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 | ||
53e043b2 SM |
459 | struct io_uring_task { |
460 | /* submission side */ | |
461 | struct xarray xa; | |
462 | struct wait_queue_head wait; | |
ee53fb2b SM |
463 | const struct io_ring_ctx *last; |
464 | struct io_wq *io_wq; | |
53e043b2 SM |
465 | struct percpu_counter inflight; |
466 | atomic_t in_idle; | |
467 | bool sqpoll; | |
468 | ||
469 | spinlock_t task_lock; | |
470 | struct io_wq_work_list task_list; | |
471 | unsigned long task_state; | |
472 | struct callback_head task_work; | |
473 | }; | |
474 | ||
09bb8394 JA |
475 | /* |
476 | * First field must be the file pointer in all the | |
477 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
478 | */ | |
221c5eb2 JA |
479 | struct io_poll_iocb { |
480 | struct file *file; | |
018043be | 481 | struct wait_queue_head *head; |
221c5eb2 | 482 | __poll_t events; |
8c838788 | 483 | bool done; |
221c5eb2 | 484 | bool canceled; |
392edb45 | 485 | struct wait_queue_entry wait; |
221c5eb2 JA |
486 | }; |
487 | ||
018043be PB |
488 | struct io_poll_remove { |
489 | struct file *file; | |
490 | u64 addr; | |
491 | }; | |
492 | ||
b5dba59e JA |
493 | struct io_close { |
494 | struct file *file; | |
b5dba59e JA |
495 | int fd; |
496 | }; | |
497 | ||
ad8a48ac JA |
498 | struct io_timeout_data { |
499 | struct io_kiocb *req; | |
500 | struct hrtimer timer; | |
501 | struct timespec64 ts; | |
502 | enum hrtimer_mode mode; | |
503 | }; | |
504 | ||
8ed8d3c3 JA |
505 | struct io_accept { |
506 | struct file *file; | |
507 | struct sockaddr __user *addr; | |
508 | int __user *addr_len; | |
509 | int flags; | |
09952e3e | 510 | unsigned long nofile; |
8ed8d3c3 JA |
511 | }; |
512 | ||
513 | struct io_sync { | |
514 | struct file *file; | |
515 | loff_t len; | |
516 | loff_t off; | |
517 | int flags; | |
d63d1b5e | 518 | int mode; |
8ed8d3c3 JA |
519 | }; |
520 | ||
fbf23849 JA |
521 | struct io_cancel { |
522 | struct file *file; | |
523 | u64 addr; | |
524 | }; | |
525 | ||
b29472ee JA |
526 | struct io_timeout { |
527 | struct file *file; | |
bfe68a22 PB |
528 | u32 off; |
529 | u32 target_seq; | |
135fcde8 | 530 | struct list_head list; |
90cd7e42 PB |
531 | /* head of the link, used by linked timeouts only */ |
532 | struct io_kiocb *head; | |
b29472ee JA |
533 | }; |
534 | ||
0bdf7a2d PB |
535 | struct io_timeout_rem { |
536 | struct file *file; | |
537 | u64 addr; | |
9c8e11b3 PB |
538 | |
539 | /* timeout update */ | |
540 | struct timespec64 ts; | |
541 | u32 flags; | |
0bdf7a2d PB |
542 | }; |
543 | ||
9adbd45d JA |
544 | struct io_rw { |
545 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
546 | struct kiocb kiocb; | |
547 | u64 addr; | |
548 | u64 len; | |
549 | }; | |
550 | ||
3fbb51c1 JA |
551 | struct io_connect { |
552 | struct file *file; | |
553 | struct sockaddr __user *addr; | |
554 | int addr_len; | |
555 | }; | |
556 | ||
e47293fd JA |
557 | struct io_sr_msg { |
558 | struct file *file; | |
fddaface | 559 | union { |
270a5940 | 560 | struct user_msghdr __user *umsg; |
fddaface JA |
561 | void __user *buf; |
562 | }; | |
e47293fd | 563 | int msg_flags; |
bcda7baa | 564 | int bgid; |
fddaface | 565 | size_t len; |
bcda7baa | 566 | struct io_buffer *kbuf; |
e47293fd JA |
567 | }; |
568 | ||
15b71abe JA |
569 | struct io_open { |
570 | struct file *file; | |
571 | int dfd; | |
15b71abe | 572 | struct filename *filename; |
c12cedf2 | 573 | struct open_how how; |
4022e7af | 574 | unsigned long nofile; |
15b71abe JA |
575 | }; |
576 | ||
269bbe5f | 577 | struct io_rsrc_update { |
05f3fb3c JA |
578 | struct file *file; |
579 | u64 arg; | |
580 | u32 nr_args; | |
581 | u32 offset; | |
582 | }; | |
583 | ||
4840e418 JA |
584 | struct io_fadvise { |
585 | struct file *file; | |
586 | u64 offset; | |
587 | u32 len; | |
588 | u32 advice; | |
589 | }; | |
590 | ||
c1ca757b JA |
591 | struct io_madvise { |
592 | struct file *file; | |
593 | u64 addr; | |
594 | u32 len; | |
595 | u32 advice; | |
596 | }; | |
597 | ||
3e4827b0 JA |
598 | struct io_epoll { |
599 | struct file *file; | |
600 | int epfd; | |
601 | int op; | |
602 | int fd; | |
603 | struct epoll_event event; | |
e47293fd JA |
604 | }; |
605 | ||
7d67af2c PB |
606 | struct io_splice { |
607 | struct file *file_out; | |
608 | struct file *file_in; | |
609 | loff_t off_out; | |
610 | loff_t off_in; | |
611 | u64 len; | |
612 | unsigned int flags; | |
613 | }; | |
614 | ||
ddf0322d JA |
615 | struct io_provide_buf { |
616 | struct file *file; | |
617 | __u64 addr; | |
618 | __s32 len; | |
619 | __u32 bgid; | |
620 | __u16 nbufs; | |
621 | __u16 bid; | |
622 | }; | |
623 | ||
1d9e1288 BM |
624 | struct io_statx { |
625 | struct file *file; | |
626 | int dfd; | |
627 | unsigned int mask; | |
628 | unsigned int flags; | |
e62753e4 | 629 | const char __user *filename; |
1d9e1288 BM |
630 | struct statx __user *buffer; |
631 | }; | |
632 | ||
36f4fa68 JA |
633 | struct io_shutdown { |
634 | struct file *file; | |
635 | int how; | |
636 | }; | |
637 | ||
80a261fd JA |
638 | struct io_rename { |
639 | struct file *file; | |
640 | int old_dfd; | |
641 | int new_dfd; | |
642 | struct filename *oldpath; | |
643 | struct filename *newpath; | |
644 | int flags; | |
645 | }; | |
646 | ||
14a1143b JA |
647 | struct io_unlink { |
648 | struct file *file; | |
649 | int dfd; | |
650 | int flags; | |
651 | struct filename *filename; | |
652 | }; | |
653 | ||
3ca405eb PB |
654 | struct io_completion { |
655 | struct file *file; | |
656 | struct list_head list; | |
0f7e466b | 657 | int cflags; |
3ca405eb PB |
658 | }; |
659 | ||
f499a021 JA |
660 | struct io_async_connect { |
661 | struct sockaddr_storage address; | |
662 | }; | |
663 | ||
03b1230c JA |
664 | struct io_async_msghdr { |
665 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
666 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
667 | struct iovec *free_iov; | |
03b1230c JA |
668 | struct sockaddr __user *uaddr; |
669 | struct msghdr msg; | |
b537916c | 670 | struct sockaddr_storage addr; |
03b1230c JA |
671 | }; |
672 | ||
f67676d1 JA |
673 | struct io_async_rw { |
674 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
675 | const struct iovec *free_iovec; |
676 | struct iov_iter iter; | |
227c0c96 | 677 | size_t bytes_done; |
bcf5a063 | 678 | struct wait_page_queue wpq; |
f67676d1 JA |
679 | }; |
680 | ||
6b47ee6e PB |
681 | enum { |
682 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
683 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
684 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
685 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
686 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 687 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 688 | |
6b47ee6e PB |
689 | REQ_F_FAIL_LINK_BIT, |
690 | REQ_F_INFLIGHT_BIT, | |
691 | REQ_F_CUR_POS_BIT, | |
692 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 693 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 694 | REQ_F_ISREG_BIT, |
99bc4c38 | 695 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 696 | REQ_F_POLLED_BIT, |
bcda7baa | 697 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 698 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 699 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 700 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
701 | |
702 | /* not a real bit, just to check we're not overflowing the space */ | |
703 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
704 | }; |
705 | ||
706 | enum { | |
707 | /* ctx owns file */ | |
708 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
709 | /* drain existing IO first */ | |
710 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
711 | /* linked sqes */ | |
712 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
713 | /* doesn't sever on completion < 0 */ | |
714 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
715 | /* IOSQE_ASYNC */ | |
716 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
717 | /* IOSQE_BUFFER_SELECT */ |
718 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 719 | |
6b47ee6e PB |
720 | /* fail rest of links */ |
721 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 722 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
723 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
724 | /* read/write uses file position */ | |
725 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
726 | /* must not punt to workers */ | |
727 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 728 | /* has or had linked timeout */ |
6b47ee6e | 729 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
730 | /* regular file */ |
731 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
732 | /* needs cleanup */ |
733 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
734 | /* already went through poll handler */ |
735 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
736 | /* buffer already selected */ |
737 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
738 | /* doesn't need file table for this request */ |
739 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
740 | /* linked timeout is active, i.e. prepared by link's head */ |
741 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
742 | /* completion is deferred through io_comp_state */ |
743 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
744 | }; |
745 | ||
746 | struct async_poll { | |
747 | struct io_poll_iocb poll; | |
807abcb0 | 748 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
749 | }; |
750 | ||
7cbf1722 JA |
751 | struct io_task_work { |
752 | struct io_wq_work_node node; | |
753 | task_work_func_t func; | |
754 | }; | |
755 | ||
09bb8394 JA |
756 | /* |
757 | * NOTE! Each of the iocb union members has the file pointer | |
758 | * as the first entry in their struct definition. So you can | |
759 | * access the file pointer through any of the sub-structs, | |
760 | * or directly as just 'ki_filp' in this struct. | |
761 | */ | |
2b188cc1 | 762 | struct io_kiocb { |
221c5eb2 | 763 | union { |
09bb8394 | 764 | struct file *file; |
9adbd45d | 765 | struct io_rw rw; |
221c5eb2 | 766 | struct io_poll_iocb poll; |
018043be | 767 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
768 | struct io_accept accept; |
769 | struct io_sync sync; | |
fbf23849 | 770 | struct io_cancel cancel; |
b29472ee | 771 | struct io_timeout timeout; |
0bdf7a2d | 772 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 773 | struct io_connect connect; |
e47293fd | 774 | struct io_sr_msg sr_msg; |
15b71abe | 775 | struct io_open open; |
b5dba59e | 776 | struct io_close close; |
269bbe5f | 777 | struct io_rsrc_update rsrc_update; |
4840e418 | 778 | struct io_fadvise fadvise; |
c1ca757b | 779 | struct io_madvise madvise; |
3e4827b0 | 780 | struct io_epoll epoll; |
7d67af2c | 781 | struct io_splice splice; |
ddf0322d | 782 | struct io_provide_buf pbuf; |
1d9e1288 | 783 | struct io_statx statx; |
36f4fa68 | 784 | struct io_shutdown shutdown; |
80a261fd | 785 | struct io_rename rename; |
14a1143b | 786 | struct io_unlink unlink; |
3ca405eb PB |
787 | /* use only after cleaning per-op data, see io_clean_op() */ |
788 | struct io_completion compl; | |
221c5eb2 | 789 | }; |
2b188cc1 | 790 | |
e8c2bc1f JA |
791 | /* opcode allocated if it needs to store data for async defer */ |
792 | void *async_data; | |
d625c6ee | 793 | u8 opcode; |
65a6543d XW |
794 | /* polled IO has completed */ |
795 | u8 iopoll_completed; | |
2b188cc1 | 796 | |
4f4eeba8 | 797 | u16 buf_index; |
9cf7c104 | 798 | u32 result; |
4f4eeba8 | 799 | |
010e8e6b PB |
800 | struct io_ring_ctx *ctx; |
801 | unsigned int flags; | |
802 | refcount_t refs; | |
803 | struct task_struct *task; | |
804 | u64 user_data; | |
d7718a9d | 805 | |
f2f87370 | 806 | struct io_kiocb *link; |
269bbe5f | 807 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 808 | |
d21ffe7e PB |
809 | /* |
810 | * 1. used with ctx->iopoll_list with reads/writes | |
811 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
812 | */ | |
010e8e6b | 813 | struct list_head inflight_entry; |
7cbf1722 JA |
814 | union { |
815 | struct io_task_work io_task_work; | |
816 | struct callback_head task_work; | |
817 | }; | |
010e8e6b PB |
818 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
819 | struct hlist_node hash_node; | |
820 | struct async_poll *apoll; | |
821 | struct io_wq_work work; | |
2b188cc1 | 822 | }; |
05589553 | 823 | |
13bf43f5 PB |
824 | struct io_tctx_node { |
825 | struct list_head ctx_node; | |
826 | struct task_struct *task; | |
13bf43f5 PB |
827 | struct io_ring_ctx *ctx; |
828 | }; | |
829 | ||
27dc8338 PB |
830 | struct io_defer_entry { |
831 | struct list_head list; | |
832 | struct io_kiocb *req; | |
9cf7c104 | 833 | u32 seq; |
2b188cc1 JA |
834 | }; |
835 | ||
d3656344 | 836 | struct io_op_def { |
d3656344 JA |
837 | /* needs req->file assigned */ |
838 | unsigned needs_file : 1; | |
d3656344 JA |
839 | /* hash wq insertion if file is a regular file */ |
840 | unsigned hash_reg_file : 1; | |
841 | /* unbound wq insertion if file is a non-regular file */ | |
842 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
843 | /* opcode is not supported by this kernel */ |
844 | unsigned not_supported : 1; | |
8a72758c JA |
845 | /* set if opcode supports polled "wait" */ |
846 | unsigned pollin : 1; | |
847 | unsigned pollout : 1; | |
bcda7baa JA |
848 | /* op supports buffer selection */ |
849 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
850 | /* must always have async data allocated */ |
851 | unsigned needs_async_data : 1; | |
27926b68 JA |
852 | /* should block plug */ |
853 | unsigned plug : 1; | |
e8c2bc1f JA |
854 | /* size of async data needed, if any */ |
855 | unsigned short async_size; | |
d3656344 JA |
856 | }; |
857 | ||
0918682b | 858 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
859 | [IORING_OP_NOP] = {}, |
860 | [IORING_OP_READV] = { | |
d3656344 JA |
861 | .needs_file = 1, |
862 | .unbound_nonreg_file = 1, | |
8a72758c | 863 | .pollin = 1, |
4d954c25 | 864 | .buffer_select = 1, |
e8c2bc1f | 865 | .needs_async_data = 1, |
27926b68 | 866 | .plug = 1, |
e8c2bc1f | 867 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 868 | }, |
0463b6c5 | 869 | [IORING_OP_WRITEV] = { |
d3656344 JA |
870 | .needs_file = 1, |
871 | .hash_reg_file = 1, | |
872 | .unbound_nonreg_file = 1, | |
8a72758c | 873 | .pollout = 1, |
e8c2bc1f | 874 | .needs_async_data = 1, |
27926b68 | 875 | .plug = 1, |
e8c2bc1f | 876 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_FSYNC] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | }, | |
0463b6c5 | 881 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
882 | .needs_file = 1, |
883 | .unbound_nonreg_file = 1, | |
8a72758c | 884 | .pollin = 1, |
27926b68 | 885 | .plug = 1, |
e8c2bc1f | 886 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 887 | }, |
0463b6c5 | 888 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
889 | .needs_file = 1, |
890 | .hash_reg_file = 1, | |
891 | .unbound_nonreg_file = 1, | |
8a72758c | 892 | .pollout = 1, |
27926b68 | 893 | .plug = 1, |
e8c2bc1f | 894 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 895 | }, |
0463b6c5 | 896 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
897 | .needs_file = 1, |
898 | .unbound_nonreg_file = 1, | |
899 | }, | |
0463b6c5 PB |
900 | [IORING_OP_POLL_REMOVE] = {}, |
901 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
902 | .needs_file = 1, |
903 | }, | |
0463b6c5 | 904 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
905 | .needs_file = 1, |
906 | .unbound_nonreg_file = 1, | |
8a72758c | 907 | .pollout = 1, |
e8c2bc1f JA |
908 | .needs_async_data = 1, |
909 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 910 | }, |
0463b6c5 | 911 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
912 | .needs_file = 1, |
913 | .unbound_nonreg_file = 1, | |
8a72758c | 914 | .pollin = 1, |
52de1fe1 | 915 | .buffer_select = 1, |
e8c2bc1f JA |
916 | .needs_async_data = 1, |
917 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 918 | }, |
0463b6c5 | 919 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
920 | .needs_async_data = 1, |
921 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 922 | }, |
9c8e11b3 PB |
923 | [IORING_OP_TIMEOUT_REMOVE] = { |
924 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 925 | }, |
0463b6c5 | 926 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
927 | .needs_file = 1, |
928 | .unbound_nonreg_file = 1, | |
8a72758c | 929 | .pollin = 1, |
d3656344 | 930 | }, |
0463b6c5 PB |
931 | [IORING_OP_ASYNC_CANCEL] = {}, |
932 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
933 | .needs_async_data = 1, |
934 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 935 | }, |
0463b6c5 | 936 | [IORING_OP_CONNECT] = { |
d3656344 JA |
937 | .needs_file = 1, |
938 | .unbound_nonreg_file = 1, | |
8a72758c | 939 | .pollout = 1, |
e8c2bc1f JA |
940 | .needs_async_data = 1, |
941 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_FALLOCATE] = { |
d3656344 | 944 | .needs_file = 1, |
d3656344 | 945 | }, |
44526bed JA |
946 | [IORING_OP_OPENAT] = {}, |
947 | [IORING_OP_CLOSE] = {}, | |
948 | [IORING_OP_FILES_UPDATE] = {}, | |
949 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 950 | [IORING_OP_READ] = { |
3a6820f2 JA |
951 | .needs_file = 1, |
952 | .unbound_nonreg_file = 1, | |
8a72758c | 953 | .pollin = 1, |
bcda7baa | 954 | .buffer_select = 1, |
27926b68 | 955 | .plug = 1, |
e8c2bc1f | 956 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 957 | }, |
0463b6c5 | 958 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
959 | .needs_file = 1, |
960 | .unbound_nonreg_file = 1, | |
8a72758c | 961 | .pollout = 1, |
27926b68 | 962 | .plug = 1, |
e8c2bc1f | 963 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 964 | }, |
0463b6c5 | 965 | [IORING_OP_FADVISE] = { |
4840e418 | 966 | .needs_file = 1, |
c1ca757b | 967 | }, |
44526bed | 968 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 969 | [IORING_OP_SEND] = { |
fddaface JA |
970 | .needs_file = 1, |
971 | .unbound_nonreg_file = 1, | |
8a72758c | 972 | .pollout = 1, |
fddaface | 973 | }, |
0463b6c5 | 974 | [IORING_OP_RECV] = { |
fddaface JA |
975 | .needs_file = 1, |
976 | .unbound_nonreg_file = 1, | |
8a72758c | 977 | .pollin = 1, |
bcda7baa | 978 | .buffer_select = 1, |
fddaface | 979 | }, |
0463b6c5 | 980 | [IORING_OP_OPENAT2] = { |
cebdb986 | 981 | }, |
3e4827b0 JA |
982 | [IORING_OP_EPOLL_CTL] = { |
983 | .unbound_nonreg_file = 1, | |
3e4827b0 | 984 | }, |
7d67af2c PB |
985 | [IORING_OP_SPLICE] = { |
986 | .needs_file = 1, | |
987 | .hash_reg_file = 1, | |
988 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
989 | }, |
990 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 991 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
992 | [IORING_OP_TEE] = { |
993 | .needs_file = 1, | |
994 | .hash_reg_file = 1, | |
995 | .unbound_nonreg_file = 1, | |
996 | }, | |
36f4fa68 JA |
997 | [IORING_OP_SHUTDOWN] = { |
998 | .needs_file = 1, | |
999 | }, | |
44526bed JA |
1000 | [IORING_OP_RENAMEAT] = {}, |
1001 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1002 | }; |
1003 | ||
7a612350 | 1004 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1005 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1006 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1007 | struct task_struct *task, | |
1008 | struct files_struct *files); | |
37d1e2e3 | 1009 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 1010 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1011 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1012 | struct io_ring_ctx *ctx); |
f2303b1f | 1013 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 1014 | |
23faba36 | 1015 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1016 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1017 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1018 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1019 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1020 | static void io_dismantle_req(struct io_kiocb *req); |
1021 | static void io_put_task(struct task_struct *task, int nr); | |
1022 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1023 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1024 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1025 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1026 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1027 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1028 | unsigned nr_args); |
3ca405eb | 1029 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1030 | static struct file *io_file_get(struct io_submit_state *state, |
1031 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1032 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1033 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1034 | |
847595de PB |
1035 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1036 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1037 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1038 | const struct iovec *fast_iov, | |
227c0c96 | 1039 | struct iov_iter *iter, bool force); |
907d1df3 | 1040 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1041 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1042 | struct io_ring_ctx *ctx); | |
de0617e4 | 1043 | |
2b188cc1 JA |
1044 | static struct kmem_cache *req_cachep; |
1045 | ||
0918682b | 1046 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1047 | |
1048 | struct sock *io_uring_get_socket(struct file *file) | |
1049 | { | |
1050 | #if defined(CONFIG_UNIX) | |
1051 | if (file->f_op == &io_uring_fops) { | |
1052 | struct io_ring_ctx *ctx = file->private_data; | |
1053 | ||
1054 | return ctx->ring_sock->sk; | |
1055 | } | |
1056 | #endif | |
1057 | return NULL; | |
1058 | } | |
1059 | EXPORT_SYMBOL(io_uring_get_socket); | |
1060 | ||
f2f87370 PB |
1061 | #define io_for_each_link(pos, head) \ |
1062 | for (pos = (head); pos; pos = pos->link) | |
1063 | ||
3ca405eb PB |
1064 | static inline void io_clean_op(struct io_kiocb *req) |
1065 | { | |
9d5c8190 | 1066 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1067 | __io_clean_op(req); |
1068 | } | |
1069 | ||
36f72fe2 PB |
1070 | static inline void io_set_resource_node(struct io_kiocb *req) |
1071 | { | |
1072 | struct io_ring_ctx *ctx = req->ctx; | |
1073 | ||
269bbe5f BM |
1074 | if (!req->fixed_rsrc_refs) { |
1075 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1076 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1077 | } |
1078 | } | |
1079 | ||
08d23634 PB |
1080 | static bool io_match_task(struct io_kiocb *head, |
1081 | struct task_struct *task, | |
1082 | struct files_struct *files) | |
1083 | { | |
1084 | struct io_kiocb *req; | |
1085 | ||
84965ff8 JA |
1086 | if (task && head->task != task) { |
1087 | /* in terms of cancelation, always match if req task is dead */ | |
1088 | if (head->task->flags & PF_EXITING) | |
1089 | return true; | |
08d23634 | 1090 | return false; |
84965ff8 | 1091 | } |
08d23634 PB |
1092 | if (!files) |
1093 | return true; | |
1094 | ||
1095 | io_for_each_link(req, head) { | |
b05a1bcd | 1096 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1097 | return true; |
4379bf8b | 1098 | if (req->task->files == files) |
08d23634 PB |
1099 | return true; |
1100 | } | |
1101 | return false; | |
1102 | } | |
1103 | ||
c40f6379 JA |
1104 | static inline void req_set_fail_links(struct io_kiocb *req) |
1105 | { | |
1106 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1107 | req->flags |= REQ_F_FAIL_LINK; | |
1108 | } | |
4a38aed2 | 1109 | |
2b188cc1 JA |
1110 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1111 | { | |
1112 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1113 | ||
0f158b4c | 1114 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1115 | } |
1116 | ||
8eb7e2d0 PB |
1117 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1118 | { | |
1119 | return !req->timeout.off; | |
1120 | } | |
1121 | ||
2b188cc1 JA |
1122 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1123 | { | |
1124 | struct io_ring_ctx *ctx; | |
78076bb6 | 1125 | int hash_bits; |
2b188cc1 JA |
1126 | |
1127 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1128 | if (!ctx) | |
1129 | return NULL; | |
1130 | ||
78076bb6 JA |
1131 | /* |
1132 | * Use 5 bits less than the max cq entries, that should give us around | |
1133 | * 32 entries per hash list if totally full and uniformly spread. | |
1134 | */ | |
1135 | hash_bits = ilog2(p->cq_entries); | |
1136 | hash_bits -= 5; | |
1137 | if (hash_bits <= 0) | |
1138 | hash_bits = 1; | |
1139 | ctx->cancel_hash_bits = hash_bits; | |
1140 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1141 | GFP_KERNEL); | |
1142 | if (!ctx->cancel_hash) | |
1143 | goto err; | |
1144 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1145 | ||
21482896 | 1146 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1147 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1148 | goto err; | |
2b188cc1 JA |
1149 | |
1150 | ctx->flags = p->flags; | |
90554200 | 1151 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1152 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1153 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1154 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1155 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1156 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1157 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1158 | mutex_init(&ctx->uring_lock); |
1159 | init_waitqueue_head(&ctx->wait); | |
1160 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1161 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1162 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1163 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1164 | spin_lock_init(&ctx->inflight_lock); |
1165 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1166 | spin_lock_init(&ctx->rsrc_ref_lock); |
1167 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1168 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1169 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1170 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1171 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1172 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1173 | return ctx; |
206aefde | 1174 | err: |
78076bb6 | 1175 | kfree(ctx->cancel_hash); |
206aefde JA |
1176 | kfree(ctx); |
1177 | return NULL; | |
2b188cc1 JA |
1178 | } |
1179 | ||
9cf7c104 | 1180 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1181 | { |
2bc9930e JA |
1182 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1183 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1184 | |
9cf7c104 | 1185 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1186 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1187 | } |
de0617e4 | 1188 | |
9d858b21 | 1189 | return false; |
de0617e4 JA |
1190 | } |
1191 | ||
ce3d5aae PB |
1192 | static void io_req_track_inflight(struct io_kiocb *req) |
1193 | { | |
1194 | struct io_ring_ctx *ctx = req->ctx; | |
1195 | ||
1196 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1197 | req->flags |= REQ_F_INFLIGHT; |
1198 | ||
1199 | spin_lock_irq(&ctx->inflight_lock); | |
1200 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1201 | spin_unlock_irq(&ctx->inflight_lock); | |
1202 | } | |
1203 | } | |
1204 | ||
1e6fa521 JA |
1205 | static void io_prep_async_work(struct io_kiocb *req) |
1206 | { | |
1207 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1208 | struct io_ring_ctx *ctx = req->ctx; |
1209 | ||
003e8dcc JA |
1210 | if (!req->work.creds) |
1211 | req->work.creds = get_current_cred(); | |
1212 | ||
feaadc4f PB |
1213 | if (req->flags & REQ_F_FORCE_ASYNC) |
1214 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1215 | ||
1e6fa521 JA |
1216 | if (req->flags & REQ_F_ISREG) { |
1217 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1218 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1219 | } else { | |
1220 | if (def->unbound_nonreg_file) | |
1221 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1222 | } | |
561fb04a | 1223 | } |
cccf0ee8 | 1224 | |
cbdcb435 | 1225 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1226 | { |
cbdcb435 | 1227 | struct io_kiocb *cur; |
54a91f3b | 1228 | |
f2f87370 PB |
1229 | io_for_each_link(cur, req) |
1230 | io_prep_async_work(cur); | |
561fb04a JA |
1231 | } |
1232 | ||
ebf93667 | 1233 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1234 | { |
a197f664 | 1235 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1236 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1237 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1238 | |
3bfe6106 JA |
1239 | BUG_ON(!tctx); |
1240 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1241 | |
8766dd51 PB |
1242 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1243 | &req->work, req->flags); | |
cbdcb435 PB |
1244 | /* init ->work of the whole link before punting */ |
1245 | io_prep_async_link(req); | |
ebf93667 | 1246 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1247 | if (link) |
1248 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1249 | } |
1250 | ||
5262f567 JA |
1251 | static void io_kill_timeout(struct io_kiocb *req) |
1252 | { | |
e8c2bc1f | 1253 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1254 | int ret; |
1255 | ||
e8c2bc1f | 1256 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1257 | if (ret != -1) { |
01cec8c1 PB |
1258 | atomic_set(&req->ctx->cq_timeouts, |
1259 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1260 | list_del_init(&req->timeout.list); |
78e19bbe | 1261 | io_cqring_fill_event(req, 0); |
216578e5 | 1262 | io_put_req_deferred(req, 1); |
5262f567 JA |
1263 | } |
1264 | } | |
1265 | ||
76e1b642 JA |
1266 | /* |
1267 | * Returns true if we found and killed one or more timeouts | |
1268 | */ | |
6b81928d PB |
1269 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1270 | struct files_struct *files) | |
5262f567 JA |
1271 | { |
1272 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1273 | int canceled = 0; |
5262f567 JA |
1274 | |
1275 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1276 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1277 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1278 | io_kill_timeout(req); |
76e1b642 JA |
1279 | canceled++; |
1280 | } | |
f3606e3a | 1281 | } |
5262f567 | 1282 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1283 | return canceled != 0; |
5262f567 JA |
1284 | } |
1285 | ||
04518945 | 1286 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1287 | { |
04518945 | 1288 | do { |
27dc8338 PB |
1289 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1290 | struct io_defer_entry, list); | |
de0617e4 | 1291 | |
9cf7c104 | 1292 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1293 | break; |
27dc8338 | 1294 | list_del_init(&de->list); |
907d1df3 | 1295 | io_req_task_queue(de->req); |
27dc8338 | 1296 | kfree(de); |
04518945 PB |
1297 | } while (!list_empty(&ctx->defer_list)); |
1298 | } | |
1299 | ||
360428f8 | 1300 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1301 | { |
f010505b MDG |
1302 | u32 seq; |
1303 | ||
1304 | if (list_empty(&ctx->timeout_list)) | |
1305 | return; | |
1306 | ||
1307 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1308 | ||
1309 | do { | |
1310 | u32 events_needed, events_got; | |
360428f8 | 1311 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1312 | struct io_kiocb, timeout.list); |
de0617e4 | 1313 | |
8eb7e2d0 | 1314 | if (io_is_timeout_noseq(req)) |
360428f8 | 1315 | break; |
f010505b MDG |
1316 | |
1317 | /* | |
1318 | * Since seq can easily wrap around over time, subtract | |
1319 | * the last seq at which timeouts were flushed before comparing. | |
1320 | * Assuming not more than 2^31-1 events have happened since, | |
1321 | * these subtractions won't have wrapped, so we can check if | |
1322 | * target is in [last_seq, current_seq] by comparing the two. | |
1323 | */ | |
1324 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1325 | events_got = seq - ctx->cq_last_tm_flush; | |
1326 | if (events_got < events_needed) | |
360428f8 | 1327 | break; |
bfe68a22 | 1328 | |
135fcde8 | 1329 | list_del_init(&req->timeout.list); |
5262f567 | 1330 | io_kill_timeout(req); |
f010505b MDG |
1331 | } while (!list_empty(&ctx->timeout_list)); |
1332 | ||
1333 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1334 | } |
5262f567 | 1335 | |
360428f8 PB |
1336 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1337 | { | |
1338 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1339 | |
1340 | /* order cqe stores with ring update */ | |
1341 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1342 | |
04518945 PB |
1343 | if (unlikely(!list_empty(&ctx->defer_list))) |
1344 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1345 | } |
1346 | ||
90554200 JA |
1347 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1348 | { | |
1349 | struct io_rings *r = ctx->rings; | |
1350 | ||
1351 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1352 | } | |
1353 | ||
888aae2e PB |
1354 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1355 | { | |
1356 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1357 | } | |
1358 | ||
2b188cc1 JA |
1359 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1360 | { | |
75b28aff | 1361 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1362 | unsigned tail; |
1363 | ||
115e12e5 SB |
1364 | /* |
1365 | * writes to the cq entry need to come after reading head; the | |
1366 | * control dependency is enough as we're using WRITE_ONCE to | |
1367 | * fill the cq entry | |
1368 | */ | |
888aae2e | 1369 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1370 | return NULL; |
1371 | ||
888aae2e | 1372 | tail = ctx->cached_cq_tail++; |
75b28aff | 1373 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1374 | } |
1375 | ||
f2842ab5 JA |
1376 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1377 | { | |
f0b493e6 JA |
1378 | if (!ctx->cq_ev_fd) |
1379 | return false; | |
7e55a19c SG |
1380 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1381 | return false; | |
f2842ab5 JA |
1382 | if (!ctx->eventfd_async) |
1383 | return true; | |
b41e9852 | 1384 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1385 | } |
1386 | ||
b41e9852 | 1387 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1388 | { |
b1445e59 PB |
1389 | /* see waitqueue_active() comment */ |
1390 | smp_mb(); | |
1391 | ||
1d7bb1d5 JA |
1392 | if (waitqueue_active(&ctx->wait)) |
1393 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1394 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1395 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1396 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1397 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1398 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1399 | wake_up_interruptible(&ctx->cq_wait); |
1400 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1401 | } | |
1d7bb1d5 JA |
1402 | } |
1403 | ||
80c18e4a PB |
1404 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1405 | { | |
b1445e59 PB |
1406 | /* see waitqueue_active() comment */ |
1407 | smp_mb(); | |
1408 | ||
80c18e4a PB |
1409 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1410 | if (waitqueue_active(&ctx->wait)) | |
1411 | wake_up(&ctx->wait); | |
1412 | } | |
1413 | if (io_should_trigger_evfd(ctx)) | |
1414 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1415 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1416 | wake_up_interruptible(&ctx->cq_wait); |
1417 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1418 | } | |
80c18e4a PB |
1419 | } |
1420 | ||
c4a2ed72 | 1421 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1422 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1423 | struct task_struct *tsk, | |
1424 | struct files_struct *files) | |
1d7bb1d5 JA |
1425 | { |
1426 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1427 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1428 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1429 | unsigned long flags; |
b18032bb | 1430 | bool all_flushed, posted; |
1d7bb1d5 JA |
1431 | LIST_HEAD(list); |
1432 | ||
e23de15f PB |
1433 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1434 | return false; | |
1d7bb1d5 | 1435 | |
b18032bb | 1436 | posted = false; |
1d7bb1d5 | 1437 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1438 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1439 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1440 | continue; |
1441 | ||
1d7bb1d5 JA |
1442 | cqe = io_get_cqring(ctx); |
1443 | if (!cqe && !force) | |
1444 | break; | |
1445 | ||
40d8ddd4 | 1446 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1447 | if (cqe) { |
1448 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1449 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1450 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1451 | } else { |
2c3bac6d | 1452 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1453 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1454 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1455 | } |
b18032bb | 1456 | posted = true; |
1d7bb1d5 JA |
1457 | } |
1458 | ||
09e88404 PB |
1459 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1460 | if (all_flushed) { | |
1461 | clear_bit(0, &ctx->sq_check_overflow); | |
1462 | clear_bit(0, &ctx->cq_check_overflow); | |
1463 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1464 | } | |
46930143 | 1465 | |
b18032bb JA |
1466 | if (posted) |
1467 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1468 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1469 | if (posted) |
1470 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1471 | |
1472 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1473 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1474 | list_del(&req->compl.list); | |
ec9c02ad | 1475 | io_put_req(req); |
1d7bb1d5 | 1476 | } |
c4a2ed72 | 1477 | |
09e88404 | 1478 | return all_flushed; |
1d7bb1d5 JA |
1479 | } |
1480 | ||
ca0a2651 | 1481 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1482 | struct task_struct *tsk, |
1483 | struct files_struct *files) | |
1484 | { | |
ca0a2651 JA |
1485 | bool ret = true; |
1486 | ||
6c503150 PB |
1487 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1488 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1489 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1490 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1491 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1492 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1493 | mutex_unlock(&ctx->uring_lock); | |
1494 | } | |
ca0a2651 JA |
1495 | |
1496 | return ret; | |
6c503150 PB |
1497 | } |
1498 | ||
bcda7baa | 1499 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1500 | { |
78e19bbe | 1501 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1502 | struct io_uring_cqe *cqe; |
1503 | ||
78e19bbe | 1504 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1505 | |
2b188cc1 JA |
1506 | /* |
1507 | * If we can't get a cq entry, userspace overflowed the | |
1508 | * submission (by quite a lot). Increment the overflow count in | |
1509 | * the ring. | |
1510 | */ | |
1511 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1512 | if (likely(cqe)) { |
78e19bbe | 1513 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1514 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1515 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1516 | } else if (ctx->cq_overflow_flushed || |
1517 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1518 | /* |
1519 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1520 | * then we cannot store the request for later flushing, we need | |
1521 | * to drop it on the floor. | |
1522 | */ | |
2c3bac6d PB |
1523 | ctx->cached_cq_overflow++; |
1524 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1525 | } else { |
ad3eb2c8 JA |
1526 | if (list_empty(&ctx->cq_overflow_list)) { |
1527 | set_bit(0, &ctx->sq_check_overflow); | |
1528 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1529 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1530 | } |
40d8ddd4 | 1531 | io_clean_op(req); |
1d7bb1d5 | 1532 | req->result = res; |
0f7e466b | 1533 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1534 | refcount_inc(&req->refs); |
1535 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1536 | } |
1537 | } | |
1538 | ||
bcda7baa JA |
1539 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1540 | { | |
1541 | __io_cqring_fill_event(req, res, 0); | |
1542 | } | |
1543 | ||
7a612350 PB |
1544 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1545 | unsigned int cflags) | |
2b188cc1 | 1546 | { |
78e19bbe | 1547 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1548 | unsigned long flags; |
1549 | ||
1550 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1551 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1552 | /* |
1553 | * If we're the last reference to this request, add to our locked | |
1554 | * free_list cache. | |
1555 | */ | |
1556 | if (refcount_dec_and_test(&req->refs)) { | |
1557 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1558 | ||
7a612350 PB |
1559 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1560 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1561 | io_disarm_next(req); | |
1562 | if (req->link) { | |
1563 | io_req_task_queue(req->link); | |
1564 | req->link = NULL; | |
1565 | } | |
1566 | } | |
c7dae4ba JA |
1567 | io_dismantle_req(req); |
1568 | io_put_task(req->task, 1); | |
1569 | list_add(&req->compl.list, &cs->locked_free_list); | |
1570 | cs->locked_free_nr++; | |
180f829f PB |
1571 | } else { |
1572 | if (!percpu_ref_tryget(&ctx->refs)) | |
1573 | req = NULL; | |
1574 | } | |
7a612350 | 1575 | io_commit_cqring(ctx); |
2b188cc1 | 1576 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1577 | |
180f829f PB |
1578 | if (req) { |
1579 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1580 | percpu_ref_put(&ctx->refs); |
180f829f | 1581 | } |
229a7b63 JA |
1582 | } |
1583 | ||
a38d68db | 1584 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1585 | unsigned int cflags) |
229a7b63 | 1586 | { |
a38d68db PB |
1587 | io_clean_op(req); |
1588 | req->result = res; | |
1589 | req->compl.cflags = cflags; | |
e342c807 | 1590 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1591 | } |
1592 | ||
889fca73 PB |
1593 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1594 | long res, unsigned cflags) | |
bcda7baa | 1595 | { |
889fca73 PB |
1596 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1597 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1598 | else |
c7dae4ba | 1599 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1600 | } |
1601 | ||
a38d68db | 1602 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1603 | { |
889fca73 | 1604 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1605 | } |
1606 | ||
c7dae4ba | 1607 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1608 | { |
c7dae4ba JA |
1609 | struct io_submit_state *state = &ctx->submit_state; |
1610 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1611 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1612 | |
c7dae4ba JA |
1613 | /* |
1614 | * If we have more than a batch's worth of requests in our IRQ side | |
1615 | * locked cache, grab the lock and move them over to our submission | |
1616 | * side cache. | |
1617 | */ | |
1618 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1619 | spin_lock_irq(&ctx->completion_lock); | |
1620 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1621 | cs->locked_free_nr = 0; | |
1622 | spin_unlock_irq(&ctx->completion_lock); | |
1623 | } | |
0ddf92e8 | 1624 | |
c7dae4ba JA |
1625 | while (!list_empty(&cs->free_list)) { |
1626 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1627 | compl.list); |
1628 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1629 | state->reqs[state->free_reqs++] = req; |
1630 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1631 | break; | |
1b4c351f JA |
1632 | } |
1633 | ||
e5d1bc0a | 1634 | return req != NULL; |
0ddf92e8 JA |
1635 | } |
1636 | ||
e5d1bc0a | 1637 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1638 | { |
e5d1bc0a PB |
1639 | struct io_submit_state *state = &ctx->submit_state; |
1640 | ||
1641 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1642 | ||
f6b6c7d6 | 1643 | if (!state->free_reqs) { |
291b2821 | 1644 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1645 | int ret; |
1646 | ||
c7dae4ba | 1647 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1648 | goto got_req; |
1649 | ||
bf019da7 PB |
1650 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1651 | state->reqs); | |
fd6fab2c JA |
1652 | |
1653 | /* | |
1654 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1655 | * retry single alloc to be on the safe side. | |
1656 | */ | |
1657 | if (unlikely(ret <= 0)) { | |
1658 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1659 | if (!state->reqs[0]) | |
3893f39f | 1660 | return NULL; |
fd6fab2c JA |
1661 | ret = 1; |
1662 | } | |
291b2821 | 1663 | state->free_reqs = ret; |
2b188cc1 | 1664 | } |
e5d1bc0a | 1665 | got_req: |
291b2821 PB |
1666 | state->free_reqs--; |
1667 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1668 | } |
1669 | ||
8da11c19 PB |
1670 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1671 | bool fixed) | |
1672 | { | |
36f72fe2 | 1673 | if (!fixed) |
8da11c19 PB |
1674 | fput(file); |
1675 | } | |
1676 | ||
4edf20f9 | 1677 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1678 | { |
3ca405eb | 1679 | io_clean_op(req); |
929a3af9 | 1680 | |
e8c2bc1f JA |
1681 | if (req->async_data) |
1682 | kfree(req->async_data); | |
8da11c19 PB |
1683 | if (req->file) |
1684 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1685 | if (req->fixed_rsrc_refs) |
1686 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1687 | if (req->work.creds) { |
1688 | put_cred(req->work.creds); | |
1689 | req->work.creds = NULL; | |
1690 | } | |
f85c310a PB |
1691 | |
1692 | if (req->flags & REQ_F_INFLIGHT) { | |
1693 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1694 | unsigned long flags; |
1695 | ||
1696 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1697 | list_del(&req->inflight_entry); | |
1698 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1699 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1700 | } |
e65ef56d JA |
1701 | } |
1702 | ||
b23fcf47 | 1703 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1704 | static inline void io_put_task(struct task_struct *task, int nr) |
1705 | { | |
1706 | struct io_uring_task *tctx = task->io_uring; | |
1707 | ||
1708 | percpu_counter_sub(&tctx->inflight, nr); | |
1709 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1710 | wake_up(&tctx->wait); | |
1711 | put_task_struct_many(task, nr); | |
1712 | } | |
1713 | ||
216578e5 | 1714 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1715 | { |
51a4cc11 | 1716 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1717 | |
216578e5 | 1718 | io_dismantle_req(req); |
7c660731 | 1719 | io_put_task(req->task, 1); |
c6ca97b3 | 1720 | |
3893f39f | 1721 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1722 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1723 | } |
1724 | ||
f2f87370 PB |
1725 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1726 | { | |
1727 | struct io_kiocb *nxt = req->link; | |
1728 | ||
1729 | req->link = nxt->link; | |
1730 | nxt->link = NULL; | |
1731 | } | |
1732 | ||
33cc89a9 PB |
1733 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1734 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1735 | { |
33cc89a9 | 1736 | struct io_kiocb *link = req->link; |
c9abd7ad | 1737 | bool cancelled = false; |
f2f87370 | 1738 | |
900fad45 PB |
1739 | /* |
1740 | * Can happen if a linked timeout fired and link had been like | |
1741 | * req -> link t-out -> link t-out [-> ...] | |
1742 | */ | |
c9abd7ad PB |
1743 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1744 | struct io_timeout_data *io = link->async_data; | |
1745 | int ret; | |
7c86ffee | 1746 | |
f2f87370 | 1747 | io_remove_next_linked(req); |
90cd7e42 | 1748 | link->timeout.head = NULL; |
c9abd7ad PB |
1749 | ret = hrtimer_try_to_cancel(&io->timer); |
1750 | if (ret != -1) { | |
1751 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1752 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1753 | cancelled = true; |
1754 | } | |
1755 | } | |
7c86ffee | 1756 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1757 | return cancelled; |
7c86ffee PB |
1758 | } |
1759 | ||
d148ca4b | 1760 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1761 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1762 | { |
33cc89a9 | 1763 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1764 | |
f2f87370 | 1765 | req->link = NULL; |
f2f87370 PB |
1766 | while (link) { |
1767 | nxt = link->link; | |
1768 | link->link = NULL; | |
2665abfd | 1769 | |
f2f87370 | 1770 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1771 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1772 | io_put_req_deferred(link, 2); |
f2f87370 | 1773 | link = nxt; |
9e645e11 | 1774 | } |
33cc89a9 | 1775 | } |
9e645e11 | 1776 | |
33cc89a9 PB |
1777 | static bool io_disarm_next(struct io_kiocb *req) |
1778 | __must_hold(&req->ctx->completion_lock) | |
1779 | { | |
1780 | bool posted = false; | |
1781 | ||
1782 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1783 | posted = io_kill_linked_timeout(req); | |
1784 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1785 | posted |= (req->link != NULL); | |
1786 | io_fail_links(req); | |
1787 | } | |
1788 | return posted; | |
9e645e11 JA |
1789 | } |
1790 | ||
3fa5e0f3 | 1791 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1792 | { |
33cc89a9 | 1793 | struct io_kiocb *nxt; |
944e58bf | 1794 | |
9e645e11 JA |
1795 | /* |
1796 | * If LINK is set, we have dependent requests in this chain. If we | |
1797 | * didn't fail this request, queue the first one up, moving any other | |
1798 | * dependencies to the next request. In case of failure, fail the rest | |
1799 | * of the chain. | |
1800 | */ | |
33cc89a9 PB |
1801 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1802 | struct io_ring_ctx *ctx = req->ctx; | |
1803 | unsigned long flags; | |
1804 | bool posted; | |
1805 | ||
1806 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1807 | posted = io_disarm_next(req); | |
1808 | if (posted) | |
1809 | io_commit_cqring(req->ctx); | |
1810 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1811 | if (posted) | |
1812 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1813 | } |
33cc89a9 PB |
1814 | nxt = req->link; |
1815 | req->link = NULL; | |
1816 | return nxt; | |
4d7dd462 | 1817 | } |
9e645e11 | 1818 | |
f2f87370 | 1819 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1820 | { |
cdbff982 | 1821 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1822 | return NULL; |
1823 | return __io_req_find_next(req); | |
1824 | } | |
1825 | ||
2c32395d PB |
1826 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1827 | { | |
1828 | if (!ctx) | |
1829 | return; | |
1830 | if (ctx->submit_state.comp.nr) { | |
1831 | mutex_lock(&ctx->uring_lock); | |
1832 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1833 | mutex_unlock(&ctx->uring_lock); | |
1834 | } | |
1835 | percpu_ref_put(&ctx->refs); | |
1836 | } | |
1837 | ||
7cbf1722 | 1838 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1839 | { |
65453d1e | 1840 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1841 | struct io_wq_work_list list; |
1842 | struct io_wq_work_node *node; | |
c2c4c83c | 1843 | |
7cbf1722 JA |
1844 | if (wq_list_empty(&tctx->task_list)) |
1845 | return false; | |
6200b0ae | 1846 | |
0b81e80c | 1847 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1848 | list = tctx->task_list; |
1849 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1850 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1851 | |
7cbf1722 JA |
1852 | node = list.first; |
1853 | while (node) { | |
1854 | struct io_wq_work_node *next = node->next; | |
1855 | struct io_kiocb *req; | |
0ba9c9ed | 1856 | |
7cbf1722 | 1857 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1858 | if (req->ctx != ctx) { |
1859 | ctx_flush_and_put(ctx); | |
1860 | ctx = req->ctx; | |
1861 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1862 | } |
65453d1e | 1863 | |
2c32395d PB |
1864 | req->task_work.func(&req->task_work); |
1865 | node = next; | |
7cbf1722 JA |
1866 | } |
1867 | ||
2c32395d | 1868 | ctx_flush_and_put(ctx); |
7cbf1722 | 1869 | return list.first != NULL; |
c2c4c83c JA |
1870 | } |
1871 | ||
7cbf1722 | 1872 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1873 | { |
7cbf1722 | 1874 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1875 | |
1d5f360d JA |
1876 | clear_bit(0, &tctx->task_state); |
1877 | ||
7cbf1722 JA |
1878 | while (__tctx_task_work(tctx)) |
1879 | cond_resched(); | |
7cbf1722 JA |
1880 | } |
1881 | ||
1882 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1883 | enum task_work_notify_mode notify) | |
1884 | { | |
1885 | struct io_uring_task *tctx = tsk->io_uring; | |
1886 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1887 | unsigned long flags; |
7cbf1722 JA |
1888 | int ret; |
1889 | ||
1890 | WARN_ON_ONCE(!tctx); | |
1891 | ||
0b81e80c | 1892 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1893 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1894 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1895 | |
1896 | /* task_work already pending, we're done */ | |
1897 | if (test_bit(0, &tctx->task_state) || | |
1898 | test_and_set_bit(0, &tctx->task_state)) | |
1899 | return 0; | |
1900 | ||
1901 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1902 | return 0; | |
1903 | ||
1904 | /* | |
1905 | * Slow path - we failed, find and delete work. if the work is not | |
1906 | * in the list, it got run and we're fine. | |
1907 | */ | |
1908 | ret = 0; | |
0b81e80c | 1909 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1910 | wq_list_for_each(node, prev, &tctx->task_list) { |
1911 | if (&req->io_task_work.node == node) { | |
1912 | wq_list_del(&tctx->task_list, node, prev); | |
1913 | ret = 1; | |
1914 | break; | |
1915 | } | |
1916 | } | |
0b81e80c | 1917 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1918 | clear_bit(0, &tctx->task_state); |
1919 | return ret; | |
1920 | } | |
1921 | ||
355fb9e2 | 1922 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1923 | { |
1924 | struct task_struct *tsk = req->task; | |
1925 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1926 | enum task_work_notify_mode notify; |
1927 | int ret; | |
c2c4c83c | 1928 | |
6200b0ae JA |
1929 | if (tsk->flags & PF_EXITING) |
1930 | return -ESRCH; | |
1931 | ||
c2c4c83c | 1932 | /* |
0ba9c9ed JA |
1933 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1934 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1935 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1936 | * will do the job. | |
c2c4c83c | 1937 | */ |
91989c70 | 1938 | notify = TWA_NONE; |
355fb9e2 | 1939 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1940 | notify = TWA_SIGNAL; |
1941 | ||
7cbf1722 | 1942 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1943 | if (!ret) |
1944 | wake_up_process(tsk); | |
0ba9c9ed | 1945 | |
c2c4c83c JA |
1946 | return ret; |
1947 | } | |
1948 | ||
9b465711 PB |
1949 | static bool io_run_task_work_head(struct callback_head **work_head) |
1950 | { | |
1951 | struct callback_head *work, *next; | |
1952 | bool executed = false; | |
1953 | ||
1954 | do { | |
1955 | work = xchg(work_head, NULL); | |
1956 | if (!work) | |
1957 | break; | |
1958 | ||
1959 | do { | |
1960 | next = work->next; | |
1961 | work->func(work); | |
1962 | work = next; | |
1963 | cond_resched(); | |
1964 | } while (work); | |
1965 | executed = true; | |
1966 | } while (1); | |
1967 | ||
1968 | return executed; | |
1969 | } | |
1970 | ||
1971 | static void io_task_work_add_head(struct callback_head **work_head, | |
1972 | struct callback_head *task_work) | |
eab30c4d | 1973 | { |
7c25c0d1 | 1974 | struct callback_head *head; |
eab30c4d | 1975 | |
7c25c0d1 | 1976 | do { |
9b465711 PB |
1977 | head = READ_ONCE(*work_head); |
1978 | task_work->next = head; | |
1979 | } while (cmpxchg(work_head, head, task_work) != head); | |
1980 | } | |
1981 | ||
1982 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
1983 | task_work_func_t cb) | |
1984 | { | |
1985 | init_task_work(&req->task_work, cb); | |
1986 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
1987 | } |
1988 | ||
c40f6379 JA |
1989 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1990 | { | |
1991 | struct io_ring_ctx *ctx = req->ctx; | |
1992 | ||
1993 | spin_lock_irq(&ctx->completion_lock); | |
1994 | io_cqring_fill_event(req, error); | |
1995 | io_commit_cqring(ctx); | |
1996 | spin_unlock_irq(&ctx->completion_lock); | |
1997 | ||
1998 | io_cqring_ev_posted(ctx); | |
1999 | req_set_fail_links(req); | |
2000 | io_double_put_req(req); | |
2001 | } | |
2002 | ||
2003 | static void io_req_task_cancel(struct callback_head *cb) | |
2004 | { | |
2005 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2006 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2007 | |
792bb6eb | 2008 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 2009 | __io_req_task_cancel(req, req->result); |
792bb6eb | 2010 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 2011 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2012 | } |
2013 | ||
2014 | static void __io_req_task_submit(struct io_kiocb *req) | |
2015 | { | |
2016 | struct io_ring_ctx *ctx = req->ctx; | |
2017 | ||
04fc6c80 | 2018 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2019 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2020 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2021 | __io_queue_sqe(req); |
81b6d05c | 2022 | else |
c40f6379 | 2023 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2024 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2025 | } |
2026 | ||
2027 | static void io_req_task_submit(struct callback_head *cb) | |
2028 | { | |
2029 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2030 | ||
2031 | __io_req_task_submit(req); | |
2032 | } | |
2033 | ||
2034 | static void io_req_task_queue(struct io_kiocb *req) | |
2035 | { | |
c40f6379 JA |
2036 | int ret; |
2037 | ||
7cbf1722 | 2038 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2039 | ret = io_req_task_work_add(req); |
c40f6379 | 2040 | if (unlikely(ret)) { |
a3df7698 | 2041 | req->result = -ECANCELED; |
04fc6c80 | 2042 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 2043 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 2044 | } |
c40f6379 JA |
2045 | } |
2046 | ||
a3df7698 PB |
2047 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2048 | { | |
2049 | percpu_ref_get(&req->ctx->refs); | |
2050 | req->result = ret; | |
2051 | req->task_work.func = io_req_task_cancel; | |
2052 | ||
2053 | if (unlikely(io_req_task_work_add(req))) | |
2054 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2055 | } | |
2056 | ||
f2f87370 | 2057 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2058 | { |
9b5f7bd9 | 2059 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2060 | |
2061 | if (nxt) | |
906a8c3f | 2062 | io_req_task_queue(nxt); |
c69f8dbe JL |
2063 | } |
2064 | ||
c3524383 | 2065 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2066 | { |
c3524383 PB |
2067 | io_queue_next(req); |
2068 | __io_free_req(req); | |
2069 | } | |
8766dd51 | 2070 | |
2d6500d4 | 2071 | struct req_batch { |
5af1d13e PB |
2072 | struct task_struct *task; |
2073 | int task_refs; | |
1b4c351f | 2074 | int ctx_refs; |
2d6500d4 PB |
2075 | }; |
2076 | ||
5af1d13e PB |
2077 | static inline void io_init_req_batch(struct req_batch *rb) |
2078 | { | |
5af1d13e | 2079 | rb->task_refs = 0; |
9ae72463 | 2080 | rb->ctx_refs = 0; |
5af1d13e PB |
2081 | rb->task = NULL; |
2082 | } | |
2083 | ||
2d6500d4 PB |
2084 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2085 | struct req_batch *rb) | |
2086 | { | |
6e833d53 | 2087 | if (rb->task) |
7c660731 | 2088 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2089 | if (rb->ctx_refs) |
2090 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2091 | } |
2092 | ||
6ff119a6 PB |
2093 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2094 | struct io_submit_state *state) | |
2d6500d4 | 2095 | { |
f2f87370 | 2096 | io_queue_next(req); |
2d6500d4 | 2097 | |
e3bc8e9d | 2098 | if (req->task != rb->task) { |
7c660731 PB |
2099 | if (rb->task) |
2100 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2101 | rb->task = req->task; |
2102 | rb->task_refs = 0; | |
5af1d13e | 2103 | } |
e3bc8e9d | 2104 | rb->task_refs++; |
9ae72463 | 2105 | rb->ctx_refs++; |
5af1d13e | 2106 | |
4edf20f9 | 2107 | io_dismantle_req(req); |
bd759045 | 2108 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2109 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2110 | else |
2111 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2112 | } |
2113 | ||
905c172f PB |
2114 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2115 | struct io_ring_ctx *ctx) | |
2116 | { | |
2117 | int i, nr = cs->nr; | |
2118 | struct io_kiocb *req; | |
2119 | struct req_batch rb; | |
2120 | ||
2121 | io_init_req_batch(&rb); | |
2122 | spin_lock_irq(&ctx->completion_lock); | |
2123 | for (i = 0; i < nr; i++) { | |
2124 | req = cs->reqs[i]; | |
2125 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2126 | } | |
2127 | io_commit_cqring(ctx); | |
2128 | spin_unlock_irq(&ctx->completion_lock); | |
2129 | ||
2130 | io_cqring_ev_posted(ctx); | |
2131 | for (i = 0; i < nr; i++) { | |
2132 | req = cs->reqs[i]; | |
2133 | ||
2134 | /* submission and completion refs */ | |
2135 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2136 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2137 | } |
2138 | ||
2139 | io_req_free_batch_finish(ctx, &rb); | |
2140 | cs->nr = 0; | |
7a743e22 PB |
2141 | } |
2142 | ||
ba816ad6 JA |
2143 | /* |
2144 | * Drop reference to request, return next in chain (if there is one) if this | |
2145 | * was the last reference to this request. | |
2146 | */ | |
9b5f7bd9 | 2147 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2148 | { |
9b5f7bd9 PB |
2149 | struct io_kiocb *nxt = NULL; |
2150 | ||
2a44f467 | 2151 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2152 | nxt = io_req_find_next(req); |
4d7dd462 | 2153 | __io_free_req(req); |
2a44f467 | 2154 | } |
9b5f7bd9 | 2155 | return nxt; |
2b188cc1 JA |
2156 | } |
2157 | ||
e65ef56d JA |
2158 | static void io_put_req(struct io_kiocb *req) |
2159 | { | |
2160 | if (refcount_dec_and_test(&req->refs)) | |
2161 | io_free_req(req); | |
2b188cc1 JA |
2162 | } |
2163 | ||
216578e5 PB |
2164 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2165 | { | |
2166 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2167 | ||
2168 | io_free_req(req); | |
2169 | } | |
2170 | ||
2171 | static void io_free_req_deferred(struct io_kiocb *req) | |
2172 | { | |
2173 | int ret; | |
2174 | ||
7cbf1722 | 2175 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2176 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2177 | if (unlikely(ret)) |
2178 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2179 | } |
2180 | ||
2181 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2182 | { | |
2183 | if (refcount_sub_and_test(refs, &req->refs)) | |
2184 | io_free_req_deferred(req); | |
2185 | } | |
2186 | ||
978db57e JA |
2187 | static void io_double_put_req(struct io_kiocb *req) |
2188 | { | |
2189 | /* drop both submit and complete references */ | |
2190 | if (refcount_sub_and_test(2, &req->refs)) | |
2191 | io_free_req(req); | |
2192 | } | |
2193 | ||
6c503150 | 2194 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2195 | { |
2196 | /* See comment at the top of this file */ | |
2197 | smp_rmb(); | |
e23de15f | 2198 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2199 | } |
2200 | ||
fb5ccc98 PB |
2201 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2202 | { | |
2203 | struct io_rings *rings = ctx->rings; | |
2204 | ||
2205 | /* make sure SQ entry isn't read before tail */ | |
2206 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2207 | } | |
2208 | ||
8ff069bf | 2209 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2210 | { |
8ff069bf | 2211 | unsigned int cflags; |
e94f141b | 2212 | |
bcda7baa JA |
2213 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2214 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2215 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2216 | kfree(kbuf); |
2217 | return cflags; | |
e94f141b JA |
2218 | } |
2219 | ||
8ff069bf | 2220 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2221 | { |
4d954c25 | 2222 | struct io_buffer *kbuf; |
bcda7baa | 2223 | |
4d954c25 | 2224 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2225 | return io_put_kbuf(req, kbuf); |
2226 | } | |
2227 | ||
4c6e277c JA |
2228 | static inline bool io_run_task_work(void) |
2229 | { | |
6200b0ae JA |
2230 | /* |
2231 | * Not safe to run on exiting task, and the task_work handling will | |
2232 | * not add work to such a task. | |
2233 | */ | |
2234 | if (unlikely(current->flags & PF_EXITING)) | |
2235 | return false; | |
4c6e277c JA |
2236 | if (current->task_works) { |
2237 | __set_current_state(TASK_RUNNING); | |
2238 | task_work_run(); | |
2239 | return true; | |
2240 | } | |
2241 | ||
2242 | return false; | |
bcda7baa JA |
2243 | } |
2244 | ||
def596e9 JA |
2245 | /* |
2246 | * Find and free completed poll iocbs | |
2247 | */ | |
2248 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2249 | struct list_head *done) | |
2250 | { | |
8237e045 | 2251 | struct req_batch rb; |
def596e9 | 2252 | struct io_kiocb *req; |
bbde017a XW |
2253 | |
2254 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2255 | smp_rmb(); | |
def596e9 | 2256 | |
5af1d13e | 2257 | io_init_req_batch(&rb); |
def596e9 | 2258 | while (!list_empty(done)) { |
bcda7baa JA |
2259 | int cflags = 0; |
2260 | ||
d21ffe7e | 2261 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2262 | list_del(&req->inflight_entry); |
2263 | ||
bbde017a XW |
2264 | if (READ_ONCE(req->result) == -EAGAIN) { |
2265 | req->iopoll_completed = 0; | |
23faba36 | 2266 | if (io_rw_reissue(req)) |
f161340d | 2267 | continue; |
bbde017a | 2268 | } |
def596e9 | 2269 | |
bcda7baa | 2270 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2271 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2272 | |
2273 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2274 | (*nr_events)++; |
2275 | ||
c3524383 | 2276 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2277 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2278 | } |
def596e9 | 2279 | |
09bb8394 | 2280 | io_commit_cqring(ctx); |
80c18e4a | 2281 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2282 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2283 | } |
2284 | ||
def596e9 JA |
2285 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2286 | long min) | |
2287 | { | |
2288 | struct io_kiocb *req, *tmp; | |
2289 | LIST_HEAD(done); | |
2290 | bool spin; | |
2291 | int ret; | |
2292 | ||
2293 | /* | |
2294 | * Only spin for completions if we don't have multiple devices hanging | |
2295 | * off our complete list, and we're under the requested amount. | |
2296 | */ | |
2297 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2298 | ||
2299 | ret = 0; | |
d21ffe7e | 2300 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2301 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2302 | |
2303 | /* | |
581f9810 BM |
2304 | * Move completed and retryable entries to our local lists. |
2305 | * If we find a request that requires polling, break out | |
2306 | * and complete those lists first, if we have entries there. | |
def596e9 | 2307 | */ |
65a6543d | 2308 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2309 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2310 | continue; |
2311 | } | |
2312 | if (!list_empty(&done)) | |
2313 | break; | |
2314 | ||
2315 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2316 | if (ret < 0) | |
2317 | break; | |
2318 | ||
3aadc23e PB |
2319 | /* iopoll may have completed current req */ |
2320 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2321 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2322 | |
def596e9 JA |
2323 | if (ret && spin) |
2324 | spin = false; | |
2325 | ret = 0; | |
2326 | } | |
2327 | ||
2328 | if (!list_empty(&done)) | |
2329 | io_iopoll_complete(ctx, nr_events, &done); | |
2330 | ||
2331 | return ret; | |
2332 | } | |
2333 | ||
2334 | /* | |
d195a66e | 2335 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2336 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2337 | * as a non-spinning completion check. | |
2338 | */ | |
2339 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2340 | long min) | |
2341 | { | |
540e32a0 | 2342 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2343 | int ret; |
2344 | ||
2345 | ret = io_do_iopoll(ctx, nr_events, min); | |
2346 | if (ret < 0) | |
2347 | return ret; | |
eba0a4dd | 2348 | if (*nr_events >= min) |
def596e9 JA |
2349 | return 0; |
2350 | } | |
2351 | ||
2352 | return 1; | |
2353 | } | |
2354 | ||
2355 | /* | |
2356 | * We can't just wait for polled events to come to us, we have to actively | |
2357 | * find and complete them. | |
2358 | */ | |
b2edc0a7 | 2359 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2360 | { |
2361 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2362 | return; | |
2363 | ||
2364 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2365 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2366 | unsigned int nr_events = 0; |
2367 | ||
b2edc0a7 | 2368 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2369 | |
b2edc0a7 PB |
2370 | /* let it sleep and repeat later if can't complete a request */ |
2371 | if (nr_events == 0) | |
2372 | break; | |
08f5439f JA |
2373 | /* |
2374 | * Ensure we allow local-to-the-cpu processing to take place, | |
2375 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2376 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2377 | */ |
3fcee5a6 PB |
2378 | if (need_resched()) { |
2379 | mutex_unlock(&ctx->uring_lock); | |
2380 | cond_resched(); | |
2381 | mutex_lock(&ctx->uring_lock); | |
2382 | } | |
def596e9 JA |
2383 | } |
2384 | mutex_unlock(&ctx->uring_lock); | |
2385 | } | |
2386 | ||
7668b92a | 2387 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2388 | { |
7668b92a | 2389 | unsigned int nr_events = 0; |
2b2ed975 | 2390 | int iters = 0, ret = 0; |
500f9fba | 2391 | |
c7849be9 XW |
2392 | /* |
2393 | * We disallow the app entering submit/complete with polling, but we | |
2394 | * still need to lock the ring to prevent racing with polled issue | |
2395 | * that got punted to a workqueue. | |
2396 | */ | |
2397 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2398 | do { |
a3a0e43f JA |
2399 | /* |
2400 | * Don't enter poll loop if we already have events pending. | |
2401 | * If we do, we can potentially be spinning for commands that | |
2402 | * already triggered a CQE (eg in error). | |
2403 | */ | |
6c503150 PB |
2404 | if (test_bit(0, &ctx->cq_check_overflow)) |
2405 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2406 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2407 | break; |
2408 | ||
500f9fba JA |
2409 | /* |
2410 | * If a submit got punted to a workqueue, we can have the | |
2411 | * application entering polling for a command before it gets | |
2412 | * issued. That app will hold the uring_lock for the duration | |
2413 | * of the poll right here, so we need to take a breather every | |
2414 | * now and then to ensure that the issue has a chance to add | |
2415 | * the poll to the issued list. Otherwise we can spin here | |
2416 | * forever, while the workqueue is stuck trying to acquire the | |
2417 | * very same mutex. | |
2418 | */ | |
2419 | if (!(++iters & 7)) { | |
2420 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2421 | io_run_task_work(); |
500f9fba JA |
2422 | mutex_lock(&ctx->uring_lock); |
2423 | } | |
2424 | ||
7668b92a | 2425 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2426 | if (ret <= 0) |
2427 | break; | |
2428 | ret = 0; | |
7668b92a | 2429 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2430 | |
500f9fba | 2431 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2432 | return ret; |
2433 | } | |
2434 | ||
491381ce | 2435 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2436 | { |
491381ce JA |
2437 | /* |
2438 | * Tell lockdep we inherited freeze protection from submission | |
2439 | * thread. | |
2440 | */ | |
2441 | if (req->flags & REQ_F_ISREG) { | |
2442 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2443 | |
491381ce | 2444 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2445 | } |
491381ce | 2446 | file_end_write(req->file); |
2b188cc1 JA |
2447 | } |
2448 | ||
b63534c4 | 2449 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2450 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2451 | { |
2452 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2453 | int rw, ret; |
b63534c4 | 2454 | struct iov_iter iter; |
b63534c4 | 2455 | |
dc2a6e9a PB |
2456 | /* already prepared */ |
2457 | if (req->async_data) | |
2458 | return true; | |
b63534c4 JA |
2459 | |
2460 | switch (req->opcode) { | |
2461 | case IORING_OP_READV: | |
2462 | case IORING_OP_READ_FIXED: | |
2463 | case IORING_OP_READ: | |
2464 | rw = READ; | |
2465 | break; | |
2466 | case IORING_OP_WRITEV: | |
2467 | case IORING_OP_WRITE_FIXED: | |
2468 | case IORING_OP_WRITE: | |
2469 | rw = WRITE; | |
2470 | break; | |
2471 | default: | |
2472 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2473 | req->opcode); | |
dc2a6e9a | 2474 | return false; |
b63534c4 JA |
2475 | } |
2476 | ||
dc2a6e9a PB |
2477 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2478 | if (ret < 0) | |
2479 | return false; | |
6bf985dc | 2480 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2481 | } |
b63534c4 | 2482 | |
3e6a0d3c | 2483 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2484 | { |
355afaeb | 2485 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2486 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2487 | |
355afaeb JA |
2488 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2489 | return false; | |
3e6a0d3c JA |
2490 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2491 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2492 | return false; |
7c977a58 JA |
2493 | /* |
2494 | * If ref is dying, we might be running poll reap from the exit work. | |
2495 | * Don't attempt to reissue from that path, just let it fail with | |
2496 | * -EAGAIN. | |
2497 | */ | |
3e6a0d3c JA |
2498 | if (percpu_ref_is_dying(&ctx->refs)) |
2499 | return false; | |
2500 | return true; | |
2501 | } | |
2502 | #endif | |
2503 | ||
2504 | static bool io_rw_reissue(struct io_kiocb *req) | |
2505 | { | |
2506 | #ifdef CONFIG_BLOCK | |
2507 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2508 | return false; |
b63534c4 | 2509 | |
55e6ac1e PB |
2510 | lockdep_assert_held(&req->ctx->uring_lock); |
2511 | ||
37d1e2e3 | 2512 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2513 | refcount_inc(&req->refs); |
2514 | io_queue_async_work(req); | |
b63534c4 | 2515 | return true; |
fdee946d | 2516 | } |
dc2a6e9a | 2517 | req_set_fail_links(req); |
b63534c4 JA |
2518 | #endif |
2519 | return false; | |
2520 | } | |
2521 | ||
a1d7c393 | 2522 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2523 | unsigned int issue_flags) |
a1d7c393 | 2524 | { |
2f8e45f1 PB |
2525 | int cflags = 0; |
2526 | ||
23faba36 PB |
2527 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2528 | return; | |
2f8e45f1 PB |
2529 | if (res != req->result) |
2530 | req_set_fail_links(req); | |
23faba36 | 2531 | |
2f8e45f1 PB |
2532 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2533 | kiocb_end_write(req); | |
2534 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2535 | cflags = io_put_rw_kbuf(req); | |
2536 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2537 | } |
2538 | ||
2539 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2540 | { | |
9adbd45d | 2541 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2542 | |
889fca73 | 2543 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2544 | } |
2545 | ||
def596e9 JA |
2546 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2547 | { | |
9adbd45d | 2548 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2549 | |
3e6a0d3c JA |
2550 | #ifdef CONFIG_BLOCK |
2551 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2552 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2553 | struct io_async_rw *rw = req->async_data; | |
2554 | ||
2555 | if (rw) | |
2556 | iov_iter_revert(&rw->iter, | |
2557 | req->result - iov_iter_count(&rw->iter)); | |
2558 | else if (!io_resubmit_prep(req)) | |
2559 | res = -EIO; | |
2560 | } | |
2561 | #endif | |
2562 | ||
491381ce JA |
2563 | if (kiocb->ki_flags & IOCB_WRITE) |
2564 | kiocb_end_write(req); | |
def596e9 | 2565 | |
2d7d6792 | 2566 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2567 | req_set_fail_links(req); |
bbde017a XW |
2568 | |
2569 | WRITE_ONCE(req->result, res); | |
2570 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2571 | smp_wmb(); |
2572 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2573 | } |
2574 | ||
2575 | /* | |
2576 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2577 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2578 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2579 | * accessing the kiocb cookie. | |
2580 | */ | |
2e9dbe90 | 2581 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2582 | { |
2583 | struct io_ring_ctx *ctx = req->ctx; | |
2584 | ||
2585 | /* | |
2586 | * Track whether we have multiple files in our lists. This will impact | |
2587 | * how we do polling eventually, not spinning if we're on potentially | |
2588 | * different devices. | |
2589 | */ | |
540e32a0 | 2590 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2591 | ctx->poll_multi_file = false; |
2592 | } else if (!ctx->poll_multi_file) { | |
2593 | struct io_kiocb *list_req; | |
2594 | ||
540e32a0 | 2595 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2596 | inflight_entry); |
9adbd45d | 2597 | if (list_req->file != req->file) |
def596e9 JA |
2598 | ctx->poll_multi_file = true; |
2599 | } | |
2600 | ||
2601 | /* | |
2602 | * For fast devices, IO may have already completed. If it has, add | |
2603 | * it to the front so we find it first. | |
2604 | */ | |
65a6543d | 2605 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2606 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2607 | else |
d21ffe7e | 2608 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2609 | |
2e9dbe90 XW |
2610 | /* |
2611 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2612 | * task context or in io worker task context. If current task context is | |
2613 | * sq thread, we don't need to check whether should wake up sq thread. | |
2614 | */ | |
2615 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2616 | wq_has_sleeper(&ctx->sq_data->wait)) |
2617 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2618 | } |
2619 | ||
9f13c35b PB |
2620 | static inline void io_state_file_put(struct io_submit_state *state) |
2621 | { | |
02b23a9a PB |
2622 | if (state->file_refs) { |
2623 | fput_many(state->file, state->file_refs); | |
2624 | state->file_refs = 0; | |
2625 | } | |
9a56a232 JA |
2626 | } |
2627 | ||
2628 | /* | |
2629 | * Get as many references to a file as we have IOs left in this submission, | |
2630 | * assuming most submissions are for one file, or at least that each file | |
2631 | * has more than one submission. | |
2632 | */ | |
8da11c19 | 2633 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2634 | { |
2635 | if (!state) | |
2636 | return fget(fd); | |
2637 | ||
6e1271e6 | 2638 | if (state->file_refs) { |
9a56a232 | 2639 | if (state->fd == fd) { |
6e1271e6 | 2640 | state->file_refs--; |
9a56a232 JA |
2641 | return state->file; |
2642 | } | |
02b23a9a | 2643 | io_state_file_put(state); |
9a56a232 JA |
2644 | } |
2645 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2646 | if (unlikely(!state->file)) |
9a56a232 JA |
2647 | return NULL; |
2648 | ||
2649 | state->fd = fd; | |
6e1271e6 | 2650 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2651 | return state->file; |
2652 | } | |
2653 | ||
4503b767 JA |
2654 | static bool io_bdev_nowait(struct block_device *bdev) |
2655 | { | |
9ba0d0c8 | 2656 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2657 | } |
2658 | ||
2b188cc1 JA |
2659 | /* |
2660 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2661 | * any file. For now, just ensure that anything potentially problematic is done | |
2662 | * inline. | |
2663 | */ | |
af197f50 | 2664 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2665 | { |
2666 | umode_t mode = file_inode(file)->i_mode; | |
2667 | ||
4503b767 | 2668 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2669 | if (IS_ENABLED(CONFIG_BLOCK) && |
2670 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2671 | return true; |
2672 | return false; | |
2673 | } | |
2674 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2675 | return true; |
4503b767 | 2676 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2677 | if (IS_ENABLED(CONFIG_BLOCK) && |
2678 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2679 | file->f_op != &io_uring_fops) |
2680 | return true; | |
2681 | return false; | |
2682 | } | |
2b188cc1 | 2683 | |
c5b85625 JA |
2684 | /* any ->read/write should understand O_NONBLOCK */ |
2685 | if (file->f_flags & O_NONBLOCK) | |
2686 | return true; | |
2687 | ||
af197f50 JA |
2688 | if (!(file->f_mode & FMODE_NOWAIT)) |
2689 | return false; | |
2690 | ||
2691 | if (rw == READ) | |
2692 | return file->f_op->read_iter != NULL; | |
2693 | ||
2694 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2695 | } |
2696 | ||
a88fc400 | 2697 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2698 | { |
def596e9 | 2699 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2700 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2701 | struct file *file = req->file; |
09bb8394 JA |
2702 | unsigned ioprio; |
2703 | int ret; | |
2b188cc1 | 2704 | |
75c668cd | 2705 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2706 | req->flags |= REQ_F_ISREG; |
2707 | ||
2b188cc1 | 2708 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2709 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2710 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2711 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2712 | } |
2b188cc1 | 2713 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2714 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2715 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2716 | if (unlikely(ret)) | |
2717 | return ret; | |
2b188cc1 | 2718 | |
75c668cd PB |
2719 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2720 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2721 | req->flags |= REQ_F_NOWAIT; | |
2722 | ||
2b188cc1 JA |
2723 | ioprio = READ_ONCE(sqe->ioprio); |
2724 | if (ioprio) { | |
2725 | ret = ioprio_check_cap(ioprio); | |
2726 | if (ret) | |
09bb8394 | 2727 | return ret; |
2b188cc1 JA |
2728 | |
2729 | kiocb->ki_ioprio = ioprio; | |
2730 | } else | |
2731 | kiocb->ki_ioprio = get_current_ioprio(); | |
2732 | ||
def596e9 | 2733 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2734 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2735 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2736 | return -EOPNOTSUPP; |
2b188cc1 | 2737 | |
def596e9 JA |
2738 | kiocb->ki_flags |= IOCB_HIPRI; |
2739 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2740 | req->iopoll_completed = 0; |
def596e9 | 2741 | } else { |
09bb8394 JA |
2742 | if (kiocb->ki_flags & IOCB_HIPRI) |
2743 | return -EINVAL; | |
def596e9 JA |
2744 | kiocb->ki_complete = io_complete_rw; |
2745 | } | |
9adbd45d | 2746 | |
3529d8c2 JA |
2747 | req->rw.addr = READ_ONCE(sqe->addr); |
2748 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2749 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2750 | return 0; |
2b188cc1 JA |
2751 | } |
2752 | ||
2753 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2754 | { | |
2755 | switch (ret) { | |
2756 | case -EIOCBQUEUED: | |
2757 | break; | |
2758 | case -ERESTARTSYS: | |
2759 | case -ERESTARTNOINTR: | |
2760 | case -ERESTARTNOHAND: | |
2761 | case -ERESTART_RESTARTBLOCK: | |
2762 | /* | |
2763 | * We can't just restart the syscall, since previously | |
2764 | * submitted sqes may already be in progress. Just fail this | |
2765 | * IO with EINTR. | |
2766 | */ | |
2767 | ret = -EINTR; | |
df561f66 | 2768 | fallthrough; |
2b188cc1 JA |
2769 | default: |
2770 | kiocb->ki_complete(kiocb, ret, 0); | |
2771 | } | |
2772 | } | |
2773 | ||
a1d7c393 | 2774 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2775 | unsigned int issue_flags) |
ba816ad6 | 2776 | { |
ba04291e | 2777 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2778 | struct io_async_rw *io = req->async_data; |
ba04291e | 2779 | |
227c0c96 | 2780 | /* add previously done IO, if any */ |
e8c2bc1f | 2781 | if (io && io->bytes_done > 0) { |
227c0c96 | 2782 | if (ret < 0) |
e8c2bc1f | 2783 | ret = io->bytes_done; |
227c0c96 | 2784 | else |
e8c2bc1f | 2785 | ret += io->bytes_done; |
227c0c96 JA |
2786 | } |
2787 | ||
ba04291e JA |
2788 | if (req->flags & REQ_F_CUR_POS) |
2789 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2790 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2791 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2792 | else |
2793 | io_rw_done(kiocb, ret); | |
2794 | } | |
2795 | ||
847595de | 2796 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2797 | { |
9adbd45d JA |
2798 | struct io_ring_ctx *ctx = req->ctx; |
2799 | size_t len = req->rw.len; | |
edafccee | 2800 | struct io_mapped_ubuf *imu; |
4be1c615 | 2801 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2802 | size_t offset; |
2803 | u64 buf_addr; | |
2804 | ||
edafccee JA |
2805 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2806 | return -EFAULT; | |
edafccee JA |
2807 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2808 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2809 | buf_addr = req->rw.addr; |
edafccee JA |
2810 | |
2811 | /* overflow */ | |
2812 | if (buf_addr + len < buf_addr) | |
2813 | return -EFAULT; | |
2814 | /* not inside the mapped region */ | |
2815 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2816 | return -EFAULT; | |
2817 | ||
2818 | /* | |
2819 | * May not be a start of buffer, set size appropriately | |
2820 | * and advance us to the beginning. | |
2821 | */ | |
2822 | offset = buf_addr - imu->ubuf; | |
2823 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2824 | |
2825 | if (offset) { | |
2826 | /* | |
2827 | * Don't use iov_iter_advance() here, as it's really slow for | |
2828 | * using the latter parts of a big fixed buffer - it iterates | |
2829 | * over each segment manually. We can cheat a bit here, because | |
2830 | * we know that: | |
2831 | * | |
2832 | * 1) it's a BVEC iter, we set it up | |
2833 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2834 | * first and last bvec | |
2835 | * | |
2836 | * So just find our index, and adjust the iterator afterwards. | |
2837 | * If the offset is within the first bvec (or the whole first | |
2838 | * bvec, just use iov_iter_advance(). This makes it easier | |
2839 | * since we can just skip the first segment, which may not | |
2840 | * be PAGE_SIZE aligned. | |
2841 | */ | |
2842 | const struct bio_vec *bvec = imu->bvec; | |
2843 | ||
2844 | if (offset <= bvec->bv_len) { | |
2845 | iov_iter_advance(iter, offset); | |
2846 | } else { | |
2847 | unsigned long seg_skip; | |
2848 | ||
2849 | /* skip first vec */ | |
2850 | offset -= bvec->bv_len; | |
2851 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2852 | ||
2853 | iter->bvec = bvec + seg_skip; | |
2854 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2855 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2856 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2857 | } |
2858 | } | |
2859 | ||
847595de | 2860 | return 0; |
edafccee JA |
2861 | } |
2862 | ||
bcda7baa JA |
2863 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2864 | { | |
2865 | if (needs_lock) | |
2866 | mutex_unlock(&ctx->uring_lock); | |
2867 | } | |
2868 | ||
2869 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2870 | { | |
2871 | /* | |
2872 | * "Normal" inline submissions always hold the uring_lock, since we | |
2873 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2874 | * The only exception is when we've detached the request and issue it | |
2875 | * from an async worker thread, grab the lock for that case. | |
2876 | */ | |
2877 | if (needs_lock) | |
2878 | mutex_lock(&ctx->uring_lock); | |
2879 | } | |
2880 | ||
2881 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2882 | int bgid, struct io_buffer *kbuf, | |
2883 | bool needs_lock) | |
2884 | { | |
2885 | struct io_buffer *head; | |
2886 | ||
2887 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2888 | return kbuf; | |
2889 | ||
2890 | io_ring_submit_lock(req->ctx, needs_lock); | |
2891 | ||
2892 | lockdep_assert_held(&req->ctx->uring_lock); | |
2893 | ||
9e15c3a0 | 2894 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2895 | if (head) { |
2896 | if (!list_empty(&head->list)) { | |
2897 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2898 | list); | |
2899 | list_del(&kbuf->list); | |
2900 | } else { | |
2901 | kbuf = head; | |
9e15c3a0 | 2902 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2903 | } |
2904 | if (*len > kbuf->len) | |
2905 | *len = kbuf->len; | |
2906 | } else { | |
2907 | kbuf = ERR_PTR(-ENOBUFS); | |
2908 | } | |
2909 | ||
2910 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2911 | ||
2912 | return kbuf; | |
2913 | } | |
2914 | ||
4d954c25 JA |
2915 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2916 | bool needs_lock) | |
2917 | { | |
2918 | struct io_buffer *kbuf; | |
4f4eeba8 | 2919 | u16 bgid; |
4d954c25 JA |
2920 | |
2921 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2922 | bgid = req->buf_index; |
4d954c25 JA |
2923 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2924 | if (IS_ERR(kbuf)) | |
2925 | return kbuf; | |
2926 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2927 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2928 | return u64_to_user_ptr(kbuf->addr); | |
2929 | } | |
2930 | ||
2931 | #ifdef CONFIG_COMPAT | |
2932 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2933 | bool needs_lock) | |
2934 | { | |
2935 | struct compat_iovec __user *uiov; | |
2936 | compat_ssize_t clen; | |
2937 | void __user *buf; | |
2938 | ssize_t len; | |
2939 | ||
2940 | uiov = u64_to_user_ptr(req->rw.addr); | |
2941 | if (!access_ok(uiov, sizeof(*uiov))) | |
2942 | return -EFAULT; | |
2943 | if (__get_user(clen, &uiov->iov_len)) | |
2944 | return -EFAULT; | |
2945 | if (clen < 0) | |
2946 | return -EINVAL; | |
2947 | ||
2948 | len = clen; | |
2949 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2950 | if (IS_ERR(buf)) | |
2951 | return PTR_ERR(buf); | |
2952 | iov[0].iov_base = buf; | |
2953 | iov[0].iov_len = (compat_size_t) len; | |
2954 | return 0; | |
2955 | } | |
2956 | #endif | |
2957 | ||
2958 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2959 | bool needs_lock) | |
2960 | { | |
2961 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2962 | void __user *buf; | |
2963 | ssize_t len; | |
2964 | ||
2965 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2966 | return -EFAULT; | |
2967 | ||
2968 | len = iov[0].iov_len; | |
2969 | if (len < 0) | |
2970 | return -EINVAL; | |
2971 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2972 | if (IS_ERR(buf)) | |
2973 | return PTR_ERR(buf); | |
2974 | iov[0].iov_base = buf; | |
2975 | iov[0].iov_len = len; | |
2976 | return 0; | |
2977 | } | |
2978 | ||
2979 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2980 | bool needs_lock) | |
2981 | { | |
dddb3e26 JA |
2982 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2983 | struct io_buffer *kbuf; | |
2984 | ||
2985 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2986 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2987 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2988 | return 0; |
dddb3e26 | 2989 | } |
dd201662 | 2990 | if (req->rw.len != 1) |
4d954c25 JA |
2991 | return -EINVAL; |
2992 | ||
2993 | #ifdef CONFIG_COMPAT | |
2994 | if (req->ctx->compat) | |
2995 | return io_compat_import(req, iov, needs_lock); | |
2996 | #endif | |
2997 | ||
2998 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2999 | } | |
3000 | ||
847595de PB |
3001 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3002 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3003 | { |
9adbd45d JA |
3004 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3005 | size_t sqe_len = req->rw.len; | |
847595de | 3006 | u8 opcode = req->opcode; |
4d954c25 | 3007 | ssize_t ret; |
edafccee | 3008 | |
7d009165 | 3009 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3010 | *iovec = NULL; |
9adbd45d | 3011 | return io_import_fixed(req, rw, iter); |
edafccee | 3012 | } |
2b188cc1 | 3013 | |
bcda7baa | 3014 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3015 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3016 | return -EINVAL; |
3017 | ||
3a6820f2 | 3018 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3019 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3020 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3021 | if (IS_ERR(buf)) |
4d954c25 | 3022 | return PTR_ERR(buf); |
3f9d6441 | 3023 | req->rw.len = sqe_len; |
bcda7baa JA |
3024 | } |
3025 | ||
3a6820f2 JA |
3026 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3027 | *iovec = NULL; | |
10fc72e4 | 3028 | return ret; |
3a6820f2 JA |
3029 | } |
3030 | ||
4d954c25 JA |
3031 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3032 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3033 | if (!ret) |
3034 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3035 | *iovec = NULL; |
3036 | return ret; | |
3037 | } | |
3038 | ||
89cd35c5 CH |
3039 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3040 | req->ctx->compat); | |
2b188cc1 JA |
3041 | } |
3042 | ||
0fef9483 JA |
3043 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3044 | { | |
5b09e37e | 3045 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3046 | } |
3047 | ||
31b51510 | 3048 | /* |
32960613 JA |
3049 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3050 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3051 | */ |
4017eb91 | 3052 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3053 | { |
4017eb91 JA |
3054 | struct kiocb *kiocb = &req->rw.kiocb; |
3055 | struct file *file = req->file; | |
32960613 JA |
3056 | ssize_t ret = 0; |
3057 | ||
3058 | /* | |
3059 | * Don't support polled IO through this interface, and we can't | |
3060 | * support non-blocking either. For the latter, this just causes | |
3061 | * the kiocb to be handled from an async context. | |
3062 | */ | |
3063 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3064 | return -EOPNOTSUPP; | |
3065 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3066 | return -EAGAIN; | |
3067 | ||
3068 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3069 | struct iovec iovec; |
32960613 JA |
3070 | ssize_t nr; |
3071 | ||
311ae9e1 PB |
3072 | if (!iov_iter_is_bvec(iter)) { |
3073 | iovec = iov_iter_iovec(iter); | |
3074 | } else { | |
4017eb91 JA |
3075 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3076 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3077 | } |
3078 | ||
32960613 JA |
3079 | if (rw == READ) { |
3080 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3081 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3082 | } else { |
3083 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3084 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3085 | } |
3086 | ||
3087 | if (nr < 0) { | |
3088 | if (!ret) | |
3089 | ret = nr; | |
3090 | break; | |
3091 | } | |
3092 | ret += nr; | |
3093 | if (nr != iovec.iov_len) | |
3094 | break; | |
4017eb91 JA |
3095 | req->rw.len -= nr; |
3096 | req->rw.addr += nr; | |
32960613 JA |
3097 | iov_iter_advance(iter, nr); |
3098 | } | |
3099 | ||
3100 | return ret; | |
3101 | } | |
3102 | ||
ff6165b2 JA |
3103 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3104 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3105 | { |
e8c2bc1f | 3106 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3107 | |
ff6165b2 | 3108 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3109 | rw->free_iovec = iovec; |
227c0c96 | 3110 | rw->bytes_done = 0; |
ff6165b2 | 3111 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3112 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3113 | return; |
b64e3444 | 3114 | if (!iovec) { |
ff6165b2 JA |
3115 | unsigned iov_off = 0; |
3116 | ||
3117 | rw->iter.iov = rw->fast_iov; | |
3118 | if (iter->iov != fast_iov) { | |
3119 | iov_off = iter->iov - fast_iov; | |
3120 | rw->iter.iov += iov_off; | |
3121 | } | |
3122 | if (rw->fast_iov != fast_iov) | |
3123 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3124 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3125 | } else { |
3126 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3127 | } |
3128 | } | |
3129 | ||
e8c2bc1f | 3130 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3131 | { |
e8c2bc1f JA |
3132 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3133 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3134 | return req->async_data == NULL; | |
3d9932a8 XW |
3135 | } |
3136 | ||
e8c2bc1f | 3137 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3138 | { |
e8c2bc1f | 3139 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3140 | return 0; |
3d9932a8 | 3141 | |
e8c2bc1f | 3142 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3143 | } |
3144 | ||
ff6165b2 JA |
3145 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3146 | const struct iovec *fast_iov, | |
227c0c96 | 3147 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3148 | { |
e8c2bc1f | 3149 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3150 | return 0; |
e8c2bc1f | 3151 | if (!req->async_data) { |
6bf985dc PB |
3152 | if (__io_alloc_async_data(req)) { |
3153 | kfree(iovec); | |
5d204bcf | 3154 | return -ENOMEM; |
6bf985dc | 3155 | } |
b7bb4f7d | 3156 | |
ff6165b2 | 3157 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3158 | } |
b7bb4f7d | 3159 | return 0; |
f67676d1 JA |
3160 | } |
3161 | ||
73debe68 | 3162 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3163 | { |
e8c2bc1f | 3164 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3165 | struct iovec *iov = iorw->fast_iov; |
847595de | 3166 | int ret; |
c3e330a4 | 3167 | |
2846c481 | 3168 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3169 | if (unlikely(ret < 0)) |
3170 | return ret; | |
3171 | ||
ab0b196c PB |
3172 | iorw->bytes_done = 0; |
3173 | iorw->free_iovec = iov; | |
3174 | if (iov) | |
3175 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3176 | return 0; |
3177 | } | |
3178 | ||
73debe68 | 3179 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3180 | { |
3529d8c2 JA |
3181 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3182 | return -EBADF; | |
93642ef8 | 3183 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3184 | } |
3185 | ||
c1dd91d1 JA |
3186 | /* |
3187 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3188 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3189 | * This gets called when the page is unlocked, and we generally expect that to | |
3190 | * happen when the page IO is completed and the page is now uptodate. This will | |
3191 | * queue a task_work based retry of the operation, attempting to copy the data | |
3192 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3193 | * do a thread based blocking retry of the operation. That's the unexpected | |
3194 | * slow path. | |
3195 | */ | |
bcf5a063 JA |
3196 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3197 | int sync, void *arg) | |
3198 | { | |
3199 | struct wait_page_queue *wpq; | |
3200 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3201 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3202 | |
3203 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3204 | ||
cdc8fcb4 LT |
3205 | if (!wake_page_match(wpq, key)) |
3206 | return 0; | |
3207 | ||
c8d317aa | 3208 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3209 | list_del_init(&wait->entry); |
3210 | ||
bcf5a063 JA |
3211 | /* submit ref gets dropped, acquire a new one */ |
3212 | refcount_inc(&req->refs); | |
921b9054 | 3213 | io_req_task_queue(req); |
bcf5a063 JA |
3214 | return 1; |
3215 | } | |
3216 | ||
c1dd91d1 JA |
3217 | /* |
3218 | * This controls whether a given IO request should be armed for async page | |
3219 | * based retry. If we return false here, the request is handed to the async | |
3220 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3221 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3222 | * will either succeed because the page is now uptodate and unlocked, or it | |
3223 | * will register a callback when the page is unlocked at IO completion. Through | |
3224 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3225 | * That retry will attempt the buffered read again. The retry will generally | |
3226 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3227 | * async worker threads for a blocking retry. | |
3228 | */ | |
227c0c96 | 3229 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3230 | { |
e8c2bc1f JA |
3231 | struct io_async_rw *rw = req->async_data; |
3232 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3233 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3234 | |
bcf5a063 JA |
3235 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3236 | if (req->flags & REQ_F_NOWAIT) | |
3237 | return false; | |
f67676d1 | 3238 | |
227c0c96 | 3239 | /* Only for buffered IO */ |
3b2a4439 | 3240 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3241 | return false; |
3b2a4439 | 3242 | |
bcf5a063 JA |
3243 | /* |
3244 | * just use poll if we can, and don't attempt if the fs doesn't | |
3245 | * support callback based unlocks | |
3246 | */ | |
3247 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3248 | return false; | |
f67676d1 | 3249 | |
3b2a4439 JA |
3250 | wait->wait.func = io_async_buf_func; |
3251 | wait->wait.private = req; | |
3252 | wait->wait.flags = 0; | |
3253 | INIT_LIST_HEAD(&wait->wait.entry); | |
3254 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3255 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3256 | kiocb->ki_waitq = wait; |
3b2a4439 | 3257 | return true; |
bcf5a063 JA |
3258 | } |
3259 | ||
3260 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3261 | { | |
3262 | if (req->file->f_op->read_iter) | |
3263 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3264 | else if (req->file->f_op->read) |
4017eb91 | 3265 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3266 | else |
3267 | return -EINVAL; | |
f67676d1 JA |
3268 | } |
3269 | ||
889fca73 | 3270 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3271 | { |
3272 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3273 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3274 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3275 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3276 | ssize_t io_size, ret, ret2; |
45d189c6 | 3277 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3278 | |
2846c481 | 3279 | if (rw) { |
e8c2bc1f | 3280 | iter = &rw->iter; |
2846c481 PB |
3281 | iovec = NULL; |
3282 | } else { | |
3283 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3284 | if (ret < 0) | |
3285 | return ret; | |
3286 | } | |
632546c4 | 3287 | io_size = iov_iter_count(iter); |
fa15bafb | 3288 | req->result = io_size; |
2b188cc1 | 3289 | |
fd6c2e4c JA |
3290 | /* Ensure we clear previously set non-block flag */ |
3291 | if (!force_nonblock) | |
29de5f6a | 3292 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3293 | else |
3294 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3295 | ||
24c74678 | 3296 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3297 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3298 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3299 | return ret ?: -EAGAIN; |
6713e7a6 | 3300 | } |
9e645e11 | 3301 | |
632546c4 | 3302 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3303 | if (unlikely(ret)) { |
3304 | kfree(iovec); | |
3305 | return ret; | |
3306 | } | |
2b188cc1 | 3307 | |
227c0c96 | 3308 | ret = io_iter_do_read(req, iter); |
32960613 | 3309 | |
57cd657b | 3310 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3311 | if (req->async_data) |
3312 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3313 | goto out_free; |
227c0c96 | 3314 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3315 | /* IOPOLL retry should happen for io-wq threads */ |
3316 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3317 | goto done; |
75c668cd PB |
3318 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3319 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3320 | goto done; |
84216315 | 3321 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3322 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3323 | ret = 0; |
7335e3bf | 3324 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3325 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3326 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3327 | goto done; |
227c0c96 JA |
3328 | } |
3329 | ||
227c0c96 | 3330 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3331 | if (ret2) |
3332 | return ret2; | |
3333 | ||
fe1cdd55 | 3334 | iovec = NULL; |
e8c2bc1f | 3335 | rw = req->async_data; |
227c0c96 | 3336 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3337 | iter = &rw->iter; |
227c0c96 | 3338 | |
b23df91b PB |
3339 | do { |
3340 | io_size -= ret; | |
3341 | rw->bytes_done += ret; | |
3342 | /* if we can retry, do so with the callbacks armed */ | |
3343 | if (!io_rw_should_retry(req)) { | |
3344 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3345 | return -EAGAIN; | |
3346 | } | |
3347 | ||
3348 | /* | |
3349 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3350 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3351 | * desired page gets unlocked. We can also get a partial read | |
3352 | * here, and if we do, then just retry at the new offset. | |
3353 | */ | |
3354 | ret = io_iter_do_read(req, iter); | |
3355 | if (ret == -EIOCBQUEUED) | |
3356 | return 0; | |
227c0c96 | 3357 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3358 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3359 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3360 | done: |
889fca73 | 3361 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3362 | out_free: |
3363 | /* it's faster to check here then delegate to kfree */ | |
3364 | if (iovec) | |
3365 | kfree(iovec); | |
5ea5dd45 | 3366 | return 0; |
2b188cc1 JA |
3367 | } |
3368 | ||
73debe68 | 3369 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3370 | { |
3529d8c2 JA |
3371 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3372 | return -EBADF; | |
93642ef8 | 3373 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3374 | } |
3375 | ||
889fca73 | 3376 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3377 | { |
3378 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3379 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3380 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3381 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3382 | ssize_t ret, ret2, io_size; |
45d189c6 | 3383 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3384 | |
2846c481 | 3385 | if (rw) { |
e8c2bc1f | 3386 | iter = &rw->iter; |
2846c481 PB |
3387 | iovec = NULL; |
3388 | } else { | |
3389 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3390 | if (ret < 0) | |
3391 | return ret; | |
3392 | } | |
632546c4 | 3393 | io_size = iov_iter_count(iter); |
fa15bafb | 3394 | req->result = io_size; |
2b188cc1 | 3395 | |
fd6c2e4c JA |
3396 | /* Ensure we clear previously set non-block flag */ |
3397 | if (!force_nonblock) | |
a88fc400 PB |
3398 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3399 | else | |
3400 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3401 | |
24c74678 | 3402 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3403 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3404 | goto copy_iov; |
31b51510 | 3405 | |
10d59345 JA |
3406 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3407 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3408 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3409 | goto copy_iov; |
31b51510 | 3410 | |
632546c4 | 3411 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3412 | if (unlikely(ret)) |
3413 | goto out_free; | |
4ed734b0 | 3414 | |
fa15bafb PB |
3415 | /* |
3416 | * Open-code file_start_write here to grab freeze protection, | |
3417 | * which will be released by another thread in | |
3418 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3419 | * released so that it doesn't complain about the held lock when | |
3420 | * we return to userspace. | |
3421 | */ | |
3422 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3423 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3424 | __sb_writers_release(file_inode(req->file)->i_sb, |
3425 | SB_FREEZE_WRITE); | |
3426 | } | |
3427 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3428 | |
fa15bafb | 3429 | if (req->file->f_op->write_iter) |
ff6165b2 | 3430 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3431 | else if (req->file->f_op->write) |
4017eb91 | 3432 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3433 | else |
3434 | ret2 = -EINVAL; | |
4ed734b0 | 3435 | |
fa15bafb PB |
3436 | /* |
3437 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3438 | * retry them without IOCB_NOWAIT. | |
3439 | */ | |
3440 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3441 | ret2 = -EAGAIN; | |
75c668cd PB |
3442 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3443 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3444 | goto done; |
3e6a0d3c JA |
3445 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3446 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3447 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3448 | /* IOPOLL retry should happen for io-wq threads */ |
3449 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3450 | goto copy_iov; | |
355afaeb | 3451 | done: |
889fca73 | 3452 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3453 | } else { |
f67676d1 | 3454 | copy_iov: |
84216315 | 3455 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3456 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3457 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3458 | return ret ?: -EAGAIN; |
2b188cc1 | 3459 | } |
31b51510 | 3460 | out_free: |
f261c168 | 3461 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3462 | if (iovec) |
6f2cc166 | 3463 | kfree(iovec); |
2b188cc1 JA |
3464 | return ret; |
3465 | } | |
3466 | ||
80a261fd JA |
3467 | static int io_renameat_prep(struct io_kiocb *req, |
3468 | const struct io_uring_sqe *sqe) | |
3469 | { | |
3470 | struct io_rename *ren = &req->rename; | |
3471 | const char __user *oldf, *newf; | |
3472 | ||
3473 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3474 | return -EBADF; | |
3475 | ||
3476 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3477 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3478 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3479 | ren->new_dfd = READ_ONCE(sqe->len); | |
3480 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3481 | ||
3482 | ren->oldpath = getname(oldf); | |
3483 | if (IS_ERR(ren->oldpath)) | |
3484 | return PTR_ERR(ren->oldpath); | |
3485 | ||
3486 | ren->newpath = getname(newf); | |
3487 | if (IS_ERR(ren->newpath)) { | |
3488 | putname(ren->oldpath); | |
3489 | return PTR_ERR(ren->newpath); | |
3490 | } | |
3491 | ||
3492 | req->flags |= REQ_F_NEED_CLEANUP; | |
3493 | return 0; | |
3494 | } | |
3495 | ||
45d189c6 | 3496 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3497 | { |
3498 | struct io_rename *ren = &req->rename; | |
3499 | int ret; | |
3500 | ||
45d189c6 | 3501 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3502 | return -EAGAIN; |
3503 | ||
3504 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3505 | ren->newpath, ren->flags); | |
3506 | ||
3507 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3508 | if (ret < 0) | |
3509 | req_set_fail_links(req); | |
3510 | io_req_complete(req, ret); | |
3511 | return 0; | |
3512 | } | |
3513 | ||
14a1143b JA |
3514 | static int io_unlinkat_prep(struct io_kiocb *req, |
3515 | const struct io_uring_sqe *sqe) | |
3516 | { | |
3517 | struct io_unlink *un = &req->unlink; | |
3518 | const char __user *fname; | |
3519 | ||
3520 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3521 | return -EBADF; | |
3522 | ||
3523 | un->dfd = READ_ONCE(sqe->fd); | |
3524 | ||
3525 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3526 | if (un->flags & ~AT_REMOVEDIR) | |
3527 | return -EINVAL; | |
3528 | ||
3529 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3530 | un->filename = getname(fname); | |
3531 | if (IS_ERR(un->filename)) | |
3532 | return PTR_ERR(un->filename); | |
3533 | ||
3534 | req->flags |= REQ_F_NEED_CLEANUP; | |
3535 | return 0; | |
3536 | } | |
3537 | ||
45d189c6 | 3538 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3539 | { |
3540 | struct io_unlink *un = &req->unlink; | |
3541 | int ret; | |
3542 | ||
45d189c6 | 3543 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3544 | return -EAGAIN; |
3545 | ||
3546 | if (un->flags & AT_REMOVEDIR) | |
3547 | ret = do_rmdir(un->dfd, un->filename); | |
3548 | else | |
3549 | ret = do_unlinkat(un->dfd, un->filename); | |
3550 | ||
3551 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3552 | if (ret < 0) | |
3553 | req_set_fail_links(req); | |
3554 | io_req_complete(req, ret); | |
3555 | return 0; | |
3556 | } | |
3557 | ||
36f4fa68 JA |
3558 | static int io_shutdown_prep(struct io_kiocb *req, |
3559 | const struct io_uring_sqe *sqe) | |
3560 | { | |
3561 | #if defined(CONFIG_NET) | |
3562 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3563 | return -EINVAL; | |
3564 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3565 | sqe->buf_index) | |
3566 | return -EINVAL; | |
3567 | ||
3568 | req->shutdown.how = READ_ONCE(sqe->len); | |
3569 | return 0; | |
3570 | #else | |
3571 | return -EOPNOTSUPP; | |
3572 | #endif | |
3573 | } | |
3574 | ||
45d189c6 | 3575 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3576 | { |
3577 | #if defined(CONFIG_NET) | |
3578 | struct socket *sock; | |
3579 | int ret; | |
3580 | ||
45d189c6 | 3581 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3582 | return -EAGAIN; |
3583 | ||
48aba79b | 3584 | sock = sock_from_file(req->file); |
36f4fa68 | 3585 | if (unlikely(!sock)) |
48aba79b | 3586 | return -ENOTSOCK; |
36f4fa68 JA |
3587 | |
3588 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3589 | if (ret < 0) |
3590 | req_set_fail_links(req); | |
36f4fa68 JA |
3591 | io_req_complete(req, ret); |
3592 | return 0; | |
3593 | #else | |
3594 | return -EOPNOTSUPP; | |
3595 | #endif | |
3596 | } | |
3597 | ||
f2a8d5c7 PB |
3598 | static int __io_splice_prep(struct io_kiocb *req, |
3599 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3600 | { |
3601 | struct io_splice* sp = &req->splice; | |
3602 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3603 | |
3232dd02 PB |
3604 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3605 | return -EINVAL; | |
7d67af2c PB |
3606 | |
3607 | sp->file_in = NULL; | |
7d67af2c PB |
3608 | sp->len = READ_ONCE(sqe->len); |
3609 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3610 | ||
3611 | if (unlikely(sp->flags & ~valid_flags)) | |
3612 | return -EINVAL; | |
3613 | ||
8371adf5 PB |
3614 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3615 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3616 | if (!sp->file_in) | |
3617 | return -EBADF; | |
7d67af2c PB |
3618 | req->flags |= REQ_F_NEED_CLEANUP; |
3619 | ||
7cdaf587 XW |
3620 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3621 | /* | |
3622 | * Splice operation will be punted aync, and here need to | |
3623 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3624 | */ | |
7d67af2c | 3625 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3626 | } |
7d67af2c PB |
3627 | |
3628 | return 0; | |
3629 | } | |
3630 | ||
f2a8d5c7 PB |
3631 | static int io_tee_prep(struct io_kiocb *req, |
3632 | const struct io_uring_sqe *sqe) | |
3633 | { | |
3634 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3635 | return -EINVAL; | |
3636 | return __io_splice_prep(req, sqe); | |
3637 | } | |
3638 | ||
45d189c6 | 3639 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3640 | { |
3641 | struct io_splice *sp = &req->splice; | |
3642 | struct file *in = sp->file_in; | |
3643 | struct file *out = sp->file_out; | |
3644 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3645 | long ret = 0; | |
3646 | ||
45d189c6 | 3647 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3648 | return -EAGAIN; |
3649 | if (sp->len) | |
3650 | ret = do_tee(in, out, sp->len, flags); | |
3651 | ||
3652 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3653 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3654 | ||
f2a8d5c7 PB |
3655 | if (ret != sp->len) |
3656 | req_set_fail_links(req); | |
e1e16097 | 3657 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3658 | return 0; |
3659 | } | |
3660 | ||
3661 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3662 | { | |
3663 | struct io_splice* sp = &req->splice; | |
3664 | ||
3665 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3666 | sp->off_out = READ_ONCE(sqe->off); | |
3667 | return __io_splice_prep(req, sqe); | |
3668 | } | |
3669 | ||
45d189c6 | 3670 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3671 | { |
3672 | struct io_splice *sp = &req->splice; | |
3673 | struct file *in = sp->file_in; | |
3674 | struct file *out = sp->file_out; | |
3675 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3676 | loff_t *poff_in, *poff_out; | |
c9687426 | 3677 | long ret = 0; |
7d67af2c | 3678 | |
45d189c6 | 3679 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3680 | return -EAGAIN; |
7d67af2c PB |
3681 | |
3682 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3683 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3684 | |
948a7749 | 3685 | if (sp->len) |
c9687426 | 3686 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3687 | |
3688 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3689 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3690 | ||
7d67af2c PB |
3691 | if (ret != sp->len) |
3692 | req_set_fail_links(req); | |
e1e16097 | 3693 | io_req_complete(req, ret); |
7d67af2c PB |
3694 | return 0; |
3695 | } | |
3696 | ||
2b188cc1 JA |
3697 | /* |
3698 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3699 | */ | |
889fca73 | 3700 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3701 | { |
3702 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3703 | |
def596e9 JA |
3704 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3705 | return -EINVAL; | |
3706 | ||
889fca73 | 3707 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3708 | return 0; |
3709 | } | |
3710 | ||
1155c76a | 3711 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3712 | { |
6b06314c | 3713 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3714 | |
09bb8394 JA |
3715 | if (!req->file) |
3716 | return -EBADF; | |
c992fe29 | 3717 | |
6b06314c | 3718 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3719 | return -EINVAL; |
edafccee | 3720 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3721 | return -EINVAL; |
3722 | ||
8ed8d3c3 JA |
3723 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3724 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3725 | return -EINVAL; | |
3726 | ||
3727 | req->sync.off = READ_ONCE(sqe->off); | |
3728 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3729 | return 0; |
3730 | } | |
3731 | ||
45d189c6 | 3732 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3733 | { |
8ed8d3c3 | 3734 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3735 | int ret; |
3736 | ||
ac45abc0 | 3737 | /* fsync always requires a blocking context */ |
45d189c6 | 3738 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3739 | return -EAGAIN; |
3740 | ||
9adbd45d | 3741 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3742 | end > 0 ? end : LLONG_MAX, |
3743 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3744 | if (ret < 0) | |
3745 | req_set_fail_links(req); | |
e1e16097 | 3746 | io_req_complete(req, ret); |
c992fe29 CH |
3747 | return 0; |
3748 | } | |
3749 | ||
d63d1b5e JA |
3750 | static int io_fallocate_prep(struct io_kiocb *req, |
3751 | const struct io_uring_sqe *sqe) | |
3752 | { | |
3753 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3754 | return -EINVAL; | |
3232dd02 PB |
3755 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3756 | return -EINVAL; | |
d63d1b5e JA |
3757 | |
3758 | req->sync.off = READ_ONCE(sqe->off); | |
3759 | req->sync.len = READ_ONCE(sqe->addr); | |
3760 | req->sync.mode = READ_ONCE(sqe->len); | |
3761 | return 0; | |
3762 | } | |
3763 | ||
45d189c6 | 3764 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3765 | { |
ac45abc0 PB |
3766 | int ret; |
3767 | ||
d63d1b5e | 3768 | /* fallocate always requiring blocking context */ |
45d189c6 | 3769 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3770 | return -EAGAIN; |
ac45abc0 PB |
3771 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3772 | req->sync.len); | |
ac45abc0 PB |
3773 | if (ret < 0) |
3774 | req_set_fail_links(req); | |
e1e16097 | 3775 | io_req_complete(req, ret); |
5d17b4a4 JA |
3776 | return 0; |
3777 | } | |
3778 | ||
ec65fea5 | 3779 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3780 | { |
f8748881 | 3781 | const char __user *fname; |
15b71abe | 3782 | int ret; |
b7bb4f7d | 3783 | |
ec65fea5 | 3784 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3785 | return -EINVAL; |
ec65fea5 | 3786 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3787 | return -EBADF; |
03b1230c | 3788 | |
ec65fea5 PB |
3789 | /* open.how should be already initialised */ |
3790 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3791 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3792 | |
25e72d10 PB |
3793 | req->open.dfd = READ_ONCE(sqe->fd); |
3794 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3795 | req->open.filename = getname(fname); |
15b71abe JA |
3796 | if (IS_ERR(req->open.filename)) { |
3797 | ret = PTR_ERR(req->open.filename); | |
3798 | req->open.filename = NULL; | |
3799 | return ret; | |
3800 | } | |
4022e7af | 3801 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3802 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3803 | return 0; |
03b1230c JA |
3804 | } |
3805 | ||
ec65fea5 PB |
3806 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3807 | { | |
3808 | u64 flags, mode; | |
3809 | ||
14587a46 | 3810 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3811 | return -EINVAL; |
ec65fea5 PB |
3812 | mode = READ_ONCE(sqe->len); |
3813 | flags = READ_ONCE(sqe->open_flags); | |
3814 | req->open.how = build_open_how(flags, mode); | |
3815 | return __io_openat_prep(req, sqe); | |
3816 | } | |
3817 | ||
cebdb986 | 3818 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3819 | { |
cebdb986 | 3820 | struct open_how __user *how; |
cebdb986 | 3821 | size_t len; |
0fa03c62 JA |
3822 | int ret; |
3823 | ||
14587a46 | 3824 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3825 | return -EINVAL; |
cebdb986 JA |
3826 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3827 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3828 | if (len < OPEN_HOW_SIZE_VER0) |
3829 | return -EINVAL; | |
3529d8c2 | 3830 | |
cebdb986 JA |
3831 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3832 | len); | |
3833 | if (ret) | |
3834 | return ret; | |
3529d8c2 | 3835 | |
ec65fea5 | 3836 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3837 | } |
3838 | ||
45d189c6 | 3839 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3840 | { |
3841 | struct open_flags op; | |
15b71abe | 3842 | struct file *file; |
3a81fd02 JA |
3843 | bool nonblock_set; |
3844 | bool resolve_nonblock; | |
15b71abe JA |
3845 | int ret; |
3846 | ||
cebdb986 | 3847 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3848 | if (ret) |
3849 | goto err; | |
3a81fd02 JA |
3850 | nonblock_set = op.open_flag & O_NONBLOCK; |
3851 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3852 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3853 | /* |
3854 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3855 | * it'll always -EAGAIN | |
3856 | */ | |
3857 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3858 | return -EAGAIN; | |
3859 | op.lookup_flags |= LOOKUP_CACHED; | |
3860 | op.open_flag |= O_NONBLOCK; | |
3861 | } | |
15b71abe | 3862 | |
4022e7af | 3863 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3864 | if (ret < 0) |
3865 | goto err; | |
3866 | ||
3867 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3868 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3869 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3870 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3871 | /* |
3a81fd02 JA |
3872 | * We could hang on to this 'fd', but seems like marginal |
3873 | * gain for something that is now known to be a slower path. | |
3874 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3875 | */ |
3a81fd02 JA |
3876 | put_unused_fd(ret); |
3877 | return -EAGAIN; | |
3878 | } | |
3879 | ||
15b71abe JA |
3880 | if (IS_ERR(file)) { |
3881 | put_unused_fd(ret); | |
3882 | ret = PTR_ERR(file); | |
3883 | } else { | |
45d189c6 | 3884 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3885 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3886 | fsnotify_open(file); |
3887 | fd_install(ret, file); | |
3888 | } | |
3889 | err: | |
3890 | putname(req->open.filename); | |
8fef80bf | 3891 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3892 | if (ret < 0) |
3893 | req_set_fail_links(req); | |
e1e16097 | 3894 | io_req_complete(req, ret); |
15b71abe JA |
3895 | return 0; |
3896 | } | |
3897 | ||
45d189c6 | 3898 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3899 | { |
e45cff58 | 3900 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3901 | } |
3902 | ||
067524e9 JA |
3903 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3904 | const struct io_uring_sqe *sqe) | |
3905 | { | |
3906 | struct io_provide_buf *p = &req->pbuf; | |
3907 | u64 tmp; | |
3908 | ||
3909 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3910 | return -EINVAL; | |
3911 | ||
3912 | tmp = READ_ONCE(sqe->fd); | |
3913 | if (!tmp || tmp > USHRT_MAX) | |
3914 | return -EINVAL; | |
3915 | ||
3916 | memset(p, 0, sizeof(*p)); | |
3917 | p->nbufs = tmp; | |
3918 | p->bgid = READ_ONCE(sqe->buf_group); | |
3919 | return 0; | |
3920 | } | |
3921 | ||
3922 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3923 | int bgid, unsigned nbufs) | |
3924 | { | |
3925 | unsigned i = 0; | |
3926 | ||
3927 | /* shouldn't happen */ | |
3928 | if (!nbufs) | |
3929 | return 0; | |
3930 | ||
3931 | /* the head kbuf is the list itself */ | |
3932 | while (!list_empty(&buf->list)) { | |
3933 | struct io_buffer *nxt; | |
3934 | ||
3935 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3936 | list_del(&nxt->list); | |
3937 | kfree(nxt); | |
3938 | if (++i == nbufs) | |
3939 | return i; | |
3940 | } | |
3941 | i++; | |
3942 | kfree(buf); | |
9e15c3a0 | 3943 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3944 | |
3945 | return i; | |
3946 | } | |
3947 | ||
889fca73 | 3948 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3949 | { |
3950 | struct io_provide_buf *p = &req->pbuf; | |
3951 | struct io_ring_ctx *ctx = req->ctx; | |
3952 | struct io_buffer *head; | |
3953 | int ret = 0; | |
45d189c6 | 3954 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3955 | |
3956 | io_ring_submit_lock(ctx, !force_nonblock); | |
3957 | ||
3958 | lockdep_assert_held(&ctx->uring_lock); | |
3959 | ||
3960 | ret = -ENOENT; | |
9e15c3a0 | 3961 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3962 | if (head) |
3963 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3964 | if (ret < 0) |
3965 | req_set_fail_links(req); | |
067524e9 | 3966 | |
31bff9a5 PB |
3967 | /* need to hold the lock to complete IOPOLL requests */ |
3968 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3969 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3970 | io_ring_submit_unlock(ctx, !force_nonblock); |
3971 | } else { | |
3972 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3973 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3974 | } |
067524e9 JA |
3975 | return 0; |
3976 | } | |
3977 | ||
ddf0322d JA |
3978 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3979 | const struct io_uring_sqe *sqe) | |
3980 | { | |
3981 | struct io_provide_buf *p = &req->pbuf; | |
3982 | u64 tmp; | |
3983 | ||
3984 | if (sqe->ioprio || sqe->rw_flags) | |
3985 | return -EINVAL; | |
3986 | ||
3987 | tmp = READ_ONCE(sqe->fd); | |
3988 | if (!tmp || tmp > USHRT_MAX) | |
3989 | return -E2BIG; | |
3990 | p->nbufs = tmp; | |
3991 | p->addr = READ_ONCE(sqe->addr); | |
3992 | p->len = READ_ONCE(sqe->len); | |
3993 | ||
efe68c1c | 3994 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3995 | return -EFAULT; |
3996 | ||
3997 | p->bgid = READ_ONCE(sqe->buf_group); | |
3998 | tmp = READ_ONCE(sqe->off); | |
3999 | if (tmp > USHRT_MAX) | |
4000 | return -E2BIG; | |
4001 | p->bid = tmp; | |
4002 | return 0; | |
4003 | } | |
4004 | ||
4005 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4006 | { | |
4007 | struct io_buffer *buf; | |
4008 | u64 addr = pbuf->addr; | |
4009 | int i, bid = pbuf->bid; | |
4010 | ||
4011 | for (i = 0; i < pbuf->nbufs; i++) { | |
4012 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4013 | if (!buf) | |
4014 | break; | |
4015 | ||
4016 | buf->addr = addr; | |
4017 | buf->len = pbuf->len; | |
4018 | buf->bid = bid; | |
4019 | addr += pbuf->len; | |
4020 | bid++; | |
4021 | if (!*head) { | |
4022 | INIT_LIST_HEAD(&buf->list); | |
4023 | *head = buf; | |
4024 | } else { | |
4025 | list_add_tail(&buf->list, &(*head)->list); | |
4026 | } | |
4027 | } | |
4028 | ||
4029 | return i ? i : -ENOMEM; | |
4030 | } | |
4031 | ||
889fca73 | 4032 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4033 | { |
4034 | struct io_provide_buf *p = &req->pbuf; | |
4035 | struct io_ring_ctx *ctx = req->ctx; | |
4036 | struct io_buffer *head, *list; | |
4037 | int ret = 0; | |
45d189c6 | 4038 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4039 | |
4040 | io_ring_submit_lock(ctx, !force_nonblock); | |
4041 | ||
4042 | lockdep_assert_held(&ctx->uring_lock); | |
4043 | ||
9e15c3a0 | 4044 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4045 | |
4046 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4047 | if (ret >= 0 && !list) { |
4048 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4049 | if (ret < 0) | |
067524e9 | 4050 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4051 | } |
ddf0322d JA |
4052 | if (ret < 0) |
4053 | req_set_fail_links(req); | |
31bff9a5 PB |
4054 | |
4055 | /* need to hold the lock to complete IOPOLL requests */ | |
4056 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4057 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4058 | io_ring_submit_unlock(ctx, !force_nonblock); |
4059 | } else { | |
4060 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4061 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4062 | } |
ddf0322d | 4063 | return 0; |
cebdb986 JA |
4064 | } |
4065 | ||
3e4827b0 JA |
4066 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4067 | const struct io_uring_sqe *sqe) | |
4068 | { | |
4069 | #if defined(CONFIG_EPOLL) | |
4070 | if (sqe->ioprio || sqe->buf_index) | |
4071 | return -EINVAL; | |
6ca56f84 | 4072 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4073 | return -EINVAL; |
3e4827b0 JA |
4074 | |
4075 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4076 | req->epoll.op = READ_ONCE(sqe->len); | |
4077 | req->epoll.fd = READ_ONCE(sqe->off); | |
4078 | ||
4079 | if (ep_op_has_event(req->epoll.op)) { | |
4080 | struct epoll_event __user *ev; | |
4081 | ||
4082 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4083 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4084 | return -EFAULT; | |
4085 | } | |
4086 | ||
4087 | return 0; | |
4088 | #else | |
4089 | return -EOPNOTSUPP; | |
4090 | #endif | |
4091 | } | |
4092 | ||
889fca73 | 4093 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4094 | { |
4095 | #if defined(CONFIG_EPOLL) | |
4096 | struct io_epoll *ie = &req->epoll; | |
4097 | int ret; | |
45d189c6 | 4098 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4099 | |
4100 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4101 | if (force_nonblock && ret == -EAGAIN) | |
4102 | return -EAGAIN; | |
4103 | ||
4104 | if (ret < 0) | |
4105 | req_set_fail_links(req); | |
889fca73 | 4106 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4107 | return 0; |
4108 | #else | |
4109 | return -EOPNOTSUPP; | |
4110 | #endif | |
4111 | } | |
4112 | ||
c1ca757b JA |
4113 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4114 | { | |
4115 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4116 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4117 | return -EINVAL; | |
3232dd02 PB |
4118 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4119 | return -EINVAL; | |
c1ca757b JA |
4120 | |
4121 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4122 | req->madvise.len = READ_ONCE(sqe->len); | |
4123 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4124 | return 0; | |
4125 | #else | |
4126 | return -EOPNOTSUPP; | |
4127 | #endif | |
4128 | } | |
4129 | ||
45d189c6 | 4130 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4131 | { |
4132 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4133 | struct io_madvise *ma = &req->madvise; | |
4134 | int ret; | |
4135 | ||
45d189c6 | 4136 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4137 | return -EAGAIN; |
4138 | ||
0726b01e | 4139 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4140 | if (ret < 0) |
4141 | req_set_fail_links(req); | |
e1e16097 | 4142 | io_req_complete(req, ret); |
c1ca757b JA |
4143 | return 0; |
4144 | #else | |
4145 | return -EOPNOTSUPP; | |
4146 | #endif | |
4147 | } | |
4148 | ||
4840e418 JA |
4149 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4150 | { | |
4151 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4152 | return -EINVAL; | |
3232dd02 PB |
4153 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4154 | return -EINVAL; | |
4840e418 JA |
4155 | |
4156 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4157 | req->fadvise.len = READ_ONCE(sqe->len); | |
4158 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4159 | return 0; | |
4160 | } | |
4161 | ||
45d189c6 | 4162 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4163 | { |
4164 | struct io_fadvise *fa = &req->fadvise; | |
4165 | int ret; | |
4166 | ||
45d189c6 | 4167 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4168 | switch (fa->advice) { |
4169 | case POSIX_FADV_NORMAL: | |
4170 | case POSIX_FADV_RANDOM: | |
4171 | case POSIX_FADV_SEQUENTIAL: | |
4172 | break; | |
4173 | default: | |
4174 | return -EAGAIN; | |
4175 | } | |
4176 | } | |
4840e418 JA |
4177 | |
4178 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4179 | if (ret < 0) | |
4180 | req_set_fail_links(req); | |
e1e16097 | 4181 | io_req_complete(req, ret); |
4840e418 JA |
4182 | return 0; |
4183 | } | |
4184 | ||
eddc7ef5 JA |
4185 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4186 | { | |
6ca56f84 | 4187 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4188 | return -EINVAL; |
eddc7ef5 JA |
4189 | if (sqe->ioprio || sqe->buf_index) |
4190 | return -EINVAL; | |
9c280f90 | 4191 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4192 | return -EBADF; |
eddc7ef5 | 4193 | |
1d9e1288 BM |
4194 | req->statx.dfd = READ_ONCE(sqe->fd); |
4195 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4196 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4197 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4198 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4199 | |
4200 | return 0; | |
4201 | } | |
4202 | ||
45d189c6 | 4203 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4204 | { |
1d9e1288 | 4205 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4206 | int ret; |
4207 | ||
45d189c6 | 4208 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4209 | /* only need file table for an actual valid fd */ |
4210 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4211 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4212 | return -EAGAIN; |
5b0bbee4 | 4213 | } |
eddc7ef5 | 4214 | |
e62753e4 BM |
4215 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4216 | ctx->buffer); | |
eddc7ef5 | 4217 | |
eddc7ef5 JA |
4218 | if (ret < 0) |
4219 | req_set_fail_links(req); | |
e1e16097 | 4220 | io_req_complete(req, ret); |
eddc7ef5 JA |
4221 | return 0; |
4222 | } | |
4223 | ||
b5dba59e JA |
4224 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4225 | { | |
14587a46 | 4226 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4227 | return -EINVAL; |
b5dba59e JA |
4228 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4229 | sqe->rw_flags || sqe->buf_index) | |
4230 | return -EINVAL; | |
9c280f90 | 4231 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4232 | return -EBADF; |
b5dba59e JA |
4233 | |
4234 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4235 | return 0; |
b5dba59e JA |
4236 | } |
4237 | ||
889fca73 | 4238 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4239 | { |
9eac1904 | 4240 | struct files_struct *files = current->files; |
3af73b28 | 4241 | struct io_close *close = &req->close; |
9eac1904 JA |
4242 | struct fdtable *fdt; |
4243 | struct file *file; | |
b5dba59e JA |
4244 | int ret; |
4245 | ||
9eac1904 JA |
4246 | file = NULL; |
4247 | ret = -EBADF; | |
4248 | spin_lock(&files->file_lock); | |
4249 | fdt = files_fdtable(files); | |
4250 | if (close->fd >= fdt->max_fds) { | |
4251 | spin_unlock(&files->file_lock); | |
4252 | goto err; | |
4253 | } | |
4254 | file = fdt->fd[close->fd]; | |
4255 | if (!file) { | |
4256 | spin_unlock(&files->file_lock); | |
4257 | goto err; | |
4258 | } | |
4259 | ||
4260 | if (file->f_op == &io_uring_fops) { | |
4261 | spin_unlock(&files->file_lock); | |
4262 | file = NULL; | |
4263 | goto err; | |
3af73b28 | 4264 | } |
b5dba59e JA |
4265 | |
4266 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4267 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4268 | spin_unlock(&files->file_lock); |
0bf0eefd | 4269 | return -EAGAIN; |
a2100672 | 4270 | } |
b5dba59e | 4271 | |
9eac1904 JA |
4272 | ret = __close_fd_get_file(close->fd, &file); |
4273 | spin_unlock(&files->file_lock); | |
4274 | if (ret < 0) { | |
4275 | if (ret == -ENOENT) | |
4276 | ret = -EBADF; | |
4277 | goto err; | |
4278 | } | |
4279 | ||
3af73b28 | 4280 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4281 | ret = filp_close(file, current->files); |
4282 | err: | |
3af73b28 PB |
4283 | if (ret < 0) |
4284 | req_set_fail_links(req); | |
9eac1904 JA |
4285 | if (file) |
4286 | fput(file); | |
889fca73 | 4287 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4288 | return 0; |
b5dba59e JA |
4289 | } |
4290 | ||
1155c76a | 4291 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4292 | { |
4293 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4294 | |
5d17b4a4 JA |
4295 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4296 | return -EINVAL; | |
4297 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4298 | return -EINVAL; | |
4299 | ||
8ed8d3c3 JA |
4300 | req->sync.off = READ_ONCE(sqe->off); |
4301 | req->sync.len = READ_ONCE(sqe->len); | |
4302 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4303 | return 0; |
4304 | } | |
4305 | ||
45d189c6 | 4306 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4307 | { |
8ed8d3c3 JA |
4308 | int ret; |
4309 | ||
ac45abc0 | 4310 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4311 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4312 | return -EAGAIN; |
4313 | ||
9adbd45d | 4314 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4315 | req->sync.flags); |
4316 | if (ret < 0) | |
4317 | req_set_fail_links(req); | |
e1e16097 | 4318 | io_req_complete(req, ret); |
5d17b4a4 JA |
4319 | return 0; |
4320 | } | |
4321 | ||
469956e8 | 4322 | #if defined(CONFIG_NET) |
02d27d89 PB |
4323 | static int io_setup_async_msg(struct io_kiocb *req, |
4324 | struct io_async_msghdr *kmsg) | |
4325 | { | |
e8c2bc1f JA |
4326 | struct io_async_msghdr *async_msg = req->async_data; |
4327 | ||
4328 | if (async_msg) | |
02d27d89 | 4329 | return -EAGAIN; |
e8c2bc1f | 4330 | if (io_alloc_async_data(req)) { |
257e84a5 | 4331 | kfree(kmsg->free_iov); |
02d27d89 PB |
4332 | return -ENOMEM; |
4333 | } | |
e8c2bc1f | 4334 | async_msg = req->async_data; |
02d27d89 | 4335 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4336 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4337 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4338 | /* if were using fast_iov, set it to the new one */ |
4339 | if (!async_msg->free_iov) | |
4340 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4341 | ||
02d27d89 PB |
4342 | return -EAGAIN; |
4343 | } | |
4344 | ||
2ae523ed PB |
4345 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4346 | struct io_async_msghdr *iomsg) | |
4347 | { | |
2ae523ed | 4348 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4349 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4350 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4351 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4352 | } |
4353 | ||
93642ef8 PB |
4354 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4355 | { | |
4356 | int ret; | |
4357 | ||
4358 | if (!io_op_defs[req->opcode].needs_async_data) | |
4359 | return 0; | |
4360 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4361 | if (!ret) | |
4362 | req->flags |= REQ_F_NEED_CLEANUP; | |
4363 | return ret; | |
4364 | } | |
4365 | ||
3529d8c2 | 4366 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4367 | { |
e47293fd | 4368 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4369 | |
d2b6f48b PB |
4370 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4371 | return -EINVAL; | |
4372 | ||
e47293fd | 4373 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4374 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4375 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4376 | |
d8768362 JA |
4377 | #ifdef CONFIG_COMPAT |
4378 | if (req->ctx->compat) | |
4379 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4380 | #endif | |
93642ef8 | 4381 | return 0; |
03b1230c JA |
4382 | } |
4383 | ||
889fca73 | 4384 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4385 | { |
6b754c8b | 4386 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4387 | struct socket *sock; |
7a7cacba | 4388 | unsigned flags; |
0fa03c62 JA |
4389 | int ret; |
4390 | ||
dba4a925 | 4391 | sock = sock_from_file(req->file); |
7a7cacba | 4392 | if (unlikely(!sock)) |
dba4a925 | 4393 | return -ENOTSOCK; |
3529d8c2 | 4394 | |
257e84a5 PB |
4395 | kmsg = req->async_data; |
4396 | if (!kmsg) { | |
7a7cacba PB |
4397 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4398 | if (ret) | |
4399 | return ret; | |
4400 | kmsg = &iomsg; | |
0fa03c62 | 4401 | } |
0fa03c62 | 4402 | |
76cd979f | 4403 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4404 | if (flags & MSG_DONTWAIT) |
4405 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4406 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4407 | flags |= MSG_DONTWAIT; |
e47293fd | 4408 | |
7a7cacba | 4409 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4410 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4411 | return io_setup_async_msg(req, kmsg); |
4412 | if (ret == -ERESTARTSYS) | |
4413 | ret = -EINTR; | |
0fa03c62 | 4414 | |
257e84a5 PB |
4415 | /* fast path, check for non-NULL to avoid function call */ |
4416 | if (kmsg->free_iov) | |
4417 | kfree(kmsg->free_iov); | |
99bc4c38 | 4418 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4419 | if (ret < 0) |
4420 | req_set_fail_links(req); | |
889fca73 | 4421 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4422 | return 0; |
03b1230c | 4423 | } |
aa1fa28f | 4424 | |
889fca73 | 4425 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4426 | { |
7a7cacba PB |
4427 | struct io_sr_msg *sr = &req->sr_msg; |
4428 | struct msghdr msg; | |
4429 | struct iovec iov; | |
fddaface | 4430 | struct socket *sock; |
7a7cacba | 4431 | unsigned flags; |
fddaface JA |
4432 | int ret; |
4433 | ||
dba4a925 | 4434 | sock = sock_from_file(req->file); |
7a7cacba | 4435 | if (unlikely(!sock)) |
dba4a925 | 4436 | return -ENOTSOCK; |
fddaface | 4437 | |
7a7cacba PB |
4438 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4439 | if (unlikely(ret)) | |
14db8411 | 4440 | return ret; |
fddaface | 4441 | |
7a7cacba PB |
4442 | msg.msg_name = NULL; |
4443 | msg.msg_control = NULL; | |
4444 | msg.msg_controllen = 0; | |
4445 | msg.msg_namelen = 0; | |
fddaface | 4446 | |
76cd979f | 4447 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4448 | if (flags & MSG_DONTWAIT) |
4449 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4450 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4451 | flags |= MSG_DONTWAIT; |
fddaface | 4452 | |
7a7cacba PB |
4453 | msg.msg_flags = flags; |
4454 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4455 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4456 | return -EAGAIN; |
4457 | if (ret == -ERESTARTSYS) | |
4458 | ret = -EINTR; | |
fddaface | 4459 | |
fddaface JA |
4460 | if (ret < 0) |
4461 | req_set_fail_links(req); | |
889fca73 | 4462 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4463 | return 0; |
fddaface JA |
4464 | } |
4465 | ||
1400e697 PB |
4466 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4467 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4468 | { |
4469 | struct io_sr_msg *sr = &req->sr_msg; | |
4470 | struct iovec __user *uiov; | |
4471 | size_t iov_len; | |
4472 | int ret; | |
4473 | ||
1400e697 PB |
4474 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4475 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4476 | if (ret) |
4477 | return ret; | |
4478 | ||
4479 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4480 | if (iov_len > 1) | |
4481 | return -EINVAL; | |
5476dfed | 4482 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4483 | return -EFAULT; |
5476dfed | 4484 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4485 | iomsg->free_iov = NULL; |
52de1fe1 | 4486 | } else { |
257e84a5 | 4487 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4488 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4489 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4490 | false); |
52de1fe1 JA |
4491 | if (ret > 0) |
4492 | ret = 0; | |
4493 | } | |
4494 | ||
4495 | return ret; | |
4496 | } | |
4497 | ||
4498 | #ifdef CONFIG_COMPAT | |
4499 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4500 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4501 | { |
4502 | struct compat_msghdr __user *msg_compat; | |
4503 | struct io_sr_msg *sr = &req->sr_msg; | |
4504 | struct compat_iovec __user *uiov; | |
4505 | compat_uptr_t ptr; | |
4506 | compat_size_t len; | |
4507 | int ret; | |
4508 | ||
270a5940 | 4509 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4510 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4511 | &ptr, &len); |
4512 | if (ret) | |
4513 | return ret; | |
4514 | ||
4515 | uiov = compat_ptr(ptr); | |
4516 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4517 | compat_ssize_t clen; | |
4518 | ||
4519 | if (len > 1) | |
4520 | return -EINVAL; | |
4521 | if (!access_ok(uiov, sizeof(*uiov))) | |
4522 | return -EFAULT; | |
4523 | if (__get_user(clen, &uiov->iov_len)) | |
4524 | return -EFAULT; | |
4525 | if (clen < 0) | |
4526 | return -EINVAL; | |
2d280bc8 | 4527 | sr->len = clen; |
257e84a5 | 4528 | iomsg->free_iov = NULL; |
52de1fe1 | 4529 | } else { |
257e84a5 | 4530 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4531 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4532 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4533 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4534 | if (ret < 0) |
4535 | return ret; | |
4536 | } | |
4537 | ||
4538 | return 0; | |
4539 | } | |
4540 | #endif | |
4541 | ||
1400e697 PB |
4542 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4543 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4544 | { |
1400e697 | 4545 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4546 | |
4547 | #ifdef CONFIG_COMPAT | |
4548 | if (req->ctx->compat) | |
1400e697 | 4549 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4550 | #endif |
52de1fe1 | 4551 | |
1400e697 | 4552 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4553 | } |
4554 | ||
bcda7baa | 4555 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4556 | bool needs_lock) |
bcda7baa JA |
4557 | { |
4558 | struct io_sr_msg *sr = &req->sr_msg; | |
4559 | struct io_buffer *kbuf; | |
4560 | ||
bcda7baa JA |
4561 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4562 | if (IS_ERR(kbuf)) | |
4563 | return kbuf; | |
4564 | ||
4565 | sr->kbuf = kbuf; | |
4566 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4567 | return kbuf; |
fddaface JA |
4568 | } |
4569 | ||
7fbb1b54 PB |
4570 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4571 | { | |
4572 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4573 | } | |
4574 | ||
93642ef8 | 4575 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4576 | { |
99bc4c38 | 4577 | int ret; |
3529d8c2 | 4578 | |
93642ef8 PB |
4579 | if (!io_op_defs[req->opcode].needs_async_data) |
4580 | return 0; | |
4581 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4582 | if (!ret) | |
4583 | req->flags |= REQ_F_NEED_CLEANUP; | |
4584 | return ret; | |
4585 | } | |
4586 | ||
4587 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4588 | { | |
4589 | struct io_sr_msg *sr = &req->sr_msg; | |
4590 | ||
d2b6f48b PB |
4591 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4592 | return -EINVAL; | |
4593 | ||
3529d8c2 | 4594 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4595 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4596 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4597 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4598 | |
d8768362 JA |
4599 | #ifdef CONFIG_COMPAT |
4600 | if (req->ctx->compat) | |
4601 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4602 | #endif | |
93642ef8 | 4603 | return 0; |
aa1fa28f JA |
4604 | } |
4605 | ||
889fca73 | 4606 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4607 | { |
6b754c8b | 4608 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4609 | struct socket *sock; |
7fbb1b54 | 4610 | struct io_buffer *kbuf; |
7a7cacba | 4611 | unsigned flags; |
52de1fe1 | 4612 | int ret, cflags = 0; |
45d189c6 | 4613 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4614 | |
dba4a925 | 4615 | sock = sock_from_file(req->file); |
7a7cacba | 4616 | if (unlikely(!sock)) |
dba4a925 | 4617 | return -ENOTSOCK; |
3529d8c2 | 4618 | |
257e84a5 PB |
4619 | kmsg = req->async_data; |
4620 | if (!kmsg) { | |
7a7cacba PB |
4621 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4622 | if (ret) | |
681fda8d | 4623 | return ret; |
7a7cacba PB |
4624 | kmsg = &iomsg; |
4625 | } | |
03b1230c | 4626 | |
bc02ef33 | 4627 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4628 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4629 | if (IS_ERR(kbuf)) |
52de1fe1 | 4630 | return PTR_ERR(kbuf); |
7a7cacba | 4631 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4632 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4633 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4634 | 1, req->sr_msg.len); |
4635 | } | |
52de1fe1 | 4636 | |
76cd979f | 4637 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4638 | if (flags & MSG_DONTWAIT) |
4639 | req->flags |= REQ_F_NOWAIT; | |
4640 | else if (force_nonblock) | |
4641 | flags |= MSG_DONTWAIT; | |
e47293fd | 4642 | |
7a7cacba PB |
4643 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4644 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4645 | if (force_nonblock && ret == -EAGAIN) |
4646 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4647 | if (ret == -ERESTARTSYS) |
4648 | ret = -EINTR; | |
03b1230c | 4649 | |
7fbb1b54 PB |
4650 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4651 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4652 | /* fast path, check for non-NULL to avoid function call */ |
4653 | if (kmsg->free_iov) | |
4654 | kfree(kmsg->free_iov); | |
99bc4c38 | 4655 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4656 | if (ret < 0) |
4657 | req_set_fail_links(req); | |
889fca73 | 4658 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4659 | return 0; |
0fa03c62 | 4660 | } |
5d17b4a4 | 4661 | |
889fca73 | 4662 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4663 | { |
6b754c8b | 4664 | struct io_buffer *kbuf; |
7a7cacba PB |
4665 | struct io_sr_msg *sr = &req->sr_msg; |
4666 | struct msghdr msg; | |
4667 | void __user *buf = sr->buf; | |
fddaface | 4668 | struct socket *sock; |
7a7cacba PB |
4669 | struct iovec iov; |
4670 | unsigned flags; | |
bcda7baa | 4671 | int ret, cflags = 0; |
45d189c6 | 4672 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4673 | |
dba4a925 | 4674 | sock = sock_from_file(req->file); |
7a7cacba | 4675 | if (unlikely(!sock)) |
dba4a925 | 4676 | return -ENOTSOCK; |
fddaface | 4677 | |
bc02ef33 | 4678 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4679 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4680 | if (IS_ERR(kbuf)) |
4681 | return PTR_ERR(kbuf); | |
7a7cacba | 4682 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4683 | } |
bcda7baa | 4684 | |
7a7cacba | 4685 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4686 | if (unlikely(ret)) |
4687 | goto out_free; | |
fddaface | 4688 | |
7a7cacba PB |
4689 | msg.msg_name = NULL; |
4690 | msg.msg_control = NULL; | |
4691 | msg.msg_controllen = 0; | |
4692 | msg.msg_namelen = 0; | |
4693 | msg.msg_iocb = NULL; | |
4694 | msg.msg_flags = 0; | |
fddaface | 4695 | |
76cd979f | 4696 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4697 | if (flags & MSG_DONTWAIT) |
4698 | req->flags |= REQ_F_NOWAIT; | |
4699 | else if (force_nonblock) | |
4700 | flags |= MSG_DONTWAIT; | |
4701 | ||
4702 | ret = sock_recvmsg(sock, &msg, flags); | |
4703 | if (force_nonblock && ret == -EAGAIN) | |
4704 | return -EAGAIN; | |
4705 | if (ret == -ERESTARTSYS) | |
4706 | ret = -EINTR; | |
14c32eee | 4707 | out_free: |
7fbb1b54 PB |
4708 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4709 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4710 | if (ret < 0) |
4711 | req_set_fail_links(req); | |
889fca73 | 4712 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4713 | return 0; |
fddaface JA |
4714 | } |
4715 | ||
3529d8c2 | 4716 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4717 | { |
8ed8d3c3 JA |
4718 | struct io_accept *accept = &req->accept; |
4719 | ||
14587a46 | 4720 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4721 | return -EINVAL; |
8042d6ce | 4722 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4723 | return -EINVAL; |
4724 | ||
d55e5f5b JA |
4725 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4726 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4727 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4728 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4729 | return 0; |
8ed8d3c3 | 4730 | } |
17f2fe35 | 4731 | |
889fca73 | 4732 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4733 | { |
4734 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4735 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4736 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4737 | int ret; |
4738 | ||
e697deed JX |
4739 | if (req->file->f_flags & O_NONBLOCK) |
4740 | req->flags |= REQ_F_NOWAIT; | |
4741 | ||
8ed8d3c3 | 4742 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4743 | accept->addr_len, accept->flags, |
4744 | accept->nofile); | |
8ed8d3c3 | 4745 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4746 | return -EAGAIN; |
ac45abc0 PB |
4747 | if (ret < 0) { |
4748 | if (ret == -ERESTARTSYS) | |
4749 | ret = -EINTR; | |
4e88d6e7 | 4750 | req_set_fail_links(req); |
ac45abc0 | 4751 | } |
889fca73 | 4752 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4753 | return 0; |
8ed8d3c3 JA |
4754 | } |
4755 | ||
93642ef8 PB |
4756 | static int io_connect_prep_async(struct io_kiocb *req) |
4757 | { | |
4758 | struct io_async_connect *io = req->async_data; | |
4759 | struct io_connect *conn = &req->connect; | |
4760 | ||
4761 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4762 | } | |
4763 | ||
3529d8c2 | 4764 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4765 | { |
3529d8c2 | 4766 | struct io_connect *conn = &req->connect; |
f499a021 | 4767 | |
14587a46 | 4768 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4769 | return -EINVAL; |
4770 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4771 | return -EINVAL; | |
4772 | ||
3529d8c2 JA |
4773 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4774 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4775 | return 0; |
f499a021 JA |
4776 | } |
4777 | ||
889fca73 | 4778 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4779 | { |
e8c2bc1f | 4780 | struct io_async_connect __io, *io; |
f8e85cf2 | 4781 | unsigned file_flags; |
3fbb51c1 | 4782 | int ret; |
45d189c6 | 4783 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4784 | |
e8c2bc1f JA |
4785 | if (req->async_data) { |
4786 | io = req->async_data; | |
f499a021 | 4787 | } else { |
3529d8c2 JA |
4788 | ret = move_addr_to_kernel(req->connect.addr, |
4789 | req->connect.addr_len, | |
e8c2bc1f | 4790 | &__io.address); |
f499a021 JA |
4791 | if (ret) |
4792 | goto out; | |
4793 | io = &__io; | |
4794 | } | |
4795 | ||
3fbb51c1 JA |
4796 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4797 | ||
e8c2bc1f | 4798 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4799 | req->connect.addr_len, file_flags); |
87f80d62 | 4800 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4801 | if (req->async_data) |
b7bb4f7d | 4802 | return -EAGAIN; |
e8c2bc1f | 4803 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4804 | ret = -ENOMEM; |
4805 | goto out; | |
4806 | } | |
e8c2bc1f JA |
4807 | io = req->async_data; |
4808 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4809 | return -EAGAIN; |
f499a021 | 4810 | } |
f8e85cf2 JA |
4811 | if (ret == -ERESTARTSYS) |
4812 | ret = -EINTR; | |
f499a021 | 4813 | out: |
4e88d6e7 JA |
4814 | if (ret < 0) |
4815 | req_set_fail_links(req); | |
889fca73 | 4816 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4817 | return 0; |
469956e8 Y |
4818 | } |
4819 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4820 | #define IO_NETOP_FN(op) \ |
4821 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4822 | { \ | |
4823 | return -EOPNOTSUPP; \ | |
4824 | } | |
4825 | ||
4826 | #define IO_NETOP_PREP(op) \ | |
4827 | IO_NETOP_FN(op) \ | |
4828 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4829 | { \ | |
4830 | return -EOPNOTSUPP; \ | |
4831 | } \ | |
4832 | ||
4833 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4834 | IO_NETOP_PREP(op) \ | |
4835 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4836 | { \ | |
4837 | return -EOPNOTSUPP; \ | |
4838 | } | |
4839 | ||
4840 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4841 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4842 | IO_NETOP_PREP_ASYNC(connect); | |
4843 | IO_NETOP_PREP(accept); | |
4844 | IO_NETOP_FN(send); | |
4845 | IO_NETOP_FN(recv); | |
469956e8 | 4846 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4847 | |
d7718a9d JA |
4848 | struct io_poll_table { |
4849 | struct poll_table_struct pt; | |
4850 | struct io_kiocb *req; | |
4851 | int error; | |
4852 | }; | |
ce593a6c | 4853 | |
d7718a9d JA |
4854 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4855 | __poll_t mask, task_work_func_t func) | |
4856 | { | |
aa96bf8a | 4857 | int ret; |
d7718a9d JA |
4858 | |
4859 | /* for instances that support it check for an event match first: */ | |
4860 | if (mask && !(mask & poll->events)) | |
4861 | return 0; | |
4862 | ||
4863 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4864 | ||
4865 | list_del_init(&poll->wait.entry); | |
4866 | ||
d7718a9d | 4867 | req->result = mask; |
7cbf1722 | 4868 | req->task_work.func = func; |
6d816e08 JA |
4869 | percpu_ref_get(&req->ctx->refs); |
4870 | ||
d7718a9d | 4871 | /* |
e3aabf95 JA |
4872 | * If this fails, then the task is exiting. When a task exits, the |
4873 | * work gets canceled, so just cancel this request as well instead | |
4874 | * of executing it. We can't safely execute it anyway, as we may not | |
4875 | * have the needed state needed for it anyway. | |
d7718a9d | 4876 | */ |
355fb9e2 | 4877 | ret = io_req_task_work_add(req); |
aa96bf8a | 4878 | if (unlikely(ret)) { |
e3aabf95 | 4879 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4880 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4881 | } |
d7718a9d JA |
4882 | return 1; |
4883 | } | |
4884 | ||
74ce6ce4 JA |
4885 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4886 | __acquires(&req->ctx->completion_lock) | |
4887 | { | |
4888 | struct io_ring_ctx *ctx = req->ctx; | |
4889 | ||
4890 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4891 | struct poll_table_struct pt = { ._key = poll->events }; | |
4892 | ||
4893 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4894 | } | |
4895 | ||
4896 | spin_lock_irq(&ctx->completion_lock); | |
4897 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4898 | add_wait_queue(poll->head, &poll->wait); | |
4899 | return true; | |
4900 | } | |
4901 | ||
4902 | return false; | |
4903 | } | |
4904 | ||
d4e7cd36 | 4905 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4906 | { |
e8c2bc1f | 4907 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4908 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4909 | return req->async_data; |
d4e7cd36 JA |
4910 | return req->apoll->double_poll; |
4911 | } | |
4912 | ||
4913 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4914 | { | |
4915 | if (req->opcode == IORING_OP_POLL_ADD) | |
4916 | return &req->poll; | |
4917 | return &req->apoll->poll; | |
4918 | } | |
4919 | ||
4920 | static void io_poll_remove_double(struct io_kiocb *req) | |
4921 | { | |
4922 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4923 | |
4924 | lockdep_assert_held(&req->ctx->completion_lock); | |
4925 | ||
4926 | if (poll && poll->head) { | |
4927 | struct wait_queue_head *head = poll->head; | |
4928 | ||
4929 | spin_lock(&head->lock); | |
4930 | list_del_init(&poll->wait.entry); | |
4931 | if (poll->wait.private) | |
4932 | refcount_dec(&req->refs); | |
4933 | poll->head = NULL; | |
4934 | spin_unlock(&head->lock); | |
4935 | } | |
4936 | } | |
4937 | ||
4938 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4939 | { | |
4940 | struct io_ring_ctx *ctx = req->ctx; | |
4941 | ||
d4e7cd36 | 4942 | io_poll_remove_double(req); |
18bceab1 JA |
4943 | req->poll.done = true; |
4944 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4945 | io_commit_cqring(ctx); | |
4946 | } | |
4947 | ||
dd221f46 | 4948 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4949 | { |
dd221f46 | 4950 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4951 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4952 | struct io_kiocb *nxt; |
18bceab1 JA |
4953 | |
4954 | if (io_poll_rewait(req, &req->poll)) { | |
4955 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4956 | } else { |
4957 | hash_del(&req->hash_node); | |
4958 | io_poll_complete(req, req->result, 0); | |
4959 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4960 | |
dd221f46 PB |
4961 | nxt = io_put_req_find_next(req); |
4962 | io_cqring_ev_posted(ctx); | |
4963 | if (nxt) | |
4964 | __io_req_task_submit(nxt); | |
4965 | } | |
18bceab1 | 4966 | |
6d816e08 | 4967 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4968 | } |
4969 | ||
4970 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4971 | int sync, void *key) | |
4972 | { | |
4973 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4974 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4975 | __poll_t mask = key_to_poll(key); |
4976 | ||
4977 | /* for instances that support it check for an event match first: */ | |
4978 | if (mask && !(mask & poll->events)) | |
4979 | return 0; | |
4980 | ||
8706e04e JA |
4981 | list_del_init(&wait->entry); |
4982 | ||
807abcb0 | 4983 | if (poll && poll->head) { |
18bceab1 JA |
4984 | bool done; |
4985 | ||
807abcb0 JA |
4986 | spin_lock(&poll->head->lock); |
4987 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4988 | if (!done) |
807abcb0 | 4989 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4990 | /* make sure double remove sees this as being gone */ |
4991 | wait->private = NULL; | |
807abcb0 | 4992 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4993 | if (!done) { |
4994 | /* use wait func handler, so it matches the rq type */ | |
4995 | poll->wait.func(&poll->wait, mode, sync, key); | |
4996 | } | |
18bceab1 JA |
4997 | } |
4998 | refcount_dec(&req->refs); | |
4999 | return 1; | |
5000 | } | |
5001 | ||
5002 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5003 | wait_queue_func_t wake_func) | |
5004 | { | |
5005 | poll->head = NULL; | |
5006 | poll->done = false; | |
5007 | poll->canceled = false; | |
5008 | poll->events = events; | |
5009 | INIT_LIST_HEAD(&poll->wait.entry); | |
5010 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5011 | } | |
5012 | ||
5013 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5014 | struct wait_queue_head *head, |
5015 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5016 | { |
5017 | struct io_kiocb *req = pt->req; | |
5018 | ||
5019 | /* | |
5020 | * If poll->head is already set, it's because the file being polled | |
5021 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5022 | * for write). Setup a separate io_poll_iocb if this happens. | |
5023 | */ | |
5024 | if (unlikely(poll->head)) { | |
58852d4d PB |
5025 | struct io_poll_iocb *poll_one = poll; |
5026 | ||
18bceab1 | 5027 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5028 | if (*poll_ptr) { |
18bceab1 JA |
5029 | pt->error = -EINVAL; |
5030 | return; | |
5031 | } | |
1c3b3e65 JA |
5032 | /* double add on the same waitqueue head, ignore */ |
5033 | if (poll->head == head) | |
5034 | return; | |
18bceab1 JA |
5035 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5036 | if (!poll) { | |
5037 | pt->error = -ENOMEM; | |
5038 | return; | |
5039 | } | |
58852d4d | 5040 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5041 | refcount_inc(&req->refs); |
5042 | poll->wait.private = req; | |
807abcb0 | 5043 | *poll_ptr = poll; |
18bceab1 JA |
5044 | } |
5045 | ||
5046 | pt->error = 0; | |
5047 | poll->head = head; | |
a31eb4a2 JX |
5048 | |
5049 | if (poll->events & EPOLLEXCLUSIVE) | |
5050 | add_wait_queue_exclusive(head, &poll->wait); | |
5051 | else | |
5052 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5053 | } |
5054 | ||
5055 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5056 | struct poll_table_struct *p) | |
5057 | { | |
5058 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5059 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5060 | |
807abcb0 | 5061 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5062 | } |
5063 | ||
d7718a9d JA |
5064 | static void io_async_task_func(struct callback_head *cb) |
5065 | { | |
5066 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5067 | struct async_poll *apoll = req->apoll; | |
5068 | struct io_ring_ctx *ctx = req->ctx; | |
5069 | ||
5070 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5071 | ||
74ce6ce4 | 5072 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5073 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5074 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5075 | return; |
d7718a9d JA |
5076 | } |
5077 | ||
31067255 | 5078 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5079 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5080 | hash_del(&req->hash_node); |
2bae047e | 5081 | |
d4e7cd36 | 5082 | io_poll_remove_double(req); |
74ce6ce4 JA |
5083 | spin_unlock_irq(&ctx->completion_lock); |
5084 | ||
0be0b0e3 PB |
5085 | if (!READ_ONCE(apoll->poll.canceled)) |
5086 | __io_req_task_submit(req); | |
5087 | else | |
5088 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5089 | |
6d816e08 | 5090 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5091 | kfree(apoll->double_poll); |
31067255 | 5092 | kfree(apoll); |
d7718a9d JA |
5093 | } |
5094 | ||
5095 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5096 | void *key) | |
5097 | { | |
5098 | struct io_kiocb *req = wait->private; | |
5099 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5100 | ||
5101 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5102 | key_to_poll(key)); | |
5103 | ||
5104 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5105 | } | |
5106 | ||
5107 | static void io_poll_req_insert(struct io_kiocb *req) | |
5108 | { | |
5109 | struct io_ring_ctx *ctx = req->ctx; | |
5110 | struct hlist_head *list; | |
5111 | ||
5112 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5113 | hlist_add_head(&req->hash_node, list); | |
5114 | } | |
5115 | ||
5116 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5117 | struct io_poll_iocb *poll, | |
5118 | struct io_poll_table *ipt, __poll_t mask, | |
5119 | wait_queue_func_t wake_func) | |
5120 | __acquires(&ctx->completion_lock) | |
5121 | { | |
5122 | struct io_ring_ctx *ctx = req->ctx; | |
5123 | bool cancel = false; | |
5124 | ||
4d52f338 | 5125 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5126 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5127 | poll->file = req->file; |
18bceab1 | 5128 | poll->wait.private = req; |
d7718a9d JA |
5129 | |
5130 | ipt->pt._key = mask; | |
5131 | ipt->req = req; | |
5132 | ipt->error = -EINVAL; | |
5133 | ||
d7718a9d JA |
5134 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5135 | ||
5136 | spin_lock_irq(&ctx->completion_lock); | |
5137 | if (likely(poll->head)) { | |
5138 | spin_lock(&poll->head->lock); | |
5139 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5140 | if (ipt->error) | |
5141 | cancel = true; | |
5142 | ipt->error = 0; | |
5143 | mask = 0; | |
5144 | } | |
5145 | if (mask || ipt->error) | |
5146 | list_del_init(&poll->wait.entry); | |
5147 | else if (cancel) | |
5148 | WRITE_ONCE(poll->canceled, true); | |
5149 | else if (!poll->done) /* actually waiting for an event */ | |
5150 | io_poll_req_insert(req); | |
5151 | spin_unlock(&poll->head->lock); | |
5152 | } | |
5153 | ||
5154 | return mask; | |
5155 | } | |
5156 | ||
5157 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5158 | { | |
5159 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5160 | struct io_ring_ctx *ctx = req->ctx; | |
5161 | struct async_poll *apoll; | |
5162 | struct io_poll_table ipt; | |
5163 | __poll_t mask, ret; | |
9dab14b8 | 5164 | int rw; |
d7718a9d JA |
5165 | |
5166 | if (!req->file || !file_can_poll(req->file)) | |
5167 | return false; | |
24c74678 | 5168 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5169 | return false; |
9dab14b8 JA |
5170 | if (def->pollin) |
5171 | rw = READ; | |
5172 | else if (def->pollout) | |
5173 | rw = WRITE; | |
5174 | else | |
5175 | return false; | |
5176 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5177 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5178 | return false; |
5179 | ||
5180 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5181 | if (unlikely(!apoll)) | |
5182 | return false; | |
807abcb0 | 5183 | apoll->double_poll = NULL; |
d7718a9d JA |
5184 | |
5185 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5186 | req->apoll = apoll; |
d7718a9d | 5187 | |
8755d97a | 5188 | mask = 0; |
d7718a9d | 5189 | if (def->pollin) |
8755d97a | 5190 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5191 | if (def->pollout) |
5192 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5193 | |
5194 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5195 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5196 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5197 | mask &= ~POLLIN; | |
5198 | ||
d7718a9d JA |
5199 | mask |= POLLERR | POLLPRI; |
5200 | ||
5201 | ipt.pt._qproc = io_async_queue_proc; | |
5202 | ||
5203 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5204 | io_async_wake); | |
a36da65c | 5205 | if (ret || ipt.error) { |
d4e7cd36 | 5206 | io_poll_remove_double(req); |
d7718a9d | 5207 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5208 | kfree(apoll->double_poll); |
d7718a9d JA |
5209 | kfree(apoll); |
5210 | return false; | |
5211 | } | |
5212 | spin_unlock_irq(&ctx->completion_lock); | |
5213 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5214 | apoll->poll.events); | |
5215 | return true; | |
5216 | } | |
5217 | ||
5218 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5219 | struct io_poll_iocb *poll) | |
221c5eb2 | 5220 | { |
b41e9852 | 5221 | bool do_complete = false; |
221c5eb2 JA |
5222 | |
5223 | spin_lock(&poll->head->lock); | |
5224 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5225 | if (!list_empty(&poll->wait.entry)) { |
5226 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5227 | do_complete = true; |
221c5eb2 JA |
5228 | } |
5229 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5230 | hash_del(&req->hash_node); |
d7718a9d JA |
5231 | return do_complete; |
5232 | } | |
5233 | ||
5234 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5235 | { | |
5236 | bool do_complete; | |
5237 | ||
d4e7cd36 JA |
5238 | io_poll_remove_double(req); |
5239 | ||
d7718a9d JA |
5240 | if (req->opcode == IORING_OP_POLL_ADD) { |
5241 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5242 | } else { | |
3bfa5bcb JA |
5243 | struct async_poll *apoll = req->apoll; |
5244 | ||
d7718a9d | 5245 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5246 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5247 | if (do_complete) { | |
d7718a9d | 5248 | io_put_req(req); |
807abcb0 | 5249 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5250 | kfree(apoll); |
5251 | } | |
b1f573bd XW |
5252 | } |
5253 | ||
b41e9852 JA |
5254 | if (do_complete) { |
5255 | io_cqring_fill_event(req, -ECANCELED); | |
5256 | io_commit_cqring(req->ctx); | |
f254ac04 | 5257 | req_set_fail_links(req); |
216578e5 | 5258 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5259 | } |
5260 | ||
5261 | return do_complete; | |
221c5eb2 JA |
5262 | } |
5263 | ||
76e1b642 JA |
5264 | /* |
5265 | * Returns true if we found and killed one or more poll requests | |
5266 | */ | |
6b81928d PB |
5267 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5268 | struct files_struct *files) | |
221c5eb2 | 5269 | { |
78076bb6 | 5270 | struct hlist_node *tmp; |
221c5eb2 | 5271 | struct io_kiocb *req; |
8e2e1faf | 5272 | int posted = 0, i; |
221c5eb2 JA |
5273 | |
5274 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5275 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5276 | struct hlist_head *list; | |
5277 | ||
5278 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5279 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5280 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5281 | posted += io_poll_remove_one(req); |
5282 | } | |
221c5eb2 JA |
5283 | } |
5284 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5285 | |
8e2e1faf JA |
5286 | if (posted) |
5287 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5288 | |
5289 | return posted != 0; | |
221c5eb2 JA |
5290 | } |
5291 | ||
47f46768 JA |
5292 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5293 | { | |
78076bb6 | 5294 | struct hlist_head *list; |
47f46768 JA |
5295 | struct io_kiocb *req; |
5296 | ||
78076bb6 JA |
5297 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5298 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5299 | if (sqe_addr != req->user_data) |
5300 | continue; | |
5301 | if (io_poll_remove_one(req)) | |
eac406c6 | 5302 | return 0; |
b41e9852 | 5303 | return -EALREADY; |
47f46768 JA |
5304 | } |
5305 | ||
5306 | return -ENOENT; | |
5307 | } | |
5308 | ||
3529d8c2 JA |
5309 | static int io_poll_remove_prep(struct io_kiocb *req, |
5310 | const struct io_uring_sqe *sqe) | |
0969e783 | 5311 | { |
0969e783 JA |
5312 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5313 | return -EINVAL; | |
5314 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5315 | sqe->poll_events) | |
5316 | return -EINVAL; | |
5317 | ||
018043be | 5318 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5319 | return 0; |
5320 | } | |
5321 | ||
221c5eb2 JA |
5322 | /* |
5323 | * Find a running poll command that matches one specified in sqe->addr, | |
5324 | * and remove it if found. | |
5325 | */ | |
61e98203 | 5326 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5327 | { |
5328 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5329 | int ret; |
221c5eb2 | 5330 | |
221c5eb2 | 5331 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5332 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5333 | spin_unlock_irq(&ctx->completion_lock); |
5334 | ||
4e88d6e7 JA |
5335 | if (ret < 0) |
5336 | req_set_fail_links(req); | |
e1e16097 | 5337 | io_req_complete(req, ret); |
221c5eb2 JA |
5338 | return 0; |
5339 | } | |
5340 | ||
221c5eb2 JA |
5341 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5342 | void *key) | |
5343 | { | |
c2f2eb7d JA |
5344 | struct io_kiocb *req = wait->private; |
5345 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5346 | |
d7718a9d | 5347 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5348 | } |
5349 | ||
221c5eb2 JA |
5350 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5351 | struct poll_table_struct *p) | |
5352 | { | |
5353 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5354 | ||
e8c2bc1f | 5355 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5356 | } |
5357 | ||
3529d8c2 | 5358 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5359 | { |
5360 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5361 | u32 events; |
221c5eb2 JA |
5362 | |
5363 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5364 | return -EINVAL; | |
5365 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5366 | return -EINVAL; | |
5367 | ||
5769a351 JX |
5368 | events = READ_ONCE(sqe->poll32_events); |
5369 | #ifdef __BIG_ENDIAN | |
5370 | events = swahw32(events); | |
5371 | #endif | |
a31eb4a2 JX |
5372 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5373 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5374 | return 0; |
5375 | } | |
5376 | ||
61e98203 | 5377 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5378 | { |
5379 | struct io_poll_iocb *poll = &req->poll; | |
5380 | struct io_ring_ctx *ctx = req->ctx; | |
5381 | struct io_poll_table ipt; | |
0969e783 | 5382 | __poll_t mask; |
0969e783 | 5383 | |
d7718a9d | 5384 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5385 | |
d7718a9d JA |
5386 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5387 | io_poll_wake); | |
221c5eb2 | 5388 | |
8c838788 | 5389 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5390 | ipt.error = 0; |
b0dd8a41 | 5391 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5392 | } |
221c5eb2 JA |
5393 | spin_unlock_irq(&ctx->completion_lock); |
5394 | ||
8c838788 JA |
5395 | if (mask) { |
5396 | io_cqring_ev_posted(ctx); | |
014db007 | 5397 | io_put_req(req); |
221c5eb2 | 5398 | } |
8c838788 | 5399 | return ipt.error; |
221c5eb2 JA |
5400 | } |
5401 | ||
5262f567 JA |
5402 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5403 | { | |
ad8a48ac JA |
5404 | struct io_timeout_data *data = container_of(timer, |
5405 | struct io_timeout_data, timer); | |
5406 | struct io_kiocb *req = data->req; | |
5407 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5408 | unsigned long flags; |
5409 | ||
5262f567 | 5410 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5411 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5412 | atomic_set(&req->ctx->cq_timeouts, |
5413 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5414 | ||
78e19bbe | 5415 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5416 | io_commit_cqring(ctx); |
5417 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5418 | ||
5419 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5420 | req_set_fail_links(req); |
5262f567 JA |
5421 | io_put_req(req); |
5422 | return HRTIMER_NORESTART; | |
5423 | } | |
5424 | ||
fbd15848 PB |
5425 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5426 | __u64 user_data) | |
f254ac04 | 5427 | { |
fbd15848 | 5428 | struct io_timeout_data *io; |
47f46768 JA |
5429 | struct io_kiocb *req; |
5430 | int ret = -ENOENT; | |
f254ac04 | 5431 | |
135fcde8 | 5432 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5433 | if (user_data == req->user_data) { |
47f46768 JA |
5434 | ret = 0; |
5435 | break; | |
5436 | } | |
5437 | } | |
5438 | ||
5439 | if (ret == -ENOENT) | |
fbd15848 PB |
5440 | return ERR_PTR(ret); |
5441 | ||
5442 | io = req->async_data; | |
e8c2bc1f | 5443 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5444 | if (ret == -1) |
fbd15848 | 5445 | return ERR_PTR(-EALREADY); |
a71976f3 | 5446 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5447 | return req; |
5448 | } | |
47f46768 | 5449 | |
fbd15848 PB |
5450 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5451 | { | |
5452 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5453 | ||
5454 | if (IS_ERR(req)) | |
5455 | return PTR_ERR(req); | |
f254ac04 JA |
5456 | |
5457 | req_set_fail_links(req); | |
f254ac04 | 5458 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5459 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5460 | return 0; |
5461 | } | |
5462 | ||
9c8e11b3 PB |
5463 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5464 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5465 | { |
9c8e11b3 PB |
5466 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5467 | struct io_timeout_data *data; | |
47f46768 | 5468 | |
9c8e11b3 PB |
5469 | if (IS_ERR(req)) |
5470 | return PTR_ERR(req); | |
47f46768 | 5471 | |
9c8e11b3 PB |
5472 | req->timeout.off = 0; /* noseq */ |
5473 | data = req->async_data; | |
5474 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5475 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5476 | data->timer.function = io_timeout_fn; | |
5477 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5478 | return 0; | |
47f46768 JA |
5479 | } |
5480 | ||
3529d8c2 JA |
5481 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5482 | const struct io_uring_sqe *sqe) | |
b29472ee | 5483 | { |
9c8e11b3 PB |
5484 | struct io_timeout_rem *tr = &req->timeout_rem; |
5485 | ||
b29472ee JA |
5486 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5487 | return -EINVAL; | |
61710e43 DA |
5488 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5489 | return -EINVAL; | |
9c8e11b3 | 5490 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5491 | return -EINVAL; |
5492 | ||
9c8e11b3 PB |
5493 | tr->addr = READ_ONCE(sqe->addr); |
5494 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5495 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5496 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5497 | return -EINVAL; | |
5498 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5499 | return -EFAULT; | |
5500 | } else if (tr->flags) { | |
5501 | /* timeout removal doesn't support flags */ | |
b29472ee | 5502 | return -EINVAL; |
9c8e11b3 | 5503 | } |
b29472ee | 5504 | |
b29472ee JA |
5505 | return 0; |
5506 | } | |
5507 | ||
8662daec PB |
5508 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5509 | { | |
5510 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5511 | : HRTIMER_MODE_REL; | |
5512 | } | |
5513 | ||
11365043 JA |
5514 | /* |
5515 | * Remove or update an existing timeout command | |
5516 | */ | |
61e98203 | 5517 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5518 | { |
9c8e11b3 | 5519 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5520 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5521 | int ret; |
11365043 | 5522 | |
11365043 | 5523 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5524 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5525 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5526 | else |
5527 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5528 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5529 | |
47f46768 | 5530 | io_cqring_fill_event(req, ret); |
11365043 JA |
5531 | io_commit_cqring(ctx); |
5532 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5533 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5534 | if (ret < 0) |
5535 | req_set_fail_links(req); | |
ec9c02ad | 5536 | io_put_req(req); |
11365043 | 5537 | return 0; |
5262f567 JA |
5538 | } |
5539 | ||
3529d8c2 | 5540 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5541 | bool is_timeout_link) |
5262f567 | 5542 | { |
ad8a48ac | 5543 | struct io_timeout_data *data; |
a41525ab | 5544 | unsigned flags; |
56080b02 | 5545 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5546 | |
ad8a48ac | 5547 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5548 | return -EINVAL; |
ad8a48ac | 5549 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5550 | return -EINVAL; |
56080b02 | 5551 | if (off && is_timeout_link) |
2d28390a | 5552 | return -EINVAL; |
a41525ab JA |
5553 | flags = READ_ONCE(sqe->timeout_flags); |
5554 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5555 | return -EINVAL; |
bdf20073 | 5556 | |
bfe68a22 | 5557 | req->timeout.off = off; |
26a61679 | 5558 | |
e8c2bc1f | 5559 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5560 | return -ENOMEM; |
5561 | ||
e8c2bc1f | 5562 | data = req->async_data; |
ad8a48ac | 5563 | data->req = req; |
ad8a48ac JA |
5564 | |
5565 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5566 | return -EFAULT; |
5567 | ||
8662daec | 5568 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5569 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
dd59a3d5 | 5570 | io_req_track_inflight(req); |
ad8a48ac JA |
5571 | return 0; |
5572 | } | |
5573 | ||
61e98203 | 5574 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5575 | { |
ad8a48ac | 5576 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5577 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5578 | struct list_head *entry; |
bfe68a22 | 5579 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5580 | |
733f5c95 | 5581 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5582 | |
5262f567 JA |
5583 | /* |
5584 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5585 | * timeout event to be satisfied. If it isn't set, then this is |
5586 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5587 | */ |
8eb7e2d0 | 5588 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5589 | entry = ctx->timeout_list.prev; |
5590 | goto add; | |
5591 | } | |
5262f567 | 5592 | |
bfe68a22 PB |
5593 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5594 | req->timeout.target_seq = tail + off; | |
5262f567 | 5595 | |
f010505b MDG |
5596 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5597 | * This is safe because ->completion_lock is held, and submissions | |
5598 | * and completions are never mixed in the same ->completion_lock section. | |
5599 | */ | |
5600 | ctx->cq_last_tm_flush = tail; | |
5601 | ||
5262f567 JA |
5602 | /* |
5603 | * Insertion sort, ensuring the first entry in the list is always | |
5604 | * the one we need first. | |
5605 | */ | |
5262f567 | 5606 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5607 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5608 | timeout.list); | |
5262f567 | 5609 | |
8eb7e2d0 | 5610 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5611 | continue; |
bfe68a22 PB |
5612 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5613 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5614 | break; |
5615 | } | |
93bd25bb | 5616 | add: |
135fcde8 | 5617 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5618 | data->timer.function = io_timeout_fn; |
5619 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5620 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5621 | return 0; |
5622 | } | |
5262f567 | 5623 | |
f458dd84 PB |
5624 | struct io_cancel_data { |
5625 | struct io_ring_ctx *ctx; | |
5626 | u64 user_data; | |
5627 | }; | |
5628 | ||
62755e35 JA |
5629 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5630 | { | |
5631 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5632 | struct io_cancel_data *cd = data; |
62755e35 | 5633 | |
f458dd84 | 5634 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5635 | } |
5636 | ||
f458dd84 PB |
5637 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5638 | struct io_ring_ctx *ctx) | |
62755e35 | 5639 | { |
f458dd84 | 5640 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5641 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5642 | int ret = 0; |
5643 | ||
f458dd84 | 5644 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5645 | return -ENOENT; |
5646 | ||
f458dd84 | 5647 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5648 | switch (cancel_ret) { |
5649 | case IO_WQ_CANCEL_OK: | |
5650 | ret = 0; | |
5651 | break; | |
5652 | case IO_WQ_CANCEL_RUNNING: | |
5653 | ret = -EALREADY; | |
5654 | break; | |
5655 | case IO_WQ_CANCEL_NOTFOUND: | |
5656 | ret = -ENOENT; | |
5657 | break; | |
5658 | } | |
5659 | ||
e977d6d3 JA |
5660 | return ret; |
5661 | } | |
5662 | ||
47f46768 JA |
5663 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5664 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5665 | int success_ret) |
47f46768 JA |
5666 | { |
5667 | unsigned long flags; | |
5668 | int ret; | |
5669 | ||
f458dd84 | 5670 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5671 | if (ret != -ENOENT) { |
5672 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5673 | goto done; | |
5674 | } | |
5675 | ||
5676 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5677 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5678 | if (ret != -ENOENT) | |
5679 | goto done; | |
5680 | ret = io_poll_cancel(ctx, sqe_addr); | |
5681 | done: | |
b0dd8a41 JA |
5682 | if (!ret) |
5683 | ret = success_ret; | |
47f46768 JA |
5684 | io_cqring_fill_event(req, ret); |
5685 | io_commit_cqring(ctx); | |
5686 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5687 | io_cqring_ev_posted(ctx); | |
5688 | ||
4e88d6e7 JA |
5689 | if (ret < 0) |
5690 | req_set_fail_links(req); | |
014db007 | 5691 | io_put_req(req); |
47f46768 JA |
5692 | } |
5693 | ||
3529d8c2 JA |
5694 | static int io_async_cancel_prep(struct io_kiocb *req, |
5695 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5696 | { |
fbf23849 | 5697 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5698 | return -EINVAL; |
61710e43 DA |
5699 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5700 | return -EINVAL; | |
5701 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5702 | return -EINVAL; |
5703 | ||
fbf23849 JA |
5704 | req->cancel.addr = READ_ONCE(sqe->addr); |
5705 | return 0; | |
5706 | } | |
5707 | ||
61e98203 | 5708 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5709 | { |
5710 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5711 | u64 sqe_addr = req->cancel.addr; |
5712 | struct io_tctx_node *node; | |
5713 | int ret; | |
5714 | ||
5715 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5716 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5717 | spin_lock_irq(&ctx->completion_lock); | |
5718 | if (ret != -ENOENT) | |
5719 | goto done; | |
5720 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5721 | if (ret != -ENOENT) | |
5722 | goto done; | |
5723 | ret = io_poll_cancel(ctx, sqe_addr); | |
5724 | if (ret != -ENOENT) | |
5725 | goto done; | |
5726 | spin_unlock_irq(&ctx->completion_lock); | |
5727 | ||
5728 | /* slow path, try all io-wq's */ | |
5729 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5730 | ret = -ENOENT; | |
5731 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5732 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5733 | |
58f99373 PB |
5734 | if (!tctx || !tctx->io_wq) |
5735 | continue; | |
5736 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5737 | if (ret != -ENOENT) | |
5738 | break; | |
5739 | } | |
5740 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5741 | ||
5742 | spin_lock_irq(&ctx->completion_lock); | |
5743 | done: | |
5744 | io_cqring_fill_event(req, ret); | |
5745 | io_commit_cqring(ctx); | |
5746 | spin_unlock_irq(&ctx->completion_lock); | |
5747 | io_cqring_ev_posted(ctx); | |
5748 | ||
5749 | if (ret < 0) | |
5750 | req_set_fail_links(req); | |
5751 | io_put_req(req); | |
5262f567 JA |
5752 | return 0; |
5753 | } | |
5754 | ||
269bbe5f | 5755 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5756 | const struct io_uring_sqe *sqe) |
5757 | { | |
6ca56f84 JA |
5758 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5759 | return -EINVAL; | |
61710e43 DA |
5760 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5761 | return -EINVAL; | |
5762 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5763 | return -EINVAL; |
5764 | ||
269bbe5f BM |
5765 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5766 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5767 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5768 | return -EINVAL; |
269bbe5f | 5769 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5770 | return 0; |
5771 | } | |
5772 | ||
889fca73 | 5773 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5774 | { |
5775 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5776 | struct io_uring_rsrc_update up; |
05f3fb3c | 5777 | int ret; |
fbf23849 | 5778 | |
45d189c6 | 5779 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5780 | return -EAGAIN; |
05f3fb3c | 5781 | |
269bbe5f BM |
5782 | up.offset = req->rsrc_update.offset; |
5783 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5784 | |
5785 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5786 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5787 | mutex_unlock(&ctx->uring_lock); |
5788 | ||
5789 | if (ret < 0) | |
5790 | req_set_fail_links(req); | |
889fca73 | 5791 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5792 | return 0; |
5793 | } | |
5794 | ||
bfe76559 | 5795 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5796 | { |
d625c6ee | 5797 | switch (req->opcode) { |
e781573e | 5798 | case IORING_OP_NOP: |
bfe76559 | 5799 | return 0; |
f67676d1 JA |
5800 | case IORING_OP_READV: |
5801 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5802 | case IORING_OP_READ: |
bfe76559 | 5803 | return io_read_prep(req, sqe); |
f67676d1 JA |
5804 | case IORING_OP_WRITEV: |
5805 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5806 | case IORING_OP_WRITE: |
bfe76559 | 5807 | return io_write_prep(req, sqe); |
0969e783 | 5808 | case IORING_OP_POLL_ADD: |
bfe76559 | 5809 | return io_poll_add_prep(req, sqe); |
0969e783 | 5810 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5811 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5812 | case IORING_OP_FSYNC: |
1155c76a | 5813 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5814 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5815 | return io_sfr_prep(req, sqe); |
03b1230c | 5816 | case IORING_OP_SENDMSG: |
fddaface | 5817 | case IORING_OP_SEND: |
bfe76559 | 5818 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5819 | case IORING_OP_RECVMSG: |
fddaface | 5820 | case IORING_OP_RECV: |
bfe76559 | 5821 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5822 | case IORING_OP_CONNECT: |
bfe76559 | 5823 | return io_connect_prep(req, sqe); |
2d28390a | 5824 | case IORING_OP_TIMEOUT: |
bfe76559 | 5825 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5826 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5827 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5828 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5829 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5830 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5831 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5832 | case IORING_OP_ACCEPT: |
bfe76559 | 5833 | return io_accept_prep(req, sqe); |
d63d1b5e | 5834 | case IORING_OP_FALLOCATE: |
bfe76559 | 5835 | return io_fallocate_prep(req, sqe); |
15b71abe | 5836 | case IORING_OP_OPENAT: |
bfe76559 | 5837 | return io_openat_prep(req, sqe); |
b5dba59e | 5838 | case IORING_OP_CLOSE: |
bfe76559 | 5839 | return io_close_prep(req, sqe); |
05f3fb3c | 5840 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5841 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5842 | case IORING_OP_STATX: |
bfe76559 | 5843 | return io_statx_prep(req, sqe); |
4840e418 | 5844 | case IORING_OP_FADVISE: |
bfe76559 | 5845 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5846 | case IORING_OP_MADVISE: |
bfe76559 | 5847 | return io_madvise_prep(req, sqe); |
cebdb986 | 5848 | case IORING_OP_OPENAT2: |
bfe76559 | 5849 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5850 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5851 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5852 | case IORING_OP_SPLICE: |
bfe76559 | 5853 | return io_splice_prep(req, sqe); |
ddf0322d | 5854 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5855 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5856 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5857 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5858 | case IORING_OP_TEE: |
bfe76559 | 5859 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5860 | case IORING_OP_SHUTDOWN: |
5861 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5862 | case IORING_OP_RENAMEAT: |
5863 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5864 | case IORING_OP_UNLINKAT: |
5865 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5866 | } |
5867 | ||
bfe76559 PB |
5868 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5869 | req->opcode); | |
5870 | return-EINVAL; | |
5871 | } | |
5872 | ||
93642ef8 | 5873 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5874 | { |
93642ef8 PB |
5875 | switch (req->opcode) { |
5876 | case IORING_OP_READV: | |
5877 | case IORING_OP_READ_FIXED: | |
5878 | case IORING_OP_READ: | |
5879 | return io_rw_prep_async(req, READ); | |
5880 | case IORING_OP_WRITEV: | |
5881 | case IORING_OP_WRITE_FIXED: | |
5882 | case IORING_OP_WRITE: | |
5883 | return io_rw_prep_async(req, WRITE); | |
5884 | case IORING_OP_SENDMSG: | |
5885 | case IORING_OP_SEND: | |
5886 | return io_sendmsg_prep_async(req); | |
5887 | case IORING_OP_RECVMSG: | |
5888 | case IORING_OP_RECV: | |
5889 | return io_recvmsg_prep_async(req); | |
5890 | case IORING_OP_CONNECT: | |
5891 | return io_connect_prep_async(req); | |
5892 | } | |
5893 | return 0; | |
5894 | } | |
5895 | ||
be7053b7 | 5896 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5897 | { |
be7053b7 | 5898 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5899 | return 0; |
be7053b7 | 5900 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5901 | if (req->async_data) |
be7053b7 PB |
5902 | return 0; |
5903 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5904 | return -EAGAIN; |
be7053b7 | 5905 | return io_req_prep_async(req); |
f67676d1 JA |
5906 | } |
5907 | ||
9cf7c104 PB |
5908 | static u32 io_get_sequence(struct io_kiocb *req) |
5909 | { | |
5910 | struct io_kiocb *pos; | |
5911 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5912 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5913 | |
f2f87370 PB |
5914 | io_for_each_link(pos, req) |
5915 | nr_reqs++; | |
9cf7c104 PB |
5916 | |
5917 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5918 | return total_submitted - nr_reqs; | |
5919 | } | |
5920 | ||
be7053b7 | 5921 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5922 | { |
a197f664 | 5923 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5924 | struct io_defer_entry *de; |
f67676d1 | 5925 | int ret; |
9cf7c104 | 5926 | u32 seq; |
de0617e4 | 5927 | |
9d858b21 | 5928 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5929 | if (likely(list_empty_careful(&ctx->defer_list) && |
5930 | !(req->flags & REQ_F_IO_DRAIN))) | |
5931 | return 0; | |
5932 | ||
5933 | seq = io_get_sequence(req); | |
5934 | /* Still a chance to pass the sequence check */ | |
5935 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5936 | return 0; |
5937 | ||
be7053b7 PB |
5938 | ret = io_req_defer_prep(req); |
5939 | if (ret) | |
5940 | return ret; | |
cbdcb435 | 5941 | io_prep_async_link(req); |
27dc8338 PB |
5942 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5943 | if (!de) | |
5944 | return -ENOMEM; | |
2d28390a | 5945 | |
de0617e4 | 5946 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5947 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5948 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5949 | kfree(de); |
ae34817b PB |
5950 | io_queue_async_work(req); |
5951 | return -EIOCBQUEUED; | |
de0617e4 JA |
5952 | } |
5953 | ||
915967f6 | 5954 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5955 | de->req = req; |
9cf7c104 | 5956 | de->seq = seq; |
27dc8338 | 5957 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5958 | spin_unlock_irq(&ctx->completion_lock); |
5959 | return -EIOCBQUEUED; | |
5960 | } | |
5961 | ||
3ca405eb | 5962 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5963 | { |
0e1b6fe3 PB |
5964 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5965 | switch (req->opcode) { | |
5966 | case IORING_OP_READV: | |
5967 | case IORING_OP_READ_FIXED: | |
5968 | case IORING_OP_READ: | |
bcda7baa | 5969 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5970 | break; |
5971 | case IORING_OP_RECVMSG: | |
5972 | case IORING_OP_RECV: | |
bcda7baa | 5973 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5974 | break; |
5975 | } | |
5976 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5977 | } |
5978 | ||
0e1b6fe3 PB |
5979 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5980 | switch (req->opcode) { | |
5981 | case IORING_OP_READV: | |
5982 | case IORING_OP_READ_FIXED: | |
5983 | case IORING_OP_READ: | |
5984 | case IORING_OP_WRITEV: | |
5985 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5986 | case IORING_OP_WRITE: { |
5987 | struct io_async_rw *io = req->async_data; | |
5988 | if (io->free_iovec) | |
5989 | kfree(io->free_iovec); | |
0e1b6fe3 | 5990 | break; |
e8c2bc1f | 5991 | } |
0e1b6fe3 | 5992 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5993 | case IORING_OP_SENDMSG: { |
5994 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5995 | |
5996 | kfree(io->free_iov); | |
0e1b6fe3 | 5997 | break; |
e8c2bc1f | 5998 | } |
0e1b6fe3 PB |
5999 | case IORING_OP_SPLICE: |
6000 | case IORING_OP_TEE: | |
6001 | io_put_file(req, req->splice.file_in, | |
6002 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6003 | break; | |
f3cd4850 JA |
6004 | case IORING_OP_OPENAT: |
6005 | case IORING_OP_OPENAT2: | |
6006 | if (req->open.filename) | |
6007 | putname(req->open.filename); | |
6008 | break; | |
80a261fd JA |
6009 | case IORING_OP_RENAMEAT: |
6010 | putname(req->rename.oldpath); | |
6011 | putname(req->rename.newpath); | |
6012 | break; | |
14a1143b JA |
6013 | case IORING_OP_UNLINKAT: |
6014 | putname(req->unlink.filename); | |
6015 | break; | |
0e1b6fe3 PB |
6016 | } |
6017 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6018 | } |
99bc4c38 PB |
6019 | } |
6020 | ||
889fca73 | 6021 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6022 | { |
a197f664 | 6023 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6024 | const struct cred *creds = NULL; |
d625c6ee | 6025 | int ret; |
2b188cc1 | 6026 | |
003e8dcc JA |
6027 | if (req->work.creds && req->work.creds != current_cred()) |
6028 | creds = override_creds(req->work.creds); | |
5730b27e | 6029 | |
d625c6ee | 6030 | switch (req->opcode) { |
2b188cc1 | 6031 | case IORING_OP_NOP: |
889fca73 | 6032 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6033 | break; |
6034 | case IORING_OP_READV: | |
edafccee | 6035 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6036 | case IORING_OP_READ: |
889fca73 | 6037 | ret = io_read(req, issue_flags); |
edafccee | 6038 | break; |
3529d8c2 | 6039 | case IORING_OP_WRITEV: |
edafccee | 6040 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6041 | case IORING_OP_WRITE: |
889fca73 | 6042 | ret = io_write(req, issue_flags); |
2b188cc1 | 6043 | break; |
c992fe29 | 6044 | case IORING_OP_FSYNC: |
45d189c6 | 6045 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6046 | break; |
221c5eb2 | 6047 | case IORING_OP_POLL_ADD: |
61e98203 | 6048 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6049 | break; |
6050 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6051 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6052 | break; |
5d17b4a4 | 6053 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6054 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6055 | break; |
0fa03c62 | 6056 | case IORING_OP_SENDMSG: |
889fca73 | 6057 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6058 | break; |
fddaface | 6059 | case IORING_OP_SEND: |
889fca73 | 6060 | ret = io_send(req, issue_flags); |
0fa03c62 | 6061 | break; |
aa1fa28f | 6062 | case IORING_OP_RECVMSG: |
889fca73 | 6063 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6064 | break; |
fddaface | 6065 | case IORING_OP_RECV: |
889fca73 | 6066 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6067 | break; |
5262f567 | 6068 | case IORING_OP_TIMEOUT: |
61e98203 | 6069 | ret = io_timeout(req, issue_flags); |
5262f567 | 6070 | break; |
11365043 | 6071 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6072 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6073 | break; |
17f2fe35 | 6074 | case IORING_OP_ACCEPT: |
889fca73 | 6075 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6076 | break; |
f8e85cf2 | 6077 | case IORING_OP_CONNECT: |
889fca73 | 6078 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6079 | break; |
62755e35 | 6080 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6081 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6082 | break; |
d63d1b5e | 6083 | case IORING_OP_FALLOCATE: |
45d189c6 | 6084 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6085 | break; |
15b71abe | 6086 | case IORING_OP_OPENAT: |
45d189c6 | 6087 | ret = io_openat(req, issue_flags); |
15b71abe | 6088 | break; |
b5dba59e | 6089 | case IORING_OP_CLOSE: |
889fca73 | 6090 | ret = io_close(req, issue_flags); |
b5dba59e | 6091 | break; |
05f3fb3c | 6092 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6093 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6094 | break; |
eddc7ef5 | 6095 | case IORING_OP_STATX: |
45d189c6 | 6096 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6097 | break; |
4840e418 | 6098 | case IORING_OP_FADVISE: |
45d189c6 | 6099 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6100 | break; |
c1ca757b | 6101 | case IORING_OP_MADVISE: |
45d189c6 | 6102 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6103 | break; |
cebdb986 | 6104 | case IORING_OP_OPENAT2: |
45d189c6 | 6105 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6106 | break; |
3e4827b0 | 6107 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6108 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6109 | break; |
7d67af2c | 6110 | case IORING_OP_SPLICE: |
45d189c6 | 6111 | ret = io_splice(req, issue_flags); |
7d67af2c | 6112 | break; |
ddf0322d | 6113 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6114 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6115 | break; |
067524e9 | 6116 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6117 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6118 | break; |
f2a8d5c7 | 6119 | case IORING_OP_TEE: |
45d189c6 | 6120 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6121 | break; |
36f4fa68 | 6122 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6123 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6124 | break; |
80a261fd | 6125 | case IORING_OP_RENAMEAT: |
45d189c6 | 6126 | ret = io_renameat(req, issue_flags); |
80a261fd | 6127 | break; |
14a1143b | 6128 | case IORING_OP_UNLINKAT: |
45d189c6 | 6129 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6130 | break; |
2b188cc1 JA |
6131 | default: |
6132 | ret = -EINVAL; | |
6133 | break; | |
6134 | } | |
6135 | ||
5730b27e JA |
6136 | if (creds) |
6137 | revert_creds(creds); | |
6138 | ||
def596e9 JA |
6139 | if (ret) |
6140 | return ret; | |
6141 | ||
b532576e JA |
6142 | /* If the op doesn't have a file, we're not polling for it */ |
6143 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6144 | const bool in_async = io_wq_current_is_worker(); |
6145 | ||
11ba820b JA |
6146 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6147 | if (in_async) | |
6148 | mutex_lock(&ctx->uring_lock); | |
6149 | ||
2e9dbe90 | 6150 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6151 | |
6152 | if (in_async) | |
6153 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6154 | } |
6155 | ||
6156 | return 0; | |
2b188cc1 JA |
6157 | } |
6158 | ||
5280f7e5 | 6159 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6160 | { |
6161 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6162 | struct io_kiocb *timeout; |
561fb04a | 6163 | int ret = 0; |
2b188cc1 | 6164 | |
6df1db6b PB |
6165 | timeout = io_prep_linked_timeout(req); |
6166 | if (timeout) | |
6167 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6168 | |
4014d943 | 6169 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6170 | ret = -ECANCELED; |
31b51510 | 6171 | |
561fb04a | 6172 | if (!ret) { |
561fb04a | 6173 | do { |
889fca73 | 6174 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6175 | /* |
6176 | * We can get EAGAIN for polled IO even though we're | |
6177 | * forcing a sync submission from here, since we can't | |
6178 | * wait for request slots on the block side. | |
6179 | */ | |
6180 | if (ret != -EAGAIN) | |
6181 | break; | |
6182 | cond_resched(); | |
6183 | } while (1); | |
6184 | } | |
31b51510 | 6185 | |
a3df7698 | 6186 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6187 | if (ret) { |
a3df7698 PB |
6188 | /* io-wq is going to take one down */ |
6189 | refcount_inc(&req->refs); | |
6190 | io_req_task_queue_fail(req, ret); | |
edafccee | 6191 | } |
2b188cc1 JA |
6192 | } |
6193 | ||
65e19f54 JA |
6194 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6195 | int index) | |
6196 | { | |
269bbe5f | 6197 | struct fixed_rsrc_table *table; |
65e19f54 | 6198 | |
05f3fb3c | 6199 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6200 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6201 | } |
6202 | ||
8371adf5 PB |
6203 | static struct file *io_file_get(struct io_submit_state *state, |
6204 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6205 | { |
a197f664 | 6206 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6207 | struct file *file; |
09bb8394 | 6208 | |
8da11c19 | 6209 | if (fixed) { |
479f517b | 6210 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6211 | return NULL; |
b7620121 | 6212 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6213 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6214 | io_set_resource_node(req); |
09bb8394 | 6215 | } else { |
c826bd7a | 6216 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6217 | file = __io_file_get(state, fd); |
09bb8394 JA |
6218 | } |
6219 | ||
ce3d5aae PB |
6220 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6221 | io_req_track_inflight(req); | |
8371adf5 | 6222 | return file; |
09bb8394 JA |
6223 | } |
6224 | ||
2665abfd | 6225 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6226 | { |
ad8a48ac JA |
6227 | struct io_timeout_data *data = container_of(timer, |
6228 | struct io_timeout_data, timer); | |
90cd7e42 | 6229 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6230 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6231 | unsigned long flags; |
2665abfd JA |
6232 | |
6233 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6234 | prev = req->timeout.head; |
6235 | req->timeout.head = NULL; | |
2665abfd JA |
6236 | |
6237 | /* | |
6238 | * We don't expect the list to be empty, that will only happen if we | |
6239 | * race with the completion of the linked work. | |
6240 | */ | |
90cd7e42 | 6241 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6242 | io_remove_next_linked(prev); |
90cd7e42 PB |
6243 | else |
6244 | prev = NULL; | |
2665abfd JA |
6245 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6246 | ||
6247 | if (prev) { | |
014db007 | 6248 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6249 | io_put_req_deferred(prev, 1); |
47f46768 | 6250 | } else { |
9ae1f8dd PB |
6251 | io_req_complete_post(req, -ETIME, 0); |
6252 | io_put_req_deferred(req, 1); | |
2665abfd | 6253 | } |
2665abfd JA |
6254 | return HRTIMER_NORESTART; |
6255 | } | |
6256 | ||
7271ef3a | 6257 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6258 | { |
76a46e06 | 6259 | /* |
f2f87370 PB |
6260 | * If the back reference is NULL, then our linked request finished |
6261 | * before we got a chance to setup the timer | |
76a46e06 | 6262 | */ |
90cd7e42 | 6263 | if (req->timeout.head) { |
e8c2bc1f | 6264 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6265 | |
ad8a48ac JA |
6266 | data->timer.function = io_link_timeout_fn; |
6267 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6268 | data->mode); | |
2665abfd | 6269 | } |
7271ef3a JA |
6270 | } |
6271 | ||
6272 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6273 | { | |
6274 | struct io_ring_ctx *ctx = req->ctx; | |
6275 | ||
6276 | spin_lock_irq(&ctx->completion_lock); | |
6277 | __io_queue_linked_timeout(req); | |
76a46e06 | 6278 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6279 | |
2665abfd | 6280 | /* drop submission reference */ |
76a46e06 JA |
6281 | io_put_req(req); |
6282 | } | |
2665abfd | 6283 | |
ad8a48ac | 6284 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6285 | { |
f2f87370 | 6286 | struct io_kiocb *nxt = req->link; |
2665abfd | 6287 | |
f2f87370 PB |
6288 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6289 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6290 | return NULL; |
2665abfd | 6291 | |
90cd7e42 | 6292 | nxt->timeout.head = req; |
900fad45 | 6293 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6294 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6295 | return nxt; |
2665abfd JA |
6296 | } |
6297 | ||
c5eef2b9 | 6298 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6299 | { |
d3d7298d | 6300 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6301 | int ret; |
2b188cc1 | 6302 | |
c5eef2b9 | 6303 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6304 | |
491381ce JA |
6305 | /* |
6306 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6307 | * doesn't support non-blocking read/write attempts | |
6308 | */ | |
24c74678 | 6309 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6310 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6311 | /* |
6312 | * Queued up for async execution, worker will release | |
6313 | * submit reference when the iocb is actually submitted. | |
6314 | */ | |
6315 | io_queue_async_work(req); | |
2b188cc1 | 6316 | } |
0d63c148 PB |
6317 | } else if (likely(!ret)) { |
6318 | /* drop submission reference */ | |
e342c807 | 6319 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6320 | struct io_ring_ctx *ctx = req->ctx; |
6321 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6322 | |
6dd0be1e | 6323 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6324 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6325 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6326 | } else { |
d3d7298d | 6327 | io_put_req(req); |
0d63c148 PB |
6328 | } |
6329 | } else { | |
4e88d6e7 | 6330 | req_set_fail_links(req); |
e65ef56d | 6331 | io_put_req(req); |
e1e16097 | 6332 | io_req_complete(req, ret); |
9e645e11 | 6333 | } |
d3d7298d PB |
6334 | if (linked_timeout) |
6335 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6336 | } |
6337 | ||
be7053b7 | 6338 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6339 | { |
6340 | int ret; | |
6341 | ||
be7053b7 | 6342 | ret = io_req_defer(req); |
4fe2c963 JL |
6343 | if (ret) { |
6344 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6345 | fail_req: |
4e88d6e7 | 6346 | req_set_fail_links(req); |
e1e16097 JA |
6347 | io_put_req(req); |
6348 | io_req_complete(req, ret); | |
4fe2c963 | 6349 | } |
2550878f | 6350 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6351 | ret = io_req_defer_prep(req); |
6352 | if (unlikely(ret)) | |
6353 | goto fail_req; | |
ce35a47a JA |
6354 | io_queue_async_work(req); |
6355 | } else { | |
c5eef2b9 | 6356 | __io_queue_sqe(req); |
ce35a47a | 6357 | } |
4fe2c963 JL |
6358 | } |
6359 | ||
b16fed66 PB |
6360 | /* |
6361 | * Check SQE restrictions (opcode and flags). | |
6362 | * | |
6363 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6364 | */ | |
6365 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6366 | struct io_kiocb *req, | |
6367 | unsigned int sqe_flags) | |
4fe2c963 | 6368 | { |
b16fed66 PB |
6369 | if (!ctx->restricted) |
6370 | return true; | |
6371 | ||
6372 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6373 | return false; | |
6374 | ||
6375 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6376 | ctx->restrictions.sqe_flags_required) | |
6377 | return false; | |
6378 | ||
6379 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6380 | ctx->restrictions.sqe_flags_required)) | |
6381 | return false; | |
6382 | ||
6383 | return true; | |
4fe2c963 JL |
6384 | } |
6385 | ||
b16fed66 PB |
6386 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6387 | const struct io_uring_sqe *sqe) | |
6388 | { | |
6389 | struct io_submit_state *state; | |
6390 | unsigned int sqe_flags; | |
003e8dcc | 6391 | int personality, ret = 0; |
b16fed66 PB |
6392 | |
6393 | req->opcode = READ_ONCE(sqe->opcode); | |
6394 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6395 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6396 | req->user_data = READ_ONCE(sqe->user_data); | |
6397 | req->async_data = NULL; | |
6398 | req->file = NULL; | |
6399 | req->ctx = ctx; | |
6400 | req->link = NULL; | |
6401 | req->fixed_rsrc_refs = NULL; | |
6402 | /* one is dropped after submission, the other at completion */ | |
6403 | refcount_set(&req->refs, 2); | |
6404 | req->task = current; | |
6405 | req->result = 0; | |
93e68e03 JA |
6406 | req->work.list.next = NULL; |
6407 | req->work.creds = NULL; | |
6408 | req->work.flags = 0; | |
b16fed66 PB |
6409 | |
6410 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6411 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6412 | req->flags = 0; | |
b16fed66 | 6413 | return -EINVAL; |
ebf4a5db | 6414 | } |
b16fed66 PB |
6415 | |
6416 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6417 | return -EINVAL; | |
6418 | ||
b16fed66 PB |
6419 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6420 | return -EACCES; | |
6421 | ||
6422 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6423 | !io_op_defs[req->opcode].buffer_select) | |
6424 | return -EOPNOTSUPP; | |
863e0560 | 6425 | |
003e8dcc JA |
6426 | personality = READ_ONCE(sqe->personality); |
6427 | if (personality) { | |
61cf9370 | 6428 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6429 | if (!req->work.creds) |
6430 | return -EINVAL; | |
6431 | get_cred(req->work.creds); | |
003e8dcc | 6432 | } |
b16fed66 PB |
6433 | state = &ctx->submit_state; |
6434 | ||
6435 | /* | |
6436 | * Plug now if we have more than 1 IO left after this, and the target | |
6437 | * is potentially a read/write to block based storage. | |
6438 | */ | |
6439 | if (!state->plug_started && state->ios_left > 1 && | |
6440 | io_op_defs[req->opcode].plug) { | |
6441 | blk_start_plug(&state->plug); | |
6442 | state->plug_started = true; | |
6443 | } | |
6444 | ||
6445 | if (io_op_defs[req->opcode].needs_file) { | |
6446 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6447 | ||
6448 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6449 | if (unlikely(!req->file)) | |
6450 | ret = -EBADF; | |
6451 | } | |
6452 | ||
6453 | state->ios_left--; | |
6454 | return ret; | |
6455 | } | |
6456 | ||
a6b8cadc | 6457 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6458 | const struct io_uring_sqe *sqe) |
9e645e11 | 6459 | { |
a1ab7b35 | 6460 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6461 | int ret; |
9e645e11 | 6462 | |
a6b8cadc PB |
6463 | ret = io_init_req(ctx, req, sqe); |
6464 | if (unlikely(ret)) { | |
6465 | fail_req: | |
6466 | io_put_req(req); | |
6467 | io_req_complete(req, ret); | |
de59bc10 PB |
6468 | if (link->head) { |
6469 | /* fail even hard links since we don't submit */ | |
cf109604 | 6470 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6471 | io_put_req(link->head); |
6472 | io_req_complete(link->head, -ECANCELED); | |
6473 | link->head = NULL; | |
6474 | } | |
a6b8cadc PB |
6475 | return ret; |
6476 | } | |
be7053b7 PB |
6477 | ret = io_req_prep(req, sqe); |
6478 | if (unlikely(ret)) | |
6479 | goto fail_req; | |
a6b8cadc | 6480 | |
be7053b7 | 6481 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6482 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6483 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6484 | ||
9e645e11 JA |
6485 | /* |
6486 | * If we already have a head request, queue this one for async | |
6487 | * submittal once the head completes. If we don't have a head but | |
6488 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6489 | * submitted sync once the chain is complete. If none of those | |
6490 | * conditions are true (normal request), then just queue it. | |
6491 | */ | |
863e0560 PB |
6492 | if (link->head) { |
6493 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6494 | |
8cdf2193 PB |
6495 | /* |
6496 | * Taking sequential execution of a link, draining both sides | |
6497 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6498 | * requests in the link. So, it drains the head and the | |
6499 | * next after the link request. The last one is done via | |
6500 | * drain_next flag to persist the effect across calls. | |
6501 | */ | |
ef4ff581 | 6502 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6503 | head->flags |= REQ_F_IO_DRAIN; |
6504 | ctx->drain_next = 1; | |
6505 | } | |
be7053b7 | 6506 | ret = io_req_defer_prep(req); |
cf109604 | 6507 | if (unlikely(ret)) |
a6b8cadc | 6508 | goto fail_req; |
9d76377f | 6509 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6510 | link->last->link = req; |
863e0560 | 6511 | link->last = req; |
32fe525b PB |
6512 | |
6513 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6514 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6515 | io_queue_sqe(head); |
863e0560 | 6516 | link->head = NULL; |
32fe525b | 6517 | } |
9e645e11 | 6518 | } else { |
711be031 PB |
6519 | if (unlikely(ctx->drain_next)) { |
6520 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6521 | ctx->drain_next = 0; |
711be031 | 6522 | } |
ef4ff581 | 6523 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6524 | link->head = req; |
6525 | link->last = req; | |
711be031 | 6526 | } else { |
be7053b7 | 6527 | io_queue_sqe(req); |
711be031 | 6528 | } |
9e645e11 | 6529 | } |
2e6e1fde | 6530 | |
1d4240cc | 6531 | return 0; |
9e645e11 JA |
6532 | } |
6533 | ||
9a56a232 JA |
6534 | /* |
6535 | * Batched submission is done, ensure local IO is flushed out. | |
6536 | */ | |
ba88ff11 PB |
6537 | static void io_submit_state_end(struct io_submit_state *state, |
6538 | struct io_ring_ctx *ctx) | |
9a56a232 | 6539 | { |
a1ab7b35 | 6540 | if (state->link.head) |
de59bc10 | 6541 | io_queue_sqe(state->link.head); |
6dd0be1e | 6542 | if (state->comp.nr) |
ba88ff11 | 6543 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6544 | if (state->plug_started) |
6545 | blk_finish_plug(&state->plug); | |
9f13c35b | 6546 | io_state_file_put(state); |
9a56a232 JA |
6547 | } |
6548 | ||
6549 | /* | |
6550 | * Start submission side cache. | |
6551 | */ | |
6552 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6553 | unsigned int max_ios) |
9a56a232 | 6554 | { |
27926b68 | 6555 | state->plug_started = false; |
9a56a232 | 6556 | state->ios_left = max_ios; |
a1ab7b35 PB |
6557 | /* set only head, no need to init link_last in advance */ |
6558 | state->link.head = NULL; | |
9a56a232 JA |
6559 | } |
6560 | ||
2b188cc1 JA |
6561 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6562 | { | |
75b28aff | 6563 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6564 | |
caf582c6 PB |
6565 | /* |
6566 | * Ensure any loads from the SQEs are done at this point, | |
6567 | * since once we write the new head, the application could | |
6568 | * write new data to them. | |
6569 | */ | |
6570 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6571 | } |
6572 | ||
2b188cc1 | 6573 | /* |
3529d8c2 | 6574 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6575 | * that is mapped by userspace. This means that care needs to be taken to |
6576 | * ensure that reads are stable, as we cannot rely on userspace always | |
6577 | * being a good citizen. If members of the sqe are validated and then later | |
6578 | * used, it's important that those reads are done through READ_ONCE() to | |
6579 | * prevent a re-load down the line. | |
6580 | */ | |
709b302f | 6581 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6582 | { |
75b28aff | 6583 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6584 | unsigned head; |
6585 | ||
6586 | /* | |
6587 | * The cached sq head (or cq tail) serves two purposes: | |
6588 | * | |
6589 | * 1) allows us to batch the cost of updating the user visible | |
6590 | * head updates. | |
6591 | * 2) allows the kernel side to track the head on its own, even | |
6592 | * though the application is the one updating it. | |
6593 | */ | |
4fccfcbb | 6594 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6595 | if (likely(head < ctx->sq_entries)) |
6596 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6597 | |
6598 | /* drop invalid entries */ | |
498ccd9e | 6599 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6600 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6601 | return NULL; |
6602 | } | |
6603 | ||
0f212204 | 6604 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6605 | { |
46c4e16a | 6606 | int submitted = 0; |
6c271ce2 | 6607 | |
c4a2ed72 | 6608 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6609 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6610 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6611 | return -EBUSY; |
6612 | } | |
6c271ce2 | 6613 | |
ee7d46d9 PB |
6614 | /* make sure SQ entry isn't read before tail */ |
6615 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6616 | |
2b85edfc PB |
6617 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6618 | return -EAGAIN; | |
6c271ce2 | 6619 | |
d8a6df10 | 6620 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6621 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6622 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6623 | |
46c4e16a | 6624 | while (submitted < nr) { |
3529d8c2 | 6625 | const struct io_uring_sqe *sqe; |
196be95c | 6626 | struct io_kiocb *req; |
fb5ccc98 | 6627 | |
258b29a9 | 6628 | req = io_alloc_req(ctx); |
196be95c PB |
6629 | if (unlikely(!req)) { |
6630 | if (!submitted) | |
6631 | submitted = -EAGAIN; | |
fb5ccc98 | 6632 | break; |
196be95c | 6633 | } |
4fccfcbb PB |
6634 | sqe = io_get_sqe(ctx); |
6635 | if (unlikely(!sqe)) { | |
6636 | kmem_cache_free(req_cachep, req); | |
6637 | break; | |
6638 | } | |
d3656344 JA |
6639 | /* will complete beyond this point, count as submitted */ |
6640 | submitted++; | |
a1ab7b35 | 6641 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6642 | break; |
6c271ce2 JA |
6643 | } |
6644 | ||
9466f437 PB |
6645 | if (unlikely(submitted != nr)) { |
6646 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6647 | struct io_uring_task *tctx = current->io_uring; |
6648 | int unused = nr - ref_used; | |
9466f437 | 6649 | |
d8a6df10 JA |
6650 | percpu_ref_put_many(&ctx->refs, unused); |
6651 | percpu_counter_sub(&tctx->inflight, unused); | |
6652 | put_task_struct_many(current, unused); | |
9466f437 | 6653 | } |
6c271ce2 | 6654 | |
a1ab7b35 | 6655 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6656 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6657 | io_commit_sqring(ctx); | |
6658 | ||
6c271ce2 JA |
6659 | return submitted; |
6660 | } | |
6661 | ||
23b3628e XW |
6662 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6663 | { | |
6664 | /* Tell userspace we may need a wakeup call */ | |
6665 | spin_lock_irq(&ctx->completion_lock); | |
6666 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6667 | spin_unlock_irq(&ctx->completion_lock); | |
6668 | } | |
6669 | ||
6670 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6671 | { | |
6672 | spin_lock_irq(&ctx->completion_lock); | |
6673 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6674 | spin_unlock_irq(&ctx->completion_lock); | |
6675 | } | |
6676 | ||
08369246 | 6677 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6678 | { |
c8d1ba58 | 6679 | unsigned int to_submit; |
bdcd3eab | 6680 | int ret = 0; |
6c271ce2 | 6681 | |
c8d1ba58 | 6682 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6683 | /* if we're handling multiple rings, cap submit size for fairness */ |
6684 | if (cap_entries && to_submit > 8) | |
6685 | to_submit = 8; | |
6686 | ||
906a3c6f | 6687 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6688 | unsigned nr_events = 0; |
a4c0b3de | 6689 | |
c8d1ba58 | 6690 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6691 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6692 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6693 | |
0298ef96 PB |
6694 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6695 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6696 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6697 | mutex_unlock(&ctx->uring_lock); |
6698 | } | |
6c271ce2 | 6699 | |
90554200 JA |
6700 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6701 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6702 | |
08369246 XW |
6703 | return ret; |
6704 | } | |
6c271ce2 | 6705 | |
08369246 XW |
6706 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6707 | { | |
6708 | struct io_ring_ctx *ctx; | |
6709 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6710 | |
08369246 XW |
6711 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6712 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6713 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6714 | } |
c1edbf5f | 6715 | |
08369246 | 6716 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6717 | } |
6c271ce2 | 6718 | |
c8d1ba58 JA |
6719 | static int io_sq_thread(void *data) |
6720 | { | |
69fb2131 JA |
6721 | struct io_sq_data *sqd = data; |
6722 | struct io_ring_ctx *ctx; | |
a0d9205f | 6723 | unsigned long timeout = 0; |
37d1e2e3 | 6724 | char buf[TASK_COMM_LEN]; |
08369246 | 6725 | DEFINE_WAIT(wait); |
6c271ce2 | 6726 | |
37d1e2e3 JA |
6727 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6728 | set_task_comm(current, buf); | |
37d1e2e3 JA |
6729 | current->pf_io_worker = NULL; |
6730 | ||
6731 | if (sqd->sq_cpu != -1) | |
6732 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6733 | else | |
6734 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6735 | current->flags |= PF_NO_SETAFFINITY; | |
6736 | ||
09a6f4ef | 6737 | mutex_lock(&sqd->lock); |
05962f95 | 6738 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6739 | int ret; |
6740 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6741 | |
05962f95 | 6742 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
09a6f4ef | 6743 | mutex_unlock(&sqd->lock); |
05962f95 | 6744 | cond_resched(); |
09a6f4ef | 6745 | mutex_lock(&sqd->lock); |
521d6a73 | 6746 | io_run_task_work(); |
b7f5a0bf | 6747 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6748 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6749 | continue; |
08369246 | 6750 | } |
37d1e2e3 JA |
6751 | if (fatal_signal_pending(current)) |
6752 | break; | |
08369246 | 6753 | sqt_spin = false; |
e95eee2d | 6754 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6755 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6756 | const struct cred *creds = NULL; |
6757 | ||
6758 | if (ctx->sq_creds != current_cred()) | |
6759 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6760 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6761 | if (creds) |
6762 | revert_creds(creds); | |
08369246 XW |
6763 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6764 | sqt_spin = true; | |
69fb2131 | 6765 | } |
6c271ce2 | 6766 | |
08369246 | 6767 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6768 | io_run_task_work(); |
6769 | cond_resched(); | |
08369246 XW |
6770 | if (sqt_spin) |
6771 | timeout = jiffies + sqd->sq_thread_idle; | |
6772 | continue; | |
6773 | } | |
6774 | ||
08369246 XW |
6775 | needs_sched = true; |
6776 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6777 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6778 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6779 | !list_empty_careful(&ctx->iopoll_list)) { | |
6780 | needs_sched = false; | |
6781 | break; | |
6782 | } | |
6783 | if (io_sqring_entries(ctx)) { | |
6784 | needs_sched = false; | |
6785 | break; | |
6786 | } | |
6787 | } | |
6788 | ||
05962f95 | 6789 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6790 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6791 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6792 | |
09a6f4ef | 6793 | mutex_unlock(&sqd->lock); |
69fb2131 | 6794 | schedule(); |
16efa4fc | 6795 | try_to_freeze(); |
09a6f4ef | 6796 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6797 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6798 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6799 | } |
08369246 XW |
6800 | |
6801 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6802 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6803 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6804 | } |
28cea78a | 6805 | |
521d6a73 PB |
6806 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6807 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6808 | sqd->thread = NULL; |
05962f95 | 6809 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6810 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6811 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6812 | |
6813 | io_run_task_work(); | |
b7f5a0bf | 6814 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6815 | complete(&sqd->exited); |
6816 | do_exit(0); | |
6c271ce2 JA |
6817 | } |
6818 | ||
bda52162 JA |
6819 | struct io_wait_queue { |
6820 | struct wait_queue_entry wq; | |
6821 | struct io_ring_ctx *ctx; | |
6822 | unsigned to_wait; | |
6823 | unsigned nr_timeouts; | |
6824 | }; | |
6825 | ||
6c503150 | 6826 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6827 | { |
6828 | struct io_ring_ctx *ctx = iowq->ctx; | |
6829 | ||
6830 | /* | |
d195a66e | 6831 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6832 | * started waiting. For timeouts, we always want to return to userspace, |
6833 | * regardless of event count. | |
6834 | */ | |
6c503150 | 6835 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6836 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6837 | } | |
6838 | ||
6839 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6840 | int wake_flags, void *key) | |
6841 | { | |
6842 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6843 | wq); | |
6844 | ||
6c503150 PB |
6845 | /* |
6846 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6847 | * the task, and the next invocation will do it. | |
6848 | */ | |
6849 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6850 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6851 | return -1; | |
bda52162 JA |
6852 | } |
6853 | ||
af9c1a44 JA |
6854 | static int io_run_task_work_sig(void) |
6855 | { | |
6856 | if (io_run_task_work()) | |
6857 | return 1; | |
6858 | if (!signal_pending(current)) | |
6859 | return 0; | |
792ee0f6 JA |
6860 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6861 | return -ERESTARTSYS; | |
af9c1a44 JA |
6862 | return -EINTR; |
6863 | } | |
6864 | ||
eeb60b9a PB |
6865 | /* when returns >0, the caller should retry */ |
6866 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6867 | struct io_wait_queue *iowq, | |
6868 | signed long *timeout) | |
6869 | { | |
6870 | int ret; | |
6871 | ||
6872 | /* make sure we run task_work before checking for signals */ | |
6873 | ret = io_run_task_work_sig(); | |
6874 | if (ret || io_should_wake(iowq)) | |
6875 | return ret; | |
6876 | /* let the caller flush overflows, retry */ | |
6877 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6878 | return 1; | |
6879 | ||
6880 | *timeout = schedule_timeout(*timeout); | |
6881 | return !*timeout ? -ETIME : 1; | |
6882 | } | |
6883 | ||
2b188cc1 JA |
6884 | /* |
6885 | * Wait until events become available, if we don't already have some. The | |
6886 | * application must reap them itself, as they reside on the shared cq ring. | |
6887 | */ | |
6888 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6889 | const sigset_t __user *sig, size_t sigsz, |
6890 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6891 | { |
bda52162 JA |
6892 | struct io_wait_queue iowq = { |
6893 | .wq = { | |
6894 | .private = current, | |
6895 | .func = io_wake_function, | |
6896 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6897 | }, | |
6898 | .ctx = ctx, | |
6899 | .to_wait = min_events, | |
6900 | }; | |
75b28aff | 6901 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6902 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6903 | int ret; | |
2b188cc1 | 6904 | |
b41e9852 | 6905 | do { |
6c503150 PB |
6906 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6907 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6908 | return 0; |
4c6e277c | 6909 | if (!io_run_task_work()) |
b41e9852 | 6910 | break; |
b41e9852 | 6911 | } while (1); |
2b188cc1 JA |
6912 | |
6913 | if (sig) { | |
9e75ad5d AB |
6914 | #ifdef CONFIG_COMPAT |
6915 | if (in_compat_syscall()) | |
6916 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6917 | sigsz); |
9e75ad5d AB |
6918 | else |
6919 | #endif | |
b772434b | 6920 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6921 | |
2b188cc1 JA |
6922 | if (ret) |
6923 | return ret; | |
6924 | } | |
6925 | ||
c73ebb68 | 6926 | if (uts) { |
c1d5a224 PB |
6927 | struct timespec64 ts; |
6928 | ||
c73ebb68 HX |
6929 | if (get_timespec64(&ts, uts)) |
6930 | return -EFAULT; | |
6931 | timeout = timespec64_to_jiffies(&ts); | |
6932 | } | |
6933 | ||
bda52162 | 6934 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6935 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6936 | do { |
ca0a2651 JA |
6937 | /* if we can't even flush overflow, don't wait for more */ |
6938 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6939 | ret = -EBUSY; | |
6940 | break; | |
6941 | } | |
bda52162 JA |
6942 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6943 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6944 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6945 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6946 | cond_resched(); |
eeb60b9a | 6947 | } while (ret > 0); |
bda52162 | 6948 | |
b7db41c9 | 6949 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6950 | |
75b28aff | 6951 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6952 | } |
6953 | ||
6b06314c JA |
6954 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6955 | { | |
6956 | #if defined(CONFIG_UNIX) | |
6957 | if (ctx->ring_sock) { | |
6958 | struct sock *sock = ctx->ring_sock->sk; | |
6959 | struct sk_buff *skb; | |
6960 | ||
6961 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6962 | kfree_skb(skb); | |
6963 | } | |
6964 | #else | |
6965 | int i; | |
6966 | ||
65e19f54 JA |
6967 | for (i = 0; i < ctx->nr_user_files; i++) { |
6968 | struct file *file; | |
6969 | ||
6970 | file = io_file_from_index(ctx, i); | |
6971 | if (file) | |
6972 | fput(file); | |
6973 | } | |
6b06314c JA |
6974 | #endif |
6975 | } | |
6976 | ||
00835dce | 6977 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6978 | { |
269bbe5f | 6979 | struct fixed_rsrc_data *data; |
05f3fb3c | 6980 | |
269bbe5f | 6981 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6982 | complete(&data->done); |
6983 | } | |
6984 | ||
2a63b2d9 | 6985 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6986 | { |
2a63b2d9 | 6987 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6988 | } |
6989 | ||
2a63b2d9 | 6990 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6991 | { |
2a63b2d9 BM |
6992 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6993 | } | |
65e19f54 | 6994 | |
d67d2263 BM |
6995 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6996 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6997 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6998 | { |
2a63b2d9 | 6999 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7000 | rsrc_data->node = ref_node; |
d67d2263 | 7001 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7002 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7003 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7004 | } |
7005 | ||
8bad28d8 | 7006 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 7007 | { |
8bad28d8 | 7008 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 7009 | |
2a63b2d9 | 7010 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7011 | ref_node = data->node; |
e6cb007c | 7012 | data->node = NULL; |
2a63b2d9 | 7013 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7014 | if (ref_node) |
7015 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
7016 | } |
7017 | ||
7018 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
7019 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
7020 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7021 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7022 | { |
f2303b1f | 7023 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 7024 | int ret; |
05589553 | 7025 | |
8bad28d8 HX |
7026 | if (data->quiesce) |
7027 | return -ENXIO; | |
05589553 | 7028 | |
8bad28d8 | 7029 | data->quiesce = true; |
1ffc5422 | 7030 | do { |
f2303b1f PB |
7031 | ret = -ENOMEM; |
7032 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7033 | if (!backup_node) | |
7034 | break; | |
7035 | backup_node->rsrc_data = data; | |
7036 | backup_node->rsrc_put = rsrc_put; | |
7037 | ||
8bad28d8 HX |
7038 | io_sqe_rsrc_kill_node(ctx, data); |
7039 | percpu_ref_kill(&data->refs); | |
7040 | flush_delayed_work(&ctx->rsrc_put_work); | |
7041 | ||
1ffc5422 PB |
7042 | ret = wait_for_completion_interruptible(&data->done); |
7043 | if (!ret) | |
7044 | break; | |
8bad28d8 | 7045 | |
cb5e1b81 | 7046 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7047 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7048 | backup_node = NULL; | |
cb5e1b81 | 7049 | reinit_completion(&data->done); |
8bad28d8 | 7050 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7051 | ret = io_run_task_work_sig(); |
8bad28d8 | 7052 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7053 | } while (ret >= 0); |
8bad28d8 | 7054 | data->quiesce = false; |
05f3fb3c | 7055 | |
8bad28d8 HX |
7056 | if (backup_node) |
7057 | destroy_fixed_rsrc_ref_node(backup_node); | |
7058 | return ret; | |
d7954b2b BM |
7059 | } |
7060 | ||
1ad555c6 BM |
7061 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7062 | { | |
7063 | struct fixed_rsrc_data *data; | |
7064 | ||
7065 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7066 | if (!data) | |
7067 | return NULL; | |
7068 | ||
00835dce | 7069 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7070 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7071 | kfree(data); | |
7072 | return NULL; | |
7073 | } | |
7074 | data->ctx = ctx; | |
7075 | init_completion(&data->done); | |
7076 | return data; | |
7077 | } | |
7078 | ||
7079 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7080 | { | |
7081 | percpu_ref_exit(&data->refs); | |
7082 | kfree(data->table); | |
7083 | kfree(data); | |
7084 | } | |
7085 | ||
d7954b2b BM |
7086 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7087 | { | |
7088 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7089 | unsigned nr_tables, i; |
7090 | int ret; | |
7091 | ||
8bad28d8 HX |
7092 | /* |
7093 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7094 | * Since we possibly drop uring lock later in this function to | |
7095 | * run task work. | |
7096 | */ | |
7097 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7098 | return -ENXIO; |
f2303b1f | 7099 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7100 | if (ret) |
7101 | return ret; | |
7102 | ||
6b06314c | 7103 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7104 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7105 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7106 | kfree(data->table[i].files); |
1ad555c6 | 7107 | free_fixed_rsrc_data(data); |
05f3fb3c | 7108 | ctx->file_data = NULL; |
6b06314c JA |
7109 | ctx->nr_user_files = 0; |
7110 | return 0; | |
7111 | } | |
7112 | ||
37d1e2e3 | 7113 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7114 | __releases(&sqd->lock) |
37d1e2e3 | 7115 | { |
521d6a73 PB |
7116 | WARN_ON_ONCE(sqd->thread == current); |
7117 | ||
9e138a48 PB |
7118 | /* |
7119 | * Do the dance but not conditional clear_bit() because it'd race with | |
7120 | * other threads incrementing park_pending and setting the bit. | |
7121 | */ | |
37d1e2e3 | 7122 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7123 | if (atomic_dec_return(&sqd->park_pending)) |
7124 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7125 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7126 | } |
7127 | ||
86e0d676 | 7128 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7129 | __acquires(&sqd->lock) |
37d1e2e3 | 7130 | { |
521d6a73 PB |
7131 | WARN_ON_ONCE(sqd->thread == current); |
7132 | ||
9e138a48 | 7133 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7134 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7135 | mutex_lock(&sqd->lock); |
05962f95 | 7136 | if (sqd->thread) |
86e0d676 | 7137 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7138 | } |
7139 | ||
7140 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7141 | { | |
521d6a73 PB |
7142 | WARN_ON_ONCE(sqd->thread == current); |
7143 | ||
09a6f4ef | 7144 | mutex_lock(&sqd->lock); |
05962f95 | 7145 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7146 | if (sqd->thread) |
7147 | wake_up_process(sqd->thread); | |
09a6f4ef | 7148 | mutex_unlock(&sqd->lock); |
05962f95 | 7149 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7150 | } |
7151 | ||
534ca6d6 | 7152 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7153 | { |
534ca6d6 | 7154 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7155 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7156 | ||
37d1e2e3 JA |
7157 | io_sq_thread_stop(sqd); |
7158 | kfree(sqd); | |
7159 | } | |
7160 | } | |
7161 | ||
7162 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7163 | { | |
7164 | struct io_sq_data *sqd = ctx->sq_data; | |
7165 | ||
7166 | if (sqd) { | |
05962f95 | 7167 | io_sq_thread_park(sqd); |
521d6a73 | 7168 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7169 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7170 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7171 | |
7172 | io_put_sq_data(sqd); | |
7173 | ctx->sq_data = NULL; | |
7c30f36a SM |
7174 | if (ctx->sq_creds) |
7175 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7176 | } |
7177 | } | |
7178 | ||
aa06165d JA |
7179 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7180 | { | |
7181 | struct io_ring_ctx *ctx_attach; | |
7182 | struct io_sq_data *sqd; | |
7183 | struct fd f; | |
7184 | ||
7185 | f = fdget(p->wq_fd); | |
7186 | if (!f.file) | |
7187 | return ERR_PTR(-ENXIO); | |
7188 | if (f.file->f_op != &io_uring_fops) { | |
7189 | fdput(f); | |
7190 | return ERR_PTR(-EINVAL); | |
7191 | } | |
7192 | ||
7193 | ctx_attach = f.file->private_data; | |
7194 | sqd = ctx_attach->sq_data; | |
7195 | if (!sqd) { | |
7196 | fdput(f); | |
7197 | return ERR_PTR(-EINVAL); | |
7198 | } | |
5c2469e0 JA |
7199 | if (sqd->task_tgid != current->tgid) { |
7200 | fdput(f); | |
7201 | return ERR_PTR(-EPERM); | |
7202 | } | |
aa06165d JA |
7203 | |
7204 | refcount_inc(&sqd->refs); | |
7205 | fdput(f); | |
7206 | return sqd; | |
7207 | } | |
7208 | ||
26984fbf PB |
7209 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7210 | bool *attached) | |
534ca6d6 JA |
7211 | { |
7212 | struct io_sq_data *sqd; | |
7213 | ||
26984fbf | 7214 | *attached = false; |
5c2469e0 JA |
7215 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7216 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7217 | if (!IS_ERR(sqd)) { |
7218 | *attached = true; | |
5c2469e0 | 7219 | return sqd; |
26984fbf | 7220 | } |
5c2469e0 JA |
7221 | /* fall through for EPERM case, setup new sqd/task */ |
7222 | if (PTR_ERR(sqd) != -EPERM) | |
7223 | return sqd; | |
7224 | } | |
aa06165d | 7225 | |
534ca6d6 JA |
7226 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7227 | if (!sqd) | |
7228 | return ERR_PTR(-ENOMEM); | |
7229 | ||
9e138a48 | 7230 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7231 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7232 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7233 | mutex_init(&sqd->lock); |
534ca6d6 | 7234 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7235 | init_completion(&sqd->exited); |
534ca6d6 JA |
7236 | return sqd; |
7237 | } | |
7238 | ||
6b06314c | 7239 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7240 | /* |
7241 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7242 | * the io_uring can be safely unregistered on process exit, even if we have | |
7243 | * loops in the file referencing. | |
7244 | */ | |
7245 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7246 | { | |
7247 | struct sock *sk = ctx->ring_sock->sk; | |
7248 | struct scm_fp_list *fpl; | |
7249 | struct sk_buff *skb; | |
08a45173 | 7250 | int i, nr_files; |
6b06314c | 7251 | |
6b06314c JA |
7252 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7253 | if (!fpl) | |
7254 | return -ENOMEM; | |
7255 | ||
7256 | skb = alloc_skb(0, GFP_KERNEL); | |
7257 | if (!skb) { | |
7258 | kfree(fpl); | |
7259 | return -ENOMEM; | |
7260 | } | |
7261 | ||
7262 | skb->sk = sk; | |
6b06314c | 7263 | |
08a45173 | 7264 | nr_files = 0; |
62e398be | 7265 | fpl->user = get_uid(current_user()); |
6b06314c | 7266 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7267 | struct file *file = io_file_from_index(ctx, i + offset); |
7268 | ||
7269 | if (!file) | |
08a45173 | 7270 | continue; |
65e19f54 | 7271 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7272 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7273 | nr_files++; | |
6b06314c JA |
7274 | } |
7275 | ||
08a45173 JA |
7276 | if (nr_files) { |
7277 | fpl->max = SCM_MAX_FD; | |
7278 | fpl->count = nr_files; | |
7279 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7280 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7281 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7282 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7283 | |
08a45173 JA |
7284 | for (i = 0; i < nr_files; i++) |
7285 | fput(fpl->fp[i]); | |
7286 | } else { | |
7287 | kfree_skb(skb); | |
7288 | kfree(fpl); | |
7289 | } | |
6b06314c JA |
7290 | |
7291 | return 0; | |
7292 | } | |
7293 | ||
7294 | /* | |
7295 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7296 | * causes regular reference counting to break down. We rely on the UNIX | |
7297 | * garbage collection to take care of this problem for us. | |
7298 | */ | |
7299 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7300 | { | |
7301 | unsigned left, total; | |
7302 | int ret = 0; | |
7303 | ||
7304 | total = 0; | |
7305 | left = ctx->nr_user_files; | |
7306 | while (left) { | |
7307 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7308 | |
7309 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7310 | if (ret) | |
7311 | break; | |
7312 | left -= this_files; | |
7313 | total += this_files; | |
7314 | } | |
7315 | ||
7316 | if (!ret) | |
7317 | return 0; | |
7318 | ||
7319 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7320 | struct file *file = io_file_from_index(ctx, total); |
7321 | ||
7322 | if (file) | |
7323 | fput(file); | |
6b06314c JA |
7324 | total++; |
7325 | } | |
7326 | ||
7327 | return ret; | |
7328 | } | |
7329 | #else | |
7330 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7331 | { | |
7332 | return 0; | |
7333 | } | |
7334 | #endif | |
7335 | ||
269bbe5f | 7336 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7337 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7338 | { |
7339 | int i; | |
7340 | ||
7341 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7342 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7343 | unsigned this_files; |
7344 | ||
7345 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7346 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7347 | GFP_KERNEL); | |
7348 | if (!table->files) | |
7349 | break; | |
7350 | nr_files -= this_files; | |
7351 | } | |
7352 | ||
7353 | if (i == nr_tables) | |
7354 | return 0; | |
7355 | ||
7356 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7357 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7358 | kfree(table->files); |
7359 | } | |
7360 | return 1; | |
7361 | } | |
7362 | ||
50238531 | 7363 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7364 | { |
50238531 | 7365 | struct file *file = prsrc->file; |
05f3fb3c JA |
7366 | #if defined(CONFIG_UNIX) |
7367 | struct sock *sock = ctx->ring_sock->sk; | |
7368 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7369 | struct sk_buff *skb; | |
7370 | int i; | |
7371 | ||
7372 | __skb_queue_head_init(&list); | |
7373 | ||
7374 | /* | |
7375 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7376 | * remove this entry and rearrange the file array. | |
7377 | */ | |
7378 | skb = skb_dequeue(head); | |
7379 | while (skb) { | |
7380 | struct scm_fp_list *fp; | |
7381 | ||
7382 | fp = UNIXCB(skb).fp; | |
7383 | for (i = 0; i < fp->count; i++) { | |
7384 | int left; | |
7385 | ||
7386 | if (fp->fp[i] != file) | |
7387 | continue; | |
7388 | ||
7389 | unix_notinflight(fp->user, fp->fp[i]); | |
7390 | left = fp->count - 1 - i; | |
7391 | if (left) { | |
7392 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7393 | left * sizeof(struct file *)); | |
7394 | } | |
7395 | fp->count--; | |
7396 | if (!fp->count) { | |
7397 | kfree_skb(skb); | |
7398 | skb = NULL; | |
7399 | } else { | |
7400 | __skb_queue_tail(&list, skb); | |
7401 | } | |
7402 | fput(file); | |
7403 | file = NULL; | |
7404 | break; | |
7405 | } | |
7406 | ||
7407 | if (!file) | |
7408 | break; | |
7409 | ||
7410 | __skb_queue_tail(&list, skb); | |
7411 | ||
7412 | skb = skb_dequeue(head); | |
7413 | } | |
7414 | ||
7415 | if (skb_peek(&list)) { | |
7416 | spin_lock_irq(&head->lock); | |
7417 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7418 | __skb_queue_tail(head, skb); | |
7419 | spin_unlock_irq(&head->lock); | |
7420 | } | |
7421 | #else | |
7422 | fput(file); | |
7423 | #endif | |
7424 | } | |
7425 | ||
269bbe5f | 7426 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7427 | { |
269bbe5f BM |
7428 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7429 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7430 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7431 | |
269bbe5f BM |
7432 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7433 | list_del(&prsrc->list); | |
50238531 | 7434 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7435 | kfree(prsrc); |
65e19f54 | 7436 | } |
05589553 | 7437 | |
05589553 XW |
7438 | percpu_ref_exit(&ref_node->refs); |
7439 | kfree(ref_node); | |
269bbe5f | 7440 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7441 | } |
65e19f54 | 7442 | |
269bbe5f | 7443 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7444 | { |
7445 | struct io_ring_ctx *ctx; | |
7446 | struct llist_node *node; | |
7447 | ||
269bbe5f BM |
7448 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7449 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7450 | |
7451 | while (node) { | |
269bbe5f | 7452 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7453 | struct llist_node *next = node->next; |
7454 | ||
269bbe5f BM |
7455 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7456 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7457 | node = next; |
7458 | } | |
7459 | } | |
7460 | ||
ea64ec02 PB |
7461 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7462 | unsigned i) | |
2faf852d | 7463 | { |
ea64ec02 PB |
7464 | struct fixed_rsrc_table *table; |
7465 | ||
7466 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7467 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7468 | } | |
7469 | ||
00835dce | 7470 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7471 | { |
269bbe5f BM |
7472 | struct fixed_rsrc_ref_node *ref_node; |
7473 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7474 | struct io_ring_ctx *ctx; |
e297822b | 7475 | bool first_add = false; |
4a38aed2 | 7476 | int delay = HZ; |
65e19f54 | 7477 | |
269bbe5f BM |
7478 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7479 | data = ref_node->rsrc_data; | |
e297822b PB |
7480 | ctx = data->ctx; |
7481 | ||
2a63b2d9 | 7482 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7483 | ref_node->done = true; |
7484 | ||
d67d2263 BM |
7485 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7486 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7487 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7488 | /* recycle ref nodes in order */ |
7489 | if (!ref_node->done) | |
7490 | break; | |
7491 | list_del(&ref_node->node); | |
269bbe5f | 7492 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7493 | } |
2a63b2d9 | 7494 | io_rsrc_ref_unlock(ctx); |
05589553 | 7495 | |
e297822b | 7496 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7497 | delay = 0; |
05589553 | 7498 | |
4a38aed2 | 7499 | if (!delay) |
269bbe5f | 7500 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7501 | else if (first_add) |
269bbe5f | 7502 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7503 | } |
65e19f54 | 7504 | |
6802535d | 7505 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7506 | struct io_ring_ctx *ctx) |
05f3fb3c | 7507 | { |
269bbe5f | 7508 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7509 | |
05589553 XW |
7510 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7511 | if (!ref_node) | |
3e2224c5 | 7512 | return NULL; |
05f3fb3c | 7513 | |
00835dce | 7514 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7515 | 0, GFP_KERNEL)) { |
7516 | kfree(ref_node); | |
3e2224c5 | 7517 | return NULL; |
05589553 XW |
7518 | } |
7519 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7520 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7521 | ref_node->done = false; |
05589553 | 7522 | return ref_node; |
05589553 XW |
7523 | } |
7524 | ||
bc9744cd PB |
7525 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7526 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7527 | { |
269bbe5f | 7528 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7529 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7530 | } |
7531 | ||
269bbe5f | 7532 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7533 | { |
7534 | percpu_ref_exit(&ref_node->refs); | |
7535 | kfree(ref_node); | |
65e19f54 JA |
7536 | } |
7537 | ||
ea64ec02 | 7538 | |
6b06314c JA |
7539 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7540 | unsigned nr_args) | |
7541 | { | |
7542 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7543 | unsigned nr_tables, i; |
05f3fb3c | 7544 | struct file *file; |
600cf3f8 | 7545 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7546 | struct fixed_rsrc_ref_node *ref_node; |
7547 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7548 | |
05f3fb3c | 7549 | if (ctx->file_data) |
6b06314c JA |
7550 | return -EBUSY; |
7551 | if (!nr_args) | |
7552 | return -EINVAL; | |
7553 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7554 | return -EMFILE; | |
7555 | ||
1ad555c6 | 7556 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7557 | if (!file_data) |
05f3fb3c | 7558 | return -ENOMEM; |
13770a71 | 7559 | ctx->file_data = file_data; |
05f3fb3c | 7560 | |
65e19f54 | 7561 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7562 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7563 | GFP_KERNEL); |
600cf3f8 PB |
7564 | if (!file_data->table) |
7565 | goto out_free; | |
05f3fb3c | 7566 | |
600cf3f8 | 7567 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7568 | goto out_free; |
65e19f54 | 7569 | |
08a45173 | 7570 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7571 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7572 | ret = -EFAULT; | |
7573 | goto out_fput; | |
7574 | } | |
08a45173 | 7575 | /* allow sparse sets */ |
600cf3f8 | 7576 | if (fd == -1) |
08a45173 | 7577 | continue; |
6b06314c | 7578 | |
05f3fb3c | 7579 | file = fget(fd); |
6b06314c | 7580 | ret = -EBADF; |
05f3fb3c | 7581 | if (!file) |
600cf3f8 | 7582 | goto out_fput; |
05f3fb3c | 7583 | |
6b06314c JA |
7584 | /* |
7585 | * Don't allow io_uring instances to be registered. If UNIX | |
7586 | * isn't enabled, then this causes a reference cycle and this | |
7587 | * instance can never get freed. If UNIX is enabled we'll | |
7588 | * handle it just fine, but there's still no point in allowing | |
7589 | * a ring fd as it doesn't support regular read/write anyway. | |
7590 | */ | |
05f3fb3c JA |
7591 | if (file->f_op == &io_uring_fops) { |
7592 | fput(file); | |
600cf3f8 | 7593 | goto out_fput; |
6b06314c | 7594 | } |
ea64ec02 | 7595 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7596 | } |
7597 | ||
6b06314c | 7598 | ret = io_sqe_files_scm(ctx); |
05589553 | 7599 | if (ret) { |
6b06314c | 7600 | io_sqe_files_unregister(ctx); |
05589553 XW |
7601 | return ret; |
7602 | } | |
6b06314c | 7603 | |
bc9744cd | 7604 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7605 | if (!ref_node) { |
05589553 | 7606 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7607 | return -ENOMEM; |
05589553 | 7608 | } |
bc9744cd | 7609 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7610 | |
d67d2263 | 7611 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7612 | return ret; |
600cf3f8 PB |
7613 | out_fput: |
7614 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7615 | file = io_file_from_index(ctx, i); | |
7616 | if (file) | |
7617 | fput(file); | |
7618 | } | |
7619 | for (i = 0; i < nr_tables; i++) | |
7620 | kfree(file_data->table[i].files); | |
7621 | ctx->nr_user_files = 0; | |
600cf3f8 | 7622 | out_free: |
1ad555c6 | 7623 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7624 | ctx->file_data = NULL; |
6b06314c JA |
7625 | return ret; |
7626 | } | |
7627 | ||
c3a31e60 JA |
7628 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7629 | int index) | |
7630 | { | |
7631 | #if defined(CONFIG_UNIX) | |
7632 | struct sock *sock = ctx->ring_sock->sk; | |
7633 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7634 | struct sk_buff *skb; | |
7635 | ||
7636 | /* | |
7637 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7638 | * file set. If there's no room, fall back to allocating a new skb | |
7639 | * and filling it in. | |
7640 | */ | |
7641 | spin_lock_irq(&head->lock); | |
7642 | skb = skb_peek(head); | |
7643 | if (skb) { | |
7644 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7645 | ||
7646 | if (fpl->count < SCM_MAX_FD) { | |
7647 | __skb_unlink(skb, head); | |
7648 | spin_unlock_irq(&head->lock); | |
7649 | fpl->fp[fpl->count] = get_file(file); | |
7650 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7651 | fpl->count++; | |
7652 | spin_lock_irq(&head->lock); | |
7653 | __skb_queue_head(head, skb); | |
7654 | } else { | |
7655 | skb = NULL; | |
7656 | } | |
7657 | } | |
7658 | spin_unlock_irq(&head->lock); | |
7659 | ||
7660 | if (skb) { | |
7661 | fput(file); | |
7662 | return 0; | |
7663 | } | |
7664 | ||
7665 | return __io_sqe_files_scm(ctx, 1, index); | |
7666 | #else | |
7667 | return 0; | |
7668 | #endif | |
7669 | } | |
7670 | ||
50238531 | 7671 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7672 | { |
269bbe5f BM |
7673 | struct io_rsrc_put *prsrc; |
7674 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7675 | |
269bbe5f BM |
7676 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7677 | if (!prsrc) | |
a5318d3c | 7678 | return -ENOMEM; |
05f3fb3c | 7679 | |
50238531 | 7680 | prsrc->rsrc = rsrc; |
269bbe5f | 7681 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7682 | |
a5318d3c | 7683 | return 0; |
05f3fb3c JA |
7684 | } |
7685 | ||
269bbe5f BM |
7686 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7687 | struct file *file) | |
7688 | { | |
50238531 | 7689 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7690 | } |
7691 | ||
05f3fb3c | 7692 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7693 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7694 | unsigned nr_args) |
7695 | { | |
269bbe5f BM |
7696 | struct fixed_rsrc_data *data = ctx->file_data; |
7697 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7698 | struct file *file, **file_slot; |
c3a31e60 JA |
7699 | __s32 __user *fds; |
7700 | int fd, i, err; | |
7701 | __u32 done; | |
05589553 | 7702 | bool needs_switch = false; |
c3a31e60 | 7703 | |
05f3fb3c | 7704 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7705 | return -EOVERFLOW; |
7706 | if (done > ctx->nr_user_files) | |
7707 | return -EINVAL; | |
7708 | ||
bc9744cd | 7709 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7710 | if (!ref_node) |
7711 | return -ENOMEM; | |
bc9744cd | 7712 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7713 | |
269bbe5f | 7714 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7715 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7716 | err = 0; |
7717 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7718 | err = -EFAULT; | |
7719 | break; | |
7720 | } | |
4e0377a1 | 7721 | if (fd == IORING_REGISTER_FILES_SKIP) |
7722 | continue; | |
7723 | ||
67973b93 | 7724 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7725 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7726 | ||
7727 | if (*file_slot) { | |
7728 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7729 | if (err) |
7730 | break; | |
ea64ec02 | 7731 | *file_slot = NULL; |
05589553 | 7732 | needs_switch = true; |
c3a31e60 JA |
7733 | } |
7734 | if (fd != -1) { | |
c3a31e60 JA |
7735 | file = fget(fd); |
7736 | if (!file) { | |
7737 | err = -EBADF; | |
7738 | break; | |
7739 | } | |
7740 | /* | |
7741 | * Don't allow io_uring instances to be registered. If | |
7742 | * UNIX isn't enabled, then this causes a reference | |
7743 | * cycle and this instance can never get freed. If UNIX | |
7744 | * is enabled we'll handle it just fine, but there's | |
7745 | * still no point in allowing a ring fd as it doesn't | |
7746 | * support regular read/write anyway. | |
7747 | */ | |
7748 | if (file->f_op == &io_uring_fops) { | |
7749 | fput(file); | |
7750 | err = -EBADF; | |
7751 | break; | |
7752 | } | |
e68a3ff8 | 7753 | *file_slot = file; |
c3a31e60 | 7754 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7755 | if (err) { |
e68a3ff8 | 7756 | *file_slot = NULL; |
f3bd9dae | 7757 | fput(file); |
c3a31e60 | 7758 | break; |
f3bd9dae | 7759 | } |
c3a31e60 | 7760 | } |
05f3fb3c JA |
7761 | } |
7762 | ||
05589553 | 7763 | if (needs_switch) { |
b2e96852 | 7764 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7765 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7766 | } else |
269bbe5f | 7767 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7768 | |
7769 | return done ? done : err; | |
7770 | } | |
05589553 | 7771 | |
05f3fb3c JA |
7772 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7773 | unsigned nr_args) | |
7774 | { | |
269bbe5f | 7775 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7776 | |
7777 | if (!ctx->file_data) | |
7778 | return -ENXIO; | |
7779 | if (!nr_args) | |
7780 | return -EINVAL; | |
7781 | if (copy_from_user(&up, arg, sizeof(up))) | |
7782 | return -EFAULT; | |
7783 | if (up.resv) | |
7784 | return -EINVAL; | |
7785 | ||
7786 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7787 | } | |
c3a31e60 | 7788 | |
5280f7e5 | 7789 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7790 | { |
7791 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7792 | ||
5280f7e5 PB |
7793 | req = io_put_req_find_next(req); |
7794 | return req ? &req->work : NULL; | |
7d723065 JA |
7795 | } |
7796 | ||
5aa75ed5 | 7797 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7798 | { |
e941894e | 7799 | struct io_wq_hash *hash; |
24369c2e | 7800 | struct io_wq_data data; |
24369c2e | 7801 | unsigned int concurrency; |
24369c2e | 7802 | |
e941894e JA |
7803 | hash = ctx->hash_map; |
7804 | if (!hash) { | |
7805 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7806 | if (!hash) | |
7807 | return ERR_PTR(-ENOMEM); | |
7808 | refcount_set(&hash->refs, 1); | |
7809 | init_waitqueue_head(&hash->wait); | |
7810 | ctx->hash_map = hash; | |
24369c2e PB |
7811 | } |
7812 | ||
e941894e | 7813 | data.hash = hash; |
e9fd9396 | 7814 | data.free_work = io_free_work; |
f5fa38c5 | 7815 | data.do_work = io_wq_submit_work; |
24369c2e | 7816 | |
d25e3a3d JA |
7817 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7818 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7819 | |
5aa75ed5 | 7820 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7821 | } |
7822 | ||
5aa75ed5 JA |
7823 | static int io_uring_alloc_task_context(struct task_struct *task, |
7824 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7825 | { |
7826 | struct io_uring_task *tctx; | |
d8a6df10 | 7827 | int ret; |
0f212204 JA |
7828 | |
7829 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7830 | if (unlikely(!tctx)) | |
7831 | return -ENOMEM; | |
7832 | ||
d8a6df10 JA |
7833 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7834 | if (unlikely(ret)) { | |
7835 | kfree(tctx); | |
7836 | return ret; | |
7837 | } | |
7838 | ||
5aa75ed5 JA |
7839 | tctx->io_wq = io_init_wq_offload(ctx); |
7840 | if (IS_ERR(tctx->io_wq)) { | |
7841 | ret = PTR_ERR(tctx->io_wq); | |
7842 | percpu_counter_destroy(&tctx->inflight); | |
7843 | kfree(tctx); | |
7844 | return ret; | |
7845 | } | |
7846 | ||
0f212204 JA |
7847 | xa_init(&tctx->xa); |
7848 | init_waitqueue_head(&tctx->wait); | |
7849 | tctx->last = NULL; | |
fdaf083c | 7850 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7851 | task->io_uring = tctx; |
7cbf1722 JA |
7852 | spin_lock_init(&tctx->task_lock); |
7853 | INIT_WQ_LIST(&tctx->task_list); | |
7854 | tctx->task_state = 0; | |
7855 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7856 | return 0; |
7857 | } | |
7858 | ||
7859 | void __io_uring_free(struct task_struct *tsk) | |
7860 | { | |
7861 | struct io_uring_task *tctx = tsk->io_uring; | |
7862 | ||
7863 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7864 | WARN_ON_ONCE(tctx->io_wq); |
7865 | ||
d8a6df10 | 7866 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7867 | kfree(tctx); |
7868 | tsk->io_uring = NULL; | |
7869 | } | |
7870 | ||
7e84e1c7 SG |
7871 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7872 | struct io_uring_params *p) | |
2b188cc1 JA |
7873 | { |
7874 | int ret; | |
7875 | ||
d25e3a3d JA |
7876 | /* Retain compatibility with failing for an invalid attach attempt */ |
7877 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7878 | IORING_SETUP_ATTACH_WQ) { | |
7879 | struct fd f; | |
7880 | ||
7881 | f = fdget(p->wq_fd); | |
7882 | if (!f.file) | |
7883 | return -ENXIO; | |
7884 | if (f.file->f_op != &io_uring_fops) { | |
7885 | fdput(f); | |
7886 | return -EINVAL; | |
7887 | } | |
7888 | fdput(f); | |
7889 | } | |
6c271ce2 | 7890 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7891 | struct task_struct *tsk; |
534ca6d6 | 7892 | struct io_sq_data *sqd; |
26984fbf | 7893 | bool attached; |
534ca6d6 | 7894 | |
3ec482d1 | 7895 | ret = -EPERM; |
ce59fc69 | 7896 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7897 | goto err; |
7898 | ||
26984fbf | 7899 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7900 | if (IS_ERR(sqd)) { |
7901 | ret = PTR_ERR(sqd); | |
7902 | goto err; | |
7903 | } | |
69fb2131 | 7904 | |
7c30f36a | 7905 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7906 | ctx->sq_data = sqd; |
917257da JA |
7907 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7908 | if (!ctx->sq_thread_idle) | |
7909 | ctx->sq_thread_idle = HZ; | |
7910 | ||
26984fbf | 7911 | ret = 0; |
78d7f6ba | 7912 | io_sq_thread_park(sqd); |
26984fbf PB |
7913 | /* don't attach to a dying SQPOLL thread, would be racy */ |
7914 | if (attached && !sqd->thread) { | |
7915 | ret = -ENXIO; | |
7916 | } else { | |
7917 | list_add(&ctx->sqd_list, &sqd->ctx_list); | |
7918 | io_sqd_update_thread_idle(sqd); | |
7919 | } | |
78d7f6ba PB |
7920 | io_sq_thread_unpark(sqd); |
7921 | ||
26984fbf PB |
7922 | if (ret < 0) { |
7923 | io_put_sq_data(sqd); | |
7924 | ctx->sq_data = NULL; | |
7925 | return ret; | |
7926 | } else if (attached) { | |
5aa75ed5 | 7927 | return 0; |
26984fbf | 7928 | } |
aa06165d | 7929 | |
6c271ce2 | 7930 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7931 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7932 | |
917257da | 7933 | ret = -EINVAL; |
44a9bd18 | 7934 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7935 | goto err_sqpoll; |
7889f44d | 7936 | if (!cpu_online(cpu)) |
e8f98f24 | 7937 | goto err_sqpoll; |
917257da | 7938 | |
37d1e2e3 | 7939 | sqd->sq_cpu = cpu; |
6c271ce2 | 7940 | } else { |
37d1e2e3 | 7941 | sqd->sq_cpu = -1; |
6c271ce2 | 7942 | } |
37d1e2e3 JA |
7943 | |
7944 | sqd->task_pid = current->pid; | |
5c2469e0 | 7945 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7946 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7947 | if (IS_ERR(tsk)) { | |
7948 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7949 | goto err_sqpoll; |
6c271ce2 | 7950 | } |
97a73a0f | 7951 | |
46fe18b1 | 7952 | sqd->thread = tsk; |
97a73a0f | 7953 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7954 | wake_up_new_task(tsk); |
0f212204 JA |
7955 | if (ret) |
7956 | goto err; | |
6c271ce2 JA |
7957 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7958 | /* Can't have SQ_AFF without SQPOLL */ | |
7959 | ret = -EINVAL; | |
7960 | goto err; | |
7961 | } | |
7962 | ||
2b188cc1 JA |
7963 | return 0; |
7964 | err: | |
37d1e2e3 | 7965 | io_sq_thread_finish(ctx); |
2b188cc1 | 7966 | return ret; |
e8f98f24 JA |
7967 | err_sqpoll: |
7968 | complete(&ctx->sq_data->exited); | |
7969 | goto err; | |
2b188cc1 JA |
7970 | } |
7971 | ||
a087e2b5 BM |
7972 | static inline void __io_unaccount_mem(struct user_struct *user, |
7973 | unsigned long nr_pages) | |
2b188cc1 JA |
7974 | { |
7975 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7976 | } | |
7977 | ||
a087e2b5 BM |
7978 | static inline int __io_account_mem(struct user_struct *user, |
7979 | unsigned long nr_pages) | |
2b188cc1 JA |
7980 | { |
7981 | unsigned long page_limit, cur_pages, new_pages; | |
7982 | ||
7983 | /* Don't allow more pages than we can safely lock */ | |
7984 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7985 | ||
7986 | do { | |
7987 | cur_pages = atomic_long_read(&user->locked_vm); | |
7988 | new_pages = cur_pages + nr_pages; | |
7989 | if (new_pages > page_limit) | |
7990 | return -ENOMEM; | |
7991 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7992 | new_pages) != cur_pages); | |
7993 | ||
7994 | return 0; | |
7995 | } | |
7996 | ||
26bfa89e | 7997 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7998 | { |
62e398be | 7999 | if (ctx->user) |
a087e2b5 | 8000 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8001 | |
26bfa89e JA |
8002 | if (ctx->mm_account) |
8003 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8004 | } |
8005 | ||
26bfa89e | 8006 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8007 | { |
30975825 BM |
8008 | int ret; |
8009 | ||
62e398be | 8010 | if (ctx->user) { |
30975825 BM |
8011 | ret = __io_account_mem(ctx->user, nr_pages); |
8012 | if (ret) | |
8013 | return ret; | |
8014 | } | |
8015 | ||
26bfa89e JA |
8016 | if (ctx->mm_account) |
8017 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8018 | |
8019 | return 0; | |
8020 | } | |
8021 | ||
2b188cc1 JA |
8022 | static void io_mem_free(void *ptr) |
8023 | { | |
52e04ef4 MR |
8024 | struct page *page; |
8025 | ||
8026 | if (!ptr) | |
8027 | return; | |
2b188cc1 | 8028 | |
52e04ef4 | 8029 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8030 | if (put_page_testzero(page)) |
8031 | free_compound_page(page); | |
8032 | } | |
8033 | ||
8034 | static void *io_mem_alloc(size_t size) | |
8035 | { | |
8036 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8037 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8038 | |
8039 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8040 | } | |
8041 | ||
75b28aff HV |
8042 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8043 | size_t *sq_offset) | |
8044 | { | |
8045 | struct io_rings *rings; | |
8046 | size_t off, sq_array_size; | |
8047 | ||
8048 | off = struct_size(rings, cqes, cq_entries); | |
8049 | if (off == SIZE_MAX) | |
8050 | return SIZE_MAX; | |
8051 | ||
8052 | #ifdef CONFIG_SMP | |
8053 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8054 | if (off == 0) | |
8055 | return SIZE_MAX; | |
8056 | #endif | |
8057 | ||
b36200f5 DV |
8058 | if (sq_offset) |
8059 | *sq_offset = off; | |
8060 | ||
75b28aff HV |
8061 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8062 | if (sq_array_size == SIZE_MAX) | |
8063 | return SIZE_MAX; | |
8064 | ||
8065 | if (check_add_overflow(off, sq_array_size, &off)) | |
8066 | return SIZE_MAX; | |
8067 | ||
75b28aff HV |
8068 | return off; |
8069 | } | |
8070 | ||
0a96bbe4 | 8071 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8072 | { |
8073 | int i, j; | |
8074 | ||
8075 | if (!ctx->user_bufs) | |
8076 | return -ENXIO; | |
8077 | ||
8078 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8079 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8080 | ||
8081 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8082 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8083 | |
de293938 | 8084 | if (imu->acct_pages) |
26bfa89e | 8085 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8086 | kvfree(imu->bvec); |
edafccee JA |
8087 | imu->nr_bvecs = 0; |
8088 | } | |
8089 | ||
8090 | kfree(ctx->user_bufs); | |
8091 | ctx->user_bufs = NULL; | |
8092 | ctx->nr_user_bufs = 0; | |
8093 | return 0; | |
8094 | } | |
8095 | ||
8096 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8097 | void __user *arg, unsigned index) | |
8098 | { | |
8099 | struct iovec __user *src; | |
8100 | ||
8101 | #ifdef CONFIG_COMPAT | |
8102 | if (ctx->compat) { | |
8103 | struct compat_iovec __user *ciovs; | |
8104 | struct compat_iovec ciov; | |
8105 | ||
8106 | ciovs = (struct compat_iovec __user *) arg; | |
8107 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8108 | return -EFAULT; | |
8109 | ||
d55e5f5b | 8110 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8111 | dst->iov_len = ciov.iov_len; |
8112 | return 0; | |
8113 | } | |
8114 | #endif | |
8115 | src = (struct iovec __user *) arg; | |
8116 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8117 | return -EFAULT; | |
8118 | return 0; | |
8119 | } | |
8120 | ||
de293938 JA |
8121 | /* |
8122 | * Not super efficient, but this is just a registration time. And we do cache | |
8123 | * the last compound head, so generally we'll only do a full search if we don't | |
8124 | * match that one. | |
8125 | * | |
8126 | * We check if the given compound head page has already been accounted, to | |
8127 | * avoid double accounting it. This allows us to account the full size of the | |
8128 | * page, not just the constituent pages of a huge page. | |
8129 | */ | |
8130 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8131 | int nr_pages, struct page *hpage) | |
8132 | { | |
8133 | int i, j; | |
8134 | ||
8135 | /* check current page array */ | |
8136 | for (i = 0; i < nr_pages; i++) { | |
8137 | if (!PageCompound(pages[i])) | |
8138 | continue; | |
8139 | if (compound_head(pages[i]) == hpage) | |
8140 | return true; | |
8141 | } | |
8142 | ||
8143 | /* check previously registered pages */ | |
8144 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8145 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8146 | ||
8147 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8148 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8149 | continue; | |
8150 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8151 | return true; | |
8152 | } | |
8153 | } | |
8154 | ||
8155 | return false; | |
8156 | } | |
8157 | ||
8158 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8159 | int nr_pages, struct io_mapped_ubuf *imu, | |
8160 | struct page **last_hpage) | |
8161 | { | |
8162 | int i, ret; | |
8163 | ||
8164 | for (i = 0; i < nr_pages; i++) { | |
8165 | if (!PageCompound(pages[i])) { | |
8166 | imu->acct_pages++; | |
8167 | } else { | |
8168 | struct page *hpage; | |
8169 | ||
8170 | hpage = compound_head(pages[i]); | |
8171 | if (hpage == *last_hpage) | |
8172 | continue; | |
8173 | *last_hpage = hpage; | |
8174 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8175 | continue; | |
8176 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8177 | } | |
8178 | } | |
8179 | ||
8180 | if (!imu->acct_pages) | |
8181 | return 0; | |
8182 | ||
26bfa89e | 8183 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8184 | if (ret) |
8185 | imu->acct_pages = 0; | |
8186 | return ret; | |
8187 | } | |
8188 | ||
0a96bbe4 BM |
8189 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8190 | struct io_mapped_ubuf *imu, | |
8191 | struct page **last_hpage) | |
edafccee JA |
8192 | { |
8193 | struct vm_area_struct **vmas = NULL; | |
8194 | struct page **pages = NULL; | |
0a96bbe4 BM |
8195 | unsigned long off, start, end, ubuf; |
8196 | size_t size; | |
8197 | int ret, pret, nr_pages, i; | |
8198 | ||
8199 | ubuf = (unsigned long) iov->iov_base; | |
8200 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8201 | start = ubuf >> PAGE_SHIFT; | |
8202 | nr_pages = end - start; | |
8203 | ||
8204 | ret = -ENOMEM; | |
8205 | ||
8206 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8207 | if (!pages) | |
8208 | goto done; | |
8209 | ||
8210 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8211 | GFP_KERNEL); | |
8212 | if (!vmas) | |
8213 | goto done; | |
edafccee | 8214 | |
0a96bbe4 BM |
8215 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8216 | GFP_KERNEL); | |
8217 | if (!imu->bvec) | |
8218 | goto done; | |
8219 | ||
8220 | ret = 0; | |
8221 | mmap_read_lock(current->mm); | |
8222 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8223 | pages, vmas); | |
8224 | if (pret == nr_pages) { | |
8225 | /* don't support file backed memory */ | |
8226 | for (i = 0; i < nr_pages; i++) { | |
8227 | struct vm_area_struct *vma = vmas[i]; | |
8228 | ||
8229 | if (vma->vm_file && | |
8230 | !is_file_hugepages(vma->vm_file)) { | |
8231 | ret = -EOPNOTSUPP; | |
8232 | break; | |
8233 | } | |
8234 | } | |
8235 | } else { | |
8236 | ret = pret < 0 ? pret : -EFAULT; | |
8237 | } | |
8238 | mmap_read_unlock(current->mm); | |
8239 | if (ret) { | |
8240 | /* | |
8241 | * if we did partial map, or found file backed vmas, | |
8242 | * release any pages we did get | |
8243 | */ | |
8244 | if (pret > 0) | |
8245 | unpin_user_pages(pages, pret); | |
8246 | kvfree(imu->bvec); | |
8247 | goto done; | |
8248 | } | |
8249 | ||
8250 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8251 | if (ret) { | |
8252 | unpin_user_pages(pages, pret); | |
8253 | kvfree(imu->bvec); | |
8254 | goto done; | |
8255 | } | |
8256 | ||
8257 | off = ubuf & ~PAGE_MASK; | |
8258 | size = iov->iov_len; | |
8259 | for (i = 0; i < nr_pages; i++) { | |
8260 | size_t vec_len; | |
8261 | ||
8262 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8263 | imu->bvec[i].bv_page = pages[i]; | |
8264 | imu->bvec[i].bv_len = vec_len; | |
8265 | imu->bvec[i].bv_offset = off; | |
8266 | off = 0; | |
8267 | size -= vec_len; | |
8268 | } | |
8269 | /* store original address for later verification */ | |
8270 | imu->ubuf = ubuf; | |
8271 | imu->len = iov->iov_len; | |
8272 | imu->nr_bvecs = nr_pages; | |
8273 | ret = 0; | |
8274 | done: | |
8275 | kvfree(pages); | |
8276 | kvfree(vmas); | |
8277 | return ret; | |
8278 | } | |
8279 | ||
2b358604 | 8280 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8281 | { |
edafccee JA |
8282 | if (ctx->user_bufs) |
8283 | return -EBUSY; | |
8284 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8285 | return -EINVAL; | |
8286 | ||
8287 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8288 | GFP_KERNEL); | |
8289 | if (!ctx->user_bufs) | |
8290 | return -ENOMEM; | |
8291 | ||
2b358604 BM |
8292 | return 0; |
8293 | } | |
edafccee | 8294 | |
2b358604 BM |
8295 | static int io_buffer_validate(struct iovec *iov) |
8296 | { | |
8297 | /* | |
8298 | * Don't impose further limits on the size and buffer | |
8299 | * constraints here, we'll -EINVAL later when IO is | |
8300 | * submitted if they are wrong. | |
8301 | */ | |
8302 | if (!iov->iov_base || !iov->iov_len) | |
8303 | return -EFAULT; | |
edafccee | 8304 | |
2b358604 BM |
8305 | /* arbitrary limit, but we need something */ |
8306 | if (iov->iov_len > SZ_1G) | |
8307 | return -EFAULT; | |
edafccee | 8308 | |
2b358604 BM |
8309 | return 0; |
8310 | } | |
edafccee | 8311 | |
2b358604 BM |
8312 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8313 | unsigned int nr_args) | |
8314 | { | |
8315 | int i, ret; | |
8316 | struct iovec iov; | |
8317 | struct page *last_hpage = NULL; | |
edafccee | 8318 | |
2b358604 BM |
8319 | ret = io_buffers_map_alloc(ctx, nr_args); |
8320 | if (ret) | |
8321 | return ret; | |
edafccee | 8322 | |
edafccee JA |
8323 | for (i = 0; i < nr_args; i++) { |
8324 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8325 | |
edafccee JA |
8326 | ret = io_copy_iov(ctx, &iov, arg, i); |
8327 | if (ret) | |
0a96bbe4 | 8328 | break; |
de293938 | 8329 | |
2b358604 BM |
8330 | ret = io_buffer_validate(&iov); |
8331 | if (ret) | |
0a96bbe4 | 8332 | break; |
edafccee | 8333 | |
0a96bbe4 BM |
8334 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8335 | if (ret) | |
8336 | break; | |
edafccee JA |
8337 | |
8338 | ctx->nr_user_bufs++; | |
8339 | } | |
0a96bbe4 BM |
8340 | |
8341 | if (ret) | |
8342 | io_sqe_buffers_unregister(ctx); | |
8343 | ||
edafccee JA |
8344 | return ret; |
8345 | } | |
8346 | ||
9b402849 JA |
8347 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8348 | { | |
8349 | __s32 __user *fds = arg; | |
8350 | int fd; | |
8351 | ||
8352 | if (ctx->cq_ev_fd) | |
8353 | return -EBUSY; | |
8354 | ||
8355 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8356 | return -EFAULT; | |
8357 | ||
8358 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8359 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8360 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8361 | ctx->cq_ev_fd = NULL; | |
8362 | return ret; | |
8363 | } | |
8364 | ||
8365 | return 0; | |
8366 | } | |
8367 | ||
8368 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8369 | { | |
8370 | if (ctx->cq_ev_fd) { | |
8371 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8372 | ctx->cq_ev_fd = NULL; | |
8373 | return 0; | |
8374 | } | |
8375 | ||
8376 | return -ENXIO; | |
8377 | } | |
8378 | ||
5a2e745d JA |
8379 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8380 | { | |
9e15c3a0 JA |
8381 | struct io_buffer *buf; |
8382 | unsigned long index; | |
8383 | ||
8384 | xa_for_each(&ctx->io_buffers, index, buf) | |
8385 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8386 | } |
8387 | ||
68e68ee6 | 8388 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8389 | { |
68e68ee6 | 8390 | struct io_kiocb *req, *nxt; |
1b4c351f | 8391 | |
68e68ee6 JA |
8392 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8393 | if (tsk && req->task != tsk) | |
8394 | continue; | |
1b4c351f JA |
8395 | list_del(&req->compl.list); |
8396 | kmem_cache_free(req_cachep, req); | |
8397 | } | |
8398 | } | |
8399 | ||
4010fec4 | 8400 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8401 | { |
bf019da7 | 8402 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8403 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8404 | |
9a4fdbd8 JA |
8405 | mutex_lock(&ctx->uring_lock); |
8406 | ||
8e5c66c4 | 8407 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8408 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8409 | submit_state->reqs); | |
8e5c66c4 PB |
8410 | submit_state->free_reqs = 0; |
8411 | } | |
9a4fdbd8 JA |
8412 | |
8413 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8414 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8415 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8416 | spin_unlock_irq(&ctx->completion_lock); |
8417 | ||
e5547d2c PB |
8418 | io_req_cache_free(&cs->free_list, NULL); |
8419 | ||
9a4fdbd8 JA |
8420 | mutex_unlock(&ctx->uring_lock); |
8421 | } | |
8422 | ||
2b188cc1 JA |
8423 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8424 | { | |
04fc6c80 PB |
8425 | /* |
8426 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8427 | * and they are free to do so while still holding uring_lock or |
8428 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8429 | */ |
8430 | mutex_lock(&ctx->uring_lock); | |
8431 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8432 | spin_lock_irq(&ctx->completion_lock); |
8433 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8434 | |
37d1e2e3 | 8435 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8436 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8437 | |
37d1e2e3 | 8438 | if (ctx->mm_account) { |
2aede0e4 JA |
8439 | mmdrop(ctx->mm_account); |
8440 | ctx->mm_account = NULL; | |
30975825 | 8441 | } |
def596e9 | 8442 | |
8bad28d8 | 8443 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8444 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8445 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8446 | io_eventfd_unregister(ctx); |
5a2e745d | 8447 | io_destroy_buffers(ctx); |
def596e9 | 8448 | |
2b188cc1 | 8449 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8450 | if (ctx->ring_sock) { |
8451 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8452 | sock_release(ctx->ring_sock); |
355e8d26 | 8453 | } |
2b188cc1 JA |
8454 | #endif |
8455 | ||
75b28aff | 8456 | io_mem_free(ctx->rings); |
2b188cc1 | 8457 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8458 | |
8459 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8460 | free_uid(ctx->user); |
4010fec4 | 8461 | io_req_caches_free(ctx); |
e941894e JA |
8462 | if (ctx->hash_map) |
8463 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8464 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8465 | kfree(ctx); |
8466 | } | |
8467 | ||
8468 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8469 | { | |
8470 | struct io_ring_ctx *ctx = file->private_data; | |
8471 | __poll_t mask = 0; | |
8472 | ||
8473 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8474 | /* |
8475 | * synchronizes with barrier from wq_has_sleeper call in | |
8476 | * io_commit_cqring | |
8477 | */ | |
2b188cc1 | 8478 | smp_rmb(); |
90554200 | 8479 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8480 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8481 | |
8482 | /* | |
8483 | * Don't flush cqring overflow list here, just do a simple check. | |
8484 | * Otherwise there could possible be ABBA deadlock: | |
8485 | * CPU0 CPU1 | |
8486 | * ---- ---- | |
8487 | * lock(&ctx->uring_lock); | |
8488 | * lock(&ep->mtx); | |
8489 | * lock(&ctx->uring_lock); | |
8490 | * lock(&ep->mtx); | |
8491 | * | |
8492 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8493 | * pushs them to do the flush. | |
8494 | */ | |
8495 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8496 | mask |= EPOLLIN | EPOLLRDNORM; |
8497 | ||
8498 | return mask; | |
8499 | } | |
8500 | ||
8501 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8502 | { | |
8503 | struct io_ring_ctx *ctx = file->private_data; | |
8504 | ||
8505 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8506 | } | |
8507 | ||
0bead8cd | 8508 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8509 | { |
4379bf8b | 8510 | const struct cred *creds; |
071698e1 | 8511 | |
61cf9370 | 8512 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8513 | if (creds) { |
8514 | put_cred(creds); | |
0bead8cd | 8515 | return 0; |
1e6fa521 | 8516 | } |
0bead8cd YD |
8517 | |
8518 | return -EINVAL; | |
8519 | } | |
8520 | ||
9b465711 | 8521 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8522 | { |
9b465711 | 8523 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8524 | } |
8525 | ||
d56d938b PB |
8526 | struct io_tctx_exit { |
8527 | struct callback_head task_work; | |
8528 | struct completion completion; | |
baf186c4 | 8529 | struct io_ring_ctx *ctx; |
d56d938b PB |
8530 | }; |
8531 | ||
8532 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8533 | { | |
8534 | struct io_uring_task *tctx = current->io_uring; | |
8535 | struct io_tctx_exit *work; | |
8536 | ||
8537 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8538 | /* | |
8539 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8540 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8541 | */ | |
8542 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8543 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8544 | complete(&work->completion); |
8545 | } | |
8546 | ||
85faa7b8 JA |
8547 | static void io_ring_exit_work(struct work_struct *work) |
8548 | { | |
d56d938b | 8549 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8550 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8551 | struct io_tctx_exit exit; |
8552 | struct io_tctx_node *node; | |
8553 | int ret; | |
85faa7b8 | 8554 | |
56952e91 JA |
8555 | /* |
8556 | * If we're doing polled IO and end up having requests being | |
8557 | * submitted async (out-of-line), then completions can come in while | |
8558 | * we're waiting for refs to drop. We need to reap these manually, | |
8559 | * as nobody else will be looking for them. | |
8560 | */ | |
b2edc0a7 | 8561 | do { |
9936c7c2 | 8562 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8563 | |
8564 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8565 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8566 | |
8567 | mutex_lock(&ctx->uring_lock); | |
8568 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8569 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8570 | ||
d56d938b PB |
8571 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8572 | ctx_node); | |
baf186c4 | 8573 | exit.ctx = ctx; |
d56d938b PB |
8574 | init_completion(&exit.completion); |
8575 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8576 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8577 | if (WARN_ON_ONCE(ret)) | |
8578 | continue; | |
8579 | wake_up_process(node->task); | |
8580 | ||
8581 | mutex_unlock(&ctx->uring_lock); | |
8582 | wait_for_completion(&exit.completion); | |
8583 | cond_resched(); | |
8584 | mutex_lock(&ctx->uring_lock); | |
8585 | } | |
8586 | mutex_unlock(&ctx->uring_lock); | |
8587 | ||
85faa7b8 JA |
8588 | io_ring_ctx_free(ctx); |
8589 | } | |
8590 | ||
2b188cc1 JA |
8591 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8592 | { | |
61cf9370 MWO |
8593 | unsigned long index; |
8594 | struct creds *creds; | |
8595 | ||
2b188cc1 JA |
8596 | mutex_lock(&ctx->uring_lock); |
8597 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8598 | /* if force is set, the ring is going away. always drop after that */ |
8599 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8600 | if (ctx->rings) |
6c503150 | 8601 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8602 | xa_for_each(&ctx->personalities, index, creds) |
8603 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8604 | mutex_unlock(&ctx->uring_lock); |
8605 | ||
f6d54255 PB |
8606 | /* prevent SQPOLL from submitting new requests */ |
8607 | if (ctx->sq_data) { | |
8608 | io_sq_thread_park(ctx->sq_data); | |
8609 | list_del_init(&ctx->sqd_list); | |
8610 | io_sqd_update_thread_idle(ctx->sq_data); | |
8611 | io_sq_thread_unpark(ctx->sq_data); | |
8612 | } | |
8613 | ||
6b81928d PB |
8614 | io_kill_timeouts(ctx, NULL, NULL); |
8615 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8616 | |
15dff286 | 8617 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8618 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8619 | |
85faa7b8 | 8620 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8621 | /* |
8622 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8623 | * if we're exiting a ton of rings at the same time. It just adds | |
8624 | * noise and overhead, there's no discernable change in runtime | |
8625 | * over using system_wq. | |
8626 | */ | |
8627 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8628 | } |
8629 | ||
8630 | static int io_uring_release(struct inode *inode, struct file *file) | |
8631 | { | |
8632 | struct io_ring_ctx *ctx = file->private_data; | |
8633 | ||
8634 | file->private_data = NULL; | |
8635 | io_ring_ctx_wait_and_kill(ctx); | |
8636 | return 0; | |
8637 | } | |
8638 | ||
f6edbabb PB |
8639 | struct io_task_cancel { |
8640 | struct task_struct *task; | |
8641 | struct files_struct *files; | |
8642 | }; | |
f254ac04 | 8643 | |
f6edbabb | 8644 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8645 | { |
9a472ef7 | 8646 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8647 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8648 | bool ret; |
8649 | ||
f6edbabb | 8650 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8651 | unsigned long flags; |
8652 | struct io_ring_ctx *ctx = req->ctx; | |
8653 | ||
8654 | /* protect against races with linked timeouts */ | |
8655 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8656 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8657 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8658 | } else { | |
f6edbabb | 8659 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8660 | } |
8661 | return ret; | |
b711d4ea JA |
8662 | } |
8663 | ||
e1915f76 | 8664 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8665 | struct task_struct *task, |
b7ddce3c PB |
8666 | struct files_struct *files) |
8667 | { | |
e1915f76 | 8668 | struct io_defer_entry *de; |
b7ddce3c PB |
8669 | LIST_HEAD(list); |
8670 | ||
8671 | spin_lock_irq(&ctx->completion_lock); | |
8672 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8673 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8674 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8675 | break; | |
8676 | } | |
8677 | } | |
8678 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8679 | if (list_empty(&list)) |
8680 | return false; | |
b7ddce3c PB |
8681 | |
8682 | while (!list_empty(&list)) { | |
8683 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8684 | list_del_init(&de->list); | |
8685 | req_set_fail_links(de->req); | |
8686 | io_put_req(de->req); | |
8687 | io_req_complete(de->req, -ECANCELED); | |
8688 | kfree(de); | |
8689 | } | |
e1915f76 | 8690 | return true; |
b7ddce3c PB |
8691 | } |
8692 | ||
1b00764f PB |
8693 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8694 | { | |
8695 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8696 | ||
8697 | return req->ctx == data; | |
8698 | } | |
8699 | ||
8700 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8701 | { | |
8702 | struct io_tctx_node *node; | |
8703 | enum io_wq_cancel cret; | |
8704 | bool ret = false; | |
8705 | ||
8706 | mutex_lock(&ctx->uring_lock); | |
8707 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8708 | struct io_uring_task *tctx = node->task->io_uring; | |
8709 | ||
8710 | /* | |
8711 | * io_wq will stay alive while we hold uring_lock, because it's | |
8712 | * killed after ctx nodes, which requires to take the lock. | |
8713 | */ | |
8714 | if (!tctx || !tctx->io_wq) | |
8715 | continue; | |
8716 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8717 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8718 | } | |
8719 | mutex_unlock(&ctx->uring_lock); | |
8720 | ||
8721 | return ret; | |
8722 | } | |
8723 | ||
9936c7c2 PB |
8724 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8725 | struct task_struct *task, | |
8726 | struct files_struct *files) | |
8727 | { | |
8728 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8729 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8730 | |
8731 | while (1) { | |
8732 | enum io_wq_cancel cret; | |
8733 | bool ret = false; | |
8734 | ||
1b00764f PB |
8735 | if (!task) { |
8736 | ret |= io_uring_try_cancel_iowq(ctx); | |
8737 | } else if (tctx && tctx->io_wq) { | |
8738 | /* | |
8739 | * Cancels requests of all rings, not only @ctx, but | |
8740 | * it's fine as the task is in exit/exec. | |
8741 | */ | |
5aa75ed5 | 8742 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8743 | &cancel, true); |
8744 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8745 | } | |
8746 | ||
8747 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8748 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8749 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8750 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8751 | io_iopoll_try_reap_events(ctx); | |
8752 | ret = true; | |
8753 | } | |
8754 | } | |
8755 | ||
e1915f76 | 8756 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8757 | ret |= io_poll_remove_all(ctx, task, files); |
8758 | ret |= io_kill_timeouts(ctx, task, files); | |
8759 | ret |= io_run_task_work(); | |
ba50a036 | 8760 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8761 | io_cqring_overflow_flush(ctx, true, task, files); |
8762 | if (!ret) | |
8763 | break; | |
8764 | cond_resched(); | |
8765 | } | |
8766 | } | |
8767 | ||
ca70f00b PB |
8768 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8769 | struct task_struct *task, | |
8770 | struct files_struct *files) | |
8771 | { | |
8772 | struct io_kiocb *req; | |
8773 | int cnt = 0; | |
8774 | ||
8775 | spin_lock_irq(&ctx->inflight_lock); | |
8776 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8777 | cnt += io_match_task(req, task, files); | |
8778 | spin_unlock_irq(&ctx->inflight_lock); | |
8779 | return cnt; | |
8780 | } | |
8781 | ||
b52fda00 | 8782 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8783 | struct task_struct *task, |
fcb323cc JA |
8784 | struct files_struct *files) |
8785 | { | |
fcb323cc | 8786 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8787 | DEFINE_WAIT(wait); |
ca70f00b | 8788 | int inflight; |
fcb323cc | 8789 | |
ca70f00b PB |
8790 | inflight = io_uring_count_inflight(ctx, task, files); |
8791 | if (!inflight) | |
fcb323cc | 8792 | break; |
f6edbabb | 8793 | |
9936c7c2 | 8794 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8795 | |
8796 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8797 | TASK_UNINTERRUPTIBLE); | |
8798 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8799 | schedule(); | |
c98de08c | 8800 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8801 | } |
0f212204 JA |
8802 | } |
8803 | ||
0f212204 JA |
8804 | /* |
8805 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8806 | */ | |
baf186c4 | 8807 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8808 | { |
236434c3 | 8809 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8810 | struct io_tctx_node *node; |
a528b04e | 8811 | int ret; |
236434c3 MWO |
8812 | |
8813 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8814 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8815 | if (unlikely(ret)) |
8816 | return ret; | |
236434c3 | 8817 | tctx = current->io_uring; |
0f212204 | 8818 | } |
baf186c4 PB |
8819 | if (tctx->last != ctx) { |
8820 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8821 | |
236434c3 | 8822 | if (!old) { |
13bf43f5 PB |
8823 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8824 | if (!node) | |
8825 | return -ENOMEM; | |
8826 | node->ctx = ctx; | |
13bf43f5 PB |
8827 | node->task = current; |
8828 | ||
baf186c4 | 8829 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8830 | node, GFP_KERNEL)); |
a528b04e | 8831 | if (ret) { |
13bf43f5 | 8832 | kfree(node); |
a528b04e PB |
8833 | return ret; |
8834 | } | |
13bf43f5 PB |
8835 | |
8836 | mutex_lock(&ctx->uring_lock); | |
8837 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8838 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8839 | } |
baf186c4 | 8840 | tctx->last = ctx; |
0f212204 | 8841 | } |
0f212204 JA |
8842 | return 0; |
8843 | } | |
8844 | ||
8845 | /* | |
8846 | * Remove this io_uring_file -> task mapping. | |
8847 | */ | |
2941267b | 8848 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8849 | { |
8850 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8851 | struct io_tctx_node *node; |
2941267b | 8852 | |
eebd2e37 PB |
8853 | if (!tctx) |
8854 | return; | |
13bf43f5 PB |
8855 | node = xa_erase(&tctx->xa, index); |
8856 | if (!node) | |
2941267b | 8857 | return; |
0f212204 | 8858 | |
13bf43f5 PB |
8859 | WARN_ON_ONCE(current != node->task); |
8860 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8861 | ||
8862 | mutex_lock(&node->ctx->uring_lock); | |
8863 | list_del(&node->ctx_node); | |
8864 | mutex_unlock(&node->ctx->uring_lock); | |
8865 | ||
baf186c4 | 8866 | if (tctx->last == node->ctx) |
0f212204 | 8867 | tctx->last = NULL; |
13bf43f5 | 8868 | kfree(node); |
0f212204 JA |
8869 | } |
8870 | ||
8452d4a6 | 8871 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8872 | { |
13bf43f5 | 8873 | struct io_tctx_node *node; |
de7f1d9e PB |
8874 | unsigned long index; |
8875 | ||
13bf43f5 | 8876 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8877 | io_uring_del_task_file(index); |
8452d4a6 PB |
8878 | if (tctx->io_wq) { |
8879 | io_wq_put_and_exit(tctx->io_wq); | |
8880 | tctx->io_wq = NULL; | |
8881 | } | |
de7f1d9e PB |
8882 | } |
8883 | ||
521d6a73 PB |
8884 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8885 | { | |
8886 | return percpu_counter_sum(&tctx->inflight); | |
8887 | } | |
8888 | ||
8889 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8890 | { | |
8891 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8892 | struct io_ring_ctx *ctx = work->ctx; | |
8893 | struct io_sq_data *sqd = ctx->sq_data; | |
8894 | ||
8895 | if (sqd->thread) | |
8896 | io_uring_cancel_sqpoll(ctx); | |
8897 | complete(&work->completion); | |
8898 | } | |
8899 | ||
8900 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8901 | { | |
8902 | struct io_sq_data *sqd = ctx->sq_data; | |
8903 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8904 | struct task_struct *task; | |
8905 | ||
8906 | io_sq_thread_park(sqd); | |
8907 | list_del_init(&ctx->sqd_list); | |
8908 | io_sqd_update_thread_idle(sqd); | |
8909 | task = sqd->thread; | |
8910 | if (task) { | |
8911 | init_completion(&work.completion); | |
8912 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8913 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8914 | wake_up_process(task); |
8915 | } | |
8916 | io_sq_thread_unpark(sqd); | |
8917 | ||
8918 | if (task) | |
8919 | wait_for_completion(&work.completion); | |
8920 | } | |
8921 | ||
0f212204 JA |
8922 | void __io_uring_files_cancel(struct files_struct *files) |
8923 | { | |
8924 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8925 | struct io_tctx_node *node; |
ce765372 | 8926 | unsigned long index; |
0f212204 JA |
8927 | |
8928 | /* make sure overflow events are dropped */ | |
fdaf083c | 8929 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8930 | xa_for_each(&tctx->xa, index, node) { |
8931 | struct io_ring_ctx *ctx = node->ctx; | |
8932 | ||
8933 | if (ctx->sq_data) { | |
8934 | io_sqpoll_cancel_sync(ctx); | |
8935 | continue; | |
8936 | } | |
8937 | io_uring_cancel_files(ctx, current, files); | |
8938 | if (!files) | |
8939 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8940 | } | |
fdaf083c | 8941 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8942 | |
8452d4a6 PB |
8943 | if (files) |
8944 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8945 | } |
8946 | ||
521d6a73 | 8947 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8948 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8949 | { | |
37d1e2e3 | 8950 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8951 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8952 | s64 inflight; |
8953 | DEFINE_WAIT(wait); | |
fdaf083c | 8954 | |
521d6a73 PB |
8955 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8956 | ||
0e9ddb39 PB |
8957 | atomic_inc(&tctx->in_idle); |
8958 | do { | |
8959 | /* read completions before cancelations */ | |
8960 | inflight = tctx_inflight(tctx); | |
8961 | if (!inflight) | |
8962 | break; | |
521d6a73 | 8963 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8964 | |
0e9ddb39 PB |
8965 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8966 | /* | |
8967 | * If we've seen completions, retry without waiting. This | |
8968 | * avoids a race where a completion comes in before we did | |
8969 | * prepare_to_wait(). | |
8970 | */ | |
8971 | if (inflight == tctx_inflight(tctx)) | |
8972 | schedule(); | |
8973 | finish_wait(&tctx->wait, &wait); | |
8974 | } while (1); | |
8975 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8976 | } |
8977 | ||
0f212204 JA |
8978 | /* |
8979 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8980 | * requests. | |
8981 | */ | |
8982 | void __io_uring_task_cancel(void) | |
8983 | { | |
8984 | struct io_uring_task *tctx = current->io_uring; | |
8985 | DEFINE_WAIT(wait); | |
d8a6df10 | 8986 | s64 inflight; |
0f212204 JA |
8987 | |
8988 | /* make sure overflow events are dropped */ | |
fdaf083c | 8989 | atomic_inc(&tctx->in_idle); |
d8a6df10 | 8990 | do { |
0f212204 | 8991 | /* read completions before cancelations */ |
fdaf083c | 8992 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8993 | if (!inflight) |
8994 | break; | |
0f212204 JA |
8995 | __io_uring_files_cancel(NULL); |
8996 | ||
8997 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8998 | ||
8999 | /* | |
a1bb3cd5 PB |
9000 | * If we've seen completions, retry without waiting. This |
9001 | * avoids a race where a completion comes in before we did | |
9002 | * prepare_to_wait(). | |
0f212204 | 9003 | */ |
a1bb3cd5 PB |
9004 | if (inflight == tctx_inflight(tctx)) |
9005 | schedule(); | |
f57555ed | 9006 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9007 | } while (1); |
0f212204 | 9008 | |
fdaf083c | 9009 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9010 | |
8452d4a6 PB |
9011 | io_uring_clean_tctx(tctx); |
9012 | /* all current's requests should be gone, we can kill tctx */ | |
9013 | __io_uring_free(current); | |
44e728b8 PB |
9014 | } |
9015 | ||
6c5c240e RP |
9016 | static void *io_uring_validate_mmap_request(struct file *file, |
9017 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9018 | { |
2b188cc1 | 9019 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9020 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9021 | struct page *page; |
9022 | void *ptr; | |
9023 | ||
9024 | switch (offset) { | |
9025 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9026 | case IORING_OFF_CQ_RING: |
9027 | ptr = ctx->rings; | |
2b188cc1 JA |
9028 | break; |
9029 | case IORING_OFF_SQES: | |
9030 | ptr = ctx->sq_sqes; | |
9031 | break; | |
2b188cc1 | 9032 | default: |
6c5c240e | 9033 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9034 | } |
9035 | ||
9036 | page = virt_to_head_page(ptr); | |
a50b854e | 9037 | if (sz > page_size(page)) |
6c5c240e RP |
9038 | return ERR_PTR(-EINVAL); |
9039 | ||
9040 | return ptr; | |
9041 | } | |
9042 | ||
9043 | #ifdef CONFIG_MMU | |
9044 | ||
9045 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9046 | { | |
9047 | size_t sz = vma->vm_end - vma->vm_start; | |
9048 | unsigned long pfn; | |
9049 | void *ptr; | |
9050 | ||
9051 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9052 | if (IS_ERR(ptr)) | |
9053 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9054 | |
9055 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9056 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9057 | } | |
9058 | ||
6c5c240e RP |
9059 | #else /* !CONFIG_MMU */ |
9060 | ||
9061 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9062 | { | |
9063 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9064 | } | |
9065 | ||
9066 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9067 | { | |
9068 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9069 | } | |
9070 | ||
9071 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9072 | unsigned long addr, unsigned long len, | |
9073 | unsigned long pgoff, unsigned long flags) | |
9074 | { | |
9075 | void *ptr; | |
9076 | ||
9077 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9078 | if (IS_ERR(ptr)) | |
9079 | return PTR_ERR(ptr); | |
9080 | ||
9081 | return (unsigned long) ptr; | |
9082 | } | |
9083 | ||
9084 | #endif /* !CONFIG_MMU */ | |
9085 | ||
d9d05217 | 9086 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9087 | { |
9088 | DEFINE_WAIT(wait); | |
9089 | ||
9090 | do { | |
9091 | if (!io_sqring_full(ctx)) | |
9092 | break; | |
90554200 JA |
9093 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9094 | ||
9095 | if (!io_sqring_full(ctx)) | |
9096 | break; | |
90554200 JA |
9097 | schedule(); |
9098 | } while (!signal_pending(current)); | |
9099 | ||
9100 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9101 | return 0; |
90554200 JA |
9102 | } |
9103 | ||
c73ebb68 HX |
9104 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9105 | struct __kernel_timespec __user **ts, | |
9106 | const sigset_t __user **sig) | |
9107 | { | |
9108 | struct io_uring_getevents_arg arg; | |
9109 | ||
9110 | /* | |
9111 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9112 | * is just a pointer to the sigset_t. | |
9113 | */ | |
9114 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9115 | *sig = (const sigset_t __user *) argp; | |
9116 | *ts = NULL; | |
9117 | return 0; | |
9118 | } | |
9119 | ||
9120 | /* | |
9121 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9122 | * timespec and sigset_t pointers if good. | |
9123 | */ | |
9124 | if (*argsz != sizeof(arg)) | |
9125 | return -EINVAL; | |
9126 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9127 | return -EFAULT; | |
9128 | *sig = u64_to_user_ptr(arg.sigmask); | |
9129 | *argsz = arg.sigmask_sz; | |
9130 | *ts = u64_to_user_ptr(arg.ts); | |
9131 | return 0; | |
9132 | } | |
9133 | ||
2b188cc1 | 9134 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9135 | u32, min_complete, u32, flags, const void __user *, argp, |
9136 | size_t, argsz) | |
2b188cc1 JA |
9137 | { |
9138 | struct io_ring_ctx *ctx; | |
9139 | long ret = -EBADF; | |
9140 | int submitted = 0; | |
9141 | struct fd f; | |
9142 | ||
4c6e277c | 9143 | io_run_task_work(); |
b41e9852 | 9144 | |
90554200 | 9145 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9146 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9147 | return -EINVAL; |
9148 | ||
9149 | f = fdget(fd); | |
9150 | if (!f.file) | |
9151 | return -EBADF; | |
9152 | ||
9153 | ret = -EOPNOTSUPP; | |
9154 | if (f.file->f_op != &io_uring_fops) | |
9155 | goto out_fput; | |
9156 | ||
9157 | ret = -ENXIO; | |
9158 | ctx = f.file->private_data; | |
9159 | if (!percpu_ref_tryget(&ctx->refs)) | |
9160 | goto out_fput; | |
9161 | ||
7e84e1c7 SG |
9162 | ret = -EBADFD; |
9163 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9164 | goto out; | |
9165 | ||
6c271ce2 JA |
9166 | /* |
9167 | * For SQ polling, the thread will do all submissions and completions. | |
9168 | * Just return the requested submit count, and wake the thread if | |
9169 | * we were asked to. | |
9170 | */ | |
b2a9eada | 9171 | ret = 0; |
6c271ce2 | 9172 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9173 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9174 | |
d9d05217 | 9175 | ret = -EOWNERDEAD; |
04147488 SM |
9176 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9177 | goto out; | |
9178 | } | |
6c271ce2 | 9179 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9180 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9181 | if (flags & IORING_ENTER_SQ_WAIT) { |
9182 | ret = io_sqpoll_wait_sq(ctx); | |
9183 | if (ret) | |
9184 | goto out; | |
9185 | } | |
6c271ce2 | 9186 | submitted = to_submit; |
b2a9eada | 9187 | } else if (to_submit) { |
baf186c4 | 9188 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9189 | if (unlikely(ret)) |
9190 | goto out; | |
2b188cc1 | 9191 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9192 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9193 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9194 | |
9195 | if (submitted != to_submit) | |
9196 | goto out; | |
2b188cc1 JA |
9197 | } |
9198 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9199 | const sigset_t __user *sig; |
9200 | struct __kernel_timespec __user *ts; | |
9201 | ||
9202 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9203 | if (unlikely(ret)) | |
9204 | goto out; | |
9205 | ||
2b188cc1 JA |
9206 | min_complete = min(min_complete, ctx->cq_entries); |
9207 | ||
32b2244a XW |
9208 | /* |
9209 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9210 | * space applications don't need to do io completion events | |
9211 | * polling again, they can rely on io_sq_thread to do polling | |
9212 | * work, which can reduce cpu usage and uring_lock contention. | |
9213 | */ | |
9214 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9215 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9216 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9217 | } else { |
c73ebb68 | 9218 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9219 | } |
2b188cc1 JA |
9220 | } |
9221 | ||
7c504e65 | 9222 | out: |
6805b32e | 9223 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9224 | out_fput: |
9225 | fdput(f); | |
9226 | return submitted ? submitted : ret; | |
9227 | } | |
9228 | ||
bebdb65e | 9229 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9230 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9231 | const struct cred *cred) | |
87ce955b | 9232 | { |
87ce955b JA |
9233 | struct user_namespace *uns = seq_user_ns(m); |
9234 | struct group_info *gi; | |
9235 | kernel_cap_t cap; | |
9236 | unsigned __capi; | |
9237 | int g; | |
9238 | ||
9239 | seq_printf(m, "%5d\n", id); | |
9240 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9241 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9242 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9243 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9244 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9245 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9246 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9247 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9248 | seq_puts(m, "\n\tGroups:\t"); | |
9249 | gi = cred->group_info; | |
9250 | for (g = 0; g < gi->ngroups; g++) { | |
9251 | seq_put_decimal_ull(m, g ? " " : "", | |
9252 | from_kgid_munged(uns, gi->gid[g])); | |
9253 | } | |
9254 | seq_puts(m, "\n\tCapEff:\t"); | |
9255 | cap = cred->cap_effective; | |
9256 | CAP_FOR_EACH_U32(__capi) | |
9257 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9258 | seq_putc(m, '\n'); | |
9259 | return 0; | |
9260 | } | |
9261 | ||
9262 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9263 | { | |
dbbe9c64 | 9264 | struct io_sq_data *sq = NULL; |
fad8e0de | 9265 | bool has_lock; |
87ce955b JA |
9266 | int i; |
9267 | ||
fad8e0de JA |
9268 | /* |
9269 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9270 | * since fdinfo case grabs it in the opposite direction of normal use | |
9271 | * cases. If we fail to get the lock, we just don't iterate any | |
9272 | * structures that could be going away outside the io_uring mutex. | |
9273 | */ | |
9274 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9275 | ||
5f3f26f9 | 9276 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9277 | sq = ctx->sq_data; |
5f3f26f9 JA |
9278 | if (!sq->thread) |
9279 | sq = NULL; | |
9280 | } | |
dbbe9c64 JQ |
9281 | |
9282 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9283 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9284 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9285 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9286 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9287 | |
87ce955b JA |
9288 | if (f) |
9289 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9290 | else | |
9291 | seq_printf(m, "%5u: <none>\n", i); | |
9292 | } | |
9293 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9294 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9295 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9296 | ||
9297 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9298 | (unsigned int) buf->len); | |
9299 | } | |
61cf9370 MWO |
9300 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9301 | unsigned long index; | |
9302 | const struct cred *cred; | |
9303 | ||
87ce955b | 9304 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9305 | xa_for_each(&ctx->personalities, index, cred) |
9306 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9307 | } |
d7718a9d JA |
9308 | seq_printf(m, "PollList:\n"); |
9309 | spin_lock_irq(&ctx->completion_lock); | |
9310 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9311 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9312 | struct io_kiocb *req; | |
9313 | ||
9314 | hlist_for_each_entry(req, list, hash_node) | |
9315 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9316 | req->task->task_works != NULL); | |
9317 | } | |
9318 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9319 | if (has_lock) |
9320 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9321 | } |
9322 | ||
9323 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9324 | { | |
9325 | struct io_ring_ctx *ctx = f->private_data; | |
9326 | ||
9327 | if (percpu_ref_tryget(&ctx->refs)) { | |
9328 | __io_uring_show_fdinfo(ctx, m); | |
9329 | percpu_ref_put(&ctx->refs); | |
9330 | } | |
9331 | } | |
bebdb65e | 9332 | #endif |
87ce955b | 9333 | |
2b188cc1 JA |
9334 | static const struct file_operations io_uring_fops = { |
9335 | .release = io_uring_release, | |
9336 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9337 | #ifndef CONFIG_MMU |
9338 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9339 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9340 | #endif | |
2b188cc1 JA |
9341 | .poll = io_uring_poll, |
9342 | .fasync = io_uring_fasync, | |
bebdb65e | 9343 | #ifdef CONFIG_PROC_FS |
87ce955b | 9344 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9345 | #endif |
2b188cc1 JA |
9346 | }; |
9347 | ||
9348 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9349 | struct io_uring_params *p) | |
9350 | { | |
75b28aff HV |
9351 | struct io_rings *rings; |
9352 | size_t size, sq_array_offset; | |
2b188cc1 | 9353 | |
bd740481 JA |
9354 | /* make sure these are sane, as we already accounted them */ |
9355 | ctx->sq_entries = p->sq_entries; | |
9356 | ctx->cq_entries = p->cq_entries; | |
9357 | ||
75b28aff HV |
9358 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9359 | if (size == SIZE_MAX) | |
9360 | return -EOVERFLOW; | |
9361 | ||
9362 | rings = io_mem_alloc(size); | |
9363 | if (!rings) | |
2b188cc1 JA |
9364 | return -ENOMEM; |
9365 | ||
75b28aff HV |
9366 | ctx->rings = rings; |
9367 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9368 | rings->sq_ring_mask = p->sq_entries - 1; | |
9369 | rings->cq_ring_mask = p->cq_entries - 1; | |
9370 | rings->sq_ring_entries = p->sq_entries; | |
9371 | rings->cq_ring_entries = p->cq_entries; | |
9372 | ctx->sq_mask = rings->sq_ring_mask; | |
9373 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9374 | |
9375 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9376 | if (size == SIZE_MAX) { |
9377 | io_mem_free(ctx->rings); | |
9378 | ctx->rings = NULL; | |
2b188cc1 | 9379 | return -EOVERFLOW; |
eb065d30 | 9380 | } |
2b188cc1 JA |
9381 | |
9382 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9383 | if (!ctx->sq_sqes) { |
9384 | io_mem_free(ctx->rings); | |
9385 | ctx->rings = NULL; | |
2b188cc1 | 9386 | return -ENOMEM; |
eb065d30 | 9387 | } |
2b188cc1 | 9388 | |
2b188cc1 JA |
9389 | return 0; |
9390 | } | |
9391 | ||
9faadcc8 PB |
9392 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9393 | { | |
9394 | int ret, fd; | |
9395 | ||
9396 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9397 | if (fd < 0) | |
9398 | return fd; | |
9399 | ||
baf186c4 | 9400 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9401 | if (ret) { |
9402 | put_unused_fd(fd); | |
9403 | return ret; | |
9404 | } | |
9405 | fd_install(fd, file); | |
9406 | return fd; | |
9407 | } | |
9408 | ||
2b188cc1 JA |
9409 | /* |
9410 | * Allocate an anonymous fd, this is what constitutes the application | |
9411 | * visible backing of an io_uring instance. The application mmaps this | |
9412 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9413 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9414 | */ | |
9faadcc8 | 9415 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9416 | { |
9417 | struct file *file; | |
9faadcc8 | 9418 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9419 | int ret; |
9420 | ||
2b188cc1 JA |
9421 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9422 | &ctx->ring_sock); | |
9423 | if (ret) | |
9faadcc8 | 9424 | return ERR_PTR(ret); |
2b188cc1 JA |
9425 | #endif |
9426 | ||
2b188cc1 JA |
9427 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9428 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9429 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9430 | if (IS_ERR(file)) { |
9431 | sock_release(ctx->ring_sock); | |
9432 | ctx->ring_sock = NULL; | |
9433 | } else { | |
9434 | ctx->ring_sock->file = file; | |
0f212204 | 9435 | } |
2b188cc1 | 9436 | #endif |
9faadcc8 | 9437 | return file; |
2b188cc1 JA |
9438 | } |
9439 | ||
7f13657d XW |
9440 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9441 | struct io_uring_params __user *params) | |
2b188cc1 | 9442 | { |
2b188cc1 | 9443 | struct io_ring_ctx *ctx; |
9faadcc8 | 9444 | struct file *file; |
2b188cc1 JA |
9445 | int ret; |
9446 | ||
8110c1a6 | 9447 | if (!entries) |
2b188cc1 | 9448 | return -EINVAL; |
8110c1a6 JA |
9449 | if (entries > IORING_MAX_ENTRIES) { |
9450 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9451 | return -EINVAL; | |
9452 | entries = IORING_MAX_ENTRIES; | |
9453 | } | |
2b188cc1 JA |
9454 | |
9455 | /* | |
9456 | * Use twice as many entries for the CQ ring. It's possible for the | |
9457 | * application to drive a higher depth than the size of the SQ ring, | |
9458 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9459 | * some flexibility in overcommitting a bit. If the application has |
9460 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9461 | * of CQ ring entries manually. | |
2b188cc1 JA |
9462 | */ |
9463 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9464 | if (p->flags & IORING_SETUP_CQSIZE) { |
9465 | /* | |
9466 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9467 | * to a power-of-two, if it isn't already. We do NOT impose | |
9468 | * any cq vs sq ring sizing. | |
9469 | */ | |
eb2667b3 | 9470 | if (!p->cq_entries) |
33a107f0 | 9471 | return -EINVAL; |
8110c1a6 JA |
9472 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9473 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9474 | return -EINVAL; | |
9475 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9476 | } | |
eb2667b3 JQ |
9477 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9478 | if (p->cq_entries < p->sq_entries) | |
9479 | return -EINVAL; | |
33a107f0 JA |
9480 | } else { |
9481 | p->cq_entries = 2 * p->sq_entries; | |
9482 | } | |
2b188cc1 | 9483 | |
2b188cc1 | 9484 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9485 | if (!ctx) |
2b188cc1 | 9486 | return -ENOMEM; |
2b188cc1 | 9487 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9488 | if (!capable(CAP_IPC_LOCK)) |
9489 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9490 | |
9491 | /* | |
9492 | * This is just grabbed for accounting purposes. When a process exits, | |
9493 | * the mm is exited and dropped before the files, hence we need to hang | |
9494 | * on to this mm purely for the purposes of being able to unaccount | |
9495 | * memory (locked/pinned vm). It's not used for anything else. | |
9496 | */ | |
6b7898eb | 9497 | mmgrab(current->mm); |
2aede0e4 | 9498 | ctx->mm_account = current->mm; |
6b7898eb | 9499 | |
2b188cc1 JA |
9500 | ret = io_allocate_scq_urings(ctx, p); |
9501 | if (ret) | |
9502 | goto err; | |
9503 | ||
7e84e1c7 | 9504 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9505 | if (ret) |
9506 | goto err; | |
9507 | ||
2b188cc1 | 9508 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9509 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9510 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9511 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9512 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9513 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9514 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9515 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9516 | |
9517 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9518 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9519 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9520 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9521 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9522 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9523 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9524 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9525 | |
7f13657d XW |
9526 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9527 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9528 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9529 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9530 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9531 | |
9532 | if (copy_to_user(params, p, sizeof(*p))) { | |
9533 | ret = -EFAULT; | |
9534 | goto err; | |
9535 | } | |
d1719f70 | 9536 | |
9faadcc8 PB |
9537 | file = io_uring_get_file(ctx); |
9538 | if (IS_ERR(file)) { | |
9539 | ret = PTR_ERR(file); | |
9540 | goto err; | |
9541 | } | |
9542 | ||
044c1ab3 JA |
9543 | /* |
9544 | * Install ring fd as the very last thing, so we don't risk someone | |
9545 | * having closed it before we finish setup | |
9546 | */ | |
9faadcc8 PB |
9547 | ret = io_uring_install_fd(ctx, file); |
9548 | if (ret < 0) { | |
9549 | /* fput will clean it up */ | |
9550 | fput(file); | |
9551 | return ret; | |
9552 | } | |
044c1ab3 | 9553 | |
c826bd7a | 9554 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9555 | return ret; |
9556 | err: | |
9557 | io_ring_ctx_wait_and_kill(ctx); | |
9558 | return ret; | |
9559 | } | |
9560 | ||
9561 | /* | |
9562 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9563 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9564 | * params structure passed in. | |
9565 | */ | |
9566 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9567 | { | |
9568 | struct io_uring_params p; | |
2b188cc1 JA |
9569 | int i; |
9570 | ||
9571 | if (copy_from_user(&p, params, sizeof(p))) | |
9572 | return -EFAULT; | |
9573 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9574 | if (p.resv[i]) | |
9575 | return -EINVAL; | |
9576 | } | |
9577 | ||
6c271ce2 | 9578 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9579 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9580 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9581 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9582 | return -EINVAL; |
9583 | ||
7f13657d | 9584 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9585 | } |
9586 | ||
9587 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9588 | struct io_uring_params __user *, params) | |
9589 | { | |
9590 | return io_uring_setup(entries, params); | |
9591 | } | |
9592 | ||
66f4af93 JA |
9593 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9594 | { | |
9595 | struct io_uring_probe *p; | |
9596 | size_t size; | |
9597 | int i, ret; | |
9598 | ||
9599 | size = struct_size(p, ops, nr_args); | |
9600 | if (size == SIZE_MAX) | |
9601 | return -EOVERFLOW; | |
9602 | p = kzalloc(size, GFP_KERNEL); | |
9603 | if (!p) | |
9604 | return -ENOMEM; | |
9605 | ||
9606 | ret = -EFAULT; | |
9607 | if (copy_from_user(p, arg, size)) | |
9608 | goto out; | |
9609 | ret = -EINVAL; | |
9610 | if (memchr_inv(p, 0, size)) | |
9611 | goto out; | |
9612 | ||
9613 | p->last_op = IORING_OP_LAST - 1; | |
9614 | if (nr_args > IORING_OP_LAST) | |
9615 | nr_args = IORING_OP_LAST; | |
9616 | ||
9617 | for (i = 0; i < nr_args; i++) { | |
9618 | p->ops[i].op = i; | |
9619 | if (!io_op_defs[i].not_supported) | |
9620 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9621 | } | |
9622 | p->ops_len = i; | |
9623 | ||
9624 | ret = 0; | |
9625 | if (copy_to_user(arg, p, size)) | |
9626 | ret = -EFAULT; | |
9627 | out: | |
9628 | kfree(p); | |
9629 | return ret; | |
9630 | } | |
9631 | ||
071698e1 JA |
9632 | static int io_register_personality(struct io_ring_ctx *ctx) |
9633 | { | |
4379bf8b | 9634 | const struct cred *creds; |
61cf9370 | 9635 | u32 id; |
1e6fa521 | 9636 | int ret; |
071698e1 | 9637 | |
4379bf8b | 9638 | creds = get_current_cred(); |
1e6fa521 | 9639 | |
61cf9370 MWO |
9640 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9641 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9642 | if (!ret) | |
9643 | return id; | |
9644 | put_cred(creds); | |
1e6fa521 | 9645 | return ret; |
071698e1 JA |
9646 | } |
9647 | ||
21b55dbc SG |
9648 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9649 | unsigned int nr_args) | |
9650 | { | |
9651 | struct io_uring_restriction *res; | |
9652 | size_t size; | |
9653 | int i, ret; | |
9654 | ||
7e84e1c7 SG |
9655 | /* Restrictions allowed only if rings started disabled */ |
9656 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9657 | return -EBADFD; | |
9658 | ||
21b55dbc | 9659 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9660 | if (ctx->restrictions.registered) |
21b55dbc SG |
9661 | return -EBUSY; |
9662 | ||
9663 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9664 | return -EINVAL; | |
9665 | ||
9666 | size = array_size(nr_args, sizeof(*res)); | |
9667 | if (size == SIZE_MAX) | |
9668 | return -EOVERFLOW; | |
9669 | ||
9670 | res = memdup_user(arg, size); | |
9671 | if (IS_ERR(res)) | |
9672 | return PTR_ERR(res); | |
9673 | ||
9674 | ret = 0; | |
9675 | ||
9676 | for (i = 0; i < nr_args; i++) { | |
9677 | switch (res[i].opcode) { | |
9678 | case IORING_RESTRICTION_REGISTER_OP: | |
9679 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9680 | ret = -EINVAL; | |
9681 | goto out; | |
9682 | } | |
9683 | ||
9684 | __set_bit(res[i].register_op, | |
9685 | ctx->restrictions.register_op); | |
9686 | break; | |
9687 | case IORING_RESTRICTION_SQE_OP: | |
9688 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9689 | ret = -EINVAL; | |
9690 | goto out; | |
9691 | } | |
9692 | ||
9693 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9694 | break; | |
9695 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9696 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9697 | break; | |
9698 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9699 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9700 | break; | |
9701 | default: | |
9702 | ret = -EINVAL; | |
9703 | goto out; | |
9704 | } | |
9705 | } | |
9706 | ||
9707 | out: | |
9708 | /* Reset all restrictions if an error happened */ | |
9709 | if (ret != 0) | |
9710 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9711 | else | |
7e84e1c7 | 9712 | ctx->restrictions.registered = true; |
21b55dbc SG |
9713 | |
9714 | kfree(res); | |
9715 | return ret; | |
9716 | } | |
9717 | ||
7e84e1c7 SG |
9718 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9719 | { | |
9720 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9721 | return -EBADFD; | |
9722 | ||
9723 | if (ctx->restrictions.registered) | |
9724 | ctx->restricted = 1; | |
9725 | ||
0298ef96 PB |
9726 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9727 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9728 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9729 | return 0; |
9730 | } | |
9731 | ||
071698e1 JA |
9732 | static bool io_register_op_must_quiesce(int op) |
9733 | { | |
9734 | switch (op) { | |
9735 | case IORING_UNREGISTER_FILES: | |
9736 | case IORING_REGISTER_FILES_UPDATE: | |
9737 | case IORING_REGISTER_PROBE: | |
9738 | case IORING_REGISTER_PERSONALITY: | |
9739 | case IORING_UNREGISTER_PERSONALITY: | |
9740 | return false; | |
9741 | default: | |
9742 | return true; | |
9743 | } | |
9744 | } | |
9745 | ||
edafccee JA |
9746 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9747 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9748 | __releases(ctx->uring_lock) |
9749 | __acquires(ctx->uring_lock) | |
edafccee JA |
9750 | { |
9751 | int ret; | |
9752 | ||
35fa71a0 JA |
9753 | /* |
9754 | * We're inside the ring mutex, if the ref is already dying, then | |
9755 | * someone else killed the ctx or is already going through | |
9756 | * io_uring_register(). | |
9757 | */ | |
9758 | if (percpu_ref_is_dying(&ctx->refs)) | |
9759 | return -ENXIO; | |
9760 | ||
071698e1 | 9761 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9762 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9763 | |
05f3fb3c JA |
9764 | /* |
9765 | * Drop uring mutex before waiting for references to exit. If | |
9766 | * another thread is currently inside io_uring_enter() it might | |
9767 | * need to grab the uring_lock to make progress. If we hold it | |
9768 | * here across the drain wait, then we can deadlock. It's safe | |
9769 | * to drop the mutex here, since no new references will come in | |
9770 | * after we've killed the percpu ref. | |
9771 | */ | |
9772 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9773 | do { |
9774 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9775 | if (!ret) | |
9776 | break; | |
ed6930c9 JA |
9777 | ret = io_run_task_work_sig(); |
9778 | if (ret < 0) | |
9779 | break; | |
af9c1a44 JA |
9780 | } while (1); |
9781 | ||
05f3fb3c | 9782 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9783 | |
c150368b JA |
9784 | if (ret) { |
9785 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9786 | goto out_quiesce; |
9787 | } | |
9788 | } | |
9789 | ||
9790 | if (ctx->restricted) { | |
9791 | if (opcode >= IORING_REGISTER_LAST) { | |
9792 | ret = -EINVAL; | |
9793 | goto out; | |
9794 | } | |
9795 | ||
9796 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9797 | ret = -EACCES; | |
c150368b JA |
9798 | goto out; |
9799 | } | |
05f3fb3c | 9800 | } |
edafccee JA |
9801 | |
9802 | switch (opcode) { | |
9803 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9804 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9805 | break; |
9806 | case IORING_UNREGISTER_BUFFERS: | |
9807 | ret = -EINVAL; | |
9808 | if (arg || nr_args) | |
9809 | break; | |
0a96bbe4 | 9810 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9811 | break; |
6b06314c JA |
9812 | case IORING_REGISTER_FILES: |
9813 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9814 | break; | |
9815 | case IORING_UNREGISTER_FILES: | |
9816 | ret = -EINVAL; | |
9817 | if (arg || nr_args) | |
9818 | break; | |
9819 | ret = io_sqe_files_unregister(ctx); | |
9820 | break; | |
c3a31e60 JA |
9821 | case IORING_REGISTER_FILES_UPDATE: |
9822 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9823 | break; | |
9b402849 | 9824 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9825 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9826 | ret = -EINVAL; |
9827 | if (nr_args != 1) | |
9828 | break; | |
9829 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9830 | if (ret) |
9831 | break; | |
9832 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9833 | ctx->eventfd_async = 1; | |
9834 | else | |
9835 | ctx->eventfd_async = 0; | |
9b402849 JA |
9836 | break; |
9837 | case IORING_UNREGISTER_EVENTFD: | |
9838 | ret = -EINVAL; | |
9839 | if (arg || nr_args) | |
9840 | break; | |
9841 | ret = io_eventfd_unregister(ctx); | |
9842 | break; | |
66f4af93 JA |
9843 | case IORING_REGISTER_PROBE: |
9844 | ret = -EINVAL; | |
9845 | if (!arg || nr_args > 256) | |
9846 | break; | |
9847 | ret = io_probe(ctx, arg, nr_args); | |
9848 | break; | |
071698e1 JA |
9849 | case IORING_REGISTER_PERSONALITY: |
9850 | ret = -EINVAL; | |
9851 | if (arg || nr_args) | |
9852 | break; | |
9853 | ret = io_register_personality(ctx); | |
9854 | break; | |
9855 | case IORING_UNREGISTER_PERSONALITY: | |
9856 | ret = -EINVAL; | |
9857 | if (arg) | |
9858 | break; | |
9859 | ret = io_unregister_personality(ctx, nr_args); | |
9860 | break; | |
7e84e1c7 SG |
9861 | case IORING_REGISTER_ENABLE_RINGS: |
9862 | ret = -EINVAL; | |
9863 | if (arg || nr_args) | |
9864 | break; | |
9865 | ret = io_register_enable_rings(ctx); | |
9866 | break; | |
21b55dbc SG |
9867 | case IORING_REGISTER_RESTRICTIONS: |
9868 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9869 | break; | |
edafccee JA |
9870 | default: |
9871 | ret = -EINVAL; | |
9872 | break; | |
9873 | } | |
9874 | ||
21b55dbc | 9875 | out: |
071698e1 | 9876 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9877 | /* bring the ctx back to life */ |
05f3fb3c | 9878 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9879 | out_quiesce: |
0f158b4c | 9880 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9881 | } |
edafccee JA |
9882 | return ret; |
9883 | } | |
9884 | ||
9885 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9886 | void __user *, arg, unsigned int, nr_args) | |
9887 | { | |
9888 | struct io_ring_ctx *ctx; | |
9889 | long ret = -EBADF; | |
9890 | struct fd f; | |
9891 | ||
9892 | f = fdget(fd); | |
9893 | if (!f.file) | |
9894 | return -EBADF; | |
9895 | ||
9896 | ret = -EOPNOTSUPP; | |
9897 | if (f.file->f_op != &io_uring_fops) | |
9898 | goto out_fput; | |
9899 | ||
9900 | ctx = f.file->private_data; | |
9901 | ||
b6c23dd5 PB |
9902 | io_run_task_work(); |
9903 | ||
edafccee JA |
9904 | mutex_lock(&ctx->uring_lock); |
9905 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9906 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9907 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9908 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9909 | out_fput: |
9910 | fdput(f); | |
9911 | return ret; | |
9912 | } | |
9913 | ||
2b188cc1 JA |
9914 | static int __init io_uring_init(void) |
9915 | { | |
d7f62e82 SM |
9916 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9917 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9918 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9919 | } while (0) | |
9920 | ||
9921 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9922 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9923 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9924 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9925 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9926 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9927 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9928 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9929 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9930 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9931 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9932 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9933 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9934 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9935 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9936 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9937 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9938 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9939 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9940 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9941 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9942 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9943 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9944 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9945 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9946 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9947 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9948 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9949 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9950 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9951 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9952 | |
d3656344 | 9953 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9954 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9955 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9956 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9957 | return 0; |
9958 | }; | |
9959 | __initcall(io_uring_init); |