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 | ||
05f3fb3c JA |
189 | struct fixed_file_data { |
190 | struct fixed_file_table *table; | |
191 | struct io_ring_ctx *ctx; | |
192 | ||
193 | struct percpu_ref refs; | |
194 | struct llist_head put_llist; | |
05f3fb3c JA |
195 | struct work_struct ref_work; |
196 | struct completion done; | |
197 | }; | |
198 | ||
5a2e745d JA |
199 | struct io_buffer { |
200 | struct list_head list; | |
201 | __u64 addr; | |
202 | __s32 len; | |
203 | __u16 bid; | |
204 | }; | |
205 | ||
2b188cc1 JA |
206 | struct io_ring_ctx { |
207 | struct { | |
208 | struct percpu_ref refs; | |
209 | } ____cacheline_aligned_in_smp; | |
210 | ||
211 | struct { | |
212 | unsigned int flags; | |
e1d85334 RD |
213 | unsigned int compat: 1; |
214 | unsigned int account_mem: 1; | |
215 | unsigned int cq_overflow_flushed: 1; | |
216 | unsigned int drain_next: 1; | |
217 | unsigned int eventfd_async: 1; | |
2b188cc1 | 218 | |
75b28aff HV |
219 | /* |
220 | * Ring buffer of indices into array of io_uring_sqe, which is | |
221 | * mmapped by the application using the IORING_OFF_SQES offset. | |
222 | * | |
223 | * This indirection could e.g. be used to assign fixed | |
224 | * io_uring_sqe entries to operations and only submit them to | |
225 | * the queue when needed. | |
226 | * | |
227 | * The kernel modifies neither the indices array nor the entries | |
228 | * array. | |
229 | */ | |
230 | u32 *sq_array; | |
2b188cc1 JA |
231 | unsigned cached_sq_head; |
232 | unsigned sq_entries; | |
233 | unsigned sq_mask; | |
6c271ce2 | 234 | unsigned sq_thread_idle; |
498ccd9e | 235 | unsigned cached_sq_dropped; |
206aefde | 236 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 237 | unsigned long sq_check_overflow; |
de0617e4 JA |
238 | |
239 | struct list_head defer_list; | |
5262f567 | 240 | struct list_head timeout_list; |
1d7bb1d5 | 241 | struct list_head cq_overflow_list; |
fcb323cc JA |
242 | |
243 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 244 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
245 | } ____cacheline_aligned_in_smp; |
246 | ||
206aefde JA |
247 | struct io_rings *rings; |
248 | ||
2b188cc1 | 249 | /* IO offload */ |
561fb04a | 250 | struct io_wq *io_wq; |
6c271ce2 | 251 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 252 | struct mm_struct *sqo_mm; |
6c271ce2 | 253 | wait_queue_head_t sqo_wait; |
75b28aff | 254 | |
6b06314c JA |
255 | /* |
256 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
257 | * readers must ensure that ->refs is alive as long as the file* is | |
258 | * used. Only updated through io_uring_register(2). | |
259 | */ | |
05f3fb3c | 260 | struct fixed_file_data *file_data; |
6b06314c | 261 | unsigned nr_user_files; |
b14cca0c PB |
262 | int ring_fd; |
263 | struct file *ring_file; | |
6b06314c | 264 | |
edafccee JA |
265 | /* if used, fixed mapped user buffers */ |
266 | unsigned nr_user_bufs; | |
267 | struct io_mapped_ubuf *user_bufs; | |
268 | ||
2b188cc1 JA |
269 | struct user_struct *user; |
270 | ||
0b8c0ec7 | 271 | const struct cred *creds; |
181e448d | 272 | |
206aefde JA |
273 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
274 | struct completion *completions; | |
275 | ||
0ddf92e8 JA |
276 | /* if all else fails... */ |
277 | struct io_kiocb *fallback_req; | |
278 | ||
206aefde JA |
279 | #if defined(CONFIG_UNIX) |
280 | struct socket *ring_sock; | |
281 | #endif | |
282 | ||
5a2e745d JA |
283 | struct idr io_buffer_idr; |
284 | ||
071698e1 JA |
285 | struct idr personality_idr; |
286 | ||
206aefde JA |
287 | struct { |
288 | unsigned cached_cq_tail; | |
289 | unsigned cq_entries; | |
290 | unsigned cq_mask; | |
291 | atomic_t cq_timeouts; | |
ad3eb2c8 | 292 | unsigned long cq_check_overflow; |
206aefde JA |
293 | struct wait_queue_head cq_wait; |
294 | struct fasync_struct *cq_fasync; | |
295 | struct eventfd_ctx *cq_ev_fd; | |
296 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
297 | |
298 | struct { | |
299 | struct mutex uring_lock; | |
300 | wait_queue_head_t wait; | |
301 | } ____cacheline_aligned_in_smp; | |
302 | ||
303 | struct { | |
304 | spinlock_t completion_lock; | |
e94f141b | 305 | |
def596e9 JA |
306 | /* |
307 | * ->poll_list is protected by the ctx->uring_lock for | |
308 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
309 | * For SQPOLL, only the single threaded io_sq_thread() will | |
310 | * manipulate the list, hence no extra locking is needed there. | |
311 | */ | |
312 | struct list_head poll_list; | |
78076bb6 JA |
313 | struct hlist_head *cancel_hash; |
314 | unsigned cancel_hash_bits; | |
e94f141b | 315 | bool poll_multi_file; |
31b51510 | 316 | |
fcb323cc JA |
317 | spinlock_t inflight_lock; |
318 | struct list_head inflight_list; | |
2b188cc1 | 319 | } ____cacheline_aligned_in_smp; |
2b188cc1 JA |
320 | }; |
321 | ||
09bb8394 JA |
322 | /* |
323 | * First field must be the file pointer in all the | |
324 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
325 | */ | |
221c5eb2 JA |
326 | struct io_poll_iocb { |
327 | struct file *file; | |
0969e783 JA |
328 | union { |
329 | struct wait_queue_head *head; | |
330 | u64 addr; | |
331 | }; | |
221c5eb2 | 332 | __poll_t events; |
8c838788 | 333 | bool done; |
221c5eb2 | 334 | bool canceled; |
392edb45 | 335 | struct wait_queue_entry wait; |
221c5eb2 JA |
336 | }; |
337 | ||
b5dba59e JA |
338 | struct io_close { |
339 | struct file *file; | |
340 | struct file *put_file; | |
341 | int fd; | |
342 | }; | |
343 | ||
ad8a48ac JA |
344 | struct io_timeout_data { |
345 | struct io_kiocb *req; | |
346 | struct hrtimer timer; | |
347 | struct timespec64 ts; | |
348 | enum hrtimer_mode mode; | |
cc42e0ac | 349 | u32 seq_offset; |
ad8a48ac JA |
350 | }; |
351 | ||
8ed8d3c3 JA |
352 | struct io_accept { |
353 | struct file *file; | |
354 | struct sockaddr __user *addr; | |
355 | int __user *addr_len; | |
356 | int flags; | |
09952e3e | 357 | unsigned long nofile; |
8ed8d3c3 JA |
358 | }; |
359 | ||
360 | struct io_sync { | |
361 | struct file *file; | |
362 | loff_t len; | |
363 | loff_t off; | |
364 | int flags; | |
d63d1b5e | 365 | int mode; |
8ed8d3c3 JA |
366 | }; |
367 | ||
fbf23849 JA |
368 | struct io_cancel { |
369 | struct file *file; | |
370 | u64 addr; | |
371 | }; | |
372 | ||
b29472ee JA |
373 | struct io_timeout { |
374 | struct file *file; | |
375 | u64 addr; | |
376 | int flags; | |
26a61679 | 377 | unsigned count; |
b29472ee JA |
378 | }; |
379 | ||
9adbd45d JA |
380 | struct io_rw { |
381 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
382 | struct kiocb kiocb; | |
383 | u64 addr; | |
384 | u64 len; | |
385 | }; | |
386 | ||
3fbb51c1 JA |
387 | struct io_connect { |
388 | struct file *file; | |
389 | struct sockaddr __user *addr; | |
390 | int addr_len; | |
391 | }; | |
392 | ||
e47293fd JA |
393 | struct io_sr_msg { |
394 | struct file *file; | |
fddaface JA |
395 | union { |
396 | struct user_msghdr __user *msg; | |
397 | void __user *buf; | |
398 | }; | |
e47293fd | 399 | int msg_flags; |
bcda7baa | 400 | int bgid; |
fddaface | 401 | size_t len; |
bcda7baa | 402 | struct io_buffer *kbuf; |
e47293fd JA |
403 | }; |
404 | ||
15b71abe JA |
405 | struct io_open { |
406 | struct file *file; | |
407 | int dfd; | |
eddc7ef5 | 408 | union { |
eddc7ef5 JA |
409 | unsigned mask; |
410 | }; | |
15b71abe | 411 | struct filename *filename; |
eddc7ef5 | 412 | struct statx __user *buffer; |
c12cedf2 | 413 | struct open_how how; |
4022e7af | 414 | unsigned long nofile; |
15b71abe JA |
415 | }; |
416 | ||
05f3fb3c JA |
417 | struct io_files_update { |
418 | struct file *file; | |
419 | u64 arg; | |
420 | u32 nr_args; | |
421 | u32 offset; | |
422 | }; | |
423 | ||
4840e418 JA |
424 | struct io_fadvise { |
425 | struct file *file; | |
426 | u64 offset; | |
427 | u32 len; | |
428 | u32 advice; | |
429 | }; | |
430 | ||
c1ca757b JA |
431 | struct io_madvise { |
432 | struct file *file; | |
433 | u64 addr; | |
434 | u32 len; | |
435 | u32 advice; | |
436 | }; | |
437 | ||
3e4827b0 JA |
438 | struct io_epoll { |
439 | struct file *file; | |
440 | int epfd; | |
441 | int op; | |
442 | int fd; | |
443 | struct epoll_event event; | |
e47293fd JA |
444 | }; |
445 | ||
7d67af2c PB |
446 | struct io_splice { |
447 | struct file *file_out; | |
448 | struct file *file_in; | |
449 | loff_t off_out; | |
450 | loff_t off_in; | |
451 | u64 len; | |
452 | unsigned int flags; | |
453 | }; | |
454 | ||
ddf0322d JA |
455 | struct io_provide_buf { |
456 | struct file *file; | |
457 | __u64 addr; | |
458 | __s32 len; | |
459 | __u32 bgid; | |
460 | __u16 nbufs; | |
461 | __u16 bid; | |
462 | }; | |
463 | ||
f499a021 JA |
464 | struct io_async_connect { |
465 | struct sockaddr_storage address; | |
466 | }; | |
467 | ||
03b1230c JA |
468 | struct io_async_msghdr { |
469 | struct iovec fast_iov[UIO_FASTIOV]; | |
470 | struct iovec *iov; | |
471 | struct sockaddr __user *uaddr; | |
472 | struct msghdr msg; | |
b537916c | 473 | struct sockaddr_storage addr; |
03b1230c JA |
474 | }; |
475 | ||
f67676d1 JA |
476 | struct io_async_rw { |
477 | struct iovec fast_iov[UIO_FASTIOV]; | |
478 | struct iovec *iov; | |
479 | ssize_t nr_segs; | |
480 | ssize_t size; | |
481 | }; | |
482 | ||
1a6b74fc | 483 | struct io_async_ctx { |
f67676d1 JA |
484 | union { |
485 | struct io_async_rw rw; | |
03b1230c | 486 | struct io_async_msghdr msg; |
f499a021 | 487 | struct io_async_connect connect; |
2d28390a | 488 | struct io_timeout_data timeout; |
f67676d1 | 489 | }; |
1a6b74fc JA |
490 | }; |
491 | ||
6b47ee6e PB |
492 | enum { |
493 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
494 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
495 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
496 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
497 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 498 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e PB |
499 | |
500 | REQ_F_LINK_NEXT_BIT, | |
501 | REQ_F_FAIL_LINK_BIT, | |
502 | REQ_F_INFLIGHT_BIT, | |
503 | REQ_F_CUR_POS_BIT, | |
504 | REQ_F_NOWAIT_BIT, | |
505 | REQ_F_IOPOLL_COMPLETED_BIT, | |
506 | REQ_F_LINK_TIMEOUT_BIT, | |
507 | REQ_F_TIMEOUT_BIT, | |
508 | REQ_F_ISREG_BIT, | |
509 | REQ_F_MUST_PUNT_BIT, | |
510 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
511 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 512 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 513 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 514 | REQ_F_POLLED_BIT, |
bcda7baa | 515 | REQ_F_BUFFER_SELECTED_BIT, |
84557871 JA |
516 | |
517 | /* not a real bit, just to check we're not overflowing the space */ | |
518 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
519 | }; |
520 | ||
521 | enum { | |
522 | /* ctx owns file */ | |
523 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
524 | /* drain existing IO first */ | |
525 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
526 | /* linked sqes */ | |
527 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
528 | /* doesn't sever on completion < 0 */ | |
529 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
530 | /* IOSQE_ASYNC */ | |
531 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
532 | /* IOSQE_BUFFER_SELECT */ |
533 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e PB |
534 | |
535 | /* already grabbed next link */ | |
536 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
537 | /* fail rest of links */ | |
538 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
539 | /* on inflight list */ | |
540 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
541 | /* read/write uses file position */ | |
542 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
543 | /* must not punt to workers */ | |
544 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
545 | /* polled IO has completed */ | |
546 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
547 | /* has linked timeout */ | |
548 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
549 | /* timeout request */ | |
550 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
551 | /* regular file */ | |
552 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
553 | /* must be punted even for NONBLOCK */ | |
554 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
555 | /* no timeout sequence */ | |
556 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
557 | /* completion under lock */ | |
558 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
559 | /* needs cleanup */ |
560 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
561 | /* in overflow list */ |
562 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
563 | /* already went through poll handler */ |
564 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
565 | /* buffer already selected */ |
566 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
d7718a9d JA |
567 | }; |
568 | ||
569 | struct async_poll { | |
570 | struct io_poll_iocb poll; | |
571 | struct io_wq_work work; | |
6b47ee6e PB |
572 | }; |
573 | ||
09bb8394 JA |
574 | /* |
575 | * NOTE! Each of the iocb union members has the file pointer | |
576 | * as the first entry in their struct definition. So you can | |
577 | * access the file pointer through any of the sub-structs, | |
578 | * or directly as just 'ki_filp' in this struct. | |
579 | */ | |
2b188cc1 | 580 | struct io_kiocb { |
221c5eb2 | 581 | union { |
09bb8394 | 582 | struct file *file; |
9adbd45d | 583 | struct io_rw rw; |
221c5eb2 | 584 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
585 | struct io_accept accept; |
586 | struct io_sync sync; | |
fbf23849 | 587 | struct io_cancel cancel; |
b29472ee | 588 | struct io_timeout timeout; |
3fbb51c1 | 589 | struct io_connect connect; |
e47293fd | 590 | struct io_sr_msg sr_msg; |
15b71abe | 591 | struct io_open open; |
b5dba59e | 592 | struct io_close close; |
05f3fb3c | 593 | struct io_files_update files_update; |
4840e418 | 594 | struct io_fadvise fadvise; |
c1ca757b | 595 | struct io_madvise madvise; |
3e4827b0 | 596 | struct io_epoll epoll; |
7d67af2c | 597 | struct io_splice splice; |
ddf0322d | 598 | struct io_provide_buf pbuf; |
221c5eb2 | 599 | }; |
2b188cc1 | 600 | |
1a6b74fc | 601 | struct io_async_ctx *io; |
cf6fd4bd | 602 | bool needs_fixed_file; |
d625c6ee | 603 | u8 opcode; |
2b188cc1 JA |
604 | |
605 | struct io_ring_ctx *ctx; | |
d7718a9d | 606 | struct list_head list; |
2b188cc1 | 607 | unsigned int flags; |
c16361c1 | 608 | refcount_t refs; |
4ed734b0 JA |
609 | union { |
610 | struct task_struct *task; | |
611 | unsigned long fsize; | |
612 | }; | |
2b188cc1 | 613 | u64 user_data; |
9e645e11 | 614 | u32 result; |
de0617e4 | 615 | u32 sequence; |
2b188cc1 | 616 | |
d7718a9d JA |
617 | struct list_head link_list; |
618 | ||
fcb323cc JA |
619 | struct list_head inflight_entry; |
620 | ||
b41e9852 JA |
621 | union { |
622 | /* | |
623 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
624 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
625 | * async armed poll handlers for regular commands. The latter | |
626 | * restore the work, if needed. | |
b41e9852 JA |
627 | */ |
628 | struct { | |
b41e9852 | 629 | struct callback_head task_work; |
d7718a9d JA |
630 | struct hlist_node hash_node; |
631 | struct async_poll *apoll; | |
bcda7baa | 632 | int cflags; |
b41e9852 JA |
633 | }; |
634 | struct io_wq_work work; | |
635 | }; | |
2b188cc1 JA |
636 | }; |
637 | ||
638 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 639 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 640 | |
9a56a232 JA |
641 | struct io_submit_state { |
642 | struct blk_plug plug; | |
643 | ||
2579f913 JA |
644 | /* |
645 | * io_kiocb alloc cache | |
646 | */ | |
647 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 648 | unsigned int free_reqs; |
2579f913 | 649 | |
9a56a232 JA |
650 | /* |
651 | * File reference cache | |
652 | */ | |
653 | struct file *file; | |
654 | unsigned int fd; | |
655 | unsigned int has_refs; | |
656 | unsigned int used_refs; | |
657 | unsigned int ios_left; | |
658 | }; | |
659 | ||
d3656344 JA |
660 | struct io_op_def { |
661 | /* needs req->io allocated for deferral/async */ | |
662 | unsigned async_ctx : 1; | |
663 | /* needs current->mm setup, does mm access */ | |
664 | unsigned needs_mm : 1; | |
665 | /* needs req->file assigned */ | |
666 | unsigned needs_file : 1; | |
667 | /* needs req->file assigned IFF fd is >= 0 */ | |
668 | unsigned fd_non_neg : 1; | |
669 | /* hash wq insertion if file is a regular file */ | |
670 | unsigned hash_reg_file : 1; | |
671 | /* unbound wq insertion if file is a non-regular file */ | |
672 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
673 | /* opcode is not supported by this kernel */ |
674 | unsigned not_supported : 1; | |
f86cd20c JA |
675 | /* needs file table */ |
676 | unsigned file_table : 1; | |
ff002b30 JA |
677 | /* needs ->fs */ |
678 | unsigned needs_fs : 1; | |
8a72758c JA |
679 | /* set if opcode supports polled "wait" */ |
680 | unsigned pollin : 1; | |
681 | unsigned pollout : 1; | |
bcda7baa JA |
682 | /* op supports buffer selection */ |
683 | unsigned buffer_select : 1; | |
d3656344 JA |
684 | }; |
685 | ||
686 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
687 | [IORING_OP_NOP] = {}, |
688 | [IORING_OP_READV] = { | |
d3656344 JA |
689 | .async_ctx = 1, |
690 | .needs_mm = 1, | |
691 | .needs_file = 1, | |
692 | .unbound_nonreg_file = 1, | |
8a72758c | 693 | .pollin = 1, |
4d954c25 | 694 | .buffer_select = 1, |
d3656344 | 695 | }, |
0463b6c5 | 696 | [IORING_OP_WRITEV] = { |
d3656344 JA |
697 | .async_ctx = 1, |
698 | .needs_mm = 1, | |
699 | .needs_file = 1, | |
700 | .hash_reg_file = 1, | |
701 | .unbound_nonreg_file = 1, | |
8a72758c | 702 | .pollout = 1, |
d3656344 | 703 | }, |
0463b6c5 | 704 | [IORING_OP_FSYNC] = { |
d3656344 JA |
705 | .needs_file = 1, |
706 | }, | |
0463b6c5 | 707 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
708 | .needs_file = 1, |
709 | .unbound_nonreg_file = 1, | |
8a72758c | 710 | .pollin = 1, |
d3656344 | 711 | }, |
0463b6c5 | 712 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
713 | .needs_file = 1, |
714 | .hash_reg_file = 1, | |
715 | .unbound_nonreg_file = 1, | |
8a72758c | 716 | .pollout = 1, |
d3656344 | 717 | }, |
0463b6c5 | 718 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
719 | .needs_file = 1, |
720 | .unbound_nonreg_file = 1, | |
721 | }, | |
0463b6c5 PB |
722 | [IORING_OP_POLL_REMOVE] = {}, |
723 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
724 | .needs_file = 1, |
725 | }, | |
0463b6c5 | 726 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
727 | .async_ctx = 1, |
728 | .needs_mm = 1, | |
729 | .needs_file = 1, | |
730 | .unbound_nonreg_file = 1, | |
ff002b30 | 731 | .needs_fs = 1, |
8a72758c | 732 | .pollout = 1, |
d3656344 | 733 | }, |
0463b6c5 | 734 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
735 | .async_ctx = 1, |
736 | .needs_mm = 1, | |
737 | .needs_file = 1, | |
738 | .unbound_nonreg_file = 1, | |
ff002b30 | 739 | .needs_fs = 1, |
8a72758c | 740 | .pollin = 1, |
52de1fe1 | 741 | .buffer_select = 1, |
d3656344 | 742 | }, |
0463b6c5 | 743 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
744 | .async_ctx = 1, |
745 | .needs_mm = 1, | |
746 | }, | |
0463b6c5 PB |
747 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
748 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
749 | .needs_mm = 1, |
750 | .needs_file = 1, | |
751 | .unbound_nonreg_file = 1, | |
f86cd20c | 752 | .file_table = 1, |
8a72758c | 753 | .pollin = 1, |
d3656344 | 754 | }, |
0463b6c5 PB |
755 | [IORING_OP_ASYNC_CANCEL] = {}, |
756 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
757 | .async_ctx = 1, |
758 | .needs_mm = 1, | |
759 | }, | |
0463b6c5 | 760 | [IORING_OP_CONNECT] = { |
d3656344 JA |
761 | .async_ctx = 1, |
762 | .needs_mm = 1, | |
763 | .needs_file = 1, | |
764 | .unbound_nonreg_file = 1, | |
8a72758c | 765 | .pollout = 1, |
d3656344 | 766 | }, |
0463b6c5 | 767 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
768 | .needs_file = 1, |
769 | }, | |
0463b6c5 | 770 | [IORING_OP_OPENAT] = { |
d3656344 JA |
771 | .needs_file = 1, |
772 | .fd_non_neg = 1, | |
f86cd20c | 773 | .file_table = 1, |
ff002b30 | 774 | .needs_fs = 1, |
d3656344 | 775 | }, |
0463b6c5 | 776 | [IORING_OP_CLOSE] = { |
d3656344 | 777 | .needs_file = 1, |
f86cd20c | 778 | .file_table = 1, |
d3656344 | 779 | }, |
0463b6c5 | 780 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 781 | .needs_mm = 1, |
f86cd20c | 782 | .file_table = 1, |
d3656344 | 783 | }, |
0463b6c5 | 784 | [IORING_OP_STATX] = { |
d3656344 JA |
785 | .needs_mm = 1, |
786 | .needs_file = 1, | |
787 | .fd_non_neg = 1, | |
ff002b30 | 788 | .needs_fs = 1, |
d3656344 | 789 | }, |
0463b6c5 | 790 | [IORING_OP_READ] = { |
3a6820f2 JA |
791 | .needs_mm = 1, |
792 | .needs_file = 1, | |
793 | .unbound_nonreg_file = 1, | |
8a72758c | 794 | .pollin = 1, |
bcda7baa | 795 | .buffer_select = 1, |
3a6820f2 | 796 | }, |
0463b6c5 | 797 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
798 | .needs_mm = 1, |
799 | .needs_file = 1, | |
800 | .unbound_nonreg_file = 1, | |
8a72758c | 801 | .pollout = 1, |
3a6820f2 | 802 | }, |
0463b6c5 | 803 | [IORING_OP_FADVISE] = { |
4840e418 JA |
804 | .needs_file = 1, |
805 | }, | |
0463b6c5 | 806 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
807 | .needs_mm = 1, |
808 | }, | |
0463b6c5 | 809 | [IORING_OP_SEND] = { |
fddaface JA |
810 | .needs_mm = 1, |
811 | .needs_file = 1, | |
812 | .unbound_nonreg_file = 1, | |
8a72758c | 813 | .pollout = 1, |
fddaface | 814 | }, |
0463b6c5 | 815 | [IORING_OP_RECV] = { |
fddaface JA |
816 | .needs_mm = 1, |
817 | .needs_file = 1, | |
818 | .unbound_nonreg_file = 1, | |
8a72758c | 819 | .pollin = 1, |
bcda7baa | 820 | .buffer_select = 1, |
fddaface | 821 | }, |
0463b6c5 | 822 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
823 | .needs_file = 1, |
824 | .fd_non_neg = 1, | |
f86cd20c | 825 | .file_table = 1, |
ff002b30 | 826 | .needs_fs = 1, |
cebdb986 | 827 | }, |
3e4827b0 JA |
828 | [IORING_OP_EPOLL_CTL] = { |
829 | .unbound_nonreg_file = 1, | |
830 | .file_table = 1, | |
831 | }, | |
7d67af2c PB |
832 | [IORING_OP_SPLICE] = { |
833 | .needs_file = 1, | |
834 | .hash_reg_file = 1, | |
835 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
836 | }, |
837 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 838 | [IORING_OP_REMOVE_BUFFERS] = {}, |
d3656344 JA |
839 | }; |
840 | ||
561fb04a | 841 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 842 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 843 | static void io_put_req(struct io_kiocb *req); |
978db57e | 844 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
845 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
846 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
847 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
848 | struct io_uring_files_update *ip, | |
849 | unsigned nr_args); | |
f86cd20c | 850 | static int io_grab_files(struct io_kiocb *req); |
2faf852d | 851 | static void io_ring_file_ref_flush(struct fixed_file_data *data); |
99bc4c38 | 852 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
853 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
854 | int fd, struct file **out_file, bool fixed); | |
855 | static void __io_queue_sqe(struct io_kiocb *req, | |
856 | const struct io_uring_sqe *sqe); | |
de0617e4 | 857 | |
2b188cc1 JA |
858 | static struct kmem_cache *req_cachep; |
859 | ||
860 | static const struct file_operations io_uring_fops; | |
861 | ||
862 | struct sock *io_uring_get_socket(struct file *file) | |
863 | { | |
864 | #if defined(CONFIG_UNIX) | |
865 | if (file->f_op == &io_uring_fops) { | |
866 | struct io_ring_ctx *ctx = file->private_data; | |
867 | ||
868 | return ctx->ring_sock->sk; | |
869 | } | |
870 | #endif | |
871 | return NULL; | |
872 | } | |
873 | EXPORT_SYMBOL(io_uring_get_socket); | |
874 | ||
875 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
876 | { | |
877 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
878 | ||
206aefde | 879 | complete(&ctx->completions[0]); |
2b188cc1 JA |
880 | } |
881 | ||
882 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
883 | { | |
884 | struct io_ring_ctx *ctx; | |
78076bb6 | 885 | int hash_bits; |
2b188cc1 JA |
886 | |
887 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
888 | if (!ctx) | |
889 | return NULL; | |
890 | ||
0ddf92e8 JA |
891 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
892 | if (!ctx->fallback_req) | |
893 | goto err; | |
894 | ||
206aefde JA |
895 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
896 | if (!ctx->completions) | |
897 | goto err; | |
898 | ||
78076bb6 JA |
899 | /* |
900 | * Use 5 bits less than the max cq entries, that should give us around | |
901 | * 32 entries per hash list if totally full and uniformly spread. | |
902 | */ | |
903 | hash_bits = ilog2(p->cq_entries); | |
904 | hash_bits -= 5; | |
905 | if (hash_bits <= 0) | |
906 | hash_bits = 1; | |
907 | ctx->cancel_hash_bits = hash_bits; | |
908 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
909 | GFP_KERNEL); | |
910 | if (!ctx->cancel_hash) | |
911 | goto err; | |
912 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
913 | ||
21482896 | 914 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
915 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
916 | goto err; | |
2b188cc1 JA |
917 | |
918 | ctx->flags = p->flags; | |
919 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 920 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
921 | init_completion(&ctx->completions[0]); |
922 | init_completion(&ctx->completions[1]); | |
5a2e745d | 923 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 924 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
925 | mutex_init(&ctx->uring_lock); |
926 | init_waitqueue_head(&ctx->wait); | |
927 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 928 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 929 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 930 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
931 | init_waitqueue_head(&ctx->inflight_wait); |
932 | spin_lock_init(&ctx->inflight_lock); | |
933 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 934 | return ctx; |
206aefde | 935 | err: |
0ddf92e8 JA |
936 | if (ctx->fallback_req) |
937 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 938 | kfree(ctx->completions); |
78076bb6 | 939 | kfree(ctx->cancel_hash); |
206aefde JA |
940 | kfree(ctx); |
941 | return NULL; | |
2b188cc1 JA |
942 | } |
943 | ||
9d858b21 | 944 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 945 | { |
a197f664 JL |
946 | struct io_ring_ctx *ctx = req->ctx; |
947 | ||
498ccd9e JA |
948 | return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped |
949 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
950 | } |
951 | ||
9d858b21 | 952 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 953 | { |
87987898 | 954 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 955 | return __req_need_defer(req); |
de0617e4 | 956 | |
9d858b21 | 957 | return false; |
de0617e4 JA |
958 | } |
959 | ||
7adf4eaf | 960 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
961 | { |
962 | struct io_kiocb *req; | |
963 | ||
7adf4eaf | 964 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 965 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
966 | list_del_init(&req->list); |
967 | return req; | |
968 | } | |
969 | ||
970 | return NULL; | |
971 | } | |
972 | ||
5262f567 JA |
973 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
974 | { | |
7adf4eaf JA |
975 | struct io_kiocb *req; |
976 | ||
977 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
978 | if (req) { |
979 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
980 | return NULL; | |
fb4b3d3f | 981 | if (!__req_need_defer(req)) { |
93bd25bb JA |
982 | list_del_init(&req->list); |
983 | return req; | |
984 | } | |
7adf4eaf JA |
985 | } |
986 | ||
987 | return NULL; | |
5262f567 JA |
988 | } |
989 | ||
de0617e4 | 990 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 991 | { |
75b28aff | 992 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 993 | |
07910158 PB |
994 | /* order cqe stores with ring update */ |
995 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 996 | |
07910158 PB |
997 | if (wq_has_sleeper(&ctx->cq_wait)) { |
998 | wake_up_interruptible(&ctx->cq_wait); | |
999 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
1000 | } |
1001 | } | |
1002 | ||
cccf0ee8 JA |
1003 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
1004 | const struct io_op_def *def) | |
1005 | { | |
1006 | if (!req->work.mm && def->needs_mm) { | |
1007 | mmgrab(current->mm); | |
1008 | req->work.mm = current->mm; | |
2b188cc1 | 1009 | } |
cccf0ee8 JA |
1010 | if (!req->work.creds) |
1011 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
1012 | if (!req->work.fs && def->needs_fs) { |
1013 | spin_lock(¤t->fs->lock); | |
1014 | if (!current->fs->in_exec) { | |
1015 | req->work.fs = current->fs; | |
1016 | req->work.fs->users++; | |
1017 | } else { | |
1018 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1019 | } | |
1020 | spin_unlock(¤t->fs->lock); | |
1021 | } | |
6ab23144 JA |
1022 | if (!req->work.task_pid) |
1023 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
1024 | } |
1025 | ||
cccf0ee8 | 1026 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 1027 | { |
cccf0ee8 JA |
1028 | if (req->work.mm) { |
1029 | mmdrop(req->work.mm); | |
1030 | req->work.mm = NULL; | |
1031 | } | |
1032 | if (req->work.creds) { | |
1033 | put_cred(req->work.creds); | |
1034 | req->work.creds = NULL; | |
1035 | } | |
ff002b30 JA |
1036 | if (req->work.fs) { |
1037 | struct fs_struct *fs = req->work.fs; | |
1038 | ||
1039 | spin_lock(&req->work.fs->lock); | |
1040 | if (--fs->users) | |
1041 | fs = NULL; | |
1042 | spin_unlock(&req->work.fs->lock); | |
1043 | if (fs) | |
1044 | free_fs_struct(fs); | |
1045 | } | |
561fb04a JA |
1046 | } |
1047 | ||
8766dd51 | 1048 | static inline void io_prep_async_work(struct io_kiocb *req, |
94ae5e77 | 1049 | struct io_kiocb **link) |
18d9be1a | 1050 | { |
d3656344 | 1051 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
54a91f3b | 1052 | |
d3656344 JA |
1053 | if (req->flags & REQ_F_ISREG) { |
1054 | if (def->hash_reg_file) | |
8766dd51 | 1055 | io_wq_hash_work(&req->work, file_inode(req->file)); |
d3656344 JA |
1056 | } else { |
1057 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1058 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1059 | } |
cccf0ee8 JA |
1060 | |
1061 | io_req_work_grab_env(req, def); | |
54a91f3b | 1062 | |
94ae5e77 | 1063 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1064 | } |
1065 | ||
a197f664 | 1066 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1067 | { |
a197f664 | 1068 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 | 1069 | struct io_kiocb *link; |
94ae5e77 | 1070 | |
8766dd51 | 1071 | io_prep_async_work(req, &link); |
561fb04a | 1072 | |
8766dd51 PB |
1073 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1074 | &req->work, req->flags); | |
1075 | io_wq_enqueue(ctx->io_wq, &req->work); | |
94ae5e77 JA |
1076 | |
1077 | if (link) | |
1078 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1079 | } |
1080 | ||
5262f567 JA |
1081 | static void io_kill_timeout(struct io_kiocb *req) |
1082 | { | |
1083 | int ret; | |
1084 | ||
2d28390a | 1085 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1086 | if (ret != -1) { |
1087 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1088 | list_del_init(&req->list); |
f0e20b89 | 1089 | req->flags |= REQ_F_COMP_LOCKED; |
78e19bbe | 1090 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1091 | io_put_req(req); |
5262f567 JA |
1092 | } |
1093 | } | |
1094 | ||
1095 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1096 | { | |
1097 | struct io_kiocb *req, *tmp; | |
1098 | ||
1099 | spin_lock_irq(&ctx->completion_lock); | |
1100 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1101 | io_kill_timeout(req); | |
1102 | spin_unlock_irq(&ctx->completion_lock); | |
1103 | } | |
1104 | ||
de0617e4 JA |
1105 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1106 | { | |
1107 | struct io_kiocb *req; | |
1108 | ||
5262f567 JA |
1109 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1110 | io_kill_timeout(req); | |
1111 | ||
de0617e4 JA |
1112 | __io_commit_cqring(ctx); |
1113 | ||
87987898 | 1114 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1115 | io_queue_async_work(req); |
de0617e4 JA |
1116 | } |
1117 | ||
2b188cc1 JA |
1118 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1119 | { | |
75b28aff | 1120 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1121 | unsigned tail; |
1122 | ||
1123 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1124 | /* |
1125 | * writes to the cq entry need to come after reading head; the | |
1126 | * control dependency is enough as we're using WRITE_ONCE to | |
1127 | * fill the cq entry | |
1128 | */ | |
75b28aff | 1129 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1130 | return NULL; |
1131 | ||
1132 | ctx->cached_cq_tail++; | |
75b28aff | 1133 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1134 | } |
1135 | ||
f2842ab5 JA |
1136 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1137 | { | |
f0b493e6 JA |
1138 | if (!ctx->cq_ev_fd) |
1139 | return false; | |
f2842ab5 JA |
1140 | if (!ctx->eventfd_async) |
1141 | return true; | |
b41e9852 | 1142 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1143 | } |
1144 | ||
b41e9852 | 1145 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1146 | { |
1147 | if (waitqueue_active(&ctx->wait)) | |
1148 | wake_up(&ctx->wait); | |
1149 | if (waitqueue_active(&ctx->sqo_wait)) | |
1150 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1151 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1152 | eventfd_signal(ctx->cq_ev_fd, 1); |
1153 | } | |
1154 | ||
c4a2ed72 JA |
1155 | /* Returns true if there are no backlogged entries after the flush */ |
1156 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1157 | { |
1158 | struct io_rings *rings = ctx->rings; | |
1159 | struct io_uring_cqe *cqe; | |
1160 | struct io_kiocb *req; | |
1161 | unsigned long flags; | |
1162 | LIST_HEAD(list); | |
1163 | ||
1164 | if (!force) { | |
1165 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1166 | return true; |
1d7bb1d5 JA |
1167 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1168 | rings->cq_ring_entries)) | |
c4a2ed72 | 1169 | return false; |
1d7bb1d5 JA |
1170 | } |
1171 | ||
1172 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1173 | ||
1174 | /* if force is set, the ring is going away. always drop after that */ | |
1175 | if (force) | |
69b3e546 | 1176 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1177 | |
c4a2ed72 | 1178 | cqe = NULL; |
1d7bb1d5 JA |
1179 | while (!list_empty(&ctx->cq_overflow_list)) { |
1180 | cqe = io_get_cqring(ctx); | |
1181 | if (!cqe && !force) | |
1182 | break; | |
1183 | ||
1184 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1185 | list); | |
1186 | list_move(&req->list, &list); | |
2ca10259 | 1187 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1188 | if (cqe) { |
1189 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1190 | WRITE_ONCE(cqe->res, req->result); | |
bcda7baa | 1191 | WRITE_ONCE(cqe->flags, req->cflags); |
1d7bb1d5 JA |
1192 | } else { |
1193 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1194 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1199 | if (cqe) { |
1200 | clear_bit(0, &ctx->sq_check_overflow); | |
1201 | clear_bit(0, &ctx->cq_check_overflow); | |
1202 | } | |
1d7bb1d5 JA |
1203 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1204 | io_cqring_ev_posted(ctx); | |
1205 | ||
1206 | while (!list_empty(&list)) { | |
1207 | req = list_first_entry(&list, struct io_kiocb, list); | |
1208 | list_del(&req->list); | |
ec9c02ad | 1209 | io_put_req(req); |
1d7bb1d5 | 1210 | } |
c4a2ed72 JA |
1211 | |
1212 | return cqe != NULL; | |
1d7bb1d5 JA |
1213 | } |
1214 | ||
bcda7baa | 1215 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1216 | { |
78e19bbe | 1217 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1218 | struct io_uring_cqe *cqe; |
1219 | ||
78e19bbe | 1220 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1221 | |
2b188cc1 JA |
1222 | /* |
1223 | * If we can't get a cq entry, userspace overflowed the | |
1224 | * submission (by quite a lot). Increment the overflow count in | |
1225 | * the ring. | |
1226 | */ | |
1227 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1228 | if (likely(cqe)) { |
78e19bbe | 1229 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1230 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1231 | WRITE_ONCE(cqe->flags, cflags); |
1d7bb1d5 | 1232 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1233 | WRITE_ONCE(ctx->rings->cq_overflow, |
1234 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1235 | } else { |
ad3eb2c8 JA |
1236 | if (list_empty(&ctx->cq_overflow_list)) { |
1237 | set_bit(0, &ctx->sq_check_overflow); | |
1238 | set_bit(0, &ctx->cq_check_overflow); | |
1239 | } | |
2ca10259 | 1240 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1241 | refcount_inc(&req->refs); |
1242 | req->result = res; | |
bcda7baa | 1243 | req->cflags = cflags; |
1d7bb1d5 | 1244 | list_add_tail(&req->list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1245 | } |
1246 | } | |
1247 | ||
bcda7baa JA |
1248 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1249 | { | |
1250 | __io_cqring_fill_event(req, res, 0); | |
1251 | } | |
1252 | ||
1253 | static void __io_cqring_add_event(struct io_kiocb *req, long res, long cflags) | |
2b188cc1 | 1254 | { |
78e19bbe | 1255 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1256 | unsigned long flags; |
1257 | ||
1258 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1259 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 JA |
1260 | io_commit_cqring(ctx); |
1261 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1262 | ||
8c838788 | 1263 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1264 | } |
1265 | ||
bcda7baa JA |
1266 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
1267 | { | |
1268 | __io_cqring_add_event(req, res, 0); | |
1269 | } | |
1270 | ||
0ddf92e8 JA |
1271 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1272 | { | |
1273 | return req == (struct io_kiocb *) | |
1274 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1275 | } | |
1276 | ||
1277 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1278 | { | |
1279 | struct io_kiocb *req; | |
1280 | ||
1281 | req = ctx->fallback_req; | |
1282 | if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req)) | |
1283 | return req; | |
1284 | ||
1285 | return NULL; | |
1286 | } | |
1287 | ||
2579f913 JA |
1288 | static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx, |
1289 | struct io_submit_state *state) | |
2b188cc1 | 1290 | { |
fd6fab2c | 1291 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1292 | struct io_kiocb *req; |
1293 | ||
2579f913 | 1294 | if (!state) { |
fd6fab2c | 1295 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1296 | if (unlikely(!req)) |
0ddf92e8 | 1297 | goto fallback; |
2579f913 JA |
1298 | } else if (!state->free_reqs) { |
1299 | size_t sz; | |
1300 | int ret; | |
1301 | ||
1302 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1303 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1304 | ||
1305 | /* | |
1306 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1307 | * retry single alloc to be on the safe side. | |
1308 | */ | |
1309 | if (unlikely(ret <= 0)) { | |
1310 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1311 | if (!state->reqs[0]) | |
0ddf92e8 | 1312 | goto fallback; |
fd6fab2c JA |
1313 | ret = 1; |
1314 | } | |
2579f913 | 1315 | state->free_reqs = ret - 1; |
6c8a3134 | 1316 | req = state->reqs[ret - 1]; |
2579f913 | 1317 | } else { |
2579f913 | 1318 | state->free_reqs--; |
6c8a3134 | 1319 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1320 | } |
1321 | ||
0ddf92e8 | 1322 | got_it: |
1a6b74fc | 1323 | req->io = NULL; |
60c112b0 | 1324 | req->file = NULL; |
2579f913 JA |
1325 | req->ctx = ctx; |
1326 | req->flags = 0; | |
e65ef56d JA |
1327 | /* one is dropped after submission, the other at completion */ |
1328 | refcount_set(&req->refs, 2); | |
9e645e11 | 1329 | req->result = 0; |
561fb04a | 1330 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
2579f913 | 1331 | return req; |
0ddf92e8 JA |
1332 | fallback: |
1333 | req = io_get_fallback_req(ctx); | |
1334 | if (req) | |
1335 | goto got_it; | |
6805b32e | 1336 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
1337 | return NULL; |
1338 | } | |
1339 | ||
8da11c19 PB |
1340 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1341 | bool fixed) | |
1342 | { | |
1343 | if (fixed) | |
1344 | percpu_ref_put(&req->ctx->file_data->refs); | |
1345 | else | |
1346 | fput(file); | |
1347 | } | |
1348 | ||
2b85edfc | 1349 | static void __io_req_do_free(struct io_kiocb *req) |
def596e9 | 1350 | { |
2b85edfc PB |
1351 | if (likely(!io_is_fallback_req(req))) |
1352 | kmem_cache_free(req_cachep, req); | |
1353 | else | |
1354 | clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req); | |
1355 | } | |
1356 | ||
c6ca97b3 | 1357 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1358 | { |
929a3af9 PB |
1359 | if (req->flags & REQ_F_NEED_CLEANUP) |
1360 | io_cleanup_req(req); | |
1361 | ||
96fd84d8 | 1362 | kfree(req->io); |
8da11c19 PB |
1363 | if (req->file) |
1364 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
cccf0ee8 JA |
1365 | |
1366 | io_req_work_drop_env(req); | |
def596e9 JA |
1367 | } |
1368 | ||
9e645e11 | 1369 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1370 | { |
c6ca97b3 | 1371 | __io_req_aux_free(req); |
fcb323cc | 1372 | |
fcb323cc | 1373 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1374 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1375 | unsigned long flags; |
1376 | ||
1377 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1378 | list_del(&req->inflight_entry); | |
1379 | if (waitqueue_active(&ctx->inflight_wait)) | |
1380 | wake_up(&ctx->inflight_wait); | |
1381 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1382 | } | |
2b85edfc PB |
1383 | |
1384 | percpu_ref_put(&req->ctx->refs); | |
1385 | __io_req_do_free(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 | { | |
10fef4be JA |
1396 | int fixed_refs = rb->to_free; |
1397 | ||
c6ca97b3 JA |
1398 | if (!rb->to_free) |
1399 | return; | |
1400 | if (rb->need_iter) { | |
1401 | int i, inflight = 0; | |
1402 | unsigned long flags; | |
1403 | ||
10fef4be | 1404 | fixed_refs = 0; |
c6ca97b3 JA |
1405 | for (i = 0; i < rb->to_free; i++) { |
1406 | struct io_kiocb *req = rb->reqs[i]; | |
1407 | ||
10fef4be | 1408 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1409 | req->file = NULL; |
10fef4be JA |
1410 | fixed_refs++; |
1411 | } | |
c6ca97b3 JA |
1412 | if (req->flags & REQ_F_INFLIGHT) |
1413 | inflight++; | |
c6ca97b3 JA |
1414 | __io_req_aux_free(req); |
1415 | } | |
1416 | if (!inflight) | |
1417 | goto do_free; | |
1418 | ||
1419 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1420 | for (i = 0; i < rb->to_free; i++) { | |
1421 | struct io_kiocb *req = rb->reqs[i]; | |
1422 | ||
10fef4be | 1423 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1424 | list_del(&req->inflight_entry); |
1425 | if (!--inflight) | |
1426 | break; | |
1427 | } | |
1428 | } | |
1429 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1430 | ||
1431 | if (waitqueue_active(&ctx->inflight_wait)) | |
1432 | wake_up(&ctx->inflight_wait); | |
1433 | } | |
1434 | do_free: | |
1435 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
10fef4be JA |
1436 | if (fixed_refs) |
1437 | percpu_ref_put_many(&ctx->file_data->refs, fixed_refs); | |
c6ca97b3 | 1438 | percpu_ref_put_many(&ctx->refs, rb->to_free); |
c6ca97b3 | 1439 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1440 | } |
1441 | ||
a197f664 | 1442 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1443 | { |
a197f664 | 1444 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1445 | int ret; |
1446 | ||
2d28390a | 1447 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1448 | if (ret != -1) { |
78e19bbe | 1449 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd JA |
1450 | io_commit_cqring(ctx); |
1451 | req->flags &= ~REQ_F_LINK; | |
ec9c02ad | 1452 | io_put_req(req); |
2665abfd JA |
1453 | return true; |
1454 | } | |
1455 | ||
1456 | return false; | |
e65ef56d JA |
1457 | } |
1458 | ||
ba816ad6 | 1459 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1460 | { |
2665abfd | 1461 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1462 | bool wake_ev = false; |
9e645e11 | 1463 | |
4d7dd462 JA |
1464 | /* Already got next link */ |
1465 | if (req->flags & REQ_F_LINK_NEXT) | |
1466 | return; | |
1467 | ||
9e645e11 JA |
1468 | /* |
1469 | * The list should never be empty when we are called here. But could | |
1470 | * potentially happen if the chain is messed up, check to be on the | |
1471 | * safe side. | |
1472 | */ | |
4493233e PB |
1473 | while (!list_empty(&req->link_list)) { |
1474 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1475 | struct io_kiocb, link_list); | |
94ae5e77 | 1476 | |
4493233e PB |
1477 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1478 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1479 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1480 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1481 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1482 | continue; | |
1483 | } | |
9e645e11 | 1484 | |
4493233e PB |
1485 | list_del_init(&req->link_list); |
1486 | if (!list_empty(&nxt->link_list)) | |
1487 | nxt->flags |= REQ_F_LINK; | |
b18fdf71 | 1488 | *nxtptr = nxt; |
94ae5e77 | 1489 | break; |
9e645e11 | 1490 | } |
2665abfd | 1491 | |
4d7dd462 | 1492 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1493 | if (wake_ev) |
1494 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1495 | } |
1496 | ||
1497 | /* | |
1498 | * Called if REQ_F_LINK is set, and we fail the head request | |
1499 | */ | |
1500 | static void io_fail_links(struct io_kiocb *req) | |
1501 | { | |
2665abfd | 1502 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1503 | unsigned long flags; |
1504 | ||
1505 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1506 | |
1507 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1508 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1509 | struct io_kiocb, link_list); | |
9e645e11 | 1510 | |
4493233e | 1511 | list_del_init(&link->link_list); |
c826bd7a | 1512 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1513 | |
1514 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1515 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1516 | io_link_cancel_timeout(link); |
2665abfd | 1517 | } else { |
78e19bbe | 1518 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1519 | __io_double_put_req(link); |
2665abfd | 1520 | } |
5d960724 | 1521 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1522 | } |
2665abfd JA |
1523 | |
1524 | io_commit_cqring(ctx); | |
1525 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1526 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1527 | } |
1528 | ||
4d7dd462 | 1529 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1530 | { |
4d7dd462 | 1531 | if (likely(!(req->flags & REQ_F_LINK))) |
2665abfd | 1532 | return; |
2665abfd | 1533 | |
9e645e11 JA |
1534 | /* |
1535 | * If LINK is set, we have dependent requests in this chain. If we | |
1536 | * didn't fail this request, queue the first one up, moving any other | |
1537 | * dependencies to the next request. In case of failure, fail the rest | |
1538 | * of the chain. | |
1539 | */ | |
2665abfd JA |
1540 | if (req->flags & REQ_F_FAIL_LINK) { |
1541 | io_fail_links(req); | |
7c9e7f0f JA |
1542 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1543 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1544 | struct io_ring_ctx *ctx = req->ctx; |
1545 | unsigned long flags; | |
1546 | ||
1547 | /* | |
1548 | * If this is a timeout link, we could be racing with the | |
1549 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1550 | * protect against that. |
2665abfd JA |
1551 | */ |
1552 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1553 | io_req_link_next(req, nxt); | |
1554 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1555 | } else { | |
1556 | io_req_link_next(req, nxt); | |
9e645e11 | 1557 | } |
4d7dd462 | 1558 | } |
9e645e11 | 1559 | |
c69f8dbe JL |
1560 | static void io_free_req(struct io_kiocb *req) |
1561 | { | |
944e58bf PB |
1562 | struct io_kiocb *nxt = NULL; |
1563 | ||
1564 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1565 | __io_free_req(req); |
944e58bf PB |
1566 | |
1567 | if (nxt) | |
1568 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1569 | } |
1570 | ||
7a743e22 PB |
1571 | static void io_link_work_cb(struct io_wq_work **workptr) |
1572 | { | |
18a542ff PB |
1573 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
1574 | struct io_kiocb *link; | |
7a743e22 | 1575 | |
18a542ff | 1576 | link = list_first_entry(&req->link_list, struct io_kiocb, link_list); |
7a743e22 PB |
1577 | io_queue_linked_timeout(link); |
1578 | io_wq_submit_work(workptr); | |
1579 | } | |
1580 | ||
1581 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
1582 | { | |
1583 | struct io_kiocb *link; | |
8766dd51 PB |
1584 | const struct io_op_def *def = &io_op_defs[nxt->opcode]; |
1585 | ||
1586 | if ((nxt->flags & REQ_F_ISREG) && def->hash_reg_file) | |
1587 | io_wq_hash_work(&nxt->work, file_inode(nxt->file)); | |
7a743e22 PB |
1588 | |
1589 | *workptr = &nxt->work; | |
1590 | link = io_prep_linked_timeout(nxt); | |
18a542ff | 1591 | if (link) |
7a743e22 | 1592 | nxt->work.func = io_link_work_cb; |
7a743e22 PB |
1593 | } |
1594 | ||
ba816ad6 JA |
1595 | /* |
1596 | * Drop reference to request, return next in chain (if there is one) if this | |
1597 | * was the last reference to this request. | |
1598 | */ | |
f9bd67f6 | 1599 | __attribute__((nonnull)) |
ec9c02ad | 1600 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1601 | { |
2a44f467 JA |
1602 | if (refcount_dec_and_test(&req->refs)) { |
1603 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1604 | __io_free_req(req); |
2a44f467 | 1605 | } |
2b188cc1 JA |
1606 | } |
1607 | ||
e65ef56d JA |
1608 | static void io_put_req(struct io_kiocb *req) |
1609 | { | |
1610 | if (refcount_dec_and_test(&req->refs)) | |
1611 | io_free_req(req); | |
2b188cc1 JA |
1612 | } |
1613 | ||
e9fd9396 PB |
1614 | static void io_steal_work(struct io_kiocb *req, |
1615 | struct io_wq_work **workptr) | |
7a743e22 PB |
1616 | { |
1617 | /* | |
1618 | * It's in an io-wq worker, so there always should be at least | |
1619 | * one reference, which will be dropped in io_put_work() just | |
1620 | * after the current handler returns. | |
1621 | * | |
1622 | * It also means, that if the counter dropped to 1, then there is | |
1623 | * no asynchronous users left, so it's safe to steal the next work. | |
1624 | */ | |
7a743e22 PB |
1625 | if (refcount_read(&req->refs) == 1) { |
1626 | struct io_kiocb *nxt = NULL; | |
1627 | ||
1628 | io_req_find_next(req, &nxt); | |
1629 | if (nxt) | |
1630 | io_wq_assign_next(workptr, nxt); | |
1631 | } | |
1632 | } | |
1633 | ||
978db57e JA |
1634 | /* |
1635 | * Must only be used if we don't need to care about links, usually from | |
1636 | * within the completion handling itself. | |
1637 | */ | |
1638 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1639 | { |
1640 | /* drop both submit and complete references */ | |
1641 | if (refcount_sub_and_test(2, &req->refs)) | |
1642 | __io_free_req(req); | |
1643 | } | |
1644 | ||
978db57e JA |
1645 | static void io_double_put_req(struct io_kiocb *req) |
1646 | { | |
1647 | /* drop both submit and complete references */ | |
1648 | if (refcount_sub_and_test(2, &req->refs)) | |
1649 | io_free_req(req); | |
1650 | } | |
1651 | ||
1d7bb1d5 | 1652 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1653 | { |
84f97dc2 JA |
1654 | struct io_rings *rings = ctx->rings; |
1655 | ||
ad3eb2c8 JA |
1656 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1657 | /* | |
1658 | * noflush == true is from the waitqueue handler, just ensure | |
1659 | * we wake up the task, and the next invocation will flush the | |
1660 | * entries. We cannot safely to it from here. | |
1661 | */ | |
1662 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1663 | return -1U; | |
1d7bb1d5 | 1664 | |
ad3eb2c8 JA |
1665 | io_cqring_overflow_flush(ctx, false); |
1666 | } | |
1d7bb1d5 | 1667 | |
a3a0e43f JA |
1668 | /* See comment at the top of this file */ |
1669 | smp_rmb(); | |
ad3eb2c8 | 1670 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1671 | } |
1672 | ||
fb5ccc98 PB |
1673 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1674 | { | |
1675 | struct io_rings *rings = ctx->rings; | |
1676 | ||
1677 | /* make sure SQ entry isn't read before tail */ | |
1678 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1679 | } | |
1680 | ||
8237e045 | 1681 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1682 | { |
c6ca97b3 JA |
1683 | if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req)) |
1684 | return false; | |
e94f141b | 1685 | |
c6ca97b3 JA |
1686 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1687 | rb->need_iter++; | |
1688 | ||
1689 | rb->reqs[rb->to_free++] = req; | |
1690 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1691 | io_free_req_many(req->ctx, rb); | |
1692 | return true; | |
e94f141b JA |
1693 | } |
1694 | ||
bcda7baa JA |
1695 | static int io_put_kbuf(struct io_kiocb *req) |
1696 | { | |
4d954c25 | 1697 | struct io_buffer *kbuf; |
bcda7baa JA |
1698 | int cflags; |
1699 | ||
4d954c25 | 1700 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
bcda7baa JA |
1701 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
1702 | cflags |= IORING_CQE_F_BUFFER; | |
1703 | req->rw.addr = 0; | |
1704 | kfree(kbuf); | |
1705 | return cflags; | |
1706 | } | |
1707 | ||
def596e9 JA |
1708 | /* |
1709 | * Find and free completed poll iocbs | |
1710 | */ | |
1711 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1712 | struct list_head *done) | |
1713 | { | |
8237e045 | 1714 | struct req_batch rb; |
def596e9 | 1715 | struct io_kiocb *req; |
def596e9 | 1716 | |
c6ca97b3 | 1717 | rb.to_free = rb.need_iter = 0; |
def596e9 | 1718 | while (!list_empty(done)) { |
bcda7baa JA |
1719 | int cflags = 0; |
1720 | ||
def596e9 JA |
1721 | req = list_first_entry(done, struct io_kiocb, list); |
1722 | list_del(&req->list); | |
1723 | ||
bcda7baa JA |
1724 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1725 | cflags = io_put_kbuf(req); | |
1726 | ||
1727 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
1728 | (*nr_events)++; |
1729 | ||
8237e045 JA |
1730 | if (refcount_dec_and_test(&req->refs) && |
1731 | !io_req_multi_free(&rb, req)) | |
1732 | io_free_req(req); | |
def596e9 | 1733 | } |
def596e9 | 1734 | |
09bb8394 | 1735 | io_commit_cqring(ctx); |
32b2244a XW |
1736 | if (ctx->flags & IORING_SETUP_SQPOLL) |
1737 | io_cqring_ev_posted(ctx); | |
8237e045 | 1738 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1739 | } |
1740 | ||
1741 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1742 | long min) | |
1743 | { | |
1744 | struct io_kiocb *req, *tmp; | |
1745 | LIST_HEAD(done); | |
1746 | bool spin; | |
1747 | int ret; | |
1748 | ||
1749 | /* | |
1750 | * Only spin for completions if we don't have multiple devices hanging | |
1751 | * off our complete list, and we're under the requested amount. | |
1752 | */ | |
1753 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1754 | ||
1755 | ret = 0; | |
1756 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1757 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1758 | |
1759 | /* | |
1760 | * Move completed entries to our local list. If we find a | |
1761 | * request that requires polling, break out and complete | |
1762 | * the done list first, if we have entries there. | |
1763 | */ | |
1764 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1765 | list_move_tail(&req->list, &done); | |
1766 | continue; | |
1767 | } | |
1768 | if (!list_empty(&done)) | |
1769 | break; | |
1770 | ||
1771 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
1772 | if (ret < 0) | |
1773 | break; | |
1774 | ||
1775 | if (ret && spin) | |
1776 | spin = false; | |
1777 | ret = 0; | |
1778 | } | |
1779 | ||
1780 | if (!list_empty(&done)) | |
1781 | io_iopoll_complete(ctx, nr_events, &done); | |
1782 | ||
1783 | return ret; | |
1784 | } | |
1785 | ||
1786 | /* | |
d195a66e | 1787 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1788 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1789 | * as a non-spinning completion check. | |
1790 | */ | |
1791 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1792 | long min) | |
1793 | { | |
08f5439f | 1794 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1795 | int ret; |
1796 | ||
1797 | ret = io_do_iopoll(ctx, nr_events, min); | |
1798 | if (ret < 0) | |
1799 | return ret; | |
1800 | if (!min || *nr_events >= min) | |
1801 | return 0; | |
1802 | } | |
1803 | ||
1804 | return 1; | |
1805 | } | |
1806 | ||
1807 | /* | |
1808 | * We can't just wait for polled events to come to us, we have to actively | |
1809 | * find and complete them. | |
1810 | */ | |
1811 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1812 | { | |
1813 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1814 | return; | |
1815 | ||
1816 | mutex_lock(&ctx->uring_lock); | |
1817 | while (!list_empty(&ctx->poll_list)) { | |
1818 | unsigned int nr_events = 0; | |
1819 | ||
1820 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1821 | |
1822 | /* | |
1823 | * Ensure we allow local-to-the-cpu processing to take place, | |
1824 | * in this case we need to ensure that we reap all events. | |
1825 | */ | |
1826 | cond_resched(); | |
def596e9 JA |
1827 | } |
1828 | mutex_unlock(&ctx->uring_lock); | |
1829 | } | |
1830 | ||
c7849be9 XW |
1831 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1832 | long min) | |
def596e9 | 1833 | { |
2b2ed975 | 1834 | int iters = 0, ret = 0; |
500f9fba | 1835 | |
c7849be9 XW |
1836 | /* |
1837 | * We disallow the app entering submit/complete with polling, but we | |
1838 | * still need to lock the ring to prevent racing with polled issue | |
1839 | * that got punted to a workqueue. | |
1840 | */ | |
1841 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1842 | do { |
1843 | int tmin = 0; | |
1844 | ||
a3a0e43f JA |
1845 | /* |
1846 | * Don't enter poll loop if we already have events pending. | |
1847 | * If we do, we can potentially be spinning for commands that | |
1848 | * already triggered a CQE (eg in error). | |
1849 | */ | |
1d7bb1d5 | 1850 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1851 | break; |
1852 | ||
500f9fba JA |
1853 | /* |
1854 | * If a submit got punted to a workqueue, we can have the | |
1855 | * application entering polling for a command before it gets | |
1856 | * issued. That app will hold the uring_lock for the duration | |
1857 | * of the poll right here, so we need to take a breather every | |
1858 | * now and then to ensure that the issue has a chance to add | |
1859 | * the poll to the issued list. Otherwise we can spin here | |
1860 | * forever, while the workqueue is stuck trying to acquire the | |
1861 | * very same mutex. | |
1862 | */ | |
1863 | if (!(++iters & 7)) { | |
1864 | mutex_unlock(&ctx->uring_lock); | |
1865 | mutex_lock(&ctx->uring_lock); | |
1866 | } | |
1867 | ||
def596e9 JA |
1868 | if (*nr_events < min) |
1869 | tmin = min - *nr_events; | |
1870 | ||
1871 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1872 | if (ret <= 0) | |
1873 | break; | |
1874 | ret = 0; | |
1875 | } while (min && !*nr_events && !need_resched()); | |
1876 | ||
500f9fba | 1877 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1878 | return ret; |
1879 | } | |
1880 | ||
491381ce | 1881 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1882 | { |
491381ce JA |
1883 | /* |
1884 | * Tell lockdep we inherited freeze protection from submission | |
1885 | * thread. | |
1886 | */ | |
1887 | if (req->flags & REQ_F_ISREG) { | |
1888 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1889 | |
491381ce | 1890 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1891 | } |
491381ce | 1892 | file_end_write(req->file); |
2b188cc1 JA |
1893 | } |
1894 | ||
4e88d6e7 JA |
1895 | static inline void req_set_fail_links(struct io_kiocb *req) |
1896 | { | |
1897 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1898 | req->flags |= REQ_F_FAIL_LINK; | |
1899 | } | |
1900 | ||
ba816ad6 | 1901 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1902 | { |
9adbd45d | 1903 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
bcda7baa | 1904 | int cflags = 0; |
2b188cc1 | 1905 | |
491381ce JA |
1906 | if (kiocb->ki_flags & IOCB_WRITE) |
1907 | kiocb_end_write(req); | |
2b188cc1 | 1908 | |
4e88d6e7 JA |
1909 | if (res != req->result) |
1910 | req_set_fail_links(req); | |
bcda7baa JA |
1911 | if (req->flags & REQ_F_BUFFER_SELECTED) |
1912 | cflags = io_put_kbuf(req); | |
1913 | __io_cqring_add_event(req, res, cflags); | |
ba816ad6 JA |
1914 | } |
1915 | ||
1916 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1917 | { | |
9adbd45d | 1918 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1919 | |
1920 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1921 | io_put_req(req); |
2b188cc1 JA |
1922 | } |
1923 | ||
def596e9 JA |
1924 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1925 | { | |
9adbd45d | 1926 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1927 | |
491381ce JA |
1928 | if (kiocb->ki_flags & IOCB_WRITE) |
1929 | kiocb_end_write(req); | |
def596e9 | 1930 | |
4e88d6e7 JA |
1931 | if (res != req->result) |
1932 | req_set_fail_links(req); | |
9e645e11 | 1933 | req->result = res; |
def596e9 JA |
1934 | if (res != -EAGAIN) |
1935 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1936 | } | |
1937 | ||
1938 | /* | |
1939 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1940 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1941 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1942 | * accessing the kiocb cookie. | |
1943 | */ | |
1944 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1945 | { | |
1946 | struct io_ring_ctx *ctx = req->ctx; | |
1947 | ||
1948 | /* | |
1949 | * Track whether we have multiple files in our lists. This will impact | |
1950 | * how we do polling eventually, not spinning if we're on potentially | |
1951 | * different devices. | |
1952 | */ | |
1953 | if (list_empty(&ctx->poll_list)) { | |
1954 | ctx->poll_multi_file = false; | |
1955 | } else if (!ctx->poll_multi_file) { | |
1956 | struct io_kiocb *list_req; | |
1957 | ||
1958 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1959 | list); | |
9adbd45d | 1960 | if (list_req->file != req->file) |
def596e9 JA |
1961 | ctx->poll_multi_file = true; |
1962 | } | |
1963 | ||
1964 | /* | |
1965 | * For fast devices, IO may have already completed. If it has, add | |
1966 | * it to the front so we find it first. | |
1967 | */ | |
1968 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1969 | list_add(&req->list, &ctx->poll_list); | |
1970 | else | |
1971 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1972 | |
1973 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1974 | wq_has_sleeper(&ctx->sqo_wait)) | |
1975 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1976 | } |
1977 | ||
3d6770fb | 1978 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1979 | { |
3d6770fb | 1980 | if (state->file) { |
9a56a232 JA |
1981 | int diff = state->has_refs - state->used_refs; |
1982 | ||
1983 | if (diff) | |
1984 | fput_many(state->file, diff); | |
1985 | state->file = NULL; | |
1986 | } | |
1987 | } | |
1988 | ||
1989 | /* | |
1990 | * Get as many references to a file as we have IOs left in this submission, | |
1991 | * assuming most submissions are for one file, or at least that each file | |
1992 | * has more than one submission. | |
1993 | */ | |
8da11c19 | 1994 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
1995 | { |
1996 | if (!state) | |
1997 | return fget(fd); | |
1998 | ||
1999 | if (state->file) { | |
2000 | if (state->fd == fd) { | |
2001 | state->used_refs++; | |
2002 | state->ios_left--; | |
2003 | return state->file; | |
2004 | } | |
3d6770fb | 2005 | io_file_put(state); |
9a56a232 JA |
2006 | } |
2007 | state->file = fget_many(fd, state->ios_left); | |
2008 | if (!state->file) | |
2009 | return NULL; | |
2010 | ||
2011 | state->fd = fd; | |
2012 | state->has_refs = state->ios_left; | |
2013 | state->used_refs = 1; | |
2014 | state->ios_left--; | |
2015 | return state->file; | |
2016 | } | |
2017 | ||
2b188cc1 JA |
2018 | /* |
2019 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2020 | * any file. For now, just ensure that anything potentially problematic is done | |
2021 | * inline. | |
2022 | */ | |
2023 | static bool io_file_supports_async(struct file *file) | |
2024 | { | |
2025 | umode_t mode = file_inode(file)->i_mode; | |
2026 | ||
10d59345 | 2027 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
2028 | return true; |
2029 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
2030 | return true; | |
2031 | ||
2032 | return false; | |
2033 | } | |
2034 | ||
3529d8c2 JA |
2035 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2036 | bool force_nonblock) | |
2b188cc1 | 2037 | { |
def596e9 | 2038 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2039 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
2040 | unsigned ioprio; |
2041 | int ret; | |
2b188cc1 | 2042 | |
491381ce JA |
2043 | if (S_ISREG(file_inode(req->file)->i_mode)) |
2044 | req->flags |= REQ_F_ISREG; | |
2045 | ||
2b188cc1 | 2046 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
2047 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
2048 | req->flags |= REQ_F_CUR_POS; | |
2049 | kiocb->ki_pos = req->file->f_pos; | |
2050 | } | |
2b188cc1 | 2051 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2052 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2053 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2054 | if (unlikely(ret)) | |
2055 | return ret; | |
2b188cc1 JA |
2056 | |
2057 | ioprio = READ_ONCE(sqe->ioprio); | |
2058 | if (ioprio) { | |
2059 | ret = ioprio_check_cap(ioprio); | |
2060 | if (ret) | |
09bb8394 | 2061 | return ret; |
2b188cc1 JA |
2062 | |
2063 | kiocb->ki_ioprio = ioprio; | |
2064 | } else | |
2065 | kiocb->ki_ioprio = get_current_ioprio(); | |
2066 | ||
8449eeda | 2067 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
2068 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
2069 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
2070 | req->flags |= REQ_F_NOWAIT; |
2071 | ||
2072 | if (force_nonblock) | |
2b188cc1 | 2073 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 2074 | |
def596e9 | 2075 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2076 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2077 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2078 | return -EOPNOTSUPP; |
2b188cc1 | 2079 | |
def596e9 JA |
2080 | kiocb->ki_flags |= IOCB_HIPRI; |
2081 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 2082 | req->result = 0; |
def596e9 | 2083 | } else { |
09bb8394 JA |
2084 | if (kiocb->ki_flags & IOCB_HIPRI) |
2085 | return -EINVAL; | |
def596e9 JA |
2086 | kiocb->ki_complete = io_complete_rw; |
2087 | } | |
9adbd45d | 2088 | |
3529d8c2 JA |
2089 | req->rw.addr = READ_ONCE(sqe->addr); |
2090 | req->rw.len = READ_ONCE(sqe->len); | |
bcda7baa | 2091 | /* we own ->private, reuse it for the buffer index / buffer ID */ |
9adbd45d | 2092 | req->rw.kiocb.private = (void *) (unsigned long) |
3529d8c2 | 2093 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 2094 | return 0; |
2b188cc1 JA |
2095 | } |
2096 | ||
2097 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2098 | { | |
2099 | switch (ret) { | |
2100 | case -EIOCBQUEUED: | |
2101 | break; | |
2102 | case -ERESTARTSYS: | |
2103 | case -ERESTARTNOINTR: | |
2104 | case -ERESTARTNOHAND: | |
2105 | case -ERESTART_RESTARTBLOCK: | |
2106 | /* | |
2107 | * We can't just restart the syscall, since previously | |
2108 | * submitted sqes may already be in progress. Just fail this | |
2109 | * IO with EINTR. | |
2110 | */ | |
2111 | ret = -EINTR; | |
2112 | /* fall through */ | |
2113 | default: | |
2114 | kiocb->ki_complete(kiocb, ret, 0); | |
2115 | } | |
2116 | } | |
2117 | ||
014db007 | 2118 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret) |
ba816ad6 | 2119 | { |
ba04291e JA |
2120 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2121 | ||
2122 | if (req->flags & REQ_F_CUR_POS) | |
2123 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2124 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
014db007 | 2125 | io_complete_rw(kiocb, ret, 0); |
ba816ad6 JA |
2126 | else |
2127 | io_rw_done(kiocb, ret); | |
2128 | } | |
2129 | ||
9adbd45d | 2130 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2131 | struct iov_iter *iter) |
edafccee | 2132 | { |
9adbd45d JA |
2133 | struct io_ring_ctx *ctx = req->ctx; |
2134 | size_t len = req->rw.len; | |
edafccee JA |
2135 | struct io_mapped_ubuf *imu; |
2136 | unsigned index, buf_index; | |
2137 | size_t offset; | |
2138 | u64 buf_addr; | |
2139 | ||
2140 | /* attempt to use fixed buffers without having provided iovecs */ | |
2141 | if (unlikely(!ctx->user_bufs)) | |
2142 | return -EFAULT; | |
2143 | ||
9adbd45d | 2144 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2145 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2146 | return -EFAULT; | |
2147 | ||
2148 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2149 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2150 | buf_addr = req->rw.addr; |
edafccee JA |
2151 | |
2152 | /* overflow */ | |
2153 | if (buf_addr + len < buf_addr) | |
2154 | return -EFAULT; | |
2155 | /* not inside the mapped region */ | |
2156 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2157 | return -EFAULT; | |
2158 | ||
2159 | /* | |
2160 | * May not be a start of buffer, set size appropriately | |
2161 | * and advance us to the beginning. | |
2162 | */ | |
2163 | offset = buf_addr - imu->ubuf; | |
2164 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2165 | |
2166 | if (offset) { | |
2167 | /* | |
2168 | * Don't use iov_iter_advance() here, as it's really slow for | |
2169 | * using the latter parts of a big fixed buffer - it iterates | |
2170 | * over each segment manually. We can cheat a bit here, because | |
2171 | * we know that: | |
2172 | * | |
2173 | * 1) it's a BVEC iter, we set it up | |
2174 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2175 | * first and last bvec | |
2176 | * | |
2177 | * So just find our index, and adjust the iterator afterwards. | |
2178 | * If the offset is within the first bvec (or the whole first | |
2179 | * bvec, just use iov_iter_advance(). This makes it easier | |
2180 | * since we can just skip the first segment, which may not | |
2181 | * be PAGE_SIZE aligned. | |
2182 | */ | |
2183 | const struct bio_vec *bvec = imu->bvec; | |
2184 | ||
2185 | if (offset <= bvec->bv_len) { | |
2186 | iov_iter_advance(iter, offset); | |
2187 | } else { | |
2188 | unsigned long seg_skip; | |
2189 | ||
2190 | /* skip first vec */ | |
2191 | offset -= bvec->bv_len; | |
2192 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2193 | ||
2194 | iter->bvec = bvec + seg_skip; | |
2195 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2196 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2197 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2198 | } |
2199 | } | |
2200 | ||
5e559561 | 2201 | return len; |
edafccee JA |
2202 | } |
2203 | ||
bcda7baa JA |
2204 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2205 | { | |
2206 | if (needs_lock) | |
2207 | mutex_unlock(&ctx->uring_lock); | |
2208 | } | |
2209 | ||
2210 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2211 | { | |
2212 | /* | |
2213 | * "Normal" inline submissions always hold the uring_lock, since we | |
2214 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2215 | * The only exception is when we've detached the request and issue it | |
2216 | * from an async worker thread, grab the lock for that case. | |
2217 | */ | |
2218 | if (needs_lock) | |
2219 | mutex_lock(&ctx->uring_lock); | |
2220 | } | |
2221 | ||
2222 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2223 | int bgid, struct io_buffer *kbuf, | |
2224 | bool needs_lock) | |
2225 | { | |
2226 | struct io_buffer *head; | |
2227 | ||
2228 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2229 | return kbuf; | |
2230 | ||
2231 | io_ring_submit_lock(req->ctx, needs_lock); | |
2232 | ||
2233 | lockdep_assert_held(&req->ctx->uring_lock); | |
2234 | ||
2235 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2236 | if (head) { | |
2237 | if (!list_empty(&head->list)) { | |
2238 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2239 | list); | |
2240 | list_del(&kbuf->list); | |
2241 | } else { | |
2242 | kbuf = head; | |
2243 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2244 | } | |
2245 | if (*len > kbuf->len) | |
2246 | *len = kbuf->len; | |
2247 | } else { | |
2248 | kbuf = ERR_PTR(-ENOBUFS); | |
2249 | } | |
2250 | ||
2251 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2252 | ||
2253 | return kbuf; | |
2254 | } | |
2255 | ||
4d954c25 JA |
2256 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2257 | bool needs_lock) | |
2258 | { | |
2259 | struct io_buffer *kbuf; | |
2260 | int bgid; | |
2261 | ||
2262 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2263 | bgid = (int) (unsigned long) req->rw.kiocb.private; | |
2264 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); | |
2265 | if (IS_ERR(kbuf)) | |
2266 | return kbuf; | |
2267 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2268 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2269 | return u64_to_user_ptr(kbuf->addr); | |
2270 | } | |
2271 | ||
2272 | #ifdef CONFIG_COMPAT | |
2273 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2274 | bool needs_lock) | |
2275 | { | |
2276 | struct compat_iovec __user *uiov; | |
2277 | compat_ssize_t clen; | |
2278 | void __user *buf; | |
2279 | ssize_t len; | |
2280 | ||
2281 | uiov = u64_to_user_ptr(req->rw.addr); | |
2282 | if (!access_ok(uiov, sizeof(*uiov))) | |
2283 | return -EFAULT; | |
2284 | if (__get_user(clen, &uiov->iov_len)) | |
2285 | return -EFAULT; | |
2286 | if (clen < 0) | |
2287 | return -EINVAL; | |
2288 | ||
2289 | len = clen; | |
2290 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2291 | if (IS_ERR(buf)) | |
2292 | return PTR_ERR(buf); | |
2293 | iov[0].iov_base = buf; | |
2294 | iov[0].iov_len = (compat_size_t) len; | |
2295 | return 0; | |
2296 | } | |
2297 | #endif | |
2298 | ||
2299 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2300 | bool needs_lock) | |
2301 | { | |
2302 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2303 | void __user *buf; | |
2304 | ssize_t len; | |
2305 | ||
2306 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2307 | return -EFAULT; | |
2308 | ||
2309 | len = iov[0].iov_len; | |
2310 | if (len < 0) | |
2311 | return -EINVAL; | |
2312 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2313 | if (IS_ERR(buf)) | |
2314 | return PTR_ERR(buf); | |
2315 | iov[0].iov_base = buf; | |
2316 | iov[0].iov_len = len; | |
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2321 | bool needs_lock) | |
2322 | { | |
2323 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2324 | return 0; | |
2325 | if (!req->rw.len) | |
2326 | return 0; | |
2327 | else if (req->rw.len > 1) | |
2328 | return -EINVAL; | |
2329 | ||
2330 | #ifdef CONFIG_COMPAT | |
2331 | if (req->ctx->compat) | |
2332 | return io_compat_import(req, iov, needs_lock); | |
2333 | #endif | |
2334 | ||
2335 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2336 | } | |
2337 | ||
cf6fd4bd | 2338 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
bcda7baa JA |
2339 | struct iovec **iovec, struct iov_iter *iter, |
2340 | bool needs_lock) | |
2b188cc1 | 2341 | { |
9adbd45d JA |
2342 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2343 | size_t sqe_len = req->rw.len; | |
4d954c25 | 2344 | ssize_t ret; |
edafccee JA |
2345 | u8 opcode; |
2346 | ||
d625c6ee | 2347 | opcode = req->opcode; |
7d009165 | 2348 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2349 | *iovec = NULL; |
9adbd45d | 2350 | return io_import_fixed(req, rw, iter); |
edafccee | 2351 | } |
2b188cc1 | 2352 | |
bcda7baa JA |
2353 | /* buffer index only valid with fixed read/write, or buffer select */ |
2354 | if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT)) | |
9adbd45d JA |
2355 | return -EINVAL; |
2356 | ||
3a6820f2 | 2357 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2358 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 JA |
2359 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
2360 | if (IS_ERR(buf)) { | |
bcda7baa | 2361 | *iovec = NULL; |
4d954c25 | 2362 | return PTR_ERR(buf); |
bcda7baa | 2363 | } |
3f9d6441 | 2364 | req->rw.len = sqe_len; |
bcda7baa JA |
2365 | } |
2366 | ||
3a6820f2 JA |
2367 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2368 | *iovec = NULL; | |
3a901598 | 2369 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2370 | } |
2371 | ||
f67676d1 JA |
2372 | if (req->io) { |
2373 | struct io_async_rw *iorw = &req->io->rw; | |
2374 | ||
2375 | *iovec = iorw->iov; | |
2376 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2377 | if (iorw->iov == iorw->fast_iov) | |
2378 | *iovec = NULL; | |
2379 | return iorw->size; | |
2380 | } | |
2381 | ||
4d954c25 JA |
2382 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2383 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
3f9d6441 JA |
2384 | if (!ret) { |
2385 | ret = (*iovec)->iov_len; | |
2386 | iov_iter_init(iter, rw, *iovec, 1, ret); | |
2387 | } | |
4d954c25 JA |
2388 | *iovec = NULL; |
2389 | return ret; | |
2390 | } | |
2391 | ||
2b188cc1 | 2392 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2393 | if (req->ctx->compat) |
2b188cc1 JA |
2394 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2395 | iovec, iter); | |
2396 | #endif | |
2397 | ||
2398 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2399 | } | |
2400 | ||
31b51510 | 2401 | /* |
32960613 JA |
2402 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2403 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2404 | */ |
32960613 JA |
2405 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2406 | struct iov_iter *iter) | |
2407 | { | |
2408 | ssize_t ret = 0; | |
2409 | ||
2410 | /* | |
2411 | * Don't support polled IO through this interface, and we can't | |
2412 | * support non-blocking either. For the latter, this just causes | |
2413 | * the kiocb to be handled from an async context. | |
2414 | */ | |
2415 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2416 | return -EOPNOTSUPP; | |
2417 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2418 | return -EAGAIN; | |
2419 | ||
2420 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2421 | struct iovec iovec; |
32960613 JA |
2422 | ssize_t nr; |
2423 | ||
311ae9e1 PB |
2424 | if (!iov_iter_is_bvec(iter)) { |
2425 | iovec = iov_iter_iovec(iter); | |
2426 | } else { | |
2427 | /* fixed buffers import bvec */ | |
2428 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2429 | + iter->iov_offset; | |
2430 | iovec.iov_len = min(iter->count, | |
2431 | iter->bvec->bv_len - iter->iov_offset); | |
2432 | } | |
2433 | ||
32960613 JA |
2434 | if (rw == READ) { |
2435 | nr = file->f_op->read(file, iovec.iov_base, | |
2436 | iovec.iov_len, &kiocb->ki_pos); | |
2437 | } else { | |
2438 | nr = file->f_op->write(file, iovec.iov_base, | |
2439 | iovec.iov_len, &kiocb->ki_pos); | |
2440 | } | |
2441 | ||
311ae9e1 PB |
2442 | if (iov_iter_is_bvec(iter)) |
2443 | kunmap(iter->bvec->bv_page); | |
2444 | ||
32960613 JA |
2445 | if (nr < 0) { |
2446 | if (!ret) | |
2447 | ret = nr; | |
2448 | break; | |
2449 | } | |
2450 | ret += nr; | |
2451 | if (nr != iovec.iov_len) | |
2452 | break; | |
2453 | iov_iter_advance(iter, nr); | |
2454 | } | |
2455 | ||
2456 | return ret; | |
2457 | } | |
2458 | ||
b7bb4f7d | 2459 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2460 | struct iovec *iovec, struct iovec *fast_iov, |
2461 | struct iov_iter *iter) | |
2462 | { | |
2463 | req->io->rw.nr_segs = iter->nr_segs; | |
2464 | req->io->rw.size = io_size; | |
2465 | req->io->rw.iov = iovec; | |
2466 | if (!req->io->rw.iov) { | |
2467 | req->io->rw.iov = req->io->rw.fast_iov; | |
2468 | memcpy(req->io->rw.iov, fast_iov, | |
2469 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2470 | } else { |
2471 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2472 | } |
2473 | } | |
2474 | ||
3d9932a8 XW |
2475 | static inline int __io_alloc_async_ctx(struct io_kiocb *req) |
2476 | { | |
2477 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); | |
2478 | return req->io == NULL; | |
2479 | } | |
2480 | ||
b7bb4f7d | 2481 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2482 | { |
d3656344 JA |
2483 | if (!io_op_defs[req->opcode].async_ctx) |
2484 | return 0; | |
3d9932a8 XW |
2485 | |
2486 | return __io_alloc_async_ctx(req); | |
b7bb4f7d JA |
2487 | } |
2488 | ||
b7bb4f7d JA |
2489 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2490 | struct iovec *iovec, struct iovec *fast_iov, | |
2491 | struct iov_iter *iter) | |
2492 | { | |
980ad263 | 2493 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2494 | return 0; |
5d204bcf | 2495 | if (!req->io) { |
3d9932a8 | 2496 | if (__io_alloc_async_ctx(req)) |
5d204bcf | 2497 | return -ENOMEM; |
b7bb4f7d | 2498 | |
5d204bcf JA |
2499 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2500 | } | |
b7bb4f7d | 2501 | return 0; |
f67676d1 JA |
2502 | } |
2503 | ||
3529d8c2 JA |
2504 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2505 | bool force_nonblock) | |
f67676d1 | 2506 | { |
3529d8c2 JA |
2507 | struct io_async_ctx *io; |
2508 | struct iov_iter iter; | |
f67676d1 JA |
2509 | ssize_t ret; |
2510 | ||
3529d8c2 JA |
2511 | ret = io_prep_rw(req, sqe, force_nonblock); |
2512 | if (ret) | |
2513 | return ret; | |
f67676d1 | 2514 | |
3529d8c2 JA |
2515 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2516 | return -EBADF; | |
f67676d1 | 2517 | |
5f798bea PB |
2518 | /* either don't need iovec imported or already have it */ |
2519 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2520 | return 0; |
2521 | ||
2522 | io = req->io; | |
2523 | io->rw.iov = io->rw.fast_iov; | |
2524 | req->io = NULL; | |
bcda7baa | 2525 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2526 | req->io = io; |
2527 | if (ret < 0) | |
2528 | return ret; | |
2529 | ||
2530 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2531 | return 0; | |
f67676d1 JA |
2532 | } |
2533 | ||
014db007 | 2534 | static int io_read(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2535 | { |
2536 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2537 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2538 | struct iov_iter iter; |
31b51510 | 2539 | size_t iov_count; |
f67676d1 | 2540 | ssize_t io_size, ret; |
2b188cc1 | 2541 | |
bcda7baa | 2542 | ret = io_import_iovec(READ, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2543 | if (ret < 0) |
2544 | return ret; | |
2b188cc1 | 2545 | |
fd6c2e4c JA |
2546 | /* Ensure we clear previously set non-block flag */ |
2547 | if (!force_nonblock) | |
29de5f6a | 2548 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2549 | |
797f3f53 | 2550 | req->result = 0; |
f67676d1 | 2551 | io_size = ret; |
9e645e11 | 2552 | if (req->flags & REQ_F_LINK) |
f67676d1 JA |
2553 | req->result = io_size; |
2554 | ||
2555 | /* | |
2556 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2557 | * we know to async punt it even if it was opened O_NONBLOCK | |
2558 | */ | |
29de5f6a | 2559 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2560 | goto copy_iov; |
9e645e11 | 2561 | |
31b51510 | 2562 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2563 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2564 | if (!ret) { |
2565 | ssize_t ret2; | |
2566 | ||
9adbd45d JA |
2567 | if (req->file->f_op->read_iter) |
2568 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2569 | else |
9adbd45d | 2570 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2571 | |
9d93a3f5 | 2572 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2573 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2574 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2575 | } else { |
2576 | copy_iov: | |
b7bb4f7d | 2577 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2578 | inline_vecs, &iter); |
2579 | if (ret) | |
2580 | goto out_free; | |
29de5f6a JA |
2581 | /* any defer here is final, must blocking retry */ |
2582 | if (!(req->flags & REQ_F_NOWAIT)) | |
2583 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2584 | return -EAGAIN; |
2585 | } | |
2b188cc1 | 2586 | } |
f67676d1 | 2587 | out_free: |
1e95081c | 2588 | kfree(iovec); |
99bc4c38 | 2589 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2590 | return ret; |
2591 | } | |
2592 | ||
3529d8c2 JA |
2593 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2594 | bool force_nonblock) | |
f67676d1 | 2595 | { |
3529d8c2 JA |
2596 | struct io_async_ctx *io; |
2597 | struct iov_iter iter; | |
f67676d1 JA |
2598 | ssize_t ret; |
2599 | ||
3529d8c2 JA |
2600 | ret = io_prep_rw(req, sqe, force_nonblock); |
2601 | if (ret) | |
2602 | return ret; | |
f67676d1 | 2603 | |
3529d8c2 JA |
2604 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2605 | return -EBADF; | |
f67676d1 | 2606 | |
4ed734b0 JA |
2607 | req->fsize = rlimit(RLIMIT_FSIZE); |
2608 | ||
5f798bea PB |
2609 | /* either don't need iovec imported or already have it */ |
2610 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2611 | return 0; |
2612 | ||
2613 | io = req->io; | |
2614 | io->rw.iov = io->rw.fast_iov; | |
2615 | req->io = NULL; | |
bcda7baa | 2616 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter, !force_nonblock); |
3529d8c2 JA |
2617 | req->io = io; |
2618 | if (ret < 0) | |
2619 | return ret; | |
2620 | ||
2621 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2622 | return 0; | |
f67676d1 JA |
2623 | } |
2624 | ||
014db007 | 2625 | static int io_write(struct io_kiocb *req, bool force_nonblock) |
2b188cc1 JA |
2626 | { |
2627 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2628 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2629 | struct iov_iter iter; |
31b51510 | 2630 | size_t iov_count; |
f67676d1 | 2631 | ssize_t ret, io_size; |
2b188cc1 | 2632 | |
bcda7baa | 2633 | ret = io_import_iovec(WRITE, req, &iovec, &iter, !force_nonblock); |
06b76d44 JA |
2634 | if (ret < 0) |
2635 | return ret; | |
2b188cc1 | 2636 | |
fd6c2e4c JA |
2637 | /* Ensure we clear previously set non-block flag */ |
2638 | if (!force_nonblock) | |
9adbd45d | 2639 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2640 | |
797f3f53 | 2641 | req->result = 0; |
f67676d1 | 2642 | io_size = ret; |
9e645e11 | 2643 | if (req->flags & REQ_F_LINK) |
f67676d1 | 2644 | req->result = io_size; |
9e645e11 | 2645 | |
f67676d1 JA |
2646 | /* |
2647 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2648 | * we know to async punt it even if it was opened O_NONBLOCK | |
2649 | */ | |
29de5f6a | 2650 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2651 | goto copy_iov; |
31b51510 | 2652 | |
10d59345 JA |
2653 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2654 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2655 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2656 | goto copy_iov; |
31b51510 | 2657 | |
f67676d1 | 2658 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2659 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2660 | if (!ret) { |
9bf7933f RP |
2661 | ssize_t ret2; |
2662 | ||
2b188cc1 JA |
2663 | /* |
2664 | * Open-code file_start_write here to grab freeze protection, | |
2665 | * which will be released by another thread in | |
2666 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2667 | * released so that it doesn't complain about the held lock when | |
2668 | * we return to userspace. | |
2669 | */ | |
491381ce | 2670 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2671 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2672 | SB_FREEZE_WRITE, true); |
9adbd45d | 2673 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2674 | SB_FREEZE_WRITE); |
2675 | } | |
2676 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2677 | |
4ed734b0 JA |
2678 | if (!force_nonblock) |
2679 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; | |
2680 | ||
9adbd45d JA |
2681 | if (req->file->f_op->write_iter) |
2682 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2683 | else |
9adbd45d | 2684 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
4ed734b0 JA |
2685 | |
2686 | if (!force_nonblock) | |
2687 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
2688 | ||
faac996c | 2689 | /* |
bff6035d | 2690 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just |
faac996c JA |
2691 | * retry them without IOCB_NOWAIT. |
2692 | */ | |
2693 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2694 | ret2 = -EAGAIN; | |
f67676d1 | 2695 | if (!force_nonblock || ret2 != -EAGAIN) { |
014db007 | 2696 | kiocb_done(kiocb, ret2); |
f67676d1 JA |
2697 | } else { |
2698 | copy_iov: | |
b7bb4f7d | 2699 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2700 | inline_vecs, &iter); |
2701 | if (ret) | |
2702 | goto out_free; | |
29de5f6a JA |
2703 | /* any defer here is final, must blocking retry */ |
2704 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2705 | return -EAGAIN; |
2706 | } | |
2b188cc1 | 2707 | } |
31b51510 | 2708 | out_free: |
99bc4c38 | 2709 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2710 | kfree(iovec); |
2b188cc1 JA |
2711 | return ret; |
2712 | } | |
2713 | ||
7d67af2c PB |
2714 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2715 | { | |
2716 | struct io_splice* sp = &req->splice; | |
2717 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2718 | int ret; | |
2719 | ||
2720 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2721 | return 0; | |
2722 | ||
2723 | sp->file_in = NULL; | |
2724 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2725 | sp->off_out = READ_ONCE(sqe->off); | |
2726 | sp->len = READ_ONCE(sqe->len); | |
2727 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2728 | ||
2729 | if (unlikely(sp->flags & ~valid_flags)) | |
2730 | return -EINVAL; | |
2731 | ||
2732 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2733 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2734 | if (ret) | |
2735 | return ret; | |
2736 | req->flags |= REQ_F_NEED_CLEANUP; | |
2737 | ||
2738 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2739 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2740 | ||
2741 | return 0; | |
2742 | } | |
2743 | ||
2744 | static bool io_splice_punt(struct file *file) | |
2745 | { | |
2746 | if (get_pipe_info(file)) | |
2747 | return false; | |
2748 | if (!io_file_supports_async(file)) | |
2749 | return true; | |
2750 | return !(file->f_mode & O_NONBLOCK); | |
2751 | } | |
2752 | ||
014db007 | 2753 | static int io_splice(struct io_kiocb *req, bool force_nonblock) |
7d67af2c PB |
2754 | { |
2755 | struct io_splice *sp = &req->splice; | |
2756 | struct file *in = sp->file_in; | |
2757 | struct file *out = sp->file_out; | |
2758 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2759 | loff_t *poff_in, *poff_out; | |
2760 | long ret; | |
2761 | ||
2762 | if (force_nonblock) { | |
2763 | if (io_splice_punt(in) || io_splice_punt(out)) | |
2764 | return -EAGAIN; | |
2765 | flags |= SPLICE_F_NONBLOCK; | |
2766 | } | |
2767 | ||
2768 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2769 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2770 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2771 | if (force_nonblock && ret == -EAGAIN) | |
2772 | return -EAGAIN; | |
2773 | ||
2774 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2775 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2776 | ||
2777 | io_cqring_add_event(req, ret); | |
2778 | if (ret != sp->len) | |
2779 | req_set_fail_links(req); | |
014db007 | 2780 | io_put_req(req); |
7d67af2c PB |
2781 | return 0; |
2782 | } | |
2783 | ||
2b188cc1 JA |
2784 | /* |
2785 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2786 | */ | |
78e19bbe | 2787 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2788 | { |
2789 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2790 | |
def596e9 JA |
2791 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2792 | return -EINVAL; | |
2793 | ||
78e19bbe | 2794 | io_cqring_add_event(req, 0); |
e65ef56d | 2795 | io_put_req(req); |
2b188cc1 JA |
2796 | return 0; |
2797 | } | |
2798 | ||
3529d8c2 | 2799 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2800 | { |
6b06314c | 2801 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2802 | |
09bb8394 JA |
2803 | if (!req->file) |
2804 | return -EBADF; | |
c992fe29 | 2805 | |
6b06314c | 2806 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2807 | return -EINVAL; |
edafccee | 2808 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2809 | return -EINVAL; |
2810 | ||
8ed8d3c3 JA |
2811 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2812 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2813 | return -EINVAL; | |
2814 | ||
2815 | req->sync.off = READ_ONCE(sqe->off); | |
2816 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2817 | return 0; |
2818 | } | |
2819 | ||
8ed8d3c3 JA |
2820 | static bool io_req_cancelled(struct io_kiocb *req) |
2821 | { | |
2822 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2823 | req_set_fail_links(req); | |
2824 | io_cqring_add_event(req, -ECANCELED); | |
2825 | io_put_req(req); | |
2826 | return true; | |
2827 | } | |
2828 | ||
2829 | return false; | |
2830 | } | |
2831 | ||
014db007 | 2832 | static void __io_fsync(struct io_kiocb *req) |
8ed8d3c3 | 2833 | { |
8ed8d3c3 | 2834 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2835 | int ret; |
2836 | ||
9adbd45d | 2837 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2838 | end > 0 ? end : LLONG_MAX, |
2839 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2840 | if (ret < 0) | |
2841 | req_set_fail_links(req); | |
2842 | io_cqring_add_event(req, ret); | |
014db007 | 2843 | io_put_req(req); |
8ed8d3c3 JA |
2844 | } |
2845 | ||
5ea62161 | 2846 | static void io_fsync_finish(struct io_wq_work **workptr) |
c992fe29 | 2847 | { |
5ea62161 | 2848 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); |
5ea62161 PB |
2849 | |
2850 | if (io_req_cancelled(req)) | |
2851 | return; | |
014db007 | 2852 | __io_fsync(req); |
e9fd9396 | 2853 | io_steal_work(req, workptr); |
8ed8d3c3 | 2854 | } |
c992fe29 | 2855 | |
014db007 | 2856 | static int io_fsync(struct io_kiocb *req, bool force_nonblock) |
c992fe29 | 2857 | { |
c992fe29 | 2858 | /* fsync always requires a blocking context */ |
8ed8d3c3 | 2859 | if (force_nonblock) { |
8ed8d3c3 | 2860 | req->work.func = io_fsync_finish; |
c992fe29 | 2861 | return -EAGAIN; |
8ed8d3c3 | 2862 | } |
014db007 | 2863 | __io_fsync(req); |
c992fe29 CH |
2864 | return 0; |
2865 | } | |
2866 | ||
014db007 | 2867 | static void __io_fallocate(struct io_kiocb *req) |
8ed8d3c3 | 2868 | { |
8ed8d3c3 JA |
2869 | int ret; |
2870 | ||
4ed734b0 | 2871 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize; |
d63d1b5e JA |
2872 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2873 | req->sync.len); | |
4ed734b0 | 2874 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; |
8ed8d3c3 JA |
2875 | if (ret < 0) |
2876 | req_set_fail_links(req); | |
2877 | io_cqring_add_event(req, ret); | |
014db007 | 2878 | io_put_req(req); |
5ea62161 PB |
2879 | } |
2880 | ||
2881 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2882 | { | |
2883 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
5ea62161 | 2884 | |
594506fe PB |
2885 | if (io_req_cancelled(req)) |
2886 | return; | |
014db007 | 2887 | __io_fallocate(req); |
e9fd9396 | 2888 | io_steal_work(req, workptr); |
5d17b4a4 JA |
2889 | } |
2890 | ||
d63d1b5e JA |
2891 | static int io_fallocate_prep(struct io_kiocb *req, |
2892 | const struct io_uring_sqe *sqe) | |
2893 | { | |
2894 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2895 | return -EINVAL; | |
2896 | ||
2897 | req->sync.off = READ_ONCE(sqe->off); | |
2898 | req->sync.len = READ_ONCE(sqe->addr); | |
2899 | req->sync.mode = READ_ONCE(sqe->len); | |
4ed734b0 | 2900 | req->fsize = rlimit(RLIMIT_FSIZE); |
d63d1b5e JA |
2901 | return 0; |
2902 | } | |
2903 | ||
014db007 | 2904 | static int io_fallocate(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 2905 | { |
d63d1b5e | 2906 | /* fallocate always requiring blocking context */ |
8ed8d3c3 | 2907 | if (force_nonblock) { |
d63d1b5e | 2908 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2909 | return -EAGAIN; |
8ed8d3c3 | 2910 | } |
5d17b4a4 | 2911 | |
014db007 | 2912 | __io_fallocate(req); |
5d17b4a4 JA |
2913 | return 0; |
2914 | } | |
2915 | ||
15b71abe | 2916 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2917 | { |
f8748881 | 2918 | const char __user *fname; |
15b71abe | 2919 | int ret; |
b7bb4f7d | 2920 | |
15b71abe JA |
2921 | if (sqe->ioprio || sqe->buf_index) |
2922 | return -EINVAL; | |
cf3040ca JA |
2923 | if (sqe->flags & IOSQE_FIXED_FILE) |
2924 | return -EBADF; | |
0bdbdd08 PB |
2925 | if (req->flags & REQ_F_NEED_CLEANUP) |
2926 | return 0; | |
03b1230c | 2927 | |
15b71abe | 2928 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2929 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2930 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2931 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
3529d8c2 | 2932 | |
f8748881 | 2933 | req->open.filename = getname(fname); |
15b71abe JA |
2934 | if (IS_ERR(req->open.filename)) { |
2935 | ret = PTR_ERR(req->open.filename); | |
2936 | req->open.filename = NULL; | |
2937 | return ret; | |
2938 | } | |
3529d8c2 | 2939 | |
4022e7af | 2940 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2941 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2942 | return 0; |
03b1230c JA |
2943 | } |
2944 | ||
cebdb986 | 2945 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2946 | { |
cebdb986 JA |
2947 | struct open_how __user *how; |
2948 | const char __user *fname; | |
2949 | size_t len; | |
0fa03c62 JA |
2950 | int ret; |
2951 | ||
cebdb986 | 2952 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2953 | return -EINVAL; |
cf3040ca JA |
2954 | if (sqe->flags & IOSQE_FIXED_FILE) |
2955 | return -EBADF; | |
0bdbdd08 PB |
2956 | if (req->flags & REQ_F_NEED_CLEANUP) |
2957 | return 0; | |
0fa03c62 | 2958 | |
cebdb986 JA |
2959 | req->open.dfd = READ_ONCE(sqe->fd); |
2960 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2961 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2962 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2963 | |
cebdb986 JA |
2964 | if (len < OPEN_HOW_SIZE_VER0) |
2965 | return -EINVAL; | |
3529d8c2 | 2966 | |
cebdb986 JA |
2967 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2968 | len); | |
2969 | if (ret) | |
2970 | return ret; | |
3529d8c2 | 2971 | |
cebdb986 JA |
2972 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2973 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2974 | |
cebdb986 JA |
2975 | req->open.filename = getname(fname); |
2976 | if (IS_ERR(req->open.filename)) { | |
2977 | ret = PTR_ERR(req->open.filename); | |
2978 | req->open.filename = NULL; | |
2979 | return ret; | |
2980 | } | |
2981 | ||
4022e7af | 2982 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 2983 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2984 | return 0; |
2985 | } | |
2986 | ||
014db007 | 2987 | static int io_openat2(struct io_kiocb *req, bool force_nonblock) |
15b71abe JA |
2988 | { |
2989 | struct open_flags op; | |
15b71abe JA |
2990 | struct file *file; |
2991 | int ret; | |
2992 | ||
f86cd20c | 2993 | if (force_nonblock) |
15b71abe | 2994 | return -EAGAIN; |
15b71abe | 2995 | |
cebdb986 | 2996 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
2997 | if (ret) |
2998 | goto err; | |
2999 | ||
4022e7af | 3000 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3001 | if (ret < 0) |
3002 | goto err; | |
3003 | ||
3004 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3005 | if (IS_ERR(file)) { | |
3006 | put_unused_fd(ret); | |
3007 | ret = PTR_ERR(file); | |
3008 | } else { | |
3009 | fsnotify_open(file); | |
3010 | fd_install(ret, file); | |
3011 | } | |
3012 | err: | |
3013 | putname(req->open.filename); | |
8fef80bf | 3014 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3015 | if (ret < 0) |
3016 | req_set_fail_links(req); | |
3017 | io_cqring_add_event(req, ret); | |
014db007 | 3018 | io_put_req(req); |
15b71abe JA |
3019 | return 0; |
3020 | } | |
3021 | ||
014db007 | 3022 | static int io_openat(struct io_kiocb *req, bool force_nonblock) |
cebdb986 JA |
3023 | { |
3024 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
014db007 | 3025 | return io_openat2(req, force_nonblock); |
cebdb986 JA |
3026 | } |
3027 | ||
067524e9 JA |
3028 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3029 | const struct io_uring_sqe *sqe) | |
3030 | { | |
3031 | struct io_provide_buf *p = &req->pbuf; | |
3032 | u64 tmp; | |
3033 | ||
3034 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3035 | return -EINVAL; | |
3036 | ||
3037 | tmp = READ_ONCE(sqe->fd); | |
3038 | if (!tmp || tmp > USHRT_MAX) | |
3039 | return -EINVAL; | |
3040 | ||
3041 | memset(p, 0, sizeof(*p)); | |
3042 | p->nbufs = tmp; | |
3043 | p->bgid = READ_ONCE(sqe->buf_group); | |
3044 | return 0; | |
3045 | } | |
3046 | ||
3047 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3048 | int bgid, unsigned nbufs) | |
3049 | { | |
3050 | unsigned i = 0; | |
3051 | ||
3052 | /* shouldn't happen */ | |
3053 | if (!nbufs) | |
3054 | return 0; | |
3055 | ||
3056 | /* the head kbuf is the list itself */ | |
3057 | while (!list_empty(&buf->list)) { | |
3058 | struct io_buffer *nxt; | |
3059 | ||
3060 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3061 | list_del(&nxt->list); | |
3062 | kfree(nxt); | |
3063 | if (++i == nbufs) | |
3064 | return i; | |
3065 | } | |
3066 | i++; | |
3067 | kfree(buf); | |
3068 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3069 | ||
3070 | return i; | |
3071 | } | |
3072 | ||
3073 | static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock) | |
3074 | { | |
3075 | struct io_provide_buf *p = &req->pbuf; | |
3076 | struct io_ring_ctx *ctx = req->ctx; | |
3077 | struct io_buffer *head; | |
3078 | int ret = 0; | |
3079 | ||
3080 | io_ring_submit_lock(ctx, !force_nonblock); | |
3081 | ||
3082 | lockdep_assert_held(&ctx->uring_lock); | |
3083 | ||
3084 | ret = -ENOENT; | |
3085 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3086 | if (head) | |
3087 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
3088 | ||
3089 | io_ring_submit_lock(ctx, !force_nonblock); | |
3090 | if (ret < 0) | |
3091 | req_set_fail_links(req); | |
3092 | io_cqring_add_event(req, ret); | |
3093 | io_put_req(req); | |
3094 | return 0; | |
3095 | } | |
3096 | ||
ddf0322d JA |
3097 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3098 | const struct io_uring_sqe *sqe) | |
3099 | { | |
3100 | struct io_provide_buf *p = &req->pbuf; | |
3101 | u64 tmp; | |
3102 | ||
3103 | if (sqe->ioprio || sqe->rw_flags) | |
3104 | return -EINVAL; | |
3105 | ||
3106 | tmp = READ_ONCE(sqe->fd); | |
3107 | if (!tmp || tmp > USHRT_MAX) | |
3108 | return -E2BIG; | |
3109 | p->nbufs = tmp; | |
3110 | p->addr = READ_ONCE(sqe->addr); | |
3111 | p->len = READ_ONCE(sqe->len); | |
3112 | ||
3113 | if (!access_ok(u64_to_user_ptr(p->addr), p->len)) | |
3114 | return -EFAULT; | |
3115 | ||
3116 | p->bgid = READ_ONCE(sqe->buf_group); | |
3117 | tmp = READ_ONCE(sqe->off); | |
3118 | if (tmp > USHRT_MAX) | |
3119 | return -E2BIG; | |
3120 | p->bid = tmp; | |
3121 | return 0; | |
3122 | } | |
3123 | ||
3124 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3125 | { | |
3126 | struct io_buffer *buf; | |
3127 | u64 addr = pbuf->addr; | |
3128 | int i, bid = pbuf->bid; | |
3129 | ||
3130 | for (i = 0; i < pbuf->nbufs; i++) { | |
3131 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3132 | if (!buf) | |
3133 | break; | |
3134 | ||
3135 | buf->addr = addr; | |
3136 | buf->len = pbuf->len; | |
3137 | buf->bid = bid; | |
3138 | addr += pbuf->len; | |
3139 | bid++; | |
3140 | if (!*head) { | |
3141 | INIT_LIST_HEAD(&buf->list); | |
3142 | *head = buf; | |
3143 | } else { | |
3144 | list_add_tail(&buf->list, &(*head)->list); | |
3145 | } | |
3146 | } | |
3147 | ||
3148 | return i ? i : -ENOMEM; | |
3149 | } | |
3150 | ||
ddf0322d JA |
3151 | static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock) |
3152 | { | |
3153 | struct io_provide_buf *p = &req->pbuf; | |
3154 | struct io_ring_ctx *ctx = req->ctx; | |
3155 | struct io_buffer *head, *list; | |
3156 | int ret = 0; | |
3157 | ||
3158 | io_ring_submit_lock(ctx, !force_nonblock); | |
3159 | ||
3160 | lockdep_assert_held(&ctx->uring_lock); | |
3161 | ||
3162 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3163 | ||
3164 | ret = io_add_buffers(p, &head); | |
3165 | if (ret < 0) | |
3166 | goto out; | |
3167 | ||
3168 | if (!list) { | |
3169 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3170 | GFP_KERNEL); | |
3171 | if (ret < 0) { | |
067524e9 | 3172 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3173 | goto out; |
3174 | } | |
3175 | } | |
3176 | out: | |
3177 | io_ring_submit_unlock(ctx, !force_nonblock); | |
3178 | if (ret < 0) | |
3179 | req_set_fail_links(req); | |
3180 | io_cqring_add_event(req, ret); | |
3181 | io_put_req(req); | |
3182 | return 0; | |
cebdb986 JA |
3183 | } |
3184 | ||
3e4827b0 JA |
3185 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3186 | const struct io_uring_sqe *sqe) | |
3187 | { | |
3188 | #if defined(CONFIG_EPOLL) | |
3189 | if (sqe->ioprio || sqe->buf_index) | |
3190 | return -EINVAL; | |
3191 | ||
3192 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
3193 | req->epoll.op = READ_ONCE(sqe->len); | |
3194 | req->epoll.fd = READ_ONCE(sqe->off); | |
3195 | ||
3196 | if (ep_op_has_event(req->epoll.op)) { | |
3197 | struct epoll_event __user *ev; | |
3198 | ||
3199 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3200 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
3201 | return -EFAULT; | |
3202 | } | |
3203 | ||
3204 | return 0; | |
3205 | #else | |
3206 | return -EOPNOTSUPP; | |
3207 | #endif | |
3208 | } | |
3209 | ||
014db007 | 3210 | static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock) |
3e4827b0 JA |
3211 | { |
3212 | #if defined(CONFIG_EPOLL) | |
3213 | struct io_epoll *ie = &req->epoll; | |
3214 | int ret; | |
3215 | ||
3216 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
3217 | if (force_nonblock && ret == -EAGAIN) | |
3218 | return -EAGAIN; | |
3219 | ||
3220 | if (ret < 0) | |
3221 | req_set_fail_links(req); | |
3222 | io_cqring_add_event(req, ret); | |
014db007 | 3223 | io_put_req(req); |
3e4827b0 JA |
3224 | return 0; |
3225 | #else | |
3226 | return -EOPNOTSUPP; | |
3227 | #endif | |
3228 | } | |
3229 | ||
c1ca757b JA |
3230 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3231 | { | |
3232 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3233 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
3234 | return -EINVAL; | |
3235 | ||
3236 | req->madvise.addr = READ_ONCE(sqe->addr); | |
3237 | req->madvise.len = READ_ONCE(sqe->len); | |
3238 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3239 | return 0; | |
3240 | #else | |
3241 | return -EOPNOTSUPP; | |
3242 | #endif | |
3243 | } | |
3244 | ||
014db007 | 3245 | static int io_madvise(struct io_kiocb *req, bool force_nonblock) |
c1ca757b JA |
3246 | { |
3247 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
3248 | struct io_madvise *ma = &req->madvise; | |
3249 | int ret; | |
3250 | ||
3251 | if (force_nonblock) | |
3252 | return -EAGAIN; | |
3253 | ||
3254 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
3255 | if (ret < 0) | |
3256 | req_set_fail_links(req); | |
3257 | io_cqring_add_event(req, ret); | |
014db007 | 3258 | io_put_req(req); |
c1ca757b JA |
3259 | return 0; |
3260 | #else | |
3261 | return -EOPNOTSUPP; | |
3262 | #endif | |
3263 | } | |
3264 | ||
4840e418 JA |
3265 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3266 | { | |
3267 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
3268 | return -EINVAL; | |
3269 | ||
3270 | req->fadvise.offset = READ_ONCE(sqe->off); | |
3271 | req->fadvise.len = READ_ONCE(sqe->len); | |
3272 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
3273 | return 0; | |
3274 | } | |
3275 | ||
014db007 | 3276 | static int io_fadvise(struct io_kiocb *req, bool force_nonblock) |
4840e418 JA |
3277 | { |
3278 | struct io_fadvise *fa = &req->fadvise; | |
3279 | int ret; | |
3280 | ||
3e69426d JA |
3281 | if (force_nonblock) { |
3282 | switch (fa->advice) { | |
3283 | case POSIX_FADV_NORMAL: | |
3284 | case POSIX_FADV_RANDOM: | |
3285 | case POSIX_FADV_SEQUENTIAL: | |
3286 | break; | |
3287 | default: | |
3288 | return -EAGAIN; | |
3289 | } | |
3290 | } | |
4840e418 JA |
3291 | |
3292 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
3293 | if (ret < 0) | |
3294 | req_set_fail_links(req); | |
3295 | io_cqring_add_event(req, ret); | |
014db007 | 3296 | io_put_req(req); |
4840e418 JA |
3297 | return 0; |
3298 | } | |
3299 | ||
eddc7ef5 JA |
3300 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3301 | { | |
f8748881 | 3302 | const char __user *fname; |
eddc7ef5 JA |
3303 | unsigned lookup_flags; |
3304 | int ret; | |
3305 | ||
3306 | if (sqe->ioprio || sqe->buf_index) | |
3307 | return -EINVAL; | |
cf3040ca JA |
3308 | if (sqe->flags & IOSQE_FIXED_FILE) |
3309 | return -EBADF; | |
0bdbdd08 PB |
3310 | if (req->flags & REQ_F_NEED_CLEANUP) |
3311 | return 0; | |
eddc7ef5 JA |
3312 | |
3313 | req->open.dfd = READ_ONCE(sqe->fd); | |
3314 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 3315 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 3316 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 3317 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 3318 | |
c12cedf2 | 3319 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
3320 | return -EINVAL; |
3321 | ||
f8748881 | 3322 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
3323 | if (IS_ERR(req->open.filename)) { |
3324 | ret = PTR_ERR(req->open.filename); | |
3325 | req->open.filename = NULL; | |
3326 | return ret; | |
3327 | } | |
3328 | ||
8fef80bf | 3329 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3330 | return 0; |
3331 | } | |
3332 | ||
014db007 | 3333 | static int io_statx(struct io_kiocb *req, bool force_nonblock) |
eddc7ef5 JA |
3334 | { |
3335 | struct io_open *ctx = &req->open; | |
3336 | unsigned lookup_flags; | |
3337 | struct path path; | |
3338 | struct kstat stat; | |
3339 | int ret; | |
3340 | ||
3341 | if (force_nonblock) | |
3342 | return -EAGAIN; | |
3343 | ||
c12cedf2 | 3344 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
3345 | return -EINVAL; |
3346 | ||
3347 | retry: | |
3348 | /* filename_lookup() drops it, keep a reference */ | |
3349 | ctx->filename->refcnt++; | |
3350 | ||
3351 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
3352 | NULL); | |
3353 | if (ret) | |
3354 | goto err; | |
3355 | ||
c12cedf2 | 3356 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
3357 | path_put(&path); |
3358 | if (retry_estale(ret, lookup_flags)) { | |
3359 | lookup_flags |= LOOKUP_REVAL; | |
3360 | goto retry; | |
3361 | } | |
3362 | if (!ret) | |
3363 | ret = cp_statx(&stat, ctx->buffer); | |
3364 | err: | |
3365 | putname(ctx->filename); | |
8fef80bf | 3366 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
3367 | if (ret < 0) |
3368 | req_set_fail_links(req); | |
3369 | io_cqring_add_event(req, ret); | |
014db007 | 3370 | io_put_req(req); |
eddc7ef5 JA |
3371 | return 0; |
3372 | } | |
3373 | ||
b5dba59e JA |
3374 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3375 | { | |
3376 | /* | |
3377 | * If we queue this for async, it must not be cancellable. That would | |
3378 | * leave the 'file' in an undeterminate state. | |
3379 | */ | |
3380 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
3381 | ||
3382 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
3383 | sqe->rw_flags || sqe->buf_index) | |
3384 | return -EINVAL; | |
3385 | if (sqe->flags & IOSQE_FIXED_FILE) | |
cf3040ca | 3386 | return -EBADF; |
b5dba59e JA |
3387 | |
3388 | req->close.fd = READ_ONCE(sqe->fd); | |
3389 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 3390 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
3391 | return -EBADF; |
3392 | ||
3393 | return 0; | |
3394 | } | |
3395 | ||
a93b3331 | 3396 | /* only called when __close_fd_get_file() is done */ |
014db007 | 3397 | static void __io_close_finish(struct io_kiocb *req) |
a93b3331 PB |
3398 | { |
3399 | int ret; | |
3400 | ||
3401 | ret = filp_close(req->close.put_file, req->work.files); | |
3402 | if (ret < 0) | |
3403 | req_set_fail_links(req); | |
3404 | io_cqring_add_event(req, ret); | |
3405 | fput(req->close.put_file); | |
014db007 | 3406 | io_put_req(req); |
a93b3331 PB |
3407 | } |
3408 | ||
b5dba59e JA |
3409 | static void io_close_finish(struct io_wq_work **workptr) |
3410 | { | |
3411 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
b5dba59e | 3412 | |
7fbeb95d | 3413 | /* not cancellable, don't do io_req_cancelled() */ |
014db007 | 3414 | __io_close_finish(req); |
e9fd9396 | 3415 | io_steal_work(req, workptr); |
b5dba59e JA |
3416 | } |
3417 | ||
014db007 | 3418 | static int io_close(struct io_kiocb *req, bool force_nonblock) |
b5dba59e JA |
3419 | { |
3420 | int ret; | |
3421 | ||
3422 | req->close.put_file = NULL; | |
3423 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3424 | if (ret < 0) | |
3425 | return ret; | |
3426 | ||
3427 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 | 3428 | if (req->close.put_file->f_op->flush && force_nonblock) { |
594506fe PB |
3429 | /* submission ref will be dropped, take it for async */ |
3430 | refcount_inc(&req->refs); | |
3431 | ||
a2100672 PB |
3432 | req->work.func = io_close_finish; |
3433 | /* | |
3434 | * Do manual async queue here to avoid grabbing files - we don't | |
3435 | * need the files, and it'll cause io_close_finish() to close | |
3436 | * the file again and cause a double CQE entry for this request | |
3437 | */ | |
3438 | io_queue_async_work(req); | |
3439 | return 0; | |
3440 | } | |
b5dba59e JA |
3441 | |
3442 | /* | |
3443 | * No ->flush(), safely close from here and just punt the | |
3444 | * fput() to async context. | |
3445 | */ | |
014db007 | 3446 | __io_close_finish(req); |
1a417f4e | 3447 | return 0; |
b5dba59e JA |
3448 | } |
3449 | ||
3529d8c2 | 3450 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3451 | { |
3452 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3453 | |
3454 | if (!req->file) | |
3455 | return -EBADF; | |
5d17b4a4 JA |
3456 | |
3457 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3458 | return -EINVAL; | |
3459 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3460 | return -EINVAL; | |
3461 | ||
8ed8d3c3 JA |
3462 | req->sync.off = READ_ONCE(sqe->off); |
3463 | req->sync.len = READ_ONCE(sqe->len); | |
3464 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3465 | return 0; |
3466 | } | |
3467 | ||
014db007 | 3468 | static void __io_sync_file_range(struct io_kiocb *req) |
8ed8d3c3 | 3469 | { |
8ed8d3c3 JA |
3470 | int ret; |
3471 | ||
9adbd45d | 3472 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3473 | req->sync.flags); |
3474 | if (ret < 0) | |
3475 | req_set_fail_links(req); | |
3476 | io_cqring_add_event(req, ret); | |
014db007 | 3477 | io_put_req(req); |
5ea62161 PB |
3478 | } |
3479 | ||
3480 | ||
3481 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3482 | { | |
3483 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3484 | struct io_kiocb *nxt = NULL; | |
3485 | ||
3486 | if (io_req_cancelled(req)) | |
3487 | return; | |
014db007 | 3488 | __io_sync_file_range(req); |
594506fe | 3489 | io_put_req(req); /* put submission ref */ |
8ed8d3c3 | 3490 | if (nxt) |
78912934 | 3491 | io_wq_assign_next(workptr, nxt); |
5d17b4a4 JA |
3492 | } |
3493 | ||
014db007 | 3494 | static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock) |
5d17b4a4 | 3495 | { |
5d17b4a4 | 3496 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 | 3497 | if (force_nonblock) { |
8ed8d3c3 | 3498 | req->work.func = io_sync_file_range_finish; |
5d17b4a4 | 3499 | return -EAGAIN; |
8ed8d3c3 | 3500 | } |
5d17b4a4 | 3501 | |
014db007 | 3502 | __io_sync_file_range(req); |
5d17b4a4 JA |
3503 | return 0; |
3504 | } | |
3505 | ||
469956e8 | 3506 | #if defined(CONFIG_NET) |
02d27d89 PB |
3507 | static int io_setup_async_msg(struct io_kiocb *req, |
3508 | struct io_async_msghdr *kmsg) | |
3509 | { | |
3510 | if (req->io) | |
3511 | return -EAGAIN; | |
3512 | if (io_alloc_async_ctx(req)) { | |
3513 | if (kmsg->iov != kmsg->fast_iov) | |
3514 | kfree(kmsg->iov); | |
3515 | return -ENOMEM; | |
3516 | } | |
3517 | req->flags |= REQ_F_NEED_CLEANUP; | |
3518 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3519 | return -EAGAIN; | |
3520 | } | |
3521 | ||
3529d8c2 | 3522 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3523 | { |
e47293fd | 3524 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3525 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3526 | int ret; |
03b1230c | 3527 | |
e47293fd JA |
3528 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3529 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3530 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3531 | |
d8768362 JA |
3532 | #ifdef CONFIG_COMPAT |
3533 | if (req->ctx->compat) | |
3534 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3535 | #endif | |
3536 | ||
fddaface | 3537 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3538 | return 0; |
5f798bea PB |
3539 | /* iovec is already imported */ |
3540 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3541 | return 0; | |
3529d8c2 | 3542 | |
d9688565 | 3543 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3544 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3545 | &io->msg.iov); |
99bc4c38 PB |
3546 | if (!ret) |
3547 | req->flags |= REQ_F_NEED_CLEANUP; | |
3548 | return ret; | |
03b1230c JA |
3549 | } |
3550 | ||
014db007 | 3551 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3552 | { |
0b416c3e | 3553 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3554 | struct socket *sock; |
3555 | int ret; | |
3556 | ||
3557 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3558 | return -EINVAL; | |
3559 | ||
3560 | sock = sock_from_file(req->file, &ret); | |
3561 | if (sock) { | |
b7bb4f7d | 3562 | struct io_async_ctx io; |
0fa03c62 JA |
3563 | unsigned flags; |
3564 | ||
03b1230c | 3565 | if (req->io) { |
0b416c3e | 3566 | kmsg = &req->io->msg; |
b537916c | 3567 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3568 | /* if iov is set, it's allocated already */ |
3569 | if (!kmsg->iov) | |
3570 | kmsg->iov = kmsg->fast_iov; | |
3571 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3572 | } else { |
3529d8c2 JA |
3573 | struct io_sr_msg *sr = &req->sr_msg; |
3574 | ||
0b416c3e | 3575 | kmsg = &io.msg; |
b537916c | 3576 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3577 | |
3578 | io.msg.iov = io.msg.fast_iov; | |
3579 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3580 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3581 | if (ret) |
3529d8c2 | 3582 | return ret; |
03b1230c | 3583 | } |
0fa03c62 | 3584 | |
e47293fd JA |
3585 | flags = req->sr_msg.msg_flags; |
3586 | if (flags & MSG_DONTWAIT) | |
3587 | req->flags |= REQ_F_NOWAIT; | |
3588 | else if (force_nonblock) | |
3589 | flags |= MSG_DONTWAIT; | |
3590 | ||
0b416c3e | 3591 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3592 | if (force_nonblock && ret == -EAGAIN) |
3593 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3594 | if (ret == -ERESTARTSYS) |
3595 | ret = -EINTR; | |
0fa03c62 JA |
3596 | } |
3597 | ||
1e95081c | 3598 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3599 | kfree(kmsg->iov); |
99bc4c38 | 3600 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3601 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3602 | if (ret < 0) |
3603 | req_set_fail_links(req); | |
014db007 | 3604 | io_put_req(req); |
5d17b4a4 | 3605 | return 0; |
03b1230c | 3606 | } |
aa1fa28f | 3607 | |
014db007 | 3608 | static int io_send(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3609 | { |
fddaface JA |
3610 | struct socket *sock; |
3611 | int ret; | |
3612 | ||
3613 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3614 | return -EINVAL; | |
3615 | ||
3616 | sock = sock_from_file(req->file, &ret); | |
3617 | if (sock) { | |
3618 | struct io_sr_msg *sr = &req->sr_msg; | |
3619 | struct msghdr msg; | |
3620 | struct iovec iov; | |
3621 | unsigned flags; | |
3622 | ||
3623 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3624 | &msg.msg_iter); | |
3625 | if (ret) | |
3626 | return ret; | |
3627 | ||
3628 | msg.msg_name = NULL; | |
3629 | msg.msg_control = NULL; | |
3630 | msg.msg_controllen = 0; | |
3631 | msg.msg_namelen = 0; | |
3632 | ||
3633 | flags = req->sr_msg.msg_flags; | |
3634 | if (flags & MSG_DONTWAIT) | |
3635 | req->flags |= REQ_F_NOWAIT; | |
3636 | else if (force_nonblock) | |
3637 | flags |= MSG_DONTWAIT; | |
3638 | ||
0b7b21e4 JA |
3639 | msg.msg_flags = flags; |
3640 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3641 | if (force_nonblock && ret == -EAGAIN) |
3642 | return -EAGAIN; | |
3643 | if (ret == -ERESTARTSYS) | |
3644 | ret = -EINTR; | |
3645 | } | |
3646 | ||
3647 | io_cqring_add_event(req, ret); | |
3648 | if (ret < 0) | |
3649 | req_set_fail_links(req); | |
014db007 | 3650 | io_put_req(req); |
fddaface | 3651 | return 0; |
fddaface JA |
3652 | } |
3653 | ||
52de1fe1 JA |
3654 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) |
3655 | { | |
3656 | struct io_sr_msg *sr = &req->sr_msg; | |
3657 | struct iovec __user *uiov; | |
3658 | size_t iov_len; | |
3659 | int ret; | |
3660 | ||
3661 | ret = __copy_msghdr_from_user(&io->msg.msg, sr->msg, &io->msg.uaddr, | |
3662 | &uiov, &iov_len); | |
3663 | if (ret) | |
3664 | return ret; | |
3665 | ||
3666 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3667 | if (iov_len > 1) | |
3668 | return -EINVAL; | |
3669 | if (copy_from_user(io->msg.iov, uiov, sizeof(*uiov))) | |
3670 | return -EFAULT; | |
3671 | sr->len = io->msg.iov[0].iov_len; | |
3672 | iov_iter_init(&io->msg.msg.msg_iter, READ, io->msg.iov, 1, | |
3673 | sr->len); | |
3674 | io->msg.iov = NULL; | |
3675 | } else { | |
3676 | ret = import_iovec(READ, uiov, iov_len, UIO_FASTIOV, | |
3677 | &io->msg.iov, &io->msg.msg.msg_iter); | |
3678 | if (ret > 0) | |
3679 | ret = 0; | |
3680 | } | |
3681 | ||
3682 | return ret; | |
3683 | } | |
3684 | ||
3685 | #ifdef CONFIG_COMPAT | |
3686 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
3687 | struct io_async_ctx *io) | |
3688 | { | |
3689 | struct compat_msghdr __user *msg_compat; | |
3690 | struct io_sr_msg *sr = &req->sr_msg; | |
3691 | struct compat_iovec __user *uiov; | |
3692 | compat_uptr_t ptr; | |
3693 | compat_size_t len; | |
3694 | int ret; | |
3695 | ||
3696 | msg_compat = (struct compat_msghdr __user *) sr->msg; | |
3697 | ret = __get_compat_msghdr(&io->msg.msg, msg_compat, &io->msg.uaddr, | |
3698 | &ptr, &len); | |
3699 | if (ret) | |
3700 | return ret; | |
3701 | ||
3702 | uiov = compat_ptr(ptr); | |
3703 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
3704 | compat_ssize_t clen; | |
3705 | ||
3706 | if (len > 1) | |
3707 | return -EINVAL; | |
3708 | if (!access_ok(uiov, sizeof(*uiov))) | |
3709 | return -EFAULT; | |
3710 | if (__get_user(clen, &uiov->iov_len)) | |
3711 | return -EFAULT; | |
3712 | if (clen < 0) | |
3713 | return -EINVAL; | |
3714 | sr->len = io->msg.iov[0].iov_len; | |
3715 | io->msg.iov = NULL; | |
3716 | } else { | |
3717 | ret = compat_import_iovec(READ, uiov, len, UIO_FASTIOV, | |
3718 | &io->msg.iov, | |
3719 | &io->msg.msg.msg_iter); | |
3720 | if (ret < 0) | |
3721 | return ret; | |
3722 | } | |
3723 | ||
3724 | return 0; | |
3725 | } | |
3726 | #endif | |
3727 | ||
3728 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, struct io_async_ctx *io) | |
3729 | { | |
3730 | io->msg.iov = io->msg.fast_iov; | |
3731 | ||
3732 | #ifdef CONFIG_COMPAT | |
3733 | if (req->ctx->compat) | |
3734 | return __io_compat_recvmsg_copy_hdr(req, io); | |
fddaface | 3735 | #endif |
52de1fe1 JA |
3736 | |
3737 | return __io_recvmsg_copy_hdr(req, io); | |
3738 | } | |
3739 | ||
bcda7baa JA |
3740 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
3741 | int *cflags, bool needs_lock) | |
3742 | { | |
3743 | struct io_sr_msg *sr = &req->sr_msg; | |
3744 | struct io_buffer *kbuf; | |
3745 | ||
3746 | if (!(req->flags & REQ_F_BUFFER_SELECT)) | |
3747 | return NULL; | |
3748 | ||
3749 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); | |
3750 | if (IS_ERR(kbuf)) | |
3751 | return kbuf; | |
3752 | ||
3753 | sr->kbuf = kbuf; | |
3754 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3755 | ||
3756 | *cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; | |
3757 | *cflags |= IORING_CQE_F_BUFFER; | |
3758 | return kbuf; | |
fddaface JA |
3759 | } |
3760 | ||
3529d8c2 JA |
3761 | static int io_recvmsg_prep(struct io_kiocb *req, |
3762 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 3763 | { |
e47293fd | 3764 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3765 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3766 | int ret; |
3529d8c2 JA |
3767 | |
3768 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3769 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3770 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 3771 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 3772 | |
d8768362 JA |
3773 | #ifdef CONFIG_COMPAT |
3774 | if (req->ctx->compat) | |
3775 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3776 | #endif | |
3777 | ||
fddaface | 3778 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3779 | return 0; |
5f798bea PB |
3780 | /* iovec is already imported */ |
3781 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3782 | return 0; | |
03b1230c | 3783 | |
52de1fe1 | 3784 | ret = io_recvmsg_copy_hdr(req, io); |
99bc4c38 PB |
3785 | if (!ret) |
3786 | req->flags |= REQ_F_NEED_CLEANUP; | |
3787 | return ret; | |
aa1fa28f JA |
3788 | } |
3789 | ||
014db007 | 3790 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) |
aa1fa28f | 3791 | { |
0b416c3e | 3792 | struct io_async_msghdr *kmsg = NULL; |
03b1230c | 3793 | struct socket *sock; |
52de1fe1 | 3794 | int ret, cflags = 0; |
03b1230c JA |
3795 | |
3796 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3797 | return -EINVAL; | |
3798 | ||
3799 | sock = sock_from_file(req->file, &ret); | |
3800 | if (sock) { | |
52de1fe1 | 3801 | struct io_buffer *kbuf; |
b7bb4f7d | 3802 | struct io_async_ctx io; |
03b1230c JA |
3803 | unsigned flags; |
3804 | ||
03b1230c | 3805 | if (req->io) { |
0b416c3e | 3806 | kmsg = &req->io->msg; |
b537916c | 3807 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3808 | /* if iov is set, it's allocated already */ |
3809 | if (!kmsg->iov) | |
3810 | kmsg->iov = kmsg->fast_iov; | |
3811 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3812 | } else { |
0b416c3e | 3813 | kmsg = &io.msg; |
b537916c | 3814 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 | 3815 | |
52de1fe1 | 3816 | ret = io_recvmsg_copy_hdr(req, &io); |
03b1230c | 3817 | if (ret) |
3529d8c2 | 3818 | return ret; |
03b1230c JA |
3819 | } |
3820 | ||
52de1fe1 JA |
3821 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3822 | if (IS_ERR(kbuf)) { | |
3823 | return PTR_ERR(kbuf); | |
3824 | } else if (kbuf) { | |
3825 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3826 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov, | |
3827 | 1, req->sr_msg.len); | |
3828 | } | |
3829 | ||
e47293fd JA |
3830 | flags = req->sr_msg.msg_flags; |
3831 | if (flags & MSG_DONTWAIT) | |
3832 | req->flags |= REQ_F_NOWAIT; | |
3833 | else if (force_nonblock) | |
3834 | flags |= MSG_DONTWAIT; | |
3835 | ||
3836 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3837 | kmsg->uaddr, flags); | |
02d27d89 PB |
3838 | if (force_nonblock && ret == -EAGAIN) |
3839 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3840 | if (ret == -ERESTARTSYS) |
3841 | ret = -EINTR; | |
3842 | } | |
3843 | ||
1e95081c | 3844 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3845 | kfree(kmsg->iov); |
99bc4c38 | 3846 | req->flags &= ~REQ_F_NEED_CLEANUP; |
52de1fe1 | 3847 | __io_cqring_add_event(req, ret, cflags); |
4e88d6e7 JA |
3848 | if (ret < 0) |
3849 | req_set_fail_links(req); | |
014db007 | 3850 | io_put_req(req); |
03b1230c | 3851 | return 0; |
0fa03c62 | 3852 | } |
5d17b4a4 | 3853 | |
014db007 | 3854 | static int io_recv(struct io_kiocb *req, bool force_nonblock) |
fddaface | 3855 | { |
bcda7baa | 3856 | struct io_buffer *kbuf = NULL; |
fddaface | 3857 | struct socket *sock; |
bcda7baa | 3858 | int ret, cflags = 0; |
fddaface JA |
3859 | |
3860 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3861 | return -EINVAL; | |
3862 | ||
3863 | sock = sock_from_file(req->file, &ret); | |
3864 | if (sock) { | |
3865 | struct io_sr_msg *sr = &req->sr_msg; | |
bcda7baa | 3866 | void __user *buf = sr->buf; |
fddaface JA |
3867 | struct msghdr msg; |
3868 | struct iovec iov; | |
3869 | unsigned flags; | |
3870 | ||
bcda7baa JA |
3871 | kbuf = io_recv_buffer_select(req, &cflags, !force_nonblock); |
3872 | if (IS_ERR(kbuf)) | |
3873 | return PTR_ERR(kbuf); | |
3874 | else if (kbuf) | |
3875 | buf = u64_to_user_ptr(kbuf->addr); | |
3876 | ||
3877 | ret = import_single_range(READ, buf, sr->len, &iov, | |
fddaface | 3878 | &msg.msg_iter); |
bcda7baa JA |
3879 | if (ret) { |
3880 | kfree(kbuf); | |
fddaface | 3881 | return ret; |
bcda7baa | 3882 | } |
fddaface | 3883 | |
bcda7baa | 3884 | req->flags |= REQ_F_NEED_CLEANUP; |
fddaface JA |
3885 | msg.msg_name = NULL; |
3886 | msg.msg_control = NULL; | |
3887 | msg.msg_controllen = 0; | |
3888 | msg.msg_namelen = 0; | |
3889 | msg.msg_iocb = NULL; | |
3890 | msg.msg_flags = 0; | |
3891 | ||
3892 | flags = req->sr_msg.msg_flags; | |
3893 | if (flags & MSG_DONTWAIT) | |
3894 | req->flags |= REQ_F_NOWAIT; | |
3895 | else if (force_nonblock) | |
3896 | flags |= MSG_DONTWAIT; | |
3897 | ||
0b7b21e4 | 3898 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3899 | if (force_nonblock && ret == -EAGAIN) |
3900 | return -EAGAIN; | |
3901 | if (ret == -ERESTARTSYS) | |
3902 | ret = -EINTR; | |
3903 | } | |
3904 | ||
bcda7baa JA |
3905 | kfree(kbuf); |
3906 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3907 | __io_cqring_add_event(req, ret, cflags); | |
fddaface JA |
3908 | if (ret < 0) |
3909 | req_set_fail_links(req); | |
014db007 | 3910 | io_put_req(req); |
fddaface | 3911 | return 0; |
fddaface JA |
3912 | } |
3913 | ||
3529d8c2 | 3914 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 3915 | { |
8ed8d3c3 JA |
3916 | struct io_accept *accept = &req->accept; |
3917 | ||
17f2fe35 JA |
3918 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3919 | return -EINVAL; | |
8042d6ce | 3920 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3921 | return -EINVAL; |
3922 | ||
d55e5f5b JA |
3923 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3924 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3925 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 3926 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 3927 | return 0; |
8ed8d3c3 | 3928 | } |
17f2fe35 | 3929 | |
014db007 | 3930 | static int __io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 JA |
3931 | { |
3932 | struct io_accept *accept = &req->accept; | |
3933 | unsigned file_flags; | |
3934 | int ret; | |
3935 | ||
3936 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3937 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
09952e3e JA |
3938 | accept->addr_len, accept->flags, |
3939 | accept->nofile); | |
8ed8d3c3 | 3940 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 3941 | return -EAGAIN; |
8e3cca12 JA |
3942 | if (ret == -ERESTARTSYS) |
3943 | ret = -EINTR; | |
4e88d6e7 JA |
3944 | if (ret < 0) |
3945 | req_set_fail_links(req); | |
78e19bbe | 3946 | io_cqring_add_event(req, ret); |
014db007 | 3947 | io_put_req(req); |
17f2fe35 | 3948 | return 0; |
8ed8d3c3 JA |
3949 | } |
3950 | ||
3951 | static void io_accept_finish(struct io_wq_work **workptr) | |
3952 | { | |
3953 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
8ed8d3c3 JA |
3954 | |
3955 | if (io_req_cancelled(req)) | |
3956 | return; | |
014db007 | 3957 | __io_accept(req, false); |
e9fd9396 | 3958 | io_steal_work(req, workptr); |
8ed8d3c3 | 3959 | } |
8ed8d3c3 | 3960 | |
014db007 | 3961 | static int io_accept(struct io_kiocb *req, bool force_nonblock) |
8ed8d3c3 | 3962 | { |
8ed8d3c3 JA |
3963 | int ret; |
3964 | ||
014db007 | 3965 | ret = __io_accept(req, force_nonblock); |
8ed8d3c3 JA |
3966 | if (ret == -EAGAIN && force_nonblock) { |
3967 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3968 | return -EAGAIN; |
3969 | } | |
3970 | return 0; | |
0fa03c62 | 3971 | } |
5d17b4a4 | 3972 | |
3529d8c2 | 3973 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 3974 | { |
3529d8c2 JA |
3975 | struct io_connect *conn = &req->connect; |
3976 | struct io_async_ctx *io = req->io; | |
f499a021 | 3977 | |
3fbb51c1 JA |
3978 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3979 | return -EINVAL; | |
3980 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3981 | return -EINVAL; | |
3982 | ||
3529d8c2 JA |
3983 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3984 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3985 | ||
3986 | if (!io) | |
3987 | return 0; | |
3988 | ||
3989 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3990 | &io->connect.address); |
f499a021 JA |
3991 | } |
3992 | ||
014db007 | 3993 | static int io_connect(struct io_kiocb *req, bool force_nonblock) |
f8e85cf2 | 3994 | { |
f499a021 | 3995 | struct io_async_ctx __io, *io; |
f8e85cf2 | 3996 | unsigned file_flags; |
3fbb51c1 | 3997 | int ret; |
f8e85cf2 | 3998 | |
f499a021 JA |
3999 | if (req->io) { |
4000 | io = req->io; | |
4001 | } else { | |
3529d8c2 JA |
4002 | ret = move_addr_to_kernel(req->connect.addr, |
4003 | req->connect.addr_len, | |
4004 | &__io.connect.address); | |
f499a021 JA |
4005 | if (ret) |
4006 | goto out; | |
4007 | io = &__io; | |
4008 | } | |
4009 | ||
3fbb51c1 JA |
4010 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4011 | ||
4012 | ret = __sys_connect_file(req->file, &io->connect.address, | |
4013 | req->connect.addr_len, file_flags); | |
87f80d62 | 4014 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
4015 | if (req->io) |
4016 | return -EAGAIN; | |
4017 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
4018 | ret = -ENOMEM; |
4019 | goto out; | |
4020 | } | |
b7bb4f7d | 4021 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 4022 | return -EAGAIN; |
f499a021 | 4023 | } |
f8e85cf2 JA |
4024 | if (ret == -ERESTARTSYS) |
4025 | ret = -EINTR; | |
f499a021 | 4026 | out: |
4e88d6e7 JA |
4027 | if (ret < 0) |
4028 | req_set_fail_links(req); | |
f8e85cf2 | 4029 | io_cqring_add_event(req, ret); |
014db007 | 4030 | io_put_req(req); |
f8e85cf2 | 4031 | return 0; |
469956e8 Y |
4032 | } |
4033 | #else /* !CONFIG_NET */ | |
4034 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4035 | { | |
f8e85cf2 | 4036 | return -EOPNOTSUPP; |
f8e85cf2 JA |
4037 | } |
4038 | ||
469956e8 Y |
4039 | static int io_sendmsg(struct io_kiocb *req, bool force_nonblock) |
4040 | { | |
4041 | return -EOPNOTSUPP; | |
4042 | } | |
4043 | ||
4044 | static int io_send(struct io_kiocb *req, bool force_nonblock) | |
4045 | { | |
4046 | return -EOPNOTSUPP; | |
4047 | } | |
4048 | ||
4049 | static int io_recvmsg_prep(struct io_kiocb *req, | |
4050 | const struct io_uring_sqe *sqe) | |
4051 | { | |
4052 | return -EOPNOTSUPP; | |
4053 | } | |
4054 | ||
4055 | static int io_recvmsg(struct io_kiocb *req, bool force_nonblock) | |
4056 | { | |
4057 | return -EOPNOTSUPP; | |
4058 | } | |
4059 | ||
4060 | static int io_recv(struct io_kiocb *req, bool force_nonblock) | |
4061 | { | |
4062 | return -EOPNOTSUPP; | |
4063 | } | |
4064 | ||
4065 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4066 | { | |
4067 | return -EOPNOTSUPP; | |
4068 | } | |
4069 | ||
4070 | static int io_accept(struct io_kiocb *req, bool force_nonblock) | |
4071 | { | |
4072 | return -EOPNOTSUPP; | |
4073 | } | |
4074 | ||
4075 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4076 | { | |
4077 | return -EOPNOTSUPP; | |
4078 | } | |
4079 | ||
4080 | static int io_connect(struct io_kiocb *req, bool force_nonblock) | |
4081 | { | |
f8e85cf2 | 4082 | return -EOPNOTSUPP; |
f8e85cf2 | 4083 | } |
469956e8 | 4084 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4085 | |
d7718a9d JA |
4086 | struct io_poll_table { |
4087 | struct poll_table_struct pt; | |
4088 | struct io_kiocb *req; | |
4089 | int error; | |
4090 | }; | |
4091 | ||
4092 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
4093 | struct wait_queue_head *head) | |
4094 | { | |
4095 | if (unlikely(poll->head)) { | |
4096 | pt->error = -EINVAL; | |
4097 | return; | |
4098 | } | |
4099 | ||
4100 | pt->error = 0; | |
4101 | poll->head = head; | |
4102 | add_wait_queue(head, &poll->wait); | |
4103 | } | |
4104 | ||
4105 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4106 | struct poll_table_struct *p) | |
4107 | { | |
4108 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4109 | ||
4110 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
4111 | } | |
4112 | ||
4113 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
4114 | __poll_t mask, task_work_func_t func) | |
4115 | { | |
4116 | struct task_struct *tsk; | |
4117 | ||
4118 | /* for instances that support it check for an event match first: */ | |
4119 | if (mask && !(mask & poll->events)) | |
4120 | return 0; | |
4121 | ||
4122 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4123 | ||
4124 | list_del_init(&poll->wait.entry); | |
4125 | ||
4126 | tsk = req->task; | |
4127 | req->result = mask; | |
4128 | init_task_work(&req->task_work, func); | |
4129 | /* | |
4130 | * If this fails, then the task is exiting. If that is the case, then | |
4131 | * the exit check will ultimately cancel these work items. Hence we | |
4132 | * don't need to check here and handle it specifically. | |
4133 | */ | |
4134 | task_work_add(tsk, &req->task_work, true); | |
4135 | wake_up_process(tsk); | |
4136 | return 1; | |
4137 | } | |
4138 | ||
4139 | static void io_async_task_func(struct callback_head *cb) | |
4140 | { | |
4141 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4142 | struct async_poll *apoll = req->apoll; | |
4143 | struct io_ring_ctx *ctx = req->ctx; | |
4144 | ||
4145 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4146 | ||
4147 | WARN_ON_ONCE(!list_empty(&req->apoll->poll.wait.entry)); | |
4148 | ||
4149 | if (hash_hashed(&req->hash_node)) { | |
4150 | spin_lock_irq(&ctx->completion_lock); | |
4151 | hash_del(&req->hash_node); | |
4152 | spin_unlock_irq(&ctx->completion_lock); | |
4153 | } | |
4154 | ||
4155 | /* restore ->work in case we need to retry again */ | |
4156 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4157 | ||
4158 | __set_current_state(TASK_RUNNING); | |
4159 | mutex_lock(&ctx->uring_lock); | |
4160 | __io_queue_sqe(req, NULL); | |
4161 | mutex_unlock(&ctx->uring_lock); | |
4162 | ||
4163 | kfree(apoll); | |
4164 | } | |
4165 | ||
4166 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
4167 | void *key) | |
4168 | { | |
4169 | struct io_kiocb *req = wait->private; | |
4170 | struct io_poll_iocb *poll = &req->apoll->poll; | |
4171 | ||
4172 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
4173 | key_to_poll(key)); | |
4174 | ||
4175 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
4176 | } | |
4177 | ||
4178 | static void io_poll_req_insert(struct io_kiocb *req) | |
4179 | { | |
4180 | struct io_ring_ctx *ctx = req->ctx; | |
4181 | struct hlist_head *list; | |
4182 | ||
4183 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
4184 | hlist_add_head(&req->hash_node, list); | |
4185 | } | |
4186 | ||
4187 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
4188 | struct io_poll_iocb *poll, | |
4189 | struct io_poll_table *ipt, __poll_t mask, | |
4190 | wait_queue_func_t wake_func) | |
4191 | __acquires(&ctx->completion_lock) | |
4192 | { | |
4193 | struct io_ring_ctx *ctx = req->ctx; | |
4194 | bool cancel = false; | |
4195 | ||
4196 | poll->file = req->file; | |
4197 | poll->head = NULL; | |
4198 | poll->done = poll->canceled = false; | |
4199 | poll->events = mask; | |
4200 | ||
4201 | ipt->pt._key = mask; | |
4202 | ipt->req = req; | |
4203 | ipt->error = -EINVAL; | |
4204 | ||
4205 | INIT_LIST_HEAD(&poll->wait.entry); | |
4206 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4207 | poll->wait.private = req; | |
4208 | ||
4209 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
4210 | ||
4211 | spin_lock_irq(&ctx->completion_lock); | |
4212 | if (likely(poll->head)) { | |
4213 | spin_lock(&poll->head->lock); | |
4214 | if (unlikely(list_empty(&poll->wait.entry))) { | |
4215 | if (ipt->error) | |
4216 | cancel = true; | |
4217 | ipt->error = 0; | |
4218 | mask = 0; | |
4219 | } | |
4220 | if (mask || ipt->error) | |
4221 | list_del_init(&poll->wait.entry); | |
4222 | else if (cancel) | |
4223 | WRITE_ONCE(poll->canceled, true); | |
4224 | else if (!poll->done) /* actually waiting for an event */ | |
4225 | io_poll_req_insert(req); | |
4226 | spin_unlock(&poll->head->lock); | |
4227 | } | |
4228 | ||
4229 | return mask; | |
4230 | } | |
4231 | ||
4232 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
4233 | { | |
4234 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
4235 | struct io_ring_ctx *ctx = req->ctx; | |
4236 | struct async_poll *apoll; | |
4237 | struct io_poll_table ipt; | |
4238 | __poll_t mask, ret; | |
4239 | ||
4240 | if (!req->file || !file_can_poll(req->file)) | |
4241 | return false; | |
4242 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
4243 | return false; | |
4244 | if (!def->pollin && !def->pollout) | |
4245 | return false; | |
4246 | ||
4247 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
4248 | if (unlikely(!apoll)) | |
4249 | return false; | |
4250 | ||
4251 | req->flags |= REQ_F_POLLED; | |
4252 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
4253 | ||
4254 | /* | |
4255 | * Don't need a reference here, as we're adding it to the task | |
4256 | * task_works list. If the task exits, the list is pruned. | |
4257 | */ | |
4258 | req->task = current; | |
4259 | req->apoll = apoll; | |
4260 | INIT_HLIST_NODE(&req->hash_node); | |
4261 | ||
8755d97a | 4262 | mask = 0; |
d7718a9d | 4263 | if (def->pollin) |
8755d97a | 4264 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
4265 | if (def->pollout) |
4266 | mask |= POLLOUT | POLLWRNORM; | |
4267 | mask |= POLLERR | POLLPRI; | |
4268 | ||
4269 | ipt.pt._qproc = io_async_queue_proc; | |
4270 | ||
4271 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
4272 | io_async_wake); | |
4273 | if (ret) { | |
4274 | ipt.error = 0; | |
4275 | apoll->poll.done = true; | |
4276 | spin_unlock_irq(&ctx->completion_lock); | |
4277 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
4278 | kfree(apoll); | |
4279 | return false; | |
4280 | } | |
4281 | spin_unlock_irq(&ctx->completion_lock); | |
4282 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
4283 | apoll->poll.events); | |
4284 | return true; | |
4285 | } | |
4286 | ||
4287 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
4288 | struct io_poll_iocb *poll) | |
221c5eb2 | 4289 | { |
b41e9852 | 4290 | bool do_complete = false; |
221c5eb2 JA |
4291 | |
4292 | spin_lock(&poll->head->lock); | |
4293 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
4294 | if (!list_empty(&poll->wait.entry)) { |
4295 | list_del_init(&poll->wait.entry); | |
b41e9852 | 4296 | do_complete = true; |
221c5eb2 JA |
4297 | } |
4298 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
4299 | return do_complete; |
4300 | } | |
4301 | ||
4302 | static bool io_poll_remove_one(struct io_kiocb *req) | |
4303 | { | |
4304 | bool do_complete; | |
4305 | ||
4306 | if (req->opcode == IORING_OP_POLL_ADD) { | |
4307 | do_complete = __io_poll_remove_one(req, &req->poll); | |
4308 | } else { | |
4309 | /* non-poll requests have submit ref still */ | |
4310 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
4311 | if (do_complete) | |
4312 | io_put_req(req); | |
4313 | } | |
4314 | ||
78076bb6 | 4315 | hash_del(&req->hash_node); |
d7718a9d | 4316 | |
b41e9852 JA |
4317 | if (do_complete) { |
4318 | io_cqring_fill_event(req, -ECANCELED); | |
4319 | io_commit_cqring(req->ctx); | |
4320 | req->flags |= REQ_F_COMP_LOCKED; | |
4321 | io_put_req(req); | |
4322 | } | |
4323 | ||
4324 | return do_complete; | |
221c5eb2 JA |
4325 | } |
4326 | ||
4327 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
4328 | { | |
78076bb6 | 4329 | struct hlist_node *tmp; |
221c5eb2 | 4330 | struct io_kiocb *req; |
78076bb6 | 4331 | int i; |
221c5eb2 JA |
4332 | |
4333 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
4334 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
4335 | struct hlist_head *list; | |
4336 | ||
4337 | list = &ctx->cancel_hash[i]; | |
4338 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
4339 | io_poll_remove_one(req); | |
221c5eb2 JA |
4340 | } |
4341 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 JA |
4342 | |
4343 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
4344 | } |
4345 | ||
47f46768 JA |
4346 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
4347 | { | |
78076bb6 | 4348 | struct hlist_head *list; |
47f46768 JA |
4349 | struct io_kiocb *req; |
4350 | ||
78076bb6 JA |
4351 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
4352 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
4353 | if (sqe_addr != req->user_data) |
4354 | continue; | |
4355 | if (io_poll_remove_one(req)) | |
eac406c6 | 4356 | return 0; |
b41e9852 | 4357 | return -EALREADY; |
47f46768 JA |
4358 | } |
4359 | ||
4360 | return -ENOENT; | |
4361 | } | |
4362 | ||
3529d8c2 JA |
4363 | static int io_poll_remove_prep(struct io_kiocb *req, |
4364 | const struct io_uring_sqe *sqe) | |
0969e783 | 4365 | { |
0969e783 JA |
4366 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4367 | return -EINVAL; | |
4368 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
4369 | sqe->poll_events) | |
4370 | return -EINVAL; | |
4371 | ||
4372 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
4373 | return 0; |
4374 | } | |
4375 | ||
221c5eb2 JA |
4376 | /* |
4377 | * Find a running poll command that matches one specified in sqe->addr, | |
4378 | * and remove it if found. | |
4379 | */ | |
fc4df999 | 4380 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
4381 | { |
4382 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 4383 | u64 addr; |
47f46768 | 4384 | int ret; |
221c5eb2 | 4385 | |
0969e783 | 4386 | addr = req->poll.addr; |
221c5eb2 | 4387 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 4388 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
4389 | spin_unlock_irq(&ctx->completion_lock); |
4390 | ||
78e19bbe | 4391 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
4392 | if (ret < 0) |
4393 | req_set_fail_links(req); | |
e65ef56d | 4394 | io_put_req(req); |
221c5eb2 JA |
4395 | return 0; |
4396 | } | |
4397 | ||
b0dd8a41 | 4398 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 4399 | { |
a197f664 JL |
4400 | struct io_ring_ctx *ctx = req->ctx; |
4401 | ||
8c838788 | 4402 | req->poll.done = true; |
b0a20349 | 4403 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 4404 | io_commit_cqring(ctx); |
221c5eb2 JA |
4405 | } |
4406 | ||
b41e9852 | 4407 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 4408 | { |
221c5eb2 | 4409 | struct io_ring_ctx *ctx = req->ctx; |
221c5eb2 | 4410 | |
221c5eb2 | 4411 | spin_lock_irq(&ctx->completion_lock); |
78076bb6 | 4412 | hash_del(&req->hash_node); |
b41e9852 JA |
4413 | io_poll_complete(req, req->result, 0); |
4414 | req->flags |= REQ_F_COMP_LOCKED; | |
4415 | io_put_req_find_next(req, nxt); | |
221c5eb2 JA |
4416 | spin_unlock_irq(&ctx->completion_lock); |
4417 | ||
8c838788 | 4418 | io_cqring_ev_posted(ctx); |
221c5eb2 JA |
4419 | } |
4420 | ||
b41e9852 | 4421 | static void io_poll_task_func(struct callback_head *cb) |
e94f141b | 4422 | { |
b41e9852 JA |
4423 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
4424 | struct io_kiocb *nxt = NULL; | |
e94f141b | 4425 | |
b41e9852 | 4426 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
4427 | if (nxt) { |
4428 | struct io_ring_ctx *ctx = nxt->ctx; | |
e94f141b | 4429 | |
d7718a9d | 4430 | mutex_lock(&ctx->uring_lock); |
b41e9852 | 4431 | __io_queue_sqe(nxt, NULL); |
d7718a9d | 4432 | mutex_unlock(&ctx->uring_lock); |
e94f141b | 4433 | } |
f0b493e6 JA |
4434 | } |
4435 | ||
221c5eb2 JA |
4436 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
4437 | void *key) | |
4438 | { | |
c2f2eb7d JA |
4439 | struct io_kiocb *req = wait->private; |
4440 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4441 | |
d7718a9d | 4442 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
4443 | } |
4444 | ||
221c5eb2 JA |
4445 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
4446 | struct poll_table_struct *p) | |
4447 | { | |
4448 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
4449 | ||
d7718a9d | 4450 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
4451 | } |
4452 | ||
3529d8c2 | 4453 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
4454 | { |
4455 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 4456 | u16 events; |
221c5eb2 JA |
4457 | |
4458 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
4459 | return -EINVAL; | |
4460 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
4461 | return -EINVAL; | |
09bb8394 JA |
4462 | if (!poll->file) |
4463 | return -EBADF; | |
221c5eb2 | 4464 | |
221c5eb2 JA |
4465 | events = READ_ONCE(sqe->poll_events); |
4466 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 4467 | |
d7718a9d JA |
4468 | /* |
4469 | * Don't need a reference here, as we're adding it to the task | |
4470 | * task_works list. If the task exits, the list is pruned. | |
4471 | */ | |
b41e9852 | 4472 | req->task = current; |
0969e783 JA |
4473 | return 0; |
4474 | } | |
4475 | ||
014db007 | 4476 | static int io_poll_add(struct io_kiocb *req) |
0969e783 JA |
4477 | { |
4478 | struct io_poll_iocb *poll = &req->poll; | |
4479 | struct io_ring_ctx *ctx = req->ctx; | |
4480 | struct io_poll_table ipt; | |
0969e783 | 4481 | __poll_t mask; |
0969e783 | 4482 | |
78076bb6 | 4483 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 4484 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 4485 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 4486 | |
d7718a9d JA |
4487 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
4488 | io_poll_wake); | |
221c5eb2 | 4489 | |
8c838788 | 4490 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 4491 | ipt.error = 0; |
b0dd8a41 | 4492 | io_poll_complete(req, mask, 0); |
221c5eb2 | 4493 | } |
221c5eb2 JA |
4494 | spin_unlock_irq(&ctx->completion_lock); |
4495 | ||
8c838788 JA |
4496 | if (mask) { |
4497 | io_cqring_ev_posted(ctx); | |
014db007 | 4498 | io_put_req(req); |
221c5eb2 | 4499 | } |
8c838788 | 4500 | return ipt.error; |
221c5eb2 JA |
4501 | } |
4502 | ||
5262f567 JA |
4503 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
4504 | { | |
ad8a48ac JA |
4505 | struct io_timeout_data *data = container_of(timer, |
4506 | struct io_timeout_data, timer); | |
4507 | struct io_kiocb *req = data->req; | |
4508 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
4509 | unsigned long flags; |
4510 | ||
5262f567 JA |
4511 | atomic_inc(&ctx->cq_timeouts); |
4512 | ||
4513 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 4514 | /* |
11365043 JA |
4515 | * We could be racing with timeout deletion. If the list is empty, |
4516 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 4517 | */ |
842f9612 | 4518 | if (!list_empty(&req->list)) { |
11365043 | 4519 | struct io_kiocb *prev; |
5262f567 | 4520 | |
11365043 JA |
4521 | /* |
4522 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4523 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4524 | * pointer will be increased, otherwise other timeout reqs may |
4525 | * return in advance without waiting for enough wait_nr. | |
4526 | */ | |
4527 | prev = req; | |
4528 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4529 | prev->sequence++; | |
11365043 | 4530 | list_del_init(&req->list); |
11365043 | 4531 | } |
5262f567 | 4532 | |
78e19bbe | 4533 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4534 | io_commit_cqring(ctx); |
4535 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4536 | ||
4537 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4538 | req_set_fail_links(req); |
5262f567 JA |
4539 | io_put_req(req); |
4540 | return HRTIMER_NORESTART; | |
4541 | } | |
4542 | ||
47f46768 JA |
4543 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4544 | { | |
4545 | struct io_kiocb *req; | |
4546 | int ret = -ENOENT; | |
4547 | ||
4548 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4549 | if (user_data == req->user_data) { | |
4550 | list_del_init(&req->list); | |
4551 | ret = 0; | |
4552 | break; | |
4553 | } | |
4554 | } | |
4555 | ||
4556 | if (ret == -ENOENT) | |
4557 | return ret; | |
4558 | ||
2d28390a | 4559 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4560 | if (ret == -1) |
4561 | return -EALREADY; | |
4562 | ||
4e88d6e7 | 4563 | req_set_fail_links(req); |
47f46768 JA |
4564 | io_cqring_fill_event(req, -ECANCELED); |
4565 | io_put_req(req); | |
4566 | return 0; | |
4567 | } | |
4568 | ||
3529d8c2 JA |
4569 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4570 | const struct io_uring_sqe *sqe) | |
b29472ee | 4571 | { |
b29472ee JA |
4572 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4573 | return -EINVAL; | |
4574 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4575 | return -EINVAL; | |
4576 | ||
4577 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4578 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4579 | if (req->timeout.flags) | |
4580 | return -EINVAL; | |
4581 | ||
b29472ee JA |
4582 | return 0; |
4583 | } | |
4584 | ||
11365043 JA |
4585 | /* |
4586 | * Remove or update an existing timeout command | |
4587 | */ | |
fc4df999 | 4588 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4589 | { |
4590 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4591 | int ret; |
11365043 | 4592 | |
11365043 | 4593 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4594 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4595 | |
47f46768 | 4596 | io_cqring_fill_event(req, ret); |
11365043 JA |
4597 | io_commit_cqring(ctx); |
4598 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4599 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4600 | if (ret < 0) |
4601 | req_set_fail_links(req); | |
ec9c02ad | 4602 | io_put_req(req); |
11365043 | 4603 | return 0; |
5262f567 JA |
4604 | } |
4605 | ||
3529d8c2 | 4606 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4607 | bool is_timeout_link) |
5262f567 | 4608 | { |
ad8a48ac | 4609 | struct io_timeout_data *data; |
a41525ab | 4610 | unsigned flags; |
5262f567 | 4611 | |
ad8a48ac | 4612 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4613 | return -EINVAL; |
ad8a48ac | 4614 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4615 | return -EINVAL; |
2d28390a JA |
4616 | if (sqe->off && is_timeout_link) |
4617 | return -EINVAL; | |
a41525ab JA |
4618 | flags = READ_ONCE(sqe->timeout_flags); |
4619 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4620 | return -EINVAL; |
bdf20073 | 4621 | |
26a61679 JA |
4622 | req->timeout.count = READ_ONCE(sqe->off); |
4623 | ||
3529d8c2 | 4624 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4625 | return -ENOMEM; |
4626 | ||
4627 | data = &req->io->timeout; | |
ad8a48ac | 4628 | data->req = req; |
ad8a48ac JA |
4629 | req->flags |= REQ_F_TIMEOUT; |
4630 | ||
4631 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4632 | return -EFAULT; |
4633 | ||
11365043 | 4634 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4635 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4636 | else |
ad8a48ac | 4637 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4638 | |
ad8a48ac JA |
4639 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4640 | return 0; | |
4641 | } | |
4642 | ||
fc4df999 | 4643 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac JA |
4644 | { |
4645 | unsigned count; | |
4646 | struct io_ring_ctx *ctx = req->ctx; | |
4647 | struct io_timeout_data *data; | |
4648 | struct list_head *entry; | |
4649 | unsigned span = 0; | |
ad8a48ac | 4650 | |
2d28390a | 4651 | data = &req->io->timeout; |
93bd25bb | 4652 | |
5262f567 JA |
4653 | /* |
4654 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4655 | * timeout event to be satisfied. If it isn't set, then this is |
4656 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4657 | */ |
26a61679 | 4658 | count = req->timeout.count; |
93bd25bb JA |
4659 | if (!count) { |
4660 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4661 | spin_lock_irq(&ctx->completion_lock); | |
4662 | entry = ctx->timeout_list.prev; | |
4663 | goto add; | |
4664 | } | |
5262f567 JA |
4665 | |
4666 | req->sequence = ctx->cached_sq_head + count - 1; | |
2d28390a | 4667 | data->seq_offset = count; |
5262f567 JA |
4668 | |
4669 | /* | |
4670 | * Insertion sort, ensuring the first entry in the list is always | |
4671 | * the one we need first. | |
4672 | */ | |
5262f567 JA |
4673 | spin_lock_irq(&ctx->completion_lock); |
4674 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4675 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5da0fb1a | 4676 | unsigned nxt_sq_head; |
4677 | long long tmp, tmp_nxt; | |
2d28390a | 4678 | u32 nxt_offset = nxt->io->timeout.seq_offset; |
5262f567 | 4679 | |
93bd25bb JA |
4680 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4681 | continue; | |
4682 | ||
5da0fb1a | 4683 | /* |
4684 | * Since cached_sq_head + count - 1 can overflow, use type long | |
4685 | * long to store it. | |
4686 | */ | |
4687 | tmp = (long long)ctx->cached_sq_head + count - 1; | |
cc42e0ac PB |
4688 | nxt_sq_head = nxt->sequence - nxt_offset + 1; |
4689 | tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1; | |
5da0fb1a | 4690 | |
4691 | /* | |
4692 | * cached_sq_head may overflow, and it will never overflow twice | |
4693 | * once there is some timeout req still be valid. | |
4694 | */ | |
4695 | if (ctx->cached_sq_head < nxt_sq_head) | |
8b07a65a | 4696 | tmp += UINT_MAX; |
5da0fb1a | 4697 | |
a1f58ba4 | 4698 | if (tmp > tmp_nxt) |
5262f567 | 4699 | break; |
a1f58ba4 | 4700 | |
4701 | /* | |
4702 | * Sequence of reqs after the insert one and itself should | |
4703 | * be adjusted because each timeout req consumes a slot. | |
4704 | */ | |
4705 | span++; | |
4706 | nxt->sequence++; | |
5262f567 | 4707 | } |
a1f58ba4 | 4708 | req->sequence -= span; |
93bd25bb | 4709 | add: |
5262f567 | 4710 | list_add(&req->list, entry); |
ad8a48ac JA |
4711 | data->timer.function = io_timeout_fn; |
4712 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4713 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4714 | return 0; |
4715 | } | |
5262f567 | 4716 | |
62755e35 JA |
4717 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4718 | { | |
4719 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4720 | ||
4721 | return req->user_data == (unsigned long) data; | |
4722 | } | |
4723 | ||
e977d6d3 | 4724 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4725 | { |
62755e35 | 4726 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4727 | int ret = 0; |
4728 | ||
62755e35 JA |
4729 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4730 | switch (cancel_ret) { | |
4731 | case IO_WQ_CANCEL_OK: | |
4732 | ret = 0; | |
4733 | break; | |
4734 | case IO_WQ_CANCEL_RUNNING: | |
4735 | ret = -EALREADY; | |
4736 | break; | |
4737 | case IO_WQ_CANCEL_NOTFOUND: | |
4738 | ret = -ENOENT; | |
4739 | break; | |
4740 | } | |
4741 | ||
e977d6d3 JA |
4742 | return ret; |
4743 | } | |
4744 | ||
47f46768 JA |
4745 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4746 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 4747 | int success_ret) |
47f46768 JA |
4748 | { |
4749 | unsigned long flags; | |
4750 | int ret; | |
4751 | ||
4752 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4753 | if (ret != -ENOENT) { | |
4754 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4755 | goto done; | |
4756 | } | |
4757 | ||
4758 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4759 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4760 | if (ret != -ENOENT) | |
4761 | goto done; | |
4762 | ret = io_poll_cancel(ctx, sqe_addr); | |
4763 | done: | |
b0dd8a41 JA |
4764 | if (!ret) |
4765 | ret = success_ret; | |
47f46768 JA |
4766 | io_cqring_fill_event(req, ret); |
4767 | io_commit_cqring(ctx); | |
4768 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4769 | io_cqring_ev_posted(ctx); | |
4770 | ||
4e88d6e7 JA |
4771 | if (ret < 0) |
4772 | req_set_fail_links(req); | |
014db007 | 4773 | io_put_req(req); |
47f46768 JA |
4774 | } |
4775 | ||
3529d8c2 JA |
4776 | static int io_async_cancel_prep(struct io_kiocb *req, |
4777 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4778 | { |
fbf23849 | 4779 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4780 | return -EINVAL; |
4781 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4782 | sqe->cancel_flags) | |
4783 | return -EINVAL; | |
4784 | ||
fbf23849 JA |
4785 | req->cancel.addr = READ_ONCE(sqe->addr); |
4786 | return 0; | |
4787 | } | |
4788 | ||
014db007 | 4789 | static int io_async_cancel(struct io_kiocb *req) |
fbf23849 JA |
4790 | { |
4791 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 4792 | |
014db007 | 4793 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
4794 | return 0; |
4795 | } | |
4796 | ||
05f3fb3c JA |
4797 | static int io_files_update_prep(struct io_kiocb *req, |
4798 | const struct io_uring_sqe *sqe) | |
4799 | { | |
4800 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4801 | return -EINVAL; | |
4802 | ||
4803 | req->files_update.offset = READ_ONCE(sqe->off); | |
4804 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4805 | if (!req->files_update.nr_args) | |
4806 | return -EINVAL; | |
4807 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4808 | return 0; | |
4809 | } | |
4810 | ||
4811 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4812 | { |
4813 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4814 | struct io_uring_files_update up; |
4815 | int ret; | |
fbf23849 | 4816 | |
f86cd20c | 4817 | if (force_nonblock) |
05f3fb3c | 4818 | return -EAGAIN; |
05f3fb3c JA |
4819 | |
4820 | up.offset = req->files_update.offset; | |
4821 | up.fds = req->files_update.arg; | |
4822 | ||
4823 | mutex_lock(&ctx->uring_lock); | |
4824 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4825 | mutex_unlock(&ctx->uring_lock); | |
4826 | ||
4827 | if (ret < 0) | |
4828 | req_set_fail_links(req); | |
4829 | io_cqring_add_event(req, ret); | |
4830 | io_put_req(req); | |
5262f567 JA |
4831 | return 0; |
4832 | } | |
4833 | ||
3529d8c2 JA |
4834 | static int io_req_defer_prep(struct io_kiocb *req, |
4835 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4836 | { |
e781573e | 4837 | ssize_t ret = 0; |
f67676d1 | 4838 | |
f1d96a8f PB |
4839 | if (!sqe) |
4840 | return 0; | |
4841 | ||
f86cd20c JA |
4842 | if (io_op_defs[req->opcode].file_table) { |
4843 | ret = io_grab_files(req); | |
4844 | if (unlikely(ret)) | |
4845 | return ret; | |
4846 | } | |
4847 | ||
cccf0ee8 JA |
4848 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4849 | ||
d625c6ee | 4850 | switch (req->opcode) { |
e781573e JA |
4851 | case IORING_OP_NOP: |
4852 | break; | |
f67676d1 JA |
4853 | case IORING_OP_READV: |
4854 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4855 | case IORING_OP_READ: |
3529d8c2 | 4856 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4857 | break; |
4858 | case IORING_OP_WRITEV: | |
4859 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4860 | case IORING_OP_WRITE: |
3529d8c2 | 4861 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4862 | break; |
0969e783 | 4863 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4864 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4865 | break; |
4866 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4867 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4868 | break; |
8ed8d3c3 | 4869 | case IORING_OP_FSYNC: |
3529d8c2 | 4870 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4871 | break; |
4872 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4873 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4874 | break; |
03b1230c | 4875 | case IORING_OP_SENDMSG: |
fddaface | 4876 | case IORING_OP_SEND: |
3529d8c2 | 4877 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4878 | break; |
4879 | case IORING_OP_RECVMSG: | |
fddaface | 4880 | case IORING_OP_RECV: |
3529d8c2 | 4881 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4882 | break; |
f499a021 | 4883 | case IORING_OP_CONNECT: |
3529d8c2 | 4884 | ret = io_connect_prep(req, sqe); |
f499a021 | 4885 | break; |
2d28390a | 4886 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4887 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4888 | break; |
b29472ee | 4889 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4890 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4891 | break; |
fbf23849 | 4892 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4893 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4894 | break; |
2d28390a | 4895 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4896 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4897 | break; |
8ed8d3c3 | 4898 | case IORING_OP_ACCEPT: |
3529d8c2 | 4899 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4900 | break; |
d63d1b5e JA |
4901 | case IORING_OP_FALLOCATE: |
4902 | ret = io_fallocate_prep(req, sqe); | |
4903 | break; | |
15b71abe JA |
4904 | case IORING_OP_OPENAT: |
4905 | ret = io_openat_prep(req, sqe); | |
4906 | break; | |
b5dba59e JA |
4907 | case IORING_OP_CLOSE: |
4908 | ret = io_close_prep(req, sqe); | |
4909 | break; | |
05f3fb3c JA |
4910 | case IORING_OP_FILES_UPDATE: |
4911 | ret = io_files_update_prep(req, sqe); | |
4912 | break; | |
eddc7ef5 JA |
4913 | case IORING_OP_STATX: |
4914 | ret = io_statx_prep(req, sqe); | |
4915 | break; | |
4840e418 JA |
4916 | case IORING_OP_FADVISE: |
4917 | ret = io_fadvise_prep(req, sqe); | |
4918 | break; | |
c1ca757b JA |
4919 | case IORING_OP_MADVISE: |
4920 | ret = io_madvise_prep(req, sqe); | |
4921 | break; | |
cebdb986 JA |
4922 | case IORING_OP_OPENAT2: |
4923 | ret = io_openat2_prep(req, sqe); | |
4924 | break; | |
3e4827b0 JA |
4925 | case IORING_OP_EPOLL_CTL: |
4926 | ret = io_epoll_ctl_prep(req, sqe); | |
4927 | break; | |
7d67af2c PB |
4928 | case IORING_OP_SPLICE: |
4929 | ret = io_splice_prep(req, sqe); | |
4930 | break; | |
ddf0322d JA |
4931 | case IORING_OP_PROVIDE_BUFFERS: |
4932 | ret = io_provide_buffers_prep(req, sqe); | |
4933 | break; | |
067524e9 JA |
4934 | case IORING_OP_REMOVE_BUFFERS: |
4935 | ret = io_remove_buffers_prep(req, sqe); | |
4936 | break; | |
f67676d1 | 4937 | default: |
e781573e JA |
4938 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4939 | req->opcode); | |
4940 | ret = -EINVAL; | |
b7bb4f7d | 4941 | break; |
f67676d1 JA |
4942 | } |
4943 | ||
b7bb4f7d | 4944 | return ret; |
f67676d1 JA |
4945 | } |
4946 | ||
3529d8c2 | 4947 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 4948 | { |
a197f664 | 4949 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 4950 | int ret; |
de0617e4 | 4951 | |
9d858b21 BL |
4952 | /* Still need defer if there is pending req in defer list. */ |
4953 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
4954 | return 0; |
4955 | ||
3529d8c2 | 4956 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
4957 | return -EAGAIN; |
4958 | ||
3529d8c2 | 4959 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 4960 | if (ret < 0) |
2d28390a | 4961 | return ret; |
2d28390a | 4962 | |
de0617e4 | 4963 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 4964 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 4965 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
4966 | return 0; |
4967 | } | |
4968 | ||
915967f6 | 4969 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
4970 | list_add_tail(&req->list, &ctx->defer_list); |
4971 | spin_unlock_irq(&ctx->completion_lock); | |
4972 | return -EIOCBQUEUED; | |
4973 | } | |
4974 | ||
99bc4c38 PB |
4975 | static void io_cleanup_req(struct io_kiocb *req) |
4976 | { | |
4977 | struct io_async_ctx *io = req->io; | |
4978 | ||
4979 | switch (req->opcode) { | |
4980 | case IORING_OP_READV: | |
4981 | case IORING_OP_READ_FIXED: | |
4982 | case IORING_OP_READ: | |
bcda7baa JA |
4983 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4984 | kfree((void *)(unsigned long)req->rw.addr); | |
4985 | /* fallthrough */ | |
99bc4c38 PB |
4986 | case IORING_OP_WRITEV: |
4987 | case IORING_OP_WRITE_FIXED: | |
4988 | case IORING_OP_WRITE: | |
4989 | if (io->rw.iov != io->rw.fast_iov) | |
4990 | kfree(io->rw.iov); | |
4991 | break; | |
99bc4c38 | 4992 | case IORING_OP_RECVMSG: |
52de1fe1 JA |
4993 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4994 | kfree(req->sr_msg.kbuf); | |
4995 | /* fallthrough */ | |
4996 | case IORING_OP_SENDMSG: | |
99bc4c38 PB |
4997 | if (io->msg.iov != io->msg.fast_iov) |
4998 | kfree(io->msg.iov); | |
4999 | break; | |
bcda7baa JA |
5000 | case IORING_OP_RECV: |
5001 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
5002 | kfree(req->sr_msg.kbuf); | |
5003 | break; | |
8fef80bf PB |
5004 | case IORING_OP_OPENAT: |
5005 | case IORING_OP_OPENAT2: | |
5006 | case IORING_OP_STATX: | |
5007 | putname(req->open.filename); | |
5008 | break; | |
7d67af2c PB |
5009 | case IORING_OP_SPLICE: |
5010 | io_put_file(req, req->splice.file_in, | |
5011 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5012 | break; | |
99bc4c38 PB |
5013 | } |
5014 | ||
5015 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
5016 | } | |
5017 | ||
3529d8c2 | 5018 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
014db007 | 5019 | bool force_nonblock) |
2b188cc1 | 5020 | { |
a197f664 | 5021 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 5022 | int ret; |
2b188cc1 | 5023 | |
d625c6ee | 5024 | switch (req->opcode) { |
2b188cc1 | 5025 | case IORING_OP_NOP: |
78e19bbe | 5026 | ret = io_nop(req); |
2b188cc1 JA |
5027 | break; |
5028 | case IORING_OP_READV: | |
edafccee | 5029 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5030 | case IORING_OP_READ: |
3529d8c2 JA |
5031 | if (sqe) { |
5032 | ret = io_read_prep(req, sqe, force_nonblock); | |
5033 | if (ret < 0) | |
5034 | break; | |
5035 | } | |
014db007 | 5036 | ret = io_read(req, force_nonblock); |
edafccee | 5037 | break; |
3529d8c2 | 5038 | case IORING_OP_WRITEV: |
edafccee | 5039 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5040 | case IORING_OP_WRITE: |
3529d8c2 JA |
5041 | if (sqe) { |
5042 | ret = io_write_prep(req, sqe, force_nonblock); | |
5043 | if (ret < 0) | |
5044 | break; | |
5045 | } | |
014db007 | 5046 | ret = io_write(req, force_nonblock); |
2b188cc1 | 5047 | break; |
c992fe29 | 5048 | case IORING_OP_FSYNC: |
3529d8c2 JA |
5049 | if (sqe) { |
5050 | ret = io_prep_fsync(req, sqe); | |
5051 | if (ret < 0) | |
5052 | break; | |
5053 | } | |
014db007 | 5054 | ret = io_fsync(req, force_nonblock); |
c992fe29 | 5055 | break; |
221c5eb2 | 5056 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
5057 | if (sqe) { |
5058 | ret = io_poll_add_prep(req, sqe); | |
5059 | if (ret) | |
5060 | break; | |
5061 | } | |
014db007 | 5062 | ret = io_poll_add(req); |
221c5eb2 JA |
5063 | break; |
5064 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
5065 | if (sqe) { |
5066 | ret = io_poll_remove_prep(req, sqe); | |
5067 | if (ret < 0) | |
5068 | break; | |
5069 | } | |
fc4df999 | 5070 | ret = io_poll_remove(req); |
221c5eb2 | 5071 | break; |
5d17b4a4 | 5072 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
5073 | if (sqe) { |
5074 | ret = io_prep_sfr(req, sqe); | |
5075 | if (ret < 0) | |
5076 | break; | |
5077 | } | |
014db007 | 5078 | ret = io_sync_file_range(req, force_nonblock); |
5d17b4a4 | 5079 | break; |
0fa03c62 | 5080 | case IORING_OP_SENDMSG: |
fddaface | 5081 | case IORING_OP_SEND: |
3529d8c2 JA |
5082 | if (sqe) { |
5083 | ret = io_sendmsg_prep(req, sqe); | |
5084 | if (ret < 0) | |
5085 | break; | |
5086 | } | |
fddaface | 5087 | if (req->opcode == IORING_OP_SENDMSG) |
014db007 | 5088 | ret = io_sendmsg(req, force_nonblock); |
fddaface | 5089 | else |
014db007 | 5090 | ret = io_send(req, force_nonblock); |
0fa03c62 | 5091 | break; |
aa1fa28f | 5092 | case IORING_OP_RECVMSG: |
fddaface | 5093 | case IORING_OP_RECV: |
3529d8c2 JA |
5094 | if (sqe) { |
5095 | ret = io_recvmsg_prep(req, sqe); | |
5096 | if (ret) | |
5097 | break; | |
5098 | } | |
fddaface | 5099 | if (req->opcode == IORING_OP_RECVMSG) |
014db007 | 5100 | ret = io_recvmsg(req, force_nonblock); |
fddaface | 5101 | else |
014db007 | 5102 | ret = io_recv(req, force_nonblock); |
aa1fa28f | 5103 | break; |
5262f567 | 5104 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
5105 | if (sqe) { |
5106 | ret = io_timeout_prep(req, sqe, false); | |
5107 | if (ret) | |
5108 | break; | |
5109 | } | |
fc4df999 | 5110 | ret = io_timeout(req); |
5262f567 | 5111 | break; |
11365043 | 5112 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
5113 | if (sqe) { |
5114 | ret = io_timeout_remove_prep(req, sqe); | |
5115 | if (ret) | |
5116 | break; | |
5117 | } | |
fc4df999 | 5118 | ret = io_timeout_remove(req); |
11365043 | 5119 | break; |
17f2fe35 | 5120 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
5121 | if (sqe) { |
5122 | ret = io_accept_prep(req, sqe); | |
5123 | if (ret) | |
5124 | break; | |
5125 | } | |
014db007 | 5126 | ret = io_accept(req, force_nonblock); |
17f2fe35 | 5127 | break; |
f8e85cf2 | 5128 | case IORING_OP_CONNECT: |
3529d8c2 JA |
5129 | if (sqe) { |
5130 | ret = io_connect_prep(req, sqe); | |
5131 | if (ret) | |
5132 | break; | |
5133 | } | |
014db007 | 5134 | ret = io_connect(req, force_nonblock); |
f8e85cf2 | 5135 | break; |
62755e35 | 5136 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
5137 | if (sqe) { |
5138 | ret = io_async_cancel_prep(req, sqe); | |
5139 | if (ret) | |
5140 | break; | |
5141 | } | |
014db007 | 5142 | ret = io_async_cancel(req); |
62755e35 | 5143 | break; |
d63d1b5e JA |
5144 | case IORING_OP_FALLOCATE: |
5145 | if (sqe) { | |
5146 | ret = io_fallocate_prep(req, sqe); | |
5147 | if (ret) | |
5148 | break; | |
5149 | } | |
014db007 | 5150 | ret = io_fallocate(req, force_nonblock); |
d63d1b5e | 5151 | break; |
15b71abe JA |
5152 | case IORING_OP_OPENAT: |
5153 | if (sqe) { | |
5154 | ret = io_openat_prep(req, sqe); | |
5155 | if (ret) | |
5156 | break; | |
5157 | } | |
014db007 | 5158 | ret = io_openat(req, force_nonblock); |
15b71abe | 5159 | break; |
b5dba59e JA |
5160 | case IORING_OP_CLOSE: |
5161 | if (sqe) { | |
5162 | ret = io_close_prep(req, sqe); | |
5163 | if (ret) | |
5164 | break; | |
5165 | } | |
014db007 | 5166 | ret = io_close(req, force_nonblock); |
b5dba59e | 5167 | break; |
05f3fb3c JA |
5168 | case IORING_OP_FILES_UPDATE: |
5169 | if (sqe) { | |
5170 | ret = io_files_update_prep(req, sqe); | |
5171 | if (ret) | |
5172 | break; | |
5173 | } | |
5174 | ret = io_files_update(req, force_nonblock); | |
5175 | break; | |
eddc7ef5 JA |
5176 | case IORING_OP_STATX: |
5177 | if (sqe) { | |
5178 | ret = io_statx_prep(req, sqe); | |
5179 | if (ret) | |
5180 | break; | |
5181 | } | |
014db007 | 5182 | ret = io_statx(req, force_nonblock); |
eddc7ef5 | 5183 | break; |
4840e418 JA |
5184 | case IORING_OP_FADVISE: |
5185 | if (sqe) { | |
5186 | ret = io_fadvise_prep(req, sqe); | |
5187 | if (ret) | |
5188 | break; | |
5189 | } | |
014db007 | 5190 | ret = io_fadvise(req, force_nonblock); |
4840e418 | 5191 | break; |
c1ca757b JA |
5192 | case IORING_OP_MADVISE: |
5193 | if (sqe) { | |
5194 | ret = io_madvise_prep(req, sqe); | |
5195 | if (ret) | |
5196 | break; | |
5197 | } | |
014db007 | 5198 | ret = io_madvise(req, force_nonblock); |
c1ca757b | 5199 | break; |
cebdb986 JA |
5200 | case IORING_OP_OPENAT2: |
5201 | if (sqe) { | |
5202 | ret = io_openat2_prep(req, sqe); | |
5203 | if (ret) | |
5204 | break; | |
5205 | } | |
014db007 | 5206 | ret = io_openat2(req, force_nonblock); |
cebdb986 | 5207 | break; |
3e4827b0 JA |
5208 | case IORING_OP_EPOLL_CTL: |
5209 | if (sqe) { | |
5210 | ret = io_epoll_ctl_prep(req, sqe); | |
5211 | if (ret) | |
5212 | break; | |
5213 | } | |
014db007 | 5214 | ret = io_epoll_ctl(req, force_nonblock); |
3e4827b0 | 5215 | break; |
7d67af2c PB |
5216 | case IORING_OP_SPLICE: |
5217 | if (sqe) { | |
5218 | ret = io_splice_prep(req, sqe); | |
5219 | if (ret < 0) | |
5220 | break; | |
5221 | } | |
014db007 | 5222 | ret = io_splice(req, force_nonblock); |
7d67af2c | 5223 | break; |
ddf0322d JA |
5224 | case IORING_OP_PROVIDE_BUFFERS: |
5225 | if (sqe) { | |
5226 | ret = io_provide_buffers_prep(req, sqe); | |
5227 | if (ret) | |
5228 | break; | |
5229 | } | |
5230 | ret = io_provide_buffers(req, force_nonblock); | |
5231 | break; | |
067524e9 JA |
5232 | case IORING_OP_REMOVE_BUFFERS: |
5233 | if (sqe) { | |
5234 | ret = io_remove_buffers_prep(req, sqe); | |
5235 | if (ret) | |
5236 | break; | |
5237 | } | |
5238 | ret = io_remove_buffers(req, force_nonblock); | |
3e4827b0 | 5239 | break; |
2b188cc1 JA |
5240 | default: |
5241 | ret = -EINVAL; | |
5242 | break; | |
5243 | } | |
5244 | ||
def596e9 JA |
5245 | if (ret) |
5246 | return ret; | |
5247 | ||
5248 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
5249 | const bool in_async = io_wq_current_is_worker(); |
5250 | ||
9e645e11 | 5251 | if (req->result == -EAGAIN) |
def596e9 JA |
5252 | return -EAGAIN; |
5253 | ||
11ba820b JA |
5254 | /* workqueue context doesn't hold uring_lock, grab it now */ |
5255 | if (in_async) | |
5256 | mutex_lock(&ctx->uring_lock); | |
5257 | ||
def596e9 | 5258 | io_iopoll_req_issued(req); |
11ba820b JA |
5259 | |
5260 | if (in_async) | |
5261 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
5262 | } |
5263 | ||
5264 | return 0; | |
2b188cc1 JA |
5265 | } |
5266 | ||
561fb04a | 5267 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 5268 | { |
561fb04a | 5269 | struct io_wq_work *work = *workptr; |
2b188cc1 | 5270 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a | 5271 | int ret = 0; |
2b188cc1 | 5272 | |
0c9d5ccd JA |
5273 | /* if NO_CANCEL is set, we must still run the work */ |
5274 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
5275 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 5276 | ret = -ECANCELED; |
0c9d5ccd | 5277 | } |
31b51510 | 5278 | |
561fb04a | 5279 | if (!ret) { |
561fb04a | 5280 | do { |
014db007 | 5281 | ret = io_issue_sqe(req, NULL, false); |
561fb04a JA |
5282 | /* |
5283 | * We can get EAGAIN for polled IO even though we're | |
5284 | * forcing a sync submission from here, since we can't | |
5285 | * wait for request slots on the block side. | |
5286 | */ | |
5287 | if (ret != -EAGAIN) | |
5288 | break; | |
5289 | cond_resched(); | |
5290 | } while (1); | |
5291 | } | |
31b51510 | 5292 | |
561fb04a | 5293 | if (ret) { |
4e88d6e7 | 5294 | req_set_fail_links(req); |
78e19bbe | 5295 | io_cqring_add_event(req, ret); |
817869d2 | 5296 | io_put_req(req); |
edafccee | 5297 | } |
2b188cc1 | 5298 | |
e9fd9396 | 5299 | io_steal_work(req, workptr); |
2b188cc1 JA |
5300 | } |
5301 | ||
15b71abe | 5302 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 5303 | { |
d3656344 | 5304 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 5305 | return 0; |
0b5faf6b | 5306 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
5307 | return 0; |
5308 | return 1; | |
09bb8394 JA |
5309 | } |
5310 | ||
65e19f54 JA |
5311 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
5312 | int index) | |
5313 | { | |
5314 | struct fixed_file_table *table; | |
5315 | ||
05f3fb3c JA |
5316 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
5317 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
5318 | } |
5319 | ||
8da11c19 PB |
5320 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
5321 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 5322 | { |
a197f664 | 5323 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 5324 | struct file *file; |
09bb8394 | 5325 | |
8da11c19 | 5326 | if (fixed) { |
05f3fb3c | 5327 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
5328 | (unsigned) fd >= ctx->nr_user_files)) |
5329 | return -EBADF; | |
b7620121 | 5330 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
5331 | file = io_file_from_index(ctx, fd); |
5332 | if (!file) | |
08a45173 | 5333 | return -EBADF; |
05f3fb3c | 5334 | percpu_ref_get(&ctx->file_data->refs); |
09bb8394 | 5335 | } else { |
c826bd7a | 5336 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
5337 | file = __io_file_get(state, fd); |
5338 | if (unlikely(!file)) | |
09bb8394 JA |
5339 | return -EBADF; |
5340 | } | |
5341 | ||
8da11c19 | 5342 | *out_file = file; |
09bb8394 JA |
5343 | return 0; |
5344 | } | |
5345 | ||
8da11c19 PB |
5346 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
5347 | const struct io_uring_sqe *sqe) | |
5348 | { | |
5349 | unsigned flags; | |
5350 | int fd; | |
5351 | bool fixed; | |
5352 | ||
5353 | flags = READ_ONCE(sqe->flags); | |
5354 | fd = READ_ONCE(sqe->fd); | |
5355 | ||
5356 | if (!io_req_needs_file(req, fd)) | |
5357 | return 0; | |
5358 | ||
5359 | fixed = (flags & IOSQE_FIXED_FILE); | |
5360 | if (unlikely(!fixed && req->needs_fixed_file)) | |
5361 | return -EBADF; | |
5362 | ||
5363 | return io_file_get(state, req, fd, &req->file, fixed); | |
5364 | } | |
5365 | ||
a197f664 | 5366 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
5367 | { |
5368 | int ret = -EBADF; | |
a197f664 | 5369 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 5370 | |
f86cd20c JA |
5371 | if (req->work.files) |
5372 | return 0; | |
b14cca0c | 5373 | if (!ctx->ring_file) |
b5dba59e JA |
5374 | return -EBADF; |
5375 | ||
fcb323cc JA |
5376 | rcu_read_lock(); |
5377 | spin_lock_irq(&ctx->inflight_lock); | |
5378 | /* | |
5379 | * We use the f_ops->flush() handler to ensure that we can flush | |
5380 | * out work accessing these files if the fd is closed. Check if | |
5381 | * the fd has changed since we started down this path, and disallow | |
5382 | * this operation if it has. | |
5383 | */ | |
b14cca0c | 5384 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
5385 | list_add(&req->inflight_entry, &ctx->inflight_list); |
5386 | req->flags |= REQ_F_INFLIGHT; | |
5387 | req->work.files = current->files; | |
5388 | ret = 0; | |
5389 | } | |
5390 | spin_unlock_irq(&ctx->inflight_lock); | |
5391 | rcu_read_unlock(); | |
5392 | ||
5393 | return ret; | |
5394 | } | |
5395 | ||
2665abfd | 5396 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 5397 | { |
ad8a48ac JA |
5398 | struct io_timeout_data *data = container_of(timer, |
5399 | struct io_timeout_data, timer); | |
5400 | struct io_kiocb *req = data->req; | |
2665abfd JA |
5401 | struct io_ring_ctx *ctx = req->ctx; |
5402 | struct io_kiocb *prev = NULL; | |
5403 | unsigned long flags; | |
2665abfd JA |
5404 | |
5405 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5406 | ||
5407 | /* | |
5408 | * We don't expect the list to be empty, that will only happen if we | |
5409 | * race with the completion of the linked work. | |
5410 | */ | |
4493233e PB |
5411 | if (!list_empty(&req->link_list)) { |
5412 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
5413 | link_list); | |
5d960724 | 5414 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 5415 | list_del_init(&req->link_list); |
5d960724 JA |
5416 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
5417 | } else | |
76a46e06 | 5418 | prev = NULL; |
2665abfd JA |
5419 | } |
5420 | ||
5421 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5422 | ||
5423 | if (prev) { | |
4e88d6e7 | 5424 | req_set_fail_links(prev); |
014db007 | 5425 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
76a46e06 | 5426 | io_put_req(prev); |
47f46768 JA |
5427 | } else { |
5428 | io_cqring_add_event(req, -ETIME); | |
5429 | io_put_req(req); | |
2665abfd | 5430 | } |
2665abfd JA |
5431 | return HRTIMER_NORESTART; |
5432 | } | |
5433 | ||
ad8a48ac | 5434 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 5435 | { |
76a46e06 | 5436 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 5437 | |
76a46e06 JA |
5438 | /* |
5439 | * If the list is now empty, then our linked request finished before | |
5440 | * we got a chance to setup the timer | |
5441 | */ | |
5442 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 5443 | if (!list_empty(&req->link_list)) { |
2d28390a | 5444 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 5445 | |
ad8a48ac JA |
5446 | data->timer.function = io_link_timeout_fn; |
5447 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
5448 | data->mode); | |
2665abfd | 5449 | } |
76a46e06 | 5450 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 5451 | |
2665abfd | 5452 | /* drop submission reference */ |
76a46e06 JA |
5453 | io_put_req(req); |
5454 | } | |
2665abfd | 5455 | |
ad8a48ac | 5456 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
5457 | { |
5458 | struct io_kiocb *nxt; | |
5459 | ||
5460 | if (!(req->flags & REQ_F_LINK)) | |
5461 | return NULL; | |
d7718a9d JA |
5462 | /* for polled retry, if flag is set, we already went through here */ |
5463 | if (req->flags & REQ_F_POLLED) | |
5464 | return NULL; | |
2665abfd | 5465 | |
4493233e PB |
5466 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
5467 | link_list); | |
d625c6ee | 5468 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 5469 | return NULL; |
2665abfd | 5470 | |
76a46e06 | 5471 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 5472 | return nxt; |
2665abfd JA |
5473 | } |
5474 | ||
3529d8c2 | 5475 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 5476 | { |
4a0a7a18 | 5477 | struct io_kiocb *linked_timeout; |
4bc4494e | 5478 | struct io_kiocb *nxt; |
193155c8 | 5479 | const struct cred *old_creds = NULL; |
e0c5c576 | 5480 | int ret; |
2b188cc1 | 5481 | |
4a0a7a18 JA |
5482 | again: |
5483 | linked_timeout = io_prep_linked_timeout(req); | |
5484 | ||
193155c8 JA |
5485 | if (req->work.creds && req->work.creds != current_cred()) { |
5486 | if (old_creds) | |
5487 | revert_creds(old_creds); | |
5488 | if (old_creds == req->work.creds) | |
5489 | old_creds = NULL; /* restored original creds */ | |
5490 | else | |
5491 | old_creds = override_creds(req->work.creds); | |
5492 | } | |
5493 | ||
014db007 | 5494 | ret = io_issue_sqe(req, sqe, true); |
491381ce JA |
5495 | |
5496 | /* | |
5497 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
5498 | * doesn't support non-blocking read/write attempts | |
5499 | */ | |
5500 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
5501 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
5502 | if (io_arm_poll_handler(req)) { |
5503 | if (linked_timeout) | |
5504 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 5505 | goto exit; |
d7718a9d | 5506 | } |
86a761f8 | 5507 | punt: |
f86cd20c | 5508 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
5509 | ret = io_grab_files(req); |
5510 | if (ret) | |
5511 | goto err; | |
2b188cc1 | 5512 | } |
bbad27b2 PB |
5513 | |
5514 | /* | |
5515 | * Queued up for async execution, worker will release | |
5516 | * submit reference when the iocb is actually submitted. | |
5517 | */ | |
5518 | io_queue_async_work(req); | |
4bc4494e | 5519 | goto exit; |
2b188cc1 | 5520 | } |
e65ef56d | 5521 | |
fcb323cc | 5522 | err: |
4bc4494e | 5523 | nxt = NULL; |
76a46e06 | 5524 | /* drop submission reference */ |
2a44f467 | 5525 | io_put_req_find_next(req, &nxt); |
e65ef56d | 5526 | |
f9bd67f6 | 5527 | if (linked_timeout) { |
76a46e06 | 5528 | if (!ret) |
f9bd67f6 | 5529 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 5530 | else |
f9bd67f6 | 5531 | io_put_req(linked_timeout); |
76a46e06 JA |
5532 | } |
5533 | ||
e65ef56d | 5534 | /* and drop final reference, if we failed */ |
9e645e11 | 5535 | if (ret) { |
78e19bbe | 5536 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5537 | req_set_fail_links(req); |
e65ef56d | 5538 | io_put_req(req); |
9e645e11 | 5539 | } |
4a0a7a18 JA |
5540 | if (nxt) { |
5541 | req = nxt; | |
86a761f8 PB |
5542 | |
5543 | if (req->flags & REQ_F_FORCE_ASYNC) | |
5544 | goto punt; | |
4a0a7a18 JA |
5545 | goto again; |
5546 | } | |
4bc4494e | 5547 | exit: |
193155c8 JA |
5548 | if (old_creds) |
5549 | revert_creds(old_creds); | |
2b188cc1 JA |
5550 | } |
5551 | ||
3529d8c2 | 5552 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5553 | { |
5554 | int ret; | |
5555 | ||
3529d8c2 | 5556 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5557 | if (ret) { |
5558 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5559 | fail_req: |
78e19bbe | 5560 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5561 | req_set_fail_links(req); |
78e19bbe | 5562 | io_double_put_req(req); |
4fe2c963 | 5563 | } |
2550878f | 5564 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5565 | ret = io_req_defer_prep(req, sqe); |
5566 | if (unlikely(ret < 0)) | |
5567 | goto fail_req; | |
ce35a47a JA |
5568 | /* |
5569 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5570 | * to async execution. | |
5571 | */ | |
5572 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5573 | io_queue_async_work(req); | |
5574 | } else { | |
3529d8c2 | 5575 | __io_queue_sqe(req, sqe); |
ce35a47a | 5576 | } |
4fe2c963 JL |
5577 | } |
5578 | ||
1b4a51b6 | 5579 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5580 | { |
94ae5e77 | 5581 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5582 | io_cqring_add_event(req, -ECANCELED); |
5583 | io_double_put_req(req); | |
5584 | } else | |
3529d8c2 | 5585 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5586 | } |
5587 | ||
4e88d6e7 | 5588 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
bcda7baa JA |
5589 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ |
5590 | IOSQE_BUFFER_SELECT) | |
9e645e11 | 5591 | |
3529d8c2 JA |
5592 | static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5593 | struct io_submit_state *state, struct io_kiocb **link) | |
9e645e11 | 5594 | { |
a197f664 | 5595 | struct io_ring_ctx *ctx = req->ctx; |
32fe525b | 5596 | unsigned int sqe_flags; |
75c6a039 | 5597 | int ret, id; |
9e645e11 | 5598 | |
32fe525b | 5599 | sqe_flags = READ_ONCE(sqe->flags); |
9e645e11 JA |
5600 | |
5601 | /* enforce forwards compatibility on users */ | |
32fe525b | 5602 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
9e645e11 | 5603 | ret = -EINVAL; |
196be95c | 5604 | goto err_req; |
9e645e11 JA |
5605 | } |
5606 | ||
bcda7baa JA |
5607 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
5608 | !io_op_defs[req->opcode].buffer_select) { | |
5609 | ret = -EOPNOTSUPP; | |
5610 | goto err_req; | |
5611 | } | |
5612 | ||
75c6a039 JA |
5613 | id = READ_ONCE(sqe->personality); |
5614 | if (id) { | |
193155c8 JA |
5615 | req->work.creds = idr_find(&ctx->personality_idr, id); |
5616 | if (unlikely(!req->work.creds)) { | |
75c6a039 JA |
5617 | ret = -EINVAL; |
5618 | goto err_req; | |
5619 | } | |
193155c8 | 5620 | get_cred(req->work.creds); |
75c6a039 JA |
5621 | } |
5622 | ||
6b47ee6e | 5623 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
8da11c19 | 5624 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | |
bcda7baa JA |
5625 | IOSQE_ASYNC | IOSQE_FIXED_FILE | |
5626 | IOSQE_BUFFER_SELECT); | |
9e645e11 | 5627 | |
3529d8c2 | 5628 | ret = io_req_set_file(state, req, sqe); |
9e645e11 JA |
5629 | if (unlikely(ret)) { |
5630 | err_req: | |
78e19bbe JA |
5631 | io_cqring_add_event(req, ret); |
5632 | io_double_put_req(req); | |
2e6e1fde | 5633 | return false; |
9e645e11 JA |
5634 | } |
5635 | ||
9e645e11 JA |
5636 | /* |
5637 | * If we already have a head request, queue this one for async | |
5638 | * submittal once the head completes. If we don't have a head but | |
5639 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5640 | * submitted sync once the chain is complete. If none of those | |
5641 | * conditions are true (normal request), then just queue it. | |
5642 | */ | |
5643 | if (*link) { | |
9d76377f | 5644 | struct io_kiocb *head = *link; |
4e88d6e7 | 5645 | |
8cdf2193 PB |
5646 | /* |
5647 | * Taking sequential execution of a link, draining both sides | |
5648 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5649 | * requests in the link. So, it drains the head and the | |
5650 | * next after the link request. The last one is done via | |
5651 | * drain_next flag to persist the effect across calls. | |
5652 | */ | |
711be031 PB |
5653 | if (sqe_flags & IOSQE_IO_DRAIN) { |
5654 | head->flags |= REQ_F_IO_DRAIN; | |
5655 | ctx->drain_next = 1; | |
5656 | } | |
b7bb4f7d | 5657 | if (io_alloc_async_ctx(req)) { |
9e645e11 JA |
5658 | ret = -EAGAIN; |
5659 | goto err_req; | |
5660 | } | |
5661 | ||
3529d8c2 | 5662 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5663 | if (ret) { |
4e88d6e7 | 5664 | /* fail even hard links since we don't submit */ |
9d76377f | 5665 | head->flags |= REQ_F_FAIL_LINK; |
f67676d1 | 5666 | goto err_req; |
2d28390a | 5667 | } |
9d76377f PB |
5668 | trace_io_uring_link(ctx, req, head); |
5669 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5670 | |
5671 | /* last request of a link, enqueue the link */ | |
5672 | if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) { | |
5673 | io_queue_link_head(head); | |
5674 | *link = NULL; | |
5675 | } | |
9e645e11 | 5676 | } else { |
711be031 PB |
5677 | if (unlikely(ctx->drain_next)) { |
5678 | req->flags |= REQ_F_IO_DRAIN; | |
5679 | req->ctx->drain_next = 0; | |
5680 | } | |
5681 | if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) { | |
5682 | req->flags |= REQ_F_LINK; | |
711be031 | 5683 | INIT_LIST_HEAD(&req->link_list); |
f1d96a8f PB |
5684 | |
5685 | if (io_alloc_async_ctx(req)) { | |
5686 | ret = -EAGAIN; | |
5687 | goto err_req; | |
5688 | } | |
711be031 PB |
5689 | ret = io_req_defer_prep(req, sqe); |
5690 | if (ret) | |
5691 | req->flags |= REQ_F_FAIL_LINK; | |
5692 | *link = req; | |
5693 | } else { | |
5694 | io_queue_sqe(req, sqe); | |
5695 | } | |
9e645e11 | 5696 | } |
2e6e1fde PB |
5697 | |
5698 | return true; | |
9e645e11 JA |
5699 | } |
5700 | ||
9a56a232 JA |
5701 | /* |
5702 | * Batched submission is done, ensure local IO is flushed out. | |
5703 | */ | |
5704 | static void io_submit_state_end(struct io_submit_state *state) | |
5705 | { | |
5706 | blk_finish_plug(&state->plug); | |
3d6770fb | 5707 | io_file_put(state); |
2579f913 | 5708 | if (state->free_reqs) |
6c8a3134 | 5709 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5710 | } |
5711 | ||
5712 | /* | |
5713 | * Start submission side cache. | |
5714 | */ | |
5715 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5716 | unsigned int max_ios) |
9a56a232 JA |
5717 | { |
5718 | blk_start_plug(&state->plug); | |
2579f913 | 5719 | state->free_reqs = 0; |
9a56a232 JA |
5720 | state->file = NULL; |
5721 | state->ios_left = max_ios; | |
5722 | } | |
5723 | ||
2b188cc1 JA |
5724 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5725 | { | |
75b28aff | 5726 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5727 | |
caf582c6 PB |
5728 | /* |
5729 | * Ensure any loads from the SQEs are done at this point, | |
5730 | * since once we write the new head, the application could | |
5731 | * write new data to them. | |
5732 | */ | |
5733 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5734 | } |
5735 | ||
2b188cc1 | 5736 | /* |
3529d8c2 | 5737 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5738 | * that is mapped by userspace. This means that care needs to be taken to |
5739 | * ensure that reads are stable, as we cannot rely on userspace always | |
5740 | * being a good citizen. If members of the sqe are validated and then later | |
5741 | * used, it's important that those reads are done through READ_ONCE() to | |
5742 | * prevent a re-load down the line. | |
5743 | */ | |
3529d8c2 JA |
5744 | static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req, |
5745 | const struct io_uring_sqe **sqe_ptr) | |
2b188cc1 | 5746 | { |
75b28aff | 5747 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5748 | unsigned head; |
5749 | ||
5750 | /* | |
5751 | * The cached sq head (or cq tail) serves two purposes: | |
5752 | * | |
5753 | * 1) allows us to batch the cost of updating the user visible | |
5754 | * head updates. | |
5755 | * 2) allows the kernel side to track the head on its own, even | |
5756 | * though the application is the one updating it. | |
5757 | */ | |
ee7d46d9 | 5758 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
9835d6fa | 5759 | if (likely(head < ctx->sq_entries)) { |
cf6fd4bd PB |
5760 | /* |
5761 | * All io need record the previous position, if LINK vs DARIN, | |
5762 | * it can be used to mark the position of the first IO in the | |
5763 | * link list. | |
5764 | */ | |
5765 | req->sequence = ctx->cached_sq_head; | |
3529d8c2 JA |
5766 | *sqe_ptr = &ctx->sq_sqes[head]; |
5767 | req->opcode = READ_ONCE((*sqe_ptr)->opcode); | |
5768 | req->user_data = READ_ONCE((*sqe_ptr)->user_data); | |
2b188cc1 JA |
5769 | ctx->cached_sq_head++; |
5770 | return true; | |
5771 | } | |
5772 | ||
5773 | /* drop invalid entries */ | |
5774 | ctx->cached_sq_head++; | |
498ccd9e | 5775 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5776 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
2b188cc1 JA |
5777 | return false; |
5778 | } | |
5779 | ||
fb5ccc98 | 5780 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
ae9428ca PB |
5781 | struct file *ring_file, int ring_fd, |
5782 | struct mm_struct **mm, bool async) | |
6c271ce2 JA |
5783 | { |
5784 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5785 | struct io_kiocb *link = NULL; |
9e645e11 | 5786 | int i, submitted = 0; |
95a1b3ff | 5787 | bool mm_fault = false; |
6c271ce2 | 5788 | |
c4a2ed72 | 5789 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5790 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5791 | if (!list_empty(&ctx->cq_overflow_list) && | |
5792 | !io_cqring_overflow_flush(ctx, false)) | |
5793 | return -EBUSY; | |
5794 | } | |
6c271ce2 | 5795 | |
ee7d46d9 PB |
5796 | /* make sure SQ entry isn't read before tail */ |
5797 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5798 | |
2b85edfc PB |
5799 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5800 | return -EAGAIN; | |
6c271ce2 JA |
5801 | |
5802 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5803 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5804 | statep = &state; |
5805 | } | |
5806 | ||
b14cca0c PB |
5807 | ctx->ring_fd = ring_fd; |
5808 | ctx->ring_file = ring_file; | |
5809 | ||
6c271ce2 | 5810 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5811 | const struct io_uring_sqe *sqe; |
196be95c | 5812 | struct io_kiocb *req; |
1cb1edb2 | 5813 | int err; |
fb5ccc98 | 5814 | |
196be95c PB |
5815 | req = io_get_req(ctx, statep); |
5816 | if (unlikely(!req)) { | |
5817 | if (!submitted) | |
5818 | submitted = -EAGAIN; | |
fb5ccc98 | 5819 | break; |
196be95c | 5820 | } |
3529d8c2 | 5821 | if (!io_get_sqring(ctx, req, &sqe)) { |
2b85edfc | 5822 | __io_req_do_free(req); |
196be95c PB |
5823 | break; |
5824 | } | |
fb5ccc98 | 5825 | |
d3656344 JA |
5826 | /* will complete beyond this point, count as submitted */ |
5827 | submitted++; | |
5828 | ||
5829 | if (unlikely(req->opcode >= IORING_OP_LAST)) { | |
1cb1edb2 PB |
5830 | err = -EINVAL; |
5831 | fail_req: | |
5832 | io_cqring_add_event(req, err); | |
d3656344 | 5833 | io_double_put_req(req); |
196be95c PB |
5834 | break; |
5835 | } | |
fb5ccc98 | 5836 | |
d3656344 | 5837 | if (io_op_defs[req->opcode].needs_mm && !*mm) { |
95a1b3ff | 5838 | mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm); |
1cb1edb2 PB |
5839 | if (unlikely(mm_fault)) { |
5840 | err = -EFAULT; | |
5841 | goto fail_req; | |
95a1b3ff | 5842 | } |
1cb1edb2 PB |
5843 | use_mm(ctx->sqo_mm); |
5844 | *mm = ctx->sqo_mm; | |
9e645e11 | 5845 | } |
9e645e11 | 5846 | |
cf6fd4bd | 5847 | req->needs_fixed_file = async; |
354420f7 JA |
5848 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5849 | true, async); | |
3529d8c2 | 5850 | if (!io_submit_sqe(req, sqe, statep, &link)) |
2e6e1fde | 5851 | break; |
6c271ce2 JA |
5852 | } |
5853 | ||
9466f437 PB |
5854 | if (unlikely(submitted != nr)) { |
5855 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5856 | ||
5857 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5858 | } | |
9e645e11 | 5859 | if (link) |
1b4a51b6 | 5860 | io_queue_link_head(link); |
6c271ce2 JA |
5861 | if (statep) |
5862 | io_submit_state_end(&state); | |
5863 | ||
ae9428ca PB |
5864 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5865 | io_commit_sqring(ctx); | |
5866 | ||
6c271ce2 JA |
5867 | return submitted; |
5868 | } | |
5869 | ||
5870 | static int io_sq_thread(void *data) | |
5871 | { | |
6c271ce2 JA |
5872 | struct io_ring_ctx *ctx = data; |
5873 | struct mm_struct *cur_mm = NULL; | |
181e448d | 5874 | const struct cred *old_cred; |
6c271ce2 JA |
5875 | mm_segment_t old_fs; |
5876 | DEFINE_WAIT(wait); | |
6c271ce2 | 5877 | unsigned long timeout; |
bdcd3eab | 5878 | int ret = 0; |
6c271ce2 | 5879 | |
206aefde | 5880 | complete(&ctx->completions[1]); |
a4c0b3de | 5881 | |
6c271ce2 JA |
5882 | old_fs = get_fs(); |
5883 | set_fs(USER_DS); | |
181e448d | 5884 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5885 | |
bdcd3eab | 5886 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5887 | while (!kthread_should_park()) { |
fb5ccc98 | 5888 | unsigned int to_submit; |
6c271ce2 | 5889 | |
bdcd3eab | 5890 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5891 | unsigned nr_events = 0; |
5892 | ||
bdcd3eab XW |
5893 | mutex_lock(&ctx->uring_lock); |
5894 | if (!list_empty(&ctx->poll_list)) | |
5895 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5896 | else | |
6c271ce2 | 5897 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5898 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5899 | } |
5900 | ||
fb5ccc98 | 5901 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5902 | |
5903 | /* | |
5904 | * If submit got -EBUSY, flag us as needing the application | |
5905 | * to enter the kernel to reap and flush events. | |
5906 | */ | |
5907 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5908 | /* |
5909 | * Drop cur_mm before scheduling, we can't hold it for | |
5910 | * long periods (or over schedule()). Do this before | |
5911 | * adding ourselves to the waitqueue, as the unuse/drop | |
5912 | * may sleep. | |
5913 | */ | |
5914 | if (cur_mm) { | |
5915 | unuse_mm(cur_mm); | |
5916 | mmput(cur_mm); | |
5917 | cur_mm = NULL; | |
5918 | } | |
5919 | ||
6c271ce2 JA |
5920 | /* |
5921 | * We're polling. If we're within the defined idle | |
5922 | * period, then let us spin without work before going | |
c1edbf5f JA |
5923 | * to sleep. The exception is if we got EBUSY doing |
5924 | * more IO, we should wait for the application to | |
5925 | * reap events and wake us up. | |
6c271ce2 | 5926 | */ |
bdcd3eab | 5927 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5928 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5929 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5930 | if (current->task_works) |
5931 | task_work_run(); | |
9831a90c | 5932 | cond_resched(); |
6c271ce2 JA |
5933 | continue; |
5934 | } | |
5935 | ||
6c271ce2 JA |
5936 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5937 | TASK_INTERRUPTIBLE); | |
5938 | ||
bdcd3eab XW |
5939 | /* |
5940 | * While doing polled IO, before going to sleep, we need | |
5941 | * to check if there are new reqs added to poll_list, it | |
5942 | * is because reqs may have been punted to io worker and | |
5943 | * will be added to poll_list later, hence check the | |
5944 | * poll_list again. | |
5945 | */ | |
5946 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5947 | !list_empty_careful(&ctx->poll_list)) { | |
5948 | finish_wait(&ctx->sqo_wait, &wait); | |
5949 | continue; | |
5950 | } | |
5951 | ||
6c271ce2 | 5952 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 5953 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
5954 | /* make sure to read SQ tail after writing flags */ |
5955 | smp_mb(); | |
6c271ce2 | 5956 | |
fb5ccc98 | 5957 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 5958 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 5959 | if (kthread_should_park()) { |
6c271ce2 JA |
5960 | finish_wait(&ctx->sqo_wait, &wait); |
5961 | break; | |
5962 | } | |
b41e9852 JA |
5963 | if (current->task_works) { |
5964 | task_work_run(); | |
5965 | continue; | |
5966 | } | |
6c271ce2 JA |
5967 | if (signal_pending(current)) |
5968 | flush_signals(current); | |
5969 | schedule(); | |
5970 | finish_wait(&ctx->sqo_wait, &wait); | |
5971 | ||
75b28aff | 5972 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5973 | continue; |
5974 | } | |
5975 | finish_wait(&ctx->sqo_wait, &wait); | |
5976 | ||
75b28aff | 5977 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5978 | } |
5979 | ||
8a4955ff | 5980 | mutex_lock(&ctx->uring_lock); |
1d7bb1d5 | 5981 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true); |
8a4955ff | 5982 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 5983 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
5984 | } |
5985 | ||
b41e9852 JA |
5986 | if (current->task_works) |
5987 | task_work_run(); | |
5988 | ||
6c271ce2 JA |
5989 | set_fs(old_fs); |
5990 | if (cur_mm) { | |
5991 | unuse_mm(cur_mm); | |
5992 | mmput(cur_mm); | |
5993 | } | |
181e448d | 5994 | revert_creds(old_cred); |
06058632 | 5995 | |
2bbcd6d3 | 5996 | kthread_parkme(); |
06058632 | 5997 | |
6c271ce2 JA |
5998 | return 0; |
5999 | } | |
6000 | ||
bda52162 JA |
6001 | struct io_wait_queue { |
6002 | struct wait_queue_entry wq; | |
6003 | struct io_ring_ctx *ctx; | |
6004 | unsigned to_wait; | |
6005 | unsigned nr_timeouts; | |
6006 | }; | |
6007 | ||
1d7bb1d5 | 6008 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
6009 | { |
6010 | struct io_ring_ctx *ctx = iowq->ctx; | |
6011 | ||
6012 | /* | |
d195a66e | 6013 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6014 | * started waiting. For timeouts, we always want to return to userspace, |
6015 | * regardless of event count. | |
6016 | */ | |
1d7bb1d5 | 6017 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
6018 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6019 | } | |
6020 | ||
6021 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6022 | int wake_flags, void *key) | |
6023 | { | |
6024 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6025 | wq); | |
6026 | ||
1d7bb1d5 JA |
6027 | /* use noflush == true, as we can't safely rely on locking context */ |
6028 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
6029 | return -1; |
6030 | ||
6031 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6032 | } | |
6033 | ||
2b188cc1 JA |
6034 | /* |
6035 | * Wait until events become available, if we don't already have some. The | |
6036 | * application must reap them itself, as they reside on the shared cq ring. | |
6037 | */ | |
6038 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
6039 | const sigset_t __user *sig, size_t sigsz) | |
6040 | { | |
bda52162 JA |
6041 | struct io_wait_queue iowq = { |
6042 | .wq = { | |
6043 | .private = current, | |
6044 | .func = io_wake_function, | |
6045 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6046 | }, | |
6047 | .ctx = ctx, | |
6048 | .to_wait = min_events, | |
6049 | }; | |
75b28aff | 6050 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 6051 | int ret = 0; |
2b188cc1 | 6052 | |
b41e9852 JA |
6053 | do { |
6054 | if (io_cqring_events(ctx, false) >= min_events) | |
6055 | return 0; | |
6056 | if (!current->task_works) | |
6057 | break; | |
6058 | task_work_run(); | |
6059 | } while (1); | |
2b188cc1 JA |
6060 | |
6061 | if (sig) { | |
9e75ad5d AB |
6062 | #ifdef CONFIG_COMPAT |
6063 | if (in_compat_syscall()) | |
6064 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6065 | sigsz); |
9e75ad5d AB |
6066 | else |
6067 | #endif | |
b772434b | 6068 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6069 | |
2b188cc1 JA |
6070 | if (ret) |
6071 | return ret; | |
6072 | } | |
6073 | ||
bda52162 | 6074 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6075 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
6076 | do { |
6077 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
6078 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
6079 | if (current->task_works) |
6080 | task_work_run(); | |
1d7bb1d5 | 6081 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
6082 | break; |
6083 | schedule(); | |
6084 | if (signal_pending(current)) { | |
e9ffa5c2 | 6085 | ret = -EINTR; |
bda52162 JA |
6086 | break; |
6087 | } | |
6088 | } while (1); | |
6089 | finish_wait(&ctx->wait, &iowq.wq); | |
6090 | ||
e9ffa5c2 | 6091 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6092 | |
75b28aff | 6093 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6094 | } |
6095 | ||
6b06314c JA |
6096 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6097 | { | |
6098 | #if defined(CONFIG_UNIX) | |
6099 | if (ctx->ring_sock) { | |
6100 | struct sock *sock = ctx->ring_sock->sk; | |
6101 | struct sk_buff *skb; | |
6102 | ||
6103 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6104 | kfree_skb(skb); | |
6105 | } | |
6106 | #else | |
6107 | int i; | |
6108 | ||
65e19f54 JA |
6109 | for (i = 0; i < ctx->nr_user_files; i++) { |
6110 | struct file *file; | |
6111 | ||
6112 | file = io_file_from_index(ctx, i); | |
6113 | if (file) | |
6114 | fput(file); | |
6115 | } | |
6b06314c JA |
6116 | #endif |
6117 | } | |
6118 | ||
05f3fb3c JA |
6119 | static void io_file_ref_kill(struct percpu_ref *ref) |
6120 | { | |
6121 | struct fixed_file_data *data; | |
6122 | ||
6123 | data = container_of(ref, struct fixed_file_data, refs); | |
6124 | complete(&data->done); | |
6125 | } | |
6126 | ||
805b13ad | 6127 | static void io_file_ref_exit_and_free(struct work_struct *work) |
c1e2148f | 6128 | { |
805b13ad JA |
6129 | struct fixed_file_data *data; |
6130 | ||
6131 | data = container_of(work, struct fixed_file_data, ref_work); | |
c1e2148f | 6132 | |
c1e2148f | 6133 | /* |
805b13ad JA |
6134 | * Ensure any percpu-ref atomic switch callback has run, it could have |
6135 | * been in progress when the files were being unregistered. Once | |
6136 | * that's done, we can safely exit and free the ref and containing | |
6137 | * data structure. | |
c1e2148f | 6138 | */ |
805b13ad JA |
6139 | rcu_barrier(); |
6140 | percpu_ref_exit(&data->refs); | |
6141 | kfree(data); | |
c1e2148f JA |
6142 | } |
6143 | ||
6b06314c JA |
6144 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6145 | { | |
05f3fb3c | 6146 | struct fixed_file_data *data = ctx->file_data; |
65e19f54 JA |
6147 | unsigned nr_tables, i; |
6148 | ||
05f3fb3c | 6149 | if (!data) |
6b06314c JA |
6150 | return -ENXIO; |
6151 | ||
05f3fb3c | 6152 | percpu_ref_kill_and_confirm(&data->refs, io_file_ref_kill); |
e46a7950 | 6153 | flush_work(&data->ref_work); |
2faf852d JA |
6154 | wait_for_completion(&data->done); |
6155 | io_ring_file_ref_flush(data); | |
05f3fb3c | 6156 | |
6b06314c | 6157 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
6158 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
6159 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
6160 | kfree(data->table[i].files); |
6161 | kfree(data->table); | |
805b13ad JA |
6162 | INIT_WORK(&data->ref_work, io_file_ref_exit_and_free); |
6163 | queue_work(system_wq, &data->ref_work); | |
05f3fb3c | 6164 | ctx->file_data = NULL; |
6b06314c JA |
6165 | ctx->nr_user_files = 0; |
6166 | return 0; | |
6167 | } | |
6168 | ||
6c271ce2 JA |
6169 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
6170 | { | |
6171 | if (ctx->sqo_thread) { | |
206aefde | 6172 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
6173 | /* |
6174 | * The park is a bit of a work-around, without it we get | |
6175 | * warning spews on shutdown with SQPOLL set and affinity | |
6176 | * set to a single CPU. | |
6177 | */ | |
06058632 | 6178 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
6179 | kthread_stop(ctx->sqo_thread); |
6180 | ctx->sqo_thread = NULL; | |
6181 | } | |
6182 | } | |
6183 | ||
6b06314c JA |
6184 | static void io_finish_async(struct io_ring_ctx *ctx) |
6185 | { | |
6c271ce2 JA |
6186 | io_sq_thread_stop(ctx); |
6187 | ||
561fb04a JA |
6188 | if (ctx->io_wq) { |
6189 | io_wq_destroy(ctx->io_wq); | |
6190 | ctx->io_wq = NULL; | |
6b06314c JA |
6191 | } |
6192 | } | |
6193 | ||
6194 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
6195 | /* |
6196 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
6197 | * the io_uring can be safely unregistered on process exit, even if we have | |
6198 | * loops in the file referencing. | |
6199 | */ | |
6200 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
6201 | { | |
6202 | struct sock *sk = ctx->ring_sock->sk; | |
6203 | struct scm_fp_list *fpl; | |
6204 | struct sk_buff *skb; | |
08a45173 | 6205 | int i, nr_files; |
6b06314c JA |
6206 | |
6207 | if (!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) { | |
6208 | unsigned long inflight = ctx->user->unix_inflight + nr; | |
6209 | ||
6210 | if (inflight > task_rlimit(current, RLIMIT_NOFILE)) | |
6211 | return -EMFILE; | |
6212 | } | |
6213 | ||
6214 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); | |
6215 | if (!fpl) | |
6216 | return -ENOMEM; | |
6217 | ||
6218 | skb = alloc_skb(0, GFP_KERNEL); | |
6219 | if (!skb) { | |
6220 | kfree(fpl); | |
6221 | return -ENOMEM; | |
6222 | } | |
6223 | ||
6224 | skb->sk = sk; | |
6b06314c | 6225 | |
08a45173 | 6226 | nr_files = 0; |
6b06314c JA |
6227 | fpl->user = get_uid(ctx->user); |
6228 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
6229 | struct file *file = io_file_from_index(ctx, i + offset); |
6230 | ||
6231 | if (!file) | |
08a45173 | 6232 | continue; |
65e19f54 | 6233 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
6234 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
6235 | nr_files++; | |
6b06314c JA |
6236 | } |
6237 | ||
08a45173 JA |
6238 | if (nr_files) { |
6239 | fpl->max = SCM_MAX_FD; | |
6240 | fpl->count = nr_files; | |
6241 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 6242 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
6243 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
6244 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 6245 | |
08a45173 JA |
6246 | for (i = 0; i < nr_files; i++) |
6247 | fput(fpl->fp[i]); | |
6248 | } else { | |
6249 | kfree_skb(skb); | |
6250 | kfree(fpl); | |
6251 | } | |
6b06314c JA |
6252 | |
6253 | return 0; | |
6254 | } | |
6255 | ||
6256 | /* | |
6257 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
6258 | * causes regular reference counting to break down. We rely on the UNIX | |
6259 | * garbage collection to take care of this problem for us. | |
6260 | */ | |
6261 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6262 | { | |
6263 | unsigned left, total; | |
6264 | int ret = 0; | |
6265 | ||
6266 | total = 0; | |
6267 | left = ctx->nr_user_files; | |
6268 | while (left) { | |
6269 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
6270 | |
6271 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
6272 | if (ret) | |
6273 | break; | |
6274 | left -= this_files; | |
6275 | total += this_files; | |
6276 | } | |
6277 | ||
6278 | if (!ret) | |
6279 | return 0; | |
6280 | ||
6281 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
6282 | struct file *file = io_file_from_index(ctx, total); |
6283 | ||
6284 | if (file) | |
6285 | fput(file); | |
6b06314c JA |
6286 | total++; |
6287 | } | |
6288 | ||
6289 | return ret; | |
6290 | } | |
6291 | #else | |
6292 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
6293 | { | |
6294 | return 0; | |
6295 | } | |
6296 | #endif | |
6297 | ||
65e19f54 JA |
6298 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
6299 | unsigned nr_files) | |
6300 | { | |
6301 | int i; | |
6302 | ||
6303 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6304 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6305 | unsigned this_files; |
6306 | ||
6307 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
6308 | table->files = kcalloc(this_files, sizeof(struct file *), | |
6309 | GFP_KERNEL); | |
6310 | if (!table->files) | |
6311 | break; | |
6312 | nr_files -= this_files; | |
6313 | } | |
6314 | ||
6315 | if (i == nr_tables) | |
6316 | return 0; | |
6317 | ||
6318 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 6319 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
6320 | kfree(table->files); |
6321 | } | |
6322 | return 1; | |
6323 | } | |
6324 | ||
05f3fb3c JA |
6325 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
6326 | { | |
6327 | #if defined(CONFIG_UNIX) | |
6328 | struct sock *sock = ctx->ring_sock->sk; | |
6329 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
6330 | struct sk_buff *skb; | |
6331 | int i; | |
6332 | ||
6333 | __skb_queue_head_init(&list); | |
6334 | ||
6335 | /* | |
6336 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
6337 | * remove this entry and rearrange the file array. | |
6338 | */ | |
6339 | skb = skb_dequeue(head); | |
6340 | while (skb) { | |
6341 | struct scm_fp_list *fp; | |
6342 | ||
6343 | fp = UNIXCB(skb).fp; | |
6344 | for (i = 0; i < fp->count; i++) { | |
6345 | int left; | |
6346 | ||
6347 | if (fp->fp[i] != file) | |
6348 | continue; | |
6349 | ||
6350 | unix_notinflight(fp->user, fp->fp[i]); | |
6351 | left = fp->count - 1 - i; | |
6352 | if (left) { | |
6353 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
6354 | left * sizeof(struct file *)); | |
6355 | } | |
6356 | fp->count--; | |
6357 | if (!fp->count) { | |
6358 | kfree_skb(skb); | |
6359 | skb = NULL; | |
6360 | } else { | |
6361 | __skb_queue_tail(&list, skb); | |
6362 | } | |
6363 | fput(file); | |
6364 | file = NULL; | |
6365 | break; | |
6366 | } | |
6367 | ||
6368 | if (!file) | |
6369 | break; | |
6370 | ||
6371 | __skb_queue_tail(&list, skb); | |
6372 | ||
6373 | skb = skb_dequeue(head); | |
6374 | } | |
6375 | ||
6376 | if (skb_peek(&list)) { | |
6377 | spin_lock_irq(&head->lock); | |
6378 | while ((skb = __skb_dequeue(&list)) != NULL) | |
6379 | __skb_queue_tail(head, skb); | |
6380 | spin_unlock_irq(&head->lock); | |
6381 | } | |
6382 | #else | |
6383 | fput(file); | |
6384 | #endif | |
6385 | } | |
6386 | ||
6387 | struct io_file_put { | |
6388 | struct llist_node llist; | |
6389 | struct file *file; | |
05f3fb3c JA |
6390 | }; |
6391 | ||
2faf852d | 6392 | static void io_ring_file_ref_flush(struct fixed_file_data *data) |
65e19f54 | 6393 | { |
05f3fb3c | 6394 | struct io_file_put *pfile, *tmp; |
05f3fb3c | 6395 | struct llist_node *node; |
65e19f54 | 6396 | |
05f3fb3c JA |
6397 | while ((node = llist_del_all(&data->put_llist)) != NULL) { |
6398 | llist_for_each_entry_safe(pfile, tmp, node, llist) { | |
6399 | io_ring_file_put(data->ctx, pfile->file); | |
a5318d3c | 6400 | kfree(pfile); |
05f3fb3c | 6401 | } |
65e19f54 | 6402 | } |
2faf852d | 6403 | } |
65e19f54 | 6404 | |
2faf852d JA |
6405 | static void io_ring_file_ref_switch(struct work_struct *work) |
6406 | { | |
6407 | struct fixed_file_data *data; | |
65e19f54 | 6408 | |
2faf852d JA |
6409 | data = container_of(work, struct fixed_file_data, ref_work); |
6410 | io_ring_file_ref_flush(data); | |
05f3fb3c JA |
6411 | percpu_ref_switch_to_percpu(&data->refs); |
6412 | } | |
65e19f54 | 6413 | |
05f3fb3c JA |
6414 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
6415 | { | |
6416 | struct fixed_file_data *data; | |
6417 | ||
6418 | data = container_of(ref, struct fixed_file_data, refs); | |
6419 | ||
2faf852d JA |
6420 | /* |
6421 | * We can't safely switch from inside this context, punt to wq. If | |
6422 | * the table ref is going away, the table is being unregistered. | |
6423 | * Don't queue up the async work for that case, the caller will | |
6424 | * handle it. | |
6425 | */ | |
6426 | if (!percpu_ref_is_dying(&data->refs)) | |
6427 | queue_work(system_wq, &data->ref_work); | |
65e19f54 JA |
6428 | } |
6429 | ||
6b06314c JA |
6430 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
6431 | unsigned nr_args) | |
6432 | { | |
6433 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 6434 | unsigned nr_tables; |
05f3fb3c | 6435 | struct file *file; |
6b06314c JA |
6436 | int fd, ret = 0; |
6437 | unsigned i; | |
6438 | ||
05f3fb3c | 6439 | if (ctx->file_data) |
6b06314c JA |
6440 | return -EBUSY; |
6441 | if (!nr_args) | |
6442 | return -EINVAL; | |
6443 | if (nr_args > IORING_MAX_FIXED_FILES) | |
6444 | return -EMFILE; | |
6445 | ||
05f3fb3c JA |
6446 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
6447 | if (!ctx->file_data) | |
6448 | return -ENOMEM; | |
6449 | ctx->file_data->ctx = ctx; | |
6450 | init_completion(&ctx->file_data->done); | |
6451 | ||
65e19f54 | 6452 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
6453 | ctx->file_data->table = kcalloc(nr_tables, |
6454 | sizeof(struct fixed_file_table), | |
65e19f54 | 6455 | GFP_KERNEL); |
05f3fb3c JA |
6456 | if (!ctx->file_data->table) { |
6457 | kfree(ctx->file_data); | |
6458 | ctx->file_data = NULL; | |
6b06314c | 6459 | return -ENOMEM; |
05f3fb3c JA |
6460 | } |
6461 | ||
6462 | if (percpu_ref_init(&ctx->file_data->refs, io_file_data_ref_zero, | |
6463 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { | |
6464 | kfree(ctx->file_data->table); | |
6465 | kfree(ctx->file_data); | |
6466 | ctx->file_data = NULL; | |
6b06314c | 6467 | return -ENOMEM; |
05f3fb3c JA |
6468 | } |
6469 | ctx->file_data->put_llist.first = NULL; | |
6470 | INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch); | |
6b06314c | 6471 | |
65e19f54 | 6472 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
6473 | percpu_ref_exit(&ctx->file_data->refs); |
6474 | kfree(ctx->file_data->table); | |
6475 | kfree(ctx->file_data); | |
6476 | ctx->file_data = NULL; | |
65e19f54 JA |
6477 | return -ENOMEM; |
6478 | } | |
6479 | ||
08a45173 | 6480 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
6481 | struct fixed_file_table *table; |
6482 | unsigned index; | |
6483 | ||
6b06314c JA |
6484 | ret = -EFAULT; |
6485 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
6486 | break; | |
08a45173 JA |
6487 | /* allow sparse sets */ |
6488 | if (fd == -1) { | |
6489 | ret = 0; | |
6490 | continue; | |
6491 | } | |
6b06314c | 6492 | |
05f3fb3c | 6493 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 6494 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 6495 | file = fget(fd); |
6b06314c JA |
6496 | |
6497 | ret = -EBADF; | |
05f3fb3c | 6498 | if (!file) |
6b06314c | 6499 | break; |
05f3fb3c | 6500 | |
6b06314c JA |
6501 | /* |
6502 | * Don't allow io_uring instances to be registered. If UNIX | |
6503 | * isn't enabled, then this causes a reference cycle and this | |
6504 | * instance can never get freed. If UNIX is enabled we'll | |
6505 | * handle it just fine, but there's still no point in allowing | |
6506 | * a ring fd as it doesn't support regular read/write anyway. | |
6507 | */ | |
05f3fb3c JA |
6508 | if (file->f_op == &io_uring_fops) { |
6509 | fput(file); | |
6b06314c JA |
6510 | break; |
6511 | } | |
6b06314c | 6512 | ret = 0; |
05f3fb3c | 6513 | table->files[index] = file; |
6b06314c JA |
6514 | } |
6515 | ||
6516 | if (ret) { | |
65e19f54 | 6517 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
6518 | file = io_file_from_index(ctx, i); |
6519 | if (file) | |
6520 | fput(file); | |
6521 | } | |
6522 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 6523 | kfree(ctx->file_data->table[i].files); |
6b06314c | 6524 | |
05f3fb3c JA |
6525 | kfree(ctx->file_data->table); |
6526 | kfree(ctx->file_data); | |
6527 | ctx->file_data = NULL; | |
6b06314c JA |
6528 | ctx->nr_user_files = 0; |
6529 | return ret; | |
6530 | } | |
6531 | ||
6532 | ret = io_sqe_files_scm(ctx); | |
6533 | if (ret) | |
6534 | io_sqe_files_unregister(ctx); | |
6535 | ||
6536 | return ret; | |
6537 | } | |
6538 | ||
c3a31e60 JA |
6539 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
6540 | int index) | |
6541 | { | |
6542 | #if defined(CONFIG_UNIX) | |
6543 | struct sock *sock = ctx->ring_sock->sk; | |
6544 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
6545 | struct sk_buff *skb; | |
6546 | ||
6547 | /* | |
6548 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
6549 | * file set. If there's no room, fall back to allocating a new skb | |
6550 | * and filling it in. | |
6551 | */ | |
6552 | spin_lock_irq(&head->lock); | |
6553 | skb = skb_peek(head); | |
6554 | if (skb) { | |
6555 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
6556 | ||
6557 | if (fpl->count < SCM_MAX_FD) { | |
6558 | __skb_unlink(skb, head); | |
6559 | spin_unlock_irq(&head->lock); | |
6560 | fpl->fp[fpl->count] = get_file(file); | |
6561 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
6562 | fpl->count++; | |
6563 | spin_lock_irq(&head->lock); | |
6564 | __skb_queue_head(head, skb); | |
6565 | } else { | |
6566 | skb = NULL; | |
6567 | } | |
6568 | } | |
6569 | spin_unlock_irq(&head->lock); | |
6570 | ||
6571 | if (skb) { | |
6572 | fput(file); | |
6573 | return 0; | |
6574 | } | |
6575 | ||
6576 | return __io_sqe_files_scm(ctx, 1, index); | |
6577 | #else | |
6578 | return 0; | |
6579 | #endif | |
6580 | } | |
6581 | ||
05f3fb3c | 6582 | static void io_atomic_switch(struct percpu_ref *ref) |
c3a31e60 | 6583 | { |
05f3fb3c JA |
6584 | struct fixed_file_data *data; |
6585 | ||
dd3db2a3 JA |
6586 | /* |
6587 | * Juggle reference to ensure we hit zero, if needed, so we can | |
6588 | * switch back to percpu mode | |
6589 | */ | |
05f3fb3c | 6590 | data = container_of(ref, struct fixed_file_data, refs); |
dd3db2a3 JA |
6591 | percpu_ref_put(&data->refs); |
6592 | percpu_ref_get(&data->refs); | |
05f3fb3c JA |
6593 | } |
6594 | ||
a5318d3c | 6595 | static int io_queue_file_removal(struct fixed_file_data *data, |
05f3fb3c JA |
6596 | struct file *file) |
6597 | { | |
a5318d3c | 6598 | struct io_file_put *pfile; |
05f3fb3c | 6599 | |
05f3fb3c | 6600 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); |
a5318d3c HD |
6601 | if (!pfile) |
6602 | return -ENOMEM; | |
05f3fb3c JA |
6603 | |
6604 | pfile->file = file; | |
6605 | llist_add(&pfile->llist, &data->put_llist); | |
a5318d3c | 6606 | return 0; |
05f3fb3c JA |
6607 | } |
6608 | ||
6609 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6610 | struct io_uring_files_update *up, | |
6611 | unsigned nr_args) | |
6612 | { | |
6613 | struct fixed_file_data *data = ctx->file_data; | |
6614 | bool ref_switch = false; | |
6615 | struct file *file; | |
c3a31e60 JA |
6616 | __s32 __user *fds; |
6617 | int fd, i, err; | |
6618 | __u32 done; | |
6619 | ||
05f3fb3c | 6620 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6621 | return -EOVERFLOW; |
6622 | if (done > ctx->nr_user_files) | |
6623 | return -EINVAL; | |
6624 | ||
6625 | done = 0; | |
05f3fb3c | 6626 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6627 | while (nr_args) { |
65e19f54 JA |
6628 | struct fixed_file_table *table; |
6629 | unsigned index; | |
6630 | ||
c3a31e60 JA |
6631 | err = 0; |
6632 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6633 | err = -EFAULT; | |
6634 | break; | |
6635 | } | |
05f3fb3c JA |
6636 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6637 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6638 | index = i & IORING_FILE_TABLE_MASK; |
6639 | if (table->files[index]) { | |
05f3fb3c | 6640 | file = io_file_from_index(ctx, index); |
a5318d3c HD |
6641 | err = io_queue_file_removal(data, file); |
6642 | if (err) | |
6643 | break; | |
65e19f54 | 6644 | table->files[index] = NULL; |
a5318d3c | 6645 | ref_switch = true; |
c3a31e60 JA |
6646 | } |
6647 | if (fd != -1) { | |
c3a31e60 JA |
6648 | file = fget(fd); |
6649 | if (!file) { | |
6650 | err = -EBADF; | |
6651 | break; | |
6652 | } | |
6653 | /* | |
6654 | * Don't allow io_uring instances to be registered. If | |
6655 | * UNIX isn't enabled, then this causes a reference | |
6656 | * cycle and this instance can never get freed. If UNIX | |
6657 | * is enabled we'll handle it just fine, but there's | |
6658 | * still no point in allowing a ring fd as it doesn't | |
6659 | * support regular read/write anyway. | |
6660 | */ | |
6661 | if (file->f_op == &io_uring_fops) { | |
6662 | fput(file); | |
6663 | err = -EBADF; | |
6664 | break; | |
6665 | } | |
65e19f54 | 6666 | table->files[index] = file; |
c3a31e60 JA |
6667 | err = io_sqe_file_register(ctx, file, i); |
6668 | if (err) | |
6669 | break; | |
6670 | } | |
6671 | nr_args--; | |
6672 | done++; | |
05f3fb3c JA |
6673 | up->offset++; |
6674 | } | |
6675 | ||
dd3db2a3 | 6676 | if (ref_switch) |
05f3fb3c | 6677 | percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch); |
c3a31e60 JA |
6678 | |
6679 | return done ? done : err; | |
6680 | } | |
05f3fb3c JA |
6681 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6682 | unsigned nr_args) | |
6683 | { | |
6684 | struct io_uring_files_update up; | |
6685 | ||
6686 | if (!ctx->file_data) | |
6687 | return -ENXIO; | |
6688 | if (!nr_args) | |
6689 | return -EINVAL; | |
6690 | if (copy_from_user(&up, arg, sizeof(up))) | |
6691 | return -EFAULT; | |
6692 | if (up.resv) | |
6693 | return -EINVAL; | |
6694 | ||
6695 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6696 | } | |
c3a31e60 | 6697 | |
e9fd9396 | 6698 | static void io_free_work(struct io_wq_work *work) |
7d723065 JA |
6699 | { |
6700 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6701 | ||
e9fd9396 | 6702 | /* Consider that io_steal_work() relies on this ref */ |
7d723065 JA |
6703 | io_put_req(req); |
6704 | } | |
6705 | ||
24369c2e PB |
6706 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6707 | struct io_uring_params *p) | |
6708 | { | |
6709 | struct io_wq_data data; | |
6710 | struct fd f; | |
6711 | struct io_ring_ctx *ctx_attach; | |
6712 | unsigned int concurrency; | |
6713 | int ret = 0; | |
6714 | ||
6715 | data.user = ctx->user; | |
e9fd9396 | 6716 | data.free_work = io_free_work; |
24369c2e PB |
6717 | |
6718 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6719 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6720 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6721 | ||
6722 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6723 | if (IS_ERR(ctx->io_wq)) { | |
6724 | ret = PTR_ERR(ctx->io_wq); | |
6725 | ctx->io_wq = NULL; | |
6726 | } | |
6727 | return ret; | |
6728 | } | |
6729 | ||
6730 | f = fdget(p->wq_fd); | |
6731 | if (!f.file) | |
6732 | return -EBADF; | |
6733 | ||
6734 | if (f.file->f_op != &io_uring_fops) { | |
6735 | ret = -EINVAL; | |
6736 | goto out_fput; | |
6737 | } | |
6738 | ||
6739 | ctx_attach = f.file->private_data; | |
6740 | /* @io_wq is protected by holding the fd */ | |
6741 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6742 | ret = -EINVAL; | |
6743 | goto out_fput; | |
6744 | } | |
6745 | ||
6746 | ctx->io_wq = ctx_attach->io_wq; | |
6747 | out_fput: | |
6748 | fdput(f); | |
6749 | return ret; | |
6750 | } | |
6751 | ||
6c271ce2 JA |
6752 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6753 | struct io_uring_params *p) | |
2b188cc1 JA |
6754 | { |
6755 | int ret; | |
6756 | ||
6c271ce2 | 6757 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6758 | mmgrab(current->mm); |
6759 | ctx->sqo_mm = current->mm; | |
6760 | ||
6c271ce2 | 6761 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6762 | ret = -EPERM; |
6763 | if (!capable(CAP_SYS_ADMIN)) | |
6764 | goto err; | |
6765 | ||
917257da JA |
6766 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6767 | if (!ctx->sq_thread_idle) | |
6768 | ctx->sq_thread_idle = HZ; | |
6769 | ||
6c271ce2 | 6770 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6771 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6772 | |
917257da | 6773 | ret = -EINVAL; |
44a9bd18 JA |
6774 | if (cpu >= nr_cpu_ids) |
6775 | goto err; | |
7889f44d | 6776 | if (!cpu_online(cpu)) |
917257da JA |
6777 | goto err; |
6778 | ||
6c271ce2 JA |
6779 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6780 | ctx, cpu, | |
6781 | "io_uring-sq"); | |
6782 | } else { | |
6783 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6784 | "io_uring-sq"); | |
6785 | } | |
6786 | if (IS_ERR(ctx->sqo_thread)) { | |
6787 | ret = PTR_ERR(ctx->sqo_thread); | |
6788 | ctx->sqo_thread = NULL; | |
6789 | goto err; | |
6790 | } | |
6791 | wake_up_process(ctx->sqo_thread); | |
6792 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6793 | /* Can't have SQ_AFF without SQPOLL */ | |
6794 | ret = -EINVAL; | |
6795 | goto err; | |
6796 | } | |
6797 | ||
24369c2e PB |
6798 | ret = io_init_wq_offload(ctx, p); |
6799 | if (ret) | |
2b188cc1 | 6800 | goto err; |
2b188cc1 JA |
6801 | |
6802 | return 0; | |
6803 | err: | |
54a91f3b | 6804 | io_finish_async(ctx); |
2b188cc1 JA |
6805 | mmdrop(ctx->sqo_mm); |
6806 | ctx->sqo_mm = NULL; | |
6807 | return ret; | |
6808 | } | |
6809 | ||
6810 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6811 | { | |
6812 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6813 | } | |
6814 | ||
6815 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6816 | { | |
6817 | unsigned long page_limit, cur_pages, new_pages; | |
6818 | ||
6819 | /* Don't allow more pages than we can safely lock */ | |
6820 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6821 | ||
6822 | do { | |
6823 | cur_pages = atomic_long_read(&user->locked_vm); | |
6824 | new_pages = cur_pages + nr_pages; | |
6825 | if (new_pages > page_limit) | |
6826 | return -ENOMEM; | |
6827 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6828 | new_pages) != cur_pages); | |
6829 | ||
6830 | return 0; | |
6831 | } | |
6832 | ||
6833 | static void io_mem_free(void *ptr) | |
6834 | { | |
52e04ef4 MR |
6835 | struct page *page; |
6836 | ||
6837 | if (!ptr) | |
6838 | return; | |
2b188cc1 | 6839 | |
52e04ef4 | 6840 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6841 | if (put_page_testzero(page)) |
6842 | free_compound_page(page); | |
6843 | } | |
6844 | ||
6845 | static void *io_mem_alloc(size_t size) | |
6846 | { | |
6847 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6848 | __GFP_NORETRY; | |
6849 | ||
6850 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6851 | } | |
6852 | ||
75b28aff HV |
6853 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6854 | size_t *sq_offset) | |
6855 | { | |
6856 | struct io_rings *rings; | |
6857 | size_t off, sq_array_size; | |
6858 | ||
6859 | off = struct_size(rings, cqes, cq_entries); | |
6860 | if (off == SIZE_MAX) | |
6861 | return SIZE_MAX; | |
6862 | ||
6863 | #ifdef CONFIG_SMP | |
6864 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6865 | if (off == 0) | |
6866 | return SIZE_MAX; | |
6867 | #endif | |
6868 | ||
6869 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6870 | if (sq_array_size == SIZE_MAX) | |
6871 | return SIZE_MAX; | |
6872 | ||
6873 | if (check_add_overflow(off, sq_array_size, &off)) | |
6874 | return SIZE_MAX; | |
6875 | ||
6876 | if (sq_offset) | |
6877 | *sq_offset = off; | |
6878 | ||
6879 | return off; | |
6880 | } | |
6881 | ||
2b188cc1 JA |
6882 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6883 | { | |
75b28aff | 6884 | size_t pages; |
2b188cc1 | 6885 | |
75b28aff HV |
6886 | pages = (size_t)1 << get_order( |
6887 | rings_size(sq_entries, cq_entries, NULL)); | |
6888 | pages += (size_t)1 << get_order( | |
6889 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6890 | |
75b28aff | 6891 | return pages; |
2b188cc1 JA |
6892 | } |
6893 | ||
edafccee JA |
6894 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6895 | { | |
6896 | int i, j; | |
6897 | ||
6898 | if (!ctx->user_bufs) | |
6899 | return -ENXIO; | |
6900 | ||
6901 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6902 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6903 | ||
6904 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6905 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6906 | |
6907 | if (ctx->account_mem) | |
6908 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6909 | kvfree(imu->bvec); |
edafccee JA |
6910 | imu->nr_bvecs = 0; |
6911 | } | |
6912 | ||
6913 | kfree(ctx->user_bufs); | |
6914 | ctx->user_bufs = NULL; | |
6915 | ctx->nr_user_bufs = 0; | |
6916 | return 0; | |
6917 | } | |
6918 | ||
6919 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6920 | void __user *arg, unsigned index) | |
6921 | { | |
6922 | struct iovec __user *src; | |
6923 | ||
6924 | #ifdef CONFIG_COMPAT | |
6925 | if (ctx->compat) { | |
6926 | struct compat_iovec __user *ciovs; | |
6927 | struct compat_iovec ciov; | |
6928 | ||
6929 | ciovs = (struct compat_iovec __user *) arg; | |
6930 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
6931 | return -EFAULT; | |
6932 | ||
d55e5f5b | 6933 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
6934 | dst->iov_len = ciov.iov_len; |
6935 | return 0; | |
6936 | } | |
6937 | #endif | |
6938 | src = (struct iovec __user *) arg; | |
6939 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
6940 | return -EFAULT; | |
6941 | return 0; | |
6942 | } | |
6943 | ||
6944 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
6945 | unsigned nr_args) | |
6946 | { | |
6947 | struct vm_area_struct **vmas = NULL; | |
6948 | struct page **pages = NULL; | |
6949 | int i, j, got_pages = 0; | |
6950 | int ret = -EINVAL; | |
6951 | ||
6952 | if (ctx->user_bufs) | |
6953 | return -EBUSY; | |
6954 | if (!nr_args || nr_args > UIO_MAXIOV) | |
6955 | return -EINVAL; | |
6956 | ||
6957 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
6958 | GFP_KERNEL); | |
6959 | if (!ctx->user_bufs) | |
6960 | return -ENOMEM; | |
6961 | ||
6962 | for (i = 0; i < nr_args; i++) { | |
6963 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6964 | unsigned long off, start, end, ubuf; | |
6965 | int pret, nr_pages; | |
6966 | struct iovec iov; | |
6967 | size_t size; | |
6968 | ||
6969 | ret = io_copy_iov(ctx, &iov, arg, i); | |
6970 | if (ret) | |
a278682d | 6971 | goto err; |
edafccee JA |
6972 | |
6973 | /* | |
6974 | * Don't impose further limits on the size and buffer | |
6975 | * constraints here, we'll -EINVAL later when IO is | |
6976 | * submitted if they are wrong. | |
6977 | */ | |
6978 | ret = -EFAULT; | |
6979 | if (!iov.iov_base || !iov.iov_len) | |
6980 | goto err; | |
6981 | ||
6982 | /* arbitrary limit, but we need something */ | |
6983 | if (iov.iov_len > SZ_1G) | |
6984 | goto err; | |
6985 | ||
6986 | ubuf = (unsigned long) iov.iov_base; | |
6987 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
6988 | start = ubuf >> PAGE_SHIFT; | |
6989 | nr_pages = end - start; | |
6990 | ||
6991 | if (ctx->account_mem) { | |
6992 | ret = io_account_mem(ctx->user, nr_pages); | |
6993 | if (ret) | |
6994 | goto err; | |
6995 | } | |
6996 | ||
6997 | ret = 0; | |
6998 | if (!pages || nr_pages > got_pages) { | |
6999 | kfree(vmas); | |
7000 | kfree(pages); | |
d4ef6475 | 7001 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 7002 | GFP_KERNEL); |
d4ef6475 | 7003 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
7004 | sizeof(struct vm_area_struct *), |
7005 | GFP_KERNEL); | |
7006 | if (!pages || !vmas) { | |
7007 | ret = -ENOMEM; | |
7008 | if (ctx->account_mem) | |
7009 | io_unaccount_mem(ctx->user, nr_pages); | |
7010 | goto err; | |
7011 | } | |
7012 | got_pages = nr_pages; | |
7013 | } | |
7014 | ||
d4ef6475 | 7015 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
7016 | GFP_KERNEL); |
7017 | ret = -ENOMEM; | |
7018 | if (!imu->bvec) { | |
7019 | if (ctx->account_mem) | |
7020 | io_unaccount_mem(ctx->user, nr_pages); | |
7021 | goto err; | |
7022 | } | |
7023 | ||
7024 | ret = 0; | |
7025 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 7026 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
7027 | FOLL_WRITE | FOLL_LONGTERM, |
7028 | pages, vmas); | |
edafccee JA |
7029 | if (pret == nr_pages) { |
7030 | /* don't support file backed memory */ | |
7031 | for (j = 0; j < nr_pages; j++) { | |
7032 | struct vm_area_struct *vma = vmas[j]; | |
7033 | ||
7034 | if (vma->vm_file && | |
7035 | !is_file_hugepages(vma->vm_file)) { | |
7036 | ret = -EOPNOTSUPP; | |
7037 | break; | |
7038 | } | |
7039 | } | |
7040 | } else { | |
7041 | ret = pret < 0 ? pret : -EFAULT; | |
7042 | } | |
7043 | up_read(¤t->mm->mmap_sem); | |
7044 | if (ret) { | |
7045 | /* | |
7046 | * if we did partial map, or found file backed vmas, | |
7047 | * release any pages we did get | |
7048 | */ | |
27c4d3a3 | 7049 | if (pret > 0) |
f1f6a7dd | 7050 | unpin_user_pages(pages, pret); |
edafccee JA |
7051 | if (ctx->account_mem) |
7052 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 7053 | kvfree(imu->bvec); |
edafccee JA |
7054 | goto err; |
7055 | } | |
7056 | ||
7057 | off = ubuf & ~PAGE_MASK; | |
7058 | size = iov.iov_len; | |
7059 | for (j = 0; j < nr_pages; j++) { | |
7060 | size_t vec_len; | |
7061 | ||
7062 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
7063 | imu->bvec[j].bv_page = pages[j]; | |
7064 | imu->bvec[j].bv_len = vec_len; | |
7065 | imu->bvec[j].bv_offset = off; | |
7066 | off = 0; | |
7067 | size -= vec_len; | |
7068 | } | |
7069 | /* store original address for later verification */ | |
7070 | imu->ubuf = ubuf; | |
7071 | imu->len = iov.iov_len; | |
7072 | imu->nr_bvecs = nr_pages; | |
7073 | ||
7074 | ctx->nr_user_bufs++; | |
7075 | } | |
d4ef6475 MR |
7076 | kvfree(pages); |
7077 | kvfree(vmas); | |
edafccee JA |
7078 | return 0; |
7079 | err: | |
d4ef6475 MR |
7080 | kvfree(pages); |
7081 | kvfree(vmas); | |
edafccee JA |
7082 | io_sqe_buffer_unregister(ctx); |
7083 | return ret; | |
7084 | } | |
7085 | ||
9b402849 JA |
7086 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
7087 | { | |
7088 | __s32 __user *fds = arg; | |
7089 | int fd; | |
7090 | ||
7091 | if (ctx->cq_ev_fd) | |
7092 | return -EBUSY; | |
7093 | ||
7094 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
7095 | return -EFAULT; | |
7096 | ||
7097 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
7098 | if (IS_ERR(ctx->cq_ev_fd)) { | |
7099 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
7100 | ctx->cq_ev_fd = NULL; | |
7101 | return ret; | |
7102 | } | |
7103 | ||
7104 | return 0; | |
7105 | } | |
7106 | ||
7107 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
7108 | { | |
7109 | if (ctx->cq_ev_fd) { | |
7110 | eventfd_ctx_put(ctx->cq_ev_fd); | |
7111 | ctx->cq_ev_fd = NULL; | |
7112 | return 0; | |
7113 | } | |
7114 | ||
7115 | return -ENXIO; | |
7116 | } | |
7117 | ||
5a2e745d JA |
7118 | static int __io_destroy_buffers(int id, void *p, void *data) |
7119 | { | |
7120 | struct io_ring_ctx *ctx = data; | |
7121 | struct io_buffer *buf = p; | |
7122 | ||
067524e9 | 7123 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
7124 | return 0; |
7125 | } | |
7126 | ||
7127 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
7128 | { | |
7129 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
7130 | idr_destroy(&ctx->io_buffer_idr); | |
7131 | } | |
7132 | ||
2b188cc1 JA |
7133 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
7134 | { | |
6b06314c | 7135 | io_finish_async(ctx); |
2b188cc1 JA |
7136 | if (ctx->sqo_mm) |
7137 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
7138 | |
7139 | io_iopoll_reap_events(ctx); | |
edafccee | 7140 | io_sqe_buffer_unregister(ctx); |
6b06314c | 7141 | io_sqe_files_unregister(ctx); |
9b402849 | 7142 | io_eventfd_unregister(ctx); |
5a2e745d | 7143 | io_destroy_buffers(ctx); |
41726c9a | 7144 | idr_destroy(&ctx->personality_idr); |
def596e9 | 7145 | |
2b188cc1 | 7146 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
7147 | if (ctx->ring_sock) { |
7148 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 7149 | sock_release(ctx->ring_sock); |
355e8d26 | 7150 | } |
2b188cc1 JA |
7151 | #endif |
7152 | ||
75b28aff | 7153 | io_mem_free(ctx->rings); |
2b188cc1 | 7154 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
7155 | |
7156 | percpu_ref_exit(&ctx->refs); | |
7157 | if (ctx->account_mem) | |
7158 | io_unaccount_mem(ctx->user, | |
7159 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
7160 | free_uid(ctx->user); | |
181e448d | 7161 | put_cred(ctx->creds); |
206aefde | 7162 | kfree(ctx->completions); |
78076bb6 | 7163 | kfree(ctx->cancel_hash); |
0ddf92e8 | 7164 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
7165 | kfree(ctx); |
7166 | } | |
7167 | ||
7168 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
7169 | { | |
7170 | struct io_ring_ctx *ctx = file->private_data; | |
7171 | __poll_t mask = 0; | |
7172 | ||
7173 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
7174 | /* |
7175 | * synchronizes with barrier from wq_has_sleeper call in | |
7176 | * io_commit_cqring | |
7177 | */ | |
2b188cc1 | 7178 | smp_rmb(); |
75b28aff HV |
7179 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
7180 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 7181 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 7182 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
7183 | mask |= EPOLLIN | EPOLLRDNORM; |
7184 | ||
7185 | return mask; | |
7186 | } | |
7187 | ||
7188 | static int io_uring_fasync(int fd, struct file *file, int on) | |
7189 | { | |
7190 | struct io_ring_ctx *ctx = file->private_data; | |
7191 | ||
7192 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
7193 | } | |
7194 | ||
071698e1 JA |
7195 | static int io_remove_personalities(int id, void *p, void *data) |
7196 | { | |
7197 | struct io_ring_ctx *ctx = data; | |
7198 | const struct cred *cred; | |
7199 | ||
7200 | cred = idr_remove(&ctx->personality_idr, id); | |
7201 | if (cred) | |
7202 | put_cred(cred); | |
7203 | return 0; | |
7204 | } | |
7205 | ||
2b188cc1 JA |
7206 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
7207 | { | |
7208 | mutex_lock(&ctx->uring_lock); | |
7209 | percpu_ref_kill(&ctx->refs); | |
7210 | mutex_unlock(&ctx->uring_lock); | |
7211 | ||
df069d80 JA |
7212 | /* |
7213 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
7214 | * work after we've killed the ctx ref above. This is important to do | |
7215 | * before we cancel existing commands, as the thread could otherwise | |
7216 | * be queueing new work post that. If that's work we need to cancel, | |
7217 | * it could cause shutdown to hang. | |
7218 | */ | |
7219 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
7220 | cpu_relax(); | |
7221 | ||
5262f567 | 7222 | io_kill_timeouts(ctx); |
221c5eb2 | 7223 | io_poll_remove_all(ctx); |
561fb04a JA |
7224 | |
7225 | if (ctx->io_wq) | |
7226 | io_wq_cancel_all(ctx->io_wq); | |
7227 | ||
def596e9 | 7228 | io_iopoll_reap_events(ctx); |
15dff286 JA |
7229 | /* if we failed setting up the ctx, we might not have any rings */ |
7230 | if (ctx->rings) | |
7231 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 7232 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
206aefde | 7233 | wait_for_completion(&ctx->completions[0]); |
2b188cc1 JA |
7234 | io_ring_ctx_free(ctx); |
7235 | } | |
7236 | ||
7237 | static int io_uring_release(struct inode *inode, struct file *file) | |
7238 | { | |
7239 | struct io_ring_ctx *ctx = file->private_data; | |
7240 | ||
7241 | file->private_data = NULL; | |
7242 | io_ring_ctx_wait_and_kill(ctx); | |
7243 | return 0; | |
7244 | } | |
7245 | ||
fcb323cc JA |
7246 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
7247 | struct files_struct *files) | |
7248 | { | |
7249 | struct io_kiocb *req; | |
7250 | DEFINE_WAIT(wait); | |
7251 | ||
7252 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 7253 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
7254 | |
7255 | spin_lock_irq(&ctx->inflight_lock); | |
7256 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
7257 | if (req->work.files != files) |
7258 | continue; | |
7259 | /* req is being completed, ignore */ | |
7260 | if (!refcount_inc_not_zero(&req->refs)) | |
7261 | continue; | |
7262 | cancel_req = req; | |
7263 | break; | |
fcb323cc | 7264 | } |
768134d4 | 7265 | if (cancel_req) |
fcb323cc | 7266 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 7267 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
7268 | spin_unlock_irq(&ctx->inflight_lock); |
7269 | ||
768134d4 JA |
7270 | /* We need to keep going until we don't find a matching req */ |
7271 | if (!cancel_req) | |
fcb323cc | 7272 | break; |
2f6d9b9d | 7273 | |
2ca10259 JA |
7274 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
7275 | spin_lock_irq(&ctx->completion_lock); | |
7276 | list_del(&cancel_req->list); | |
7277 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
7278 | if (list_empty(&ctx->cq_overflow_list)) { | |
7279 | clear_bit(0, &ctx->sq_check_overflow); | |
7280 | clear_bit(0, &ctx->cq_check_overflow); | |
7281 | } | |
7282 | spin_unlock_irq(&ctx->completion_lock); | |
7283 | ||
7284 | WRITE_ONCE(ctx->rings->cq_overflow, | |
7285 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
7286 | ||
7287 | /* | |
7288 | * Put inflight ref and overflow ref. If that's | |
7289 | * all we had, then we're done with this request. | |
7290 | */ | |
7291 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
7292 | io_put_req(cancel_req); | |
7293 | continue; | |
7294 | } | |
7295 | } | |
7296 | ||
2f6d9b9d BL |
7297 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
7298 | io_put_req(cancel_req); | |
fcb323cc JA |
7299 | schedule(); |
7300 | } | |
768134d4 | 7301 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
7302 | } |
7303 | ||
7304 | static int io_uring_flush(struct file *file, void *data) | |
7305 | { | |
7306 | struct io_ring_ctx *ctx = file->private_data; | |
7307 | ||
7308 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
7309 | |
7310 | /* | |
7311 | * If the task is going away, cancel work it may have pending | |
7312 | */ | |
7313 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
7314 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
7315 | ||
fcb323cc JA |
7316 | return 0; |
7317 | } | |
7318 | ||
6c5c240e RP |
7319 | static void *io_uring_validate_mmap_request(struct file *file, |
7320 | loff_t pgoff, size_t sz) | |
2b188cc1 | 7321 | { |
2b188cc1 | 7322 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 7323 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
7324 | struct page *page; |
7325 | void *ptr; | |
7326 | ||
7327 | switch (offset) { | |
7328 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
7329 | case IORING_OFF_CQ_RING: |
7330 | ptr = ctx->rings; | |
2b188cc1 JA |
7331 | break; |
7332 | case IORING_OFF_SQES: | |
7333 | ptr = ctx->sq_sqes; | |
7334 | break; | |
2b188cc1 | 7335 | default: |
6c5c240e | 7336 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
7337 | } |
7338 | ||
7339 | page = virt_to_head_page(ptr); | |
a50b854e | 7340 | if (sz > page_size(page)) |
6c5c240e RP |
7341 | return ERR_PTR(-EINVAL); |
7342 | ||
7343 | return ptr; | |
7344 | } | |
7345 | ||
7346 | #ifdef CONFIG_MMU | |
7347 | ||
7348 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7349 | { | |
7350 | size_t sz = vma->vm_end - vma->vm_start; | |
7351 | unsigned long pfn; | |
7352 | void *ptr; | |
7353 | ||
7354 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
7355 | if (IS_ERR(ptr)) | |
7356 | return PTR_ERR(ptr); | |
2b188cc1 JA |
7357 | |
7358 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
7359 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
7360 | } | |
7361 | ||
6c5c240e RP |
7362 | #else /* !CONFIG_MMU */ |
7363 | ||
7364 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
7365 | { | |
7366 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
7367 | } | |
7368 | ||
7369 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
7370 | { | |
7371 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
7372 | } | |
7373 | ||
7374 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
7375 | unsigned long addr, unsigned long len, | |
7376 | unsigned long pgoff, unsigned long flags) | |
7377 | { | |
7378 | void *ptr; | |
7379 | ||
7380 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
7381 | if (IS_ERR(ptr)) | |
7382 | return PTR_ERR(ptr); | |
7383 | ||
7384 | return (unsigned long) ptr; | |
7385 | } | |
7386 | ||
7387 | #endif /* !CONFIG_MMU */ | |
7388 | ||
2b188cc1 JA |
7389 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
7390 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
7391 | size_t, sigsz) | |
7392 | { | |
7393 | struct io_ring_ctx *ctx; | |
7394 | long ret = -EBADF; | |
7395 | int submitted = 0; | |
7396 | struct fd f; | |
7397 | ||
b41e9852 JA |
7398 | if (current->task_works) |
7399 | task_work_run(); | |
7400 | ||
6c271ce2 | 7401 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
7402 | return -EINVAL; |
7403 | ||
7404 | f = fdget(fd); | |
7405 | if (!f.file) | |
7406 | return -EBADF; | |
7407 | ||
7408 | ret = -EOPNOTSUPP; | |
7409 | if (f.file->f_op != &io_uring_fops) | |
7410 | goto out_fput; | |
7411 | ||
7412 | ret = -ENXIO; | |
7413 | ctx = f.file->private_data; | |
7414 | if (!percpu_ref_tryget(&ctx->refs)) | |
7415 | goto out_fput; | |
7416 | ||
6c271ce2 JA |
7417 | /* |
7418 | * For SQ polling, the thread will do all submissions and completions. | |
7419 | * Just return the requested submit count, and wake the thread if | |
7420 | * we were asked to. | |
7421 | */ | |
b2a9eada | 7422 | ret = 0; |
6c271ce2 | 7423 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
7424 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
7425 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
7426 | if (flags & IORING_ENTER_SQ_WAKEUP) |
7427 | wake_up(&ctx->sqo_wait); | |
7428 | submitted = to_submit; | |
b2a9eada | 7429 | } else if (to_submit) { |
ae9428ca | 7430 | struct mm_struct *cur_mm; |
2b188cc1 JA |
7431 | |
7432 | mutex_lock(&ctx->uring_lock); | |
ae9428ca PB |
7433 | /* already have mm, so io_submit_sqes() won't try to grab it */ |
7434 | cur_mm = ctx->sqo_mm; | |
7435 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, | |
7436 | &cur_mm, false); | |
2b188cc1 | 7437 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
7438 | |
7439 | if (submitted != to_submit) | |
7440 | goto out; | |
2b188cc1 JA |
7441 | } |
7442 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
7443 | unsigned nr_events = 0; |
7444 | ||
2b188cc1 JA |
7445 | min_complete = min(min_complete, ctx->cq_entries); |
7446 | ||
32b2244a XW |
7447 | /* |
7448 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
7449 | * space applications don't need to do io completion events | |
7450 | * polling again, they can rely on io_sq_thread to do polling | |
7451 | * work, which can reduce cpu usage and uring_lock contention. | |
7452 | */ | |
7453 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
7454 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
def596e9 | 7455 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
7456 | } else { |
7457 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
7458 | } | |
2b188cc1 JA |
7459 | } |
7460 | ||
7c504e65 | 7461 | out: |
6805b32e | 7462 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
7463 | out_fput: |
7464 | fdput(f); | |
7465 | return submitted ? submitted : ret; | |
7466 | } | |
7467 | ||
bebdb65e | 7468 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
7469 | static int io_uring_show_cred(int id, void *p, void *data) |
7470 | { | |
7471 | const struct cred *cred = p; | |
7472 | struct seq_file *m = data; | |
7473 | struct user_namespace *uns = seq_user_ns(m); | |
7474 | struct group_info *gi; | |
7475 | kernel_cap_t cap; | |
7476 | unsigned __capi; | |
7477 | int g; | |
7478 | ||
7479 | seq_printf(m, "%5d\n", id); | |
7480 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
7481 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
7482 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
7483 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
7484 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
7485 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
7486 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
7487 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
7488 | seq_puts(m, "\n\tGroups:\t"); | |
7489 | gi = cred->group_info; | |
7490 | for (g = 0; g < gi->ngroups; g++) { | |
7491 | seq_put_decimal_ull(m, g ? " " : "", | |
7492 | from_kgid_munged(uns, gi->gid[g])); | |
7493 | } | |
7494 | seq_puts(m, "\n\tCapEff:\t"); | |
7495 | cap = cred->cap_effective; | |
7496 | CAP_FOR_EACH_U32(__capi) | |
7497 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
7498 | seq_putc(m, '\n'); | |
7499 | return 0; | |
7500 | } | |
7501 | ||
7502 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
7503 | { | |
7504 | int i; | |
7505 | ||
7506 | mutex_lock(&ctx->uring_lock); | |
7507 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
7508 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7509 | struct fixed_file_table *table; | |
7510 | struct file *f; | |
7511 | ||
7512 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7513 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
7514 | if (f) | |
7515 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
7516 | else | |
7517 | seq_printf(m, "%5u: <none>\n", i); | |
7518 | } | |
7519 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
7520 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
7521 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
7522 | ||
7523 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
7524 | (unsigned int) buf->len); | |
7525 | } | |
7526 | if (!idr_is_empty(&ctx->personality_idr)) { | |
7527 | seq_printf(m, "Personalities:\n"); | |
7528 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
7529 | } | |
d7718a9d JA |
7530 | seq_printf(m, "PollList:\n"); |
7531 | spin_lock_irq(&ctx->completion_lock); | |
7532 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
7533 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
7534 | struct io_kiocb *req; | |
7535 | ||
7536 | hlist_for_each_entry(req, list, hash_node) | |
7537 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
7538 | req->task->task_works != NULL); | |
7539 | } | |
7540 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
7541 | mutex_unlock(&ctx->uring_lock); |
7542 | } | |
7543 | ||
7544 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
7545 | { | |
7546 | struct io_ring_ctx *ctx = f->private_data; | |
7547 | ||
7548 | if (percpu_ref_tryget(&ctx->refs)) { | |
7549 | __io_uring_show_fdinfo(ctx, m); | |
7550 | percpu_ref_put(&ctx->refs); | |
7551 | } | |
7552 | } | |
bebdb65e | 7553 | #endif |
87ce955b | 7554 | |
2b188cc1 JA |
7555 | static const struct file_operations io_uring_fops = { |
7556 | .release = io_uring_release, | |
fcb323cc | 7557 | .flush = io_uring_flush, |
2b188cc1 | 7558 | .mmap = io_uring_mmap, |
6c5c240e RP |
7559 | #ifndef CONFIG_MMU |
7560 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
7561 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
7562 | #endif | |
2b188cc1 JA |
7563 | .poll = io_uring_poll, |
7564 | .fasync = io_uring_fasync, | |
bebdb65e | 7565 | #ifdef CONFIG_PROC_FS |
87ce955b | 7566 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 7567 | #endif |
2b188cc1 JA |
7568 | }; |
7569 | ||
7570 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
7571 | struct io_uring_params *p) | |
7572 | { | |
75b28aff HV |
7573 | struct io_rings *rings; |
7574 | size_t size, sq_array_offset; | |
2b188cc1 | 7575 | |
75b28aff HV |
7576 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
7577 | if (size == SIZE_MAX) | |
7578 | return -EOVERFLOW; | |
7579 | ||
7580 | rings = io_mem_alloc(size); | |
7581 | if (!rings) | |
2b188cc1 JA |
7582 | return -ENOMEM; |
7583 | ||
75b28aff HV |
7584 | ctx->rings = rings; |
7585 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7586 | rings->sq_ring_mask = p->sq_entries - 1; | |
7587 | rings->cq_ring_mask = p->cq_entries - 1; | |
7588 | rings->sq_ring_entries = p->sq_entries; | |
7589 | rings->cq_ring_entries = p->cq_entries; | |
7590 | ctx->sq_mask = rings->sq_ring_mask; | |
7591 | ctx->cq_mask = rings->cq_ring_mask; | |
7592 | ctx->sq_entries = rings->sq_ring_entries; | |
7593 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7594 | |
7595 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7596 | if (size == SIZE_MAX) { |
7597 | io_mem_free(ctx->rings); | |
7598 | ctx->rings = NULL; | |
2b188cc1 | 7599 | return -EOVERFLOW; |
eb065d30 | 7600 | } |
2b188cc1 JA |
7601 | |
7602 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7603 | if (!ctx->sq_sqes) { |
7604 | io_mem_free(ctx->rings); | |
7605 | ctx->rings = NULL; | |
2b188cc1 | 7606 | return -ENOMEM; |
eb065d30 | 7607 | } |
2b188cc1 | 7608 | |
2b188cc1 JA |
7609 | return 0; |
7610 | } | |
7611 | ||
7612 | /* | |
7613 | * Allocate an anonymous fd, this is what constitutes the application | |
7614 | * visible backing of an io_uring instance. The application mmaps this | |
7615 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7616 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7617 | */ | |
7618 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7619 | { | |
7620 | struct file *file; | |
7621 | int ret; | |
7622 | ||
7623 | #if defined(CONFIG_UNIX) | |
7624 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7625 | &ctx->ring_sock); | |
7626 | if (ret) | |
7627 | return ret; | |
7628 | #endif | |
7629 | ||
7630 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7631 | if (ret < 0) | |
7632 | goto err; | |
7633 | ||
7634 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7635 | O_RDWR | O_CLOEXEC); | |
7636 | if (IS_ERR(file)) { | |
7637 | put_unused_fd(ret); | |
7638 | ret = PTR_ERR(file); | |
7639 | goto err; | |
7640 | } | |
7641 | ||
7642 | #if defined(CONFIG_UNIX) | |
7643 | ctx->ring_sock->file = file; | |
7644 | #endif | |
7645 | fd_install(ret, file); | |
7646 | return ret; | |
7647 | err: | |
7648 | #if defined(CONFIG_UNIX) | |
7649 | sock_release(ctx->ring_sock); | |
7650 | ctx->ring_sock = NULL; | |
7651 | #endif | |
7652 | return ret; | |
7653 | } | |
7654 | ||
7655 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7656 | { | |
7657 | struct user_struct *user = NULL; | |
7658 | struct io_ring_ctx *ctx; | |
7659 | bool account_mem; | |
7660 | int ret; | |
7661 | ||
8110c1a6 | 7662 | if (!entries) |
2b188cc1 | 7663 | return -EINVAL; |
8110c1a6 JA |
7664 | if (entries > IORING_MAX_ENTRIES) { |
7665 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7666 | return -EINVAL; | |
7667 | entries = IORING_MAX_ENTRIES; | |
7668 | } | |
2b188cc1 JA |
7669 | |
7670 | /* | |
7671 | * Use twice as many entries for the CQ ring. It's possible for the | |
7672 | * application to drive a higher depth than the size of the SQ ring, | |
7673 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7674 | * some flexibility in overcommitting a bit. If the application has |
7675 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7676 | * of CQ ring entries manually. | |
2b188cc1 JA |
7677 | */ |
7678 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7679 | if (p->flags & IORING_SETUP_CQSIZE) { |
7680 | /* | |
7681 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7682 | * to a power-of-two, if it isn't already. We do NOT impose | |
7683 | * any cq vs sq ring sizing. | |
7684 | */ | |
8110c1a6 | 7685 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7686 | return -EINVAL; |
8110c1a6 JA |
7687 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7688 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7689 | return -EINVAL; | |
7690 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7691 | } | |
33a107f0 JA |
7692 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7693 | } else { | |
7694 | p->cq_entries = 2 * p->sq_entries; | |
7695 | } | |
2b188cc1 JA |
7696 | |
7697 | user = get_uid(current_user()); | |
7698 | account_mem = !capable(CAP_IPC_LOCK); | |
7699 | ||
7700 | if (account_mem) { | |
7701 | ret = io_account_mem(user, | |
7702 | ring_pages(p->sq_entries, p->cq_entries)); | |
7703 | if (ret) { | |
7704 | free_uid(user); | |
7705 | return ret; | |
7706 | } | |
7707 | } | |
7708 | ||
7709 | ctx = io_ring_ctx_alloc(p); | |
7710 | if (!ctx) { | |
7711 | if (account_mem) | |
7712 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7713 | p->cq_entries)); | |
7714 | free_uid(user); | |
7715 | return -ENOMEM; | |
7716 | } | |
7717 | ctx->compat = in_compat_syscall(); | |
7718 | ctx->account_mem = account_mem; | |
7719 | ctx->user = user; | |
0b8c0ec7 | 7720 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7721 | |
7722 | ret = io_allocate_scq_urings(ctx, p); | |
7723 | if (ret) | |
7724 | goto err; | |
7725 | ||
6c271ce2 | 7726 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7727 | if (ret) |
7728 | goto err; | |
7729 | ||
2b188cc1 | 7730 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7731 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7732 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7733 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7734 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7735 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7736 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7737 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7738 | |
7739 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7740 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7741 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7742 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7743 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7744 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7745 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7746 | |
044c1ab3 JA |
7747 | /* |
7748 | * Install ring fd as the very last thing, so we don't risk someone | |
7749 | * having closed it before we finish setup | |
7750 | */ | |
7751 | ret = io_uring_get_fd(ctx); | |
7752 | if (ret < 0) | |
7753 | goto err; | |
7754 | ||
da8c9690 | 7755 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7756 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7757 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7758 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7759 | return ret; |
7760 | err: | |
7761 | io_ring_ctx_wait_and_kill(ctx); | |
7762 | return ret; | |
7763 | } | |
7764 | ||
7765 | /* | |
7766 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7767 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7768 | * params structure passed in. | |
7769 | */ | |
7770 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7771 | { | |
7772 | struct io_uring_params p; | |
7773 | long ret; | |
7774 | int i; | |
7775 | ||
7776 | if (copy_from_user(&p, params, sizeof(p))) | |
7777 | return -EFAULT; | |
7778 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7779 | if (p.resv[i]) | |
7780 | return -EINVAL; | |
7781 | } | |
7782 | ||
6c271ce2 | 7783 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7784 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7785 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7786 | return -EINVAL; |
7787 | ||
7788 | ret = io_uring_create(entries, &p); | |
7789 | if (ret < 0) | |
7790 | return ret; | |
7791 | ||
7792 | if (copy_to_user(params, &p, sizeof(p))) | |
7793 | return -EFAULT; | |
7794 | ||
7795 | return ret; | |
7796 | } | |
7797 | ||
7798 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7799 | struct io_uring_params __user *, params) | |
7800 | { | |
7801 | return io_uring_setup(entries, params); | |
7802 | } | |
7803 | ||
66f4af93 JA |
7804 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7805 | { | |
7806 | struct io_uring_probe *p; | |
7807 | size_t size; | |
7808 | int i, ret; | |
7809 | ||
7810 | size = struct_size(p, ops, nr_args); | |
7811 | if (size == SIZE_MAX) | |
7812 | return -EOVERFLOW; | |
7813 | p = kzalloc(size, GFP_KERNEL); | |
7814 | if (!p) | |
7815 | return -ENOMEM; | |
7816 | ||
7817 | ret = -EFAULT; | |
7818 | if (copy_from_user(p, arg, size)) | |
7819 | goto out; | |
7820 | ret = -EINVAL; | |
7821 | if (memchr_inv(p, 0, size)) | |
7822 | goto out; | |
7823 | ||
7824 | p->last_op = IORING_OP_LAST - 1; | |
7825 | if (nr_args > IORING_OP_LAST) | |
7826 | nr_args = IORING_OP_LAST; | |
7827 | ||
7828 | for (i = 0; i < nr_args; i++) { | |
7829 | p->ops[i].op = i; | |
7830 | if (!io_op_defs[i].not_supported) | |
7831 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7832 | } | |
7833 | p->ops_len = i; | |
7834 | ||
7835 | ret = 0; | |
7836 | if (copy_to_user(arg, p, size)) | |
7837 | ret = -EFAULT; | |
7838 | out: | |
7839 | kfree(p); | |
7840 | return ret; | |
7841 | } | |
7842 | ||
071698e1 JA |
7843 | static int io_register_personality(struct io_ring_ctx *ctx) |
7844 | { | |
7845 | const struct cred *creds = get_current_cred(); | |
7846 | int id; | |
7847 | ||
7848 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7849 | USHRT_MAX, GFP_KERNEL); | |
7850 | if (id < 0) | |
7851 | put_cred(creds); | |
7852 | return id; | |
7853 | } | |
7854 | ||
7855 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7856 | { | |
7857 | const struct cred *old_creds; | |
7858 | ||
7859 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7860 | if (old_creds) { | |
7861 | put_cred(old_creds); | |
7862 | return 0; | |
7863 | } | |
7864 | ||
7865 | return -EINVAL; | |
7866 | } | |
7867 | ||
7868 | static bool io_register_op_must_quiesce(int op) | |
7869 | { | |
7870 | switch (op) { | |
7871 | case IORING_UNREGISTER_FILES: | |
7872 | case IORING_REGISTER_FILES_UPDATE: | |
7873 | case IORING_REGISTER_PROBE: | |
7874 | case IORING_REGISTER_PERSONALITY: | |
7875 | case IORING_UNREGISTER_PERSONALITY: | |
7876 | return false; | |
7877 | default: | |
7878 | return true; | |
7879 | } | |
7880 | } | |
7881 | ||
edafccee JA |
7882 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7883 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7884 | __releases(ctx->uring_lock) |
7885 | __acquires(ctx->uring_lock) | |
edafccee JA |
7886 | { |
7887 | int ret; | |
7888 | ||
35fa71a0 JA |
7889 | /* |
7890 | * We're inside the ring mutex, if the ref is already dying, then | |
7891 | * someone else killed the ctx or is already going through | |
7892 | * io_uring_register(). | |
7893 | */ | |
7894 | if (percpu_ref_is_dying(&ctx->refs)) | |
7895 | return -ENXIO; | |
7896 | ||
071698e1 | 7897 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7898 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7899 | |
05f3fb3c JA |
7900 | /* |
7901 | * Drop uring mutex before waiting for references to exit. If | |
7902 | * another thread is currently inside io_uring_enter() it might | |
7903 | * need to grab the uring_lock to make progress. If we hold it | |
7904 | * here across the drain wait, then we can deadlock. It's safe | |
7905 | * to drop the mutex here, since no new references will come in | |
7906 | * after we've killed the percpu ref. | |
7907 | */ | |
7908 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7909 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7910 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7911 | if (ret) { |
7912 | percpu_ref_resurrect(&ctx->refs); | |
7913 | ret = -EINTR; | |
7914 | goto out; | |
7915 | } | |
05f3fb3c | 7916 | } |
edafccee JA |
7917 | |
7918 | switch (opcode) { | |
7919 | case IORING_REGISTER_BUFFERS: | |
7920 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
7921 | break; | |
7922 | case IORING_UNREGISTER_BUFFERS: | |
7923 | ret = -EINVAL; | |
7924 | if (arg || nr_args) | |
7925 | break; | |
7926 | ret = io_sqe_buffer_unregister(ctx); | |
7927 | break; | |
6b06314c JA |
7928 | case IORING_REGISTER_FILES: |
7929 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
7930 | break; | |
7931 | case IORING_UNREGISTER_FILES: | |
7932 | ret = -EINVAL; | |
7933 | if (arg || nr_args) | |
7934 | break; | |
7935 | ret = io_sqe_files_unregister(ctx); | |
7936 | break; | |
c3a31e60 JA |
7937 | case IORING_REGISTER_FILES_UPDATE: |
7938 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
7939 | break; | |
9b402849 | 7940 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 7941 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
7942 | ret = -EINVAL; |
7943 | if (nr_args != 1) | |
7944 | break; | |
7945 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
7946 | if (ret) |
7947 | break; | |
7948 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
7949 | ctx->eventfd_async = 1; | |
7950 | else | |
7951 | ctx->eventfd_async = 0; | |
9b402849 JA |
7952 | break; |
7953 | case IORING_UNREGISTER_EVENTFD: | |
7954 | ret = -EINVAL; | |
7955 | if (arg || nr_args) | |
7956 | break; | |
7957 | ret = io_eventfd_unregister(ctx); | |
7958 | break; | |
66f4af93 JA |
7959 | case IORING_REGISTER_PROBE: |
7960 | ret = -EINVAL; | |
7961 | if (!arg || nr_args > 256) | |
7962 | break; | |
7963 | ret = io_probe(ctx, arg, nr_args); | |
7964 | break; | |
071698e1 JA |
7965 | case IORING_REGISTER_PERSONALITY: |
7966 | ret = -EINVAL; | |
7967 | if (arg || nr_args) | |
7968 | break; | |
7969 | ret = io_register_personality(ctx); | |
7970 | break; | |
7971 | case IORING_UNREGISTER_PERSONALITY: | |
7972 | ret = -EINVAL; | |
7973 | if (arg) | |
7974 | break; | |
7975 | ret = io_unregister_personality(ctx, nr_args); | |
7976 | break; | |
edafccee JA |
7977 | default: |
7978 | ret = -EINVAL; | |
7979 | break; | |
7980 | } | |
7981 | ||
071698e1 | 7982 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7983 | /* bring the ctx back to life */ |
05f3fb3c | 7984 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
7985 | out: |
7986 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 7987 | } |
edafccee JA |
7988 | return ret; |
7989 | } | |
7990 | ||
7991 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
7992 | void __user *, arg, unsigned int, nr_args) | |
7993 | { | |
7994 | struct io_ring_ctx *ctx; | |
7995 | long ret = -EBADF; | |
7996 | struct fd f; | |
7997 | ||
7998 | f = fdget(fd); | |
7999 | if (!f.file) | |
8000 | return -EBADF; | |
8001 | ||
8002 | ret = -EOPNOTSUPP; | |
8003 | if (f.file->f_op != &io_uring_fops) | |
8004 | goto out_fput; | |
8005 | ||
8006 | ctx = f.file->private_data; | |
8007 | ||
8008 | mutex_lock(&ctx->uring_lock); | |
8009 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
8010 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
8011 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
8012 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
8013 | out_fput: |
8014 | fdput(f); | |
8015 | return ret; | |
8016 | } | |
8017 | ||
2b188cc1 JA |
8018 | static int __init io_uring_init(void) |
8019 | { | |
d7f62e82 SM |
8020 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
8021 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
8022 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
8023 | } while (0) | |
8024 | ||
8025 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
8026 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
8027 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
8028 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
8029 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
8030 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
8031 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
8032 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
8033 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
8034 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 8035 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
8036 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
8037 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
8038 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
8039 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
8040 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
8041 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
8042 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
8043 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
8044 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
8045 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
8046 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
8047 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
8048 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
8049 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 8050 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
8051 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
8052 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
8053 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 8054 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 8055 | |
d3656344 | 8056 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 8057 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
2b188cc1 JA |
8058 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
8059 | return 0; | |
8060 | }; | |
8061 | __initcall(io_uring_init); |