1356 | ||
96fd84d8 | 1357 | kfree(req->io); |
8da11c19 PB |
1358 | if (req->file) |
1359 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
3537b6a7 JA |
1360 | if (req->task) |
1361 | put_task_struct(req->task); | |
cccf0ee8 JA |
1362 | |
1363 | io_req_work_drop_env(req); | |
def596e9 JA |
1364 | } |
1365 | ||
9e645e11 | 1366 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1367 | { |
c6ca97b3 | 1368 | __io_req_aux_free(req); |
fcb323cc | 1369 | |
fcb323cc | 1370 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1371 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1372 | unsigned long flags; |
1373 | ||
1374 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1375 | list_del(&req->inflight_entry); | |
1376 | if (waitqueue_active(&ctx->inflight_wait)) | |
1377 | wake_up(&ctx->inflight_wait); | |
1378 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1379 | } | |
2b85edfc PB |
1380 | |
1381 | percpu_ref_put(&req->ctx->refs); | |
b1e50e54 PB |
1382 | if (likely(!io_is_fallback_req(req))) |
1383 | kmem_cache_free(req_cachep, req); | |
1384 | else | |
dd461af6 | 1385 | clear_bit_unlock(0, (unsigned long *) &req->ctx->fallback_req); |
e65ef56d JA |
1386 | } |
1387 | ||
c6ca97b3 JA |
1388 | struct req_batch { |
1389 | void *reqs[IO_IOPOLL_BATCH]; | |
1390 | int to_free; | |
1391 | int need_iter; | |
1392 | }; | |
1393 | ||
1394 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1395 | { | |
1396 | if (!rb->to_free) | |
1397 | return; | |
1398 | if (rb->need_iter) { | |
1399 | int i, inflight = 0; | |
1400 | unsigned long flags; | |
1401 | ||
1402 | for (i = 0; i < rb->to_free; i++) { | |
1403 | struct io_kiocb *req = rb->reqs[i]; | |
1404 | ||
10fef4be | 1405 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1406 | req->file = NULL; |
05589553 | 1407 | percpu_ref_put(req->fixed_file_refs); |
10fef4be | 1408 | } |
c6ca97b3 JA |
1409 | if (req->flags & REQ_F_INFLIGHT) |
1410 | inflight++; | |
c6ca97b3 JA |
1411 | __io_req_aux_free(req); |
1412 | } | |
1413 | if (!inflight) | |
1414 | goto do_free; | |
1415 | ||
1416 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1417 | for (i = 0; i < rb->to_free; i++) { | |
1418 | struct io_kiocb *req = rb->reqs[i]; | |
1419 | ||
10fef4be | 1420 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1421 | list_del(&req->inflight_entry); |
1422 | if (!--inflight) | |
1423 | break; | |
1424 | } | |
1425 | } | |
1426 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1427 | ||
1428 | if (waitqueue_active(&ctx->inflight_wait)) | |
1429 | wake_up(&ctx->inflight_wait); | |
1430 | } | |
1431 | do_free: | |
1432 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
1433 | percpu_ref_put_many(&ctx->refs, rb->to_free); | |
c6ca97b3 | 1434 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1435 | } |
1436 | ||
a197f664 | 1437 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1438 | { |
a197f664 | 1439 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1440 | int ret; |
1441 | ||
2d28390a | 1442 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1443 | if (ret != -1) { |
78e19bbe | 1444 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd | 1445 | io_commit_cqring(ctx); |
dea3b49c | 1446 | req->flags &= ~REQ_F_LINK_HEAD; |
ec9c02ad | 1447 | io_put_req(req); |
2665abfd JA |
1448 | return true; |
1449 | } | |
1450 | ||
1451 | return false; | |
e65ef56d JA |
1452 | } |
1453 | ||
ba816ad6 | 1454 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1455 | { |
2665abfd | 1456 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1457 | bool wake_ev = false; |
9e645e11 | 1458 | |
4d7dd462 JA |
1459 | /* Already got next link */ |
1460 | if (req->flags & REQ_F_LINK_NEXT) | |
1461 | return; | |
1462 | ||
9e645e11 JA |
1463 | /* |
1464 | * The list should never be empty when we are called here. But could | |
1465 | * potentially happen if the chain is messed up, check to be on the | |
1466 | * safe side. | |
1467 | */ | |
4493233e PB |
1468 | while (!list_empty(&req->link_list)) { |
1469 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1470 | struct io_kiocb, link_list); | |
94ae5e77 | 1471 | |
4493233e PB |
1472 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1473 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1474 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1475 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1476 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1477 | continue; | |
1478 | } | |
9e645e11 | 1479 | |
4493233e PB |
1480 | list_del_init(&req->link_list); |
1481 | if (!list_empty(&nxt->link_list)) | |
dea3b49c | 1482 | nxt->flags |= REQ_F_LINK_HEAD; |
b18fdf71 | 1483 | *nxtptr = nxt; |
94ae5e77 | 1484 | break; |
9e645e11 | 1485 | } |
2665abfd | 1486 | |
4d7dd462 | 1487 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1488 | if (wake_ev) |
1489 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1490 | } |
1491 | ||
1492 | /* | |
dea3b49c | 1493 | * Called if REQ_F_LINK_HEAD is set, and we fail the head request |
9e645e11 JA |
1494 | */ |
1495 | static void io_fail_links(struct io_kiocb *req) | |
1496 | { | |
2665abfd | 1497 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1498 | unsigned long flags; |
1499 | ||
1500 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1501 | |
1502 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1503 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1504 | struct io_kiocb, link_list); | |
9e645e11 | 1505 | |
4493233e | 1506 | list_del_init(&link->link_list); |
c826bd7a | 1507 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1508 | |
1509 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1510 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1511 | io_link_cancel_timeout(link); |
2665abfd | 1512 | } else { |
78e19bbe | 1513 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1514 | __io_double_put_req(link); |
2665abfd | 1515 | } |
5d960724 | 1516 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1517 | } |
2665abfd JA |
1518 | |
1519 | io_commit_cqring(ctx); | |
1520 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1521 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1522 | } |
1523 | ||
4d7dd462 | 1524 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1525 | { |
dea3b49c | 1526 | if (likely(!(req->flags & REQ_F_LINK_HEAD))) |
2665abfd | 1527 | return; |
2665abfd | 1528 | |
9e645e11 JA |
1529 | /* |
1530 | * If LINK is set, we have dependent requests in this chain. If we | |
1531 | * didn't fail this request, queue the first one up, moving any other | |
1532 | * dependencies to the next request. In case of failure, fail the rest | |
1533 | * of the chain. | |
1534 | */ | |
2665abfd JA |
1535 | if (req->flags & REQ_F_FAIL_LINK) { |
1536 | io_fail_links(req); | |
7c9e7f0f JA |
1537 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1538 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1539 | struct io_ring_ctx *ctx = req->ctx; |
1540 | unsigned long flags; | |
1541 | ||
1542 | /* | |
1543 | * If this is a timeout link, we could be racing with the | |
1544 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1545 | * protect against that. |
2665abfd JA |
1546 | */ |
1547 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1548 | io_req_link_next(req, nxt); | |
1549 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1550 | } else { | |
1551 | io_req_link_next(req, nxt); | |
9e645e11 | 1552 | } |
4d7dd462 | 1553 | } |
9e645e11 | 1554 | |
c69f8dbe JL |
1555 | static void io_free_req(struct io_kiocb *req) |
1556 | { | |
944e58bf PB |
1557 | struct io_kiocb *nxt = NULL; |
1558 | ||
1559 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1560 | __io_free_req(req); |
944e58bf PB |
1561 | |
1562 | if (nxt) | |
1563 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1564 | } |
1565 | ||
7a743e22 PB |
1566 | static void io_link_work_cb(struct io_wq_work **workptr) |
1567 | { | |
18a542ff PB |
1568 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1569 | struct io_kiocb *link; | |
7a743e22 | 1570 | |
18a542ff | 1571 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1572 | io_queue_linked_timeout(link); |
1573 | io_wq_submit_work(workptr); | |
1574 | } | |
1575 | ||
1576 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1577 | { | |
1578 | struct io_kiocb *link; | |
8766dd51 PB |
1579 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1580 | ||
1581 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1582 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1583 | |
1584 | *workptr = &nxt->work; | |
1585 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1586 | if (link) |
7a743e22 | 1587 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1588 | } |
1589 | ||
ba816ad6 JA |
1590 | /* |
1591 | * Drop reference to request, return next in chain (if there is one) if this | |
1592 | * was the last reference to this request. | |
1593 | */ | |
f9bd67f6 | 1594 | __attribute__((nonnull)) |
ec9c02ad | 1595 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1596 | { |
2a44f467 JA |
1597 | if (refcount_dec_and_test(&req->refs)) { |
1598 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1599 | __io_free_req(req); |
2a44f467 | 1600 | } |
2b188cc1 JA |
1601 | } |
1602 | ||
e65ef56d JA |
1603 | static void io_put_req(struct io_kiocb *req) |
1604 | { | |
1605 | if (refcount_dec_and_test(&req->refs)) | |
1606 | io_free_req(req); | |
2b188cc1 JA |
1607 | } |
1608 | ||
e9fd9396 PB |
1609 | static void io_steal_work(struct io_kiocb *req, |
1610 | struct io_wq_work **workptr) | |
7a743e22 PB |
1611 | { |
1612 | /* | |
1613 | * It's in an io-wq worker, so there always should be at least | |
1614 | * one reference, which will be dropped in io_put_work() just | |
1615 | * after the current handler returns. | |
1616 | * | |
1617 | * It also means, that if the counter dropped to 1, then there is | |
1618 | * no asynchronous users left, so it's safe to steal the next work. | |
1619 | */ | |
7a743e22 PB |
1620 | if (refcount_read(&req->refs) == 1) { |
1621 | struct io_kiocb *nxt = NULL; | |
1622 | ||
1623 | io_req_find_next(req, &nxt); | |
1624 | if (nxt) | |
1625 | io_wq_assign_next(workptr, nxt); | |
1626 | } | |
1627 | } | |
1628 | ||
978db57e JA |
1629 | /* |
1630 | * Must only be used if we don't need to care about links, usually from | |
1631 | * within the completion handling itself. | |
1632 | */ | |
1633 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1634 | { |
1635 | /* drop both submit and complete references */ | |
1636 | if (refcount_sub_and_test(2, &req->refs)) | |
1637 | __io_free_req(req); | |
1638 | } | |
1639 | ||
978db57e JA |
1640 | static void io_double_put_req(struct io_kiocb *req) |
1641 | { | |
1642 | /* drop both submit and complete references */ | |
1643 | if (refcount_sub_and_test(2, &req->refs)) | |
1644 | io_free_req(req); | |
1645 | } | |
1646 | ||
1d7bb1d5 | 1647 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1648 | { |
84f97dc2 JA |
1649 | struct io_rings *rings = ctx->rings; |
1650 | ||
ad3eb2c8 JA |
1651 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1652 | /* | |
1653 | * noflush == true is from the waitqueue handler, just ensure | |
1654 | * we wake up the task, and the next invocation will flush the | |
1655 | * entries. We cannot safely to it from here. | |
1656 | */ | |
1657 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1658 | return -1U; | |
1d7bb1d5 | 1659 | |
ad3eb2c8 JA |
1660 | io_cqring_overflow_flush(ctx, false); |
1661 | } | |
1d7bb1d5 | 1662 | |
a3a0e43f JA |
1663 | /* See comment at the top of this file */ |
1664 | smp_rmb(); | |
ad3eb2c8 | 1665 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1666 | } |
1667 | ||
fb5ccc98 PB |
1668 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1669 | { | |
1670 | struct io_rings *rings = ctx->rings; | |
1671 | ||
1672 | /* make sure SQ entry isn't read before tail */ | |
1673 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1674 | } | |
1675 | ||
8237e045 | 1676 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1677 | { |
dea3b49c | 1678 | if ((req->flags & REQ_F_LINK_HEAD) || io_is_fallback_req(req)) |
c6ca97b3 | 1679 | return false; |
e94f141b | 1680 | |
c6ca97b3 JA |
1681 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1682 | rb->need_iter++; | |
1683 | ||
1684 | rb->reqs[rb->to_free++] = req; | |
1685 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1686 | io_free_req_many(req->ctx, rb); | |
1687 | return true; | |
e94f141b JA |
1688 | } |
1689 | ||
bcda7baa JA |
1690 | static int io_put_kbuf(struct io_kiocb *req) |
1691 | { | |
4d954c25 | 1692 | struct io_buffer *kbuf; |
bcda7baa JA |
1693 | int cflags; |
1694 | ||
4d954c25 | 1695 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1696 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1697 | cflags |= IORING_CQE_F_BUFFER; | |
1698 | req->rw.addr = 0; | |
1699 | kfree(kbuf); | |
1700 | return cflags; | |
1701 | } | |
1702 | ||
def596e9 JA |
1703 | /* |
1704 | * Find and free completed poll iocbs | |
1705 | */ | |
1706 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1707 | struct list_head *done) | |
1708 | { | |
8237e045 | 1709 | struct req_batch rb; |
def596e9 | 1710 | struct io_kiocb *req; |
def596e9 | 1711 | |
c6ca97b3 | 1712 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1713 | while (!list_empty(done)) { |
bcda7baa JA |
1714 | int cflags = 0; |
1715 | ||
def596e9 JA |
1716 | req = list_first_entry(done, struct io_kiocb, list); |
1717 | list_del(&req->list); | |
1718 | ||
bcda7baa JA |
1719 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1720 | cflags = io_put_kbuf(req); | |
1721 | ||
1722 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1723 | (*nr_events)++; |
1724 | ||
8237e045 JA |
1725 | if (refcount_dec_and_test(&req->refs) && |
1726 | !io_req_multi_free(&rb, req)) | |
1727 | io_free_req(req); | |
def596e9 | 1728 | } |
def596e9 | 1729 | |
09bb8394 | 1730 | io_commit_cqring(ctx); |
32b2244a XW |
1731 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1732 | io_cqring_ev_posted(ctx); | |
8237e045 | 1733 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1734 | } |
1735 | ||
581f9810 BM |
1736 | static void io_iopoll_queue(struct list_head *again) |
1737 | { | |
1738 | struct io_kiocb *req; | |
1739 | ||
1740 | do { | |
1741 | req = list_first_entry(again, struct io_kiocb, list); | |
1742 | list_del(&req->list); | |
1743 | refcount_inc(&req->refs); | |
1744 | io_queue_async_work(req); | |
1745 | } while (!list_empty(again)); | |
1746 | } | |
1747 | ||
def596e9 JA |
1748 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
1749 | long min) | |
1750 | { | |
1751 | struct io_kiocb *req, *tmp; | |
1752 | LIST_HEAD(done); | |
581f9810 | 1753 | LIST_HEAD(again); |
def596e9 JA |
1754 | bool spin; |
1755 | int ret; | |
1756 | ||
1757 | /* | |
1758 | * Only spin for completions if we don't have multiple devices hanging | |
1759 | * off our complete list, and we're under the requested amount. | |
1760 | */ | |
1761 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1762 | ||
1763 | ret = 0; | |
1764 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1765 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1766 | |
1767 | /* | |
581f9810 BM |
1768 | * Move completed and retryable entries to our local lists. |
1769 | * If we find a request that requires polling, break out | |
1770 | * and complete those lists first, if we have entries there. | |
def596e9 JA |
1771 | */ |
1772 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1773 | list_move_tail(&req->list, &done); | |
1774 | continue; | |
1775 | } | |
1776 | if (!list_empty(&done)) | |
1777 | break; | |
1778 | ||
581f9810 BM |
1779 | if (req->result == -EAGAIN) { |
1780 | list_move_tail(&req->list, &again); | |
1781 | continue; | |
1782 | } | |
1783 | if (!list_empty(&again)) | |
1784 | break; | |
1785 | ||
def596e9 JA |
1786 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); |
1787 | if (ret < 0) | |
1788 | break; | |
1789 | ||
1790 | if (ret && spin) | |
1791 | spin = false; | |
1792 | ret = 0; | |
1793 | } | |
1794 | ||
1795 | if (!list_empty(&done)) | |
1796 | io_iopoll_complete(ctx, nr_events, &done); | |
1797 | ||
581f9810 BM |
1798 | if (!list_empty(&again)) |
1799 | io_iopoll_queue(&again); | |
1800 | ||
def596e9 JA |
1801 | return ret; |
1802 | } | |
1803 | ||
1804 | /* | |
d195a66e | 1805 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1806 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1807 | * as a non-spinning completion check. | |
1808 | */ | |
1809 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1810 | long min) | |
1811 | { | |
08f5439f | 1812 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1813 | int ret; |
1814 | ||
1815 | ret = io_do_iopoll(ctx, nr_events, min); | |
1816 | if (ret < 0) | |
1817 | return ret; | |
1818 | if (!min || *nr_events >= min) | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | return 1; | |
1823 | } | |
1824 | ||
1825 | /* | |
1826 | * We can't just wait for polled events to come to us, we have to actively | |
1827 | * find and complete them. | |
1828 | */ | |
1829 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1830 | { | |
1831 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1832 | return; | |
1833 | ||
1834 | mutex_lock(&ctx->uring_lock); | |
1835 | while (!list_empty(&ctx->poll_list)) { | |
1836 | unsigned int nr_events = 0; | |
1837 | ||
1838 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1839 | |
1840 | /* | |
1841 | * Ensure we allow local-to-the-cpu processing to take place, | |
1842 | * in this case we need to ensure that we reap all events. | |
1843 | */ | |
1844 | cond_resched(); | |
def596e9 JA |
1845 | } |
1846 | mutex_unlock(&ctx->uring_lock); | |
1847 | } | |
1848 | ||
c7849be9 XW |
1849 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1850 | long min) | |
def596e9 | 1851 | { |
2b2ed975 | 1852 | int iters = 0, ret = 0; |
500f9fba | 1853 | |
c7849be9 XW |
1854 | /* |
1855 | * We disallow the app entering submit/complete with polling, but we | |
1856 | * still need to lock the ring to prevent racing with polled issue | |
1857 | * that got punted to a workqueue. | |
1858 | */ | |
1859 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1860 | do { |
1861 | int tmin = 0; | |
1862 | ||
a3a0e43f JA |
1863 | /* |
1864 | * Don't enter poll loop if we already have events pending. | |
1865 | * If we do, we can potentially be spinning for commands that | |
1866 | * already triggered a CQE (eg in error). | |
1867 | */ | |
1d7bb1d5 | 1868 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1869 | break; |
1870 | ||
500f9fba JA |
1871 | /* |
1872 | * If a submit got punted to a workqueue, we can have the | |
1873 | * application entering polling for a command before it gets | |
1874 | * issued. That app will hold the uring_lock for the duration | |
1875 | * of the poll right here, so we need to take a breather every | |
1876 | * now and then to ensure that the issue has a chance to add | |
1877 | * the poll to the issued list. Otherwise we can spin here | |
1878 | * forever, while the workqueue is stuck trying to acquire the | |
1879 | * very same mutex. | |
1880 | */ | |
1881 | if (!(++iters & 7)) { | |
1882 | mutex_unlock(&ctx->uring_lock); | |
1883 | mutex_lock(&ctx->uring_lock); | |
1884 | } | |
1885 | ||
def596e9 JA |
1886 | if (*nr_events < min) |
1887 | tmin = min - *nr_events; | |
1888 | ||
1889 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1890 | if (ret <= 0) | |
1891 | break; | |
1892 | ret = 0; | |
1893 | } while (min && !*nr_events && !need_resched()); | |
1894 | ||
500f9fba | 1895 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1896 | return ret; |
1897 | } | |
1898 | ||
491381ce | 1899 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1900 | { |
491381ce JA |
1901 | /* |
1902 | * Tell lockdep we inherited freeze protection from submission | |
1903 | * thread. | |
1904 | */ | |
1905 | if (req->flags & REQ_F_ISREG) { | |
1906 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1907 | |
491381ce | 1908 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1909 | } |
491381ce | 1910 | file_end_write(req->file); |
2b188cc1 JA |
1911 | } |
1912 | ||
4e88d6e7 JA |
1913 | static inline void req_set_fail_links(struct io_kiocb *req) |
1914 | { | |
1915 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1916 | req->flags |= REQ_F_FAIL_LINK; | |
1917 | } | |
1918 | ||
ba816ad6 | 1919 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1920 | { |
9adbd45d | 1921 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1922 | int cflags = 0; |
2b188cc1 | 1923 | |
491381ce JA |
1924 | if (kiocb->ki_flags & IOCB_WRITE) |
1925 | kiocb_end_write(req); | |
2b188cc1 | 1926 | |
4e88d6e7 JA |
1927 | if (res != req->result) |
1928 | req_set_fail_links(req); | |
bcda7baa JA |
1929 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1930 | cflags = io_put_kbuf(req); | |
1931 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1932 | } |
1933 | ||
1934 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1935 | { | |
9adbd45d | 1936 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1937 | |
1938 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1939 | io_put_req(req); |
2b188cc1 JA |
1940 | } |
1941 | ||
def596e9 JA |
1942 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1943 | { | |
9adbd45d | 1944 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1945 | |
491381ce JA |
1946 | if (kiocb->ki_flags & IOCB_WRITE) |
1947 | kiocb_end_write(req); | |
def596e9 | 1948 | |
4e88d6e7 JA |
1949 | if (res != req->result) |
1950 | req_set_fail_links(req); | |
9e645e11 | 1951 | req->result = res; |
def596e9 JA |
1952 | if (res != -EAGAIN) |
1953 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1954 | } | |
1955 | ||
1956 | /* | |
1957 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1958 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1959 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1960 | * accessing the kiocb cookie. | |
1961 | */ | |
1962 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1963 | { | |
1964 | struct io_ring_ctx *ctx = req->ctx; | |
1965 | ||
1966 | /* | |
1967 | * Track whether we have multiple files in our lists. This will impact | |
1968 | * how we do polling eventually, not spinning if we're on potentially | |
1969 | * different devices. | |
1970 | */ | |
1971 | if (list_empty(&ctx->poll_list)) { | |
1972 | ctx->poll_multi_file = false; | |
1973 | } else if (!ctx->poll_multi_file) { | |
1974 | struct io_kiocb *list_req; | |
1975 | ||
1976 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1977 | list); | |
9adbd45d | 1978 | if (list_req->file != req->file) |
def596e9 JA |
1979 | ctx->poll_multi_file = true; |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * For fast devices, IO may have already completed. If it has, add | |
1984 | * it to the front so we find it first. | |
1985 | */ | |
1986 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1987 | list_add(&req->list, &ctx->poll_list); | |
1988 | else | |
1989 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1990 | |
1991 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1992 | wq_has_sleeper(&ctx->sqo_wait)) | |
1993 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1994 | } |
1995 | ||
3d6770fb | 1996 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1997 | { |
3d6770fb | 1998 | if (state->file) { |
9a56a232 JA |
1999 | int diff = state->has_refs - state->used_refs; |
2000 | ||
2001 | if (diff) | |
2002 | fput_many(state->file, diff); | |
2003 | state->file = NULL; | |
2004 | } | |
2005 | } | |
2006 | ||
2007 | /* | |
2008 | * Get as many references to a file as we have IOs left in this submission, | |
2009 | * assuming most submissions are for one file, or at least that each file | |
2010 | * has more than one submission. | |
2011 | */ | |
8da11c19 | 2012 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2013 | { |
2014 | if (!state) | |
2015 | return fget(fd); | |
2016 | ||
2017 | if (state->file) { | |
2018 | if (state->fd == fd) { | |
2019 | state->used_refs++; | |
2020 | state->ios_left--; | |
2021 | return state->file; | |
2022 | } | |
3d6770fb | 2023 | io_file_put(state); |
9a56a232 JA |
2024 | } |
2025 | state->file = fget_many(fd, state->ios_left); | |
2026 | if (!state->file) | |
2027 | return NULL; | |
2028 | ||
2029 | state->fd = fd; | |
2030 | state->has_refs = state->ios_left; | |
2031 | state->used_refs = 1; | |
2032 | state->ios_left--; | |
2033 | return state->file; | |
2034 | } | |
2035 | ||
2b188cc1 JA |
2036 | /* |
2037 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2038 | * any file. For now, just ensure that anything potentially problematic is done | |
2039 | * inline. | |
2040 | */ | |
af197f50 | 2041 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2042 | { |
2043 | umode_t mode = file_inode(file)->i_mode; | |
2044 | ||
10d59345 | 2045 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2046 | return true; |
2047 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2048 | return true; | |
2049 | ||
af197f50 JA |
2050 | if (!(file->f_mode & FMODE_NOWAIT)) |
2051 | return false; | |
2052 | ||
2053 | if (rw == READ) | |
2054 | return file->f_op->read_iter != NULL; | |
2055 | ||
2056 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2057 | } |
2058 | ||
3529d8c2 JA |
2059 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2060 | bool force_nonblock) | |
2b188cc1 | 2061 | { |
def596e9 | 2062 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2063 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2064 | unsigned ioprio; |
2065 | int ret; | |
2b188cc1 | 2066 | |
491381ce JA |
2067 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2068 | req->flags |= REQ_F_ISREG; | |
2069 | ||
2b188cc1 | 2070 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2071 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2072 | req->flags |= REQ_F_CUR_POS; | |
2073 | kiocb->ki_pos = req->file->f_pos; | |
2074 | } | |
2b188cc1 | 2075 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2076 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2077 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2078 | if (unlikely(ret)) | |
2079 | return ret; | |
2b188cc1 JA |
2080 | |
2081 | ioprio = READ_ONCE(sqe->ioprio); | |
2082 | if (ioprio) { | |
2083 | ret = ioprio_check_cap(ioprio); | |
2084 | if (ret) | |
09bb8394 | 2085 | return ret; |
2b188cc1 JA |
2086 | |
2087 | kiocb->ki_ioprio = ioprio; | |
2088 | } else | |
2089 | kiocb->ki_ioprio = get_current_ioprio(); | |
2090 | ||
8449eeda | 2091 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2092 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2093 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2094 | req->flags |= REQ_F_NOWAIT; |
2095 | ||
2096 | if (force_nonblock) | |
2b188cc1 | 2097 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2098 | |
def596e9 | 2099 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2100 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2101 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2102 | return -EOPNOTSUPP; |
2b188cc1 | 2103 | |
def596e9 JA |
2104 | kiocb->ki_flags |= IOCB_HIPRI; |
2105 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2106 | req->result = 0; |
def596e9 | 2107 | } else { |
09bb8394 JA |
2108 | if (kiocb->ki_flags & IOCB_HIPRI) |
2109 | return -EINVAL; | |
def596e9 JA |
2110 | kiocb->ki_complete = io_complete_rw; |
2111 | } | |
9adbd45d | 2112 | |
3529d8c2 JA |
2113 | req->rw.addr = READ_ONCE(sqe->addr); |
2114 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2115 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2116 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2117 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2118 | return 0; |
2b188cc1 JA |
2119 | } |
2120 | ||
2121 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2122 | { | |
2123 | switch (ret) { | |
2124 | case -EIOCBQUEUED: | |
2125 | break; | |
2126 | case -ERESTARTSYS: | |
2127 | case -ERESTARTNOINTR: | |
2128 | case -ERESTARTNOHAND: | |
2129 | case -ERESTART_RESTARTBLOCK: | |
2130 | /* | |
2131 | * We can't just restart the syscall, since previously | |
2132 | * submitted sqes may already be in progress. Just fail this | |
2133 | * IO with EINTR. | |
2134 | */ | |
2135 | ret = -EINTR; | |
2136 | /* fall through */ | |
2137 | default: | |
2138 | kiocb->ki_complete(kiocb, ret, 0); | |
2139 | } | |
2140 | } | |
2141 | ||
014db007 | 2142 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2143 | { |
ba04291e JA |
2144 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2145 | ||
2146 | if (req->flags & REQ_F_CUR_POS) | |
2147 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2148 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2149 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2150 | else |
2151 | io_rw_done(kiocb, ret); | |
2152 | } | |
2153 | ||
9adbd45d | 2154 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2155 | struct iov_iter *iter) |
edafccee | 2156 | { |
9adbd45d JA |
2157 | struct io_ring_ctx *ctx = req->ctx; |
2158 | size_t len = req->rw.len; | |
edafccee JA |
2159 | struct io_mapped_ubuf *imu; |
2160 | unsigned index, buf_index; | |
2161 | size_t offset; | |
2162 | u64 buf_addr; | |
2163 | ||
2164 | /* attempt to use fixed buffers without having provided iovecs */ | |
2165 | if (unlikely(!ctx->user_bufs)) | |
2166 | return -EFAULT; | |
2167 | ||
9adbd45d | 2168 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2169 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2170 | return -EFAULT; | |
2171 | ||
2172 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2173 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2174 | buf_addr = req->rw.addr; |
edafccee JA |
2175 | |
2176 | /* overflow */ | |
2177 | if (buf_addr + len < buf_addr) | |
2178 | return -EFAULT; | |
2179 | /* not inside the mapped region */ | |
2180 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2181 | return -EFAULT; | |
2182 | ||
2183 | /* | |
2184 | * May not be a start of buffer, set size appropriately | |
2185 | * and advance us to the beginning. | |
2186 | */ | |
2187 | offset = buf_addr - imu->ubuf; | |
2188 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2189 | |
2190 | if (offset) { | |
2191 | /* | |
2192 | * Don't use iov_iter_advance() here, as it's really slow for | |
2193 | * using the latter parts of a big fixed buffer - it iterates | |
2194 | * over each segment manually. We can cheat a bit here, because | |
2195 | * we know that: | |
2196 | * | |
2197 | * 1) it's a BVEC iter, we set it up | |
2198 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2199 | * first and last bvec | |
2200 | * | |
2201 | * So just find our index, and adjust the iterator afterwards. | |
2202 | * If the offset is within the first bvec (or the whole first | |
2203 | * bvec, just use iov_iter_advance(). This makes it easier | |
2204 | * since we can just skip the first segment, which may not | |
2205 | * be PAGE_SIZE aligned. | |
2206 | */ | |
2207 | const struct bio_vec *bvec = imu->bvec; | |
2208 | ||
2209 | if (offset <= bvec->bv_len) { | |
2210 | iov_iter_advance(iter, offset); | |
2211 | } else { | |
2212 | unsigned long seg_skip; | |
2213 | ||
2214 | /* skip first vec */ | |
2215 | offset -= bvec->bv_len; | |
2216 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2217 | ||
2218 | iter->bvec = bvec + seg_skip; | |
2219 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2220 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2221 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2222 | } |
2223 | } | |
2224 | ||
5e559561 | 2225 | return len; |
edafccee JA |
2226 | } |
2227 | ||
bcda7baa JA |
2228 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2229 | { | |
2230 | if (needs_lock) | |
2231 | mutex_unlock(&ctx->uring_lock); | |
2232 | } | |
2233 | ||
2234 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2235 | { | |
2236 | /* | |
2237 | * "Normal" inline submissions always hold the uring_lock, since we | |
2238 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2239 | * The only exception is when we've detached the request and issue it | |
2240 | * from an async worker thread, grab the lock for that case. | |
2241 | */ | |
2242 | if (needs_lock) | |
2243 | mutex_lock(&ctx->uring_lock); | |
2244 | } | |
2245 | ||
2246 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2247 | int bgid, struct io_buffer *kbuf, | |
2248 | bool needs_lock) | |
2249 | { | |
2250 | struct io_buffer *head; | |
2251 | ||
2252 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2253 | return kbuf; | |
2254 | ||
2255 | io_ring_submit_lock(req->ctx, needs_lock); | |
2256 | ||
2257 | lockdep_assert_held(&req->ctx->uring_lock); | |
2258 | ||
2259 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2260 | if (head) { | |
2261 | if (!list_empty(&head->list)) { | |
2262 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2263 | list); | |
2264 | list_del(&kbuf->list); | |
2265 | } else { | |
2266 | kbuf = head; | |
2267 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2268 | } | |
2269 | if (*len > kbuf->len) | |
2270 | *len = kbuf->len; | |
2271 | } else { | |
2272 | kbuf = ERR_PTR(-ENOBUFS); | |
2273 | } | |
2274 | ||
2275 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2276 | ||
2277 | return kbuf; | |
2278 | } | |
2279 | ||
4d954c25 JA |
2280 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2281 | bool needs_lock) | |
2282 | { | |
2283 | struct io_buffer *kbuf; | |
2284 | int bgid; | |
2285 | ||
2286 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2287 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2288 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2289 | if (IS_ERR(kbuf)) | |
2290 | return kbuf; | |
2291 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2292 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2293 | return u64_to_user_ptr(kbuf->addr); | |
2294 | } | |
2295 | ||
2296 | #ifdef CONFIG_COMPAT | |
2297 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2298 | bool needs_lock) | |
2299 | { | |
2300 | struct compat_iovec __user *uiov; | |
2301 | compat_ssize_t clen; | |
2302 | void __user *buf; | |
2303 | ssize_t len; | |
2304 | ||
2305 | uiov = u64_to_user_ptr(req->rw.addr); | |
2306 | if (!access_ok(uiov, sizeof(*uiov))) | |
2307 | return -EFAULT; | |
2308 | if (__get_user(clen, &uiov->iov_len)) | |
2309 | return -EFAULT; | |
2310 | if (clen < 0) | |
2311 | return -EINVAL; | |
2312 | ||
2313 | len = clen; | |
2314 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2315 | if (IS_ERR(buf)) | |
2316 | return PTR_ERR(buf); | |
2317 | iov[0].iov_base = buf; | |
2318 | iov[0].iov_len = (compat_size_t) len; | |
2319 | return 0; | |
2320 | } | |
2321 | #endif | |
2322 | ||
2323 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2324 | bool needs_lock) | |
2325 | { | |
2326 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2327 | void __user *buf; | |
2328 | ssize_t len; | |
2329 | ||
2330 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2331 | return -EFAULT; | |
2332 | ||
2333 | len = iov[0].iov_len; | |
2334 | if (len < 0) | |
2335 | return -EINVAL; | |
2336 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2337 | if (IS_ERR(buf)) | |
2338 | return PTR_ERR(buf); | |
2339 | iov[0].iov_base = buf; | |
2340 | iov[0].iov_len = len; | |
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2345 | bool needs_lock) | |
2346 | { | |
2347 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2348 | return 0; | |
2349 | if (!req->rw.len) | |
2350 | return 0; | |
2351 | else if (req->rw.len > 1) | |
2352 | return -EINVAL; | |
2353 | ||
2354 | #ifdef CONFIG_COMPAT | |
2355 | if (req->ctx->compat) | |
2356 | return io_compat_import(req, iov, needs_lock); | |
2357 | #endif | |
2358 | ||
2359 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2360 | } | |
2361 | ||
cf6fd4bd | 2362 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2363 | struct iovec **iovec, struct iov_iter *iter, |
2364 | bool needs_lock) | |
2b188cc1 | 2365 | { |
9adbd45d JA |
2366 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2367 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2368 | ssize_t ret; |
edafccee JA |
2369 | u8 opcode; |
2370 | ||
d625c6ee | 2371 | opcode = req->opcode; |
7d009165 | 2372 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2373 | *iovec = NULL; |
9adbd45d | 2374 | return io_import_fixed(req, rw, iter); |
edafccee | 2375 | } |
2b188cc1 | 2376 | |
bcda7baa JA |
2377 | /* buffer index only valid with fixed read/write, or buffer select */ |
2378 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2379 | return -EINVAL; |
2380 | ||
3a6820f2 | 2381 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2382 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2383 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2384 | if (IS_ERR(buf)) { | |
bcda7baa | 2385 | *iovec = NULL; |
4d954c25 | 2386 | return PTR_ERR(buf); |
bcda7baa | 2387 | } |
3f9d6441 | 2388 | req->rw.len = sqe_len; |
bcda7baa JA |
2389 | } |
2390 | ||
3a6820f2 JA |
2391 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2392 | *iovec = NULL; | |
3a901598 | 2393 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2394 | } |
2395 | ||
f67676d1 JA |
2396 | if (req->io) { |
2397 | struct io_async_rw *iorw = &req->io->rw; | |
2398 | ||
2399 | *iovec = iorw->iov; | |
2400 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2401 | if (iorw->iov == iorw->fast_iov) | |
2402 | *iovec = NULL; | |
2403 | return iorw->size; | |
2404 | } | |
2405 | ||
4d954c25 JA |
2406 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2407 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2408 | if (!ret) { |
2409 | ret = (*iovec)->iov_len; | |
2410 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2411 | } | |
4d954c25 JA |
2412 | *iovec = NULL; |
2413 | return ret; | |
2414 | } | |
2415 | ||
2b188cc1 | 2416 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2417 | if (req->ctx->compat) |
2b188cc1 JA |
2418 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2419 | iovec, iter); | |
2420 | #endif | |
2421 | ||
2422 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2423 | } | |
2424 | ||
31b51510 | 2425 | /* |
32960613 JA |
2426 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2427 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2428 | */ |
32960613 JA |
2429 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2430 | struct iov_iter *iter) | |
2431 | { | |
2432 | ssize_t ret = 0; | |
2433 | ||
2434 | /* | |
2435 | * Don't support polled IO through this interface, and we can't | |
2436 | * support non-blocking either. For the latter, this just causes | |
2437 | * the kiocb to be handled from an async context. | |
2438 | */ | |
2439 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2440 | return -EOPNOTSUPP; | |
2441 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2442 | return -EAGAIN; | |
2443 | ||
2444 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2445 | struct iovec iovec; |
32960613 JA |
2446 | ssize_t nr; |
2447 | ||
311ae9e1 PB |
2448 | if (!iov_iter_is_bvec(iter)) { |
2449 | iovec = iov_iter_iovec(iter); | |
2450 | } else { | |
2451 | /* fixed buffers import bvec */ | |
2452 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2453 | + iter->iov_offset; | |
2454 | iovec.iov_len = min(iter->count, | |
2455 | iter->bvec->bv_len - iter->iov_offset); | |
2456 | } | |
2457 | ||
32960613 JA |
2458 | if (rw == READ) { |
2459 | nr = file->f_op->read(file, iovec.iov_base, | |
2460 | iovec.iov_len, &kiocb->ki_pos); | |
2461 | } else { | |
2462 | nr = file->f_op->write(file, iovec.iov_base, | |
2463 | iovec.iov_len, &kiocb->ki_pos); | |
2464 | } | |
2465 | ||
311ae9e1 PB |
2466 | if (iov_iter_is_bvec(iter)) |
2467 | kunmap(iter->bvec->bv_page); | |
2468 | ||
32960613 JA |
2469 | if (nr < 0) { |
2470 | if (!ret) | |
2471 | ret = nr; | |
2472 | break; | |
2473 | } | |
2474 | ret += nr; | |
2475 | if (nr != iovec.iov_len) | |
2476 | break; | |
2477 | iov_iter_advance(iter, nr); | |
2478 | } | |
2479 | ||
2480 | return ret; | |
2481 | } | |
2482 | ||
b7bb4f7d | 2483 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2484 | struct iovec *iovec, struct iovec *fast_iov, |
2485 | struct iov_iter *iter) | |
2486 | { | |
2487 | req->io->rw.nr_segs = iter->nr_segs; | |
2488 | req->io->rw.size = io_size; | |
2489 | req->io->rw.iov = iovec; | |
2490 | if (!req->io->rw.iov) { | |
2491 | req->io->rw.iov = req->io->rw.fast_iov; | |
45097dae XW |
2492 | if (req->io->rw.iov != fast_iov) |
2493 | memcpy(req->io->rw.iov, fast_iov, | |
2494 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2495 | } else { |
2496 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2497 | } |
2498 | } | |
2499 | ||
3d9932a8 XW |
2500 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2501 | { | |
2502 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2503 | return req->io == NULL; | |
2504 | } | |
2505 | ||
b7bb4f7d | 2506 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2507 | { |
d3656344 JA |
2508 | if (!io_op_defs[req->opcode].async_ctx) |
2509 | return 0; | |
3d9932a8 XW |
2510 | |
2511 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2512 | } |
2513 | ||
b7bb4f7d JA |
2514 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2515 | struct iovec *iovec, struct iovec *fast_iov, | |
2516 | struct iov_iter *iter) | |
2517 | { | |
980ad263 | 2518 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2519 | return 0; |
5d204bcf | 2520 | if (!req->io) { |
3d9932a8 | 2521 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2522 | return -ENOMEM; |
b7bb4f7d | 2523 | |
5d204bcf JA |
2524 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2525 | } | |
b7bb4f7d | 2526 | return 0; |
f67676d1 JA |
2527 | } |
2528 | ||
3529d8c2 JA |
2529 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2530 | bool force_nonblock) | |
f67676d1 | 2531 | { |
3529d8c2 JA |
2532 | struct io_async_ctx *io; |
2533 | struct iov_iter iter; | |
f67676d1 JA |
2534 | ssize_t ret; |
2535 | ||
3529d8c2 JA |
2536 | ret = io_prep_rw(req, sqe, force_nonblock); |
2537 | if (ret) | |
2538 | return ret; | |
f67676d1 | 2539 | |
3529d8c2 JA |
2540 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2541 | return -EBADF; | |
f67676d1 | 2542 | |
5f798bea PB |
2543 | /* either don't need iovec imported or already have it */ |
2544 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2545 | return 0; |
2546 | ||
2547 | io = req->io; | |
2548 | io->rw.iov = io->rw.fast_iov; | |
2549 | req->io = NULL; | |
bcda7baa | 2550 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2551 | req->io = io; |
2552 | if (ret < 0) | |
2553 | return ret; | |
2554 | ||
2555 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2556 | return 0; | |
f67676d1 JA |
2557 | } |
2558 | ||
014db007 | 2559 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2560 | { |
2561 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2562 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2563 | struct iov_iter iter; |
31b51510 | 2564 | size_t iov_count; |
f67676d1 | 2565 | ssize_t io_size, ret; |
2b188cc1 | 2566 | |
bcda7baa | 2567 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2568 | if (ret < 0) |
2569 | return ret; | |
2b188cc1 | 2570 | |
fd6c2e4c JA |
2571 | /* Ensure we clear previously set non-block flag */ |
2572 | if (!force_nonblock) | |
29de5f6a | 2573 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2574 | |
797f3f53 | 2575 | req->result = 0; |
f67676d1 | 2576 | io_size = ret; |
dea3b49c | 2577 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 JA |
2578 | req->result = io_size; |
2579 | ||
2580 | /* | |
2581 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2582 | * we know to async punt it even if it was opened O_NONBLOCK | |
2583 | */ | |
af197f50 | 2584 | if (force_nonblock && !io_file_supports_async(req->file, READ)) |
f67676d1 | 2585 | goto copy_iov; |
9e645e11 | 2586 | |
31b51510 | 2587 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2588 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2589 | if (!ret) { |
2590 | ssize_t ret2; | |
2591 | ||
9adbd45d JA |
2592 | if (req->file->f_op->read_iter) |
2593 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2594 | else |
9adbd45d | 2595 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2596 | |
9d93a3f5 | 2597 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2598 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2599 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2600 | } else { |
2601 | copy_iov: | |
b7bb4f7d | 2602 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2603 | inline_vecs, &iter); |
2604 | if (ret) | |
2605 | goto out_free; | |
29de5f6a | 2606 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2607 | if (!(req->flags & REQ_F_NOWAIT) && |
2608 | !file_can_poll(req->file)) | |
29de5f6a | 2609 | req->flags |= REQ_F_MUST_PUNT; |
f67676d1 JA |
2610 | return -EAGAIN; |
2611 | } | |
2b188cc1 | 2612 | } |
f67676d1 | 2613 | out_free: |
1e95081c | 2614 | kfree(iovec); |
99bc4c38 | 2615 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2616 | return ret; |
2617 | } | |
2618 | ||
3529d8c2 JA |
2619 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2620 | bool force_nonblock) | |
f67676d1 | 2621 | { |
3529d8c2 JA |
2622 | struct io_async_ctx *io; |
2623 | struct iov_iter iter; | |
f67676d1 JA |
2624 | ssize_t ret; |
2625 | ||
3529d8c2 JA |
2626 | ret = io_prep_rw(req, sqe, force_nonblock); |
2627 | if (ret) | |
2628 | return ret; | |
f67676d1 | 2629 | |
3529d8c2 JA |
2630 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2631 | return -EBADF; | |
f67676d1 | 2632 | |
4ed734b0 JA |
2633 | req->fsize = rlimit(RLIMIT_FSIZE); |
2634 | ||
5f798bea PB |
2635 | /* either don't need iovec imported or already have it */ |
2636 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2637 | return 0; |
2638 | ||
2639 | io = req->io; | |
2640 | io->rw.iov = io->rw.fast_iov; | |
2641 | req->io = NULL; | |
bcda7baa | 2642 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2643 | req->io = io; |
2644 | if (ret < 0) | |
2645 | return ret; | |
2646 | ||
2647 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2648 | return 0; | |
f67676d1 JA |
2649 | } |
2650 | ||
014db007 | 2651 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2652 | { |
2653 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2654 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2655 | struct iov_iter iter; |
31b51510 | 2656 | size_t iov_count; |
f67676d1 | 2657 | ssize_t ret, io_size; |
2b188cc1 | 2658 | |
bcda7baa | 2659 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2660 | if (ret < 0) |
2661 | return ret; | |
2b188cc1 | 2662 | |
fd6c2e4c JA |
2663 | /* Ensure we clear previously set non-block flag */ |
2664 | if (!force_nonblock) | |
9adbd45d | 2665 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2666 | |
797f3f53 | 2667 | req->result = 0; |
f67676d1 | 2668 | io_size = ret; |
dea3b49c | 2669 | if (req->flags & REQ_F_LINK_HEAD) |
f67676d1 | 2670 | req->result = io_size; |
9e645e11 | 2671 | |
f67676d1 JA |
2672 | /* |
2673 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2674 | * we know to async punt it even if it was opened O_NONBLOCK | |
2675 | */ | |
af197f50 | 2676 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 2677 | goto copy_iov; |
31b51510 | 2678 | |
10d59345 JA |
2679 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2680 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2681 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2682 | goto copy_iov; |
31b51510 | 2683 | |
f67676d1 | 2684 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2685 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2686 | if (!ret) { |
9bf7933f RP |
2687 | ssize_t ret2; |
2688 | ||
2b188cc1 JA |
2689 | /* |
2690 | * Open-code file_start_write here to grab freeze protection, | |
2691 | * which will be released by another thread in | |
2692 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2693 | * released so that it doesn't complain about the held lock when | |
2694 | * we return to userspace. | |
2695 | */ | |
491381ce | 2696 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2697 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2698 | SB_FREEZE_WRITE, true); |
9adbd45d | 2699 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2700 | SB_FREEZE_WRITE); |
2701 | } | |
2702 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2703 | |
4ed734b0 JA |
2704 | if (!force_nonblock) |
2705 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2706 | ||
9adbd45d JA |
2707 | if (req->file->f_op->write_iter) |
2708 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2709 | else |
9adbd45d | 2710 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2711 | |
2712 | if (!force_nonblock) | |
2713 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2714 | ||
faac996c | 2715 | /* |
bff6035d | 2716 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2717 | * retry them without IOCB_NOWAIT. |
2718 | */ | |
2719 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2720 | ret2 = -EAGAIN; | |
f67676d1 | 2721 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2722 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2723 | } else { |
2724 | copy_iov: | |
b7bb4f7d | 2725 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2726 | inline_vecs, &iter); |
2727 | if (ret) | |
2728 | goto out_free; | |
29de5f6a | 2729 | /* any defer here is final, must blocking retry */ |
490e8967 JA |
2730 | if (!file_can_poll(req->file)) |
2731 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2732 | return -EAGAIN; |
2733 | } | |
2b188cc1 | 2734 | } |
31b51510 | 2735 | out_free: |
99bc4c38 | 2736 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2737 | kfree(iovec); |
2b188cc1 JA |
2738 | return ret; |
2739 | } | |
2740 | ||
7d67af2c PB |
2741 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2742 | { | |
2743 | struct io_splice* sp = &req->splice; | |
2744 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2745 | int ret; | |
2746 | ||
2747 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2748 | return 0; | |
2749 | ||
2750 | sp->file_in = NULL; | |
2751 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2752 | sp->off_out = READ_ONCE(sqe->off); | |
2753 | sp->len = READ_ONCE(sqe->len); | |
2754 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2755 | ||
2756 | if (unlikely(sp->flags & ~valid_flags)) | |
2757 | return -EINVAL; | |
2758 | ||
2759 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2760 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2761 | if (ret) | |
2762 | return ret; | |
2763 | req->flags |= REQ_F_NEED_CLEANUP; | |
2764 | ||
2765 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2766 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2767 | ||
2768 | return 0; | |
2769 | } | |
2770 | ||
af197f50 | 2771 | static bool io_splice_punt(struct file *file, int rw) |
7d67af2c PB |
2772 | { |
2773 | if (get_pipe_info(file)) | |
2774 | return false; | |
af197f50 | 2775 | if (!io_file_supports_async(file, rw)) |
7d67af2c | 2776 | return true; |
88357580 | 2777 | return !(file->f_flags & O_NONBLOCK); |
7d67af2c PB |
2778 | } |
2779 | ||
014db007 | 2780 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2781 | { |
2782 | struct io_splice *sp = &req->splice; | |
2783 | struct file *in = sp->file_in; | |
2784 | struct file *out = sp->file_out; | |
2785 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2786 | loff_t *poff_in, *poff_out; | |
2787 | long ret; | |
2788 | ||
2789 | if (force_nonblock) { | |
af197f50 | 2790 | if (io_splice_punt(in, READ) || io_splice_punt(out, WRITE)) |
7d67af2c PB |
2791 | return -EAGAIN; |
2792 | flags |= SPLICE_F_NONBLOCK; | |
2793 | } | |
2794 | ||
2795 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2796 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2797 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2798 | if (force_nonblock && ret == -EAGAIN) | |
2799 | return -EAGAIN; | |
2800 | ||
2801 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2802 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2803 | ||
2804 | io_cqring_add_event(req, ret); | |
2805 | if (ret != sp->len) | |
2806 | req_set_fail_links(req); | |
014db007 | 2807 | io_put_req(req); |
7d67af2c PB |
2808 | return 0; |
2809 | } | |
2810 | ||
2b188cc1 JA |
2811 | /* |
2812 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2813 | */ | |
78e19bbe | 2814 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2815 | { |
2816 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2817 | |
def596e9 JA |
2818 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2819 | return -EINVAL; | |
2820 | ||
78e19bbe | 2821 | io_cqring_add_event(req, 0); |
e65ef56d | 2822 | io_put_req(req); |
2b188cc1 JA |
2823 | return 0; |
2824 | } | |
2825 | ||
3529d8c2 | 2826 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2827 | { |
6b06314c | 2828 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2829 | |
09bb8394 JA |
2830 | if (!req->file) |
2831 | return -EBADF; | |
c992fe29 | 2832 | |
6b06314c | 2833 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2834 | return -EINVAL; |
edafccee | 2835 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2836 | return -EINVAL; |
2837 | ||
8ed8d3c3 JA |
2838 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2839 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2840 | return -EINVAL; | |
2841 | ||
2842 | req->sync.off = READ_ONCE(sqe->off); | |
2843 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2844 | return 0; |
2845 | } | |
2846 | ||
8ed8d3c3 JA |
2847 | static bool io_req_cancelled(struct io_kiocb *req) |
2848 | { | |
2849 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2850 | req_set_fail_links(req); | |
2851 | io_cqring_add_event(req, -ECANCELED); | |
2852 | io_put_req(req); | |
2853 | return true; | |
2854 | } | |
2855 | ||
2856 | return false; | |
2857 | } | |
2858 | ||
014db007 | 2859 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2860 | { |
8ed8d3c3 | 2861 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2862 | int ret; |
2863 | ||
9adbd45d | 2864 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2865 | end > 0 ? end : LLONG_MAX, |
2866 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2867 | if (ret < 0) | |
2868 | req_set_fail_links(req); | |
2869 | io_cqring_add_event(req, ret); | |
014db007 | 2870 | io_put_req(req); |
8ed8d3c3 JA |
2871 | } |
2872 | ||
5ea62161 | 2873 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2874 | { |
5ea62161 | 2875 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2876 | |
2877 | if (io_req_cancelled(req)) | |
2878 | return; | |
014db007 | 2879 | __io_fsync(req); |
e9fd9396 | 2880 | io_steal_work(req, workptr); |
8ed8d3c3 | 2881 | } |
c992fe29 | 2882 | |
014db007 | 2883 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2884 | { |
c992fe29 | 2885 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2886 | if (force_nonblock) { |
8ed8d3c3 | 2887 | req->work.func = io_fsync_finish; |
c992fe29 | 2888 | return -EAGAIN; |
8ed8d3c3 | 2889 | } |
014db007 | 2890 | __io_fsync(req); |
c992fe29 CH |
2891 | return 0; |
2892 | } | |
2893 | ||
014db007 | 2894 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2895 | { |
8ed8d3c3 JA |
2896 | int ret; |
2897 | ||
4ed734b0 | 2898 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2899 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2900 | req->sync.len); | |
4ed734b0 | 2901 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2902 | if (ret < 0) |
2903 | req_set_fail_links(req); | |
2904 | io_cqring_add_event(req, ret); | |
014db007 | 2905 | io_put_req(req); |
5ea62161 PB |
2906 | } |
2907 | ||
2908 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2909 | { | |
2910 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2911 | |
594506fe PB |
2912 | if (io_req_cancelled(req)) |
2913 | return; | |
014db007 | 2914 | __io_fallocate(req); |
e9fd9396 | 2915 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2916 | } |
2917 | ||
d63d1b5e JA |
2918 | static int io_fallocate_prep(struct io_kiocb *req, |
2919 | const struct io_uring_sqe *sqe) | |
2920 | { | |
2921 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2922 | return -EINVAL; | |
2923 | ||
2924 | req->sync.off = READ_ONCE(sqe->off); | |
2925 | req->sync.len = READ_ONCE(sqe->addr); | |
2926 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2927 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2928 | return 0; |
2929 | } | |
2930 | ||
014db007 | 2931 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2932 | { |
d63d1b5e | 2933 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2934 | if (force_nonblock) { |
d63d1b5e | 2935 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2936 | return -EAGAIN; |
8ed8d3c3 | 2937 | } |
5d17b4a4 | 2938 | |
014db007 | 2939 | __io_fallocate(req); |
5d17b4a4 JA |
2940 | return 0; |
2941 | } | |
2942 | ||
15b71abe | 2943 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2944 | { |
f8748881 | 2945 | const char __user *fname; |
15b71abe | 2946 | int ret; |
b7bb4f7d | 2947 | |
15b71abe JA |
2948 | if (sqe->ioprio || sqe->buf_index) |
2949 | return -EINVAL; | |
9c280f90 | 2950 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2951 | return -EBADF; |
0bdbdd08 PB |
2952 | if (req->flags & REQ_F_NEED_CLEANUP) |
2953 | return 0; | |
03b1230c | 2954 | |
15b71abe | 2955 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2956 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2957 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2958 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
08a1d26e JA |
2959 | if (force_o_largefile()) |
2960 | req->open.how.flags |= O_LARGEFILE; | |
3529d8c2 | 2961 | |
f8748881 | 2962 | req->open.filename = getname(fname); |
15b71abe JA |
2963 | if (IS_ERR(req->open.filename)) { |
2964 | ret = PTR_ERR(req->open.filename); | |
2965 | req->open.filename = NULL; | |
2966 | return ret; | |
2967 | } | |
3529d8c2 | 2968 | |
4022e7af | 2969 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2970 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2971 | return 0; |
03b1230c JA |
2972 | } |
2973 | ||
cebdb986 | 2974 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2975 | { |
cebdb986 JA |
2976 | struct open_how __user *how; |
2977 | const char __user *fname; | |
2978 | size_t len; | |
0fa03c62 JA |
2979 | int ret; |
2980 | ||
cebdb986 | 2981 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2982 | return -EINVAL; |
9c280f90 | 2983 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 2984 | return -EBADF; |
0bdbdd08 PB |
2985 | if (req->flags & REQ_F_NEED_CLEANUP) |
2986 | return 0; | |
0fa03c62 | 2987 | |
cebdb986 JA |
2988 | req->open.dfd = READ_ONCE(sqe->fd); |
2989 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2990 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2991 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2992 | |
cebdb986 JA |
2993 | if (len < OPEN_HOW_SIZE_VER0) |
2994 | return -EINVAL; | |
3529d8c2 | 2995 | |
cebdb986 JA |
2996 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2997 | len); | |
2998 | if (ret) | |
2999 | return ret; | |
3529d8c2 | 3000 | |
cebdb986 JA |
3001 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
3002 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 3003 | |
cebdb986 JA |
3004 | req->open.filename = getname(fname); |
3005 | if (IS_ERR(req->open.filename)) { | |
3006 | ret = PTR_ERR(req->open.filename); | |
3007 | req->open.filename = NULL; | |
3008 | return ret; | |
3009 | } | |
3010 | ||
4022e7af | 3011 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3012 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
3013 | return 0; |
3014 | } | |
3015 | ||
014db007 | 3016 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
3017 | { |
3018 | struct open_flags op; | |
15b71abe JA |
3019 | struct file *file; |
3020 | int ret; | |
3021 | ||
f86cd20c | 3022 | if (force_nonblock) |
15b71abe | 3023 | return -EAGAIN; |
15b71abe | 3024 | |
cebdb986 | 3025 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3026 | if (ret) |
3027 | goto err; | |
3028 | ||
4022e7af | 3029 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3030 | if (ret < 0) |
3031 | goto err; | |
3032 | ||
3033 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3034 | if (IS_ERR(file)) { | |
3035 | put_unused_fd(ret); | |
3036 | ret = PTR_ERR(file); | |
3037 | } else { | |
3038 | fsnotify_open(file); | |
3039 | fd_install(ret, file); | |
3040 | } | |
3041 | err: | |
3042 | putname(req->open.filename); | |
8fef80bf | 3043 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3044 | if (ret < 0) |
3045 | req_set_fail_links(req); | |
3046 | io_cqring_add_event(req, ret); | |
014db007 | 3047 | io_put_req(req); |
15b71abe JA |
3048 | return 0; |
3049 | } | |
3050 | ||
014db007 | 3051 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3052 | { |
3053 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3054 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3055 | } |
3056 | ||
067524e9 JA |
3057 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3058 | const struct io_uring_sqe *sqe) | |
3059 | { | |
3060 | struct io_provide_buf *p = &req->pbuf; | |
3061 | u64 tmp; | |
3062 | ||
3063 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3064 | return -EINVAL; | |
3065 | ||
3066 | tmp = READ_ONCE(sqe->fd); | |
3067 | if (!tmp || tmp > USHRT_MAX) | |
3068 | return -EINVAL; | |
3069 | ||
3070 | memset(p, 0, sizeof(*p)); | |
3071 | p->nbufs = tmp; | |
3072 | p->bgid = READ_ONCE(sqe->buf_group); | |
3073 | return 0; | |
3074 | } | |
3075 | ||
3076 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3077 | int bgid, unsigned nbufs) | |
3078 | { | |
3079 | unsigned i = 0; | |
3080 | ||
3081 | /* shouldn't happen */ | |
3082 | if (!nbufs) | |
3083 | return 0; | |
3084 | ||
3085 | /* the head kbuf is the list itself */ | |
3086 | while (!list_empty(&buf->list)) { | |
3087 | struct io_buffer *nxt; | |
3088 | ||
3089 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3090 | list_del(&nxt->list); | |
3091 | kfree(nxt); | |
3092 | if (++i == nbufs) | |
3093 | return i; | |
3094 | } | |
3095 | i++; | |
3096 | kfree(buf); | |
3097 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3098 | ||
3099 | return i; | |
3100 | } | |
3101 | ||
3102 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3103 | { | |
3104 | struct io_provide_buf *p = &req->pbuf; | |
3105 | struct io_ring_ctx *ctx = req->ctx; | |
3106 | struct io_buffer *head; | |
3107 | int ret = 0; | |
3108 | ||
3109 | io_ring_submit_lock(ctx, !force_nonblock); | |
3110 | ||
3111 | lockdep_assert_held(&ctx->uring_lock); | |
3112 | ||
3113 | ret = -ENOENT; | |
3114 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3115 | if (head) | |
3116 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3117 | ||
3118 | io_ring_submit_lock(ctx, !force_nonblock); | |
3119 | if (ret < 0) | |
3120 | req_set_fail_links(req); | |
3121 | io_cqring_add_event(req, ret); | |
3122 | io_put_req(req); | |
3123 | return 0; | |
3124 | } | |
3125 | ||
ddf0322d JA |
3126 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3127 | const struct io_uring_sqe *sqe) | |
3128 | { | |
3129 | struct io_provide_buf *p = &req->pbuf; | |
3130 | u64 tmp; | |
3131 | ||
3132 | if (sqe->ioprio || sqe->rw_flags) | |
3133 | return -EINVAL; | |
3134 | ||
3135 | tmp = READ_ONCE(sqe->fd); | |
3136 | if (!tmp || tmp > USHRT_MAX) | |
3137 | return -E2BIG; | |
3138 | p->nbufs = tmp; | |
3139 | p->addr = READ_ONCE(sqe->addr); | |
3140 | p->len = READ_ONCE(sqe->len); | |
3141 | ||
3142 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3143 | return -EFAULT; | |
3144 | ||
3145 | p->bgid = READ_ONCE(sqe->buf_group); | |
3146 | tmp = READ_ONCE(sqe->off); | |
3147 | if (tmp > USHRT_MAX) | |
3148 | return -E2BIG; | |
3149 | p->bid = tmp; | |
3150 | return 0; | |
3151 | } | |
3152 | ||
3153 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3154 | { | |
3155 | struct io_buffer *buf; | |
3156 | u64 addr = pbuf->addr; | |
3157 | int i, bid = pbuf->bid; | |
3158 | ||
3159 | for (i = 0; i < pbuf->nbufs; i++) { | |
3160 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3161 | if (!buf) | |
3162 | break; | |
3163 | ||
3164 | buf->addr = addr; | |
3165 | buf->len = pbuf->len; | |
3166 | buf->bid = bid; | |
3167 | addr += pbuf->len; | |
3168 | bid++; | |
3169 | if (!*head) { | |
3170 | INIT_LIST_HEAD(&buf->list); | |
3171 | *head = buf; | |
3172 | } else { | |
3173 | list_add_tail(&buf->list, &(*head)->list); | |
3174 | } | |
3175 | } | |
3176 | ||
3177 | return i ? i : -ENOMEM; | |
3178 | } | |
3179 | ||
ddf0322d JA |
3180 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3181 | { | |
3182 | struct io_provide_buf *p = &req->pbuf; | |
3183 | struct io_ring_ctx *ctx = req->ctx; | |
3184 | struct io_buffer *head, *list; | |
3185 | int ret = 0; | |
3186 | ||
3187 | io_ring_submit_lock(ctx, !force_nonblock); | |
3188 | ||
3189 | lockdep_assert_held(&ctx->uring_lock); | |
3190 | ||
3191 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3192 | ||
3193 | ret = io_add_buffers(p, &head); | |
3194 | if (ret < 0) | |
3195 | goto out; | |
3196 | ||
3197 | if (!list) { | |
3198 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3199 | GFP_KERNEL); | |
3200 | if (ret < 0) { | |
067524e9 | 3201 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3202 | goto out; |
3203 | } | |
3204 | } | |
3205 | out: | |
3206 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3207 | if (ret < 0) | |
3208 | req_set_fail_links(req); | |
3209 | io_cqring_add_event(req, ret); | |
3210 | io_put_req(req); | |
3211 | return 0; | |
cebdb986 JA |
3212 | } |
3213 | ||
3e4827b0 JA |
3214 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3215 | const struct io_uring_sqe *sqe) | |
3216 | { | |
3217 | #if defined(CONFIG_EPOLL) | |
3218 | if (sqe->ioprio || sqe->buf_index) | |
3219 | return -EINVAL; | |
3220 | ||
3221 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3222 | req->epoll.op = READ_ONCE(sqe->len); | |
3223 | req->epoll.fd = READ_ONCE(sqe->off); | |
3224 | ||
3225 | if (ep_op_has_event(req->epoll.op)) { | |
3226 | struct epoll_event __user *ev; | |
3227 | ||
3228 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3229 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3230 | return -EFAULT; | |
3231 | } | |
3232 | ||
3233 | return 0; | |
3234 | #else | |
3235 | return -EOPNOTSUPP; | |
3236 | #endif | |
3237 | } | |
3238 | ||
014db007 | 3239 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3240 | { |
3241 | #if defined(CONFIG_EPOLL) | |
3242 | struct io_epoll *ie = &req->epoll; | |
3243 | int ret; | |
3244 | ||
3245 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3246 | if (force_nonblock && ret == -EAGAIN) | |
3247 | return -EAGAIN; | |
3248 | ||
3249 | if (ret < 0) | |
3250 | req_set_fail_links(req); | |
3251 | io_cqring_add_event(req, ret); | |
014db007 | 3252 | io_put_req(req); |
3e4827b0 JA |
3253 | return 0; |
3254 | #else | |
3255 | return -EOPNOTSUPP; | |
3256 | #endif | |
3257 | } | |
3258 | ||
c1ca757b JA |
3259 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3260 | { | |
3261 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3262 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3263 | return -EINVAL; | |
3264 | ||
3265 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3266 | req->madvise.len = READ_ONCE(sqe->len); | |
3267 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3268 | return 0; | |
3269 | #else | |
3270 | return -EOPNOTSUPP; | |
3271 | #endif | |
3272 | } | |
3273 | ||
014db007 | 3274 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3275 | { |
3276 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3277 | struct io_madvise *ma = &req->madvise; | |
3278 | int ret; | |
3279 | ||
3280 | if (force_nonblock) | |
3281 | return -EAGAIN; | |
3282 | ||
3283 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3284 | if (ret < 0) | |
3285 | req_set_fail_links(req); | |
3286 | io_cqring_add_event(req, ret); | |
014db007 | 3287 | io_put_req(req); |
c1ca757b JA |
3288 | return 0; |
3289 | #else | |
3290 | return -EOPNOTSUPP; | |
3291 | #endif | |
3292 | } | |
3293 | ||
4840e418 JA |
3294 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3295 | { | |
3296 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3297 | return -EINVAL; | |
3298 | ||
3299 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3300 | req->fadvise.len = READ_ONCE(sqe->len); | |
3301 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3302 | return 0; | |
3303 | } | |
3304 | ||
014db007 | 3305 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3306 | { |
3307 | struct io_fadvise *fa = &req->fadvise; | |
3308 | int ret; | |
3309 | ||
3e69426d JA |
3310 | if (force_nonblock) { |
3311 | switch (fa->advice) { | |
3312 | case POSIX_FADV_NORMAL: | |
3313 | case POSIX_FADV_RANDOM: | |
3314 | case POSIX_FADV_SEQUENTIAL: | |
3315 | break; | |
3316 | default: | |
3317 | return -EAGAIN; | |
3318 | } | |
3319 | } | |
4840e418 JA |
3320 | |
3321 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3322 | if (ret < 0) | |
3323 | req_set_fail_links(req); | |
3324 | io_cqring_add_event(req, ret); | |
014db007 | 3325 | io_put_req(req); |
4840e418 JA |
3326 | return 0; |
3327 | } | |
3328 | ||
eddc7ef5 JA |
3329 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3330 | { | |
f8748881 | 3331 | const char __user *fname; |
eddc7ef5 JA |
3332 | unsigned lookup_flags; |
3333 | int ret; | |
3334 | ||
3335 | if (sqe->ioprio || sqe->buf_index) | |
3336 | return -EINVAL; | |
9c280f90 | 3337 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3338 | return -EBADF; |
0bdbdd08 PB |
3339 | if (req->flags & REQ_F_NEED_CLEANUP) |
3340 | return 0; | |
eddc7ef5 JA |
3341 | |
3342 | req->open.dfd = READ_ONCE(sqe->fd); | |
3343 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3344 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3345 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3346 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3347 | |
c12cedf2 | 3348 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3349 | return -EINVAL; |
3350 | ||
f8748881 | 3351 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3352 | if (IS_ERR(req->open.filename)) { |
3353 | ret = PTR_ERR(req->open.filename); | |
3354 | req->open.filename = NULL; | |
3355 | return ret; | |
3356 | } | |
3357 | ||
8fef80bf | 3358 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3359 | return 0; |
3360 | } | |
3361 | ||
014db007 | 3362 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3363 | { |
3364 | struct io_open *ctx = &req->open; | |
3365 | unsigned lookup_flags; | |
3366 | struct path path; | |
3367 | struct kstat stat; | |
3368 | int ret; | |
3369 | ||
5b0bbee4 JA |
3370 | if (force_nonblock) { |
3371 | /* only need file table for an actual valid fd */ | |
3372 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
3373 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 3374 | return -EAGAIN; |
5b0bbee4 | 3375 | } |
eddc7ef5 | 3376 | |
c12cedf2 | 3377 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3378 | return -EINVAL; |
3379 | ||
3380 | retry: | |
3381 | /* filename_lookup() drops it, keep a reference */ | |
3382 | ctx->filename->refcnt++; | |
3383 | ||
3384 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3385 | NULL); | |
3386 | if (ret) | |
3387 | goto err; | |
3388 | ||
c12cedf2 | 3389 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3390 | path_put(&path); |
3391 | if (retry_estale(ret, lookup_flags)) { | |
3392 | lookup_flags |= LOOKUP_REVAL; | |
3393 | goto retry; | |
3394 | } | |
3395 | if (!ret) | |
3396 | ret = cp_statx(&stat, ctx->buffer); | |
3397 | err: | |
3398 | putname(ctx->filename); | |
8fef80bf | 3399 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3400 | if (ret < 0) |
3401 | req_set_fail_links(req); | |
3402 | io_cqring_add_event(req, ret); | |
014db007 | 3403 | io_put_req(req); |
eddc7ef5 JA |
3404 | return 0; |
3405 | } | |
3406 | ||
b5dba59e JA |
3407 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3408 | { | |
3409 | /* | |
3410 | * If we queue this for async, it must not be cancellable. That would | |
3411 | * leave the 'file' in an undeterminate state. | |
3412 | */ | |
3413 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3414 | ||
3415 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3416 | sqe->rw_flags || sqe->buf_index) | |
3417 | return -EINVAL; | |
9c280f90 | 3418 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 3419 | return -EBADF; |
b5dba59e JA |
3420 | |
3421 | req->close.fd = READ_ONCE(sqe->fd); | |
3422 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3423 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3424 | return -EBADF; |
3425 | ||
3426 | return 0; | |
3427 | } | |
3428 | ||
a93b3331 | 3429 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3430 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3431 | { |
3432 | int ret; | |
3433 | ||
3434 | ret = filp_close(req->close.put_file, req->work.files); | |
3435 | if (ret < 0) | |
3436 | req_set_fail_links(req); | |
3437 | io_cqring_add_event(req, ret); | |
3438 | fput(req->close.put_file); | |
014db007 | 3439 | io_put_req(req); |
a93b3331 PB |
3440 | } |
3441 | ||
b5dba59e JA |
3442 | static void io_close_finish(struct io_wq_work **workptr) |
3443 | { | |
3444 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3445 | |
7fbeb95d | 3446 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3447 | __io_close_finish(req); |
e9fd9396 | 3448 | io_steal_work(req, workptr); |
b5dba59e JA |
3449 | } |
3450 | ||
014db007 | 3451 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3452 | { |
3453 | int ret; | |
3454 | ||
3455 | req->close.put_file = NULL; | |
3456 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3457 | if (ret < 0) | |
3458 | return ret; | |
3459 | ||
3460 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3461 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3462 | /* submission ref will be dropped, take it for async */ |
3463 | refcount_inc(&req->refs); | |
3464 | ||
a2100672 PB |
3465 | req->work.func = io_close_finish; |
3466 | /* | |
3467 | * Do manual async queue here to avoid grabbing files - we don't | |
3468 | * need the files, and it'll cause io_close_finish() to close | |
3469 | * the file again and cause a double CQE entry for this request | |
3470 | */ | |
3471 | io_queue_async_work(req); | |
3472 | return 0; | |
3473 | } | |
b5dba59e JA |
3474 | |
3475 | /* | |
3476 | * No ->flush(), safely close from here and just punt the | |
3477 | * fput() to async context. | |
3478 | */ | |
014db007 | 3479 | __io_close_finish(req); |
1a417f4e | 3480 | return 0; |
b5dba59e JA |
3481 | } |
3482 | ||
3529d8c2 | 3483 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3484 | { |
3485 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3486 | |
3487 | if (!req->file) | |
3488 | return -EBADF; | |
5d17b4a4 JA |
3489 | |
3490 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3491 | return -EINVAL; | |
3492 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3493 | return -EINVAL; | |
3494 | ||
8ed8d3c3 JA |
3495 | req->sync.off = READ_ONCE(sqe->off); |
3496 | req->sync.len = READ_ONCE(sqe->len); | |
3497 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3498 | return 0; |
3499 | } | |
3500 | ||
014db007 | 3501 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3502 | { |
8ed8d3c3 JA |
3503 | int ret; |
3504 | ||
9adbd45d | 3505 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3506 | req->sync.flags); |
3507 | if (ret < 0) | |
3508 | req_set_fail_links(req); | |
3509 | io_cqring_add_event(req, ret); | |
014db007 | 3510 | io_put_req(req); |
5ea62161 PB |
3511 | } |
3512 | ||
3513 | ||
3514 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3515 | { | |
3516 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 PB |
3517 | |
3518 | if (io_req_cancelled(req)) | |
3519 | return; | |
014db007 | 3520 | __io_sync_file_range(req); |
594506fe | 3521 | io_put_req(req); /* put submission ref */ |
5d17b4a4 JA |
3522 | } |
3523 | ||
014db007 | 3524 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3525 | { |
5d17b4a4 | 3526 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3527 | if (force_nonblock) { |
8ed8d3c3 | 3528 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3529 | return -EAGAIN; |
8ed8d3c3 | 3530 | } |
5d17b4a4 | 3531 | |
014db007 | 3532 | __io_sync_file_range(req); |
5d17b4a4 JA |
3533 | return 0; |
3534 | } | |
3535 | ||
469956e8 | 3536 | #if defined(CONFIG_NET) |
02d27d89 PB |
3537 | static int io_setup_async_msg(struct io_kiocb *req, |
3538 | struct io_async_msghdr *kmsg) | |
3539 | { | |
3540 | if (req->io) | |
3541 | return -EAGAIN; | |
3542 | if (io_alloc_async_ctx(req)) { | |
3543 | if (kmsg->iov != kmsg->fast_iov) | |
3544 | kfree(kmsg->iov); | |
3545 | return -ENOMEM; | |
3546 | } | |
3547 | req->flags |= REQ_F_NEED_CLEANUP; | |
3548 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3549 | return -EAGAIN; | |
3550 | } | |
3551 | ||
3529d8c2 | 3552 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3553 | { |
e47293fd | 3554 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3555 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3556 | int ret; |
03b1230c | 3557 | |
e47293fd JA |
3558 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3559 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3560 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3561 | |
d8768362 JA |
3562 | #ifdef CONFIG_COMPAT |
3563 | if (req->ctx->compat) | |
3564 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3565 | #endif | |
3566 | ||
fddaface | 3567 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3568 | return 0; |
5f798bea PB |
3569 | /* iovec is already imported */ |
3570 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3571 | return 0; | |
3529d8c2 | 3572 | |
d9688565 | 3573 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3574 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3575 | &io->msg.iov); |
99bc4c38 PB |
3576 | if (!ret) |
3577 | req->flags |= REQ_F_NEED_CLEANUP; | |
3578 | return ret; | |
03b1230c JA |
3579 | } |
3580 | ||
014db007 | 3581 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3582 | { |
0b416c3e | 3583 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3584 | struct socket *sock; |
3585 | int ret; | |
3586 | ||
3587 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3588 | return -EINVAL; | |
3589 | ||
3590 | sock = sock_from_file(req->file, &ret); | |
3591 | if (sock) { | |
b7bb4f7d | 3592 | struct io_async_ctx io; |
0fa03c62 JA |
3593 | unsigned flags; |
3594 | ||
03b1230c | 3595 | if (req->io) { |
0b416c3e | 3596 | kmsg = &req->io->msg; |
b537916c | 3597 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3598 | /* if iov is set, it's allocated already */ |
3599 | if (!kmsg->iov) | |
3600 | kmsg->iov = kmsg->fast_iov; | |
3601 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3602 | } else { |
3529d8c2 JA |
3603 | struct io_sr_msg *sr = &req->sr_msg; |
3604 | ||
0b416c3e | 3605 | kmsg = &io.msg; |
b537916c | 3606 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3607 | |
3608 | io.msg.iov = io.msg.fast_iov; | |
3609 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3610 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3611 | if (ret) |
3529d8c2 | 3612 | return ret; |
03b1230c | 3613 | } |
0fa03c62 | 3614 | |
e47293fd JA |
3615 | flags = req->sr_msg.msg_flags; |
3616 | if (flags & MSG_DONTWAIT) | |
3617 | req->flags |= REQ_F_NOWAIT; | |
3618 | else if (force_nonblock) | |
3619 | flags |= MSG_DONTWAIT; | |
3620 | ||
0b416c3e | 3621 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3622 | if (force_nonblock && ret == -EAGAIN) |
3623 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3624 | if (ret == -ERESTARTSYS) |
3625 | ret = -EINTR; | |
0fa03c62 JA |
3626 | } |
3627 | ||
1e95081c | 3628 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3629 | kfree(kmsg->iov); |
99bc4c38 | 3630 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3631 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3632 | if (ret < 0) |
3633 | req_set_fail_links(req); | |
014db007 | 3634 | io_put_req(req); |
5d17b4a4 | 3635 | return 0; |
03b1230c | 3636 | } |
aa1fa28f | 3637 | |
014db007 | 3638 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3639 | { |
fddaface JA |
3640 | struct socket *sock; |
3641 | int ret; | |
3642 | ||
3643 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3644 | return -EINVAL; | |
3645 | ||
3646 | sock = sock_from_file(req->file, &ret); | |
3647 | if (sock) { | |
3648 | struct io_sr_msg *sr = &req->sr_msg; | |
3649 | struct msghdr msg; | |
3650 | struct iovec iov; | |
3651 | unsigned flags; | |
3652 | ||
3653 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3654 | &msg.msg_iter); | |
3655 | if (ret) | |
3656 | return ret; | |
3657 | ||
3658 | msg.msg_name = NULL; | |
3659 | msg.msg_control = NULL; | |
3660 | msg.msg_controllen = 0; | |
3661 | msg.msg_namelen = 0; | |
3662 | ||
3663 | flags = req->sr_msg.msg_flags; | |
3664 | if (flags & MSG_DONTWAIT) | |
3665 | req->flags |= REQ_F_NOWAIT; | |
3666 | else if (force_nonblock) | |
3667 | flags |= MSG_DONTWAIT; | |
3668 | ||
0b7b21e4 JA |
3669 | msg.msg_flags = flags; |
3670 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3671 | if (force_nonblock && ret == -EAGAIN) |
3672 | return -EAGAIN; | |
3673 | if (ret == -ERESTARTSYS) | |
3674 | ret = -EINTR; | |
3675 | } | |
3676 | ||
3677 | io_cqring_add_event(req, ret); | |
3678 | if (ret < 0) | |
3679 | req_set_fail_links(req); | |
014db007 | 3680 | io_put_req(req); |
fddaface | 3681 | return 0; |
fddaface JA |
3682 | } |
3683 | ||
52de1fe1 JA |
3684 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3685 | { | |
3686 | struct io_sr_msg *sr = &req->sr_msg; | |
3687 | struct iovec __user *uiov; | |
3688 | size_t iov_len; | |
3689 | int ret; | |
3690 | ||
3691 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3692 | &uiov, &iov_len); | |
3693 | if (ret) | |
3694 | return ret; | |
3695 | ||
3696 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3697 | if (iov_len > 1) | |
3698 | return -EINVAL; | |
3699 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3700 | return -EFAULT; | |
3701 | sr->len = io->msg.iov[0].iov_len; | |
3702 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3703 | sr->len); | |
3704 | io->msg.iov = NULL; | |
3705 | } else { | |
3706 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3707 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3708 | if (ret > 0) | |
3709 | ret = 0; | |
3710 | } | |
3711 | ||
3712 | return ret; | |
3713 | } | |
3714 | ||
3715 | #ifdef CONFIG_COMPAT | |
3716 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3717 | struct io_async_ctx *io) | |
3718 | { | |
3719 | struct compat_msghdr __user *msg_compat; | |
3720 | struct io_sr_msg *sr = &req->sr_msg; | |
3721 | struct compat_iovec __user *uiov; | |
3722 | compat_uptr_t ptr; | |
3723 | compat_size_t len; | |
3724 | int ret; | |
3725 | ||
3726 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3727 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3728 | &ptr, &len); | |
3729 | if (ret) | |
3730 | return ret; | |
3731 | ||
3732 | uiov = compat_ptr(ptr); | |
3733 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3734 | compat_ssize_t clen; | |
3735 | ||
3736 | if (len > 1) | |
3737 | return -EINVAL; | |
3738 | if (!access_ok(uiov, sizeof(*uiov))) | |
3739 | return -EFAULT; | |
3740 | if (__get_user(clen, &uiov->iov_len)) | |
3741 | return -EFAULT; | |
3742 | if (clen < 0) | |
3743 | return -EINVAL; | |
3744 | sr->len = io->msg.iov[0].iov_len; | |
3745 | io->msg.iov = NULL; | |
3746 | } else { | |
3747 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3748 | &io->msg.iov, | |
3749 | &io->msg.msg.msg_iter); | |
3750 | if (ret < 0) | |
3751 | return ret; | |
3752 | } | |
3753 | ||
3754 | return 0; | |
3755 | } | |
3756 | #endif | |
3757 | ||
3758 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3759 | { | |
3760 | io->msg.iov = io->msg.fast_iov; | |
3761 | ||
3762 | #ifdef CONFIG_COMPAT | |
3763 | if (req->ctx->compat) | |
3764 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3765 | #endif |
52de1fe1 JA |
3766 | |
3767 | return __io_recvmsg_copy_hdr(req, io); | |
3768 | } | |
3769 | ||
bcda7baa JA |
3770 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3771 | int *cflags, bool needs_lock) | |
3772 | { | |
3773 | struct io_sr_msg *sr = &req->sr_msg; | |
3774 | struct io_buffer *kbuf; | |
3775 | ||
3776 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3777 | return NULL; | |
3778 | ||
3779 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3780 | if (IS_ERR(kbuf)) | |
3781 | return kbuf; | |
3782 | ||
3783 | sr->kbuf = kbuf; | |
3784 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3785 | ||
3786 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3787 | *cflags |= IORING_CQE_F_BUFFER; | |
3788 | return kbuf; | |
fddaface JA |
3789 | } |
3790 | ||
3529d8c2 JA |
3791 | static int io_recvmsg_prep(struct io_kiocb *req, |
3792 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3793 | { |
e47293fd | 3794 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3795 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3796 | int ret; |
3529d8c2 JA |
3797 | |
3798 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3799 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3800 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3801 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3802 | |
d8768362 JA |
3803 | #ifdef CONFIG_COMPAT |
3804 | if (req->ctx->compat) | |
3805 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3806 | #endif | |
3807 | ||
fddaface | 3808 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3809 | return 0; |
5f798bea PB |
3810 | /* iovec is already imported */ |
3811 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3812 | return 0; | |
03b1230c | 3813 | |
52de1fe1 | 3814 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3815 | if (!ret) |
3816 | req->flags |= REQ_F_NEED_CLEANUP; | |
3817 | return ret; | |
aa1fa28f JA |
3818 | } |
3819 | ||
014db007 | 3820 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3821 | { |
0b416c3e | 3822 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3823 | struct socket *sock; |
52de1fe1 | 3824 | int ret, cflags = 0; |
03b1230c JA |
3825 | |
3826 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3827 | return -EINVAL; | |
3828 | ||
3829 | sock = sock_from_file(req->file, &ret); | |
3830 | if (sock) { | |
52de1fe1 | 3831 | struct io_buffer *kbuf; |
b7bb4f7d | 3832 | struct io_async_ctx io; |
03b1230c JA |
3833 | unsigned flags; |
3834 | ||
03b1230c | 3835 | if (req->io) { |
0b416c3e | 3836 | kmsg = &req->io->msg; |
b537916c | 3837 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3838 | /* if iov is set, it's allocated already */ |
3839 | if (!kmsg->iov) | |
3840 | kmsg->iov = kmsg->fast_iov; | |
3841 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3842 | } else { |
0b416c3e | 3843 | kmsg = &io.msg; |
b537916c | 3844 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3845 | |
52de1fe1 | 3846 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3847 | if (ret) |
3529d8c2 | 3848 | return ret; |
03b1230c JA |
3849 | } |
3850 | ||
52de1fe1 JA |
3851 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3852 | if (IS_ERR(kbuf)) { | |
3853 | return PTR_ERR(kbuf); | |
3854 | } else if (kbuf) { | |
3855 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3856 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3857 | 1, req->sr_msg.len); | |
3858 | } | |
3859 | ||
e47293fd JA |
3860 | flags = req->sr_msg.msg_flags; |
3861 | if (flags & MSG_DONTWAIT) | |
3862 | req->flags |= REQ_F_NOWAIT; | |
3863 | else if (force_nonblock) | |
3864 | flags |= MSG_DONTWAIT; | |
3865 | ||
3866 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3867 | kmsg->uaddr, flags); | |
02d27d89 PB |
3868 | if (force_nonblock && ret == -EAGAIN) |
3869 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3870 | if (ret == -ERESTARTSYS) |
3871 | ret = -EINTR; | |
3872 | } | |
3873 | ||
1e95081c | 3874 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3875 | kfree(kmsg->iov); |
99bc4c38 | 3876 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3877 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3878 | if (ret < 0) |
3879 | req_set_fail_links(req); | |
014db007 | 3880 | io_put_req(req); |
03b1230c | 3881 | return 0; |
0fa03c62 | 3882 | } |
5d17b4a4 | 3883 | |
014db007 | 3884 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3885 | { |
bcda7baa | 3886 | struct io_buffer *kbuf = NULL; |
fddaface | 3887 | struct socket *sock; |
bcda7baa | 3888 | int ret, cflags = 0; |
fddaface JA |
3889 | |
3890 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3891 | return -EINVAL; | |
3892 | ||
3893 | sock = sock_from_file(req->file, &ret); | |
3894 | if (sock) { | |
3895 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3896 | void __user *buf = sr->buf; |
fddaface JA |
3897 | struct msghdr msg; |
3898 | struct iovec iov; | |
3899 | unsigned flags; | |
3900 | ||
bcda7baa JA |
3901 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3902 | if (IS_ERR(kbuf)) | |
3903 | return PTR_ERR(kbuf); | |
3904 | else if (kbuf) | |
3905 | buf = u64_to_user_ptr(kbuf->addr); | |
3906 | ||
3907 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3908 | &msg.msg_iter); |
bcda7baa JA |
3909 | if (ret) { |
3910 | kfree(kbuf); | |
fddaface | 3911 | return ret; |
bcda7baa | 3912 | } |
fddaface | 3913 | |
bcda7baa | 3914 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3915 | msg.msg_name = NULL; |
3916 | msg.msg_control = NULL; | |
3917 | msg.msg_controllen = 0; | |
3918 | msg.msg_namelen = 0; | |
3919 | msg.msg_iocb = NULL; | |
3920 | msg.msg_flags = 0; | |
3921 | ||
3922 | flags = req->sr_msg.msg_flags; | |
3923 | if (flags & MSG_DONTWAIT) | |
3924 | req->flags |= REQ_F_NOWAIT; | |
3925 | else if (force_nonblock) | |
3926 | flags |= MSG_DONTWAIT; | |
3927 | ||
0b7b21e4 | 3928 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3929 | if (force_nonblock && ret == -EAGAIN) |
3930 | return -EAGAIN; | |
3931 | if (ret == -ERESTARTSYS) | |
3932 | ret = -EINTR; | |
3933 | } | |
3934 | ||
bcda7baa JA |
3935 | kfree(kbuf); |
3936 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3937 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3938 | if (ret < 0) |
3939 | req_set_fail_links(req); | |
014db007 | 3940 | io_put_req(req); |
fddaface | 3941 | return 0; |
fddaface JA |
3942 | } |
3943 | ||
3529d8c2 | 3944 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3945 | { |
8ed8d3c3 JA |
3946 | struct io_accept *accept = &req->accept; |
3947 | ||
17f2fe35 JA |
3948 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3949 | return -EINVAL; | |
8042d6ce | 3950 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3951 | return -EINVAL; |
3952 | ||
d55e5f5b JA |
3953 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3954 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3955 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3956 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3957 | return 0; |
8ed8d3c3 | 3958 | } |
17f2fe35 | 3959 | |
014db007 | 3960 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3961 | { |
3962 | struct io_accept *accept = &req->accept; | |
3963 | unsigned file_flags; | |
3964 | int ret; | |
3965 | ||
3966 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3967 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3968 | accept->addr_len, accept->flags, |
3969 | accept->nofile); | |
8ed8d3c3 | 3970 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3971 | return -EAGAIN; |
8e3cca12 JA |
3972 | if (ret == -ERESTARTSYS) |
3973 | ret = -EINTR; | |
4e88d6e7 JA |
3974 | if (ret < 0) |
3975 | req_set_fail_links(req); | |
78e19bbe | 3976 | io_cqring_add_event(req, ret); |
014db007 | 3977 | io_put_req(req); |
17f2fe35 | 3978 | return 0; |
8ed8d3c3 JA |
3979 | } |
3980 | ||
3981 | static void io_accept_finish(struct io_wq_work **workptr) | |
3982 | { | |
3983 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3984 | |
3985 | if (io_req_cancelled(req)) | |
3986 | return; | |
014db007 | 3987 | __io_accept(req, false); |
e9fd9396 | 3988 | io_steal_work(req, workptr); |
8ed8d3c3 | 3989 | } |
8ed8d3c3 | 3990 | |
014db007 | 3991 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3992 | { |
8ed8d3c3 JA |
3993 | int ret; |
3994 | ||
014db007 | 3995 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3996 | if (ret == -EAGAIN && force_nonblock) { |
3997 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3998 | return -EAGAIN; |
3999 | } | |
4000 | return 0; | |
0fa03c62 | 4001 | } |
5d17b4a4 | 4002 | |
3529d8c2 | 4003 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4004 | { |
3529d8c2 JA |
4005 | struct io_connect *conn = &req->connect; |
4006 | struct io_async_ctx *io = req->io; | |
f499a021 | 4007 | |
3fbb51c1 JA |
4008 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
4009 | return -EINVAL; | |
4010 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4011 | return -EINVAL; | |
4012 | ||
3529d8c2 JA |
4013 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4014 | conn->addr_len = READ_ONCE(sqe->addr2); | |
4015 | ||
4016 | if (!io) | |
4017 | return 0; | |
4018 | ||
4019 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 4020 | &io->connect.address); |
f499a021 JA |
4021 | } |
4022 | ||
014db007 | 4023 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 4024 | { |
f499a021 | 4025 | struct io_async_ctx __io, *io; |
f8e85cf2 | 4026 | unsigned file_flags; |
3fbb51c1 | 4027 | int ret; |
f8e85cf2 | 4028 | |
f499a021 JA |
4029 | if (req->io) { |
4030 | io = req->io; | |
4031 | } else { | |
3529d8c2 JA |
4032 | ret = move_addr_to_kernel(req->connect.addr, |
4033 | req->connect.addr_len, | |
4034 | &__io.connect.address); | |
f499a021 JA |
4035 | if (ret) |
4036 | goto out; | |
4037 | io = &__io; | |
4038 | } | |
4039 | ||
3fbb51c1 JA |
4040 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4041 | ||
4042 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4043 | req->connect.addr_len, file_flags); | |
87f80d62 | 4044 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4045 | if (req->io) |
4046 | return -EAGAIN; | |
4047 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4048 | ret = -ENOMEM; |
4049 | goto out; | |
4050 | } | |
b7bb4f7d | 4051 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4052 | return -EAGAIN; |
f499a021 | 4053 | } |
f8e85cf2 JA |
4054 | if (ret == -ERESTARTSYS) |
4055 | ret = -EINTR; | |
f499a021 | 4056 | out: |
4e88d6e7 JA |
4057 | if (ret < 0) |
4058 | req_set_fail_links(req); | |
f8e85cf2 | 4059 | io_cqring_add_event(req, ret); |
014db007 | 4060 | io_put_req(req); |
f8e85cf2 | 4061 | return 0; |
469956e8 Y |
4062 | } |
4063 | #else /* !CONFIG_NET */ | |
4064 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4065 | { | |
f8e85cf2 | 4066 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4067 | } |
4068 | ||
469956e8 Y |
4069 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4070 | { | |
4071 | return -EOPNOTSUPP; | |
4072 | } | |
4073 | ||
4074 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4075 | { | |
4076 | return -EOPNOTSUPP; | |
4077 | } | |
4078 | ||
4079 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4080 | const struct io_uring_sqe *sqe) | |
4081 | { | |
4082 | return -EOPNOTSUPP; | |
4083 | } | |
4084 | ||
4085 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4086 | { | |
4087 | return -EOPNOTSUPP; | |
4088 | } | |
4089 | ||
4090 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4091 | { | |
4092 | return -EOPNOTSUPP; | |
4093 | } | |
4094 | ||
4095 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4096 | { | |
4097 | return -EOPNOTSUPP; | |
4098 | } | |
4099 | ||
4100 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4101 | { | |
4102 | return -EOPNOTSUPP; | |
4103 | } | |
4104 | ||
4105 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4106 | { | |
4107 | return -EOPNOTSUPP; | |
4108 | } | |
4109 | ||
4110 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4111 | { | |
f8e85cf2 | 4112 | return -EOPNOTSUPP; |
f8e85cf2 | 4113 | } |
469956e8 | 4114 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4115 | |
d7718a9d JA |
4116 | struct io_poll_table { |
4117 | struct poll_table_struct pt; | |
4118 | struct io_kiocb *req; | |
4119 | int error; | |
4120 | }; | |
4121 | ||
4122 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4123 | struct wait_queue_head *head) | |
4124 | { | |
4125 | if (unlikely(poll->head)) { | |
4126 | pt->error = -EINVAL; | |
4127 | return; | |
4128 | } | |
4129 | ||
4130 | pt->error = 0; | |
4131 | poll->head = head; | |
4132 | add_wait_queue(head, &poll->wait); | |
4133 | } | |
4134 | ||
4135 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4136 | struct poll_table_struct *p) | |
4137 | { | |
4138 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4139 | ||
4140 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4141 | } | |
4142 | ||
4143 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4144 | __poll_t mask, task_work_func_t func) | |
4145 | { | |
4146 | struct task_struct *tsk; | |
aa96bf8a | 4147 | int ret; |
d7718a9d JA |
4148 | |
4149 | /* for instances that support it check for an event match first: */ | |
4150 | if (mask && !(mask & poll->events)) | |
4151 | return 0; | |
4152 | ||
4153 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4154 | ||
4155 | list_del_init(&poll->wait.entry); | |
4156 | ||
4157 | tsk = req->task; | |
4158 | req->result = mask; | |
4159 | init_task_work(&req->task_work, func); | |
4160 | /* | |
aa96bf8a JA |
4161 | * If this fails, then the task is exiting. Punt to one of the io-wq |
4162 | * threads to ensure the work gets run, we can't always rely on exit | |
4163 | * cancelation taking care of this. | |
d7718a9d | 4164 | */ |
aa96bf8a JA |
4165 | ret = task_work_add(tsk, &req->task_work, true); |
4166 | if (unlikely(ret)) { | |
4167 | tsk = io_wq_get_task(req->ctx->io_wq); | |
4168 | task_work_add(tsk, &req->task_work, true); | |
4169 | } | |
d7718a9d JA |
4170 | wake_up_process(tsk); |
4171 | return 1; | |
4172 | } | |
4173 | ||
74ce6ce4 JA |
4174 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4175 | __acquires(&req->ctx->completion_lock) | |
4176 | { | |
4177 | struct io_ring_ctx *ctx = req->ctx; | |
4178 | ||
4179 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4180 | struct poll_table_struct pt = { ._key = poll->events }; | |
4181 | ||
4182 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4183 | } | |
4184 | ||
4185 | spin_lock_irq(&ctx->completion_lock); | |
4186 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4187 | add_wait_queue(poll->head, &poll->wait); | |
4188 | return true; | |
4189 | } | |
4190 | ||
4191 | return false; | |
4192 | } | |
4193 | ||
d7718a9d JA |
4194 | static void io_async_task_func(struct callback_head *cb) |
4195 | { | |
4196 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4197 | struct async_poll *apoll = req->apoll; | |
4198 | struct io_ring_ctx *ctx = req->ctx; | |
2bae047e | 4199 | bool canceled; |
d7718a9d JA |
4200 | |
4201 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4202 | ||
74ce6ce4 | 4203 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 4204 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 4205 | return; |
d7718a9d JA |
4206 | } |
4207 | ||
74ce6ce4 JA |
4208 | if (hash_hashed(&req->hash_node)) |
4209 | hash_del(&req->hash_node); | |
4210 | ||
2bae047e JA |
4211 | canceled = READ_ONCE(apoll->poll.canceled); |
4212 | if (canceled) { | |
4213 | io_cqring_fill_event(req, -ECANCELED); | |
4214 | io_commit_cqring(ctx); | |
4215 | } | |
4216 | ||
74ce6ce4 JA |
4217 | spin_unlock_irq(&ctx->completion_lock); |
4218 | ||
44575a67 XW |
4219 | /* restore ->work in case we need to retry again */ |
4220 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4221 | ||
2bae047e JA |
4222 | if (canceled) { |
4223 | kfree(apoll); | |
4224 | io_cqring_ev_posted(ctx); | |
4225 | req_set_fail_links(req); | |
44575a67 | 4226 | io_double_put_req(req); |
2bae047e JA |
4227 | return; |
4228 | } | |
4229 | ||
d7718a9d JA |
4230 | __set_current_state(TASK_RUNNING); |
4231 | mutex_lock(&ctx->uring_lock); | |
4232 | __io_queue_sqe(req, NULL); | |
4233 | mutex_unlock(&ctx->uring_lock); | |
4234 | ||
4235 | kfree(apoll); | |
4236 | } | |
4237 | ||
4238 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4239 | void *key) | |
4240 | { | |
4241 | struct io_kiocb *req = wait->private; | |
4242 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4243 | ||
4244 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4245 | key_to_poll(key)); | |
4246 | ||
4247 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4248 | } | |
4249 | ||
4250 | static void io_poll_req_insert(struct io_kiocb *req) | |
4251 | { | |
4252 | struct io_ring_ctx *ctx = req->ctx; | |
4253 | struct hlist_head *list; | |
4254 | ||
4255 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4256 | hlist_add_head(&req->hash_node, list); | |
4257 | } | |
4258 | ||
4259 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4260 | struct io_poll_iocb *poll, | |
4261 | struct io_poll_table *ipt, __poll_t mask, | |
4262 | wait_queue_func_t wake_func) | |
4263 | __acquires(&ctx->completion_lock) | |
4264 | { | |
4265 | struct io_ring_ctx *ctx = req->ctx; | |
4266 | bool cancel = false; | |
4267 | ||
4268 | poll->file = req->file; | |
4269 | poll->head = NULL; | |
4270 | poll->done = poll->canceled = false; | |
4271 | poll->events = mask; | |
4272 | ||
4273 | ipt->pt._key = mask; | |
4274 | ipt->req = req; | |
4275 | ipt->error = -EINVAL; | |
4276 | ||
4277 | INIT_LIST_HEAD(&poll->wait.entry); | |
4278 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4279 | poll->wait.private = req; | |
4280 | ||
4281 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4282 | ||
4283 | spin_lock_irq(&ctx->completion_lock); | |
4284 | if (likely(poll->head)) { | |
4285 | spin_lock(&poll->head->lock); | |
4286 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4287 | if (ipt->error) | |
4288 | cancel = true; | |
4289 | ipt->error = 0; | |
4290 | mask = 0; | |
4291 | } | |
4292 | if (mask || ipt->error) | |
4293 | list_del_init(&poll->wait.entry); | |
4294 | else if (cancel) | |
4295 | WRITE_ONCE(poll->canceled, true); | |
4296 | else if (!poll->done) /* actually waiting for an event */ | |
4297 | io_poll_req_insert(req); | |
4298 | spin_unlock(&poll->head->lock); | |
4299 | } | |
4300 | ||
4301 | return mask; | |
4302 | } | |
4303 | ||
4304 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4305 | { | |
4306 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4307 | struct io_ring_ctx *ctx = req->ctx; | |
4308 | struct async_poll *apoll; | |
4309 | struct io_poll_table ipt; | |
4310 | __poll_t mask, ret; | |
4311 | ||
4312 | if (!req->file || !file_can_poll(req->file)) | |
4313 | return false; | |
4314 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4315 | return false; | |
4316 | if (!def->pollin && !def->pollout) | |
4317 | return false; | |
4318 | ||
4319 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4320 | if (unlikely(!apoll)) | |
4321 | return false; | |
4322 | ||
4323 | req->flags |= REQ_F_POLLED; | |
4324 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4325 | ||
3537b6a7 | 4326 | get_task_struct(current); |
d7718a9d JA |
4327 | req->task = current; |
4328 | req->apoll = apoll; | |
4329 | INIT_HLIST_NODE(&req->hash_node); | |
4330 | ||
8755d97a | 4331 | mask = 0; |
d7718a9d | 4332 | if (def->pollin) |
8755d97a | 4333 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4334 | if (def->pollout) |
4335 | mask |= POLLOUT | POLLWRNORM; | |
4336 | mask |= POLLERR | POLLPRI; | |
4337 | ||
4338 | ipt.pt._qproc = io_async_queue_proc; | |
4339 | ||
4340 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4341 | io_async_wake); | |
4342 | if (ret) { | |
4343 | ipt.error = 0; | |
4344 | apoll->poll.done = true; | |
4345 | spin_unlock_irq(&ctx->completion_lock); | |
4346 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4347 | kfree(apoll); | |
4348 | return false; | |
4349 | } | |
4350 | spin_unlock_irq(&ctx->completion_lock); | |
4351 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4352 | apoll->poll.events); | |
4353 | return true; | |
4354 | } | |
4355 | ||
4356 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4357 | struct io_poll_iocb *poll) | |
221c5eb2 | 4358 | { |
b41e9852 | 4359 | bool do_complete = false; |
221c5eb2 JA |
4360 | |
4361 | spin_lock(&poll->head->lock); | |
4362 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4363 | if (!list_empty(&poll->wait.entry)) { |
4364 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4365 | do_complete = true; |
221c5eb2 JA |
4366 | } |
4367 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4368 | return do_complete; |
4369 | } | |
4370 | ||
4371 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4372 | { | |
b1f573bd | 4373 | struct async_poll *apoll = NULL; |
d7718a9d JA |
4374 | bool do_complete; |
4375 | ||
4376 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4377 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4378 | } else { | |
b1f573bd | 4379 | apoll = req->apoll; |
d7718a9d JA |
4380 | /* non-poll requests have submit ref still */ |
4381 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4382 | if (do_complete) | |
4383 | io_put_req(req); | |
4384 | } | |
4385 | ||
78076bb6 | 4386 | hash_del(&req->hash_node); |
d7718a9d | 4387 | |
44575a67 | 4388 | if (do_complete && apoll) { |
b1f573bd XW |
4389 | /* |
4390 | * restore ->work because we need to call io_req_work_drop_env. | |
4391 | */ | |
4392 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4393 | kfree(apoll); | |
4394 | } | |
4395 | ||
b41e9852 JA |
4396 | if (do_complete) { |
4397 | io_cqring_fill_event(req, -ECANCELED); | |
4398 | io_commit_cqring(req->ctx); | |
4399 | req->flags |= REQ_F_COMP_LOCKED; | |
4400 | io_put_req(req); | |
4401 | } | |
4402 | ||
4403 | return do_complete; | |
221c5eb2 JA |
4404 | } |
4405 | ||
4406 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4407 | { | |
78076bb6 | 4408 | struct hlist_node *tmp; |
221c5eb2 | 4409 | struct io_kiocb *req; |
8e2e1faf | 4410 | int posted = 0, i; |
221c5eb2 JA |
4411 | |
4412 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4413 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4414 | struct hlist_head *list; | |
4415 | ||
4416 | list = &ctx->cancel_hash[i]; | |
4417 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
8e2e1faf | 4418 | posted += io_poll_remove_one(req); |
221c5eb2 JA |
4419 | } |
4420 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 4421 | |
8e2e1faf JA |
4422 | if (posted) |
4423 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4424 | } |
4425 | ||
47f46768 JA |
4426 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4427 | { | |
78076bb6 | 4428 | struct hlist_head *list; |
47f46768 JA |
4429 | struct io_kiocb *req; |
4430 | ||
78076bb6 JA |
4431 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4432 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4433 | if (sqe_addr != req->user_data) |
4434 | continue; | |
4435 | if (io_poll_remove_one(req)) | |
eac406c6 | 4436 | return 0; |
b41e9852 | 4437 | return -EALREADY; |
47f46768 JA |
4438 | } |
4439 | ||
4440 | return -ENOENT; | |
4441 | } | |
4442 | ||
3529d8c2 JA |
4443 | static int io_poll_remove_prep(struct io_kiocb *req, |
4444 | const struct io_uring_sqe *sqe) | |
0969e783 | 4445 | { |
0969e783 JA |
4446 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4447 | return -EINVAL; | |
4448 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4449 | sqe->poll_events) | |
4450 | return -EINVAL; | |
4451 | ||
4452 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4453 | return 0; |
4454 | } | |
4455 | ||
221c5eb2 JA |
4456 | /* |
4457 | * Find a running poll command that matches one specified in sqe->addr, | |
4458 | * and remove it if found. | |
4459 | */ | |
fc4df999 | 4460 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4461 | { |
4462 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4463 | u64 addr; |
47f46768 | 4464 | int ret; |
221c5eb2 | 4465 | |
0969e783 | 4466 | addr = req->poll.addr; |
221c5eb2 | 4467 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4468 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4469 | spin_unlock_irq(&ctx->completion_lock); |
4470 | ||
78e19bbe | 4471 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4472 | if (ret < 0) |
4473 | req_set_fail_links(req); | |
e65ef56d | 4474 | io_put_req(req); |
221c5eb2 JA |
4475 | return 0; |
4476 | } | |
4477 | ||
b0dd8a41 | 4478 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4479 | { |
a197f664 JL |
4480 | struct io_ring_ctx *ctx = req->ctx; |
4481 | ||
8c838788 | 4482 | req->poll.done = true; |
b0a20349 | 4483 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4484 | io_commit_cqring(ctx); |
221c5eb2 JA |
4485 | } |
4486 | ||
b41e9852 | 4487 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4488 | { |
221c5eb2 | 4489 | struct io_ring_ctx *ctx = req->ctx; |
a6ba632d JA |
4490 | struct io_poll_iocb *poll = &req->poll; |
4491 | ||
74ce6ce4 | 4492 | if (io_poll_rewait(req, poll)) { |
a6ba632d JA |
4493 | spin_unlock_irq(&ctx->completion_lock); |
4494 | return; | |
4495 | } | |
74ce6ce4 | 4496 | |
78076bb6 | 4497 | hash_del(&req->hash_node); |
b41e9852 JA |
4498 | io_poll_complete(req, req->result, 0); |
4499 | req->flags |= REQ_F_COMP_LOCKED; | |
4500 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4501 | spin_unlock_irq(&ctx->completion_lock); |
4502 | ||
8c838788 | 4503 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4504 | } |
4505 | ||
b41e9852 | 4506 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4507 | { |
b41e9852 JA |
4508 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4509 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4510 | |
b41e9852 | 4511 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4512 | if (nxt) { |
4513 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4514 | |
d7718a9d | 4515 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4516 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4517 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4518 | } |
f0b493e6 JA |
4519 | } |
4520 | ||
221c5eb2 JA |
4521 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4522 | void *key) | |
4523 | { | |
c2f2eb7d JA |
4524 | struct io_kiocb *req = wait->private; |
4525 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4526 | |
d7718a9d | 4527 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4528 | } |
4529 | ||
221c5eb2 JA |
4530 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4531 | struct poll_table_struct *p) | |
4532 | { | |
4533 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4534 | ||
d7718a9d | 4535 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4536 | } |
4537 | ||
3529d8c2 | 4538 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4539 | { |
4540 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4541 | u16 events; |
221c5eb2 JA |
4542 | |
4543 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4544 | return -EINVAL; | |
4545 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4546 | return -EINVAL; | |
09bb8394 JA |
4547 | if (!poll->file) |
4548 | return -EBADF; | |
221c5eb2 | 4549 | |
221c5eb2 JA |
4550 | events = READ_ONCE(sqe->poll_events); |
4551 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4552 | |
3537b6a7 | 4553 | get_task_struct(current); |
b41e9852 | 4554 | req->task = current; |
0969e783 JA |
4555 | return 0; |
4556 | } | |
4557 | ||
014db007 | 4558 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4559 | { |
4560 | struct io_poll_iocb *poll = &req->poll; | |
4561 | struct io_ring_ctx *ctx = req->ctx; | |
4562 | struct io_poll_table ipt; | |
0969e783 | 4563 | __poll_t mask; |
0969e783 | 4564 | |
78076bb6 | 4565 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4566 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4567 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4568 | |
d7718a9d JA |
4569 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4570 | io_poll_wake); | |
221c5eb2 | 4571 | |
8c838788 | 4572 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4573 | ipt.error = 0; |
b0dd8a41 | 4574 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4575 | } |
221c5eb2 JA |
4576 | spin_unlock_irq(&ctx->completion_lock); |
4577 | ||
8c838788 JA |
4578 | if (mask) { |
4579 | io_cqring_ev_posted(ctx); | |
014db007 | 4580 | io_put_req(req); |
221c5eb2 | 4581 | } |
8c838788 | 4582 | return ipt.error; |
221c5eb2 JA |
4583 | } |
4584 | ||
5262f567 JA |
4585 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4586 | { | |
ad8a48ac JA |
4587 | struct io_timeout_data *data = container_of(timer, |
4588 | struct io_timeout_data, timer); | |
4589 | struct io_kiocb *req = data->req; | |
4590 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4591 | unsigned long flags; |
4592 | ||
5262f567 JA |
4593 | atomic_inc(&ctx->cq_timeouts); |
4594 | ||
4595 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4596 | /* |
11365043 JA |
4597 | * We could be racing with timeout deletion. If the list is empty, |
4598 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4599 | */ |
842f9612 | 4600 | if (!list_empty(&req->list)) { |
11365043 | 4601 | struct io_kiocb *prev; |
5262f567 | 4602 | |
11365043 JA |
4603 | /* |
4604 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4605 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4606 | * pointer will be increased, otherwise other timeout reqs may |
4607 | * return in advance without waiting for enough wait_nr. | |
4608 | */ | |
4609 | prev = req; | |
4610 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4611 | prev->sequence++; | |
11365043 | 4612 | list_del_init(&req->list); |
11365043 | 4613 | } |
5262f567 | 4614 | |
78e19bbe | 4615 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4616 | io_commit_cqring(ctx); |
4617 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4618 | ||
4619 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4620 | req_set_fail_links(req); |
5262f567 JA |
4621 | io_put_req(req); |
4622 | return HRTIMER_NORESTART; | |
4623 | } | |
4624 | ||
47f46768 JA |
4625 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4626 | { | |
4627 | struct io_kiocb *req; | |
4628 | int ret = -ENOENT; | |
4629 | ||
4630 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4631 | if (user_data == req->user_data) { | |
4632 | list_del_init(&req->list); | |
4633 | ret = 0; | |
4634 | break; | |
4635 | } | |
4636 | } | |
4637 | ||
4638 | if (ret == -ENOENT) | |
4639 | return ret; | |
4640 | ||
2d28390a | 4641 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4642 | if (ret == -1) |
4643 | return -EALREADY; | |
4644 | ||
4e88d6e7 | 4645 | req_set_fail_links(req); |
47f46768 JA |
4646 | io_cqring_fill_event(req, -ECANCELED); |
4647 | io_put_req(req); | |
4648 | return 0; | |
4649 | } | |
4650 | ||
3529d8c2 JA |
4651 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4652 | const struct io_uring_sqe *sqe) | |
b29472ee | 4653 | { |
b29472ee JA |
4654 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4655 | return -EINVAL; | |
4656 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4657 | return -EINVAL; | |
4658 | ||
4659 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4660 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4661 | if (req->timeout.flags) | |
4662 | return -EINVAL; | |
4663 | ||
b29472ee JA |
4664 | return 0; |
4665 | } | |
4666 | ||
11365043 JA |
4667 | /* |
4668 | * Remove or update an existing timeout command | |
4669 | */ | |
fc4df999 | 4670 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4671 | { |
4672 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4673 | int ret; |
11365043 | 4674 | |
11365043 | 4675 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4676 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4677 | |
47f46768 | 4678 | io_cqring_fill_event(req, ret); |
11365043 JA |
4679 | io_commit_cqring(ctx); |
4680 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4681 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4682 | if (ret < 0) |
4683 | req_set_fail_links(req); | |
ec9c02ad | 4684 | io_put_req(req); |
11365043 | 4685 | return 0; |
5262f567 JA |
4686 | } |
4687 | ||
3529d8c2 | 4688 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4689 | bool is_timeout_link) |
5262f567 | 4690 | { |
ad8a48ac | 4691 | struct io_timeout_data *data; |
a41525ab | 4692 | unsigned flags; |
5262f567 | 4693 | |
ad8a48ac | 4694 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4695 | return -EINVAL; |
ad8a48ac | 4696 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4697 | return -EINVAL; |
2d28390a JA |
4698 | if (sqe->off && is_timeout_link) |
4699 | return -EINVAL; | |
a41525ab JA |
4700 | flags = READ_ONCE(sqe->timeout_flags); |
4701 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4702 | return -EINVAL; |
bdf20073 | 4703 | |
26a61679 JA |
4704 | req->timeout.count = READ_ONCE(sqe->off); |
4705 | ||
3529d8c2 | 4706 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4707 | return -ENOMEM; |
4708 | ||
4709 | data = &req->io->timeout; | |
ad8a48ac | 4710 | data->req = req; |
ad8a48ac JA |
4711 | req->flags |= REQ_F_TIMEOUT; |
4712 | ||
4713 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4714 | return -EFAULT; |
4715 | ||
11365043 | 4716 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4717 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4718 | else |
ad8a48ac | 4719 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4720 | |
ad8a48ac JA |
4721 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4722 | return 0; | |
4723 | } | |
4724 | ||
fc4df999 | 4725 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac | 4726 | { |
ad8a48ac JA |
4727 | struct io_ring_ctx *ctx = req->ctx; |
4728 | struct io_timeout_data *data; | |
4729 | struct list_head *entry; | |
4730 | unsigned span = 0; | |
b55ce732 | 4731 | u32 count = req->timeout.count; |
22cad158 | 4732 | u32 seq = req->sequence; |
ad8a48ac | 4733 | |
2d28390a | 4734 | data = &req->io->timeout; |
93bd25bb | 4735 | |
5262f567 JA |
4736 | /* |
4737 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4738 | * timeout event to be satisfied. If it isn't set, then this is |
4739 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4740 | */ |
93bd25bb JA |
4741 | if (!count) { |
4742 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4743 | spin_lock_irq(&ctx->completion_lock); | |
4744 | entry = ctx->timeout_list.prev; | |
4745 | goto add; | |
4746 | } | |
5262f567 | 4747 | |
22cad158 | 4748 | req->sequence = seq + count; |
5262f567 JA |
4749 | |
4750 | /* | |
4751 | * Insertion sort, ensuring the first entry in the list is always | |
4752 | * the one we need first. | |
4753 | */ | |
5262f567 JA |
4754 | spin_lock_irq(&ctx->completion_lock); |
4755 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4756 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
22cad158 | 4757 | unsigned nxt_seq; |
5da0fb1a | 4758 | long long tmp, tmp_nxt; |
b55ce732 | 4759 | u32 nxt_offset = nxt->timeout.count; |
5262f567 | 4760 | |
93bd25bb JA |
4761 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4762 | continue; | |
4763 | ||
5da0fb1a | 4764 | /* |
22cad158 | 4765 | * Since seq + count can overflow, use type long |
5da0fb1a | 4766 | * long to store it. |
4767 | */ | |
22cad158 PB |
4768 | tmp = (long long)seq + count; |
4769 | nxt_seq = nxt->sequence - nxt_offset; | |
4770 | tmp_nxt = (long long)nxt_seq + nxt_offset; | |
5da0fb1a | 4771 | |
4772 | /* | |
4773 | * cached_sq_head may overflow, and it will never overflow twice | |
4774 | * once there is some timeout req still be valid. | |
4775 | */ | |
22cad158 | 4776 | if (seq < nxt_seq) |
8b07a65a | 4777 | tmp += UINT_MAX; |
5da0fb1a | 4778 | |
a1f58ba4 | 4779 | if (tmp > tmp_nxt) |
5262f567 | 4780 | break; |
a1f58ba4 | 4781 | |
4782 | /* | |
4783 | * Sequence of reqs after the insert one and itself should | |
4784 | * be adjusted because each timeout req consumes a slot. | |
4785 | */ | |
4786 | span++; | |
4787 | nxt->sequence++; | |
5262f567 | 4788 | } |
a1f58ba4 | 4789 | req->sequence -= span; |
93bd25bb | 4790 | add: |
5262f567 | 4791 | list_add(&req->list, entry); |
ad8a48ac JA |
4792 | data->timer.function = io_timeout_fn; |
4793 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4794 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4795 | return 0; |
4796 | } | |
5262f567 | 4797 | |
62755e35 JA |
4798 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4799 | { | |
4800 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4801 | ||
4802 | return req->user_data == (unsigned long) data; | |
4803 | } | |
4804 | ||
e977d6d3 | 4805 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4806 | { |
62755e35 | 4807 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4808 | int ret = 0; |
4809 | ||
62755e35 JA |
4810 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4811 | switch (cancel_ret) { | |
4812 | case IO_WQ_CANCEL_OK: | |
4813 | ret = 0; | |
4814 | break; | |
4815 | case IO_WQ_CANCEL_RUNNING: | |
4816 | ret = -EALREADY; | |
4817 | break; | |
4818 | case IO_WQ_CANCEL_NOTFOUND: | |
4819 | ret = -ENOENT; | |
4820 | break; | |
4821 | } | |
4822 | ||
e977d6d3 JA |
4823 | return ret; |
4824 | } | |
4825 | ||
47f46768 JA |
4826 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4827 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4828 | int success_ret) |
47f46768 JA |
4829 | { |
4830 | unsigned long flags; | |
4831 | int ret; | |
4832 | ||
4833 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4834 | if (ret != -ENOENT) { | |
4835 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4836 | goto done; | |
4837 | } | |
4838 | ||
4839 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4840 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4841 | if (ret != -ENOENT) | |
4842 | goto done; | |
4843 | ret = io_poll_cancel(ctx, sqe_addr); | |
4844 | done: | |
b0dd8a41 JA |
4845 | if (!ret) |
4846 | ret = success_ret; | |
47f46768 JA |
4847 | io_cqring_fill_event(req, ret); |
4848 | io_commit_cqring(ctx); | |
4849 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4850 | io_cqring_ev_posted(ctx); | |
4851 | ||
4e88d6e7 JA |
4852 | if (ret < 0) |
4853 | req_set_fail_links(req); | |
014db007 | 4854 | io_put_req(req); |
47f46768 JA |
4855 | } |
4856 | ||
3529d8c2 JA |
4857 | static int io_async_cancel_prep(struct io_kiocb *req, |
4858 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4859 | { |
fbf23849 | 4860 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4861 | return -EINVAL; |
4862 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4863 | sqe->cancel_flags) | |
4864 | return -EINVAL; | |
4865 | ||
fbf23849 JA |
4866 | req->cancel.addr = READ_ONCE(sqe->addr); |
4867 | return 0; | |
4868 | } | |
4869 | ||
014db007 | 4870 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4871 | { |
4872 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4873 | |
014db007 | 4874 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4875 | return 0; |
4876 | } | |
4877 | ||
05f3fb3c JA |
4878 | static int io_files_update_prep(struct io_kiocb *req, |
4879 | const struct io_uring_sqe *sqe) | |
4880 | { | |
4881 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4882 | return -EINVAL; | |
4883 | ||
4884 | req->files_update.offset = READ_ONCE(sqe->off); | |
4885 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4886 | if (!req->files_update.nr_args) | |
4887 | return -EINVAL; | |
4888 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4889 | return 0; | |
4890 | } | |
4891 | ||
4892 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4893 | { |
4894 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4895 | struct io_uring_files_update up; |
4896 | int ret; | |
fbf23849 | 4897 | |
f86cd20c | 4898 | if (force_nonblock) |
05f3fb3c | 4899 | return -EAGAIN; |
05f3fb3c JA |
4900 | |
4901 | up.offset = req->files_update.offset; | |
4902 | up.fds = req->files_update.arg; | |
4903 | ||
4904 | mutex_lock(&ctx->uring_lock); | |
4905 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4906 | mutex_unlock(&ctx->uring_lock); | |
4907 | ||
4908 | if (ret < 0) | |
4909 | req_set_fail_links(req); | |
4910 | io_cqring_add_event(req, ret); | |
4911 | io_put_req(req); | |
5262f567 JA |
4912 | return 0; |
4913 | } | |
4914 | ||
3529d8c2 JA |
4915 | static int io_req_defer_prep(struct io_kiocb *req, |
4916 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4917 | { |
e781573e | 4918 | ssize_t ret = 0; |
f67676d1 | 4919 | |
f1d96a8f PB |
4920 | if (!sqe) |
4921 | return 0; | |
4922 | ||
f86cd20c JA |
4923 | if (io_op_defs[req->opcode].file_table) { |
4924 | ret = io_grab_files(req); | |
4925 | if (unlikely(ret)) | |
4926 | return ret; | |
4927 | } | |
4928 | ||
cccf0ee8 JA |
4929 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4930 | ||
d625c6ee | 4931 | switch (req->opcode) { |
e781573e JA |
4932 | case IORING_OP_NOP: |
4933 | break; | |
f67676d1 JA |
4934 | case IORING_OP_READV: |
4935 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4936 | case IORING_OP_READ: |
3529d8c2 | 4937 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4938 | break; |
4939 | case IORING_OP_WRITEV: | |
4940 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4941 | case IORING_OP_WRITE: |
3529d8c2 | 4942 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4943 | break; |
0969e783 | 4944 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4945 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4946 | break; |
4947 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4948 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4949 | break; |
8ed8d3c3 | 4950 | case IORING_OP_FSYNC: |
3529d8c2 | 4951 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4952 | break; |
4953 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4954 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4955 | break; |
03b1230c | 4956 | case IORING_OP_SENDMSG: |
fddaface | 4957 | case IORING_OP_SEND: |
3529d8c2 | 4958 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4959 | break; |
4960 | case IORING_OP_RECVMSG: | |
fddaface | 4961 | case IORING_OP_RECV: |
3529d8c2 | 4962 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4963 | break; |
f499a021 | 4964 | case IORING_OP_CONNECT: |
3529d8c2 | 4965 | ret = io_connect_prep(req, sqe); |
f499a021 | 4966 | break; |
2d28390a | 4967 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4968 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4969 | break; |
b29472ee | 4970 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4971 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4972 | break; |
fbf23849 | 4973 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4974 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4975 | break; |
2d28390a | 4976 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4977 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4978 | break; |
8ed8d3c3 | 4979 | case IORING_OP_ACCEPT: |
3529d8c2 | 4980 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4981 | break; |
d63d1b5e JA |
4982 | case IORING_OP_FALLOCATE: |
4983 | ret = io_fallocate_prep(req, sqe); | |
4984 | break; | |
15b71abe JA |
4985 | case IORING_OP_OPENAT: |
4986 | ret = io_openat_prep(req, sqe); | |
4987 | break; | |
b5dba59e JA |
4988 | case IORING_OP_CLOSE: |
4989 | ret = io_close_prep(req, sqe); | |
4990 | break; | |
05f3fb3c JA |
4991 | case IORING_OP_FILES_UPDATE: |
4992 | ret = io_files_update_prep(req, sqe); | |
4993 | break; | |
eddc7ef5 JA |
4994 | case IORING_OP_STATX: |
4995 | ret = io_statx_prep(req, sqe); | |
4996 | break; | |
4840e418 JA |
4997 | case IORING_OP_FADVISE: |
4998 | ret = io_fadvise_prep(req, sqe); | |
4999 | break; | |
c1ca757b JA |
5000 | case IORING_OP_MADVISE: |
5001 | ret = io_madvise_prep(req, sqe); | |
5002 | break; | |
cebdb986 JA |
5003 | case IORING_OP_OPENAT2: |
5004 | ret = io_openat2_prep(req, sqe); | |
5005 | break; | |
3e4827b0 JA |
5006 | case IORING_OP_EPOLL_CTL: |
5007 | ret = io_epoll_ctl_prep(req, sqe); | |
5008 | break; | |
7d67af2c PB |
5009 | case IORING_OP_SPLICE: |
5010 | ret = io_splice_prep(req, sqe); | |
5011 | break; | |
ddf0322d JA |
5012 | case IORING_OP_PROVIDE_BUFFERS: |
5013 | ret = io_provide_buffers_prep(req, sqe); | |
5014 | break; | |
067524e9 JA |
5015 | case IORING_OP_REMOVE_BUFFERS: |
5016 | ret = io_remove_buffers_prep(req, sqe); | |
5017 | break; | |
f67676d1 | 5018 | default: |
e781573e JA |
5019 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5020 | req->opcode); | |
5021 | ret = -EINVAL; | |
b7bb4f7d | 5022 | break; |
f67676d1 JA |
5023 | } |
5024 | ||
b7bb4f7d | 5025 | return ret; |
f67676d1 JA |
5026 | } |
5027 | ||
3529d8c2 | 5028 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 5029 | { |
a197f664 | 5030 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 5031 | int ret; |
de0617e4 | 5032 | |
9d858b21 BL |
5033 | /* Still need defer if there is pending req in defer list. */ |
5034 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
5035 | return 0; |
5036 | ||
3529d8c2 | 5037 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
5038 | return -EAGAIN; |
5039 | ||
3529d8c2 | 5040 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 5041 | if (ret < 0) |
2d28390a | 5042 | return ret; |
2d28390a | 5043 | |
de0617e4 | 5044 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 5045 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5046 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
5047 | return 0; |
5048 | } | |
5049 | ||
915967f6 | 5050 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
5051 | list_add_tail(&req->list, &ctx->defer_list); |
5052 | spin_unlock_irq(&ctx->completion_lock); | |
5053 | return -EIOCBQUEUED; | |
5054 | } | |
5055 | ||
99bc4c38 PB |
5056 | static void io_cleanup_req(struct io_kiocb *req) |
5057 | { | |
5058 | struct io_async_ctx *io = req->io; | |
5059 | ||
5060 | switch (req->opcode) { | |
5061 | case IORING_OP_READV: | |
5062 | case IORING_OP_READ_FIXED: | |
5063 | case IORING_OP_READ: | |
bcda7baa JA |
5064 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5065 | kfree((void *)(unsigned long)req->rw.addr); | |
5066 | /* fallthrough */ | |
99bc4c38 PB |
5067 | case IORING_OP_WRITEV: |
5068 | case IORING_OP_WRITE_FIXED: | |
5069 | case IORING_OP_WRITE: | |
5070 | if (io->rw.iov != io->rw.fast_iov) | |
5071 | kfree(io->rw.iov); | |
5072 | break; | |
99bc4c38 | 5073 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
5074 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5075 | kfree(req->sr_msg.kbuf); | |
5076 | /* fallthrough */ | |
5077 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
5078 | if (io->msg.iov != io->msg.fast_iov) |
5079 | kfree(io->msg.iov); | |
5080 | break; | |
bcda7baa JA |
5081 | case IORING_OP_RECV: |
5082 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5083 | kfree(req->sr_msg.kbuf); | |
5084 | break; | |
8fef80bf PB |
5085 | case IORING_OP_OPENAT: |
5086 | case IORING_OP_OPENAT2: | |
5087 | case IORING_OP_STATX: | |
5088 | putname(req->open.filename); | |
5089 | break; | |
7d67af2c PB |
5090 | case IORING_OP_SPLICE: |
5091 | io_put_file(req, req->splice.file_in, | |
5092 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5093 | break; | |
99bc4c38 PB |
5094 | } |
5095 | ||
5096 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5097 | } | |
5098 | ||
3529d8c2 | 5099 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5100 | bool force_nonblock) |
2b188cc1 | 5101 | { |
a197f664 | 5102 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5103 | int ret; |
2b188cc1 | 5104 | |
d625c6ee | 5105 | switch (req->opcode) { |
2b188cc1 | 5106 | case IORING_OP_NOP: |
78e19bbe | 5107 | ret = io_nop(req); |
2b188cc1 JA |
5108 | break; |
5109 | case IORING_OP_READV: | |
edafccee | 5110 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5111 | case IORING_OP_READ: |
3529d8c2 JA |
5112 | if (sqe) { |
5113 | ret = io_read_prep(req, sqe, force_nonblock); | |
5114 | if (ret < 0) | |
5115 | break; | |
5116 | } | |
014db007 | 5117 | ret = io_read(req, force_nonblock); |
edafccee | 5118 | break; |
3529d8c2 | 5119 | case IORING_OP_WRITEV: |
edafccee | 5120 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5121 | case IORING_OP_WRITE: |
3529d8c2 JA |
5122 | if (sqe) { |
5123 | ret = io_write_prep(req, sqe, force_nonblock); | |
5124 | if (ret < 0) | |
5125 | break; | |
5126 | } | |
014db007 | 5127 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5128 | break; |
c992fe29 | 5129 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5130 | if (sqe) { |
5131 | ret = io_prep_fsync(req, sqe); | |
5132 | if (ret < 0) | |
5133 | break; | |
5134 | } | |
014db007 | 5135 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5136 | break; |
221c5eb2 | 5137 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5138 | if (sqe) { |
5139 | ret = io_poll_add_prep(req, sqe); | |
5140 | if (ret) | |
5141 | break; | |
5142 | } | |
014db007 | 5143 | ret = io_poll_add(req); |
221c5eb2 JA |
5144 | break; |
5145 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5146 | if (sqe) { |
5147 | ret = io_poll_remove_prep(req, sqe); | |
5148 | if (ret < 0) | |
5149 | break; | |
5150 | } | |
fc4df999 | 5151 | ret = io_poll_remove(req); |
221c5eb2 | 5152 | break; |
5d17b4a4 | 5153 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5154 | if (sqe) { |
5155 | ret = io_prep_sfr(req, sqe); | |
5156 | if (ret < 0) | |
5157 | break; | |
5158 | } | |
014db007 | 5159 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5160 | break; |
0fa03c62 | 5161 | case IORING_OP_SENDMSG: |
fddaface | 5162 | case IORING_OP_SEND: |
3529d8c2 JA |
5163 | if (sqe) { |
5164 | ret = io_sendmsg_prep(req, sqe); | |
5165 | if (ret < 0) | |
5166 | break; | |
5167 | } | |
fddaface | 5168 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5169 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5170 | else |
014db007 | 5171 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5172 | break; |
aa1fa28f | 5173 | case IORING_OP_RECVMSG: |
fddaface | 5174 | case IORING_OP_RECV: |
3529d8c2 JA |
5175 | if (sqe) { |
5176 | ret = io_recvmsg_prep(req, sqe); | |
5177 | if (ret) | |
5178 | break; | |
5179 | } | |
fddaface | 5180 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5181 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5182 | else |
014db007 | 5183 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5184 | break; |
5262f567 | 5185 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5186 | if (sqe) { |
5187 | ret = io_timeout_prep(req, sqe, false); | |
5188 | if (ret) | |
5189 | break; | |
5190 | } | |
fc4df999 | 5191 | ret = io_timeout(req); |
5262f567 | 5192 | break; |
11365043 | 5193 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5194 | if (sqe) { |
5195 | ret = io_timeout_remove_prep(req, sqe); | |
5196 | if (ret) | |
5197 | break; | |
5198 | } | |
fc4df999 | 5199 | ret = io_timeout_remove(req); |
11365043 | 5200 | break; |
17f2fe35 | 5201 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5202 | if (sqe) { |
5203 | ret = io_accept_prep(req, sqe); | |
5204 | if (ret) | |
5205 | break; | |
5206 | } | |
014db007 | 5207 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5208 | break; |
f8e85cf2 | 5209 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5210 | if (sqe) { |
5211 | ret = io_connect_prep(req, sqe); | |
5212 | if (ret) | |
5213 | break; | |
5214 | } | |
014db007 | 5215 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5216 | break; |
62755e35 | 5217 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5218 | if (sqe) { |
5219 | ret = io_async_cancel_prep(req, sqe); | |
5220 | if (ret) | |
5221 | break; | |
5222 | } | |
014db007 | 5223 | ret = io_async_cancel(req); |
62755e35 | 5224 | break; |
d63d1b5e JA |
5225 | case IORING_OP_FALLOCATE: |
5226 | if (sqe) { | |
5227 | ret = io_fallocate_prep(req, sqe); | |
5228 | if (ret) | |
5229 | break; | |
5230 | } | |
014db007 | 5231 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5232 | break; |
15b71abe JA |
5233 | case IORING_OP_OPENAT: |
5234 | if (sqe) { | |
5235 | ret = io_openat_prep(req, sqe); | |
5236 | if (ret) | |
5237 | break; | |
5238 | } | |
014db007 | 5239 | ret = io_openat(req, force_nonblock); |
15b71abe | 5240 | break; |
b5dba59e JA |
5241 | case IORING_OP_CLOSE: |
5242 | if (sqe) { | |
5243 | ret = io_close_prep(req, sqe); | |
5244 | if (ret) | |
5245 | break; | |
5246 | } | |
014db007 | 5247 | ret = io_close(req, force_nonblock); |
b5dba59e | 5248 | break; |
05f3fb3c JA |
5249 | case IORING_OP_FILES_UPDATE: |
5250 | if (sqe) { | |
5251 | ret = io_files_update_prep(req, sqe); | |
5252 | if (ret) | |
5253 | break; | |
5254 | } | |
5255 | ret = io_files_update(req, force_nonblock); | |
5256 | break; | |
eddc7ef5 JA |
5257 | case IORING_OP_STATX: |
5258 | if (sqe) { | |
5259 | ret = io_statx_prep(req, sqe); | |
5260 | if (ret) | |
5261 | break; | |
5262 | } | |
014db007 | 5263 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5264 | break; |
4840e418 JA |
5265 | case IORING_OP_FADVISE: |
5266 | if (sqe) { | |
5267 | ret = io_fadvise_prep(req, sqe); | |
5268 | if (ret) | |
5269 | break; | |
5270 | } | |
014db007 | 5271 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5272 | break; |
c1ca757b JA |
5273 | case IORING_OP_MADVISE: |
5274 | if (sqe) { | |
5275 | ret = io_madvise_prep(req, sqe); | |
5276 | if (ret) | |
5277 | break; | |
5278 | } | |
014db007 | 5279 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5280 | break; |
cebdb986 JA |
5281 | case IORING_OP_OPENAT2: |
5282 | if (sqe) { | |
5283 | ret = io_openat2_prep(req, sqe); | |
5284 | if (ret) | |
5285 | break; | |
5286 | } | |
014db007 | 5287 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5288 | break; |
3e4827b0 JA |
5289 | case IORING_OP_EPOLL_CTL: |
5290 | if (sqe) { | |
5291 | ret = io_epoll_ctl_prep(req, sqe); | |
5292 | if (ret) | |
5293 | break; | |
5294 | } | |
014db007 | 5295 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5296 | break; |
7d67af2c PB |
5297 | case IORING_OP_SPLICE: |
5298 | if (sqe) { | |
5299 | ret = io_splice_prep(req, sqe); | |
5300 | if (ret < 0) | |
5301 | break; | |
5302 | } | |
014db007 | 5303 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5304 | break; |
ddf0322d JA |
5305 | case IORING_OP_PROVIDE_BUFFERS: |
5306 | if (sqe) { | |
5307 | ret = io_provide_buffers_prep(req, sqe); | |
5308 | if (ret) | |
5309 | break; | |
5310 | } | |
5311 | ret = io_provide_buffers(req, force_nonblock); | |
5312 | break; | |
067524e9 JA |
5313 | case IORING_OP_REMOVE_BUFFERS: |
5314 | if (sqe) { | |
5315 | ret = io_remove_buffers_prep(req, sqe); | |
5316 | if (ret) | |
5317 | break; | |
5318 | } | |
5319 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5320 | break; |
2b188cc1 JA |
5321 | default: |
5322 | ret = -EINVAL; | |
5323 | break; | |
5324 | } | |
5325 | ||
def596e9 JA |
5326 | if (ret) |
5327 | return ret; | |
5328 | ||
5329 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5330 | const bool in_async = io_wq_current_is_worker(); |
5331 | ||
9e645e11 | 5332 | if (req->result == -EAGAIN) |
def596e9 JA |
5333 | return -EAGAIN; |
5334 | ||
11ba820b JA |
5335 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5336 | if (in_async) | |
5337 | mutex_lock(&ctx->uring_lock); | |
5338 | ||
def596e9 | 5339 | io_iopoll_req_issued(req); |
11ba820b JA |
5340 | |
5341 | if (in_async) | |
5342 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5343 | } |
5344 | ||
5345 | return 0; | |
2b188cc1 JA |
5346 | } |
5347 | ||
561fb04a | 5348 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5349 | { |
561fb04a | 5350 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5351 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5352 | int ret = 0; |
2b188cc1 | 5353 | |
0c9d5ccd JA |
5354 | /* if NO_CANCEL is set, we must still run the work */ |
5355 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5356 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5357 | ret = -ECANCELED; |
0c9d5ccd | 5358 | } |
31b51510 | 5359 | |
561fb04a | 5360 | if (!ret) { |
561fb04a | 5361 | do { |
014db007 | 5362 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5363 | /* |
5364 | * We can get EAGAIN for polled IO even though we're | |
5365 | * forcing a sync submission from here, since we can't | |
5366 | * wait for request slots on the block side. | |
5367 | */ | |
5368 | if (ret != -EAGAIN) | |
5369 | break; | |
5370 | cond_resched(); | |
5371 | } while (1); | |
5372 | } | |
31b51510 | 5373 | |
561fb04a | 5374 | if (ret) { |
4e88d6e7 | 5375 | req_set_fail_links(req); |
78e19bbe | 5376 | io_cqring_add_event(req, ret); |
817869d2 | 5377 | io_put_req(req); |
edafccee | 5378 | } |
2b188cc1 | 5379 | |
e9fd9396 | 5380 | io_steal_work(req, workptr); |
2b188cc1 JA |
5381 | } |
5382 | ||
15b71abe | 5383 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 5384 | { |
d3656344 | 5385 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 5386 | return 0; |
0b5faf6b | 5387 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
5388 | return 0; |
5389 | return 1; | |
09bb8394 JA |
5390 | } |
5391 | ||
65e19f54 JA |
5392 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5393 | int index) | |
5394 | { | |
5395 | struct fixed_file_table *table; | |
5396 | ||
05f3fb3c JA |
5397 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5398 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5399 | } |
5400 | ||
8da11c19 PB |
5401 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5402 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5403 | { |
a197f664 | 5404 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5405 | struct file *file; |
09bb8394 | 5406 | |
8da11c19 | 5407 | if (fixed) { |
05f3fb3c | 5408 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5409 | (unsigned) fd >= ctx->nr_user_files)) |
5410 | return -EBADF; | |
b7620121 | 5411 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5412 | file = io_file_from_index(ctx, fd); |
5413 | if (!file) | |
08a45173 | 5414 | return -EBADF; |
05589553 XW |
5415 | req->fixed_file_refs = ctx->file_data->cur_refs; |
5416 | percpu_ref_get(req->fixed_file_refs); | |
09bb8394 | 5417 | } else { |
c826bd7a | 5418 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5419 | file = __io_file_get(state, fd); |
5420 | if (unlikely(!file)) | |
09bb8394 JA |
5421 | return -EBADF; |
5422 | } | |
5423 | ||
8da11c19 | 5424 | *out_file = file; |
09bb8394 JA |
5425 | return 0; |
5426 | } | |
5427 | ||
8da11c19 | 5428 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
9c280f90 | 5429 | int fd, unsigned int flags) |
8da11c19 | 5430 | { |
8da11c19 PB |
5431 | bool fixed; |
5432 | ||
8da11c19 PB |
5433 | if (!io_req_needs_file(req, fd)) |
5434 | return 0; | |
5435 | ||
5436 | fixed = (flags & IOSQE_FIXED_FILE); | |
5437 | if (unlikely(!fixed && req->needs_fixed_file)) | |
5438 | return -EBADF; | |
5439 | ||
5440 | return io_file_get(state, req, fd, &req->file, fixed); | |
5441 | } | |
5442 | ||
a197f664 | 5443 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5444 | { |
5445 | int ret = -EBADF; | |
a197f664 | 5446 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5447 | |
5b0bbee4 | 5448 | if (req->work.files || (req->flags & REQ_F_NO_FILE_TABLE)) |
f86cd20c | 5449 | return 0; |
b14cca0c | 5450 | if (!ctx->ring_file) |
b5dba59e JA |
5451 | return -EBADF; |
5452 | ||
fcb323cc JA |
5453 | rcu_read_lock(); |
5454 | spin_lock_irq(&ctx->inflight_lock); | |
5455 | /* | |
5456 | * We use the f_ops->flush() handler to ensure that we can flush | |
5457 | * out work accessing these files if the fd is closed. Check if | |
5458 | * the fd has changed since we started down this path, and disallow | |
5459 | * this operation if it has. | |
5460 | */ | |
b14cca0c | 5461 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5462 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5463 | req->flags |= REQ_F_INFLIGHT; | |
5464 | req->work.files = current->files; | |
5465 | ret = 0; | |
5466 | } | |
5467 | spin_unlock_irq(&ctx->inflight_lock); | |
5468 | rcu_read_unlock(); | |
5469 | ||
5470 | return ret; | |
5471 | } | |
5472 | ||
2665abfd | 5473 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5474 | { |
ad8a48ac JA |
5475 | struct io_timeout_data *data = container_of(timer, |
5476 | struct io_timeout_data, timer); | |
5477 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5478 | struct io_ring_ctx *ctx = req->ctx; |
5479 | struct io_kiocb *prev = NULL; | |
5480 | unsigned long flags; | |
2665abfd JA |
5481 | |
5482 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5483 | ||
5484 | /* | |
5485 | * We don't expect the list to be empty, that will only happen if we | |
5486 | * race with the completion of the linked work. | |
5487 | */ | |
4493233e PB |
5488 | if (!list_empty(&req->link_list)) { |
5489 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5490 | link_list); | |
5d960724 | 5491 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5492 | list_del_init(&req->link_list); |
5d960724 JA |
5493 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5494 | } else | |
76a46e06 | 5495 | prev = NULL; |
2665abfd JA |
5496 | } |
5497 | ||
5498 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5499 | ||
5500 | if (prev) { | |
4e88d6e7 | 5501 | req_set_fail_links(prev); |
014db007 | 5502 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5503 | io_put_req(prev); |
47f46768 JA |
5504 | } else { |
5505 | io_cqring_add_event(req, -ETIME); | |
5506 | io_put_req(req); | |
2665abfd | 5507 | } |
2665abfd JA |
5508 | return HRTIMER_NORESTART; |
5509 | } | |
5510 | ||
ad8a48ac | 5511 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5512 | { |
76a46e06 | 5513 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5514 | |
76a46e06 JA |
5515 | /* |
5516 | * If the list is now empty, then our linked request finished before | |
5517 | * we got a chance to setup the timer | |
5518 | */ | |
5519 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5520 | if (!list_empty(&req->link_list)) { |
2d28390a | 5521 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5522 | |
ad8a48ac JA |
5523 | data->timer.function = io_link_timeout_fn; |
5524 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5525 | data->mode); | |
2665abfd | 5526 | } |
76a46e06 | 5527 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5528 | |
2665abfd | 5529 | /* drop submission reference */ |
76a46e06 JA |
5530 | io_put_req(req); |
5531 | } | |
2665abfd | 5532 | |
ad8a48ac | 5533 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5534 | { |
5535 | struct io_kiocb *nxt; | |
5536 | ||
dea3b49c | 5537 | if (!(req->flags & REQ_F_LINK_HEAD)) |
2665abfd | 5538 | return NULL; |
d7718a9d JA |
5539 | /* for polled retry, if flag is set, we already went through here */ |
5540 | if (req->flags & REQ_F_POLLED) | |
5541 | return NULL; | |
2665abfd | 5542 | |
4493233e PB |
5543 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5544 | link_list); | |
d625c6ee | 5545 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5546 | return NULL; |
2665abfd | 5547 | |
76a46e06 | 5548 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5549 | return nxt; |
2665abfd JA |
5550 | } |
5551 | ||
3529d8c2 | 5552 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5553 | { |
4a0a7a18 | 5554 | struct io_kiocb *linked_timeout; |
4bc4494e | 5555 | struct io_kiocb *nxt; |
193155c8 | 5556 | const struct cred *old_creds = NULL; |
e0c5c576 | 5557 | int ret; |
2b188cc1 | 5558 | |
4a0a7a18 JA |
5559 | again: |
5560 | linked_timeout = io_prep_linked_timeout(req); | |
5561 | ||
193155c8 JA |
5562 | if (req->work.creds && req->work.creds != current_cred()) { |
5563 | if (old_creds) | |
5564 | revert_creds(old_creds); | |
5565 | if (old_creds == req->work.creds) | |
5566 | old_creds = NULL; /* restored original creds */ | |
5567 | else | |
5568 | old_creds = override_creds(req->work.creds); | |
5569 | } | |
5570 | ||
014db007 | 5571 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5572 | |
5573 | /* | |
5574 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5575 | * doesn't support non-blocking read/write attempts | |
5576 | */ | |
5577 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5578 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5579 | if (io_arm_poll_handler(req)) { |
5580 | if (linked_timeout) | |
5581 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5582 | goto exit; |
d7718a9d | 5583 | } |
86a761f8 | 5584 | punt: |
f86cd20c | 5585 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5586 | ret = io_grab_files(req); |
5587 | if (ret) | |
5588 | goto err; | |
2b188cc1 | 5589 | } |
bbad27b2 PB |
5590 | |
5591 | /* | |
5592 | * Queued up for async execution, worker will release | |
5593 | * submit reference when the iocb is actually submitted. | |
5594 | */ | |
5595 | io_queue_async_work(req); | |
4bc4494e | 5596 | goto exit; |
2b188cc1 | 5597 | } |
e65ef56d | 5598 | |
fcb323cc | 5599 | err: |
4bc4494e | 5600 | nxt = NULL; |
76a46e06 | 5601 | /* drop submission reference */ |
2a44f467 | 5602 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5603 | |
f9bd67f6 | 5604 | if (linked_timeout) { |
76a46e06 | 5605 | if (!ret) |
f9bd67f6 | 5606 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5607 | else |
f9bd67f6 | 5608 | io_put_req(linked_timeout); |
76a46e06 JA |
5609 | } |
5610 | ||
e65ef56d | 5611 | /* and drop final reference, if we failed */ |
9e645e11 | 5612 | if (ret) { |
78e19bbe | 5613 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5614 | req_set_fail_links(req); |
e65ef56d | 5615 | io_put_req(req); |
9e645e11 | 5616 | } |
4a0a7a18 JA |
5617 | if (nxt) { |
5618 | req = nxt; | |
86a761f8 PB |
5619 | |
5620 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5621 | goto punt; | |
4a0a7a18 JA |
5622 | goto again; |
5623 | } | |
4bc4494e | 5624 | exit: |
193155c8 JA |
5625 | if (old_creds) |
5626 | revert_creds(old_creds); | |
2b188cc1 JA |
5627 | } |
5628 | ||
3529d8c2 | 5629 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5630 | { |
5631 | int ret; | |
5632 | ||
3529d8c2 | 5633 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5634 | if (ret) { |
5635 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5636 | fail_req: |
78e19bbe | 5637 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5638 | req_set_fail_links(req); |
78e19bbe | 5639 | io_double_put_req(req); |
4fe2c963 | 5640 | } |
2550878f | 5641 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5642 | ret = io_req_defer_prep(req, sqe); |
5643 | if (unlikely(ret < 0)) | |
5644 | goto fail_req; | |
ce35a47a JA |
5645 | /* |
5646 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5647 | * to async execution. | |
5648 | */ | |
5649 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5650 | io_queue_async_work(req); | |
5651 | } else { | |
3529d8c2 | 5652 | __io_queue_sqe(req, sqe); |
ce35a47a | 5653 | } |
4fe2c963 JL |
5654 | } |
5655 | ||
1b4a51b6 | 5656 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5657 | { |
94ae5e77 | 5658 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5659 | io_cqring_add_event(req, -ECANCELED); |
5660 | io_double_put_req(req); | |
5661 | } else | |
3529d8c2 | 5662 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5663 | } |
5664 | ||
1d4240cc | 5665 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
3529d8c2 | 5666 | struct io_submit_state *state, struct io_kiocb **link) |
9e645e11 | 5667 | { |
a197f664 | 5668 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 5669 | int ret; |
9e645e11 | 5670 | |
9e645e11 JA |
5671 | /* |
5672 | * If we already have a head request, queue this one for async | |
5673 | * submittal once the head completes. If we don't have a head but | |
5674 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5675 | * submitted sync once the chain is complete. If none of those | |
5676 | * conditions are true (normal request), then just queue it. | |
5677 | */ | |
5678 | if (*link) { | |
9d76377f | 5679 | struct io_kiocb *head = *link; |
4e88d6e7 | 5680 | |
8cdf2193 PB |
5681 | /* |
5682 | * Taking sequential execution of a link, draining both sides | |
5683 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5684 | * requests in the link. So, it drains the head and the | |
5685 | * next after the link request. The last one is done via | |
5686 | * drain_next flag to persist the effect across calls. | |
5687 | */ | |
ef4ff581 | 5688 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
5689 | head->flags |= REQ_F_IO_DRAIN; |
5690 | ctx->drain_next = 1; | |
5691 | } | |
1d4240cc PB |
5692 | if (io_alloc_async_ctx(req)) |
5693 | return -EAGAIN; | |
9e645e11 | 5694 | |
3529d8c2 | 5695 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5696 | if (ret) { |
4e88d6e7 | 5697 | /* fail even hard links since we don't submit */ |
9d76377f | 5698 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 5699 | return ret; |
2d28390a | 5700 | } |
9d76377f PB |
5701 | trace_io_uring_link(ctx, req, head); |
5702 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5703 | |
5704 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 5705 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
32fe525b PB |
5706 | io_queue_link_head(head); |
5707 | *link = NULL; | |
5708 | } | |
9e645e11 | 5709 | } else { |
711be031 PB |
5710 | if (unlikely(ctx->drain_next)) { |
5711 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 5712 | ctx->drain_next = 0; |
711be031 | 5713 | } |
ef4ff581 | 5714 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
dea3b49c | 5715 | req->flags |= REQ_F_LINK_HEAD; |
711be031 | 5716 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f | 5717 | |
1d4240cc PB |
5718 | if (io_alloc_async_ctx(req)) |
5719 | return -EAGAIN; | |
5720 | ||
711be031 PB |
5721 | ret = io_req_defer_prep(req, sqe); |
5722 | if (ret) | |
5723 | req->flags |= REQ_F_FAIL_LINK; | |
5724 | *link = req; | |
5725 | } else { | |
5726 | io_queue_sqe(req, sqe); | |
5727 | } | |
9e645e11 | 5728 | } |
2e6e1fde | 5729 | |
1d4240cc | 5730 | return 0; |
9e645e11 JA |
5731 | } |
5732 | ||
9a56a232 JA |
5733 | /* |
5734 | * Batched submission is done, ensure local IO is flushed out. | |
5735 | */ | |
5736 | static void io_submit_state_end(struct io_submit_state *state) | |
5737 | { | |
5738 | blk_finish_plug(&state->plug); | |
3d6770fb | 5739 | io_file_put(state); |
2579f913 | 5740 | if (state->free_reqs) |
6c8a3134 | 5741 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5742 | } |
5743 | ||
5744 | /* | |
5745 | * Start submission side cache. | |
5746 | */ | |
5747 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5748 | unsigned int max_ios) |
9a56a232 JA |
5749 | { |
5750 | blk_start_plug(&state->plug); | |
2579f913 | 5751 | state->free_reqs = 0; |
9a56a232 JA |
5752 | state->file = NULL; |
5753 | state->ios_left = max_ios; | |
5754 | } | |
5755 | ||
2b188cc1 JA |
5756 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5757 | { | |
75b28aff | 5758 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5759 | |
caf582c6 PB |
5760 | /* |
5761 | * Ensure any loads from the SQEs are done at this point, | |
5762 | * since once we write the new head, the application could | |
5763 | * write new data to them. | |
5764 | */ | |
5765 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5766 | } |
5767 | ||
2b188cc1 | 5768 | /* |
3529d8c2 | 5769 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5770 | * that is mapped by userspace. This means that care needs to be taken to |
5771 | * ensure that reads are stable, as we cannot rely on userspace always | |
5772 | * being a good citizen. If members of the sqe are validated and then later | |
5773 | * used, it's important that those reads are done through READ_ONCE() to | |
5774 | * prevent a re-load down the line. | |
5775 | */ | |
709b302f | 5776 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 5777 | { |
75b28aff | 5778 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5779 | unsigned head; |
5780 | ||
5781 | /* | |
5782 | * The cached sq head (or cq tail) serves two purposes: | |
5783 | * | |
5784 | * 1) allows us to batch the cost of updating the user visible | |
5785 | * head updates. | |
5786 | * 2) allows the kernel side to track the head on its own, even | |
5787 | * though the application is the one updating it. | |
5788 | */ | |
ee7d46d9 | 5789 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
709b302f PB |
5790 | if (likely(head < ctx->sq_entries)) |
5791 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
5792 | |
5793 | /* drop invalid entries */ | |
498ccd9e | 5794 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5795 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
5796 | return NULL; |
5797 | } | |
5798 | ||
5799 | static inline void io_consume_sqe(struct io_ring_ctx *ctx) | |
5800 | { | |
5801 | ctx->cached_sq_head++; | |
2b188cc1 JA |
5802 | } |
5803 | ||
ef4ff581 PB |
5804 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
5805 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
5806 | IOSQE_BUFFER_SELECT) | |
5807 | ||
5808 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
5809 | const struct io_uring_sqe *sqe, | |
5810 | struct io_submit_state *state, bool async) | |
0553b8bd | 5811 | { |
ef4ff581 PB |
5812 | unsigned int sqe_flags; |
5813 | int id, fd; | |
5814 | ||
0553b8bd PB |
5815 | /* |
5816 | * All io need record the previous position, if LINK vs DARIN, | |
5817 | * it can be used to mark the position of the first IO in the | |
5818 | * link list. | |
5819 | */ | |
31af27c7 | 5820 | req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; |
0553b8bd PB |
5821 | req->opcode = READ_ONCE(sqe->opcode); |
5822 | req->user_data = READ_ONCE(sqe->user_data); | |
5823 | req->io = NULL; | |
5824 | req->file = NULL; | |
5825 | req->ctx = ctx; | |
5826 | req->flags = 0; | |
5827 | /* one is dropped after submission, the other at completion */ | |
5828 | refcount_set(&req->refs, 2); | |
5829 | req->task = NULL; | |
5830 | req->result = 0; | |
ef4ff581 | 5831 | req->needs_fixed_file = async; |
0553b8bd | 5832 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
ef4ff581 PB |
5833 | |
5834 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
5835 | return -EINVAL; | |
5836 | ||
5837 | if (io_op_defs[req->opcode].needs_mm && !current->mm) { | |
5838 | if (unlikely(!mmget_not_zero(ctx->sqo_mm))) | |
5839 | return -EFAULT; | |
5840 | use_mm(ctx->sqo_mm); | |
5841 | } | |
5842 | ||
5843 | sqe_flags = READ_ONCE(sqe->flags); | |
5844 | /* enforce forwards compatibility on users */ | |
5845 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
5846 | return -EINVAL; | |
5847 | ||
5848 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
5849 | !io_op_defs[req->opcode].buffer_select) | |
5850 | return -EOPNOTSUPP; | |
5851 | ||
5852 | id = READ_ONCE(sqe->personality); | |
5853 | if (id) { | |
5854 | req->work.creds = idr_find(&ctx->personality_idr, id); | |
5855 | if (unlikely(!req->work.creds)) | |
5856 | return -EINVAL; | |
5857 | get_cred(req->work.creds); | |
5858 | } | |
5859 | ||
5860 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
5861 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | | |
5862 | IOSQE_ASYNC | IOSQE_FIXED_FILE | | |
5863 | IOSQE_BUFFER_SELECT | IOSQE_IO_LINK); | |
5864 | ||
5865 | fd = READ_ONCE(sqe->fd); | |
5866 | return io_req_set_file(state, req, fd, sqe_flags); | |
0553b8bd PB |
5867 | } |
5868 | ||
fb5ccc98 | 5869 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
bf9c2f1c | 5870 | struct file *ring_file, int ring_fd, bool async) |
6c271ce2 JA |
5871 | { |
5872 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5873 | struct io_kiocb *link = NULL; |
9e645e11 | 5874 | int i, submitted = 0; |
6c271ce2 | 5875 | |
c4a2ed72 | 5876 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5877 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5878 | if (!list_empty(&ctx->cq_overflow_list) && | |
5879 | !io_cqring_overflow_flush(ctx, false)) | |
5880 | return -EBUSY; | |
5881 | } | |
6c271ce2 | 5882 | |
ee7d46d9 PB |
5883 | /* make sure SQ entry isn't read before tail */ |
5884 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5885 | |
2b85edfc PB |
5886 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5887 | return -EAGAIN; | |
6c271ce2 JA |
5888 | |
5889 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5890 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5891 | statep = &state; |
5892 | } | |
5893 | ||
b14cca0c PB |
5894 | ctx->ring_fd = ring_fd; |
5895 | ctx->ring_file = ring_file; | |
5896 | ||
6c271ce2 | 5897 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5898 | const struct io_uring_sqe *sqe; |
196be95c | 5899 | struct io_kiocb *req; |
1cb1edb2 | 5900 | int err; |
fb5ccc98 | 5901 | |
b1e50e54 PB |
5902 | sqe = io_get_sqe(ctx); |
5903 | if (unlikely(!sqe)) { | |
5904 | io_consume_sqe(ctx); | |
5905 | break; | |
5906 | } | |
0553b8bd | 5907 | req = io_alloc_req(ctx, statep); |
196be95c PB |
5908 | if (unlikely(!req)) { |
5909 | if (!submitted) | |
5910 | submitted = -EAGAIN; | |
fb5ccc98 | 5911 | break; |
196be95c | 5912 | } |
fb5ccc98 | 5913 | |
ef4ff581 | 5914 | err = io_init_req(ctx, req, sqe, statep, async); |
709b302f | 5915 | io_consume_sqe(ctx); |
d3656344 JA |
5916 | /* will complete beyond this point, count as submitted */ |
5917 | submitted++; | |
5918 | ||
ef4ff581 | 5919 | if (unlikely(err)) { |
1cb1edb2 PB |
5920 | fail_req: |
5921 | io_cqring_add_event(req, err); | |
d3656344 | 5922 | io_double_put_req(req); |
196be95c PB |
5923 | break; |
5924 | } | |
fb5ccc98 | 5925 | |
354420f7 JA |
5926 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5927 | true, async); | |
1d4240cc PB |
5928 | err = io_submit_sqe(req, sqe, statep, &link); |
5929 | if (err) | |
5930 | goto fail_req; | |
6c271ce2 JA |
5931 | } |
5932 | ||
9466f437 PB |
5933 | if (unlikely(submitted != nr)) { |
5934 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5935 | ||
5936 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5937 | } | |
9e645e11 | 5938 | if (link) |
1b4a51b6 | 5939 | io_queue_link_head(link); |
6c271ce2 JA |
5940 | if (statep) |
5941 | io_submit_state_end(&state); | |
5942 | ||
ae9428ca PB |
5943 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5944 | io_commit_sqring(ctx); | |
5945 | ||
6c271ce2 JA |
5946 | return submitted; |
5947 | } | |
5948 | ||
bf9c2f1c PB |
5949 | static inline void io_sq_thread_drop_mm(struct io_ring_ctx *ctx) |
5950 | { | |
5951 | struct mm_struct *mm = current->mm; | |
5952 | ||
5953 | if (mm) { | |
5954 | unuse_mm(mm); | |
5955 | mmput(mm); | |
5956 | } | |
5957 | } | |
5958 | ||
6c271ce2 JA |
5959 | static int io_sq_thread(void *data) |
5960 | { | |
6c271ce2 | 5961 | struct io_ring_ctx *ctx = data; |
181e448d | 5962 | const struct cred *old_cred; |
6c271ce2 JA |
5963 | mm_segment_t old_fs; |
5964 | DEFINE_WAIT(wait); | |
6c271ce2 | 5965 | unsigned long timeout; |
bdcd3eab | 5966 | int ret = 0; |
6c271ce2 | 5967 | |
206aefde | 5968 | complete(&ctx->completions[1]); |
a4c0b3de | 5969 | |
6c271ce2 JA |
5970 | old_fs = get_fs(); |
5971 | set_fs(USER_DS); | |
181e448d | 5972 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5973 | |
bdcd3eab | 5974 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5975 | while (!kthread_should_park()) { |
fb5ccc98 | 5976 | unsigned int to_submit; |
6c271ce2 | 5977 | |
bdcd3eab | 5978 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5979 | unsigned nr_events = 0; |
5980 | ||
bdcd3eab XW |
5981 | mutex_lock(&ctx->uring_lock); |
5982 | if (!list_empty(&ctx->poll_list)) | |
5983 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5984 | else | |
6c271ce2 | 5985 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5986 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5987 | } |
5988 | ||
fb5ccc98 | 5989 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5990 | |
5991 | /* | |
5992 | * If submit got -EBUSY, flag us as needing the application | |
5993 | * to enter the kernel to reap and flush events. | |
5994 | */ | |
5995 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5996 | /* |
5997 | * Drop cur_mm before scheduling, we can't hold it for | |
5998 | * long periods (or over schedule()). Do this before | |
5999 | * adding ourselves to the waitqueue, as the unuse/drop | |
6000 | * may sleep. | |
6001 | */ | |
bf9c2f1c | 6002 | io_sq_thread_drop_mm(ctx); |
7143b5ac | 6003 | |
6c271ce2 JA |
6004 | /* |
6005 | * We're polling. If we're within the defined idle | |
6006 | * period, then let us spin without work before going | |
c1edbf5f JA |
6007 | * to sleep. The exception is if we got EBUSY doing |
6008 | * more IO, we should wait for the application to | |
6009 | * reap events and wake us up. | |
6c271ce2 | 6010 | */ |
bdcd3eab | 6011 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
6012 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
6013 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
6014 | if (current->task_works) |
6015 | task_work_run(); | |
9831a90c | 6016 | cond_resched(); |
6c271ce2 JA |
6017 | continue; |
6018 | } | |
6019 | ||
6c271ce2 JA |
6020 | prepare_to_wait(&ctx->sqo_wait, &wait, |
6021 | TASK_INTERRUPTIBLE); | |
6022 | ||
bdcd3eab XW |
6023 | /* |
6024 | * While doing polled IO, before going to sleep, we need | |
6025 | * to check if there are new reqs added to poll_list, it | |
6026 | * is because reqs may have been punted to io worker and | |
6027 | * will be added to poll_list later, hence check the | |
6028 | * poll_list again. | |
6029 | */ | |
6030 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6031 | !list_empty_careful(&ctx->poll_list)) { | |
6032 | finish_wait(&ctx->sqo_wait, &wait); | |
6033 | continue; | |
6034 | } | |
6035 | ||
6c271ce2 | 6036 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 6037 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
6038 | /* make sure to read SQ tail after writing flags */ |
6039 | smp_mb(); | |
6c271ce2 | 6040 | |
fb5ccc98 | 6041 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 6042 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 6043 | if (kthread_should_park()) { |
6c271ce2 JA |
6044 | finish_wait(&ctx->sqo_wait, &wait); |
6045 | break; | |
6046 | } | |
b41e9852 JA |
6047 | if (current->task_works) { |
6048 | task_work_run(); | |
10bea96d | 6049 | finish_wait(&ctx->sqo_wait, &wait); |
b41e9852 JA |
6050 | continue; |
6051 | } | |
6c271ce2 JA |
6052 | if (signal_pending(current)) |
6053 | flush_signals(current); | |
6054 | schedule(); | |
6055 | finish_wait(&ctx->sqo_wait, &wait); | |
6056 | ||
75b28aff | 6057 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6058 | continue; |
6059 | } | |
6060 | finish_wait(&ctx->sqo_wait, &wait); | |
6061 | ||
75b28aff | 6062 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
6063 | } |
6064 | ||
8a4955ff | 6065 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 6066 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, true); |
8a4955ff | 6067 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 6068 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
6069 | } |
6070 | ||
b41e9852 JA |
6071 | if (current->task_works) |
6072 | task_work_run(); | |
6073 | ||
6c271ce2 | 6074 | set_fs(old_fs); |
bf9c2f1c | 6075 | io_sq_thread_drop_mm(ctx); |
181e448d | 6076 | revert_creds(old_cred); |
06058632 | 6077 | |
2bbcd6d3 | 6078 | kthread_parkme(); |
06058632 | 6079 | |
6c271ce2 JA |
6080 | return 0; |
6081 | } | |
6082 | ||
bda52162 JA |
6083 | struct io_wait_queue { |
6084 | struct wait_queue_entry wq; | |
6085 | struct io_ring_ctx *ctx; | |
6086 | unsigned to_wait; | |
6087 | unsigned nr_timeouts; | |
6088 | }; | |
6089 | ||
1d7bb1d5 | 6090 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6091 | { |
6092 | struct io_ring_ctx *ctx = iowq->ctx; | |
6093 | ||
6094 | /* | |
d195a66e | 6095 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6096 | * started waiting. For timeouts, we always want to return to userspace, |
6097 | * regardless of event count. | |
6098 | */ | |
1d7bb1d5 | 6099 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6100 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6101 | } | |
6102 | ||
6103 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6104 | int wake_flags, void *key) | |
6105 | { | |
6106 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6107 | wq); | |
6108 | ||
1d7bb1d5 JA |
6109 | /* use noflush == true, as we can't safely rely on locking context */ |
6110 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6111 | return -1; |
6112 | ||
6113 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6114 | } | |
6115 | ||
2b188cc1 JA |
6116 | /* |
6117 | * Wait until events become available, if we don't already have some. The | |
6118 | * application must reap them itself, as they reside on the shared cq ring. | |
6119 | */ | |
6120 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6121 | const sigset_t __user *sig, size_t sigsz) | |
6122 | { | |
bda52162 JA |
6123 | struct io_wait_queue iowq = { |
6124 | .wq = { | |
6125 | .private = current, | |
6126 | .func = io_wake_function, | |
6127 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6128 | }, | |
6129 | .ctx = ctx, | |
6130 | .to_wait = min_events, | |
6131 | }; | |
75b28aff | 6132 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6133 | int ret = 0; |
2b188cc1 | 6134 | |
b41e9852 JA |
6135 | do { |
6136 | if (io_cqring_events(ctx, false) >= min_events) | |
6137 | return 0; | |
6138 | if (!current->task_works) | |
6139 | break; | |
6140 | task_work_run(); | |
6141 | } while (1); | |
2b188cc1 JA |
6142 | |
6143 | if (sig) { | |
9e75ad5d AB |
6144 | #ifdef CONFIG_COMPAT |
6145 | if (in_compat_syscall()) | |
6146 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6147 | sigsz); |
9e75ad5d AB |
6148 | else |
6149 | #endif | |
b772434b | 6150 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6151 | |
2b188cc1 JA |
6152 | if (ret) |
6153 | return ret; | |
6154 | } | |
6155 | ||
bda52162 | 6156 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6157 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6158 | do { |
6159 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6160 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6161 | if (current->task_works) |
6162 | task_work_run(); | |
1d7bb1d5 | 6163 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6164 | break; |
6165 | schedule(); | |
6166 | if (signal_pending(current)) { | |
e9ffa5c2 | 6167 | ret = -EINTR; |
bda52162 JA |
6168 | break; |
6169 | } | |
6170 | } while (1); | |
6171 | finish_wait(&ctx->wait, &iowq.wq); | |
6172 | ||
e9ffa5c2 | 6173 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6174 | |
75b28aff | 6175 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6176 | } |
6177 | ||
6b06314c JA |
6178 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6179 | { | |
6180 | #if defined(CONFIG_UNIX) | |
6181 | if (ctx->ring_sock) { | |
6182 | struct sock *sock = ctx->ring_sock->sk; | |
6183 | struct sk_buff *skb; | |
6184 | ||
6185 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6186 | kfree_skb(skb); | |
6187 | } | |
6188 | #else | |
6189 | int i; | |
6190 | ||
65e19f54 JA |
6191 | for (i = 0; i < ctx->nr_user_files; i++) { |
6192 | struct file *file; | |
6193 | ||
6194 | file = io_file_from_index(ctx, i); | |
6195 | if (file) | |
6196 | fput(file); | |
6197 | } | |
6b06314c JA |
6198 | #endif |
6199 | } | |
6200 | ||
05f3fb3c JA |
6201 | static void io_file_ref_kill(struct percpu_ref *ref) |
6202 | { | |
6203 | struct fixed_file_data *data; | |
6204 | ||
6205 | data = container_of(ref, struct fixed_file_data, refs); | |
6206 | complete(&data->done); | |
6207 | } | |
6208 | ||
6b06314c JA |
6209 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6210 | { | |
05f3fb3c | 6211 | struct fixed_file_data *data = ctx->file_data; |
05589553 | 6212 | struct fixed_file_ref_node *ref_node = NULL; |
65e19f54 | 6213 | unsigned nr_tables, i; |
05589553 | 6214 | unsigned long flags; |
65e19f54 | 6215 | |
05f3fb3c | 6216 | if (!data) |
6b06314c JA |
6217 | return -ENXIO; |
6218 | ||
05589553 XW |
6219 | spin_lock_irqsave(&data->lock, flags); |
6220 | if (!list_empty(&data->ref_list)) | |
6221 | ref_node = list_first_entry(&data->ref_list, | |
6222 | struct fixed_file_ref_node, node); | |
6223 | spin_unlock_irqrestore(&data->lock, flags); | |
6224 | if (ref_node) | |
6225 | percpu_ref_kill(&ref_node->refs); | |
6226 | ||
6227 | percpu_ref_kill(&data->refs); | |
6228 | ||
6229 | /* wait for all refs nodes to complete */ | |
2faf852d | 6230 | wait_for_completion(&data->done); |
05f3fb3c | 6231 | |
6b06314c | 6232 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6233 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6234 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6235 | kfree(data->table[i].files); |
6236 | kfree(data->table); | |
05589553 XW |
6237 | percpu_ref_exit(&data->refs); |
6238 | kfree(data); | |
05f3fb3c | 6239 | ctx->file_data = NULL; |
6b06314c JA |
6240 | ctx->nr_user_files = 0; |
6241 | return 0; | |
6242 | } | |
6243 | ||
6c271ce2 JA |
6244 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6245 | { | |
6246 | if (ctx->sqo_thread) { | |
206aefde | 6247 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6248 | /* |
6249 | * The park is a bit of a work-around, without it we get | |
6250 | * warning spews on shutdown with SQPOLL set and affinity | |
6251 | * set to a single CPU. | |
6252 | */ | |
06058632 | 6253 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6254 | kthread_stop(ctx->sqo_thread); |
6255 | ctx->sqo_thread = NULL; | |
6256 | } | |
6257 | } | |
6258 | ||
6b06314c JA |
6259 | static void io_finish_async(struct io_ring_ctx *ctx) |
6260 | { | |
6c271ce2 JA |
6261 | io_sq_thread_stop(ctx); |
6262 | ||
561fb04a JA |
6263 | if (ctx->io_wq) { |
6264 | io_wq_destroy(ctx->io_wq); | |
6265 | ctx->io_wq = NULL; | |
6b06314c JA |
6266 | } |
6267 | } | |
6268 | ||
6269 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6270 | /* |
6271 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6272 | * the io_uring can be safely unregistered on process exit, even if we have | |
6273 | * loops in the file referencing. | |
6274 | */ | |
6275 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6276 | { | |
6277 | struct sock *sk = ctx->ring_sock->sk; | |
6278 | struct scm_fp_list *fpl; | |
6279 | struct sk_buff *skb; | |
08a45173 | 6280 | int i, nr_files; |
6b06314c | 6281 | |
6b06314c JA |
6282 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
6283 | if (!fpl) | |
6284 | return -ENOMEM; | |
6285 | ||
6286 | skb = alloc_skb(0, GFP_KERNEL); | |
6287 | if (!skb) { | |
6288 | kfree(fpl); | |
6289 | return -ENOMEM; | |
6290 | } | |
6291 | ||
6292 | skb->sk = sk; | |
6b06314c | 6293 | |
08a45173 | 6294 | nr_files = 0; |
6b06314c JA |
6295 | fpl->user = get_uid(ctx->user); |
6296 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6297 | struct file *file = io_file_from_index(ctx, i + offset); |
6298 | ||
6299 | if (!file) | |
08a45173 | 6300 | continue; |
65e19f54 | 6301 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6302 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6303 | nr_files++; | |
6b06314c JA |
6304 | } |
6305 | ||
08a45173 JA |
6306 | if (nr_files) { |
6307 | fpl->max = SCM_MAX_FD; | |
6308 | fpl->count = nr_files; | |
6309 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6310 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6311 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6312 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6313 | |
08a45173 JA |
6314 | for (i = 0; i < nr_files; i++) |
6315 | fput(fpl->fp[i]); | |
6316 | } else { | |
6317 | kfree_skb(skb); | |
6318 | kfree(fpl); | |
6319 | } | |
6b06314c JA |
6320 | |
6321 | return 0; | |
6322 | } | |
6323 | ||
6324 | /* | |
6325 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6326 | * causes regular reference counting to break down. We rely on the UNIX | |
6327 | * garbage collection to take care of this problem for us. | |
6328 | */ | |
6329 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6330 | { | |
6331 | unsigned left, total; | |
6332 | int ret = 0; | |
6333 | ||
6334 | total = 0; | |
6335 | left = ctx->nr_user_files; | |
6336 | while (left) { | |
6337 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6338 | |
6339 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6340 | if (ret) | |
6341 | break; | |
6342 | left -= this_files; | |
6343 | total += this_files; | |
6344 | } | |
6345 | ||
6346 | if (!ret) | |
6347 | return 0; | |
6348 | ||
6349 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6350 | struct file *file = io_file_from_index(ctx, total); |
6351 | ||
6352 | if (file) | |
6353 | fput(file); | |
6b06314c JA |
6354 | total++; |
6355 | } | |
6356 | ||
6357 | return ret; | |
6358 | } | |
6359 | #else | |
6360 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6361 | { | |
6362 | return 0; | |
6363 | } | |
6364 | #endif | |
6365 | ||
65e19f54 JA |
6366 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6367 | unsigned nr_files) | |
6368 | { | |
6369 | int i; | |
6370 | ||
6371 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6372 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6373 | unsigned this_files; |
6374 | ||
6375 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6376 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6377 | GFP_KERNEL); | |
6378 | if (!table->files) | |
6379 | break; | |
6380 | nr_files -= this_files; | |
6381 | } | |
6382 | ||
6383 | if (i == nr_tables) | |
6384 | return 0; | |
6385 | ||
6386 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6387 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6388 | kfree(table->files); |
6389 | } | |
6390 | return 1; | |
6391 | } | |
6392 | ||
05f3fb3c JA |
6393 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6394 | { | |
6395 | #if defined(CONFIG_UNIX) | |
6396 | struct sock *sock = ctx->ring_sock->sk; | |
6397 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6398 | struct sk_buff *skb; | |
6399 | int i; | |
6400 | ||
6401 | __skb_queue_head_init(&list); | |
6402 | ||
6403 | /* | |
6404 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6405 | * remove this entry and rearrange the file array. | |
6406 | */ | |
6407 | skb = skb_dequeue(head); | |
6408 | while (skb) { | |
6409 | struct scm_fp_list *fp; | |
6410 | ||
6411 | fp = UNIXCB(skb).fp; | |
6412 | for (i = 0; i < fp->count; i++) { | |
6413 | int left; | |
6414 | ||
6415 | if (fp->fp[i] != file) | |
6416 | continue; | |
6417 | ||
6418 | unix_notinflight(fp->user, fp->fp[i]); | |
6419 | left = fp->count - 1 - i; | |
6420 | if (left) { | |
6421 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6422 | left * sizeof(struct file *)); | |
6423 | } | |
6424 | fp->count--; | |
6425 | if (!fp->count) { | |
6426 | kfree_skb(skb); | |
6427 | skb = NULL; | |
6428 | } else { | |
6429 | __skb_queue_tail(&list, skb); | |
6430 | } | |
6431 | fput(file); | |
6432 | file = NULL; | |
6433 | break; | |
6434 | } | |
6435 | ||
6436 | if (!file) | |
6437 | break; | |
6438 | ||
6439 | __skb_queue_tail(&list, skb); | |
6440 | ||
6441 | skb = skb_dequeue(head); | |
6442 | } | |
6443 | ||
6444 | if (skb_peek(&list)) { | |
6445 | spin_lock_irq(&head->lock); | |
6446 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6447 | __skb_queue_tail(head, skb); | |
6448 | spin_unlock_irq(&head->lock); | |
6449 | } | |
6450 | #else | |
6451 | fput(file); | |
6452 | #endif | |
6453 | } | |
6454 | ||
6455 | struct io_file_put { | |
05589553 | 6456 | struct list_head list; |
05f3fb3c | 6457 | struct file *file; |
05f3fb3c JA |
6458 | }; |
6459 | ||
05589553 | 6460 | static void io_file_put_work(struct work_struct *work) |
65e19f54 | 6461 | { |
05589553 XW |
6462 | struct fixed_file_ref_node *ref_node; |
6463 | struct fixed_file_data *file_data; | |
6464 | struct io_ring_ctx *ctx; | |
05f3fb3c | 6465 | struct io_file_put *pfile, *tmp; |
05589553 | 6466 | unsigned long flags; |
65e19f54 | 6467 | |
05589553 XW |
6468 | ref_node = container_of(work, struct fixed_file_ref_node, work); |
6469 | file_data = ref_node->file_data; | |
6470 | ctx = file_data->ctx; | |
6471 | ||
6472 | list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) { | |
6473 | list_del_init(&pfile->list); | |
6474 | io_ring_file_put(ctx, pfile->file); | |
6475 | kfree(pfile); | |
65e19f54 | 6476 | } |
05589553 XW |
6477 | |
6478 | spin_lock_irqsave(&file_data->lock, flags); | |
6479 | list_del_init(&ref_node->node); | |
6480 | spin_unlock_irqrestore(&file_data->lock, flags); | |
6481 | ||
6482 | percpu_ref_exit(&ref_node->refs); | |
6483 | kfree(ref_node); | |
6484 | percpu_ref_put(&file_data->refs); | |
2faf852d | 6485 | } |
65e19f54 | 6486 | |
05589553 | 6487 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
2faf852d | 6488 | { |
05589553 | 6489 | struct fixed_file_ref_node *ref_node; |
65e19f54 | 6490 | |
05589553 XW |
6491 | ref_node = container_of(ref, struct fixed_file_ref_node, refs); |
6492 | ||
6493 | queue_work(system_wq, &ref_node->work); | |
05f3fb3c | 6494 | } |
65e19f54 | 6495 | |
05589553 XW |
6496 | static struct fixed_file_ref_node *alloc_fixed_file_ref_node( |
6497 | struct io_ring_ctx *ctx) | |
05f3fb3c | 6498 | { |
05589553 | 6499 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6500 | |
05589553 XW |
6501 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
6502 | if (!ref_node) | |
6503 | return ERR_PTR(-ENOMEM); | |
05f3fb3c | 6504 | |
05589553 XW |
6505 | if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero, |
6506 | 0, GFP_KERNEL)) { | |
6507 | kfree(ref_node); | |
6508 | return ERR_PTR(-ENOMEM); | |
6509 | } | |
6510 | INIT_LIST_HEAD(&ref_node->node); | |
6511 | INIT_LIST_HEAD(&ref_node->file_list); | |
6512 | INIT_WORK(&ref_node->work, io_file_put_work); | |
6513 | ref_node->file_data = ctx->file_data; | |
6514 | return ref_node; | |
6515 | ||
6516 | } | |
6517 | ||
6518 | static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node) | |
6519 | { | |
6520 | percpu_ref_exit(&ref_node->refs); | |
6521 | kfree(ref_node); | |
65e19f54 JA |
6522 | } |
6523 | ||
6b06314c JA |
6524 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6525 | unsigned nr_args) | |
6526 | { | |
6527 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6528 | unsigned nr_tables; |
05f3fb3c | 6529 | struct file *file; |
6b06314c JA |
6530 | int fd, ret = 0; |
6531 | unsigned i; | |
05589553 XW |
6532 | struct fixed_file_ref_node *ref_node; |
6533 | unsigned long flags; | |
6b06314c | 6534 | |
05f3fb3c | 6535 | if (ctx->file_data) |
6b06314c JA |
6536 | return -EBUSY; |
6537 | if (!nr_args) | |
6538 | return -EINVAL; | |
6539 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6540 | return -EMFILE; | |
6541 | ||
05f3fb3c JA |
6542 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6543 | if (!ctx->file_data) | |
6544 | return -ENOMEM; | |
6545 | ctx->file_data->ctx = ctx; | |
6546 | init_completion(&ctx->file_data->done); | |
05589553 | 6547 | INIT_LIST_HEAD(&ctx->file_data->ref_list); |
f7fe9346 | 6548 | spin_lock_init(&ctx->file_data->lock); |
05f3fb3c | 6549 | |
65e19f54 | 6550 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6551 | ctx->file_data->table = kcalloc(nr_tables, |
6552 | sizeof(struct fixed_file_table), | |
65e19f54 | 6553 | GFP_KERNEL); |
05f3fb3c JA |
6554 | if (!ctx->file_data->table) { |
6555 | kfree(ctx->file_data); | |
6556 | ctx->file_data = NULL; | |
6b06314c | 6557 | return -ENOMEM; |
05f3fb3c JA |
6558 | } |
6559 | ||
05589553 | 6560 | if (percpu_ref_init(&ctx->file_data->refs, io_file_ref_kill, |
05f3fb3c JA |
6561 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
6562 | kfree(ctx->file_data->table); | |
6563 | kfree(ctx->file_data); | |
6564 | ctx->file_data = NULL; | |
6b06314c | 6565 | return -ENOMEM; |
05f3fb3c | 6566 | } |
6b06314c | 6567 | |
65e19f54 | 6568 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6569 | percpu_ref_exit(&ctx->file_data->refs); |
6570 | kfree(ctx->file_data->table); | |
6571 | kfree(ctx->file_data); | |
6572 | ctx->file_data = NULL; | |
65e19f54 JA |
6573 | return -ENOMEM; |
6574 | } | |
6575 | ||
08a45173 | 6576 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6577 | struct fixed_file_table *table; |
6578 | unsigned index; | |
6579 | ||
6b06314c JA |
6580 | ret = -EFAULT; |
6581 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6582 | break; | |
08a45173 JA |
6583 | /* allow sparse sets */ |
6584 | if (fd == -1) { | |
6585 | ret = 0; | |
6586 | continue; | |
6587 | } | |
6b06314c | 6588 | |
05f3fb3c | 6589 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6590 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6591 | file = fget(fd); |
6b06314c JA |
6592 | |
6593 | ret = -EBADF; | |
05f3fb3c | 6594 | if (!file) |
6b06314c | 6595 | break; |
05f3fb3c | 6596 | |
6b06314c JA |
6597 | /* |
6598 | * Don't allow io_uring instances to be registered. If UNIX | |
6599 | * isn't enabled, then this causes a reference cycle and this | |
6600 | * instance can never get freed. If UNIX is enabled we'll | |
6601 | * handle it just fine, but there's still no point in allowing | |
6602 | * a ring fd as it doesn't support regular read/write anyway. | |
6603 | */ | |
05f3fb3c JA |
6604 | if (file->f_op == &io_uring_fops) { |
6605 | fput(file); | |
6b06314c JA |
6606 | break; |
6607 | } | |
6b06314c | 6608 | ret = 0; |
05f3fb3c | 6609 | table->files[index] = file; |
6b06314c JA |
6610 | } |
6611 | ||
6612 | if (ret) { | |
65e19f54 | 6613 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6614 | file = io_file_from_index(ctx, i); |
6615 | if (file) | |
6616 | fput(file); | |
6617 | } | |
6618 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6619 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6620 | |
05f3fb3c JA |
6621 | kfree(ctx->file_data->table); |
6622 | kfree(ctx->file_data); | |
6623 | ctx->file_data = NULL; | |
6b06314c JA |
6624 | ctx->nr_user_files = 0; |
6625 | return ret; | |
6626 | } | |
6627 | ||
6628 | ret = io_sqe_files_scm(ctx); | |
05589553 | 6629 | if (ret) { |
6b06314c | 6630 | io_sqe_files_unregister(ctx); |
05589553 XW |
6631 | return ret; |
6632 | } | |
6b06314c | 6633 | |
05589553 XW |
6634 | ref_node = alloc_fixed_file_ref_node(ctx); |
6635 | if (IS_ERR(ref_node)) { | |
6636 | io_sqe_files_unregister(ctx); | |
6637 | return PTR_ERR(ref_node); | |
6638 | } | |
6639 | ||
6640 | ctx->file_data->cur_refs = &ref_node->refs; | |
6641 | spin_lock_irqsave(&ctx->file_data->lock, flags); | |
6642 | list_add(&ref_node->node, &ctx->file_data->ref_list); | |
6643 | spin_unlock_irqrestore(&ctx->file_data->lock, flags); | |
6644 | percpu_ref_get(&ctx->file_data->refs); | |
6b06314c JA |
6645 | return ret; |
6646 | } | |
6647 | ||
c3a31e60 JA |
6648 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6649 | int index) | |
6650 | { | |
6651 | #if defined(CONFIG_UNIX) | |
6652 | struct sock *sock = ctx->ring_sock->sk; | |
6653 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6654 | struct sk_buff *skb; | |
6655 | ||
6656 | /* | |
6657 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6658 | * file set. If there's no room, fall back to allocating a new skb | |
6659 | * and filling it in. | |
6660 | */ | |
6661 | spin_lock_irq(&head->lock); | |
6662 | skb = skb_peek(head); | |
6663 | if (skb) { | |
6664 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6665 | ||
6666 | if (fpl->count < SCM_MAX_FD) { | |
6667 | __skb_unlink(skb, head); | |
6668 | spin_unlock_irq(&head->lock); | |
6669 | fpl->fp[fpl->count] = get_file(file); | |
6670 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6671 | fpl->count++; | |
6672 | spin_lock_irq(&head->lock); | |
6673 | __skb_queue_head(head, skb); | |
6674 | } else { | |
6675 | skb = NULL; | |
6676 | } | |
6677 | } | |
6678 | spin_unlock_irq(&head->lock); | |
6679 | ||
6680 | if (skb) { | |
6681 | fput(file); | |
6682 | return 0; | |
6683 | } | |
6684 | ||
6685 | return __io_sqe_files_scm(ctx, 1, index); | |
6686 | #else | |
6687 | return 0; | |
6688 | #endif | |
6689 | } | |
6690 | ||
a5318d3c | 6691 | static int io_queue_file_removal(struct fixed_file_data *data, |
05589553 | 6692 | struct file *file) |
05f3fb3c | 6693 | { |
a5318d3c | 6694 | struct io_file_put *pfile; |
05589553 XW |
6695 | struct percpu_ref *refs = data->cur_refs; |
6696 | struct fixed_file_ref_node *ref_node; | |
05f3fb3c | 6697 | |
05f3fb3c | 6698 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6699 | if (!pfile) |
6700 | return -ENOMEM; | |
05f3fb3c | 6701 | |
05589553 | 6702 | ref_node = container_of(refs, struct fixed_file_ref_node, refs); |
05f3fb3c | 6703 | pfile->file = file; |
05589553 XW |
6704 | list_add(&pfile->list, &ref_node->file_list); |
6705 | ||
a5318d3c | 6706 | return 0; |
05f3fb3c JA |
6707 | } |
6708 | ||
6709 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6710 | struct io_uring_files_update *up, | |
6711 | unsigned nr_args) | |
6712 | { | |
6713 | struct fixed_file_data *data = ctx->file_data; | |
05589553 | 6714 | struct fixed_file_ref_node *ref_node; |
05f3fb3c | 6715 | struct file *file; |
c3a31e60 JA |
6716 | __s32 __user *fds; |
6717 | int fd, i, err; | |
6718 | __u32 done; | |
05589553 XW |
6719 | unsigned long flags; |
6720 | bool needs_switch = false; | |
c3a31e60 | 6721 | |
05f3fb3c | 6722 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6723 | return -EOVERFLOW; |
6724 | if (done > ctx->nr_user_files) | |
6725 | return -EINVAL; | |
6726 | ||
05589553 XW |
6727 | ref_node = alloc_fixed_file_ref_node(ctx); |
6728 | if (IS_ERR(ref_node)) | |
6729 | return PTR_ERR(ref_node); | |
6730 | ||
c3a31e60 | 6731 | done = 0; |
05f3fb3c | 6732 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6733 | while (nr_args) { |
65e19f54 JA |
6734 | struct fixed_file_table *table; |
6735 | unsigned index; | |
6736 | ||
c3a31e60 JA |
6737 | err = 0; |
6738 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6739 | err = -EFAULT; | |
6740 | break; | |
6741 | } | |
05f3fb3c JA |
6742 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6743 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6744 | index = i & IORING_FILE_TABLE_MASK; |
6745 | if (table->files[index]) { | |
05f3fb3c | 6746 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6747 | err = io_queue_file_removal(data, file); |
6748 | if (err) | |
6749 | break; | |
65e19f54 | 6750 | table->files[index] = NULL; |
05589553 | 6751 | needs_switch = true; |
c3a31e60 JA |
6752 | } |
6753 | if (fd != -1) { | |
c3a31e60 JA |
6754 | file = fget(fd); |
6755 | if (!file) { | |
6756 | err = -EBADF; | |
6757 | break; | |
6758 | } | |
6759 | /* | |
6760 | * Don't allow io_uring instances to be registered. If | |
6761 | * UNIX isn't enabled, then this causes a reference | |
6762 | * cycle and this instance can never get freed. If UNIX | |
6763 | * is enabled we'll handle it just fine, but there's | |
6764 | * still no point in allowing a ring fd as it doesn't | |
6765 | * support regular read/write anyway. | |
6766 | */ | |
6767 | if (file->f_op == &io_uring_fops) { | |
6768 | fput(file); | |
6769 | err = -EBADF; | |
6770 | break; | |
6771 | } | |
65e19f54 | 6772 | table->files[index] = file; |
c3a31e60 JA |
6773 | err = io_sqe_file_register(ctx, file, i); |
6774 | if (err) | |
6775 | break; | |
6776 | } | |
6777 | nr_args--; | |
6778 | done++; | |
05f3fb3c JA |
6779 | up->offset++; |
6780 | } | |
6781 | ||
05589553 XW |
6782 | if (needs_switch) { |
6783 | percpu_ref_kill(data->cur_refs); | |
6784 | spin_lock_irqsave(&data->lock, flags); | |
6785 | list_add(&ref_node->node, &data->ref_list); | |
6786 | data->cur_refs = &ref_node->refs; | |
6787 | spin_unlock_irqrestore(&data->lock, flags); | |
6788 | percpu_ref_get(&ctx->file_data->refs); | |
6789 | } else | |
6790 | destroy_fixed_file_ref_node(ref_node); | |
c3a31e60 JA |
6791 | |
6792 | return done ? done : err; | |
6793 | } | |
05589553 | 6794 | |
05f3fb3c JA |
6795 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6796 | unsigned nr_args) | |
6797 | { | |
6798 | struct io_uring_files_update up; | |
6799 | ||
6800 | if (!ctx->file_data) | |
6801 | return -ENXIO; | |
6802 | if (!nr_args) | |
6803 | return -EINVAL; | |
6804 | if (copy_from_user(&up, arg, sizeof(up))) | |
6805 | return -EFAULT; | |
6806 | if (up.resv) | |
6807 | return -EINVAL; | |
6808 | ||
6809 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6810 | } | |
c3a31e60 | 6811 | |
e9fd9396 | 6812 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6813 | { |
6814 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6815 | ||
e9fd9396 | 6816 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6817 | io_put_req(req); |
6818 | } | |
6819 | ||
24369c2e PB |
6820 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6821 | struct io_uring_params *p) | |
6822 | { | |
6823 | struct io_wq_data data; | |
6824 | struct fd f; | |
6825 | struct io_ring_ctx *ctx_attach; | |
6826 | unsigned int concurrency; | |
6827 | int ret = 0; | |
6828 | ||
6829 | data.user = ctx->user; | |
e9fd9396 | 6830 | data.free_work = io_free_work; |
24369c2e PB |
6831 | |
6832 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6833 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6834 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6835 | ||
6836 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6837 | if (IS_ERR(ctx->io_wq)) { | |
6838 | ret = PTR_ERR(ctx->io_wq); | |
6839 | ctx->io_wq = NULL; | |
6840 | } | |
6841 | return ret; | |
6842 | } | |
6843 | ||
6844 | f = fdget(p->wq_fd); | |
6845 | if (!f.file) | |
6846 | return -EBADF; | |
6847 | ||
6848 | if (f.file->f_op != &io_uring_fops) { | |
6849 | ret = -EINVAL; | |
6850 | goto out_fput; | |
6851 | } | |
6852 | ||
6853 | ctx_attach = f.file->private_data; | |
6854 | /* @io_wq is protected by holding the fd */ | |
6855 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6856 | ret = -EINVAL; | |
6857 | goto out_fput; | |
6858 | } | |
6859 | ||
6860 | ctx->io_wq = ctx_attach->io_wq; | |
6861 | out_fput: | |
6862 | fdput(f); | |
6863 | return ret; | |
6864 | } | |
6865 | ||
6c271ce2 JA |
6866 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6867 | struct io_uring_params *p) | |
2b188cc1 JA |
6868 | { |
6869 | int ret; | |
6870 | ||
6c271ce2 | 6871 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6872 | mmgrab(current->mm); |
6873 | ctx->sqo_mm = current->mm; | |
6874 | ||
6c271ce2 | 6875 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6876 | ret = -EPERM; |
6877 | if (!capable(CAP_SYS_ADMIN)) | |
6878 | goto err; | |
6879 | ||
917257da JA |
6880 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6881 | if (!ctx->sq_thread_idle) | |
6882 | ctx->sq_thread_idle = HZ; | |
6883 | ||
6c271ce2 | 6884 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6885 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6886 | |
917257da | 6887 | ret = -EINVAL; |
44a9bd18 JA |
6888 | if (cpu >= nr_cpu_ids) |
6889 | goto err; | |
7889f44d | 6890 | if (!cpu_online(cpu)) |
917257da JA |
6891 | goto err; |
6892 | ||
6c271ce2 JA |
6893 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6894 | ctx, cpu, | |
6895 | "io_uring-sq"); | |
6896 | } else { | |
6897 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6898 | "io_uring-sq"); | |
6899 | } | |
6900 | if (IS_ERR(ctx->sqo_thread)) { | |
6901 | ret = PTR_ERR(ctx->sqo_thread); | |
6902 | ctx->sqo_thread = NULL; | |
6903 | goto err; | |
6904 | } | |
6905 | wake_up_process(ctx->sqo_thread); | |
6906 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6907 | /* Can't have SQ_AFF without SQPOLL */ | |
6908 | ret = -EINVAL; | |
6909 | goto err; | |
6910 | } | |
6911 | ||
24369c2e PB |
6912 | ret = io_init_wq_offload(ctx, p); |
6913 | if (ret) | |
2b188cc1 | 6914 | goto err; |
2b188cc1 JA |
6915 | |
6916 | return 0; | |
6917 | err: | |
54a91f3b | 6918 | io_finish_async(ctx); |
2b188cc1 JA |
6919 | mmdrop(ctx->sqo_mm); |
6920 | ctx->sqo_mm = NULL; | |
6921 | return ret; | |
6922 | } | |
6923 | ||
6924 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6925 | { | |
6926 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6927 | } | |
6928 | ||
6929 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6930 | { | |
6931 | unsigned long page_limit, cur_pages, new_pages; | |
6932 | ||
6933 | /* Don't allow more pages than we can safely lock */ | |
6934 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6935 | ||
6936 | do { | |
6937 | cur_pages = atomic_long_read(&user->locked_vm); | |
6938 | new_pages = cur_pages + nr_pages; | |
6939 | if (new_pages > page_limit) | |
6940 | return -ENOMEM; | |
6941 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6942 | new_pages) != cur_pages); | |
6943 | ||
6944 | return 0; | |
6945 | } | |
6946 | ||
6947 | static void io_mem_free(void *ptr) | |
6948 | { | |
52e04ef4 MR |
6949 | struct page *page; |
6950 | ||
6951 | if (!ptr) | |
6952 | return; | |
2b188cc1 | 6953 | |
52e04ef4 | 6954 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6955 | if (put_page_testzero(page)) |
6956 | free_compound_page(page); | |
6957 | } | |
6958 | ||
6959 | static void *io_mem_alloc(size_t size) | |
6960 | { | |
6961 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6962 | __GFP_NORETRY; | |
6963 | ||
6964 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6965 | } | |
6966 | ||
75b28aff HV |
6967 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6968 | size_t *sq_offset) | |
6969 | { | |
6970 | struct io_rings *rings; | |
6971 | size_t off, sq_array_size; | |
6972 | ||
6973 | off = struct_size(rings, cqes, cq_entries); | |
6974 | if (off == SIZE_MAX) | |
6975 | return SIZE_MAX; | |
6976 | ||
6977 | #ifdef CONFIG_SMP | |
6978 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6979 | if (off == 0) | |
6980 | return SIZE_MAX; | |
6981 | #endif | |
6982 | ||
6983 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6984 | if (sq_array_size == SIZE_MAX) | |
6985 | return SIZE_MAX; | |
6986 | ||
6987 | if (check_add_overflow(off, sq_array_size, &off)) | |
6988 | return SIZE_MAX; | |
6989 | ||
6990 | if (sq_offset) | |
6991 | *sq_offset = off; | |
6992 | ||
6993 | return off; | |
6994 | } | |
6995 | ||
2b188cc1 JA |
6996 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6997 | { | |
75b28aff | 6998 | size_t pages; |
2b188cc1 | 6999 | |
75b28aff HV |
7000 | pages = (size_t)1 << get_order( |
7001 | rings_size(sq_entries, cq_entries, NULL)); | |
7002 | pages += (size_t)1 << get_order( | |
7003 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 7004 | |
75b28aff | 7005 | return pages; |
2b188cc1 JA |
7006 | } |
7007 | ||
edafccee JA |
7008 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
7009 | { | |
7010 | int i, j; | |
7011 | ||
7012 | if (!ctx->user_bufs) | |
7013 | return -ENXIO; | |
7014 | ||
7015 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7016 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7017 | ||
7018 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 7019 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
7020 | |
7021 | if (ctx->account_mem) | |
7022 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 7023 | kvfree(imu->bvec); |
edafccee JA |
7024 | imu->nr_bvecs = 0; |
7025 | } | |
7026 | ||
7027 | kfree(ctx->user_bufs); | |
7028 | ctx->user_bufs = NULL; | |
7029 | ctx->nr_user_bufs = 0; | |
7030 | return 0; | |
7031 | } | |
7032 | ||
7033 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
7034 | void __user *arg, unsigned index) | |
7035 | { | |
7036 | struct iovec __user *src; | |
7037 | ||
7038 | #ifdef CONFIG_COMPAT | |
7039 | if (ctx->compat) { | |
7040 | struct compat_iovec __user *ciovs; | |
7041 | struct compat_iovec ciov; | |
7042 | ||
7043 | ciovs = (struct compat_iovec __user *) arg; | |
7044 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
7045 | return -EFAULT; | |
7046 | ||
d55e5f5b | 7047 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
7048 | dst->iov_len = ciov.iov_len; |
7049 | return 0; | |
7050 | } | |
7051 | #endif | |
7052 | src = (struct iovec __user *) arg; | |
7053 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
7054 | return -EFAULT; | |
7055 | return 0; | |
7056 | } | |
7057 | ||
7058 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
7059 | unsigned nr_args) | |
7060 | { | |
7061 | struct vm_area_struct **vmas = NULL; | |
7062 | struct page **pages = NULL; | |
7063 | int i, j, got_pages = 0; | |
7064 | int ret = -EINVAL; | |
7065 | ||
7066 | if (ctx->user_bufs) | |
7067 | return -EBUSY; | |
7068 | if (!nr_args || nr_args > UIO_MAXIOV) | |
7069 | return -EINVAL; | |
7070 | ||
7071 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
7072 | GFP_KERNEL); | |
7073 | if (!ctx->user_bufs) | |
7074 | return -ENOMEM; | |
7075 | ||
7076 | for (i = 0; i < nr_args; i++) { | |
7077 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
7078 | unsigned long off, start, end, ubuf; | |
7079 | int pret, nr_pages; | |
7080 | struct iovec iov; | |
7081 | size_t size; | |
7082 | ||
7083 | ret = io_copy_iov(ctx, &iov, arg, i); | |
7084 | if (ret) | |
a278682d | 7085 | goto err; |
edafccee JA |
7086 | |
7087 | /* | |
7088 | * Don't impose further limits on the size and buffer | |
7089 | * constraints here, we'll -EINVAL later when IO is | |
7090 | * submitted if they are wrong. | |
7091 | */ | |
7092 | ret = -EFAULT; | |
7093 | if (!iov.iov_base || !iov.iov_len) | |
7094 | goto err; | |
7095 | ||
7096 | /* arbitrary limit, but we need something */ | |
7097 | if (iov.iov_len > SZ_1G) | |
7098 | goto err; | |
7099 | ||
7100 | ubuf = (unsigned long) iov.iov_base; | |
7101 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
7102 | start = ubuf >> PAGE_SHIFT; | |
7103 | nr_pages = end - start; | |
7104 | ||
7105 | if (ctx->account_mem) { | |
7106 | ret = io_account_mem(ctx->user, nr_pages); | |
7107 | if (ret) | |
7108 | goto err; | |
7109 | } | |
7110 | ||
7111 | ret = 0; | |
7112 | if (!pages || nr_pages > got_pages) { | |
7113 | kfree(vmas); | |
7114 | kfree(pages); | |
d4ef6475 | 7115 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7116 | GFP_KERNEL); |
d4ef6475 | 7117 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7118 | sizeof(struct vm_area_struct *), |
7119 | GFP_KERNEL); | |
7120 | if (!pages || !vmas) { | |
7121 | ret = -ENOMEM; | |
7122 | if (ctx->account_mem) | |
7123 | io_unaccount_mem(ctx->user, nr_pages); | |
7124 | goto err; | |
7125 | } | |
7126 | got_pages = nr_pages; | |
7127 | } | |
7128 | ||
d4ef6475 | 7129 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7130 | GFP_KERNEL); |
7131 | ret = -ENOMEM; | |
7132 | if (!imu->bvec) { | |
7133 | if (ctx->account_mem) | |
7134 | io_unaccount_mem(ctx->user, nr_pages); | |
7135 | goto err; | |
7136 | } | |
7137 | ||
7138 | ret = 0; | |
7139 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7140 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7141 | FOLL_WRITE | FOLL_LONGTERM, |
7142 | pages, vmas); | |
edafccee JA |
7143 | if (pret == nr_pages) { |
7144 | /* don't support file backed memory */ | |
7145 | for (j = 0; j < nr_pages; j++) { | |
7146 | struct vm_area_struct *vma = vmas[j]; | |
7147 | ||
7148 | if (vma->vm_file && | |
7149 | !is_file_hugepages(vma->vm_file)) { | |
7150 | ret = -EOPNOTSUPP; | |
7151 | break; | |
7152 | } | |
7153 | } | |
7154 | } else { | |
7155 | ret = pret < 0 ? pret : -EFAULT; | |
7156 | } | |
7157 | up_read(¤t->mm->mmap_sem); | |
7158 | if (ret) { | |
7159 | /* | |
7160 | * if we did partial map, or found file backed vmas, | |
7161 | * release any pages we did get | |
7162 | */ | |
27c4d3a3 | 7163 | if (pret > 0) |
f1f6a7dd | 7164 | unpin_user_pages(pages, pret); |
edafccee JA |
7165 | if (ctx->account_mem) |
7166 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7167 | kvfree(imu->bvec); |
edafccee JA |
7168 | goto err; |
7169 | } | |
7170 | ||
7171 | off = ubuf & ~PAGE_MASK; | |
7172 | size = iov.iov_len; | |
7173 | for (j = 0; j < nr_pages; j++) { | |
7174 | size_t vec_len; | |
7175 | ||
7176 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7177 | imu->bvec[j].bv_page = pages[j]; | |
7178 | imu->bvec[j].bv_len = vec_len; | |
7179 | imu->bvec[j].bv_offset = off; | |
7180 | off = 0; | |
7181 | size -= vec_len; | |
7182 | } | |
7183 | /* store original address for later verification */ | |
7184 | imu->ubuf = ubuf; | |
7185 | imu->len = iov.iov_len; | |
7186 | imu->nr_bvecs = nr_pages; | |
7187 | ||
7188 | ctx->nr_user_bufs++; | |
7189 | } | |
d4ef6475 MR |
7190 | kvfree(pages); |
7191 | kvfree(vmas); | |
edafccee JA |
7192 | return 0; |
7193 | err: | |
d4ef6475 MR |
7194 | kvfree(pages); |
7195 | kvfree(vmas); | |
edafccee JA |
7196 | io_sqe_buffer_unregister(ctx); |
7197 | return ret; | |
7198 | } | |
7199 | ||
9b402849 JA |
7200 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7201 | { | |
7202 | __s32 __user *fds = arg; | |
7203 | int fd; | |
7204 | ||
7205 | if (ctx->cq_ev_fd) | |
7206 | return -EBUSY; | |
7207 | ||
7208 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7209 | return -EFAULT; | |
7210 | ||
7211 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7212 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7213 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7214 | ctx->cq_ev_fd = NULL; | |
7215 | return ret; | |
7216 | } | |
7217 | ||
7218 | return 0; | |
7219 | } | |
7220 | ||
7221 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7222 | { | |
7223 | if (ctx->cq_ev_fd) { | |
7224 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7225 | ctx->cq_ev_fd = NULL; | |
7226 | return 0; | |
7227 | } | |
7228 | ||
7229 | return -ENXIO; | |
7230 | } | |
7231 | ||
5a2e745d JA |
7232 | static int __io_destroy_buffers(int id, void *p, void *data) |
7233 | { | |
7234 | struct io_ring_ctx *ctx = data; | |
7235 | struct io_buffer *buf = p; | |
7236 | ||
067524e9 | 7237 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7238 | return 0; |
7239 | } | |
7240 | ||
7241 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7242 | { | |
7243 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7244 | idr_destroy(&ctx->io_buffer_idr); | |
7245 | } | |
7246 | ||
2b188cc1 JA |
7247 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7248 | { | |
6b06314c | 7249 | io_finish_async(ctx); |
2b188cc1 JA |
7250 | if (ctx->sqo_mm) |
7251 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7252 | |
7253 | io_iopoll_reap_events(ctx); | |
edafccee | 7254 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7255 | io_sqe_files_unregister(ctx); |
9b402849 | 7256 | io_eventfd_unregister(ctx); |
5a2e745d | 7257 | io_destroy_buffers(ctx); |
41726c9a | 7258 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7259 | |
2b188cc1 | 7260 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7261 | if (ctx->ring_sock) { |
7262 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7263 | sock_release(ctx->ring_sock); |
355e8d26 | 7264 | } |
2b188cc1 JA |
7265 | #endif |
7266 | ||
75b28aff | 7267 | io_mem_free(ctx->rings); |
2b188cc1 | 7268 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7269 | |
7270 | percpu_ref_exit(&ctx->refs); | |
7271 | if (ctx->account_mem) | |
7272 | io_unaccount_mem(ctx->user, | |
7273 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7274 | free_uid(ctx->user); | |
181e448d | 7275 | put_cred(ctx->creds); |
206aefde | 7276 | kfree(ctx->completions); |
78076bb6 | 7277 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7278 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7279 | kfree(ctx); |
7280 | } | |
7281 | ||
7282 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7283 | { | |
7284 | struct io_ring_ctx *ctx = file->private_data; | |
7285 | __poll_t mask = 0; | |
7286 | ||
7287 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7288 | /* |
7289 | * synchronizes with barrier from wq_has_sleeper call in | |
7290 | * io_commit_cqring | |
7291 | */ | |
2b188cc1 | 7292 | smp_rmb(); |
75b28aff HV |
7293 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7294 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7295 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7296 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7297 | mask |= EPOLLIN | EPOLLRDNORM; |
7298 | ||
7299 | return mask; | |
7300 | } | |
7301 | ||
7302 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7303 | { | |
7304 | struct io_ring_ctx *ctx = file->private_data; | |
7305 | ||
7306 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7307 | } | |
7308 | ||
071698e1 JA |
7309 | static int io_remove_personalities(int id, void *p, void *data) |
7310 | { | |
7311 | struct io_ring_ctx *ctx = data; | |
7312 | const struct cred *cred; | |
7313 | ||
7314 | cred = idr_remove(&ctx->personality_idr, id); | |
7315 | if (cred) | |
7316 | put_cred(cred); | |
7317 | return 0; | |
7318 | } | |
7319 | ||
85faa7b8 JA |
7320 | static void io_ring_exit_work(struct work_struct *work) |
7321 | { | |
7322 | struct io_ring_ctx *ctx; | |
7323 | ||
7324 | ctx = container_of(work, struct io_ring_ctx, exit_work); | |
7325 | if (ctx->rings) | |
7326 | io_cqring_overflow_flush(ctx, true); | |
7327 | ||
7328 | wait_for_completion(&ctx->completions[0]); | |
7329 | io_ring_ctx_free(ctx); | |
7330 | } | |
7331 | ||
2b188cc1 JA |
7332 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7333 | { | |
7334 | mutex_lock(&ctx->uring_lock); | |
7335 | percpu_ref_kill(&ctx->refs); | |
7336 | mutex_unlock(&ctx->uring_lock); | |
7337 | ||
df069d80 JA |
7338 | /* |
7339 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7340 | * work after we've killed the ctx ref above. This is important to do | |
7341 | * before we cancel existing commands, as the thread could otherwise | |
7342 | * be queueing new work post that. If that's work we need to cancel, | |
7343 | * it could cause shutdown to hang. | |
7344 | */ | |
7345 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
7346 | cpu_relax(); | |
7347 | ||
5262f567 | 7348 | io_kill_timeouts(ctx); |
221c5eb2 | 7349 | io_poll_remove_all(ctx); |
561fb04a JA |
7350 | |
7351 | if (ctx->io_wq) | |
7352 | io_wq_cancel_all(ctx->io_wq); | |
7353 | ||
def596e9 | 7354 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7355 | /* if we failed setting up the ctx, we might not have any rings */ |
7356 | if (ctx->rings) | |
7357 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7358 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
85faa7b8 JA |
7359 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
7360 | queue_work(system_wq, &ctx->exit_work); | |
2b188cc1 JA |
7361 | } |
7362 | ||
7363 | static int io_uring_release(struct inode *inode, struct file *file) | |
7364 | { | |
7365 | struct io_ring_ctx *ctx = file->private_data; | |
7366 | ||
7367 | file->private_data = NULL; | |
7368 | io_ring_ctx_wait_and_kill(ctx); | |
7369 | return 0; | |
7370 | } | |
7371 | ||
fcb323cc JA |
7372 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7373 | struct files_struct *files) | |
7374 | { | |
7375 | struct io_kiocb *req; | |
7376 | DEFINE_WAIT(wait); | |
7377 | ||
7378 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 7379 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
7380 | |
7381 | spin_lock_irq(&ctx->inflight_lock); | |
7382 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7383 | if (req->work.files != files) |
7384 | continue; | |
7385 | /* req is being completed, ignore */ | |
7386 | if (!refcount_inc_not_zero(&req->refs)) | |
7387 | continue; | |
7388 | cancel_req = req; | |
7389 | break; | |
fcb323cc | 7390 | } |
768134d4 | 7391 | if (cancel_req) |
fcb323cc | 7392 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7393 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7394 | spin_unlock_irq(&ctx->inflight_lock); |
7395 | ||
768134d4 JA |
7396 | /* We need to keep going until we don't find a matching req */ |
7397 | if (!cancel_req) | |
fcb323cc | 7398 | break; |
2f6d9b9d | 7399 | |
2ca10259 JA |
7400 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7401 | spin_lock_irq(&ctx->completion_lock); | |
7402 | list_del(&cancel_req->list); | |
7403 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7404 | if (list_empty(&ctx->cq_overflow_list)) { | |
7405 | clear_bit(0, &ctx->sq_check_overflow); | |
7406 | clear_bit(0, &ctx->cq_check_overflow); | |
7407 | } | |
7408 | spin_unlock_irq(&ctx->completion_lock); | |
7409 | ||
7410 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7411 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7412 | ||
7413 | /* | |
7414 | * Put inflight ref and overflow ref. If that's | |
7415 | * all we had, then we're done with this request. | |
7416 | */ | |
7417 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7418 | io_put_req(cancel_req); | |
7419 | continue; | |
7420 | } | |
7421 | } | |
7422 | ||
2f6d9b9d BL |
7423 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7424 | io_put_req(cancel_req); | |
fcb323cc JA |
7425 | schedule(); |
7426 | } | |
768134d4 | 7427 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7428 | } |
7429 | ||
7430 | static int io_uring_flush(struct file *file, void *data) | |
7431 | { | |
7432 | struct io_ring_ctx *ctx = file->private_data; | |
7433 | ||
7434 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7435 | |
7436 | /* | |
7437 | * If the task is going away, cancel work it may have pending | |
7438 | */ | |
7439 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7440 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7441 | ||
fcb323cc JA |
7442 | return 0; |
7443 | } | |
7444 | ||
6c5c240e RP |
7445 | static void *io_uring_validate_mmap_request(struct file *file, |
7446 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7447 | { |
2b188cc1 | 7448 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7449 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7450 | struct page *page; |
7451 | void *ptr; | |
7452 | ||
7453 | switch (offset) { | |
7454 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7455 | case IORING_OFF_CQ_RING: |
7456 | ptr = ctx->rings; | |
2b188cc1 JA |
7457 | break; |
7458 | case IORING_OFF_SQES: | |
7459 | ptr = ctx->sq_sqes; | |
7460 | break; | |
2b188cc1 | 7461 | default: |
6c5c240e | 7462 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7463 | } |
7464 | ||
7465 | page = virt_to_head_page(ptr); | |
a50b854e | 7466 | if (sz > page_size(page)) |
6c5c240e RP |
7467 | return ERR_PTR(-EINVAL); |
7468 | ||
7469 | return ptr; | |
7470 | } | |
7471 | ||
7472 | #ifdef CONFIG_MMU | |
7473 | ||
7474 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7475 | { | |
7476 | size_t sz = vma->vm_end - vma->vm_start; | |
7477 | unsigned long pfn; | |
7478 | void *ptr; | |
7479 | ||
7480 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7481 | if (IS_ERR(ptr)) | |
7482 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7483 | |
7484 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7485 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7486 | } | |
7487 | ||
6c5c240e RP |
7488 | #else /* !CONFIG_MMU */ |
7489 | ||
7490 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7491 | { | |
7492 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7493 | } | |
7494 | ||
7495 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7496 | { | |
7497 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7498 | } | |
7499 | ||
7500 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7501 | unsigned long addr, unsigned long len, | |
7502 | unsigned long pgoff, unsigned long flags) | |
7503 | { | |
7504 | void *ptr; | |
7505 | ||
7506 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7507 | if (IS_ERR(ptr)) | |
7508 | return PTR_ERR(ptr); | |
7509 | ||
7510 | return (unsigned long) ptr; | |
7511 | } | |
7512 | ||
7513 | #endif /* !CONFIG_MMU */ | |
7514 | ||
2b188cc1 JA |
7515 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7516 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7517 | size_t, sigsz) | |
7518 | { | |
7519 | struct io_ring_ctx *ctx; | |
7520 | long ret = -EBADF; | |
7521 | int submitted = 0; | |
7522 | struct fd f; | |
7523 | ||
b41e9852 JA |
7524 | if (current->task_works) |
7525 | task_work_run(); | |
7526 | ||
6c271ce2 | 7527 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7528 | return -EINVAL; |
7529 | ||
7530 | f = fdget(fd); | |
7531 | if (!f.file) | |
7532 | return -EBADF; | |
7533 | ||
7534 | ret = -EOPNOTSUPP; | |
7535 | if (f.file->f_op != &io_uring_fops) | |
7536 | goto out_fput; | |
7537 | ||
7538 | ret = -ENXIO; | |
7539 | ctx = f.file->private_data; | |
7540 | if (!percpu_ref_tryget(&ctx->refs)) | |
7541 | goto out_fput; | |
7542 | ||
6c271ce2 JA |
7543 | /* |
7544 | * For SQ polling, the thread will do all submissions and completions. | |
7545 | * Just return the requested submit count, and wake the thread if | |
7546 | * we were asked to. | |
7547 | */ | |
b2a9eada | 7548 | ret = 0; |
6c271ce2 | 7549 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7550 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7551 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7552 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7553 | wake_up(&ctx->sqo_wait); | |
7554 | submitted = to_submit; | |
b2a9eada | 7555 | } else if (to_submit) { |
2b188cc1 | 7556 | mutex_lock(&ctx->uring_lock); |
bf9c2f1c | 7557 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, false); |
2b188cc1 | 7558 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7559 | |
7560 | if (submitted != to_submit) | |
7561 | goto out; | |
2b188cc1 JA |
7562 | } |
7563 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7564 | unsigned nr_events = 0; |
7565 | ||
2b188cc1 JA |
7566 | min_complete = min(min_complete, ctx->cq_entries); |
7567 | ||
32b2244a XW |
7568 | /* |
7569 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7570 | * space applications don't need to do io completion events | |
7571 | * polling again, they can rely on io_sq_thread to do polling | |
7572 | * work, which can reduce cpu usage and uring_lock contention. | |
7573 | */ | |
7574 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7575 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7576 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7577 | } else { |
7578 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7579 | } | |
2b188cc1 JA |
7580 | } |
7581 | ||
7c504e65 | 7582 | out: |
6805b32e | 7583 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7584 | out_fput: |
7585 | fdput(f); | |
7586 | return submitted ? submitted : ret; | |
7587 | } | |
7588 | ||
bebdb65e | 7589 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7590 | static int io_uring_show_cred(int id, void *p, void *data) |
7591 | { | |
7592 | const struct cred *cred = p; | |
7593 | struct seq_file *m = data; | |
7594 | struct user_namespace *uns = seq_user_ns(m); | |
7595 | struct group_info *gi; | |
7596 | kernel_cap_t cap; | |
7597 | unsigned __capi; | |
7598 | int g; | |
7599 | ||
7600 | seq_printf(m, "%5d\n", id); | |
7601 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7602 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7603 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7604 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7605 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7606 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7607 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7608 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7609 | seq_puts(m, "\n\tGroups:\t"); | |
7610 | gi = cred->group_info; | |
7611 | for (g = 0; g < gi->ngroups; g++) { | |
7612 | seq_put_decimal_ull(m, g ? " " : "", | |
7613 | from_kgid_munged(uns, gi->gid[g])); | |
7614 | } | |
7615 | seq_puts(m, "\n\tCapEff:\t"); | |
7616 | cap = cred->cap_effective; | |
7617 | CAP_FOR_EACH_U32(__capi) | |
7618 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7619 | seq_putc(m, '\n'); | |
7620 | return 0; | |
7621 | } | |
7622 | ||
7623 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7624 | { | |
7625 | int i; | |
7626 | ||
7627 | mutex_lock(&ctx->uring_lock); | |
7628 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7629 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7630 | struct fixed_file_table *table; | |
7631 | struct file *f; | |
7632 | ||
7633 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7634 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7635 | if (f) | |
7636 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7637 | else | |
7638 | seq_printf(m, "%5u: <none>\n", i); | |
7639 | } | |
7640 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7641 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7642 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7643 | ||
7644 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7645 | (unsigned int) buf->len); | |
7646 | } | |
7647 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7648 | seq_printf(m, "Personalities:\n"); | |
7649 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7650 | } | |
d7718a9d JA |
7651 | seq_printf(m, "PollList:\n"); |
7652 | spin_lock_irq(&ctx->completion_lock); | |
7653 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7654 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7655 | struct io_kiocb *req; | |
7656 | ||
7657 | hlist_for_each_entry(req, list, hash_node) | |
7658 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7659 | req->task->task_works != NULL); | |
7660 | } | |
7661 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7662 | mutex_unlock(&ctx->uring_lock); |
7663 | } | |
7664 | ||
7665 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7666 | { | |
7667 | struct io_ring_ctx *ctx = f->private_data; | |
7668 | ||
7669 | if (percpu_ref_tryget(&ctx->refs)) { | |
7670 | __io_uring_show_fdinfo(ctx, m); | |
7671 | percpu_ref_put(&ctx->refs); | |
7672 | } | |
7673 | } | |
bebdb65e | 7674 | #endif |
87ce955b | 7675 | |
2b188cc1 JA |
7676 | static const struct file_operations io_uring_fops = { |
7677 | .release = io_uring_release, | |
fcb323cc | 7678 | .flush = io_uring_flush, |
2b188cc1 | 7679 | .mmap = io_uring_mmap, |
6c5c240e RP |
7680 | #ifndef CONFIG_MMU |
7681 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7682 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7683 | #endif | |
2b188cc1 JA |
7684 | .poll = io_uring_poll, |
7685 | .fasync = io_uring_fasync, | |
bebdb65e | 7686 | #ifdef CONFIG_PROC_FS |
87ce955b | 7687 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7688 | #endif |
2b188cc1 JA |
7689 | }; |
7690 | ||
7691 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7692 | struct io_uring_params *p) | |
7693 | { | |
75b28aff HV |
7694 | struct io_rings *rings; |
7695 | size_t size, sq_array_offset; | |
2b188cc1 | 7696 | |
75b28aff HV |
7697 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7698 | if (size == SIZE_MAX) | |
7699 | return -EOVERFLOW; | |
7700 | ||
7701 | rings = io_mem_alloc(size); | |
7702 | if (!rings) | |
2b188cc1 JA |
7703 | return -ENOMEM; |
7704 | ||
75b28aff HV |
7705 | ctx->rings = rings; |
7706 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7707 | rings->sq_ring_mask = p->sq_entries - 1; | |
7708 | rings->cq_ring_mask = p->cq_entries - 1; | |
7709 | rings->sq_ring_entries = p->sq_entries; | |
7710 | rings->cq_ring_entries = p->cq_entries; | |
7711 | ctx->sq_mask = rings->sq_ring_mask; | |
7712 | ctx->cq_mask = rings->cq_ring_mask; | |
7713 | ctx->sq_entries = rings->sq_ring_entries; | |
7714 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7715 | |
7716 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7717 | if (size == SIZE_MAX) { |
7718 | io_mem_free(ctx->rings); | |
7719 | ctx->rings = NULL; | |
2b188cc1 | 7720 | return -EOVERFLOW; |
eb065d30 | 7721 | } |
2b188cc1 JA |
7722 | |
7723 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7724 | if (!ctx->sq_sqes) { |
7725 | io_mem_free(ctx->rings); | |
7726 | ctx->rings = NULL; | |
2b188cc1 | 7727 | return -ENOMEM; |
eb065d30 | 7728 | } |
2b188cc1 | 7729 | |
2b188cc1 JA |
7730 | return 0; |
7731 | } | |
7732 | ||
7733 | /* | |
7734 | * Allocate an anonymous fd, this is what constitutes the application | |
7735 | * visible backing of an io_uring instance. The application mmaps this | |
7736 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7737 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7738 | */ | |
7739 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7740 | { | |
7741 | struct file *file; | |
7742 | int ret; | |
7743 | ||
7744 | #if defined(CONFIG_UNIX) | |
7745 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7746 | &ctx->ring_sock); | |
7747 | if (ret) | |
7748 | return ret; | |
7749 | #endif | |
7750 | ||
7751 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7752 | if (ret < 0) | |
7753 | goto err; | |
7754 | ||
7755 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7756 | O_RDWR | O_CLOEXEC); | |
7757 | if (IS_ERR(file)) { | |
7758 | put_unused_fd(ret); | |
7759 | ret = PTR_ERR(file); | |
7760 | goto err; | |
7761 | } | |
7762 | ||
7763 | #if defined(CONFIG_UNIX) | |
7764 | ctx->ring_sock->file = file; | |
7765 | #endif | |
7766 | fd_install(ret, file); | |
7767 | return ret; | |
7768 | err: | |
7769 | #if defined(CONFIG_UNIX) | |
7770 | sock_release(ctx->ring_sock); | |
7771 | ctx->ring_sock = NULL; | |
7772 | #endif | |
7773 | return ret; | |
7774 | } | |
7775 | ||
7776 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7777 | { | |
7778 | struct user_struct *user = NULL; | |
7779 | struct io_ring_ctx *ctx; | |
7780 | bool account_mem; | |
7781 | int ret; | |
7782 | ||
8110c1a6 | 7783 | if (!entries) |
2b188cc1 | 7784 | return -EINVAL; |
8110c1a6 JA |
7785 | if (entries > IORING_MAX_ENTRIES) { |
7786 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7787 | return -EINVAL; | |
7788 | entries = IORING_MAX_ENTRIES; | |
7789 | } | |
2b188cc1 JA |
7790 | |
7791 | /* | |
7792 | * Use twice as many entries for the CQ ring. It's possible for the | |
7793 | * application to drive a higher depth than the size of the SQ ring, | |
7794 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7795 | * some flexibility in overcommitting a bit. If the application has |
7796 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7797 | * of CQ ring entries manually. | |
2b188cc1 JA |
7798 | */ |
7799 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7800 | if (p->flags & IORING_SETUP_CQSIZE) { |
7801 | /* | |
7802 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7803 | * to a power-of-two, if it isn't already. We do NOT impose | |
7804 | * any cq vs sq ring sizing. | |
7805 | */ | |
8110c1a6 | 7806 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7807 | return -EINVAL; |
8110c1a6 JA |
7808 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7809 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7810 | return -EINVAL; | |
7811 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7812 | } | |
33a107f0 JA |
7813 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7814 | } else { | |
7815 | p->cq_entries = 2 * p->sq_entries; | |
7816 | } | |
2b188cc1 JA |
7817 | |
7818 | user = get_uid(current_user()); | |
7819 | account_mem = !capable(CAP_IPC_LOCK); | |
7820 | ||
7821 | if (account_mem) { | |
7822 | ret = io_account_mem(user, | |
7823 | ring_pages(p->sq_entries, p->cq_entries)); | |
7824 | if (ret) { | |
7825 | free_uid(user); | |
7826 | return ret; | |
7827 | } | |
7828 | } | |
7829 | ||
7830 | ctx = io_ring_ctx_alloc(p); | |
7831 | if (!ctx) { | |
7832 | if (account_mem) | |
7833 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7834 | p->cq_entries)); | |
7835 | free_uid(user); | |
7836 | return -ENOMEM; | |
7837 | } | |
7838 | ctx->compat = in_compat_syscall(); | |
7839 | ctx->account_mem = account_mem; | |
7840 | ctx->user = user; | |
0b8c0ec7 | 7841 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7842 | |
7843 | ret = io_allocate_scq_urings(ctx, p); | |
7844 | if (ret) | |
7845 | goto err; | |
7846 | ||
6c271ce2 | 7847 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7848 | if (ret) |
7849 | goto err; | |
7850 | ||
2b188cc1 | 7851 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7852 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7853 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7854 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7855 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7856 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7857 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7858 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7859 | |
7860 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7861 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7862 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7863 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7864 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7865 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7866 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7867 | |
044c1ab3 JA |
7868 | /* |
7869 | * Install ring fd as the very last thing, so we don't risk someone | |
7870 | * having closed it before we finish setup | |
7871 | */ | |
7872 | ret = io_uring_get_fd(ctx); | |
7873 | if (ret < 0) | |
7874 | goto err; | |
7875 | ||
da8c9690 | 7876 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7877 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7878 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7879 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7880 | return ret; |
7881 | err: | |
7882 | io_ring_ctx_wait_and_kill(ctx); | |
7883 | return ret; | |
7884 | } | |
7885 | ||
7886 | /* | |
7887 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7888 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7889 | * params structure passed in. | |
7890 | */ | |
7891 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7892 | { | |
7893 | struct io_uring_params p; | |
7894 | long ret; | |
7895 | int i; | |
7896 | ||
7897 | if (copy_from_user(&p, params, sizeof(p))) | |
7898 | return -EFAULT; | |
7899 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7900 | if (p.resv[i]) | |
7901 | return -EINVAL; | |
7902 | } | |
7903 | ||
6c271ce2 | 7904 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7905 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7906 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7907 | return -EINVAL; |
7908 | ||
7909 | ret = io_uring_create(entries, &p); | |
7910 | if (ret < 0) | |
7911 | return ret; | |
7912 | ||
7913 | if (copy_to_user(params, &p, sizeof(p))) | |
7914 | return -EFAULT; | |
7915 | ||
7916 | return ret; | |
7917 | } | |
7918 | ||
7919 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7920 | struct io_uring_params __user *, params) | |
7921 | { | |
7922 | return io_uring_setup(entries, params); | |
7923 | } | |
7924 | ||
66f4af93 JA |
7925 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7926 | { | |
7927 | struct io_uring_probe *p; | |
7928 | size_t size; | |
7929 | int i, ret; | |
7930 | ||
7931 | size = struct_size(p, ops, nr_args); | |
7932 | if (size == SIZE_MAX) | |
7933 | return -EOVERFLOW; | |
7934 | p = kzalloc(size, GFP_KERNEL); | |
7935 | if (!p) | |
7936 | return -ENOMEM; | |
7937 | ||
7938 | ret = -EFAULT; | |
7939 | if (copy_from_user(p, arg, size)) | |
7940 | goto out; | |
7941 | ret = -EINVAL; | |
7942 | if (memchr_inv(p, 0, size)) | |
7943 | goto out; | |
7944 | ||
7945 | p->last_op = IORING_OP_LAST - 1; | |
7946 | if (nr_args > IORING_OP_LAST) | |
7947 | nr_args = IORING_OP_LAST; | |
7948 | ||
7949 | for (i = 0; i < nr_args; i++) { | |
7950 | p->ops[i].op = i; | |
7951 | if (!io_op_defs[i].not_supported) | |
7952 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7953 | } | |
7954 | p->ops_len = i; | |
7955 | ||
7956 | ret = 0; | |
7957 | if (copy_to_user(arg, p, size)) | |
7958 | ret = -EFAULT; | |
7959 | out: | |
7960 | kfree(p); | |
7961 | return ret; | |
7962 | } | |
7963 | ||
071698e1 JA |
7964 | static int io_register_personality(struct io_ring_ctx *ctx) |
7965 | { | |
7966 | const struct cred *creds = get_current_cred(); | |
7967 | int id; | |
7968 | ||
7969 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7970 | USHRT_MAX, GFP_KERNEL); | |
7971 | if (id < 0) | |
7972 | put_cred(creds); | |
7973 | return id; | |
7974 | } | |
7975 | ||
7976 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7977 | { | |
7978 | const struct cred *old_creds; | |
7979 | ||
7980 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7981 | if (old_creds) { | |
7982 | put_cred(old_creds); | |
7983 | return 0; | |
7984 | } | |
7985 | ||
7986 | return -EINVAL; | |
7987 | } | |
7988 | ||
7989 | static bool io_register_op_must_quiesce(int op) | |
7990 | { | |
7991 | switch (op) { | |
7992 | case IORING_UNREGISTER_FILES: | |
7993 | case IORING_REGISTER_FILES_UPDATE: | |
7994 | case IORING_REGISTER_PROBE: | |
7995 | case IORING_REGISTER_PERSONALITY: | |
7996 | case IORING_UNREGISTER_PERSONALITY: | |
7997 | return false; | |
7998 | default: | |
7999 | return true; | |
8000 | } | |
8001 | } | |
8002 | ||
edafccee JA |
8003 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
8004 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
8005 | __releases(ctx->uring_lock) |
8006 | __acquires(ctx->uring_lock) | |
edafccee JA |
8007 | { |
8008 | int ret; | |
8009 | ||
35fa71a0 JA |
8010 | /* |
8011 | * We're inside the ring mutex, if the ref is already dying, then | |
8012 | * someone else killed the ctx or is already going through | |
8013 | * io_uring_register(). | |
8014 | */ | |
8015 | if (percpu_ref_is_dying(&ctx->refs)) | |
8016 | return -ENXIO; | |
8017 | ||
071698e1 | 8018 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8019 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 8020 | |
05f3fb3c JA |
8021 | /* |
8022 | * Drop uring mutex before waiting for references to exit. If | |
8023 | * another thread is currently inside io_uring_enter() it might | |
8024 | * need to grab the uring_lock to make progress. If we hold it | |
8025 | * here across the drain wait, then we can deadlock. It's safe | |
8026 | * to drop the mutex here, since no new references will come in | |
8027 | * after we've killed the percpu ref. | |
8028 | */ | |
8029 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 8030 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 8031 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
8032 | if (ret) { |
8033 | percpu_ref_resurrect(&ctx->refs); | |
8034 | ret = -EINTR; | |
8035 | goto out; | |
8036 | } | |
05f3fb3c | 8037 | } |
edafccee JA |
8038 | |
8039 | switch (opcode) { | |
8040 | case IORING_REGISTER_BUFFERS: | |
8041 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
8042 | break; | |
8043 | case IORING_UNREGISTER_BUFFERS: | |
8044 | ret = -EINVAL; | |
8045 | if (arg || nr_args) | |
8046 | break; | |
8047 | ret = io_sqe_buffer_unregister(ctx); | |
8048 | break; | |
6b06314c JA |
8049 | case IORING_REGISTER_FILES: |
8050 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
8051 | break; | |
8052 | case IORING_UNREGISTER_FILES: | |
8053 | ret = -EINVAL; | |
8054 | if (arg || nr_args) | |
8055 | break; | |
8056 | ret = io_sqe_files_unregister(ctx); | |
8057 | break; | |
c3a31e60 JA |
8058 | case IORING_REGISTER_FILES_UPDATE: |
8059 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
8060 | break; | |
9b402849 | 8061 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 8062 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
8063 | ret = -EINVAL; |
8064 | if (nr_args != 1) | |
8065 | break; | |
8066 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
8067 | if (ret) |
8068 | break; | |
8069 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
8070 | ctx->eventfd_async = 1; | |
8071 | else | |
8072 | ctx->eventfd_async = 0; | |
9b402849 JA |
8073 | break; |
8074 | case IORING_UNREGISTER_EVENTFD: | |
8075 | ret = -EINVAL; | |
8076 | if (arg || nr_args) | |
8077 | break; | |
8078 | ret = io_eventfd_unregister(ctx); | |
8079 | break; | |
66f4af93 JA |
8080 | case IORING_REGISTER_PROBE: |
8081 | ret = -EINVAL; | |
8082 | if (!arg || nr_args > 256) | |
8083 | break; | |
8084 | ret = io_probe(ctx, arg, nr_args); | |
8085 | break; | |
071698e1 JA |
8086 | case IORING_REGISTER_PERSONALITY: |
8087 | ret = -EINVAL; | |
8088 | if (arg || nr_args) | |
8089 | break; | |
8090 | ret = io_register_personality(ctx); | |
8091 | break; | |
8092 | case IORING_UNREGISTER_PERSONALITY: | |
8093 | ret = -EINVAL; | |
8094 | if (arg) | |
8095 | break; | |
8096 | ret = io_unregister_personality(ctx, nr_args); | |
8097 | break; | |
edafccee JA |
8098 | default: |
8099 | ret = -EINVAL; | |
8100 | break; | |
8101 | } | |
8102 | ||
071698e1 | 8103 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 8104 | /* bring the ctx back to life */ |
05f3fb3c | 8105 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
8106 | out: |
8107 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 8108 | } |
edafccee JA |
8109 | return ret; |
8110 | } | |
8111 | ||
8112 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
8113 | void __user *, arg, unsigned int, nr_args) | |
8114 | { | |
8115 | struct io_ring_ctx *ctx; | |
8116 | long ret = -EBADF; | |
8117 | struct fd f; | |
8118 | ||
8119 | f = fdget(fd); | |
8120 | if (!f.file) | |
8121 | return -EBADF; | |
8122 | ||
8123 | ret = -EOPNOTSUPP; | |
8124 | if (f.file->f_op != &io_uring_fops) | |
8125 | goto out_fput; | |
8126 | ||
8127 | ctx = f.file->private_data; | |
8128 | ||
8129 | mutex_lock(&ctx->uring_lock); | |
8130 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8131 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8132 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8133 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8134 | out_fput: |
8135 | fdput(f); | |
8136 | return ret; | |
8137 | } | |
8138 | ||
2b188cc1 JA |
8139 | static int __init io_uring_init(void) |
8140 | { | |
d7f62e82 SM |
8141 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8142 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8143 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8144 | } while (0) | |
8145 | ||
8146 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8147 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8148 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8149 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8150 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8151 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8152 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8153 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8154 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8155 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8156 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8157 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8158 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8159 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8160 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8161 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8162 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8163 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8164 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8165 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8166 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8167 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8168 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8169 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8170 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8171 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8172 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8173 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8174 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8175 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8176 | |
d3656344 | 8177 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8178 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8179 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8180 | return 0; | |
8181 | }; | |
8182 | __initcall(io_uring_init); |