]>
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> | |
47 | #include <linux/refcount.h> | |
48 | #include <linux/uio.h> | |
6b47ee6e | 49 | #include <linux/bits.h> |
2b188cc1 JA |
50 | |
51 | #include <linux/sched/signal.h> | |
52 | #include <linux/fs.h> | |
53 | #include <linux/file.h> | |
54 | #include <linux/fdtable.h> | |
55 | #include <linux/mm.h> | |
56 | #include <linux/mman.h> | |
57 | #include <linux/mmu_context.h> | |
58 | #include <linux/percpu.h> | |
59 | #include <linux/slab.h> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
2b188cc1 JA |
100 | |
101 | struct io_uring { | |
102 | u32 head ____cacheline_aligned_in_smp; | |
103 | u32 tail ____cacheline_aligned_in_smp; | |
104 | }; | |
105 | ||
1e84b97b | 106 | /* |
75b28aff HV |
107 | * This data is shared with the application through the mmap at offsets |
108 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
109 | * |
110 | * The offsets to the member fields are published through struct | |
111 | * io_sqring_offsets when calling io_uring_setup. | |
112 | */ | |
75b28aff | 113 | struct io_rings { |
1e84b97b SB |
114 | /* |
115 | * Head and tail offsets into the ring; the offsets need to be | |
116 | * masked to get valid indices. | |
117 | * | |
75b28aff HV |
118 | * The kernel controls head of the sq ring and the tail of the cq ring, |
119 | * and the application controls tail of the sq ring and the head of the | |
120 | * cq ring. | |
1e84b97b | 121 | */ |
75b28aff | 122 | struct io_uring sq, cq; |
1e84b97b | 123 | /* |
75b28aff | 124 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
125 | * ring_entries - 1) |
126 | */ | |
75b28aff HV |
127 | u32 sq_ring_mask, cq_ring_mask; |
128 | /* Ring sizes (constant, power of 2) */ | |
129 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
130 | /* |
131 | * Number of invalid entries dropped by the kernel due to | |
132 | * invalid index stored in array | |
133 | * | |
134 | * Written by the kernel, shouldn't be modified by the | |
135 | * application (i.e. get number of "new events" by comparing to | |
136 | * cached value). | |
137 | * | |
138 | * After a new SQ head value was read by the application this | |
139 | * counter includes all submissions that were dropped reaching | |
140 | * the new SQ head (and possibly more). | |
141 | */ | |
75b28aff | 142 | u32 sq_dropped; |
1e84b97b SB |
143 | /* |
144 | * Runtime flags | |
145 | * | |
146 | * Written by the kernel, shouldn't be modified by the | |
147 | * application. | |
148 | * | |
149 | * The application needs a full memory barrier before checking | |
150 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
151 | */ | |
75b28aff | 152 | u32 sq_flags; |
1e84b97b SB |
153 | /* |
154 | * Number of completion events lost because the queue was full; | |
155 | * this should be avoided by the application by making sure | |
0b4295b5 | 156 | * there are not more requests pending than there is space in |
1e84b97b SB |
157 | * the completion queue. |
158 | * | |
159 | * Written by the kernel, shouldn't be modified by the | |
160 | * application (i.e. get number of "new events" by comparing to | |
161 | * cached value). | |
162 | * | |
163 | * As completion events come in out of order this counter is not | |
164 | * ordered with any other data. | |
165 | */ | |
75b28aff | 166 | u32 cq_overflow; |
1e84b97b SB |
167 | /* |
168 | * Ring buffer of completion events. | |
169 | * | |
170 | * The kernel writes completion events fresh every time they are | |
171 | * produced, so the application is allowed to modify pending | |
172 | * entries. | |
173 | */ | |
75b28aff | 174 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
175 | }; |
176 | ||
edafccee JA |
177 | struct io_mapped_ubuf { |
178 | u64 ubuf; | |
179 | size_t len; | |
180 | struct bio_vec *bvec; | |
181 | unsigned int nr_bvecs; | |
182 | }; | |
183 | ||
65e19f54 JA |
184 | struct fixed_file_table { |
185 | struct file **files; | |
31b51510 JA |
186 | }; |
187 | ||
05f3fb3c JA |
188 | struct fixed_file_data { |
189 | struct fixed_file_table *table; | |
190 | struct io_ring_ctx *ctx; | |
191 | ||
192 | struct percpu_ref refs; | |
193 | struct llist_head put_llist; | |
05f3fb3c JA |
194 | struct work_struct ref_work; |
195 | struct completion done; | |
196 | }; | |
197 | ||
2b188cc1 JA |
198 | struct io_ring_ctx { |
199 | struct { | |
200 | struct percpu_ref refs; | |
201 | } ____cacheline_aligned_in_smp; | |
202 | ||
203 | struct { | |
204 | unsigned int flags; | |
e1d85334 RD |
205 | unsigned int compat: 1; |
206 | unsigned int account_mem: 1; | |
207 | unsigned int cq_overflow_flushed: 1; | |
208 | unsigned int drain_next: 1; | |
209 | unsigned int eventfd_async: 1; | |
2b188cc1 | 210 | |
75b28aff HV |
211 | /* |
212 | * Ring buffer of indices into array of io_uring_sqe, which is | |
213 | * mmapped by the application using the IORING_OFF_SQES offset. | |
214 | * | |
215 | * This indirection could e.g. be used to assign fixed | |
216 | * io_uring_sqe entries to operations and only submit them to | |
217 | * the queue when needed. | |
218 | * | |
219 | * The kernel modifies neither the indices array nor the entries | |
220 | * array. | |
221 | */ | |
222 | u32 *sq_array; | |
2b188cc1 JA |
223 | unsigned cached_sq_head; |
224 | unsigned sq_entries; | |
225 | unsigned sq_mask; | |
6c271ce2 | 226 | unsigned sq_thread_idle; |
498ccd9e | 227 | unsigned cached_sq_dropped; |
206aefde | 228 | atomic_t cached_cq_overflow; |
ad3eb2c8 | 229 | unsigned long sq_check_overflow; |
de0617e4 JA |
230 | |
231 | struct list_head defer_list; | |
5262f567 | 232 | struct list_head timeout_list; |
1d7bb1d5 | 233 | struct list_head cq_overflow_list; |
fcb323cc JA |
234 | |
235 | wait_queue_head_t inflight_wait; | |
ad3eb2c8 | 236 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
237 | } ____cacheline_aligned_in_smp; |
238 | ||
206aefde JA |
239 | struct io_rings *rings; |
240 | ||
2b188cc1 | 241 | /* IO offload */ |
561fb04a | 242 | struct io_wq *io_wq; |
6c271ce2 | 243 | struct task_struct *sqo_thread; /* if using sq thread polling */ |
2b188cc1 | 244 | struct mm_struct *sqo_mm; |
6c271ce2 | 245 | wait_queue_head_t sqo_wait; |
75b28aff | 246 | |
6b06314c JA |
247 | /* |
248 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
249 | * readers must ensure that ->refs is alive as long as the file* is | |
250 | * used. Only updated through io_uring_register(2). | |
251 | */ | |
05f3fb3c | 252 | struct fixed_file_data *file_data; |
6b06314c | 253 | unsigned nr_user_files; |
b14cca0c PB |
254 | int ring_fd; |
255 | struct file *ring_file; | |
6b06314c | 256 | |
edafccee JA |
257 | /* if used, fixed mapped user buffers */ |
258 | unsigned nr_user_bufs; | |
259 | struct io_mapped_ubuf *user_bufs; | |
260 | ||
2b188cc1 JA |
261 | struct user_struct *user; |
262 | ||
0b8c0ec7 | 263 | const struct cred *creds; |
181e448d | 264 | |
206aefde JA |
265 | /* 0 is for ctx quiesce/reinit/free, 1 is for sqo_thread started */ |
266 | struct completion *completions; | |
267 | ||
0ddf92e8 JA |
268 | /* if all else fails... */ |
269 | struct io_kiocb *fallback_req; | |
270 | ||
206aefde JA |
271 | #if defined(CONFIG_UNIX) |
272 | struct socket *ring_sock; | |
273 | #endif | |
274 | ||
071698e1 JA |
275 | struct idr personality_idr; |
276 | ||
206aefde JA |
277 | struct { |
278 | unsigned cached_cq_tail; | |
279 | unsigned cq_entries; | |
280 | unsigned cq_mask; | |
281 | atomic_t cq_timeouts; | |
ad3eb2c8 | 282 | unsigned long cq_check_overflow; |
206aefde JA |
283 | struct wait_queue_head cq_wait; |
284 | struct fasync_struct *cq_fasync; | |
285 | struct eventfd_ctx *cq_ev_fd; | |
286 | } ____cacheline_aligned_in_smp; | |
2b188cc1 JA |
287 | |
288 | struct { | |
289 | struct mutex uring_lock; | |
290 | wait_queue_head_t wait; | |
291 | } ____cacheline_aligned_in_smp; | |
292 | ||
293 | struct { | |
294 | spinlock_t completion_lock; | |
e94f141b | 295 | |
def596e9 JA |
296 | /* |
297 | * ->poll_list is protected by the ctx->uring_lock for | |
298 | * io_uring instances that don't use IORING_SETUP_SQPOLL. | |
299 | * For SQPOLL, only the single threaded io_sq_thread() will | |
300 | * manipulate the list, hence no extra locking is needed there. | |
301 | */ | |
302 | struct list_head poll_list; | |
78076bb6 JA |
303 | struct hlist_head *cancel_hash; |
304 | unsigned cancel_hash_bits; | |
e94f141b | 305 | bool poll_multi_file; |
31b51510 | 306 | |
fcb323cc JA |
307 | spinlock_t inflight_lock; |
308 | struct list_head inflight_list; | |
2b188cc1 | 309 | } ____cacheline_aligned_in_smp; |
2b188cc1 JA |
310 | }; |
311 | ||
09bb8394 JA |
312 | /* |
313 | * First field must be the file pointer in all the | |
314 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
315 | */ | |
221c5eb2 JA |
316 | struct io_poll_iocb { |
317 | struct file *file; | |
0969e783 JA |
318 | union { |
319 | struct wait_queue_head *head; | |
320 | u64 addr; | |
321 | }; | |
221c5eb2 | 322 | __poll_t events; |
8c838788 | 323 | bool done; |
221c5eb2 | 324 | bool canceled; |
392edb45 | 325 | struct wait_queue_entry wait; |
221c5eb2 JA |
326 | }; |
327 | ||
b5dba59e JA |
328 | struct io_close { |
329 | struct file *file; | |
330 | struct file *put_file; | |
331 | int fd; | |
332 | }; | |
333 | ||
ad8a48ac JA |
334 | struct io_timeout_data { |
335 | struct io_kiocb *req; | |
336 | struct hrtimer timer; | |
337 | struct timespec64 ts; | |
338 | enum hrtimer_mode mode; | |
cc42e0ac | 339 | u32 seq_offset; |
ad8a48ac JA |
340 | }; |
341 | ||
8ed8d3c3 JA |
342 | struct io_accept { |
343 | struct file *file; | |
344 | struct sockaddr __user *addr; | |
345 | int __user *addr_len; | |
346 | int flags; | |
347 | }; | |
348 | ||
349 | struct io_sync { | |
350 | struct file *file; | |
351 | loff_t len; | |
352 | loff_t off; | |
353 | int flags; | |
d63d1b5e | 354 | int mode; |
8ed8d3c3 JA |
355 | }; |
356 | ||
fbf23849 JA |
357 | struct io_cancel { |
358 | struct file *file; | |
359 | u64 addr; | |
360 | }; | |
361 | ||
b29472ee JA |
362 | struct io_timeout { |
363 | struct file *file; | |
364 | u64 addr; | |
365 | int flags; | |
26a61679 | 366 | unsigned count; |
b29472ee JA |
367 | }; |
368 | ||
9adbd45d JA |
369 | struct io_rw { |
370 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
371 | struct kiocb kiocb; | |
372 | u64 addr; | |
373 | u64 len; | |
374 | }; | |
375 | ||
3fbb51c1 JA |
376 | struct io_connect { |
377 | struct file *file; | |
378 | struct sockaddr __user *addr; | |
379 | int addr_len; | |
380 | }; | |
381 | ||
e47293fd JA |
382 | struct io_sr_msg { |
383 | struct file *file; | |
fddaface JA |
384 | union { |
385 | struct user_msghdr __user *msg; | |
386 | void __user *buf; | |
387 | }; | |
e47293fd | 388 | int msg_flags; |
fddaface | 389 | size_t len; |
e47293fd JA |
390 | }; |
391 | ||
15b71abe JA |
392 | struct io_open { |
393 | struct file *file; | |
394 | int dfd; | |
eddc7ef5 | 395 | union { |
eddc7ef5 JA |
396 | unsigned mask; |
397 | }; | |
15b71abe | 398 | struct filename *filename; |
eddc7ef5 | 399 | struct statx __user *buffer; |
c12cedf2 | 400 | struct open_how how; |
15b71abe JA |
401 | }; |
402 | ||
05f3fb3c JA |
403 | struct io_files_update { |
404 | struct file *file; | |
405 | u64 arg; | |
406 | u32 nr_args; | |
407 | u32 offset; | |
408 | }; | |
409 | ||
4840e418 JA |
410 | struct io_fadvise { |
411 | struct file *file; | |
412 | u64 offset; | |
413 | u32 len; | |
414 | u32 advice; | |
415 | }; | |
416 | ||
c1ca757b JA |
417 | struct io_madvise { |
418 | struct file *file; | |
419 | u64 addr; | |
420 | u32 len; | |
421 | u32 advice; | |
422 | }; | |
423 | ||
3e4827b0 JA |
424 | struct io_epoll { |
425 | struct file *file; | |
426 | int epfd; | |
427 | int op; | |
428 | int fd; | |
429 | struct epoll_event event; | |
e47293fd JA |
430 | }; |
431 | ||
7d67af2c PB |
432 | struct io_splice { |
433 | struct file *file_out; | |
434 | struct file *file_in; | |
435 | loff_t off_out; | |
436 | loff_t off_in; | |
437 | u64 len; | |
438 | unsigned int flags; | |
439 | }; | |
440 | ||
f499a021 JA |
441 | struct io_async_connect { |
442 | struct sockaddr_storage address; | |
443 | }; | |
444 | ||
03b1230c JA |
445 | struct io_async_msghdr { |
446 | struct iovec fast_iov[UIO_FASTIOV]; | |
447 | struct iovec *iov; | |
448 | struct sockaddr __user *uaddr; | |
449 | struct msghdr msg; | |
b537916c | 450 | struct sockaddr_storage addr; |
03b1230c JA |
451 | }; |
452 | ||
f67676d1 JA |
453 | struct io_async_rw { |
454 | struct iovec fast_iov[UIO_FASTIOV]; | |
455 | struct iovec *iov; | |
456 | ssize_t nr_segs; | |
457 | ssize_t size; | |
458 | }; | |
459 | ||
1a6b74fc | 460 | struct io_async_ctx { |
f67676d1 JA |
461 | union { |
462 | struct io_async_rw rw; | |
03b1230c | 463 | struct io_async_msghdr msg; |
f499a021 | 464 | struct io_async_connect connect; |
2d28390a | 465 | struct io_timeout_data timeout; |
f67676d1 | 466 | }; |
1a6b74fc JA |
467 | }; |
468 | ||
6b47ee6e PB |
469 | enum { |
470 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
471 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
472 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
473 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
474 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
475 | ||
476 | REQ_F_LINK_NEXT_BIT, | |
477 | REQ_F_FAIL_LINK_BIT, | |
478 | REQ_F_INFLIGHT_BIT, | |
479 | REQ_F_CUR_POS_BIT, | |
480 | REQ_F_NOWAIT_BIT, | |
481 | REQ_F_IOPOLL_COMPLETED_BIT, | |
482 | REQ_F_LINK_TIMEOUT_BIT, | |
483 | REQ_F_TIMEOUT_BIT, | |
484 | REQ_F_ISREG_BIT, | |
485 | REQ_F_MUST_PUNT_BIT, | |
486 | REQ_F_TIMEOUT_NOSEQ_BIT, | |
487 | REQ_F_COMP_LOCKED_BIT, | |
99bc4c38 | 488 | REQ_F_NEED_CLEANUP_BIT, |
2ca10259 | 489 | REQ_F_OVERFLOW_BIT, |
d7718a9d | 490 | REQ_F_POLLED_BIT, |
6b47ee6e PB |
491 | }; |
492 | ||
493 | enum { | |
494 | /* ctx owns file */ | |
495 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
496 | /* drain existing IO first */ | |
497 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
498 | /* linked sqes */ | |
499 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
500 | /* doesn't sever on completion < 0 */ | |
501 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
502 | /* IOSQE_ASYNC */ | |
503 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
504 | ||
505 | /* already grabbed next link */ | |
506 | REQ_F_LINK_NEXT = BIT(REQ_F_LINK_NEXT_BIT), | |
507 | /* fail rest of links */ | |
508 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
509 | /* on inflight list */ | |
510 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
511 | /* read/write uses file position */ | |
512 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
513 | /* must not punt to workers */ | |
514 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
515 | /* polled IO has completed */ | |
516 | REQ_F_IOPOLL_COMPLETED = BIT(REQ_F_IOPOLL_COMPLETED_BIT), | |
517 | /* has linked timeout */ | |
518 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), | |
519 | /* timeout request */ | |
520 | REQ_F_TIMEOUT = BIT(REQ_F_TIMEOUT_BIT), | |
521 | /* regular file */ | |
522 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
523 | /* must be punted even for NONBLOCK */ | |
524 | REQ_F_MUST_PUNT = BIT(REQ_F_MUST_PUNT_BIT), | |
525 | /* no timeout sequence */ | |
526 | REQ_F_TIMEOUT_NOSEQ = BIT(REQ_F_TIMEOUT_NOSEQ_BIT), | |
527 | /* completion under lock */ | |
528 | REQ_F_COMP_LOCKED = BIT(REQ_F_COMP_LOCKED_BIT), | |
99bc4c38 PB |
529 | /* needs cleanup */ |
530 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
2ca10259 JA |
531 | /* in overflow list */ |
532 | REQ_F_OVERFLOW = BIT(REQ_F_OVERFLOW_BIT), | |
d7718a9d JA |
533 | /* already went through poll handler */ |
534 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
535 | }; | |
536 | ||
537 | struct async_poll { | |
538 | struct io_poll_iocb poll; | |
539 | struct io_wq_work work; | |
6b47ee6e PB |
540 | }; |
541 | ||
09bb8394 JA |
542 | /* |
543 | * NOTE! Each of the iocb union members has the file pointer | |
544 | * as the first entry in their struct definition. So you can | |
545 | * access the file pointer through any of the sub-structs, | |
546 | * or directly as just 'ki_filp' in this struct. | |
547 | */ | |
2b188cc1 | 548 | struct io_kiocb { |
221c5eb2 | 549 | union { |
09bb8394 | 550 | struct file *file; |
9adbd45d | 551 | struct io_rw rw; |
221c5eb2 | 552 | struct io_poll_iocb poll; |
8ed8d3c3 JA |
553 | struct io_accept accept; |
554 | struct io_sync sync; | |
fbf23849 | 555 | struct io_cancel cancel; |
b29472ee | 556 | struct io_timeout timeout; |
3fbb51c1 | 557 | struct io_connect connect; |
e47293fd | 558 | struct io_sr_msg sr_msg; |
15b71abe | 559 | struct io_open open; |
b5dba59e | 560 | struct io_close close; |
05f3fb3c | 561 | struct io_files_update files_update; |
4840e418 | 562 | struct io_fadvise fadvise; |
c1ca757b | 563 | struct io_madvise madvise; |
3e4827b0 | 564 | struct io_epoll epoll; |
7d67af2c | 565 | struct io_splice splice; |
221c5eb2 | 566 | }; |
2b188cc1 | 567 | |
1a6b74fc | 568 | struct io_async_ctx *io; |
cf6fd4bd | 569 | bool needs_fixed_file; |
d625c6ee | 570 | u8 opcode; |
2b188cc1 JA |
571 | |
572 | struct io_ring_ctx *ctx; | |
d7718a9d | 573 | struct list_head list; |
2b188cc1 | 574 | unsigned int flags; |
c16361c1 | 575 | refcount_t refs; |
d7718a9d | 576 | struct task_struct *task; |
2b188cc1 | 577 | u64 user_data; |
9e645e11 | 578 | u32 result; |
de0617e4 | 579 | u32 sequence; |
2b188cc1 | 580 | |
d7718a9d JA |
581 | struct list_head link_list; |
582 | ||
fcb323cc JA |
583 | struct list_head inflight_entry; |
584 | ||
b41e9852 JA |
585 | union { |
586 | /* | |
587 | * Only commands that never go async can use the below fields, | |
d7718a9d JA |
588 | * obviously. Right now only IORING_OP_POLL_ADD uses them, and |
589 | * async armed poll handlers for regular commands. The latter | |
590 | * restore the work, if needed. | |
b41e9852 JA |
591 | */ |
592 | struct { | |
b41e9852 | 593 | struct callback_head task_work; |
d7718a9d JA |
594 | struct hlist_node hash_node; |
595 | struct async_poll *apoll; | |
b41e9852 JA |
596 | }; |
597 | struct io_wq_work work; | |
598 | }; | |
2b188cc1 JA |
599 | }; |
600 | ||
601 | #define IO_PLUG_THRESHOLD 2 | |
def596e9 | 602 | #define IO_IOPOLL_BATCH 8 |
2b188cc1 | 603 | |
9a56a232 JA |
604 | struct io_submit_state { |
605 | struct blk_plug plug; | |
606 | ||
2579f913 JA |
607 | /* |
608 | * io_kiocb alloc cache | |
609 | */ | |
610 | void *reqs[IO_IOPOLL_BATCH]; | |
6c8a3134 | 611 | unsigned int free_reqs; |
2579f913 | 612 | |
9a56a232 JA |
613 | /* |
614 | * File reference cache | |
615 | */ | |
616 | struct file *file; | |
617 | unsigned int fd; | |
618 | unsigned int has_refs; | |
619 | unsigned int used_refs; | |
620 | unsigned int ios_left; | |
621 | }; | |
622 | ||
d3656344 JA |
623 | struct io_op_def { |
624 | /* needs req->io allocated for deferral/async */ | |
625 | unsigned async_ctx : 1; | |
626 | /* needs current->mm setup, does mm access */ | |
627 | unsigned needs_mm : 1; | |
628 | /* needs req->file assigned */ | |
629 | unsigned needs_file : 1; | |
630 | /* needs req->file assigned IFF fd is >= 0 */ | |
631 | unsigned fd_non_neg : 1; | |
632 | /* hash wq insertion if file is a regular file */ | |
633 | unsigned hash_reg_file : 1; | |
634 | /* unbound wq insertion if file is a non-regular file */ | |
635 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
636 | /* opcode is not supported by this kernel */ |
637 | unsigned not_supported : 1; | |
f86cd20c JA |
638 | /* needs file table */ |
639 | unsigned file_table : 1; | |
ff002b30 JA |
640 | /* needs ->fs */ |
641 | unsigned needs_fs : 1; | |
8a72758c JA |
642 | /* set if opcode supports polled "wait" */ |
643 | unsigned pollin : 1; | |
644 | unsigned pollout : 1; | |
d3656344 JA |
645 | }; |
646 | ||
647 | static const struct io_op_def io_op_defs[] = { | |
0463b6c5 PB |
648 | [IORING_OP_NOP] = {}, |
649 | [IORING_OP_READV] = { | |
d3656344 JA |
650 | .async_ctx = 1, |
651 | .needs_mm = 1, | |
652 | .needs_file = 1, | |
653 | .unbound_nonreg_file = 1, | |
8a72758c | 654 | .pollin = 1, |
d3656344 | 655 | }, |
0463b6c5 | 656 | [IORING_OP_WRITEV] = { |
d3656344 JA |
657 | .async_ctx = 1, |
658 | .needs_mm = 1, | |
659 | .needs_file = 1, | |
660 | .hash_reg_file = 1, | |
661 | .unbound_nonreg_file = 1, | |
8a72758c | 662 | .pollout = 1, |
d3656344 | 663 | }, |
0463b6c5 | 664 | [IORING_OP_FSYNC] = { |
d3656344 JA |
665 | .needs_file = 1, |
666 | }, | |
0463b6c5 | 667 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
668 | .needs_file = 1, |
669 | .unbound_nonreg_file = 1, | |
8a72758c | 670 | .pollin = 1, |
d3656344 | 671 | }, |
0463b6c5 | 672 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
673 | .needs_file = 1, |
674 | .hash_reg_file = 1, | |
675 | .unbound_nonreg_file = 1, | |
8a72758c | 676 | .pollout = 1, |
d3656344 | 677 | }, |
0463b6c5 | 678 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
679 | .needs_file = 1, |
680 | .unbound_nonreg_file = 1, | |
681 | }, | |
0463b6c5 PB |
682 | [IORING_OP_POLL_REMOVE] = {}, |
683 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
684 | .needs_file = 1, |
685 | }, | |
0463b6c5 | 686 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
687 | .async_ctx = 1, |
688 | .needs_mm = 1, | |
689 | .needs_file = 1, | |
690 | .unbound_nonreg_file = 1, | |
ff002b30 | 691 | .needs_fs = 1, |
8a72758c | 692 | .pollout = 1, |
d3656344 | 693 | }, |
0463b6c5 | 694 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
695 | .async_ctx = 1, |
696 | .needs_mm = 1, | |
697 | .needs_file = 1, | |
698 | .unbound_nonreg_file = 1, | |
ff002b30 | 699 | .needs_fs = 1, |
8a72758c | 700 | .pollin = 1, |
d3656344 | 701 | }, |
0463b6c5 | 702 | [IORING_OP_TIMEOUT] = { |
d3656344 JA |
703 | .async_ctx = 1, |
704 | .needs_mm = 1, | |
705 | }, | |
0463b6c5 PB |
706 | [IORING_OP_TIMEOUT_REMOVE] = {}, |
707 | [IORING_OP_ACCEPT] = { | |
d3656344 JA |
708 | .needs_mm = 1, |
709 | .needs_file = 1, | |
710 | .unbound_nonreg_file = 1, | |
f86cd20c | 711 | .file_table = 1, |
8a72758c | 712 | .pollin = 1, |
d3656344 | 713 | }, |
0463b6c5 PB |
714 | [IORING_OP_ASYNC_CANCEL] = {}, |
715 | [IORING_OP_LINK_TIMEOUT] = { | |
d3656344 JA |
716 | .async_ctx = 1, |
717 | .needs_mm = 1, | |
718 | }, | |
0463b6c5 | 719 | [IORING_OP_CONNECT] = { |
d3656344 JA |
720 | .async_ctx = 1, |
721 | .needs_mm = 1, | |
722 | .needs_file = 1, | |
723 | .unbound_nonreg_file = 1, | |
8a72758c | 724 | .pollout = 1, |
d3656344 | 725 | }, |
0463b6c5 | 726 | [IORING_OP_FALLOCATE] = { |
d3656344 JA |
727 | .needs_file = 1, |
728 | }, | |
0463b6c5 | 729 | [IORING_OP_OPENAT] = { |
d3656344 JA |
730 | .needs_file = 1, |
731 | .fd_non_neg = 1, | |
f86cd20c | 732 | .file_table = 1, |
ff002b30 | 733 | .needs_fs = 1, |
d3656344 | 734 | }, |
0463b6c5 | 735 | [IORING_OP_CLOSE] = { |
d3656344 | 736 | .needs_file = 1, |
f86cd20c | 737 | .file_table = 1, |
d3656344 | 738 | }, |
0463b6c5 | 739 | [IORING_OP_FILES_UPDATE] = { |
d3656344 | 740 | .needs_mm = 1, |
f86cd20c | 741 | .file_table = 1, |
d3656344 | 742 | }, |
0463b6c5 | 743 | [IORING_OP_STATX] = { |
d3656344 JA |
744 | .needs_mm = 1, |
745 | .needs_file = 1, | |
746 | .fd_non_neg = 1, | |
ff002b30 | 747 | .needs_fs = 1, |
d3656344 | 748 | }, |
0463b6c5 | 749 | [IORING_OP_READ] = { |
3a6820f2 JA |
750 | .needs_mm = 1, |
751 | .needs_file = 1, | |
752 | .unbound_nonreg_file = 1, | |
8a72758c | 753 | .pollin = 1, |
3a6820f2 | 754 | }, |
0463b6c5 | 755 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
756 | .needs_mm = 1, |
757 | .needs_file = 1, | |
758 | .unbound_nonreg_file = 1, | |
8a72758c | 759 | .pollout = 1, |
3a6820f2 | 760 | }, |
0463b6c5 | 761 | [IORING_OP_FADVISE] = { |
4840e418 JA |
762 | .needs_file = 1, |
763 | }, | |
0463b6c5 | 764 | [IORING_OP_MADVISE] = { |
c1ca757b JA |
765 | .needs_mm = 1, |
766 | }, | |
0463b6c5 | 767 | [IORING_OP_SEND] = { |
fddaface JA |
768 | .needs_mm = 1, |
769 | .needs_file = 1, | |
770 | .unbound_nonreg_file = 1, | |
8a72758c | 771 | .pollout = 1, |
fddaface | 772 | }, |
0463b6c5 | 773 | [IORING_OP_RECV] = { |
fddaface JA |
774 | .needs_mm = 1, |
775 | .needs_file = 1, | |
776 | .unbound_nonreg_file = 1, | |
8a72758c | 777 | .pollin = 1, |
fddaface | 778 | }, |
0463b6c5 | 779 | [IORING_OP_OPENAT2] = { |
cebdb986 JA |
780 | .needs_file = 1, |
781 | .fd_non_neg = 1, | |
f86cd20c | 782 | .file_table = 1, |
ff002b30 | 783 | .needs_fs = 1, |
cebdb986 | 784 | }, |
3e4827b0 JA |
785 | [IORING_OP_EPOLL_CTL] = { |
786 | .unbound_nonreg_file = 1, | |
787 | .file_table = 1, | |
788 | }, | |
7d67af2c PB |
789 | [IORING_OP_SPLICE] = { |
790 | .needs_file = 1, | |
791 | .hash_reg_file = 1, | |
792 | .unbound_nonreg_file = 1, | |
793 | } | |
d3656344 JA |
794 | }; |
795 | ||
561fb04a | 796 | static void io_wq_submit_work(struct io_wq_work **workptr); |
78e19bbe | 797 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 798 | static void io_put_req(struct io_kiocb *req); |
978db57e | 799 | static void __io_double_put_req(struct io_kiocb *req); |
94ae5e77 JA |
800 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
801 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c JA |
802 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
803 | struct io_uring_files_update *ip, | |
804 | unsigned nr_args); | |
f86cd20c | 805 | static int io_grab_files(struct io_kiocb *req); |
2faf852d | 806 | static void io_ring_file_ref_flush(struct fixed_file_data *data); |
99bc4c38 | 807 | static void io_cleanup_req(struct io_kiocb *req); |
b41e9852 JA |
808 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
809 | int fd, struct file **out_file, bool fixed); | |
810 | static void __io_queue_sqe(struct io_kiocb *req, | |
811 | const struct io_uring_sqe *sqe); | |
de0617e4 | 812 | |
2b188cc1 JA |
813 | static struct kmem_cache *req_cachep; |
814 | ||
815 | static const struct file_operations io_uring_fops; | |
816 | ||
817 | struct sock *io_uring_get_socket(struct file *file) | |
818 | { | |
819 | #if defined(CONFIG_UNIX) | |
820 | if (file->f_op == &io_uring_fops) { | |
821 | struct io_ring_ctx *ctx = file->private_data; | |
822 | ||
823 | return ctx->ring_sock->sk; | |
824 | } | |
825 | #endif | |
826 | return NULL; | |
827 | } | |
828 | EXPORT_SYMBOL(io_uring_get_socket); | |
829 | ||
830 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) | |
831 | { | |
832 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
833 | ||
206aefde | 834 | complete(&ctx->completions[0]); |
2b188cc1 JA |
835 | } |
836 | ||
837 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) | |
838 | { | |
839 | struct io_ring_ctx *ctx; | |
78076bb6 | 840 | int hash_bits; |
2b188cc1 JA |
841 | |
842 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
843 | if (!ctx) | |
844 | return NULL; | |
845 | ||
0ddf92e8 JA |
846 | ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL); |
847 | if (!ctx->fallback_req) | |
848 | goto err; | |
849 | ||
206aefde JA |
850 | ctx->completions = kmalloc(2 * sizeof(struct completion), GFP_KERNEL); |
851 | if (!ctx->completions) | |
852 | goto err; | |
853 | ||
78076bb6 JA |
854 | /* |
855 | * Use 5 bits less than the max cq entries, that should give us around | |
856 | * 32 entries per hash list if totally full and uniformly spread. | |
857 | */ | |
858 | hash_bits = ilog2(p->cq_entries); | |
859 | hash_bits -= 5; | |
860 | if (hash_bits <= 0) | |
861 | hash_bits = 1; | |
862 | ctx->cancel_hash_bits = hash_bits; | |
863 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
864 | GFP_KERNEL); | |
865 | if (!ctx->cancel_hash) | |
866 | goto err; | |
867 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
868 | ||
21482896 | 869 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
870 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
871 | goto err; | |
2b188cc1 JA |
872 | |
873 | ctx->flags = p->flags; | |
874 | init_waitqueue_head(&ctx->cq_wait); | |
1d7bb1d5 | 875 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
206aefde JA |
876 | init_completion(&ctx->completions[0]); |
877 | init_completion(&ctx->completions[1]); | |
071698e1 | 878 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
879 | mutex_init(&ctx->uring_lock); |
880 | init_waitqueue_head(&ctx->wait); | |
881 | spin_lock_init(&ctx->completion_lock); | |
def596e9 | 882 | INIT_LIST_HEAD(&ctx->poll_list); |
de0617e4 | 883 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 884 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
885 | init_waitqueue_head(&ctx->inflight_wait); |
886 | spin_lock_init(&ctx->inflight_lock); | |
887 | INIT_LIST_HEAD(&ctx->inflight_list); | |
2b188cc1 | 888 | return ctx; |
206aefde | 889 | err: |
0ddf92e8 JA |
890 | if (ctx->fallback_req) |
891 | kmem_cache_free(req_cachep, ctx->fallback_req); | |
206aefde | 892 | kfree(ctx->completions); |
78076bb6 | 893 | kfree(ctx->cancel_hash); |
206aefde JA |
894 | kfree(ctx); |
895 | return NULL; | |
2b188cc1 JA |
896 | } |
897 | ||
9d858b21 | 898 | static inline bool __req_need_defer(struct io_kiocb *req) |
7adf4eaf | 899 | { |
a197f664 JL |
900 | struct io_ring_ctx *ctx = req->ctx; |
901 | ||
498ccd9e JA |
902 | return req->sequence != ctx->cached_cq_tail + ctx->cached_sq_dropped |
903 | + atomic_read(&ctx->cached_cq_overflow); | |
7adf4eaf JA |
904 | } |
905 | ||
9d858b21 | 906 | static inline bool req_need_defer(struct io_kiocb *req) |
de0617e4 | 907 | { |
87987898 | 908 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) |
9d858b21 | 909 | return __req_need_defer(req); |
de0617e4 | 910 | |
9d858b21 | 911 | return false; |
de0617e4 JA |
912 | } |
913 | ||
7adf4eaf | 914 | static struct io_kiocb *io_get_deferred_req(struct io_ring_ctx *ctx) |
de0617e4 JA |
915 | { |
916 | struct io_kiocb *req; | |
917 | ||
7adf4eaf | 918 | req = list_first_entry_or_null(&ctx->defer_list, struct io_kiocb, list); |
9d858b21 | 919 | if (req && !req_need_defer(req)) { |
de0617e4 JA |
920 | list_del_init(&req->list); |
921 | return req; | |
922 | } | |
923 | ||
924 | return NULL; | |
925 | } | |
926 | ||
5262f567 JA |
927 | static struct io_kiocb *io_get_timeout_req(struct io_ring_ctx *ctx) |
928 | { | |
7adf4eaf JA |
929 | struct io_kiocb *req; |
930 | ||
931 | req = list_first_entry_or_null(&ctx->timeout_list, struct io_kiocb, list); | |
93bd25bb JA |
932 | if (req) { |
933 | if (req->flags & REQ_F_TIMEOUT_NOSEQ) | |
934 | return NULL; | |
fb4b3d3f | 935 | if (!__req_need_defer(req)) { |
93bd25bb JA |
936 | list_del_init(&req->list); |
937 | return req; | |
938 | } | |
7adf4eaf JA |
939 | } |
940 | ||
941 | return NULL; | |
5262f567 JA |
942 | } |
943 | ||
de0617e4 | 944 | static void __io_commit_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 945 | { |
75b28aff | 946 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 947 | |
07910158 PB |
948 | /* order cqe stores with ring update */ |
949 | smp_store_release(&rings->cq.tail, ctx->cached_cq_tail); | |
2b188cc1 | 950 | |
07910158 PB |
951 | if (wq_has_sleeper(&ctx->cq_wait)) { |
952 | wake_up_interruptible(&ctx->cq_wait); | |
953 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
2b188cc1 JA |
954 | } |
955 | } | |
956 | ||
cccf0ee8 JA |
957 | static inline void io_req_work_grab_env(struct io_kiocb *req, |
958 | const struct io_op_def *def) | |
959 | { | |
960 | if (!req->work.mm && def->needs_mm) { | |
961 | mmgrab(current->mm); | |
962 | req->work.mm = current->mm; | |
2b188cc1 | 963 | } |
cccf0ee8 JA |
964 | if (!req->work.creds) |
965 | req->work.creds = get_current_cred(); | |
ff002b30 JA |
966 | if (!req->work.fs && def->needs_fs) { |
967 | spin_lock(¤t->fs->lock); | |
968 | if (!current->fs->in_exec) { | |
969 | req->work.fs = current->fs; | |
970 | req->work.fs->users++; | |
971 | } else { | |
972 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
973 | } | |
974 | spin_unlock(¤t->fs->lock); | |
975 | } | |
6ab23144 JA |
976 | if (!req->work.task_pid) |
977 | req->work.task_pid = task_pid_vnr(current); | |
2b188cc1 JA |
978 | } |
979 | ||
cccf0ee8 | 980 | static inline void io_req_work_drop_env(struct io_kiocb *req) |
18d9be1a | 981 | { |
cccf0ee8 JA |
982 | if (req->work.mm) { |
983 | mmdrop(req->work.mm); | |
984 | req->work.mm = NULL; | |
985 | } | |
986 | if (req->work.creds) { | |
987 | put_cred(req->work.creds); | |
988 | req->work.creds = NULL; | |
989 | } | |
ff002b30 JA |
990 | if (req->work.fs) { |
991 | struct fs_struct *fs = req->work.fs; | |
992 | ||
993 | spin_lock(&req->work.fs->lock); | |
994 | if (--fs->users) | |
995 | fs = NULL; | |
996 | spin_unlock(&req->work.fs->lock); | |
997 | if (fs) | |
998 | free_fs_struct(fs); | |
999 | } | |
561fb04a JA |
1000 | } |
1001 | ||
94ae5e77 JA |
1002 | static inline bool io_prep_async_work(struct io_kiocb *req, |
1003 | struct io_kiocb **link) | |
18d9be1a | 1004 | { |
d3656344 | 1005 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
561fb04a | 1006 | bool do_hashed = false; |
54a91f3b | 1007 | |
d3656344 JA |
1008 | if (req->flags & REQ_F_ISREG) { |
1009 | if (def->hash_reg_file) | |
3529d8c2 | 1010 | do_hashed = true; |
d3656344 JA |
1011 | } else { |
1012 | if (def->unbound_nonreg_file) | |
3529d8c2 | 1013 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
54a91f3b | 1014 | } |
cccf0ee8 JA |
1015 | |
1016 | io_req_work_grab_env(req, def); | |
54a91f3b | 1017 | |
94ae5e77 | 1018 | *link = io_prep_linked_timeout(req); |
561fb04a JA |
1019 | return do_hashed; |
1020 | } | |
1021 | ||
a197f664 | 1022 | static inline void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1023 | { |
a197f664 | 1024 | struct io_ring_ctx *ctx = req->ctx; |
94ae5e77 JA |
1025 | struct io_kiocb *link; |
1026 | bool do_hashed; | |
1027 | ||
1028 | do_hashed = io_prep_async_work(req, &link); | |
561fb04a JA |
1029 | |
1030 | trace_io_uring_queue_async_work(ctx, do_hashed, req, &req->work, | |
1031 | req->flags); | |
1032 | if (!do_hashed) { | |
1033 | io_wq_enqueue(ctx->io_wq, &req->work); | |
1034 | } else { | |
1035 | io_wq_enqueue_hashed(ctx->io_wq, &req->work, | |
1036 | file_inode(req->file)); | |
1037 | } | |
94ae5e77 JA |
1038 | |
1039 | if (link) | |
1040 | io_queue_linked_timeout(link); | |
18d9be1a JA |
1041 | } |
1042 | ||
5262f567 JA |
1043 | static void io_kill_timeout(struct io_kiocb *req) |
1044 | { | |
1045 | int ret; | |
1046 | ||
2d28390a | 1047 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
5262f567 JA |
1048 | if (ret != -1) { |
1049 | atomic_inc(&req->ctx->cq_timeouts); | |
842f9612 | 1050 | list_del_init(&req->list); |
78e19bbe | 1051 | io_cqring_fill_event(req, 0); |
ec9c02ad | 1052 | io_put_req(req); |
5262f567 JA |
1053 | } |
1054 | } | |
1055 | ||
1056 | static void io_kill_timeouts(struct io_ring_ctx *ctx) | |
1057 | { | |
1058 | struct io_kiocb *req, *tmp; | |
1059 | ||
1060 | spin_lock_irq(&ctx->completion_lock); | |
1061 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, list) | |
1062 | io_kill_timeout(req); | |
1063 | spin_unlock_irq(&ctx->completion_lock); | |
1064 | } | |
1065 | ||
de0617e4 JA |
1066 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1067 | { | |
1068 | struct io_kiocb *req; | |
1069 | ||
5262f567 JA |
1070 | while ((req = io_get_timeout_req(ctx)) != NULL) |
1071 | io_kill_timeout(req); | |
1072 | ||
de0617e4 JA |
1073 | __io_commit_cqring(ctx); |
1074 | ||
87987898 | 1075 | while ((req = io_get_deferred_req(ctx)) != NULL) |
a197f664 | 1076 | io_queue_async_work(req); |
de0617e4 JA |
1077 | } |
1078 | ||
2b188cc1 JA |
1079 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1080 | { | |
75b28aff | 1081 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1082 | unsigned tail; |
1083 | ||
1084 | tail = ctx->cached_cq_tail; | |
115e12e5 SB |
1085 | /* |
1086 | * writes to the cq entry need to come after reading head; the | |
1087 | * control dependency is enough as we're using WRITE_ONCE to | |
1088 | * fill the cq entry | |
1089 | */ | |
75b28aff | 1090 | if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries) |
2b188cc1 JA |
1091 | return NULL; |
1092 | ||
1093 | ctx->cached_cq_tail++; | |
75b28aff | 1094 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1095 | } |
1096 | ||
f2842ab5 JA |
1097 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1098 | { | |
f0b493e6 JA |
1099 | if (!ctx->cq_ev_fd) |
1100 | return false; | |
f2842ab5 JA |
1101 | if (!ctx->eventfd_async) |
1102 | return true; | |
b41e9852 | 1103 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1104 | } |
1105 | ||
b41e9852 | 1106 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 JA |
1107 | { |
1108 | if (waitqueue_active(&ctx->wait)) | |
1109 | wake_up(&ctx->wait); | |
1110 | if (waitqueue_active(&ctx->sqo_wait)) | |
1111 | wake_up(&ctx->sqo_wait); | |
b41e9852 | 1112 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 JA |
1113 | eventfd_signal(ctx->cq_ev_fd, 1); |
1114 | } | |
1115 | ||
c4a2ed72 JA |
1116 | /* Returns true if there are no backlogged entries after the flush */ |
1117 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) | |
1d7bb1d5 JA |
1118 | { |
1119 | struct io_rings *rings = ctx->rings; | |
1120 | struct io_uring_cqe *cqe; | |
1121 | struct io_kiocb *req; | |
1122 | unsigned long flags; | |
1123 | LIST_HEAD(list); | |
1124 | ||
1125 | if (!force) { | |
1126 | if (list_empty_careful(&ctx->cq_overflow_list)) | |
c4a2ed72 | 1127 | return true; |
1d7bb1d5 JA |
1128 | if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) == |
1129 | rings->cq_ring_entries)) | |
c4a2ed72 | 1130 | return false; |
1d7bb1d5 JA |
1131 | } |
1132 | ||
1133 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1134 | ||
1135 | /* if force is set, the ring is going away. always drop after that */ | |
1136 | if (force) | |
69b3e546 | 1137 | ctx->cq_overflow_flushed = 1; |
1d7bb1d5 | 1138 | |
c4a2ed72 | 1139 | cqe = NULL; |
1d7bb1d5 JA |
1140 | while (!list_empty(&ctx->cq_overflow_list)) { |
1141 | cqe = io_get_cqring(ctx); | |
1142 | if (!cqe && !force) | |
1143 | break; | |
1144 | ||
1145 | req = list_first_entry(&ctx->cq_overflow_list, struct io_kiocb, | |
1146 | list); | |
1147 | list_move(&req->list, &list); | |
2ca10259 | 1148 | req->flags &= ~REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1149 | if (cqe) { |
1150 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1151 | WRITE_ONCE(cqe->res, req->result); | |
1152 | WRITE_ONCE(cqe->flags, 0); | |
1153 | } else { | |
1154 | WRITE_ONCE(ctx->rings->cq_overflow, | |
1155 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | io_commit_cqring(ctx); | |
ad3eb2c8 JA |
1160 | if (cqe) { |
1161 | clear_bit(0, &ctx->sq_check_overflow); | |
1162 | clear_bit(0, &ctx->cq_check_overflow); | |
1163 | } | |
1d7bb1d5 JA |
1164 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1165 | io_cqring_ev_posted(ctx); | |
1166 | ||
1167 | while (!list_empty(&list)) { | |
1168 | req = list_first_entry(&list, struct io_kiocb, list); | |
1169 | list_del(&req->list); | |
ec9c02ad | 1170 | io_put_req(req); |
1d7bb1d5 | 1171 | } |
c4a2ed72 JA |
1172 | |
1173 | return cqe != NULL; | |
1d7bb1d5 JA |
1174 | } |
1175 | ||
78e19bbe | 1176 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
2b188cc1 | 1177 | { |
78e19bbe | 1178 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1179 | struct io_uring_cqe *cqe; |
1180 | ||
78e19bbe | 1181 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1182 | |
2b188cc1 JA |
1183 | /* |
1184 | * If we can't get a cq entry, userspace overflowed the | |
1185 | * submission (by quite a lot). Increment the overflow count in | |
1186 | * the ring. | |
1187 | */ | |
1188 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1189 | if (likely(cqe)) { |
78e19bbe | 1190 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1191 | WRITE_ONCE(cqe->res, res); |
c71ffb67 | 1192 | WRITE_ONCE(cqe->flags, 0); |
1d7bb1d5 | 1193 | } else if (ctx->cq_overflow_flushed) { |
498ccd9e JA |
1194 | WRITE_ONCE(ctx->rings->cq_overflow, |
1195 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
1d7bb1d5 | 1196 | } else { |
ad3eb2c8 JA |
1197 | if (list_empty(&ctx->cq_overflow_list)) { |
1198 | set_bit(0, &ctx->sq_check_overflow); | |
1199 | set_bit(0, &ctx->cq_check_overflow); | |
1200 | } | |
2ca10259 | 1201 | req->flags |= REQ_F_OVERFLOW; |
1d7bb1d5 JA |
1202 | refcount_inc(&req->refs); |
1203 | req->result = res; | |
1204 | list_add_tail(&req->list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1205 | } |
1206 | } | |
1207 | ||
78e19bbe | 1208 | static void io_cqring_add_event(struct io_kiocb *req, long res) |
2b188cc1 | 1209 | { |
78e19bbe | 1210 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1211 | unsigned long flags; |
1212 | ||
1213 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
78e19bbe | 1214 | io_cqring_fill_event(req, res); |
2b188cc1 JA |
1215 | io_commit_cqring(ctx); |
1216 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1217 | ||
8c838788 | 1218 | io_cqring_ev_posted(ctx); |
2b188cc1 JA |
1219 | } |
1220 | ||
0ddf92e8 JA |
1221 | static inline bool io_is_fallback_req(struct io_kiocb *req) |
1222 | { | |
1223 | return req == (struct io_kiocb *) | |
1224 | ((unsigned long) req->ctx->fallback_req & ~1UL); | |
1225 | } | |
1226 | ||
1227 | static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx) | |
1228 | { | |
1229 | struct io_kiocb *req; | |
1230 | ||
1231 | req = ctx->fallback_req; | |
1232 | if (!test_and_set_bit_lock(0, (unsigned long *) ctx->fallback_req)) | |
1233 | return req; | |
1234 | ||
1235 | return NULL; | |
1236 | } | |
1237 | ||
2579f913 JA |
1238 | static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx, |
1239 | struct io_submit_state *state) | |
2b188cc1 | 1240 | { |
fd6fab2c | 1241 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2b188cc1 JA |
1242 | struct io_kiocb *req; |
1243 | ||
2579f913 | 1244 | if (!state) { |
fd6fab2c | 1245 | req = kmem_cache_alloc(req_cachep, gfp); |
2579f913 | 1246 | if (unlikely(!req)) |
0ddf92e8 | 1247 | goto fallback; |
2579f913 JA |
1248 | } else if (!state->free_reqs) { |
1249 | size_t sz; | |
1250 | int ret; | |
1251 | ||
1252 | sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs)); | |
fd6fab2c JA |
1253 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs); |
1254 | ||
1255 | /* | |
1256 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1257 | * retry single alloc to be on the safe side. | |
1258 | */ | |
1259 | if (unlikely(ret <= 0)) { | |
1260 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1261 | if (!state->reqs[0]) | |
0ddf92e8 | 1262 | goto fallback; |
fd6fab2c JA |
1263 | ret = 1; |
1264 | } | |
2579f913 | 1265 | state->free_reqs = ret - 1; |
6c8a3134 | 1266 | req = state->reqs[ret - 1]; |
2579f913 | 1267 | } else { |
2579f913 | 1268 | state->free_reqs--; |
6c8a3134 | 1269 | req = state->reqs[state->free_reqs]; |
2b188cc1 JA |
1270 | } |
1271 | ||
0ddf92e8 | 1272 | got_it: |
1a6b74fc | 1273 | req->io = NULL; |
60c112b0 | 1274 | req->file = NULL; |
2579f913 JA |
1275 | req->ctx = ctx; |
1276 | req->flags = 0; | |
e65ef56d JA |
1277 | /* one is dropped after submission, the other at completion */ |
1278 | refcount_set(&req->refs, 2); | |
9e645e11 | 1279 | req->result = 0; |
561fb04a | 1280 | INIT_IO_WORK(&req->work, io_wq_submit_work); |
2579f913 | 1281 | return req; |
0ddf92e8 JA |
1282 | fallback: |
1283 | req = io_get_fallback_req(ctx); | |
1284 | if (req) | |
1285 | goto got_it; | |
6805b32e | 1286 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
1287 | return NULL; |
1288 | } | |
1289 | ||
8da11c19 PB |
1290 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1291 | bool fixed) | |
1292 | { | |
1293 | if (fixed) | |
1294 | percpu_ref_put(&req->ctx->file_data->refs); | |
1295 | else | |
1296 | fput(file); | |
1297 | } | |
1298 | ||
2b85edfc | 1299 | static void __io_req_do_free(struct io_kiocb *req) |
def596e9 | 1300 | { |
2b85edfc PB |
1301 | if (likely(!io_is_fallback_req(req))) |
1302 | kmem_cache_free(req_cachep, req); | |
1303 | else | |
1304 | clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req); | |
1305 | } | |
1306 | ||
c6ca97b3 | 1307 | static void __io_req_aux_free(struct io_kiocb *req) |
2b188cc1 | 1308 | { |
929a3af9 PB |
1309 | if (req->flags & REQ_F_NEED_CLEANUP) |
1310 | io_cleanup_req(req); | |
1311 | ||
96fd84d8 | 1312 | kfree(req->io); |
8da11c19 PB |
1313 | if (req->file) |
1314 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
cccf0ee8 JA |
1315 | |
1316 | io_req_work_drop_env(req); | |
def596e9 JA |
1317 | } |
1318 | ||
9e645e11 | 1319 | static void __io_free_req(struct io_kiocb *req) |
2b188cc1 | 1320 | { |
c6ca97b3 | 1321 | __io_req_aux_free(req); |
fcb323cc | 1322 | |
fcb323cc | 1323 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 | 1324 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc JA |
1325 | unsigned long flags; |
1326 | ||
1327 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1328 | list_del(&req->inflight_entry); | |
1329 | if (waitqueue_active(&ctx->inflight_wait)) | |
1330 | wake_up(&ctx->inflight_wait); | |
1331 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1332 | } | |
2b85edfc PB |
1333 | |
1334 | percpu_ref_put(&req->ctx->refs); | |
1335 | __io_req_do_free(req); | |
e65ef56d JA |
1336 | } |
1337 | ||
c6ca97b3 JA |
1338 | struct req_batch { |
1339 | void *reqs[IO_IOPOLL_BATCH]; | |
1340 | int to_free; | |
1341 | int need_iter; | |
1342 | }; | |
1343 | ||
1344 | static void io_free_req_many(struct io_ring_ctx *ctx, struct req_batch *rb) | |
1345 | { | |
10fef4be JA |
1346 | int fixed_refs = rb->to_free; |
1347 | ||
c6ca97b3 JA |
1348 | if (!rb->to_free) |
1349 | return; | |
1350 | if (rb->need_iter) { | |
1351 | int i, inflight = 0; | |
1352 | unsigned long flags; | |
1353 | ||
10fef4be | 1354 | fixed_refs = 0; |
c6ca97b3 JA |
1355 | for (i = 0; i < rb->to_free; i++) { |
1356 | struct io_kiocb *req = rb->reqs[i]; | |
1357 | ||
10fef4be | 1358 | if (req->flags & REQ_F_FIXED_FILE) { |
c6ca97b3 | 1359 | req->file = NULL; |
10fef4be JA |
1360 | fixed_refs++; |
1361 | } | |
c6ca97b3 JA |
1362 | if (req->flags & REQ_F_INFLIGHT) |
1363 | inflight++; | |
c6ca97b3 JA |
1364 | __io_req_aux_free(req); |
1365 | } | |
1366 | if (!inflight) | |
1367 | goto do_free; | |
1368 | ||
1369 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1370 | for (i = 0; i < rb->to_free; i++) { | |
1371 | struct io_kiocb *req = rb->reqs[i]; | |
1372 | ||
10fef4be | 1373 | if (req->flags & REQ_F_INFLIGHT) { |
c6ca97b3 JA |
1374 | list_del(&req->inflight_entry); |
1375 | if (!--inflight) | |
1376 | break; | |
1377 | } | |
1378 | } | |
1379 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1380 | ||
1381 | if (waitqueue_active(&ctx->inflight_wait)) | |
1382 | wake_up(&ctx->inflight_wait); | |
1383 | } | |
1384 | do_free: | |
1385 | kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs); | |
10fef4be JA |
1386 | if (fixed_refs) |
1387 | percpu_ref_put_many(&ctx->file_data->refs, fixed_refs); | |
c6ca97b3 | 1388 | percpu_ref_put_many(&ctx->refs, rb->to_free); |
c6ca97b3 | 1389 | rb->to_free = rb->need_iter = 0; |
e65ef56d JA |
1390 | } |
1391 | ||
a197f664 | 1392 | static bool io_link_cancel_timeout(struct io_kiocb *req) |
2665abfd | 1393 | { |
a197f664 | 1394 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1395 | int ret; |
1396 | ||
2d28390a | 1397 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
2665abfd | 1398 | if (ret != -1) { |
78e19bbe | 1399 | io_cqring_fill_event(req, -ECANCELED); |
2665abfd JA |
1400 | io_commit_cqring(ctx); |
1401 | req->flags &= ~REQ_F_LINK; | |
ec9c02ad | 1402 | io_put_req(req); |
2665abfd JA |
1403 | return true; |
1404 | } | |
1405 | ||
1406 | return false; | |
e65ef56d JA |
1407 | } |
1408 | ||
ba816ad6 | 1409 | static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
9e645e11 | 1410 | { |
2665abfd | 1411 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 1412 | bool wake_ev = false; |
9e645e11 | 1413 | |
4d7dd462 JA |
1414 | /* Already got next link */ |
1415 | if (req->flags & REQ_F_LINK_NEXT) | |
1416 | return; | |
1417 | ||
9e645e11 JA |
1418 | /* |
1419 | * The list should never be empty when we are called here. But could | |
1420 | * potentially happen if the chain is messed up, check to be on the | |
1421 | * safe side. | |
1422 | */ | |
4493233e PB |
1423 | while (!list_empty(&req->link_list)) { |
1424 | struct io_kiocb *nxt = list_first_entry(&req->link_list, | |
1425 | struct io_kiocb, link_list); | |
94ae5e77 | 1426 | |
4493233e PB |
1427 | if (unlikely((req->flags & REQ_F_LINK_TIMEOUT) && |
1428 | (nxt->flags & REQ_F_TIMEOUT))) { | |
1429 | list_del_init(&nxt->link_list); | |
94ae5e77 | 1430 | wake_ev |= io_link_cancel_timeout(nxt); |
94ae5e77 JA |
1431 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
1432 | continue; | |
1433 | } | |
9e645e11 | 1434 | |
4493233e PB |
1435 | list_del_init(&req->link_list); |
1436 | if (!list_empty(&nxt->link_list)) | |
1437 | nxt->flags |= REQ_F_LINK; | |
b18fdf71 | 1438 | *nxtptr = nxt; |
94ae5e77 | 1439 | break; |
9e645e11 | 1440 | } |
2665abfd | 1441 | |
4d7dd462 | 1442 | req->flags |= REQ_F_LINK_NEXT; |
2665abfd JA |
1443 | if (wake_ev) |
1444 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * Called if REQ_F_LINK is set, and we fail the head request | |
1449 | */ | |
1450 | static void io_fail_links(struct io_kiocb *req) | |
1451 | { | |
2665abfd | 1452 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd JA |
1453 | unsigned long flags; |
1454 | ||
1455 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
9e645e11 JA |
1456 | |
1457 | while (!list_empty(&req->link_list)) { | |
4493233e PB |
1458 | struct io_kiocb *link = list_first_entry(&req->link_list, |
1459 | struct io_kiocb, link_list); | |
9e645e11 | 1460 | |
4493233e | 1461 | list_del_init(&link->link_list); |
c826bd7a | 1462 | trace_io_uring_fail_link(req, link); |
2665abfd JA |
1463 | |
1464 | if ((req->flags & REQ_F_LINK_TIMEOUT) && | |
d625c6ee | 1465 | link->opcode == IORING_OP_LINK_TIMEOUT) { |
a197f664 | 1466 | io_link_cancel_timeout(link); |
2665abfd | 1467 | } else { |
78e19bbe | 1468 | io_cqring_fill_event(link, -ECANCELED); |
978db57e | 1469 | __io_double_put_req(link); |
2665abfd | 1470 | } |
5d960724 | 1471 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
9e645e11 | 1472 | } |
2665abfd JA |
1473 | |
1474 | io_commit_cqring(ctx); | |
1475 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1476 | io_cqring_ev_posted(ctx); | |
9e645e11 JA |
1477 | } |
1478 | ||
4d7dd462 | 1479 | static void io_req_find_next(struct io_kiocb *req, struct io_kiocb **nxt) |
9e645e11 | 1480 | { |
4d7dd462 | 1481 | if (likely(!(req->flags & REQ_F_LINK))) |
2665abfd | 1482 | return; |
2665abfd | 1483 | |
9e645e11 JA |
1484 | /* |
1485 | * If LINK is set, we have dependent requests in this chain. If we | |
1486 | * didn't fail this request, queue the first one up, moving any other | |
1487 | * dependencies to the next request. In case of failure, fail the rest | |
1488 | * of the chain. | |
1489 | */ | |
2665abfd JA |
1490 | if (req->flags & REQ_F_FAIL_LINK) { |
1491 | io_fail_links(req); | |
7c9e7f0f JA |
1492 | } else if ((req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_COMP_LOCKED)) == |
1493 | REQ_F_LINK_TIMEOUT) { | |
2665abfd JA |
1494 | struct io_ring_ctx *ctx = req->ctx; |
1495 | unsigned long flags; | |
1496 | ||
1497 | /* | |
1498 | * If this is a timeout link, we could be racing with the | |
1499 | * timeout timer. Grab the completion lock for this case to | |
7c9e7f0f | 1500 | * protect against that. |
2665abfd JA |
1501 | */ |
1502 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1503 | io_req_link_next(req, nxt); | |
1504 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1505 | } else { | |
1506 | io_req_link_next(req, nxt); | |
9e645e11 | 1507 | } |
4d7dd462 | 1508 | } |
9e645e11 | 1509 | |
c69f8dbe JL |
1510 | static void io_free_req(struct io_kiocb *req) |
1511 | { | |
944e58bf PB |
1512 | struct io_kiocb *nxt = NULL; |
1513 | ||
1514 | io_req_find_next(req, &nxt); | |
70cf9f32 | 1515 | __io_free_req(req); |
944e58bf PB |
1516 | |
1517 | if (nxt) | |
1518 | io_queue_async_work(nxt); | |
c69f8dbe JL |
1519 | } |
1520 | ||
ba816ad6 JA |
1521 | /* |
1522 | * Drop reference to request, return next in chain (if there is one) if this | |
1523 | * was the last reference to this request. | |
1524 | */ | |
f9bd67f6 | 1525 | __attribute__((nonnull)) |
ec9c02ad | 1526 | static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr) |
e65ef56d | 1527 | { |
2a44f467 JA |
1528 | if (refcount_dec_and_test(&req->refs)) { |
1529 | io_req_find_next(req, nxtptr); | |
4d7dd462 | 1530 | __io_free_req(req); |
2a44f467 | 1531 | } |
2b188cc1 JA |
1532 | } |
1533 | ||
e65ef56d JA |
1534 | static void io_put_req(struct io_kiocb *req) |
1535 | { | |
1536 | if (refcount_dec_and_test(&req->refs)) | |
1537 | io_free_req(req); | |
2b188cc1 JA |
1538 | } |
1539 | ||
978db57e JA |
1540 | /* |
1541 | * Must only be used if we don't need to care about links, usually from | |
1542 | * within the completion handling itself. | |
1543 | */ | |
1544 | static void __io_double_put_req(struct io_kiocb *req) | |
78e19bbe JA |
1545 | { |
1546 | /* drop both submit and complete references */ | |
1547 | if (refcount_sub_and_test(2, &req->refs)) | |
1548 | __io_free_req(req); | |
1549 | } | |
1550 | ||
978db57e JA |
1551 | static void io_double_put_req(struct io_kiocb *req) |
1552 | { | |
1553 | /* drop both submit and complete references */ | |
1554 | if (refcount_sub_and_test(2, &req->refs)) | |
1555 | io_free_req(req); | |
1556 | } | |
1557 | ||
1d7bb1d5 | 1558 | static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush) |
a3a0e43f | 1559 | { |
84f97dc2 JA |
1560 | struct io_rings *rings = ctx->rings; |
1561 | ||
ad3eb2c8 JA |
1562 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1563 | /* | |
1564 | * noflush == true is from the waitqueue handler, just ensure | |
1565 | * we wake up the task, and the next invocation will flush the | |
1566 | * entries. We cannot safely to it from here. | |
1567 | */ | |
1568 | if (noflush && !list_empty(&ctx->cq_overflow_list)) | |
1569 | return -1U; | |
1d7bb1d5 | 1570 | |
ad3eb2c8 JA |
1571 | io_cqring_overflow_flush(ctx, false); |
1572 | } | |
1d7bb1d5 | 1573 | |
a3a0e43f JA |
1574 | /* See comment at the top of this file */ |
1575 | smp_rmb(); | |
ad3eb2c8 | 1576 | return ctx->cached_cq_tail - READ_ONCE(rings->cq.head); |
a3a0e43f JA |
1577 | } |
1578 | ||
fb5ccc98 PB |
1579 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
1580 | { | |
1581 | struct io_rings *rings = ctx->rings; | |
1582 | ||
1583 | /* make sure SQ entry isn't read before tail */ | |
1584 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
1585 | } | |
1586 | ||
8237e045 | 1587 | static inline bool io_req_multi_free(struct req_batch *rb, struct io_kiocb *req) |
e94f141b | 1588 | { |
c6ca97b3 JA |
1589 | if ((req->flags & REQ_F_LINK) || io_is_fallback_req(req)) |
1590 | return false; | |
e94f141b | 1591 | |
c6ca97b3 JA |
1592 | if (!(req->flags & REQ_F_FIXED_FILE) || req->io) |
1593 | rb->need_iter++; | |
1594 | ||
1595 | rb->reqs[rb->to_free++] = req; | |
1596 | if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs))) | |
1597 | io_free_req_many(req->ctx, rb); | |
1598 | return true; | |
e94f141b JA |
1599 | } |
1600 | ||
def596e9 JA |
1601 | /* |
1602 | * Find and free completed poll iocbs | |
1603 | */ | |
1604 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1605 | struct list_head *done) | |
1606 | { | |
8237e045 | 1607 | struct req_batch rb; |
def596e9 | 1608 | struct io_kiocb *req; |
def596e9 | 1609 | |
c6ca97b3 | 1610 | rb.to_free = rb.need_iter = 0; |
def596e9 JA |
1611 | while (!list_empty(done)) { |
1612 | req = list_first_entry(done, struct io_kiocb, list); | |
1613 | list_del(&req->list); | |
1614 | ||
78e19bbe | 1615 | io_cqring_fill_event(req, req->result); |
def596e9 JA |
1616 | (*nr_events)++; |
1617 | ||
8237e045 JA |
1618 | if (refcount_dec_and_test(&req->refs) && |
1619 | !io_req_multi_free(&rb, req)) | |
1620 | io_free_req(req); | |
def596e9 | 1621 | } |
def596e9 | 1622 | |
09bb8394 | 1623 | io_commit_cqring(ctx); |
8237e045 | 1624 | io_free_req_many(ctx, &rb); |
def596e9 JA |
1625 | } |
1626 | ||
1627 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1628 | long min) | |
1629 | { | |
1630 | struct io_kiocb *req, *tmp; | |
1631 | LIST_HEAD(done); | |
1632 | bool spin; | |
1633 | int ret; | |
1634 | ||
1635 | /* | |
1636 | * Only spin for completions if we don't have multiple devices hanging | |
1637 | * off our complete list, and we're under the requested amount. | |
1638 | */ | |
1639 | spin = !ctx->poll_multi_file && *nr_events < min; | |
1640 | ||
1641 | ret = 0; | |
1642 | list_for_each_entry_safe(req, tmp, &ctx->poll_list, list) { | |
9adbd45d | 1643 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
1644 | |
1645 | /* | |
1646 | * Move completed entries to our local list. If we find a | |
1647 | * request that requires polling, break out and complete | |
1648 | * the done list first, if we have entries there. | |
1649 | */ | |
1650 | if (req->flags & REQ_F_IOPOLL_COMPLETED) { | |
1651 | list_move_tail(&req->list, &done); | |
1652 | continue; | |
1653 | } | |
1654 | if (!list_empty(&done)) | |
1655 | break; | |
1656 | ||
1657 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
1658 | if (ret < 0) | |
1659 | break; | |
1660 | ||
1661 | if (ret && spin) | |
1662 | spin = false; | |
1663 | ret = 0; | |
1664 | } | |
1665 | ||
1666 | if (!list_empty(&done)) | |
1667 | io_iopoll_complete(ctx, nr_events, &done); | |
1668 | ||
1669 | return ret; | |
1670 | } | |
1671 | ||
1672 | /* | |
d195a66e | 1673 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
1674 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
1675 | * as a non-spinning completion check. | |
1676 | */ | |
1677 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
1678 | long min) | |
1679 | { | |
08f5439f | 1680 | while (!list_empty(&ctx->poll_list) && !need_resched()) { |
def596e9 JA |
1681 | int ret; |
1682 | ||
1683 | ret = io_do_iopoll(ctx, nr_events, min); | |
1684 | if (ret < 0) | |
1685 | return ret; | |
1686 | if (!min || *nr_events >= min) | |
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | return 1; | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * We can't just wait for polled events to come to us, we have to actively | |
1695 | * find and complete them. | |
1696 | */ | |
1697 | static void io_iopoll_reap_events(struct io_ring_ctx *ctx) | |
1698 | { | |
1699 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
1700 | return; | |
1701 | ||
1702 | mutex_lock(&ctx->uring_lock); | |
1703 | while (!list_empty(&ctx->poll_list)) { | |
1704 | unsigned int nr_events = 0; | |
1705 | ||
1706 | io_iopoll_getevents(ctx, &nr_events, 1); | |
08f5439f JA |
1707 | |
1708 | /* | |
1709 | * Ensure we allow local-to-the-cpu processing to take place, | |
1710 | * in this case we need to ensure that we reap all events. | |
1711 | */ | |
1712 | cond_resched(); | |
def596e9 JA |
1713 | } |
1714 | mutex_unlock(&ctx->uring_lock); | |
1715 | } | |
1716 | ||
c7849be9 XW |
1717 | static int io_iopoll_check(struct io_ring_ctx *ctx, unsigned *nr_events, |
1718 | long min) | |
def596e9 | 1719 | { |
2b2ed975 | 1720 | int iters = 0, ret = 0; |
500f9fba | 1721 | |
c7849be9 XW |
1722 | /* |
1723 | * We disallow the app entering submit/complete with polling, but we | |
1724 | * still need to lock the ring to prevent racing with polled issue | |
1725 | * that got punted to a workqueue. | |
1726 | */ | |
1727 | mutex_lock(&ctx->uring_lock); | |
def596e9 JA |
1728 | do { |
1729 | int tmin = 0; | |
1730 | ||
a3a0e43f JA |
1731 | /* |
1732 | * Don't enter poll loop if we already have events pending. | |
1733 | * If we do, we can potentially be spinning for commands that | |
1734 | * already triggered a CQE (eg in error). | |
1735 | */ | |
1d7bb1d5 | 1736 | if (io_cqring_events(ctx, false)) |
a3a0e43f JA |
1737 | break; |
1738 | ||
500f9fba JA |
1739 | /* |
1740 | * If a submit got punted to a workqueue, we can have the | |
1741 | * application entering polling for a command before it gets | |
1742 | * issued. That app will hold the uring_lock for the duration | |
1743 | * of the poll right here, so we need to take a breather every | |
1744 | * now and then to ensure that the issue has a chance to add | |
1745 | * the poll to the issued list. Otherwise we can spin here | |
1746 | * forever, while the workqueue is stuck trying to acquire the | |
1747 | * very same mutex. | |
1748 | */ | |
1749 | if (!(++iters & 7)) { | |
1750 | mutex_unlock(&ctx->uring_lock); | |
1751 | mutex_lock(&ctx->uring_lock); | |
1752 | } | |
1753 | ||
def596e9 JA |
1754 | if (*nr_events < min) |
1755 | tmin = min - *nr_events; | |
1756 | ||
1757 | ret = io_iopoll_getevents(ctx, nr_events, tmin); | |
1758 | if (ret <= 0) | |
1759 | break; | |
1760 | ret = 0; | |
1761 | } while (min && !*nr_events && !need_resched()); | |
1762 | ||
500f9fba | 1763 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
1764 | return ret; |
1765 | } | |
1766 | ||
491381ce | 1767 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 1768 | { |
491381ce JA |
1769 | /* |
1770 | * Tell lockdep we inherited freeze protection from submission | |
1771 | * thread. | |
1772 | */ | |
1773 | if (req->flags & REQ_F_ISREG) { | |
1774 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 1775 | |
491381ce | 1776 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 1777 | } |
491381ce | 1778 | file_end_write(req->file); |
2b188cc1 JA |
1779 | } |
1780 | ||
4e88d6e7 JA |
1781 | static inline void req_set_fail_links(struct io_kiocb *req) |
1782 | { | |
1783 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1784 | req->flags |= REQ_F_FAIL_LINK; | |
1785 | } | |
1786 | ||
ba816ad6 | 1787 | static void io_complete_rw_common(struct kiocb *kiocb, long res) |
2b188cc1 | 1788 | { |
9adbd45d | 1789 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2b188cc1 | 1790 | |
491381ce JA |
1791 | if (kiocb->ki_flags & IOCB_WRITE) |
1792 | kiocb_end_write(req); | |
2b188cc1 | 1793 | |
4e88d6e7 JA |
1794 | if (res != req->result) |
1795 | req_set_fail_links(req); | |
78e19bbe | 1796 | io_cqring_add_event(req, res); |
ba816ad6 JA |
1797 | } |
1798 | ||
1799 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
1800 | { | |
9adbd45d | 1801 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 JA |
1802 | |
1803 | io_complete_rw_common(kiocb, res); | |
e65ef56d | 1804 | io_put_req(req); |
2b188cc1 JA |
1805 | } |
1806 | ||
ba816ad6 JA |
1807 | static struct io_kiocb *__io_complete_rw(struct kiocb *kiocb, long res) |
1808 | { | |
9adbd45d | 1809 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ec9c02ad | 1810 | struct io_kiocb *nxt = NULL; |
ba816ad6 JA |
1811 | |
1812 | io_complete_rw_common(kiocb, res); | |
ec9c02ad JL |
1813 | io_put_req_find_next(req, &nxt); |
1814 | ||
1815 | return nxt; | |
2b188cc1 JA |
1816 | } |
1817 | ||
def596e9 JA |
1818 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
1819 | { | |
9adbd45d | 1820 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 1821 | |
491381ce JA |
1822 | if (kiocb->ki_flags & IOCB_WRITE) |
1823 | kiocb_end_write(req); | |
def596e9 | 1824 | |
4e88d6e7 JA |
1825 | if (res != req->result) |
1826 | req_set_fail_links(req); | |
9e645e11 | 1827 | req->result = res; |
def596e9 JA |
1828 | if (res != -EAGAIN) |
1829 | req->flags |= REQ_F_IOPOLL_COMPLETED; | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * After the iocb has been issued, it's safe to be found on the poll list. | |
1834 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
1835 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
1836 | * accessing the kiocb cookie. | |
1837 | */ | |
1838 | static void io_iopoll_req_issued(struct io_kiocb *req) | |
1839 | { | |
1840 | struct io_ring_ctx *ctx = req->ctx; | |
1841 | ||
1842 | /* | |
1843 | * Track whether we have multiple files in our lists. This will impact | |
1844 | * how we do polling eventually, not spinning if we're on potentially | |
1845 | * different devices. | |
1846 | */ | |
1847 | if (list_empty(&ctx->poll_list)) { | |
1848 | ctx->poll_multi_file = false; | |
1849 | } else if (!ctx->poll_multi_file) { | |
1850 | struct io_kiocb *list_req; | |
1851 | ||
1852 | list_req = list_first_entry(&ctx->poll_list, struct io_kiocb, | |
1853 | list); | |
9adbd45d | 1854 | if (list_req->file != req->file) |
def596e9 JA |
1855 | ctx->poll_multi_file = true; |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * For fast devices, IO may have already completed. If it has, add | |
1860 | * it to the front so we find it first. | |
1861 | */ | |
1862 | if (req->flags & REQ_F_IOPOLL_COMPLETED) | |
1863 | list_add(&req->list, &ctx->poll_list); | |
1864 | else | |
1865 | list_add_tail(&req->list, &ctx->poll_list); | |
bdcd3eab XW |
1866 | |
1867 | if ((ctx->flags & IORING_SETUP_SQPOLL) && | |
1868 | wq_has_sleeper(&ctx->sqo_wait)) | |
1869 | wake_up(&ctx->sqo_wait); | |
def596e9 JA |
1870 | } |
1871 | ||
3d6770fb | 1872 | static void io_file_put(struct io_submit_state *state) |
9a56a232 | 1873 | { |
3d6770fb | 1874 | if (state->file) { |
9a56a232 JA |
1875 | int diff = state->has_refs - state->used_refs; |
1876 | ||
1877 | if (diff) | |
1878 | fput_many(state->file, diff); | |
1879 | state->file = NULL; | |
1880 | } | |
1881 | } | |
1882 | ||
1883 | /* | |
1884 | * Get as many references to a file as we have IOs left in this submission, | |
1885 | * assuming most submissions are for one file, or at least that each file | |
1886 | * has more than one submission. | |
1887 | */ | |
8da11c19 | 1888 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
1889 | { |
1890 | if (!state) | |
1891 | return fget(fd); | |
1892 | ||
1893 | if (state->file) { | |
1894 | if (state->fd == fd) { | |
1895 | state->used_refs++; | |
1896 | state->ios_left--; | |
1897 | return state->file; | |
1898 | } | |
3d6770fb | 1899 | io_file_put(state); |
9a56a232 JA |
1900 | } |
1901 | state->file = fget_many(fd, state->ios_left); | |
1902 | if (!state->file) | |
1903 | return NULL; | |
1904 | ||
1905 | state->fd = fd; | |
1906 | state->has_refs = state->ios_left; | |
1907 | state->used_refs = 1; | |
1908 | state->ios_left--; | |
1909 | return state->file; | |
1910 | } | |
1911 | ||
2b188cc1 JA |
1912 | /* |
1913 | * If we tracked the file through the SCM inflight mechanism, we could support | |
1914 | * any file. For now, just ensure that anything potentially problematic is done | |
1915 | * inline. | |
1916 | */ | |
1917 | static bool io_file_supports_async(struct file *file) | |
1918 | { | |
1919 | umode_t mode = file_inode(file)->i_mode; | |
1920 | ||
10d59345 | 1921 | if (S_ISBLK(mode) || S_ISCHR(mode) || S_ISSOCK(mode)) |
2b188cc1 JA |
1922 | return true; |
1923 | if (S_ISREG(mode) && file->f_op != &io_uring_fops) | |
1924 | return true; | |
1925 | ||
1926 | return false; | |
1927 | } | |
1928 | ||
3529d8c2 JA |
1929 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
1930 | bool force_nonblock) | |
2b188cc1 | 1931 | { |
def596e9 | 1932 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 1933 | struct kiocb *kiocb = &req->rw.kiocb; |
09bb8394 JA |
1934 | unsigned ioprio; |
1935 | int ret; | |
2b188cc1 | 1936 | |
491381ce JA |
1937 | if (S_ISREG(file_inode(req->file)->i_mode)) |
1938 | req->flags |= REQ_F_ISREG; | |
1939 | ||
2b188cc1 | 1940 | kiocb->ki_pos = READ_ONCE(sqe->off); |
ba04291e JA |
1941 | if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) { |
1942 | req->flags |= REQ_F_CUR_POS; | |
1943 | kiocb->ki_pos = req->file->f_pos; | |
1944 | } | |
2b188cc1 | 1945 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
1946 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
1947 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
1948 | if (unlikely(ret)) | |
1949 | return ret; | |
2b188cc1 JA |
1950 | |
1951 | ioprio = READ_ONCE(sqe->ioprio); | |
1952 | if (ioprio) { | |
1953 | ret = ioprio_check_cap(ioprio); | |
1954 | if (ret) | |
09bb8394 | 1955 | return ret; |
2b188cc1 JA |
1956 | |
1957 | kiocb->ki_ioprio = ioprio; | |
1958 | } else | |
1959 | kiocb->ki_ioprio = get_current_ioprio(); | |
1960 | ||
8449eeda | 1961 | /* don't allow async punt if RWF_NOWAIT was requested */ |
491381ce JA |
1962 | if ((kiocb->ki_flags & IOCB_NOWAIT) || |
1963 | (req->file->f_flags & O_NONBLOCK)) | |
8449eeda SB |
1964 | req->flags |= REQ_F_NOWAIT; |
1965 | ||
1966 | if (force_nonblock) | |
2b188cc1 | 1967 | kiocb->ki_flags |= IOCB_NOWAIT; |
8449eeda | 1968 | |
def596e9 | 1969 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
1970 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
1971 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 1972 | return -EOPNOTSUPP; |
2b188cc1 | 1973 | |
def596e9 JA |
1974 | kiocb->ki_flags |= IOCB_HIPRI; |
1975 | kiocb->ki_complete = io_complete_rw_iopoll; | |
6873e0bd | 1976 | req->result = 0; |
def596e9 | 1977 | } else { |
09bb8394 JA |
1978 | if (kiocb->ki_flags & IOCB_HIPRI) |
1979 | return -EINVAL; | |
def596e9 JA |
1980 | kiocb->ki_complete = io_complete_rw; |
1981 | } | |
9adbd45d | 1982 | |
3529d8c2 JA |
1983 | req->rw.addr = READ_ONCE(sqe->addr); |
1984 | req->rw.len = READ_ONCE(sqe->len); | |
9adbd45d JA |
1985 | /* we own ->private, reuse it for the buffer index */ |
1986 | req->rw.kiocb.private = (void *) (unsigned long) | |
3529d8c2 | 1987 | READ_ONCE(sqe->buf_index); |
2b188cc1 | 1988 | return 0; |
2b188cc1 JA |
1989 | } |
1990 | ||
1991 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
1992 | { | |
1993 | switch (ret) { | |
1994 | case -EIOCBQUEUED: | |
1995 | break; | |
1996 | case -ERESTARTSYS: | |
1997 | case -ERESTARTNOINTR: | |
1998 | case -ERESTARTNOHAND: | |
1999 | case -ERESTART_RESTARTBLOCK: | |
2000 | /* | |
2001 | * We can't just restart the syscall, since previously | |
2002 | * submitted sqes may already be in progress. Just fail this | |
2003 | * IO with EINTR. | |
2004 | */ | |
2005 | ret = -EINTR; | |
2006 | /* fall through */ | |
2007 | default: | |
2008 | kiocb->ki_complete(kiocb, ret, 0); | |
2009 | } | |
2010 | } | |
2011 | ||
bcaec089 | 2012 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, struct io_kiocb **nxt) |
ba816ad6 | 2013 | { |
ba04291e JA |
2014 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
2015 | ||
2016 | if (req->flags & REQ_F_CUR_POS) | |
2017 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2018 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
ba816ad6 JA |
2019 | *nxt = __io_complete_rw(kiocb, ret); |
2020 | else | |
2021 | io_rw_done(kiocb, ret); | |
2022 | } | |
2023 | ||
9adbd45d | 2024 | static ssize_t io_import_fixed(struct io_kiocb *req, int rw, |
7d009165 | 2025 | struct iov_iter *iter) |
edafccee | 2026 | { |
9adbd45d JA |
2027 | struct io_ring_ctx *ctx = req->ctx; |
2028 | size_t len = req->rw.len; | |
edafccee JA |
2029 | struct io_mapped_ubuf *imu; |
2030 | unsigned index, buf_index; | |
2031 | size_t offset; | |
2032 | u64 buf_addr; | |
2033 | ||
2034 | /* attempt to use fixed buffers without having provided iovecs */ | |
2035 | if (unlikely(!ctx->user_bufs)) | |
2036 | return -EFAULT; | |
2037 | ||
9adbd45d | 2038 | buf_index = (unsigned long) req->rw.kiocb.private; |
edafccee JA |
2039 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2040 | return -EFAULT; | |
2041 | ||
2042 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2043 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2044 | buf_addr = req->rw.addr; |
edafccee JA |
2045 | |
2046 | /* overflow */ | |
2047 | if (buf_addr + len < buf_addr) | |
2048 | return -EFAULT; | |
2049 | /* not inside the mapped region */ | |
2050 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2051 | return -EFAULT; | |
2052 | ||
2053 | /* | |
2054 | * May not be a start of buffer, set size appropriately | |
2055 | * and advance us to the beginning. | |
2056 | */ | |
2057 | offset = buf_addr - imu->ubuf; | |
2058 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2059 | |
2060 | if (offset) { | |
2061 | /* | |
2062 | * Don't use iov_iter_advance() here, as it's really slow for | |
2063 | * using the latter parts of a big fixed buffer - it iterates | |
2064 | * over each segment manually. We can cheat a bit here, because | |
2065 | * we know that: | |
2066 | * | |
2067 | * 1) it's a BVEC iter, we set it up | |
2068 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2069 | * first and last bvec | |
2070 | * | |
2071 | * So just find our index, and adjust the iterator afterwards. | |
2072 | * If the offset is within the first bvec (or the whole first | |
2073 | * bvec, just use iov_iter_advance(). This makes it easier | |
2074 | * since we can just skip the first segment, which may not | |
2075 | * be PAGE_SIZE aligned. | |
2076 | */ | |
2077 | const struct bio_vec *bvec = imu->bvec; | |
2078 | ||
2079 | if (offset <= bvec->bv_len) { | |
2080 | iov_iter_advance(iter, offset); | |
2081 | } else { | |
2082 | unsigned long seg_skip; | |
2083 | ||
2084 | /* skip first vec */ | |
2085 | offset -= bvec->bv_len; | |
2086 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2087 | ||
2088 | iter->bvec = bvec + seg_skip; | |
2089 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2090 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2091 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2092 | } |
2093 | } | |
2094 | ||
5e559561 | 2095 | return len; |
edafccee JA |
2096 | } |
2097 | ||
cf6fd4bd PB |
2098 | static ssize_t io_import_iovec(int rw, struct io_kiocb *req, |
2099 | struct iovec **iovec, struct iov_iter *iter) | |
2b188cc1 | 2100 | { |
9adbd45d JA |
2101 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2102 | size_t sqe_len = req->rw.len; | |
edafccee JA |
2103 | u8 opcode; |
2104 | ||
d625c6ee | 2105 | opcode = req->opcode; |
7d009165 | 2106 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2107 | *iovec = NULL; |
9adbd45d | 2108 | return io_import_fixed(req, rw, iter); |
edafccee | 2109 | } |
2b188cc1 | 2110 | |
9adbd45d JA |
2111 | /* buffer index only valid with fixed read/write */ |
2112 | if (req->rw.kiocb.private) | |
2113 | return -EINVAL; | |
2114 | ||
3a6820f2 JA |
2115 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
2116 | ssize_t ret; | |
2117 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); | |
2118 | *iovec = NULL; | |
3a901598 | 2119 | return ret < 0 ? ret : sqe_len; |
3a6820f2 JA |
2120 | } |
2121 | ||
f67676d1 JA |
2122 | if (req->io) { |
2123 | struct io_async_rw *iorw = &req->io->rw; | |
2124 | ||
2125 | *iovec = iorw->iov; | |
2126 | iov_iter_init(iter, rw, *iovec, iorw->nr_segs, iorw->size); | |
2127 | if (iorw->iov == iorw->fast_iov) | |
2128 | *iovec = NULL; | |
2129 | return iorw->size; | |
2130 | } | |
2131 | ||
2b188cc1 | 2132 | #ifdef CONFIG_COMPAT |
cf6fd4bd | 2133 | if (req->ctx->compat) |
2b188cc1 JA |
2134 | return compat_import_iovec(rw, buf, sqe_len, UIO_FASTIOV, |
2135 | iovec, iter); | |
2136 | #endif | |
2137 | ||
2138 | return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter); | |
2139 | } | |
2140 | ||
31b51510 | 2141 | /* |
32960613 JA |
2142 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2143 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2144 | */ |
32960613 JA |
2145 | static ssize_t loop_rw_iter(int rw, struct file *file, struct kiocb *kiocb, |
2146 | struct iov_iter *iter) | |
2147 | { | |
2148 | ssize_t ret = 0; | |
2149 | ||
2150 | /* | |
2151 | * Don't support polled IO through this interface, and we can't | |
2152 | * support non-blocking either. For the latter, this just causes | |
2153 | * the kiocb to be handled from an async context. | |
2154 | */ | |
2155 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2156 | return -EOPNOTSUPP; | |
2157 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2158 | return -EAGAIN; | |
2159 | ||
2160 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2161 | struct iovec iovec; |
32960613 JA |
2162 | ssize_t nr; |
2163 | ||
311ae9e1 PB |
2164 | if (!iov_iter_is_bvec(iter)) { |
2165 | iovec = iov_iter_iovec(iter); | |
2166 | } else { | |
2167 | /* fixed buffers import bvec */ | |
2168 | iovec.iov_base = kmap(iter->bvec->bv_page) | |
2169 | + iter->iov_offset; | |
2170 | iovec.iov_len = min(iter->count, | |
2171 | iter->bvec->bv_len - iter->iov_offset); | |
2172 | } | |
2173 | ||
32960613 JA |
2174 | if (rw == READ) { |
2175 | nr = file->f_op->read(file, iovec.iov_base, | |
2176 | iovec.iov_len, &kiocb->ki_pos); | |
2177 | } else { | |
2178 | nr = file->f_op->write(file, iovec.iov_base, | |
2179 | iovec.iov_len, &kiocb->ki_pos); | |
2180 | } | |
2181 | ||
311ae9e1 PB |
2182 | if (iov_iter_is_bvec(iter)) |
2183 | kunmap(iter->bvec->bv_page); | |
2184 | ||
32960613 JA |
2185 | if (nr < 0) { |
2186 | if (!ret) | |
2187 | ret = nr; | |
2188 | break; | |
2189 | } | |
2190 | ret += nr; | |
2191 | if (nr != iovec.iov_len) | |
2192 | break; | |
2193 | iov_iter_advance(iter, nr); | |
2194 | } | |
2195 | ||
2196 | return ret; | |
2197 | } | |
2198 | ||
b7bb4f7d | 2199 | static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size, |
f67676d1 JA |
2200 | struct iovec *iovec, struct iovec *fast_iov, |
2201 | struct iov_iter *iter) | |
2202 | { | |
2203 | req->io->rw.nr_segs = iter->nr_segs; | |
2204 | req->io->rw.size = io_size; | |
2205 | req->io->rw.iov = iovec; | |
2206 | if (!req->io->rw.iov) { | |
2207 | req->io->rw.iov = req->io->rw.fast_iov; | |
2208 | memcpy(req->io->rw.iov, fast_iov, | |
2209 | sizeof(struct iovec) * iter->nr_segs); | |
99bc4c38 PB |
2210 | } else { |
2211 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
2212 | } |
2213 | } | |
2214 | ||
b7bb4f7d | 2215 | static int io_alloc_async_ctx(struct io_kiocb *req) |
f67676d1 | 2216 | { |
d3656344 JA |
2217 | if (!io_op_defs[req->opcode].async_ctx) |
2218 | return 0; | |
f67676d1 | 2219 | req->io = kmalloc(sizeof(*req->io), GFP_KERNEL); |
06b76d44 | 2220 | return req->io == NULL; |
b7bb4f7d JA |
2221 | } |
2222 | ||
b7bb4f7d JA |
2223 | static int io_setup_async_rw(struct io_kiocb *req, ssize_t io_size, |
2224 | struct iovec *iovec, struct iovec *fast_iov, | |
2225 | struct iov_iter *iter) | |
2226 | { | |
980ad263 | 2227 | if (!io_op_defs[req->opcode].async_ctx) |
74566df3 | 2228 | return 0; |
5d204bcf JA |
2229 | if (!req->io) { |
2230 | if (io_alloc_async_ctx(req)) | |
2231 | return -ENOMEM; | |
b7bb4f7d | 2232 | |
5d204bcf JA |
2233 | io_req_map_rw(req, io_size, iovec, fast_iov, iter); |
2234 | } | |
b7bb4f7d | 2235 | return 0; |
f67676d1 JA |
2236 | } |
2237 | ||
3529d8c2 JA |
2238 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2239 | bool force_nonblock) | |
f67676d1 | 2240 | { |
3529d8c2 JA |
2241 | struct io_async_ctx *io; |
2242 | struct iov_iter iter; | |
f67676d1 JA |
2243 | ssize_t ret; |
2244 | ||
3529d8c2 JA |
2245 | ret = io_prep_rw(req, sqe, force_nonblock); |
2246 | if (ret) | |
2247 | return ret; | |
f67676d1 | 2248 | |
3529d8c2 JA |
2249 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
2250 | return -EBADF; | |
f67676d1 | 2251 | |
5f798bea PB |
2252 | /* either don't need iovec imported or already have it */ |
2253 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2254 | return 0; |
2255 | ||
2256 | io = req->io; | |
2257 | io->rw.iov = io->rw.fast_iov; | |
2258 | req->io = NULL; | |
2259 | ret = io_import_iovec(READ, req, &io->rw.iov, &iter); | |
2260 | req->io = io; | |
2261 | if (ret < 0) | |
2262 | return ret; | |
2263 | ||
2264 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2265 | return 0; | |
f67676d1 JA |
2266 | } |
2267 | ||
267bc904 | 2268 | static int io_read(struct io_kiocb *req, struct io_kiocb **nxt, |
8358e3a8 | 2269 | bool force_nonblock) |
2b188cc1 JA |
2270 | { |
2271 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2272 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2273 | struct iov_iter iter; |
31b51510 | 2274 | size_t iov_count; |
f67676d1 | 2275 | ssize_t io_size, ret; |
2b188cc1 | 2276 | |
3529d8c2 | 2277 | ret = io_import_iovec(READ, req, &iovec, &iter); |
06b76d44 JA |
2278 | if (ret < 0) |
2279 | return ret; | |
2b188cc1 | 2280 | |
fd6c2e4c JA |
2281 | /* Ensure we clear previously set non-block flag */ |
2282 | if (!force_nonblock) | |
29de5f6a | 2283 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2284 | |
797f3f53 | 2285 | req->result = 0; |
f67676d1 | 2286 | io_size = ret; |
9e645e11 | 2287 | if (req->flags & REQ_F_LINK) |
f67676d1 JA |
2288 | req->result = io_size; |
2289 | ||
2290 | /* | |
2291 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2292 | * we know to async punt it even if it was opened O_NONBLOCK | |
2293 | */ | |
29de5f6a | 2294 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2295 | goto copy_iov; |
9e645e11 | 2296 | |
31b51510 | 2297 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2298 | ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 JA |
2299 | if (!ret) { |
2300 | ssize_t ret2; | |
2301 | ||
9adbd45d JA |
2302 | if (req->file->f_op->read_iter) |
2303 | ret2 = call_read_iter(req->file, kiocb, &iter); | |
32960613 | 2304 | else |
9adbd45d | 2305 | ret2 = loop_rw_iter(READ, req->file, kiocb, &iter); |
32960613 | 2306 | |
9d93a3f5 | 2307 | /* Catch -EAGAIN return for forced non-blocking submission */ |
f67676d1 | 2308 | if (!force_nonblock || ret2 != -EAGAIN) { |
bcaec089 | 2309 | kiocb_done(kiocb, ret2, nxt); |
f67676d1 JA |
2310 | } else { |
2311 | copy_iov: | |
b7bb4f7d | 2312 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2313 | inline_vecs, &iter); |
2314 | if (ret) | |
2315 | goto out_free; | |
29de5f6a JA |
2316 | /* any defer here is final, must blocking retry */ |
2317 | if (!(req->flags & REQ_F_NOWAIT)) | |
2318 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2319 | return -EAGAIN; |
2320 | } | |
2b188cc1 | 2321 | } |
f67676d1 | 2322 | out_free: |
1e95081c | 2323 | kfree(iovec); |
99bc4c38 | 2324 | req->flags &= ~REQ_F_NEED_CLEANUP; |
2b188cc1 JA |
2325 | return ret; |
2326 | } | |
2327 | ||
3529d8c2 JA |
2328 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2329 | bool force_nonblock) | |
f67676d1 | 2330 | { |
3529d8c2 JA |
2331 | struct io_async_ctx *io; |
2332 | struct iov_iter iter; | |
f67676d1 JA |
2333 | ssize_t ret; |
2334 | ||
3529d8c2 JA |
2335 | ret = io_prep_rw(req, sqe, force_nonblock); |
2336 | if (ret) | |
2337 | return ret; | |
f67676d1 | 2338 | |
3529d8c2 JA |
2339 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
2340 | return -EBADF; | |
f67676d1 | 2341 | |
5f798bea PB |
2342 | /* either don't need iovec imported or already have it */ |
2343 | if (!req->io || req->flags & REQ_F_NEED_CLEANUP) | |
3529d8c2 JA |
2344 | return 0; |
2345 | ||
2346 | io = req->io; | |
2347 | io->rw.iov = io->rw.fast_iov; | |
2348 | req->io = NULL; | |
2349 | ret = io_import_iovec(WRITE, req, &io->rw.iov, &iter); | |
2350 | req->io = io; | |
2351 | if (ret < 0) | |
2352 | return ret; | |
2353 | ||
2354 | io_req_map_rw(req, ret, io->rw.iov, io->rw.fast_iov, &iter); | |
2355 | return 0; | |
f67676d1 JA |
2356 | } |
2357 | ||
267bc904 | 2358 | static int io_write(struct io_kiocb *req, struct io_kiocb **nxt, |
8358e3a8 | 2359 | bool force_nonblock) |
2b188cc1 JA |
2360 | { |
2361 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 2362 | struct kiocb *kiocb = &req->rw.kiocb; |
2b188cc1 | 2363 | struct iov_iter iter; |
31b51510 | 2364 | size_t iov_count; |
f67676d1 | 2365 | ssize_t ret, io_size; |
2b188cc1 | 2366 | |
3529d8c2 | 2367 | ret = io_import_iovec(WRITE, req, &iovec, &iter); |
06b76d44 JA |
2368 | if (ret < 0) |
2369 | return ret; | |
2b188cc1 | 2370 | |
fd6c2e4c JA |
2371 | /* Ensure we clear previously set non-block flag */ |
2372 | if (!force_nonblock) | |
9adbd45d | 2373 | req->rw.kiocb.ki_flags &= ~IOCB_NOWAIT; |
fd6c2e4c | 2374 | |
797f3f53 | 2375 | req->result = 0; |
f67676d1 | 2376 | io_size = ret; |
9e645e11 | 2377 | if (req->flags & REQ_F_LINK) |
f67676d1 | 2378 | req->result = io_size; |
9e645e11 | 2379 | |
f67676d1 JA |
2380 | /* |
2381 | * If the file doesn't support async, mark it as REQ_F_MUST_PUNT so | |
2382 | * we know to async punt it even if it was opened O_NONBLOCK | |
2383 | */ | |
29de5f6a | 2384 | if (force_nonblock && !io_file_supports_async(req->file)) |
f67676d1 | 2385 | goto copy_iov; |
31b51510 | 2386 | |
10d59345 JA |
2387 | /* file path doesn't support NOWAIT for non-direct_IO */ |
2388 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
2389 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 2390 | goto copy_iov; |
31b51510 | 2391 | |
f67676d1 | 2392 | iov_count = iov_iter_count(&iter); |
9adbd45d | 2393 | ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count); |
2b188cc1 | 2394 | if (!ret) { |
9bf7933f RP |
2395 | ssize_t ret2; |
2396 | ||
2b188cc1 JA |
2397 | /* |
2398 | * Open-code file_start_write here to grab freeze protection, | |
2399 | * which will be released by another thread in | |
2400 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
2401 | * released so that it doesn't complain about the held lock when | |
2402 | * we return to userspace. | |
2403 | */ | |
491381ce | 2404 | if (req->flags & REQ_F_ISREG) { |
9adbd45d | 2405 | __sb_start_write(file_inode(req->file)->i_sb, |
2b188cc1 | 2406 | SB_FREEZE_WRITE, true); |
9adbd45d | 2407 | __sb_writers_release(file_inode(req->file)->i_sb, |
2b188cc1 JA |
2408 | SB_FREEZE_WRITE); |
2409 | } | |
2410 | kiocb->ki_flags |= IOCB_WRITE; | |
9bf7933f | 2411 | |
9adbd45d JA |
2412 | if (req->file->f_op->write_iter) |
2413 | ret2 = call_write_iter(req->file, kiocb, &iter); | |
32960613 | 2414 | else |
9adbd45d | 2415 | ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter); |
faac996c JA |
2416 | /* |
2417 | * Raw bdev writes will -EOPNOTSUPP for IOCB_NOWAIT. Just | |
2418 | * retry them without IOCB_NOWAIT. | |
2419 | */ | |
2420 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
2421 | ret2 = -EAGAIN; | |
f67676d1 | 2422 | if (!force_nonblock || ret2 != -EAGAIN) { |
bcaec089 | 2423 | kiocb_done(kiocb, ret2, nxt); |
f67676d1 JA |
2424 | } else { |
2425 | copy_iov: | |
b7bb4f7d | 2426 | ret = io_setup_async_rw(req, io_size, iovec, |
f67676d1 JA |
2427 | inline_vecs, &iter); |
2428 | if (ret) | |
2429 | goto out_free; | |
29de5f6a JA |
2430 | /* any defer here is final, must blocking retry */ |
2431 | req->flags |= REQ_F_MUST_PUNT; | |
f67676d1 JA |
2432 | return -EAGAIN; |
2433 | } | |
2b188cc1 | 2434 | } |
31b51510 | 2435 | out_free: |
99bc4c38 | 2436 | req->flags &= ~REQ_F_NEED_CLEANUP; |
1e95081c | 2437 | kfree(iovec); |
2b188cc1 JA |
2438 | return ret; |
2439 | } | |
2440 | ||
7d67af2c PB |
2441 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2442 | { | |
2443 | struct io_splice* sp = &req->splice; | |
2444 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
2445 | int ret; | |
2446 | ||
2447 | if (req->flags & REQ_F_NEED_CLEANUP) | |
2448 | return 0; | |
2449 | ||
2450 | sp->file_in = NULL; | |
2451 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
2452 | sp->off_out = READ_ONCE(sqe->off); | |
2453 | sp->len = READ_ONCE(sqe->len); | |
2454 | sp->flags = READ_ONCE(sqe->splice_flags); | |
2455 | ||
2456 | if (unlikely(sp->flags & ~valid_flags)) | |
2457 | return -EINVAL; | |
2458 | ||
2459 | ret = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), &sp->file_in, | |
2460 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2461 | if (ret) | |
2462 | return ret; | |
2463 | req->flags |= REQ_F_NEED_CLEANUP; | |
2464 | ||
2465 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) | |
2466 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
2467 | ||
2468 | return 0; | |
2469 | } | |
2470 | ||
2471 | static bool io_splice_punt(struct file *file) | |
2472 | { | |
2473 | if (get_pipe_info(file)) | |
2474 | return false; | |
2475 | if (!io_file_supports_async(file)) | |
2476 | return true; | |
2477 | return !(file->f_mode & O_NONBLOCK); | |
2478 | } | |
2479 | ||
2480 | static int io_splice(struct io_kiocb *req, struct io_kiocb **nxt, | |
2481 | bool force_nonblock) | |
2482 | { | |
2483 | struct io_splice *sp = &req->splice; | |
2484 | struct file *in = sp->file_in; | |
2485 | struct file *out = sp->file_out; | |
2486 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
2487 | loff_t *poff_in, *poff_out; | |
2488 | long ret; | |
2489 | ||
2490 | if (force_nonblock) { | |
2491 | if (io_splice_punt(in) || io_splice_punt(out)) | |
2492 | return -EAGAIN; | |
2493 | flags |= SPLICE_F_NONBLOCK; | |
2494 | } | |
2495 | ||
2496 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
2497 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
2498 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); | |
2499 | if (force_nonblock && ret == -EAGAIN) | |
2500 | return -EAGAIN; | |
2501 | ||
2502 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
2503 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
2504 | ||
2505 | io_cqring_add_event(req, ret); | |
2506 | if (ret != sp->len) | |
2507 | req_set_fail_links(req); | |
2508 | io_put_req_find_next(req, nxt); | |
2509 | return 0; | |
2510 | } | |
2511 | ||
2b188cc1 JA |
2512 | /* |
2513 | * IORING_OP_NOP just posts a completion event, nothing else. | |
2514 | */ | |
78e19bbe | 2515 | static int io_nop(struct io_kiocb *req) |
2b188cc1 JA |
2516 | { |
2517 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 2518 | |
def596e9 JA |
2519 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
2520 | return -EINVAL; | |
2521 | ||
78e19bbe | 2522 | io_cqring_add_event(req, 0); |
e65ef56d | 2523 | io_put_req(req); |
2b188cc1 JA |
2524 | return 0; |
2525 | } | |
2526 | ||
3529d8c2 | 2527 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 2528 | { |
6b06314c | 2529 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 2530 | |
09bb8394 JA |
2531 | if (!req->file) |
2532 | return -EBADF; | |
c992fe29 | 2533 | |
6b06314c | 2534 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 2535 | return -EINVAL; |
edafccee | 2536 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
2537 | return -EINVAL; |
2538 | ||
8ed8d3c3 JA |
2539 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
2540 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
2541 | return -EINVAL; | |
2542 | ||
2543 | req->sync.off = READ_ONCE(sqe->off); | |
2544 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
2545 | return 0; |
2546 | } | |
2547 | ||
8ed8d3c3 JA |
2548 | static bool io_req_cancelled(struct io_kiocb *req) |
2549 | { | |
2550 | if (req->work.flags & IO_WQ_WORK_CANCEL) { | |
2551 | req_set_fail_links(req); | |
2552 | io_cqring_add_event(req, -ECANCELED); | |
2553 | io_put_req(req); | |
2554 | return true; | |
2555 | } | |
2556 | ||
2557 | return false; | |
2558 | } | |
2559 | ||
78912934 JA |
2560 | static void io_link_work_cb(struct io_wq_work **workptr) |
2561 | { | |
2562 | struct io_wq_work *work = *workptr; | |
2563 | struct io_kiocb *link = work->data; | |
2564 | ||
2565 | io_queue_linked_timeout(link); | |
5eae8619 | 2566 | io_wq_submit_work(workptr); |
78912934 JA |
2567 | } |
2568 | ||
2569 | static void io_wq_assign_next(struct io_wq_work **workptr, struct io_kiocb *nxt) | |
2570 | { | |
2571 | struct io_kiocb *link; | |
2572 | ||
78912934 | 2573 | *workptr = &nxt->work; |
3b17cf5a | 2574 | link = io_prep_linked_timeout(nxt); |
78912934 | 2575 | if (link) { |
78912934 JA |
2576 | nxt->work.func = io_link_work_cb; |
2577 | nxt->work.data = link; | |
2578 | } | |
2579 | } | |
2580 | ||
5ea62161 | 2581 | static void __io_fsync(struct io_kiocb *req, struct io_kiocb **nxt) |
8ed8d3c3 | 2582 | { |
8ed8d3c3 | 2583 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
2584 | int ret; |
2585 | ||
9adbd45d | 2586 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
2587 | end > 0 ? end : LLONG_MAX, |
2588 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
2589 | if (ret < 0) | |
2590 | req_set_fail_links(req); | |
2591 | io_cqring_add_event(req, ret); | |
5ea62161 PB |
2592 | io_put_req_find_next(req, nxt); |
2593 | } | |
2594 | ||
2595 | static void io_fsync_finish(struct io_wq_work **workptr) | |
2596 | { | |
2597 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
2598 | struct io_kiocb *nxt = NULL; | |
2599 | ||
2600 | if (io_req_cancelled(req)) | |
2601 | return; | |
2602 | __io_fsync(req, &nxt); | |
8ed8d3c3 | 2603 | if (nxt) |
78912934 | 2604 | io_wq_assign_next(workptr, nxt); |
8ed8d3c3 JA |
2605 | } |
2606 | ||
fc4df999 JA |
2607 | static int io_fsync(struct io_kiocb *req, struct io_kiocb **nxt, |
2608 | bool force_nonblock) | |
c992fe29 | 2609 | { |
c992fe29 | 2610 | /* fsync always requires a blocking context */ |
8ed8d3c3 JA |
2611 | if (force_nonblock) { |
2612 | io_put_req(req); | |
2613 | req->work.func = io_fsync_finish; | |
c992fe29 | 2614 | return -EAGAIN; |
8ed8d3c3 | 2615 | } |
5ea62161 | 2616 | __io_fsync(req, nxt); |
c992fe29 CH |
2617 | return 0; |
2618 | } | |
2619 | ||
5ea62161 | 2620 | static void __io_fallocate(struct io_kiocb *req, struct io_kiocb **nxt) |
8ed8d3c3 | 2621 | { |
8ed8d3c3 JA |
2622 | int ret; |
2623 | ||
7fbeb95d PB |
2624 | if (io_req_cancelled(req)) |
2625 | return; | |
2626 | ||
d63d1b5e JA |
2627 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
2628 | req->sync.len); | |
8ed8d3c3 JA |
2629 | if (ret < 0) |
2630 | req_set_fail_links(req); | |
2631 | io_cqring_add_event(req, ret); | |
5ea62161 PB |
2632 | io_put_req_find_next(req, nxt); |
2633 | } | |
2634 | ||
2635 | static void io_fallocate_finish(struct io_wq_work **workptr) | |
2636 | { | |
2637 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
2638 | struct io_kiocb *nxt = NULL; | |
2639 | ||
2640 | __io_fallocate(req, &nxt); | |
8ed8d3c3 | 2641 | if (nxt) |
78912934 | 2642 | io_wq_assign_next(workptr, nxt); |
5d17b4a4 JA |
2643 | } |
2644 | ||
d63d1b5e JA |
2645 | static int io_fallocate_prep(struct io_kiocb *req, |
2646 | const struct io_uring_sqe *sqe) | |
2647 | { | |
2648 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
2649 | return -EINVAL; | |
2650 | ||
2651 | req->sync.off = READ_ONCE(sqe->off); | |
2652 | req->sync.len = READ_ONCE(sqe->addr); | |
2653 | req->sync.mode = READ_ONCE(sqe->len); | |
2654 | return 0; | |
2655 | } | |
2656 | ||
2657 | static int io_fallocate(struct io_kiocb *req, struct io_kiocb **nxt, | |
2658 | bool force_nonblock) | |
5d17b4a4 | 2659 | { |
d63d1b5e | 2660 | /* fallocate always requiring blocking context */ |
8ed8d3c3 JA |
2661 | if (force_nonblock) { |
2662 | io_put_req(req); | |
d63d1b5e | 2663 | req->work.func = io_fallocate_finish; |
5d17b4a4 | 2664 | return -EAGAIN; |
8ed8d3c3 | 2665 | } |
5d17b4a4 | 2666 | |
5ea62161 | 2667 | __io_fallocate(req, nxt); |
5d17b4a4 JA |
2668 | return 0; |
2669 | } | |
2670 | ||
15b71abe | 2671 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 2672 | { |
f8748881 | 2673 | const char __user *fname; |
15b71abe | 2674 | int ret; |
b7bb4f7d | 2675 | |
15b71abe JA |
2676 | if (sqe->ioprio || sqe->buf_index) |
2677 | return -EINVAL; | |
cf3040ca JA |
2678 | if (sqe->flags & IOSQE_FIXED_FILE) |
2679 | return -EBADF; | |
0bdbdd08 PB |
2680 | if (req->flags & REQ_F_NEED_CLEANUP) |
2681 | return 0; | |
03b1230c | 2682 | |
15b71abe | 2683 | req->open.dfd = READ_ONCE(sqe->fd); |
c12cedf2 | 2684 | req->open.how.mode = READ_ONCE(sqe->len); |
f8748881 | 2685 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
c12cedf2 | 2686 | req->open.how.flags = READ_ONCE(sqe->open_flags); |
3529d8c2 | 2687 | |
f8748881 | 2688 | req->open.filename = getname(fname); |
15b71abe JA |
2689 | if (IS_ERR(req->open.filename)) { |
2690 | ret = PTR_ERR(req->open.filename); | |
2691 | req->open.filename = NULL; | |
2692 | return ret; | |
2693 | } | |
3529d8c2 | 2694 | |
8fef80bf | 2695 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 2696 | return 0; |
03b1230c JA |
2697 | } |
2698 | ||
cebdb986 | 2699 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 2700 | { |
cebdb986 JA |
2701 | struct open_how __user *how; |
2702 | const char __user *fname; | |
2703 | size_t len; | |
0fa03c62 JA |
2704 | int ret; |
2705 | ||
cebdb986 | 2706 | if (sqe->ioprio || sqe->buf_index) |
0fa03c62 | 2707 | return -EINVAL; |
cf3040ca JA |
2708 | if (sqe->flags & IOSQE_FIXED_FILE) |
2709 | return -EBADF; | |
0bdbdd08 PB |
2710 | if (req->flags & REQ_F_NEED_CLEANUP) |
2711 | return 0; | |
0fa03c62 | 2712 | |
cebdb986 JA |
2713 | req->open.dfd = READ_ONCE(sqe->fd); |
2714 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2715 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
2716 | len = READ_ONCE(sqe->len); | |
0fa03c62 | 2717 | |
cebdb986 JA |
2718 | if (len < OPEN_HOW_SIZE_VER0) |
2719 | return -EINVAL; | |
3529d8c2 | 2720 | |
cebdb986 JA |
2721 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
2722 | len); | |
2723 | if (ret) | |
2724 | return ret; | |
3529d8c2 | 2725 | |
cebdb986 JA |
2726 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) |
2727 | req->open.how.flags |= O_LARGEFILE; | |
0fa03c62 | 2728 | |
cebdb986 JA |
2729 | req->open.filename = getname(fname); |
2730 | if (IS_ERR(req->open.filename)) { | |
2731 | ret = PTR_ERR(req->open.filename); | |
2732 | req->open.filename = NULL; | |
2733 | return ret; | |
2734 | } | |
2735 | ||
8fef80bf | 2736 | req->flags |= REQ_F_NEED_CLEANUP; |
cebdb986 JA |
2737 | return 0; |
2738 | } | |
2739 | ||
2740 | static int io_openat2(struct io_kiocb *req, struct io_kiocb **nxt, | |
2741 | bool force_nonblock) | |
15b71abe JA |
2742 | { |
2743 | struct open_flags op; | |
15b71abe JA |
2744 | struct file *file; |
2745 | int ret; | |
2746 | ||
f86cd20c | 2747 | if (force_nonblock) |
15b71abe | 2748 | return -EAGAIN; |
15b71abe | 2749 | |
cebdb986 | 2750 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
2751 | if (ret) |
2752 | goto err; | |
2753 | ||
cebdb986 | 2754 | ret = get_unused_fd_flags(req->open.how.flags); |
15b71abe JA |
2755 | if (ret < 0) |
2756 | goto err; | |
2757 | ||
2758 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
2759 | if (IS_ERR(file)) { | |
2760 | put_unused_fd(ret); | |
2761 | ret = PTR_ERR(file); | |
2762 | } else { | |
2763 | fsnotify_open(file); | |
2764 | fd_install(ret, file); | |
2765 | } | |
2766 | err: | |
2767 | putname(req->open.filename); | |
8fef80bf | 2768 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
2769 | if (ret < 0) |
2770 | req_set_fail_links(req); | |
2771 | io_cqring_add_event(req, ret); | |
2772 | io_put_req_find_next(req, nxt); | |
2773 | return 0; | |
2774 | } | |
2775 | ||
cebdb986 JA |
2776 | static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt, |
2777 | bool force_nonblock) | |
2778 | { | |
2779 | req->open.how = build_open_how(req->open.how.flags, req->open.how.mode); | |
2780 | return io_openat2(req, nxt, force_nonblock); | |
2781 | } | |
2782 | ||
3e4827b0 JA |
2783 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
2784 | const struct io_uring_sqe *sqe) | |
2785 | { | |
2786 | #if defined(CONFIG_EPOLL) | |
2787 | if (sqe->ioprio || sqe->buf_index) | |
2788 | return -EINVAL; | |
2789 | ||
2790 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
2791 | req->epoll.op = READ_ONCE(sqe->len); | |
2792 | req->epoll.fd = READ_ONCE(sqe->off); | |
2793 | ||
2794 | if (ep_op_has_event(req->epoll.op)) { | |
2795 | struct epoll_event __user *ev; | |
2796 | ||
2797 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
2798 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
2799 | return -EFAULT; | |
2800 | } | |
2801 | ||
2802 | return 0; | |
2803 | #else | |
2804 | return -EOPNOTSUPP; | |
2805 | #endif | |
2806 | } | |
2807 | ||
2808 | static int io_epoll_ctl(struct io_kiocb *req, struct io_kiocb **nxt, | |
2809 | bool force_nonblock) | |
2810 | { | |
2811 | #if defined(CONFIG_EPOLL) | |
2812 | struct io_epoll *ie = &req->epoll; | |
2813 | int ret; | |
2814 | ||
2815 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
2816 | if (force_nonblock && ret == -EAGAIN) | |
2817 | return -EAGAIN; | |
2818 | ||
2819 | if (ret < 0) | |
2820 | req_set_fail_links(req); | |
2821 | io_cqring_add_event(req, ret); | |
2822 | io_put_req_find_next(req, nxt); | |
2823 | return 0; | |
2824 | #else | |
2825 | return -EOPNOTSUPP; | |
2826 | #endif | |
2827 | } | |
2828 | ||
c1ca757b JA |
2829 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2830 | { | |
2831 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
2832 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
2833 | return -EINVAL; | |
2834 | ||
2835 | req->madvise.addr = READ_ONCE(sqe->addr); | |
2836 | req->madvise.len = READ_ONCE(sqe->len); | |
2837 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
2838 | return 0; | |
2839 | #else | |
2840 | return -EOPNOTSUPP; | |
2841 | #endif | |
2842 | } | |
2843 | ||
2844 | static int io_madvise(struct io_kiocb *req, struct io_kiocb **nxt, | |
2845 | bool force_nonblock) | |
2846 | { | |
2847 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
2848 | struct io_madvise *ma = &req->madvise; | |
2849 | int ret; | |
2850 | ||
2851 | if (force_nonblock) | |
2852 | return -EAGAIN; | |
2853 | ||
2854 | ret = do_madvise(ma->addr, ma->len, ma->advice); | |
2855 | if (ret < 0) | |
2856 | req_set_fail_links(req); | |
2857 | io_cqring_add_event(req, ret); | |
2858 | io_put_req_find_next(req, nxt); | |
2859 | return 0; | |
2860 | #else | |
2861 | return -EOPNOTSUPP; | |
2862 | #endif | |
2863 | } | |
2864 | ||
4840e418 JA |
2865 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2866 | { | |
2867 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
2868 | return -EINVAL; | |
2869 | ||
2870 | req->fadvise.offset = READ_ONCE(sqe->off); | |
2871 | req->fadvise.len = READ_ONCE(sqe->len); | |
2872 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
2873 | return 0; | |
2874 | } | |
2875 | ||
2876 | static int io_fadvise(struct io_kiocb *req, struct io_kiocb **nxt, | |
2877 | bool force_nonblock) | |
2878 | { | |
2879 | struct io_fadvise *fa = &req->fadvise; | |
2880 | int ret; | |
2881 | ||
3e69426d JA |
2882 | if (force_nonblock) { |
2883 | switch (fa->advice) { | |
2884 | case POSIX_FADV_NORMAL: | |
2885 | case POSIX_FADV_RANDOM: | |
2886 | case POSIX_FADV_SEQUENTIAL: | |
2887 | break; | |
2888 | default: | |
2889 | return -EAGAIN; | |
2890 | } | |
2891 | } | |
4840e418 JA |
2892 | |
2893 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
2894 | if (ret < 0) | |
2895 | req_set_fail_links(req); | |
2896 | io_cqring_add_event(req, ret); | |
2897 | io_put_req_find_next(req, nxt); | |
2898 | return 0; | |
2899 | } | |
2900 | ||
eddc7ef5 JA |
2901 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2902 | { | |
f8748881 | 2903 | const char __user *fname; |
eddc7ef5 JA |
2904 | unsigned lookup_flags; |
2905 | int ret; | |
2906 | ||
2907 | if (sqe->ioprio || sqe->buf_index) | |
2908 | return -EINVAL; | |
cf3040ca JA |
2909 | if (sqe->flags & IOSQE_FIXED_FILE) |
2910 | return -EBADF; | |
0bdbdd08 PB |
2911 | if (req->flags & REQ_F_NEED_CLEANUP) |
2912 | return 0; | |
eddc7ef5 JA |
2913 | |
2914 | req->open.dfd = READ_ONCE(sqe->fd); | |
2915 | req->open.mask = READ_ONCE(sqe->len); | |
f8748881 | 2916 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
eddc7ef5 | 2917 | req->open.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
c12cedf2 | 2918 | req->open.how.flags = READ_ONCE(sqe->statx_flags); |
eddc7ef5 | 2919 | |
c12cedf2 | 2920 | if (vfs_stat_set_lookup_flags(&lookup_flags, req->open.how.flags)) |
eddc7ef5 JA |
2921 | return -EINVAL; |
2922 | ||
f8748881 | 2923 | req->open.filename = getname_flags(fname, lookup_flags, NULL); |
eddc7ef5 JA |
2924 | if (IS_ERR(req->open.filename)) { |
2925 | ret = PTR_ERR(req->open.filename); | |
2926 | req->open.filename = NULL; | |
2927 | return ret; | |
2928 | } | |
2929 | ||
8fef80bf | 2930 | req->flags |= REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
2931 | return 0; |
2932 | } | |
2933 | ||
2934 | static int io_statx(struct io_kiocb *req, struct io_kiocb **nxt, | |
2935 | bool force_nonblock) | |
2936 | { | |
2937 | struct io_open *ctx = &req->open; | |
2938 | unsigned lookup_flags; | |
2939 | struct path path; | |
2940 | struct kstat stat; | |
2941 | int ret; | |
2942 | ||
2943 | if (force_nonblock) | |
2944 | return -EAGAIN; | |
2945 | ||
c12cedf2 | 2946 | if (vfs_stat_set_lookup_flags(&lookup_flags, ctx->how.flags)) |
eddc7ef5 JA |
2947 | return -EINVAL; |
2948 | ||
2949 | retry: | |
2950 | /* filename_lookup() drops it, keep a reference */ | |
2951 | ctx->filename->refcnt++; | |
2952 | ||
2953 | ret = filename_lookup(ctx->dfd, ctx->filename, lookup_flags, &path, | |
2954 | NULL); | |
2955 | if (ret) | |
2956 | goto err; | |
2957 | ||
c12cedf2 | 2958 | ret = vfs_getattr(&path, &stat, ctx->mask, ctx->how.flags); |
eddc7ef5 JA |
2959 | path_put(&path); |
2960 | if (retry_estale(ret, lookup_flags)) { | |
2961 | lookup_flags |= LOOKUP_REVAL; | |
2962 | goto retry; | |
2963 | } | |
2964 | if (!ret) | |
2965 | ret = cp_statx(&stat, ctx->buffer); | |
2966 | err: | |
2967 | putname(ctx->filename); | |
8fef80bf | 2968 | req->flags &= ~REQ_F_NEED_CLEANUP; |
eddc7ef5 JA |
2969 | if (ret < 0) |
2970 | req_set_fail_links(req); | |
2971 | io_cqring_add_event(req, ret); | |
2972 | io_put_req_find_next(req, nxt); | |
2973 | return 0; | |
2974 | } | |
2975 | ||
b5dba59e JA |
2976 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2977 | { | |
2978 | /* | |
2979 | * If we queue this for async, it must not be cancellable. That would | |
2980 | * leave the 'file' in an undeterminate state. | |
2981 | */ | |
2982 | req->work.flags |= IO_WQ_WORK_NO_CANCEL; | |
2983 | ||
2984 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || | |
2985 | sqe->rw_flags || sqe->buf_index) | |
2986 | return -EINVAL; | |
2987 | if (sqe->flags & IOSQE_FIXED_FILE) | |
cf3040ca | 2988 | return -EBADF; |
b5dba59e JA |
2989 | |
2990 | req->close.fd = READ_ONCE(sqe->fd); | |
2991 | if (req->file->f_op == &io_uring_fops || | |
b14cca0c | 2992 | req->close.fd == req->ctx->ring_fd) |
b5dba59e JA |
2993 | return -EBADF; |
2994 | ||
2995 | return 0; | |
2996 | } | |
2997 | ||
a93b3331 PB |
2998 | /* only called when __close_fd_get_file() is done */ |
2999 | static void __io_close_finish(struct io_kiocb *req, struct io_kiocb **nxt) | |
3000 | { | |
3001 | int ret; | |
3002 | ||
3003 | ret = filp_close(req->close.put_file, req->work.files); | |
3004 | if (ret < 0) | |
3005 | req_set_fail_links(req); | |
3006 | io_cqring_add_event(req, ret); | |
3007 | fput(req->close.put_file); | |
3008 | io_put_req_find_next(req, nxt); | |
3009 | } | |
3010 | ||
b5dba59e JA |
3011 | static void io_close_finish(struct io_wq_work **workptr) |
3012 | { | |
3013 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3014 | struct io_kiocb *nxt = NULL; | |
3015 | ||
7fbeb95d | 3016 | /* not cancellable, don't do io_req_cancelled() */ |
a93b3331 | 3017 | __io_close_finish(req, &nxt); |
b5dba59e JA |
3018 | if (nxt) |
3019 | io_wq_assign_next(workptr, nxt); | |
3020 | } | |
3021 | ||
3022 | static int io_close(struct io_kiocb *req, struct io_kiocb **nxt, | |
3023 | bool force_nonblock) | |
3024 | { | |
3025 | int ret; | |
3026 | ||
3027 | req->close.put_file = NULL; | |
3028 | ret = __close_fd_get_file(req->close.fd, &req->close.put_file); | |
3029 | if (ret < 0) | |
3030 | return ret; | |
3031 | ||
3032 | /* if the file has a flush method, be safe and punt to async */ | |
a2100672 PB |
3033 | if (req->close.put_file->f_op->flush && force_nonblock) { |
3034 | req->work.func = io_close_finish; | |
3035 | /* | |
3036 | * Do manual async queue here to avoid grabbing files - we don't | |
3037 | * need the files, and it'll cause io_close_finish() to close | |
3038 | * the file again and cause a double CQE entry for this request | |
3039 | */ | |
3040 | io_queue_async_work(req); | |
3041 | return 0; | |
3042 | } | |
b5dba59e JA |
3043 | |
3044 | /* | |
3045 | * No ->flush(), safely close from here and just punt the | |
3046 | * fput() to async context. | |
3047 | */ | |
a93b3331 PB |
3048 | __io_close_finish(req, nxt); |
3049 | return 0; | |
b5dba59e JA |
3050 | } |
3051 | ||
3529d8c2 | 3052 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
3053 | { |
3054 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
3055 | |
3056 | if (!req->file) | |
3057 | return -EBADF; | |
5d17b4a4 JA |
3058 | |
3059 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
3060 | return -EINVAL; | |
3061 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
3062 | return -EINVAL; | |
3063 | ||
8ed8d3c3 JA |
3064 | req->sync.off = READ_ONCE(sqe->off); |
3065 | req->sync.len = READ_ONCE(sqe->len); | |
3066 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
3067 | return 0; |
3068 | } | |
3069 | ||
5ea62161 | 3070 | static void __io_sync_file_range(struct io_kiocb *req, struct io_kiocb **nxt) |
8ed8d3c3 | 3071 | { |
8ed8d3c3 JA |
3072 | int ret; |
3073 | ||
9adbd45d | 3074 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
3075 | req->sync.flags); |
3076 | if (ret < 0) | |
3077 | req_set_fail_links(req); | |
3078 | io_cqring_add_event(req, ret); | |
5ea62161 PB |
3079 | io_put_req_find_next(req, nxt); |
3080 | } | |
3081 | ||
3082 | ||
3083 | static void io_sync_file_range_finish(struct io_wq_work **workptr) | |
3084 | { | |
3085 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3086 | struct io_kiocb *nxt = NULL; | |
3087 | ||
3088 | if (io_req_cancelled(req)) | |
3089 | return; | |
3090 | __io_sync_file_range(req, &nxt); | |
8ed8d3c3 | 3091 | if (nxt) |
78912934 | 3092 | io_wq_assign_next(workptr, nxt); |
5d17b4a4 JA |
3093 | } |
3094 | ||
fc4df999 | 3095 | static int io_sync_file_range(struct io_kiocb *req, struct io_kiocb **nxt, |
5d17b4a4 JA |
3096 | bool force_nonblock) |
3097 | { | |
5d17b4a4 | 3098 | /* sync_file_range always requires a blocking context */ |
8ed8d3c3 JA |
3099 | if (force_nonblock) { |
3100 | io_put_req(req); | |
3101 | req->work.func = io_sync_file_range_finish; | |
5d17b4a4 | 3102 | return -EAGAIN; |
8ed8d3c3 | 3103 | } |
5d17b4a4 | 3104 | |
5ea62161 | 3105 | __io_sync_file_range(req, nxt); |
5d17b4a4 JA |
3106 | return 0; |
3107 | } | |
3108 | ||
02d27d89 PB |
3109 | static int io_setup_async_msg(struct io_kiocb *req, |
3110 | struct io_async_msghdr *kmsg) | |
3111 | { | |
3112 | if (req->io) | |
3113 | return -EAGAIN; | |
3114 | if (io_alloc_async_ctx(req)) { | |
3115 | if (kmsg->iov != kmsg->fast_iov) | |
3116 | kfree(kmsg->iov); | |
3117 | return -ENOMEM; | |
3118 | } | |
3119 | req->flags |= REQ_F_NEED_CLEANUP; | |
3120 | memcpy(&req->io->msg, kmsg, sizeof(*kmsg)); | |
3121 | return -EAGAIN; | |
3122 | } | |
3123 | ||
3529d8c2 | 3124 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 3125 | { |
0fa03c62 | 3126 | #if defined(CONFIG_NET) |
e47293fd | 3127 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3128 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3129 | int ret; |
03b1230c | 3130 | |
e47293fd JA |
3131 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
3132 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
fddaface | 3133 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 3134 | |
d8768362 JA |
3135 | #ifdef CONFIG_COMPAT |
3136 | if (req->ctx->compat) | |
3137 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3138 | #endif | |
3139 | ||
fddaface | 3140 | if (!io || req->opcode == IORING_OP_SEND) |
3529d8c2 | 3141 | return 0; |
5f798bea PB |
3142 | /* iovec is already imported */ |
3143 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3144 | return 0; | |
3529d8c2 | 3145 | |
d9688565 | 3146 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3147 | ret = sendmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3148 | &io->msg.iov); |
99bc4c38 PB |
3149 | if (!ret) |
3150 | req->flags |= REQ_F_NEED_CLEANUP; | |
3151 | return ret; | |
03b1230c | 3152 | #else |
e47293fd | 3153 | return -EOPNOTSUPP; |
03b1230c JA |
3154 | #endif |
3155 | } | |
3156 | ||
fc4df999 JA |
3157 | static int io_sendmsg(struct io_kiocb *req, struct io_kiocb **nxt, |
3158 | bool force_nonblock) | |
aa1fa28f | 3159 | { |
03b1230c | 3160 | #if defined(CONFIG_NET) |
0b416c3e | 3161 | struct io_async_msghdr *kmsg = NULL; |
0fa03c62 JA |
3162 | struct socket *sock; |
3163 | int ret; | |
3164 | ||
3165 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3166 | return -EINVAL; | |
3167 | ||
3168 | sock = sock_from_file(req->file, &ret); | |
3169 | if (sock) { | |
b7bb4f7d | 3170 | struct io_async_ctx io; |
0fa03c62 JA |
3171 | unsigned flags; |
3172 | ||
03b1230c | 3173 | if (req->io) { |
0b416c3e | 3174 | kmsg = &req->io->msg; |
b537916c | 3175 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3176 | /* if iov is set, it's allocated already */ |
3177 | if (!kmsg->iov) | |
3178 | kmsg->iov = kmsg->fast_iov; | |
3179 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3180 | } else { |
3529d8c2 JA |
3181 | struct io_sr_msg *sr = &req->sr_msg; |
3182 | ||
0b416c3e | 3183 | kmsg = &io.msg; |
b537916c | 3184 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3185 | |
3186 | io.msg.iov = io.msg.fast_iov; | |
3187 | ret = sendmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3188 | sr->msg_flags, &io.msg.iov); | |
03b1230c | 3189 | if (ret) |
3529d8c2 | 3190 | return ret; |
03b1230c | 3191 | } |
0fa03c62 | 3192 | |
e47293fd JA |
3193 | flags = req->sr_msg.msg_flags; |
3194 | if (flags & MSG_DONTWAIT) | |
3195 | req->flags |= REQ_F_NOWAIT; | |
3196 | else if (force_nonblock) | |
3197 | flags |= MSG_DONTWAIT; | |
3198 | ||
0b416c3e | 3199 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
02d27d89 PB |
3200 | if (force_nonblock && ret == -EAGAIN) |
3201 | return io_setup_async_msg(req, kmsg); | |
441cdbd5 JA |
3202 | if (ret == -ERESTARTSYS) |
3203 | ret = -EINTR; | |
0fa03c62 JA |
3204 | } |
3205 | ||
1e95081c | 3206 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3207 | kfree(kmsg->iov); |
99bc4c38 | 3208 | req->flags &= ~REQ_F_NEED_CLEANUP; |
78e19bbe | 3209 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3210 | if (ret < 0) |
3211 | req_set_fail_links(req); | |
ec9c02ad | 3212 | io_put_req_find_next(req, nxt); |
5d17b4a4 | 3213 | return 0; |
03b1230c JA |
3214 | #else |
3215 | return -EOPNOTSUPP; | |
aa1fa28f | 3216 | #endif |
03b1230c | 3217 | } |
aa1fa28f | 3218 | |
fddaface JA |
3219 | static int io_send(struct io_kiocb *req, struct io_kiocb **nxt, |
3220 | bool force_nonblock) | |
3221 | { | |
3222 | #if defined(CONFIG_NET) | |
3223 | struct socket *sock; | |
3224 | int ret; | |
3225 | ||
3226 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3227 | return -EINVAL; | |
3228 | ||
3229 | sock = sock_from_file(req->file, &ret); | |
3230 | if (sock) { | |
3231 | struct io_sr_msg *sr = &req->sr_msg; | |
3232 | struct msghdr msg; | |
3233 | struct iovec iov; | |
3234 | unsigned flags; | |
3235 | ||
3236 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, | |
3237 | &msg.msg_iter); | |
3238 | if (ret) | |
3239 | return ret; | |
3240 | ||
3241 | msg.msg_name = NULL; | |
3242 | msg.msg_control = NULL; | |
3243 | msg.msg_controllen = 0; | |
3244 | msg.msg_namelen = 0; | |
3245 | ||
3246 | flags = req->sr_msg.msg_flags; | |
3247 | if (flags & MSG_DONTWAIT) | |
3248 | req->flags |= REQ_F_NOWAIT; | |
3249 | else if (force_nonblock) | |
3250 | flags |= MSG_DONTWAIT; | |
3251 | ||
0b7b21e4 JA |
3252 | msg.msg_flags = flags; |
3253 | ret = sock_sendmsg(sock, &msg); | |
fddaface JA |
3254 | if (force_nonblock && ret == -EAGAIN) |
3255 | return -EAGAIN; | |
3256 | if (ret == -ERESTARTSYS) | |
3257 | ret = -EINTR; | |
3258 | } | |
3259 | ||
3260 | io_cqring_add_event(req, ret); | |
3261 | if (ret < 0) | |
3262 | req_set_fail_links(req); | |
3263 | io_put_req_find_next(req, nxt); | |
3264 | return 0; | |
3265 | #else | |
3266 | return -EOPNOTSUPP; | |
3267 | #endif | |
3268 | } | |
3269 | ||
3529d8c2 JA |
3270 | static int io_recvmsg_prep(struct io_kiocb *req, |
3271 | const struct io_uring_sqe *sqe) | |
aa1fa28f JA |
3272 | { |
3273 | #if defined(CONFIG_NET) | |
e47293fd | 3274 | struct io_sr_msg *sr = &req->sr_msg; |
3529d8c2 | 3275 | struct io_async_ctx *io = req->io; |
99bc4c38 | 3276 | int ret; |
3529d8c2 JA |
3277 | |
3278 | sr->msg_flags = READ_ONCE(sqe->msg_flags); | |
3279 | sr->msg = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
0b7b21e4 | 3280 | sr->len = READ_ONCE(sqe->len); |
06b76d44 | 3281 | |
d8768362 JA |
3282 | #ifdef CONFIG_COMPAT |
3283 | if (req->ctx->compat) | |
3284 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
3285 | #endif | |
3286 | ||
fddaface | 3287 | if (!io || req->opcode == IORING_OP_RECV) |
06b76d44 | 3288 | return 0; |
5f798bea PB |
3289 | /* iovec is already imported */ |
3290 | if (req->flags & REQ_F_NEED_CLEANUP) | |
3291 | return 0; | |
03b1230c | 3292 | |
d9688565 | 3293 | io->msg.iov = io->msg.fast_iov; |
99bc4c38 | 3294 | ret = recvmsg_copy_msghdr(&io->msg.msg, sr->msg, sr->msg_flags, |
e47293fd | 3295 | &io->msg.uaddr, &io->msg.iov); |
99bc4c38 PB |
3296 | if (!ret) |
3297 | req->flags |= REQ_F_NEED_CLEANUP; | |
3298 | return ret; | |
aa1fa28f | 3299 | #else |
e47293fd | 3300 | return -EOPNOTSUPP; |
aa1fa28f JA |
3301 | #endif |
3302 | } | |
3303 | ||
fc4df999 JA |
3304 | static int io_recvmsg(struct io_kiocb *req, struct io_kiocb **nxt, |
3305 | bool force_nonblock) | |
aa1fa28f JA |
3306 | { |
3307 | #if defined(CONFIG_NET) | |
0b416c3e | 3308 | struct io_async_msghdr *kmsg = NULL; |
03b1230c JA |
3309 | struct socket *sock; |
3310 | int ret; | |
3311 | ||
3312 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3313 | return -EINVAL; | |
3314 | ||
3315 | sock = sock_from_file(req->file, &ret); | |
3316 | if (sock) { | |
b7bb4f7d | 3317 | struct io_async_ctx io; |
03b1230c JA |
3318 | unsigned flags; |
3319 | ||
03b1230c | 3320 | if (req->io) { |
0b416c3e | 3321 | kmsg = &req->io->msg; |
b537916c | 3322 | kmsg->msg.msg_name = &req->io->msg.addr; |
0b416c3e JA |
3323 | /* if iov is set, it's allocated already */ |
3324 | if (!kmsg->iov) | |
3325 | kmsg->iov = kmsg->fast_iov; | |
3326 | kmsg->msg.msg_iter.iov = kmsg->iov; | |
03b1230c | 3327 | } else { |
3529d8c2 JA |
3328 | struct io_sr_msg *sr = &req->sr_msg; |
3329 | ||
0b416c3e | 3330 | kmsg = &io.msg; |
b537916c | 3331 | kmsg->msg.msg_name = &io.msg.addr; |
3529d8c2 JA |
3332 | |
3333 | io.msg.iov = io.msg.fast_iov; | |
3334 | ret = recvmsg_copy_msghdr(&io.msg.msg, sr->msg, | |
3335 | sr->msg_flags, &io.msg.uaddr, | |
3336 | &io.msg.iov); | |
03b1230c | 3337 | if (ret) |
3529d8c2 | 3338 | return ret; |
03b1230c JA |
3339 | } |
3340 | ||
e47293fd JA |
3341 | flags = req->sr_msg.msg_flags; |
3342 | if (flags & MSG_DONTWAIT) | |
3343 | req->flags |= REQ_F_NOWAIT; | |
3344 | else if (force_nonblock) | |
3345 | flags |= MSG_DONTWAIT; | |
3346 | ||
3347 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.msg, | |
3348 | kmsg->uaddr, flags); | |
02d27d89 PB |
3349 | if (force_nonblock && ret == -EAGAIN) |
3350 | return io_setup_async_msg(req, kmsg); | |
03b1230c JA |
3351 | if (ret == -ERESTARTSYS) |
3352 | ret = -EINTR; | |
3353 | } | |
3354 | ||
1e95081c | 3355 | if (kmsg && kmsg->iov != kmsg->fast_iov) |
0b416c3e | 3356 | kfree(kmsg->iov); |
99bc4c38 | 3357 | req->flags &= ~REQ_F_NEED_CLEANUP; |
03b1230c | 3358 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3359 | if (ret < 0) |
3360 | req_set_fail_links(req); | |
03b1230c JA |
3361 | io_put_req_find_next(req, nxt); |
3362 | return 0; | |
0fa03c62 JA |
3363 | #else |
3364 | return -EOPNOTSUPP; | |
3365 | #endif | |
3366 | } | |
5d17b4a4 | 3367 | |
fddaface JA |
3368 | static int io_recv(struct io_kiocb *req, struct io_kiocb **nxt, |
3369 | bool force_nonblock) | |
3370 | { | |
3371 | #if defined(CONFIG_NET) | |
3372 | struct socket *sock; | |
3373 | int ret; | |
3374 | ||
3375 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3376 | return -EINVAL; | |
3377 | ||
3378 | sock = sock_from_file(req->file, &ret); | |
3379 | if (sock) { | |
3380 | struct io_sr_msg *sr = &req->sr_msg; | |
3381 | struct msghdr msg; | |
3382 | struct iovec iov; | |
3383 | unsigned flags; | |
3384 | ||
3385 | ret = import_single_range(READ, sr->buf, sr->len, &iov, | |
3386 | &msg.msg_iter); | |
3387 | if (ret) | |
3388 | return ret; | |
3389 | ||
3390 | msg.msg_name = NULL; | |
3391 | msg.msg_control = NULL; | |
3392 | msg.msg_controllen = 0; | |
3393 | msg.msg_namelen = 0; | |
3394 | msg.msg_iocb = NULL; | |
3395 | msg.msg_flags = 0; | |
3396 | ||
3397 | flags = req->sr_msg.msg_flags; | |
3398 | if (flags & MSG_DONTWAIT) | |
3399 | req->flags |= REQ_F_NOWAIT; | |
3400 | else if (force_nonblock) | |
3401 | flags |= MSG_DONTWAIT; | |
3402 | ||
0b7b21e4 | 3403 | ret = sock_recvmsg(sock, &msg, flags); |
fddaface JA |
3404 | if (force_nonblock && ret == -EAGAIN) |
3405 | return -EAGAIN; | |
3406 | if (ret == -ERESTARTSYS) | |
3407 | ret = -EINTR; | |
3408 | } | |
3409 | ||
3410 | io_cqring_add_event(req, ret); | |
3411 | if (ret < 0) | |
3412 | req_set_fail_links(req); | |
3413 | io_put_req_find_next(req, nxt); | |
3414 | return 0; | |
3415 | #else | |
3416 | return -EOPNOTSUPP; | |
3417 | #endif | |
3418 | } | |
3419 | ||
3420 | ||
3529d8c2 | 3421 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 JA |
3422 | { |
3423 | #if defined(CONFIG_NET) | |
8ed8d3c3 JA |
3424 | struct io_accept *accept = &req->accept; |
3425 | ||
17f2fe35 JA |
3426 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3427 | return -EINVAL; | |
8042d6ce | 3428 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
3429 | return -EINVAL; |
3430 | ||
d55e5f5b JA |
3431 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3432 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 3433 | accept->flags = READ_ONCE(sqe->accept_flags); |
8ed8d3c3 JA |
3434 | return 0; |
3435 | #else | |
3436 | return -EOPNOTSUPP; | |
3437 | #endif | |
3438 | } | |
17f2fe35 | 3439 | |
8ed8d3c3 JA |
3440 | #if defined(CONFIG_NET) |
3441 | static int __io_accept(struct io_kiocb *req, struct io_kiocb **nxt, | |
3442 | bool force_nonblock) | |
3443 | { | |
3444 | struct io_accept *accept = &req->accept; | |
3445 | unsigned file_flags; | |
3446 | int ret; | |
3447 | ||
3448 | file_flags = force_nonblock ? O_NONBLOCK : 0; | |
3449 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, | |
3450 | accept->addr_len, accept->flags); | |
3451 | if (ret == -EAGAIN && force_nonblock) | |
17f2fe35 | 3452 | return -EAGAIN; |
8e3cca12 JA |
3453 | if (ret == -ERESTARTSYS) |
3454 | ret = -EINTR; | |
4e88d6e7 JA |
3455 | if (ret < 0) |
3456 | req_set_fail_links(req); | |
78e19bbe | 3457 | io_cqring_add_event(req, ret); |
ec9c02ad | 3458 | io_put_req_find_next(req, nxt); |
17f2fe35 | 3459 | return 0; |
8ed8d3c3 JA |
3460 | } |
3461 | ||
3462 | static void io_accept_finish(struct io_wq_work **workptr) | |
3463 | { | |
3464 | struct io_kiocb *req = container_of(*workptr, struct io_kiocb, work); | |
3465 | struct io_kiocb *nxt = NULL; | |
3466 | ||
e441d1cf JA |
3467 | io_put_req(req); |
3468 | ||
8ed8d3c3 JA |
3469 | if (io_req_cancelled(req)) |
3470 | return; | |
3471 | __io_accept(req, &nxt, false); | |
3472 | if (nxt) | |
78912934 | 3473 | io_wq_assign_next(workptr, nxt); |
8ed8d3c3 JA |
3474 | } |
3475 | #endif | |
3476 | ||
3477 | static int io_accept(struct io_kiocb *req, struct io_kiocb **nxt, | |
3478 | bool force_nonblock) | |
3479 | { | |
3480 | #if defined(CONFIG_NET) | |
3481 | int ret; | |
3482 | ||
8ed8d3c3 JA |
3483 | ret = __io_accept(req, nxt, force_nonblock); |
3484 | if (ret == -EAGAIN && force_nonblock) { | |
3485 | req->work.func = io_accept_finish; | |
8ed8d3c3 JA |
3486 | return -EAGAIN; |
3487 | } | |
3488 | return 0; | |
0fa03c62 JA |
3489 | #else |
3490 | return -EOPNOTSUPP; | |
3491 | #endif | |
3492 | } | |
5d17b4a4 | 3493 | |
3529d8c2 | 3494 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 JA |
3495 | { |
3496 | #if defined(CONFIG_NET) | |
3529d8c2 JA |
3497 | struct io_connect *conn = &req->connect; |
3498 | struct io_async_ctx *io = req->io; | |
f499a021 | 3499 | |
3fbb51c1 JA |
3500 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL))) |
3501 | return -EINVAL; | |
3502 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
3503 | return -EINVAL; | |
3504 | ||
3529d8c2 JA |
3505 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
3506 | conn->addr_len = READ_ONCE(sqe->addr2); | |
3507 | ||
3508 | if (!io) | |
3509 | return 0; | |
3510 | ||
3511 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
3fbb51c1 | 3512 | &io->connect.address); |
f499a021 | 3513 | #else |
3fbb51c1 | 3514 | return -EOPNOTSUPP; |
f499a021 JA |
3515 | #endif |
3516 | } | |
3517 | ||
fc4df999 JA |
3518 | static int io_connect(struct io_kiocb *req, struct io_kiocb **nxt, |
3519 | bool force_nonblock) | |
f8e85cf2 JA |
3520 | { |
3521 | #if defined(CONFIG_NET) | |
f499a021 | 3522 | struct io_async_ctx __io, *io; |
f8e85cf2 | 3523 | unsigned file_flags; |
3fbb51c1 | 3524 | int ret; |
f8e85cf2 | 3525 | |
f499a021 JA |
3526 | if (req->io) { |
3527 | io = req->io; | |
3528 | } else { | |
3529d8c2 JA |
3529 | ret = move_addr_to_kernel(req->connect.addr, |
3530 | req->connect.addr_len, | |
3531 | &__io.connect.address); | |
f499a021 JA |
3532 | if (ret) |
3533 | goto out; | |
3534 | io = &__io; | |
3535 | } | |
3536 | ||
3fbb51c1 JA |
3537 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
3538 | ||
3539 | ret = __sys_connect_file(req->file, &io->connect.address, | |
3540 | req->connect.addr_len, file_flags); | |
87f80d62 | 3541 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
b7bb4f7d JA |
3542 | if (req->io) |
3543 | return -EAGAIN; | |
3544 | if (io_alloc_async_ctx(req)) { | |
f499a021 JA |
3545 | ret = -ENOMEM; |
3546 | goto out; | |
3547 | } | |
b7bb4f7d | 3548 | memcpy(&req->io->connect, &__io.connect, sizeof(__io.connect)); |
f8e85cf2 | 3549 | return -EAGAIN; |
f499a021 | 3550 | } |
f8e85cf2 JA |
3551 | if (ret == -ERESTARTSYS) |
3552 | ret = -EINTR; | |
f499a021 | 3553 | out: |
4e88d6e7 JA |
3554 | if (ret < 0) |
3555 | req_set_fail_links(req); | |
f8e85cf2 JA |
3556 | io_cqring_add_event(req, ret); |
3557 | io_put_req_find_next(req, nxt); | |
3558 | return 0; | |
3559 | #else | |
3560 | return -EOPNOTSUPP; | |
3561 | #endif | |
3562 | } | |
3563 | ||
d7718a9d JA |
3564 | struct io_poll_table { |
3565 | struct poll_table_struct pt; | |
3566 | struct io_kiocb *req; | |
3567 | int error; | |
3568 | }; | |
3569 | ||
3570 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
3571 | struct wait_queue_head *head) | |
3572 | { | |
3573 | if (unlikely(poll->head)) { | |
3574 | pt->error = -EINVAL; | |
3575 | return; | |
3576 | } | |
3577 | ||
3578 | pt->error = 0; | |
3579 | poll->head = head; | |
3580 | add_wait_queue(head, &poll->wait); | |
3581 | } | |
3582 | ||
3583 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
3584 | struct poll_table_struct *p) | |
3585 | { | |
3586 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
3587 | ||
3588 | __io_queue_proc(&pt->req->apoll->poll, pt, head); | |
3589 | } | |
3590 | ||
3591 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, | |
3592 | __poll_t mask, task_work_func_t func) | |
3593 | { | |
3594 | struct task_struct *tsk; | |
3595 | ||
3596 | /* for instances that support it check for an event match first: */ | |
3597 | if (mask && !(mask & poll->events)) | |
3598 | return 0; | |
3599 | ||
3600 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
3601 | ||
3602 | list_del_init(&poll->wait.entry); | |
3603 | ||
3604 | tsk = req->task; | |
3605 | req->result = mask; | |
3606 | init_task_work(&req->task_work, func); | |
3607 | /* | |
3608 | * If this fails, then the task is exiting. If that is the case, then | |
3609 | * the exit check will ultimately cancel these work items. Hence we | |
3610 | * don't need to check here and handle it specifically. | |
3611 | */ | |
3612 | task_work_add(tsk, &req->task_work, true); | |
3613 | wake_up_process(tsk); | |
3614 | return 1; | |
3615 | } | |
3616 | ||
3617 | static void io_async_task_func(struct callback_head *cb) | |
3618 | { | |
3619 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
3620 | struct async_poll *apoll = req->apoll; | |
3621 | struct io_ring_ctx *ctx = req->ctx; | |
3622 | ||
3623 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
3624 | ||
3625 | WARN_ON_ONCE(!list_empty(&req->apoll->poll.wait.entry)); | |
3626 | ||
3627 | if (hash_hashed(&req->hash_node)) { | |
3628 | spin_lock_irq(&ctx->completion_lock); | |
3629 | hash_del(&req->hash_node); | |
3630 | spin_unlock_irq(&ctx->completion_lock); | |
3631 | } | |
3632 | ||
3633 | /* restore ->work in case we need to retry again */ | |
3634 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
3635 | ||
3636 | __set_current_state(TASK_RUNNING); | |
3637 | mutex_lock(&ctx->uring_lock); | |
3638 | __io_queue_sqe(req, NULL); | |
3639 | mutex_unlock(&ctx->uring_lock); | |
3640 | ||
3641 | kfree(apoll); | |
3642 | } | |
3643 | ||
3644 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
3645 | void *key) | |
3646 | { | |
3647 | struct io_kiocb *req = wait->private; | |
3648 | struct io_poll_iocb *poll = &req->apoll->poll; | |
3649 | ||
3650 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
3651 | key_to_poll(key)); | |
3652 | ||
3653 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
3654 | } | |
3655 | ||
3656 | static void io_poll_req_insert(struct io_kiocb *req) | |
3657 | { | |
3658 | struct io_ring_ctx *ctx = req->ctx; | |
3659 | struct hlist_head *list; | |
3660 | ||
3661 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
3662 | hlist_add_head(&req->hash_node, list); | |
3663 | } | |
3664 | ||
3665 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
3666 | struct io_poll_iocb *poll, | |
3667 | struct io_poll_table *ipt, __poll_t mask, | |
3668 | wait_queue_func_t wake_func) | |
3669 | __acquires(&ctx->completion_lock) | |
3670 | { | |
3671 | struct io_ring_ctx *ctx = req->ctx; | |
3672 | bool cancel = false; | |
3673 | ||
3674 | poll->file = req->file; | |
3675 | poll->head = NULL; | |
3676 | poll->done = poll->canceled = false; | |
3677 | poll->events = mask; | |
3678 | ||
3679 | ipt->pt._key = mask; | |
3680 | ipt->req = req; | |
3681 | ipt->error = -EINVAL; | |
3682 | ||
3683 | INIT_LIST_HEAD(&poll->wait.entry); | |
3684 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
3685 | poll->wait.private = req; | |
3686 | ||
3687 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; | |
3688 | ||
3689 | spin_lock_irq(&ctx->completion_lock); | |
3690 | if (likely(poll->head)) { | |
3691 | spin_lock(&poll->head->lock); | |
3692 | if (unlikely(list_empty(&poll->wait.entry))) { | |
3693 | if (ipt->error) | |
3694 | cancel = true; | |
3695 | ipt->error = 0; | |
3696 | mask = 0; | |
3697 | } | |
3698 | if (mask || ipt->error) | |
3699 | list_del_init(&poll->wait.entry); | |
3700 | else if (cancel) | |
3701 | WRITE_ONCE(poll->canceled, true); | |
3702 | else if (!poll->done) /* actually waiting for an event */ | |
3703 | io_poll_req_insert(req); | |
3704 | spin_unlock(&poll->head->lock); | |
3705 | } | |
3706 | ||
3707 | return mask; | |
3708 | } | |
3709 | ||
3710 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
3711 | { | |
3712 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
3713 | struct io_ring_ctx *ctx = req->ctx; | |
3714 | struct async_poll *apoll; | |
3715 | struct io_poll_table ipt; | |
3716 | __poll_t mask, ret; | |
3717 | ||
3718 | if (!req->file || !file_can_poll(req->file)) | |
3719 | return false; | |
3720 | if (req->flags & (REQ_F_MUST_PUNT | REQ_F_POLLED)) | |
3721 | return false; | |
3722 | if (!def->pollin && !def->pollout) | |
3723 | return false; | |
3724 | ||
3725 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
3726 | if (unlikely(!apoll)) | |
3727 | return false; | |
3728 | ||
3729 | req->flags |= REQ_F_POLLED; | |
3730 | memcpy(&apoll->work, &req->work, sizeof(req->work)); | |
3731 | ||
3732 | /* | |
3733 | * Don't need a reference here, as we're adding it to the task | |
3734 | * task_works list. If the task exits, the list is pruned. | |
3735 | */ | |
3736 | req->task = current; | |
3737 | req->apoll = apoll; | |
3738 | INIT_HLIST_NODE(&req->hash_node); | |
3739 | ||
8755d97a | 3740 | mask = 0; |
d7718a9d | 3741 | if (def->pollin) |
8755d97a | 3742 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
3743 | if (def->pollout) |
3744 | mask |= POLLOUT | POLLWRNORM; | |
3745 | mask |= POLLERR | POLLPRI; | |
3746 | ||
3747 | ipt.pt._qproc = io_async_queue_proc; | |
3748 | ||
3749 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
3750 | io_async_wake); | |
3751 | if (ret) { | |
3752 | ipt.error = 0; | |
3753 | apoll->poll.done = true; | |
3754 | spin_unlock_irq(&ctx->completion_lock); | |
3755 | memcpy(&req->work, &apoll->work, sizeof(req->work)); | |
3756 | kfree(apoll); | |
3757 | return false; | |
3758 | } | |
3759 | spin_unlock_irq(&ctx->completion_lock); | |
3760 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
3761 | apoll->poll.events); | |
3762 | return true; | |
3763 | } | |
3764 | ||
3765 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
3766 | struct io_poll_iocb *poll) | |
221c5eb2 | 3767 | { |
b41e9852 | 3768 | bool do_complete = false; |
221c5eb2 JA |
3769 | |
3770 | spin_lock(&poll->head->lock); | |
3771 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
3772 | if (!list_empty(&poll->wait.entry)) { |
3773 | list_del_init(&poll->wait.entry); | |
b41e9852 | 3774 | do_complete = true; |
221c5eb2 JA |
3775 | } |
3776 | spin_unlock(&poll->head->lock); | |
d7718a9d JA |
3777 | return do_complete; |
3778 | } | |
3779 | ||
3780 | static bool io_poll_remove_one(struct io_kiocb *req) | |
3781 | { | |
3782 | bool do_complete; | |
3783 | ||
3784 | if (req->opcode == IORING_OP_POLL_ADD) { | |
3785 | do_complete = __io_poll_remove_one(req, &req->poll); | |
3786 | } else { | |
3787 | /* non-poll requests have submit ref still */ | |
3788 | do_complete = __io_poll_remove_one(req, &req->apoll->poll); | |
3789 | if (do_complete) | |
3790 | io_put_req(req); | |
3791 | } | |
3792 | ||
78076bb6 | 3793 | hash_del(&req->hash_node); |
d7718a9d | 3794 | |
b41e9852 JA |
3795 | if (do_complete) { |
3796 | io_cqring_fill_event(req, -ECANCELED); | |
3797 | io_commit_cqring(req->ctx); | |
3798 | req->flags |= REQ_F_COMP_LOCKED; | |
3799 | io_put_req(req); | |
3800 | } | |
3801 | ||
3802 | return do_complete; | |
221c5eb2 JA |
3803 | } |
3804 | ||
3805 | static void io_poll_remove_all(struct io_ring_ctx *ctx) | |
3806 | { | |
78076bb6 | 3807 | struct hlist_node *tmp; |
221c5eb2 | 3808 | struct io_kiocb *req; |
78076bb6 | 3809 | int i; |
221c5eb2 JA |
3810 | |
3811 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
3812 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
3813 | struct hlist_head *list; | |
3814 | ||
3815 | list = &ctx->cancel_hash[i]; | |
3816 | hlist_for_each_entry_safe(req, tmp, list, hash_node) | |
3817 | io_poll_remove_one(req); | |
221c5eb2 JA |
3818 | } |
3819 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 JA |
3820 | |
3821 | io_cqring_ev_posted(ctx); | |
221c5eb2 JA |
3822 | } |
3823 | ||
47f46768 JA |
3824 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
3825 | { | |
78076bb6 | 3826 | struct hlist_head *list; |
47f46768 JA |
3827 | struct io_kiocb *req; |
3828 | ||
78076bb6 JA |
3829 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
3830 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
3831 | if (sqe_addr != req->user_data) |
3832 | continue; | |
3833 | if (io_poll_remove_one(req)) | |
eac406c6 | 3834 | return 0; |
b41e9852 | 3835 | return -EALREADY; |
47f46768 JA |
3836 | } |
3837 | ||
3838 | return -ENOENT; | |
3839 | } | |
3840 | ||
3529d8c2 JA |
3841 | static int io_poll_remove_prep(struct io_kiocb *req, |
3842 | const struct io_uring_sqe *sqe) | |
0969e783 | 3843 | { |
0969e783 JA |
3844 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3845 | return -EINVAL; | |
3846 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
3847 | sqe->poll_events) | |
3848 | return -EINVAL; | |
3849 | ||
3850 | req->poll.addr = READ_ONCE(sqe->addr); | |
0969e783 JA |
3851 | return 0; |
3852 | } | |
3853 | ||
221c5eb2 JA |
3854 | /* |
3855 | * Find a running poll command that matches one specified in sqe->addr, | |
3856 | * and remove it if found. | |
3857 | */ | |
fc4df999 | 3858 | static int io_poll_remove(struct io_kiocb *req) |
221c5eb2 JA |
3859 | { |
3860 | struct io_ring_ctx *ctx = req->ctx; | |
0969e783 | 3861 | u64 addr; |
47f46768 | 3862 | int ret; |
221c5eb2 | 3863 | |
0969e783 | 3864 | addr = req->poll.addr; |
221c5eb2 | 3865 | spin_lock_irq(&ctx->completion_lock); |
0969e783 | 3866 | ret = io_poll_cancel(ctx, addr); |
221c5eb2 JA |
3867 | spin_unlock_irq(&ctx->completion_lock); |
3868 | ||
78e19bbe | 3869 | io_cqring_add_event(req, ret); |
4e88d6e7 JA |
3870 | if (ret < 0) |
3871 | req_set_fail_links(req); | |
e65ef56d | 3872 | io_put_req(req); |
221c5eb2 JA |
3873 | return 0; |
3874 | } | |
3875 | ||
b0dd8a41 | 3876 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
221c5eb2 | 3877 | { |
a197f664 JL |
3878 | struct io_ring_ctx *ctx = req->ctx; |
3879 | ||
8c838788 | 3880 | req->poll.done = true; |
b0a20349 | 3881 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); |
8c838788 | 3882 | io_commit_cqring(ctx); |
221c5eb2 JA |
3883 | } |
3884 | ||
b41e9852 | 3885 | static void io_poll_task_handler(struct io_kiocb *req, struct io_kiocb **nxt) |
221c5eb2 | 3886 | { |
221c5eb2 | 3887 | struct io_ring_ctx *ctx = req->ctx; |
221c5eb2 | 3888 | |
221c5eb2 | 3889 | spin_lock_irq(&ctx->completion_lock); |
78076bb6 | 3890 | hash_del(&req->hash_node); |
b41e9852 JA |
3891 | io_poll_complete(req, req->result, 0); |
3892 | req->flags |= REQ_F_COMP_LOCKED; | |
3893 | io_put_req_find_next(req, nxt); | |
e94f141b JA |
3894 | spin_unlock_irq(&ctx->completion_lock); |
3895 | ||
3896 | io_cqring_ev_posted(ctx); | |
e94f141b JA |
3897 | } |
3898 | ||
b41e9852 | 3899 | static void io_poll_task_func(struct callback_head *cb) |
f0b493e6 | 3900 | { |
b41e9852 JA |
3901 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
3902 | struct io_kiocb *nxt = NULL; | |
f0b493e6 | 3903 | |
b41e9852 | 3904 | io_poll_task_handler(req, &nxt); |
d7718a9d JA |
3905 | if (nxt) { |
3906 | struct io_ring_ctx *ctx = nxt->ctx; | |
3907 | ||
3908 | mutex_lock(&ctx->uring_lock); | |
b41e9852 | 3909 | __io_queue_sqe(nxt, NULL); |
d7718a9d JA |
3910 | mutex_unlock(&ctx->uring_lock); |
3911 | } | |
f0b493e6 JA |
3912 | } |
3913 | ||
221c5eb2 JA |
3914 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
3915 | void *key) | |
3916 | { | |
c2f2eb7d JA |
3917 | struct io_kiocb *req = wait->private; |
3918 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 3919 | |
d7718a9d | 3920 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
3921 | } |
3922 | ||
221c5eb2 JA |
3923 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
3924 | struct poll_table_struct *p) | |
3925 | { | |
3926 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
3927 | ||
d7718a9d | 3928 | __io_queue_proc(&pt->req->poll, pt, head); |
eac406c6 JA |
3929 | } |
3930 | ||
3529d8c2 | 3931 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
3932 | { |
3933 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 3934 | u16 events; |
221c5eb2 JA |
3935 | |
3936 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3937 | return -EINVAL; | |
3938 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
3939 | return -EINVAL; | |
09bb8394 JA |
3940 | if (!poll->file) |
3941 | return -EBADF; | |
221c5eb2 | 3942 | |
221c5eb2 JA |
3943 | events = READ_ONCE(sqe->poll_events); |
3944 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP; | |
b41e9852 | 3945 | |
d7718a9d JA |
3946 | /* |
3947 | * Don't need a reference here, as we're adding it to the task | |
3948 | * task_works list. If the task exits, the list is pruned. | |
3949 | */ | |
b41e9852 | 3950 | req->task = current; |
0969e783 JA |
3951 | return 0; |
3952 | } | |
3953 | ||
3954 | static int io_poll_add(struct io_kiocb *req, struct io_kiocb **nxt) | |
3955 | { | |
3956 | struct io_poll_iocb *poll = &req->poll; | |
3957 | struct io_ring_ctx *ctx = req->ctx; | |
3958 | struct io_poll_table ipt; | |
0969e783 | 3959 | __poll_t mask; |
0969e783 | 3960 | |
78076bb6 | 3961 | INIT_HLIST_NODE(&req->hash_node); |
36703247 | 3962 | INIT_LIST_HEAD(&req->list); |
d7718a9d | 3963 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 3964 | |
d7718a9d JA |
3965 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
3966 | io_poll_wake); | |
221c5eb2 | 3967 | |
8c838788 | 3968 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 3969 | ipt.error = 0; |
b0dd8a41 | 3970 | io_poll_complete(req, mask, 0); |
221c5eb2 | 3971 | } |
221c5eb2 JA |
3972 | spin_unlock_irq(&ctx->completion_lock); |
3973 | ||
8c838788 JA |
3974 | if (mask) { |
3975 | io_cqring_ev_posted(ctx); | |
ec9c02ad | 3976 | io_put_req_find_next(req, nxt); |
221c5eb2 | 3977 | } |
8c838788 | 3978 | return ipt.error; |
221c5eb2 JA |
3979 | } |
3980 | ||
5262f567 JA |
3981 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
3982 | { | |
ad8a48ac JA |
3983 | struct io_timeout_data *data = container_of(timer, |
3984 | struct io_timeout_data, timer); | |
3985 | struct io_kiocb *req = data->req; | |
3986 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
3987 | unsigned long flags; |
3988 | ||
5262f567 JA |
3989 | atomic_inc(&ctx->cq_timeouts); |
3990 | ||
3991 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
ef03681a | 3992 | /* |
11365043 JA |
3993 | * We could be racing with timeout deletion. If the list is empty, |
3994 | * then timeout lookup already found it and will be handling it. | |
ef03681a | 3995 | */ |
842f9612 | 3996 | if (!list_empty(&req->list)) { |
11365043 | 3997 | struct io_kiocb *prev; |
5262f567 | 3998 | |
11365043 JA |
3999 | /* |
4000 | * Adjust the reqs sequence before the current one because it | |
d195a66e | 4001 | * will consume a slot in the cq_ring and the cq_tail |
11365043 JA |
4002 | * pointer will be increased, otherwise other timeout reqs may |
4003 | * return in advance without waiting for enough wait_nr. | |
4004 | */ | |
4005 | prev = req; | |
4006 | list_for_each_entry_continue_reverse(prev, &ctx->timeout_list, list) | |
4007 | prev->sequence++; | |
11365043 | 4008 | list_del_init(&req->list); |
11365043 | 4009 | } |
5262f567 | 4010 | |
78e19bbe | 4011 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
4012 | io_commit_cqring(ctx); |
4013 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4014 | ||
4015 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 4016 | req_set_fail_links(req); |
5262f567 JA |
4017 | io_put_req(req); |
4018 | return HRTIMER_NORESTART; | |
4019 | } | |
4020 | ||
47f46768 JA |
4021 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
4022 | { | |
4023 | struct io_kiocb *req; | |
4024 | int ret = -ENOENT; | |
4025 | ||
4026 | list_for_each_entry(req, &ctx->timeout_list, list) { | |
4027 | if (user_data == req->user_data) { | |
4028 | list_del_init(&req->list); | |
4029 | ret = 0; | |
4030 | break; | |
4031 | } | |
4032 | } | |
4033 | ||
4034 | if (ret == -ENOENT) | |
4035 | return ret; | |
4036 | ||
2d28390a | 4037 | ret = hrtimer_try_to_cancel(&req->io->timeout.timer); |
47f46768 JA |
4038 | if (ret == -1) |
4039 | return -EALREADY; | |
4040 | ||
4e88d6e7 | 4041 | req_set_fail_links(req); |
47f46768 JA |
4042 | io_cqring_fill_event(req, -ECANCELED); |
4043 | io_put_req(req); | |
4044 | return 0; | |
4045 | } | |
4046 | ||
3529d8c2 JA |
4047 | static int io_timeout_remove_prep(struct io_kiocb *req, |
4048 | const struct io_uring_sqe *sqe) | |
b29472ee | 4049 | { |
b29472ee JA |
4050 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4051 | return -EINVAL; | |
4052 | if (sqe->flags || sqe->ioprio || sqe->buf_index || sqe->len) | |
4053 | return -EINVAL; | |
4054 | ||
4055 | req->timeout.addr = READ_ONCE(sqe->addr); | |
4056 | req->timeout.flags = READ_ONCE(sqe->timeout_flags); | |
4057 | if (req->timeout.flags) | |
4058 | return -EINVAL; | |
4059 | ||
b29472ee JA |
4060 | return 0; |
4061 | } | |
4062 | ||
11365043 JA |
4063 | /* |
4064 | * Remove or update an existing timeout command | |
4065 | */ | |
fc4df999 | 4066 | static int io_timeout_remove(struct io_kiocb *req) |
11365043 JA |
4067 | { |
4068 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 4069 | int ret; |
11365043 | 4070 | |
11365043 | 4071 | spin_lock_irq(&ctx->completion_lock); |
b29472ee | 4072 | ret = io_timeout_cancel(ctx, req->timeout.addr); |
11365043 | 4073 | |
47f46768 | 4074 | io_cqring_fill_event(req, ret); |
11365043 JA |
4075 | io_commit_cqring(ctx); |
4076 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 4077 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
4078 | if (ret < 0) |
4079 | req_set_fail_links(req); | |
ec9c02ad | 4080 | io_put_req(req); |
11365043 | 4081 | return 0; |
5262f567 JA |
4082 | } |
4083 | ||
3529d8c2 | 4084 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 4085 | bool is_timeout_link) |
5262f567 | 4086 | { |
ad8a48ac | 4087 | struct io_timeout_data *data; |
a41525ab | 4088 | unsigned flags; |
5262f567 | 4089 | |
ad8a48ac | 4090 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 4091 | return -EINVAL; |
ad8a48ac | 4092 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 4093 | return -EINVAL; |
2d28390a JA |
4094 | if (sqe->off && is_timeout_link) |
4095 | return -EINVAL; | |
a41525ab JA |
4096 | flags = READ_ONCE(sqe->timeout_flags); |
4097 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 4098 | return -EINVAL; |
bdf20073 | 4099 | |
26a61679 JA |
4100 | req->timeout.count = READ_ONCE(sqe->off); |
4101 | ||
3529d8c2 | 4102 | if (!req->io && io_alloc_async_ctx(req)) |
26a61679 JA |
4103 | return -ENOMEM; |
4104 | ||
4105 | data = &req->io->timeout; | |
ad8a48ac | 4106 | data->req = req; |
ad8a48ac JA |
4107 | req->flags |= REQ_F_TIMEOUT; |
4108 | ||
4109 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
4110 | return -EFAULT; |
4111 | ||
11365043 | 4112 | if (flags & IORING_TIMEOUT_ABS) |
ad8a48ac | 4113 | data->mode = HRTIMER_MODE_ABS; |
11365043 | 4114 | else |
ad8a48ac | 4115 | data->mode = HRTIMER_MODE_REL; |
11365043 | 4116 | |
ad8a48ac JA |
4117 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
4118 | return 0; | |
4119 | } | |
4120 | ||
fc4df999 | 4121 | static int io_timeout(struct io_kiocb *req) |
ad8a48ac JA |
4122 | { |
4123 | unsigned count; | |
4124 | struct io_ring_ctx *ctx = req->ctx; | |
4125 | struct io_timeout_data *data; | |
4126 | struct list_head *entry; | |
4127 | unsigned span = 0; | |
ad8a48ac | 4128 | |
2d28390a | 4129 | data = &req->io->timeout; |
93bd25bb | 4130 | |
5262f567 JA |
4131 | /* |
4132 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
4133 | * timeout event to be satisfied. If it isn't set, then this is |
4134 | * a pure timeout request, sequence isn't used. | |
5262f567 | 4135 | */ |
26a61679 | 4136 | count = req->timeout.count; |
93bd25bb JA |
4137 | if (!count) { |
4138 | req->flags |= REQ_F_TIMEOUT_NOSEQ; | |
4139 | spin_lock_irq(&ctx->completion_lock); | |
4140 | entry = ctx->timeout_list.prev; | |
4141 | goto add; | |
4142 | } | |
5262f567 JA |
4143 | |
4144 | req->sequence = ctx->cached_sq_head + count - 1; | |
2d28390a | 4145 | data->seq_offset = count; |
5262f567 JA |
4146 | |
4147 | /* | |
4148 | * Insertion sort, ensuring the first entry in the list is always | |
4149 | * the one we need first. | |
4150 | */ | |
5262f567 JA |
4151 | spin_lock_irq(&ctx->completion_lock); |
4152 | list_for_each_prev(entry, &ctx->timeout_list) { | |
4153 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, list); | |
5da0fb1a | 4154 | unsigned nxt_sq_head; |
4155 | long long tmp, tmp_nxt; | |
2d28390a | 4156 | u32 nxt_offset = nxt->io->timeout.seq_offset; |
5262f567 | 4157 | |
93bd25bb JA |
4158 | if (nxt->flags & REQ_F_TIMEOUT_NOSEQ) |
4159 | continue; | |
4160 | ||
5da0fb1a | 4161 | /* |
4162 | * Since cached_sq_head + count - 1 can overflow, use type long | |
4163 | * long to store it. | |
4164 | */ | |
4165 | tmp = (long long)ctx->cached_sq_head + count - 1; | |
cc42e0ac PB |
4166 | nxt_sq_head = nxt->sequence - nxt_offset + 1; |
4167 | tmp_nxt = (long long)nxt_sq_head + nxt_offset - 1; | |
5da0fb1a | 4168 | |
4169 | /* | |
4170 | * cached_sq_head may overflow, and it will never overflow twice | |
4171 | * once there is some timeout req still be valid. | |
4172 | */ | |
4173 | if (ctx->cached_sq_head < nxt_sq_head) | |
8b07a65a | 4174 | tmp += UINT_MAX; |
5da0fb1a | 4175 | |
a1f58ba4 | 4176 | if (tmp > tmp_nxt) |
5262f567 | 4177 | break; |
a1f58ba4 | 4178 | |
4179 | /* | |
4180 | * Sequence of reqs after the insert one and itself should | |
4181 | * be adjusted because each timeout req consumes a slot. | |
4182 | */ | |
4183 | span++; | |
4184 | nxt->sequence++; | |
5262f567 | 4185 | } |
a1f58ba4 | 4186 | req->sequence -= span; |
93bd25bb | 4187 | add: |
5262f567 | 4188 | list_add(&req->list, entry); |
ad8a48ac JA |
4189 | data->timer.function = io_timeout_fn; |
4190 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 4191 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
4192 | return 0; |
4193 | } | |
5262f567 | 4194 | |
62755e35 JA |
4195 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
4196 | { | |
4197 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
4198 | ||
4199 | return req->user_data == (unsigned long) data; | |
4200 | } | |
4201 | ||
e977d6d3 | 4202 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 4203 | { |
62755e35 | 4204 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
4205 | int ret = 0; |
4206 | ||
62755e35 JA |
4207 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr); |
4208 | switch (cancel_ret) { | |
4209 | case IO_WQ_CANCEL_OK: | |
4210 | ret = 0; | |
4211 | break; | |
4212 | case IO_WQ_CANCEL_RUNNING: | |
4213 | ret = -EALREADY; | |
4214 | break; | |
4215 | case IO_WQ_CANCEL_NOTFOUND: | |
4216 | ret = -ENOENT; | |
4217 | break; | |
4218 | } | |
4219 | ||
e977d6d3 JA |
4220 | return ret; |
4221 | } | |
4222 | ||
47f46768 JA |
4223 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
4224 | struct io_kiocb *req, __u64 sqe_addr, | |
b0dd8a41 | 4225 | struct io_kiocb **nxt, int success_ret) |
47f46768 JA |
4226 | { |
4227 | unsigned long flags; | |
4228 | int ret; | |
4229 | ||
4230 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
4231 | if (ret != -ENOENT) { | |
4232 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4233 | goto done; | |
4234 | } | |
4235 | ||
4236 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4237 | ret = io_timeout_cancel(ctx, sqe_addr); | |
4238 | if (ret != -ENOENT) | |
4239 | goto done; | |
4240 | ret = io_poll_cancel(ctx, sqe_addr); | |
4241 | done: | |
b0dd8a41 JA |
4242 | if (!ret) |
4243 | ret = success_ret; | |
47f46768 JA |
4244 | io_cqring_fill_event(req, ret); |
4245 | io_commit_cqring(ctx); | |
4246 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4247 | io_cqring_ev_posted(ctx); | |
4248 | ||
4e88d6e7 JA |
4249 | if (ret < 0) |
4250 | req_set_fail_links(req); | |
47f46768 JA |
4251 | io_put_req_find_next(req, nxt); |
4252 | } | |
4253 | ||
3529d8c2 JA |
4254 | static int io_async_cancel_prep(struct io_kiocb *req, |
4255 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 4256 | { |
fbf23849 | 4257 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 JA |
4258 | return -EINVAL; |
4259 | if (sqe->flags || sqe->ioprio || sqe->off || sqe->len || | |
4260 | sqe->cancel_flags) | |
4261 | return -EINVAL; | |
4262 | ||
fbf23849 JA |
4263 | req->cancel.addr = READ_ONCE(sqe->addr); |
4264 | return 0; | |
4265 | } | |
4266 | ||
4267 | static int io_async_cancel(struct io_kiocb *req, struct io_kiocb **nxt) | |
4268 | { | |
4269 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 JA |
4270 | |
4271 | io_async_find_and_cancel(ctx, req, req->cancel.addr, nxt, 0); | |
5262f567 JA |
4272 | return 0; |
4273 | } | |
4274 | ||
05f3fb3c JA |
4275 | static int io_files_update_prep(struct io_kiocb *req, |
4276 | const struct io_uring_sqe *sqe) | |
4277 | { | |
4278 | if (sqe->flags || sqe->ioprio || sqe->rw_flags) | |
4279 | return -EINVAL; | |
4280 | ||
4281 | req->files_update.offset = READ_ONCE(sqe->off); | |
4282 | req->files_update.nr_args = READ_ONCE(sqe->len); | |
4283 | if (!req->files_update.nr_args) | |
4284 | return -EINVAL; | |
4285 | req->files_update.arg = READ_ONCE(sqe->addr); | |
4286 | return 0; | |
4287 | } | |
4288 | ||
4289 | static int io_files_update(struct io_kiocb *req, bool force_nonblock) | |
fbf23849 JA |
4290 | { |
4291 | struct io_ring_ctx *ctx = req->ctx; | |
05f3fb3c JA |
4292 | struct io_uring_files_update up; |
4293 | int ret; | |
fbf23849 | 4294 | |
f86cd20c | 4295 | if (force_nonblock) |
05f3fb3c | 4296 | return -EAGAIN; |
05f3fb3c JA |
4297 | |
4298 | up.offset = req->files_update.offset; | |
4299 | up.fds = req->files_update.arg; | |
4300 | ||
4301 | mutex_lock(&ctx->uring_lock); | |
4302 | ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args); | |
4303 | mutex_unlock(&ctx->uring_lock); | |
4304 | ||
4305 | if (ret < 0) | |
4306 | req_set_fail_links(req); | |
4307 | io_cqring_add_event(req, ret); | |
4308 | io_put_req(req); | |
5262f567 JA |
4309 | return 0; |
4310 | } | |
4311 | ||
3529d8c2 JA |
4312 | static int io_req_defer_prep(struct io_kiocb *req, |
4313 | const struct io_uring_sqe *sqe) | |
f67676d1 | 4314 | { |
e781573e | 4315 | ssize_t ret = 0; |
f67676d1 | 4316 | |
f86cd20c JA |
4317 | if (io_op_defs[req->opcode].file_table) { |
4318 | ret = io_grab_files(req); | |
4319 | if (unlikely(ret)) | |
4320 | return ret; | |
4321 | } | |
4322 | ||
cccf0ee8 JA |
4323 | io_req_work_grab_env(req, &io_op_defs[req->opcode]); |
4324 | ||
d625c6ee | 4325 | switch (req->opcode) { |
e781573e JA |
4326 | case IORING_OP_NOP: |
4327 | break; | |
f67676d1 JA |
4328 | case IORING_OP_READV: |
4329 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 4330 | case IORING_OP_READ: |
3529d8c2 | 4331 | ret = io_read_prep(req, sqe, true); |
f67676d1 JA |
4332 | break; |
4333 | case IORING_OP_WRITEV: | |
4334 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 4335 | case IORING_OP_WRITE: |
3529d8c2 | 4336 | ret = io_write_prep(req, sqe, true); |
f67676d1 | 4337 | break; |
0969e783 | 4338 | case IORING_OP_POLL_ADD: |
3529d8c2 | 4339 | ret = io_poll_add_prep(req, sqe); |
0969e783 JA |
4340 | break; |
4341 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 | 4342 | ret = io_poll_remove_prep(req, sqe); |
0969e783 | 4343 | break; |
8ed8d3c3 | 4344 | case IORING_OP_FSYNC: |
3529d8c2 | 4345 | ret = io_prep_fsync(req, sqe); |
8ed8d3c3 JA |
4346 | break; |
4347 | case IORING_OP_SYNC_FILE_RANGE: | |
3529d8c2 | 4348 | ret = io_prep_sfr(req, sqe); |
8ed8d3c3 | 4349 | break; |
03b1230c | 4350 | case IORING_OP_SENDMSG: |
fddaface | 4351 | case IORING_OP_SEND: |
3529d8c2 | 4352 | ret = io_sendmsg_prep(req, sqe); |
03b1230c JA |
4353 | break; |
4354 | case IORING_OP_RECVMSG: | |
fddaface | 4355 | case IORING_OP_RECV: |
3529d8c2 | 4356 | ret = io_recvmsg_prep(req, sqe); |
03b1230c | 4357 | break; |
f499a021 | 4358 | case IORING_OP_CONNECT: |
3529d8c2 | 4359 | ret = io_connect_prep(req, sqe); |
f499a021 | 4360 | break; |
2d28390a | 4361 | case IORING_OP_TIMEOUT: |
3529d8c2 | 4362 | ret = io_timeout_prep(req, sqe, false); |
b7bb4f7d | 4363 | break; |
b29472ee | 4364 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 | 4365 | ret = io_timeout_remove_prep(req, sqe); |
b29472ee | 4366 | break; |
fbf23849 | 4367 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 | 4368 | ret = io_async_cancel_prep(req, sqe); |
fbf23849 | 4369 | break; |
2d28390a | 4370 | case IORING_OP_LINK_TIMEOUT: |
3529d8c2 | 4371 | ret = io_timeout_prep(req, sqe, true); |
b7bb4f7d | 4372 | break; |
8ed8d3c3 | 4373 | case IORING_OP_ACCEPT: |
3529d8c2 | 4374 | ret = io_accept_prep(req, sqe); |
8ed8d3c3 | 4375 | break; |
d63d1b5e JA |
4376 | case IORING_OP_FALLOCATE: |
4377 | ret = io_fallocate_prep(req, sqe); | |
4378 | break; | |
15b71abe JA |
4379 | case IORING_OP_OPENAT: |
4380 | ret = io_openat_prep(req, sqe); | |
4381 | break; | |
b5dba59e JA |
4382 | case IORING_OP_CLOSE: |
4383 | ret = io_close_prep(req, sqe); | |
4384 | break; | |
05f3fb3c JA |
4385 | case IORING_OP_FILES_UPDATE: |
4386 | ret = io_files_update_prep(req, sqe); | |
4387 | break; | |
eddc7ef5 JA |
4388 | case IORING_OP_STATX: |
4389 | ret = io_statx_prep(req, sqe); | |
4390 | break; | |
4840e418 JA |
4391 | case IORING_OP_FADVISE: |
4392 | ret = io_fadvise_prep(req, sqe); | |
4393 | break; | |
c1ca757b JA |
4394 | case IORING_OP_MADVISE: |
4395 | ret = io_madvise_prep(req, sqe); | |
4396 | break; | |
cebdb986 JA |
4397 | case IORING_OP_OPENAT2: |
4398 | ret = io_openat2_prep(req, sqe); | |
4399 | break; | |
3e4827b0 JA |
4400 | case IORING_OP_EPOLL_CTL: |
4401 | ret = io_epoll_ctl_prep(req, sqe); | |
4402 | break; | |
7d67af2c PB |
4403 | case IORING_OP_SPLICE: |
4404 | ret = io_splice_prep(req, sqe); | |
4405 | break; | |
f67676d1 | 4406 | default: |
e781573e JA |
4407 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
4408 | req->opcode); | |
4409 | ret = -EINVAL; | |
b7bb4f7d | 4410 | break; |
f67676d1 JA |
4411 | } |
4412 | ||
b7bb4f7d | 4413 | return ret; |
f67676d1 JA |
4414 | } |
4415 | ||
3529d8c2 | 4416 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 4417 | { |
a197f664 | 4418 | struct io_ring_ctx *ctx = req->ctx; |
f67676d1 | 4419 | int ret; |
de0617e4 | 4420 | |
9d858b21 BL |
4421 | /* Still need defer if there is pending req in defer list. */ |
4422 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) | |
de0617e4 JA |
4423 | return 0; |
4424 | ||
3529d8c2 | 4425 | if (!req->io && io_alloc_async_ctx(req)) |
de0617e4 JA |
4426 | return -EAGAIN; |
4427 | ||
3529d8c2 | 4428 | ret = io_req_defer_prep(req, sqe); |
b7bb4f7d | 4429 | if (ret < 0) |
2d28390a | 4430 | return ret; |
2d28390a | 4431 | |
de0617e4 | 4432 | spin_lock_irq(&ctx->completion_lock); |
9d858b21 | 4433 | if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { |
de0617e4 | 4434 | spin_unlock_irq(&ctx->completion_lock); |
de0617e4 JA |
4435 | return 0; |
4436 | } | |
4437 | ||
915967f6 | 4438 | trace_io_uring_defer(ctx, req, req->user_data); |
de0617e4 JA |
4439 | list_add_tail(&req->list, &ctx->defer_list); |
4440 | spin_unlock_irq(&ctx->completion_lock); | |
4441 | return -EIOCBQUEUED; | |
4442 | } | |
4443 | ||
99bc4c38 PB |
4444 | static void io_cleanup_req(struct io_kiocb *req) |
4445 | { | |
4446 | struct io_async_ctx *io = req->io; | |
4447 | ||
4448 | switch (req->opcode) { | |
4449 | case IORING_OP_READV: | |
4450 | case IORING_OP_READ_FIXED: | |
4451 | case IORING_OP_READ: | |
4452 | case IORING_OP_WRITEV: | |
4453 | case IORING_OP_WRITE_FIXED: | |
4454 | case IORING_OP_WRITE: | |
4455 | if (io->rw.iov != io->rw.fast_iov) | |
4456 | kfree(io->rw.iov); | |
4457 | break; | |
4458 | case IORING_OP_SENDMSG: | |
4459 | case IORING_OP_RECVMSG: | |
4460 | if (io->msg.iov != io->msg.fast_iov) | |
4461 | kfree(io->msg.iov); | |
4462 | break; | |
8fef80bf PB |
4463 | case IORING_OP_OPENAT: |
4464 | case IORING_OP_OPENAT2: | |
4465 | case IORING_OP_STATX: | |
4466 | putname(req->open.filename); | |
4467 | break; | |
7d67af2c PB |
4468 | case IORING_OP_SPLICE: |
4469 | io_put_file(req, req->splice.file_in, | |
4470 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
4471 | break; | |
99bc4c38 PB |
4472 | } |
4473 | ||
4474 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4475 | } | |
4476 | ||
3529d8c2 JA |
4477 | static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
4478 | struct io_kiocb **nxt, bool force_nonblock) | |
2b188cc1 | 4479 | { |
a197f664 | 4480 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 4481 | int ret; |
2b188cc1 | 4482 | |
d625c6ee | 4483 | switch (req->opcode) { |
2b188cc1 | 4484 | case IORING_OP_NOP: |
78e19bbe | 4485 | ret = io_nop(req); |
2b188cc1 JA |
4486 | break; |
4487 | case IORING_OP_READV: | |
edafccee | 4488 | case IORING_OP_READ_FIXED: |
3a6820f2 | 4489 | case IORING_OP_READ: |
3529d8c2 JA |
4490 | if (sqe) { |
4491 | ret = io_read_prep(req, sqe, force_nonblock); | |
4492 | if (ret < 0) | |
4493 | break; | |
4494 | } | |
267bc904 | 4495 | ret = io_read(req, nxt, force_nonblock); |
edafccee | 4496 | break; |
3529d8c2 | 4497 | case IORING_OP_WRITEV: |
edafccee | 4498 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 4499 | case IORING_OP_WRITE: |
3529d8c2 JA |
4500 | if (sqe) { |
4501 | ret = io_write_prep(req, sqe, force_nonblock); | |
4502 | if (ret < 0) | |
4503 | break; | |
4504 | } | |
267bc904 | 4505 | ret = io_write(req, nxt, force_nonblock); |
2b188cc1 | 4506 | break; |
c992fe29 | 4507 | case IORING_OP_FSYNC: |
3529d8c2 JA |
4508 | if (sqe) { |
4509 | ret = io_prep_fsync(req, sqe); | |
4510 | if (ret < 0) | |
4511 | break; | |
4512 | } | |
fc4df999 | 4513 | ret = io_fsync(req, nxt, force_nonblock); |
c992fe29 | 4514 | break; |
221c5eb2 | 4515 | case IORING_OP_POLL_ADD: |
3529d8c2 JA |
4516 | if (sqe) { |
4517 | ret = io_poll_add_prep(req, sqe); | |
4518 | if (ret) | |
4519 | break; | |
4520 | } | |
fc4df999 | 4521 | ret = io_poll_add(req, nxt); |
221c5eb2 JA |
4522 | break; |
4523 | case IORING_OP_POLL_REMOVE: | |
3529d8c2 JA |
4524 | if (sqe) { |
4525 | ret = io_poll_remove_prep(req, sqe); | |
4526 | if (ret < 0) | |
4527 | break; | |
4528 | } | |
fc4df999 | 4529 | ret = io_poll_remove(req); |
221c5eb2 | 4530 | break; |
5d17b4a4 | 4531 | case IORING_OP_SYNC_FILE_RANGE: |
3529d8c2 JA |
4532 | if (sqe) { |
4533 | ret = io_prep_sfr(req, sqe); | |
4534 | if (ret < 0) | |
4535 | break; | |
4536 | } | |
fc4df999 | 4537 | ret = io_sync_file_range(req, nxt, force_nonblock); |
5d17b4a4 | 4538 | break; |
0fa03c62 | 4539 | case IORING_OP_SENDMSG: |
fddaface | 4540 | case IORING_OP_SEND: |
3529d8c2 JA |
4541 | if (sqe) { |
4542 | ret = io_sendmsg_prep(req, sqe); | |
4543 | if (ret < 0) | |
4544 | break; | |
4545 | } | |
fddaface JA |
4546 | if (req->opcode == IORING_OP_SENDMSG) |
4547 | ret = io_sendmsg(req, nxt, force_nonblock); | |
4548 | else | |
4549 | ret = io_send(req, nxt, force_nonblock); | |
0fa03c62 | 4550 | break; |
aa1fa28f | 4551 | case IORING_OP_RECVMSG: |
fddaface | 4552 | case IORING_OP_RECV: |
3529d8c2 JA |
4553 | if (sqe) { |
4554 | ret = io_recvmsg_prep(req, sqe); | |
4555 | if (ret) | |
4556 | break; | |
4557 | } | |
fddaface JA |
4558 | if (req->opcode == IORING_OP_RECVMSG) |
4559 | ret = io_recvmsg(req, nxt, force_nonblock); | |
4560 | else | |
4561 | ret = io_recv(req, nxt, force_nonblock); | |
aa1fa28f | 4562 | break; |
5262f567 | 4563 | case IORING_OP_TIMEOUT: |
3529d8c2 JA |
4564 | if (sqe) { |
4565 | ret = io_timeout_prep(req, sqe, false); | |
4566 | if (ret) | |
4567 | break; | |
4568 | } | |
fc4df999 | 4569 | ret = io_timeout(req); |
5262f567 | 4570 | break; |
11365043 | 4571 | case IORING_OP_TIMEOUT_REMOVE: |
3529d8c2 JA |
4572 | if (sqe) { |
4573 | ret = io_timeout_remove_prep(req, sqe); | |
4574 | if (ret) | |
4575 | break; | |
4576 | } | |
fc4df999 | 4577 | ret = io_timeout_remove(req); |
11365043 | 4578 | break; |
17f2fe35 | 4579 | case IORING_OP_ACCEPT: |
3529d8c2 JA |
4580 | if (sqe) { |
4581 | ret = io_accept_prep(req, sqe); | |
4582 | if (ret) | |
4583 | break; | |
4584 | } | |
fc4df999 | 4585 | ret = io_accept(req, nxt, force_nonblock); |
17f2fe35 | 4586 | break; |
f8e85cf2 | 4587 | case IORING_OP_CONNECT: |
3529d8c2 JA |
4588 | if (sqe) { |
4589 | ret = io_connect_prep(req, sqe); | |
4590 | if (ret) | |
4591 | break; | |
4592 | } | |
fc4df999 | 4593 | ret = io_connect(req, nxt, force_nonblock); |
f8e85cf2 | 4594 | break; |
62755e35 | 4595 | case IORING_OP_ASYNC_CANCEL: |
3529d8c2 JA |
4596 | if (sqe) { |
4597 | ret = io_async_cancel_prep(req, sqe); | |
4598 | if (ret) | |
4599 | break; | |
4600 | } | |
fc4df999 | 4601 | ret = io_async_cancel(req, nxt); |
62755e35 | 4602 | break; |
d63d1b5e JA |
4603 | case IORING_OP_FALLOCATE: |
4604 | if (sqe) { | |
4605 | ret = io_fallocate_prep(req, sqe); | |
4606 | if (ret) | |
4607 | break; | |
4608 | } | |
4609 | ret = io_fallocate(req, nxt, force_nonblock); | |
4610 | break; | |
15b71abe JA |
4611 | case IORING_OP_OPENAT: |
4612 | if (sqe) { | |
4613 | ret = io_openat_prep(req, sqe); | |
4614 | if (ret) | |
4615 | break; | |
4616 | } | |
4617 | ret = io_openat(req, nxt, force_nonblock); | |
4618 | break; | |
b5dba59e JA |
4619 | case IORING_OP_CLOSE: |
4620 | if (sqe) { | |
4621 | ret = io_close_prep(req, sqe); | |
4622 | if (ret) | |
4623 | break; | |
4624 | } | |
4625 | ret = io_close(req, nxt, force_nonblock); | |
4626 | break; | |
05f3fb3c JA |
4627 | case IORING_OP_FILES_UPDATE: |
4628 | if (sqe) { | |
4629 | ret = io_files_update_prep(req, sqe); | |
4630 | if (ret) | |
4631 | break; | |
4632 | } | |
4633 | ret = io_files_update(req, force_nonblock); | |
4634 | break; | |
eddc7ef5 JA |
4635 | case IORING_OP_STATX: |
4636 | if (sqe) { | |
4637 | ret = io_statx_prep(req, sqe); | |
4638 | if (ret) | |
4639 | break; | |
4640 | } | |
4641 | ret = io_statx(req, nxt, force_nonblock); | |
4642 | break; | |
4840e418 JA |
4643 | case IORING_OP_FADVISE: |
4644 | if (sqe) { | |
4645 | ret = io_fadvise_prep(req, sqe); | |
4646 | if (ret) | |
4647 | break; | |
4648 | } | |
4649 | ret = io_fadvise(req, nxt, force_nonblock); | |
4650 | break; | |
c1ca757b JA |
4651 | case IORING_OP_MADVISE: |
4652 | if (sqe) { | |
4653 | ret = io_madvise_prep(req, sqe); | |
4654 | if (ret) | |
4655 | break; | |
4656 | } | |
4657 | ret = io_madvise(req, nxt, force_nonblock); | |
4658 | break; | |
cebdb986 JA |
4659 | case IORING_OP_OPENAT2: |
4660 | if (sqe) { | |
4661 | ret = io_openat2_prep(req, sqe); | |
4662 | if (ret) | |
4663 | break; | |
4664 | } | |
4665 | ret = io_openat2(req, nxt, force_nonblock); | |
4666 | break; | |
3e4827b0 JA |
4667 | case IORING_OP_EPOLL_CTL: |
4668 | if (sqe) { | |
4669 | ret = io_epoll_ctl_prep(req, sqe); | |
4670 | if (ret) | |
4671 | break; | |
4672 | } | |
4673 | ret = io_epoll_ctl(req, nxt, force_nonblock); | |
4674 | break; | |
7d67af2c PB |
4675 | case IORING_OP_SPLICE: |
4676 | if (sqe) { | |
4677 | ret = io_splice_prep(req, sqe); | |
4678 | if (ret < 0) | |
4679 | break; | |
4680 | } | |
4681 | ret = io_splice(req, nxt, force_nonblock); | |
4682 | break; | |
2b188cc1 JA |
4683 | default: |
4684 | ret = -EINVAL; | |
4685 | break; | |
4686 | } | |
4687 | ||
def596e9 JA |
4688 | if (ret) |
4689 | return ret; | |
4690 | ||
4691 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
11ba820b JA |
4692 | const bool in_async = io_wq_current_is_worker(); |
4693 | ||
9e645e11 | 4694 | if (req->result == -EAGAIN) |
def596e9 JA |
4695 | return -EAGAIN; |
4696 | ||
11ba820b JA |
4697 | /* workqueue context doesn't hold uring_lock, grab it now */ |
4698 | if (in_async) | |
4699 | mutex_lock(&ctx->uring_lock); | |
4700 | ||
def596e9 | 4701 | io_iopoll_req_issued(req); |
11ba820b JA |
4702 | |
4703 | if (in_async) | |
4704 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
4705 | } |
4706 | ||
4707 | return 0; | |
2b188cc1 JA |
4708 | } |
4709 | ||
561fb04a | 4710 | static void io_wq_submit_work(struct io_wq_work **workptr) |
2b188cc1 | 4711 | { |
561fb04a | 4712 | struct io_wq_work *work = *workptr; |
2b188cc1 | 4713 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
561fb04a JA |
4714 | struct io_kiocb *nxt = NULL; |
4715 | int ret = 0; | |
2b188cc1 | 4716 | |
0c9d5ccd JA |
4717 | /* if NO_CANCEL is set, we must still run the work */ |
4718 | if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) == | |
4719 | IO_WQ_WORK_CANCEL) { | |
561fb04a | 4720 | ret = -ECANCELED; |
0c9d5ccd | 4721 | } |
31b51510 | 4722 | |
561fb04a | 4723 | if (!ret) { |
561fb04a | 4724 | do { |
3529d8c2 | 4725 | ret = io_issue_sqe(req, NULL, &nxt, false); |
561fb04a JA |
4726 | /* |
4727 | * We can get EAGAIN for polled IO even though we're | |
4728 | * forcing a sync submission from here, since we can't | |
4729 | * wait for request slots on the block side. | |
4730 | */ | |
4731 | if (ret != -EAGAIN) | |
4732 | break; | |
4733 | cond_resched(); | |
4734 | } while (1); | |
4735 | } | |
31b51510 | 4736 | |
561fb04a | 4737 | /* drop submission reference */ |
ec9c02ad | 4738 | io_put_req(req); |
817869d2 | 4739 | |
561fb04a | 4740 | if (ret) { |
4e88d6e7 | 4741 | req_set_fail_links(req); |
78e19bbe | 4742 | io_cqring_add_event(req, ret); |
817869d2 | 4743 | io_put_req(req); |
edafccee | 4744 | } |
2b188cc1 | 4745 | |
561fb04a | 4746 | /* if a dependent link is ready, pass it back */ |
78912934 JA |
4747 | if (!ret && nxt) |
4748 | io_wq_assign_next(workptr, nxt); | |
2b188cc1 JA |
4749 | } |
4750 | ||
15b71abe | 4751 | static int io_req_needs_file(struct io_kiocb *req, int fd) |
9e3aa61a | 4752 | { |
d3656344 | 4753 | if (!io_op_defs[req->opcode].needs_file) |
9e3aa61a | 4754 | return 0; |
0b5faf6b | 4755 | if ((fd == -1 || fd == AT_FDCWD) && io_op_defs[req->opcode].fd_non_neg) |
d3656344 JA |
4756 | return 0; |
4757 | return 1; | |
09bb8394 JA |
4758 | } |
4759 | ||
65e19f54 JA |
4760 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
4761 | int index) | |
4762 | { | |
4763 | struct fixed_file_table *table; | |
4764 | ||
05f3fb3c JA |
4765 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
4766 | return table->files[index & IORING_FILE_TABLE_MASK];; | |
65e19f54 JA |
4767 | } |
4768 | ||
8da11c19 PB |
4769 | static int io_file_get(struct io_submit_state *state, struct io_kiocb *req, |
4770 | int fd, struct file **out_file, bool fixed) | |
09bb8394 | 4771 | { |
a197f664 | 4772 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 4773 | struct file *file; |
09bb8394 | 4774 | |
8da11c19 | 4775 | if (fixed) { |
05f3fb3c | 4776 | if (unlikely(!ctx->file_data || |
09bb8394 JA |
4777 | (unsigned) fd >= ctx->nr_user_files)) |
4778 | return -EBADF; | |
b7620121 | 4779 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 PB |
4780 | file = io_file_from_index(ctx, fd); |
4781 | if (!file) | |
08a45173 | 4782 | return -EBADF; |
05f3fb3c | 4783 | percpu_ref_get(&ctx->file_data->refs); |
09bb8394 | 4784 | } else { |
c826bd7a | 4785 | trace_io_uring_file_get(ctx, fd); |
8da11c19 PB |
4786 | file = __io_file_get(state, fd); |
4787 | if (unlikely(!file)) | |
09bb8394 JA |
4788 | return -EBADF; |
4789 | } | |
4790 | ||
8da11c19 | 4791 | *out_file = file; |
09bb8394 JA |
4792 | return 0; |
4793 | } | |
4794 | ||
8da11c19 PB |
4795 | static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req, |
4796 | const struct io_uring_sqe *sqe) | |
4797 | { | |
4798 | unsigned flags; | |
4799 | int fd; | |
4800 | bool fixed; | |
4801 | ||
4802 | flags = READ_ONCE(sqe->flags); | |
4803 | fd = READ_ONCE(sqe->fd); | |
4804 | ||
4805 | if (!io_req_needs_file(req, fd)) | |
4806 | return 0; | |
4807 | ||
4808 | fixed = (flags & IOSQE_FIXED_FILE); | |
4809 | if (unlikely(!fixed && req->needs_fixed_file)) | |
4810 | return -EBADF; | |
4811 | ||
4812 | return io_file_get(state, req, fd, &req->file, fixed); | |
4813 | } | |
4814 | ||
a197f664 | 4815 | static int io_grab_files(struct io_kiocb *req) |
fcb323cc JA |
4816 | { |
4817 | int ret = -EBADF; | |
a197f664 | 4818 | struct io_ring_ctx *ctx = req->ctx; |
fcb323cc | 4819 | |
f86cd20c JA |
4820 | if (req->work.files) |
4821 | return 0; | |
b14cca0c | 4822 | if (!ctx->ring_file) |
b5dba59e JA |
4823 | return -EBADF; |
4824 | ||
fcb323cc JA |
4825 | rcu_read_lock(); |
4826 | spin_lock_irq(&ctx->inflight_lock); | |
4827 | /* | |
4828 | * We use the f_ops->flush() handler to ensure that we can flush | |
4829 | * out work accessing these files if the fd is closed. Check if | |
4830 | * the fd has changed since we started down this path, and disallow | |
4831 | * this operation if it has. | |
4832 | */ | |
b14cca0c | 4833 | if (fcheck(ctx->ring_fd) == ctx->ring_file) { |
fcb323cc JA |
4834 | list_add(&req->inflight_entry, &ctx->inflight_list); |
4835 | req->flags |= REQ_F_INFLIGHT; | |
4836 | req->work.files = current->files; | |
4837 | ret = 0; | |
4838 | } | |
4839 | spin_unlock_irq(&ctx->inflight_lock); | |
4840 | rcu_read_unlock(); | |
4841 | ||
4842 | return ret; | |
4843 | } | |
4844 | ||
2665abfd | 4845 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 4846 | { |
ad8a48ac JA |
4847 | struct io_timeout_data *data = container_of(timer, |
4848 | struct io_timeout_data, timer); | |
4849 | struct io_kiocb *req = data->req; | |
2665abfd JA |
4850 | struct io_ring_ctx *ctx = req->ctx; |
4851 | struct io_kiocb *prev = NULL; | |
4852 | unsigned long flags; | |
2665abfd JA |
4853 | |
4854 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
4855 | ||
4856 | /* | |
4857 | * We don't expect the list to be empty, that will only happen if we | |
4858 | * race with the completion of the linked work. | |
4859 | */ | |
4493233e PB |
4860 | if (!list_empty(&req->link_list)) { |
4861 | prev = list_entry(req->link_list.prev, struct io_kiocb, | |
4862 | link_list); | |
5d960724 | 4863 | if (refcount_inc_not_zero(&prev->refs)) { |
4493233e | 4864 | list_del_init(&req->link_list); |
5d960724 JA |
4865 | prev->flags &= ~REQ_F_LINK_TIMEOUT; |
4866 | } else | |
76a46e06 | 4867 | prev = NULL; |
2665abfd JA |
4868 | } |
4869 | ||
4870 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
4871 | ||
4872 | if (prev) { | |
4e88d6e7 | 4873 | req_set_fail_links(prev); |
b0dd8a41 JA |
4874 | io_async_find_and_cancel(ctx, req, prev->user_data, NULL, |
4875 | -ETIME); | |
76a46e06 | 4876 | io_put_req(prev); |
47f46768 JA |
4877 | } else { |
4878 | io_cqring_add_event(req, -ETIME); | |
4879 | io_put_req(req); | |
2665abfd | 4880 | } |
2665abfd JA |
4881 | return HRTIMER_NORESTART; |
4882 | } | |
4883 | ||
ad8a48ac | 4884 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 4885 | { |
76a46e06 | 4886 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 4887 | |
76a46e06 JA |
4888 | /* |
4889 | * If the list is now empty, then our linked request finished before | |
4890 | * we got a chance to setup the timer | |
4891 | */ | |
4892 | spin_lock_irq(&ctx->completion_lock); | |
4493233e | 4893 | if (!list_empty(&req->link_list)) { |
2d28390a | 4894 | struct io_timeout_data *data = &req->io->timeout; |
94ae5e77 | 4895 | |
ad8a48ac JA |
4896 | data->timer.function = io_link_timeout_fn; |
4897 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
4898 | data->mode); | |
2665abfd | 4899 | } |
76a46e06 | 4900 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 4901 | |
2665abfd | 4902 | /* drop submission reference */ |
76a46e06 JA |
4903 | io_put_req(req); |
4904 | } | |
2665abfd | 4905 | |
ad8a48ac | 4906 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd JA |
4907 | { |
4908 | struct io_kiocb *nxt; | |
4909 | ||
4910 | if (!(req->flags & REQ_F_LINK)) | |
4911 | return NULL; | |
d7718a9d JA |
4912 | /* for polled retry, if flag is set, we already went through here */ |
4913 | if (req->flags & REQ_F_POLLED) | |
4914 | return NULL; | |
2665abfd | 4915 | |
4493233e PB |
4916 | nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, |
4917 | link_list); | |
d625c6ee | 4918 | if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT) |
76a46e06 | 4919 | return NULL; |
2665abfd | 4920 | |
76a46e06 | 4921 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 4922 | return nxt; |
2665abfd JA |
4923 | } |
4924 | ||
3529d8c2 | 4925 | static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 4926 | { |
4a0a7a18 | 4927 | struct io_kiocb *linked_timeout; |
4bc4494e | 4928 | struct io_kiocb *nxt; |
193155c8 | 4929 | const struct cred *old_creds = NULL; |
e0c5c576 | 4930 | int ret; |
2b188cc1 | 4931 | |
4a0a7a18 JA |
4932 | again: |
4933 | linked_timeout = io_prep_linked_timeout(req); | |
4934 | ||
193155c8 JA |
4935 | if (req->work.creds && req->work.creds != current_cred()) { |
4936 | if (old_creds) | |
4937 | revert_creds(old_creds); | |
4938 | if (old_creds == req->work.creds) | |
4939 | old_creds = NULL; /* restored original creds */ | |
4940 | else | |
4941 | old_creds = override_creds(req->work.creds); | |
4942 | } | |
4943 | ||
3529d8c2 | 4944 | ret = io_issue_sqe(req, sqe, &nxt, true); |
491381ce JA |
4945 | |
4946 | /* | |
4947 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
4948 | * doesn't support non-blocking read/write attempts | |
4949 | */ | |
4950 | if (ret == -EAGAIN && (!(req->flags & REQ_F_NOWAIT) || | |
4951 | (req->flags & REQ_F_MUST_PUNT))) { | |
d7718a9d JA |
4952 | if (io_arm_poll_handler(req)) { |
4953 | if (linked_timeout) | |
4954 | io_queue_linked_timeout(linked_timeout); | |
4bc4494e | 4955 | goto exit; |
d7718a9d | 4956 | } |
86a761f8 | 4957 | punt: |
f86cd20c | 4958 | if (io_op_defs[req->opcode].file_table) { |
bbad27b2 PB |
4959 | ret = io_grab_files(req); |
4960 | if (ret) | |
4961 | goto err; | |
2b188cc1 | 4962 | } |
bbad27b2 PB |
4963 | |
4964 | /* | |
4965 | * Queued up for async execution, worker will release | |
4966 | * submit reference when the iocb is actually submitted. | |
4967 | */ | |
4968 | io_queue_async_work(req); | |
4bc4494e | 4969 | goto exit; |
2b188cc1 | 4970 | } |
e65ef56d | 4971 | |
fcb323cc | 4972 | err: |
4bc4494e | 4973 | nxt = NULL; |
76a46e06 | 4974 | /* drop submission reference */ |
2a44f467 | 4975 | io_put_req_find_next(req, &nxt); |
e65ef56d | 4976 | |
f9bd67f6 | 4977 | if (linked_timeout) { |
76a46e06 | 4978 | if (!ret) |
f9bd67f6 | 4979 | io_queue_linked_timeout(linked_timeout); |
76a46e06 | 4980 | else |
f9bd67f6 | 4981 | io_put_req(linked_timeout); |
76a46e06 JA |
4982 | } |
4983 | ||
e65ef56d | 4984 | /* and drop final reference, if we failed */ |
9e645e11 | 4985 | if (ret) { |
78e19bbe | 4986 | io_cqring_add_event(req, ret); |
4e88d6e7 | 4987 | req_set_fail_links(req); |
e65ef56d | 4988 | io_put_req(req); |
9e645e11 | 4989 | } |
4a0a7a18 JA |
4990 | if (nxt) { |
4991 | req = nxt; | |
86a761f8 PB |
4992 | |
4993 | if (req->flags & REQ_F_FORCE_ASYNC) | |
4994 | goto punt; | |
4a0a7a18 JA |
4995 | goto again; |
4996 | } | |
4bc4494e | 4997 | exit: |
193155c8 JA |
4998 | if (old_creds) |
4999 | revert_creds(old_creds); | |
2b188cc1 JA |
5000 | } |
5001 | ||
3529d8c2 | 5002 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
5003 | { |
5004 | int ret; | |
5005 | ||
3529d8c2 | 5006 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
5007 | if (ret) { |
5008 | if (ret != -EIOCBQUEUED) { | |
1118591a | 5009 | fail_req: |
78e19bbe | 5010 | io_cqring_add_event(req, ret); |
4e88d6e7 | 5011 | req_set_fail_links(req); |
78e19bbe | 5012 | io_double_put_req(req); |
4fe2c963 | 5013 | } |
2550878f | 5014 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
1118591a PB |
5015 | ret = io_req_defer_prep(req, sqe); |
5016 | if (unlikely(ret < 0)) | |
5017 | goto fail_req; | |
ce35a47a JA |
5018 | /* |
5019 | * Never try inline submit of IOSQE_ASYNC is set, go straight | |
5020 | * to async execution. | |
5021 | */ | |
5022 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
5023 | io_queue_async_work(req); | |
5024 | } else { | |
3529d8c2 | 5025 | __io_queue_sqe(req, sqe); |
ce35a47a | 5026 | } |
4fe2c963 JL |
5027 | } |
5028 | ||
1b4a51b6 | 5029 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 5030 | { |
94ae5e77 | 5031 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
1b4a51b6 PB |
5032 | io_cqring_add_event(req, -ECANCELED); |
5033 | io_double_put_req(req); | |
5034 | } else | |
3529d8c2 | 5035 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
5036 | } |
5037 | ||
4e88d6e7 | 5038 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
ce35a47a | 5039 | IOSQE_IO_HARDLINK | IOSQE_ASYNC) |
9e645e11 | 5040 | |
3529d8c2 JA |
5041 | static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
5042 | struct io_submit_state *state, struct io_kiocb **link) | |
9e645e11 | 5043 | { |
a197f664 | 5044 | struct io_ring_ctx *ctx = req->ctx; |
32fe525b | 5045 | unsigned int sqe_flags; |
75c6a039 | 5046 | int ret, id; |
9e645e11 | 5047 | |
32fe525b | 5048 | sqe_flags = READ_ONCE(sqe->flags); |
9e645e11 JA |
5049 | |
5050 | /* enforce forwards compatibility on users */ | |
32fe525b | 5051 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
9e645e11 | 5052 | ret = -EINVAL; |
196be95c | 5053 | goto err_req; |
9e645e11 JA |
5054 | } |
5055 | ||
75c6a039 JA |
5056 | id = READ_ONCE(sqe->personality); |
5057 | if (id) { | |
193155c8 JA |
5058 | req->work.creds = idr_find(&ctx->personality_idr, id); |
5059 | if (unlikely(!req->work.creds)) { | |
75c6a039 JA |
5060 | ret = -EINVAL; |
5061 | goto err_req; | |
5062 | } | |
193155c8 | 5063 | get_cred(req->work.creds); |
75c6a039 JA |
5064 | } |
5065 | ||
6b47ee6e | 5066 | /* same numerical values with corresponding REQ_F_*, safe to copy */ |
8da11c19 PB |
5067 | req->flags |= sqe_flags & (IOSQE_IO_DRAIN | IOSQE_IO_HARDLINK | |
5068 | IOSQE_ASYNC | IOSQE_FIXED_FILE); | |
9e645e11 | 5069 | |
3529d8c2 | 5070 | ret = io_req_set_file(state, req, sqe); |
9e645e11 JA |
5071 | if (unlikely(ret)) { |
5072 | err_req: | |
78e19bbe JA |
5073 | io_cqring_add_event(req, ret); |
5074 | io_double_put_req(req); | |
2e6e1fde | 5075 | return false; |
9e645e11 JA |
5076 | } |
5077 | ||
9e645e11 JA |
5078 | /* |
5079 | * If we already have a head request, queue this one for async | |
5080 | * submittal once the head completes. If we don't have a head but | |
5081 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
5082 | * submitted sync once the chain is complete. If none of those | |
5083 | * conditions are true (normal request), then just queue it. | |
5084 | */ | |
5085 | if (*link) { | |
9d76377f | 5086 | struct io_kiocb *head = *link; |
4e88d6e7 | 5087 | |
8cdf2193 PB |
5088 | /* |
5089 | * Taking sequential execution of a link, draining both sides | |
5090 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
5091 | * requests in the link. So, it drains the head and the | |
5092 | * next after the link request. The last one is done via | |
5093 | * drain_next flag to persist the effect across calls. | |
5094 | */ | |
711be031 PB |
5095 | if (sqe_flags & IOSQE_IO_DRAIN) { |
5096 | head->flags |= REQ_F_IO_DRAIN; | |
5097 | ctx->drain_next = 1; | |
5098 | } | |
b7bb4f7d | 5099 | if (io_alloc_async_ctx(req)) { |
9e645e11 JA |
5100 | ret = -EAGAIN; |
5101 | goto err_req; | |
5102 | } | |
5103 | ||
3529d8c2 | 5104 | ret = io_req_defer_prep(req, sqe); |
2d28390a | 5105 | if (ret) { |
4e88d6e7 | 5106 | /* fail even hard links since we don't submit */ |
9d76377f | 5107 | head->flags |= REQ_F_FAIL_LINK; |
f67676d1 | 5108 | goto err_req; |
2d28390a | 5109 | } |
9d76377f PB |
5110 | trace_io_uring_link(ctx, req, head); |
5111 | list_add_tail(&req->link_list, &head->link_list); | |
32fe525b PB |
5112 | |
5113 | /* last request of a link, enqueue the link */ | |
5114 | if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) { | |
5115 | io_queue_link_head(head); | |
5116 | *link = NULL; | |
5117 | } | |
9e645e11 | 5118 | } else { |
711be031 PB |
5119 | if (unlikely(ctx->drain_next)) { |
5120 | req->flags |= REQ_F_IO_DRAIN; | |
5121 | req->ctx->drain_next = 0; | |
5122 | } | |
5123 | if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) { | |
5124 | req->flags |= REQ_F_LINK; | |
711be031 PB |
5125 | INIT_LIST_HEAD(&req->link_list); |
5126 | ret = io_req_defer_prep(req, sqe); | |
5127 | if (ret) | |
5128 | req->flags |= REQ_F_FAIL_LINK; | |
5129 | *link = req; | |
5130 | } else { | |
5131 | io_queue_sqe(req, sqe); | |
5132 | } | |
9e645e11 | 5133 | } |
2e6e1fde PB |
5134 | |
5135 | return true; | |
9e645e11 JA |
5136 | } |
5137 | ||
9a56a232 JA |
5138 | /* |
5139 | * Batched submission is done, ensure local IO is flushed out. | |
5140 | */ | |
5141 | static void io_submit_state_end(struct io_submit_state *state) | |
5142 | { | |
5143 | blk_finish_plug(&state->plug); | |
3d6770fb | 5144 | io_file_put(state); |
2579f913 | 5145 | if (state->free_reqs) |
6c8a3134 | 5146 | kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); |
9a56a232 JA |
5147 | } |
5148 | ||
5149 | /* | |
5150 | * Start submission side cache. | |
5151 | */ | |
5152 | static void io_submit_state_start(struct io_submit_state *state, | |
22efde59 | 5153 | unsigned int max_ios) |
9a56a232 JA |
5154 | { |
5155 | blk_start_plug(&state->plug); | |
2579f913 | 5156 | state->free_reqs = 0; |
9a56a232 JA |
5157 | state->file = NULL; |
5158 | state->ios_left = max_ios; | |
5159 | } | |
5160 | ||
2b188cc1 JA |
5161 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
5162 | { | |
75b28aff | 5163 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 5164 | |
caf582c6 PB |
5165 | /* |
5166 | * Ensure any loads from the SQEs are done at this point, | |
5167 | * since once we write the new head, the application could | |
5168 | * write new data to them. | |
5169 | */ | |
5170 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
5171 | } |
5172 | ||
2b188cc1 | 5173 | /* |
3529d8c2 | 5174 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
5175 | * that is mapped by userspace. This means that care needs to be taken to |
5176 | * ensure that reads are stable, as we cannot rely on userspace always | |
5177 | * being a good citizen. If members of the sqe are validated and then later | |
5178 | * used, it's important that those reads are done through READ_ONCE() to | |
5179 | * prevent a re-load down the line. | |
5180 | */ | |
3529d8c2 JA |
5181 | static bool io_get_sqring(struct io_ring_ctx *ctx, struct io_kiocb *req, |
5182 | const struct io_uring_sqe **sqe_ptr) | |
2b188cc1 | 5183 | { |
75b28aff | 5184 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
5185 | unsigned head; |
5186 | ||
5187 | /* | |
5188 | * The cached sq head (or cq tail) serves two purposes: | |
5189 | * | |
5190 | * 1) allows us to batch the cost of updating the user visible | |
5191 | * head updates. | |
5192 | * 2) allows the kernel side to track the head on its own, even | |
5193 | * though the application is the one updating it. | |
5194 | */ | |
ee7d46d9 | 5195 | head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]); |
9835d6fa | 5196 | if (likely(head < ctx->sq_entries)) { |
cf6fd4bd PB |
5197 | /* |
5198 | * All io need record the previous position, if LINK vs DARIN, | |
5199 | * it can be used to mark the position of the first IO in the | |
5200 | * link list. | |
5201 | */ | |
5202 | req->sequence = ctx->cached_sq_head; | |
3529d8c2 JA |
5203 | *sqe_ptr = &ctx->sq_sqes[head]; |
5204 | req->opcode = READ_ONCE((*sqe_ptr)->opcode); | |
5205 | req->user_data = READ_ONCE((*sqe_ptr)->user_data); | |
2b188cc1 JA |
5206 | ctx->cached_sq_head++; |
5207 | return true; | |
5208 | } | |
5209 | ||
5210 | /* drop invalid entries */ | |
5211 | ctx->cached_sq_head++; | |
498ccd9e | 5212 | ctx->cached_sq_dropped++; |
ee7d46d9 | 5213 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
2b188cc1 JA |
5214 | return false; |
5215 | } | |
5216 | ||
fb5ccc98 | 5217 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr, |
ae9428ca PB |
5218 | struct file *ring_file, int ring_fd, |
5219 | struct mm_struct **mm, bool async) | |
6c271ce2 JA |
5220 | { |
5221 | struct io_submit_state state, *statep = NULL; | |
9e645e11 | 5222 | struct io_kiocb *link = NULL; |
9e645e11 | 5223 | int i, submitted = 0; |
95a1b3ff | 5224 | bool mm_fault = false; |
6c271ce2 | 5225 | |
c4a2ed72 | 5226 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 JA |
5227 | if (test_bit(0, &ctx->sq_check_overflow)) { |
5228 | if (!list_empty(&ctx->cq_overflow_list) && | |
5229 | !io_cqring_overflow_flush(ctx, false)) | |
5230 | return -EBUSY; | |
5231 | } | |
6c271ce2 | 5232 | |
ee7d46d9 PB |
5233 | /* make sure SQ entry isn't read before tail */ |
5234 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 5235 | |
2b85edfc PB |
5236 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
5237 | return -EAGAIN; | |
6c271ce2 JA |
5238 | |
5239 | if (nr > IO_PLUG_THRESHOLD) { | |
22efde59 | 5240 | io_submit_state_start(&state, nr); |
6c271ce2 JA |
5241 | statep = &state; |
5242 | } | |
5243 | ||
b14cca0c PB |
5244 | ctx->ring_fd = ring_fd; |
5245 | ctx->ring_file = ring_file; | |
5246 | ||
6c271ce2 | 5247 | for (i = 0; i < nr; i++) { |
3529d8c2 | 5248 | const struct io_uring_sqe *sqe; |
196be95c | 5249 | struct io_kiocb *req; |
1cb1edb2 | 5250 | int err; |
fb5ccc98 | 5251 | |
196be95c PB |
5252 | req = io_get_req(ctx, statep); |
5253 | if (unlikely(!req)) { | |
5254 | if (!submitted) | |
5255 | submitted = -EAGAIN; | |
fb5ccc98 | 5256 | break; |
196be95c | 5257 | } |
3529d8c2 | 5258 | if (!io_get_sqring(ctx, req, &sqe)) { |
2b85edfc | 5259 | __io_req_do_free(req); |
196be95c PB |
5260 | break; |
5261 | } | |
fb5ccc98 | 5262 | |
d3656344 JA |
5263 | /* will complete beyond this point, count as submitted */ |
5264 | submitted++; | |
5265 | ||
5266 | if (unlikely(req->opcode >= IORING_OP_LAST)) { | |
1cb1edb2 PB |
5267 | err = -EINVAL; |
5268 | fail_req: | |
5269 | io_cqring_add_event(req, err); | |
d3656344 | 5270 | io_double_put_req(req); |
196be95c PB |
5271 | break; |
5272 | } | |
fb5ccc98 | 5273 | |
d3656344 | 5274 | if (io_op_defs[req->opcode].needs_mm && !*mm) { |
95a1b3ff | 5275 | mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm); |
1cb1edb2 PB |
5276 | if (unlikely(mm_fault)) { |
5277 | err = -EFAULT; | |
5278 | goto fail_req; | |
95a1b3ff | 5279 | } |
1cb1edb2 PB |
5280 | use_mm(ctx->sqo_mm); |
5281 | *mm = ctx->sqo_mm; | |
9e645e11 | 5282 | } |
9e645e11 | 5283 | |
cf6fd4bd | 5284 | req->needs_fixed_file = async; |
354420f7 JA |
5285 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
5286 | true, async); | |
3529d8c2 | 5287 | if (!io_submit_sqe(req, sqe, statep, &link)) |
2e6e1fde | 5288 | break; |
6c271ce2 JA |
5289 | } |
5290 | ||
9466f437 PB |
5291 | if (unlikely(submitted != nr)) { |
5292 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
5293 | ||
5294 | percpu_ref_put_many(&ctx->refs, nr - ref_used); | |
5295 | } | |
9e645e11 | 5296 | if (link) |
1b4a51b6 | 5297 | io_queue_link_head(link); |
6c271ce2 JA |
5298 | if (statep) |
5299 | io_submit_state_end(&state); | |
5300 | ||
ae9428ca PB |
5301 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
5302 | io_commit_sqring(ctx); | |
5303 | ||
6c271ce2 JA |
5304 | return submitted; |
5305 | } | |
5306 | ||
5307 | static int io_sq_thread(void *data) | |
5308 | { | |
6c271ce2 JA |
5309 | struct io_ring_ctx *ctx = data; |
5310 | struct mm_struct *cur_mm = NULL; | |
181e448d | 5311 | const struct cred *old_cred; |
6c271ce2 JA |
5312 | mm_segment_t old_fs; |
5313 | DEFINE_WAIT(wait); | |
6c271ce2 | 5314 | unsigned long timeout; |
bdcd3eab | 5315 | int ret = 0; |
6c271ce2 | 5316 | |
206aefde | 5317 | complete(&ctx->completions[1]); |
a4c0b3de | 5318 | |
6c271ce2 JA |
5319 | old_fs = get_fs(); |
5320 | set_fs(USER_DS); | |
181e448d | 5321 | old_cred = override_creds(ctx->creds); |
6c271ce2 | 5322 | |
bdcd3eab | 5323 | timeout = jiffies + ctx->sq_thread_idle; |
2bbcd6d3 | 5324 | while (!kthread_should_park()) { |
fb5ccc98 | 5325 | unsigned int to_submit; |
6c271ce2 | 5326 | |
bdcd3eab | 5327 | if (!list_empty(&ctx->poll_list)) { |
6c271ce2 JA |
5328 | unsigned nr_events = 0; |
5329 | ||
bdcd3eab XW |
5330 | mutex_lock(&ctx->uring_lock); |
5331 | if (!list_empty(&ctx->poll_list)) | |
5332 | io_iopoll_getevents(ctx, &nr_events, 0); | |
5333 | else | |
6c271ce2 | 5334 | timeout = jiffies + ctx->sq_thread_idle; |
bdcd3eab | 5335 | mutex_unlock(&ctx->uring_lock); |
6c271ce2 JA |
5336 | } |
5337 | ||
fb5ccc98 | 5338 | to_submit = io_sqring_entries(ctx); |
c1edbf5f JA |
5339 | |
5340 | /* | |
5341 | * If submit got -EBUSY, flag us as needing the application | |
5342 | * to enter the kernel to reap and flush events. | |
5343 | */ | |
5344 | if (!to_submit || ret == -EBUSY) { | |
7143b5ac SG |
5345 | /* |
5346 | * Drop cur_mm before scheduling, we can't hold it for | |
5347 | * long periods (or over schedule()). Do this before | |
5348 | * adding ourselves to the waitqueue, as the unuse/drop | |
5349 | * may sleep. | |
5350 | */ | |
5351 | if (cur_mm) { | |
5352 | unuse_mm(cur_mm); | |
5353 | mmput(cur_mm); | |
5354 | cur_mm = NULL; | |
5355 | } | |
5356 | ||
6c271ce2 JA |
5357 | /* |
5358 | * We're polling. If we're within the defined idle | |
5359 | * period, then let us spin without work before going | |
c1edbf5f JA |
5360 | * to sleep. The exception is if we got EBUSY doing |
5361 | * more IO, we should wait for the application to | |
5362 | * reap events and wake us up. | |
6c271ce2 | 5363 | */ |
bdcd3eab | 5364 | if (!list_empty(&ctx->poll_list) || |
df069d80 JA |
5365 | (!time_after(jiffies, timeout) && ret != -EBUSY && |
5366 | !percpu_ref_is_dying(&ctx->refs))) { | |
b41e9852 JA |
5367 | if (current->task_works) |
5368 | task_work_run(); | |
9831a90c | 5369 | cond_resched(); |
6c271ce2 JA |
5370 | continue; |
5371 | } | |
5372 | ||
6c271ce2 JA |
5373 | prepare_to_wait(&ctx->sqo_wait, &wait, |
5374 | TASK_INTERRUPTIBLE); | |
5375 | ||
bdcd3eab XW |
5376 | /* |
5377 | * While doing polled IO, before going to sleep, we need | |
5378 | * to check if there are new reqs added to poll_list, it | |
5379 | * is because reqs may have been punted to io worker and | |
5380 | * will be added to poll_list later, hence check the | |
5381 | * poll_list again. | |
5382 | */ | |
5383 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
5384 | !list_empty_careful(&ctx->poll_list)) { | |
5385 | finish_wait(&ctx->sqo_wait, &wait); | |
5386 | continue; | |
5387 | } | |
5388 | ||
6c271ce2 | 5389 | /* Tell userspace we may need a wakeup call */ |
75b28aff | 5390 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; |
0d7bae69 SB |
5391 | /* make sure to read SQ tail after writing flags */ |
5392 | smp_mb(); | |
6c271ce2 | 5393 | |
fb5ccc98 | 5394 | to_submit = io_sqring_entries(ctx); |
c1edbf5f | 5395 | if (!to_submit || ret == -EBUSY) { |
2bbcd6d3 | 5396 | if (kthread_should_park()) { |
6c271ce2 JA |
5397 | finish_wait(&ctx->sqo_wait, &wait); |
5398 | break; | |
5399 | } | |
b41e9852 JA |
5400 | if (current->task_works) { |
5401 | task_work_run(); | |
5402 | continue; | |
5403 | } | |
6c271ce2 JA |
5404 | if (signal_pending(current)) |
5405 | flush_signals(current); | |
5406 | schedule(); | |
5407 | finish_wait(&ctx->sqo_wait, &wait); | |
5408 | ||
75b28aff | 5409 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5410 | continue; |
5411 | } | |
5412 | finish_wait(&ctx->sqo_wait, &wait); | |
5413 | ||
75b28aff | 5414 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; |
6c271ce2 JA |
5415 | } |
5416 | ||
8a4955ff | 5417 | mutex_lock(&ctx->uring_lock); |
1d7bb1d5 | 5418 | ret = io_submit_sqes(ctx, to_submit, NULL, -1, &cur_mm, true); |
8a4955ff | 5419 | mutex_unlock(&ctx->uring_lock); |
bdcd3eab | 5420 | timeout = jiffies + ctx->sq_thread_idle; |
6c271ce2 JA |
5421 | } |
5422 | ||
b41e9852 JA |
5423 | if (current->task_works) |
5424 | task_work_run(); | |
5425 | ||
6c271ce2 JA |
5426 | set_fs(old_fs); |
5427 | if (cur_mm) { | |
5428 | unuse_mm(cur_mm); | |
5429 | mmput(cur_mm); | |
5430 | } | |
181e448d | 5431 | revert_creds(old_cred); |
06058632 | 5432 | |
2bbcd6d3 | 5433 | kthread_parkme(); |
06058632 | 5434 | |
6c271ce2 JA |
5435 | return 0; |
5436 | } | |
5437 | ||
bda52162 JA |
5438 | struct io_wait_queue { |
5439 | struct wait_queue_entry wq; | |
5440 | struct io_ring_ctx *ctx; | |
5441 | unsigned to_wait; | |
5442 | unsigned nr_timeouts; | |
5443 | }; | |
5444 | ||
1d7bb1d5 | 5445 | static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush) |
bda52162 JA |
5446 | { |
5447 | struct io_ring_ctx *ctx = iowq->ctx; | |
5448 | ||
5449 | /* | |
d195a66e | 5450 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
5451 | * started waiting. For timeouts, we always want to return to userspace, |
5452 | * regardless of event count. | |
5453 | */ | |
1d7bb1d5 | 5454 | return io_cqring_events(ctx, noflush) >= iowq->to_wait || |
bda52162 JA |
5455 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
5456 | } | |
5457 | ||
5458 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
5459 | int wake_flags, void *key) | |
5460 | { | |
5461 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
5462 | wq); | |
5463 | ||
1d7bb1d5 JA |
5464 | /* use noflush == true, as we can't safely rely on locking context */ |
5465 | if (!io_should_wake(iowq, true)) | |
bda52162 JA |
5466 | return -1; |
5467 | ||
5468 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
5469 | } | |
5470 | ||
2b188cc1 JA |
5471 | /* |
5472 | * Wait until events become available, if we don't already have some. The | |
5473 | * application must reap them itself, as they reside on the shared cq ring. | |
5474 | */ | |
5475 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
5476 | const sigset_t __user *sig, size_t sigsz) | |
5477 | { | |
bda52162 JA |
5478 | struct io_wait_queue iowq = { |
5479 | .wq = { | |
5480 | .private = current, | |
5481 | .func = io_wake_function, | |
5482 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
5483 | }, | |
5484 | .ctx = ctx, | |
5485 | .to_wait = min_events, | |
5486 | }; | |
75b28aff | 5487 | struct io_rings *rings = ctx->rings; |
e9ffa5c2 | 5488 | int ret = 0; |
2b188cc1 | 5489 | |
b41e9852 JA |
5490 | do { |
5491 | if (io_cqring_events(ctx, false) >= min_events) | |
5492 | return 0; | |
5493 | if (!current->task_works) | |
5494 | break; | |
5495 | task_work_run(); | |
5496 | } while (1); | |
2b188cc1 JA |
5497 | |
5498 | if (sig) { | |
9e75ad5d AB |
5499 | #ifdef CONFIG_COMPAT |
5500 | if (in_compat_syscall()) | |
5501 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 5502 | sigsz); |
9e75ad5d AB |
5503 | else |
5504 | #endif | |
b772434b | 5505 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 5506 | |
2b188cc1 JA |
5507 | if (ret) |
5508 | return ret; | |
5509 | } | |
5510 | ||
bda52162 | 5511 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 5512 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 JA |
5513 | do { |
5514 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, | |
5515 | TASK_INTERRUPTIBLE); | |
b41e9852 JA |
5516 | if (current->task_works) |
5517 | task_work_run(); | |
1d7bb1d5 | 5518 | if (io_should_wake(&iowq, false)) |
bda52162 JA |
5519 | break; |
5520 | schedule(); | |
5521 | if (signal_pending(current)) { | |
e9ffa5c2 | 5522 | ret = -EINTR; |
bda52162 JA |
5523 | break; |
5524 | } | |
5525 | } while (1); | |
5526 | finish_wait(&ctx->wait, &iowq.wq); | |
5527 | ||
e9ffa5c2 | 5528 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 5529 | |
75b28aff | 5530 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
5531 | } |
5532 | ||
6b06314c JA |
5533 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
5534 | { | |
5535 | #if defined(CONFIG_UNIX) | |
5536 | if (ctx->ring_sock) { | |
5537 | struct sock *sock = ctx->ring_sock->sk; | |
5538 | struct sk_buff *skb; | |
5539 | ||
5540 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
5541 | kfree_skb(skb); | |
5542 | } | |
5543 | #else | |
5544 | int i; | |
5545 | ||
65e19f54 JA |
5546 | for (i = 0; i < ctx->nr_user_files; i++) { |
5547 | struct file *file; | |
5548 | ||
5549 | file = io_file_from_index(ctx, i); | |
5550 | if (file) | |
5551 | fput(file); | |
5552 | } | |
6b06314c JA |
5553 | #endif |
5554 | } | |
5555 | ||
05f3fb3c JA |
5556 | static void io_file_ref_kill(struct percpu_ref *ref) |
5557 | { | |
5558 | struct fixed_file_data *data; | |
5559 | ||
5560 | data = container_of(ref, struct fixed_file_data, refs); | |
5561 | complete(&data->done); | |
5562 | } | |
5563 | ||
6b06314c JA |
5564 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
5565 | { | |
05f3fb3c | 5566 | struct fixed_file_data *data = ctx->file_data; |
65e19f54 JA |
5567 | unsigned nr_tables, i; |
5568 | ||
05f3fb3c | 5569 | if (!data) |
6b06314c JA |
5570 | return -ENXIO; |
5571 | ||
05f3fb3c | 5572 | percpu_ref_kill_and_confirm(&data->refs, io_file_ref_kill); |
e46a7950 | 5573 | flush_work(&data->ref_work); |
2faf852d JA |
5574 | wait_for_completion(&data->done); |
5575 | io_ring_file_ref_flush(data); | |
05f3fb3c JA |
5576 | percpu_ref_exit(&data->refs); |
5577 | ||
6b06314c | 5578 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
5579 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
5580 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c JA |
5581 | kfree(data->table[i].files); |
5582 | kfree(data->table); | |
5583 | kfree(data); | |
5584 | ctx->file_data = NULL; | |
6b06314c JA |
5585 | ctx->nr_user_files = 0; |
5586 | return 0; | |
5587 | } | |
5588 | ||
6c271ce2 JA |
5589 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
5590 | { | |
5591 | if (ctx->sqo_thread) { | |
206aefde | 5592 | wait_for_completion(&ctx->completions[1]); |
2bbcd6d3 RP |
5593 | /* |
5594 | * The park is a bit of a work-around, without it we get | |
5595 | * warning spews on shutdown with SQPOLL set and affinity | |
5596 | * set to a single CPU. | |
5597 | */ | |
06058632 | 5598 | kthread_park(ctx->sqo_thread); |
6c271ce2 JA |
5599 | kthread_stop(ctx->sqo_thread); |
5600 | ctx->sqo_thread = NULL; | |
5601 | } | |
5602 | } | |
5603 | ||
6b06314c JA |
5604 | static void io_finish_async(struct io_ring_ctx *ctx) |
5605 | { | |
6c271ce2 JA |
5606 | io_sq_thread_stop(ctx); |
5607 | ||
561fb04a JA |
5608 | if (ctx->io_wq) { |
5609 | io_wq_destroy(ctx->io_wq); | |
5610 | ctx->io_wq = NULL; | |
6b06314c JA |
5611 | } |
5612 | } | |
5613 | ||
5614 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
5615 | /* |
5616 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
5617 | * the io_uring can be safely unregistered on process exit, even if we have | |
5618 | * loops in the file referencing. | |
5619 | */ | |
5620 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
5621 | { | |
5622 | struct sock *sk = ctx->ring_sock->sk; | |
5623 | struct scm_fp_list *fpl; | |
5624 | struct sk_buff *skb; | |
08a45173 | 5625 | int i, nr_files; |
6b06314c JA |
5626 | |
5627 | if (!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) { | |
5628 | unsigned long inflight = ctx->user->unix_inflight + nr; | |
5629 | ||
5630 | if (inflight > task_rlimit(current, RLIMIT_NOFILE)) | |
5631 | return -EMFILE; | |
5632 | } | |
5633 | ||
5634 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); | |
5635 | if (!fpl) | |
5636 | return -ENOMEM; | |
5637 | ||
5638 | skb = alloc_skb(0, GFP_KERNEL); | |
5639 | if (!skb) { | |
5640 | kfree(fpl); | |
5641 | return -ENOMEM; | |
5642 | } | |
5643 | ||
5644 | skb->sk = sk; | |
6b06314c | 5645 | |
08a45173 | 5646 | nr_files = 0; |
6b06314c JA |
5647 | fpl->user = get_uid(ctx->user); |
5648 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
5649 | struct file *file = io_file_from_index(ctx, i + offset); |
5650 | ||
5651 | if (!file) | |
08a45173 | 5652 | continue; |
65e19f54 | 5653 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
5654 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
5655 | nr_files++; | |
6b06314c JA |
5656 | } |
5657 | ||
08a45173 JA |
5658 | if (nr_files) { |
5659 | fpl->max = SCM_MAX_FD; | |
5660 | fpl->count = nr_files; | |
5661 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 5662 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
5663 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
5664 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 5665 | |
08a45173 JA |
5666 | for (i = 0; i < nr_files; i++) |
5667 | fput(fpl->fp[i]); | |
5668 | } else { | |
5669 | kfree_skb(skb); | |
5670 | kfree(fpl); | |
5671 | } | |
6b06314c JA |
5672 | |
5673 | return 0; | |
5674 | } | |
5675 | ||
5676 | /* | |
5677 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
5678 | * causes regular reference counting to break down. We rely on the UNIX | |
5679 | * garbage collection to take care of this problem for us. | |
5680 | */ | |
5681 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
5682 | { | |
5683 | unsigned left, total; | |
5684 | int ret = 0; | |
5685 | ||
5686 | total = 0; | |
5687 | left = ctx->nr_user_files; | |
5688 | while (left) { | |
5689 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
5690 | |
5691 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
5692 | if (ret) | |
5693 | break; | |
5694 | left -= this_files; | |
5695 | total += this_files; | |
5696 | } | |
5697 | ||
5698 | if (!ret) | |
5699 | return 0; | |
5700 | ||
5701 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
5702 | struct file *file = io_file_from_index(ctx, total); |
5703 | ||
5704 | if (file) | |
5705 | fput(file); | |
6b06314c JA |
5706 | total++; |
5707 | } | |
5708 | ||
5709 | return ret; | |
5710 | } | |
5711 | #else | |
5712 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
5713 | { | |
5714 | return 0; | |
5715 | } | |
5716 | #endif | |
5717 | ||
65e19f54 JA |
5718 | static int io_sqe_alloc_file_tables(struct io_ring_ctx *ctx, unsigned nr_tables, |
5719 | unsigned nr_files) | |
5720 | { | |
5721 | int i; | |
5722 | ||
5723 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 5724 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
5725 | unsigned this_files; |
5726 | ||
5727 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
5728 | table->files = kcalloc(this_files, sizeof(struct file *), | |
5729 | GFP_KERNEL); | |
5730 | if (!table->files) | |
5731 | break; | |
5732 | nr_files -= this_files; | |
5733 | } | |
5734 | ||
5735 | if (i == nr_tables) | |
5736 | return 0; | |
5737 | ||
5738 | for (i = 0; i < nr_tables; i++) { | |
05f3fb3c | 5739 | struct fixed_file_table *table = &ctx->file_data->table[i]; |
65e19f54 JA |
5740 | kfree(table->files); |
5741 | } | |
5742 | return 1; | |
5743 | } | |
5744 | ||
05f3fb3c JA |
5745 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file) |
5746 | { | |
5747 | #if defined(CONFIG_UNIX) | |
5748 | struct sock *sock = ctx->ring_sock->sk; | |
5749 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
5750 | struct sk_buff *skb; | |
5751 | int i; | |
5752 | ||
5753 | __skb_queue_head_init(&list); | |
5754 | ||
5755 | /* | |
5756 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
5757 | * remove this entry and rearrange the file array. | |
5758 | */ | |
5759 | skb = skb_dequeue(head); | |
5760 | while (skb) { | |
5761 | struct scm_fp_list *fp; | |
5762 | ||
5763 | fp = UNIXCB(skb).fp; | |
5764 | for (i = 0; i < fp->count; i++) { | |
5765 | int left; | |
5766 | ||
5767 | if (fp->fp[i] != file) | |
5768 | continue; | |
5769 | ||
5770 | unix_notinflight(fp->user, fp->fp[i]); | |
5771 | left = fp->count - 1 - i; | |
5772 | if (left) { | |
5773 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
5774 | left * sizeof(struct file *)); | |
5775 | } | |
5776 | fp->count--; | |
5777 | if (!fp->count) { | |
5778 | kfree_skb(skb); | |
5779 | skb = NULL; | |
5780 | } else { | |
5781 | __skb_queue_tail(&list, skb); | |
5782 | } | |
5783 | fput(file); | |
5784 | file = NULL; | |
5785 | break; | |
5786 | } | |
5787 | ||
5788 | if (!file) | |
5789 | break; | |
5790 | ||
5791 | __skb_queue_tail(&list, skb); | |
5792 | ||
5793 | skb = skb_dequeue(head); | |
5794 | } | |
5795 | ||
5796 | if (skb_peek(&list)) { | |
5797 | spin_lock_irq(&head->lock); | |
5798 | while ((skb = __skb_dequeue(&list)) != NULL) | |
5799 | __skb_queue_tail(head, skb); | |
5800 | spin_unlock_irq(&head->lock); | |
5801 | } | |
5802 | #else | |
5803 | fput(file); | |
5804 | #endif | |
5805 | } | |
5806 | ||
5807 | struct io_file_put { | |
5808 | struct llist_node llist; | |
5809 | struct file *file; | |
5810 | struct completion *done; | |
5811 | }; | |
5812 | ||
2faf852d | 5813 | static void io_ring_file_ref_flush(struct fixed_file_data *data) |
65e19f54 | 5814 | { |
05f3fb3c | 5815 | struct io_file_put *pfile, *tmp; |
05f3fb3c | 5816 | struct llist_node *node; |
65e19f54 | 5817 | |
05f3fb3c JA |
5818 | while ((node = llist_del_all(&data->put_llist)) != NULL) { |
5819 | llist_for_each_entry_safe(pfile, tmp, node, llist) { | |
5820 | io_ring_file_put(data->ctx, pfile->file); | |
5821 | if (pfile->done) | |
5822 | complete(pfile->done); | |
5823 | else | |
5824 | kfree(pfile); | |
5825 | } | |
65e19f54 | 5826 | } |
2faf852d | 5827 | } |
65e19f54 | 5828 | |
2faf852d JA |
5829 | static void io_ring_file_ref_switch(struct work_struct *work) |
5830 | { | |
5831 | struct fixed_file_data *data; | |
65e19f54 | 5832 | |
2faf852d JA |
5833 | data = container_of(work, struct fixed_file_data, ref_work); |
5834 | io_ring_file_ref_flush(data); | |
05f3fb3c JA |
5835 | percpu_ref_switch_to_percpu(&data->refs); |
5836 | } | |
65e19f54 | 5837 | |
05f3fb3c JA |
5838 | static void io_file_data_ref_zero(struct percpu_ref *ref) |
5839 | { | |
5840 | struct fixed_file_data *data; | |
5841 | ||
5842 | data = container_of(ref, struct fixed_file_data, refs); | |
5843 | ||
2faf852d JA |
5844 | /* |
5845 | * We can't safely switch from inside this context, punt to wq. If | |
5846 | * the table ref is going away, the table is being unregistered. | |
5847 | * Don't queue up the async work for that case, the caller will | |
5848 | * handle it. | |
5849 | */ | |
5850 | if (!percpu_ref_is_dying(&data->refs)) | |
5851 | queue_work(system_wq, &data->ref_work); | |
65e19f54 JA |
5852 | } |
5853 | ||
6b06314c JA |
5854 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
5855 | unsigned nr_args) | |
5856 | { | |
5857 | __s32 __user *fds = (__s32 __user *) arg; | |
65e19f54 | 5858 | unsigned nr_tables; |
05f3fb3c | 5859 | struct file *file; |
6b06314c JA |
5860 | int fd, ret = 0; |
5861 | unsigned i; | |
5862 | ||
05f3fb3c | 5863 | if (ctx->file_data) |
6b06314c JA |
5864 | return -EBUSY; |
5865 | if (!nr_args) | |
5866 | return -EINVAL; | |
5867 | if (nr_args > IORING_MAX_FIXED_FILES) | |
5868 | return -EMFILE; | |
5869 | ||
05f3fb3c JA |
5870 | ctx->file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL); |
5871 | if (!ctx->file_data) | |
5872 | return -ENOMEM; | |
5873 | ctx->file_data->ctx = ctx; | |
5874 | init_completion(&ctx->file_data->done); | |
5875 | ||
65e19f54 | 5876 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
05f3fb3c JA |
5877 | ctx->file_data->table = kcalloc(nr_tables, |
5878 | sizeof(struct fixed_file_table), | |
65e19f54 | 5879 | GFP_KERNEL); |
05f3fb3c JA |
5880 | if (!ctx->file_data->table) { |
5881 | kfree(ctx->file_data); | |
5882 | ctx->file_data = NULL; | |
6b06314c | 5883 | return -ENOMEM; |
05f3fb3c JA |
5884 | } |
5885 | ||
5886 | if (percpu_ref_init(&ctx->file_data->refs, io_file_data_ref_zero, | |
5887 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { | |
5888 | kfree(ctx->file_data->table); | |
5889 | kfree(ctx->file_data); | |
5890 | ctx->file_data = NULL; | |
6b06314c | 5891 | return -ENOMEM; |
05f3fb3c JA |
5892 | } |
5893 | ctx->file_data->put_llist.first = NULL; | |
5894 | INIT_WORK(&ctx->file_data->ref_work, io_ring_file_ref_switch); | |
6b06314c | 5895 | |
65e19f54 | 5896 | if (io_sqe_alloc_file_tables(ctx, nr_tables, nr_args)) { |
05f3fb3c JA |
5897 | percpu_ref_exit(&ctx->file_data->refs); |
5898 | kfree(ctx->file_data->table); | |
5899 | kfree(ctx->file_data); | |
5900 | ctx->file_data = NULL; | |
65e19f54 JA |
5901 | return -ENOMEM; |
5902 | } | |
5903 | ||
08a45173 | 5904 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
65e19f54 JA |
5905 | struct fixed_file_table *table; |
5906 | unsigned index; | |
5907 | ||
6b06314c JA |
5908 | ret = -EFAULT; |
5909 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) | |
5910 | break; | |
08a45173 JA |
5911 | /* allow sparse sets */ |
5912 | if (fd == -1) { | |
5913 | ret = 0; | |
5914 | continue; | |
5915 | } | |
6b06314c | 5916 | |
05f3fb3c | 5917 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
65e19f54 | 5918 | index = i & IORING_FILE_TABLE_MASK; |
05f3fb3c | 5919 | file = fget(fd); |
6b06314c JA |
5920 | |
5921 | ret = -EBADF; | |
05f3fb3c | 5922 | if (!file) |
6b06314c | 5923 | break; |
05f3fb3c | 5924 | |
6b06314c JA |
5925 | /* |
5926 | * Don't allow io_uring instances to be registered. If UNIX | |
5927 | * isn't enabled, then this causes a reference cycle and this | |
5928 | * instance can never get freed. If UNIX is enabled we'll | |
5929 | * handle it just fine, but there's still no point in allowing | |
5930 | * a ring fd as it doesn't support regular read/write anyway. | |
5931 | */ | |
05f3fb3c JA |
5932 | if (file->f_op == &io_uring_fops) { |
5933 | fput(file); | |
6b06314c JA |
5934 | break; |
5935 | } | |
6b06314c | 5936 | ret = 0; |
05f3fb3c | 5937 | table->files[index] = file; |
6b06314c JA |
5938 | } |
5939 | ||
5940 | if (ret) { | |
65e19f54 | 5941 | for (i = 0; i < ctx->nr_user_files; i++) { |
65e19f54 JA |
5942 | file = io_file_from_index(ctx, i); |
5943 | if (file) | |
5944 | fput(file); | |
5945 | } | |
5946 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 5947 | kfree(ctx->file_data->table[i].files); |
6b06314c | 5948 | |
05f3fb3c JA |
5949 | kfree(ctx->file_data->table); |
5950 | kfree(ctx->file_data); | |
5951 | ctx->file_data = NULL; | |
6b06314c JA |
5952 | ctx->nr_user_files = 0; |
5953 | return ret; | |
5954 | } | |
5955 | ||
5956 | ret = io_sqe_files_scm(ctx); | |
5957 | if (ret) | |
5958 | io_sqe_files_unregister(ctx); | |
5959 | ||
5960 | return ret; | |
5961 | } | |
5962 | ||
c3a31e60 JA |
5963 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
5964 | int index) | |
5965 | { | |
5966 | #if defined(CONFIG_UNIX) | |
5967 | struct sock *sock = ctx->ring_sock->sk; | |
5968 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
5969 | struct sk_buff *skb; | |
5970 | ||
5971 | /* | |
5972 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
5973 | * file set. If there's no room, fall back to allocating a new skb | |
5974 | * and filling it in. | |
5975 | */ | |
5976 | spin_lock_irq(&head->lock); | |
5977 | skb = skb_peek(head); | |
5978 | if (skb) { | |
5979 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
5980 | ||
5981 | if (fpl->count < SCM_MAX_FD) { | |
5982 | __skb_unlink(skb, head); | |
5983 | spin_unlock_irq(&head->lock); | |
5984 | fpl->fp[fpl->count] = get_file(file); | |
5985 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
5986 | fpl->count++; | |
5987 | spin_lock_irq(&head->lock); | |
5988 | __skb_queue_head(head, skb); | |
5989 | } else { | |
5990 | skb = NULL; | |
5991 | } | |
5992 | } | |
5993 | spin_unlock_irq(&head->lock); | |
5994 | ||
5995 | if (skb) { | |
5996 | fput(file); | |
5997 | return 0; | |
5998 | } | |
5999 | ||
6000 | return __io_sqe_files_scm(ctx, 1, index); | |
6001 | #else | |
6002 | return 0; | |
6003 | #endif | |
6004 | } | |
6005 | ||
05f3fb3c | 6006 | static void io_atomic_switch(struct percpu_ref *ref) |
c3a31e60 | 6007 | { |
05f3fb3c JA |
6008 | struct fixed_file_data *data; |
6009 | ||
dd3db2a3 JA |
6010 | /* |
6011 | * Juggle reference to ensure we hit zero, if needed, so we can | |
6012 | * switch back to percpu mode | |
6013 | */ | |
05f3fb3c | 6014 | data = container_of(ref, struct fixed_file_data, refs); |
dd3db2a3 JA |
6015 | percpu_ref_put(&data->refs); |
6016 | percpu_ref_get(&data->refs); | |
05f3fb3c JA |
6017 | } |
6018 | ||
6019 | static bool io_queue_file_removal(struct fixed_file_data *data, | |
6020 | struct file *file) | |
6021 | { | |
6022 | struct io_file_put *pfile, pfile_stack; | |
6023 | DECLARE_COMPLETION_ONSTACK(done); | |
6024 | ||
6025 | /* | |
6026 | * If we fail allocating the struct we need for doing async reomval | |
6027 | * of this file, just punt to sync and wait for it. | |
6028 | */ | |
6029 | pfile = kzalloc(sizeof(*pfile), GFP_KERNEL); | |
6030 | if (!pfile) { | |
6031 | pfile = &pfile_stack; | |
6032 | pfile->done = &done; | |
6033 | } | |
6034 | ||
6035 | pfile->file = file; | |
6036 | llist_add(&pfile->llist, &data->put_llist); | |
6037 | ||
6038 | if (pfile == &pfile_stack) { | |
dd3db2a3 | 6039 | percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch); |
05f3fb3c JA |
6040 | wait_for_completion(&done); |
6041 | flush_work(&data->ref_work); | |
6042 | return false; | |
6043 | } | |
6044 | ||
6045 | return true; | |
6046 | } | |
6047 | ||
6048 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
6049 | struct io_uring_files_update *up, | |
6050 | unsigned nr_args) | |
6051 | { | |
6052 | struct fixed_file_data *data = ctx->file_data; | |
6053 | bool ref_switch = false; | |
6054 | struct file *file; | |
c3a31e60 JA |
6055 | __s32 __user *fds; |
6056 | int fd, i, err; | |
6057 | __u32 done; | |
6058 | ||
05f3fb3c | 6059 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
6060 | return -EOVERFLOW; |
6061 | if (done > ctx->nr_user_files) | |
6062 | return -EINVAL; | |
6063 | ||
6064 | done = 0; | |
05f3fb3c | 6065 | fds = u64_to_user_ptr(up->fds); |
c3a31e60 | 6066 | while (nr_args) { |
65e19f54 JA |
6067 | struct fixed_file_table *table; |
6068 | unsigned index; | |
6069 | ||
c3a31e60 JA |
6070 | err = 0; |
6071 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
6072 | err = -EFAULT; | |
6073 | break; | |
6074 | } | |
05f3fb3c JA |
6075 | i = array_index_nospec(up->offset, ctx->nr_user_files); |
6076 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
65e19f54 JA |
6077 | index = i & IORING_FILE_TABLE_MASK; |
6078 | if (table->files[index]) { | |
05f3fb3c | 6079 | file = io_file_from_index(ctx, index); |
65e19f54 | 6080 | table->files[index] = NULL; |
05f3fb3c JA |
6081 | if (io_queue_file_removal(data, file)) |
6082 | ref_switch = true; | |
c3a31e60 JA |
6083 | } |
6084 | if (fd != -1) { | |
c3a31e60 JA |
6085 | file = fget(fd); |
6086 | if (!file) { | |
6087 | err = -EBADF; | |
6088 | break; | |
6089 | } | |
6090 | /* | |
6091 | * Don't allow io_uring instances to be registered. If | |
6092 | * UNIX isn't enabled, then this causes a reference | |
6093 | * cycle and this instance can never get freed. If UNIX | |
6094 | * is enabled we'll handle it just fine, but there's | |
6095 | * still no point in allowing a ring fd as it doesn't | |
6096 | * support regular read/write anyway. | |
6097 | */ | |
6098 | if (file->f_op == &io_uring_fops) { | |
6099 | fput(file); | |
6100 | err = -EBADF; | |
6101 | break; | |
6102 | } | |
65e19f54 | 6103 | table->files[index] = file; |
c3a31e60 JA |
6104 | err = io_sqe_file_register(ctx, file, i); |
6105 | if (err) | |
6106 | break; | |
6107 | } | |
6108 | nr_args--; | |
6109 | done++; | |
05f3fb3c JA |
6110 | up->offset++; |
6111 | } | |
6112 | ||
dd3db2a3 | 6113 | if (ref_switch) |
05f3fb3c | 6114 | percpu_ref_switch_to_atomic(&data->refs, io_atomic_switch); |
c3a31e60 JA |
6115 | |
6116 | return done ? done : err; | |
6117 | } | |
05f3fb3c JA |
6118 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
6119 | unsigned nr_args) | |
6120 | { | |
6121 | struct io_uring_files_update up; | |
6122 | ||
6123 | if (!ctx->file_data) | |
6124 | return -ENXIO; | |
6125 | if (!nr_args) | |
6126 | return -EINVAL; | |
6127 | if (copy_from_user(&up, arg, sizeof(up))) | |
6128 | return -EFAULT; | |
6129 | if (up.resv) | |
6130 | return -EINVAL; | |
6131 | ||
6132 | return __io_sqe_files_update(ctx, &up, nr_args); | |
6133 | } | |
c3a31e60 | 6134 | |
7d723065 JA |
6135 | static void io_put_work(struct io_wq_work *work) |
6136 | { | |
6137 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6138 | ||
6139 | io_put_req(req); | |
6140 | } | |
6141 | ||
6142 | static void io_get_work(struct io_wq_work *work) | |
6143 | { | |
6144 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6145 | ||
6146 | refcount_inc(&req->refs); | |
6147 | } | |
6148 | ||
24369c2e PB |
6149 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
6150 | struct io_uring_params *p) | |
6151 | { | |
6152 | struct io_wq_data data; | |
6153 | struct fd f; | |
6154 | struct io_ring_ctx *ctx_attach; | |
6155 | unsigned int concurrency; | |
6156 | int ret = 0; | |
6157 | ||
6158 | data.user = ctx->user; | |
6159 | data.get_work = io_get_work; | |
6160 | data.put_work = io_put_work; | |
6161 | ||
6162 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
6163 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
6164 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
6165 | ||
6166 | ctx->io_wq = io_wq_create(concurrency, &data); | |
6167 | if (IS_ERR(ctx->io_wq)) { | |
6168 | ret = PTR_ERR(ctx->io_wq); | |
6169 | ctx->io_wq = NULL; | |
6170 | } | |
6171 | return ret; | |
6172 | } | |
6173 | ||
6174 | f = fdget(p->wq_fd); | |
6175 | if (!f.file) | |
6176 | return -EBADF; | |
6177 | ||
6178 | if (f.file->f_op != &io_uring_fops) { | |
6179 | ret = -EINVAL; | |
6180 | goto out_fput; | |
6181 | } | |
6182 | ||
6183 | ctx_attach = f.file->private_data; | |
6184 | /* @io_wq is protected by holding the fd */ | |
6185 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
6186 | ret = -EINVAL; | |
6187 | goto out_fput; | |
6188 | } | |
6189 | ||
6190 | ctx->io_wq = ctx_attach->io_wq; | |
6191 | out_fput: | |
6192 | fdput(f); | |
6193 | return ret; | |
6194 | } | |
6195 | ||
6c271ce2 JA |
6196 | static int io_sq_offload_start(struct io_ring_ctx *ctx, |
6197 | struct io_uring_params *p) | |
2b188cc1 JA |
6198 | { |
6199 | int ret; | |
6200 | ||
6c271ce2 | 6201 | init_waitqueue_head(&ctx->sqo_wait); |
2b188cc1 JA |
6202 | mmgrab(current->mm); |
6203 | ctx->sqo_mm = current->mm; | |
6204 | ||
6c271ce2 | 6205 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
3ec482d1 JA |
6206 | ret = -EPERM; |
6207 | if (!capable(CAP_SYS_ADMIN)) | |
6208 | goto err; | |
6209 | ||
917257da JA |
6210 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
6211 | if (!ctx->sq_thread_idle) | |
6212 | ctx->sq_thread_idle = HZ; | |
6213 | ||
6c271ce2 | 6214 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 6215 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 6216 | |
917257da | 6217 | ret = -EINVAL; |
44a9bd18 JA |
6218 | if (cpu >= nr_cpu_ids) |
6219 | goto err; | |
7889f44d | 6220 | if (!cpu_online(cpu)) |
917257da JA |
6221 | goto err; |
6222 | ||
6c271ce2 JA |
6223 | ctx->sqo_thread = kthread_create_on_cpu(io_sq_thread, |
6224 | ctx, cpu, | |
6225 | "io_uring-sq"); | |
6226 | } else { | |
6227 | ctx->sqo_thread = kthread_create(io_sq_thread, ctx, | |
6228 | "io_uring-sq"); | |
6229 | } | |
6230 | if (IS_ERR(ctx->sqo_thread)) { | |
6231 | ret = PTR_ERR(ctx->sqo_thread); | |
6232 | ctx->sqo_thread = NULL; | |
6233 | goto err; | |
6234 | } | |
6235 | wake_up_process(ctx->sqo_thread); | |
6236 | } else if (p->flags & IORING_SETUP_SQ_AFF) { | |
6237 | /* Can't have SQ_AFF without SQPOLL */ | |
6238 | ret = -EINVAL; | |
6239 | goto err; | |
6240 | } | |
6241 | ||
24369c2e PB |
6242 | ret = io_init_wq_offload(ctx, p); |
6243 | if (ret) | |
2b188cc1 | 6244 | goto err; |
2b188cc1 JA |
6245 | |
6246 | return 0; | |
6247 | err: | |
54a91f3b | 6248 | io_finish_async(ctx); |
2b188cc1 JA |
6249 | mmdrop(ctx->sqo_mm); |
6250 | ctx->sqo_mm = NULL; | |
6251 | return ret; | |
6252 | } | |
6253 | ||
6254 | static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) | |
6255 | { | |
6256 | atomic_long_sub(nr_pages, &user->locked_vm); | |
6257 | } | |
6258 | ||
6259 | static int io_account_mem(struct user_struct *user, unsigned long nr_pages) | |
6260 | { | |
6261 | unsigned long page_limit, cur_pages, new_pages; | |
6262 | ||
6263 | /* Don't allow more pages than we can safely lock */ | |
6264 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
6265 | ||
6266 | do { | |
6267 | cur_pages = atomic_long_read(&user->locked_vm); | |
6268 | new_pages = cur_pages + nr_pages; | |
6269 | if (new_pages > page_limit) | |
6270 | return -ENOMEM; | |
6271 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
6272 | new_pages) != cur_pages); | |
6273 | ||
6274 | return 0; | |
6275 | } | |
6276 | ||
6277 | static void io_mem_free(void *ptr) | |
6278 | { | |
52e04ef4 MR |
6279 | struct page *page; |
6280 | ||
6281 | if (!ptr) | |
6282 | return; | |
2b188cc1 | 6283 | |
52e04ef4 | 6284 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
6285 | if (put_page_testzero(page)) |
6286 | free_compound_page(page); | |
6287 | } | |
6288 | ||
6289 | static void *io_mem_alloc(size_t size) | |
6290 | { | |
6291 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
6292 | __GFP_NORETRY; | |
6293 | ||
6294 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
6295 | } | |
6296 | ||
75b28aff HV |
6297 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
6298 | size_t *sq_offset) | |
6299 | { | |
6300 | struct io_rings *rings; | |
6301 | size_t off, sq_array_size; | |
6302 | ||
6303 | off = struct_size(rings, cqes, cq_entries); | |
6304 | if (off == SIZE_MAX) | |
6305 | return SIZE_MAX; | |
6306 | ||
6307 | #ifdef CONFIG_SMP | |
6308 | off = ALIGN(off, SMP_CACHE_BYTES); | |
6309 | if (off == 0) | |
6310 | return SIZE_MAX; | |
6311 | #endif | |
6312 | ||
6313 | sq_array_size = array_size(sizeof(u32), sq_entries); | |
6314 | if (sq_array_size == SIZE_MAX) | |
6315 | return SIZE_MAX; | |
6316 | ||
6317 | if (check_add_overflow(off, sq_array_size, &off)) | |
6318 | return SIZE_MAX; | |
6319 | ||
6320 | if (sq_offset) | |
6321 | *sq_offset = off; | |
6322 | ||
6323 | return off; | |
6324 | } | |
6325 | ||
2b188cc1 JA |
6326 | static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries) |
6327 | { | |
75b28aff | 6328 | size_t pages; |
2b188cc1 | 6329 | |
75b28aff HV |
6330 | pages = (size_t)1 << get_order( |
6331 | rings_size(sq_entries, cq_entries, NULL)); | |
6332 | pages += (size_t)1 << get_order( | |
6333 | array_size(sizeof(struct io_uring_sqe), sq_entries)); | |
2b188cc1 | 6334 | |
75b28aff | 6335 | return pages; |
2b188cc1 JA |
6336 | } |
6337 | ||
edafccee JA |
6338 | static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx) |
6339 | { | |
6340 | int i, j; | |
6341 | ||
6342 | if (!ctx->user_bufs) | |
6343 | return -ENXIO; | |
6344 | ||
6345 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6346 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6347 | ||
6348 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 6349 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee JA |
6350 | |
6351 | if (ctx->account_mem) | |
6352 | io_unaccount_mem(ctx->user, imu->nr_bvecs); | |
d4ef6475 | 6353 | kvfree(imu->bvec); |
edafccee JA |
6354 | imu->nr_bvecs = 0; |
6355 | } | |
6356 | ||
6357 | kfree(ctx->user_bufs); | |
6358 | ctx->user_bufs = NULL; | |
6359 | ctx->nr_user_bufs = 0; | |
6360 | return 0; | |
6361 | } | |
6362 | ||
6363 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
6364 | void __user *arg, unsigned index) | |
6365 | { | |
6366 | struct iovec __user *src; | |
6367 | ||
6368 | #ifdef CONFIG_COMPAT | |
6369 | if (ctx->compat) { | |
6370 | struct compat_iovec __user *ciovs; | |
6371 | struct compat_iovec ciov; | |
6372 | ||
6373 | ciovs = (struct compat_iovec __user *) arg; | |
6374 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
6375 | return -EFAULT; | |
6376 | ||
d55e5f5b | 6377 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
6378 | dst->iov_len = ciov.iov_len; |
6379 | return 0; | |
6380 | } | |
6381 | #endif | |
6382 | src = (struct iovec __user *) arg; | |
6383 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
6384 | return -EFAULT; | |
6385 | return 0; | |
6386 | } | |
6387 | ||
6388 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg, | |
6389 | unsigned nr_args) | |
6390 | { | |
6391 | struct vm_area_struct **vmas = NULL; | |
6392 | struct page **pages = NULL; | |
6393 | int i, j, got_pages = 0; | |
6394 | int ret = -EINVAL; | |
6395 | ||
6396 | if (ctx->user_bufs) | |
6397 | return -EBUSY; | |
6398 | if (!nr_args || nr_args > UIO_MAXIOV) | |
6399 | return -EINVAL; | |
6400 | ||
6401 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
6402 | GFP_KERNEL); | |
6403 | if (!ctx->user_bufs) | |
6404 | return -ENOMEM; | |
6405 | ||
6406 | for (i = 0; i < nr_args; i++) { | |
6407 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
6408 | unsigned long off, start, end, ubuf; | |
6409 | int pret, nr_pages; | |
6410 | struct iovec iov; | |
6411 | size_t size; | |
6412 | ||
6413 | ret = io_copy_iov(ctx, &iov, arg, i); | |
6414 | if (ret) | |
a278682d | 6415 | goto err; |
edafccee JA |
6416 | |
6417 | /* | |
6418 | * Don't impose further limits on the size and buffer | |
6419 | * constraints here, we'll -EINVAL later when IO is | |
6420 | * submitted if they are wrong. | |
6421 | */ | |
6422 | ret = -EFAULT; | |
6423 | if (!iov.iov_base || !iov.iov_len) | |
6424 | goto err; | |
6425 | ||
6426 | /* arbitrary limit, but we need something */ | |
6427 | if (iov.iov_len > SZ_1G) | |
6428 | goto err; | |
6429 | ||
6430 | ubuf = (unsigned long) iov.iov_base; | |
6431 | end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
6432 | start = ubuf >> PAGE_SHIFT; | |
6433 | nr_pages = end - start; | |
6434 | ||
6435 | if (ctx->account_mem) { | |
6436 | ret = io_account_mem(ctx->user, nr_pages); | |
6437 | if (ret) | |
6438 | goto err; | |
6439 | } | |
6440 | ||
6441 | ret = 0; | |
6442 | if (!pages || nr_pages > got_pages) { | |
6443 | kfree(vmas); | |
6444 | kfree(pages); | |
d4ef6475 | 6445 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), |
edafccee | 6446 | GFP_KERNEL); |
d4ef6475 | 6447 | vmas = kvmalloc_array(nr_pages, |
edafccee JA |
6448 | sizeof(struct vm_area_struct *), |
6449 | GFP_KERNEL); | |
6450 | if (!pages || !vmas) { | |
6451 | ret = -ENOMEM; | |
6452 | if (ctx->account_mem) | |
6453 | io_unaccount_mem(ctx->user, nr_pages); | |
6454 | goto err; | |
6455 | } | |
6456 | got_pages = nr_pages; | |
6457 | } | |
6458 | ||
d4ef6475 | 6459 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
edafccee JA |
6460 | GFP_KERNEL); |
6461 | ret = -ENOMEM; | |
6462 | if (!imu->bvec) { | |
6463 | if (ctx->account_mem) | |
6464 | io_unaccount_mem(ctx->user, nr_pages); | |
6465 | goto err; | |
6466 | } | |
6467 | ||
6468 | ret = 0; | |
6469 | down_read(¤t->mm->mmap_sem); | |
2113b05d | 6470 | pret = pin_user_pages(ubuf, nr_pages, |
932f4a63 IW |
6471 | FOLL_WRITE | FOLL_LONGTERM, |
6472 | pages, vmas); | |
edafccee JA |
6473 | if (pret == nr_pages) { |
6474 | /* don't support file backed memory */ | |
6475 | for (j = 0; j < nr_pages; j++) { | |
6476 | struct vm_area_struct *vma = vmas[j]; | |
6477 | ||
6478 | if (vma->vm_file && | |
6479 | !is_file_hugepages(vma->vm_file)) { | |
6480 | ret = -EOPNOTSUPP; | |
6481 | break; | |
6482 | } | |
6483 | } | |
6484 | } else { | |
6485 | ret = pret < 0 ? pret : -EFAULT; | |
6486 | } | |
6487 | up_read(¤t->mm->mmap_sem); | |
6488 | if (ret) { | |
6489 | /* | |
6490 | * if we did partial map, or found file backed vmas, | |
6491 | * release any pages we did get | |
6492 | */ | |
27c4d3a3 | 6493 | if (pret > 0) |
f1f6a7dd | 6494 | unpin_user_pages(pages, pret); |
edafccee JA |
6495 | if (ctx->account_mem) |
6496 | io_unaccount_mem(ctx->user, nr_pages); | |
d4ef6475 | 6497 | kvfree(imu->bvec); |
edafccee JA |
6498 | goto err; |
6499 | } | |
6500 | ||
6501 | off = ubuf & ~PAGE_MASK; | |
6502 | size = iov.iov_len; | |
6503 | for (j = 0; j < nr_pages; j++) { | |
6504 | size_t vec_len; | |
6505 | ||
6506 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
6507 | imu->bvec[j].bv_page = pages[j]; | |
6508 | imu->bvec[j].bv_len = vec_len; | |
6509 | imu->bvec[j].bv_offset = off; | |
6510 | off = 0; | |
6511 | size -= vec_len; | |
6512 | } | |
6513 | /* store original address for later verification */ | |
6514 | imu->ubuf = ubuf; | |
6515 | imu->len = iov.iov_len; | |
6516 | imu->nr_bvecs = nr_pages; | |
6517 | ||
6518 | ctx->nr_user_bufs++; | |
6519 | } | |
d4ef6475 MR |
6520 | kvfree(pages); |
6521 | kvfree(vmas); | |
edafccee JA |
6522 | return 0; |
6523 | err: | |
d4ef6475 MR |
6524 | kvfree(pages); |
6525 | kvfree(vmas); | |
edafccee JA |
6526 | io_sqe_buffer_unregister(ctx); |
6527 | return ret; | |
6528 | } | |
6529 | ||
9b402849 JA |
6530 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
6531 | { | |
6532 | __s32 __user *fds = arg; | |
6533 | int fd; | |
6534 | ||
6535 | if (ctx->cq_ev_fd) | |
6536 | return -EBUSY; | |
6537 | ||
6538 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
6539 | return -EFAULT; | |
6540 | ||
6541 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
6542 | if (IS_ERR(ctx->cq_ev_fd)) { | |
6543 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
6544 | ctx->cq_ev_fd = NULL; | |
6545 | return ret; | |
6546 | } | |
6547 | ||
6548 | return 0; | |
6549 | } | |
6550 | ||
6551 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
6552 | { | |
6553 | if (ctx->cq_ev_fd) { | |
6554 | eventfd_ctx_put(ctx->cq_ev_fd); | |
6555 | ctx->cq_ev_fd = NULL; | |
6556 | return 0; | |
6557 | } | |
6558 | ||
6559 | return -ENXIO; | |
6560 | } | |
6561 | ||
2b188cc1 JA |
6562 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
6563 | { | |
6b06314c | 6564 | io_finish_async(ctx); |
2b188cc1 JA |
6565 | if (ctx->sqo_mm) |
6566 | mmdrop(ctx->sqo_mm); | |
def596e9 JA |
6567 | |
6568 | io_iopoll_reap_events(ctx); | |
edafccee | 6569 | io_sqe_buffer_unregister(ctx); |
6b06314c | 6570 | io_sqe_files_unregister(ctx); |
9b402849 | 6571 | io_eventfd_unregister(ctx); |
41726c9a | 6572 | idr_destroy(&ctx->personality_idr); |
def596e9 | 6573 | |
2b188cc1 | 6574 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
6575 | if (ctx->ring_sock) { |
6576 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 6577 | sock_release(ctx->ring_sock); |
355e8d26 | 6578 | } |
2b188cc1 JA |
6579 | #endif |
6580 | ||
75b28aff | 6581 | io_mem_free(ctx->rings); |
2b188cc1 | 6582 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
6583 | |
6584 | percpu_ref_exit(&ctx->refs); | |
6585 | if (ctx->account_mem) | |
6586 | io_unaccount_mem(ctx->user, | |
6587 | ring_pages(ctx->sq_entries, ctx->cq_entries)); | |
6588 | free_uid(ctx->user); | |
181e448d | 6589 | put_cred(ctx->creds); |
206aefde | 6590 | kfree(ctx->completions); |
78076bb6 | 6591 | kfree(ctx->cancel_hash); |
0ddf92e8 | 6592 | kmem_cache_free(req_cachep, ctx->fallback_req); |
2b188cc1 JA |
6593 | kfree(ctx); |
6594 | } | |
6595 | ||
6596 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
6597 | { | |
6598 | struct io_ring_ctx *ctx = file->private_data; | |
6599 | __poll_t mask = 0; | |
6600 | ||
6601 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
6602 | /* |
6603 | * synchronizes with barrier from wq_has_sleeper call in | |
6604 | * io_commit_cqring | |
6605 | */ | |
2b188cc1 | 6606 | smp_rmb(); |
75b28aff HV |
6607 | if (READ_ONCE(ctx->rings->sq.tail) - ctx->cached_sq_head != |
6608 | ctx->rings->sq_ring_entries) | |
2b188cc1 | 6609 | mask |= EPOLLOUT | EPOLLWRNORM; |
63e5d81f | 6610 | if (io_cqring_events(ctx, false)) |
2b188cc1 JA |
6611 | mask |= EPOLLIN | EPOLLRDNORM; |
6612 | ||
6613 | return mask; | |
6614 | } | |
6615 | ||
6616 | static int io_uring_fasync(int fd, struct file *file, int on) | |
6617 | { | |
6618 | struct io_ring_ctx *ctx = file->private_data; | |
6619 | ||
6620 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
6621 | } | |
6622 | ||
071698e1 JA |
6623 | static int io_remove_personalities(int id, void *p, void *data) |
6624 | { | |
6625 | struct io_ring_ctx *ctx = data; | |
6626 | const struct cred *cred; | |
6627 | ||
6628 | cred = idr_remove(&ctx->personality_idr, id); | |
6629 | if (cred) | |
6630 | put_cred(cred); | |
6631 | return 0; | |
6632 | } | |
6633 | ||
2b188cc1 JA |
6634 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
6635 | { | |
6636 | mutex_lock(&ctx->uring_lock); | |
6637 | percpu_ref_kill(&ctx->refs); | |
6638 | mutex_unlock(&ctx->uring_lock); | |
6639 | ||
df069d80 JA |
6640 | /* |
6641 | * Wait for sq thread to idle, if we have one. It won't spin on new | |
6642 | * work after we've killed the ctx ref above. This is important to do | |
6643 | * before we cancel existing commands, as the thread could otherwise | |
6644 | * be queueing new work post that. If that's work we need to cancel, | |
6645 | * it could cause shutdown to hang. | |
6646 | */ | |
6647 | while (ctx->sqo_thread && !wq_has_sleeper(&ctx->sqo_wait)) | |
6648 | cpu_relax(); | |
6649 | ||
5262f567 | 6650 | io_kill_timeouts(ctx); |
221c5eb2 | 6651 | io_poll_remove_all(ctx); |
561fb04a JA |
6652 | |
6653 | if (ctx->io_wq) | |
6654 | io_wq_cancel_all(ctx->io_wq); | |
6655 | ||
def596e9 | 6656 | io_iopoll_reap_events(ctx); |
15dff286 JA |
6657 | /* if we failed setting up the ctx, we might not have any rings */ |
6658 | if (ctx->rings) | |
6659 | io_cqring_overflow_flush(ctx, true); | |
071698e1 | 6660 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
206aefde | 6661 | wait_for_completion(&ctx->completions[0]); |
2b188cc1 JA |
6662 | io_ring_ctx_free(ctx); |
6663 | } | |
6664 | ||
6665 | static int io_uring_release(struct inode *inode, struct file *file) | |
6666 | { | |
6667 | struct io_ring_ctx *ctx = file->private_data; | |
6668 | ||
6669 | file->private_data = NULL; | |
6670 | io_ring_ctx_wait_and_kill(ctx); | |
6671 | return 0; | |
6672 | } | |
6673 | ||
fcb323cc JA |
6674 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
6675 | struct files_struct *files) | |
6676 | { | |
6677 | struct io_kiocb *req; | |
6678 | DEFINE_WAIT(wait); | |
6679 | ||
6680 | while (!list_empty_careful(&ctx->inflight_list)) { | |
768134d4 | 6681 | struct io_kiocb *cancel_req = NULL; |
fcb323cc JA |
6682 | |
6683 | spin_lock_irq(&ctx->inflight_lock); | |
6684 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) { | |
768134d4 JA |
6685 | if (req->work.files != files) |
6686 | continue; | |
6687 | /* req is being completed, ignore */ | |
6688 | if (!refcount_inc_not_zero(&req->refs)) | |
6689 | continue; | |
6690 | cancel_req = req; | |
6691 | break; | |
fcb323cc | 6692 | } |
768134d4 | 6693 | if (cancel_req) |
fcb323cc | 6694 | prepare_to_wait(&ctx->inflight_wait, &wait, |
768134d4 | 6695 | TASK_UNINTERRUPTIBLE); |
fcb323cc JA |
6696 | spin_unlock_irq(&ctx->inflight_lock); |
6697 | ||
768134d4 JA |
6698 | /* We need to keep going until we don't find a matching req */ |
6699 | if (!cancel_req) | |
fcb323cc | 6700 | break; |
2f6d9b9d | 6701 | |
2ca10259 JA |
6702 | if (cancel_req->flags & REQ_F_OVERFLOW) { |
6703 | spin_lock_irq(&ctx->completion_lock); | |
6704 | list_del(&cancel_req->list); | |
6705 | cancel_req->flags &= ~REQ_F_OVERFLOW; | |
6706 | if (list_empty(&ctx->cq_overflow_list)) { | |
6707 | clear_bit(0, &ctx->sq_check_overflow); | |
6708 | clear_bit(0, &ctx->cq_check_overflow); | |
6709 | } | |
6710 | spin_unlock_irq(&ctx->completion_lock); | |
6711 | ||
6712 | WRITE_ONCE(ctx->rings->cq_overflow, | |
6713 | atomic_inc_return(&ctx->cached_cq_overflow)); | |
6714 | ||
6715 | /* | |
6716 | * Put inflight ref and overflow ref. If that's | |
6717 | * all we had, then we're done with this request. | |
6718 | */ | |
6719 | if (refcount_sub_and_test(2, &cancel_req->refs)) { | |
6720 | io_put_req(cancel_req); | |
6721 | continue; | |
6722 | } | |
6723 | } | |
6724 | ||
2f6d9b9d BL |
6725 | io_wq_cancel_work(ctx->io_wq, &cancel_req->work); |
6726 | io_put_req(cancel_req); | |
fcb323cc JA |
6727 | schedule(); |
6728 | } | |
768134d4 | 6729 | finish_wait(&ctx->inflight_wait, &wait); |
fcb323cc JA |
6730 | } |
6731 | ||
6732 | static int io_uring_flush(struct file *file, void *data) | |
6733 | { | |
6734 | struct io_ring_ctx *ctx = file->private_data; | |
6735 | ||
6736 | io_uring_cancel_files(ctx, data); | |
6ab23144 JA |
6737 | |
6738 | /* | |
6739 | * If the task is going away, cancel work it may have pending | |
6740 | */ | |
6741 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) | |
6742 | io_wq_cancel_pid(ctx->io_wq, task_pid_vnr(current)); | |
6743 | ||
fcb323cc JA |
6744 | return 0; |
6745 | } | |
6746 | ||
6c5c240e RP |
6747 | static void *io_uring_validate_mmap_request(struct file *file, |
6748 | loff_t pgoff, size_t sz) | |
2b188cc1 | 6749 | { |
2b188cc1 | 6750 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 6751 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
6752 | struct page *page; |
6753 | void *ptr; | |
6754 | ||
6755 | switch (offset) { | |
6756 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
6757 | case IORING_OFF_CQ_RING: |
6758 | ptr = ctx->rings; | |
2b188cc1 JA |
6759 | break; |
6760 | case IORING_OFF_SQES: | |
6761 | ptr = ctx->sq_sqes; | |
6762 | break; | |
2b188cc1 | 6763 | default: |
6c5c240e | 6764 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
6765 | } |
6766 | ||
6767 | page = virt_to_head_page(ptr); | |
a50b854e | 6768 | if (sz > page_size(page)) |
6c5c240e RP |
6769 | return ERR_PTR(-EINVAL); |
6770 | ||
6771 | return ptr; | |
6772 | } | |
6773 | ||
6774 | #ifdef CONFIG_MMU | |
6775 | ||
6776 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
6777 | { | |
6778 | size_t sz = vma->vm_end - vma->vm_start; | |
6779 | unsigned long pfn; | |
6780 | void *ptr; | |
6781 | ||
6782 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
6783 | if (IS_ERR(ptr)) | |
6784 | return PTR_ERR(ptr); | |
2b188cc1 JA |
6785 | |
6786 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
6787 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
6788 | } | |
6789 | ||
6c5c240e RP |
6790 | #else /* !CONFIG_MMU */ |
6791 | ||
6792 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
6793 | { | |
6794 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
6795 | } | |
6796 | ||
6797 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
6798 | { | |
6799 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
6800 | } | |
6801 | ||
6802 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
6803 | unsigned long addr, unsigned long len, | |
6804 | unsigned long pgoff, unsigned long flags) | |
6805 | { | |
6806 | void *ptr; | |
6807 | ||
6808 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
6809 | if (IS_ERR(ptr)) | |
6810 | return PTR_ERR(ptr); | |
6811 | ||
6812 | return (unsigned long) ptr; | |
6813 | } | |
6814 | ||
6815 | #endif /* !CONFIG_MMU */ | |
6816 | ||
2b188cc1 JA |
6817 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
6818 | u32, min_complete, u32, flags, const sigset_t __user *, sig, | |
6819 | size_t, sigsz) | |
6820 | { | |
6821 | struct io_ring_ctx *ctx; | |
6822 | long ret = -EBADF; | |
6823 | int submitted = 0; | |
6824 | struct fd f; | |
6825 | ||
b41e9852 JA |
6826 | if (current->task_works) |
6827 | task_work_run(); | |
6828 | ||
6c271ce2 | 6829 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP)) |
2b188cc1 JA |
6830 | return -EINVAL; |
6831 | ||
6832 | f = fdget(fd); | |
6833 | if (!f.file) | |
6834 | return -EBADF; | |
6835 | ||
6836 | ret = -EOPNOTSUPP; | |
6837 | if (f.file->f_op != &io_uring_fops) | |
6838 | goto out_fput; | |
6839 | ||
6840 | ret = -ENXIO; | |
6841 | ctx = f.file->private_data; | |
6842 | if (!percpu_ref_tryget(&ctx->refs)) | |
6843 | goto out_fput; | |
6844 | ||
6c271ce2 JA |
6845 | /* |
6846 | * For SQ polling, the thread will do all submissions and completions. | |
6847 | * Just return the requested submit count, and wake the thread if | |
6848 | * we were asked to. | |
6849 | */ | |
b2a9eada | 6850 | ret = 0; |
6c271ce2 | 6851 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
c1edbf5f JA |
6852 | if (!list_empty_careful(&ctx->cq_overflow_list)) |
6853 | io_cqring_overflow_flush(ctx, false); | |
6c271ce2 JA |
6854 | if (flags & IORING_ENTER_SQ_WAKEUP) |
6855 | wake_up(&ctx->sqo_wait); | |
6856 | submitted = to_submit; | |
b2a9eada | 6857 | } else if (to_submit) { |
ae9428ca | 6858 | struct mm_struct *cur_mm; |
2b188cc1 JA |
6859 | |
6860 | mutex_lock(&ctx->uring_lock); | |
ae9428ca PB |
6861 | /* already have mm, so io_submit_sqes() won't try to grab it */ |
6862 | cur_mm = ctx->sqo_mm; | |
6863 | submitted = io_submit_sqes(ctx, to_submit, f.file, fd, | |
6864 | &cur_mm, false); | |
2b188cc1 | 6865 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
6866 | |
6867 | if (submitted != to_submit) | |
6868 | goto out; | |
2b188cc1 JA |
6869 | } |
6870 | if (flags & IORING_ENTER_GETEVENTS) { | |
def596e9 JA |
6871 | unsigned nr_events = 0; |
6872 | ||
2b188cc1 JA |
6873 | min_complete = min(min_complete, ctx->cq_entries); |
6874 | ||
def596e9 | 6875 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 | 6876 | ret = io_iopoll_check(ctx, &nr_events, min_complete); |
def596e9 JA |
6877 | } else { |
6878 | ret = io_cqring_wait(ctx, min_complete, sig, sigsz); | |
6879 | } | |
2b188cc1 JA |
6880 | } |
6881 | ||
7c504e65 | 6882 | out: |
6805b32e | 6883 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
6884 | out_fput: |
6885 | fdput(f); | |
6886 | return submitted ? submitted : ret; | |
6887 | } | |
6888 | ||
bebdb65e | 6889 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
6890 | static int io_uring_show_cred(int id, void *p, void *data) |
6891 | { | |
6892 | const struct cred *cred = p; | |
6893 | struct seq_file *m = data; | |
6894 | struct user_namespace *uns = seq_user_ns(m); | |
6895 | struct group_info *gi; | |
6896 | kernel_cap_t cap; | |
6897 | unsigned __capi; | |
6898 | int g; | |
6899 | ||
6900 | seq_printf(m, "%5d\n", id); | |
6901 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
6902 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
6903 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
6904 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
6905 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
6906 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
6907 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
6908 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
6909 | seq_puts(m, "\n\tGroups:\t"); | |
6910 | gi = cred->group_info; | |
6911 | for (g = 0; g < gi->ngroups; g++) { | |
6912 | seq_put_decimal_ull(m, g ? " " : "", | |
6913 | from_kgid_munged(uns, gi->gid[g])); | |
6914 | } | |
6915 | seq_puts(m, "\n\tCapEff:\t"); | |
6916 | cap = cred->cap_effective; | |
6917 | CAP_FOR_EACH_U32(__capi) | |
6918 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
6919 | seq_putc(m, '\n'); | |
6920 | return 0; | |
6921 | } | |
6922 | ||
6923 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
6924 | { | |
6925 | int i; | |
6926 | ||
6927 | mutex_lock(&ctx->uring_lock); | |
6928 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); | |
6929 | for (i = 0; i < ctx->nr_user_files; i++) { | |
6930 | struct fixed_file_table *table; | |
6931 | struct file *f; | |
6932 | ||
6933 | table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
6934 | f = table->files[i & IORING_FILE_TABLE_MASK]; | |
6935 | if (f) | |
6936 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
6937 | else | |
6938 | seq_printf(m, "%5u: <none>\n", i); | |
6939 | } | |
6940 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
6941 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
6942 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; | |
6943 | ||
6944 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
6945 | (unsigned int) buf->len); | |
6946 | } | |
6947 | if (!idr_is_empty(&ctx->personality_idr)) { | |
6948 | seq_printf(m, "Personalities:\n"); | |
6949 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
6950 | } | |
d7718a9d JA |
6951 | seq_printf(m, "PollList:\n"); |
6952 | spin_lock_irq(&ctx->completion_lock); | |
6953 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
6954 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
6955 | struct io_kiocb *req; | |
6956 | ||
6957 | hlist_for_each_entry(req, list, hash_node) | |
6958 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
6959 | req->task->task_works != NULL); | |
6960 | } | |
6961 | spin_unlock_irq(&ctx->completion_lock); | |
87ce955b JA |
6962 | mutex_unlock(&ctx->uring_lock); |
6963 | } | |
6964 | ||
6965 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
6966 | { | |
6967 | struct io_ring_ctx *ctx = f->private_data; | |
6968 | ||
6969 | if (percpu_ref_tryget(&ctx->refs)) { | |
6970 | __io_uring_show_fdinfo(ctx, m); | |
6971 | percpu_ref_put(&ctx->refs); | |
6972 | } | |
6973 | } | |
bebdb65e | 6974 | #endif |
87ce955b | 6975 | |
2b188cc1 JA |
6976 | static const struct file_operations io_uring_fops = { |
6977 | .release = io_uring_release, | |
fcb323cc | 6978 | .flush = io_uring_flush, |
2b188cc1 | 6979 | .mmap = io_uring_mmap, |
6c5c240e RP |
6980 | #ifndef CONFIG_MMU |
6981 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
6982 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
6983 | #endif | |
2b188cc1 JA |
6984 | .poll = io_uring_poll, |
6985 | .fasync = io_uring_fasync, | |
bebdb65e | 6986 | #ifdef CONFIG_PROC_FS |
87ce955b | 6987 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 6988 | #endif |
2b188cc1 JA |
6989 | }; |
6990 | ||
6991 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
6992 | struct io_uring_params *p) | |
6993 | { | |
75b28aff HV |
6994 | struct io_rings *rings; |
6995 | size_t size, sq_array_offset; | |
2b188cc1 | 6996 | |
75b28aff HV |
6997 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
6998 | if (size == SIZE_MAX) | |
6999 | return -EOVERFLOW; | |
7000 | ||
7001 | rings = io_mem_alloc(size); | |
7002 | if (!rings) | |
2b188cc1 JA |
7003 | return -ENOMEM; |
7004 | ||
75b28aff HV |
7005 | ctx->rings = rings; |
7006 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
7007 | rings->sq_ring_mask = p->sq_entries - 1; | |
7008 | rings->cq_ring_mask = p->cq_entries - 1; | |
7009 | rings->sq_ring_entries = p->sq_entries; | |
7010 | rings->cq_ring_entries = p->cq_entries; | |
7011 | ctx->sq_mask = rings->sq_ring_mask; | |
7012 | ctx->cq_mask = rings->cq_ring_mask; | |
7013 | ctx->sq_entries = rings->sq_ring_entries; | |
7014 | ctx->cq_entries = rings->cq_ring_entries; | |
2b188cc1 JA |
7015 | |
7016 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
7017 | if (size == SIZE_MAX) { |
7018 | io_mem_free(ctx->rings); | |
7019 | ctx->rings = NULL; | |
2b188cc1 | 7020 | return -EOVERFLOW; |
eb065d30 | 7021 | } |
2b188cc1 JA |
7022 | |
7023 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
7024 | if (!ctx->sq_sqes) { |
7025 | io_mem_free(ctx->rings); | |
7026 | ctx->rings = NULL; | |
2b188cc1 | 7027 | return -ENOMEM; |
eb065d30 | 7028 | } |
2b188cc1 | 7029 | |
2b188cc1 JA |
7030 | return 0; |
7031 | } | |
7032 | ||
7033 | /* | |
7034 | * Allocate an anonymous fd, this is what constitutes the application | |
7035 | * visible backing of an io_uring instance. The application mmaps this | |
7036 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
7037 | * we have to tie this fd to a socket for file garbage collection purposes. | |
7038 | */ | |
7039 | static int io_uring_get_fd(struct io_ring_ctx *ctx) | |
7040 | { | |
7041 | struct file *file; | |
7042 | int ret; | |
7043 | ||
7044 | #if defined(CONFIG_UNIX) | |
7045 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, | |
7046 | &ctx->ring_sock); | |
7047 | if (ret) | |
7048 | return ret; | |
7049 | #endif | |
7050 | ||
7051 | ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
7052 | if (ret < 0) | |
7053 | goto err; | |
7054 | ||
7055 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, | |
7056 | O_RDWR | O_CLOEXEC); | |
7057 | if (IS_ERR(file)) { | |
7058 | put_unused_fd(ret); | |
7059 | ret = PTR_ERR(file); | |
7060 | goto err; | |
7061 | } | |
7062 | ||
7063 | #if defined(CONFIG_UNIX) | |
7064 | ctx->ring_sock->file = file; | |
7065 | #endif | |
7066 | fd_install(ret, file); | |
7067 | return ret; | |
7068 | err: | |
7069 | #if defined(CONFIG_UNIX) | |
7070 | sock_release(ctx->ring_sock); | |
7071 | ctx->ring_sock = NULL; | |
7072 | #endif | |
7073 | return ret; | |
7074 | } | |
7075 | ||
7076 | static int io_uring_create(unsigned entries, struct io_uring_params *p) | |
7077 | { | |
7078 | struct user_struct *user = NULL; | |
7079 | struct io_ring_ctx *ctx; | |
7080 | bool account_mem; | |
7081 | int ret; | |
7082 | ||
8110c1a6 | 7083 | if (!entries) |
2b188cc1 | 7084 | return -EINVAL; |
8110c1a6 JA |
7085 | if (entries > IORING_MAX_ENTRIES) { |
7086 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7087 | return -EINVAL; | |
7088 | entries = IORING_MAX_ENTRIES; | |
7089 | } | |
2b188cc1 JA |
7090 | |
7091 | /* | |
7092 | * Use twice as many entries for the CQ ring. It's possible for the | |
7093 | * application to drive a higher depth than the size of the SQ ring, | |
7094 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
7095 | * some flexibility in overcommitting a bit. If the application has |
7096 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
7097 | * of CQ ring entries manually. | |
2b188cc1 JA |
7098 | */ |
7099 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
7100 | if (p->flags & IORING_SETUP_CQSIZE) { |
7101 | /* | |
7102 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
7103 | * to a power-of-two, if it isn't already. We do NOT impose | |
7104 | * any cq vs sq ring sizing. | |
7105 | */ | |
8110c1a6 | 7106 | if (p->cq_entries < p->sq_entries) |
33a107f0 | 7107 | return -EINVAL; |
8110c1a6 JA |
7108 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
7109 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
7110 | return -EINVAL; | |
7111 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
7112 | } | |
33a107f0 JA |
7113 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
7114 | } else { | |
7115 | p->cq_entries = 2 * p->sq_entries; | |
7116 | } | |
2b188cc1 JA |
7117 | |
7118 | user = get_uid(current_user()); | |
7119 | account_mem = !capable(CAP_IPC_LOCK); | |
7120 | ||
7121 | if (account_mem) { | |
7122 | ret = io_account_mem(user, | |
7123 | ring_pages(p->sq_entries, p->cq_entries)); | |
7124 | if (ret) { | |
7125 | free_uid(user); | |
7126 | return ret; | |
7127 | } | |
7128 | } | |
7129 | ||
7130 | ctx = io_ring_ctx_alloc(p); | |
7131 | if (!ctx) { | |
7132 | if (account_mem) | |
7133 | io_unaccount_mem(user, ring_pages(p->sq_entries, | |
7134 | p->cq_entries)); | |
7135 | free_uid(user); | |
7136 | return -ENOMEM; | |
7137 | } | |
7138 | ctx->compat = in_compat_syscall(); | |
7139 | ctx->account_mem = account_mem; | |
7140 | ctx->user = user; | |
0b8c0ec7 | 7141 | ctx->creds = get_current_cred(); |
2b188cc1 JA |
7142 | |
7143 | ret = io_allocate_scq_urings(ctx, p); | |
7144 | if (ret) | |
7145 | goto err; | |
7146 | ||
6c271ce2 | 7147 | ret = io_sq_offload_start(ctx, p); |
2b188cc1 JA |
7148 | if (ret) |
7149 | goto err; | |
7150 | ||
2b188cc1 | 7151 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
7152 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
7153 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
7154 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
7155 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
7156 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
7157 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
7158 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
7159 | |
7160 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
7161 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
7162 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
7163 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
7164 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
7165 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
7166 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
ac90f249 | 7167 | |
044c1ab3 JA |
7168 | /* |
7169 | * Install ring fd as the very last thing, so we don't risk someone | |
7170 | * having closed it before we finish setup | |
7171 | */ | |
7172 | ret = io_uring_get_fd(ctx); | |
7173 | if (ret < 0) | |
7174 | goto err; | |
7175 | ||
da8c9690 | 7176 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
cccf0ee8 | 7177 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | |
d7718a9d | 7178 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL; |
c826bd7a | 7179 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
7180 | return ret; |
7181 | err: | |
7182 | io_ring_ctx_wait_and_kill(ctx); | |
7183 | return ret; | |
7184 | } | |
7185 | ||
7186 | /* | |
7187 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
7188 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
7189 | * params structure passed in. | |
7190 | */ | |
7191 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
7192 | { | |
7193 | struct io_uring_params p; | |
7194 | long ret; | |
7195 | int i; | |
7196 | ||
7197 | if (copy_from_user(&p, params, sizeof(p))) | |
7198 | return -EFAULT; | |
7199 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
7200 | if (p.resv[i]) | |
7201 | return -EINVAL; | |
7202 | } | |
7203 | ||
6c271ce2 | 7204 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 7205 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
24369c2e | 7206 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ)) |
2b188cc1 JA |
7207 | return -EINVAL; |
7208 | ||
7209 | ret = io_uring_create(entries, &p); | |
7210 | if (ret < 0) | |
7211 | return ret; | |
7212 | ||
7213 | if (copy_to_user(params, &p, sizeof(p))) | |
7214 | return -EFAULT; | |
7215 | ||
7216 | return ret; | |
7217 | } | |
7218 | ||
7219 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
7220 | struct io_uring_params __user *, params) | |
7221 | { | |
7222 | return io_uring_setup(entries, params); | |
7223 | } | |
7224 | ||
66f4af93 JA |
7225 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
7226 | { | |
7227 | struct io_uring_probe *p; | |
7228 | size_t size; | |
7229 | int i, ret; | |
7230 | ||
7231 | size = struct_size(p, ops, nr_args); | |
7232 | if (size == SIZE_MAX) | |
7233 | return -EOVERFLOW; | |
7234 | p = kzalloc(size, GFP_KERNEL); | |
7235 | if (!p) | |
7236 | return -ENOMEM; | |
7237 | ||
7238 | ret = -EFAULT; | |
7239 | if (copy_from_user(p, arg, size)) | |
7240 | goto out; | |
7241 | ret = -EINVAL; | |
7242 | if (memchr_inv(p, 0, size)) | |
7243 | goto out; | |
7244 | ||
7245 | p->last_op = IORING_OP_LAST - 1; | |
7246 | if (nr_args > IORING_OP_LAST) | |
7247 | nr_args = IORING_OP_LAST; | |
7248 | ||
7249 | for (i = 0; i < nr_args; i++) { | |
7250 | p->ops[i].op = i; | |
7251 | if (!io_op_defs[i].not_supported) | |
7252 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
7253 | } | |
7254 | p->ops_len = i; | |
7255 | ||
7256 | ret = 0; | |
7257 | if (copy_to_user(arg, p, size)) | |
7258 | ret = -EFAULT; | |
7259 | out: | |
7260 | kfree(p); | |
7261 | return ret; | |
7262 | } | |
7263 | ||
071698e1 JA |
7264 | static int io_register_personality(struct io_ring_ctx *ctx) |
7265 | { | |
7266 | const struct cred *creds = get_current_cred(); | |
7267 | int id; | |
7268 | ||
7269 | id = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, | |
7270 | USHRT_MAX, GFP_KERNEL); | |
7271 | if (id < 0) | |
7272 | put_cred(creds); | |
7273 | return id; | |
7274 | } | |
7275 | ||
7276 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) | |
7277 | { | |
7278 | const struct cred *old_creds; | |
7279 | ||
7280 | old_creds = idr_remove(&ctx->personality_idr, id); | |
7281 | if (old_creds) { | |
7282 | put_cred(old_creds); | |
7283 | return 0; | |
7284 | } | |
7285 | ||
7286 | return -EINVAL; | |
7287 | } | |
7288 | ||
7289 | static bool io_register_op_must_quiesce(int op) | |
7290 | { | |
7291 | switch (op) { | |
7292 | case IORING_UNREGISTER_FILES: | |
7293 | case IORING_REGISTER_FILES_UPDATE: | |
7294 | case IORING_REGISTER_PROBE: | |
7295 | case IORING_REGISTER_PERSONALITY: | |
7296 | case IORING_UNREGISTER_PERSONALITY: | |
7297 | return false; | |
7298 | default: | |
7299 | return true; | |
7300 | } | |
7301 | } | |
7302 | ||
edafccee JA |
7303 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
7304 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
7305 | __releases(ctx->uring_lock) |
7306 | __acquires(ctx->uring_lock) | |
edafccee JA |
7307 | { |
7308 | int ret; | |
7309 | ||
35fa71a0 JA |
7310 | /* |
7311 | * We're inside the ring mutex, if the ref is already dying, then | |
7312 | * someone else killed the ctx or is already going through | |
7313 | * io_uring_register(). | |
7314 | */ | |
7315 | if (percpu_ref_is_dying(&ctx->refs)) | |
7316 | return -ENXIO; | |
7317 | ||
071698e1 | 7318 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7319 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 7320 | |
05f3fb3c JA |
7321 | /* |
7322 | * Drop uring mutex before waiting for references to exit. If | |
7323 | * another thread is currently inside io_uring_enter() it might | |
7324 | * need to grab the uring_lock to make progress. If we hold it | |
7325 | * here across the drain wait, then we can deadlock. It's safe | |
7326 | * to drop the mutex here, since no new references will come in | |
7327 | * after we've killed the percpu ref. | |
7328 | */ | |
7329 | mutex_unlock(&ctx->uring_lock); | |
c150368b | 7330 | ret = wait_for_completion_interruptible(&ctx->completions[0]); |
05f3fb3c | 7331 | mutex_lock(&ctx->uring_lock); |
c150368b JA |
7332 | if (ret) { |
7333 | percpu_ref_resurrect(&ctx->refs); | |
7334 | ret = -EINTR; | |
7335 | goto out; | |
7336 | } | |
05f3fb3c | 7337 | } |
edafccee JA |
7338 | |
7339 | switch (opcode) { | |
7340 | case IORING_REGISTER_BUFFERS: | |
7341 | ret = io_sqe_buffer_register(ctx, arg, nr_args); | |
7342 | break; | |
7343 | case IORING_UNREGISTER_BUFFERS: | |
7344 | ret = -EINVAL; | |
7345 | if (arg || nr_args) | |
7346 | break; | |
7347 | ret = io_sqe_buffer_unregister(ctx); | |
7348 | break; | |
6b06314c JA |
7349 | case IORING_REGISTER_FILES: |
7350 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
7351 | break; | |
7352 | case IORING_UNREGISTER_FILES: | |
7353 | ret = -EINVAL; | |
7354 | if (arg || nr_args) | |
7355 | break; | |
7356 | ret = io_sqe_files_unregister(ctx); | |
7357 | break; | |
c3a31e60 JA |
7358 | case IORING_REGISTER_FILES_UPDATE: |
7359 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
7360 | break; | |
9b402849 | 7361 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 7362 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
7363 | ret = -EINVAL; |
7364 | if (nr_args != 1) | |
7365 | break; | |
7366 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
7367 | if (ret) |
7368 | break; | |
7369 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
7370 | ctx->eventfd_async = 1; | |
7371 | else | |
7372 | ctx->eventfd_async = 0; | |
9b402849 JA |
7373 | break; |
7374 | case IORING_UNREGISTER_EVENTFD: | |
7375 | ret = -EINVAL; | |
7376 | if (arg || nr_args) | |
7377 | break; | |
7378 | ret = io_eventfd_unregister(ctx); | |
7379 | break; | |
66f4af93 JA |
7380 | case IORING_REGISTER_PROBE: |
7381 | ret = -EINVAL; | |
7382 | if (!arg || nr_args > 256) | |
7383 | break; | |
7384 | ret = io_probe(ctx, arg, nr_args); | |
7385 | break; | |
071698e1 JA |
7386 | case IORING_REGISTER_PERSONALITY: |
7387 | ret = -EINVAL; | |
7388 | if (arg || nr_args) | |
7389 | break; | |
7390 | ret = io_register_personality(ctx); | |
7391 | break; | |
7392 | case IORING_UNREGISTER_PERSONALITY: | |
7393 | ret = -EINVAL; | |
7394 | if (arg) | |
7395 | break; | |
7396 | ret = io_unregister_personality(ctx, nr_args); | |
7397 | break; | |
edafccee JA |
7398 | default: |
7399 | ret = -EINVAL; | |
7400 | break; | |
7401 | } | |
7402 | ||
071698e1 | 7403 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 7404 | /* bring the ctx back to life */ |
05f3fb3c | 7405 | percpu_ref_reinit(&ctx->refs); |
c150368b JA |
7406 | out: |
7407 | reinit_completion(&ctx->completions[0]); | |
05f3fb3c | 7408 | } |
edafccee JA |
7409 | return ret; |
7410 | } | |
7411 | ||
7412 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
7413 | void __user *, arg, unsigned int, nr_args) | |
7414 | { | |
7415 | struct io_ring_ctx *ctx; | |
7416 | long ret = -EBADF; | |
7417 | struct fd f; | |
7418 | ||
7419 | f = fdget(fd); | |
7420 | if (!f.file) | |
7421 | return -EBADF; | |
7422 | ||
7423 | ret = -EOPNOTSUPP; | |
7424 | if (f.file->f_op != &io_uring_fops) | |
7425 | goto out_fput; | |
7426 | ||
7427 | ctx = f.file->private_data; | |
7428 | ||
7429 | mutex_lock(&ctx->uring_lock); | |
7430 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
7431 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
7432 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
7433 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
7434 | out_fput: |
7435 | fdput(f); | |
7436 | return ret; | |
7437 | } | |
7438 | ||
2b188cc1 JA |
7439 | static int __init io_uring_init(void) |
7440 | { | |
d7f62e82 SM |
7441 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
7442 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
7443 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
7444 | } while (0) | |
7445 | ||
7446 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
7447 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
7448 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
7449 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
7450 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
7451 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
7452 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
7453 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
7454 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
7455 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 7456 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
7457 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
7458 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
7459 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
7460 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
7461 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
7462 | BUILD_BUG_SQE_ELEM(28, __u16, poll_events); | |
7463 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); | |
7464 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
7465 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
7466 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
7467 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
7468 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
7469 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
7470 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 7471 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
7472 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
7473 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
7474 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 7475 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 7476 | |
d3656344 | 7477 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
2b188cc1 JA |
7478 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC); |
7479 | return 0; | |
7480 | }; | |
7481 | __initcall(io_uring_init); |