]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
e4b4a13f | 81 | #include <linux/freezer.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
101 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
102 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 103 | |
b16fed66 PB |
104 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
105 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
106 | IOSQE_BUFFER_SELECT) | |
107 | ||
2b188cc1 JA |
108 | struct io_uring { |
109 | u32 head ____cacheline_aligned_in_smp; | |
110 | u32 tail ____cacheline_aligned_in_smp; | |
111 | }; | |
112 | ||
1e84b97b | 113 | /* |
75b28aff HV |
114 | * This data is shared with the application through the mmap at offsets |
115 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
116 | * |
117 | * The offsets to the member fields are published through struct | |
118 | * io_sqring_offsets when calling io_uring_setup. | |
119 | */ | |
75b28aff | 120 | struct io_rings { |
1e84b97b SB |
121 | /* |
122 | * Head and tail offsets into the ring; the offsets need to be | |
123 | * masked to get valid indices. | |
124 | * | |
75b28aff HV |
125 | * The kernel controls head of the sq ring and the tail of the cq ring, |
126 | * and the application controls tail of the sq ring and the head of the | |
127 | * cq ring. | |
1e84b97b | 128 | */ |
75b28aff | 129 | struct io_uring sq, cq; |
1e84b97b | 130 | /* |
75b28aff | 131 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
132 | * ring_entries - 1) |
133 | */ | |
75b28aff HV |
134 | u32 sq_ring_mask, cq_ring_mask; |
135 | /* Ring sizes (constant, power of 2) */ | |
136 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
137 | /* |
138 | * Number of invalid entries dropped by the kernel due to | |
139 | * invalid index stored in array | |
140 | * | |
141 | * Written by the kernel, shouldn't be modified by the | |
142 | * application (i.e. get number of "new events" by comparing to | |
143 | * cached value). | |
144 | * | |
145 | * After a new SQ head value was read by the application this | |
146 | * counter includes all submissions that were dropped reaching | |
147 | * the new SQ head (and possibly more). | |
148 | */ | |
75b28aff | 149 | u32 sq_dropped; |
1e84b97b | 150 | /* |
0d9b5b3a | 151 | * Runtime SQ flags |
1e84b97b SB |
152 | * |
153 | * Written by the kernel, shouldn't be modified by the | |
154 | * application. | |
155 | * | |
156 | * The application needs a full memory barrier before checking | |
157 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
158 | */ | |
75b28aff | 159 | u32 sq_flags; |
0d9b5b3a SG |
160 | /* |
161 | * Runtime CQ flags | |
162 | * | |
163 | * Written by the application, shouldn't be modified by the | |
164 | * kernel. | |
165 | */ | |
166 | u32 cq_flags; | |
1e84b97b SB |
167 | /* |
168 | * Number of completion events lost because the queue was full; | |
169 | * this should be avoided by the application by making sure | |
0b4295b5 | 170 | * there are not more requests pending than there is space in |
1e84b97b SB |
171 | * the completion queue. |
172 | * | |
173 | * Written by the kernel, shouldn't be modified by the | |
174 | * application (i.e. get number of "new events" by comparing to | |
175 | * cached value). | |
176 | * | |
177 | * As completion events come in out of order this counter is not | |
178 | * ordered with any other data. | |
179 | */ | |
75b28aff | 180 | u32 cq_overflow; |
1e84b97b SB |
181 | /* |
182 | * Ring buffer of completion events. | |
183 | * | |
184 | * The kernel writes completion events fresh every time they are | |
185 | * produced, so the application is allowed to modify pending | |
186 | * entries. | |
187 | */ | |
75b28aff | 188 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
189 | }; |
190 | ||
45d189c6 PB |
191 | enum io_uring_cmd_flags { |
192 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 193 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
194 | }; |
195 | ||
edafccee JA |
196 | struct io_mapped_ubuf { |
197 | u64 ubuf; | |
198 | size_t len; | |
199 | struct bio_vec *bvec; | |
200 | unsigned int nr_bvecs; | |
de293938 | 201 | unsigned long acct_pages; |
edafccee JA |
202 | }; |
203 | ||
50238531 BM |
204 | struct io_ring_ctx; |
205 | ||
269bbe5f BM |
206 | struct io_rsrc_put { |
207 | struct list_head list; | |
50238531 BM |
208 | union { |
209 | void *rsrc; | |
210 | struct file *file; | |
211 | }; | |
269bbe5f BM |
212 | }; |
213 | ||
214 | struct fixed_rsrc_table { | |
65e19f54 | 215 | struct file **files; |
31b51510 JA |
216 | }; |
217 | ||
269bbe5f | 218 | struct fixed_rsrc_ref_node { |
05589553 XW |
219 | struct percpu_ref refs; |
220 | struct list_head node; | |
269bbe5f BM |
221 | struct list_head rsrc_list; |
222 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
223 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
224 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 225 | struct llist_node llist; |
e297822b | 226 | bool done; |
05589553 XW |
227 | }; |
228 | ||
269bbe5f BM |
229 | struct fixed_rsrc_data { |
230 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
231 | struct io_ring_ctx *ctx; |
232 | ||
269bbe5f | 233 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 234 | struct percpu_ref refs; |
05f3fb3c | 235 | struct completion done; |
8bad28d8 | 236 | bool quiesce; |
05f3fb3c JA |
237 | }; |
238 | ||
5a2e745d JA |
239 | struct io_buffer { |
240 | struct list_head list; | |
241 | __u64 addr; | |
242 | __s32 len; | |
243 | __u16 bid; | |
244 | }; | |
245 | ||
21b55dbc SG |
246 | struct io_restriction { |
247 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
248 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
249 | u8 sqe_flags_allowed; | |
250 | u8 sqe_flags_required; | |
7e84e1c7 | 251 | bool registered; |
21b55dbc SG |
252 | }; |
253 | ||
37d1e2e3 JA |
254 | enum { |
255 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
256 | IO_SQ_THREAD_SHOULD_PARK, | |
257 | }; | |
258 | ||
534ca6d6 JA |
259 | struct io_sq_data { |
260 | refcount_t refs; | |
9e138a48 | 261 | atomic_t park_pending; |
09a6f4ef | 262 | struct mutex lock; |
69fb2131 JA |
263 | |
264 | /* ctx's that are using this sqd */ | |
265 | struct list_head ctx_list; | |
69fb2131 | 266 | |
534ca6d6 JA |
267 | struct task_struct *thread; |
268 | struct wait_queue_head wait; | |
08369246 XW |
269 | |
270 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
271 | int sq_cpu; |
272 | pid_t task_pid; | |
5c2469e0 | 273 | pid_t task_tgid; |
37d1e2e3 JA |
274 | |
275 | unsigned long state; | |
37d1e2e3 | 276 | struct completion exited; |
b7f5a0bf | 277 | struct callback_head *park_task_work; |
534ca6d6 JA |
278 | }; |
279 | ||
258b29a9 | 280 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 281 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 282 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 283 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
284 | |
285 | struct io_comp_state { | |
6dd0be1e | 286 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 287 | unsigned int nr; |
c7dae4ba JA |
288 | unsigned int locked_free_nr; |
289 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 290 | struct list_head free_list; |
c7dae4ba JA |
291 | /* IRQ completion list, under ->completion_lock */ |
292 | struct list_head locked_free_list; | |
258b29a9 PB |
293 | }; |
294 | ||
a1ab7b35 PB |
295 | struct io_submit_link { |
296 | struct io_kiocb *head; | |
297 | struct io_kiocb *last; | |
298 | }; | |
299 | ||
258b29a9 PB |
300 | struct io_submit_state { |
301 | struct blk_plug plug; | |
a1ab7b35 | 302 | struct io_submit_link link; |
258b29a9 PB |
303 | |
304 | /* | |
305 | * io_kiocb alloc cache | |
306 | */ | |
bf019da7 | 307 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
308 | unsigned int free_reqs; |
309 | ||
310 | bool plug_started; | |
311 | ||
312 | /* | |
313 | * Batch completion logic | |
314 | */ | |
315 | struct io_comp_state comp; | |
316 | ||
317 | /* | |
318 | * File reference cache | |
319 | */ | |
320 | struct file *file; | |
321 | unsigned int fd; | |
322 | unsigned int file_refs; | |
323 | unsigned int ios_left; | |
324 | }; | |
325 | ||
2b188cc1 JA |
326 | struct io_ring_ctx { |
327 | struct { | |
328 | struct percpu_ref refs; | |
329 | } ____cacheline_aligned_in_smp; | |
330 | ||
331 | struct { | |
332 | unsigned int flags; | |
e1d85334 | 333 | unsigned int compat: 1; |
e1d85334 RD |
334 | unsigned int cq_overflow_flushed: 1; |
335 | unsigned int drain_next: 1; | |
336 | unsigned int eventfd_async: 1; | |
21b55dbc | 337 | unsigned int restricted: 1; |
2b188cc1 | 338 | |
75b28aff HV |
339 | /* |
340 | * Ring buffer of indices into array of io_uring_sqe, which is | |
341 | * mmapped by the application using the IORING_OFF_SQES offset. | |
342 | * | |
343 | * This indirection could e.g. be used to assign fixed | |
344 | * io_uring_sqe entries to operations and only submit them to | |
345 | * the queue when needed. | |
346 | * | |
347 | * The kernel modifies neither the indices array nor the entries | |
348 | * array. | |
349 | */ | |
350 | u32 *sq_array; | |
2b188cc1 JA |
351 | unsigned cached_sq_head; |
352 | unsigned sq_entries; | |
353 | unsigned sq_mask; | |
6c271ce2 | 354 | unsigned sq_thread_idle; |
498ccd9e | 355 | unsigned cached_sq_dropped; |
2c3bac6d | 356 | unsigned cached_cq_overflow; |
ad3eb2c8 | 357 | unsigned long sq_check_overflow; |
de0617e4 | 358 | |
e941894e JA |
359 | /* hashed buffered write serialization */ |
360 | struct io_wq_hash *hash_map; | |
361 | ||
de0617e4 | 362 | struct list_head defer_list; |
5262f567 | 363 | struct list_head timeout_list; |
1d7bb1d5 | 364 | struct list_head cq_overflow_list; |
fcb323cc | 365 | |
ad3eb2c8 | 366 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
367 | } ____cacheline_aligned_in_smp; |
368 | ||
3c1a2ead JA |
369 | struct { |
370 | struct mutex uring_lock; | |
371 | wait_queue_head_t wait; | |
372 | } ____cacheline_aligned_in_smp; | |
373 | ||
374 | struct io_submit_state submit_state; | |
375 | ||
206aefde JA |
376 | struct io_rings *rings; |
377 | ||
2aede0e4 JA |
378 | /* Only used for accounting purposes */ |
379 | struct mm_struct *mm_account; | |
380 | ||
7c30f36a | 381 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
382 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
383 | ||
90554200 | 384 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 385 | struct list_head sqd_list; |
75b28aff | 386 | |
6b06314c JA |
387 | /* |
388 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
389 | * readers must ensure that ->refs is alive as long as the file* is | |
390 | * used. Only updated through io_uring_register(2). | |
391 | */ | |
269bbe5f | 392 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
393 | unsigned nr_user_files; |
394 | ||
edafccee JA |
395 | /* if used, fixed mapped user buffers */ |
396 | unsigned nr_user_bufs; | |
397 | struct io_mapped_ubuf *user_bufs; | |
398 | ||
2b188cc1 JA |
399 | struct user_struct *user; |
400 | ||
0f158b4c | 401 | struct completion ref_comp; |
206aefde JA |
402 | |
403 | #if defined(CONFIG_UNIX) | |
404 | struct socket *ring_sock; | |
405 | #endif | |
406 | ||
9e15c3a0 | 407 | struct xarray io_buffers; |
5a2e745d | 408 | |
61cf9370 MWO |
409 | struct xarray personalities; |
410 | u32 pers_next; | |
071698e1 | 411 | |
206aefde JA |
412 | struct { |
413 | unsigned cached_cq_tail; | |
414 | unsigned cq_entries; | |
415 | unsigned cq_mask; | |
416 | atomic_t cq_timeouts; | |
f010505b | 417 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 418 | unsigned long cq_check_overflow; |
206aefde JA |
419 | struct wait_queue_head cq_wait; |
420 | struct fasync_struct *cq_fasync; | |
421 | struct eventfd_ctx *cq_ev_fd; | |
422 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 423 | |
2b188cc1 JA |
424 | struct { |
425 | spinlock_t completion_lock; | |
e94f141b | 426 | |
def596e9 | 427 | /* |
540e32a0 | 428 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
429 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
430 | * For SQPOLL, only the single threaded io_sq_thread() will | |
431 | * manipulate the list, hence no extra locking is needed there. | |
432 | */ | |
540e32a0 | 433 | struct list_head iopoll_list; |
78076bb6 JA |
434 | struct hlist_head *cancel_hash; |
435 | unsigned cancel_hash_bits; | |
e94f141b | 436 | bool poll_multi_file; |
31b51510 | 437 | |
fcb323cc JA |
438 | spinlock_t inflight_lock; |
439 | struct list_head inflight_list; | |
2b188cc1 | 440 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 441 | |
269bbe5f BM |
442 | struct delayed_work rsrc_put_work; |
443 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
444 | struct list_head rsrc_ref_list; |
445 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 446 | |
21b55dbc | 447 | struct io_restriction restrictions; |
3c1a2ead | 448 | |
7c25c0d1 JA |
449 | /* exit task_work */ |
450 | struct callback_head *exit_task_work; | |
451 | ||
e941894e JA |
452 | struct wait_queue_head hash_wait; |
453 | ||
3c1a2ead JA |
454 | /* Keep this last, we don't need it for the fast path */ |
455 | struct work_struct exit_work; | |
13bf43f5 | 456 | struct list_head tctx_list; |
2b188cc1 JA |
457 | }; |
458 | ||
09bb8394 JA |
459 | /* |
460 | * First field must be the file pointer in all the | |
461 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
462 | */ | |
221c5eb2 JA |
463 | struct io_poll_iocb { |
464 | struct file *file; | |
018043be | 465 | struct wait_queue_head *head; |
221c5eb2 | 466 | __poll_t events; |
8c838788 | 467 | bool done; |
221c5eb2 | 468 | bool canceled; |
392edb45 | 469 | struct wait_queue_entry wait; |
221c5eb2 JA |
470 | }; |
471 | ||
018043be PB |
472 | struct io_poll_remove { |
473 | struct file *file; | |
474 | u64 addr; | |
475 | }; | |
476 | ||
b5dba59e JA |
477 | struct io_close { |
478 | struct file *file; | |
b5dba59e JA |
479 | int fd; |
480 | }; | |
481 | ||
ad8a48ac JA |
482 | struct io_timeout_data { |
483 | struct io_kiocb *req; | |
484 | struct hrtimer timer; | |
485 | struct timespec64 ts; | |
486 | enum hrtimer_mode mode; | |
487 | }; | |
488 | ||
8ed8d3c3 JA |
489 | struct io_accept { |
490 | struct file *file; | |
491 | struct sockaddr __user *addr; | |
492 | int __user *addr_len; | |
493 | int flags; | |
09952e3e | 494 | unsigned long nofile; |
8ed8d3c3 JA |
495 | }; |
496 | ||
497 | struct io_sync { | |
498 | struct file *file; | |
499 | loff_t len; | |
500 | loff_t off; | |
501 | int flags; | |
d63d1b5e | 502 | int mode; |
8ed8d3c3 JA |
503 | }; |
504 | ||
fbf23849 JA |
505 | struct io_cancel { |
506 | struct file *file; | |
507 | u64 addr; | |
508 | }; | |
509 | ||
b29472ee JA |
510 | struct io_timeout { |
511 | struct file *file; | |
bfe68a22 PB |
512 | u32 off; |
513 | u32 target_seq; | |
135fcde8 | 514 | struct list_head list; |
90cd7e42 PB |
515 | /* head of the link, used by linked timeouts only */ |
516 | struct io_kiocb *head; | |
b29472ee JA |
517 | }; |
518 | ||
0bdf7a2d PB |
519 | struct io_timeout_rem { |
520 | struct file *file; | |
521 | u64 addr; | |
9c8e11b3 PB |
522 | |
523 | /* timeout update */ | |
524 | struct timespec64 ts; | |
525 | u32 flags; | |
0bdf7a2d PB |
526 | }; |
527 | ||
9adbd45d JA |
528 | struct io_rw { |
529 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
530 | struct kiocb kiocb; | |
531 | u64 addr; | |
532 | u64 len; | |
533 | }; | |
534 | ||
3fbb51c1 JA |
535 | struct io_connect { |
536 | struct file *file; | |
537 | struct sockaddr __user *addr; | |
538 | int addr_len; | |
539 | }; | |
540 | ||
e47293fd JA |
541 | struct io_sr_msg { |
542 | struct file *file; | |
fddaface | 543 | union { |
270a5940 | 544 | struct user_msghdr __user *umsg; |
fddaface JA |
545 | void __user *buf; |
546 | }; | |
e47293fd | 547 | int msg_flags; |
bcda7baa | 548 | int bgid; |
fddaface | 549 | size_t len; |
bcda7baa | 550 | struct io_buffer *kbuf; |
e47293fd JA |
551 | }; |
552 | ||
15b71abe JA |
553 | struct io_open { |
554 | struct file *file; | |
555 | int dfd; | |
15b71abe | 556 | struct filename *filename; |
c12cedf2 | 557 | struct open_how how; |
4022e7af | 558 | unsigned long nofile; |
15b71abe JA |
559 | }; |
560 | ||
269bbe5f | 561 | struct io_rsrc_update { |
05f3fb3c JA |
562 | struct file *file; |
563 | u64 arg; | |
564 | u32 nr_args; | |
565 | u32 offset; | |
566 | }; | |
567 | ||
4840e418 JA |
568 | struct io_fadvise { |
569 | struct file *file; | |
570 | u64 offset; | |
571 | u32 len; | |
572 | u32 advice; | |
573 | }; | |
574 | ||
c1ca757b JA |
575 | struct io_madvise { |
576 | struct file *file; | |
577 | u64 addr; | |
578 | u32 len; | |
579 | u32 advice; | |
580 | }; | |
581 | ||
3e4827b0 JA |
582 | struct io_epoll { |
583 | struct file *file; | |
584 | int epfd; | |
585 | int op; | |
586 | int fd; | |
587 | struct epoll_event event; | |
e47293fd JA |
588 | }; |
589 | ||
7d67af2c PB |
590 | struct io_splice { |
591 | struct file *file_out; | |
592 | struct file *file_in; | |
593 | loff_t off_out; | |
594 | loff_t off_in; | |
595 | u64 len; | |
596 | unsigned int flags; | |
597 | }; | |
598 | ||
ddf0322d JA |
599 | struct io_provide_buf { |
600 | struct file *file; | |
601 | __u64 addr; | |
602 | __s32 len; | |
603 | __u32 bgid; | |
604 | __u16 nbufs; | |
605 | __u16 bid; | |
606 | }; | |
607 | ||
1d9e1288 BM |
608 | struct io_statx { |
609 | struct file *file; | |
610 | int dfd; | |
611 | unsigned int mask; | |
612 | unsigned int flags; | |
e62753e4 | 613 | const char __user *filename; |
1d9e1288 BM |
614 | struct statx __user *buffer; |
615 | }; | |
616 | ||
36f4fa68 JA |
617 | struct io_shutdown { |
618 | struct file *file; | |
619 | int how; | |
620 | }; | |
621 | ||
80a261fd JA |
622 | struct io_rename { |
623 | struct file *file; | |
624 | int old_dfd; | |
625 | int new_dfd; | |
626 | struct filename *oldpath; | |
627 | struct filename *newpath; | |
628 | int flags; | |
629 | }; | |
630 | ||
14a1143b JA |
631 | struct io_unlink { |
632 | struct file *file; | |
633 | int dfd; | |
634 | int flags; | |
635 | struct filename *filename; | |
636 | }; | |
637 | ||
3ca405eb PB |
638 | struct io_completion { |
639 | struct file *file; | |
640 | struct list_head list; | |
0f7e466b | 641 | int cflags; |
3ca405eb PB |
642 | }; |
643 | ||
f499a021 JA |
644 | struct io_async_connect { |
645 | struct sockaddr_storage address; | |
646 | }; | |
647 | ||
03b1230c JA |
648 | struct io_async_msghdr { |
649 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
650 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
651 | struct iovec *free_iov; | |
03b1230c JA |
652 | struct sockaddr __user *uaddr; |
653 | struct msghdr msg; | |
b537916c | 654 | struct sockaddr_storage addr; |
03b1230c JA |
655 | }; |
656 | ||
f67676d1 JA |
657 | struct io_async_rw { |
658 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
659 | const struct iovec *free_iovec; |
660 | struct iov_iter iter; | |
227c0c96 | 661 | size_t bytes_done; |
bcf5a063 | 662 | struct wait_page_queue wpq; |
f67676d1 JA |
663 | }; |
664 | ||
6b47ee6e PB |
665 | enum { |
666 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
667 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
668 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
669 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
670 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 671 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 672 | |
6b47ee6e PB |
673 | REQ_F_FAIL_LINK_BIT, |
674 | REQ_F_INFLIGHT_BIT, | |
675 | REQ_F_CUR_POS_BIT, | |
676 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 677 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 678 | REQ_F_ISREG_BIT, |
99bc4c38 | 679 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 680 | REQ_F_POLLED_BIT, |
bcda7baa | 681 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 682 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 683 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 684 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
685 | |
686 | /* not a real bit, just to check we're not overflowing the space */ | |
687 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
688 | }; |
689 | ||
690 | enum { | |
691 | /* ctx owns file */ | |
692 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
693 | /* drain existing IO first */ | |
694 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
695 | /* linked sqes */ | |
696 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
697 | /* doesn't sever on completion < 0 */ | |
698 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
699 | /* IOSQE_ASYNC */ | |
700 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
701 | /* IOSQE_BUFFER_SELECT */ |
702 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 703 | |
6b47ee6e PB |
704 | /* fail rest of links */ |
705 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 706 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
707 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
708 | /* read/write uses file position */ | |
709 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
710 | /* must not punt to workers */ | |
711 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 712 | /* has or had linked timeout */ |
6b47ee6e | 713 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
714 | /* regular file */ |
715 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
716 | /* needs cleanup */ |
717 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
718 | /* already went through poll handler */ |
719 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
720 | /* buffer already selected */ |
721 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
722 | /* doesn't need file table for this request */ |
723 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
724 | /* linked timeout is active, i.e. prepared by link's head */ |
725 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
726 | /* completion is deferred through io_comp_state */ |
727 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
728 | }; |
729 | ||
730 | struct async_poll { | |
731 | struct io_poll_iocb poll; | |
807abcb0 | 732 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
733 | }; |
734 | ||
7cbf1722 JA |
735 | struct io_task_work { |
736 | struct io_wq_work_node node; | |
737 | task_work_func_t func; | |
738 | }; | |
739 | ||
09bb8394 JA |
740 | /* |
741 | * NOTE! Each of the iocb union members has the file pointer | |
742 | * as the first entry in their struct definition. So you can | |
743 | * access the file pointer through any of the sub-structs, | |
744 | * or directly as just 'ki_filp' in this struct. | |
745 | */ | |
2b188cc1 | 746 | struct io_kiocb { |
221c5eb2 | 747 | union { |
09bb8394 | 748 | struct file *file; |
9adbd45d | 749 | struct io_rw rw; |
221c5eb2 | 750 | struct io_poll_iocb poll; |
018043be | 751 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
752 | struct io_accept accept; |
753 | struct io_sync sync; | |
fbf23849 | 754 | struct io_cancel cancel; |
b29472ee | 755 | struct io_timeout timeout; |
0bdf7a2d | 756 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 757 | struct io_connect connect; |
e47293fd | 758 | struct io_sr_msg sr_msg; |
15b71abe | 759 | struct io_open open; |
b5dba59e | 760 | struct io_close close; |
269bbe5f | 761 | struct io_rsrc_update rsrc_update; |
4840e418 | 762 | struct io_fadvise fadvise; |
c1ca757b | 763 | struct io_madvise madvise; |
3e4827b0 | 764 | struct io_epoll epoll; |
7d67af2c | 765 | struct io_splice splice; |
ddf0322d | 766 | struct io_provide_buf pbuf; |
1d9e1288 | 767 | struct io_statx statx; |
36f4fa68 | 768 | struct io_shutdown shutdown; |
80a261fd | 769 | struct io_rename rename; |
14a1143b | 770 | struct io_unlink unlink; |
3ca405eb PB |
771 | /* use only after cleaning per-op data, see io_clean_op() */ |
772 | struct io_completion compl; | |
221c5eb2 | 773 | }; |
2b188cc1 | 774 | |
e8c2bc1f JA |
775 | /* opcode allocated if it needs to store data for async defer */ |
776 | void *async_data; | |
d625c6ee | 777 | u8 opcode; |
65a6543d XW |
778 | /* polled IO has completed */ |
779 | u8 iopoll_completed; | |
2b188cc1 | 780 | |
4f4eeba8 | 781 | u16 buf_index; |
9cf7c104 | 782 | u32 result; |
4f4eeba8 | 783 | |
010e8e6b PB |
784 | struct io_ring_ctx *ctx; |
785 | unsigned int flags; | |
786 | refcount_t refs; | |
787 | struct task_struct *task; | |
788 | u64 user_data; | |
d7718a9d | 789 | |
f2f87370 | 790 | struct io_kiocb *link; |
269bbe5f | 791 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 792 | |
d21ffe7e PB |
793 | /* |
794 | * 1. used with ctx->iopoll_list with reads/writes | |
795 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
796 | */ | |
010e8e6b | 797 | struct list_head inflight_entry; |
7cbf1722 JA |
798 | union { |
799 | struct io_task_work io_task_work; | |
800 | struct callback_head task_work; | |
801 | }; | |
010e8e6b PB |
802 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
803 | struct hlist_node hash_node; | |
804 | struct async_poll *apoll; | |
805 | struct io_wq_work work; | |
2b188cc1 | 806 | }; |
05589553 | 807 | |
13bf43f5 PB |
808 | struct io_tctx_node { |
809 | struct list_head ctx_node; | |
810 | struct task_struct *task; | |
13bf43f5 PB |
811 | struct io_ring_ctx *ctx; |
812 | }; | |
813 | ||
27dc8338 PB |
814 | struct io_defer_entry { |
815 | struct list_head list; | |
816 | struct io_kiocb *req; | |
9cf7c104 | 817 | u32 seq; |
2b188cc1 JA |
818 | }; |
819 | ||
d3656344 | 820 | struct io_op_def { |
d3656344 JA |
821 | /* needs req->file assigned */ |
822 | unsigned needs_file : 1; | |
d3656344 JA |
823 | /* hash wq insertion if file is a regular file */ |
824 | unsigned hash_reg_file : 1; | |
825 | /* unbound wq insertion if file is a non-regular file */ | |
826 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
827 | /* opcode is not supported by this kernel */ |
828 | unsigned not_supported : 1; | |
8a72758c JA |
829 | /* set if opcode supports polled "wait" */ |
830 | unsigned pollin : 1; | |
831 | unsigned pollout : 1; | |
bcda7baa JA |
832 | /* op supports buffer selection */ |
833 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
834 | /* must always have async data allocated */ |
835 | unsigned needs_async_data : 1; | |
27926b68 JA |
836 | /* should block plug */ |
837 | unsigned plug : 1; | |
e8c2bc1f JA |
838 | /* size of async data needed, if any */ |
839 | unsigned short async_size; | |
d3656344 JA |
840 | }; |
841 | ||
0918682b | 842 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
843 | [IORING_OP_NOP] = {}, |
844 | [IORING_OP_READV] = { | |
d3656344 JA |
845 | .needs_file = 1, |
846 | .unbound_nonreg_file = 1, | |
8a72758c | 847 | .pollin = 1, |
4d954c25 | 848 | .buffer_select = 1, |
e8c2bc1f | 849 | .needs_async_data = 1, |
27926b68 | 850 | .plug = 1, |
e8c2bc1f | 851 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 852 | }, |
0463b6c5 | 853 | [IORING_OP_WRITEV] = { |
d3656344 JA |
854 | .needs_file = 1, |
855 | .hash_reg_file = 1, | |
856 | .unbound_nonreg_file = 1, | |
8a72758c | 857 | .pollout = 1, |
e8c2bc1f | 858 | .needs_async_data = 1, |
27926b68 | 859 | .plug = 1, |
e8c2bc1f | 860 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 861 | }, |
0463b6c5 | 862 | [IORING_OP_FSYNC] = { |
d3656344 JA |
863 | .needs_file = 1, |
864 | }, | |
0463b6c5 | 865 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
866 | .needs_file = 1, |
867 | .unbound_nonreg_file = 1, | |
8a72758c | 868 | .pollin = 1, |
27926b68 | 869 | .plug = 1, |
e8c2bc1f | 870 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 871 | }, |
0463b6c5 | 872 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
873 | .needs_file = 1, |
874 | .hash_reg_file = 1, | |
875 | .unbound_nonreg_file = 1, | |
8a72758c | 876 | .pollout = 1, |
27926b68 | 877 | .plug = 1, |
e8c2bc1f | 878 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 879 | }, |
0463b6c5 | 880 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
881 | .needs_file = 1, |
882 | .unbound_nonreg_file = 1, | |
883 | }, | |
0463b6c5 PB |
884 | [IORING_OP_POLL_REMOVE] = {}, |
885 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
886 | .needs_file = 1, |
887 | }, | |
0463b6c5 | 888 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
889 | .needs_file = 1, |
890 | .unbound_nonreg_file = 1, | |
8a72758c | 891 | .pollout = 1, |
e8c2bc1f JA |
892 | .needs_async_data = 1, |
893 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 894 | }, |
0463b6c5 | 895 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
896 | .needs_file = 1, |
897 | .unbound_nonreg_file = 1, | |
8a72758c | 898 | .pollin = 1, |
52de1fe1 | 899 | .buffer_select = 1, |
e8c2bc1f JA |
900 | .needs_async_data = 1, |
901 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 902 | }, |
0463b6c5 | 903 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
904 | .needs_async_data = 1, |
905 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 906 | }, |
9c8e11b3 PB |
907 | [IORING_OP_TIMEOUT_REMOVE] = { |
908 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 909 | }, |
0463b6c5 | 910 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
911 | .needs_file = 1, |
912 | .unbound_nonreg_file = 1, | |
8a72758c | 913 | .pollin = 1, |
d3656344 | 914 | }, |
0463b6c5 PB |
915 | [IORING_OP_ASYNC_CANCEL] = {}, |
916 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
917 | .needs_async_data = 1, |
918 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 919 | }, |
0463b6c5 | 920 | [IORING_OP_CONNECT] = { |
d3656344 JA |
921 | .needs_file = 1, |
922 | .unbound_nonreg_file = 1, | |
8a72758c | 923 | .pollout = 1, |
e8c2bc1f JA |
924 | .needs_async_data = 1, |
925 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 926 | }, |
0463b6c5 | 927 | [IORING_OP_FALLOCATE] = { |
d3656344 | 928 | .needs_file = 1, |
d3656344 | 929 | }, |
44526bed JA |
930 | [IORING_OP_OPENAT] = {}, |
931 | [IORING_OP_CLOSE] = {}, | |
932 | [IORING_OP_FILES_UPDATE] = {}, | |
933 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 934 | [IORING_OP_READ] = { |
3a6820f2 JA |
935 | .needs_file = 1, |
936 | .unbound_nonreg_file = 1, | |
8a72758c | 937 | .pollin = 1, |
bcda7baa | 938 | .buffer_select = 1, |
27926b68 | 939 | .plug = 1, |
e8c2bc1f | 940 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 941 | }, |
0463b6c5 | 942 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
943 | .needs_file = 1, |
944 | .unbound_nonreg_file = 1, | |
8a72758c | 945 | .pollout = 1, |
27926b68 | 946 | .plug = 1, |
e8c2bc1f | 947 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 948 | }, |
0463b6c5 | 949 | [IORING_OP_FADVISE] = { |
4840e418 | 950 | .needs_file = 1, |
c1ca757b | 951 | }, |
44526bed | 952 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 953 | [IORING_OP_SEND] = { |
fddaface JA |
954 | .needs_file = 1, |
955 | .unbound_nonreg_file = 1, | |
8a72758c | 956 | .pollout = 1, |
fddaface | 957 | }, |
0463b6c5 | 958 | [IORING_OP_RECV] = { |
fddaface JA |
959 | .needs_file = 1, |
960 | .unbound_nonreg_file = 1, | |
8a72758c | 961 | .pollin = 1, |
bcda7baa | 962 | .buffer_select = 1, |
fddaface | 963 | }, |
0463b6c5 | 964 | [IORING_OP_OPENAT2] = { |
cebdb986 | 965 | }, |
3e4827b0 JA |
966 | [IORING_OP_EPOLL_CTL] = { |
967 | .unbound_nonreg_file = 1, | |
3e4827b0 | 968 | }, |
7d67af2c PB |
969 | [IORING_OP_SPLICE] = { |
970 | .needs_file = 1, | |
971 | .hash_reg_file = 1, | |
972 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
973 | }, |
974 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 975 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
976 | [IORING_OP_TEE] = { |
977 | .needs_file = 1, | |
978 | .hash_reg_file = 1, | |
979 | .unbound_nonreg_file = 1, | |
980 | }, | |
36f4fa68 JA |
981 | [IORING_OP_SHUTDOWN] = { |
982 | .needs_file = 1, | |
983 | }, | |
44526bed JA |
984 | [IORING_OP_RENAMEAT] = {}, |
985 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
986 | }; |
987 | ||
7a612350 | 988 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 989 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
990 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
991 | struct task_struct *task, | |
992 | struct files_struct *files); | |
37d1e2e3 | 993 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 994 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 995 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 996 | struct io_ring_ctx *ctx); |
f2303b1f | 997 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 998 | |
23faba36 | 999 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1000 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1001 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1002 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1003 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1004 | static void io_dismantle_req(struct io_kiocb *req); |
1005 | static void io_put_task(struct task_struct *task, int nr); | |
1006 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1007 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1008 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1009 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1010 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1011 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1012 | unsigned nr_args); |
3ca405eb | 1013 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1014 | static struct file *io_file_get(struct io_submit_state *state, |
1015 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1016 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1017 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1018 | |
847595de PB |
1019 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1020 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1021 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1022 | const struct iovec *fast_iov, | |
227c0c96 | 1023 | struct iov_iter *iter, bool force); |
907d1df3 | 1024 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1025 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1026 | struct io_ring_ctx *ctx); | |
de0617e4 | 1027 | |
2b188cc1 JA |
1028 | static struct kmem_cache *req_cachep; |
1029 | ||
0918682b | 1030 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1031 | |
1032 | struct sock *io_uring_get_socket(struct file *file) | |
1033 | { | |
1034 | #if defined(CONFIG_UNIX) | |
1035 | if (file->f_op == &io_uring_fops) { | |
1036 | struct io_ring_ctx *ctx = file->private_data; | |
1037 | ||
1038 | return ctx->ring_sock->sk; | |
1039 | } | |
1040 | #endif | |
1041 | return NULL; | |
1042 | } | |
1043 | EXPORT_SYMBOL(io_uring_get_socket); | |
1044 | ||
f2f87370 PB |
1045 | #define io_for_each_link(pos, head) \ |
1046 | for (pos = (head); pos; pos = pos->link) | |
1047 | ||
3ca405eb PB |
1048 | static inline void io_clean_op(struct io_kiocb *req) |
1049 | { | |
9d5c8190 | 1050 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1051 | __io_clean_op(req); |
1052 | } | |
1053 | ||
36f72fe2 PB |
1054 | static inline void io_set_resource_node(struct io_kiocb *req) |
1055 | { | |
1056 | struct io_ring_ctx *ctx = req->ctx; | |
1057 | ||
269bbe5f BM |
1058 | if (!req->fixed_rsrc_refs) { |
1059 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1060 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1061 | } |
1062 | } | |
1063 | ||
08d23634 PB |
1064 | static bool io_match_task(struct io_kiocb *head, |
1065 | struct task_struct *task, | |
1066 | struct files_struct *files) | |
1067 | { | |
1068 | struct io_kiocb *req; | |
1069 | ||
84965ff8 JA |
1070 | if (task && head->task != task) { |
1071 | /* in terms of cancelation, always match if req task is dead */ | |
1072 | if (head->task->flags & PF_EXITING) | |
1073 | return true; | |
08d23634 | 1074 | return false; |
84965ff8 | 1075 | } |
08d23634 PB |
1076 | if (!files) |
1077 | return true; | |
1078 | ||
1079 | io_for_each_link(req, head) { | |
b05a1bcd | 1080 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1081 | return true; |
4379bf8b | 1082 | if (req->task->files == files) |
08d23634 PB |
1083 | return true; |
1084 | } | |
1085 | return false; | |
1086 | } | |
1087 | ||
c40f6379 JA |
1088 | static inline void req_set_fail_links(struct io_kiocb *req) |
1089 | { | |
1090 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1091 | req->flags |= REQ_F_FAIL_LINK; | |
1092 | } | |
4a38aed2 | 1093 | |
2b188cc1 JA |
1094 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1095 | { | |
1096 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1097 | ||
0f158b4c | 1098 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1099 | } |
1100 | ||
8eb7e2d0 PB |
1101 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1102 | { | |
1103 | return !req->timeout.off; | |
1104 | } | |
1105 | ||
2b188cc1 JA |
1106 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1107 | { | |
1108 | struct io_ring_ctx *ctx; | |
78076bb6 | 1109 | int hash_bits; |
2b188cc1 JA |
1110 | |
1111 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1112 | if (!ctx) | |
1113 | return NULL; | |
1114 | ||
78076bb6 JA |
1115 | /* |
1116 | * Use 5 bits less than the max cq entries, that should give us around | |
1117 | * 32 entries per hash list if totally full and uniformly spread. | |
1118 | */ | |
1119 | hash_bits = ilog2(p->cq_entries); | |
1120 | hash_bits -= 5; | |
1121 | if (hash_bits <= 0) | |
1122 | hash_bits = 1; | |
1123 | ctx->cancel_hash_bits = hash_bits; | |
1124 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1125 | GFP_KERNEL); | |
1126 | if (!ctx->cancel_hash) | |
1127 | goto err; | |
1128 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1129 | ||
21482896 | 1130 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1131 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1132 | goto err; | |
2b188cc1 JA |
1133 | |
1134 | ctx->flags = p->flags; | |
90554200 | 1135 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1136 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1137 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1138 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1139 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1140 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1141 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1142 | mutex_init(&ctx->uring_lock); |
1143 | init_waitqueue_head(&ctx->wait); | |
1144 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1145 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1146 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1147 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1148 | spin_lock_init(&ctx->inflight_lock); |
1149 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1150 | spin_lock_init(&ctx->rsrc_ref_lock); |
1151 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1152 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1153 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1154 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1155 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1156 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1157 | return ctx; |
206aefde | 1158 | err: |
78076bb6 | 1159 | kfree(ctx->cancel_hash); |
206aefde JA |
1160 | kfree(ctx); |
1161 | return NULL; | |
2b188cc1 JA |
1162 | } |
1163 | ||
9cf7c104 | 1164 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1165 | { |
2bc9930e JA |
1166 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1167 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1168 | |
9cf7c104 | 1169 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1170 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1171 | } |
de0617e4 | 1172 | |
9d858b21 | 1173 | return false; |
de0617e4 JA |
1174 | } |
1175 | ||
ce3d5aae PB |
1176 | static void io_req_track_inflight(struct io_kiocb *req) |
1177 | { | |
1178 | struct io_ring_ctx *ctx = req->ctx; | |
1179 | ||
1180 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1181 | req->flags |= REQ_F_INFLIGHT; |
1182 | ||
1183 | spin_lock_irq(&ctx->inflight_lock); | |
1184 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1185 | spin_unlock_irq(&ctx->inflight_lock); | |
1186 | } | |
1187 | } | |
1188 | ||
1e6fa521 JA |
1189 | static void io_prep_async_work(struct io_kiocb *req) |
1190 | { | |
1191 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1192 | struct io_ring_ctx *ctx = req->ctx; |
1193 | ||
003e8dcc JA |
1194 | if (!req->work.creds) |
1195 | req->work.creds = get_current_cred(); | |
1196 | ||
feaadc4f PB |
1197 | if (req->flags & REQ_F_FORCE_ASYNC) |
1198 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1199 | ||
1e6fa521 JA |
1200 | if (req->flags & REQ_F_ISREG) { |
1201 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1202 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1203 | } else { | |
1204 | if (def->unbound_nonreg_file) | |
1205 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1206 | } | |
561fb04a | 1207 | } |
cccf0ee8 | 1208 | |
cbdcb435 | 1209 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1210 | { |
cbdcb435 | 1211 | struct io_kiocb *cur; |
54a91f3b | 1212 | |
f2f87370 PB |
1213 | io_for_each_link(cur, req) |
1214 | io_prep_async_work(cur); | |
561fb04a JA |
1215 | } |
1216 | ||
ebf93667 | 1217 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1218 | { |
a197f664 | 1219 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1220 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1221 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1222 | |
3bfe6106 JA |
1223 | BUG_ON(!tctx); |
1224 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1225 | |
8766dd51 PB |
1226 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1227 | &req->work, req->flags); | |
cbdcb435 PB |
1228 | /* init ->work of the whole link before punting */ |
1229 | io_prep_async_link(req); | |
ebf93667 | 1230 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1231 | if (link) |
1232 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1233 | } |
1234 | ||
5262f567 JA |
1235 | static void io_kill_timeout(struct io_kiocb *req) |
1236 | { | |
e8c2bc1f | 1237 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1238 | int ret; |
1239 | ||
e8c2bc1f | 1240 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1241 | if (ret != -1) { |
01cec8c1 PB |
1242 | atomic_set(&req->ctx->cq_timeouts, |
1243 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1244 | list_del_init(&req->timeout.list); |
78e19bbe | 1245 | io_cqring_fill_event(req, 0); |
216578e5 | 1246 | io_put_req_deferred(req, 1); |
5262f567 JA |
1247 | } |
1248 | } | |
1249 | ||
76e1b642 JA |
1250 | /* |
1251 | * Returns true if we found and killed one or more timeouts | |
1252 | */ | |
6b81928d PB |
1253 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1254 | struct files_struct *files) | |
5262f567 JA |
1255 | { |
1256 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1257 | int canceled = 0; |
5262f567 JA |
1258 | |
1259 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1260 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1261 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1262 | io_kill_timeout(req); |
76e1b642 JA |
1263 | canceled++; |
1264 | } | |
f3606e3a | 1265 | } |
5262f567 | 1266 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1267 | return canceled != 0; |
5262f567 JA |
1268 | } |
1269 | ||
04518945 | 1270 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1271 | { |
04518945 | 1272 | do { |
27dc8338 PB |
1273 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1274 | struct io_defer_entry, list); | |
de0617e4 | 1275 | |
9cf7c104 | 1276 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1277 | break; |
27dc8338 | 1278 | list_del_init(&de->list); |
907d1df3 | 1279 | io_req_task_queue(de->req); |
27dc8338 | 1280 | kfree(de); |
04518945 PB |
1281 | } while (!list_empty(&ctx->defer_list)); |
1282 | } | |
1283 | ||
360428f8 | 1284 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1285 | { |
f010505b MDG |
1286 | u32 seq; |
1287 | ||
1288 | if (list_empty(&ctx->timeout_list)) | |
1289 | return; | |
1290 | ||
1291 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1292 | ||
1293 | do { | |
1294 | u32 events_needed, events_got; | |
360428f8 | 1295 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1296 | struct io_kiocb, timeout.list); |
de0617e4 | 1297 | |
8eb7e2d0 | 1298 | if (io_is_timeout_noseq(req)) |
360428f8 | 1299 | break; |
f010505b MDG |
1300 | |
1301 | /* | |
1302 | * Since seq can easily wrap around over time, subtract | |
1303 | * the last seq at which timeouts were flushed before comparing. | |
1304 | * Assuming not more than 2^31-1 events have happened since, | |
1305 | * these subtractions won't have wrapped, so we can check if | |
1306 | * target is in [last_seq, current_seq] by comparing the two. | |
1307 | */ | |
1308 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1309 | events_got = seq - ctx->cq_last_tm_flush; | |
1310 | if (events_got < events_needed) | |
360428f8 | 1311 | break; |
bfe68a22 | 1312 | |
135fcde8 | 1313 | list_del_init(&req->timeout.list); |
5262f567 | 1314 | io_kill_timeout(req); |
f010505b MDG |
1315 | } while (!list_empty(&ctx->timeout_list)); |
1316 | ||
1317 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1318 | } |
5262f567 | 1319 | |
360428f8 PB |
1320 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1321 | { | |
1322 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1323 | |
1324 | /* order cqe stores with ring update */ | |
1325 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1326 | |
04518945 PB |
1327 | if (unlikely(!list_empty(&ctx->defer_list))) |
1328 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1329 | } |
1330 | ||
90554200 JA |
1331 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1332 | { | |
1333 | struct io_rings *r = ctx->rings; | |
1334 | ||
1335 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1336 | } | |
1337 | ||
888aae2e PB |
1338 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1339 | { | |
1340 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1341 | } | |
1342 | ||
2b188cc1 JA |
1343 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1344 | { | |
75b28aff | 1345 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1346 | unsigned tail; |
1347 | ||
115e12e5 SB |
1348 | /* |
1349 | * writes to the cq entry need to come after reading head; the | |
1350 | * control dependency is enough as we're using WRITE_ONCE to | |
1351 | * fill the cq entry | |
1352 | */ | |
888aae2e | 1353 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1354 | return NULL; |
1355 | ||
888aae2e | 1356 | tail = ctx->cached_cq_tail++; |
75b28aff | 1357 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1358 | } |
1359 | ||
f2842ab5 JA |
1360 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1361 | { | |
f0b493e6 JA |
1362 | if (!ctx->cq_ev_fd) |
1363 | return false; | |
7e55a19c SG |
1364 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1365 | return false; | |
f2842ab5 JA |
1366 | if (!ctx->eventfd_async) |
1367 | return true; | |
b41e9852 | 1368 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1369 | } |
1370 | ||
b41e9852 | 1371 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1372 | { |
b1445e59 PB |
1373 | /* see waitqueue_active() comment */ |
1374 | smp_mb(); | |
1375 | ||
1d7bb1d5 JA |
1376 | if (waitqueue_active(&ctx->wait)) |
1377 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1378 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1379 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1380 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1381 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1382 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1383 | wake_up_interruptible(&ctx->cq_wait); |
1384 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1385 | } | |
1d7bb1d5 JA |
1386 | } |
1387 | ||
80c18e4a PB |
1388 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1389 | { | |
b1445e59 PB |
1390 | /* see waitqueue_active() comment */ |
1391 | smp_mb(); | |
1392 | ||
80c18e4a PB |
1393 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1394 | if (waitqueue_active(&ctx->wait)) | |
1395 | wake_up(&ctx->wait); | |
1396 | } | |
1397 | if (io_should_trigger_evfd(ctx)) | |
1398 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1399 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1400 | wake_up_interruptible(&ctx->cq_wait); |
1401 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1402 | } | |
80c18e4a PB |
1403 | } |
1404 | ||
c4a2ed72 | 1405 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1406 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1407 | struct task_struct *tsk, | |
1408 | struct files_struct *files) | |
1d7bb1d5 JA |
1409 | { |
1410 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1411 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1412 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1413 | unsigned long flags; |
b18032bb | 1414 | bool all_flushed, posted; |
1d7bb1d5 JA |
1415 | LIST_HEAD(list); |
1416 | ||
e23de15f PB |
1417 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1418 | return false; | |
1d7bb1d5 | 1419 | |
b18032bb | 1420 | posted = false; |
1d7bb1d5 | 1421 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1422 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1423 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1424 | continue; |
1425 | ||
1d7bb1d5 JA |
1426 | cqe = io_get_cqring(ctx); |
1427 | if (!cqe && !force) | |
1428 | break; | |
1429 | ||
40d8ddd4 | 1430 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1431 | if (cqe) { |
1432 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1433 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1434 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1435 | } else { |
2c3bac6d | 1436 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1437 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1438 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1439 | } |
b18032bb | 1440 | posted = true; |
1d7bb1d5 JA |
1441 | } |
1442 | ||
09e88404 PB |
1443 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1444 | if (all_flushed) { | |
1445 | clear_bit(0, &ctx->sq_check_overflow); | |
1446 | clear_bit(0, &ctx->cq_check_overflow); | |
1447 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1448 | } | |
46930143 | 1449 | |
b18032bb JA |
1450 | if (posted) |
1451 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1452 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1453 | if (posted) |
1454 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1455 | |
1456 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1457 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1458 | list_del(&req->compl.list); | |
ec9c02ad | 1459 | io_put_req(req); |
1d7bb1d5 | 1460 | } |
c4a2ed72 | 1461 | |
09e88404 | 1462 | return all_flushed; |
1d7bb1d5 JA |
1463 | } |
1464 | ||
ca0a2651 | 1465 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1466 | struct task_struct *tsk, |
1467 | struct files_struct *files) | |
1468 | { | |
ca0a2651 JA |
1469 | bool ret = true; |
1470 | ||
6c503150 PB |
1471 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1472 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1473 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1474 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1475 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1476 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1477 | mutex_unlock(&ctx->uring_lock); | |
1478 | } | |
ca0a2651 JA |
1479 | |
1480 | return ret; | |
6c503150 PB |
1481 | } |
1482 | ||
bcda7baa | 1483 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1484 | { |
78e19bbe | 1485 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1486 | struct io_uring_cqe *cqe; |
1487 | ||
78e19bbe | 1488 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1489 | |
2b188cc1 JA |
1490 | /* |
1491 | * If we can't get a cq entry, userspace overflowed the | |
1492 | * submission (by quite a lot). Increment the overflow count in | |
1493 | * the ring. | |
1494 | */ | |
1495 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1496 | if (likely(cqe)) { |
78e19bbe | 1497 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1498 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1499 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1500 | } else if (ctx->cq_overflow_flushed || |
1501 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1502 | /* |
1503 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1504 | * then we cannot store the request for later flushing, we need | |
1505 | * to drop it on the floor. | |
1506 | */ | |
2c3bac6d PB |
1507 | ctx->cached_cq_overflow++; |
1508 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1509 | } else { |
ad3eb2c8 JA |
1510 | if (list_empty(&ctx->cq_overflow_list)) { |
1511 | set_bit(0, &ctx->sq_check_overflow); | |
1512 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1513 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1514 | } |
40d8ddd4 | 1515 | io_clean_op(req); |
1d7bb1d5 | 1516 | req->result = res; |
0f7e466b | 1517 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1518 | refcount_inc(&req->refs); |
1519 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1520 | } |
1521 | } | |
1522 | ||
bcda7baa JA |
1523 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1524 | { | |
1525 | __io_cqring_fill_event(req, res, 0); | |
1526 | } | |
1527 | ||
7a612350 PB |
1528 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1529 | unsigned int cflags) | |
2b188cc1 | 1530 | { |
78e19bbe | 1531 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1532 | unsigned long flags; |
1533 | ||
1534 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1535 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1536 | /* |
1537 | * If we're the last reference to this request, add to our locked | |
1538 | * free_list cache. | |
1539 | */ | |
1540 | if (refcount_dec_and_test(&req->refs)) { | |
1541 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1542 | ||
7a612350 PB |
1543 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1544 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1545 | io_disarm_next(req); | |
1546 | if (req->link) { | |
1547 | io_req_task_queue(req->link); | |
1548 | req->link = NULL; | |
1549 | } | |
1550 | } | |
c7dae4ba JA |
1551 | io_dismantle_req(req); |
1552 | io_put_task(req->task, 1); | |
1553 | list_add(&req->compl.list, &cs->locked_free_list); | |
1554 | cs->locked_free_nr++; | |
180f829f PB |
1555 | } else { |
1556 | if (!percpu_ref_tryget(&ctx->refs)) | |
1557 | req = NULL; | |
1558 | } | |
7a612350 | 1559 | io_commit_cqring(ctx); |
2b188cc1 | 1560 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1561 | |
180f829f PB |
1562 | if (req) { |
1563 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1564 | percpu_ref_put(&ctx->refs); |
180f829f | 1565 | } |
229a7b63 JA |
1566 | } |
1567 | ||
a38d68db | 1568 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1569 | unsigned int cflags) |
229a7b63 | 1570 | { |
a38d68db PB |
1571 | io_clean_op(req); |
1572 | req->result = res; | |
1573 | req->compl.cflags = cflags; | |
e342c807 | 1574 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1575 | } |
1576 | ||
889fca73 PB |
1577 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1578 | long res, unsigned cflags) | |
bcda7baa | 1579 | { |
889fca73 PB |
1580 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1581 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1582 | else |
c7dae4ba | 1583 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1584 | } |
1585 | ||
a38d68db | 1586 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1587 | { |
889fca73 | 1588 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1589 | } |
1590 | ||
c7dae4ba | 1591 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1592 | { |
c7dae4ba JA |
1593 | struct io_submit_state *state = &ctx->submit_state; |
1594 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1595 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1596 | |
c7dae4ba JA |
1597 | /* |
1598 | * If we have more than a batch's worth of requests in our IRQ side | |
1599 | * locked cache, grab the lock and move them over to our submission | |
1600 | * side cache. | |
1601 | */ | |
1602 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1603 | spin_lock_irq(&ctx->completion_lock); | |
1604 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1605 | cs->locked_free_nr = 0; | |
1606 | spin_unlock_irq(&ctx->completion_lock); | |
1607 | } | |
0ddf92e8 | 1608 | |
c7dae4ba JA |
1609 | while (!list_empty(&cs->free_list)) { |
1610 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1611 | compl.list); |
1612 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1613 | state->reqs[state->free_reqs++] = req; |
1614 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1615 | break; | |
1b4c351f JA |
1616 | } |
1617 | ||
e5d1bc0a | 1618 | return req != NULL; |
0ddf92e8 JA |
1619 | } |
1620 | ||
e5d1bc0a | 1621 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1622 | { |
e5d1bc0a PB |
1623 | struct io_submit_state *state = &ctx->submit_state; |
1624 | ||
1625 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1626 | ||
f6b6c7d6 | 1627 | if (!state->free_reqs) { |
291b2821 | 1628 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1629 | int ret; |
1630 | ||
c7dae4ba | 1631 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1632 | goto got_req; |
1633 | ||
bf019da7 PB |
1634 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1635 | state->reqs); | |
fd6fab2c JA |
1636 | |
1637 | /* | |
1638 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1639 | * retry single alloc to be on the safe side. | |
1640 | */ | |
1641 | if (unlikely(ret <= 0)) { | |
1642 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1643 | if (!state->reqs[0]) | |
3893f39f | 1644 | return NULL; |
fd6fab2c JA |
1645 | ret = 1; |
1646 | } | |
291b2821 | 1647 | state->free_reqs = ret; |
2b188cc1 | 1648 | } |
e5d1bc0a | 1649 | got_req: |
291b2821 PB |
1650 | state->free_reqs--; |
1651 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1652 | } |
1653 | ||
8da11c19 PB |
1654 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1655 | bool fixed) | |
1656 | { | |
36f72fe2 | 1657 | if (!fixed) |
8da11c19 PB |
1658 | fput(file); |
1659 | } | |
1660 | ||
4edf20f9 | 1661 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1662 | { |
3ca405eb | 1663 | io_clean_op(req); |
929a3af9 | 1664 | |
e8c2bc1f JA |
1665 | if (req->async_data) |
1666 | kfree(req->async_data); | |
8da11c19 PB |
1667 | if (req->file) |
1668 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1669 | if (req->fixed_rsrc_refs) |
1670 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1671 | if (req->work.creds) { |
1672 | put_cred(req->work.creds); | |
1673 | req->work.creds = NULL; | |
1674 | } | |
f85c310a PB |
1675 | |
1676 | if (req->flags & REQ_F_INFLIGHT) { | |
1677 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1678 | unsigned long flags; |
1679 | ||
1680 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1681 | list_del(&req->inflight_entry); | |
1682 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1683 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1684 | } |
e65ef56d JA |
1685 | } |
1686 | ||
b23fcf47 | 1687 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1688 | static inline void io_put_task(struct task_struct *task, int nr) |
1689 | { | |
1690 | struct io_uring_task *tctx = task->io_uring; | |
1691 | ||
1692 | percpu_counter_sub(&tctx->inflight, nr); | |
1693 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1694 | wake_up(&tctx->wait); | |
1695 | put_task_struct_many(task, nr); | |
1696 | } | |
1697 | ||
216578e5 | 1698 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1699 | { |
51a4cc11 | 1700 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1701 | |
216578e5 | 1702 | io_dismantle_req(req); |
7c660731 | 1703 | io_put_task(req->task, 1); |
c6ca97b3 | 1704 | |
3893f39f | 1705 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1706 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1707 | } |
1708 | ||
f2f87370 PB |
1709 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1710 | { | |
1711 | struct io_kiocb *nxt = req->link; | |
1712 | ||
1713 | req->link = nxt->link; | |
1714 | nxt->link = NULL; | |
1715 | } | |
1716 | ||
33cc89a9 PB |
1717 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1718 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1719 | { |
33cc89a9 | 1720 | struct io_kiocb *link = req->link; |
c9abd7ad | 1721 | bool cancelled = false; |
f2f87370 | 1722 | |
900fad45 PB |
1723 | /* |
1724 | * Can happen if a linked timeout fired and link had been like | |
1725 | * req -> link t-out -> link t-out [-> ...] | |
1726 | */ | |
c9abd7ad PB |
1727 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1728 | struct io_timeout_data *io = link->async_data; | |
1729 | int ret; | |
7c86ffee | 1730 | |
f2f87370 | 1731 | io_remove_next_linked(req); |
90cd7e42 | 1732 | link->timeout.head = NULL; |
c9abd7ad PB |
1733 | ret = hrtimer_try_to_cancel(&io->timer); |
1734 | if (ret != -1) { | |
1735 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1736 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1737 | cancelled = true; |
1738 | } | |
1739 | } | |
7c86ffee | 1740 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1741 | return cancelled; |
7c86ffee PB |
1742 | } |
1743 | ||
d148ca4b | 1744 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1745 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1746 | { |
33cc89a9 | 1747 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1748 | |
f2f87370 | 1749 | req->link = NULL; |
f2f87370 PB |
1750 | while (link) { |
1751 | nxt = link->link; | |
1752 | link->link = NULL; | |
2665abfd | 1753 | |
f2f87370 | 1754 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1755 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1756 | io_put_req_deferred(link, 2); |
f2f87370 | 1757 | link = nxt; |
9e645e11 | 1758 | } |
33cc89a9 | 1759 | } |
9e645e11 | 1760 | |
33cc89a9 PB |
1761 | static bool io_disarm_next(struct io_kiocb *req) |
1762 | __must_hold(&req->ctx->completion_lock) | |
1763 | { | |
1764 | bool posted = false; | |
1765 | ||
1766 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1767 | posted = io_kill_linked_timeout(req); | |
1768 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1769 | posted |= (req->link != NULL); | |
1770 | io_fail_links(req); | |
1771 | } | |
1772 | return posted; | |
9e645e11 JA |
1773 | } |
1774 | ||
3fa5e0f3 | 1775 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1776 | { |
33cc89a9 | 1777 | struct io_kiocb *nxt; |
944e58bf | 1778 | |
9e645e11 JA |
1779 | /* |
1780 | * If LINK is set, we have dependent requests in this chain. If we | |
1781 | * didn't fail this request, queue the first one up, moving any other | |
1782 | * dependencies to the next request. In case of failure, fail the rest | |
1783 | * of the chain. | |
1784 | */ | |
33cc89a9 PB |
1785 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1786 | struct io_ring_ctx *ctx = req->ctx; | |
1787 | unsigned long flags; | |
1788 | bool posted; | |
1789 | ||
1790 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1791 | posted = io_disarm_next(req); | |
1792 | if (posted) | |
1793 | io_commit_cqring(req->ctx); | |
1794 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1795 | if (posted) | |
1796 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1797 | } |
33cc89a9 PB |
1798 | nxt = req->link; |
1799 | req->link = NULL; | |
1800 | return nxt; | |
4d7dd462 | 1801 | } |
9e645e11 | 1802 | |
f2f87370 | 1803 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1804 | { |
cdbff982 | 1805 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1806 | return NULL; |
1807 | return __io_req_find_next(req); | |
1808 | } | |
1809 | ||
2c32395d PB |
1810 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1811 | { | |
1812 | if (!ctx) | |
1813 | return; | |
1814 | if (ctx->submit_state.comp.nr) { | |
1815 | mutex_lock(&ctx->uring_lock); | |
1816 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1817 | mutex_unlock(&ctx->uring_lock); | |
1818 | } | |
1819 | percpu_ref_put(&ctx->refs); | |
1820 | } | |
1821 | ||
7cbf1722 | 1822 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1823 | { |
65453d1e | 1824 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1825 | struct io_wq_work_list list; |
1826 | struct io_wq_work_node *node; | |
c2c4c83c | 1827 | |
7cbf1722 JA |
1828 | if (wq_list_empty(&tctx->task_list)) |
1829 | return false; | |
6200b0ae | 1830 | |
0b81e80c | 1831 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1832 | list = tctx->task_list; |
1833 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1834 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1835 | |
7cbf1722 JA |
1836 | node = list.first; |
1837 | while (node) { | |
1838 | struct io_wq_work_node *next = node->next; | |
1839 | struct io_kiocb *req; | |
0ba9c9ed | 1840 | |
7cbf1722 | 1841 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1842 | if (req->ctx != ctx) { |
1843 | ctx_flush_and_put(ctx); | |
1844 | ctx = req->ctx; | |
1845 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1846 | } |
65453d1e | 1847 | |
2c32395d PB |
1848 | req->task_work.func(&req->task_work); |
1849 | node = next; | |
7cbf1722 JA |
1850 | } |
1851 | ||
2c32395d | 1852 | ctx_flush_and_put(ctx); |
7cbf1722 | 1853 | return list.first != NULL; |
c2c4c83c JA |
1854 | } |
1855 | ||
7cbf1722 | 1856 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1857 | { |
7cbf1722 | 1858 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1859 | |
1d5f360d JA |
1860 | clear_bit(0, &tctx->task_state); |
1861 | ||
7cbf1722 JA |
1862 | while (__tctx_task_work(tctx)) |
1863 | cond_resched(); | |
7cbf1722 JA |
1864 | } |
1865 | ||
1866 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1867 | enum task_work_notify_mode notify) | |
1868 | { | |
1869 | struct io_uring_task *tctx = tsk->io_uring; | |
1870 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1871 | unsigned long flags; |
7cbf1722 JA |
1872 | int ret; |
1873 | ||
1874 | WARN_ON_ONCE(!tctx); | |
1875 | ||
0b81e80c | 1876 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1877 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1878 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1879 | |
1880 | /* task_work already pending, we're done */ | |
1881 | if (test_bit(0, &tctx->task_state) || | |
1882 | test_and_set_bit(0, &tctx->task_state)) | |
1883 | return 0; | |
1884 | ||
1885 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1886 | return 0; | |
1887 | ||
1888 | /* | |
1889 | * Slow path - we failed, find and delete work. if the work is not | |
1890 | * in the list, it got run and we're fine. | |
1891 | */ | |
1892 | ret = 0; | |
0b81e80c | 1893 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1894 | wq_list_for_each(node, prev, &tctx->task_list) { |
1895 | if (&req->io_task_work.node == node) { | |
1896 | wq_list_del(&tctx->task_list, node, prev); | |
1897 | ret = 1; | |
1898 | break; | |
1899 | } | |
1900 | } | |
0b81e80c | 1901 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1902 | clear_bit(0, &tctx->task_state); |
1903 | return ret; | |
1904 | } | |
1905 | ||
355fb9e2 | 1906 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1907 | { |
1908 | struct task_struct *tsk = req->task; | |
1909 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1910 | enum task_work_notify_mode notify; |
1911 | int ret; | |
c2c4c83c | 1912 | |
6200b0ae JA |
1913 | if (tsk->flags & PF_EXITING) |
1914 | return -ESRCH; | |
1915 | ||
c2c4c83c | 1916 | /* |
0ba9c9ed JA |
1917 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1918 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1919 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1920 | * will do the job. | |
c2c4c83c | 1921 | */ |
91989c70 | 1922 | notify = TWA_NONE; |
355fb9e2 | 1923 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1924 | notify = TWA_SIGNAL; |
1925 | ||
7cbf1722 | 1926 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1927 | if (!ret) |
1928 | wake_up_process(tsk); | |
0ba9c9ed | 1929 | |
c2c4c83c JA |
1930 | return ret; |
1931 | } | |
1932 | ||
9b465711 PB |
1933 | static bool io_run_task_work_head(struct callback_head **work_head) |
1934 | { | |
1935 | struct callback_head *work, *next; | |
1936 | bool executed = false; | |
1937 | ||
1938 | do { | |
1939 | work = xchg(work_head, NULL); | |
1940 | if (!work) | |
1941 | break; | |
1942 | ||
1943 | do { | |
1944 | next = work->next; | |
1945 | work->func(work); | |
1946 | work = next; | |
1947 | cond_resched(); | |
1948 | } while (work); | |
1949 | executed = true; | |
1950 | } while (1); | |
1951 | ||
1952 | return executed; | |
1953 | } | |
1954 | ||
1955 | static void io_task_work_add_head(struct callback_head **work_head, | |
1956 | struct callback_head *task_work) | |
eab30c4d | 1957 | { |
7c25c0d1 | 1958 | struct callback_head *head; |
eab30c4d | 1959 | |
7c25c0d1 | 1960 | do { |
9b465711 PB |
1961 | head = READ_ONCE(*work_head); |
1962 | task_work->next = head; | |
1963 | } while (cmpxchg(work_head, head, task_work) != head); | |
1964 | } | |
1965 | ||
1966 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
1967 | task_work_func_t cb) | |
1968 | { | |
1969 | init_task_work(&req->task_work, cb); | |
1970 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
1971 | } |
1972 | ||
c40f6379 JA |
1973 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1974 | { | |
1975 | struct io_ring_ctx *ctx = req->ctx; | |
1976 | ||
1977 | spin_lock_irq(&ctx->completion_lock); | |
1978 | io_cqring_fill_event(req, error); | |
1979 | io_commit_cqring(ctx); | |
1980 | spin_unlock_irq(&ctx->completion_lock); | |
1981 | ||
1982 | io_cqring_ev_posted(ctx); | |
1983 | req_set_fail_links(req); | |
1984 | io_double_put_req(req); | |
1985 | } | |
1986 | ||
1987 | static void io_req_task_cancel(struct callback_head *cb) | |
1988 | { | |
1989 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1990 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1991 | |
792bb6eb | 1992 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1993 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1994 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1995 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1996 | } |
1997 | ||
1998 | static void __io_req_task_submit(struct io_kiocb *req) | |
1999 | { | |
2000 | struct io_ring_ctx *ctx = req->ctx; | |
2001 | ||
04fc6c80 | 2002 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2003 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2004 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2005 | __io_queue_sqe(req); |
81b6d05c | 2006 | else |
c40f6379 | 2007 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2008 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2009 | } |
2010 | ||
2011 | static void io_req_task_submit(struct callback_head *cb) | |
2012 | { | |
2013 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2014 | ||
2015 | __io_req_task_submit(req); | |
2016 | } | |
2017 | ||
2018 | static void io_req_task_queue(struct io_kiocb *req) | |
2019 | { | |
c40f6379 JA |
2020 | int ret; |
2021 | ||
7cbf1722 | 2022 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2023 | ret = io_req_task_work_add(req); |
c40f6379 | 2024 | if (unlikely(ret)) { |
a3df7698 | 2025 | req->result = -ECANCELED; |
04fc6c80 | 2026 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 2027 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 2028 | } |
c40f6379 JA |
2029 | } |
2030 | ||
a3df7698 PB |
2031 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2032 | { | |
2033 | percpu_ref_get(&req->ctx->refs); | |
2034 | req->result = ret; | |
2035 | req->task_work.func = io_req_task_cancel; | |
2036 | ||
2037 | if (unlikely(io_req_task_work_add(req))) | |
2038 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2039 | } | |
2040 | ||
f2f87370 | 2041 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2042 | { |
9b5f7bd9 | 2043 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2044 | |
2045 | if (nxt) | |
906a8c3f | 2046 | io_req_task_queue(nxt); |
c69f8dbe JL |
2047 | } |
2048 | ||
c3524383 | 2049 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2050 | { |
c3524383 PB |
2051 | io_queue_next(req); |
2052 | __io_free_req(req); | |
2053 | } | |
8766dd51 | 2054 | |
2d6500d4 | 2055 | struct req_batch { |
5af1d13e PB |
2056 | struct task_struct *task; |
2057 | int task_refs; | |
1b4c351f | 2058 | int ctx_refs; |
2d6500d4 PB |
2059 | }; |
2060 | ||
5af1d13e PB |
2061 | static inline void io_init_req_batch(struct req_batch *rb) |
2062 | { | |
5af1d13e | 2063 | rb->task_refs = 0; |
9ae72463 | 2064 | rb->ctx_refs = 0; |
5af1d13e PB |
2065 | rb->task = NULL; |
2066 | } | |
2067 | ||
2d6500d4 PB |
2068 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2069 | struct req_batch *rb) | |
2070 | { | |
6e833d53 | 2071 | if (rb->task) |
7c660731 | 2072 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2073 | if (rb->ctx_refs) |
2074 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2075 | } |
2076 | ||
6ff119a6 PB |
2077 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2078 | struct io_submit_state *state) | |
2d6500d4 | 2079 | { |
f2f87370 | 2080 | io_queue_next(req); |
2d6500d4 | 2081 | |
e3bc8e9d | 2082 | if (req->task != rb->task) { |
7c660731 PB |
2083 | if (rb->task) |
2084 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2085 | rb->task = req->task; |
2086 | rb->task_refs = 0; | |
5af1d13e | 2087 | } |
e3bc8e9d | 2088 | rb->task_refs++; |
9ae72463 | 2089 | rb->ctx_refs++; |
5af1d13e | 2090 | |
4edf20f9 | 2091 | io_dismantle_req(req); |
bd759045 | 2092 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2093 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2094 | else |
2095 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2096 | } |
2097 | ||
905c172f PB |
2098 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2099 | struct io_ring_ctx *ctx) | |
2100 | { | |
2101 | int i, nr = cs->nr; | |
2102 | struct io_kiocb *req; | |
2103 | struct req_batch rb; | |
2104 | ||
2105 | io_init_req_batch(&rb); | |
2106 | spin_lock_irq(&ctx->completion_lock); | |
2107 | for (i = 0; i < nr; i++) { | |
2108 | req = cs->reqs[i]; | |
2109 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2110 | } | |
2111 | io_commit_cqring(ctx); | |
2112 | spin_unlock_irq(&ctx->completion_lock); | |
2113 | ||
2114 | io_cqring_ev_posted(ctx); | |
2115 | for (i = 0; i < nr; i++) { | |
2116 | req = cs->reqs[i]; | |
2117 | ||
2118 | /* submission and completion refs */ | |
2119 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2120 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2121 | } |
2122 | ||
2123 | io_req_free_batch_finish(ctx, &rb); | |
2124 | cs->nr = 0; | |
7a743e22 PB |
2125 | } |
2126 | ||
ba816ad6 JA |
2127 | /* |
2128 | * Drop reference to request, return next in chain (if there is one) if this | |
2129 | * was the last reference to this request. | |
2130 | */ | |
9b5f7bd9 | 2131 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2132 | { |
9b5f7bd9 PB |
2133 | struct io_kiocb *nxt = NULL; |
2134 | ||
2a44f467 | 2135 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2136 | nxt = io_req_find_next(req); |
4d7dd462 | 2137 | __io_free_req(req); |
2a44f467 | 2138 | } |
9b5f7bd9 | 2139 | return nxt; |
2b188cc1 JA |
2140 | } |
2141 | ||
e65ef56d JA |
2142 | static void io_put_req(struct io_kiocb *req) |
2143 | { | |
2144 | if (refcount_dec_and_test(&req->refs)) | |
2145 | io_free_req(req); | |
2b188cc1 JA |
2146 | } |
2147 | ||
216578e5 PB |
2148 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2149 | { | |
2150 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2151 | ||
2152 | io_free_req(req); | |
2153 | } | |
2154 | ||
2155 | static void io_free_req_deferred(struct io_kiocb *req) | |
2156 | { | |
2157 | int ret; | |
2158 | ||
7cbf1722 | 2159 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2160 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2161 | if (unlikely(ret)) |
2162 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2163 | } |
2164 | ||
2165 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2166 | { | |
2167 | if (refcount_sub_and_test(refs, &req->refs)) | |
2168 | io_free_req_deferred(req); | |
2169 | } | |
2170 | ||
978db57e JA |
2171 | static void io_double_put_req(struct io_kiocb *req) |
2172 | { | |
2173 | /* drop both submit and complete references */ | |
2174 | if (refcount_sub_and_test(2, &req->refs)) | |
2175 | io_free_req(req); | |
2176 | } | |
2177 | ||
6c503150 | 2178 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2179 | { |
2180 | /* See comment at the top of this file */ | |
2181 | smp_rmb(); | |
e23de15f | 2182 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2183 | } |
2184 | ||
fb5ccc98 PB |
2185 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2186 | { | |
2187 | struct io_rings *rings = ctx->rings; | |
2188 | ||
2189 | /* make sure SQ entry isn't read before tail */ | |
2190 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2191 | } | |
2192 | ||
8ff069bf | 2193 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2194 | { |
8ff069bf | 2195 | unsigned int cflags; |
e94f141b | 2196 | |
bcda7baa JA |
2197 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2198 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2199 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2200 | kfree(kbuf); |
2201 | return cflags; | |
e94f141b JA |
2202 | } |
2203 | ||
8ff069bf | 2204 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2205 | { |
4d954c25 | 2206 | struct io_buffer *kbuf; |
bcda7baa | 2207 | |
4d954c25 | 2208 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2209 | return io_put_kbuf(req, kbuf); |
2210 | } | |
2211 | ||
4c6e277c JA |
2212 | static inline bool io_run_task_work(void) |
2213 | { | |
6200b0ae JA |
2214 | /* |
2215 | * Not safe to run on exiting task, and the task_work handling will | |
2216 | * not add work to such a task. | |
2217 | */ | |
2218 | if (unlikely(current->flags & PF_EXITING)) | |
2219 | return false; | |
4c6e277c JA |
2220 | if (current->task_works) { |
2221 | __set_current_state(TASK_RUNNING); | |
2222 | task_work_run(); | |
2223 | return true; | |
2224 | } | |
2225 | ||
2226 | return false; | |
bcda7baa JA |
2227 | } |
2228 | ||
def596e9 JA |
2229 | /* |
2230 | * Find and free completed poll iocbs | |
2231 | */ | |
2232 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2233 | struct list_head *done) | |
2234 | { | |
8237e045 | 2235 | struct req_batch rb; |
def596e9 | 2236 | struct io_kiocb *req; |
bbde017a XW |
2237 | |
2238 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2239 | smp_rmb(); | |
def596e9 | 2240 | |
5af1d13e | 2241 | io_init_req_batch(&rb); |
def596e9 | 2242 | while (!list_empty(done)) { |
bcda7baa JA |
2243 | int cflags = 0; |
2244 | ||
d21ffe7e | 2245 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2246 | list_del(&req->inflight_entry); |
2247 | ||
bbde017a XW |
2248 | if (READ_ONCE(req->result) == -EAGAIN) { |
2249 | req->iopoll_completed = 0; | |
23faba36 | 2250 | if (io_rw_reissue(req)) |
f161340d | 2251 | continue; |
bbde017a | 2252 | } |
def596e9 | 2253 | |
bcda7baa | 2254 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2255 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2256 | |
2257 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2258 | (*nr_events)++; |
2259 | ||
c3524383 | 2260 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2261 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2262 | } |
def596e9 | 2263 | |
09bb8394 | 2264 | io_commit_cqring(ctx); |
80c18e4a | 2265 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2266 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2267 | } |
2268 | ||
def596e9 JA |
2269 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2270 | long min) | |
2271 | { | |
2272 | struct io_kiocb *req, *tmp; | |
2273 | LIST_HEAD(done); | |
2274 | bool spin; | |
2275 | int ret; | |
2276 | ||
2277 | /* | |
2278 | * Only spin for completions if we don't have multiple devices hanging | |
2279 | * off our complete list, and we're under the requested amount. | |
2280 | */ | |
2281 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2282 | ||
2283 | ret = 0; | |
d21ffe7e | 2284 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2285 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2286 | |
2287 | /* | |
581f9810 BM |
2288 | * Move completed and retryable entries to our local lists. |
2289 | * If we find a request that requires polling, break out | |
2290 | * and complete those lists first, if we have entries there. | |
def596e9 | 2291 | */ |
65a6543d | 2292 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2293 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2294 | continue; |
2295 | } | |
2296 | if (!list_empty(&done)) | |
2297 | break; | |
2298 | ||
2299 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2300 | if (ret < 0) | |
2301 | break; | |
2302 | ||
3aadc23e PB |
2303 | /* iopoll may have completed current req */ |
2304 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2305 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2306 | |
def596e9 JA |
2307 | if (ret && spin) |
2308 | spin = false; | |
2309 | ret = 0; | |
2310 | } | |
2311 | ||
2312 | if (!list_empty(&done)) | |
2313 | io_iopoll_complete(ctx, nr_events, &done); | |
2314 | ||
2315 | return ret; | |
2316 | } | |
2317 | ||
2318 | /* | |
d195a66e | 2319 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2320 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2321 | * as a non-spinning completion check. | |
2322 | */ | |
2323 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2324 | long min) | |
2325 | { | |
540e32a0 | 2326 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2327 | int ret; |
2328 | ||
2329 | ret = io_do_iopoll(ctx, nr_events, min); | |
2330 | if (ret < 0) | |
2331 | return ret; | |
eba0a4dd | 2332 | if (*nr_events >= min) |
def596e9 JA |
2333 | return 0; |
2334 | } | |
2335 | ||
2336 | return 1; | |
2337 | } | |
2338 | ||
2339 | /* | |
2340 | * We can't just wait for polled events to come to us, we have to actively | |
2341 | * find and complete them. | |
2342 | */ | |
b2edc0a7 | 2343 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2344 | { |
2345 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2346 | return; | |
2347 | ||
2348 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2349 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2350 | unsigned int nr_events = 0; |
2351 | ||
b2edc0a7 | 2352 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2353 | |
b2edc0a7 PB |
2354 | /* let it sleep and repeat later if can't complete a request */ |
2355 | if (nr_events == 0) | |
2356 | break; | |
08f5439f JA |
2357 | /* |
2358 | * Ensure we allow local-to-the-cpu processing to take place, | |
2359 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2360 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2361 | */ |
3fcee5a6 PB |
2362 | if (need_resched()) { |
2363 | mutex_unlock(&ctx->uring_lock); | |
2364 | cond_resched(); | |
2365 | mutex_lock(&ctx->uring_lock); | |
2366 | } | |
def596e9 JA |
2367 | } |
2368 | mutex_unlock(&ctx->uring_lock); | |
2369 | } | |
2370 | ||
7668b92a | 2371 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2372 | { |
7668b92a | 2373 | unsigned int nr_events = 0; |
2b2ed975 | 2374 | int iters = 0, ret = 0; |
500f9fba | 2375 | |
c7849be9 XW |
2376 | /* |
2377 | * We disallow the app entering submit/complete with polling, but we | |
2378 | * still need to lock the ring to prevent racing with polled issue | |
2379 | * that got punted to a workqueue. | |
2380 | */ | |
2381 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2382 | do { |
a3a0e43f JA |
2383 | /* |
2384 | * Don't enter poll loop if we already have events pending. | |
2385 | * If we do, we can potentially be spinning for commands that | |
2386 | * already triggered a CQE (eg in error). | |
2387 | */ | |
6c503150 PB |
2388 | if (test_bit(0, &ctx->cq_check_overflow)) |
2389 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2390 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2391 | break; |
2392 | ||
500f9fba JA |
2393 | /* |
2394 | * If a submit got punted to a workqueue, we can have the | |
2395 | * application entering polling for a command before it gets | |
2396 | * issued. That app will hold the uring_lock for the duration | |
2397 | * of the poll right here, so we need to take a breather every | |
2398 | * now and then to ensure that the issue has a chance to add | |
2399 | * the poll to the issued list. Otherwise we can spin here | |
2400 | * forever, while the workqueue is stuck trying to acquire the | |
2401 | * very same mutex. | |
2402 | */ | |
2403 | if (!(++iters & 7)) { | |
2404 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2405 | io_run_task_work(); |
500f9fba JA |
2406 | mutex_lock(&ctx->uring_lock); |
2407 | } | |
2408 | ||
7668b92a | 2409 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2410 | if (ret <= 0) |
2411 | break; | |
2412 | ret = 0; | |
7668b92a | 2413 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2414 | |
500f9fba | 2415 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2416 | return ret; |
2417 | } | |
2418 | ||
491381ce | 2419 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2420 | { |
491381ce JA |
2421 | /* |
2422 | * Tell lockdep we inherited freeze protection from submission | |
2423 | * thread. | |
2424 | */ | |
2425 | if (req->flags & REQ_F_ISREG) { | |
2426 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2427 | |
491381ce | 2428 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2429 | } |
491381ce | 2430 | file_end_write(req->file); |
2b188cc1 JA |
2431 | } |
2432 | ||
b63534c4 | 2433 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2434 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2435 | { |
2436 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2437 | int rw, ret; |
b63534c4 | 2438 | struct iov_iter iter; |
b63534c4 | 2439 | |
dc2a6e9a PB |
2440 | /* already prepared */ |
2441 | if (req->async_data) | |
2442 | return true; | |
b63534c4 JA |
2443 | |
2444 | switch (req->opcode) { | |
2445 | case IORING_OP_READV: | |
2446 | case IORING_OP_READ_FIXED: | |
2447 | case IORING_OP_READ: | |
2448 | rw = READ; | |
2449 | break; | |
2450 | case IORING_OP_WRITEV: | |
2451 | case IORING_OP_WRITE_FIXED: | |
2452 | case IORING_OP_WRITE: | |
2453 | rw = WRITE; | |
2454 | break; | |
2455 | default: | |
2456 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2457 | req->opcode); | |
dc2a6e9a | 2458 | return false; |
b63534c4 JA |
2459 | } |
2460 | ||
dc2a6e9a PB |
2461 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2462 | if (ret < 0) | |
2463 | return false; | |
6bf985dc | 2464 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2465 | } |
b63534c4 | 2466 | |
3e6a0d3c | 2467 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2468 | { |
355afaeb | 2469 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2470 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2471 | |
355afaeb JA |
2472 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2473 | return false; | |
3e6a0d3c JA |
2474 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2475 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2476 | return false; |
7c977a58 JA |
2477 | /* |
2478 | * If ref is dying, we might be running poll reap from the exit work. | |
2479 | * Don't attempt to reissue from that path, just let it fail with | |
2480 | * -EAGAIN. | |
2481 | */ | |
3e6a0d3c JA |
2482 | if (percpu_ref_is_dying(&ctx->refs)) |
2483 | return false; | |
2484 | return true; | |
2485 | } | |
2486 | #endif | |
2487 | ||
2488 | static bool io_rw_reissue(struct io_kiocb *req) | |
2489 | { | |
2490 | #ifdef CONFIG_BLOCK | |
2491 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2492 | return false; |
b63534c4 | 2493 | |
55e6ac1e PB |
2494 | lockdep_assert_held(&req->ctx->uring_lock); |
2495 | ||
37d1e2e3 | 2496 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2497 | refcount_inc(&req->refs); |
2498 | io_queue_async_work(req); | |
b63534c4 | 2499 | return true; |
fdee946d | 2500 | } |
dc2a6e9a | 2501 | req_set_fail_links(req); |
b63534c4 JA |
2502 | #endif |
2503 | return false; | |
2504 | } | |
2505 | ||
a1d7c393 | 2506 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2507 | unsigned int issue_flags) |
a1d7c393 | 2508 | { |
2f8e45f1 PB |
2509 | int cflags = 0; |
2510 | ||
23faba36 PB |
2511 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2512 | return; | |
2f8e45f1 PB |
2513 | if (res != req->result) |
2514 | req_set_fail_links(req); | |
23faba36 | 2515 | |
2f8e45f1 PB |
2516 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2517 | kiocb_end_write(req); | |
2518 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2519 | cflags = io_put_rw_kbuf(req); | |
2520 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2521 | } |
2522 | ||
2523 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2524 | { | |
9adbd45d | 2525 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2526 | |
889fca73 | 2527 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2528 | } |
2529 | ||
def596e9 JA |
2530 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2531 | { | |
9adbd45d | 2532 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2533 | |
3e6a0d3c JA |
2534 | #ifdef CONFIG_BLOCK |
2535 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2536 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2537 | struct io_async_rw *rw = req->async_data; | |
2538 | ||
2539 | if (rw) | |
2540 | iov_iter_revert(&rw->iter, | |
2541 | req->result - iov_iter_count(&rw->iter)); | |
2542 | else if (!io_resubmit_prep(req)) | |
2543 | res = -EIO; | |
2544 | } | |
2545 | #endif | |
2546 | ||
491381ce JA |
2547 | if (kiocb->ki_flags & IOCB_WRITE) |
2548 | kiocb_end_write(req); | |
def596e9 | 2549 | |
2d7d6792 | 2550 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2551 | req_set_fail_links(req); |
bbde017a XW |
2552 | |
2553 | WRITE_ONCE(req->result, res); | |
2554 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2555 | smp_wmb(); |
2556 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2557 | } |
2558 | ||
2559 | /* | |
2560 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2561 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2562 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2563 | * accessing the kiocb cookie. | |
2564 | */ | |
2e9dbe90 | 2565 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2566 | { |
2567 | struct io_ring_ctx *ctx = req->ctx; | |
2568 | ||
2569 | /* | |
2570 | * Track whether we have multiple files in our lists. This will impact | |
2571 | * how we do polling eventually, not spinning if we're on potentially | |
2572 | * different devices. | |
2573 | */ | |
540e32a0 | 2574 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2575 | ctx->poll_multi_file = false; |
2576 | } else if (!ctx->poll_multi_file) { | |
2577 | struct io_kiocb *list_req; | |
2578 | ||
540e32a0 | 2579 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2580 | inflight_entry); |
9adbd45d | 2581 | if (list_req->file != req->file) |
def596e9 JA |
2582 | ctx->poll_multi_file = true; |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * For fast devices, IO may have already completed. If it has, add | |
2587 | * it to the front so we find it first. | |
2588 | */ | |
65a6543d | 2589 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2590 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2591 | else |
d21ffe7e | 2592 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2593 | |
2e9dbe90 XW |
2594 | /* |
2595 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2596 | * task context or in io worker task context. If current task context is | |
2597 | * sq thread, we don't need to check whether should wake up sq thread. | |
2598 | */ | |
2599 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2600 | wq_has_sleeper(&ctx->sq_data->wait)) |
2601 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2602 | } |
2603 | ||
9f13c35b PB |
2604 | static inline void io_state_file_put(struct io_submit_state *state) |
2605 | { | |
02b23a9a PB |
2606 | if (state->file_refs) { |
2607 | fput_many(state->file, state->file_refs); | |
2608 | state->file_refs = 0; | |
2609 | } | |
9a56a232 JA |
2610 | } |
2611 | ||
2612 | /* | |
2613 | * Get as many references to a file as we have IOs left in this submission, | |
2614 | * assuming most submissions are for one file, or at least that each file | |
2615 | * has more than one submission. | |
2616 | */ | |
8da11c19 | 2617 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2618 | { |
2619 | if (!state) | |
2620 | return fget(fd); | |
2621 | ||
6e1271e6 | 2622 | if (state->file_refs) { |
9a56a232 | 2623 | if (state->fd == fd) { |
6e1271e6 | 2624 | state->file_refs--; |
9a56a232 JA |
2625 | return state->file; |
2626 | } | |
02b23a9a | 2627 | io_state_file_put(state); |
9a56a232 JA |
2628 | } |
2629 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2630 | if (unlikely(!state->file)) |
9a56a232 JA |
2631 | return NULL; |
2632 | ||
2633 | state->fd = fd; | |
6e1271e6 | 2634 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2635 | return state->file; |
2636 | } | |
2637 | ||
4503b767 JA |
2638 | static bool io_bdev_nowait(struct block_device *bdev) |
2639 | { | |
9ba0d0c8 | 2640 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2641 | } |
2642 | ||
2b188cc1 JA |
2643 | /* |
2644 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2645 | * any file. For now, just ensure that anything potentially problematic is done | |
2646 | * inline. | |
2647 | */ | |
af197f50 | 2648 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2649 | { |
2650 | umode_t mode = file_inode(file)->i_mode; | |
2651 | ||
4503b767 | 2652 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2653 | if (IS_ENABLED(CONFIG_BLOCK) && |
2654 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2655 | return true; |
2656 | return false; | |
2657 | } | |
2658 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2659 | return true; |
4503b767 | 2660 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2661 | if (IS_ENABLED(CONFIG_BLOCK) && |
2662 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2663 | file->f_op != &io_uring_fops) |
2664 | return true; | |
2665 | return false; | |
2666 | } | |
2b188cc1 | 2667 | |
c5b85625 JA |
2668 | /* any ->read/write should understand O_NONBLOCK */ |
2669 | if (file->f_flags & O_NONBLOCK) | |
2670 | return true; | |
2671 | ||
af197f50 JA |
2672 | if (!(file->f_mode & FMODE_NOWAIT)) |
2673 | return false; | |
2674 | ||
2675 | if (rw == READ) | |
2676 | return file->f_op->read_iter != NULL; | |
2677 | ||
2678 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2679 | } |
2680 | ||
a88fc400 | 2681 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2682 | { |
def596e9 | 2683 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2684 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2685 | struct file *file = req->file; |
09bb8394 JA |
2686 | unsigned ioprio; |
2687 | int ret; | |
2b188cc1 | 2688 | |
75c668cd | 2689 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2690 | req->flags |= REQ_F_ISREG; |
2691 | ||
2b188cc1 | 2692 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2693 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2694 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2695 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2696 | } |
2b188cc1 | 2697 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2698 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2699 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2700 | if (unlikely(ret)) | |
2701 | return ret; | |
2b188cc1 | 2702 | |
75c668cd PB |
2703 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2704 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2705 | req->flags |= REQ_F_NOWAIT; | |
2706 | ||
2b188cc1 JA |
2707 | ioprio = READ_ONCE(sqe->ioprio); |
2708 | if (ioprio) { | |
2709 | ret = ioprio_check_cap(ioprio); | |
2710 | if (ret) | |
09bb8394 | 2711 | return ret; |
2b188cc1 JA |
2712 | |
2713 | kiocb->ki_ioprio = ioprio; | |
2714 | } else | |
2715 | kiocb->ki_ioprio = get_current_ioprio(); | |
2716 | ||
def596e9 | 2717 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2718 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2719 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2720 | return -EOPNOTSUPP; |
2b188cc1 | 2721 | |
def596e9 JA |
2722 | kiocb->ki_flags |= IOCB_HIPRI; |
2723 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2724 | req->iopoll_completed = 0; |
def596e9 | 2725 | } else { |
09bb8394 JA |
2726 | if (kiocb->ki_flags & IOCB_HIPRI) |
2727 | return -EINVAL; | |
def596e9 JA |
2728 | kiocb->ki_complete = io_complete_rw; |
2729 | } | |
9adbd45d | 2730 | |
3529d8c2 JA |
2731 | req->rw.addr = READ_ONCE(sqe->addr); |
2732 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2733 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2734 | return 0; |
2b188cc1 JA |
2735 | } |
2736 | ||
2737 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2738 | { | |
2739 | switch (ret) { | |
2740 | case -EIOCBQUEUED: | |
2741 | break; | |
2742 | case -ERESTARTSYS: | |
2743 | case -ERESTARTNOINTR: | |
2744 | case -ERESTARTNOHAND: | |
2745 | case -ERESTART_RESTARTBLOCK: | |
2746 | /* | |
2747 | * We can't just restart the syscall, since previously | |
2748 | * submitted sqes may already be in progress. Just fail this | |
2749 | * IO with EINTR. | |
2750 | */ | |
2751 | ret = -EINTR; | |
df561f66 | 2752 | fallthrough; |
2b188cc1 JA |
2753 | default: |
2754 | kiocb->ki_complete(kiocb, ret, 0); | |
2755 | } | |
2756 | } | |
2757 | ||
a1d7c393 | 2758 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2759 | unsigned int issue_flags) |
ba816ad6 | 2760 | { |
ba04291e | 2761 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2762 | struct io_async_rw *io = req->async_data; |
ba04291e | 2763 | |
227c0c96 | 2764 | /* add previously done IO, if any */ |
e8c2bc1f | 2765 | if (io && io->bytes_done > 0) { |
227c0c96 | 2766 | if (ret < 0) |
e8c2bc1f | 2767 | ret = io->bytes_done; |
227c0c96 | 2768 | else |
e8c2bc1f | 2769 | ret += io->bytes_done; |
227c0c96 JA |
2770 | } |
2771 | ||
ba04291e JA |
2772 | if (req->flags & REQ_F_CUR_POS) |
2773 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2774 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2775 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2776 | else |
2777 | io_rw_done(kiocb, ret); | |
2778 | } | |
2779 | ||
847595de | 2780 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2781 | { |
9adbd45d JA |
2782 | struct io_ring_ctx *ctx = req->ctx; |
2783 | size_t len = req->rw.len; | |
edafccee | 2784 | struct io_mapped_ubuf *imu; |
4be1c615 | 2785 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2786 | size_t offset; |
2787 | u64 buf_addr; | |
2788 | ||
edafccee JA |
2789 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2790 | return -EFAULT; | |
edafccee JA |
2791 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2792 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2793 | buf_addr = req->rw.addr; |
edafccee JA |
2794 | |
2795 | /* overflow */ | |
2796 | if (buf_addr + len < buf_addr) | |
2797 | return -EFAULT; | |
2798 | /* not inside the mapped region */ | |
2799 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2800 | return -EFAULT; | |
2801 | ||
2802 | /* | |
2803 | * May not be a start of buffer, set size appropriately | |
2804 | * and advance us to the beginning. | |
2805 | */ | |
2806 | offset = buf_addr - imu->ubuf; | |
2807 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2808 | |
2809 | if (offset) { | |
2810 | /* | |
2811 | * Don't use iov_iter_advance() here, as it's really slow for | |
2812 | * using the latter parts of a big fixed buffer - it iterates | |
2813 | * over each segment manually. We can cheat a bit here, because | |
2814 | * we know that: | |
2815 | * | |
2816 | * 1) it's a BVEC iter, we set it up | |
2817 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2818 | * first and last bvec | |
2819 | * | |
2820 | * So just find our index, and adjust the iterator afterwards. | |
2821 | * If the offset is within the first bvec (or the whole first | |
2822 | * bvec, just use iov_iter_advance(). This makes it easier | |
2823 | * since we can just skip the first segment, which may not | |
2824 | * be PAGE_SIZE aligned. | |
2825 | */ | |
2826 | const struct bio_vec *bvec = imu->bvec; | |
2827 | ||
2828 | if (offset <= bvec->bv_len) { | |
2829 | iov_iter_advance(iter, offset); | |
2830 | } else { | |
2831 | unsigned long seg_skip; | |
2832 | ||
2833 | /* skip first vec */ | |
2834 | offset -= bvec->bv_len; | |
2835 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2836 | ||
2837 | iter->bvec = bvec + seg_skip; | |
2838 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2839 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2840 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2841 | } |
2842 | } | |
2843 | ||
847595de | 2844 | return 0; |
edafccee JA |
2845 | } |
2846 | ||
bcda7baa JA |
2847 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2848 | { | |
2849 | if (needs_lock) | |
2850 | mutex_unlock(&ctx->uring_lock); | |
2851 | } | |
2852 | ||
2853 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2854 | { | |
2855 | /* | |
2856 | * "Normal" inline submissions always hold the uring_lock, since we | |
2857 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2858 | * The only exception is when we've detached the request and issue it | |
2859 | * from an async worker thread, grab the lock for that case. | |
2860 | */ | |
2861 | if (needs_lock) | |
2862 | mutex_lock(&ctx->uring_lock); | |
2863 | } | |
2864 | ||
2865 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2866 | int bgid, struct io_buffer *kbuf, | |
2867 | bool needs_lock) | |
2868 | { | |
2869 | struct io_buffer *head; | |
2870 | ||
2871 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2872 | return kbuf; | |
2873 | ||
2874 | io_ring_submit_lock(req->ctx, needs_lock); | |
2875 | ||
2876 | lockdep_assert_held(&req->ctx->uring_lock); | |
2877 | ||
9e15c3a0 | 2878 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2879 | if (head) { |
2880 | if (!list_empty(&head->list)) { | |
2881 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2882 | list); | |
2883 | list_del(&kbuf->list); | |
2884 | } else { | |
2885 | kbuf = head; | |
9e15c3a0 | 2886 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2887 | } |
2888 | if (*len > kbuf->len) | |
2889 | *len = kbuf->len; | |
2890 | } else { | |
2891 | kbuf = ERR_PTR(-ENOBUFS); | |
2892 | } | |
2893 | ||
2894 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2895 | ||
2896 | return kbuf; | |
2897 | } | |
2898 | ||
4d954c25 JA |
2899 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2900 | bool needs_lock) | |
2901 | { | |
2902 | struct io_buffer *kbuf; | |
4f4eeba8 | 2903 | u16 bgid; |
4d954c25 JA |
2904 | |
2905 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2906 | bgid = req->buf_index; |
4d954c25 JA |
2907 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2908 | if (IS_ERR(kbuf)) | |
2909 | return kbuf; | |
2910 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2911 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2912 | return u64_to_user_ptr(kbuf->addr); | |
2913 | } | |
2914 | ||
2915 | #ifdef CONFIG_COMPAT | |
2916 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2917 | bool needs_lock) | |
2918 | { | |
2919 | struct compat_iovec __user *uiov; | |
2920 | compat_ssize_t clen; | |
2921 | void __user *buf; | |
2922 | ssize_t len; | |
2923 | ||
2924 | uiov = u64_to_user_ptr(req->rw.addr); | |
2925 | if (!access_ok(uiov, sizeof(*uiov))) | |
2926 | return -EFAULT; | |
2927 | if (__get_user(clen, &uiov->iov_len)) | |
2928 | return -EFAULT; | |
2929 | if (clen < 0) | |
2930 | return -EINVAL; | |
2931 | ||
2932 | len = clen; | |
2933 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2934 | if (IS_ERR(buf)) | |
2935 | return PTR_ERR(buf); | |
2936 | iov[0].iov_base = buf; | |
2937 | iov[0].iov_len = (compat_size_t) len; | |
2938 | return 0; | |
2939 | } | |
2940 | #endif | |
2941 | ||
2942 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2943 | bool needs_lock) | |
2944 | { | |
2945 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2946 | void __user *buf; | |
2947 | ssize_t len; | |
2948 | ||
2949 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2950 | return -EFAULT; | |
2951 | ||
2952 | len = iov[0].iov_len; | |
2953 | if (len < 0) | |
2954 | return -EINVAL; | |
2955 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2956 | if (IS_ERR(buf)) | |
2957 | return PTR_ERR(buf); | |
2958 | iov[0].iov_base = buf; | |
2959 | iov[0].iov_len = len; | |
2960 | return 0; | |
2961 | } | |
2962 | ||
2963 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2964 | bool needs_lock) | |
2965 | { | |
dddb3e26 JA |
2966 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2967 | struct io_buffer *kbuf; | |
2968 | ||
2969 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2970 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2971 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2972 | return 0; |
dddb3e26 | 2973 | } |
dd201662 | 2974 | if (req->rw.len != 1) |
4d954c25 JA |
2975 | return -EINVAL; |
2976 | ||
2977 | #ifdef CONFIG_COMPAT | |
2978 | if (req->ctx->compat) | |
2979 | return io_compat_import(req, iov, needs_lock); | |
2980 | #endif | |
2981 | ||
2982 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2983 | } | |
2984 | ||
847595de PB |
2985 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2986 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2987 | { |
9adbd45d JA |
2988 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2989 | size_t sqe_len = req->rw.len; | |
847595de | 2990 | u8 opcode = req->opcode; |
4d954c25 | 2991 | ssize_t ret; |
edafccee | 2992 | |
7d009165 | 2993 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2994 | *iovec = NULL; |
9adbd45d | 2995 | return io_import_fixed(req, rw, iter); |
edafccee | 2996 | } |
2b188cc1 | 2997 | |
bcda7baa | 2998 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2999 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3000 | return -EINVAL; |
3001 | ||
3a6820f2 | 3002 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3003 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3004 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3005 | if (IS_ERR(buf)) |
4d954c25 | 3006 | return PTR_ERR(buf); |
3f9d6441 | 3007 | req->rw.len = sqe_len; |
bcda7baa JA |
3008 | } |
3009 | ||
3a6820f2 JA |
3010 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3011 | *iovec = NULL; | |
10fc72e4 | 3012 | return ret; |
3a6820f2 JA |
3013 | } |
3014 | ||
4d954c25 JA |
3015 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3016 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3017 | if (!ret) |
3018 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3019 | *iovec = NULL; |
3020 | return ret; | |
3021 | } | |
3022 | ||
89cd35c5 CH |
3023 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3024 | req->ctx->compat); | |
2b188cc1 JA |
3025 | } |
3026 | ||
0fef9483 JA |
3027 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3028 | { | |
5b09e37e | 3029 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3030 | } |
3031 | ||
31b51510 | 3032 | /* |
32960613 JA |
3033 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3034 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3035 | */ |
4017eb91 | 3036 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3037 | { |
4017eb91 JA |
3038 | struct kiocb *kiocb = &req->rw.kiocb; |
3039 | struct file *file = req->file; | |
32960613 JA |
3040 | ssize_t ret = 0; |
3041 | ||
3042 | /* | |
3043 | * Don't support polled IO through this interface, and we can't | |
3044 | * support non-blocking either. For the latter, this just causes | |
3045 | * the kiocb to be handled from an async context. | |
3046 | */ | |
3047 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3048 | return -EOPNOTSUPP; | |
3049 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3050 | return -EAGAIN; | |
3051 | ||
3052 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3053 | struct iovec iovec; |
32960613 JA |
3054 | ssize_t nr; |
3055 | ||
311ae9e1 PB |
3056 | if (!iov_iter_is_bvec(iter)) { |
3057 | iovec = iov_iter_iovec(iter); | |
3058 | } else { | |
4017eb91 JA |
3059 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3060 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3061 | } |
3062 | ||
32960613 JA |
3063 | if (rw == READ) { |
3064 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3065 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3066 | } else { |
3067 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3068 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3069 | } |
3070 | ||
3071 | if (nr < 0) { | |
3072 | if (!ret) | |
3073 | ret = nr; | |
3074 | break; | |
3075 | } | |
3076 | ret += nr; | |
3077 | if (nr != iovec.iov_len) | |
3078 | break; | |
4017eb91 JA |
3079 | req->rw.len -= nr; |
3080 | req->rw.addr += nr; | |
32960613 JA |
3081 | iov_iter_advance(iter, nr); |
3082 | } | |
3083 | ||
3084 | return ret; | |
3085 | } | |
3086 | ||
ff6165b2 JA |
3087 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3088 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3089 | { |
e8c2bc1f | 3090 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3091 | |
ff6165b2 | 3092 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3093 | rw->free_iovec = iovec; |
227c0c96 | 3094 | rw->bytes_done = 0; |
ff6165b2 | 3095 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3096 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3097 | return; |
b64e3444 | 3098 | if (!iovec) { |
ff6165b2 JA |
3099 | unsigned iov_off = 0; |
3100 | ||
3101 | rw->iter.iov = rw->fast_iov; | |
3102 | if (iter->iov != fast_iov) { | |
3103 | iov_off = iter->iov - fast_iov; | |
3104 | rw->iter.iov += iov_off; | |
3105 | } | |
3106 | if (rw->fast_iov != fast_iov) | |
3107 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3108 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3109 | } else { |
3110 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3111 | } |
3112 | } | |
3113 | ||
e8c2bc1f | 3114 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3115 | { |
e8c2bc1f JA |
3116 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3117 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3118 | return req->async_data == NULL; | |
3d9932a8 XW |
3119 | } |
3120 | ||
e8c2bc1f | 3121 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3122 | { |
e8c2bc1f | 3123 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3124 | return 0; |
3d9932a8 | 3125 | |
e8c2bc1f | 3126 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3127 | } |
3128 | ||
ff6165b2 JA |
3129 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3130 | const struct iovec *fast_iov, | |
227c0c96 | 3131 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3132 | { |
e8c2bc1f | 3133 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3134 | return 0; |
e8c2bc1f | 3135 | if (!req->async_data) { |
6bf985dc PB |
3136 | if (__io_alloc_async_data(req)) { |
3137 | kfree(iovec); | |
5d204bcf | 3138 | return -ENOMEM; |
6bf985dc | 3139 | } |
b7bb4f7d | 3140 | |
ff6165b2 | 3141 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3142 | } |
b7bb4f7d | 3143 | return 0; |
f67676d1 JA |
3144 | } |
3145 | ||
73debe68 | 3146 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3147 | { |
e8c2bc1f | 3148 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3149 | struct iovec *iov = iorw->fast_iov; |
847595de | 3150 | int ret; |
c3e330a4 | 3151 | |
2846c481 | 3152 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3153 | if (unlikely(ret < 0)) |
3154 | return ret; | |
3155 | ||
ab0b196c PB |
3156 | iorw->bytes_done = 0; |
3157 | iorw->free_iovec = iov; | |
3158 | if (iov) | |
3159 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3160 | return 0; |
3161 | } | |
3162 | ||
73debe68 | 3163 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3164 | { |
3529d8c2 JA |
3165 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3166 | return -EBADF; | |
93642ef8 | 3167 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3168 | } |
3169 | ||
c1dd91d1 JA |
3170 | /* |
3171 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3172 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3173 | * This gets called when the page is unlocked, and we generally expect that to | |
3174 | * happen when the page IO is completed and the page is now uptodate. This will | |
3175 | * queue a task_work based retry of the operation, attempting to copy the data | |
3176 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3177 | * do a thread based blocking retry of the operation. That's the unexpected | |
3178 | * slow path. | |
3179 | */ | |
bcf5a063 JA |
3180 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3181 | int sync, void *arg) | |
3182 | { | |
3183 | struct wait_page_queue *wpq; | |
3184 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3185 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3186 | |
3187 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3188 | ||
cdc8fcb4 LT |
3189 | if (!wake_page_match(wpq, key)) |
3190 | return 0; | |
3191 | ||
c8d317aa | 3192 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3193 | list_del_init(&wait->entry); |
3194 | ||
bcf5a063 JA |
3195 | /* submit ref gets dropped, acquire a new one */ |
3196 | refcount_inc(&req->refs); | |
921b9054 | 3197 | io_req_task_queue(req); |
bcf5a063 JA |
3198 | return 1; |
3199 | } | |
3200 | ||
c1dd91d1 JA |
3201 | /* |
3202 | * This controls whether a given IO request should be armed for async page | |
3203 | * based retry. If we return false here, the request is handed to the async | |
3204 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3205 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3206 | * will either succeed because the page is now uptodate and unlocked, or it | |
3207 | * will register a callback when the page is unlocked at IO completion. Through | |
3208 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3209 | * That retry will attempt the buffered read again. The retry will generally | |
3210 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3211 | * async worker threads for a blocking retry. | |
3212 | */ | |
227c0c96 | 3213 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3214 | { |
e8c2bc1f JA |
3215 | struct io_async_rw *rw = req->async_data; |
3216 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3217 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3218 | |
bcf5a063 JA |
3219 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3220 | if (req->flags & REQ_F_NOWAIT) | |
3221 | return false; | |
f67676d1 | 3222 | |
227c0c96 | 3223 | /* Only for buffered IO */ |
3b2a4439 | 3224 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3225 | return false; |
3b2a4439 | 3226 | |
bcf5a063 JA |
3227 | /* |
3228 | * just use poll if we can, and don't attempt if the fs doesn't | |
3229 | * support callback based unlocks | |
3230 | */ | |
3231 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3232 | return false; | |
f67676d1 | 3233 | |
3b2a4439 JA |
3234 | wait->wait.func = io_async_buf_func; |
3235 | wait->wait.private = req; | |
3236 | wait->wait.flags = 0; | |
3237 | INIT_LIST_HEAD(&wait->wait.entry); | |
3238 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3239 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3240 | kiocb->ki_waitq = wait; |
3b2a4439 | 3241 | return true; |
bcf5a063 JA |
3242 | } |
3243 | ||
3244 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3245 | { | |
3246 | if (req->file->f_op->read_iter) | |
3247 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3248 | else if (req->file->f_op->read) |
4017eb91 | 3249 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3250 | else |
3251 | return -EINVAL; | |
f67676d1 JA |
3252 | } |
3253 | ||
889fca73 | 3254 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3255 | { |
3256 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3257 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3258 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3259 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3260 | ssize_t io_size, ret, ret2; |
45d189c6 | 3261 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3262 | |
2846c481 | 3263 | if (rw) { |
e8c2bc1f | 3264 | iter = &rw->iter; |
2846c481 PB |
3265 | iovec = NULL; |
3266 | } else { | |
3267 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3268 | if (ret < 0) | |
3269 | return ret; | |
3270 | } | |
632546c4 | 3271 | io_size = iov_iter_count(iter); |
fa15bafb | 3272 | req->result = io_size; |
2b188cc1 | 3273 | |
fd6c2e4c JA |
3274 | /* Ensure we clear previously set non-block flag */ |
3275 | if (!force_nonblock) | |
29de5f6a | 3276 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3277 | else |
3278 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3279 | ||
24c74678 | 3280 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3281 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3282 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3283 | return ret ?: -EAGAIN; |
6713e7a6 | 3284 | } |
9e645e11 | 3285 | |
632546c4 | 3286 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3287 | if (unlikely(ret)) { |
3288 | kfree(iovec); | |
3289 | return ret; | |
3290 | } | |
2b188cc1 | 3291 | |
227c0c96 | 3292 | ret = io_iter_do_read(req, iter); |
32960613 | 3293 | |
57cd657b | 3294 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3295 | if (req->async_data) |
3296 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3297 | goto out_free; |
227c0c96 | 3298 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3299 | /* IOPOLL retry should happen for io-wq threads */ |
3300 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3301 | goto done; |
75c668cd PB |
3302 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3303 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3304 | goto done; |
84216315 | 3305 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3306 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3307 | ret = 0; |
7335e3bf | 3308 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3309 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3310 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3311 | goto done; |
227c0c96 JA |
3312 | } |
3313 | ||
227c0c96 | 3314 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3315 | if (ret2) |
3316 | return ret2; | |
3317 | ||
fe1cdd55 | 3318 | iovec = NULL; |
e8c2bc1f | 3319 | rw = req->async_data; |
227c0c96 | 3320 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3321 | iter = &rw->iter; |
227c0c96 | 3322 | |
b23df91b PB |
3323 | do { |
3324 | io_size -= ret; | |
3325 | rw->bytes_done += ret; | |
3326 | /* if we can retry, do so with the callbacks armed */ | |
3327 | if (!io_rw_should_retry(req)) { | |
3328 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3329 | return -EAGAIN; | |
3330 | } | |
3331 | ||
3332 | /* | |
3333 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3334 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3335 | * desired page gets unlocked. We can also get a partial read | |
3336 | * here, and if we do, then just retry at the new offset. | |
3337 | */ | |
3338 | ret = io_iter_do_read(req, iter); | |
3339 | if (ret == -EIOCBQUEUED) | |
3340 | return 0; | |
227c0c96 | 3341 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3342 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3343 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3344 | done: |
889fca73 | 3345 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3346 | out_free: |
3347 | /* it's faster to check here then delegate to kfree */ | |
3348 | if (iovec) | |
3349 | kfree(iovec); | |
5ea5dd45 | 3350 | return 0; |
2b188cc1 JA |
3351 | } |
3352 | ||
73debe68 | 3353 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3354 | { |
3529d8c2 JA |
3355 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3356 | return -EBADF; | |
93642ef8 | 3357 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3358 | } |
3359 | ||
889fca73 | 3360 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3361 | { |
3362 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3363 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3364 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3365 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3366 | ssize_t ret, ret2, io_size; |
45d189c6 | 3367 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3368 | |
2846c481 | 3369 | if (rw) { |
e8c2bc1f | 3370 | iter = &rw->iter; |
2846c481 PB |
3371 | iovec = NULL; |
3372 | } else { | |
3373 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3374 | if (ret < 0) | |
3375 | return ret; | |
3376 | } | |
632546c4 | 3377 | io_size = iov_iter_count(iter); |
fa15bafb | 3378 | req->result = io_size; |
2b188cc1 | 3379 | |
fd6c2e4c JA |
3380 | /* Ensure we clear previously set non-block flag */ |
3381 | if (!force_nonblock) | |
a88fc400 PB |
3382 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3383 | else | |
3384 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3385 | |
24c74678 | 3386 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3387 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3388 | goto copy_iov; |
31b51510 | 3389 | |
10d59345 JA |
3390 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3391 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3392 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3393 | goto copy_iov; |
31b51510 | 3394 | |
632546c4 | 3395 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3396 | if (unlikely(ret)) |
3397 | goto out_free; | |
4ed734b0 | 3398 | |
fa15bafb PB |
3399 | /* |
3400 | * Open-code file_start_write here to grab freeze protection, | |
3401 | * which will be released by another thread in | |
3402 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3403 | * released so that it doesn't complain about the held lock when | |
3404 | * we return to userspace. | |
3405 | */ | |
3406 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3407 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3408 | __sb_writers_release(file_inode(req->file)->i_sb, |
3409 | SB_FREEZE_WRITE); | |
3410 | } | |
3411 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3412 | |
fa15bafb | 3413 | if (req->file->f_op->write_iter) |
ff6165b2 | 3414 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3415 | else if (req->file->f_op->write) |
4017eb91 | 3416 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3417 | else |
3418 | ret2 = -EINVAL; | |
4ed734b0 | 3419 | |
fa15bafb PB |
3420 | /* |
3421 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3422 | * retry them without IOCB_NOWAIT. | |
3423 | */ | |
3424 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3425 | ret2 = -EAGAIN; | |
75c668cd PB |
3426 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3427 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3428 | goto done; |
3e6a0d3c JA |
3429 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3430 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3431 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3432 | /* IOPOLL retry should happen for io-wq threads */ |
3433 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3434 | goto copy_iov; | |
355afaeb | 3435 | done: |
889fca73 | 3436 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3437 | } else { |
f67676d1 | 3438 | copy_iov: |
84216315 | 3439 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3440 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3441 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3442 | return ret ?: -EAGAIN; |
2b188cc1 | 3443 | } |
31b51510 | 3444 | out_free: |
f261c168 | 3445 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3446 | if (iovec) |
6f2cc166 | 3447 | kfree(iovec); |
2b188cc1 JA |
3448 | return ret; |
3449 | } | |
3450 | ||
80a261fd JA |
3451 | static int io_renameat_prep(struct io_kiocb *req, |
3452 | const struct io_uring_sqe *sqe) | |
3453 | { | |
3454 | struct io_rename *ren = &req->rename; | |
3455 | const char __user *oldf, *newf; | |
3456 | ||
3457 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3458 | return -EBADF; | |
3459 | ||
3460 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3461 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3462 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3463 | ren->new_dfd = READ_ONCE(sqe->len); | |
3464 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3465 | ||
3466 | ren->oldpath = getname(oldf); | |
3467 | if (IS_ERR(ren->oldpath)) | |
3468 | return PTR_ERR(ren->oldpath); | |
3469 | ||
3470 | ren->newpath = getname(newf); | |
3471 | if (IS_ERR(ren->newpath)) { | |
3472 | putname(ren->oldpath); | |
3473 | return PTR_ERR(ren->newpath); | |
3474 | } | |
3475 | ||
3476 | req->flags |= REQ_F_NEED_CLEANUP; | |
3477 | return 0; | |
3478 | } | |
3479 | ||
45d189c6 | 3480 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3481 | { |
3482 | struct io_rename *ren = &req->rename; | |
3483 | int ret; | |
3484 | ||
45d189c6 | 3485 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3486 | return -EAGAIN; |
3487 | ||
3488 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3489 | ren->newpath, ren->flags); | |
3490 | ||
3491 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3492 | if (ret < 0) | |
3493 | req_set_fail_links(req); | |
3494 | io_req_complete(req, ret); | |
3495 | return 0; | |
3496 | } | |
3497 | ||
14a1143b JA |
3498 | static int io_unlinkat_prep(struct io_kiocb *req, |
3499 | const struct io_uring_sqe *sqe) | |
3500 | { | |
3501 | struct io_unlink *un = &req->unlink; | |
3502 | const char __user *fname; | |
3503 | ||
3504 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3505 | return -EBADF; | |
3506 | ||
3507 | un->dfd = READ_ONCE(sqe->fd); | |
3508 | ||
3509 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3510 | if (un->flags & ~AT_REMOVEDIR) | |
3511 | return -EINVAL; | |
3512 | ||
3513 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3514 | un->filename = getname(fname); | |
3515 | if (IS_ERR(un->filename)) | |
3516 | return PTR_ERR(un->filename); | |
3517 | ||
3518 | req->flags |= REQ_F_NEED_CLEANUP; | |
3519 | return 0; | |
3520 | } | |
3521 | ||
45d189c6 | 3522 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3523 | { |
3524 | struct io_unlink *un = &req->unlink; | |
3525 | int ret; | |
3526 | ||
45d189c6 | 3527 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3528 | return -EAGAIN; |
3529 | ||
3530 | if (un->flags & AT_REMOVEDIR) | |
3531 | ret = do_rmdir(un->dfd, un->filename); | |
3532 | else | |
3533 | ret = do_unlinkat(un->dfd, un->filename); | |
3534 | ||
3535 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3536 | if (ret < 0) | |
3537 | req_set_fail_links(req); | |
3538 | io_req_complete(req, ret); | |
3539 | return 0; | |
3540 | } | |
3541 | ||
36f4fa68 JA |
3542 | static int io_shutdown_prep(struct io_kiocb *req, |
3543 | const struct io_uring_sqe *sqe) | |
3544 | { | |
3545 | #if defined(CONFIG_NET) | |
3546 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3547 | return -EINVAL; | |
3548 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3549 | sqe->buf_index) | |
3550 | return -EINVAL; | |
3551 | ||
3552 | req->shutdown.how = READ_ONCE(sqe->len); | |
3553 | return 0; | |
3554 | #else | |
3555 | return -EOPNOTSUPP; | |
3556 | #endif | |
3557 | } | |
3558 | ||
45d189c6 | 3559 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3560 | { |
3561 | #if defined(CONFIG_NET) | |
3562 | struct socket *sock; | |
3563 | int ret; | |
3564 | ||
45d189c6 | 3565 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3566 | return -EAGAIN; |
3567 | ||
48aba79b | 3568 | sock = sock_from_file(req->file); |
36f4fa68 | 3569 | if (unlikely(!sock)) |
48aba79b | 3570 | return -ENOTSOCK; |
36f4fa68 JA |
3571 | |
3572 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3573 | if (ret < 0) |
3574 | req_set_fail_links(req); | |
36f4fa68 JA |
3575 | io_req_complete(req, ret); |
3576 | return 0; | |
3577 | #else | |
3578 | return -EOPNOTSUPP; | |
3579 | #endif | |
3580 | } | |
3581 | ||
f2a8d5c7 PB |
3582 | static int __io_splice_prep(struct io_kiocb *req, |
3583 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3584 | { |
3585 | struct io_splice* sp = &req->splice; | |
3586 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3587 | |
3232dd02 PB |
3588 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3589 | return -EINVAL; | |
7d67af2c PB |
3590 | |
3591 | sp->file_in = NULL; | |
7d67af2c PB |
3592 | sp->len = READ_ONCE(sqe->len); |
3593 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3594 | ||
3595 | if (unlikely(sp->flags & ~valid_flags)) | |
3596 | return -EINVAL; | |
3597 | ||
8371adf5 PB |
3598 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3599 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3600 | if (!sp->file_in) | |
3601 | return -EBADF; | |
7d67af2c PB |
3602 | req->flags |= REQ_F_NEED_CLEANUP; |
3603 | ||
7cdaf587 XW |
3604 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3605 | /* | |
3606 | * Splice operation will be punted aync, and here need to | |
3607 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3608 | */ | |
7d67af2c | 3609 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3610 | } |
7d67af2c PB |
3611 | |
3612 | return 0; | |
3613 | } | |
3614 | ||
f2a8d5c7 PB |
3615 | static int io_tee_prep(struct io_kiocb *req, |
3616 | const struct io_uring_sqe *sqe) | |
3617 | { | |
3618 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3619 | return -EINVAL; | |
3620 | return __io_splice_prep(req, sqe); | |
3621 | } | |
3622 | ||
45d189c6 | 3623 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3624 | { |
3625 | struct io_splice *sp = &req->splice; | |
3626 | struct file *in = sp->file_in; | |
3627 | struct file *out = sp->file_out; | |
3628 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3629 | long ret = 0; | |
3630 | ||
45d189c6 | 3631 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3632 | return -EAGAIN; |
3633 | if (sp->len) | |
3634 | ret = do_tee(in, out, sp->len, flags); | |
3635 | ||
3636 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3637 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3638 | ||
f2a8d5c7 PB |
3639 | if (ret != sp->len) |
3640 | req_set_fail_links(req); | |
e1e16097 | 3641 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3642 | return 0; |
3643 | } | |
3644 | ||
3645 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3646 | { | |
3647 | struct io_splice* sp = &req->splice; | |
3648 | ||
3649 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3650 | sp->off_out = READ_ONCE(sqe->off); | |
3651 | return __io_splice_prep(req, sqe); | |
3652 | } | |
3653 | ||
45d189c6 | 3654 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3655 | { |
3656 | struct io_splice *sp = &req->splice; | |
3657 | struct file *in = sp->file_in; | |
3658 | struct file *out = sp->file_out; | |
3659 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3660 | loff_t *poff_in, *poff_out; | |
c9687426 | 3661 | long ret = 0; |
7d67af2c | 3662 | |
45d189c6 | 3663 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3664 | return -EAGAIN; |
7d67af2c PB |
3665 | |
3666 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3667 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3668 | |
948a7749 | 3669 | if (sp->len) |
c9687426 | 3670 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3671 | |
3672 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3673 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3674 | ||
7d67af2c PB |
3675 | if (ret != sp->len) |
3676 | req_set_fail_links(req); | |
e1e16097 | 3677 | io_req_complete(req, ret); |
7d67af2c PB |
3678 | return 0; |
3679 | } | |
3680 | ||
2b188cc1 JA |
3681 | /* |
3682 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3683 | */ | |
889fca73 | 3684 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3685 | { |
3686 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3687 | |
def596e9 JA |
3688 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3689 | return -EINVAL; | |
3690 | ||
889fca73 | 3691 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3692 | return 0; |
3693 | } | |
3694 | ||
1155c76a | 3695 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3696 | { |
6b06314c | 3697 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3698 | |
09bb8394 JA |
3699 | if (!req->file) |
3700 | return -EBADF; | |
c992fe29 | 3701 | |
6b06314c | 3702 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3703 | return -EINVAL; |
edafccee | 3704 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3705 | return -EINVAL; |
3706 | ||
8ed8d3c3 JA |
3707 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3708 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3709 | return -EINVAL; | |
3710 | ||
3711 | req->sync.off = READ_ONCE(sqe->off); | |
3712 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3713 | return 0; |
3714 | } | |
3715 | ||
45d189c6 | 3716 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3717 | { |
8ed8d3c3 | 3718 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3719 | int ret; |
3720 | ||
ac45abc0 | 3721 | /* fsync always requires a blocking context */ |
45d189c6 | 3722 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3723 | return -EAGAIN; |
3724 | ||
9adbd45d | 3725 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3726 | end > 0 ? end : LLONG_MAX, |
3727 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3728 | if (ret < 0) | |
3729 | req_set_fail_links(req); | |
e1e16097 | 3730 | io_req_complete(req, ret); |
c992fe29 CH |
3731 | return 0; |
3732 | } | |
3733 | ||
d63d1b5e JA |
3734 | static int io_fallocate_prep(struct io_kiocb *req, |
3735 | const struct io_uring_sqe *sqe) | |
3736 | { | |
3737 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3738 | return -EINVAL; | |
3232dd02 PB |
3739 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3740 | return -EINVAL; | |
d63d1b5e JA |
3741 | |
3742 | req->sync.off = READ_ONCE(sqe->off); | |
3743 | req->sync.len = READ_ONCE(sqe->addr); | |
3744 | req->sync.mode = READ_ONCE(sqe->len); | |
3745 | return 0; | |
3746 | } | |
3747 | ||
45d189c6 | 3748 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3749 | { |
ac45abc0 PB |
3750 | int ret; |
3751 | ||
d63d1b5e | 3752 | /* fallocate always requiring blocking context */ |
45d189c6 | 3753 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3754 | return -EAGAIN; |
ac45abc0 PB |
3755 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3756 | req->sync.len); | |
ac45abc0 PB |
3757 | if (ret < 0) |
3758 | req_set_fail_links(req); | |
e1e16097 | 3759 | io_req_complete(req, ret); |
5d17b4a4 JA |
3760 | return 0; |
3761 | } | |
3762 | ||
ec65fea5 | 3763 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3764 | { |
f8748881 | 3765 | const char __user *fname; |
15b71abe | 3766 | int ret; |
b7bb4f7d | 3767 | |
ec65fea5 | 3768 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3769 | return -EINVAL; |
ec65fea5 | 3770 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3771 | return -EBADF; |
03b1230c | 3772 | |
ec65fea5 PB |
3773 | /* open.how should be already initialised */ |
3774 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3775 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3776 | |
25e72d10 PB |
3777 | req->open.dfd = READ_ONCE(sqe->fd); |
3778 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3779 | req->open.filename = getname(fname); |
15b71abe JA |
3780 | if (IS_ERR(req->open.filename)) { |
3781 | ret = PTR_ERR(req->open.filename); | |
3782 | req->open.filename = NULL; | |
3783 | return ret; | |
3784 | } | |
4022e7af | 3785 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3786 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3787 | return 0; |
03b1230c JA |
3788 | } |
3789 | ||
ec65fea5 PB |
3790 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3791 | { | |
3792 | u64 flags, mode; | |
3793 | ||
14587a46 | 3794 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3795 | return -EINVAL; |
ec65fea5 PB |
3796 | mode = READ_ONCE(sqe->len); |
3797 | flags = READ_ONCE(sqe->open_flags); | |
3798 | req->open.how = build_open_how(flags, mode); | |
3799 | return __io_openat_prep(req, sqe); | |
3800 | } | |
3801 | ||
cebdb986 | 3802 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3803 | { |
cebdb986 | 3804 | struct open_how __user *how; |
cebdb986 | 3805 | size_t len; |
0fa03c62 JA |
3806 | int ret; |
3807 | ||
14587a46 | 3808 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3809 | return -EINVAL; |
cebdb986 JA |
3810 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3811 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3812 | if (len < OPEN_HOW_SIZE_VER0) |
3813 | return -EINVAL; | |
3529d8c2 | 3814 | |
cebdb986 JA |
3815 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3816 | len); | |
3817 | if (ret) | |
3818 | return ret; | |
3529d8c2 | 3819 | |
ec65fea5 | 3820 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3821 | } |
3822 | ||
45d189c6 | 3823 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3824 | { |
3825 | struct open_flags op; | |
15b71abe | 3826 | struct file *file; |
3a81fd02 JA |
3827 | bool nonblock_set; |
3828 | bool resolve_nonblock; | |
15b71abe JA |
3829 | int ret; |
3830 | ||
cebdb986 | 3831 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3832 | if (ret) |
3833 | goto err; | |
3a81fd02 JA |
3834 | nonblock_set = op.open_flag & O_NONBLOCK; |
3835 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3836 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3837 | /* |
3838 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3839 | * it'll always -EAGAIN | |
3840 | */ | |
3841 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3842 | return -EAGAIN; | |
3843 | op.lookup_flags |= LOOKUP_CACHED; | |
3844 | op.open_flag |= O_NONBLOCK; | |
3845 | } | |
15b71abe | 3846 | |
4022e7af | 3847 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3848 | if (ret < 0) |
3849 | goto err; | |
3850 | ||
3851 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3852 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3853 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3854 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3855 | /* |
3a81fd02 JA |
3856 | * We could hang on to this 'fd', but seems like marginal |
3857 | * gain for something that is now known to be a slower path. | |
3858 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3859 | */ |
3a81fd02 JA |
3860 | put_unused_fd(ret); |
3861 | return -EAGAIN; | |
3862 | } | |
3863 | ||
15b71abe JA |
3864 | if (IS_ERR(file)) { |
3865 | put_unused_fd(ret); | |
3866 | ret = PTR_ERR(file); | |
3867 | } else { | |
45d189c6 | 3868 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3869 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3870 | fsnotify_open(file); |
3871 | fd_install(ret, file); | |
3872 | } | |
3873 | err: | |
3874 | putname(req->open.filename); | |
8fef80bf | 3875 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3876 | if (ret < 0) |
3877 | req_set_fail_links(req); | |
e1e16097 | 3878 | io_req_complete(req, ret); |
15b71abe JA |
3879 | return 0; |
3880 | } | |
3881 | ||
45d189c6 | 3882 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3883 | { |
e45cff58 | 3884 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3885 | } |
3886 | ||
067524e9 JA |
3887 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3888 | const struct io_uring_sqe *sqe) | |
3889 | { | |
3890 | struct io_provide_buf *p = &req->pbuf; | |
3891 | u64 tmp; | |
3892 | ||
3893 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3894 | return -EINVAL; | |
3895 | ||
3896 | tmp = READ_ONCE(sqe->fd); | |
3897 | if (!tmp || tmp > USHRT_MAX) | |
3898 | return -EINVAL; | |
3899 | ||
3900 | memset(p, 0, sizeof(*p)); | |
3901 | p->nbufs = tmp; | |
3902 | p->bgid = READ_ONCE(sqe->buf_group); | |
3903 | return 0; | |
3904 | } | |
3905 | ||
3906 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3907 | int bgid, unsigned nbufs) | |
3908 | { | |
3909 | unsigned i = 0; | |
3910 | ||
3911 | /* shouldn't happen */ | |
3912 | if (!nbufs) | |
3913 | return 0; | |
3914 | ||
3915 | /* the head kbuf is the list itself */ | |
3916 | while (!list_empty(&buf->list)) { | |
3917 | struct io_buffer *nxt; | |
3918 | ||
3919 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3920 | list_del(&nxt->list); | |
3921 | kfree(nxt); | |
3922 | if (++i == nbufs) | |
3923 | return i; | |
3924 | } | |
3925 | i++; | |
3926 | kfree(buf); | |
9e15c3a0 | 3927 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3928 | |
3929 | return i; | |
3930 | } | |
3931 | ||
889fca73 | 3932 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3933 | { |
3934 | struct io_provide_buf *p = &req->pbuf; | |
3935 | struct io_ring_ctx *ctx = req->ctx; | |
3936 | struct io_buffer *head; | |
3937 | int ret = 0; | |
45d189c6 | 3938 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3939 | |
3940 | io_ring_submit_lock(ctx, !force_nonblock); | |
3941 | ||
3942 | lockdep_assert_held(&ctx->uring_lock); | |
3943 | ||
3944 | ret = -ENOENT; | |
9e15c3a0 | 3945 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3946 | if (head) |
3947 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3948 | if (ret < 0) |
3949 | req_set_fail_links(req); | |
067524e9 | 3950 | |
31bff9a5 PB |
3951 | /* need to hold the lock to complete IOPOLL requests */ |
3952 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3953 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3954 | io_ring_submit_unlock(ctx, !force_nonblock); |
3955 | } else { | |
3956 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3957 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3958 | } |
067524e9 JA |
3959 | return 0; |
3960 | } | |
3961 | ||
ddf0322d JA |
3962 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3963 | const struct io_uring_sqe *sqe) | |
3964 | { | |
3965 | struct io_provide_buf *p = &req->pbuf; | |
3966 | u64 tmp; | |
3967 | ||
3968 | if (sqe->ioprio || sqe->rw_flags) | |
3969 | return -EINVAL; | |
3970 | ||
3971 | tmp = READ_ONCE(sqe->fd); | |
3972 | if (!tmp || tmp > USHRT_MAX) | |
3973 | return -E2BIG; | |
3974 | p->nbufs = tmp; | |
3975 | p->addr = READ_ONCE(sqe->addr); | |
3976 | p->len = READ_ONCE(sqe->len); | |
3977 | ||
efe68c1c | 3978 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3979 | return -EFAULT; |
3980 | ||
3981 | p->bgid = READ_ONCE(sqe->buf_group); | |
3982 | tmp = READ_ONCE(sqe->off); | |
3983 | if (tmp > USHRT_MAX) | |
3984 | return -E2BIG; | |
3985 | p->bid = tmp; | |
3986 | return 0; | |
3987 | } | |
3988 | ||
3989 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3990 | { | |
3991 | struct io_buffer *buf; | |
3992 | u64 addr = pbuf->addr; | |
3993 | int i, bid = pbuf->bid; | |
3994 | ||
3995 | for (i = 0; i < pbuf->nbufs; i++) { | |
3996 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3997 | if (!buf) | |
3998 | break; | |
3999 | ||
4000 | buf->addr = addr; | |
4001 | buf->len = pbuf->len; | |
4002 | buf->bid = bid; | |
4003 | addr += pbuf->len; | |
4004 | bid++; | |
4005 | if (!*head) { | |
4006 | INIT_LIST_HEAD(&buf->list); | |
4007 | *head = buf; | |
4008 | } else { | |
4009 | list_add_tail(&buf->list, &(*head)->list); | |
4010 | } | |
4011 | } | |
4012 | ||
4013 | return i ? i : -ENOMEM; | |
4014 | } | |
4015 | ||
889fca73 | 4016 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4017 | { |
4018 | struct io_provide_buf *p = &req->pbuf; | |
4019 | struct io_ring_ctx *ctx = req->ctx; | |
4020 | struct io_buffer *head, *list; | |
4021 | int ret = 0; | |
45d189c6 | 4022 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4023 | |
4024 | io_ring_submit_lock(ctx, !force_nonblock); | |
4025 | ||
4026 | lockdep_assert_held(&ctx->uring_lock); | |
4027 | ||
9e15c3a0 | 4028 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4029 | |
4030 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4031 | if (ret >= 0 && !list) { |
4032 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4033 | if (ret < 0) | |
067524e9 | 4034 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4035 | } |
ddf0322d JA |
4036 | if (ret < 0) |
4037 | req_set_fail_links(req); | |
31bff9a5 PB |
4038 | |
4039 | /* need to hold the lock to complete IOPOLL requests */ | |
4040 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4041 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4042 | io_ring_submit_unlock(ctx, !force_nonblock); |
4043 | } else { | |
4044 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4045 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4046 | } |
ddf0322d | 4047 | return 0; |
cebdb986 JA |
4048 | } |
4049 | ||
3e4827b0 JA |
4050 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4051 | const struct io_uring_sqe *sqe) | |
4052 | { | |
4053 | #if defined(CONFIG_EPOLL) | |
4054 | if (sqe->ioprio || sqe->buf_index) | |
4055 | return -EINVAL; | |
6ca56f84 | 4056 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4057 | return -EINVAL; |
3e4827b0 JA |
4058 | |
4059 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4060 | req->epoll.op = READ_ONCE(sqe->len); | |
4061 | req->epoll.fd = READ_ONCE(sqe->off); | |
4062 | ||
4063 | if (ep_op_has_event(req->epoll.op)) { | |
4064 | struct epoll_event __user *ev; | |
4065 | ||
4066 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4067 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4068 | return -EFAULT; | |
4069 | } | |
4070 | ||
4071 | return 0; | |
4072 | #else | |
4073 | return -EOPNOTSUPP; | |
4074 | #endif | |
4075 | } | |
4076 | ||
889fca73 | 4077 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4078 | { |
4079 | #if defined(CONFIG_EPOLL) | |
4080 | struct io_epoll *ie = &req->epoll; | |
4081 | int ret; | |
45d189c6 | 4082 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4083 | |
4084 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4085 | if (force_nonblock && ret == -EAGAIN) | |
4086 | return -EAGAIN; | |
4087 | ||
4088 | if (ret < 0) | |
4089 | req_set_fail_links(req); | |
889fca73 | 4090 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4091 | return 0; |
4092 | #else | |
4093 | return -EOPNOTSUPP; | |
4094 | #endif | |
4095 | } | |
4096 | ||
c1ca757b JA |
4097 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4098 | { | |
4099 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4100 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4101 | return -EINVAL; | |
3232dd02 PB |
4102 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4103 | return -EINVAL; | |
c1ca757b JA |
4104 | |
4105 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4106 | req->madvise.len = READ_ONCE(sqe->len); | |
4107 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4108 | return 0; | |
4109 | #else | |
4110 | return -EOPNOTSUPP; | |
4111 | #endif | |
4112 | } | |
4113 | ||
45d189c6 | 4114 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4115 | { |
4116 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4117 | struct io_madvise *ma = &req->madvise; | |
4118 | int ret; | |
4119 | ||
45d189c6 | 4120 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4121 | return -EAGAIN; |
4122 | ||
0726b01e | 4123 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4124 | if (ret < 0) |
4125 | req_set_fail_links(req); | |
e1e16097 | 4126 | io_req_complete(req, ret); |
c1ca757b JA |
4127 | return 0; |
4128 | #else | |
4129 | return -EOPNOTSUPP; | |
4130 | #endif | |
4131 | } | |
4132 | ||
4840e418 JA |
4133 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4134 | { | |
4135 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4136 | return -EINVAL; | |
3232dd02 PB |
4137 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4138 | return -EINVAL; | |
4840e418 JA |
4139 | |
4140 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4141 | req->fadvise.len = READ_ONCE(sqe->len); | |
4142 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4143 | return 0; | |
4144 | } | |
4145 | ||
45d189c6 | 4146 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4147 | { |
4148 | struct io_fadvise *fa = &req->fadvise; | |
4149 | int ret; | |
4150 | ||
45d189c6 | 4151 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4152 | switch (fa->advice) { |
4153 | case POSIX_FADV_NORMAL: | |
4154 | case POSIX_FADV_RANDOM: | |
4155 | case POSIX_FADV_SEQUENTIAL: | |
4156 | break; | |
4157 | default: | |
4158 | return -EAGAIN; | |
4159 | } | |
4160 | } | |
4840e418 JA |
4161 | |
4162 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4163 | if (ret < 0) | |
4164 | req_set_fail_links(req); | |
e1e16097 | 4165 | io_req_complete(req, ret); |
4840e418 JA |
4166 | return 0; |
4167 | } | |
4168 | ||
eddc7ef5 JA |
4169 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4170 | { | |
6ca56f84 | 4171 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4172 | return -EINVAL; |
eddc7ef5 JA |
4173 | if (sqe->ioprio || sqe->buf_index) |
4174 | return -EINVAL; | |
9c280f90 | 4175 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4176 | return -EBADF; |
eddc7ef5 | 4177 | |
1d9e1288 BM |
4178 | req->statx.dfd = READ_ONCE(sqe->fd); |
4179 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4180 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4181 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4182 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4183 | |
4184 | return 0; | |
4185 | } | |
4186 | ||
45d189c6 | 4187 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4188 | { |
1d9e1288 | 4189 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4190 | int ret; |
4191 | ||
45d189c6 | 4192 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4193 | /* only need file table for an actual valid fd */ |
4194 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4195 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4196 | return -EAGAIN; |
5b0bbee4 | 4197 | } |
eddc7ef5 | 4198 | |
e62753e4 BM |
4199 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4200 | ctx->buffer); | |
eddc7ef5 | 4201 | |
eddc7ef5 JA |
4202 | if (ret < 0) |
4203 | req_set_fail_links(req); | |
e1e16097 | 4204 | io_req_complete(req, ret); |
eddc7ef5 JA |
4205 | return 0; |
4206 | } | |
4207 | ||
b5dba59e JA |
4208 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4209 | { | |
14587a46 | 4210 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4211 | return -EINVAL; |
b5dba59e JA |
4212 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4213 | sqe->rw_flags || sqe->buf_index) | |
4214 | return -EINVAL; | |
9c280f90 | 4215 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4216 | return -EBADF; |
b5dba59e JA |
4217 | |
4218 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4219 | return 0; |
b5dba59e JA |
4220 | } |
4221 | ||
889fca73 | 4222 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4223 | { |
9eac1904 | 4224 | struct files_struct *files = current->files; |
3af73b28 | 4225 | struct io_close *close = &req->close; |
9eac1904 JA |
4226 | struct fdtable *fdt; |
4227 | struct file *file; | |
b5dba59e JA |
4228 | int ret; |
4229 | ||
9eac1904 JA |
4230 | file = NULL; |
4231 | ret = -EBADF; | |
4232 | spin_lock(&files->file_lock); | |
4233 | fdt = files_fdtable(files); | |
4234 | if (close->fd >= fdt->max_fds) { | |
4235 | spin_unlock(&files->file_lock); | |
4236 | goto err; | |
4237 | } | |
4238 | file = fdt->fd[close->fd]; | |
4239 | if (!file) { | |
4240 | spin_unlock(&files->file_lock); | |
4241 | goto err; | |
4242 | } | |
4243 | ||
4244 | if (file->f_op == &io_uring_fops) { | |
4245 | spin_unlock(&files->file_lock); | |
4246 | file = NULL; | |
4247 | goto err; | |
3af73b28 | 4248 | } |
b5dba59e JA |
4249 | |
4250 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4251 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4252 | spin_unlock(&files->file_lock); |
0bf0eefd | 4253 | return -EAGAIN; |
a2100672 | 4254 | } |
b5dba59e | 4255 | |
9eac1904 JA |
4256 | ret = __close_fd_get_file(close->fd, &file); |
4257 | spin_unlock(&files->file_lock); | |
4258 | if (ret < 0) { | |
4259 | if (ret == -ENOENT) | |
4260 | ret = -EBADF; | |
4261 | goto err; | |
4262 | } | |
4263 | ||
3af73b28 | 4264 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4265 | ret = filp_close(file, current->files); |
4266 | err: | |
3af73b28 PB |
4267 | if (ret < 0) |
4268 | req_set_fail_links(req); | |
9eac1904 JA |
4269 | if (file) |
4270 | fput(file); | |
889fca73 | 4271 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4272 | return 0; |
b5dba59e JA |
4273 | } |
4274 | ||
1155c76a | 4275 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4276 | { |
4277 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4278 | |
5d17b4a4 JA |
4279 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4280 | return -EINVAL; | |
4281 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4282 | return -EINVAL; | |
4283 | ||
8ed8d3c3 JA |
4284 | req->sync.off = READ_ONCE(sqe->off); |
4285 | req->sync.len = READ_ONCE(sqe->len); | |
4286 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4287 | return 0; |
4288 | } | |
4289 | ||
45d189c6 | 4290 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4291 | { |
8ed8d3c3 JA |
4292 | int ret; |
4293 | ||
ac45abc0 | 4294 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4295 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4296 | return -EAGAIN; |
4297 | ||
9adbd45d | 4298 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4299 | req->sync.flags); |
4300 | if (ret < 0) | |
4301 | req_set_fail_links(req); | |
e1e16097 | 4302 | io_req_complete(req, ret); |
5d17b4a4 JA |
4303 | return 0; |
4304 | } | |
4305 | ||
469956e8 | 4306 | #if defined(CONFIG_NET) |
02d27d89 PB |
4307 | static int io_setup_async_msg(struct io_kiocb *req, |
4308 | struct io_async_msghdr *kmsg) | |
4309 | { | |
e8c2bc1f JA |
4310 | struct io_async_msghdr *async_msg = req->async_data; |
4311 | ||
4312 | if (async_msg) | |
02d27d89 | 4313 | return -EAGAIN; |
e8c2bc1f | 4314 | if (io_alloc_async_data(req)) { |
257e84a5 | 4315 | kfree(kmsg->free_iov); |
02d27d89 PB |
4316 | return -ENOMEM; |
4317 | } | |
e8c2bc1f | 4318 | async_msg = req->async_data; |
02d27d89 | 4319 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4320 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4321 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4322 | /* if were using fast_iov, set it to the new one */ |
4323 | if (!async_msg->free_iov) | |
4324 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4325 | ||
02d27d89 PB |
4326 | return -EAGAIN; |
4327 | } | |
4328 | ||
2ae523ed PB |
4329 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4330 | struct io_async_msghdr *iomsg) | |
4331 | { | |
2ae523ed | 4332 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4333 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4334 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4335 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4336 | } |
4337 | ||
93642ef8 PB |
4338 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4339 | { | |
4340 | int ret; | |
4341 | ||
4342 | if (!io_op_defs[req->opcode].needs_async_data) | |
4343 | return 0; | |
4344 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4345 | if (!ret) | |
4346 | req->flags |= REQ_F_NEED_CLEANUP; | |
4347 | return ret; | |
4348 | } | |
4349 | ||
3529d8c2 | 4350 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4351 | { |
e47293fd | 4352 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4353 | |
d2b6f48b PB |
4354 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4355 | return -EINVAL; | |
4356 | ||
e47293fd | 4357 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4358 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4359 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4360 | |
d8768362 JA |
4361 | #ifdef CONFIG_COMPAT |
4362 | if (req->ctx->compat) | |
4363 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4364 | #endif | |
93642ef8 | 4365 | return 0; |
03b1230c JA |
4366 | } |
4367 | ||
889fca73 | 4368 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4369 | { |
6b754c8b | 4370 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4371 | struct socket *sock; |
7a7cacba | 4372 | unsigned flags; |
0fa03c62 JA |
4373 | int ret; |
4374 | ||
dba4a925 | 4375 | sock = sock_from_file(req->file); |
7a7cacba | 4376 | if (unlikely(!sock)) |
dba4a925 | 4377 | return -ENOTSOCK; |
3529d8c2 | 4378 | |
257e84a5 PB |
4379 | kmsg = req->async_data; |
4380 | if (!kmsg) { | |
7a7cacba PB |
4381 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4382 | if (ret) | |
4383 | return ret; | |
4384 | kmsg = &iomsg; | |
0fa03c62 | 4385 | } |
0fa03c62 | 4386 | |
76cd979f | 4387 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4388 | if (flags & MSG_DONTWAIT) |
4389 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4390 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4391 | flags |= MSG_DONTWAIT; |
e47293fd | 4392 | |
7a7cacba | 4393 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4394 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4395 | return io_setup_async_msg(req, kmsg); |
4396 | if (ret == -ERESTARTSYS) | |
4397 | ret = -EINTR; | |
0fa03c62 | 4398 | |
257e84a5 PB |
4399 | /* fast path, check for non-NULL to avoid function call */ |
4400 | if (kmsg->free_iov) | |
4401 | kfree(kmsg->free_iov); | |
99bc4c38 | 4402 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4403 | if (ret < 0) |
4404 | req_set_fail_links(req); | |
889fca73 | 4405 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4406 | return 0; |
03b1230c | 4407 | } |
aa1fa28f | 4408 | |
889fca73 | 4409 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4410 | { |
7a7cacba PB |
4411 | struct io_sr_msg *sr = &req->sr_msg; |
4412 | struct msghdr msg; | |
4413 | struct iovec iov; | |
fddaface | 4414 | struct socket *sock; |
7a7cacba | 4415 | unsigned flags; |
fddaface JA |
4416 | int ret; |
4417 | ||
dba4a925 | 4418 | sock = sock_from_file(req->file); |
7a7cacba | 4419 | if (unlikely(!sock)) |
dba4a925 | 4420 | return -ENOTSOCK; |
fddaface | 4421 | |
7a7cacba PB |
4422 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4423 | if (unlikely(ret)) | |
14db8411 | 4424 | return ret; |
fddaface | 4425 | |
7a7cacba PB |
4426 | msg.msg_name = NULL; |
4427 | msg.msg_control = NULL; | |
4428 | msg.msg_controllen = 0; | |
4429 | msg.msg_namelen = 0; | |
fddaface | 4430 | |
76cd979f | 4431 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4432 | if (flags & MSG_DONTWAIT) |
4433 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4434 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4435 | flags |= MSG_DONTWAIT; |
fddaface | 4436 | |
7a7cacba PB |
4437 | msg.msg_flags = flags; |
4438 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4439 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4440 | return -EAGAIN; |
4441 | if (ret == -ERESTARTSYS) | |
4442 | ret = -EINTR; | |
fddaface | 4443 | |
fddaface JA |
4444 | if (ret < 0) |
4445 | req_set_fail_links(req); | |
889fca73 | 4446 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4447 | return 0; |
fddaface JA |
4448 | } |
4449 | ||
1400e697 PB |
4450 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4451 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4452 | { |
4453 | struct io_sr_msg *sr = &req->sr_msg; | |
4454 | struct iovec __user *uiov; | |
4455 | size_t iov_len; | |
4456 | int ret; | |
4457 | ||
1400e697 PB |
4458 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4459 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4460 | if (ret) |
4461 | return ret; | |
4462 | ||
4463 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4464 | if (iov_len > 1) | |
4465 | return -EINVAL; | |
5476dfed | 4466 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4467 | return -EFAULT; |
5476dfed | 4468 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4469 | iomsg->free_iov = NULL; |
52de1fe1 | 4470 | } else { |
257e84a5 | 4471 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4472 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4473 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4474 | false); |
52de1fe1 JA |
4475 | if (ret > 0) |
4476 | ret = 0; | |
4477 | } | |
4478 | ||
4479 | return ret; | |
4480 | } | |
4481 | ||
4482 | #ifdef CONFIG_COMPAT | |
4483 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4484 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4485 | { |
4486 | struct compat_msghdr __user *msg_compat; | |
4487 | struct io_sr_msg *sr = &req->sr_msg; | |
4488 | struct compat_iovec __user *uiov; | |
4489 | compat_uptr_t ptr; | |
4490 | compat_size_t len; | |
4491 | int ret; | |
4492 | ||
270a5940 | 4493 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4494 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4495 | &ptr, &len); |
4496 | if (ret) | |
4497 | return ret; | |
4498 | ||
4499 | uiov = compat_ptr(ptr); | |
4500 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4501 | compat_ssize_t clen; | |
4502 | ||
4503 | if (len > 1) | |
4504 | return -EINVAL; | |
4505 | if (!access_ok(uiov, sizeof(*uiov))) | |
4506 | return -EFAULT; | |
4507 | if (__get_user(clen, &uiov->iov_len)) | |
4508 | return -EFAULT; | |
4509 | if (clen < 0) | |
4510 | return -EINVAL; | |
2d280bc8 | 4511 | sr->len = clen; |
257e84a5 | 4512 | iomsg->free_iov = NULL; |
52de1fe1 | 4513 | } else { |
257e84a5 | 4514 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4515 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4516 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4517 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4518 | if (ret < 0) |
4519 | return ret; | |
4520 | } | |
4521 | ||
4522 | return 0; | |
4523 | } | |
4524 | #endif | |
4525 | ||
1400e697 PB |
4526 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4527 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4528 | { |
1400e697 | 4529 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4530 | |
4531 | #ifdef CONFIG_COMPAT | |
4532 | if (req->ctx->compat) | |
1400e697 | 4533 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4534 | #endif |
52de1fe1 | 4535 | |
1400e697 | 4536 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4537 | } |
4538 | ||
bcda7baa | 4539 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4540 | bool needs_lock) |
bcda7baa JA |
4541 | { |
4542 | struct io_sr_msg *sr = &req->sr_msg; | |
4543 | struct io_buffer *kbuf; | |
4544 | ||
bcda7baa JA |
4545 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4546 | if (IS_ERR(kbuf)) | |
4547 | return kbuf; | |
4548 | ||
4549 | sr->kbuf = kbuf; | |
4550 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4551 | return kbuf; |
fddaface JA |
4552 | } |
4553 | ||
7fbb1b54 PB |
4554 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4555 | { | |
4556 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4557 | } | |
4558 | ||
93642ef8 | 4559 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4560 | { |
99bc4c38 | 4561 | int ret; |
3529d8c2 | 4562 | |
93642ef8 PB |
4563 | if (!io_op_defs[req->opcode].needs_async_data) |
4564 | return 0; | |
4565 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4566 | if (!ret) | |
4567 | req->flags |= REQ_F_NEED_CLEANUP; | |
4568 | return ret; | |
4569 | } | |
4570 | ||
4571 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4572 | { | |
4573 | struct io_sr_msg *sr = &req->sr_msg; | |
4574 | ||
d2b6f48b PB |
4575 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4576 | return -EINVAL; | |
4577 | ||
3529d8c2 | 4578 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4579 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4580 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4581 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4582 | |
d8768362 JA |
4583 | #ifdef CONFIG_COMPAT |
4584 | if (req->ctx->compat) | |
4585 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4586 | #endif | |
93642ef8 | 4587 | return 0; |
aa1fa28f JA |
4588 | } |
4589 | ||
889fca73 | 4590 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4591 | { |
6b754c8b | 4592 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4593 | struct socket *sock; |
7fbb1b54 | 4594 | struct io_buffer *kbuf; |
7a7cacba | 4595 | unsigned flags; |
52de1fe1 | 4596 | int ret, cflags = 0; |
45d189c6 | 4597 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4598 | |
dba4a925 | 4599 | sock = sock_from_file(req->file); |
7a7cacba | 4600 | if (unlikely(!sock)) |
dba4a925 | 4601 | return -ENOTSOCK; |
3529d8c2 | 4602 | |
257e84a5 PB |
4603 | kmsg = req->async_data; |
4604 | if (!kmsg) { | |
7a7cacba PB |
4605 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4606 | if (ret) | |
681fda8d | 4607 | return ret; |
7a7cacba PB |
4608 | kmsg = &iomsg; |
4609 | } | |
03b1230c | 4610 | |
bc02ef33 | 4611 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4612 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4613 | if (IS_ERR(kbuf)) |
52de1fe1 | 4614 | return PTR_ERR(kbuf); |
7a7cacba | 4615 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4616 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4617 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4618 | 1, req->sr_msg.len); |
4619 | } | |
52de1fe1 | 4620 | |
76cd979f | 4621 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4622 | if (flags & MSG_DONTWAIT) |
4623 | req->flags |= REQ_F_NOWAIT; | |
4624 | else if (force_nonblock) | |
4625 | flags |= MSG_DONTWAIT; | |
e47293fd | 4626 | |
7a7cacba PB |
4627 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4628 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4629 | if (force_nonblock && ret == -EAGAIN) |
4630 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4631 | if (ret == -ERESTARTSYS) |
4632 | ret = -EINTR; | |
03b1230c | 4633 | |
7fbb1b54 PB |
4634 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4635 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4636 | /* fast path, check for non-NULL to avoid function call */ |
4637 | if (kmsg->free_iov) | |
4638 | kfree(kmsg->free_iov); | |
99bc4c38 | 4639 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4640 | if (ret < 0) |
4641 | req_set_fail_links(req); | |
889fca73 | 4642 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4643 | return 0; |
0fa03c62 | 4644 | } |
5d17b4a4 | 4645 | |
889fca73 | 4646 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4647 | { |
6b754c8b | 4648 | struct io_buffer *kbuf; |
7a7cacba PB |
4649 | struct io_sr_msg *sr = &req->sr_msg; |
4650 | struct msghdr msg; | |
4651 | void __user *buf = sr->buf; | |
fddaface | 4652 | struct socket *sock; |
7a7cacba PB |
4653 | struct iovec iov; |
4654 | unsigned flags; | |
bcda7baa | 4655 | int ret, cflags = 0; |
45d189c6 | 4656 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4657 | |
dba4a925 | 4658 | sock = sock_from_file(req->file); |
7a7cacba | 4659 | if (unlikely(!sock)) |
dba4a925 | 4660 | return -ENOTSOCK; |
fddaface | 4661 | |
bc02ef33 | 4662 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4663 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4664 | if (IS_ERR(kbuf)) |
4665 | return PTR_ERR(kbuf); | |
7a7cacba | 4666 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4667 | } |
bcda7baa | 4668 | |
7a7cacba | 4669 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4670 | if (unlikely(ret)) |
4671 | goto out_free; | |
fddaface | 4672 | |
7a7cacba PB |
4673 | msg.msg_name = NULL; |
4674 | msg.msg_control = NULL; | |
4675 | msg.msg_controllen = 0; | |
4676 | msg.msg_namelen = 0; | |
4677 | msg.msg_iocb = NULL; | |
4678 | msg.msg_flags = 0; | |
fddaface | 4679 | |
76cd979f | 4680 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4681 | if (flags & MSG_DONTWAIT) |
4682 | req->flags |= REQ_F_NOWAIT; | |
4683 | else if (force_nonblock) | |
4684 | flags |= MSG_DONTWAIT; | |
4685 | ||
4686 | ret = sock_recvmsg(sock, &msg, flags); | |
4687 | if (force_nonblock && ret == -EAGAIN) | |
4688 | return -EAGAIN; | |
4689 | if (ret == -ERESTARTSYS) | |
4690 | ret = -EINTR; | |
14c32eee | 4691 | out_free: |
7fbb1b54 PB |
4692 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4693 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4694 | if (ret < 0) |
4695 | req_set_fail_links(req); | |
889fca73 | 4696 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4697 | return 0; |
fddaface JA |
4698 | } |
4699 | ||
3529d8c2 | 4700 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4701 | { |
8ed8d3c3 JA |
4702 | struct io_accept *accept = &req->accept; |
4703 | ||
14587a46 | 4704 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4705 | return -EINVAL; |
8042d6ce | 4706 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4707 | return -EINVAL; |
4708 | ||
d55e5f5b JA |
4709 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4710 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4711 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4712 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4713 | return 0; |
8ed8d3c3 | 4714 | } |
17f2fe35 | 4715 | |
889fca73 | 4716 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4717 | { |
4718 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4719 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4720 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4721 | int ret; |
4722 | ||
e697deed JX |
4723 | if (req->file->f_flags & O_NONBLOCK) |
4724 | req->flags |= REQ_F_NOWAIT; | |
4725 | ||
8ed8d3c3 | 4726 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4727 | accept->addr_len, accept->flags, |
4728 | accept->nofile); | |
8ed8d3c3 | 4729 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4730 | return -EAGAIN; |
ac45abc0 PB |
4731 | if (ret < 0) { |
4732 | if (ret == -ERESTARTSYS) | |
4733 | ret = -EINTR; | |
4e88d6e7 | 4734 | req_set_fail_links(req); |
ac45abc0 | 4735 | } |
889fca73 | 4736 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4737 | return 0; |
8ed8d3c3 JA |
4738 | } |
4739 | ||
93642ef8 PB |
4740 | static int io_connect_prep_async(struct io_kiocb *req) |
4741 | { | |
4742 | struct io_async_connect *io = req->async_data; | |
4743 | struct io_connect *conn = &req->connect; | |
4744 | ||
4745 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4746 | } | |
4747 | ||
3529d8c2 | 4748 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4749 | { |
3529d8c2 | 4750 | struct io_connect *conn = &req->connect; |
f499a021 | 4751 | |
14587a46 | 4752 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4753 | return -EINVAL; |
4754 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4755 | return -EINVAL; | |
4756 | ||
3529d8c2 JA |
4757 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4758 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4759 | return 0; |
f499a021 JA |
4760 | } |
4761 | ||
889fca73 | 4762 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4763 | { |
e8c2bc1f | 4764 | struct io_async_connect __io, *io; |
f8e85cf2 | 4765 | unsigned file_flags; |
3fbb51c1 | 4766 | int ret; |
45d189c6 | 4767 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4768 | |
e8c2bc1f JA |
4769 | if (req->async_data) { |
4770 | io = req->async_data; | |
f499a021 | 4771 | } else { |
3529d8c2 JA |
4772 | ret = move_addr_to_kernel(req->connect.addr, |
4773 | req->connect.addr_len, | |
e8c2bc1f | 4774 | &__io.address); |
f499a021 JA |
4775 | if (ret) |
4776 | goto out; | |
4777 | io = &__io; | |
4778 | } | |
4779 | ||
3fbb51c1 JA |
4780 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4781 | ||
e8c2bc1f | 4782 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4783 | req->connect.addr_len, file_flags); |
87f80d62 | 4784 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4785 | if (req->async_data) |
b7bb4f7d | 4786 | return -EAGAIN; |
e8c2bc1f | 4787 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4788 | ret = -ENOMEM; |
4789 | goto out; | |
4790 | } | |
e8c2bc1f JA |
4791 | io = req->async_data; |
4792 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4793 | return -EAGAIN; |
f499a021 | 4794 | } |
f8e85cf2 JA |
4795 | if (ret == -ERESTARTSYS) |
4796 | ret = -EINTR; | |
f499a021 | 4797 | out: |
4e88d6e7 JA |
4798 | if (ret < 0) |
4799 | req_set_fail_links(req); | |
889fca73 | 4800 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4801 | return 0; |
469956e8 Y |
4802 | } |
4803 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4804 | #define IO_NETOP_FN(op) \ |
4805 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4806 | { \ | |
4807 | return -EOPNOTSUPP; \ | |
4808 | } | |
4809 | ||
4810 | #define IO_NETOP_PREP(op) \ | |
4811 | IO_NETOP_FN(op) \ | |
4812 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4813 | { \ | |
4814 | return -EOPNOTSUPP; \ | |
4815 | } \ | |
4816 | ||
4817 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4818 | IO_NETOP_PREP(op) \ | |
4819 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4820 | { \ | |
4821 | return -EOPNOTSUPP; \ | |
4822 | } | |
4823 | ||
4824 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4825 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4826 | IO_NETOP_PREP_ASYNC(connect); | |
4827 | IO_NETOP_PREP(accept); | |
4828 | IO_NETOP_FN(send); | |
4829 | IO_NETOP_FN(recv); | |
469956e8 | 4830 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4831 | |
d7718a9d JA |
4832 | struct io_poll_table { |
4833 | struct poll_table_struct pt; | |
4834 | struct io_kiocb *req; | |
4835 | int error; | |
4836 | }; | |
ce593a6c | 4837 | |
d7718a9d JA |
4838 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4839 | __poll_t mask, task_work_func_t func) | |
4840 | { | |
aa96bf8a | 4841 | int ret; |
d7718a9d JA |
4842 | |
4843 | /* for instances that support it check for an event match first: */ | |
4844 | if (mask && !(mask & poll->events)) | |
4845 | return 0; | |
4846 | ||
4847 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4848 | ||
4849 | list_del_init(&poll->wait.entry); | |
4850 | ||
d7718a9d | 4851 | req->result = mask; |
7cbf1722 | 4852 | req->task_work.func = func; |
6d816e08 JA |
4853 | percpu_ref_get(&req->ctx->refs); |
4854 | ||
d7718a9d | 4855 | /* |
e3aabf95 JA |
4856 | * If this fails, then the task is exiting. When a task exits, the |
4857 | * work gets canceled, so just cancel this request as well instead | |
4858 | * of executing it. We can't safely execute it anyway, as we may not | |
4859 | * have the needed state needed for it anyway. | |
d7718a9d | 4860 | */ |
355fb9e2 | 4861 | ret = io_req_task_work_add(req); |
aa96bf8a | 4862 | if (unlikely(ret)) { |
e3aabf95 | 4863 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4864 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4865 | } |
d7718a9d JA |
4866 | return 1; |
4867 | } | |
4868 | ||
74ce6ce4 JA |
4869 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4870 | __acquires(&req->ctx->completion_lock) | |
4871 | { | |
4872 | struct io_ring_ctx *ctx = req->ctx; | |
4873 | ||
4874 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4875 | struct poll_table_struct pt = { ._key = poll->events }; | |
4876 | ||
4877 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4878 | } | |
4879 | ||
4880 | spin_lock_irq(&ctx->completion_lock); | |
4881 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4882 | add_wait_queue(poll->head, &poll->wait); | |
4883 | return true; | |
4884 | } | |
4885 | ||
4886 | return false; | |
4887 | } | |
4888 | ||
d4e7cd36 | 4889 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4890 | { |
e8c2bc1f | 4891 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4892 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4893 | return req->async_data; |
d4e7cd36 JA |
4894 | return req->apoll->double_poll; |
4895 | } | |
4896 | ||
4897 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4898 | { | |
4899 | if (req->opcode == IORING_OP_POLL_ADD) | |
4900 | return &req->poll; | |
4901 | return &req->apoll->poll; | |
4902 | } | |
4903 | ||
4904 | static void io_poll_remove_double(struct io_kiocb *req) | |
4905 | { | |
4906 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4907 | |
4908 | lockdep_assert_held(&req->ctx->completion_lock); | |
4909 | ||
4910 | if (poll && poll->head) { | |
4911 | struct wait_queue_head *head = poll->head; | |
4912 | ||
4913 | spin_lock(&head->lock); | |
4914 | list_del_init(&poll->wait.entry); | |
4915 | if (poll->wait.private) | |
4916 | refcount_dec(&req->refs); | |
4917 | poll->head = NULL; | |
4918 | spin_unlock(&head->lock); | |
4919 | } | |
4920 | } | |
4921 | ||
4922 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4923 | { | |
4924 | struct io_ring_ctx *ctx = req->ctx; | |
4925 | ||
d4e7cd36 | 4926 | io_poll_remove_double(req); |
18bceab1 JA |
4927 | req->poll.done = true; |
4928 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4929 | io_commit_cqring(ctx); | |
4930 | } | |
4931 | ||
dd221f46 | 4932 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4933 | { |
dd221f46 | 4934 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4935 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4936 | struct io_kiocb *nxt; |
18bceab1 JA |
4937 | |
4938 | if (io_poll_rewait(req, &req->poll)) { | |
4939 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4940 | } else { |
4941 | hash_del(&req->hash_node); | |
4942 | io_poll_complete(req, req->result, 0); | |
4943 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4944 | |
dd221f46 PB |
4945 | nxt = io_put_req_find_next(req); |
4946 | io_cqring_ev_posted(ctx); | |
4947 | if (nxt) | |
4948 | __io_req_task_submit(nxt); | |
4949 | } | |
18bceab1 | 4950 | |
6d816e08 | 4951 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4952 | } |
4953 | ||
4954 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4955 | int sync, void *key) | |
4956 | { | |
4957 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4958 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4959 | __poll_t mask = key_to_poll(key); |
4960 | ||
4961 | /* for instances that support it check for an event match first: */ | |
4962 | if (mask && !(mask & poll->events)) | |
4963 | return 0; | |
4964 | ||
8706e04e JA |
4965 | list_del_init(&wait->entry); |
4966 | ||
807abcb0 | 4967 | if (poll && poll->head) { |
18bceab1 JA |
4968 | bool done; |
4969 | ||
807abcb0 JA |
4970 | spin_lock(&poll->head->lock); |
4971 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4972 | if (!done) |
807abcb0 | 4973 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4974 | /* make sure double remove sees this as being gone */ |
4975 | wait->private = NULL; | |
807abcb0 | 4976 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4977 | if (!done) { |
4978 | /* use wait func handler, so it matches the rq type */ | |
4979 | poll->wait.func(&poll->wait, mode, sync, key); | |
4980 | } | |
18bceab1 JA |
4981 | } |
4982 | refcount_dec(&req->refs); | |
4983 | return 1; | |
4984 | } | |
4985 | ||
4986 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4987 | wait_queue_func_t wake_func) | |
4988 | { | |
4989 | poll->head = NULL; | |
4990 | poll->done = false; | |
4991 | poll->canceled = false; | |
4992 | poll->events = events; | |
4993 | INIT_LIST_HEAD(&poll->wait.entry); | |
4994 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4995 | } | |
4996 | ||
4997 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4998 | struct wait_queue_head *head, |
4999 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5000 | { |
5001 | struct io_kiocb *req = pt->req; | |
5002 | ||
5003 | /* | |
5004 | * If poll->head is already set, it's because the file being polled | |
5005 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5006 | * for write). Setup a separate io_poll_iocb if this happens. | |
5007 | */ | |
5008 | if (unlikely(poll->head)) { | |
58852d4d PB |
5009 | struct io_poll_iocb *poll_one = poll; |
5010 | ||
18bceab1 | 5011 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5012 | if (*poll_ptr) { |
18bceab1 JA |
5013 | pt->error = -EINVAL; |
5014 | return; | |
5015 | } | |
1c3b3e65 JA |
5016 | /* double add on the same waitqueue head, ignore */ |
5017 | if (poll->head == head) | |
5018 | return; | |
18bceab1 JA |
5019 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5020 | if (!poll) { | |
5021 | pt->error = -ENOMEM; | |
5022 | return; | |
5023 | } | |
58852d4d | 5024 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5025 | refcount_inc(&req->refs); |
5026 | poll->wait.private = req; | |
807abcb0 | 5027 | *poll_ptr = poll; |
18bceab1 JA |
5028 | } |
5029 | ||
5030 | pt->error = 0; | |
5031 | poll->head = head; | |
a31eb4a2 JX |
5032 | |
5033 | if (poll->events & EPOLLEXCLUSIVE) | |
5034 | add_wait_queue_exclusive(head, &poll->wait); | |
5035 | else | |
5036 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5037 | } |
5038 | ||
5039 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5040 | struct poll_table_struct *p) | |
5041 | { | |
5042 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5043 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5044 | |
807abcb0 | 5045 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5046 | } |
5047 | ||
d7718a9d JA |
5048 | static void io_async_task_func(struct callback_head *cb) |
5049 | { | |
5050 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5051 | struct async_poll *apoll = req->apoll; | |
5052 | struct io_ring_ctx *ctx = req->ctx; | |
5053 | ||
5054 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5055 | ||
74ce6ce4 | 5056 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5057 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5058 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5059 | return; |
d7718a9d JA |
5060 | } |
5061 | ||
31067255 | 5062 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5063 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5064 | hash_del(&req->hash_node); |
2bae047e | 5065 | |
d4e7cd36 | 5066 | io_poll_remove_double(req); |
74ce6ce4 JA |
5067 | spin_unlock_irq(&ctx->completion_lock); |
5068 | ||
0be0b0e3 PB |
5069 | if (!READ_ONCE(apoll->poll.canceled)) |
5070 | __io_req_task_submit(req); | |
5071 | else | |
5072 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5073 | |
6d816e08 | 5074 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5075 | kfree(apoll->double_poll); |
31067255 | 5076 | kfree(apoll); |
d7718a9d JA |
5077 | } |
5078 | ||
5079 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5080 | void *key) | |
5081 | { | |
5082 | struct io_kiocb *req = wait->private; | |
5083 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5084 | ||
5085 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5086 | key_to_poll(key)); | |
5087 | ||
5088 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5089 | } | |
5090 | ||
5091 | static void io_poll_req_insert(struct io_kiocb *req) | |
5092 | { | |
5093 | struct io_ring_ctx *ctx = req->ctx; | |
5094 | struct hlist_head *list; | |
5095 | ||
5096 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5097 | hlist_add_head(&req->hash_node, list); | |
5098 | } | |
5099 | ||
5100 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5101 | struct io_poll_iocb *poll, | |
5102 | struct io_poll_table *ipt, __poll_t mask, | |
5103 | wait_queue_func_t wake_func) | |
5104 | __acquires(&ctx->completion_lock) | |
5105 | { | |
5106 | struct io_ring_ctx *ctx = req->ctx; | |
5107 | bool cancel = false; | |
5108 | ||
4d52f338 | 5109 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5110 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5111 | poll->file = req->file; |
18bceab1 | 5112 | poll->wait.private = req; |
d7718a9d JA |
5113 | |
5114 | ipt->pt._key = mask; | |
5115 | ipt->req = req; | |
5116 | ipt->error = -EINVAL; | |
5117 | ||
d7718a9d JA |
5118 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5119 | ||
5120 | spin_lock_irq(&ctx->completion_lock); | |
5121 | if (likely(poll->head)) { | |
5122 | spin_lock(&poll->head->lock); | |
5123 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5124 | if (ipt->error) | |
5125 | cancel = true; | |
5126 | ipt->error = 0; | |
5127 | mask = 0; | |
5128 | } | |
5129 | if (mask || ipt->error) | |
5130 | list_del_init(&poll->wait.entry); | |
5131 | else if (cancel) | |
5132 | WRITE_ONCE(poll->canceled, true); | |
5133 | else if (!poll->done) /* actually waiting for an event */ | |
5134 | io_poll_req_insert(req); | |
5135 | spin_unlock(&poll->head->lock); | |
5136 | } | |
5137 | ||
5138 | return mask; | |
5139 | } | |
5140 | ||
5141 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5142 | { | |
5143 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5144 | struct io_ring_ctx *ctx = req->ctx; | |
5145 | struct async_poll *apoll; | |
5146 | struct io_poll_table ipt; | |
5147 | __poll_t mask, ret; | |
9dab14b8 | 5148 | int rw; |
d7718a9d JA |
5149 | |
5150 | if (!req->file || !file_can_poll(req->file)) | |
5151 | return false; | |
24c74678 | 5152 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5153 | return false; |
9dab14b8 JA |
5154 | if (def->pollin) |
5155 | rw = READ; | |
5156 | else if (def->pollout) | |
5157 | rw = WRITE; | |
5158 | else | |
5159 | return false; | |
5160 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5161 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5162 | return false; |
5163 | ||
5164 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5165 | if (unlikely(!apoll)) | |
5166 | return false; | |
807abcb0 | 5167 | apoll->double_poll = NULL; |
d7718a9d JA |
5168 | |
5169 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5170 | req->apoll = apoll; |
d7718a9d | 5171 | |
8755d97a | 5172 | mask = 0; |
d7718a9d | 5173 | if (def->pollin) |
8755d97a | 5174 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5175 | if (def->pollout) |
5176 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5177 | |
5178 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5179 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5180 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5181 | mask &= ~POLLIN; | |
5182 | ||
d7718a9d JA |
5183 | mask |= POLLERR | POLLPRI; |
5184 | ||
5185 | ipt.pt._qproc = io_async_queue_proc; | |
5186 | ||
5187 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5188 | io_async_wake); | |
a36da65c | 5189 | if (ret || ipt.error) { |
d4e7cd36 | 5190 | io_poll_remove_double(req); |
d7718a9d | 5191 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5192 | kfree(apoll->double_poll); |
d7718a9d JA |
5193 | kfree(apoll); |
5194 | return false; | |
5195 | } | |
5196 | spin_unlock_irq(&ctx->completion_lock); | |
5197 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5198 | apoll->poll.events); | |
5199 | return true; | |
5200 | } | |
5201 | ||
5202 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5203 | struct io_poll_iocb *poll) | |
221c5eb2 | 5204 | { |
b41e9852 | 5205 | bool do_complete = false; |
221c5eb2 JA |
5206 | |
5207 | spin_lock(&poll->head->lock); | |
5208 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5209 | if (!list_empty(&poll->wait.entry)) { |
5210 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5211 | do_complete = true; |
221c5eb2 JA |
5212 | } |
5213 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5214 | hash_del(&req->hash_node); |
d7718a9d JA |
5215 | return do_complete; |
5216 | } | |
5217 | ||
5218 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5219 | { | |
5220 | bool do_complete; | |
5221 | ||
d4e7cd36 JA |
5222 | io_poll_remove_double(req); |
5223 | ||
d7718a9d JA |
5224 | if (req->opcode == IORING_OP_POLL_ADD) { |
5225 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5226 | } else { | |
3bfa5bcb JA |
5227 | struct async_poll *apoll = req->apoll; |
5228 | ||
d7718a9d | 5229 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5230 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5231 | if (do_complete) { | |
d7718a9d | 5232 | io_put_req(req); |
807abcb0 | 5233 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5234 | kfree(apoll); |
5235 | } | |
b1f573bd XW |
5236 | } |
5237 | ||
b41e9852 JA |
5238 | if (do_complete) { |
5239 | io_cqring_fill_event(req, -ECANCELED); | |
5240 | io_commit_cqring(req->ctx); | |
f254ac04 | 5241 | req_set_fail_links(req); |
216578e5 | 5242 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5243 | } |
5244 | ||
5245 | return do_complete; | |
221c5eb2 JA |
5246 | } |
5247 | ||
76e1b642 JA |
5248 | /* |
5249 | * Returns true if we found and killed one or more poll requests | |
5250 | */ | |
6b81928d PB |
5251 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5252 | struct files_struct *files) | |
221c5eb2 | 5253 | { |
78076bb6 | 5254 | struct hlist_node *tmp; |
221c5eb2 | 5255 | struct io_kiocb *req; |
8e2e1faf | 5256 | int posted = 0, i; |
221c5eb2 JA |
5257 | |
5258 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5259 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5260 | struct hlist_head *list; | |
5261 | ||
5262 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5263 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5264 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5265 | posted += io_poll_remove_one(req); |
5266 | } | |
221c5eb2 JA |
5267 | } |
5268 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5269 | |
8e2e1faf JA |
5270 | if (posted) |
5271 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5272 | |
5273 | return posted != 0; | |
221c5eb2 JA |
5274 | } |
5275 | ||
47f46768 JA |
5276 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5277 | { | |
78076bb6 | 5278 | struct hlist_head *list; |
47f46768 JA |
5279 | struct io_kiocb *req; |
5280 | ||
78076bb6 JA |
5281 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5282 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5283 | if (sqe_addr != req->user_data) |
5284 | continue; | |
5285 | if (io_poll_remove_one(req)) | |
eac406c6 | 5286 | return 0; |
b41e9852 | 5287 | return -EALREADY; |
47f46768 JA |
5288 | } |
5289 | ||
5290 | return -ENOENT; | |
5291 | } | |
5292 | ||
3529d8c2 JA |
5293 | static int io_poll_remove_prep(struct io_kiocb *req, |
5294 | const struct io_uring_sqe *sqe) | |
0969e783 | 5295 | { |
0969e783 JA |
5296 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5297 | return -EINVAL; | |
5298 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5299 | sqe->poll_events) | |
5300 | return -EINVAL; | |
5301 | ||
018043be | 5302 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5303 | return 0; |
5304 | } | |
5305 | ||
221c5eb2 JA |
5306 | /* |
5307 | * Find a running poll command that matches one specified in sqe->addr, | |
5308 | * and remove it if found. | |
5309 | */ | |
61e98203 | 5310 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5311 | { |
5312 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5313 | int ret; |
221c5eb2 | 5314 | |
221c5eb2 | 5315 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5316 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5317 | spin_unlock_irq(&ctx->completion_lock); |
5318 | ||
4e88d6e7 JA |
5319 | if (ret < 0) |
5320 | req_set_fail_links(req); | |
e1e16097 | 5321 | io_req_complete(req, ret); |
221c5eb2 JA |
5322 | return 0; |
5323 | } | |
5324 | ||
221c5eb2 JA |
5325 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5326 | void *key) | |
5327 | { | |
c2f2eb7d JA |
5328 | struct io_kiocb *req = wait->private; |
5329 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5330 | |
d7718a9d | 5331 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5332 | } |
5333 | ||
221c5eb2 JA |
5334 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5335 | struct poll_table_struct *p) | |
5336 | { | |
5337 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5338 | ||
e8c2bc1f | 5339 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5340 | } |
5341 | ||
3529d8c2 | 5342 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5343 | { |
5344 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5345 | u32 events; |
221c5eb2 JA |
5346 | |
5347 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5348 | return -EINVAL; | |
5349 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5350 | return -EINVAL; | |
5351 | ||
5769a351 JX |
5352 | events = READ_ONCE(sqe->poll32_events); |
5353 | #ifdef __BIG_ENDIAN | |
5354 | events = swahw32(events); | |
5355 | #endif | |
a31eb4a2 JX |
5356 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5357 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5358 | return 0; |
5359 | } | |
5360 | ||
61e98203 | 5361 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5362 | { |
5363 | struct io_poll_iocb *poll = &req->poll; | |
5364 | struct io_ring_ctx *ctx = req->ctx; | |
5365 | struct io_poll_table ipt; | |
0969e783 | 5366 | __poll_t mask; |
0969e783 | 5367 | |
d7718a9d | 5368 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5369 | |
d7718a9d JA |
5370 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5371 | io_poll_wake); | |
221c5eb2 | 5372 | |
8c838788 | 5373 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5374 | ipt.error = 0; |
b0dd8a41 | 5375 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5376 | } |
221c5eb2 JA |
5377 | spin_unlock_irq(&ctx->completion_lock); |
5378 | ||
8c838788 JA |
5379 | if (mask) { |
5380 | io_cqring_ev_posted(ctx); | |
014db007 | 5381 | io_put_req(req); |
221c5eb2 | 5382 | } |
8c838788 | 5383 | return ipt.error; |
221c5eb2 JA |
5384 | } |
5385 | ||
5262f567 JA |
5386 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5387 | { | |
ad8a48ac JA |
5388 | struct io_timeout_data *data = container_of(timer, |
5389 | struct io_timeout_data, timer); | |
5390 | struct io_kiocb *req = data->req; | |
5391 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5392 | unsigned long flags; |
5393 | ||
5262f567 | 5394 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5395 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5396 | atomic_set(&req->ctx->cq_timeouts, |
5397 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5398 | ||
78e19bbe | 5399 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5400 | io_commit_cqring(ctx); |
5401 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5402 | ||
5403 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5404 | req_set_fail_links(req); |
5262f567 JA |
5405 | io_put_req(req); |
5406 | return HRTIMER_NORESTART; | |
5407 | } | |
5408 | ||
fbd15848 PB |
5409 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5410 | __u64 user_data) | |
f254ac04 | 5411 | { |
fbd15848 | 5412 | struct io_timeout_data *io; |
47f46768 JA |
5413 | struct io_kiocb *req; |
5414 | int ret = -ENOENT; | |
f254ac04 | 5415 | |
135fcde8 | 5416 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5417 | if (user_data == req->user_data) { |
47f46768 JA |
5418 | ret = 0; |
5419 | break; | |
5420 | } | |
5421 | } | |
5422 | ||
5423 | if (ret == -ENOENT) | |
fbd15848 PB |
5424 | return ERR_PTR(ret); |
5425 | ||
5426 | io = req->async_data; | |
e8c2bc1f | 5427 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5428 | if (ret == -1) |
fbd15848 | 5429 | return ERR_PTR(-EALREADY); |
a71976f3 | 5430 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5431 | return req; |
5432 | } | |
47f46768 | 5433 | |
fbd15848 PB |
5434 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5435 | { | |
5436 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5437 | ||
5438 | if (IS_ERR(req)) | |
5439 | return PTR_ERR(req); | |
f254ac04 JA |
5440 | |
5441 | req_set_fail_links(req); | |
f254ac04 | 5442 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5443 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5444 | return 0; |
5445 | } | |
5446 | ||
9c8e11b3 PB |
5447 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5448 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5449 | { |
9c8e11b3 PB |
5450 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5451 | struct io_timeout_data *data; | |
47f46768 | 5452 | |
9c8e11b3 PB |
5453 | if (IS_ERR(req)) |
5454 | return PTR_ERR(req); | |
47f46768 | 5455 | |
9c8e11b3 PB |
5456 | req->timeout.off = 0; /* noseq */ |
5457 | data = req->async_data; | |
5458 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5459 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5460 | data->timer.function = io_timeout_fn; | |
5461 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5462 | return 0; | |
47f46768 JA |
5463 | } |
5464 | ||
3529d8c2 JA |
5465 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5466 | const struct io_uring_sqe *sqe) | |
b29472ee | 5467 | { |
9c8e11b3 PB |
5468 | struct io_timeout_rem *tr = &req->timeout_rem; |
5469 | ||
b29472ee JA |
5470 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5471 | return -EINVAL; | |
61710e43 DA |
5472 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5473 | return -EINVAL; | |
9c8e11b3 | 5474 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5475 | return -EINVAL; |
5476 | ||
9c8e11b3 PB |
5477 | tr->addr = READ_ONCE(sqe->addr); |
5478 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5479 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5480 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5481 | return -EINVAL; | |
5482 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5483 | return -EFAULT; | |
5484 | } else if (tr->flags) { | |
5485 | /* timeout removal doesn't support flags */ | |
b29472ee | 5486 | return -EINVAL; |
9c8e11b3 | 5487 | } |
b29472ee | 5488 | |
b29472ee JA |
5489 | return 0; |
5490 | } | |
5491 | ||
8662daec PB |
5492 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5493 | { | |
5494 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5495 | : HRTIMER_MODE_REL; | |
5496 | } | |
5497 | ||
11365043 JA |
5498 | /* |
5499 | * Remove or update an existing timeout command | |
5500 | */ | |
61e98203 | 5501 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5502 | { |
9c8e11b3 | 5503 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5504 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5505 | int ret; |
11365043 | 5506 | |
11365043 | 5507 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5508 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5509 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5510 | else |
5511 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5512 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5513 | |
47f46768 | 5514 | io_cqring_fill_event(req, ret); |
11365043 JA |
5515 | io_commit_cqring(ctx); |
5516 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5517 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5518 | if (ret < 0) |
5519 | req_set_fail_links(req); | |
ec9c02ad | 5520 | io_put_req(req); |
11365043 | 5521 | return 0; |
5262f567 JA |
5522 | } |
5523 | ||
3529d8c2 | 5524 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5525 | bool is_timeout_link) |
5262f567 | 5526 | { |
ad8a48ac | 5527 | struct io_timeout_data *data; |
a41525ab | 5528 | unsigned flags; |
56080b02 | 5529 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5530 | |
ad8a48ac | 5531 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5532 | return -EINVAL; |
ad8a48ac | 5533 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5534 | return -EINVAL; |
56080b02 | 5535 | if (off && is_timeout_link) |
2d28390a | 5536 | return -EINVAL; |
a41525ab JA |
5537 | flags = READ_ONCE(sqe->timeout_flags); |
5538 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5539 | return -EINVAL; |
bdf20073 | 5540 | |
bfe68a22 | 5541 | req->timeout.off = off; |
26a61679 | 5542 | |
e8c2bc1f | 5543 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5544 | return -ENOMEM; |
5545 | ||
e8c2bc1f | 5546 | data = req->async_data; |
ad8a48ac | 5547 | data->req = req; |
ad8a48ac JA |
5548 | |
5549 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5550 | return -EFAULT; |
5551 | ||
8662daec | 5552 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5553 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
dd59a3d5 | 5554 | io_req_track_inflight(req); |
ad8a48ac JA |
5555 | return 0; |
5556 | } | |
5557 | ||
61e98203 | 5558 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5559 | { |
ad8a48ac | 5560 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5561 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5562 | struct list_head *entry; |
bfe68a22 | 5563 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5564 | |
733f5c95 | 5565 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5566 | |
5262f567 JA |
5567 | /* |
5568 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5569 | * timeout event to be satisfied. If it isn't set, then this is |
5570 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5571 | */ |
8eb7e2d0 | 5572 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5573 | entry = ctx->timeout_list.prev; |
5574 | goto add; | |
5575 | } | |
5262f567 | 5576 | |
bfe68a22 PB |
5577 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5578 | req->timeout.target_seq = tail + off; | |
5262f567 | 5579 | |
f010505b MDG |
5580 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5581 | * This is safe because ->completion_lock is held, and submissions | |
5582 | * and completions are never mixed in the same ->completion_lock section. | |
5583 | */ | |
5584 | ctx->cq_last_tm_flush = tail; | |
5585 | ||
5262f567 JA |
5586 | /* |
5587 | * Insertion sort, ensuring the first entry in the list is always | |
5588 | * the one we need first. | |
5589 | */ | |
5262f567 | 5590 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5591 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5592 | timeout.list); | |
5262f567 | 5593 | |
8eb7e2d0 | 5594 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5595 | continue; |
bfe68a22 PB |
5596 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5597 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5598 | break; |
5599 | } | |
93bd25bb | 5600 | add: |
135fcde8 | 5601 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5602 | data->timer.function = io_timeout_fn; |
5603 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5604 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5605 | return 0; |
5606 | } | |
5262f567 | 5607 | |
f458dd84 PB |
5608 | struct io_cancel_data { |
5609 | struct io_ring_ctx *ctx; | |
5610 | u64 user_data; | |
5611 | }; | |
5612 | ||
62755e35 JA |
5613 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5614 | { | |
5615 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5616 | struct io_cancel_data *cd = data; |
62755e35 | 5617 | |
f458dd84 | 5618 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5619 | } |
5620 | ||
f458dd84 PB |
5621 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5622 | struct io_ring_ctx *ctx) | |
62755e35 | 5623 | { |
f458dd84 | 5624 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5625 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5626 | int ret = 0; |
5627 | ||
f458dd84 | 5628 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5629 | return -ENOENT; |
5630 | ||
f458dd84 | 5631 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5632 | switch (cancel_ret) { |
5633 | case IO_WQ_CANCEL_OK: | |
5634 | ret = 0; | |
5635 | break; | |
5636 | case IO_WQ_CANCEL_RUNNING: | |
5637 | ret = -EALREADY; | |
5638 | break; | |
5639 | case IO_WQ_CANCEL_NOTFOUND: | |
5640 | ret = -ENOENT; | |
5641 | break; | |
5642 | } | |
5643 | ||
e977d6d3 JA |
5644 | return ret; |
5645 | } | |
5646 | ||
47f46768 JA |
5647 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5648 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5649 | int success_ret) |
47f46768 JA |
5650 | { |
5651 | unsigned long flags; | |
5652 | int ret; | |
5653 | ||
f458dd84 | 5654 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5655 | if (ret != -ENOENT) { |
5656 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5657 | goto done; | |
5658 | } | |
5659 | ||
5660 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5661 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5662 | if (ret != -ENOENT) | |
5663 | goto done; | |
5664 | ret = io_poll_cancel(ctx, sqe_addr); | |
5665 | done: | |
b0dd8a41 JA |
5666 | if (!ret) |
5667 | ret = success_ret; | |
47f46768 JA |
5668 | io_cqring_fill_event(req, ret); |
5669 | io_commit_cqring(ctx); | |
5670 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5671 | io_cqring_ev_posted(ctx); | |
5672 | ||
4e88d6e7 JA |
5673 | if (ret < 0) |
5674 | req_set_fail_links(req); | |
014db007 | 5675 | io_put_req(req); |
47f46768 JA |
5676 | } |
5677 | ||
3529d8c2 JA |
5678 | static int io_async_cancel_prep(struct io_kiocb *req, |
5679 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5680 | { |
fbf23849 | 5681 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5682 | return -EINVAL; |
61710e43 DA |
5683 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5684 | return -EINVAL; | |
5685 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5686 | return -EINVAL; |
5687 | ||
fbf23849 JA |
5688 | req->cancel.addr = READ_ONCE(sqe->addr); |
5689 | return 0; | |
5690 | } | |
5691 | ||
61e98203 | 5692 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5693 | { |
5694 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5695 | u64 sqe_addr = req->cancel.addr; |
5696 | struct io_tctx_node *node; | |
5697 | int ret; | |
5698 | ||
5699 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5700 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5701 | spin_lock_irq(&ctx->completion_lock); | |
5702 | if (ret != -ENOENT) | |
5703 | goto done; | |
5704 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5705 | if (ret != -ENOENT) | |
5706 | goto done; | |
5707 | ret = io_poll_cancel(ctx, sqe_addr); | |
5708 | if (ret != -ENOENT) | |
5709 | goto done; | |
5710 | spin_unlock_irq(&ctx->completion_lock); | |
5711 | ||
5712 | /* slow path, try all io-wq's */ | |
5713 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5714 | ret = -ENOENT; | |
5715 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5716 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5717 | |
58f99373 PB |
5718 | if (!tctx || !tctx->io_wq) |
5719 | continue; | |
5720 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5721 | if (ret != -ENOENT) | |
5722 | break; | |
5723 | } | |
5724 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5725 | ||
5726 | spin_lock_irq(&ctx->completion_lock); | |
5727 | done: | |
5728 | io_cqring_fill_event(req, ret); | |
5729 | io_commit_cqring(ctx); | |
5730 | spin_unlock_irq(&ctx->completion_lock); | |
5731 | io_cqring_ev_posted(ctx); | |
5732 | ||
5733 | if (ret < 0) | |
5734 | req_set_fail_links(req); | |
5735 | io_put_req(req); | |
5262f567 JA |
5736 | return 0; |
5737 | } | |
5738 | ||
269bbe5f | 5739 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5740 | const struct io_uring_sqe *sqe) |
5741 | { | |
6ca56f84 JA |
5742 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5743 | return -EINVAL; | |
61710e43 DA |
5744 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5745 | return -EINVAL; | |
5746 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5747 | return -EINVAL; |
5748 | ||
269bbe5f BM |
5749 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5750 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5751 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5752 | return -EINVAL; |
269bbe5f | 5753 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5754 | return 0; |
5755 | } | |
5756 | ||
889fca73 | 5757 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5758 | { |
5759 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5760 | struct io_uring_rsrc_update up; |
05f3fb3c | 5761 | int ret; |
fbf23849 | 5762 | |
45d189c6 | 5763 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5764 | return -EAGAIN; |
05f3fb3c | 5765 | |
269bbe5f BM |
5766 | up.offset = req->rsrc_update.offset; |
5767 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5768 | |
5769 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5770 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5771 | mutex_unlock(&ctx->uring_lock); |
5772 | ||
5773 | if (ret < 0) | |
5774 | req_set_fail_links(req); | |
889fca73 | 5775 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5776 | return 0; |
5777 | } | |
5778 | ||
bfe76559 | 5779 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5780 | { |
d625c6ee | 5781 | switch (req->opcode) { |
e781573e | 5782 | case IORING_OP_NOP: |
bfe76559 | 5783 | return 0; |
f67676d1 JA |
5784 | case IORING_OP_READV: |
5785 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5786 | case IORING_OP_READ: |
bfe76559 | 5787 | return io_read_prep(req, sqe); |
f67676d1 JA |
5788 | case IORING_OP_WRITEV: |
5789 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5790 | case IORING_OP_WRITE: |
bfe76559 | 5791 | return io_write_prep(req, sqe); |
0969e783 | 5792 | case IORING_OP_POLL_ADD: |
bfe76559 | 5793 | return io_poll_add_prep(req, sqe); |
0969e783 | 5794 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5795 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5796 | case IORING_OP_FSYNC: |
1155c76a | 5797 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5798 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5799 | return io_sfr_prep(req, sqe); |
03b1230c | 5800 | case IORING_OP_SENDMSG: |
fddaface | 5801 | case IORING_OP_SEND: |
bfe76559 | 5802 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5803 | case IORING_OP_RECVMSG: |
fddaface | 5804 | case IORING_OP_RECV: |
bfe76559 | 5805 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5806 | case IORING_OP_CONNECT: |
bfe76559 | 5807 | return io_connect_prep(req, sqe); |
2d28390a | 5808 | case IORING_OP_TIMEOUT: |
bfe76559 | 5809 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5810 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5811 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5812 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5813 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5814 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5815 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5816 | case IORING_OP_ACCEPT: |
bfe76559 | 5817 | return io_accept_prep(req, sqe); |
d63d1b5e | 5818 | case IORING_OP_FALLOCATE: |
bfe76559 | 5819 | return io_fallocate_prep(req, sqe); |
15b71abe | 5820 | case IORING_OP_OPENAT: |
bfe76559 | 5821 | return io_openat_prep(req, sqe); |
b5dba59e | 5822 | case IORING_OP_CLOSE: |
bfe76559 | 5823 | return io_close_prep(req, sqe); |
05f3fb3c | 5824 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5825 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5826 | case IORING_OP_STATX: |
bfe76559 | 5827 | return io_statx_prep(req, sqe); |
4840e418 | 5828 | case IORING_OP_FADVISE: |
bfe76559 | 5829 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5830 | case IORING_OP_MADVISE: |
bfe76559 | 5831 | return io_madvise_prep(req, sqe); |
cebdb986 | 5832 | case IORING_OP_OPENAT2: |
bfe76559 | 5833 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5834 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5835 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5836 | case IORING_OP_SPLICE: |
bfe76559 | 5837 | return io_splice_prep(req, sqe); |
ddf0322d | 5838 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5839 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5840 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5841 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5842 | case IORING_OP_TEE: |
bfe76559 | 5843 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5844 | case IORING_OP_SHUTDOWN: |
5845 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5846 | case IORING_OP_RENAMEAT: |
5847 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5848 | case IORING_OP_UNLINKAT: |
5849 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5850 | } |
5851 | ||
bfe76559 PB |
5852 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5853 | req->opcode); | |
5854 | return-EINVAL; | |
5855 | } | |
5856 | ||
93642ef8 | 5857 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5858 | { |
93642ef8 PB |
5859 | switch (req->opcode) { |
5860 | case IORING_OP_READV: | |
5861 | case IORING_OP_READ_FIXED: | |
5862 | case IORING_OP_READ: | |
5863 | return io_rw_prep_async(req, READ); | |
5864 | case IORING_OP_WRITEV: | |
5865 | case IORING_OP_WRITE_FIXED: | |
5866 | case IORING_OP_WRITE: | |
5867 | return io_rw_prep_async(req, WRITE); | |
5868 | case IORING_OP_SENDMSG: | |
5869 | case IORING_OP_SEND: | |
5870 | return io_sendmsg_prep_async(req); | |
5871 | case IORING_OP_RECVMSG: | |
5872 | case IORING_OP_RECV: | |
5873 | return io_recvmsg_prep_async(req); | |
5874 | case IORING_OP_CONNECT: | |
5875 | return io_connect_prep_async(req); | |
5876 | } | |
5877 | return 0; | |
5878 | } | |
5879 | ||
be7053b7 | 5880 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5881 | { |
be7053b7 | 5882 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5883 | return 0; |
be7053b7 | 5884 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5885 | if (req->async_data) |
be7053b7 PB |
5886 | return 0; |
5887 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5888 | return -EAGAIN; |
be7053b7 | 5889 | return io_req_prep_async(req); |
f67676d1 JA |
5890 | } |
5891 | ||
9cf7c104 PB |
5892 | static u32 io_get_sequence(struct io_kiocb *req) |
5893 | { | |
5894 | struct io_kiocb *pos; | |
5895 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5896 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5897 | |
f2f87370 PB |
5898 | io_for_each_link(pos, req) |
5899 | nr_reqs++; | |
9cf7c104 PB |
5900 | |
5901 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5902 | return total_submitted - nr_reqs; | |
5903 | } | |
5904 | ||
be7053b7 | 5905 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5906 | { |
a197f664 | 5907 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5908 | struct io_defer_entry *de; |
f67676d1 | 5909 | int ret; |
9cf7c104 | 5910 | u32 seq; |
de0617e4 | 5911 | |
9d858b21 | 5912 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5913 | if (likely(list_empty_careful(&ctx->defer_list) && |
5914 | !(req->flags & REQ_F_IO_DRAIN))) | |
5915 | return 0; | |
5916 | ||
5917 | seq = io_get_sequence(req); | |
5918 | /* Still a chance to pass the sequence check */ | |
5919 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5920 | return 0; |
5921 | ||
be7053b7 PB |
5922 | ret = io_req_defer_prep(req); |
5923 | if (ret) | |
5924 | return ret; | |
cbdcb435 | 5925 | io_prep_async_link(req); |
27dc8338 PB |
5926 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5927 | if (!de) | |
5928 | return -ENOMEM; | |
2d28390a | 5929 | |
de0617e4 | 5930 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5931 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5932 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5933 | kfree(de); |
ae34817b PB |
5934 | io_queue_async_work(req); |
5935 | return -EIOCBQUEUED; | |
de0617e4 JA |
5936 | } |
5937 | ||
915967f6 | 5938 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5939 | de->req = req; |
9cf7c104 | 5940 | de->seq = seq; |
27dc8338 | 5941 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5942 | spin_unlock_irq(&ctx->completion_lock); |
5943 | return -EIOCBQUEUED; | |
5944 | } | |
5945 | ||
3ca405eb | 5946 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5947 | { |
0e1b6fe3 PB |
5948 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5949 | switch (req->opcode) { | |
5950 | case IORING_OP_READV: | |
5951 | case IORING_OP_READ_FIXED: | |
5952 | case IORING_OP_READ: | |
bcda7baa | 5953 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5954 | break; |
5955 | case IORING_OP_RECVMSG: | |
5956 | case IORING_OP_RECV: | |
bcda7baa | 5957 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5958 | break; |
5959 | } | |
5960 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5961 | } |
5962 | ||
0e1b6fe3 PB |
5963 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5964 | switch (req->opcode) { | |
5965 | case IORING_OP_READV: | |
5966 | case IORING_OP_READ_FIXED: | |
5967 | case IORING_OP_READ: | |
5968 | case IORING_OP_WRITEV: | |
5969 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5970 | case IORING_OP_WRITE: { |
5971 | struct io_async_rw *io = req->async_data; | |
5972 | if (io->free_iovec) | |
5973 | kfree(io->free_iovec); | |
0e1b6fe3 | 5974 | break; |
e8c2bc1f | 5975 | } |
0e1b6fe3 | 5976 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5977 | case IORING_OP_SENDMSG: { |
5978 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5979 | |
5980 | kfree(io->free_iov); | |
0e1b6fe3 | 5981 | break; |
e8c2bc1f | 5982 | } |
0e1b6fe3 PB |
5983 | case IORING_OP_SPLICE: |
5984 | case IORING_OP_TEE: | |
5985 | io_put_file(req, req->splice.file_in, | |
5986 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5987 | break; | |
f3cd4850 JA |
5988 | case IORING_OP_OPENAT: |
5989 | case IORING_OP_OPENAT2: | |
5990 | if (req->open.filename) | |
5991 | putname(req->open.filename); | |
5992 | break; | |
80a261fd JA |
5993 | case IORING_OP_RENAMEAT: |
5994 | putname(req->rename.oldpath); | |
5995 | putname(req->rename.newpath); | |
5996 | break; | |
14a1143b JA |
5997 | case IORING_OP_UNLINKAT: |
5998 | putname(req->unlink.filename); | |
5999 | break; | |
0e1b6fe3 PB |
6000 | } |
6001 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6002 | } |
99bc4c38 PB |
6003 | } |
6004 | ||
889fca73 | 6005 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6006 | { |
a197f664 | 6007 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6008 | const struct cred *creds = NULL; |
d625c6ee | 6009 | int ret; |
2b188cc1 | 6010 | |
003e8dcc JA |
6011 | if (req->work.creds && req->work.creds != current_cred()) |
6012 | creds = override_creds(req->work.creds); | |
5730b27e | 6013 | |
d625c6ee | 6014 | switch (req->opcode) { |
2b188cc1 | 6015 | case IORING_OP_NOP: |
889fca73 | 6016 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6017 | break; |
6018 | case IORING_OP_READV: | |
edafccee | 6019 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6020 | case IORING_OP_READ: |
889fca73 | 6021 | ret = io_read(req, issue_flags); |
edafccee | 6022 | break; |
3529d8c2 | 6023 | case IORING_OP_WRITEV: |
edafccee | 6024 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6025 | case IORING_OP_WRITE: |
889fca73 | 6026 | ret = io_write(req, issue_flags); |
2b188cc1 | 6027 | break; |
c992fe29 | 6028 | case IORING_OP_FSYNC: |
45d189c6 | 6029 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6030 | break; |
221c5eb2 | 6031 | case IORING_OP_POLL_ADD: |
61e98203 | 6032 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6033 | break; |
6034 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6035 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6036 | break; |
5d17b4a4 | 6037 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6038 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6039 | break; |
0fa03c62 | 6040 | case IORING_OP_SENDMSG: |
889fca73 | 6041 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6042 | break; |
fddaface | 6043 | case IORING_OP_SEND: |
889fca73 | 6044 | ret = io_send(req, issue_flags); |
0fa03c62 | 6045 | break; |
aa1fa28f | 6046 | case IORING_OP_RECVMSG: |
889fca73 | 6047 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6048 | break; |
fddaface | 6049 | case IORING_OP_RECV: |
889fca73 | 6050 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6051 | break; |
5262f567 | 6052 | case IORING_OP_TIMEOUT: |
61e98203 | 6053 | ret = io_timeout(req, issue_flags); |
5262f567 | 6054 | break; |
11365043 | 6055 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6056 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6057 | break; |
17f2fe35 | 6058 | case IORING_OP_ACCEPT: |
889fca73 | 6059 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6060 | break; |
f8e85cf2 | 6061 | case IORING_OP_CONNECT: |
889fca73 | 6062 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6063 | break; |
62755e35 | 6064 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6065 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6066 | break; |
d63d1b5e | 6067 | case IORING_OP_FALLOCATE: |
45d189c6 | 6068 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6069 | break; |
15b71abe | 6070 | case IORING_OP_OPENAT: |
45d189c6 | 6071 | ret = io_openat(req, issue_flags); |
15b71abe | 6072 | break; |
b5dba59e | 6073 | case IORING_OP_CLOSE: |
889fca73 | 6074 | ret = io_close(req, issue_flags); |
b5dba59e | 6075 | break; |
05f3fb3c | 6076 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6077 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6078 | break; |
eddc7ef5 | 6079 | case IORING_OP_STATX: |
45d189c6 | 6080 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6081 | break; |
4840e418 | 6082 | case IORING_OP_FADVISE: |
45d189c6 | 6083 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6084 | break; |
c1ca757b | 6085 | case IORING_OP_MADVISE: |
45d189c6 | 6086 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6087 | break; |
cebdb986 | 6088 | case IORING_OP_OPENAT2: |
45d189c6 | 6089 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6090 | break; |
3e4827b0 | 6091 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6092 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6093 | break; |
7d67af2c | 6094 | case IORING_OP_SPLICE: |
45d189c6 | 6095 | ret = io_splice(req, issue_flags); |
7d67af2c | 6096 | break; |
ddf0322d | 6097 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6098 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6099 | break; |
067524e9 | 6100 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6101 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6102 | break; |
f2a8d5c7 | 6103 | case IORING_OP_TEE: |
45d189c6 | 6104 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6105 | break; |
36f4fa68 | 6106 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6107 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6108 | break; |
80a261fd | 6109 | case IORING_OP_RENAMEAT: |
45d189c6 | 6110 | ret = io_renameat(req, issue_flags); |
80a261fd | 6111 | break; |
14a1143b | 6112 | case IORING_OP_UNLINKAT: |
45d189c6 | 6113 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6114 | break; |
2b188cc1 JA |
6115 | default: |
6116 | ret = -EINVAL; | |
6117 | break; | |
6118 | } | |
6119 | ||
5730b27e JA |
6120 | if (creds) |
6121 | revert_creds(creds); | |
6122 | ||
def596e9 JA |
6123 | if (ret) |
6124 | return ret; | |
6125 | ||
b532576e JA |
6126 | /* If the op doesn't have a file, we're not polling for it */ |
6127 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6128 | const bool in_async = io_wq_current_is_worker(); |
6129 | ||
11ba820b JA |
6130 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6131 | if (in_async) | |
6132 | mutex_lock(&ctx->uring_lock); | |
6133 | ||
2e9dbe90 | 6134 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6135 | |
6136 | if (in_async) | |
6137 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6138 | } |
6139 | ||
6140 | return 0; | |
2b188cc1 JA |
6141 | } |
6142 | ||
5280f7e5 | 6143 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6144 | { |
6145 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6146 | struct io_kiocb *timeout; |
561fb04a | 6147 | int ret = 0; |
2b188cc1 | 6148 | |
6df1db6b PB |
6149 | timeout = io_prep_linked_timeout(req); |
6150 | if (timeout) | |
6151 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6152 | |
4014d943 | 6153 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6154 | ret = -ECANCELED; |
31b51510 | 6155 | |
561fb04a | 6156 | if (!ret) { |
561fb04a | 6157 | do { |
889fca73 | 6158 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6159 | /* |
6160 | * We can get EAGAIN for polled IO even though we're | |
6161 | * forcing a sync submission from here, since we can't | |
6162 | * wait for request slots on the block side. | |
6163 | */ | |
6164 | if (ret != -EAGAIN) | |
6165 | break; | |
6166 | cond_resched(); | |
6167 | } while (1); | |
6168 | } | |
31b51510 | 6169 | |
a3df7698 | 6170 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6171 | if (ret) { |
a3df7698 PB |
6172 | /* io-wq is going to take one down */ |
6173 | refcount_inc(&req->refs); | |
6174 | io_req_task_queue_fail(req, ret); | |
edafccee | 6175 | } |
2b188cc1 JA |
6176 | } |
6177 | ||
65e19f54 JA |
6178 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6179 | int index) | |
6180 | { | |
269bbe5f | 6181 | struct fixed_rsrc_table *table; |
65e19f54 | 6182 | |
05f3fb3c | 6183 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6184 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6185 | } |
6186 | ||
8371adf5 PB |
6187 | static struct file *io_file_get(struct io_submit_state *state, |
6188 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6189 | { |
a197f664 | 6190 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6191 | struct file *file; |
09bb8394 | 6192 | |
8da11c19 | 6193 | if (fixed) { |
479f517b | 6194 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6195 | return NULL; |
b7620121 | 6196 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6197 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6198 | io_set_resource_node(req); |
09bb8394 | 6199 | } else { |
c826bd7a | 6200 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6201 | file = __io_file_get(state, fd); |
09bb8394 JA |
6202 | } |
6203 | ||
ce3d5aae PB |
6204 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6205 | io_req_track_inflight(req); | |
8371adf5 | 6206 | return file; |
09bb8394 JA |
6207 | } |
6208 | ||
2665abfd | 6209 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6210 | { |
ad8a48ac JA |
6211 | struct io_timeout_data *data = container_of(timer, |
6212 | struct io_timeout_data, timer); | |
90cd7e42 | 6213 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6214 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6215 | unsigned long flags; |
2665abfd JA |
6216 | |
6217 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6218 | prev = req->timeout.head; |
6219 | req->timeout.head = NULL; | |
2665abfd JA |
6220 | |
6221 | /* | |
6222 | * We don't expect the list to be empty, that will only happen if we | |
6223 | * race with the completion of the linked work. | |
6224 | */ | |
90cd7e42 | 6225 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6226 | io_remove_next_linked(prev); |
90cd7e42 PB |
6227 | else |
6228 | prev = NULL; | |
2665abfd JA |
6229 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6230 | ||
6231 | if (prev) { | |
014db007 | 6232 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6233 | io_put_req_deferred(prev, 1); |
47f46768 | 6234 | } else { |
9ae1f8dd PB |
6235 | io_req_complete_post(req, -ETIME, 0); |
6236 | io_put_req_deferred(req, 1); | |
2665abfd | 6237 | } |
2665abfd JA |
6238 | return HRTIMER_NORESTART; |
6239 | } | |
6240 | ||
7271ef3a | 6241 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6242 | { |
76a46e06 | 6243 | /* |
f2f87370 PB |
6244 | * If the back reference is NULL, then our linked request finished |
6245 | * before we got a chance to setup the timer | |
76a46e06 | 6246 | */ |
90cd7e42 | 6247 | if (req->timeout.head) { |
e8c2bc1f | 6248 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6249 | |
ad8a48ac JA |
6250 | data->timer.function = io_link_timeout_fn; |
6251 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6252 | data->mode); | |
2665abfd | 6253 | } |
7271ef3a JA |
6254 | } |
6255 | ||
6256 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6257 | { | |
6258 | struct io_ring_ctx *ctx = req->ctx; | |
6259 | ||
6260 | spin_lock_irq(&ctx->completion_lock); | |
6261 | __io_queue_linked_timeout(req); | |
76a46e06 | 6262 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6263 | |
2665abfd | 6264 | /* drop submission reference */ |
76a46e06 JA |
6265 | io_put_req(req); |
6266 | } | |
2665abfd | 6267 | |
ad8a48ac | 6268 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6269 | { |
f2f87370 | 6270 | struct io_kiocb *nxt = req->link; |
2665abfd | 6271 | |
f2f87370 PB |
6272 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6273 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6274 | return NULL; |
2665abfd | 6275 | |
90cd7e42 | 6276 | nxt->timeout.head = req; |
900fad45 | 6277 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6278 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6279 | return nxt; |
2665abfd JA |
6280 | } |
6281 | ||
c5eef2b9 | 6282 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6283 | { |
d3d7298d | 6284 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6285 | int ret; |
2b188cc1 | 6286 | |
c5eef2b9 | 6287 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6288 | |
491381ce JA |
6289 | /* |
6290 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6291 | * doesn't support non-blocking read/write attempts | |
6292 | */ | |
24c74678 | 6293 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6294 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6295 | /* |
6296 | * Queued up for async execution, worker will release | |
6297 | * submit reference when the iocb is actually submitted. | |
6298 | */ | |
6299 | io_queue_async_work(req); | |
2b188cc1 | 6300 | } |
0d63c148 PB |
6301 | } else if (likely(!ret)) { |
6302 | /* drop submission reference */ | |
e342c807 | 6303 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6304 | struct io_ring_ctx *ctx = req->ctx; |
6305 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6306 | |
6dd0be1e | 6307 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6308 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6309 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6310 | } else { |
d3d7298d | 6311 | io_put_req(req); |
0d63c148 PB |
6312 | } |
6313 | } else { | |
4e88d6e7 | 6314 | req_set_fail_links(req); |
e65ef56d | 6315 | io_put_req(req); |
e1e16097 | 6316 | io_req_complete(req, ret); |
9e645e11 | 6317 | } |
d3d7298d PB |
6318 | if (linked_timeout) |
6319 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6320 | } |
6321 | ||
be7053b7 | 6322 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6323 | { |
6324 | int ret; | |
6325 | ||
be7053b7 | 6326 | ret = io_req_defer(req); |
4fe2c963 JL |
6327 | if (ret) { |
6328 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6329 | fail_req: |
4e88d6e7 | 6330 | req_set_fail_links(req); |
e1e16097 JA |
6331 | io_put_req(req); |
6332 | io_req_complete(req, ret); | |
4fe2c963 | 6333 | } |
2550878f | 6334 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6335 | ret = io_req_defer_prep(req); |
6336 | if (unlikely(ret)) | |
6337 | goto fail_req; | |
ce35a47a JA |
6338 | io_queue_async_work(req); |
6339 | } else { | |
c5eef2b9 | 6340 | __io_queue_sqe(req); |
ce35a47a | 6341 | } |
4fe2c963 JL |
6342 | } |
6343 | ||
b16fed66 PB |
6344 | /* |
6345 | * Check SQE restrictions (opcode and flags). | |
6346 | * | |
6347 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6348 | */ | |
6349 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6350 | struct io_kiocb *req, | |
6351 | unsigned int sqe_flags) | |
4fe2c963 | 6352 | { |
b16fed66 PB |
6353 | if (!ctx->restricted) |
6354 | return true; | |
6355 | ||
6356 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6357 | return false; | |
6358 | ||
6359 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6360 | ctx->restrictions.sqe_flags_required) | |
6361 | return false; | |
6362 | ||
6363 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6364 | ctx->restrictions.sqe_flags_required)) | |
6365 | return false; | |
6366 | ||
6367 | return true; | |
4fe2c963 JL |
6368 | } |
6369 | ||
b16fed66 PB |
6370 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6371 | const struct io_uring_sqe *sqe) | |
6372 | { | |
6373 | struct io_submit_state *state; | |
6374 | unsigned int sqe_flags; | |
003e8dcc | 6375 | int personality, ret = 0; |
b16fed66 PB |
6376 | |
6377 | req->opcode = READ_ONCE(sqe->opcode); | |
6378 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6379 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6380 | req->user_data = READ_ONCE(sqe->user_data); | |
6381 | req->async_data = NULL; | |
6382 | req->file = NULL; | |
6383 | req->ctx = ctx; | |
6384 | req->link = NULL; | |
6385 | req->fixed_rsrc_refs = NULL; | |
6386 | /* one is dropped after submission, the other at completion */ | |
6387 | refcount_set(&req->refs, 2); | |
6388 | req->task = current; | |
6389 | req->result = 0; | |
93e68e03 JA |
6390 | req->work.list.next = NULL; |
6391 | req->work.creds = NULL; | |
6392 | req->work.flags = 0; | |
b16fed66 PB |
6393 | |
6394 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6395 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6396 | req->flags = 0; | |
b16fed66 | 6397 | return -EINVAL; |
ebf4a5db | 6398 | } |
b16fed66 PB |
6399 | |
6400 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6401 | return -EINVAL; | |
6402 | ||
b16fed66 PB |
6403 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6404 | return -EACCES; | |
6405 | ||
6406 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6407 | !io_op_defs[req->opcode].buffer_select) | |
6408 | return -EOPNOTSUPP; | |
863e0560 | 6409 | |
003e8dcc JA |
6410 | personality = READ_ONCE(sqe->personality); |
6411 | if (personality) { | |
61cf9370 | 6412 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6413 | if (!req->work.creds) |
6414 | return -EINVAL; | |
6415 | get_cred(req->work.creds); | |
003e8dcc | 6416 | } |
b16fed66 PB |
6417 | state = &ctx->submit_state; |
6418 | ||
6419 | /* | |
6420 | * Plug now if we have more than 1 IO left after this, and the target | |
6421 | * is potentially a read/write to block based storage. | |
6422 | */ | |
6423 | if (!state->plug_started && state->ios_left > 1 && | |
6424 | io_op_defs[req->opcode].plug) { | |
6425 | blk_start_plug(&state->plug); | |
6426 | state->plug_started = true; | |
6427 | } | |
6428 | ||
6429 | if (io_op_defs[req->opcode].needs_file) { | |
6430 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6431 | ||
6432 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6433 | if (unlikely(!req->file)) | |
6434 | ret = -EBADF; | |
6435 | } | |
6436 | ||
6437 | state->ios_left--; | |
6438 | return ret; | |
6439 | } | |
6440 | ||
a6b8cadc | 6441 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6442 | const struct io_uring_sqe *sqe) |
9e645e11 | 6443 | { |
a1ab7b35 | 6444 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6445 | int ret; |
9e645e11 | 6446 | |
a6b8cadc PB |
6447 | ret = io_init_req(ctx, req, sqe); |
6448 | if (unlikely(ret)) { | |
6449 | fail_req: | |
6450 | io_put_req(req); | |
6451 | io_req_complete(req, ret); | |
de59bc10 PB |
6452 | if (link->head) { |
6453 | /* fail even hard links since we don't submit */ | |
cf109604 | 6454 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6455 | io_put_req(link->head); |
6456 | io_req_complete(link->head, -ECANCELED); | |
6457 | link->head = NULL; | |
6458 | } | |
a6b8cadc PB |
6459 | return ret; |
6460 | } | |
be7053b7 PB |
6461 | ret = io_req_prep(req, sqe); |
6462 | if (unlikely(ret)) | |
6463 | goto fail_req; | |
a6b8cadc | 6464 | |
be7053b7 | 6465 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6466 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6467 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6468 | ||
9e645e11 JA |
6469 | /* |
6470 | * If we already have a head request, queue this one for async | |
6471 | * submittal once the head completes. If we don't have a head but | |
6472 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6473 | * submitted sync once the chain is complete. If none of those | |
6474 | * conditions are true (normal request), then just queue it. | |
6475 | */ | |
863e0560 PB |
6476 | if (link->head) { |
6477 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6478 | |
8cdf2193 PB |
6479 | /* |
6480 | * Taking sequential execution of a link, draining both sides | |
6481 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6482 | * requests in the link. So, it drains the head and the | |
6483 | * next after the link request. The last one is done via | |
6484 | * drain_next flag to persist the effect across calls. | |
6485 | */ | |
ef4ff581 | 6486 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6487 | head->flags |= REQ_F_IO_DRAIN; |
6488 | ctx->drain_next = 1; | |
6489 | } | |
be7053b7 | 6490 | ret = io_req_defer_prep(req); |
cf109604 | 6491 | if (unlikely(ret)) |
a6b8cadc | 6492 | goto fail_req; |
9d76377f | 6493 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6494 | link->last->link = req; |
863e0560 | 6495 | link->last = req; |
32fe525b PB |
6496 | |
6497 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6498 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6499 | io_queue_sqe(head); |
863e0560 | 6500 | link->head = NULL; |
32fe525b | 6501 | } |
9e645e11 | 6502 | } else { |
711be031 PB |
6503 | if (unlikely(ctx->drain_next)) { |
6504 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6505 | ctx->drain_next = 0; |
711be031 | 6506 | } |
ef4ff581 | 6507 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6508 | link->head = req; |
6509 | link->last = req; | |
711be031 | 6510 | } else { |
be7053b7 | 6511 | io_queue_sqe(req); |
711be031 | 6512 | } |
9e645e11 | 6513 | } |
2e6e1fde | 6514 | |
1d4240cc | 6515 | return 0; |
9e645e11 JA |
6516 | } |
6517 | ||
9a56a232 JA |
6518 | /* |
6519 | * Batched submission is done, ensure local IO is flushed out. | |
6520 | */ | |
ba88ff11 PB |
6521 | static void io_submit_state_end(struct io_submit_state *state, |
6522 | struct io_ring_ctx *ctx) | |
9a56a232 | 6523 | { |
a1ab7b35 | 6524 | if (state->link.head) |
de59bc10 | 6525 | io_queue_sqe(state->link.head); |
6dd0be1e | 6526 | if (state->comp.nr) |
ba88ff11 | 6527 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6528 | if (state->plug_started) |
6529 | blk_finish_plug(&state->plug); | |
9f13c35b | 6530 | io_state_file_put(state); |
9a56a232 JA |
6531 | } |
6532 | ||
6533 | /* | |
6534 | * Start submission side cache. | |
6535 | */ | |
6536 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6537 | unsigned int max_ios) |
9a56a232 | 6538 | { |
27926b68 | 6539 | state->plug_started = false; |
9a56a232 | 6540 | state->ios_left = max_ios; |
a1ab7b35 PB |
6541 | /* set only head, no need to init link_last in advance */ |
6542 | state->link.head = NULL; | |
9a56a232 JA |
6543 | } |
6544 | ||
2b188cc1 JA |
6545 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6546 | { | |
75b28aff | 6547 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6548 | |
caf582c6 PB |
6549 | /* |
6550 | * Ensure any loads from the SQEs are done at this point, | |
6551 | * since once we write the new head, the application could | |
6552 | * write new data to them. | |
6553 | */ | |
6554 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6555 | } |
6556 | ||
2b188cc1 | 6557 | /* |
3529d8c2 | 6558 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6559 | * that is mapped by userspace. This means that care needs to be taken to |
6560 | * ensure that reads are stable, as we cannot rely on userspace always | |
6561 | * being a good citizen. If members of the sqe are validated and then later | |
6562 | * used, it's important that those reads are done through READ_ONCE() to | |
6563 | * prevent a re-load down the line. | |
6564 | */ | |
709b302f | 6565 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6566 | { |
75b28aff | 6567 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6568 | unsigned head; |
6569 | ||
6570 | /* | |
6571 | * The cached sq head (or cq tail) serves two purposes: | |
6572 | * | |
6573 | * 1) allows us to batch the cost of updating the user visible | |
6574 | * head updates. | |
6575 | * 2) allows the kernel side to track the head on its own, even | |
6576 | * though the application is the one updating it. | |
6577 | */ | |
4fccfcbb | 6578 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6579 | if (likely(head < ctx->sq_entries)) |
6580 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6581 | |
6582 | /* drop invalid entries */ | |
498ccd9e | 6583 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6584 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6585 | return NULL; |
6586 | } | |
6587 | ||
0f212204 | 6588 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6589 | { |
46c4e16a | 6590 | int submitted = 0; |
6c271ce2 | 6591 | |
c4a2ed72 | 6592 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6593 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6594 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6595 | return -EBUSY; |
6596 | } | |
6c271ce2 | 6597 | |
ee7d46d9 PB |
6598 | /* make sure SQ entry isn't read before tail */ |
6599 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6600 | |
2b85edfc PB |
6601 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6602 | return -EAGAIN; | |
6c271ce2 | 6603 | |
d8a6df10 | 6604 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6605 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6606 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6607 | |
46c4e16a | 6608 | while (submitted < nr) { |
3529d8c2 | 6609 | const struct io_uring_sqe *sqe; |
196be95c | 6610 | struct io_kiocb *req; |
fb5ccc98 | 6611 | |
258b29a9 | 6612 | req = io_alloc_req(ctx); |
196be95c PB |
6613 | if (unlikely(!req)) { |
6614 | if (!submitted) | |
6615 | submitted = -EAGAIN; | |
fb5ccc98 | 6616 | break; |
196be95c | 6617 | } |
4fccfcbb PB |
6618 | sqe = io_get_sqe(ctx); |
6619 | if (unlikely(!sqe)) { | |
6620 | kmem_cache_free(req_cachep, req); | |
6621 | break; | |
6622 | } | |
d3656344 JA |
6623 | /* will complete beyond this point, count as submitted */ |
6624 | submitted++; | |
a1ab7b35 | 6625 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6626 | break; |
6c271ce2 JA |
6627 | } |
6628 | ||
9466f437 PB |
6629 | if (unlikely(submitted != nr)) { |
6630 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6631 | struct io_uring_task *tctx = current->io_uring; |
6632 | int unused = nr - ref_used; | |
9466f437 | 6633 | |
d8a6df10 JA |
6634 | percpu_ref_put_many(&ctx->refs, unused); |
6635 | percpu_counter_sub(&tctx->inflight, unused); | |
6636 | put_task_struct_many(current, unused); | |
9466f437 | 6637 | } |
6c271ce2 | 6638 | |
a1ab7b35 | 6639 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6640 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6641 | io_commit_sqring(ctx); | |
6642 | ||
6c271ce2 JA |
6643 | return submitted; |
6644 | } | |
6645 | ||
23b3628e XW |
6646 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6647 | { | |
6648 | /* Tell userspace we may need a wakeup call */ | |
6649 | spin_lock_irq(&ctx->completion_lock); | |
6650 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6651 | spin_unlock_irq(&ctx->completion_lock); | |
6652 | } | |
6653 | ||
6654 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6655 | { | |
6656 | spin_lock_irq(&ctx->completion_lock); | |
6657 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6658 | spin_unlock_irq(&ctx->completion_lock); | |
6659 | } | |
6660 | ||
08369246 | 6661 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6662 | { |
c8d1ba58 | 6663 | unsigned int to_submit; |
bdcd3eab | 6664 | int ret = 0; |
6c271ce2 | 6665 | |
c8d1ba58 | 6666 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6667 | /* if we're handling multiple rings, cap submit size for fairness */ |
6668 | if (cap_entries && to_submit > 8) | |
6669 | to_submit = 8; | |
6670 | ||
906a3c6f | 6671 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6672 | unsigned nr_events = 0; |
a4c0b3de | 6673 | |
c8d1ba58 | 6674 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6675 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6676 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6677 | |
0298ef96 PB |
6678 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6679 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6680 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6681 | mutex_unlock(&ctx->uring_lock); |
6682 | } | |
6c271ce2 | 6683 | |
90554200 JA |
6684 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6685 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6686 | |
08369246 XW |
6687 | return ret; |
6688 | } | |
6c271ce2 | 6689 | |
08369246 XW |
6690 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6691 | { | |
6692 | struct io_ring_ctx *ctx; | |
6693 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6694 | |
08369246 XW |
6695 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6696 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6697 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6698 | } |
c1edbf5f | 6699 | |
08369246 | 6700 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6701 | } |
6c271ce2 | 6702 | |
c8d1ba58 JA |
6703 | static int io_sq_thread(void *data) |
6704 | { | |
69fb2131 JA |
6705 | struct io_sq_data *sqd = data; |
6706 | struct io_ring_ctx *ctx; | |
a0d9205f | 6707 | unsigned long timeout = 0; |
37d1e2e3 | 6708 | char buf[TASK_COMM_LEN]; |
08369246 | 6709 | DEFINE_WAIT(wait); |
6c271ce2 | 6710 | |
37d1e2e3 JA |
6711 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6712 | set_task_comm(current, buf); | |
37d1e2e3 JA |
6713 | current->pf_io_worker = NULL; |
6714 | ||
6715 | if (sqd->sq_cpu != -1) | |
6716 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6717 | else | |
6718 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6719 | current->flags |= PF_NO_SETAFFINITY; | |
6720 | ||
09a6f4ef | 6721 | mutex_lock(&sqd->lock); |
05962f95 | 6722 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6723 | int ret; |
6724 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6725 | |
05962f95 | 6726 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
09a6f4ef | 6727 | mutex_unlock(&sqd->lock); |
05962f95 | 6728 | cond_resched(); |
09a6f4ef | 6729 | mutex_lock(&sqd->lock); |
521d6a73 | 6730 | io_run_task_work(); |
b7f5a0bf | 6731 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6732 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6733 | continue; |
08369246 | 6734 | } |
37d1e2e3 JA |
6735 | if (fatal_signal_pending(current)) |
6736 | break; | |
08369246 | 6737 | sqt_spin = false; |
e95eee2d | 6738 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6739 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6740 | const struct cred *creds = NULL; |
6741 | ||
6742 | if (ctx->sq_creds != current_cred()) | |
6743 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6744 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6745 | if (creds) |
6746 | revert_creds(creds); | |
08369246 XW |
6747 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6748 | sqt_spin = true; | |
69fb2131 | 6749 | } |
6c271ce2 | 6750 | |
08369246 | 6751 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6752 | io_run_task_work(); |
6753 | cond_resched(); | |
08369246 XW |
6754 | if (sqt_spin) |
6755 | timeout = jiffies + sqd->sq_thread_idle; | |
6756 | continue; | |
6757 | } | |
6758 | ||
08369246 XW |
6759 | needs_sched = true; |
6760 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6761 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6762 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6763 | !list_empty_careful(&ctx->iopoll_list)) { | |
6764 | needs_sched = false; | |
6765 | break; | |
6766 | } | |
6767 | if (io_sqring_entries(ctx)) { | |
6768 | needs_sched = false; | |
6769 | break; | |
6770 | } | |
6771 | } | |
6772 | ||
05962f95 | 6773 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6774 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6775 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6776 | |
09a6f4ef | 6777 | mutex_unlock(&sqd->lock); |
69fb2131 | 6778 | schedule(); |
16efa4fc | 6779 | try_to_freeze(); |
09a6f4ef | 6780 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6781 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6782 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6783 | } |
08369246 XW |
6784 | |
6785 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6786 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6787 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6788 | } |
28cea78a | 6789 | |
521d6a73 PB |
6790 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6791 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6792 | sqd->thread = NULL; |
05962f95 | 6793 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6794 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6795 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6796 | |
6797 | io_run_task_work(); | |
b7f5a0bf | 6798 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6799 | complete(&sqd->exited); |
6800 | do_exit(0); | |
6c271ce2 JA |
6801 | } |
6802 | ||
bda52162 JA |
6803 | struct io_wait_queue { |
6804 | struct wait_queue_entry wq; | |
6805 | struct io_ring_ctx *ctx; | |
6806 | unsigned to_wait; | |
6807 | unsigned nr_timeouts; | |
6808 | }; | |
6809 | ||
6c503150 | 6810 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6811 | { |
6812 | struct io_ring_ctx *ctx = iowq->ctx; | |
6813 | ||
6814 | /* | |
d195a66e | 6815 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6816 | * started waiting. For timeouts, we always want to return to userspace, |
6817 | * regardless of event count. | |
6818 | */ | |
6c503150 | 6819 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6820 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6821 | } | |
6822 | ||
6823 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6824 | int wake_flags, void *key) | |
6825 | { | |
6826 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6827 | wq); | |
6828 | ||
6c503150 PB |
6829 | /* |
6830 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6831 | * the task, and the next invocation will do it. | |
6832 | */ | |
6833 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6834 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6835 | return -1; | |
bda52162 JA |
6836 | } |
6837 | ||
af9c1a44 JA |
6838 | static int io_run_task_work_sig(void) |
6839 | { | |
6840 | if (io_run_task_work()) | |
6841 | return 1; | |
6842 | if (!signal_pending(current)) | |
6843 | return 0; | |
792ee0f6 JA |
6844 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6845 | return -ERESTARTSYS; | |
af9c1a44 JA |
6846 | return -EINTR; |
6847 | } | |
6848 | ||
eeb60b9a PB |
6849 | /* when returns >0, the caller should retry */ |
6850 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6851 | struct io_wait_queue *iowq, | |
6852 | signed long *timeout) | |
6853 | { | |
6854 | int ret; | |
6855 | ||
6856 | /* make sure we run task_work before checking for signals */ | |
6857 | ret = io_run_task_work_sig(); | |
6858 | if (ret || io_should_wake(iowq)) | |
6859 | return ret; | |
6860 | /* let the caller flush overflows, retry */ | |
6861 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6862 | return 1; | |
6863 | ||
6864 | *timeout = schedule_timeout(*timeout); | |
6865 | return !*timeout ? -ETIME : 1; | |
6866 | } | |
6867 | ||
2b188cc1 JA |
6868 | /* |
6869 | * Wait until events become available, if we don't already have some. The | |
6870 | * application must reap them itself, as they reside on the shared cq ring. | |
6871 | */ | |
6872 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6873 | const sigset_t __user *sig, size_t sigsz, |
6874 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6875 | { |
bda52162 JA |
6876 | struct io_wait_queue iowq = { |
6877 | .wq = { | |
6878 | .private = current, | |
6879 | .func = io_wake_function, | |
6880 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6881 | }, | |
6882 | .ctx = ctx, | |
6883 | .to_wait = min_events, | |
6884 | }; | |
75b28aff | 6885 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6886 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6887 | int ret; | |
2b188cc1 | 6888 | |
b41e9852 | 6889 | do { |
6c503150 PB |
6890 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6891 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6892 | return 0; |
4c6e277c | 6893 | if (!io_run_task_work()) |
b41e9852 | 6894 | break; |
b41e9852 | 6895 | } while (1); |
2b188cc1 JA |
6896 | |
6897 | if (sig) { | |
9e75ad5d AB |
6898 | #ifdef CONFIG_COMPAT |
6899 | if (in_compat_syscall()) | |
6900 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6901 | sigsz); |
9e75ad5d AB |
6902 | else |
6903 | #endif | |
b772434b | 6904 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6905 | |
2b188cc1 JA |
6906 | if (ret) |
6907 | return ret; | |
6908 | } | |
6909 | ||
c73ebb68 | 6910 | if (uts) { |
c1d5a224 PB |
6911 | struct timespec64 ts; |
6912 | ||
c73ebb68 HX |
6913 | if (get_timespec64(&ts, uts)) |
6914 | return -EFAULT; | |
6915 | timeout = timespec64_to_jiffies(&ts); | |
6916 | } | |
6917 | ||
bda52162 | 6918 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6919 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6920 | do { |
ca0a2651 JA |
6921 | /* if we can't even flush overflow, don't wait for more */ |
6922 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6923 | ret = -EBUSY; | |
6924 | break; | |
6925 | } | |
bda52162 JA |
6926 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6927 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6928 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6929 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6930 | cond_resched(); |
eeb60b9a | 6931 | } while (ret > 0); |
bda52162 | 6932 | |
b7db41c9 | 6933 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6934 | |
75b28aff | 6935 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6936 | } |
6937 | ||
6b06314c JA |
6938 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6939 | { | |
6940 | #if defined(CONFIG_UNIX) | |
6941 | if (ctx->ring_sock) { | |
6942 | struct sock *sock = ctx->ring_sock->sk; | |
6943 | struct sk_buff *skb; | |
6944 | ||
6945 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6946 | kfree_skb(skb); | |
6947 | } | |
6948 | #else | |
6949 | int i; | |
6950 | ||
65e19f54 JA |
6951 | for (i = 0; i < ctx->nr_user_files; i++) { |
6952 | struct file *file; | |
6953 | ||
6954 | file = io_file_from_index(ctx, i); | |
6955 | if (file) | |
6956 | fput(file); | |
6957 | } | |
6b06314c JA |
6958 | #endif |
6959 | } | |
6960 | ||
00835dce | 6961 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6962 | { |
269bbe5f | 6963 | struct fixed_rsrc_data *data; |
05f3fb3c | 6964 | |
269bbe5f | 6965 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6966 | complete(&data->done); |
6967 | } | |
6968 | ||
2a63b2d9 | 6969 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6970 | { |
2a63b2d9 | 6971 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6972 | } |
6973 | ||
2a63b2d9 | 6974 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6975 | { |
2a63b2d9 BM |
6976 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6977 | } | |
65e19f54 | 6978 | |
d67d2263 BM |
6979 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6980 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6981 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6982 | { |
2a63b2d9 | 6983 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6984 | rsrc_data->node = ref_node; |
d67d2263 | 6985 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6986 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6987 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6988 | } |
6989 | ||
8bad28d8 | 6990 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6991 | { |
8bad28d8 | 6992 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6993 | |
2a63b2d9 | 6994 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6995 | ref_node = data->node; |
e6cb007c | 6996 | data->node = NULL; |
2a63b2d9 | 6997 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6998 | if (ref_node) |
6999 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
7000 | } |
7001 | ||
7002 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
7003 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
7004 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7005 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7006 | { |
f2303b1f | 7007 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 7008 | int ret; |
05589553 | 7009 | |
8bad28d8 HX |
7010 | if (data->quiesce) |
7011 | return -ENXIO; | |
05589553 | 7012 | |
8bad28d8 | 7013 | data->quiesce = true; |
1ffc5422 | 7014 | do { |
f2303b1f PB |
7015 | ret = -ENOMEM; |
7016 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7017 | if (!backup_node) | |
7018 | break; | |
7019 | backup_node->rsrc_data = data; | |
7020 | backup_node->rsrc_put = rsrc_put; | |
7021 | ||
8bad28d8 HX |
7022 | io_sqe_rsrc_kill_node(ctx, data); |
7023 | percpu_ref_kill(&data->refs); | |
7024 | flush_delayed_work(&ctx->rsrc_put_work); | |
7025 | ||
1ffc5422 PB |
7026 | ret = wait_for_completion_interruptible(&data->done); |
7027 | if (!ret) | |
7028 | break; | |
8bad28d8 | 7029 | |
cb5e1b81 | 7030 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7031 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7032 | backup_node = NULL; | |
cb5e1b81 | 7033 | reinit_completion(&data->done); |
8bad28d8 | 7034 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7035 | ret = io_run_task_work_sig(); |
8bad28d8 | 7036 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7037 | } while (ret >= 0); |
8bad28d8 | 7038 | data->quiesce = false; |
05f3fb3c | 7039 | |
8bad28d8 HX |
7040 | if (backup_node) |
7041 | destroy_fixed_rsrc_ref_node(backup_node); | |
7042 | return ret; | |
d7954b2b BM |
7043 | } |
7044 | ||
1ad555c6 BM |
7045 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7046 | { | |
7047 | struct fixed_rsrc_data *data; | |
7048 | ||
7049 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7050 | if (!data) | |
7051 | return NULL; | |
7052 | ||
00835dce | 7053 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7054 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7055 | kfree(data); | |
7056 | return NULL; | |
7057 | } | |
7058 | data->ctx = ctx; | |
7059 | init_completion(&data->done); | |
7060 | return data; | |
7061 | } | |
7062 | ||
7063 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7064 | { | |
7065 | percpu_ref_exit(&data->refs); | |
7066 | kfree(data->table); | |
7067 | kfree(data); | |
7068 | } | |
7069 | ||
d7954b2b BM |
7070 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7071 | { | |
7072 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7073 | unsigned nr_tables, i; |
7074 | int ret; | |
7075 | ||
8bad28d8 HX |
7076 | /* |
7077 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7078 | * Since we possibly drop uring lock later in this function to | |
7079 | * run task work. | |
7080 | */ | |
7081 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7082 | return -ENXIO; |
f2303b1f | 7083 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7084 | if (ret) |
7085 | return ret; | |
7086 | ||
6b06314c | 7087 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7088 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7089 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7090 | kfree(data->table[i].files); |
1ad555c6 | 7091 | free_fixed_rsrc_data(data); |
05f3fb3c | 7092 | ctx->file_data = NULL; |
6b06314c JA |
7093 | ctx->nr_user_files = 0; |
7094 | return 0; | |
7095 | } | |
7096 | ||
37d1e2e3 | 7097 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7098 | __releases(&sqd->lock) |
37d1e2e3 | 7099 | { |
521d6a73 PB |
7100 | WARN_ON_ONCE(sqd->thread == current); |
7101 | ||
9e138a48 PB |
7102 | /* |
7103 | * Do the dance but not conditional clear_bit() because it'd race with | |
7104 | * other threads incrementing park_pending and setting the bit. | |
7105 | */ | |
37d1e2e3 | 7106 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7107 | if (atomic_dec_return(&sqd->park_pending)) |
7108 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7109 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7110 | } |
7111 | ||
86e0d676 | 7112 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7113 | __acquires(&sqd->lock) |
37d1e2e3 | 7114 | { |
521d6a73 PB |
7115 | WARN_ON_ONCE(sqd->thread == current); |
7116 | ||
9e138a48 | 7117 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7118 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7119 | mutex_lock(&sqd->lock); |
05962f95 | 7120 | if (sqd->thread) |
86e0d676 | 7121 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7122 | } |
7123 | ||
7124 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7125 | { | |
521d6a73 PB |
7126 | WARN_ON_ONCE(sqd->thread == current); |
7127 | ||
09a6f4ef | 7128 | mutex_lock(&sqd->lock); |
05962f95 | 7129 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7130 | if (sqd->thread) |
7131 | wake_up_process(sqd->thread); | |
09a6f4ef | 7132 | mutex_unlock(&sqd->lock); |
05962f95 | 7133 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7134 | } |
7135 | ||
534ca6d6 | 7136 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7137 | { |
534ca6d6 | 7138 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7139 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7140 | ||
37d1e2e3 JA |
7141 | io_sq_thread_stop(sqd); |
7142 | kfree(sqd); | |
7143 | } | |
7144 | } | |
7145 | ||
7146 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7147 | { | |
7148 | struct io_sq_data *sqd = ctx->sq_data; | |
7149 | ||
7150 | if (sqd) { | |
05962f95 | 7151 | io_sq_thread_park(sqd); |
521d6a73 | 7152 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7153 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7154 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7155 | |
7156 | io_put_sq_data(sqd); | |
7157 | ctx->sq_data = NULL; | |
7c30f36a SM |
7158 | if (ctx->sq_creds) |
7159 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7160 | } |
7161 | } | |
7162 | ||
aa06165d JA |
7163 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7164 | { | |
7165 | struct io_ring_ctx *ctx_attach; | |
7166 | struct io_sq_data *sqd; | |
7167 | struct fd f; | |
7168 | ||
7169 | f = fdget(p->wq_fd); | |
7170 | if (!f.file) | |
7171 | return ERR_PTR(-ENXIO); | |
7172 | if (f.file->f_op != &io_uring_fops) { | |
7173 | fdput(f); | |
7174 | return ERR_PTR(-EINVAL); | |
7175 | } | |
7176 | ||
7177 | ctx_attach = f.file->private_data; | |
7178 | sqd = ctx_attach->sq_data; | |
7179 | if (!sqd) { | |
7180 | fdput(f); | |
7181 | return ERR_PTR(-EINVAL); | |
7182 | } | |
5c2469e0 JA |
7183 | if (sqd->task_tgid != current->tgid) { |
7184 | fdput(f); | |
7185 | return ERR_PTR(-EPERM); | |
7186 | } | |
aa06165d JA |
7187 | |
7188 | refcount_inc(&sqd->refs); | |
7189 | fdput(f); | |
7190 | return sqd; | |
7191 | } | |
7192 | ||
26984fbf PB |
7193 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7194 | bool *attached) | |
534ca6d6 JA |
7195 | { |
7196 | struct io_sq_data *sqd; | |
7197 | ||
26984fbf | 7198 | *attached = false; |
5c2469e0 JA |
7199 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7200 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7201 | if (!IS_ERR(sqd)) { |
7202 | *attached = true; | |
5c2469e0 | 7203 | return sqd; |
26984fbf | 7204 | } |
5c2469e0 JA |
7205 | /* fall through for EPERM case, setup new sqd/task */ |
7206 | if (PTR_ERR(sqd) != -EPERM) | |
7207 | return sqd; | |
7208 | } | |
aa06165d | 7209 | |
534ca6d6 JA |
7210 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7211 | if (!sqd) | |
7212 | return ERR_PTR(-ENOMEM); | |
7213 | ||
9e138a48 | 7214 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7215 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7216 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7217 | mutex_init(&sqd->lock); |
534ca6d6 | 7218 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7219 | init_completion(&sqd->exited); |
534ca6d6 JA |
7220 | return sqd; |
7221 | } | |
7222 | ||
6b06314c | 7223 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7224 | /* |
7225 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7226 | * the io_uring can be safely unregistered on process exit, even if we have | |
7227 | * loops in the file referencing. | |
7228 | */ | |
7229 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7230 | { | |
7231 | struct sock *sk = ctx->ring_sock->sk; | |
7232 | struct scm_fp_list *fpl; | |
7233 | struct sk_buff *skb; | |
08a45173 | 7234 | int i, nr_files; |
6b06314c | 7235 | |
6b06314c JA |
7236 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7237 | if (!fpl) | |
7238 | return -ENOMEM; | |
7239 | ||
7240 | skb = alloc_skb(0, GFP_KERNEL); | |
7241 | if (!skb) { | |
7242 | kfree(fpl); | |
7243 | return -ENOMEM; | |
7244 | } | |
7245 | ||
7246 | skb->sk = sk; | |
6b06314c | 7247 | |
08a45173 | 7248 | nr_files = 0; |
62e398be | 7249 | fpl->user = get_uid(current_user()); |
6b06314c | 7250 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7251 | struct file *file = io_file_from_index(ctx, i + offset); |
7252 | ||
7253 | if (!file) | |
08a45173 | 7254 | continue; |
65e19f54 | 7255 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7256 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7257 | nr_files++; | |
6b06314c JA |
7258 | } |
7259 | ||
08a45173 JA |
7260 | if (nr_files) { |
7261 | fpl->max = SCM_MAX_FD; | |
7262 | fpl->count = nr_files; | |
7263 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7264 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7265 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7266 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7267 | |
08a45173 JA |
7268 | for (i = 0; i < nr_files; i++) |
7269 | fput(fpl->fp[i]); | |
7270 | } else { | |
7271 | kfree_skb(skb); | |
7272 | kfree(fpl); | |
7273 | } | |
6b06314c JA |
7274 | |
7275 | return 0; | |
7276 | } | |
7277 | ||
7278 | /* | |
7279 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7280 | * causes regular reference counting to break down. We rely on the UNIX | |
7281 | * garbage collection to take care of this problem for us. | |
7282 | */ | |
7283 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7284 | { | |
7285 | unsigned left, total; | |
7286 | int ret = 0; | |
7287 | ||
7288 | total = 0; | |
7289 | left = ctx->nr_user_files; | |
7290 | while (left) { | |
7291 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7292 | |
7293 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7294 | if (ret) | |
7295 | break; | |
7296 | left -= this_files; | |
7297 | total += this_files; | |
7298 | } | |
7299 | ||
7300 | if (!ret) | |
7301 | return 0; | |
7302 | ||
7303 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7304 | struct file *file = io_file_from_index(ctx, total); |
7305 | ||
7306 | if (file) | |
7307 | fput(file); | |
6b06314c JA |
7308 | total++; |
7309 | } | |
7310 | ||
7311 | return ret; | |
7312 | } | |
7313 | #else | |
7314 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7315 | { | |
7316 | return 0; | |
7317 | } | |
7318 | #endif | |
7319 | ||
269bbe5f | 7320 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7321 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7322 | { |
7323 | int i; | |
7324 | ||
7325 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7326 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7327 | unsigned this_files; |
7328 | ||
7329 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7330 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7331 | GFP_KERNEL); | |
7332 | if (!table->files) | |
7333 | break; | |
7334 | nr_files -= this_files; | |
7335 | } | |
7336 | ||
7337 | if (i == nr_tables) | |
7338 | return 0; | |
7339 | ||
7340 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7341 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7342 | kfree(table->files); |
7343 | } | |
7344 | return 1; | |
7345 | } | |
7346 | ||
50238531 | 7347 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7348 | { |
50238531 | 7349 | struct file *file = prsrc->file; |
05f3fb3c JA |
7350 | #if defined(CONFIG_UNIX) |
7351 | struct sock *sock = ctx->ring_sock->sk; | |
7352 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7353 | struct sk_buff *skb; | |
7354 | int i; | |
7355 | ||
7356 | __skb_queue_head_init(&list); | |
7357 | ||
7358 | /* | |
7359 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7360 | * remove this entry and rearrange the file array. | |
7361 | */ | |
7362 | skb = skb_dequeue(head); | |
7363 | while (skb) { | |
7364 | struct scm_fp_list *fp; | |
7365 | ||
7366 | fp = UNIXCB(skb).fp; | |
7367 | for (i = 0; i < fp->count; i++) { | |
7368 | int left; | |
7369 | ||
7370 | if (fp->fp[i] != file) | |
7371 | continue; | |
7372 | ||
7373 | unix_notinflight(fp->user, fp->fp[i]); | |
7374 | left = fp->count - 1 - i; | |
7375 | if (left) { | |
7376 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7377 | left * sizeof(struct file *)); | |
7378 | } | |
7379 | fp->count--; | |
7380 | if (!fp->count) { | |
7381 | kfree_skb(skb); | |
7382 | skb = NULL; | |
7383 | } else { | |
7384 | __skb_queue_tail(&list, skb); | |
7385 | } | |
7386 | fput(file); | |
7387 | file = NULL; | |
7388 | break; | |
7389 | } | |
7390 | ||
7391 | if (!file) | |
7392 | break; | |
7393 | ||
7394 | __skb_queue_tail(&list, skb); | |
7395 | ||
7396 | skb = skb_dequeue(head); | |
7397 | } | |
7398 | ||
7399 | if (skb_peek(&list)) { | |
7400 | spin_lock_irq(&head->lock); | |
7401 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7402 | __skb_queue_tail(head, skb); | |
7403 | spin_unlock_irq(&head->lock); | |
7404 | } | |
7405 | #else | |
7406 | fput(file); | |
7407 | #endif | |
7408 | } | |
7409 | ||
269bbe5f | 7410 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7411 | { |
269bbe5f BM |
7412 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7413 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7414 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7415 | |
269bbe5f BM |
7416 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7417 | list_del(&prsrc->list); | |
50238531 | 7418 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7419 | kfree(prsrc); |
65e19f54 | 7420 | } |
05589553 | 7421 | |
05589553 XW |
7422 | percpu_ref_exit(&ref_node->refs); |
7423 | kfree(ref_node); | |
269bbe5f | 7424 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7425 | } |
65e19f54 | 7426 | |
269bbe5f | 7427 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7428 | { |
7429 | struct io_ring_ctx *ctx; | |
7430 | struct llist_node *node; | |
7431 | ||
269bbe5f BM |
7432 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7433 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7434 | |
7435 | while (node) { | |
269bbe5f | 7436 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7437 | struct llist_node *next = node->next; |
7438 | ||
269bbe5f BM |
7439 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7440 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7441 | node = next; |
7442 | } | |
7443 | } | |
7444 | ||
ea64ec02 PB |
7445 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7446 | unsigned i) | |
2faf852d | 7447 | { |
ea64ec02 PB |
7448 | struct fixed_rsrc_table *table; |
7449 | ||
7450 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7451 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7452 | } | |
7453 | ||
00835dce | 7454 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7455 | { |
269bbe5f BM |
7456 | struct fixed_rsrc_ref_node *ref_node; |
7457 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7458 | struct io_ring_ctx *ctx; |
e297822b | 7459 | bool first_add = false; |
4a38aed2 | 7460 | int delay = HZ; |
65e19f54 | 7461 | |
269bbe5f BM |
7462 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7463 | data = ref_node->rsrc_data; | |
e297822b PB |
7464 | ctx = data->ctx; |
7465 | ||
2a63b2d9 | 7466 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7467 | ref_node->done = true; |
7468 | ||
d67d2263 BM |
7469 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7470 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7471 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7472 | /* recycle ref nodes in order */ |
7473 | if (!ref_node->done) | |
7474 | break; | |
7475 | list_del(&ref_node->node); | |
269bbe5f | 7476 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7477 | } |
2a63b2d9 | 7478 | io_rsrc_ref_unlock(ctx); |
05589553 | 7479 | |
e297822b | 7480 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7481 | delay = 0; |
05589553 | 7482 | |
4a38aed2 | 7483 | if (!delay) |
269bbe5f | 7484 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7485 | else if (first_add) |
269bbe5f | 7486 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7487 | } |
65e19f54 | 7488 | |
6802535d | 7489 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7490 | struct io_ring_ctx *ctx) |
05f3fb3c | 7491 | { |
269bbe5f | 7492 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7493 | |
05589553 XW |
7494 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7495 | if (!ref_node) | |
3e2224c5 | 7496 | return NULL; |
05f3fb3c | 7497 | |
00835dce | 7498 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7499 | 0, GFP_KERNEL)) { |
7500 | kfree(ref_node); | |
3e2224c5 | 7501 | return NULL; |
05589553 XW |
7502 | } |
7503 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7504 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7505 | ref_node->done = false; |
05589553 | 7506 | return ref_node; |
05589553 XW |
7507 | } |
7508 | ||
bc9744cd PB |
7509 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7510 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7511 | { |
269bbe5f | 7512 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7513 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7514 | } |
7515 | ||
269bbe5f | 7516 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7517 | { |
7518 | percpu_ref_exit(&ref_node->refs); | |
7519 | kfree(ref_node); | |
65e19f54 JA |
7520 | } |
7521 | ||
ea64ec02 | 7522 | |
6b06314c JA |
7523 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7524 | unsigned nr_args) | |
7525 | { | |
7526 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7527 | unsigned nr_tables, i; |
05f3fb3c | 7528 | struct file *file; |
600cf3f8 | 7529 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7530 | struct fixed_rsrc_ref_node *ref_node; |
7531 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7532 | |
05f3fb3c | 7533 | if (ctx->file_data) |
6b06314c JA |
7534 | return -EBUSY; |
7535 | if (!nr_args) | |
7536 | return -EINVAL; | |
7537 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7538 | return -EMFILE; | |
7539 | ||
1ad555c6 | 7540 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7541 | if (!file_data) |
05f3fb3c | 7542 | return -ENOMEM; |
13770a71 | 7543 | ctx->file_data = file_data; |
05f3fb3c | 7544 | |
65e19f54 | 7545 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7546 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7547 | GFP_KERNEL); |
600cf3f8 PB |
7548 | if (!file_data->table) |
7549 | goto out_free; | |
05f3fb3c | 7550 | |
600cf3f8 | 7551 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7552 | goto out_free; |
65e19f54 | 7553 | |
08a45173 | 7554 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7555 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7556 | ret = -EFAULT; | |
7557 | goto out_fput; | |
7558 | } | |
08a45173 | 7559 | /* allow sparse sets */ |
600cf3f8 | 7560 | if (fd == -1) |
08a45173 | 7561 | continue; |
6b06314c | 7562 | |
05f3fb3c | 7563 | file = fget(fd); |
6b06314c | 7564 | ret = -EBADF; |
05f3fb3c | 7565 | if (!file) |
600cf3f8 | 7566 | goto out_fput; |
05f3fb3c | 7567 | |
6b06314c JA |
7568 | /* |
7569 | * Don't allow io_uring instances to be registered. If UNIX | |
7570 | * isn't enabled, then this causes a reference cycle and this | |
7571 | * instance can never get freed. If UNIX is enabled we'll | |
7572 | * handle it just fine, but there's still no point in allowing | |
7573 | * a ring fd as it doesn't support regular read/write anyway. | |
7574 | */ | |
05f3fb3c JA |
7575 | if (file->f_op == &io_uring_fops) { |
7576 | fput(file); | |
600cf3f8 | 7577 | goto out_fput; |
6b06314c | 7578 | } |
ea64ec02 | 7579 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7580 | } |
7581 | ||
6b06314c | 7582 | ret = io_sqe_files_scm(ctx); |
05589553 | 7583 | if (ret) { |
6b06314c | 7584 | io_sqe_files_unregister(ctx); |
05589553 XW |
7585 | return ret; |
7586 | } | |
6b06314c | 7587 | |
bc9744cd | 7588 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7589 | if (!ref_node) { |
05589553 | 7590 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7591 | return -ENOMEM; |
05589553 | 7592 | } |
bc9744cd | 7593 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7594 | |
d67d2263 | 7595 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7596 | return ret; |
600cf3f8 PB |
7597 | out_fput: |
7598 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7599 | file = io_file_from_index(ctx, i); | |
7600 | if (file) | |
7601 | fput(file); | |
7602 | } | |
7603 | for (i = 0; i < nr_tables; i++) | |
7604 | kfree(file_data->table[i].files); | |
7605 | ctx->nr_user_files = 0; | |
600cf3f8 | 7606 | out_free: |
1ad555c6 | 7607 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7608 | ctx->file_data = NULL; |
6b06314c JA |
7609 | return ret; |
7610 | } | |
7611 | ||
c3a31e60 JA |
7612 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7613 | int index) | |
7614 | { | |
7615 | #if defined(CONFIG_UNIX) | |
7616 | struct sock *sock = ctx->ring_sock->sk; | |
7617 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7618 | struct sk_buff *skb; | |
7619 | ||
7620 | /* | |
7621 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7622 | * file set. If there's no room, fall back to allocating a new skb | |
7623 | * and filling it in. | |
7624 | */ | |
7625 | spin_lock_irq(&head->lock); | |
7626 | skb = skb_peek(head); | |
7627 | if (skb) { | |
7628 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7629 | ||
7630 | if (fpl->count < SCM_MAX_FD) { | |
7631 | __skb_unlink(skb, head); | |
7632 | spin_unlock_irq(&head->lock); | |
7633 | fpl->fp[fpl->count] = get_file(file); | |
7634 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7635 | fpl->count++; | |
7636 | spin_lock_irq(&head->lock); | |
7637 | __skb_queue_head(head, skb); | |
7638 | } else { | |
7639 | skb = NULL; | |
7640 | } | |
7641 | } | |
7642 | spin_unlock_irq(&head->lock); | |
7643 | ||
7644 | if (skb) { | |
7645 | fput(file); | |
7646 | return 0; | |
7647 | } | |
7648 | ||
7649 | return __io_sqe_files_scm(ctx, 1, index); | |
7650 | #else | |
7651 | return 0; | |
7652 | #endif | |
7653 | } | |
7654 | ||
50238531 | 7655 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7656 | { |
269bbe5f BM |
7657 | struct io_rsrc_put *prsrc; |
7658 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7659 | |
269bbe5f BM |
7660 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7661 | if (!prsrc) | |
a5318d3c | 7662 | return -ENOMEM; |
05f3fb3c | 7663 | |
50238531 | 7664 | prsrc->rsrc = rsrc; |
269bbe5f | 7665 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7666 | |
a5318d3c | 7667 | return 0; |
05f3fb3c JA |
7668 | } |
7669 | ||
269bbe5f BM |
7670 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7671 | struct file *file) | |
7672 | { | |
50238531 | 7673 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7674 | } |
7675 | ||
05f3fb3c | 7676 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7677 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7678 | unsigned nr_args) |
7679 | { | |
269bbe5f BM |
7680 | struct fixed_rsrc_data *data = ctx->file_data; |
7681 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7682 | struct file *file, **file_slot; |
c3a31e60 JA |
7683 | __s32 __user *fds; |
7684 | int fd, i, err; | |
7685 | __u32 done; | |
05589553 | 7686 | bool needs_switch = false; |
c3a31e60 | 7687 | |
05f3fb3c | 7688 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7689 | return -EOVERFLOW; |
7690 | if (done > ctx->nr_user_files) | |
7691 | return -EINVAL; | |
7692 | ||
bc9744cd | 7693 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7694 | if (!ref_node) |
7695 | return -ENOMEM; | |
bc9744cd | 7696 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7697 | |
269bbe5f | 7698 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7699 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7700 | err = 0; |
7701 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7702 | err = -EFAULT; | |
7703 | break; | |
7704 | } | |
4e0377a1 | 7705 | if (fd == IORING_REGISTER_FILES_SKIP) |
7706 | continue; | |
7707 | ||
67973b93 | 7708 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7709 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7710 | ||
7711 | if (*file_slot) { | |
7712 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7713 | if (err) |
7714 | break; | |
ea64ec02 | 7715 | *file_slot = NULL; |
05589553 | 7716 | needs_switch = true; |
c3a31e60 JA |
7717 | } |
7718 | if (fd != -1) { | |
c3a31e60 JA |
7719 | file = fget(fd); |
7720 | if (!file) { | |
7721 | err = -EBADF; | |
7722 | break; | |
7723 | } | |
7724 | /* | |
7725 | * Don't allow io_uring instances to be registered. If | |
7726 | * UNIX isn't enabled, then this causes a reference | |
7727 | * cycle and this instance can never get freed. If UNIX | |
7728 | * is enabled we'll handle it just fine, but there's | |
7729 | * still no point in allowing a ring fd as it doesn't | |
7730 | * support regular read/write anyway. | |
7731 | */ | |
7732 | if (file->f_op == &io_uring_fops) { | |
7733 | fput(file); | |
7734 | err = -EBADF; | |
7735 | break; | |
7736 | } | |
e68a3ff8 | 7737 | *file_slot = file; |
c3a31e60 | 7738 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7739 | if (err) { |
e68a3ff8 | 7740 | *file_slot = NULL; |
f3bd9dae | 7741 | fput(file); |
c3a31e60 | 7742 | break; |
f3bd9dae | 7743 | } |
c3a31e60 | 7744 | } |
05f3fb3c JA |
7745 | } |
7746 | ||
05589553 | 7747 | if (needs_switch) { |
b2e96852 | 7748 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7749 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7750 | } else |
269bbe5f | 7751 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7752 | |
7753 | return done ? done : err; | |
7754 | } | |
05589553 | 7755 | |
05f3fb3c JA |
7756 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7757 | unsigned nr_args) | |
7758 | { | |
269bbe5f | 7759 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7760 | |
7761 | if (!ctx->file_data) | |
7762 | return -ENXIO; | |
7763 | if (!nr_args) | |
7764 | return -EINVAL; | |
7765 | if (copy_from_user(&up, arg, sizeof(up))) | |
7766 | return -EFAULT; | |
7767 | if (up.resv) | |
7768 | return -EINVAL; | |
7769 | ||
7770 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7771 | } | |
c3a31e60 | 7772 | |
5280f7e5 | 7773 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7774 | { |
7775 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7776 | ||
5280f7e5 PB |
7777 | req = io_put_req_find_next(req); |
7778 | return req ? &req->work : NULL; | |
7d723065 JA |
7779 | } |
7780 | ||
5aa75ed5 | 7781 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7782 | { |
e941894e | 7783 | struct io_wq_hash *hash; |
24369c2e | 7784 | struct io_wq_data data; |
24369c2e | 7785 | unsigned int concurrency; |
24369c2e | 7786 | |
e941894e JA |
7787 | hash = ctx->hash_map; |
7788 | if (!hash) { | |
7789 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7790 | if (!hash) | |
7791 | return ERR_PTR(-ENOMEM); | |
7792 | refcount_set(&hash->refs, 1); | |
7793 | init_waitqueue_head(&hash->wait); | |
7794 | ctx->hash_map = hash; | |
24369c2e PB |
7795 | } |
7796 | ||
e941894e | 7797 | data.hash = hash; |
e9fd9396 | 7798 | data.free_work = io_free_work; |
f5fa38c5 | 7799 | data.do_work = io_wq_submit_work; |
24369c2e | 7800 | |
d25e3a3d JA |
7801 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7802 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7803 | |
5aa75ed5 | 7804 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7805 | } |
7806 | ||
5aa75ed5 JA |
7807 | static int io_uring_alloc_task_context(struct task_struct *task, |
7808 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7809 | { |
7810 | struct io_uring_task *tctx; | |
d8a6df10 | 7811 | int ret; |
0f212204 JA |
7812 | |
7813 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7814 | if (unlikely(!tctx)) | |
7815 | return -ENOMEM; | |
7816 | ||
d8a6df10 JA |
7817 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7818 | if (unlikely(ret)) { | |
7819 | kfree(tctx); | |
7820 | return ret; | |
7821 | } | |
7822 | ||
5aa75ed5 JA |
7823 | tctx->io_wq = io_init_wq_offload(ctx); |
7824 | if (IS_ERR(tctx->io_wq)) { | |
7825 | ret = PTR_ERR(tctx->io_wq); | |
7826 | percpu_counter_destroy(&tctx->inflight); | |
7827 | kfree(tctx); | |
7828 | return ret; | |
7829 | } | |
7830 | ||
0f212204 JA |
7831 | xa_init(&tctx->xa); |
7832 | init_waitqueue_head(&tctx->wait); | |
7833 | tctx->last = NULL; | |
fdaf083c | 7834 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7835 | task->io_uring = tctx; |
7cbf1722 JA |
7836 | spin_lock_init(&tctx->task_lock); |
7837 | INIT_WQ_LIST(&tctx->task_list); | |
7838 | tctx->task_state = 0; | |
7839 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7840 | return 0; |
7841 | } | |
7842 | ||
7843 | void __io_uring_free(struct task_struct *tsk) | |
7844 | { | |
7845 | struct io_uring_task *tctx = tsk->io_uring; | |
7846 | ||
7847 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7848 | WARN_ON_ONCE(tctx->io_wq); |
7849 | ||
d8a6df10 | 7850 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7851 | kfree(tctx); |
7852 | tsk->io_uring = NULL; | |
7853 | } | |
7854 | ||
7e84e1c7 SG |
7855 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7856 | struct io_uring_params *p) | |
2b188cc1 JA |
7857 | { |
7858 | int ret; | |
7859 | ||
d25e3a3d JA |
7860 | /* Retain compatibility with failing for an invalid attach attempt */ |
7861 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7862 | IORING_SETUP_ATTACH_WQ) { | |
7863 | struct fd f; | |
7864 | ||
7865 | f = fdget(p->wq_fd); | |
7866 | if (!f.file) | |
7867 | return -ENXIO; | |
7868 | if (f.file->f_op != &io_uring_fops) { | |
7869 | fdput(f); | |
7870 | return -EINVAL; | |
7871 | } | |
7872 | fdput(f); | |
7873 | } | |
6c271ce2 | 7874 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7875 | struct task_struct *tsk; |
534ca6d6 | 7876 | struct io_sq_data *sqd; |
26984fbf | 7877 | bool attached; |
534ca6d6 | 7878 | |
3ec482d1 | 7879 | ret = -EPERM; |
ce59fc69 | 7880 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7881 | goto err; |
7882 | ||
26984fbf | 7883 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7884 | if (IS_ERR(sqd)) { |
7885 | ret = PTR_ERR(sqd); | |
7886 | goto err; | |
7887 | } | |
69fb2131 | 7888 | |
7c30f36a | 7889 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7890 | ctx->sq_data = sqd; |
917257da JA |
7891 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7892 | if (!ctx->sq_thread_idle) | |
7893 | ctx->sq_thread_idle = HZ; | |
7894 | ||
26984fbf | 7895 | ret = 0; |
78d7f6ba | 7896 | io_sq_thread_park(sqd); |
26984fbf PB |
7897 | /* don't attach to a dying SQPOLL thread, would be racy */ |
7898 | if (attached && !sqd->thread) { | |
7899 | ret = -ENXIO; | |
7900 | } else { | |
7901 | list_add(&ctx->sqd_list, &sqd->ctx_list); | |
7902 | io_sqd_update_thread_idle(sqd); | |
7903 | } | |
78d7f6ba PB |
7904 | io_sq_thread_unpark(sqd); |
7905 | ||
26984fbf PB |
7906 | if (ret < 0) { |
7907 | io_put_sq_data(sqd); | |
7908 | ctx->sq_data = NULL; | |
7909 | return ret; | |
7910 | } else if (attached) { | |
5aa75ed5 | 7911 | return 0; |
26984fbf | 7912 | } |
aa06165d | 7913 | |
6c271ce2 | 7914 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7915 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7916 | |
917257da | 7917 | ret = -EINVAL; |
44a9bd18 | 7918 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7919 | goto err_sqpoll; |
7889f44d | 7920 | if (!cpu_online(cpu)) |
e8f98f24 | 7921 | goto err_sqpoll; |
917257da | 7922 | |
37d1e2e3 | 7923 | sqd->sq_cpu = cpu; |
6c271ce2 | 7924 | } else { |
37d1e2e3 | 7925 | sqd->sq_cpu = -1; |
6c271ce2 | 7926 | } |
37d1e2e3 JA |
7927 | |
7928 | sqd->task_pid = current->pid; | |
5c2469e0 | 7929 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7930 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7931 | if (IS_ERR(tsk)) { | |
7932 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7933 | goto err_sqpoll; |
6c271ce2 | 7934 | } |
97a73a0f | 7935 | |
46fe18b1 | 7936 | sqd->thread = tsk; |
97a73a0f | 7937 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7938 | wake_up_new_task(tsk); |
0f212204 JA |
7939 | if (ret) |
7940 | goto err; | |
6c271ce2 JA |
7941 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7942 | /* Can't have SQ_AFF without SQPOLL */ | |
7943 | ret = -EINVAL; | |
7944 | goto err; | |
7945 | } | |
7946 | ||
2b188cc1 JA |
7947 | return 0; |
7948 | err: | |
37d1e2e3 | 7949 | io_sq_thread_finish(ctx); |
2b188cc1 | 7950 | return ret; |
e8f98f24 JA |
7951 | err_sqpoll: |
7952 | complete(&ctx->sq_data->exited); | |
7953 | goto err; | |
2b188cc1 JA |
7954 | } |
7955 | ||
a087e2b5 BM |
7956 | static inline void __io_unaccount_mem(struct user_struct *user, |
7957 | unsigned long nr_pages) | |
2b188cc1 JA |
7958 | { |
7959 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7960 | } | |
7961 | ||
a087e2b5 BM |
7962 | static inline int __io_account_mem(struct user_struct *user, |
7963 | unsigned long nr_pages) | |
2b188cc1 JA |
7964 | { |
7965 | unsigned long page_limit, cur_pages, new_pages; | |
7966 | ||
7967 | /* Don't allow more pages than we can safely lock */ | |
7968 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7969 | ||
7970 | do { | |
7971 | cur_pages = atomic_long_read(&user->locked_vm); | |
7972 | new_pages = cur_pages + nr_pages; | |
7973 | if (new_pages > page_limit) | |
7974 | return -ENOMEM; | |
7975 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7976 | new_pages) != cur_pages); | |
7977 | ||
7978 | return 0; | |
7979 | } | |
7980 | ||
26bfa89e | 7981 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7982 | { |
62e398be | 7983 | if (ctx->user) |
a087e2b5 | 7984 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7985 | |
26bfa89e JA |
7986 | if (ctx->mm_account) |
7987 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7988 | } |
7989 | ||
26bfa89e | 7990 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7991 | { |
30975825 BM |
7992 | int ret; |
7993 | ||
62e398be | 7994 | if (ctx->user) { |
30975825 BM |
7995 | ret = __io_account_mem(ctx->user, nr_pages); |
7996 | if (ret) | |
7997 | return ret; | |
7998 | } | |
7999 | ||
26bfa89e JA |
8000 | if (ctx->mm_account) |
8001 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8002 | |
8003 | return 0; | |
8004 | } | |
8005 | ||
2b188cc1 JA |
8006 | static void io_mem_free(void *ptr) |
8007 | { | |
52e04ef4 MR |
8008 | struct page *page; |
8009 | ||
8010 | if (!ptr) | |
8011 | return; | |
2b188cc1 | 8012 | |
52e04ef4 | 8013 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8014 | if (put_page_testzero(page)) |
8015 | free_compound_page(page); | |
8016 | } | |
8017 | ||
8018 | static void *io_mem_alloc(size_t size) | |
8019 | { | |
8020 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8021 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8022 | |
8023 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8024 | } | |
8025 | ||
75b28aff HV |
8026 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8027 | size_t *sq_offset) | |
8028 | { | |
8029 | struct io_rings *rings; | |
8030 | size_t off, sq_array_size; | |
8031 | ||
8032 | off = struct_size(rings, cqes, cq_entries); | |
8033 | if (off == SIZE_MAX) | |
8034 | return SIZE_MAX; | |
8035 | ||
8036 | #ifdef CONFIG_SMP | |
8037 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8038 | if (off == 0) | |
8039 | return SIZE_MAX; | |
8040 | #endif | |
8041 | ||
b36200f5 DV |
8042 | if (sq_offset) |
8043 | *sq_offset = off; | |
8044 | ||
75b28aff HV |
8045 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8046 | if (sq_array_size == SIZE_MAX) | |
8047 | return SIZE_MAX; | |
8048 | ||
8049 | if (check_add_overflow(off, sq_array_size, &off)) | |
8050 | return SIZE_MAX; | |
8051 | ||
75b28aff HV |
8052 | return off; |
8053 | } | |
8054 | ||
0a96bbe4 | 8055 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8056 | { |
8057 | int i, j; | |
8058 | ||
8059 | if (!ctx->user_bufs) | |
8060 | return -ENXIO; | |
8061 | ||
8062 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8063 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8064 | ||
8065 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8066 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8067 | |
de293938 | 8068 | if (imu->acct_pages) |
26bfa89e | 8069 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8070 | kvfree(imu->bvec); |
edafccee JA |
8071 | imu->nr_bvecs = 0; |
8072 | } | |
8073 | ||
8074 | kfree(ctx->user_bufs); | |
8075 | ctx->user_bufs = NULL; | |
8076 | ctx->nr_user_bufs = 0; | |
8077 | return 0; | |
8078 | } | |
8079 | ||
8080 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8081 | void __user *arg, unsigned index) | |
8082 | { | |
8083 | struct iovec __user *src; | |
8084 | ||
8085 | #ifdef CONFIG_COMPAT | |
8086 | if (ctx->compat) { | |
8087 | struct compat_iovec __user *ciovs; | |
8088 | struct compat_iovec ciov; | |
8089 | ||
8090 | ciovs = (struct compat_iovec __user *) arg; | |
8091 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8092 | return -EFAULT; | |
8093 | ||
d55e5f5b | 8094 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8095 | dst->iov_len = ciov.iov_len; |
8096 | return 0; | |
8097 | } | |
8098 | #endif | |
8099 | src = (struct iovec __user *) arg; | |
8100 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8101 | return -EFAULT; | |
8102 | return 0; | |
8103 | } | |
8104 | ||
de293938 JA |
8105 | /* |
8106 | * Not super efficient, but this is just a registration time. And we do cache | |
8107 | * the last compound head, so generally we'll only do a full search if we don't | |
8108 | * match that one. | |
8109 | * | |
8110 | * We check if the given compound head page has already been accounted, to | |
8111 | * avoid double accounting it. This allows us to account the full size of the | |
8112 | * page, not just the constituent pages of a huge page. | |
8113 | */ | |
8114 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8115 | int nr_pages, struct page *hpage) | |
8116 | { | |
8117 | int i, j; | |
8118 | ||
8119 | /* check current page array */ | |
8120 | for (i = 0; i < nr_pages; i++) { | |
8121 | if (!PageCompound(pages[i])) | |
8122 | continue; | |
8123 | if (compound_head(pages[i]) == hpage) | |
8124 | return true; | |
8125 | } | |
8126 | ||
8127 | /* check previously registered pages */ | |
8128 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8129 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8130 | ||
8131 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8132 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8133 | continue; | |
8134 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8135 | return true; | |
8136 | } | |
8137 | } | |
8138 | ||
8139 | return false; | |
8140 | } | |
8141 | ||
8142 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8143 | int nr_pages, struct io_mapped_ubuf *imu, | |
8144 | struct page **last_hpage) | |
8145 | { | |
8146 | int i, ret; | |
8147 | ||
8148 | for (i = 0; i < nr_pages; i++) { | |
8149 | if (!PageCompound(pages[i])) { | |
8150 | imu->acct_pages++; | |
8151 | } else { | |
8152 | struct page *hpage; | |
8153 | ||
8154 | hpage = compound_head(pages[i]); | |
8155 | if (hpage == *last_hpage) | |
8156 | continue; | |
8157 | *last_hpage = hpage; | |
8158 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8159 | continue; | |
8160 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8161 | } | |
8162 | } | |
8163 | ||
8164 | if (!imu->acct_pages) | |
8165 | return 0; | |
8166 | ||
26bfa89e | 8167 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8168 | if (ret) |
8169 | imu->acct_pages = 0; | |
8170 | return ret; | |
8171 | } | |
8172 | ||
0a96bbe4 BM |
8173 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8174 | struct io_mapped_ubuf *imu, | |
8175 | struct page **last_hpage) | |
edafccee JA |
8176 | { |
8177 | struct vm_area_struct **vmas = NULL; | |
8178 | struct page **pages = NULL; | |
0a96bbe4 BM |
8179 | unsigned long off, start, end, ubuf; |
8180 | size_t size; | |
8181 | int ret, pret, nr_pages, i; | |
8182 | ||
8183 | ubuf = (unsigned long) iov->iov_base; | |
8184 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8185 | start = ubuf >> PAGE_SHIFT; | |
8186 | nr_pages = end - start; | |
8187 | ||
8188 | ret = -ENOMEM; | |
8189 | ||
8190 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8191 | if (!pages) | |
8192 | goto done; | |
8193 | ||
8194 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8195 | GFP_KERNEL); | |
8196 | if (!vmas) | |
8197 | goto done; | |
edafccee | 8198 | |
0a96bbe4 BM |
8199 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8200 | GFP_KERNEL); | |
8201 | if (!imu->bvec) | |
8202 | goto done; | |
8203 | ||
8204 | ret = 0; | |
8205 | mmap_read_lock(current->mm); | |
8206 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8207 | pages, vmas); | |
8208 | if (pret == nr_pages) { | |
8209 | /* don't support file backed memory */ | |
8210 | for (i = 0; i < nr_pages; i++) { | |
8211 | struct vm_area_struct *vma = vmas[i]; | |
8212 | ||
8213 | if (vma->vm_file && | |
8214 | !is_file_hugepages(vma->vm_file)) { | |
8215 | ret = -EOPNOTSUPP; | |
8216 | break; | |
8217 | } | |
8218 | } | |
8219 | } else { | |
8220 | ret = pret < 0 ? pret : -EFAULT; | |
8221 | } | |
8222 | mmap_read_unlock(current->mm); | |
8223 | if (ret) { | |
8224 | /* | |
8225 | * if we did partial map, or found file backed vmas, | |
8226 | * release any pages we did get | |
8227 | */ | |
8228 | if (pret > 0) | |
8229 | unpin_user_pages(pages, pret); | |
8230 | kvfree(imu->bvec); | |
8231 | goto done; | |
8232 | } | |
8233 | ||
8234 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8235 | if (ret) { | |
8236 | unpin_user_pages(pages, pret); | |
8237 | kvfree(imu->bvec); | |
8238 | goto done; | |
8239 | } | |
8240 | ||
8241 | off = ubuf & ~PAGE_MASK; | |
8242 | size = iov->iov_len; | |
8243 | for (i = 0; i < nr_pages; i++) { | |
8244 | size_t vec_len; | |
8245 | ||
8246 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8247 | imu->bvec[i].bv_page = pages[i]; | |
8248 | imu->bvec[i].bv_len = vec_len; | |
8249 | imu->bvec[i].bv_offset = off; | |
8250 | off = 0; | |
8251 | size -= vec_len; | |
8252 | } | |
8253 | /* store original address for later verification */ | |
8254 | imu->ubuf = ubuf; | |
8255 | imu->len = iov->iov_len; | |
8256 | imu->nr_bvecs = nr_pages; | |
8257 | ret = 0; | |
8258 | done: | |
8259 | kvfree(pages); | |
8260 | kvfree(vmas); | |
8261 | return ret; | |
8262 | } | |
8263 | ||
2b358604 | 8264 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8265 | { |
edafccee JA |
8266 | if (ctx->user_bufs) |
8267 | return -EBUSY; | |
8268 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8269 | return -EINVAL; | |
8270 | ||
8271 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8272 | GFP_KERNEL); | |
8273 | if (!ctx->user_bufs) | |
8274 | return -ENOMEM; | |
8275 | ||
2b358604 BM |
8276 | return 0; |
8277 | } | |
edafccee | 8278 | |
2b358604 BM |
8279 | static int io_buffer_validate(struct iovec *iov) |
8280 | { | |
8281 | /* | |
8282 | * Don't impose further limits on the size and buffer | |
8283 | * constraints here, we'll -EINVAL later when IO is | |
8284 | * submitted if they are wrong. | |
8285 | */ | |
8286 | if (!iov->iov_base || !iov->iov_len) | |
8287 | return -EFAULT; | |
edafccee | 8288 | |
2b358604 BM |
8289 | /* arbitrary limit, but we need something */ |
8290 | if (iov->iov_len > SZ_1G) | |
8291 | return -EFAULT; | |
edafccee | 8292 | |
2b358604 BM |
8293 | return 0; |
8294 | } | |
edafccee | 8295 | |
2b358604 BM |
8296 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8297 | unsigned int nr_args) | |
8298 | { | |
8299 | int i, ret; | |
8300 | struct iovec iov; | |
8301 | struct page *last_hpage = NULL; | |
edafccee | 8302 | |
2b358604 BM |
8303 | ret = io_buffers_map_alloc(ctx, nr_args); |
8304 | if (ret) | |
8305 | return ret; | |
edafccee | 8306 | |
edafccee JA |
8307 | for (i = 0; i < nr_args; i++) { |
8308 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8309 | |
edafccee JA |
8310 | ret = io_copy_iov(ctx, &iov, arg, i); |
8311 | if (ret) | |
0a96bbe4 | 8312 | break; |
de293938 | 8313 | |
2b358604 BM |
8314 | ret = io_buffer_validate(&iov); |
8315 | if (ret) | |
0a96bbe4 | 8316 | break; |
edafccee | 8317 | |
0a96bbe4 BM |
8318 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8319 | if (ret) | |
8320 | break; | |
edafccee JA |
8321 | |
8322 | ctx->nr_user_bufs++; | |
8323 | } | |
0a96bbe4 BM |
8324 | |
8325 | if (ret) | |
8326 | io_sqe_buffers_unregister(ctx); | |
8327 | ||
edafccee JA |
8328 | return ret; |
8329 | } | |
8330 | ||
9b402849 JA |
8331 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8332 | { | |
8333 | __s32 __user *fds = arg; | |
8334 | int fd; | |
8335 | ||
8336 | if (ctx->cq_ev_fd) | |
8337 | return -EBUSY; | |
8338 | ||
8339 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8340 | return -EFAULT; | |
8341 | ||
8342 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8343 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8344 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8345 | ctx->cq_ev_fd = NULL; | |
8346 | return ret; | |
8347 | } | |
8348 | ||
8349 | return 0; | |
8350 | } | |
8351 | ||
8352 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8353 | { | |
8354 | if (ctx->cq_ev_fd) { | |
8355 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8356 | ctx->cq_ev_fd = NULL; | |
8357 | return 0; | |
8358 | } | |
8359 | ||
8360 | return -ENXIO; | |
8361 | } | |
8362 | ||
5a2e745d JA |
8363 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8364 | { | |
9e15c3a0 JA |
8365 | struct io_buffer *buf; |
8366 | unsigned long index; | |
8367 | ||
8368 | xa_for_each(&ctx->io_buffers, index, buf) | |
8369 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8370 | } |
8371 | ||
68e68ee6 | 8372 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8373 | { |
68e68ee6 | 8374 | struct io_kiocb *req, *nxt; |
1b4c351f | 8375 | |
68e68ee6 JA |
8376 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8377 | if (tsk && req->task != tsk) | |
8378 | continue; | |
1b4c351f JA |
8379 | list_del(&req->compl.list); |
8380 | kmem_cache_free(req_cachep, req); | |
8381 | } | |
8382 | } | |
8383 | ||
4010fec4 | 8384 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8385 | { |
bf019da7 | 8386 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8387 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8388 | |
9a4fdbd8 JA |
8389 | mutex_lock(&ctx->uring_lock); |
8390 | ||
8e5c66c4 | 8391 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8392 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8393 | submit_state->reqs); | |
8e5c66c4 PB |
8394 | submit_state->free_reqs = 0; |
8395 | } | |
9a4fdbd8 JA |
8396 | |
8397 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8398 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8399 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8400 | spin_unlock_irq(&ctx->completion_lock); |
8401 | ||
e5547d2c PB |
8402 | io_req_cache_free(&cs->free_list, NULL); |
8403 | ||
9a4fdbd8 JA |
8404 | mutex_unlock(&ctx->uring_lock); |
8405 | } | |
8406 | ||
2b188cc1 JA |
8407 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8408 | { | |
04fc6c80 PB |
8409 | /* |
8410 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8411 | * and they are free to do so while still holding uring_lock or |
8412 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8413 | */ |
8414 | mutex_lock(&ctx->uring_lock); | |
8415 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8416 | spin_lock_irq(&ctx->completion_lock); |
8417 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8418 | |
37d1e2e3 | 8419 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8420 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8421 | |
37d1e2e3 | 8422 | if (ctx->mm_account) { |
2aede0e4 JA |
8423 | mmdrop(ctx->mm_account); |
8424 | ctx->mm_account = NULL; | |
30975825 | 8425 | } |
def596e9 | 8426 | |
8bad28d8 | 8427 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8428 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8429 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8430 | io_eventfd_unregister(ctx); |
5a2e745d | 8431 | io_destroy_buffers(ctx); |
def596e9 | 8432 | |
2b188cc1 | 8433 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8434 | if (ctx->ring_sock) { |
8435 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8436 | sock_release(ctx->ring_sock); |
355e8d26 | 8437 | } |
2b188cc1 JA |
8438 | #endif |
8439 | ||
75b28aff | 8440 | io_mem_free(ctx->rings); |
2b188cc1 | 8441 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8442 | |
8443 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8444 | free_uid(ctx->user); |
4010fec4 | 8445 | io_req_caches_free(ctx); |
e941894e JA |
8446 | if (ctx->hash_map) |
8447 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8448 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8449 | kfree(ctx); |
8450 | } | |
8451 | ||
8452 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8453 | { | |
8454 | struct io_ring_ctx *ctx = file->private_data; | |
8455 | __poll_t mask = 0; | |
8456 | ||
8457 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8458 | /* |
8459 | * synchronizes with barrier from wq_has_sleeper call in | |
8460 | * io_commit_cqring | |
8461 | */ | |
2b188cc1 | 8462 | smp_rmb(); |
90554200 | 8463 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8464 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8465 | |
8466 | /* | |
8467 | * Don't flush cqring overflow list here, just do a simple check. | |
8468 | * Otherwise there could possible be ABBA deadlock: | |
8469 | * CPU0 CPU1 | |
8470 | * ---- ---- | |
8471 | * lock(&ctx->uring_lock); | |
8472 | * lock(&ep->mtx); | |
8473 | * lock(&ctx->uring_lock); | |
8474 | * lock(&ep->mtx); | |
8475 | * | |
8476 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8477 | * pushs them to do the flush. | |
8478 | */ | |
8479 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8480 | mask |= EPOLLIN | EPOLLRDNORM; |
8481 | ||
8482 | return mask; | |
8483 | } | |
8484 | ||
8485 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8486 | { | |
8487 | struct io_ring_ctx *ctx = file->private_data; | |
8488 | ||
8489 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8490 | } | |
8491 | ||
0bead8cd | 8492 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8493 | { |
4379bf8b | 8494 | const struct cred *creds; |
071698e1 | 8495 | |
61cf9370 | 8496 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8497 | if (creds) { |
8498 | put_cred(creds); | |
0bead8cd | 8499 | return 0; |
1e6fa521 | 8500 | } |
0bead8cd YD |
8501 | |
8502 | return -EINVAL; | |
8503 | } | |
8504 | ||
9b465711 | 8505 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8506 | { |
9b465711 | 8507 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8508 | } |
8509 | ||
d56d938b PB |
8510 | struct io_tctx_exit { |
8511 | struct callback_head task_work; | |
8512 | struct completion completion; | |
baf186c4 | 8513 | struct io_ring_ctx *ctx; |
d56d938b PB |
8514 | }; |
8515 | ||
8516 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8517 | { | |
8518 | struct io_uring_task *tctx = current->io_uring; | |
8519 | struct io_tctx_exit *work; | |
8520 | ||
8521 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8522 | /* | |
8523 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8524 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8525 | */ | |
8526 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8527 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8528 | complete(&work->completion); |
8529 | } | |
8530 | ||
85faa7b8 JA |
8531 | static void io_ring_exit_work(struct work_struct *work) |
8532 | { | |
d56d938b | 8533 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8534 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8535 | struct io_tctx_exit exit; |
8536 | struct io_tctx_node *node; | |
8537 | int ret; | |
85faa7b8 | 8538 | |
56952e91 JA |
8539 | /* |
8540 | * If we're doing polled IO and end up having requests being | |
8541 | * submitted async (out-of-line), then completions can come in while | |
8542 | * we're waiting for refs to drop. We need to reap these manually, | |
8543 | * as nobody else will be looking for them. | |
8544 | */ | |
b2edc0a7 | 8545 | do { |
9936c7c2 | 8546 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8547 | |
8548 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8549 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8550 | |
8551 | mutex_lock(&ctx->uring_lock); | |
8552 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8553 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8554 | ||
d56d938b PB |
8555 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8556 | ctx_node); | |
baf186c4 | 8557 | exit.ctx = ctx; |
d56d938b PB |
8558 | init_completion(&exit.completion); |
8559 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8560 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8561 | if (WARN_ON_ONCE(ret)) | |
8562 | continue; | |
8563 | wake_up_process(node->task); | |
8564 | ||
8565 | mutex_unlock(&ctx->uring_lock); | |
8566 | wait_for_completion(&exit.completion); | |
8567 | cond_resched(); | |
8568 | mutex_lock(&ctx->uring_lock); | |
8569 | } | |
8570 | mutex_unlock(&ctx->uring_lock); | |
8571 | ||
85faa7b8 JA |
8572 | io_ring_ctx_free(ctx); |
8573 | } | |
8574 | ||
2b188cc1 JA |
8575 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8576 | { | |
61cf9370 MWO |
8577 | unsigned long index; |
8578 | struct creds *creds; | |
8579 | ||
2b188cc1 JA |
8580 | mutex_lock(&ctx->uring_lock); |
8581 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8582 | /* if force is set, the ring is going away. always drop after that */ |
8583 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8584 | if (ctx->rings) |
6c503150 | 8585 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8586 | xa_for_each(&ctx->personalities, index, creds) |
8587 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8588 | mutex_unlock(&ctx->uring_lock); |
8589 | ||
f6d54255 PB |
8590 | /* prevent SQPOLL from submitting new requests */ |
8591 | if (ctx->sq_data) { | |
8592 | io_sq_thread_park(ctx->sq_data); | |
8593 | list_del_init(&ctx->sqd_list); | |
8594 | io_sqd_update_thread_idle(ctx->sq_data); | |
8595 | io_sq_thread_unpark(ctx->sq_data); | |
8596 | } | |
8597 | ||
6b81928d PB |
8598 | io_kill_timeouts(ctx, NULL, NULL); |
8599 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8600 | |
15dff286 | 8601 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8602 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8603 | |
85faa7b8 | 8604 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8605 | /* |
8606 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8607 | * if we're exiting a ton of rings at the same time. It just adds | |
8608 | * noise and overhead, there's no discernable change in runtime | |
8609 | * over using system_wq. | |
8610 | */ | |
8611 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8612 | } |
8613 | ||
8614 | static int io_uring_release(struct inode *inode, struct file *file) | |
8615 | { | |
8616 | struct io_ring_ctx *ctx = file->private_data; | |
8617 | ||
8618 | file->private_data = NULL; | |
8619 | io_ring_ctx_wait_and_kill(ctx); | |
8620 | return 0; | |
8621 | } | |
8622 | ||
f6edbabb PB |
8623 | struct io_task_cancel { |
8624 | struct task_struct *task; | |
8625 | struct files_struct *files; | |
8626 | }; | |
f254ac04 | 8627 | |
f6edbabb | 8628 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8629 | { |
9a472ef7 | 8630 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8631 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8632 | bool ret; |
8633 | ||
f6edbabb | 8634 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8635 | unsigned long flags; |
8636 | struct io_ring_ctx *ctx = req->ctx; | |
8637 | ||
8638 | /* protect against races with linked timeouts */ | |
8639 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8640 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8641 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8642 | } else { | |
f6edbabb | 8643 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8644 | } |
8645 | return ret; | |
b711d4ea JA |
8646 | } |
8647 | ||
e1915f76 | 8648 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8649 | struct task_struct *task, |
b7ddce3c PB |
8650 | struct files_struct *files) |
8651 | { | |
e1915f76 | 8652 | struct io_defer_entry *de; |
b7ddce3c PB |
8653 | LIST_HEAD(list); |
8654 | ||
8655 | spin_lock_irq(&ctx->completion_lock); | |
8656 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8657 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8658 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8659 | break; | |
8660 | } | |
8661 | } | |
8662 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8663 | if (list_empty(&list)) |
8664 | return false; | |
b7ddce3c PB |
8665 | |
8666 | while (!list_empty(&list)) { | |
8667 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8668 | list_del_init(&de->list); | |
8669 | req_set_fail_links(de->req); | |
8670 | io_put_req(de->req); | |
8671 | io_req_complete(de->req, -ECANCELED); | |
8672 | kfree(de); | |
8673 | } | |
e1915f76 | 8674 | return true; |
b7ddce3c PB |
8675 | } |
8676 | ||
1b00764f PB |
8677 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8678 | { | |
8679 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8680 | ||
8681 | return req->ctx == data; | |
8682 | } | |
8683 | ||
8684 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8685 | { | |
8686 | struct io_tctx_node *node; | |
8687 | enum io_wq_cancel cret; | |
8688 | bool ret = false; | |
8689 | ||
8690 | mutex_lock(&ctx->uring_lock); | |
8691 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8692 | struct io_uring_task *tctx = node->task->io_uring; | |
8693 | ||
8694 | /* | |
8695 | * io_wq will stay alive while we hold uring_lock, because it's | |
8696 | * killed after ctx nodes, which requires to take the lock. | |
8697 | */ | |
8698 | if (!tctx || !tctx->io_wq) | |
8699 | continue; | |
8700 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8701 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8702 | } | |
8703 | mutex_unlock(&ctx->uring_lock); | |
8704 | ||
8705 | return ret; | |
8706 | } | |
8707 | ||
9936c7c2 PB |
8708 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8709 | struct task_struct *task, | |
8710 | struct files_struct *files) | |
8711 | { | |
8712 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8713 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8714 | |
8715 | while (1) { | |
8716 | enum io_wq_cancel cret; | |
8717 | bool ret = false; | |
8718 | ||
1b00764f PB |
8719 | if (!task) { |
8720 | ret |= io_uring_try_cancel_iowq(ctx); | |
8721 | } else if (tctx && tctx->io_wq) { | |
8722 | /* | |
8723 | * Cancels requests of all rings, not only @ctx, but | |
8724 | * it's fine as the task is in exit/exec. | |
8725 | */ | |
5aa75ed5 | 8726 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8727 | &cancel, true); |
8728 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8729 | } | |
8730 | ||
8731 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8732 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8733 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8734 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8735 | io_iopoll_try_reap_events(ctx); | |
8736 | ret = true; | |
8737 | } | |
8738 | } | |
8739 | ||
e1915f76 | 8740 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8741 | ret |= io_poll_remove_all(ctx, task, files); |
8742 | ret |= io_kill_timeouts(ctx, task, files); | |
8743 | ret |= io_run_task_work(); | |
ba50a036 | 8744 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8745 | io_cqring_overflow_flush(ctx, true, task, files); |
8746 | if (!ret) | |
8747 | break; | |
8748 | cond_resched(); | |
8749 | } | |
8750 | } | |
8751 | ||
ca70f00b PB |
8752 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8753 | struct task_struct *task, | |
8754 | struct files_struct *files) | |
8755 | { | |
8756 | struct io_kiocb *req; | |
8757 | int cnt = 0; | |
8758 | ||
8759 | spin_lock_irq(&ctx->inflight_lock); | |
8760 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8761 | cnt += io_match_task(req, task, files); | |
8762 | spin_unlock_irq(&ctx->inflight_lock); | |
8763 | return cnt; | |
8764 | } | |
8765 | ||
b52fda00 | 8766 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8767 | struct task_struct *task, |
fcb323cc JA |
8768 | struct files_struct *files) |
8769 | { | |
fcb323cc | 8770 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8771 | DEFINE_WAIT(wait); |
ca70f00b | 8772 | int inflight; |
fcb323cc | 8773 | |
ca70f00b PB |
8774 | inflight = io_uring_count_inflight(ctx, task, files); |
8775 | if (!inflight) | |
fcb323cc | 8776 | break; |
f6edbabb | 8777 | |
9936c7c2 | 8778 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8779 | |
8780 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8781 | TASK_UNINTERRUPTIBLE); | |
8782 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8783 | schedule(); | |
c98de08c | 8784 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8785 | } |
0f212204 JA |
8786 | } |
8787 | ||
0f212204 JA |
8788 | /* |
8789 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8790 | */ | |
baf186c4 | 8791 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8792 | { |
236434c3 | 8793 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8794 | struct io_tctx_node *node; |
a528b04e | 8795 | int ret; |
236434c3 MWO |
8796 | |
8797 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8798 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8799 | if (unlikely(ret)) |
8800 | return ret; | |
236434c3 | 8801 | tctx = current->io_uring; |
0f212204 | 8802 | } |
baf186c4 PB |
8803 | if (tctx->last != ctx) { |
8804 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8805 | |
236434c3 | 8806 | if (!old) { |
13bf43f5 PB |
8807 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8808 | if (!node) | |
8809 | return -ENOMEM; | |
8810 | node->ctx = ctx; | |
13bf43f5 PB |
8811 | node->task = current; |
8812 | ||
baf186c4 | 8813 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8814 | node, GFP_KERNEL)); |
a528b04e | 8815 | if (ret) { |
13bf43f5 | 8816 | kfree(node); |
a528b04e PB |
8817 | return ret; |
8818 | } | |
13bf43f5 PB |
8819 | |
8820 | mutex_lock(&ctx->uring_lock); | |
8821 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8822 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8823 | } |
baf186c4 | 8824 | tctx->last = ctx; |
0f212204 | 8825 | } |
0f212204 JA |
8826 | return 0; |
8827 | } | |
8828 | ||
8829 | /* | |
8830 | * Remove this io_uring_file -> task mapping. | |
8831 | */ | |
2941267b | 8832 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8833 | { |
8834 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8835 | struct io_tctx_node *node; |
2941267b | 8836 | |
eebd2e37 PB |
8837 | if (!tctx) |
8838 | return; | |
13bf43f5 PB |
8839 | node = xa_erase(&tctx->xa, index); |
8840 | if (!node) | |
2941267b | 8841 | return; |
0f212204 | 8842 | |
13bf43f5 PB |
8843 | WARN_ON_ONCE(current != node->task); |
8844 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8845 | ||
8846 | mutex_lock(&node->ctx->uring_lock); | |
8847 | list_del(&node->ctx_node); | |
8848 | mutex_unlock(&node->ctx->uring_lock); | |
8849 | ||
baf186c4 | 8850 | if (tctx->last == node->ctx) |
0f212204 | 8851 | tctx->last = NULL; |
13bf43f5 | 8852 | kfree(node); |
0f212204 JA |
8853 | } |
8854 | ||
8452d4a6 | 8855 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8856 | { |
13bf43f5 | 8857 | struct io_tctx_node *node; |
de7f1d9e PB |
8858 | unsigned long index; |
8859 | ||
13bf43f5 | 8860 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8861 | io_uring_del_task_file(index); |
8452d4a6 PB |
8862 | if (tctx->io_wq) { |
8863 | io_wq_put_and_exit(tctx->io_wq); | |
8864 | tctx->io_wq = NULL; | |
8865 | } | |
de7f1d9e PB |
8866 | } |
8867 | ||
521d6a73 PB |
8868 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8869 | { | |
8870 | return percpu_counter_sum(&tctx->inflight); | |
8871 | } | |
8872 | ||
8873 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8874 | { | |
8875 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8876 | struct io_ring_ctx *ctx = work->ctx; | |
8877 | struct io_sq_data *sqd = ctx->sq_data; | |
8878 | ||
8879 | if (sqd->thread) | |
8880 | io_uring_cancel_sqpoll(ctx); | |
8881 | complete(&work->completion); | |
8882 | } | |
8883 | ||
8884 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8885 | { | |
8886 | struct io_sq_data *sqd = ctx->sq_data; | |
8887 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8888 | struct task_struct *task; | |
8889 | ||
8890 | io_sq_thread_park(sqd); | |
8891 | list_del_init(&ctx->sqd_list); | |
8892 | io_sqd_update_thread_idle(sqd); | |
8893 | task = sqd->thread; | |
8894 | if (task) { | |
8895 | init_completion(&work.completion); | |
8896 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8897 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8898 | wake_up_process(task); |
8899 | } | |
8900 | io_sq_thread_unpark(sqd); | |
8901 | ||
8902 | if (task) | |
8903 | wait_for_completion(&work.completion); | |
8904 | } | |
8905 | ||
0f212204 JA |
8906 | void __io_uring_files_cancel(struct files_struct *files) |
8907 | { | |
8908 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8909 | struct io_tctx_node *node; |
ce765372 | 8910 | unsigned long index; |
0f212204 JA |
8911 | |
8912 | /* make sure overflow events are dropped */ | |
fdaf083c | 8913 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8914 | xa_for_each(&tctx->xa, index, node) { |
8915 | struct io_ring_ctx *ctx = node->ctx; | |
8916 | ||
8917 | if (ctx->sq_data) { | |
8918 | io_sqpoll_cancel_sync(ctx); | |
8919 | continue; | |
8920 | } | |
8921 | io_uring_cancel_files(ctx, current, files); | |
8922 | if (!files) | |
8923 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8924 | } | |
fdaf083c | 8925 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8926 | |
8452d4a6 PB |
8927 | if (files) |
8928 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8929 | } |
8930 | ||
521d6a73 | 8931 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8932 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8933 | { | |
37d1e2e3 | 8934 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8935 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8936 | s64 inflight; |
8937 | DEFINE_WAIT(wait); | |
fdaf083c | 8938 | |
521d6a73 PB |
8939 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8940 | ||
0e9ddb39 PB |
8941 | atomic_inc(&tctx->in_idle); |
8942 | do { | |
8943 | /* read completions before cancelations */ | |
8944 | inflight = tctx_inflight(tctx); | |
8945 | if (!inflight) | |
8946 | break; | |
521d6a73 | 8947 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8948 | |
0e9ddb39 PB |
8949 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8950 | /* | |
8951 | * If we've seen completions, retry without waiting. This | |
8952 | * avoids a race where a completion comes in before we did | |
8953 | * prepare_to_wait(). | |
8954 | */ | |
8955 | if (inflight == tctx_inflight(tctx)) | |
8956 | schedule(); | |
8957 | finish_wait(&tctx->wait, &wait); | |
8958 | } while (1); | |
8959 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8960 | } |
8961 | ||
0f212204 JA |
8962 | /* |
8963 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8964 | * requests. | |
8965 | */ | |
8966 | void __io_uring_task_cancel(void) | |
8967 | { | |
8968 | struct io_uring_task *tctx = current->io_uring; | |
8969 | DEFINE_WAIT(wait); | |
d8a6df10 | 8970 | s64 inflight; |
0f212204 JA |
8971 | |
8972 | /* make sure overflow events are dropped */ | |
fdaf083c | 8973 | atomic_inc(&tctx->in_idle); |
d8a6df10 | 8974 | do { |
0f212204 | 8975 | /* read completions before cancelations */ |
fdaf083c | 8976 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8977 | if (!inflight) |
8978 | break; | |
0f212204 JA |
8979 | __io_uring_files_cancel(NULL); |
8980 | ||
8981 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8982 | ||
8983 | /* | |
a1bb3cd5 PB |
8984 | * If we've seen completions, retry without waiting. This |
8985 | * avoids a race where a completion comes in before we did | |
8986 | * prepare_to_wait(). | |
0f212204 | 8987 | */ |
a1bb3cd5 PB |
8988 | if (inflight == tctx_inflight(tctx)) |
8989 | schedule(); | |
f57555ed | 8990 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8991 | } while (1); |
0f212204 | 8992 | |
fdaf083c | 8993 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8994 | |
8452d4a6 PB |
8995 | io_uring_clean_tctx(tctx); |
8996 | /* all current's requests should be gone, we can kill tctx */ | |
8997 | __io_uring_free(current); | |
44e728b8 PB |
8998 | } |
8999 | ||
6c5c240e RP |
9000 | static void *io_uring_validate_mmap_request(struct file *file, |
9001 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9002 | { |
2b188cc1 | 9003 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9004 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9005 | struct page *page; |
9006 | void *ptr; | |
9007 | ||
9008 | switch (offset) { | |
9009 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9010 | case IORING_OFF_CQ_RING: |
9011 | ptr = ctx->rings; | |
2b188cc1 JA |
9012 | break; |
9013 | case IORING_OFF_SQES: | |
9014 | ptr = ctx->sq_sqes; | |
9015 | break; | |
2b188cc1 | 9016 | default: |
6c5c240e | 9017 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9018 | } |
9019 | ||
9020 | page = virt_to_head_page(ptr); | |
a50b854e | 9021 | if (sz > page_size(page)) |
6c5c240e RP |
9022 | return ERR_PTR(-EINVAL); |
9023 | ||
9024 | return ptr; | |
9025 | } | |
9026 | ||
9027 | #ifdef CONFIG_MMU | |
9028 | ||
9029 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9030 | { | |
9031 | size_t sz = vma->vm_end - vma->vm_start; | |
9032 | unsigned long pfn; | |
9033 | void *ptr; | |
9034 | ||
9035 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9036 | if (IS_ERR(ptr)) | |
9037 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9038 | |
9039 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9040 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9041 | } | |
9042 | ||
6c5c240e RP |
9043 | #else /* !CONFIG_MMU */ |
9044 | ||
9045 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9046 | { | |
9047 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9048 | } | |
9049 | ||
9050 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9051 | { | |
9052 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9053 | } | |
9054 | ||
9055 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9056 | unsigned long addr, unsigned long len, | |
9057 | unsigned long pgoff, unsigned long flags) | |
9058 | { | |
9059 | void *ptr; | |
9060 | ||
9061 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9062 | if (IS_ERR(ptr)) | |
9063 | return PTR_ERR(ptr); | |
9064 | ||
9065 | return (unsigned long) ptr; | |
9066 | } | |
9067 | ||
9068 | #endif /* !CONFIG_MMU */ | |
9069 | ||
d9d05217 | 9070 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9071 | { |
9072 | DEFINE_WAIT(wait); | |
9073 | ||
9074 | do { | |
9075 | if (!io_sqring_full(ctx)) | |
9076 | break; | |
90554200 JA |
9077 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9078 | ||
9079 | if (!io_sqring_full(ctx)) | |
9080 | break; | |
90554200 JA |
9081 | schedule(); |
9082 | } while (!signal_pending(current)); | |
9083 | ||
9084 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9085 | return 0; |
90554200 JA |
9086 | } |
9087 | ||
c73ebb68 HX |
9088 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9089 | struct __kernel_timespec __user **ts, | |
9090 | const sigset_t __user **sig) | |
9091 | { | |
9092 | struct io_uring_getevents_arg arg; | |
9093 | ||
9094 | /* | |
9095 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9096 | * is just a pointer to the sigset_t. | |
9097 | */ | |
9098 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9099 | *sig = (const sigset_t __user *) argp; | |
9100 | *ts = NULL; | |
9101 | return 0; | |
9102 | } | |
9103 | ||
9104 | /* | |
9105 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9106 | * timespec and sigset_t pointers if good. | |
9107 | */ | |
9108 | if (*argsz != sizeof(arg)) | |
9109 | return -EINVAL; | |
9110 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9111 | return -EFAULT; | |
9112 | *sig = u64_to_user_ptr(arg.sigmask); | |
9113 | *argsz = arg.sigmask_sz; | |
9114 | *ts = u64_to_user_ptr(arg.ts); | |
9115 | return 0; | |
9116 | } | |
9117 | ||
2b188cc1 | 9118 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9119 | u32, min_complete, u32, flags, const void __user *, argp, |
9120 | size_t, argsz) | |
2b188cc1 JA |
9121 | { |
9122 | struct io_ring_ctx *ctx; | |
9123 | long ret = -EBADF; | |
9124 | int submitted = 0; | |
9125 | struct fd f; | |
9126 | ||
4c6e277c | 9127 | io_run_task_work(); |
b41e9852 | 9128 | |
90554200 | 9129 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9130 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9131 | return -EINVAL; |
9132 | ||
9133 | f = fdget(fd); | |
9134 | if (!f.file) | |
9135 | return -EBADF; | |
9136 | ||
9137 | ret = -EOPNOTSUPP; | |
9138 | if (f.file->f_op != &io_uring_fops) | |
9139 | goto out_fput; | |
9140 | ||
9141 | ret = -ENXIO; | |
9142 | ctx = f.file->private_data; | |
9143 | if (!percpu_ref_tryget(&ctx->refs)) | |
9144 | goto out_fput; | |
9145 | ||
7e84e1c7 SG |
9146 | ret = -EBADFD; |
9147 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9148 | goto out; | |
9149 | ||
6c271ce2 JA |
9150 | /* |
9151 | * For SQ polling, the thread will do all submissions and completions. | |
9152 | * Just return the requested submit count, and wake the thread if | |
9153 | * we were asked to. | |
9154 | */ | |
b2a9eada | 9155 | ret = 0; |
6c271ce2 | 9156 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9157 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9158 | |
d9d05217 | 9159 | ret = -EOWNERDEAD; |
04147488 SM |
9160 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9161 | goto out; | |
9162 | } | |
6c271ce2 | 9163 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9164 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9165 | if (flags & IORING_ENTER_SQ_WAIT) { |
9166 | ret = io_sqpoll_wait_sq(ctx); | |
9167 | if (ret) | |
9168 | goto out; | |
9169 | } | |
6c271ce2 | 9170 | submitted = to_submit; |
b2a9eada | 9171 | } else if (to_submit) { |
baf186c4 | 9172 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9173 | if (unlikely(ret)) |
9174 | goto out; | |
2b188cc1 | 9175 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9176 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9177 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9178 | |
9179 | if (submitted != to_submit) | |
9180 | goto out; | |
2b188cc1 JA |
9181 | } |
9182 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9183 | const sigset_t __user *sig; |
9184 | struct __kernel_timespec __user *ts; | |
9185 | ||
9186 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9187 | if (unlikely(ret)) | |
9188 | goto out; | |
9189 | ||
2b188cc1 JA |
9190 | min_complete = min(min_complete, ctx->cq_entries); |
9191 | ||
32b2244a XW |
9192 | /* |
9193 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9194 | * space applications don't need to do io completion events | |
9195 | * polling again, they can rely on io_sq_thread to do polling | |
9196 | * work, which can reduce cpu usage and uring_lock contention. | |
9197 | */ | |
9198 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9199 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9200 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9201 | } else { |
c73ebb68 | 9202 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9203 | } |
2b188cc1 JA |
9204 | } |
9205 | ||
7c504e65 | 9206 | out: |
6805b32e | 9207 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9208 | out_fput: |
9209 | fdput(f); | |
9210 | return submitted ? submitted : ret; | |
9211 | } | |
9212 | ||
bebdb65e | 9213 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9214 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9215 | const struct cred *cred) | |
87ce955b | 9216 | { |
87ce955b JA |
9217 | struct user_namespace *uns = seq_user_ns(m); |
9218 | struct group_info *gi; | |
9219 | kernel_cap_t cap; | |
9220 | unsigned __capi; | |
9221 | int g; | |
9222 | ||
9223 | seq_printf(m, "%5d\n", id); | |
9224 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9225 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9226 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9227 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9228 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9229 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9230 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9231 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9232 | seq_puts(m, "\n\tGroups:\t"); | |
9233 | gi = cred->group_info; | |
9234 | for (g = 0; g < gi->ngroups; g++) { | |
9235 | seq_put_decimal_ull(m, g ? " " : "", | |
9236 | from_kgid_munged(uns, gi->gid[g])); | |
9237 | } | |
9238 | seq_puts(m, "\n\tCapEff:\t"); | |
9239 | cap = cred->cap_effective; | |
9240 | CAP_FOR_EACH_U32(__capi) | |
9241 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9242 | seq_putc(m, '\n'); | |
9243 | return 0; | |
9244 | } | |
9245 | ||
9246 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9247 | { | |
dbbe9c64 | 9248 | struct io_sq_data *sq = NULL; |
fad8e0de | 9249 | bool has_lock; |
87ce955b JA |
9250 | int i; |
9251 | ||
fad8e0de JA |
9252 | /* |
9253 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9254 | * since fdinfo case grabs it in the opposite direction of normal use | |
9255 | * cases. If we fail to get the lock, we just don't iterate any | |
9256 | * structures that could be going away outside the io_uring mutex. | |
9257 | */ | |
9258 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9259 | ||
5f3f26f9 | 9260 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9261 | sq = ctx->sq_data; |
5f3f26f9 JA |
9262 | if (!sq->thread) |
9263 | sq = NULL; | |
9264 | } | |
dbbe9c64 JQ |
9265 | |
9266 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9267 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9268 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9269 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9270 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9271 | |
87ce955b JA |
9272 | if (f) |
9273 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9274 | else | |
9275 | seq_printf(m, "%5u: <none>\n", i); | |
9276 | } | |
9277 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9278 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9279 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9280 | ||
9281 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9282 | (unsigned int) buf->len); | |
9283 | } | |
61cf9370 MWO |
9284 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9285 | unsigned long index; | |
9286 | const struct cred *cred; | |
9287 | ||
87ce955b | 9288 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9289 | xa_for_each(&ctx->personalities, index, cred) |
9290 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9291 | } |
d7718a9d JA |
9292 | seq_printf(m, "PollList:\n"); |
9293 | spin_lock_irq(&ctx->completion_lock); | |
9294 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9295 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9296 | struct io_kiocb *req; | |
9297 | ||
9298 | hlist_for_each_entry(req, list, hash_node) | |
9299 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9300 | req->task->task_works != NULL); | |
9301 | } | |
9302 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9303 | if (has_lock) |
9304 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9305 | } |
9306 | ||
9307 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9308 | { | |
9309 | struct io_ring_ctx *ctx = f->private_data; | |
9310 | ||
9311 | if (percpu_ref_tryget(&ctx->refs)) { | |
9312 | __io_uring_show_fdinfo(ctx, m); | |
9313 | percpu_ref_put(&ctx->refs); | |
9314 | } | |
9315 | } | |
bebdb65e | 9316 | #endif |
87ce955b | 9317 | |
2b188cc1 JA |
9318 | static const struct file_operations io_uring_fops = { |
9319 | .release = io_uring_release, | |
9320 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9321 | #ifndef CONFIG_MMU |
9322 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9323 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9324 | #endif | |
2b188cc1 JA |
9325 | .poll = io_uring_poll, |
9326 | .fasync = io_uring_fasync, | |
bebdb65e | 9327 | #ifdef CONFIG_PROC_FS |
87ce955b | 9328 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9329 | #endif |
2b188cc1 JA |
9330 | }; |
9331 | ||
9332 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9333 | struct io_uring_params *p) | |
9334 | { | |
75b28aff HV |
9335 | struct io_rings *rings; |
9336 | size_t size, sq_array_offset; | |
2b188cc1 | 9337 | |
bd740481 JA |
9338 | /* make sure these are sane, as we already accounted them */ |
9339 | ctx->sq_entries = p->sq_entries; | |
9340 | ctx->cq_entries = p->cq_entries; | |
9341 | ||
75b28aff HV |
9342 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9343 | if (size == SIZE_MAX) | |
9344 | return -EOVERFLOW; | |
9345 | ||
9346 | rings = io_mem_alloc(size); | |
9347 | if (!rings) | |
2b188cc1 JA |
9348 | return -ENOMEM; |
9349 | ||
75b28aff HV |
9350 | ctx->rings = rings; |
9351 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9352 | rings->sq_ring_mask = p->sq_entries - 1; | |
9353 | rings->cq_ring_mask = p->cq_entries - 1; | |
9354 | rings->sq_ring_entries = p->sq_entries; | |
9355 | rings->cq_ring_entries = p->cq_entries; | |
9356 | ctx->sq_mask = rings->sq_ring_mask; | |
9357 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9358 | |
9359 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9360 | if (size == SIZE_MAX) { |
9361 | io_mem_free(ctx->rings); | |
9362 | ctx->rings = NULL; | |
2b188cc1 | 9363 | return -EOVERFLOW; |
eb065d30 | 9364 | } |
2b188cc1 JA |
9365 | |
9366 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9367 | if (!ctx->sq_sqes) { |
9368 | io_mem_free(ctx->rings); | |
9369 | ctx->rings = NULL; | |
2b188cc1 | 9370 | return -ENOMEM; |
eb065d30 | 9371 | } |
2b188cc1 | 9372 | |
2b188cc1 JA |
9373 | return 0; |
9374 | } | |
9375 | ||
9faadcc8 PB |
9376 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9377 | { | |
9378 | int ret, fd; | |
9379 | ||
9380 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9381 | if (fd < 0) | |
9382 | return fd; | |
9383 | ||
baf186c4 | 9384 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9385 | if (ret) { |
9386 | put_unused_fd(fd); | |
9387 | return ret; | |
9388 | } | |
9389 | fd_install(fd, file); | |
9390 | return fd; | |
9391 | } | |
9392 | ||
2b188cc1 JA |
9393 | /* |
9394 | * Allocate an anonymous fd, this is what constitutes the application | |
9395 | * visible backing of an io_uring instance. The application mmaps this | |
9396 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9397 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9398 | */ | |
9faadcc8 | 9399 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9400 | { |
9401 | struct file *file; | |
9faadcc8 | 9402 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9403 | int ret; |
9404 | ||
2b188cc1 JA |
9405 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9406 | &ctx->ring_sock); | |
9407 | if (ret) | |
9faadcc8 | 9408 | return ERR_PTR(ret); |
2b188cc1 JA |
9409 | #endif |
9410 | ||
2b188cc1 JA |
9411 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9412 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9413 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9414 | if (IS_ERR(file)) { |
9415 | sock_release(ctx->ring_sock); | |
9416 | ctx->ring_sock = NULL; | |
9417 | } else { | |
9418 | ctx->ring_sock->file = file; | |
0f212204 | 9419 | } |
2b188cc1 | 9420 | #endif |
9faadcc8 | 9421 | return file; |
2b188cc1 JA |
9422 | } |
9423 | ||
7f13657d XW |
9424 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9425 | struct io_uring_params __user *params) | |
2b188cc1 | 9426 | { |
2b188cc1 | 9427 | struct io_ring_ctx *ctx; |
9faadcc8 | 9428 | struct file *file; |
2b188cc1 JA |
9429 | int ret; |
9430 | ||
8110c1a6 | 9431 | if (!entries) |
2b188cc1 | 9432 | return -EINVAL; |
8110c1a6 JA |
9433 | if (entries > IORING_MAX_ENTRIES) { |
9434 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9435 | return -EINVAL; | |
9436 | entries = IORING_MAX_ENTRIES; | |
9437 | } | |
2b188cc1 JA |
9438 | |
9439 | /* | |
9440 | * Use twice as many entries for the CQ ring. It's possible for the | |
9441 | * application to drive a higher depth than the size of the SQ ring, | |
9442 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9443 | * some flexibility in overcommitting a bit. If the application has |
9444 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9445 | * of CQ ring entries manually. | |
2b188cc1 JA |
9446 | */ |
9447 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9448 | if (p->flags & IORING_SETUP_CQSIZE) { |
9449 | /* | |
9450 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9451 | * to a power-of-two, if it isn't already. We do NOT impose | |
9452 | * any cq vs sq ring sizing. | |
9453 | */ | |
eb2667b3 | 9454 | if (!p->cq_entries) |
33a107f0 | 9455 | return -EINVAL; |
8110c1a6 JA |
9456 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9457 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9458 | return -EINVAL; | |
9459 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9460 | } | |
eb2667b3 JQ |
9461 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9462 | if (p->cq_entries < p->sq_entries) | |
9463 | return -EINVAL; | |
33a107f0 JA |
9464 | } else { |
9465 | p->cq_entries = 2 * p->sq_entries; | |
9466 | } | |
2b188cc1 | 9467 | |
2b188cc1 | 9468 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9469 | if (!ctx) |
2b188cc1 | 9470 | return -ENOMEM; |
2b188cc1 | 9471 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9472 | if (!capable(CAP_IPC_LOCK)) |
9473 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9474 | |
9475 | /* | |
9476 | * This is just grabbed for accounting purposes. When a process exits, | |
9477 | * the mm is exited and dropped before the files, hence we need to hang | |
9478 | * on to this mm purely for the purposes of being able to unaccount | |
9479 | * memory (locked/pinned vm). It's not used for anything else. | |
9480 | */ | |
6b7898eb | 9481 | mmgrab(current->mm); |
2aede0e4 | 9482 | ctx->mm_account = current->mm; |
6b7898eb | 9483 | |
2b188cc1 JA |
9484 | ret = io_allocate_scq_urings(ctx, p); |
9485 | if (ret) | |
9486 | goto err; | |
9487 | ||
7e84e1c7 | 9488 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9489 | if (ret) |
9490 | goto err; | |
9491 | ||
2b188cc1 | 9492 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9493 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9494 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9495 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9496 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9497 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9498 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9499 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9500 | |
9501 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9502 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9503 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9504 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9505 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9506 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9507 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9508 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9509 | |
7f13657d XW |
9510 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9511 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9512 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9513 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9514 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9515 | |
9516 | if (copy_to_user(params, p, sizeof(*p))) { | |
9517 | ret = -EFAULT; | |
9518 | goto err; | |
9519 | } | |
d1719f70 | 9520 | |
9faadcc8 PB |
9521 | file = io_uring_get_file(ctx); |
9522 | if (IS_ERR(file)) { | |
9523 | ret = PTR_ERR(file); | |
9524 | goto err; | |
9525 | } | |
9526 | ||
044c1ab3 JA |
9527 | /* |
9528 | * Install ring fd as the very last thing, so we don't risk someone | |
9529 | * having closed it before we finish setup | |
9530 | */ | |
9faadcc8 PB |
9531 | ret = io_uring_install_fd(ctx, file); |
9532 | if (ret < 0) { | |
9533 | /* fput will clean it up */ | |
9534 | fput(file); | |
9535 | return ret; | |
9536 | } | |
044c1ab3 | 9537 | |
c826bd7a | 9538 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9539 | return ret; |
9540 | err: | |
9541 | io_ring_ctx_wait_and_kill(ctx); | |
9542 | return ret; | |
9543 | } | |
9544 | ||
9545 | /* | |
9546 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9547 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9548 | * params structure passed in. | |
9549 | */ | |
9550 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9551 | { | |
9552 | struct io_uring_params p; | |
2b188cc1 JA |
9553 | int i; |
9554 | ||
9555 | if (copy_from_user(&p, params, sizeof(p))) | |
9556 | return -EFAULT; | |
9557 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9558 | if (p.resv[i]) | |
9559 | return -EINVAL; | |
9560 | } | |
9561 | ||
6c271ce2 | 9562 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9563 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9564 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9565 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9566 | return -EINVAL; |
9567 | ||
7f13657d | 9568 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9569 | } |
9570 | ||
9571 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9572 | struct io_uring_params __user *, params) | |
9573 | { | |
9574 | return io_uring_setup(entries, params); | |
9575 | } | |
9576 | ||
66f4af93 JA |
9577 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9578 | { | |
9579 | struct io_uring_probe *p; | |
9580 | size_t size; | |
9581 | int i, ret; | |
9582 | ||
9583 | size = struct_size(p, ops, nr_args); | |
9584 | if (size == SIZE_MAX) | |
9585 | return -EOVERFLOW; | |
9586 | p = kzalloc(size, GFP_KERNEL); | |
9587 | if (!p) | |
9588 | return -ENOMEM; | |
9589 | ||
9590 | ret = -EFAULT; | |
9591 | if (copy_from_user(p, arg, size)) | |
9592 | goto out; | |
9593 | ret = -EINVAL; | |
9594 | if (memchr_inv(p, 0, size)) | |
9595 | goto out; | |
9596 | ||
9597 | p->last_op = IORING_OP_LAST - 1; | |
9598 | if (nr_args > IORING_OP_LAST) | |
9599 | nr_args = IORING_OP_LAST; | |
9600 | ||
9601 | for (i = 0; i < nr_args; i++) { | |
9602 | p->ops[i].op = i; | |
9603 | if (!io_op_defs[i].not_supported) | |
9604 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9605 | } | |
9606 | p->ops_len = i; | |
9607 | ||
9608 | ret = 0; | |
9609 | if (copy_to_user(arg, p, size)) | |
9610 | ret = -EFAULT; | |
9611 | out: | |
9612 | kfree(p); | |
9613 | return ret; | |
9614 | } | |
9615 | ||
071698e1 JA |
9616 | static int io_register_personality(struct io_ring_ctx *ctx) |
9617 | { | |
4379bf8b | 9618 | const struct cred *creds; |
61cf9370 | 9619 | u32 id; |
1e6fa521 | 9620 | int ret; |
071698e1 | 9621 | |
4379bf8b | 9622 | creds = get_current_cred(); |
1e6fa521 | 9623 | |
61cf9370 MWO |
9624 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9625 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9626 | if (!ret) | |
9627 | return id; | |
9628 | put_cred(creds); | |
1e6fa521 | 9629 | return ret; |
071698e1 JA |
9630 | } |
9631 | ||
21b55dbc SG |
9632 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9633 | unsigned int nr_args) | |
9634 | { | |
9635 | struct io_uring_restriction *res; | |
9636 | size_t size; | |
9637 | int i, ret; | |
9638 | ||
7e84e1c7 SG |
9639 | /* Restrictions allowed only if rings started disabled */ |
9640 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9641 | return -EBADFD; | |
9642 | ||
21b55dbc | 9643 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9644 | if (ctx->restrictions.registered) |
21b55dbc SG |
9645 | return -EBUSY; |
9646 | ||
9647 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9648 | return -EINVAL; | |
9649 | ||
9650 | size = array_size(nr_args, sizeof(*res)); | |
9651 | if (size == SIZE_MAX) | |
9652 | return -EOVERFLOW; | |
9653 | ||
9654 | res = memdup_user(arg, size); | |
9655 | if (IS_ERR(res)) | |
9656 | return PTR_ERR(res); | |
9657 | ||
9658 | ret = 0; | |
9659 | ||
9660 | for (i = 0; i < nr_args; i++) { | |
9661 | switch (res[i].opcode) { | |
9662 | case IORING_RESTRICTION_REGISTER_OP: | |
9663 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9664 | ret = -EINVAL; | |
9665 | goto out; | |
9666 | } | |
9667 | ||
9668 | __set_bit(res[i].register_op, | |
9669 | ctx->restrictions.register_op); | |
9670 | break; | |
9671 | case IORING_RESTRICTION_SQE_OP: | |
9672 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9673 | ret = -EINVAL; | |
9674 | goto out; | |
9675 | } | |
9676 | ||
9677 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9678 | break; | |
9679 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9680 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9681 | break; | |
9682 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9683 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9684 | break; | |
9685 | default: | |
9686 | ret = -EINVAL; | |
9687 | goto out; | |
9688 | } | |
9689 | } | |
9690 | ||
9691 | out: | |
9692 | /* Reset all restrictions if an error happened */ | |
9693 | if (ret != 0) | |
9694 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9695 | else | |
7e84e1c7 | 9696 | ctx->restrictions.registered = true; |
21b55dbc SG |
9697 | |
9698 | kfree(res); | |
9699 | return ret; | |
9700 | } | |
9701 | ||
7e84e1c7 SG |
9702 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9703 | { | |
9704 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9705 | return -EBADFD; | |
9706 | ||
9707 | if (ctx->restrictions.registered) | |
9708 | ctx->restricted = 1; | |
9709 | ||
0298ef96 PB |
9710 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9711 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9712 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9713 | return 0; |
9714 | } | |
9715 | ||
071698e1 JA |
9716 | static bool io_register_op_must_quiesce(int op) |
9717 | { | |
9718 | switch (op) { | |
9719 | case IORING_UNREGISTER_FILES: | |
9720 | case IORING_REGISTER_FILES_UPDATE: | |
9721 | case IORING_REGISTER_PROBE: | |
9722 | case IORING_REGISTER_PERSONALITY: | |
9723 | case IORING_UNREGISTER_PERSONALITY: | |
9724 | return false; | |
9725 | default: | |
9726 | return true; | |
9727 | } | |
9728 | } | |
9729 | ||
edafccee JA |
9730 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9731 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9732 | __releases(ctx->uring_lock) |
9733 | __acquires(ctx->uring_lock) | |
edafccee JA |
9734 | { |
9735 | int ret; | |
9736 | ||
35fa71a0 JA |
9737 | /* |
9738 | * We're inside the ring mutex, if the ref is already dying, then | |
9739 | * someone else killed the ctx or is already going through | |
9740 | * io_uring_register(). | |
9741 | */ | |
9742 | if (percpu_ref_is_dying(&ctx->refs)) | |
9743 | return -ENXIO; | |
9744 | ||
071698e1 | 9745 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9746 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9747 | |
05f3fb3c JA |
9748 | /* |
9749 | * Drop uring mutex before waiting for references to exit. If | |
9750 | * another thread is currently inside io_uring_enter() it might | |
9751 | * need to grab the uring_lock to make progress. If we hold it | |
9752 | * here across the drain wait, then we can deadlock. It's safe | |
9753 | * to drop the mutex here, since no new references will come in | |
9754 | * after we've killed the percpu ref. | |
9755 | */ | |
9756 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9757 | do { |
9758 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9759 | if (!ret) | |
9760 | break; | |
ed6930c9 JA |
9761 | ret = io_run_task_work_sig(); |
9762 | if (ret < 0) | |
9763 | break; | |
af9c1a44 JA |
9764 | } while (1); |
9765 | ||
05f3fb3c | 9766 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9767 | |
c150368b JA |
9768 | if (ret) { |
9769 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9770 | goto out_quiesce; |
9771 | } | |
9772 | } | |
9773 | ||
9774 | if (ctx->restricted) { | |
9775 | if (opcode >= IORING_REGISTER_LAST) { | |
9776 | ret = -EINVAL; | |
9777 | goto out; | |
9778 | } | |
9779 | ||
9780 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9781 | ret = -EACCES; | |
c150368b JA |
9782 | goto out; |
9783 | } | |
05f3fb3c | 9784 | } |
edafccee JA |
9785 | |
9786 | switch (opcode) { | |
9787 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9788 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9789 | break; |
9790 | case IORING_UNREGISTER_BUFFERS: | |
9791 | ret = -EINVAL; | |
9792 | if (arg || nr_args) | |
9793 | break; | |
0a96bbe4 | 9794 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9795 | break; |
6b06314c JA |
9796 | case IORING_REGISTER_FILES: |
9797 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9798 | break; | |
9799 | case IORING_UNREGISTER_FILES: | |
9800 | ret = -EINVAL; | |
9801 | if (arg || nr_args) | |
9802 | break; | |
9803 | ret = io_sqe_files_unregister(ctx); | |
9804 | break; | |
c3a31e60 JA |
9805 | case IORING_REGISTER_FILES_UPDATE: |
9806 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9807 | break; | |
9b402849 | 9808 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9809 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9810 | ret = -EINVAL; |
9811 | if (nr_args != 1) | |
9812 | break; | |
9813 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9814 | if (ret) |
9815 | break; | |
9816 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9817 | ctx->eventfd_async = 1; | |
9818 | else | |
9819 | ctx->eventfd_async = 0; | |
9b402849 JA |
9820 | break; |
9821 | case IORING_UNREGISTER_EVENTFD: | |
9822 | ret = -EINVAL; | |
9823 | if (arg || nr_args) | |
9824 | break; | |
9825 | ret = io_eventfd_unregister(ctx); | |
9826 | break; | |
66f4af93 JA |
9827 | case IORING_REGISTER_PROBE: |
9828 | ret = -EINVAL; | |
9829 | if (!arg || nr_args > 256) | |
9830 | break; | |
9831 | ret = io_probe(ctx, arg, nr_args); | |
9832 | break; | |
071698e1 JA |
9833 | case IORING_REGISTER_PERSONALITY: |
9834 | ret = -EINVAL; | |
9835 | if (arg || nr_args) | |
9836 | break; | |
9837 | ret = io_register_personality(ctx); | |
9838 | break; | |
9839 | case IORING_UNREGISTER_PERSONALITY: | |
9840 | ret = -EINVAL; | |
9841 | if (arg) | |
9842 | break; | |
9843 | ret = io_unregister_personality(ctx, nr_args); | |
9844 | break; | |
7e84e1c7 SG |
9845 | case IORING_REGISTER_ENABLE_RINGS: |
9846 | ret = -EINVAL; | |
9847 | if (arg || nr_args) | |
9848 | break; | |
9849 | ret = io_register_enable_rings(ctx); | |
9850 | break; | |
21b55dbc SG |
9851 | case IORING_REGISTER_RESTRICTIONS: |
9852 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9853 | break; | |
edafccee JA |
9854 | default: |
9855 | ret = -EINVAL; | |
9856 | break; | |
9857 | } | |
9858 | ||
21b55dbc | 9859 | out: |
071698e1 | 9860 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9861 | /* bring the ctx back to life */ |
05f3fb3c | 9862 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9863 | out_quiesce: |
0f158b4c | 9864 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9865 | } |
edafccee JA |
9866 | return ret; |
9867 | } | |
9868 | ||
9869 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9870 | void __user *, arg, unsigned int, nr_args) | |
9871 | { | |
9872 | struct io_ring_ctx *ctx; | |
9873 | long ret = -EBADF; | |
9874 | struct fd f; | |
9875 | ||
9876 | f = fdget(fd); | |
9877 | if (!f.file) | |
9878 | return -EBADF; | |
9879 | ||
9880 | ret = -EOPNOTSUPP; | |
9881 | if (f.file->f_op != &io_uring_fops) | |
9882 | goto out_fput; | |
9883 | ||
9884 | ctx = f.file->private_data; | |
9885 | ||
b6c23dd5 PB |
9886 | io_run_task_work(); |
9887 | ||
edafccee JA |
9888 | mutex_lock(&ctx->uring_lock); |
9889 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9890 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9891 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9892 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9893 | out_fput: |
9894 | fdput(f); | |
9895 | return ret; | |
9896 | } | |
9897 | ||
2b188cc1 JA |
9898 | static int __init io_uring_init(void) |
9899 | { | |
d7f62e82 SM |
9900 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9901 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9902 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9903 | } while (0) | |
9904 | ||
9905 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9906 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9907 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9908 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9909 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9910 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9911 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9912 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9913 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9914 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9915 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9916 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9917 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9918 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9919 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9920 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9921 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9922 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9923 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9924 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9925 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9926 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9927 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9928 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9929 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9930 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9931 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9932 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9933 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9934 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9935 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9936 | |
d3656344 | 9937 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9938 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9939 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9940 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9941 | return 0; |
9942 | }; | |
9943 | __initcall(io_uring_init); |