]>
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> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
bcf5a063 | 81 | #include <linux/pagemap.h> |
0f212204 | 82 | #include <linux/io_uring.h> |
91d8f519 | 83 | #include <linux/blk-cgroup.h> |
4ea33a97 | 84 | #include <linux/audit.h> |
2b188cc1 | 85 | |
c826bd7a DD |
86 | #define CREATE_TRACE_POINTS |
87 | #include <trace/events/io_uring.h> | |
88 | ||
2b188cc1 JA |
89 | #include <uapi/linux/io_uring.h> |
90 | ||
91 | #include "internal.h" | |
561fb04a | 92 | #include "io-wq.h" |
2b188cc1 | 93 | |
5277deaa | 94 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 95 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
96 | |
97 | /* | |
98 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
99 | */ | |
100 | #define IORING_FILE_TABLE_SHIFT 9 | |
101 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
102 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
103 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
104 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
105 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 JA |
106 | |
107 | struct io_uring { | |
108 | u32 head ____cacheline_aligned_in_smp; | |
109 | u32 tail ____cacheline_aligned_in_smp; | |
110 | }; | |
111 | ||
1e84b97b | 112 | /* |
75b28aff HV |
113 | * This data is shared with the application through the mmap at offsets |
114 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
115 | * |
116 | * The offsets to the member fields are published through struct | |
117 | * io_sqring_offsets when calling io_uring_setup. | |
118 | */ | |
75b28aff | 119 | struct io_rings { |
1e84b97b SB |
120 | /* |
121 | * Head and tail offsets into the ring; the offsets need to be | |
122 | * masked to get valid indices. | |
123 | * | |
75b28aff HV |
124 | * The kernel controls head of the sq ring and the tail of the cq ring, |
125 | * and the application controls tail of the sq ring and the head of the | |
126 | * cq ring. | |
1e84b97b | 127 | */ |
75b28aff | 128 | struct io_uring sq, cq; |
1e84b97b | 129 | /* |
75b28aff | 130 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
131 | * ring_entries - 1) |
132 | */ | |
75b28aff HV |
133 | u32 sq_ring_mask, cq_ring_mask; |
134 | /* Ring sizes (constant, power of 2) */ | |
135 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
136 | /* |
137 | * Number of invalid entries dropped by the kernel due to | |
138 | * invalid index stored in array | |
139 | * | |
140 | * Written by the kernel, shouldn't be modified by the | |
141 | * application (i.e. get number of "new events" by comparing to | |
142 | * cached value). | |
143 | * | |
144 | * After a new SQ head value was read by the application this | |
145 | * counter includes all submissions that were dropped reaching | |
146 | * the new SQ head (and possibly more). | |
147 | */ | |
75b28aff | 148 | u32 sq_dropped; |
1e84b97b | 149 | /* |
0d9b5b3a | 150 | * Runtime SQ flags |
1e84b97b SB |
151 | * |
152 | * Written by the kernel, shouldn't be modified by the | |
153 | * application. | |
154 | * | |
155 | * The application needs a full memory barrier before checking | |
156 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
157 | */ | |
75b28aff | 158 | u32 sq_flags; |
0d9b5b3a SG |
159 | /* |
160 | * Runtime CQ flags | |
161 | * | |
162 | * Written by the application, shouldn't be modified by the | |
163 | * kernel. | |
164 | */ | |
165 | u32 cq_flags; | |
1e84b97b SB |
166 | /* |
167 | * Number of completion events lost because the queue was full; | |
168 | * this should be avoided by the application by making sure | |
0b4295b5 | 169 | * there are not more requests pending than there is space in |
1e84b97b SB |
170 | * the completion queue. |
171 | * | |
172 | * Written by the kernel, shouldn't be modified by the | |
173 | * application (i.e. get number of "new events" by comparing to | |
174 | * cached value). | |
175 | * | |
176 | * As completion events come in out of order this counter is not | |
177 | * ordered with any other data. | |
178 | */ | |
75b28aff | 179 | u32 cq_overflow; |
1e84b97b SB |
180 | /* |
181 | * Ring buffer of completion events. | |
182 | * | |
183 | * The kernel writes completion events fresh every time they are | |
184 | * produced, so the application is allowed to modify pending | |
185 | * entries. | |
186 | */ | |
75b28aff | 187 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
188 | }; |
189 | ||
45d189c6 PB |
190 | enum io_uring_cmd_flags { |
191 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 192 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
193 | }; |
194 | ||
edafccee JA |
195 | struct io_mapped_ubuf { |
196 | u64 ubuf; | |
197 | size_t len; | |
198 | struct bio_vec *bvec; | |
199 | unsigned int nr_bvecs; | |
de293938 | 200 | unsigned long acct_pages; |
edafccee JA |
201 | }; |
202 | ||
50238531 BM |
203 | struct io_ring_ctx; |
204 | ||
269bbe5f BM |
205 | struct io_rsrc_put { |
206 | struct list_head list; | |
50238531 BM |
207 | union { |
208 | void *rsrc; | |
209 | struct file *file; | |
210 | }; | |
269bbe5f BM |
211 | }; |
212 | ||
213 | struct fixed_rsrc_table { | |
65e19f54 | 214 | struct file **files; |
31b51510 JA |
215 | }; |
216 | ||
269bbe5f | 217 | struct fixed_rsrc_ref_node { |
05589553 XW |
218 | struct percpu_ref refs; |
219 | struct list_head node; | |
269bbe5f BM |
220 | struct list_head rsrc_list; |
221 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
222 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
223 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 224 | struct llist_node llist; |
e297822b | 225 | bool done; |
05589553 XW |
226 | }; |
227 | ||
269bbe5f BM |
228 | struct fixed_rsrc_data { |
229 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
230 | struct io_ring_ctx *ctx; |
231 | ||
269bbe5f | 232 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 233 | struct percpu_ref refs; |
05f3fb3c JA |
234 | struct completion done; |
235 | }; | |
236 | ||
5a2e745d JA |
237 | struct io_buffer { |
238 | struct list_head list; | |
239 | __u64 addr; | |
240 | __s32 len; | |
241 | __u16 bid; | |
242 | }; | |
243 | ||
21b55dbc SG |
244 | struct io_restriction { |
245 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
246 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
247 | u8 sqe_flags_allowed; | |
248 | u8 sqe_flags_required; | |
7e84e1c7 | 249 | bool registered; |
21b55dbc SG |
250 | }; |
251 | ||
534ca6d6 JA |
252 | struct io_sq_data { |
253 | refcount_t refs; | |
69fb2131 JA |
254 | struct mutex lock; |
255 | ||
256 | /* ctx's that are using this sqd */ | |
257 | struct list_head ctx_list; | |
258 | struct list_head ctx_new_list; | |
259 | struct mutex ctx_lock; | |
260 | ||
534ca6d6 JA |
261 | struct task_struct *thread; |
262 | struct wait_queue_head wait; | |
08369246 XW |
263 | |
264 | unsigned sq_thread_idle; | |
534ca6d6 JA |
265 | }; |
266 | ||
258b29a9 | 267 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 268 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 269 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 270 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
271 | |
272 | struct io_comp_state { | |
6dd0be1e | 273 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 274 | unsigned int nr; |
c7dae4ba JA |
275 | unsigned int locked_free_nr; |
276 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 277 | struct list_head free_list; |
c7dae4ba JA |
278 | /* IRQ completion list, under ->completion_lock */ |
279 | struct list_head locked_free_list; | |
258b29a9 PB |
280 | }; |
281 | ||
282 | struct io_submit_state { | |
283 | struct blk_plug plug; | |
284 | ||
285 | /* | |
286 | * io_kiocb alloc cache | |
287 | */ | |
bf019da7 | 288 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
289 | unsigned int free_reqs; |
290 | ||
291 | bool plug_started; | |
292 | ||
293 | /* | |
294 | * Batch completion logic | |
295 | */ | |
296 | struct io_comp_state comp; | |
297 | ||
298 | /* | |
299 | * File reference cache | |
300 | */ | |
301 | struct file *file; | |
302 | unsigned int fd; | |
303 | unsigned int file_refs; | |
304 | unsigned int ios_left; | |
305 | }; | |
306 | ||
2b188cc1 JA |
307 | struct io_ring_ctx { |
308 | struct { | |
309 | struct percpu_ref refs; | |
310 | } ____cacheline_aligned_in_smp; | |
311 | ||
312 | struct { | |
313 | unsigned int flags; | |
e1d85334 | 314 | unsigned int compat: 1; |
aad5d8da | 315 | unsigned int limit_mem: 1; |
e1d85334 RD |
316 | unsigned int cq_overflow_flushed: 1; |
317 | unsigned int drain_next: 1; | |
318 | unsigned int eventfd_async: 1; | |
21b55dbc | 319 | unsigned int restricted: 1; |
d9d05217 | 320 | unsigned int sqo_dead: 1; |
2b188cc1 | 321 | |
75b28aff HV |
322 | /* |
323 | * Ring buffer of indices into array of io_uring_sqe, which is | |
324 | * mmapped by the application using the IORING_OFF_SQES offset. | |
325 | * | |
326 | * This indirection could e.g. be used to assign fixed | |
327 | * io_uring_sqe entries to operations and only submit them to | |
328 | * the queue when needed. | |
329 | * | |
330 | * The kernel modifies neither the indices array nor the entries | |
331 | * array. | |
332 | */ | |
333 | u32 *sq_array; | |
2b188cc1 JA |
334 | unsigned cached_sq_head; |
335 | unsigned sq_entries; | |
336 | unsigned sq_mask; | |
6c271ce2 | 337 | unsigned sq_thread_idle; |
498ccd9e | 338 | unsigned cached_sq_dropped; |
2c3bac6d | 339 | unsigned cached_cq_overflow; |
ad3eb2c8 | 340 | unsigned long sq_check_overflow; |
de0617e4 JA |
341 | |
342 | struct list_head defer_list; | |
5262f567 | 343 | struct list_head timeout_list; |
1d7bb1d5 | 344 | struct list_head cq_overflow_list; |
fcb323cc | 345 | |
ad3eb2c8 | 346 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
347 | } ____cacheline_aligned_in_smp; |
348 | ||
3c1a2ead JA |
349 | struct { |
350 | struct mutex uring_lock; | |
351 | wait_queue_head_t wait; | |
352 | } ____cacheline_aligned_in_smp; | |
353 | ||
354 | struct io_submit_state submit_state; | |
355 | ||
206aefde JA |
356 | struct io_rings *rings; |
357 | ||
2b188cc1 | 358 | /* IO offload */ |
561fb04a | 359 | struct io_wq *io_wq; |
2aede0e4 JA |
360 | |
361 | /* | |
362 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
363 | * have access to the ->files | |
364 | */ | |
365 | struct task_struct *sqo_task; | |
366 | ||
367 | /* Only used for accounting purposes */ | |
368 | struct mm_struct *mm_account; | |
369 | ||
91d8f519 DZ |
370 | #ifdef CONFIG_BLK_CGROUP |
371 | struct cgroup_subsys_state *sqo_blkcg_css; | |
372 | #endif | |
373 | ||
534ca6d6 JA |
374 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
375 | ||
90554200 | 376 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 377 | struct list_head sqd_list; |
75b28aff | 378 | |
6b06314c JA |
379 | /* |
380 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
381 | * readers must ensure that ->refs is alive as long as the file* is | |
382 | * used. Only updated through io_uring_register(2). | |
383 | */ | |
269bbe5f | 384 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
385 | unsigned nr_user_files; |
386 | ||
edafccee JA |
387 | /* if used, fixed mapped user buffers */ |
388 | unsigned nr_user_bufs; | |
389 | struct io_mapped_ubuf *user_bufs; | |
390 | ||
2b188cc1 JA |
391 | struct user_struct *user; |
392 | ||
0b8c0ec7 | 393 | const struct cred *creds; |
181e448d | 394 | |
4ea33a97 JA |
395 | #ifdef CONFIG_AUDIT |
396 | kuid_t loginuid; | |
397 | unsigned int sessionid; | |
398 | #endif | |
399 | ||
0f158b4c JA |
400 | struct completion ref_comp; |
401 | struct completion sq_thread_comp; | |
206aefde JA |
402 | |
403 | #if defined(CONFIG_UNIX) | |
404 | struct socket *ring_sock; | |
405 | #endif | |
406 | ||
5a2e745d JA |
407 | struct idr io_buffer_idr; |
408 | ||
071698e1 JA |
409 | struct idr personality_idr; |
410 | ||
206aefde JA |
411 | struct { |
412 | unsigned cached_cq_tail; | |
413 | unsigned cq_entries; | |
414 | unsigned cq_mask; | |
415 | atomic_t cq_timeouts; | |
f010505b | 416 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 417 | unsigned long cq_check_overflow; |
206aefde JA |
418 | struct wait_queue_head cq_wait; |
419 | struct fasync_struct *cq_fasync; | |
420 | struct eventfd_ctx *cq_ev_fd; | |
421 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 422 | |
2b188cc1 JA |
423 | struct { |
424 | spinlock_t completion_lock; | |
e94f141b | 425 | |
def596e9 | 426 | /* |
540e32a0 | 427 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
428 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
429 | * For SQPOLL, only the single threaded io_sq_thread() will | |
430 | * manipulate the list, hence no extra locking is needed there. | |
431 | */ | |
540e32a0 | 432 | struct list_head iopoll_list; |
78076bb6 JA |
433 | struct hlist_head *cancel_hash; |
434 | unsigned cancel_hash_bits; | |
e94f141b | 435 | bool poll_multi_file; |
31b51510 | 436 | |
fcb323cc JA |
437 | spinlock_t inflight_lock; |
438 | struct list_head inflight_list; | |
2b188cc1 | 439 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 440 | |
269bbe5f BM |
441 | struct delayed_work rsrc_put_work; |
442 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
443 | struct list_head rsrc_ref_list; |
444 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 445 | |
21b55dbc | 446 | struct io_restriction restrictions; |
3c1a2ead JA |
447 | |
448 | /* Keep this last, we don't need it for the fast path */ | |
449 | struct work_struct exit_work; | |
2b188cc1 JA |
450 | }; |
451 | ||
09bb8394 JA |
452 | /* |
453 | * First field must be the file pointer in all the | |
454 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
455 | */ | |
221c5eb2 JA |
456 | struct io_poll_iocb { |
457 | struct file *file; | |
018043be | 458 | struct wait_queue_head *head; |
221c5eb2 | 459 | __poll_t events; |
8c838788 | 460 | bool done; |
221c5eb2 | 461 | bool canceled; |
392edb45 | 462 | struct wait_queue_entry wait; |
221c5eb2 JA |
463 | }; |
464 | ||
018043be PB |
465 | struct io_poll_remove { |
466 | struct file *file; | |
467 | u64 addr; | |
468 | }; | |
469 | ||
b5dba59e JA |
470 | struct io_close { |
471 | struct file *file; | |
b5dba59e JA |
472 | int fd; |
473 | }; | |
474 | ||
ad8a48ac JA |
475 | struct io_timeout_data { |
476 | struct io_kiocb *req; | |
477 | struct hrtimer timer; | |
478 | struct timespec64 ts; | |
479 | enum hrtimer_mode mode; | |
480 | }; | |
481 | ||
8ed8d3c3 JA |
482 | struct io_accept { |
483 | struct file *file; | |
484 | struct sockaddr __user *addr; | |
485 | int __user *addr_len; | |
486 | int flags; | |
09952e3e | 487 | unsigned long nofile; |
8ed8d3c3 JA |
488 | }; |
489 | ||
490 | struct io_sync { | |
491 | struct file *file; | |
492 | loff_t len; | |
493 | loff_t off; | |
494 | int flags; | |
d63d1b5e | 495 | int mode; |
8ed8d3c3 JA |
496 | }; |
497 | ||
fbf23849 JA |
498 | struct io_cancel { |
499 | struct file *file; | |
500 | u64 addr; | |
501 | }; | |
502 | ||
b29472ee JA |
503 | struct io_timeout { |
504 | struct file *file; | |
bfe68a22 PB |
505 | u32 off; |
506 | u32 target_seq; | |
135fcde8 | 507 | struct list_head list; |
90cd7e42 PB |
508 | /* head of the link, used by linked timeouts only */ |
509 | struct io_kiocb *head; | |
b29472ee JA |
510 | }; |
511 | ||
0bdf7a2d PB |
512 | struct io_timeout_rem { |
513 | struct file *file; | |
514 | u64 addr; | |
9c8e11b3 PB |
515 | |
516 | /* timeout update */ | |
517 | struct timespec64 ts; | |
518 | u32 flags; | |
0bdf7a2d PB |
519 | }; |
520 | ||
9adbd45d JA |
521 | struct io_rw { |
522 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
523 | struct kiocb kiocb; | |
524 | u64 addr; | |
525 | u64 len; | |
526 | }; | |
527 | ||
3fbb51c1 JA |
528 | struct io_connect { |
529 | struct file *file; | |
530 | struct sockaddr __user *addr; | |
531 | int addr_len; | |
532 | }; | |
533 | ||
e47293fd JA |
534 | struct io_sr_msg { |
535 | struct file *file; | |
fddaface | 536 | union { |
270a5940 | 537 | struct user_msghdr __user *umsg; |
fddaface JA |
538 | void __user *buf; |
539 | }; | |
e47293fd | 540 | int msg_flags; |
bcda7baa | 541 | int bgid; |
fddaface | 542 | size_t len; |
bcda7baa | 543 | struct io_buffer *kbuf; |
e47293fd JA |
544 | }; |
545 | ||
15b71abe JA |
546 | struct io_open { |
547 | struct file *file; | |
548 | int dfd; | |
15b71abe | 549 | struct filename *filename; |
c12cedf2 | 550 | struct open_how how; |
4022e7af | 551 | unsigned long nofile; |
15b71abe JA |
552 | }; |
553 | ||
269bbe5f | 554 | struct io_rsrc_update { |
05f3fb3c JA |
555 | struct file *file; |
556 | u64 arg; | |
557 | u32 nr_args; | |
558 | u32 offset; | |
559 | }; | |
560 | ||
4840e418 JA |
561 | struct io_fadvise { |
562 | struct file *file; | |
563 | u64 offset; | |
564 | u32 len; | |
565 | u32 advice; | |
566 | }; | |
567 | ||
c1ca757b JA |
568 | struct io_madvise { |
569 | struct file *file; | |
570 | u64 addr; | |
571 | u32 len; | |
572 | u32 advice; | |
573 | }; | |
574 | ||
3e4827b0 JA |
575 | struct io_epoll { |
576 | struct file *file; | |
577 | int epfd; | |
578 | int op; | |
579 | int fd; | |
580 | struct epoll_event event; | |
e47293fd JA |
581 | }; |
582 | ||
7d67af2c PB |
583 | struct io_splice { |
584 | struct file *file_out; | |
585 | struct file *file_in; | |
586 | loff_t off_out; | |
587 | loff_t off_in; | |
588 | u64 len; | |
589 | unsigned int flags; | |
590 | }; | |
591 | ||
ddf0322d JA |
592 | struct io_provide_buf { |
593 | struct file *file; | |
594 | __u64 addr; | |
595 | __s32 len; | |
596 | __u32 bgid; | |
597 | __u16 nbufs; | |
598 | __u16 bid; | |
599 | }; | |
600 | ||
1d9e1288 BM |
601 | struct io_statx { |
602 | struct file *file; | |
603 | int dfd; | |
604 | unsigned int mask; | |
605 | unsigned int flags; | |
e62753e4 | 606 | const char __user *filename; |
1d9e1288 BM |
607 | struct statx __user *buffer; |
608 | }; | |
609 | ||
36f4fa68 JA |
610 | struct io_shutdown { |
611 | struct file *file; | |
612 | int how; | |
613 | }; | |
614 | ||
80a261fd JA |
615 | struct io_rename { |
616 | struct file *file; | |
617 | int old_dfd; | |
618 | int new_dfd; | |
619 | struct filename *oldpath; | |
620 | struct filename *newpath; | |
621 | int flags; | |
622 | }; | |
623 | ||
14a1143b JA |
624 | struct io_unlink { |
625 | struct file *file; | |
626 | int dfd; | |
627 | int flags; | |
628 | struct filename *filename; | |
629 | }; | |
630 | ||
3ca405eb PB |
631 | struct io_completion { |
632 | struct file *file; | |
633 | struct list_head list; | |
0f7e466b | 634 | int cflags; |
3ca405eb PB |
635 | }; |
636 | ||
f499a021 JA |
637 | struct io_async_connect { |
638 | struct sockaddr_storage address; | |
639 | }; | |
640 | ||
03b1230c JA |
641 | struct io_async_msghdr { |
642 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
643 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
644 | struct iovec *free_iov; | |
03b1230c JA |
645 | struct sockaddr __user *uaddr; |
646 | struct msghdr msg; | |
b537916c | 647 | struct sockaddr_storage addr; |
03b1230c JA |
648 | }; |
649 | ||
f67676d1 JA |
650 | struct io_async_rw { |
651 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
652 | const struct iovec *free_iovec; |
653 | struct iov_iter iter; | |
227c0c96 | 654 | size_t bytes_done; |
bcf5a063 | 655 | struct wait_page_queue wpq; |
f67676d1 JA |
656 | }; |
657 | ||
6b47ee6e PB |
658 | enum { |
659 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
660 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
661 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
662 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
663 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 664 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 665 | |
6b47ee6e PB |
666 | REQ_F_FAIL_LINK_BIT, |
667 | REQ_F_INFLIGHT_BIT, | |
668 | REQ_F_CUR_POS_BIT, | |
669 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 670 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 671 | REQ_F_ISREG_BIT, |
99bc4c38 | 672 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 673 | REQ_F_POLLED_BIT, |
bcda7baa | 674 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 675 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 676 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 677 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 678 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
679 | |
680 | /* not a real bit, just to check we're not overflowing the space */ | |
681 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
682 | }; |
683 | ||
684 | enum { | |
685 | /* ctx owns file */ | |
686 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
687 | /* drain existing IO first */ | |
688 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
689 | /* linked sqes */ | |
690 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
691 | /* doesn't sever on completion < 0 */ | |
692 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
693 | /* IOSQE_ASYNC */ | |
694 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
695 | /* IOSQE_BUFFER_SELECT */ |
696 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 697 | |
6b47ee6e PB |
698 | /* fail rest of links */ |
699 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
700 | /* on inflight list */ | |
701 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
702 | /* read/write uses file position */ | |
703 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
704 | /* must not punt to workers */ | |
705 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 706 | /* has or had linked timeout */ |
6b47ee6e | 707 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
708 | /* regular file */ |
709 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
710 | /* needs cleanup */ |
711 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
712 | /* already went through poll handler */ |
713 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
714 | /* buffer already selected */ |
715 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
716 | /* doesn't need file table for this request */ |
717 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
718 | /* io_wq_work is initialized */ |
719 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
720 | /* linked timeout is active, i.e. prepared by link's head */ |
721 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
722 | /* completion is deferred through io_comp_state */ |
723 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
724 | }; |
725 | ||
726 | struct async_poll { | |
727 | struct io_poll_iocb poll; | |
807abcb0 | 728 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
729 | }; |
730 | ||
7cbf1722 JA |
731 | struct io_task_work { |
732 | struct io_wq_work_node node; | |
733 | task_work_func_t func; | |
734 | }; | |
735 | ||
09bb8394 JA |
736 | /* |
737 | * NOTE! Each of the iocb union members has the file pointer | |
738 | * as the first entry in their struct definition. So you can | |
739 | * access the file pointer through any of the sub-structs, | |
740 | * or directly as just 'ki_filp' in this struct. | |
741 | */ | |
2b188cc1 | 742 | struct io_kiocb { |
221c5eb2 | 743 | union { |
09bb8394 | 744 | struct file *file; |
9adbd45d | 745 | struct io_rw rw; |
221c5eb2 | 746 | struct io_poll_iocb poll; |
018043be | 747 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
748 | struct io_accept accept; |
749 | struct io_sync sync; | |
fbf23849 | 750 | struct io_cancel cancel; |
b29472ee | 751 | struct io_timeout timeout; |
0bdf7a2d | 752 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 753 | struct io_connect connect; |
e47293fd | 754 | struct io_sr_msg sr_msg; |
15b71abe | 755 | struct io_open open; |
b5dba59e | 756 | struct io_close close; |
269bbe5f | 757 | struct io_rsrc_update rsrc_update; |
4840e418 | 758 | struct io_fadvise fadvise; |
c1ca757b | 759 | struct io_madvise madvise; |
3e4827b0 | 760 | struct io_epoll epoll; |
7d67af2c | 761 | struct io_splice splice; |
ddf0322d | 762 | struct io_provide_buf pbuf; |
1d9e1288 | 763 | struct io_statx statx; |
36f4fa68 | 764 | struct io_shutdown shutdown; |
80a261fd | 765 | struct io_rename rename; |
14a1143b | 766 | struct io_unlink unlink; |
3ca405eb PB |
767 | /* use only after cleaning per-op data, see io_clean_op() */ |
768 | struct io_completion compl; | |
221c5eb2 | 769 | }; |
2b188cc1 | 770 | |
e8c2bc1f JA |
771 | /* opcode allocated if it needs to store data for async defer */ |
772 | void *async_data; | |
d625c6ee | 773 | u8 opcode; |
65a6543d XW |
774 | /* polled IO has completed */ |
775 | u8 iopoll_completed; | |
2b188cc1 | 776 | |
4f4eeba8 | 777 | u16 buf_index; |
9cf7c104 | 778 | u32 result; |
4f4eeba8 | 779 | |
010e8e6b PB |
780 | struct io_ring_ctx *ctx; |
781 | unsigned int flags; | |
782 | refcount_t refs; | |
783 | struct task_struct *task; | |
784 | u64 user_data; | |
d7718a9d | 785 | |
f2f87370 | 786 | struct io_kiocb *link; |
269bbe5f | 787 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 788 | |
d21ffe7e PB |
789 | /* |
790 | * 1. used with ctx->iopoll_list with reads/writes | |
791 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
792 | */ | |
010e8e6b | 793 | struct list_head inflight_entry; |
7cbf1722 JA |
794 | union { |
795 | struct io_task_work io_task_work; | |
796 | struct callback_head task_work; | |
797 | }; | |
010e8e6b PB |
798 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
799 | struct hlist_node hash_node; | |
800 | struct async_poll *apoll; | |
801 | struct io_wq_work work; | |
2b188cc1 | 802 | }; |
05589553 | 803 | |
27dc8338 PB |
804 | struct io_defer_entry { |
805 | struct list_head list; | |
806 | struct io_kiocb *req; | |
9cf7c104 | 807 | u32 seq; |
2b188cc1 JA |
808 | }; |
809 | ||
d3656344 | 810 | struct io_op_def { |
d3656344 JA |
811 | /* needs req->file assigned */ |
812 | unsigned needs_file : 1; | |
d3656344 JA |
813 | /* hash wq insertion if file is a regular file */ |
814 | unsigned hash_reg_file : 1; | |
815 | /* unbound wq insertion if file is a non-regular file */ | |
816 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
817 | /* opcode is not supported by this kernel */ |
818 | unsigned not_supported : 1; | |
8a72758c JA |
819 | /* set if opcode supports polled "wait" */ |
820 | unsigned pollin : 1; | |
821 | unsigned pollout : 1; | |
bcda7baa JA |
822 | /* op supports buffer selection */ |
823 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
824 | /* must always have async data allocated */ |
825 | unsigned needs_async_data : 1; | |
27926b68 JA |
826 | /* should block plug */ |
827 | unsigned plug : 1; | |
e8c2bc1f JA |
828 | /* size of async data needed, if any */ |
829 | unsigned short async_size; | |
0f203765 | 830 | unsigned work_flags; |
d3656344 JA |
831 | }; |
832 | ||
0918682b | 833 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
834 | [IORING_OP_NOP] = {}, |
835 | [IORING_OP_READV] = { | |
d3656344 JA |
836 | .needs_file = 1, |
837 | .unbound_nonreg_file = 1, | |
8a72758c | 838 | .pollin = 1, |
4d954c25 | 839 | .buffer_select = 1, |
e8c2bc1f | 840 | .needs_async_data = 1, |
27926b68 | 841 | .plug = 1, |
e8c2bc1f | 842 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 843 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 844 | }, |
0463b6c5 | 845 | [IORING_OP_WRITEV] = { |
d3656344 JA |
846 | .needs_file = 1, |
847 | .hash_reg_file = 1, | |
848 | .unbound_nonreg_file = 1, | |
8a72758c | 849 | .pollout = 1, |
e8c2bc1f | 850 | .needs_async_data = 1, |
27926b68 | 851 | .plug = 1, |
e8c2bc1f | 852 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
853 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
854 | IO_WQ_WORK_FSIZE, | |
d3656344 | 855 | }, |
0463b6c5 | 856 | [IORING_OP_FSYNC] = { |
d3656344 | 857 | .needs_file = 1, |
0f203765 | 858 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 859 | }, |
0463b6c5 | 860 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
861 | .needs_file = 1, |
862 | .unbound_nonreg_file = 1, | |
8a72758c | 863 | .pollin = 1, |
27926b68 | 864 | .plug = 1, |
e8c2bc1f | 865 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 866 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 867 | }, |
0463b6c5 | 868 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
869 | .needs_file = 1, |
870 | .hash_reg_file = 1, | |
871 | .unbound_nonreg_file = 1, | |
8a72758c | 872 | .pollout = 1, |
27926b68 | 873 | .plug = 1, |
e8c2bc1f | 874 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
875 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
876 | IO_WQ_WORK_MM, | |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | .unbound_nonreg_file = 1, | |
881 | }, | |
0463b6c5 PB |
882 | [IORING_OP_POLL_REMOVE] = {}, |
883 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 884 | .needs_file = 1, |
0f203765 | 885 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 886 | }, |
0463b6c5 | 887 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
888 | .needs_file = 1, |
889 | .unbound_nonreg_file = 1, | |
8a72758c | 890 | .pollout = 1, |
e8c2bc1f JA |
891 | .needs_async_data = 1, |
892 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 893 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 894 | }, |
0463b6c5 | 895 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
896 | .needs_file = 1, |
897 | .unbound_nonreg_file = 1, | |
8a72758c | 898 | .pollin = 1, |
52de1fe1 | 899 | .buffer_select = 1, |
e8c2bc1f JA |
900 | .needs_async_data = 1, |
901 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 902 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 903 | }, |
0463b6c5 | 904 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
905 | .needs_async_data = 1, |
906 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 907 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 908 | }, |
9c8e11b3 PB |
909 | [IORING_OP_TIMEOUT_REMOVE] = { |
910 | /* used by timeout updates' prep() */ | |
911 | .work_flags = IO_WQ_WORK_MM, | |
912 | }, | |
0463b6c5 | 913 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
914 | .needs_file = 1, |
915 | .unbound_nonreg_file = 1, | |
8a72758c | 916 | .pollin = 1, |
0f203765 | 917 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 918 | }, |
0463b6c5 PB |
919 | [IORING_OP_ASYNC_CANCEL] = {}, |
920 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
921 | .needs_async_data = 1, |
922 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 923 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 924 | }, |
0463b6c5 | 925 | [IORING_OP_CONNECT] = { |
d3656344 JA |
926 | .needs_file = 1, |
927 | .unbound_nonreg_file = 1, | |
8a72758c | 928 | .pollout = 1, |
e8c2bc1f JA |
929 | .needs_async_data = 1, |
930 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 931 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 932 | }, |
0463b6c5 | 933 | [IORING_OP_FALLOCATE] = { |
d3656344 | 934 | .needs_file = 1, |
69228338 | 935 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 936 | }, |
0463b6c5 | 937 | [IORING_OP_OPENAT] = { |
0f203765 | 938 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
14587a46 | 939 | IO_WQ_WORK_FS | IO_WQ_WORK_MM, |
d3656344 | 940 | }, |
0463b6c5 | 941 | [IORING_OP_CLOSE] = { |
0f203765 | 942 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 943 | }, |
0463b6c5 | 944 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 945 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 946 | }, |
0463b6c5 | 947 | [IORING_OP_STATX] = { |
0f203765 JA |
948 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
949 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 950 | }, |
0463b6c5 | 951 | [IORING_OP_READ] = { |
3a6820f2 JA |
952 | .needs_file = 1, |
953 | .unbound_nonreg_file = 1, | |
8a72758c | 954 | .pollin = 1, |
bcda7baa | 955 | .buffer_select = 1, |
27926b68 | 956 | .plug = 1, |
e8c2bc1f | 957 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 958 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 959 | }, |
0463b6c5 | 960 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
961 | .needs_file = 1, |
962 | .unbound_nonreg_file = 1, | |
8a72758c | 963 | .pollout = 1, |
27926b68 | 964 | .plug = 1, |
e8c2bc1f | 965 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
966 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
967 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 968 | }, |
0463b6c5 | 969 | [IORING_OP_FADVISE] = { |
4840e418 | 970 | .needs_file = 1, |
0f203765 | 971 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 972 | }, |
0463b6c5 | 973 | [IORING_OP_MADVISE] = { |
0f203765 | 974 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 975 | }, |
0463b6c5 | 976 | [IORING_OP_SEND] = { |
fddaface JA |
977 | .needs_file = 1, |
978 | .unbound_nonreg_file = 1, | |
8a72758c | 979 | .pollout = 1, |
0f203765 | 980 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 981 | }, |
0463b6c5 | 982 | [IORING_OP_RECV] = { |
fddaface JA |
983 | .needs_file = 1, |
984 | .unbound_nonreg_file = 1, | |
8a72758c | 985 | .pollin = 1, |
bcda7baa | 986 | .buffer_select = 1, |
0f203765 | 987 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 988 | }, |
0463b6c5 | 989 | [IORING_OP_OPENAT2] = { |
0f203765 | 990 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
14587a46 | 991 | IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
cebdb986 | 992 | }, |
3e4827b0 JA |
993 | [IORING_OP_EPOLL_CTL] = { |
994 | .unbound_nonreg_file = 1, | |
0f203765 | 995 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 996 | }, |
7d67af2c PB |
997 | [IORING_OP_SPLICE] = { |
998 | .needs_file = 1, | |
999 | .hash_reg_file = 1, | |
1000 | .unbound_nonreg_file = 1, | |
0f203765 | 1001 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
1002 | }, |
1003 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1004 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1005 | [IORING_OP_TEE] = { |
1006 | .needs_file = 1, | |
1007 | .hash_reg_file = 1, | |
1008 | .unbound_nonreg_file = 1, | |
1009 | }, | |
36f4fa68 JA |
1010 | [IORING_OP_SHUTDOWN] = { |
1011 | .needs_file = 1, | |
1012 | }, | |
80a261fd JA |
1013 | [IORING_OP_RENAMEAT] = { |
1014 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1015 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1016 | }, | |
14a1143b JA |
1017 | [IORING_OP_UNLINKAT] = { |
1018 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1019 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1020 | }, | |
d3656344 JA |
1021 | }; |
1022 | ||
9936c7c2 PB |
1023 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1024 | struct task_struct *task, | |
1025 | struct files_struct *files); | |
269bbe5f | 1026 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1027 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1028 | struct io_ring_ctx *ctx); |
bc9744cd PB |
1029 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
1030 | struct fixed_rsrc_ref_node *ref_node); | |
1ffc5422 | 1031 | |
23faba36 | 1032 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1033 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1034 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1035 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1036 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1037 | static void io_dismantle_req(struct io_kiocb *req); |
1038 | static void io_put_task(struct task_struct *task, int nr); | |
1039 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1040 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1041 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1042 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1043 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1044 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1045 | unsigned nr_args); |
3ca405eb | 1046 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1047 | static struct file *io_file_get(struct io_submit_state *state, |
1048 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1049 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1050 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1051 | |
847595de PB |
1052 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1053 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1054 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1055 | const struct iovec *fast_iov, | |
227c0c96 | 1056 | struct iov_iter *iter, bool force); |
907d1df3 | 1057 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1058 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1059 | struct io_ring_ctx *ctx); | |
de0617e4 | 1060 | |
2b188cc1 JA |
1061 | static struct kmem_cache *req_cachep; |
1062 | ||
0918682b | 1063 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1064 | |
1065 | struct sock *io_uring_get_socket(struct file *file) | |
1066 | { | |
1067 | #if defined(CONFIG_UNIX) | |
1068 | if (file->f_op == &io_uring_fops) { | |
1069 | struct io_ring_ctx *ctx = file->private_data; | |
1070 | ||
1071 | return ctx->ring_sock->sk; | |
1072 | } | |
1073 | #endif | |
1074 | return NULL; | |
1075 | } | |
1076 | EXPORT_SYMBOL(io_uring_get_socket); | |
1077 | ||
f2f87370 PB |
1078 | #define io_for_each_link(pos, head) \ |
1079 | for (pos = (head); pos; pos = pos->link) | |
1080 | ||
3ca405eb PB |
1081 | static inline void io_clean_op(struct io_kiocb *req) |
1082 | { | |
9d5c8190 | 1083 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1084 | __io_clean_op(req); |
1085 | } | |
1086 | ||
36f72fe2 PB |
1087 | static inline void io_set_resource_node(struct io_kiocb *req) |
1088 | { | |
1089 | struct io_ring_ctx *ctx = req->ctx; | |
1090 | ||
269bbe5f BM |
1091 | if (!req->fixed_rsrc_refs) { |
1092 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1093 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1094 | } |
1095 | } | |
1096 | ||
08d23634 PB |
1097 | static bool io_match_task(struct io_kiocb *head, |
1098 | struct task_struct *task, | |
1099 | struct files_struct *files) | |
1100 | { | |
1101 | struct io_kiocb *req; | |
1102 | ||
84965ff8 JA |
1103 | if (task && head->task != task) { |
1104 | /* in terms of cancelation, always match if req task is dead */ | |
1105 | if (head->task->flags & PF_EXITING) | |
1106 | return true; | |
08d23634 | 1107 | return false; |
84965ff8 | 1108 | } |
08d23634 PB |
1109 | if (!files) |
1110 | return true; | |
1111 | ||
1112 | io_for_each_link(req, head) { | |
02a13674 JA |
1113 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1114 | continue; | |
1115 | if (req->file && req->file->f_op == &io_uring_fops) | |
1116 | return true; | |
1117 | if ((req->work.flags & IO_WQ_WORK_FILES) && | |
08d23634 PB |
1118 | req->work.identity->files == files) |
1119 | return true; | |
1120 | } | |
1121 | return false; | |
1122 | } | |
1123 | ||
28cea78a | 1124 | static void io_sq_thread_drop_mm_files(void) |
c40f6379 | 1125 | { |
28cea78a | 1126 | struct files_struct *files = current->files; |
c40f6379 JA |
1127 | struct mm_struct *mm = current->mm; |
1128 | ||
1129 | if (mm) { | |
1130 | kthread_unuse_mm(mm); | |
1131 | mmput(mm); | |
4b70cf9d | 1132 | current->mm = NULL; |
c40f6379 | 1133 | } |
28cea78a JA |
1134 | if (files) { |
1135 | struct nsproxy *nsproxy = current->nsproxy; | |
1136 | ||
1137 | task_lock(current); | |
1138 | current->files = NULL; | |
1139 | current->nsproxy = NULL; | |
1140 | task_unlock(current); | |
1141 | put_files_struct(files); | |
1142 | put_nsproxy(nsproxy); | |
1143 | } | |
1144 | } | |
1145 | ||
1a38ffc9 | 1146 | static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx) |
28cea78a JA |
1147 | { |
1148 | if (!current->files) { | |
1149 | struct files_struct *files; | |
1150 | struct nsproxy *nsproxy; | |
1151 | ||
1152 | task_lock(ctx->sqo_task); | |
1153 | files = ctx->sqo_task->files; | |
1154 | if (!files) { | |
1155 | task_unlock(ctx->sqo_task); | |
1a38ffc9 | 1156 | return -EOWNERDEAD; |
28cea78a JA |
1157 | } |
1158 | atomic_inc(&files->count); | |
1159 | get_nsproxy(ctx->sqo_task->nsproxy); | |
1160 | nsproxy = ctx->sqo_task->nsproxy; | |
1161 | task_unlock(ctx->sqo_task); | |
1162 | ||
1163 | task_lock(current); | |
1164 | current->files = files; | |
1165 | current->nsproxy = nsproxy; | |
1166 | task_unlock(current); | |
1167 | } | |
1a38ffc9 | 1168 | return 0; |
c40f6379 JA |
1169 | } |
1170 | ||
1171 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1172 | { | |
4b70cf9d JA |
1173 | struct mm_struct *mm; |
1174 | ||
1175 | if (current->mm) | |
1176 | return 0; | |
1177 | ||
4b70cf9d JA |
1178 | task_lock(ctx->sqo_task); |
1179 | mm = ctx->sqo_task->mm; | |
1180 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1181 | mm = NULL; | |
1182 | task_unlock(ctx->sqo_task); | |
1183 | ||
1184 | if (mm) { | |
1185 | kthread_use_mm(mm); | |
1186 | return 0; | |
c40f6379 JA |
1187 | } |
1188 | ||
4b70cf9d | 1189 | return -EFAULT; |
c40f6379 JA |
1190 | } |
1191 | ||
4e326358 PB |
1192 | static int __io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1193 | struct io_kiocb *req) | |
c40f6379 | 1194 | { |
28cea78a | 1195 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1a38ffc9 | 1196 | int ret; |
28cea78a JA |
1197 | |
1198 | if (def->work_flags & IO_WQ_WORK_MM) { | |
1a38ffc9 | 1199 | ret = __io_sq_thread_acquire_mm(ctx); |
28cea78a JA |
1200 | if (unlikely(ret)) |
1201 | return ret; | |
1202 | } | |
1203 | ||
1a38ffc9 PB |
1204 | if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) { |
1205 | ret = __io_sq_thread_acquire_files(ctx); | |
1206 | if (unlikely(ret)) | |
1207 | return ret; | |
1208 | } | |
28cea78a JA |
1209 | |
1210 | return 0; | |
c40f6379 JA |
1211 | } |
1212 | ||
4e326358 PB |
1213 | static inline int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1214 | struct io_kiocb *req) | |
1215 | { | |
4e326358 PB |
1216 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
1217 | return 0; | |
1218 | return __io_sq_thread_acquire_mm_files(ctx, req); | |
1219 | } | |
1220 | ||
91d8f519 DZ |
1221 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1222 | struct cgroup_subsys_state **cur_css) | |
1223 | ||
1224 | { | |
1225 | #ifdef CONFIG_BLK_CGROUP | |
1226 | /* puts the old one when swapping */ | |
1227 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1228 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1229 | *cur_css = ctx->sqo_blkcg_css; | |
1230 | } | |
1231 | #endif | |
1232 | } | |
1233 | ||
1234 | static void io_sq_thread_unassociate_blkcg(void) | |
1235 | { | |
1236 | #ifdef CONFIG_BLK_CGROUP | |
1237 | kthread_associate_blkcg(NULL); | |
1238 | #endif | |
1239 | } | |
1240 | ||
c40f6379 JA |
1241 | static inline void req_set_fail_links(struct io_kiocb *req) |
1242 | { | |
1243 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1244 | req->flags |= REQ_F_FAIL_LINK; | |
1245 | } | |
4a38aed2 | 1246 | |
1e6fa521 JA |
1247 | /* |
1248 | * None of these are dereferenced, they are simply used to check if any of | |
1249 | * them have changed. If we're under current and check they are still the | |
1250 | * same, we're fine to grab references to them for actual out-of-line use. | |
1251 | */ | |
1252 | static void io_init_identity(struct io_identity *id) | |
1253 | { | |
1254 | id->files = current->files; | |
1255 | id->mm = current->mm; | |
1256 | #ifdef CONFIG_BLK_CGROUP | |
1257 | rcu_read_lock(); | |
1258 | id->blkcg_css = blkcg_css(); | |
1259 | rcu_read_unlock(); | |
1260 | #endif | |
1261 | id->creds = current_cred(); | |
1262 | id->nsproxy = current->nsproxy; | |
1263 | id->fs = current->fs; | |
1264 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1265 | #ifdef CONFIG_AUDIT |
1266 | id->loginuid = current->loginuid; | |
1267 | id->sessionid = current->sessionid; | |
1268 | #endif | |
1e6fa521 JA |
1269 | refcount_set(&id->count, 1); |
1270 | } | |
1271 | ||
ec99ca6c PB |
1272 | static inline void __io_req_init_async(struct io_kiocb *req) |
1273 | { | |
1274 | memset(&req->work, 0, sizeof(req->work)); | |
1275 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1276 | } | |
1277 | ||
7cdaf587 XW |
1278 | /* |
1279 | * Note: must call io_req_init_async() for the first time you | |
1280 | * touch any members of io_wq_work. | |
1281 | */ | |
1282 | static inline void io_req_init_async(struct io_kiocb *req) | |
1283 | { | |
500a373d JA |
1284 | struct io_uring_task *tctx = current->io_uring; |
1285 | ||
7cdaf587 XW |
1286 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1287 | return; | |
1288 | ||
ec99ca6c | 1289 | __io_req_init_async(req); |
500a373d JA |
1290 | |
1291 | /* Grab a ref if this isn't our static identity */ | |
1292 | req->work.identity = tctx->identity; | |
1293 | if (tctx->identity != &tctx->__identity) | |
1294 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1295 | } |
1296 | ||
2b188cc1 JA |
1297 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1298 | { | |
1299 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1300 | ||
0f158b4c | 1301 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1302 | } |
1303 | ||
8eb7e2d0 PB |
1304 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1305 | { | |
1306 | return !req->timeout.off; | |
1307 | } | |
1308 | ||
2b188cc1 JA |
1309 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1310 | { | |
1311 | struct io_ring_ctx *ctx; | |
78076bb6 | 1312 | int hash_bits; |
2b188cc1 JA |
1313 | |
1314 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1315 | if (!ctx) | |
1316 | return NULL; | |
1317 | ||
78076bb6 JA |
1318 | /* |
1319 | * Use 5 bits less than the max cq entries, that should give us around | |
1320 | * 32 entries per hash list if totally full and uniformly spread. | |
1321 | */ | |
1322 | hash_bits = ilog2(p->cq_entries); | |
1323 | hash_bits -= 5; | |
1324 | if (hash_bits <= 0) | |
1325 | hash_bits = 1; | |
1326 | ctx->cancel_hash_bits = hash_bits; | |
1327 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1328 | GFP_KERNEL); | |
1329 | if (!ctx->cancel_hash) | |
1330 | goto err; | |
1331 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1332 | ||
21482896 | 1333 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1334 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1335 | goto err; | |
2b188cc1 JA |
1336 | |
1337 | ctx->flags = p->flags; | |
90554200 | 1338 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1339 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1340 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1341 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1342 | init_completion(&ctx->ref_comp); |
1343 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1344 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1345 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1346 | mutex_init(&ctx->uring_lock); |
1347 | init_waitqueue_head(&ctx->wait); | |
1348 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1349 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1350 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1351 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1352 | spin_lock_init(&ctx->inflight_lock); |
1353 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1354 | spin_lock_init(&ctx->rsrc_ref_lock); |
1355 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1356 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1357 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1358 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1359 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1360 | return ctx; |
206aefde | 1361 | err: |
78076bb6 | 1362 | kfree(ctx->cancel_hash); |
206aefde JA |
1363 | kfree(ctx); |
1364 | return NULL; | |
2b188cc1 JA |
1365 | } |
1366 | ||
9cf7c104 | 1367 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1368 | { |
2bc9930e JA |
1369 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1370 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1371 | |
9cf7c104 | 1372 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1373 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1374 | } |
de0617e4 | 1375 | |
9d858b21 | 1376 | return false; |
de0617e4 JA |
1377 | } |
1378 | ||
5c3462cf | 1379 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1380 | { |
500a373d | 1381 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1382 | return; |
1383 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1384 | kfree(req->work.identity); | |
1385 | } | |
1386 | ||
4edf20f9 | 1387 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1388 | { |
7cdaf587 | 1389 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1390 | return; |
51a4cc11 | 1391 | |
e86d0047 | 1392 | if (req->work.flags & IO_WQ_WORK_MM) |
98447d65 | 1393 | mmdrop(req->work.identity->mm); |
91d8f519 | 1394 | #ifdef CONFIG_BLK_CGROUP |
e86d0047 | 1395 | if (req->work.flags & IO_WQ_WORK_BLKCG) |
98447d65 | 1396 | css_put(req->work.identity->blkcg_css); |
91d8f519 | 1397 | #endif |
e86d0047 | 1398 | if (req->work.flags & IO_WQ_WORK_CREDS) |
98447d65 | 1399 | put_cred(req->work.identity->creds); |
dfead8a8 | 1400 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1401 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1402 | |
98447d65 | 1403 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1404 | if (--fs->users) |
1405 | fs = NULL; | |
98447d65 | 1406 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1407 | if (fs) |
1408 | free_fs_struct(fs); | |
1409 | } | |
34e08fed PB |
1410 | if (req->work.flags & IO_WQ_WORK_FILES) { |
1411 | put_files_struct(req->work.identity->files); | |
1412 | put_nsproxy(req->work.identity->nsproxy); | |
34e08fed PB |
1413 | } |
1414 | if (req->flags & REQ_F_INFLIGHT) { | |
1415 | struct io_ring_ctx *ctx = req->ctx; | |
1416 | struct io_uring_task *tctx = req->task->io_uring; | |
1417 | unsigned long flags; | |
1418 | ||
1419 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1420 | list_del(&req->inflight_entry); | |
1421 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1422 | req->flags &= ~REQ_F_INFLIGHT; | |
1423 | if (atomic_read(&tctx->in_idle)) | |
1424 | wake_up(&tctx->wait); | |
1425 | } | |
51a4cc11 | 1426 | |
e86d0047 PB |
1427 | req->flags &= ~REQ_F_WORK_INITIALIZED; |
1428 | req->work.flags &= ~(IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | IO_WQ_WORK_FS | | |
1429 | IO_WQ_WORK_CREDS | IO_WQ_WORK_FILES); | |
5c3462cf | 1430 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1431 | } |
1432 | ||
1e6fa521 JA |
1433 | /* |
1434 | * Create a private copy of io_identity, since some fields don't match | |
1435 | * the current context. | |
1436 | */ | |
1437 | static bool io_identity_cow(struct io_kiocb *req) | |
1438 | { | |
5c3462cf | 1439 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1440 | const struct cred *creds = NULL; |
1441 | struct io_identity *id; | |
1442 | ||
1443 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1444 | creds = req->work.identity->creds; | |
1445 | ||
1446 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1447 | if (unlikely(!id)) { | |
1448 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1449 | return false; | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * We can safely just re-init the creds we copied Either the field | |
1454 | * matches the current one, or we haven't grabbed it yet. The only | |
1455 | * exception is ->creds, through registered personalities, so handle | |
1456 | * that one separately. | |
1457 | */ | |
1458 | io_init_identity(id); | |
1459 | if (creds) | |
e8c954df | 1460 | id->creds = creds; |
1e6fa521 JA |
1461 | |
1462 | /* add one for this request */ | |
1463 | refcount_inc(&id->count); | |
1464 | ||
cb8a8ae3 JA |
1465 | /* drop tctx and req identity references, if needed */ |
1466 | if (tctx->identity != &tctx->__identity && | |
1467 | refcount_dec_and_test(&tctx->identity->count)) | |
1468 | kfree(tctx->identity); | |
1469 | if (req->work.identity != &tctx->__identity && | |
1470 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1471 | kfree(req->work.identity); |
1472 | ||
1473 | req->work.identity = id; | |
500a373d | 1474 | tctx->identity = id; |
1e6fa521 JA |
1475 | return true; |
1476 | } | |
1477 | ||
ce3d5aae PB |
1478 | static void io_req_track_inflight(struct io_kiocb *req) |
1479 | { | |
1480 | struct io_ring_ctx *ctx = req->ctx; | |
1481 | ||
1482 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
1483 | io_req_init_async(req); | |
1484 | req->flags |= REQ_F_INFLIGHT; | |
1485 | ||
1486 | spin_lock_irq(&ctx->inflight_lock); | |
1487 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1488 | spin_unlock_irq(&ctx->inflight_lock); | |
1489 | } | |
1490 | } | |
1491 | ||
1e6fa521 | 1492 | static bool io_grab_identity(struct io_kiocb *req) |
18d9be1a | 1493 | { |
d3656344 | 1494 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1495 | struct io_identity *id = req->work.identity; |
54a91f3b | 1496 | |
69228338 JA |
1497 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1498 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1499 | return false; | |
1500 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1501 | } | |
91d8f519 | 1502 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1503 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1504 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1505 | rcu_read_lock(); |
1e6fa521 JA |
1506 | if (id->blkcg_css != blkcg_css()) { |
1507 | rcu_read_unlock(); | |
1508 | return false; | |
1509 | } | |
91d8f519 DZ |
1510 | /* |
1511 | * This should be rare, either the cgroup is dying or the task | |
1512 | * is moving cgroups. Just punt to root for the handful of ios. | |
1513 | */ | |
1e6fa521 | 1514 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1515 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1516 | rcu_read_unlock(); |
1517 | } | |
1518 | #endif | |
dfead8a8 | 1519 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1520 | if (id->creds != current_cred()) |
1521 | return false; | |
1522 | get_cred(id->creds); | |
dfead8a8 JA |
1523 | req->work.flags |= IO_WQ_WORK_CREDS; |
1524 | } | |
4ea33a97 JA |
1525 | #ifdef CONFIG_AUDIT |
1526 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1527 | current->sessionid != id->sessionid) | |
1528 | return false; | |
1529 | #endif | |
dfead8a8 JA |
1530 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1531 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1532 | if (current->fs != id->fs) |
1533 | return false; | |
1534 | spin_lock(&id->fs->lock); | |
1535 | if (!id->fs->in_exec) { | |
1536 | id->fs->users++; | |
dfead8a8 | 1537 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1538 | } else { |
1539 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1540 | } | |
1541 | spin_unlock(¤t->fs->lock); | |
1542 | } | |
af604703 PB |
1543 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1544 | (def->work_flags & IO_WQ_WORK_FILES) && | |
1545 | !(req->flags & REQ_F_NO_FILE_TABLE)) { | |
1546 | if (id->files != current->files || | |
1547 | id->nsproxy != current->nsproxy) | |
1548 | return false; | |
1549 | atomic_inc(&id->files->count); | |
1550 | get_nsproxy(id->nsproxy); | |
af604703 | 1551 | req->work.flags |= IO_WQ_WORK_FILES; |
ce3d5aae | 1552 | io_req_track_inflight(req); |
af604703 | 1553 | } |
77788775 JA |
1554 | if (!(req->work.flags & IO_WQ_WORK_MM) && |
1555 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1556 | if (id->mm != current->mm) | |
1557 | return false; | |
1558 | mmgrab(id->mm); | |
1559 | req->work.flags |= IO_WQ_WORK_MM; | |
1560 | } | |
1e6fa521 JA |
1561 | |
1562 | return true; | |
1563 | } | |
1564 | ||
1565 | static void io_prep_async_work(struct io_kiocb *req) | |
1566 | { | |
1567 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1568 | struct io_ring_ctx *ctx = req->ctx; |
1569 | ||
1570 | io_req_init_async(req); | |
1571 | ||
feaadc4f PB |
1572 | if (req->flags & REQ_F_FORCE_ASYNC) |
1573 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1574 | ||
1e6fa521 JA |
1575 | if (req->flags & REQ_F_ISREG) { |
1576 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1577 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1578 | } else { | |
1579 | if (def->unbound_nonreg_file) | |
1580 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1581 | } | |
1582 | ||
1e6fa521 JA |
1583 | /* if we fail grabbing identity, we must COW, regrab, and retry */ |
1584 | if (io_grab_identity(req)) | |
1585 | return; | |
1586 | ||
1587 | if (!io_identity_cow(req)) | |
1588 | return; | |
1589 | ||
1590 | /* can't fail at this point */ | |
1591 | if (!io_grab_identity(req)) | |
1592 | WARN_ON(1); | |
561fb04a | 1593 | } |
cccf0ee8 | 1594 | |
cbdcb435 | 1595 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1596 | { |
cbdcb435 | 1597 | struct io_kiocb *cur; |
54a91f3b | 1598 | |
f2f87370 PB |
1599 | io_for_each_link(cur, req) |
1600 | io_prep_async_work(cur); | |
561fb04a JA |
1601 | } |
1602 | ||
7271ef3a | 1603 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1604 | { |
a197f664 | 1605 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1606 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1607 | |
8766dd51 PB |
1608 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1609 | &req->work, req->flags); | |
1610 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1611 | return link; |
18d9be1a JA |
1612 | } |
1613 | ||
cbdcb435 PB |
1614 | static void io_queue_async_work(struct io_kiocb *req) |
1615 | { | |
7271ef3a JA |
1616 | struct io_kiocb *link; |
1617 | ||
cbdcb435 PB |
1618 | /* init ->work of the whole link before punting */ |
1619 | io_prep_async_link(req); | |
7271ef3a JA |
1620 | link = __io_queue_async_work(req); |
1621 | ||
1622 | if (link) | |
1623 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1624 | } |
1625 | ||
5262f567 JA |
1626 | static void io_kill_timeout(struct io_kiocb *req) |
1627 | { | |
e8c2bc1f | 1628 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1629 | int ret; |
1630 | ||
e8c2bc1f | 1631 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1632 | if (ret != -1) { |
01cec8c1 PB |
1633 | atomic_set(&req->ctx->cq_timeouts, |
1634 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1635 | list_del_init(&req->timeout.list); |
78e19bbe | 1636 | io_cqring_fill_event(req, 0); |
216578e5 | 1637 | io_put_req_deferred(req, 1); |
5262f567 JA |
1638 | } |
1639 | } | |
1640 | ||
76e1b642 JA |
1641 | /* |
1642 | * Returns true if we found and killed one or more timeouts | |
1643 | */ | |
6b81928d PB |
1644 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1645 | struct files_struct *files) | |
5262f567 JA |
1646 | { |
1647 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1648 | int canceled = 0; |
5262f567 JA |
1649 | |
1650 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1651 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1652 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1653 | io_kill_timeout(req); |
76e1b642 JA |
1654 | canceled++; |
1655 | } | |
f3606e3a | 1656 | } |
5262f567 | 1657 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1658 | return canceled != 0; |
5262f567 JA |
1659 | } |
1660 | ||
04518945 | 1661 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1662 | { |
04518945 | 1663 | do { |
27dc8338 PB |
1664 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1665 | struct io_defer_entry, list); | |
de0617e4 | 1666 | |
9cf7c104 | 1667 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1668 | break; |
27dc8338 | 1669 | list_del_init(&de->list); |
907d1df3 | 1670 | io_req_task_queue(de->req); |
27dc8338 | 1671 | kfree(de); |
04518945 PB |
1672 | } while (!list_empty(&ctx->defer_list)); |
1673 | } | |
1674 | ||
360428f8 | 1675 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1676 | { |
f010505b MDG |
1677 | u32 seq; |
1678 | ||
1679 | if (list_empty(&ctx->timeout_list)) | |
1680 | return; | |
1681 | ||
1682 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1683 | ||
1684 | do { | |
1685 | u32 events_needed, events_got; | |
360428f8 | 1686 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1687 | struct io_kiocb, timeout.list); |
de0617e4 | 1688 | |
8eb7e2d0 | 1689 | if (io_is_timeout_noseq(req)) |
360428f8 | 1690 | break; |
f010505b MDG |
1691 | |
1692 | /* | |
1693 | * Since seq can easily wrap around over time, subtract | |
1694 | * the last seq at which timeouts were flushed before comparing. | |
1695 | * Assuming not more than 2^31-1 events have happened since, | |
1696 | * these subtractions won't have wrapped, so we can check if | |
1697 | * target is in [last_seq, current_seq] by comparing the two. | |
1698 | */ | |
1699 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1700 | events_got = seq - ctx->cq_last_tm_flush; | |
1701 | if (events_got < events_needed) | |
360428f8 | 1702 | break; |
bfe68a22 | 1703 | |
135fcde8 | 1704 | list_del_init(&req->timeout.list); |
5262f567 | 1705 | io_kill_timeout(req); |
f010505b MDG |
1706 | } while (!list_empty(&ctx->timeout_list)); |
1707 | ||
1708 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1709 | } |
5262f567 | 1710 | |
360428f8 PB |
1711 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1712 | { | |
1713 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1714 | |
1715 | /* order cqe stores with ring update */ | |
1716 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1717 | |
04518945 PB |
1718 | if (unlikely(!list_empty(&ctx->defer_list))) |
1719 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1720 | } |
1721 | ||
90554200 JA |
1722 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1723 | { | |
1724 | struct io_rings *r = ctx->rings; | |
1725 | ||
1726 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1727 | } | |
1728 | ||
888aae2e PB |
1729 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1730 | { | |
1731 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1732 | } | |
1733 | ||
2b188cc1 JA |
1734 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1735 | { | |
75b28aff | 1736 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1737 | unsigned tail; |
1738 | ||
115e12e5 SB |
1739 | /* |
1740 | * writes to the cq entry need to come after reading head; the | |
1741 | * control dependency is enough as we're using WRITE_ONCE to | |
1742 | * fill the cq entry | |
1743 | */ | |
888aae2e | 1744 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1745 | return NULL; |
1746 | ||
888aae2e | 1747 | tail = ctx->cached_cq_tail++; |
75b28aff | 1748 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1749 | } |
1750 | ||
f2842ab5 JA |
1751 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1752 | { | |
f0b493e6 JA |
1753 | if (!ctx->cq_ev_fd) |
1754 | return false; | |
7e55a19c SG |
1755 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1756 | return false; | |
f2842ab5 JA |
1757 | if (!ctx->eventfd_async) |
1758 | return true; | |
b41e9852 | 1759 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1760 | } |
1761 | ||
b41e9852 | 1762 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1763 | { |
b1445e59 PB |
1764 | /* see waitqueue_active() comment */ |
1765 | smp_mb(); | |
1766 | ||
1d7bb1d5 JA |
1767 | if (waitqueue_active(&ctx->wait)) |
1768 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1769 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1770 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1771 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1772 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1773 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1774 | wake_up_interruptible(&ctx->cq_wait); |
1775 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1776 | } | |
1d7bb1d5 JA |
1777 | } |
1778 | ||
80c18e4a PB |
1779 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1780 | { | |
b1445e59 PB |
1781 | /* see waitqueue_active() comment */ |
1782 | smp_mb(); | |
1783 | ||
80c18e4a PB |
1784 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1785 | if (waitqueue_active(&ctx->wait)) | |
1786 | wake_up(&ctx->wait); | |
1787 | } | |
1788 | if (io_should_trigger_evfd(ctx)) | |
1789 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1790 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1791 | wake_up_interruptible(&ctx->cq_wait); |
1792 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1793 | } | |
80c18e4a PB |
1794 | } |
1795 | ||
c4a2ed72 | 1796 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1797 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1798 | struct task_struct *tsk, | |
1799 | struct files_struct *files) | |
1d7bb1d5 JA |
1800 | { |
1801 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1802 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1803 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1804 | unsigned long flags; |
b18032bb | 1805 | bool all_flushed, posted; |
1d7bb1d5 JA |
1806 | LIST_HEAD(list); |
1807 | ||
e23de15f PB |
1808 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1809 | return false; | |
1d7bb1d5 | 1810 | |
b18032bb | 1811 | posted = false; |
1d7bb1d5 | 1812 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1813 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1814 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1815 | continue; |
1816 | ||
1d7bb1d5 JA |
1817 | cqe = io_get_cqring(ctx); |
1818 | if (!cqe && !force) | |
1819 | break; | |
1820 | ||
40d8ddd4 | 1821 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1822 | if (cqe) { |
1823 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1824 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1825 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1826 | } else { |
2c3bac6d | 1827 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1828 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1829 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1830 | } |
b18032bb | 1831 | posted = true; |
1d7bb1d5 JA |
1832 | } |
1833 | ||
09e88404 PB |
1834 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1835 | if (all_flushed) { | |
1836 | clear_bit(0, &ctx->sq_check_overflow); | |
1837 | clear_bit(0, &ctx->cq_check_overflow); | |
1838 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1839 | } | |
46930143 | 1840 | |
b18032bb JA |
1841 | if (posted) |
1842 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1843 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1844 | if (posted) |
1845 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1846 | |
1847 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1848 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1849 | list_del(&req->compl.list); | |
ec9c02ad | 1850 | io_put_req(req); |
1d7bb1d5 | 1851 | } |
c4a2ed72 | 1852 | |
09e88404 | 1853 | return all_flushed; |
1d7bb1d5 JA |
1854 | } |
1855 | ||
6c503150 PB |
1856 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1857 | struct task_struct *tsk, | |
1858 | struct files_struct *files) | |
1859 | { | |
1860 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1861 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1862 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1863 | mutex_lock(&ctx->uring_lock); | |
1864 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1865 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1866 | mutex_unlock(&ctx->uring_lock); | |
1867 | } | |
1868 | } | |
1869 | ||
bcda7baa | 1870 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1871 | { |
78e19bbe | 1872 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1873 | struct io_uring_cqe *cqe; |
1874 | ||
78e19bbe | 1875 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1876 | |
2b188cc1 JA |
1877 | /* |
1878 | * If we can't get a cq entry, userspace overflowed the | |
1879 | * submission (by quite a lot). Increment the overflow count in | |
1880 | * the ring. | |
1881 | */ | |
1882 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1883 | if (likely(cqe)) { |
78e19bbe | 1884 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1885 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1886 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1887 | } else if (ctx->cq_overflow_flushed || |
1888 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1889 | /* |
1890 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1891 | * then we cannot store the request for later flushing, we need | |
1892 | * to drop it on the floor. | |
1893 | */ | |
2c3bac6d PB |
1894 | ctx->cached_cq_overflow++; |
1895 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1896 | } else { |
ad3eb2c8 JA |
1897 | if (list_empty(&ctx->cq_overflow_list)) { |
1898 | set_bit(0, &ctx->sq_check_overflow); | |
1899 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1900 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1901 | } |
40d8ddd4 | 1902 | io_clean_op(req); |
1d7bb1d5 | 1903 | req->result = res; |
0f7e466b | 1904 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1905 | refcount_inc(&req->refs); |
1906 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1907 | } |
1908 | } | |
1909 | ||
bcda7baa JA |
1910 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1911 | { | |
1912 | __io_cqring_fill_event(req, res, 0); | |
1913 | } | |
1914 | ||
c7dae4ba JA |
1915 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1916 | unsigned int cflags) | |
2b188cc1 | 1917 | { |
78e19bbe | 1918 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1919 | unsigned long flags; |
1920 | ||
1921 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1922 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1923 | io_commit_cqring(ctx); |
c7dae4ba JA |
1924 | /* |
1925 | * If we're the last reference to this request, add to our locked | |
1926 | * free_list cache. | |
1927 | */ | |
1928 | if (refcount_dec_and_test(&req->refs)) { | |
1929 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1930 | ||
1931 | io_dismantle_req(req); | |
1932 | io_put_task(req->task, 1); | |
1933 | list_add(&req->compl.list, &cs->locked_free_list); | |
1934 | cs->locked_free_nr++; | |
1935 | } else | |
1936 | req = NULL; | |
2b188cc1 JA |
1937 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1938 | ||
8c838788 | 1939 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1940 | if (req) { |
1941 | io_queue_next(req); | |
1942 | percpu_ref_put(&ctx->refs); | |
1943 | } | |
2b188cc1 JA |
1944 | } |
1945 | ||
a38d68db | 1946 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1947 | unsigned int cflags) |
229a7b63 | 1948 | { |
a38d68db PB |
1949 | io_clean_op(req); |
1950 | req->result = res; | |
1951 | req->compl.cflags = cflags; | |
e342c807 | 1952 | req->flags |= REQ_F_COMPLETE_INLINE; |
a38d68db PB |
1953 | } |
1954 | ||
889fca73 PB |
1955 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1956 | long res, unsigned cflags) | |
a38d68db | 1957 | { |
889fca73 PB |
1958 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1959 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1960 | else |
c7dae4ba | 1961 | io_req_complete_post(req, res, cflags); |
e1e16097 JA |
1962 | } |
1963 | ||
a38d68db | 1964 | static inline void io_req_complete(struct io_kiocb *req, long res) |
bcda7baa | 1965 | { |
889fca73 | 1966 | __io_req_complete(req, 0, res, 0); |
bcda7baa JA |
1967 | } |
1968 | ||
c7dae4ba | 1969 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
2b188cc1 | 1970 | { |
c7dae4ba JA |
1971 | struct io_submit_state *state = &ctx->submit_state; |
1972 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1973 | struct io_kiocb *req = NULL; |
1b4c351f | 1974 | |
c7dae4ba JA |
1975 | /* |
1976 | * If we have more than a batch's worth of requests in our IRQ side | |
1977 | * locked cache, grab the lock and move them over to our submission | |
1978 | * side cache. | |
1979 | */ | |
1980 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1981 | spin_lock_irq(&ctx->completion_lock); | |
1982 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1983 | cs->locked_free_nr = 0; | |
1984 | spin_unlock_irq(&ctx->completion_lock); | |
1985 | } | |
1986 | ||
1987 | while (!list_empty(&cs->free_list)) { | |
1988 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1989 | compl.list); |
1990 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1991 | state->reqs[state->free_reqs++] = req; |
1992 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1993 | break; | |
1b4c351f JA |
1994 | } |
1995 | ||
e5d1bc0a PB |
1996 | return req != NULL; |
1997 | } | |
1998 | ||
1999 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
2000 | { | |
2001 | struct io_submit_state *state = &ctx->submit_state; | |
2002 | ||
2003 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
2004 | ||
f6b6c7d6 | 2005 | if (!state->free_reqs) { |
291b2821 | 2006 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
2007 | int ret; |
2008 | ||
c7dae4ba | 2009 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
2010 | goto got_req; |
2011 | ||
bf019da7 PB |
2012 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
2013 | state->reqs); | |
fd6fab2c JA |
2014 | |
2015 | /* | |
2016 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2017 | * retry single alloc to be on the safe side. | |
2018 | */ | |
2019 | if (unlikely(ret <= 0)) { | |
2020 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
2021 | if (!state->reqs[0]) | |
3893f39f | 2022 | return NULL; |
fd6fab2c JA |
2023 | ret = 1; |
2024 | } | |
291b2821 | 2025 | state->free_reqs = ret; |
2b188cc1 | 2026 | } |
e5d1bc0a | 2027 | got_req: |
291b2821 PB |
2028 | state->free_reqs--; |
2029 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
2030 | } |
2031 | ||
8da11c19 PB |
2032 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
2033 | bool fixed) | |
2034 | { | |
36f72fe2 | 2035 | if (!fixed) |
8da11c19 PB |
2036 | fput(file); |
2037 | } | |
2038 | ||
4edf20f9 | 2039 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2040 | { |
3ca405eb | 2041 | io_clean_op(req); |
929a3af9 | 2042 | |
e8c2bc1f JA |
2043 | if (req->async_data) |
2044 | kfree(req->async_data); | |
8da11c19 PB |
2045 | if (req->file) |
2046 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
2047 | if (req->fixed_rsrc_refs) |
2048 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 2049 | io_req_clean_work(req); |
e65ef56d JA |
2050 | } |
2051 | ||
7c660731 PB |
2052 | static inline void io_put_task(struct task_struct *task, int nr) |
2053 | { | |
2054 | struct io_uring_task *tctx = task->io_uring; | |
2055 | ||
2056 | percpu_counter_sub(&tctx->inflight, nr); | |
2057 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2058 | wake_up(&tctx->wait); | |
2059 | put_task_struct_many(task, nr); | |
2060 | } | |
2061 | ||
216578e5 | 2062 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2063 | { |
51a4cc11 | 2064 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2065 | |
216578e5 | 2066 | io_dismantle_req(req); |
7c660731 | 2067 | io_put_task(req->task, 1); |
e3bc8e9d | 2068 | |
3893f39f | 2069 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 2070 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
2071 | } |
2072 | ||
f2f87370 PB |
2073 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2074 | { | |
2075 | struct io_kiocb *nxt = req->link; | |
2076 | ||
2077 | req->link = nxt->link; | |
2078 | nxt->link = NULL; | |
2079 | } | |
2080 | ||
c9abd7ad | 2081 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 2082 | { |
a197f664 | 2083 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 2084 | struct io_kiocb *link; |
c9abd7ad PB |
2085 | bool cancelled = false; |
2086 | unsigned long flags; | |
7c86ffee | 2087 | |
c9abd7ad | 2088 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2089 | link = req->link; |
2090 | ||
900fad45 PB |
2091 | /* |
2092 | * Can happen if a linked timeout fired and link had been like | |
2093 | * req -> link t-out -> link t-out [-> ...] | |
2094 | */ | |
c9abd7ad PB |
2095 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
2096 | struct io_timeout_data *io = link->async_data; | |
2097 | int ret; | |
7c86ffee | 2098 | |
f2f87370 | 2099 | io_remove_next_linked(req); |
90cd7e42 | 2100 | link->timeout.head = NULL; |
c9abd7ad PB |
2101 | ret = hrtimer_try_to_cancel(&io->timer); |
2102 | if (ret != -1) { | |
2103 | io_cqring_fill_event(link, -ECANCELED); | |
2104 | io_commit_cqring(ctx); | |
2105 | cancelled = true; | |
2106 | } | |
2107 | } | |
7c86ffee | 2108 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 2109 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 2110 | |
c9abd7ad | 2111 | if (cancelled) { |
7c86ffee | 2112 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
2113 | io_put_req(link); |
2114 | } | |
7c86ffee PB |
2115 | } |
2116 | ||
9e645e11 | 2117 | |
d148ca4b | 2118 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 2119 | { |
f2f87370 | 2120 | struct io_kiocb *link, *nxt; |
2665abfd | 2121 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 2122 | unsigned long flags; |
9e645e11 | 2123 | |
d148ca4b | 2124 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2125 | link = req->link; |
2126 | req->link = NULL; | |
9e645e11 | 2127 | |
f2f87370 PB |
2128 | while (link) { |
2129 | nxt = link->link; | |
2130 | link->link = NULL; | |
2665abfd | 2131 | |
f2f87370 | 2132 | trace_io_uring_fail_link(req, link); |
7c86ffee | 2133 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
2134 | |
2135 | /* | |
2136 | * It's ok to free under spinlock as they're not linked anymore, | |
2137 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
2138 | * work.fs->lock. | |
2139 | */ | |
2140 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
2141 | io_put_req_deferred(link, 2); | |
2142 | else | |
2143 | io_double_put_req(link); | |
f2f87370 | 2144 | link = nxt; |
9e645e11 | 2145 | } |
2665abfd | 2146 | io_commit_cqring(ctx); |
216578e5 | 2147 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 2148 | |
2665abfd | 2149 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
2150 | } |
2151 | ||
3fa5e0f3 | 2152 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2153 | { |
7c86ffee PB |
2154 | if (req->flags & REQ_F_LINK_TIMEOUT) |
2155 | io_kill_linked_timeout(req); | |
944e58bf | 2156 | |
9e645e11 JA |
2157 | /* |
2158 | * If LINK is set, we have dependent requests in this chain. If we | |
2159 | * didn't fail this request, queue the first one up, moving any other | |
2160 | * dependencies to the next request. In case of failure, fail the rest | |
2161 | * of the chain. | |
2162 | */ | |
f2f87370 PB |
2163 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
2164 | struct io_kiocb *nxt = req->link; | |
2165 | ||
2166 | req->link = NULL; | |
2167 | return nxt; | |
2168 | } | |
9b5f7bd9 PB |
2169 | io_fail_links(req); |
2170 | return NULL; | |
4d7dd462 | 2171 | } |
9e645e11 | 2172 | |
f2f87370 | 2173 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2174 | { |
f2f87370 | 2175 | if (likely(!(req->link) && !(req->flags & REQ_F_LINK_TIMEOUT))) |
3fa5e0f3 PB |
2176 | return NULL; |
2177 | return __io_req_find_next(req); | |
2178 | } | |
2179 | ||
7cbf1722 JA |
2180 | static bool __tctx_task_work(struct io_uring_task *tctx) |
2181 | { | |
65453d1e | 2182 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
2183 | struct io_wq_work_list list; |
2184 | struct io_wq_work_node *node; | |
2185 | ||
2186 | if (wq_list_empty(&tctx->task_list)) | |
2187 | return false; | |
2188 | ||
2189 | spin_lock(&tctx->task_lock); | |
2190 | list = tctx->task_list; | |
2191 | INIT_WQ_LIST(&tctx->task_list); | |
2192 | spin_unlock(&tctx->task_lock); | |
2193 | ||
2194 | node = list.first; | |
2195 | while (node) { | |
2196 | struct io_wq_work_node *next = node->next; | |
65453d1e | 2197 | struct io_ring_ctx *this_ctx; |
7cbf1722 JA |
2198 | struct io_kiocb *req; |
2199 | ||
2200 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
65453d1e | 2201 | this_ctx = req->ctx; |
7cbf1722 JA |
2202 | req->task_work.func(&req->task_work); |
2203 | node = next; | |
65453d1e JA |
2204 | |
2205 | if (!ctx) { | |
2206 | ctx = this_ctx; | |
2207 | } else if (ctx != this_ctx) { | |
2208 | mutex_lock(&ctx->uring_lock); | |
2209 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2210 | mutex_unlock(&ctx->uring_lock); | |
2211 | ctx = this_ctx; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | if (ctx && ctx->submit_state.comp.nr) { | |
2216 | mutex_lock(&ctx->uring_lock); | |
2217 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2218 | mutex_unlock(&ctx->uring_lock); | |
7cbf1722 JA |
2219 | } |
2220 | ||
2221 | return list.first != NULL; | |
2222 | } | |
2223 | ||
2224 | static void tctx_task_work(struct callback_head *cb) | |
2225 | { | |
2226 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); | |
2227 | ||
2228 | while (__tctx_task_work(tctx)) | |
2229 | cond_resched(); | |
2230 | ||
2231 | clear_bit(0, &tctx->task_state); | |
2232 | } | |
2233 | ||
2234 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
2235 | enum task_work_notify_mode notify) | |
2236 | { | |
2237 | struct io_uring_task *tctx = tsk->io_uring; | |
2238 | struct io_wq_work_node *node, *prev; | |
2239 | int ret; | |
2240 | ||
2241 | WARN_ON_ONCE(!tctx); | |
2242 | ||
2243 | spin_lock(&tctx->task_lock); | |
2244 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); | |
2245 | spin_unlock(&tctx->task_lock); | |
2246 | ||
2247 | /* task_work already pending, we're done */ | |
2248 | if (test_bit(0, &tctx->task_state) || | |
2249 | test_and_set_bit(0, &tctx->task_state)) | |
2250 | return 0; | |
2251 | ||
2252 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
2253 | return 0; | |
2254 | ||
2255 | /* | |
2256 | * Slow path - we failed, find and delete work. if the work is not | |
2257 | * in the list, it got run and we're fine. | |
2258 | */ | |
2259 | ret = 0; | |
2260 | spin_lock(&tctx->task_lock); | |
2261 | wq_list_for_each(node, prev, &tctx->task_list) { | |
2262 | if (&req->io_task_work.node == node) { | |
2263 | wq_list_del(&tctx->task_list, node, prev); | |
2264 | ret = 1; | |
2265 | break; | |
2266 | } | |
2267 | } | |
2268 | spin_unlock(&tctx->task_lock); | |
2269 | clear_bit(0, &tctx->task_state); | |
2270 | return ret; | |
2271 | } | |
2272 | ||
355fb9e2 | 2273 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
2274 | { |
2275 | struct task_struct *tsk = req->task; | |
2276 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2277 | enum task_work_notify_mode notify; |
2278 | int ret; | |
c2c4c83c | 2279 | |
6200b0ae JA |
2280 | if (tsk->flags & PF_EXITING) |
2281 | return -ESRCH; | |
2282 | ||
c2c4c83c | 2283 | /* |
0ba9c9ed JA |
2284 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2285 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2286 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2287 | * will do the job. | |
c2c4c83c | 2288 | */ |
91989c70 | 2289 | notify = TWA_NONE; |
355fb9e2 | 2290 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
2291 | notify = TWA_SIGNAL; |
2292 | ||
7cbf1722 | 2293 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
2294 | if (!ret) |
2295 | wake_up_process(tsk); | |
0ba9c9ed | 2296 | |
c2c4c83c JA |
2297 | return ret; |
2298 | } | |
2299 | ||
eab30c4d | 2300 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 2301 | task_work_func_t cb) |
eab30c4d PB |
2302 | { |
2303 | struct task_struct *tsk = io_wq_get_task(req->ctx->io_wq); | |
2304 | ||
2305 | init_task_work(&req->task_work, cb); | |
2306 | task_work_add(tsk, &req->task_work, TWA_NONE); | |
2307 | wake_up_process(tsk); | |
2308 | } | |
2309 | ||
c40f6379 JA |
2310 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2311 | { | |
2312 | struct io_ring_ctx *ctx = req->ctx; | |
2313 | ||
2314 | spin_lock_irq(&ctx->completion_lock); | |
2315 | io_cqring_fill_event(req, error); | |
2316 | io_commit_cqring(ctx); | |
2317 | spin_unlock_irq(&ctx->completion_lock); | |
2318 | ||
2319 | io_cqring_ev_posted(ctx); | |
2320 | req_set_fail_links(req); | |
2321 | io_double_put_req(req); | |
2322 | } | |
2323 | ||
2324 | static void io_req_task_cancel(struct callback_head *cb) | |
2325 | { | |
2326 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2327 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2328 | |
2329 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2330 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2331 | } |
2332 | ||
2333 | static void __io_req_task_submit(struct io_kiocb *req) | |
2334 | { | |
2335 | struct io_ring_ctx *ctx = req->ctx; | |
2336 | ||
04fc6c80 | 2337 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2338 | mutex_lock(&ctx->uring_lock); |
dc0eced5 PB |
2339 | if (!ctx->sqo_dead && !(current->flags & PF_EXITING) && |
2340 | !io_sq_thread_acquire_mm_files(ctx, req)) | |
c5eef2b9 | 2341 | __io_queue_sqe(req); |
81b6d05c | 2342 | else |
c40f6379 | 2343 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2344 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2345 | } |
2346 | ||
2347 | static void io_req_task_submit(struct callback_head *cb) | |
2348 | { | |
2349 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2350 | ||
2351 | __io_req_task_submit(req); | |
2352 | } | |
2353 | ||
2354 | static void io_req_task_queue(struct io_kiocb *req) | |
2355 | { | |
c40f6379 JA |
2356 | int ret; |
2357 | ||
7cbf1722 | 2358 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2359 | ret = io_req_task_work_add(req); |
04fc6c80 PB |
2360 | if (unlikely(ret)) { |
2361 | percpu_ref_get(&req->ctx->refs); | |
eab30c4d | 2362 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
04fc6c80 | 2363 | } |
c40f6379 JA |
2364 | } |
2365 | ||
f2f87370 | 2366 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2367 | { |
9b5f7bd9 | 2368 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2369 | |
2370 | if (nxt) | |
906a8c3f | 2371 | io_req_task_queue(nxt); |
c69f8dbe JL |
2372 | } |
2373 | ||
c3524383 | 2374 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2375 | { |
c3524383 PB |
2376 | io_queue_next(req); |
2377 | __io_free_req(req); | |
2378 | } | |
8766dd51 | 2379 | |
2d6500d4 | 2380 | struct req_batch { |
5af1d13e PB |
2381 | struct task_struct *task; |
2382 | int task_refs; | |
1b4c351f | 2383 | int ctx_refs; |
2d6500d4 PB |
2384 | }; |
2385 | ||
5af1d13e PB |
2386 | static inline void io_init_req_batch(struct req_batch *rb) |
2387 | { | |
5af1d13e | 2388 | rb->task_refs = 0; |
9ae72463 | 2389 | rb->ctx_refs = 0; |
5af1d13e PB |
2390 | rb->task = NULL; |
2391 | } | |
2392 | ||
2d6500d4 PB |
2393 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2394 | struct req_batch *rb) | |
2395 | { | |
6e833d53 | 2396 | if (rb->task) |
7c660731 | 2397 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2398 | if (rb->ctx_refs) |
2399 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2400 | } |
2401 | ||
6ff119a6 PB |
2402 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2403 | struct io_submit_state *state) | |
2d6500d4 | 2404 | { |
f2f87370 | 2405 | io_queue_next(req); |
2d6500d4 | 2406 | |
e3bc8e9d | 2407 | if (req->task != rb->task) { |
7c660731 PB |
2408 | if (rb->task) |
2409 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2410 | rb->task = req->task; |
2411 | rb->task_refs = 0; | |
5af1d13e | 2412 | } |
e3bc8e9d | 2413 | rb->task_refs++; |
9ae72463 | 2414 | rb->ctx_refs++; |
5af1d13e | 2415 | |
4edf20f9 | 2416 | io_dismantle_req(req); |
bd759045 | 2417 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2418 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2419 | else |
2420 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2421 | } |
2422 | ||
905c172f PB |
2423 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2424 | struct io_ring_ctx *ctx) | |
2425 | { | |
2426 | int i, nr = cs->nr; | |
2427 | struct io_kiocb *req; | |
2428 | struct req_batch rb; | |
2429 | ||
2430 | io_init_req_batch(&rb); | |
2431 | spin_lock_irq(&ctx->completion_lock); | |
2432 | for (i = 0; i < nr; i++) { | |
2433 | req = cs->reqs[i]; | |
2434 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2435 | } | |
2436 | io_commit_cqring(ctx); | |
2437 | spin_unlock_irq(&ctx->completion_lock); | |
2438 | ||
2439 | io_cqring_ev_posted(ctx); | |
2440 | for (i = 0; i < nr; i++) { | |
2441 | req = cs->reqs[i]; | |
2442 | ||
2443 | /* submission and completion refs */ | |
2444 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2445 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2446 | } |
2447 | ||
2448 | io_req_free_batch_finish(ctx, &rb); | |
2449 | cs->nr = 0; | |
2450 | } | |
2451 | ||
ba816ad6 JA |
2452 | /* |
2453 | * Drop reference to request, return next in chain (if there is one) if this | |
2454 | * was the last reference to this request. | |
2455 | */ | |
9b5f7bd9 | 2456 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2457 | { |
9b5f7bd9 PB |
2458 | struct io_kiocb *nxt = NULL; |
2459 | ||
2a44f467 | 2460 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2461 | nxt = io_req_find_next(req); |
4d7dd462 | 2462 | __io_free_req(req); |
2a44f467 | 2463 | } |
9b5f7bd9 | 2464 | return nxt; |
2b188cc1 JA |
2465 | } |
2466 | ||
e65ef56d JA |
2467 | static void io_put_req(struct io_kiocb *req) |
2468 | { | |
2469 | if (refcount_dec_and_test(&req->refs)) | |
2470 | io_free_req(req); | |
2b188cc1 JA |
2471 | } |
2472 | ||
216578e5 PB |
2473 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2474 | { | |
2475 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2476 | ||
2477 | io_free_req(req); | |
2478 | } | |
2479 | ||
2480 | static void io_free_req_deferred(struct io_kiocb *req) | |
2481 | { | |
2482 | int ret; | |
2483 | ||
7cbf1722 | 2484 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2485 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2486 | if (unlikely(ret)) |
2487 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2488 | } |
2489 | ||
2490 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2491 | { | |
2492 | if (refcount_sub_and_test(refs, &req->refs)) | |
2493 | io_free_req_deferred(req); | |
2494 | } | |
2495 | ||
978db57e JA |
2496 | static void io_double_put_req(struct io_kiocb *req) |
2497 | { | |
2498 | /* drop both submit and complete references */ | |
2499 | if (refcount_sub_and_test(2, &req->refs)) | |
2500 | io_free_req(req); | |
2501 | } | |
2502 | ||
6c503150 | 2503 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2504 | { |
2505 | /* See comment at the top of this file */ | |
2506 | smp_rmb(); | |
e23de15f | 2507 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2508 | } |
2509 | ||
fb5ccc98 PB |
2510 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2511 | { | |
2512 | struct io_rings *rings = ctx->rings; | |
2513 | ||
2514 | /* make sure SQ entry isn't read before tail */ | |
2515 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2516 | } | |
2517 | ||
8ff069bf | 2518 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2519 | { |
8ff069bf | 2520 | unsigned int cflags; |
e94f141b | 2521 | |
bcda7baa JA |
2522 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2523 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2524 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2525 | kfree(kbuf); |
2526 | return cflags; | |
e94f141b JA |
2527 | } |
2528 | ||
8ff069bf | 2529 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2530 | { |
4d954c25 | 2531 | struct io_buffer *kbuf; |
bcda7baa | 2532 | |
4d954c25 | 2533 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2534 | return io_put_kbuf(req, kbuf); |
2535 | } | |
2536 | ||
4c6e277c JA |
2537 | static inline bool io_run_task_work(void) |
2538 | { | |
6200b0ae JA |
2539 | /* |
2540 | * Not safe to run on exiting task, and the task_work handling will | |
2541 | * not add work to such a task. | |
2542 | */ | |
2543 | if (unlikely(current->flags & PF_EXITING)) | |
2544 | return false; | |
4c6e277c JA |
2545 | if (current->task_works) { |
2546 | __set_current_state(TASK_RUNNING); | |
2547 | task_work_run(); | |
2548 | return true; | |
2549 | } | |
2550 | ||
2551 | return false; | |
bcda7baa JA |
2552 | } |
2553 | ||
def596e9 JA |
2554 | /* |
2555 | * Find and free completed poll iocbs | |
2556 | */ | |
2557 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2558 | struct list_head *done) | |
2559 | { | |
8237e045 | 2560 | struct req_batch rb; |
def596e9 | 2561 | struct io_kiocb *req; |
bbde017a XW |
2562 | |
2563 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2564 | smp_rmb(); | |
def596e9 | 2565 | |
5af1d13e | 2566 | io_init_req_batch(&rb); |
def596e9 | 2567 | while (!list_empty(done)) { |
bcda7baa JA |
2568 | int cflags = 0; |
2569 | ||
d21ffe7e | 2570 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2571 | list_del(&req->inflight_entry); |
2572 | ||
bbde017a XW |
2573 | if (READ_ONCE(req->result) == -EAGAIN) { |
2574 | req->iopoll_completed = 0; | |
23faba36 | 2575 | if (io_rw_reissue(req)) |
f161340d | 2576 | continue; |
bbde017a | 2577 | } |
def596e9 | 2578 | |
bcda7baa | 2579 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2580 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2581 | |
2582 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2583 | (*nr_events)++; |
2584 | ||
c3524383 | 2585 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2586 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2587 | } |
def596e9 | 2588 | |
09bb8394 | 2589 | io_commit_cqring(ctx); |
80c18e4a | 2590 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2591 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2592 | } |
2593 | ||
def596e9 JA |
2594 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2595 | long min) | |
2596 | { | |
2597 | struct io_kiocb *req, *tmp; | |
2598 | LIST_HEAD(done); | |
2599 | bool spin; | |
2600 | int ret; | |
2601 | ||
2602 | /* | |
2603 | * Only spin for completions if we don't have multiple devices hanging | |
2604 | * off our complete list, and we're under the requested amount. | |
2605 | */ | |
2606 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2607 | ||
2608 | ret = 0; | |
d21ffe7e | 2609 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2610 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2611 | |
2612 | /* | |
581f9810 BM |
2613 | * Move completed and retryable entries to our local lists. |
2614 | * If we find a request that requires polling, break out | |
2615 | * and complete those lists first, if we have entries there. | |
def596e9 | 2616 | */ |
65a6543d | 2617 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2618 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2619 | continue; |
2620 | } | |
2621 | if (!list_empty(&done)) | |
2622 | break; | |
2623 | ||
2624 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2625 | if (ret < 0) | |
2626 | break; | |
2627 | ||
3aadc23e PB |
2628 | /* iopoll may have completed current req */ |
2629 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2630 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2631 | |
def596e9 JA |
2632 | if (ret && spin) |
2633 | spin = false; | |
2634 | ret = 0; | |
2635 | } | |
2636 | ||
2637 | if (!list_empty(&done)) | |
2638 | io_iopoll_complete(ctx, nr_events, &done); | |
2639 | ||
2640 | return ret; | |
2641 | } | |
2642 | ||
2643 | /* | |
d195a66e | 2644 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2645 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2646 | * as a non-spinning completion check. | |
2647 | */ | |
2648 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2649 | long min) | |
2650 | { | |
540e32a0 | 2651 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2652 | int ret; |
2653 | ||
2654 | ret = io_do_iopoll(ctx, nr_events, min); | |
2655 | if (ret < 0) | |
2656 | return ret; | |
eba0a4dd | 2657 | if (*nr_events >= min) |
def596e9 JA |
2658 | return 0; |
2659 | } | |
2660 | ||
2661 | return 1; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * We can't just wait for polled events to come to us, we have to actively | |
2666 | * find and complete them. | |
2667 | */ | |
b2edc0a7 | 2668 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2669 | { |
2670 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2671 | return; | |
2672 | ||
2673 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2674 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2675 | unsigned int nr_events = 0; |
2676 | ||
b2edc0a7 | 2677 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2678 | |
b2edc0a7 PB |
2679 | /* let it sleep and repeat later if can't complete a request */ |
2680 | if (nr_events == 0) | |
2681 | break; | |
08f5439f JA |
2682 | /* |
2683 | * Ensure we allow local-to-the-cpu processing to take place, | |
2684 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2685 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2686 | */ |
3fcee5a6 PB |
2687 | if (need_resched()) { |
2688 | mutex_unlock(&ctx->uring_lock); | |
2689 | cond_resched(); | |
2690 | mutex_lock(&ctx->uring_lock); | |
2691 | } | |
def596e9 JA |
2692 | } |
2693 | mutex_unlock(&ctx->uring_lock); | |
2694 | } | |
2695 | ||
7668b92a | 2696 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2697 | { |
7668b92a | 2698 | unsigned int nr_events = 0; |
2b2ed975 | 2699 | int iters = 0, ret = 0; |
500f9fba | 2700 | |
c7849be9 XW |
2701 | /* |
2702 | * We disallow the app entering submit/complete with polling, but we | |
2703 | * still need to lock the ring to prevent racing with polled issue | |
2704 | * that got punted to a workqueue. | |
2705 | */ | |
2706 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2707 | do { |
a3a0e43f JA |
2708 | /* |
2709 | * Don't enter poll loop if we already have events pending. | |
2710 | * If we do, we can potentially be spinning for commands that | |
2711 | * already triggered a CQE (eg in error). | |
2712 | */ | |
6c503150 PB |
2713 | if (test_bit(0, &ctx->cq_check_overflow)) |
2714 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2715 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2716 | break; |
2717 | ||
500f9fba JA |
2718 | /* |
2719 | * If a submit got punted to a workqueue, we can have the | |
2720 | * application entering polling for a command before it gets | |
2721 | * issued. That app will hold the uring_lock for the duration | |
2722 | * of the poll right here, so we need to take a breather every | |
2723 | * now and then to ensure that the issue has a chance to add | |
2724 | * the poll to the issued list. Otherwise we can spin here | |
2725 | * forever, while the workqueue is stuck trying to acquire the | |
2726 | * very same mutex. | |
2727 | */ | |
2728 | if (!(++iters & 7)) { | |
2729 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2730 | io_run_task_work(); |
500f9fba JA |
2731 | mutex_lock(&ctx->uring_lock); |
2732 | } | |
2733 | ||
7668b92a | 2734 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2735 | if (ret <= 0) |
2736 | break; | |
2737 | ret = 0; | |
7668b92a | 2738 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2739 | |
500f9fba | 2740 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2741 | return ret; |
2742 | } | |
2743 | ||
491381ce | 2744 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2745 | { |
491381ce JA |
2746 | /* |
2747 | * Tell lockdep we inherited freeze protection from submission | |
2748 | * thread. | |
2749 | */ | |
2750 | if (req->flags & REQ_F_ISREG) { | |
2751 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2752 | |
491381ce | 2753 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2754 | } |
491381ce | 2755 | file_end_write(req->file); |
2b188cc1 JA |
2756 | } |
2757 | ||
b63534c4 | 2758 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2759 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2760 | { |
2761 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2762 | int rw, ret; |
b63534c4 | 2763 | struct iov_iter iter; |
b63534c4 | 2764 | |
dc2a6e9a PB |
2765 | /* already prepared */ |
2766 | if (req->async_data) | |
2767 | return true; | |
b63534c4 JA |
2768 | |
2769 | switch (req->opcode) { | |
2770 | case IORING_OP_READV: | |
2771 | case IORING_OP_READ_FIXED: | |
2772 | case IORING_OP_READ: | |
2773 | rw = READ; | |
2774 | break; | |
2775 | case IORING_OP_WRITEV: | |
2776 | case IORING_OP_WRITE_FIXED: | |
2777 | case IORING_OP_WRITE: | |
2778 | rw = WRITE; | |
2779 | break; | |
2780 | default: | |
2781 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2782 | req->opcode); | |
dc2a6e9a | 2783 | return false; |
b63534c4 JA |
2784 | } |
2785 | ||
dc2a6e9a PB |
2786 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2787 | if (ret < 0) | |
2788 | return false; | |
6bf985dc | 2789 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2790 | } |
b63534c4 JA |
2791 | #endif |
2792 | ||
23faba36 | 2793 | static bool io_rw_reissue(struct io_kiocb *req) |
b63534c4 JA |
2794 | { |
2795 | #ifdef CONFIG_BLOCK | |
23faba36 | 2796 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2797 | int ret; |
2798 | ||
75c668cd PB |
2799 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2800 | return false; | |
2801 | if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker()) | |
b63534c4 JA |
2802 | return false; |
2803 | ||
55e6ac1e PB |
2804 | lockdep_assert_held(&req->ctx->uring_lock); |
2805 | ||
28cea78a | 2806 | ret = io_sq_thread_acquire_mm_files(req->ctx, req); |
6d816e08 | 2807 | |
dc2a6e9a | 2808 | if (!ret && io_resubmit_prep(req)) { |
fdee946d JA |
2809 | refcount_inc(&req->refs); |
2810 | io_queue_async_work(req); | |
b63534c4 | 2811 | return true; |
fdee946d | 2812 | } |
dc2a6e9a | 2813 | req_set_fail_links(req); |
b63534c4 JA |
2814 | #endif |
2815 | return false; | |
2816 | } | |
2817 | ||
a1d7c393 | 2818 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2819 | unsigned int issue_flags) |
a1d7c393 | 2820 | { |
2f8e45f1 PB |
2821 | int cflags = 0; |
2822 | ||
23faba36 PB |
2823 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2824 | return; | |
2f8e45f1 PB |
2825 | if (res != req->result) |
2826 | req_set_fail_links(req); | |
23faba36 | 2827 | |
2f8e45f1 PB |
2828 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2829 | kiocb_end_write(req); | |
2830 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2831 | cflags = io_put_rw_kbuf(req); | |
2832 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2833 | } |
2834 | ||
2835 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2836 | { | |
9adbd45d | 2837 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2838 | |
889fca73 | 2839 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2840 | } |
2841 | ||
def596e9 JA |
2842 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2843 | { | |
9adbd45d | 2844 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2845 | |
491381ce JA |
2846 | if (kiocb->ki_flags & IOCB_WRITE) |
2847 | kiocb_end_write(req); | |
def596e9 | 2848 | |
2d7d6792 | 2849 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2850 | req_set_fail_links(req); |
bbde017a XW |
2851 | |
2852 | WRITE_ONCE(req->result, res); | |
2853 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2854 | smp_wmb(); |
2855 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2856 | } |
2857 | ||
2858 | /* | |
2859 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2860 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2861 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2862 | * accessing the kiocb cookie. | |
2863 | */ | |
2e9dbe90 | 2864 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2865 | { |
2866 | struct io_ring_ctx *ctx = req->ctx; | |
2867 | ||
2868 | /* | |
2869 | * Track whether we have multiple files in our lists. This will impact | |
2870 | * how we do polling eventually, not spinning if we're on potentially | |
2871 | * different devices. | |
2872 | */ | |
540e32a0 | 2873 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2874 | ctx->poll_multi_file = false; |
2875 | } else if (!ctx->poll_multi_file) { | |
2876 | struct io_kiocb *list_req; | |
2877 | ||
540e32a0 | 2878 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2879 | inflight_entry); |
9adbd45d | 2880 | if (list_req->file != req->file) |
def596e9 JA |
2881 | ctx->poll_multi_file = true; |
2882 | } | |
2883 | ||
2884 | /* | |
2885 | * For fast devices, IO may have already completed. If it has, add | |
2886 | * it to the front so we find it first. | |
2887 | */ | |
65a6543d | 2888 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2889 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2890 | else |
d21ffe7e | 2891 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2892 | |
2e9dbe90 XW |
2893 | /* |
2894 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2895 | * task context or in io worker task context. If current task context is | |
2896 | * sq thread, we don't need to check whether should wake up sq thread. | |
2897 | */ | |
2898 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2899 | wq_has_sleeper(&ctx->sq_data->wait)) |
2900 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2901 | } |
2902 | ||
9f13c35b PB |
2903 | static inline void io_state_file_put(struct io_submit_state *state) |
2904 | { | |
02b23a9a PB |
2905 | if (state->file_refs) { |
2906 | fput_many(state->file, state->file_refs); | |
2907 | state->file_refs = 0; | |
2908 | } | |
9a56a232 JA |
2909 | } |
2910 | ||
2911 | /* | |
2912 | * Get as many references to a file as we have IOs left in this submission, | |
2913 | * assuming most submissions are for one file, or at least that each file | |
2914 | * has more than one submission. | |
2915 | */ | |
8da11c19 | 2916 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2917 | { |
2918 | if (!state) | |
2919 | return fget(fd); | |
2920 | ||
6e1271e6 | 2921 | if (state->file_refs) { |
9a56a232 | 2922 | if (state->fd == fd) { |
6e1271e6 | 2923 | state->file_refs--; |
9a56a232 JA |
2924 | return state->file; |
2925 | } | |
02b23a9a | 2926 | io_state_file_put(state); |
9a56a232 JA |
2927 | } |
2928 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2929 | if (unlikely(!state->file)) |
9a56a232 JA |
2930 | return NULL; |
2931 | ||
2932 | state->fd = fd; | |
6e1271e6 | 2933 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2934 | return state->file; |
2935 | } | |
2936 | ||
4503b767 JA |
2937 | static bool io_bdev_nowait(struct block_device *bdev) |
2938 | { | |
9ba0d0c8 | 2939 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2940 | } |
2941 | ||
2b188cc1 JA |
2942 | /* |
2943 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2944 | * any file. For now, just ensure that anything potentially problematic is done | |
2945 | * inline. | |
2946 | */ | |
af197f50 | 2947 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2948 | { |
2949 | umode_t mode = file_inode(file)->i_mode; | |
2950 | ||
4503b767 | 2951 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2952 | if (IS_ENABLED(CONFIG_BLOCK) && |
2953 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2954 | return true; |
2955 | return false; | |
2956 | } | |
2957 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2958 | return true; |
4503b767 | 2959 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2960 | if (IS_ENABLED(CONFIG_BLOCK) && |
2961 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2962 | file->f_op != &io_uring_fops) |
2963 | return true; | |
2964 | return false; | |
2965 | } | |
2b188cc1 | 2966 | |
c5b85625 JA |
2967 | /* any ->read/write should understand O_NONBLOCK */ |
2968 | if (file->f_flags & O_NONBLOCK) | |
2969 | return true; | |
2970 | ||
af197f50 JA |
2971 | if (!(file->f_mode & FMODE_NOWAIT)) |
2972 | return false; | |
2973 | ||
2974 | if (rw == READ) | |
2975 | return file->f_op->read_iter != NULL; | |
2976 | ||
2977 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2978 | } |
2979 | ||
a88fc400 | 2980 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2981 | { |
def596e9 | 2982 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2983 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2984 | struct file *file = req->file; |
09bb8394 JA |
2985 | unsigned ioprio; |
2986 | int ret; | |
2b188cc1 | 2987 | |
75c668cd | 2988 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2989 | req->flags |= REQ_F_ISREG; |
2990 | ||
2b188cc1 | 2991 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2992 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2993 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2994 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2995 | } |
2b188cc1 | 2996 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2997 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2998 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2999 | if (unlikely(ret)) | |
3000 | return ret; | |
2b188cc1 | 3001 | |
75c668cd PB |
3002 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
3003 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
3004 | req->flags |= REQ_F_NOWAIT; | |
3005 | ||
2b188cc1 JA |
3006 | ioprio = READ_ONCE(sqe->ioprio); |
3007 | if (ioprio) { | |
3008 | ret = ioprio_check_cap(ioprio); | |
3009 | if (ret) | |
09bb8394 | 3010 | return ret; |
2b188cc1 JA |
3011 | |
3012 | kiocb->ki_ioprio = ioprio; | |
3013 | } else | |
3014 | kiocb->ki_ioprio = get_current_ioprio(); | |
3015 | ||
def596e9 | 3016 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
3017 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
3018 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 3019 | return -EOPNOTSUPP; |
2b188cc1 | 3020 | |
def596e9 JA |
3021 | kiocb->ki_flags |= IOCB_HIPRI; |
3022 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 3023 | req->iopoll_completed = 0; |
def596e9 | 3024 | } else { |
09bb8394 JA |
3025 | if (kiocb->ki_flags & IOCB_HIPRI) |
3026 | return -EINVAL; | |
def596e9 JA |
3027 | kiocb->ki_complete = io_complete_rw; |
3028 | } | |
9adbd45d | 3029 | |
3529d8c2 JA |
3030 | req->rw.addr = READ_ONCE(sqe->addr); |
3031 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 3032 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 3033 | return 0; |
2b188cc1 JA |
3034 | } |
3035 | ||
3036 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3037 | { | |
3038 | switch (ret) { | |
3039 | case -EIOCBQUEUED: | |
3040 | break; | |
3041 | case -ERESTARTSYS: | |
3042 | case -ERESTARTNOINTR: | |
3043 | case -ERESTARTNOHAND: | |
3044 | case -ERESTART_RESTARTBLOCK: | |
3045 | /* | |
3046 | * We can't just restart the syscall, since previously | |
3047 | * submitted sqes may already be in progress. Just fail this | |
3048 | * IO with EINTR. | |
3049 | */ | |
3050 | ret = -EINTR; | |
df561f66 | 3051 | fallthrough; |
2b188cc1 JA |
3052 | default: |
3053 | kiocb->ki_complete(kiocb, ret, 0); | |
3054 | } | |
3055 | } | |
3056 | ||
a1d7c393 | 3057 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 3058 | unsigned int issue_flags) |
ba816ad6 | 3059 | { |
ba04291e | 3060 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 3061 | struct io_async_rw *io = req->async_data; |
ba04291e | 3062 | |
227c0c96 | 3063 | /* add previously done IO, if any */ |
e8c2bc1f | 3064 | if (io && io->bytes_done > 0) { |
227c0c96 | 3065 | if (ret < 0) |
e8c2bc1f | 3066 | ret = io->bytes_done; |
227c0c96 | 3067 | else |
e8c2bc1f | 3068 | ret += io->bytes_done; |
227c0c96 JA |
3069 | } |
3070 | ||
ba04291e JA |
3071 | if (req->flags & REQ_F_CUR_POS) |
3072 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 3073 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 3074 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
3075 | else |
3076 | io_rw_done(kiocb, ret); | |
3077 | } | |
3078 | ||
847595de | 3079 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 3080 | { |
9adbd45d JA |
3081 | struct io_ring_ctx *ctx = req->ctx; |
3082 | size_t len = req->rw.len; | |
edafccee | 3083 | struct io_mapped_ubuf *imu; |
4be1c615 | 3084 | u16 index, buf_index = req->buf_index; |
edafccee JA |
3085 | size_t offset; |
3086 | u64 buf_addr; | |
3087 | ||
edafccee JA |
3088 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3089 | return -EFAULT; | |
edafccee JA |
3090 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3091 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 3092 | buf_addr = req->rw.addr; |
edafccee JA |
3093 | |
3094 | /* overflow */ | |
3095 | if (buf_addr + len < buf_addr) | |
3096 | return -EFAULT; | |
3097 | /* not inside the mapped region */ | |
3098 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
3099 | return -EFAULT; | |
3100 | ||
3101 | /* | |
3102 | * May not be a start of buffer, set size appropriately | |
3103 | * and advance us to the beginning. | |
3104 | */ | |
3105 | offset = buf_addr - imu->ubuf; | |
3106 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3107 | |
3108 | if (offset) { | |
3109 | /* | |
3110 | * Don't use iov_iter_advance() here, as it's really slow for | |
3111 | * using the latter parts of a big fixed buffer - it iterates | |
3112 | * over each segment manually. We can cheat a bit here, because | |
3113 | * we know that: | |
3114 | * | |
3115 | * 1) it's a BVEC iter, we set it up | |
3116 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3117 | * first and last bvec | |
3118 | * | |
3119 | * So just find our index, and adjust the iterator afterwards. | |
3120 | * If the offset is within the first bvec (or the whole first | |
3121 | * bvec, just use iov_iter_advance(). This makes it easier | |
3122 | * since we can just skip the first segment, which may not | |
3123 | * be PAGE_SIZE aligned. | |
3124 | */ | |
3125 | const struct bio_vec *bvec = imu->bvec; | |
3126 | ||
3127 | if (offset <= bvec->bv_len) { | |
3128 | iov_iter_advance(iter, offset); | |
3129 | } else { | |
3130 | unsigned long seg_skip; | |
3131 | ||
3132 | /* skip first vec */ | |
3133 | offset -= bvec->bv_len; | |
3134 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3135 | ||
3136 | iter->bvec = bvec + seg_skip; | |
3137 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3138 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3139 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3140 | } |
3141 | } | |
3142 | ||
847595de | 3143 | return 0; |
edafccee JA |
3144 | } |
3145 | ||
bcda7baa JA |
3146 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3147 | { | |
3148 | if (needs_lock) | |
3149 | mutex_unlock(&ctx->uring_lock); | |
3150 | } | |
3151 | ||
3152 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3153 | { | |
3154 | /* | |
3155 | * "Normal" inline submissions always hold the uring_lock, since we | |
3156 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3157 | * The only exception is when we've detached the request and issue it | |
3158 | * from an async worker thread, grab the lock for that case. | |
3159 | */ | |
3160 | if (needs_lock) | |
3161 | mutex_lock(&ctx->uring_lock); | |
3162 | } | |
3163 | ||
3164 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3165 | int bgid, struct io_buffer *kbuf, | |
3166 | bool needs_lock) | |
3167 | { | |
3168 | struct io_buffer *head; | |
3169 | ||
3170 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3171 | return kbuf; | |
3172 | ||
3173 | io_ring_submit_lock(req->ctx, needs_lock); | |
3174 | ||
3175 | lockdep_assert_held(&req->ctx->uring_lock); | |
3176 | ||
3177 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
3178 | if (head) { | |
3179 | if (!list_empty(&head->list)) { | |
3180 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3181 | list); | |
3182 | list_del(&kbuf->list); | |
3183 | } else { | |
3184 | kbuf = head; | |
3185 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
3186 | } | |
3187 | if (*len > kbuf->len) | |
3188 | *len = kbuf->len; | |
3189 | } else { | |
3190 | kbuf = ERR_PTR(-ENOBUFS); | |
3191 | } | |
3192 | ||
3193 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3194 | ||
3195 | return kbuf; | |
3196 | } | |
3197 | ||
4d954c25 JA |
3198 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3199 | bool needs_lock) | |
3200 | { | |
3201 | struct io_buffer *kbuf; | |
4f4eeba8 | 3202 | u16 bgid; |
4d954c25 JA |
3203 | |
3204 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3205 | bgid = req->buf_index; |
4d954c25 JA |
3206 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3207 | if (IS_ERR(kbuf)) | |
3208 | return kbuf; | |
3209 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3210 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3211 | return u64_to_user_ptr(kbuf->addr); | |
3212 | } | |
3213 | ||
3214 | #ifdef CONFIG_COMPAT | |
3215 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3216 | bool needs_lock) | |
3217 | { | |
3218 | struct compat_iovec __user *uiov; | |
3219 | compat_ssize_t clen; | |
3220 | void __user *buf; | |
3221 | ssize_t len; | |
3222 | ||
3223 | uiov = u64_to_user_ptr(req->rw.addr); | |
3224 | if (!access_ok(uiov, sizeof(*uiov))) | |
3225 | return -EFAULT; | |
3226 | if (__get_user(clen, &uiov->iov_len)) | |
3227 | return -EFAULT; | |
3228 | if (clen < 0) | |
3229 | return -EINVAL; | |
3230 | ||
3231 | len = clen; | |
3232 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3233 | if (IS_ERR(buf)) | |
3234 | return PTR_ERR(buf); | |
3235 | iov[0].iov_base = buf; | |
3236 | iov[0].iov_len = (compat_size_t) len; | |
3237 | return 0; | |
3238 | } | |
3239 | #endif | |
3240 | ||
3241 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3242 | bool needs_lock) | |
3243 | { | |
3244 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3245 | void __user *buf; | |
3246 | ssize_t len; | |
3247 | ||
3248 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3249 | return -EFAULT; | |
3250 | ||
3251 | len = iov[0].iov_len; | |
3252 | if (len < 0) | |
3253 | return -EINVAL; | |
3254 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3255 | if (IS_ERR(buf)) | |
3256 | return PTR_ERR(buf); | |
3257 | iov[0].iov_base = buf; | |
3258 | iov[0].iov_len = len; | |
3259 | return 0; | |
3260 | } | |
3261 | ||
3262 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3263 | bool needs_lock) | |
3264 | { | |
dddb3e26 JA |
3265 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3266 | struct io_buffer *kbuf; | |
3267 | ||
3268 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3269 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3270 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3271 | return 0; |
dddb3e26 | 3272 | } |
dd201662 | 3273 | if (req->rw.len != 1) |
4d954c25 JA |
3274 | return -EINVAL; |
3275 | ||
3276 | #ifdef CONFIG_COMPAT | |
3277 | if (req->ctx->compat) | |
3278 | return io_compat_import(req, iov, needs_lock); | |
3279 | #endif | |
3280 | ||
3281 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3282 | } | |
3283 | ||
847595de PB |
3284 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3285 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3286 | { |
9adbd45d JA |
3287 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3288 | size_t sqe_len = req->rw.len; | |
847595de | 3289 | u8 opcode = req->opcode; |
4d954c25 | 3290 | ssize_t ret; |
edafccee | 3291 | |
7d009165 | 3292 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3293 | *iovec = NULL; |
9adbd45d | 3294 | return io_import_fixed(req, rw, iter); |
edafccee | 3295 | } |
2b188cc1 | 3296 | |
bcda7baa | 3297 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3298 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3299 | return -EINVAL; |
3300 | ||
3a6820f2 | 3301 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3302 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3303 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3304 | if (IS_ERR(buf)) |
4d954c25 | 3305 | return PTR_ERR(buf); |
3f9d6441 | 3306 | req->rw.len = sqe_len; |
bcda7baa JA |
3307 | } |
3308 | ||
3a6820f2 JA |
3309 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3310 | *iovec = NULL; | |
10fc72e4 | 3311 | return ret; |
3a6820f2 JA |
3312 | } |
3313 | ||
4d954c25 JA |
3314 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3315 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3316 | if (!ret) |
3317 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3318 | *iovec = NULL; |
3319 | return ret; | |
3320 | } | |
3321 | ||
89cd35c5 CH |
3322 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3323 | req->ctx->compat); | |
2b188cc1 JA |
3324 | } |
3325 | ||
0fef9483 JA |
3326 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3327 | { | |
5b09e37e | 3328 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3329 | } |
3330 | ||
31b51510 | 3331 | /* |
32960613 JA |
3332 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3333 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3334 | */ |
4017eb91 | 3335 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3336 | { |
4017eb91 JA |
3337 | struct kiocb *kiocb = &req->rw.kiocb; |
3338 | struct file *file = req->file; | |
32960613 JA |
3339 | ssize_t ret = 0; |
3340 | ||
3341 | /* | |
3342 | * Don't support polled IO through this interface, and we can't | |
3343 | * support non-blocking either. For the latter, this just causes | |
3344 | * the kiocb to be handled from an async context. | |
3345 | */ | |
3346 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3347 | return -EOPNOTSUPP; | |
3348 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3349 | return -EAGAIN; | |
3350 | ||
3351 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3352 | struct iovec iovec; |
32960613 JA |
3353 | ssize_t nr; |
3354 | ||
311ae9e1 PB |
3355 | if (!iov_iter_is_bvec(iter)) { |
3356 | iovec = iov_iter_iovec(iter); | |
3357 | } else { | |
4017eb91 JA |
3358 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3359 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3360 | } |
3361 | ||
32960613 JA |
3362 | if (rw == READ) { |
3363 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3364 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3365 | } else { |
3366 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3367 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3368 | } |
3369 | ||
3370 | if (nr < 0) { | |
3371 | if (!ret) | |
3372 | ret = nr; | |
3373 | break; | |
3374 | } | |
3375 | ret += nr; | |
3376 | if (nr != iovec.iov_len) | |
3377 | break; | |
4017eb91 JA |
3378 | req->rw.len -= nr; |
3379 | req->rw.addr += nr; | |
32960613 JA |
3380 | iov_iter_advance(iter, nr); |
3381 | } | |
3382 | ||
3383 | return ret; | |
3384 | } | |
3385 | ||
ff6165b2 JA |
3386 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3387 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3388 | { |
e8c2bc1f | 3389 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3390 | |
ff6165b2 | 3391 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3392 | rw->free_iovec = iovec; |
227c0c96 | 3393 | rw->bytes_done = 0; |
ff6165b2 | 3394 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3395 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3396 | return; |
b64e3444 | 3397 | if (!iovec) { |
ff6165b2 JA |
3398 | unsigned iov_off = 0; |
3399 | ||
3400 | rw->iter.iov = rw->fast_iov; | |
3401 | if (iter->iov != fast_iov) { | |
3402 | iov_off = iter->iov - fast_iov; | |
3403 | rw->iter.iov += iov_off; | |
3404 | } | |
3405 | if (rw->fast_iov != fast_iov) | |
3406 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3407 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3408 | } else { |
3409 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3410 | } |
3411 | } | |
3412 | ||
e8c2bc1f | 3413 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3414 | { |
e8c2bc1f JA |
3415 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3416 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3417 | return req->async_data == NULL; | |
3d9932a8 XW |
3418 | } |
3419 | ||
e8c2bc1f | 3420 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3421 | { |
e8c2bc1f | 3422 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3423 | return 0; |
3d9932a8 | 3424 | |
e8c2bc1f | 3425 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3426 | } |
3427 | ||
ff6165b2 JA |
3428 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3429 | const struct iovec *fast_iov, | |
227c0c96 | 3430 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3431 | { |
e8c2bc1f | 3432 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3433 | return 0; |
e8c2bc1f | 3434 | if (!req->async_data) { |
6bf985dc PB |
3435 | if (__io_alloc_async_data(req)) { |
3436 | kfree(iovec); | |
5d204bcf | 3437 | return -ENOMEM; |
6bf985dc | 3438 | } |
b7bb4f7d | 3439 | |
ff6165b2 | 3440 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3441 | } |
b7bb4f7d | 3442 | return 0; |
f67676d1 JA |
3443 | } |
3444 | ||
73debe68 | 3445 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3446 | { |
e8c2bc1f | 3447 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3448 | struct iovec *iov = iorw->fast_iov; |
847595de | 3449 | int ret; |
c3e330a4 | 3450 | |
2846c481 | 3451 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3452 | if (unlikely(ret < 0)) |
3453 | return ret; | |
3454 | ||
ab0b196c PB |
3455 | iorw->bytes_done = 0; |
3456 | iorw->free_iovec = iov; | |
3457 | if (iov) | |
3458 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3459 | return 0; |
3460 | } | |
3461 | ||
73debe68 | 3462 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3463 | { |
3464 | ssize_t ret; | |
3465 | ||
a88fc400 | 3466 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3467 | if (ret) |
3468 | return ret; | |
f67676d1 | 3469 | |
3529d8c2 JA |
3470 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3471 | return -EBADF; | |
f67676d1 | 3472 | |
5f798bea | 3473 | /* either don't need iovec imported or already have it */ |
2d199895 | 3474 | if (!req->async_data) |
3529d8c2 | 3475 | return 0; |
73debe68 | 3476 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3477 | } |
3478 | ||
c1dd91d1 JA |
3479 | /* |
3480 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3481 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3482 | * This gets called when the page is unlocked, and we generally expect that to | |
3483 | * happen when the page IO is completed and the page is now uptodate. This will | |
3484 | * queue a task_work based retry of the operation, attempting to copy the data | |
3485 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3486 | * do a thread based blocking retry of the operation. That's the unexpected | |
3487 | * slow path. | |
3488 | */ | |
bcf5a063 JA |
3489 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3490 | int sync, void *arg) | |
3491 | { | |
3492 | struct wait_page_queue *wpq; | |
3493 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3494 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3495 | |
3496 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3497 | ||
cdc8fcb4 LT |
3498 | if (!wake_page_match(wpq, key)) |
3499 | return 0; | |
3500 | ||
c8d317aa | 3501 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3502 | list_del_init(&wait->entry); |
3503 | ||
bcf5a063 JA |
3504 | /* submit ref gets dropped, acquire a new one */ |
3505 | refcount_inc(&req->refs); | |
921b9054 | 3506 | io_req_task_queue(req); |
bcf5a063 JA |
3507 | return 1; |
3508 | } | |
3509 | ||
c1dd91d1 JA |
3510 | /* |
3511 | * This controls whether a given IO request should be armed for async page | |
3512 | * based retry. If we return false here, the request is handed to the async | |
3513 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3514 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3515 | * will either succeed because the page is now uptodate and unlocked, or it | |
3516 | * will register a callback when the page is unlocked at IO completion. Through | |
3517 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3518 | * That retry will attempt the buffered read again. The retry will generally | |
3519 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3520 | * async worker threads for a blocking retry. | |
3521 | */ | |
227c0c96 | 3522 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3523 | { |
e8c2bc1f JA |
3524 | struct io_async_rw *rw = req->async_data; |
3525 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3526 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3527 | |
bcf5a063 JA |
3528 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3529 | if (req->flags & REQ_F_NOWAIT) | |
3530 | return false; | |
f67676d1 | 3531 | |
227c0c96 | 3532 | /* Only for buffered IO */ |
3b2a4439 | 3533 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3534 | return false; |
3b2a4439 | 3535 | |
bcf5a063 JA |
3536 | /* |
3537 | * just use poll if we can, and don't attempt if the fs doesn't | |
3538 | * support callback based unlocks | |
3539 | */ | |
3540 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3541 | return false; | |
f67676d1 | 3542 | |
3b2a4439 JA |
3543 | wait->wait.func = io_async_buf_func; |
3544 | wait->wait.private = req; | |
3545 | wait->wait.flags = 0; | |
3546 | INIT_LIST_HEAD(&wait->wait.entry); | |
3547 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3548 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3549 | kiocb->ki_waitq = wait; |
3b2a4439 | 3550 | return true; |
bcf5a063 JA |
3551 | } |
3552 | ||
3553 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3554 | { | |
3555 | if (req->file->f_op->read_iter) | |
3556 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3557 | else if (req->file->f_op->read) |
4017eb91 | 3558 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3559 | else |
3560 | return -EINVAL; | |
f67676d1 JA |
3561 | } |
3562 | ||
889fca73 | 3563 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3564 | { |
3565 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3566 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3567 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3568 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3569 | ssize_t io_size, ret, ret2; |
45d189c6 | 3570 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3571 | |
2846c481 | 3572 | if (rw) { |
e8c2bc1f | 3573 | iter = &rw->iter; |
2846c481 PB |
3574 | iovec = NULL; |
3575 | } else { | |
3576 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3577 | if (ret < 0) | |
3578 | return ret; | |
3579 | } | |
632546c4 | 3580 | io_size = iov_iter_count(iter); |
fa15bafb | 3581 | req->result = io_size; |
2b188cc1 | 3582 | |
fd6c2e4c JA |
3583 | /* Ensure we clear previously set non-block flag */ |
3584 | if (!force_nonblock) | |
29de5f6a | 3585 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3586 | else |
3587 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3588 | ||
24c74678 | 3589 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3590 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3591 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3592 | return ret ?: -EAGAIN; |
6713e7a6 | 3593 | } |
9e645e11 | 3594 | |
632546c4 | 3595 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3596 | if (unlikely(ret)) { |
3597 | kfree(iovec); | |
3598 | return ret; | |
3599 | } | |
2b188cc1 | 3600 | |
227c0c96 | 3601 | ret = io_iter_do_read(req, iter); |
32960613 | 3602 | |
57cd657b | 3603 | if (ret == -EIOCBQUEUED) { |
5ea5dd45 PB |
3604 | /* it's faster to check here then delegate to kfree */ |
3605 | if (iovec) | |
3606 | kfree(iovec); | |
3607 | return 0; | |
227c0c96 | 3608 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3609 | /* IOPOLL retry should happen for io-wq threads */ |
3610 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3611 | goto done; |
75c668cd PB |
3612 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3613 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3614 | goto done; |
84216315 | 3615 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3616 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3617 | ret = 0; |
7335e3bf | 3618 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3619 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3620 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3621 | goto done; |
1a2cc0ce | 3622 | } |
227c0c96 | 3623 | |
227c0c96 | 3624 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3625 | if (ret2) |
3626 | return ret2; | |
3627 | ||
e8c2bc1f | 3628 | rw = req->async_data; |
227c0c96 | 3629 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3630 | iter = &rw->iter; |
227c0c96 | 3631 | |
b23df91b PB |
3632 | do { |
3633 | io_size -= ret; | |
3634 | rw->bytes_done += ret; | |
3635 | /* if we can retry, do so with the callbacks armed */ | |
3636 | if (!io_rw_should_retry(req)) { | |
3637 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3638 | return -EAGAIN; | |
3639 | } | |
3640 | ||
3641 | /* | |
3642 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3643 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3644 | * desired page gets unlocked. We can also get a partial read | |
3645 | * here, and if we do, then just retry at the new offset. | |
3646 | */ | |
3647 | ret = io_iter_do_read(req, iter); | |
3648 | if (ret == -EIOCBQUEUED) | |
3649 | return 0; | |
3650 | /* we got some bytes, but not all. retry. */ | |
3651 | } while (ret > 0 && ret < io_size); | |
227c0c96 | 3652 | done: |
889fca73 | 3653 | kiocb_done(kiocb, ret, issue_flags); |
5ea5dd45 | 3654 | return 0; |
2b188cc1 JA |
3655 | } |
3656 | ||
73debe68 | 3657 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3658 | { |
3659 | ssize_t ret; | |
3660 | ||
a88fc400 | 3661 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3662 | if (ret) |
3663 | return ret; | |
f67676d1 | 3664 | |
3529d8c2 JA |
3665 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3666 | return -EBADF; | |
f67676d1 | 3667 | |
5f798bea | 3668 | /* either don't need iovec imported or already have it */ |
2d199895 | 3669 | if (!req->async_data) |
3529d8c2 | 3670 | return 0; |
73debe68 | 3671 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3672 | } |
3673 | ||
889fca73 | 3674 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3675 | { |
3676 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3677 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3678 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3679 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3680 | ssize_t ret, ret2, io_size; |
45d189c6 | 3681 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3682 | |
2846c481 | 3683 | if (rw) { |
e8c2bc1f | 3684 | iter = &rw->iter; |
2846c481 PB |
3685 | iovec = NULL; |
3686 | } else { | |
3687 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3688 | if (ret < 0) | |
3689 | return ret; | |
3690 | } | |
632546c4 | 3691 | io_size = iov_iter_count(iter); |
fa15bafb | 3692 | req->result = io_size; |
2b188cc1 | 3693 | |
fd6c2e4c JA |
3694 | /* Ensure we clear previously set non-block flag */ |
3695 | if (!force_nonblock) | |
a88fc400 PB |
3696 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3697 | else | |
3698 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3699 | |
24c74678 | 3700 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3701 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3702 | goto copy_iov; |
31b51510 | 3703 | |
10d59345 JA |
3704 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3705 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3706 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3707 | goto copy_iov; |
31b51510 | 3708 | |
632546c4 | 3709 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3710 | if (unlikely(ret)) |
3711 | goto out_free; | |
4ed734b0 | 3712 | |
fa15bafb PB |
3713 | /* |
3714 | * Open-code file_start_write here to grab freeze protection, | |
3715 | * which will be released by another thread in | |
3716 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3717 | * released so that it doesn't complain about the held lock when | |
3718 | * we return to userspace. | |
3719 | */ | |
3720 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3721 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3722 | __sb_writers_release(file_inode(req->file)->i_sb, |
3723 | SB_FREEZE_WRITE); | |
3724 | } | |
3725 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3726 | |
fa15bafb | 3727 | if (req->file->f_op->write_iter) |
ff6165b2 | 3728 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3729 | else if (req->file->f_op->write) |
4017eb91 | 3730 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3731 | else |
3732 | ret2 = -EINVAL; | |
4ed734b0 | 3733 | |
fa15bafb PB |
3734 | /* |
3735 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3736 | * retry them without IOCB_NOWAIT. | |
3737 | */ | |
3738 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3739 | ret2 = -EAGAIN; | |
75c668cd PB |
3740 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3741 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3742 | goto done; |
fa15bafb | 3743 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3744 | /* IOPOLL retry should happen for io-wq threads */ |
3745 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3746 | goto copy_iov; | |
355afaeb | 3747 | done: |
889fca73 | 3748 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3749 | } else { |
f67676d1 | 3750 | copy_iov: |
84216315 | 3751 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3752 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3753 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3754 | return ret ?: -EAGAIN; |
2b188cc1 | 3755 | } |
31b51510 | 3756 | out_free: |
f261c168 | 3757 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3758 | if (iovec) |
6f2cc166 | 3759 | kfree(iovec); |
2b188cc1 JA |
3760 | return ret; |
3761 | } | |
3762 | ||
80a261fd JA |
3763 | static int io_renameat_prep(struct io_kiocb *req, |
3764 | const struct io_uring_sqe *sqe) | |
3765 | { | |
3766 | struct io_rename *ren = &req->rename; | |
3767 | const char __user *oldf, *newf; | |
3768 | ||
3769 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3770 | return -EBADF; | |
3771 | ||
3772 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3773 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3774 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3775 | ren->new_dfd = READ_ONCE(sqe->len); | |
3776 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3777 | ||
3778 | ren->oldpath = getname(oldf); | |
3779 | if (IS_ERR(ren->oldpath)) | |
3780 | return PTR_ERR(ren->oldpath); | |
3781 | ||
3782 | ren->newpath = getname(newf); | |
3783 | if (IS_ERR(ren->newpath)) { | |
3784 | putname(ren->oldpath); | |
3785 | return PTR_ERR(ren->newpath); | |
3786 | } | |
3787 | ||
3788 | req->flags |= REQ_F_NEED_CLEANUP; | |
3789 | return 0; | |
3790 | } | |
3791 | ||
45d189c6 | 3792 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3793 | { |
3794 | struct io_rename *ren = &req->rename; | |
3795 | int ret; | |
3796 | ||
45d189c6 | 3797 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3798 | return -EAGAIN; |
3799 | ||
3800 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3801 | ren->newpath, ren->flags); | |
3802 | ||
3803 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3804 | if (ret < 0) | |
3805 | req_set_fail_links(req); | |
3806 | io_req_complete(req, ret); | |
3807 | return 0; | |
3808 | } | |
3809 | ||
14a1143b JA |
3810 | static int io_unlinkat_prep(struct io_kiocb *req, |
3811 | const struct io_uring_sqe *sqe) | |
3812 | { | |
3813 | struct io_unlink *un = &req->unlink; | |
3814 | const char __user *fname; | |
3815 | ||
3816 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3817 | return -EBADF; | |
3818 | ||
3819 | un->dfd = READ_ONCE(sqe->fd); | |
3820 | ||
3821 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3822 | if (un->flags & ~AT_REMOVEDIR) | |
3823 | return -EINVAL; | |
3824 | ||
3825 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3826 | un->filename = getname(fname); | |
3827 | if (IS_ERR(un->filename)) | |
3828 | return PTR_ERR(un->filename); | |
3829 | ||
3830 | req->flags |= REQ_F_NEED_CLEANUP; | |
3831 | return 0; | |
3832 | } | |
3833 | ||
45d189c6 | 3834 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3835 | { |
3836 | struct io_unlink *un = &req->unlink; | |
3837 | int ret; | |
3838 | ||
45d189c6 | 3839 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3840 | return -EAGAIN; |
3841 | ||
3842 | if (un->flags & AT_REMOVEDIR) | |
3843 | ret = do_rmdir(un->dfd, un->filename); | |
3844 | else | |
3845 | ret = do_unlinkat(un->dfd, un->filename); | |
3846 | ||
3847 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3848 | if (ret < 0) | |
3849 | req_set_fail_links(req); | |
3850 | io_req_complete(req, ret); | |
3851 | return 0; | |
3852 | } | |
3853 | ||
36f4fa68 JA |
3854 | static int io_shutdown_prep(struct io_kiocb *req, |
3855 | const struct io_uring_sqe *sqe) | |
3856 | { | |
3857 | #if defined(CONFIG_NET) | |
3858 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3859 | return -EINVAL; | |
3860 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3861 | sqe->buf_index) | |
3862 | return -EINVAL; | |
3863 | ||
3864 | req->shutdown.how = READ_ONCE(sqe->len); | |
3865 | return 0; | |
3866 | #else | |
3867 | return -EOPNOTSUPP; | |
3868 | #endif | |
3869 | } | |
3870 | ||
45d189c6 | 3871 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3872 | { |
3873 | #if defined(CONFIG_NET) | |
3874 | struct socket *sock; | |
3875 | int ret; | |
3876 | ||
45d189c6 | 3877 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3878 | return -EAGAIN; |
3879 | ||
48aba79b | 3880 | sock = sock_from_file(req->file); |
36f4fa68 | 3881 | if (unlikely(!sock)) |
48aba79b | 3882 | return -ENOTSOCK; |
36f4fa68 JA |
3883 | |
3884 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3885 | if (ret < 0) |
3886 | req_set_fail_links(req); | |
36f4fa68 JA |
3887 | io_req_complete(req, ret); |
3888 | return 0; | |
3889 | #else | |
3890 | return -EOPNOTSUPP; | |
3891 | #endif | |
3892 | } | |
3893 | ||
f2a8d5c7 PB |
3894 | static int __io_splice_prep(struct io_kiocb *req, |
3895 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3896 | { |
3897 | struct io_splice* sp = &req->splice; | |
3898 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3899 | |
3232dd02 PB |
3900 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3901 | return -EINVAL; | |
7d67af2c PB |
3902 | |
3903 | sp->file_in = NULL; | |
7d67af2c PB |
3904 | sp->len = READ_ONCE(sqe->len); |
3905 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3906 | ||
3907 | if (unlikely(sp->flags & ~valid_flags)) | |
3908 | return -EINVAL; | |
3909 | ||
8371adf5 PB |
3910 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3911 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3912 | if (!sp->file_in) | |
3913 | return -EBADF; | |
7d67af2c PB |
3914 | req->flags |= REQ_F_NEED_CLEANUP; |
3915 | ||
7cdaf587 XW |
3916 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3917 | /* | |
3918 | * Splice operation will be punted aync, and here need to | |
3919 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3920 | */ | |
3921 | io_req_init_async(req); | |
7d67af2c | 3922 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3923 | } |
7d67af2c PB |
3924 | |
3925 | return 0; | |
3926 | } | |
3927 | ||
f2a8d5c7 PB |
3928 | static int io_tee_prep(struct io_kiocb *req, |
3929 | const struct io_uring_sqe *sqe) | |
3930 | { | |
3931 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3932 | return -EINVAL; | |
3933 | return __io_splice_prep(req, sqe); | |
3934 | } | |
3935 | ||
45d189c6 | 3936 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3937 | { |
3938 | struct io_splice *sp = &req->splice; | |
3939 | struct file *in = sp->file_in; | |
3940 | struct file *out = sp->file_out; | |
3941 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3942 | long ret = 0; | |
3943 | ||
45d189c6 | 3944 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3945 | return -EAGAIN; |
3946 | if (sp->len) | |
3947 | ret = do_tee(in, out, sp->len, flags); | |
3948 | ||
3949 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3950 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3951 | ||
f2a8d5c7 PB |
3952 | if (ret != sp->len) |
3953 | req_set_fail_links(req); | |
e1e16097 | 3954 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3955 | return 0; |
3956 | } | |
3957 | ||
3958 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3959 | { | |
3960 | struct io_splice* sp = &req->splice; | |
3961 | ||
3962 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3963 | sp->off_out = READ_ONCE(sqe->off); | |
3964 | return __io_splice_prep(req, sqe); | |
3965 | } | |
3966 | ||
45d189c6 | 3967 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3968 | { |
3969 | struct io_splice *sp = &req->splice; | |
3970 | struct file *in = sp->file_in; | |
3971 | struct file *out = sp->file_out; | |
3972 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3973 | loff_t *poff_in, *poff_out; | |
c9687426 | 3974 | long ret = 0; |
7d67af2c | 3975 | |
45d189c6 | 3976 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3977 | return -EAGAIN; |
7d67af2c PB |
3978 | |
3979 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3980 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3981 | |
948a7749 | 3982 | if (sp->len) |
c9687426 | 3983 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3984 | |
3985 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3986 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3987 | ||
7d67af2c PB |
3988 | if (ret != sp->len) |
3989 | req_set_fail_links(req); | |
e1e16097 | 3990 | io_req_complete(req, ret); |
7d67af2c PB |
3991 | return 0; |
3992 | } | |
3993 | ||
2b188cc1 JA |
3994 | /* |
3995 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3996 | */ | |
889fca73 | 3997 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3998 | { |
3999 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4000 | |
def596e9 JA |
4001 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4002 | return -EINVAL; | |
4003 | ||
889fca73 | 4004 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4005 | return 0; |
4006 | } | |
4007 | ||
3529d8c2 | 4008 | static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4009 | { |
6b06314c | 4010 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4011 | |
09bb8394 JA |
4012 | if (!req->file) |
4013 | return -EBADF; | |
c992fe29 | 4014 | |
6b06314c | 4015 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4016 | return -EINVAL; |
edafccee | 4017 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
4018 | return -EINVAL; |
4019 | ||
8ed8d3c3 JA |
4020 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4021 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4022 | return -EINVAL; | |
4023 | ||
4024 | req->sync.off = READ_ONCE(sqe->off); | |
4025 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4026 | return 0; |
4027 | } | |
4028 | ||
45d189c6 | 4029 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4030 | { |
8ed8d3c3 | 4031 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4032 | int ret; |
4033 | ||
ac45abc0 | 4034 | /* fsync always requires a blocking context */ |
45d189c6 | 4035 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4036 | return -EAGAIN; |
4037 | ||
9adbd45d | 4038 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4039 | end > 0 ? end : LLONG_MAX, |
4040 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4041 | if (ret < 0) | |
4042 | req_set_fail_links(req); | |
e1e16097 | 4043 | io_req_complete(req, ret); |
c992fe29 CH |
4044 | return 0; |
4045 | } | |
4046 | ||
d63d1b5e JA |
4047 | static int io_fallocate_prep(struct io_kiocb *req, |
4048 | const struct io_uring_sqe *sqe) | |
4049 | { | |
4050 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
4051 | return -EINVAL; | |
3232dd02 PB |
4052 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4053 | return -EINVAL; | |
d63d1b5e JA |
4054 | |
4055 | req->sync.off = READ_ONCE(sqe->off); | |
4056 | req->sync.len = READ_ONCE(sqe->addr); | |
4057 | req->sync.mode = READ_ONCE(sqe->len); | |
4058 | return 0; | |
4059 | } | |
4060 | ||
45d189c6 | 4061 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4062 | { |
ac45abc0 PB |
4063 | int ret; |
4064 | ||
d63d1b5e | 4065 | /* fallocate always requiring blocking context */ |
45d189c6 | 4066 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4067 | return -EAGAIN; |
ac45abc0 PB |
4068 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4069 | req->sync.len); | |
ac45abc0 PB |
4070 | if (ret < 0) |
4071 | req_set_fail_links(req); | |
e1e16097 | 4072 | io_req_complete(req, ret); |
5d17b4a4 JA |
4073 | return 0; |
4074 | } | |
4075 | ||
ec65fea5 | 4076 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4077 | { |
f8748881 | 4078 | const char __user *fname; |
15b71abe | 4079 | int ret; |
b7bb4f7d | 4080 | |
ec65fea5 | 4081 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4082 | return -EINVAL; |
ec65fea5 | 4083 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4084 | return -EBADF; |
03b1230c | 4085 | |
ec65fea5 PB |
4086 | /* open.how should be already initialised */ |
4087 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4088 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4089 | |
25e72d10 PB |
4090 | req->open.dfd = READ_ONCE(sqe->fd); |
4091 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4092 | req->open.filename = getname(fname); |
15b71abe JA |
4093 | if (IS_ERR(req->open.filename)) { |
4094 | ret = PTR_ERR(req->open.filename); | |
4095 | req->open.filename = NULL; | |
4096 | return ret; | |
4097 | } | |
4022e7af | 4098 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4099 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4100 | return 0; |
03b1230c JA |
4101 | } |
4102 | ||
ec65fea5 PB |
4103 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4104 | { | |
4105 | u64 flags, mode; | |
4106 | ||
14587a46 | 4107 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4108 | return -EINVAL; |
ec65fea5 PB |
4109 | mode = READ_ONCE(sqe->len); |
4110 | flags = READ_ONCE(sqe->open_flags); | |
4111 | req->open.how = build_open_how(flags, mode); | |
4112 | return __io_openat_prep(req, sqe); | |
4113 | } | |
4114 | ||
cebdb986 | 4115 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4116 | { |
cebdb986 | 4117 | struct open_how __user *how; |
cebdb986 | 4118 | size_t len; |
0fa03c62 JA |
4119 | int ret; |
4120 | ||
14587a46 | 4121 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4122 | return -EINVAL; |
cebdb986 JA |
4123 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4124 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4125 | if (len < OPEN_HOW_SIZE_VER0) |
4126 | return -EINVAL; | |
3529d8c2 | 4127 | |
cebdb986 JA |
4128 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4129 | len); | |
4130 | if (ret) | |
4131 | return ret; | |
3529d8c2 | 4132 | |
ec65fea5 | 4133 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4134 | } |
4135 | ||
45d189c6 | 4136 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4137 | { |
4138 | struct open_flags op; | |
15b71abe | 4139 | struct file *file; |
3a81fd02 JA |
4140 | bool nonblock_set; |
4141 | bool resolve_nonblock; | |
15b71abe JA |
4142 | int ret; |
4143 | ||
cebdb986 | 4144 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4145 | if (ret) |
4146 | goto err; | |
3a81fd02 JA |
4147 | nonblock_set = op.open_flag & O_NONBLOCK; |
4148 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4149 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4150 | /* |
4151 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4152 | * it'll always -EAGAIN | |
4153 | */ | |
4154 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4155 | return -EAGAIN; | |
4156 | op.lookup_flags |= LOOKUP_CACHED; | |
4157 | op.open_flag |= O_NONBLOCK; | |
4158 | } | |
15b71abe | 4159 | |
4022e7af | 4160 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
4161 | if (ret < 0) |
4162 | goto err; | |
4163 | ||
4164 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 4165 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
4166 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
4167 | file == ERR_PTR(-EAGAIN)) { | |
3a81fd02 JA |
4168 | /* |
4169 | * We could hang on to this 'fd', but seems like marginal | |
4170 | * gain for something that is now known to be a slower path. | |
4171 | * So just put it, and we'll get a new one when we retry. | |
4172 | */ | |
4173 | put_unused_fd(ret); | |
4174 | return -EAGAIN; | |
4175 | } | |
4176 | ||
15b71abe JA |
4177 | if (IS_ERR(file)) { |
4178 | put_unused_fd(ret); | |
4179 | ret = PTR_ERR(file); | |
4180 | } else { | |
45d189c6 | 4181 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 4182 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
4183 | fsnotify_open(file); |
4184 | fd_install(ret, file); | |
4185 | } | |
4186 | err: | |
4187 | putname(req->open.filename); | |
8fef80bf | 4188 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
4189 | if (ret < 0) |
4190 | req_set_fail_links(req); | |
e1e16097 | 4191 | io_req_complete(req, ret); |
15b71abe JA |
4192 | return 0; |
4193 | } | |
4194 | ||
45d189c6 | 4195 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4196 | { |
45d189c6 | 4197 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
4198 | } |
4199 | ||
067524e9 JA |
4200 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4201 | const struct io_uring_sqe *sqe) | |
4202 | { | |
4203 | struct io_provide_buf *p = &req->pbuf; | |
4204 | u64 tmp; | |
4205 | ||
4206 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
4207 | return -EINVAL; | |
4208 | ||
4209 | tmp = READ_ONCE(sqe->fd); | |
4210 | if (!tmp || tmp > USHRT_MAX) | |
4211 | return -EINVAL; | |
4212 | ||
4213 | memset(p, 0, sizeof(*p)); | |
4214 | p->nbufs = tmp; | |
4215 | p->bgid = READ_ONCE(sqe->buf_group); | |
4216 | return 0; | |
4217 | } | |
4218 | ||
4219 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4220 | int bgid, unsigned nbufs) | |
4221 | { | |
4222 | unsigned i = 0; | |
4223 | ||
4224 | /* shouldn't happen */ | |
4225 | if (!nbufs) | |
4226 | return 0; | |
4227 | ||
4228 | /* the head kbuf is the list itself */ | |
4229 | while (!list_empty(&buf->list)) { | |
4230 | struct io_buffer *nxt; | |
4231 | ||
4232 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4233 | list_del(&nxt->list); | |
4234 | kfree(nxt); | |
4235 | if (++i == nbufs) | |
4236 | return i; | |
4237 | } | |
4238 | i++; | |
4239 | kfree(buf); | |
4240 | idr_remove(&ctx->io_buffer_idr, bgid); | |
4241 | ||
4242 | return i; | |
4243 | } | |
4244 | ||
889fca73 | 4245 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4246 | { |
4247 | struct io_provide_buf *p = &req->pbuf; | |
4248 | struct io_ring_ctx *ctx = req->ctx; | |
4249 | struct io_buffer *head; | |
4250 | int ret = 0; | |
45d189c6 | 4251 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4252 | |
4253 | io_ring_submit_lock(ctx, !force_nonblock); | |
4254 | ||
4255 | lockdep_assert_held(&ctx->uring_lock); | |
4256 | ||
4257 | ret = -ENOENT; | |
4258 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4259 | if (head) | |
4260 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
4261 | if (ret < 0) |
4262 | req_set_fail_links(req); | |
067524e9 | 4263 | |
31bff9a5 PB |
4264 | /* need to hold the lock to complete IOPOLL requests */ |
4265 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4266 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4267 | io_ring_submit_unlock(ctx, !force_nonblock); |
4268 | } else { | |
4269 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4270 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4271 | } |
067524e9 JA |
4272 | return 0; |
4273 | } | |
4274 | ||
ddf0322d JA |
4275 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4276 | const struct io_uring_sqe *sqe) | |
4277 | { | |
4278 | struct io_provide_buf *p = &req->pbuf; | |
4279 | u64 tmp; | |
4280 | ||
4281 | if (sqe->ioprio || sqe->rw_flags) | |
4282 | return -EINVAL; | |
4283 | ||
4284 | tmp = READ_ONCE(sqe->fd); | |
4285 | if (!tmp || tmp > USHRT_MAX) | |
4286 | return -E2BIG; | |
4287 | p->nbufs = tmp; | |
4288 | p->addr = READ_ONCE(sqe->addr); | |
4289 | p->len = READ_ONCE(sqe->len); | |
4290 | ||
efe68c1c | 4291 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4292 | return -EFAULT; |
4293 | ||
4294 | p->bgid = READ_ONCE(sqe->buf_group); | |
4295 | tmp = READ_ONCE(sqe->off); | |
4296 | if (tmp > USHRT_MAX) | |
4297 | return -E2BIG; | |
4298 | p->bid = tmp; | |
4299 | return 0; | |
4300 | } | |
4301 | ||
4302 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4303 | { | |
4304 | struct io_buffer *buf; | |
4305 | u64 addr = pbuf->addr; | |
4306 | int i, bid = pbuf->bid; | |
4307 | ||
4308 | for (i = 0; i < pbuf->nbufs; i++) { | |
4309 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4310 | if (!buf) | |
4311 | break; | |
4312 | ||
4313 | buf->addr = addr; | |
4314 | buf->len = pbuf->len; | |
4315 | buf->bid = bid; | |
4316 | addr += pbuf->len; | |
4317 | bid++; | |
4318 | if (!*head) { | |
4319 | INIT_LIST_HEAD(&buf->list); | |
4320 | *head = buf; | |
4321 | } else { | |
4322 | list_add_tail(&buf->list, &(*head)->list); | |
4323 | } | |
4324 | } | |
4325 | ||
4326 | return i ? i : -ENOMEM; | |
4327 | } | |
4328 | ||
889fca73 | 4329 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4330 | { |
4331 | struct io_provide_buf *p = &req->pbuf; | |
4332 | struct io_ring_ctx *ctx = req->ctx; | |
4333 | struct io_buffer *head, *list; | |
4334 | int ret = 0; | |
45d189c6 | 4335 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4336 | |
4337 | io_ring_submit_lock(ctx, !force_nonblock); | |
4338 | ||
4339 | lockdep_assert_held(&ctx->uring_lock); | |
4340 | ||
4341 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4342 | ||
4343 | ret = io_add_buffers(p, &head); | |
4344 | if (ret < 0) | |
4345 | goto out; | |
4346 | ||
4347 | if (!list) { | |
4348 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4349 | GFP_KERNEL); | |
4350 | if (ret < 0) { | |
067524e9 | 4351 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4352 | goto out; |
4353 | } | |
4354 | } | |
4355 | out: | |
ddf0322d JA |
4356 | if (ret < 0) |
4357 | req_set_fail_links(req); | |
31bff9a5 PB |
4358 | |
4359 | /* need to hold the lock to complete IOPOLL requests */ | |
4360 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4361 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4362 | io_ring_submit_unlock(ctx, !force_nonblock); |
4363 | } else { | |
4364 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4365 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4366 | } |
ddf0322d | 4367 | return 0; |
cebdb986 JA |
4368 | } |
4369 | ||
3e4827b0 JA |
4370 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4371 | const struct io_uring_sqe *sqe) | |
4372 | { | |
4373 | #if defined(CONFIG_EPOLL) | |
4374 | if (sqe->ioprio || sqe->buf_index) | |
4375 | return -EINVAL; | |
6ca56f84 | 4376 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4377 | return -EINVAL; |
3e4827b0 JA |
4378 | |
4379 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4380 | req->epoll.op = READ_ONCE(sqe->len); | |
4381 | req->epoll.fd = READ_ONCE(sqe->off); | |
4382 | ||
4383 | if (ep_op_has_event(req->epoll.op)) { | |
4384 | struct epoll_event __user *ev; | |
4385 | ||
4386 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4387 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4388 | return -EFAULT; | |
4389 | } | |
4390 | ||
4391 | return 0; | |
4392 | #else | |
4393 | return -EOPNOTSUPP; | |
4394 | #endif | |
4395 | } | |
4396 | ||
889fca73 | 4397 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4398 | { |
4399 | #if defined(CONFIG_EPOLL) | |
4400 | struct io_epoll *ie = &req->epoll; | |
4401 | int ret; | |
45d189c6 | 4402 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4403 | |
4404 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4405 | if (force_nonblock && ret == -EAGAIN) | |
4406 | return -EAGAIN; | |
4407 | ||
4408 | if (ret < 0) | |
4409 | req_set_fail_links(req); | |
889fca73 | 4410 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4411 | return 0; |
4412 | #else | |
4413 | return -EOPNOTSUPP; | |
4414 | #endif | |
4415 | } | |
4416 | ||
c1ca757b JA |
4417 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4418 | { | |
4419 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4420 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4421 | return -EINVAL; | |
3232dd02 PB |
4422 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4423 | return -EINVAL; | |
c1ca757b JA |
4424 | |
4425 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4426 | req->madvise.len = READ_ONCE(sqe->len); | |
4427 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4428 | return 0; | |
4429 | #else | |
4430 | return -EOPNOTSUPP; | |
4431 | #endif | |
4432 | } | |
4433 | ||
45d189c6 | 4434 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4435 | { |
4436 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4437 | struct io_madvise *ma = &req->madvise; | |
4438 | int ret; | |
4439 | ||
45d189c6 | 4440 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4441 | return -EAGAIN; |
4442 | ||
0726b01e | 4443 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4444 | if (ret < 0) |
4445 | req_set_fail_links(req); | |
e1e16097 | 4446 | io_req_complete(req, ret); |
c1ca757b JA |
4447 | return 0; |
4448 | #else | |
4449 | return -EOPNOTSUPP; | |
4450 | #endif | |
4451 | } | |
4452 | ||
4840e418 JA |
4453 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4454 | { | |
4455 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4456 | return -EINVAL; | |
3232dd02 PB |
4457 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4458 | return -EINVAL; | |
4840e418 JA |
4459 | |
4460 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4461 | req->fadvise.len = READ_ONCE(sqe->len); | |
4462 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4463 | return 0; | |
4464 | } | |
4465 | ||
45d189c6 | 4466 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4467 | { |
4468 | struct io_fadvise *fa = &req->fadvise; | |
4469 | int ret; | |
4470 | ||
45d189c6 | 4471 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4472 | switch (fa->advice) { |
4473 | case POSIX_FADV_NORMAL: | |
4474 | case POSIX_FADV_RANDOM: | |
4475 | case POSIX_FADV_SEQUENTIAL: | |
4476 | break; | |
4477 | default: | |
4478 | return -EAGAIN; | |
4479 | } | |
4480 | } | |
4840e418 JA |
4481 | |
4482 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4483 | if (ret < 0) | |
4484 | req_set_fail_links(req); | |
e1e16097 | 4485 | io_req_complete(req, ret); |
4840e418 JA |
4486 | return 0; |
4487 | } | |
4488 | ||
eddc7ef5 JA |
4489 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4490 | { | |
6ca56f84 | 4491 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4492 | return -EINVAL; |
eddc7ef5 JA |
4493 | if (sqe->ioprio || sqe->buf_index) |
4494 | return -EINVAL; | |
9c280f90 | 4495 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4496 | return -EBADF; |
eddc7ef5 | 4497 | |
1d9e1288 BM |
4498 | req->statx.dfd = READ_ONCE(sqe->fd); |
4499 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4500 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4501 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4502 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4503 | |
4504 | return 0; | |
4505 | } | |
4506 | ||
45d189c6 | 4507 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4508 | { |
1d9e1288 | 4509 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4510 | int ret; |
4511 | ||
45d189c6 | 4512 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4513 | /* only need file table for an actual valid fd */ |
4514 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4515 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4516 | return -EAGAIN; |
5b0bbee4 | 4517 | } |
eddc7ef5 | 4518 | |
e62753e4 BM |
4519 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4520 | ctx->buffer); | |
eddc7ef5 | 4521 | |
eddc7ef5 JA |
4522 | if (ret < 0) |
4523 | req_set_fail_links(req); | |
e1e16097 | 4524 | io_req_complete(req, ret); |
eddc7ef5 JA |
4525 | return 0; |
4526 | } | |
4527 | ||
b5dba59e JA |
4528 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4529 | { | |
14587a46 | 4530 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4531 | return -EINVAL; |
b5dba59e JA |
4532 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4533 | sqe->rw_flags || sqe->buf_index) | |
4534 | return -EINVAL; | |
9c280f90 | 4535 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4536 | return -EBADF; |
b5dba59e JA |
4537 | |
4538 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4539 | return 0; |
b5dba59e JA |
4540 | } |
4541 | ||
889fca73 | 4542 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4543 | { |
9eac1904 | 4544 | struct files_struct *files = current->files; |
3af73b28 | 4545 | struct io_close *close = &req->close; |
9eac1904 JA |
4546 | struct fdtable *fdt; |
4547 | struct file *file; | |
b5dba59e JA |
4548 | int ret; |
4549 | ||
9eac1904 JA |
4550 | file = NULL; |
4551 | ret = -EBADF; | |
4552 | spin_lock(&files->file_lock); | |
4553 | fdt = files_fdtable(files); | |
4554 | if (close->fd >= fdt->max_fds) { | |
4555 | spin_unlock(&files->file_lock); | |
4556 | goto err; | |
4557 | } | |
4558 | file = fdt->fd[close->fd]; | |
4559 | if (!file) { | |
4560 | spin_unlock(&files->file_lock); | |
4561 | goto err; | |
4562 | } | |
4563 | ||
4564 | if (file->f_op == &io_uring_fops) { | |
4565 | spin_unlock(&files->file_lock); | |
4566 | file = NULL; | |
4567 | goto err; | |
3af73b28 | 4568 | } |
b5dba59e JA |
4569 | |
4570 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4571 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4572 | spin_unlock(&files->file_lock); |
0bf0eefd | 4573 | return -EAGAIN; |
a2100672 | 4574 | } |
b5dba59e | 4575 | |
9eac1904 JA |
4576 | ret = __close_fd_get_file(close->fd, &file); |
4577 | spin_unlock(&files->file_lock); | |
4578 | if (ret < 0) { | |
4579 | if (ret == -ENOENT) | |
4580 | ret = -EBADF; | |
4581 | goto err; | |
4582 | } | |
4583 | ||
3af73b28 | 4584 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4585 | ret = filp_close(file, current->files); |
4586 | err: | |
3af73b28 PB |
4587 | if (ret < 0) |
4588 | req_set_fail_links(req); | |
9eac1904 JA |
4589 | if (file) |
4590 | fput(file); | |
889fca73 | 4591 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4592 | return 0; |
b5dba59e JA |
4593 | } |
4594 | ||
3529d8c2 | 4595 | static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4596 | { |
4597 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 JA |
4598 | |
4599 | if (!req->file) | |
4600 | return -EBADF; | |
5d17b4a4 JA |
4601 | |
4602 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) | |
4603 | return -EINVAL; | |
4604 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4605 | return -EINVAL; | |
4606 | ||
8ed8d3c3 JA |
4607 | req->sync.off = READ_ONCE(sqe->off); |
4608 | req->sync.len = READ_ONCE(sqe->len); | |
4609 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4610 | return 0; |
4611 | } | |
4612 | ||
45d189c6 | 4613 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4614 | { |
8ed8d3c3 JA |
4615 | int ret; |
4616 | ||
ac45abc0 | 4617 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4618 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4619 | return -EAGAIN; |
4620 | ||
9adbd45d | 4621 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4622 | req->sync.flags); |
4623 | if (ret < 0) | |
4624 | req_set_fail_links(req); | |
e1e16097 | 4625 | io_req_complete(req, ret); |
5d17b4a4 JA |
4626 | return 0; |
4627 | } | |
4628 | ||
469956e8 | 4629 | #if defined(CONFIG_NET) |
02d27d89 PB |
4630 | static int io_setup_async_msg(struct io_kiocb *req, |
4631 | struct io_async_msghdr *kmsg) | |
4632 | { | |
e8c2bc1f JA |
4633 | struct io_async_msghdr *async_msg = req->async_data; |
4634 | ||
4635 | if (async_msg) | |
02d27d89 | 4636 | return -EAGAIN; |
e8c2bc1f | 4637 | if (io_alloc_async_data(req)) { |
257e84a5 | 4638 | kfree(kmsg->free_iov); |
02d27d89 PB |
4639 | return -ENOMEM; |
4640 | } | |
e8c2bc1f | 4641 | async_msg = req->async_data; |
02d27d89 | 4642 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4643 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4644 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4645 | /* if were using fast_iov, set it to the new one */ |
4646 | if (!async_msg->free_iov) | |
4647 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4648 | ||
02d27d89 PB |
4649 | return -EAGAIN; |
4650 | } | |
4651 | ||
2ae523ed PB |
4652 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4653 | struct io_async_msghdr *iomsg) | |
4654 | { | |
2ae523ed | 4655 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4656 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4657 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4658 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4659 | } |
4660 | ||
3529d8c2 | 4661 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4662 | { |
e8c2bc1f | 4663 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4664 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4665 | int ret; |
03b1230c | 4666 | |
d2b6f48b PB |
4667 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4668 | return -EINVAL; | |
4669 | ||
e47293fd | 4670 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4671 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4672 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4673 | |
d8768362 JA |
4674 | #ifdef CONFIG_COMPAT |
4675 | if (req->ctx->compat) | |
4676 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4677 | #endif | |
4678 | ||
e8c2bc1f | 4679 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4680 | return 0; |
e8c2bc1f | 4681 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4682 | if (!ret) |
4683 | req->flags |= REQ_F_NEED_CLEANUP; | |
4684 | return ret; | |
03b1230c JA |
4685 | } |
4686 | ||
889fca73 | 4687 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4688 | { |
6b754c8b | 4689 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4690 | struct socket *sock; |
7a7cacba | 4691 | unsigned flags; |
0fa03c62 JA |
4692 | int ret; |
4693 | ||
dba4a925 | 4694 | sock = sock_from_file(req->file); |
7a7cacba | 4695 | if (unlikely(!sock)) |
dba4a925 | 4696 | return -ENOTSOCK; |
3529d8c2 | 4697 | |
257e84a5 PB |
4698 | kmsg = req->async_data; |
4699 | if (!kmsg) { | |
7a7cacba PB |
4700 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4701 | if (ret) | |
4702 | return ret; | |
4703 | kmsg = &iomsg; | |
0fa03c62 | 4704 | } |
0fa03c62 | 4705 | |
7a7cacba PB |
4706 | flags = req->sr_msg.msg_flags; |
4707 | if (flags & MSG_DONTWAIT) | |
4708 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4709 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4710 | flags |= MSG_DONTWAIT; |
e47293fd | 4711 | |
7a7cacba | 4712 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4713 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4714 | return io_setup_async_msg(req, kmsg); |
4715 | if (ret == -ERESTARTSYS) | |
4716 | ret = -EINTR; | |
0fa03c62 | 4717 | |
257e84a5 PB |
4718 | /* fast path, check for non-NULL to avoid function call */ |
4719 | if (kmsg->free_iov) | |
4720 | kfree(kmsg->free_iov); | |
99bc4c38 | 4721 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4722 | if (ret < 0) |
4723 | req_set_fail_links(req); | |
889fca73 | 4724 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4725 | return 0; |
03b1230c | 4726 | } |
aa1fa28f | 4727 | |
889fca73 | 4728 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4729 | { |
7a7cacba PB |
4730 | struct io_sr_msg *sr = &req->sr_msg; |
4731 | struct msghdr msg; | |
4732 | struct iovec iov; | |
fddaface | 4733 | struct socket *sock; |
7a7cacba | 4734 | unsigned flags; |
fddaface JA |
4735 | int ret; |
4736 | ||
dba4a925 | 4737 | sock = sock_from_file(req->file); |
7a7cacba | 4738 | if (unlikely(!sock)) |
dba4a925 | 4739 | return -ENOTSOCK; |
fddaface | 4740 | |
7a7cacba PB |
4741 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4742 | if (unlikely(ret)) | |
14db8411 | 4743 | return ret; |
fddaface | 4744 | |
7a7cacba PB |
4745 | msg.msg_name = NULL; |
4746 | msg.msg_control = NULL; | |
4747 | msg.msg_controllen = 0; | |
4748 | msg.msg_namelen = 0; | |
fddaface | 4749 | |
7a7cacba PB |
4750 | flags = req->sr_msg.msg_flags; |
4751 | if (flags & MSG_DONTWAIT) | |
4752 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4753 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4754 | flags |= MSG_DONTWAIT; |
fddaface | 4755 | |
7a7cacba PB |
4756 | msg.msg_flags = flags; |
4757 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4758 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4759 | return -EAGAIN; |
4760 | if (ret == -ERESTARTSYS) | |
4761 | ret = -EINTR; | |
fddaface | 4762 | |
fddaface JA |
4763 | if (ret < 0) |
4764 | req_set_fail_links(req); | |
889fca73 | 4765 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4766 | return 0; |
fddaface JA |
4767 | } |
4768 | ||
1400e697 PB |
4769 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4770 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4771 | { |
4772 | struct io_sr_msg *sr = &req->sr_msg; | |
4773 | struct iovec __user *uiov; | |
4774 | size_t iov_len; | |
4775 | int ret; | |
4776 | ||
1400e697 PB |
4777 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4778 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4779 | if (ret) |
4780 | return ret; | |
4781 | ||
4782 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4783 | if (iov_len > 1) | |
4784 | return -EINVAL; | |
5476dfed | 4785 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4786 | return -EFAULT; |
5476dfed | 4787 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4788 | iomsg->free_iov = NULL; |
52de1fe1 | 4789 | } else { |
257e84a5 | 4790 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4791 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4792 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4793 | false); |
52de1fe1 JA |
4794 | if (ret > 0) |
4795 | ret = 0; | |
4796 | } | |
4797 | ||
4798 | return ret; | |
4799 | } | |
4800 | ||
4801 | #ifdef CONFIG_COMPAT | |
4802 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4803 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4804 | { |
4805 | struct compat_msghdr __user *msg_compat; | |
4806 | struct io_sr_msg *sr = &req->sr_msg; | |
4807 | struct compat_iovec __user *uiov; | |
4808 | compat_uptr_t ptr; | |
4809 | compat_size_t len; | |
4810 | int ret; | |
4811 | ||
270a5940 | 4812 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4813 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4814 | &ptr, &len); |
4815 | if (ret) | |
4816 | return ret; | |
4817 | ||
4818 | uiov = compat_ptr(ptr); | |
4819 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4820 | compat_ssize_t clen; | |
4821 | ||
4822 | if (len > 1) | |
4823 | return -EINVAL; | |
4824 | if (!access_ok(uiov, sizeof(*uiov))) | |
4825 | return -EFAULT; | |
4826 | if (__get_user(clen, &uiov->iov_len)) | |
4827 | return -EFAULT; | |
4828 | if (clen < 0) | |
4829 | return -EINVAL; | |
2d280bc8 | 4830 | sr->len = clen; |
257e84a5 | 4831 | iomsg->free_iov = NULL; |
52de1fe1 | 4832 | } else { |
257e84a5 | 4833 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4834 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4835 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4836 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4837 | if (ret < 0) |
4838 | return ret; | |
4839 | } | |
4840 | ||
4841 | return 0; | |
4842 | } | |
4843 | #endif | |
4844 | ||
1400e697 PB |
4845 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4846 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4847 | { |
1400e697 | 4848 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4849 | |
4850 | #ifdef CONFIG_COMPAT | |
4851 | if (req->ctx->compat) | |
1400e697 | 4852 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4853 | #endif |
52de1fe1 | 4854 | |
1400e697 | 4855 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4856 | } |
4857 | ||
bcda7baa | 4858 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4859 | bool needs_lock) |
bcda7baa JA |
4860 | { |
4861 | struct io_sr_msg *sr = &req->sr_msg; | |
4862 | struct io_buffer *kbuf; | |
4863 | ||
bcda7baa JA |
4864 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4865 | if (IS_ERR(kbuf)) | |
4866 | return kbuf; | |
4867 | ||
4868 | sr->kbuf = kbuf; | |
4869 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4870 | return kbuf; |
fddaface JA |
4871 | } |
4872 | ||
7fbb1b54 PB |
4873 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4874 | { | |
4875 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4876 | } | |
4877 | ||
3529d8c2 JA |
4878 | static int io_recvmsg_prep(struct io_kiocb *req, |
4879 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4880 | { |
e8c2bc1f | 4881 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4882 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4883 | int ret; |
3529d8c2 | 4884 | |
d2b6f48b PB |
4885 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4886 | return -EINVAL; | |
4887 | ||
3529d8c2 | 4888 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4889 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4890 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4891 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4892 | |
d8768362 JA |
4893 | #ifdef CONFIG_COMPAT |
4894 | if (req->ctx->compat) | |
4895 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4896 | #endif | |
4897 | ||
e8c2bc1f | 4898 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4899 | return 0; |
e8c2bc1f | 4900 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4901 | if (!ret) |
4902 | req->flags |= REQ_F_NEED_CLEANUP; | |
4903 | return ret; | |
aa1fa28f JA |
4904 | } |
4905 | ||
889fca73 | 4906 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4907 | { |
6b754c8b | 4908 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4909 | struct socket *sock; |
7fbb1b54 | 4910 | struct io_buffer *kbuf; |
7a7cacba | 4911 | unsigned flags; |
52de1fe1 | 4912 | int ret, cflags = 0; |
45d189c6 | 4913 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4914 | |
dba4a925 | 4915 | sock = sock_from_file(req->file); |
7a7cacba | 4916 | if (unlikely(!sock)) |
dba4a925 | 4917 | return -ENOTSOCK; |
3529d8c2 | 4918 | |
257e84a5 PB |
4919 | kmsg = req->async_data; |
4920 | if (!kmsg) { | |
7a7cacba PB |
4921 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4922 | if (ret) | |
681fda8d | 4923 | return ret; |
7a7cacba PB |
4924 | kmsg = &iomsg; |
4925 | } | |
03b1230c | 4926 | |
bc02ef33 | 4927 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4928 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4929 | if (IS_ERR(kbuf)) |
52de1fe1 | 4930 | return PTR_ERR(kbuf); |
7a7cacba | 4931 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4932 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4933 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4934 | 1, req->sr_msg.len); |
4935 | } | |
52de1fe1 | 4936 | |
7a7cacba PB |
4937 | flags = req->sr_msg.msg_flags; |
4938 | if (flags & MSG_DONTWAIT) | |
4939 | req->flags |= REQ_F_NOWAIT; | |
4940 | else if (force_nonblock) | |
4941 | flags |= MSG_DONTWAIT; | |
e47293fd | 4942 | |
7a7cacba PB |
4943 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4944 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4945 | if (force_nonblock && ret == -EAGAIN) |
4946 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4947 | if (ret == -ERESTARTSYS) |
4948 | ret = -EINTR; | |
03b1230c | 4949 | |
7fbb1b54 PB |
4950 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4951 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4952 | /* fast path, check for non-NULL to avoid function call */ |
4953 | if (kmsg->free_iov) | |
4954 | kfree(kmsg->free_iov); | |
99bc4c38 | 4955 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4956 | if (ret < 0) |
4957 | req_set_fail_links(req); | |
889fca73 | 4958 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4959 | return 0; |
0fa03c62 | 4960 | } |
5d17b4a4 | 4961 | |
889fca73 | 4962 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4963 | { |
6b754c8b | 4964 | struct io_buffer *kbuf; |
7a7cacba PB |
4965 | struct io_sr_msg *sr = &req->sr_msg; |
4966 | struct msghdr msg; | |
4967 | void __user *buf = sr->buf; | |
fddaface | 4968 | struct socket *sock; |
7a7cacba PB |
4969 | struct iovec iov; |
4970 | unsigned flags; | |
bcda7baa | 4971 | int ret, cflags = 0; |
45d189c6 | 4972 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4973 | |
dba4a925 | 4974 | sock = sock_from_file(req->file); |
7a7cacba | 4975 | if (unlikely(!sock)) |
dba4a925 | 4976 | return -ENOTSOCK; |
fddaface | 4977 | |
bc02ef33 | 4978 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4979 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4980 | if (IS_ERR(kbuf)) |
4981 | return PTR_ERR(kbuf); | |
7a7cacba | 4982 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4983 | } |
bcda7baa | 4984 | |
7a7cacba | 4985 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4986 | if (unlikely(ret)) |
4987 | goto out_free; | |
fddaface | 4988 | |
7a7cacba PB |
4989 | msg.msg_name = NULL; |
4990 | msg.msg_control = NULL; | |
4991 | msg.msg_controllen = 0; | |
4992 | msg.msg_namelen = 0; | |
4993 | msg.msg_iocb = NULL; | |
4994 | msg.msg_flags = 0; | |
fddaface | 4995 | |
7a7cacba PB |
4996 | flags = req->sr_msg.msg_flags; |
4997 | if (flags & MSG_DONTWAIT) | |
4998 | req->flags |= REQ_F_NOWAIT; | |
4999 | else if (force_nonblock) | |
5000 | flags |= MSG_DONTWAIT; | |
5001 | ||
5002 | ret = sock_recvmsg(sock, &msg, flags); | |
5003 | if (force_nonblock && ret == -EAGAIN) | |
5004 | return -EAGAIN; | |
5005 | if (ret == -ERESTARTSYS) | |
5006 | ret = -EINTR; | |
14c32eee | 5007 | out_free: |
7fbb1b54 PB |
5008 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5009 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
5010 | if (ret < 0) |
5011 | req_set_fail_links(req); | |
889fca73 | 5012 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5013 | return 0; |
fddaface JA |
5014 | } |
5015 | ||
3529d8c2 | 5016 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5017 | { |
8ed8d3c3 JA |
5018 | struct io_accept *accept = &req->accept; |
5019 | ||
14587a46 | 5020 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5021 | return -EINVAL; |
8042d6ce | 5022 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5023 | return -EINVAL; |
5024 | ||
d55e5f5b JA |
5025 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5026 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5027 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5028 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 5029 | return 0; |
8ed8d3c3 | 5030 | } |
17f2fe35 | 5031 | |
889fca73 | 5032 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5033 | { |
5034 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5035 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5036 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
5037 | int ret; |
5038 | ||
e697deed JX |
5039 | if (req->file->f_flags & O_NONBLOCK) |
5040 | req->flags |= REQ_F_NOWAIT; | |
5041 | ||
8ed8d3c3 | 5042 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
5043 | accept->addr_len, accept->flags, |
5044 | accept->nofile); | |
8ed8d3c3 | 5045 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 5046 | return -EAGAIN; |
ac45abc0 PB |
5047 | if (ret < 0) { |
5048 | if (ret == -ERESTARTSYS) | |
5049 | ret = -EINTR; | |
4e88d6e7 | 5050 | req_set_fail_links(req); |
ac45abc0 | 5051 | } |
889fca73 | 5052 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5053 | return 0; |
8ed8d3c3 JA |
5054 | } |
5055 | ||
3529d8c2 | 5056 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5057 | { |
3529d8c2 | 5058 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 5059 | struct io_async_connect *io = req->async_data; |
f499a021 | 5060 | |
14587a46 | 5061 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
5062 | return -EINVAL; |
5063 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
5064 | return -EINVAL; | |
5065 | ||
3529d8c2 JA |
5066 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5067 | conn->addr_len = READ_ONCE(sqe->addr2); | |
5068 | ||
5069 | if (!io) | |
5070 | return 0; | |
5071 | ||
5072 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 5073 | &io->address); |
f499a021 JA |
5074 | } |
5075 | ||
889fca73 | 5076 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5077 | { |
e8c2bc1f | 5078 | struct io_async_connect __io, *io; |
f8e85cf2 | 5079 | unsigned file_flags; |
3fbb51c1 | 5080 | int ret; |
45d189c6 | 5081 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5082 | |
e8c2bc1f JA |
5083 | if (req->async_data) { |
5084 | io = req->async_data; | |
f499a021 | 5085 | } else { |
3529d8c2 JA |
5086 | ret = move_addr_to_kernel(req->connect.addr, |
5087 | req->connect.addr_len, | |
e8c2bc1f | 5088 | &__io.address); |
f499a021 JA |
5089 | if (ret) |
5090 | goto out; | |
5091 | io = &__io; | |
5092 | } | |
5093 | ||
3fbb51c1 JA |
5094 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5095 | ||
e8c2bc1f | 5096 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5097 | req->connect.addr_len, file_flags); |
87f80d62 | 5098 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5099 | if (req->async_data) |
b7bb4f7d | 5100 | return -EAGAIN; |
e8c2bc1f | 5101 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5102 | ret = -ENOMEM; |
5103 | goto out; | |
5104 | } | |
e8c2bc1f JA |
5105 | io = req->async_data; |
5106 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 5107 | return -EAGAIN; |
f499a021 | 5108 | } |
f8e85cf2 JA |
5109 | if (ret == -ERESTARTSYS) |
5110 | ret = -EINTR; | |
f499a021 | 5111 | out: |
4e88d6e7 JA |
5112 | if (ret < 0) |
5113 | req_set_fail_links(req); | |
889fca73 | 5114 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5115 | return 0; |
469956e8 Y |
5116 | } |
5117 | #else /* !CONFIG_NET */ | |
5118 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5119 | { | |
f8e85cf2 | 5120 | return -EOPNOTSUPP; |
f8e85cf2 JA |
5121 | } |
5122 | ||
889fca73 | 5123 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5124 | { |
5125 | return -EOPNOTSUPP; | |
5126 | } | |
5127 | ||
889fca73 | 5128 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5129 | { |
5130 | return -EOPNOTSUPP; | |
5131 | } | |
5132 | ||
5133 | static int io_recvmsg_prep(struct io_kiocb *req, | |
5134 | const struct io_uring_sqe *sqe) | |
5135 | { | |
5136 | return -EOPNOTSUPP; | |
5137 | } | |
5138 | ||
889fca73 | 5139 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5140 | { |
5141 | return -EOPNOTSUPP; | |
5142 | } | |
5143 | ||
889fca73 | 5144 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5145 | { |
5146 | return -EOPNOTSUPP; | |
5147 | } | |
5148 | ||
5149 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5150 | { | |
5151 | return -EOPNOTSUPP; | |
5152 | } | |
5153 | ||
889fca73 | 5154 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5155 | { |
5156 | return -EOPNOTSUPP; | |
5157 | } | |
ce593a6c | 5158 | |
469956e8 Y |
5159 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5160 | { | |
5161 | return -EOPNOTSUPP; | |
5162 | } | |
5163 | ||
889fca73 | 5164 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 | 5165 | { |
f8e85cf2 | 5166 | return -EOPNOTSUPP; |
ce593a6c | 5167 | } |
469956e8 | 5168 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5169 | |
d7718a9d JA |
5170 | struct io_poll_table { |
5171 | struct poll_table_struct pt; | |
5172 | struct io_kiocb *req; | |
5173 | int error; | |
5174 | }; | |
ce593a6c | 5175 | |
d7718a9d JA |
5176 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5177 | __poll_t mask, task_work_func_t func) | |
5178 | { | |
aa96bf8a | 5179 | int ret; |
d7718a9d JA |
5180 | |
5181 | /* for instances that support it check for an event match first: */ | |
5182 | if (mask && !(mask & poll->events)) | |
5183 | return 0; | |
5184 | ||
5185 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5186 | ||
5187 | list_del_init(&poll->wait.entry); | |
5188 | ||
d7718a9d | 5189 | req->result = mask; |
7cbf1722 | 5190 | req->task_work.func = func; |
6d816e08 JA |
5191 | percpu_ref_get(&req->ctx->refs); |
5192 | ||
d7718a9d | 5193 | /* |
e3aabf95 JA |
5194 | * If this fails, then the task is exiting. When a task exits, the |
5195 | * work gets canceled, so just cancel this request as well instead | |
5196 | * of executing it. We can't safely execute it anyway, as we may not | |
5197 | * have the needed state needed for it anyway. | |
d7718a9d | 5198 | */ |
355fb9e2 | 5199 | ret = io_req_task_work_add(req); |
aa96bf8a | 5200 | if (unlikely(ret)) { |
e3aabf95 | 5201 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 5202 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 5203 | } |
d7718a9d JA |
5204 | return 1; |
5205 | } | |
5206 | ||
74ce6ce4 JA |
5207 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5208 | __acquires(&req->ctx->completion_lock) | |
5209 | { | |
5210 | struct io_ring_ctx *ctx = req->ctx; | |
5211 | ||
5212 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5213 | struct poll_table_struct pt = { ._key = poll->events }; | |
5214 | ||
5215 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5216 | } | |
5217 | ||
5218 | spin_lock_irq(&ctx->completion_lock); | |
5219 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5220 | add_wait_queue(poll->head, &poll->wait); | |
5221 | return true; | |
5222 | } | |
5223 | ||
5224 | return false; | |
5225 | } | |
5226 | ||
d4e7cd36 | 5227 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5228 | { |
e8c2bc1f | 5229 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5230 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5231 | return req->async_data; |
d4e7cd36 JA |
5232 | return req->apoll->double_poll; |
5233 | } | |
5234 | ||
5235 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5236 | { | |
5237 | if (req->opcode == IORING_OP_POLL_ADD) | |
5238 | return &req->poll; | |
5239 | return &req->apoll->poll; | |
5240 | } | |
5241 | ||
5242 | static void io_poll_remove_double(struct io_kiocb *req) | |
5243 | { | |
5244 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5245 | |
5246 | lockdep_assert_held(&req->ctx->completion_lock); | |
5247 | ||
5248 | if (poll && poll->head) { | |
5249 | struct wait_queue_head *head = poll->head; | |
5250 | ||
5251 | spin_lock(&head->lock); | |
5252 | list_del_init(&poll->wait.entry); | |
5253 | if (poll->wait.private) | |
5254 | refcount_dec(&req->refs); | |
5255 | poll->head = NULL; | |
5256 | spin_unlock(&head->lock); | |
5257 | } | |
5258 | } | |
5259 | ||
5260 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5261 | { | |
5262 | struct io_ring_ctx *ctx = req->ctx; | |
5263 | ||
d4e7cd36 | 5264 | io_poll_remove_double(req); |
18bceab1 JA |
5265 | req->poll.done = true; |
5266 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5267 | io_commit_cqring(ctx); | |
5268 | } | |
5269 | ||
dd221f46 | 5270 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5271 | { |
dd221f46 | 5272 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5273 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5274 | struct io_kiocb *nxt; |
18bceab1 JA |
5275 | |
5276 | if (io_poll_rewait(req, &req->poll)) { | |
5277 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5278 | } else { |
5279 | hash_del(&req->hash_node); | |
5280 | io_poll_complete(req, req->result, 0); | |
5281 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5282 | |
dd221f46 PB |
5283 | nxt = io_put_req_find_next(req); |
5284 | io_cqring_ev_posted(ctx); | |
5285 | if (nxt) | |
5286 | __io_req_task_submit(nxt); | |
5287 | } | |
18bceab1 | 5288 | |
6d816e08 | 5289 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5290 | } |
5291 | ||
5292 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5293 | int sync, void *key) | |
5294 | { | |
5295 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5296 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5297 | __poll_t mask = key_to_poll(key); |
5298 | ||
5299 | /* for instances that support it check for an event match first: */ | |
5300 | if (mask && !(mask & poll->events)) | |
5301 | return 0; | |
5302 | ||
8706e04e JA |
5303 | list_del_init(&wait->entry); |
5304 | ||
807abcb0 | 5305 | if (poll && poll->head) { |
18bceab1 JA |
5306 | bool done; |
5307 | ||
807abcb0 JA |
5308 | spin_lock(&poll->head->lock); |
5309 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5310 | if (!done) |
807abcb0 | 5311 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5312 | /* make sure double remove sees this as being gone */ |
5313 | wait->private = NULL; | |
807abcb0 | 5314 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5315 | if (!done) { |
5316 | /* use wait func handler, so it matches the rq type */ | |
5317 | poll->wait.func(&poll->wait, mode, sync, key); | |
5318 | } | |
18bceab1 JA |
5319 | } |
5320 | refcount_dec(&req->refs); | |
5321 | return 1; | |
5322 | } | |
5323 | ||
5324 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5325 | wait_queue_func_t wake_func) | |
5326 | { | |
5327 | poll->head = NULL; | |
5328 | poll->done = false; | |
5329 | poll->canceled = false; | |
5330 | poll->events = events; | |
5331 | INIT_LIST_HEAD(&poll->wait.entry); | |
5332 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5333 | } | |
5334 | ||
5335 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5336 | struct wait_queue_head *head, |
5337 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5338 | { |
5339 | struct io_kiocb *req = pt->req; | |
5340 | ||
5341 | /* | |
5342 | * If poll->head is already set, it's because the file being polled | |
5343 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5344 | * for write). Setup a separate io_poll_iocb if this happens. | |
5345 | */ | |
5346 | if (unlikely(poll->head)) { | |
58852d4d PB |
5347 | struct io_poll_iocb *poll_one = poll; |
5348 | ||
18bceab1 | 5349 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5350 | if (*poll_ptr) { |
18bceab1 JA |
5351 | pt->error = -EINVAL; |
5352 | return; | |
5353 | } | |
5354 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5355 | if (!poll) { | |
5356 | pt->error = -ENOMEM; | |
5357 | return; | |
5358 | } | |
58852d4d | 5359 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5360 | refcount_inc(&req->refs); |
5361 | poll->wait.private = req; | |
807abcb0 | 5362 | *poll_ptr = poll; |
18bceab1 JA |
5363 | } |
5364 | ||
5365 | pt->error = 0; | |
5366 | poll->head = head; | |
a31eb4a2 JX |
5367 | |
5368 | if (poll->events & EPOLLEXCLUSIVE) | |
5369 | add_wait_queue_exclusive(head, &poll->wait); | |
5370 | else | |
5371 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5372 | } |
5373 | ||
5374 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5375 | struct poll_table_struct *p) | |
5376 | { | |
5377 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5378 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5379 | |
807abcb0 | 5380 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5381 | } |
5382 | ||
d7718a9d JA |
5383 | static void io_async_task_func(struct callback_head *cb) |
5384 | { | |
5385 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5386 | struct async_poll *apoll = req->apoll; | |
5387 | struct io_ring_ctx *ctx = req->ctx; | |
5388 | ||
5389 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5390 | ||
74ce6ce4 | 5391 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5392 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5393 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5394 | return; |
d7718a9d JA |
5395 | } |
5396 | ||
31067255 | 5397 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5398 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5399 | hash_del(&req->hash_node); |
2bae047e | 5400 | |
d4e7cd36 | 5401 | io_poll_remove_double(req); |
74ce6ce4 JA |
5402 | spin_unlock_irq(&ctx->completion_lock); |
5403 | ||
0be0b0e3 PB |
5404 | if (!READ_ONCE(apoll->poll.canceled)) |
5405 | __io_req_task_submit(req); | |
5406 | else | |
5407 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5408 | |
6d816e08 | 5409 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5410 | kfree(apoll->double_poll); |
31067255 | 5411 | kfree(apoll); |
d7718a9d JA |
5412 | } |
5413 | ||
5414 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5415 | void *key) | |
5416 | { | |
5417 | struct io_kiocb *req = wait->private; | |
5418 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5419 | ||
5420 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5421 | key_to_poll(key)); | |
5422 | ||
5423 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5424 | } | |
5425 | ||
5426 | static void io_poll_req_insert(struct io_kiocb *req) | |
5427 | { | |
5428 | struct io_ring_ctx *ctx = req->ctx; | |
5429 | struct hlist_head *list; | |
5430 | ||
5431 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5432 | hlist_add_head(&req->hash_node, list); | |
5433 | } | |
5434 | ||
5435 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5436 | struct io_poll_iocb *poll, | |
5437 | struct io_poll_table *ipt, __poll_t mask, | |
5438 | wait_queue_func_t wake_func) | |
5439 | __acquires(&ctx->completion_lock) | |
5440 | { | |
5441 | struct io_ring_ctx *ctx = req->ctx; | |
5442 | bool cancel = false; | |
5443 | ||
4d52f338 | 5444 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5445 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5446 | poll->file = req->file; |
18bceab1 | 5447 | poll->wait.private = req; |
d7718a9d JA |
5448 | |
5449 | ipt->pt._key = mask; | |
5450 | ipt->req = req; | |
5451 | ipt->error = -EINVAL; | |
5452 | ||
d7718a9d JA |
5453 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5454 | ||
5455 | spin_lock_irq(&ctx->completion_lock); | |
5456 | if (likely(poll->head)) { | |
5457 | spin_lock(&poll->head->lock); | |
5458 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5459 | if (ipt->error) | |
5460 | cancel = true; | |
5461 | ipt->error = 0; | |
5462 | mask = 0; | |
5463 | } | |
5464 | if (mask || ipt->error) | |
5465 | list_del_init(&poll->wait.entry); | |
5466 | else if (cancel) | |
5467 | WRITE_ONCE(poll->canceled, true); | |
5468 | else if (!poll->done) /* actually waiting for an event */ | |
5469 | io_poll_req_insert(req); | |
5470 | spin_unlock(&poll->head->lock); | |
5471 | } | |
5472 | ||
5473 | return mask; | |
5474 | } | |
5475 | ||
5476 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5477 | { | |
5478 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5479 | struct io_ring_ctx *ctx = req->ctx; | |
5480 | struct async_poll *apoll; | |
5481 | struct io_poll_table ipt; | |
5482 | __poll_t mask, ret; | |
9dab14b8 | 5483 | int rw; |
d7718a9d JA |
5484 | |
5485 | if (!req->file || !file_can_poll(req->file)) | |
5486 | return false; | |
24c74678 | 5487 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5488 | return false; |
9dab14b8 JA |
5489 | if (def->pollin) |
5490 | rw = READ; | |
5491 | else if (def->pollout) | |
5492 | rw = WRITE; | |
5493 | else | |
5494 | return false; | |
5495 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5496 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5497 | return false; |
5498 | ||
5499 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5500 | if (unlikely(!apoll)) | |
5501 | return false; | |
807abcb0 | 5502 | apoll->double_poll = NULL; |
d7718a9d JA |
5503 | |
5504 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5505 | req->apoll = apoll; |
d7718a9d | 5506 | |
8755d97a | 5507 | mask = 0; |
d7718a9d | 5508 | if (def->pollin) |
8755d97a | 5509 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5510 | if (def->pollout) |
5511 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5512 | |
5513 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5514 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5515 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5516 | mask &= ~POLLIN; | |
5517 | ||
d7718a9d JA |
5518 | mask |= POLLERR | POLLPRI; |
5519 | ||
5520 | ipt.pt._qproc = io_async_queue_proc; | |
5521 | ||
5522 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5523 | io_async_wake); | |
a36da65c | 5524 | if (ret || ipt.error) { |
d4e7cd36 | 5525 | io_poll_remove_double(req); |
d7718a9d | 5526 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5527 | kfree(apoll->double_poll); |
d7718a9d JA |
5528 | kfree(apoll); |
5529 | return false; | |
5530 | } | |
5531 | spin_unlock_irq(&ctx->completion_lock); | |
5532 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5533 | apoll->poll.events); | |
5534 | return true; | |
5535 | } | |
5536 | ||
5537 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5538 | struct io_poll_iocb *poll) | |
221c5eb2 | 5539 | { |
b41e9852 | 5540 | bool do_complete = false; |
221c5eb2 JA |
5541 | |
5542 | spin_lock(&poll->head->lock); | |
5543 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5544 | if (!list_empty(&poll->wait.entry)) { |
5545 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5546 | do_complete = true; |
221c5eb2 JA |
5547 | } |
5548 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5549 | hash_del(&req->hash_node); |
d7718a9d JA |
5550 | return do_complete; |
5551 | } | |
5552 | ||
5553 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5554 | { | |
5555 | bool do_complete; | |
5556 | ||
d4e7cd36 JA |
5557 | io_poll_remove_double(req); |
5558 | ||
d7718a9d JA |
5559 | if (req->opcode == IORING_OP_POLL_ADD) { |
5560 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5561 | } else { | |
3bfa5bcb JA |
5562 | struct async_poll *apoll = req->apoll; |
5563 | ||
d7718a9d | 5564 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5565 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5566 | if (do_complete) { | |
d7718a9d | 5567 | io_put_req(req); |
807abcb0 | 5568 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5569 | kfree(apoll); |
5570 | } | |
b1f573bd XW |
5571 | } |
5572 | ||
b41e9852 JA |
5573 | if (do_complete) { |
5574 | io_cqring_fill_event(req, -ECANCELED); | |
5575 | io_commit_cqring(req->ctx); | |
f254ac04 | 5576 | req_set_fail_links(req); |
216578e5 | 5577 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5578 | } |
5579 | ||
5580 | return do_complete; | |
221c5eb2 JA |
5581 | } |
5582 | ||
76e1b642 JA |
5583 | /* |
5584 | * Returns true if we found and killed one or more poll requests | |
5585 | */ | |
6b81928d PB |
5586 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5587 | struct files_struct *files) | |
221c5eb2 | 5588 | { |
78076bb6 | 5589 | struct hlist_node *tmp; |
221c5eb2 | 5590 | struct io_kiocb *req; |
8e2e1faf | 5591 | int posted = 0, i; |
221c5eb2 JA |
5592 | |
5593 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5594 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5595 | struct hlist_head *list; | |
5596 | ||
5597 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5598 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5599 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5600 | posted += io_poll_remove_one(req); |
5601 | } | |
221c5eb2 JA |
5602 | } |
5603 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5604 | |
8e2e1faf JA |
5605 | if (posted) |
5606 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5607 | |
5608 | return posted != 0; | |
221c5eb2 JA |
5609 | } |
5610 | ||
47f46768 JA |
5611 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5612 | { | |
78076bb6 | 5613 | struct hlist_head *list; |
47f46768 JA |
5614 | struct io_kiocb *req; |
5615 | ||
78076bb6 JA |
5616 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5617 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5618 | if (sqe_addr != req->user_data) |
5619 | continue; | |
5620 | if (io_poll_remove_one(req)) | |
eac406c6 | 5621 | return 0; |
b41e9852 | 5622 | return -EALREADY; |
47f46768 JA |
5623 | } |
5624 | ||
5625 | return -ENOENT; | |
5626 | } | |
5627 | ||
3529d8c2 JA |
5628 | static int io_poll_remove_prep(struct io_kiocb *req, |
5629 | const struct io_uring_sqe *sqe) | |
0969e783 | 5630 | { |
0969e783 JA |
5631 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5632 | return -EINVAL; | |
5633 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5634 | sqe->poll_events) | |
5635 | return -EINVAL; | |
5636 | ||
018043be | 5637 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5638 | return 0; |
5639 | } | |
5640 | ||
221c5eb2 JA |
5641 | /* |
5642 | * Find a running poll command that matches one specified in sqe->addr, | |
5643 | * and remove it if found. | |
5644 | */ | |
61e98203 | 5645 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5646 | { |
5647 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5648 | int ret; |
221c5eb2 | 5649 | |
221c5eb2 | 5650 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5651 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5652 | spin_unlock_irq(&ctx->completion_lock); |
5653 | ||
4e88d6e7 JA |
5654 | if (ret < 0) |
5655 | req_set_fail_links(req); | |
e1e16097 | 5656 | io_req_complete(req, ret); |
221c5eb2 JA |
5657 | return 0; |
5658 | } | |
5659 | ||
221c5eb2 JA |
5660 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5661 | void *key) | |
5662 | { | |
c2f2eb7d JA |
5663 | struct io_kiocb *req = wait->private; |
5664 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5665 | |
d7718a9d | 5666 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5667 | } |
5668 | ||
221c5eb2 JA |
5669 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5670 | struct poll_table_struct *p) | |
5671 | { | |
5672 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5673 | ||
e8c2bc1f | 5674 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5675 | } |
5676 | ||
3529d8c2 | 5677 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5678 | { |
5679 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5680 | u32 events; |
221c5eb2 JA |
5681 | |
5682 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5683 | return -EINVAL; | |
5684 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5685 | return -EINVAL; | |
5686 | ||
5769a351 JX |
5687 | events = READ_ONCE(sqe->poll32_events); |
5688 | #ifdef __BIG_ENDIAN | |
5689 | events = swahw32(events); | |
5690 | #endif | |
a31eb4a2 JX |
5691 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5692 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5693 | return 0; |
5694 | } | |
5695 | ||
61e98203 | 5696 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5697 | { |
5698 | struct io_poll_iocb *poll = &req->poll; | |
5699 | struct io_ring_ctx *ctx = req->ctx; | |
5700 | struct io_poll_table ipt; | |
0969e783 | 5701 | __poll_t mask; |
0969e783 | 5702 | |
d7718a9d | 5703 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5704 | |
d7718a9d JA |
5705 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5706 | io_poll_wake); | |
221c5eb2 | 5707 | |
8c838788 | 5708 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5709 | ipt.error = 0; |
b0dd8a41 | 5710 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5711 | } |
221c5eb2 JA |
5712 | spin_unlock_irq(&ctx->completion_lock); |
5713 | ||
8c838788 JA |
5714 | if (mask) { |
5715 | io_cqring_ev_posted(ctx); | |
014db007 | 5716 | io_put_req(req); |
221c5eb2 | 5717 | } |
8c838788 | 5718 | return ipt.error; |
221c5eb2 JA |
5719 | } |
5720 | ||
5262f567 JA |
5721 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5722 | { | |
ad8a48ac JA |
5723 | struct io_timeout_data *data = container_of(timer, |
5724 | struct io_timeout_data, timer); | |
5725 | struct io_kiocb *req = data->req; | |
5726 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5727 | unsigned long flags; |
5728 | ||
5262f567 | 5729 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5730 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5731 | atomic_set(&req->ctx->cq_timeouts, |
5732 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5733 | ||
78e19bbe | 5734 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5735 | io_commit_cqring(ctx); |
5736 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5737 | ||
5738 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5739 | req_set_fail_links(req); |
5262f567 JA |
5740 | io_put_req(req); |
5741 | return HRTIMER_NORESTART; | |
5742 | } | |
5743 | ||
fbd15848 PB |
5744 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5745 | __u64 user_data) | |
f254ac04 | 5746 | { |
fbd15848 | 5747 | struct io_timeout_data *io; |
47f46768 JA |
5748 | struct io_kiocb *req; |
5749 | int ret = -ENOENT; | |
f254ac04 | 5750 | |
135fcde8 | 5751 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5752 | if (user_data == req->user_data) { |
47f46768 JA |
5753 | ret = 0; |
5754 | break; | |
5755 | } | |
5756 | } | |
5757 | ||
5758 | if (ret == -ENOENT) | |
fbd15848 PB |
5759 | return ERR_PTR(ret); |
5760 | ||
5761 | io = req->async_data; | |
e8c2bc1f | 5762 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5763 | if (ret == -1) |
fbd15848 | 5764 | return ERR_PTR(-EALREADY); |
a71976f3 | 5765 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5766 | return req; |
5767 | } | |
47f46768 | 5768 | |
fbd15848 PB |
5769 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5770 | { | |
5771 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5772 | ||
5773 | if (IS_ERR(req)) | |
5774 | return PTR_ERR(req); | |
f254ac04 JA |
5775 | |
5776 | req_set_fail_links(req); | |
f254ac04 | 5777 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5778 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5779 | return 0; |
5780 | } | |
5781 | ||
9c8e11b3 PB |
5782 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5783 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5784 | { |
9c8e11b3 PB |
5785 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5786 | struct io_timeout_data *data; | |
47f46768 | 5787 | |
9c8e11b3 PB |
5788 | if (IS_ERR(req)) |
5789 | return PTR_ERR(req); | |
47f46768 | 5790 | |
9c8e11b3 PB |
5791 | req->timeout.off = 0; /* noseq */ |
5792 | data = req->async_data; | |
5793 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5794 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5795 | data->timer.function = io_timeout_fn; | |
5796 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5797 | return 0; | |
47f46768 JA |
5798 | } |
5799 | ||
3529d8c2 JA |
5800 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5801 | const struct io_uring_sqe *sqe) | |
b29472ee | 5802 | { |
9c8e11b3 PB |
5803 | struct io_timeout_rem *tr = &req->timeout_rem; |
5804 | ||
b29472ee JA |
5805 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5806 | return -EINVAL; | |
61710e43 DA |
5807 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5808 | return -EINVAL; | |
9c8e11b3 | 5809 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5810 | return -EINVAL; |
5811 | ||
9c8e11b3 PB |
5812 | tr->addr = READ_ONCE(sqe->addr); |
5813 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5814 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5815 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5816 | return -EINVAL; | |
5817 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5818 | return -EFAULT; | |
5819 | } else if (tr->flags) { | |
5820 | /* timeout removal doesn't support flags */ | |
b29472ee | 5821 | return -EINVAL; |
9c8e11b3 | 5822 | } |
b29472ee | 5823 | |
b29472ee JA |
5824 | return 0; |
5825 | } | |
5826 | ||
8662daec PB |
5827 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5828 | { | |
5829 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5830 | : HRTIMER_MODE_REL; | |
5831 | } | |
5832 | ||
11365043 JA |
5833 | /* |
5834 | * Remove or update an existing timeout command | |
5835 | */ | |
61e98203 | 5836 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5837 | { |
9c8e11b3 | 5838 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5839 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5840 | int ret; |
11365043 | 5841 | |
11365043 | 5842 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5843 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5844 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5845 | else |
5846 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5847 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5848 | |
47f46768 | 5849 | io_cqring_fill_event(req, ret); |
11365043 JA |
5850 | io_commit_cqring(ctx); |
5851 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5852 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5853 | if (ret < 0) |
5854 | req_set_fail_links(req); | |
ec9c02ad | 5855 | io_put_req(req); |
11365043 | 5856 | return 0; |
5262f567 JA |
5857 | } |
5858 | ||
3529d8c2 | 5859 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5860 | bool is_timeout_link) |
5262f567 | 5861 | { |
ad8a48ac | 5862 | struct io_timeout_data *data; |
a41525ab | 5863 | unsigned flags; |
56080b02 | 5864 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5865 | |
ad8a48ac | 5866 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5867 | return -EINVAL; |
ad8a48ac | 5868 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5869 | return -EINVAL; |
56080b02 | 5870 | if (off && is_timeout_link) |
2d28390a | 5871 | return -EINVAL; |
a41525ab JA |
5872 | flags = READ_ONCE(sqe->timeout_flags); |
5873 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5874 | return -EINVAL; |
bdf20073 | 5875 | |
bfe68a22 | 5876 | req->timeout.off = off; |
26a61679 | 5877 | |
e8c2bc1f | 5878 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5879 | return -ENOMEM; |
5880 | ||
e8c2bc1f | 5881 | data = req->async_data; |
ad8a48ac | 5882 | data->req = req; |
ad8a48ac JA |
5883 | |
5884 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5885 | return -EFAULT; |
5886 | ||
8662daec | 5887 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5888 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5889 | return 0; | |
5890 | } | |
5891 | ||
61e98203 | 5892 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5893 | { |
ad8a48ac | 5894 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5895 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5896 | struct list_head *entry; |
bfe68a22 | 5897 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5898 | |
733f5c95 | 5899 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5900 | |
5262f567 JA |
5901 | /* |
5902 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5903 | * timeout event to be satisfied. If it isn't set, then this is |
5904 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5905 | */ |
8eb7e2d0 | 5906 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5907 | entry = ctx->timeout_list.prev; |
5908 | goto add; | |
5909 | } | |
5262f567 | 5910 | |
bfe68a22 PB |
5911 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5912 | req->timeout.target_seq = tail + off; | |
5262f567 | 5913 | |
f010505b MDG |
5914 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5915 | * This is safe because ->completion_lock is held, and submissions | |
5916 | * and completions are never mixed in the same ->completion_lock section. | |
5917 | */ | |
5918 | ctx->cq_last_tm_flush = tail; | |
5919 | ||
5262f567 JA |
5920 | /* |
5921 | * Insertion sort, ensuring the first entry in the list is always | |
5922 | * the one we need first. | |
5923 | */ | |
5262f567 | 5924 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5925 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5926 | timeout.list); | |
5262f567 | 5927 | |
8eb7e2d0 | 5928 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5929 | continue; |
bfe68a22 PB |
5930 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5931 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5932 | break; |
5933 | } | |
93bd25bb | 5934 | add: |
135fcde8 | 5935 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5936 | data->timer.function = io_timeout_fn; |
5937 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5938 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5939 | return 0; |
5940 | } | |
5262f567 | 5941 | |
62755e35 JA |
5942 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5943 | { | |
5944 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5945 | ||
5946 | return req->user_data == (unsigned long) data; | |
5947 | } | |
5948 | ||
e977d6d3 | 5949 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5950 | { |
62755e35 | 5951 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5952 | int ret = 0; |
5953 | ||
4f26bda1 | 5954 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5955 | switch (cancel_ret) { |
5956 | case IO_WQ_CANCEL_OK: | |
5957 | ret = 0; | |
5958 | break; | |
5959 | case IO_WQ_CANCEL_RUNNING: | |
5960 | ret = -EALREADY; | |
5961 | break; | |
5962 | case IO_WQ_CANCEL_NOTFOUND: | |
5963 | ret = -ENOENT; | |
5964 | break; | |
5965 | } | |
5966 | ||
e977d6d3 JA |
5967 | return ret; |
5968 | } | |
5969 | ||
47f46768 JA |
5970 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5971 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5972 | int success_ret) |
47f46768 JA |
5973 | { |
5974 | unsigned long flags; | |
5975 | int ret; | |
5976 | ||
5977 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5978 | if (ret != -ENOENT) { | |
5979 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5980 | goto done; | |
5981 | } | |
5982 | ||
5983 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5984 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5985 | if (ret != -ENOENT) | |
5986 | goto done; | |
5987 | ret = io_poll_cancel(ctx, sqe_addr); | |
5988 | done: | |
b0dd8a41 JA |
5989 | if (!ret) |
5990 | ret = success_ret; | |
47f46768 JA |
5991 | io_cqring_fill_event(req, ret); |
5992 | io_commit_cqring(ctx); | |
5993 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5994 | io_cqring_ev_posted(ctx); | |
5995 | ||
4e88d6e7 JA |
5996 | if (ret < 0) |
5997 | req_set_fail_links(req); | |
014db007 | 5998 | io_put_req(req); |
47f46768 JA |
5999 | } |
6000 | ||
3529d8c2 JA |
6001 | static int io_async_cancel_prep(struct io_kiocb *req, |
6002 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6003 | { |
fbf23849 | 6004 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6005 | return -EINVAL; |
61710e43 DA |
6006 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6007 | return -EINVAL; | |
6008 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
6009 | return -EINVAL; |
6010 | ||
fbf23849 JA |
6011 | req->cancel.addr = READ_ONCE(sqe->addr); |
6012 | return 0; | |
6013 | } | |
6014 | ||
61e98203 | 6015 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6016 | { |
6017 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 6018 | |
014db007 | 6019 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
6020 | return 0; |
6021 | } | |
6022 | ||
269bbe5f | 6023 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6024 | const struct io_uring_sqe *sqe) |
6025 | { | |
6ca56f84 JA |
6026 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
6027 | return -EINVAL; | |
61710e43 DA |
6028 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6029 | return -EINVAL; | |
6030 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
6031 | return -EINVAL; |
6032 | ||
269bbe5f BM |
6033 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6034 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6035 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6036 | return -EINVAL; |
269bbe5f | 6037 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6038 | return 0; |
6039 | } | |
6040 | ||
889fca73 | 6041 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6042 | { |
6043 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 6044 | struct io_uring_rsrc_update up; |
05f3fb3c | 6045 | int ret; |
fbf23849 | 6046 | |
45d189c6 | 6047 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6048 | return -EAGAIN; |
05f3fb3c | 6049 | |
269bbe5f BM |
6050 | up.offset = req->rsrc_update.offset; |
6051 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
6052 | |
6053 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 6054 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6055 | mutex_unlock(&ctx->uring_lock); |
6056 | ||
6057 | if (ret < 0) | |
6058 | req_set_fail_links(req); | |
889fca73 | 6059 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6060 | return 0; |
6061 | } | |
6062 | ||
bfe76559 | 6063 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6064 | { |
d625c6ee | 6065 | switch (req->opcode) { |
e781573e | 6066 | case IORING_OP_NOP: |
bfe76559 | 6067 | return 0; |
f67676d1 JA |
6068 | case IORING_OP_READV: |
6069 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6070 | case IORING_OP_READ: |
bfe76559 | 6071 | return io_read_prep(req, sqe); |
f67676d1 JA |
6072 | case IORING_OP_WRITEV: |
6073 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6074 | case IORING_OP_WRITE: |
bfe76559 | 6075 | return io_write_prep(req, sqe); |
0969e783 | 6076 | case IORING_OP_POLL_ADD: |
bfe76559 | 6077 | return io_poll_add_prep(req, sqe); |
0969e783 | 6078 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 6079 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 6080 | case IORING_OP_FSYNC: |
bfe76559 | 6081 | return io_prep_fsync(req, sqe); |
8ed8d3c3 | 6082 | case IORING_OP_SYNC_FILE_RANGE: |
bfe76559 | 6083 | return io_prep_sfr(req, sqe); |
03b1230c | 6084 | case IORING_OP_SENDMSG: |
fddaface | 6085 | case IORING_OP_SEND: |
bfe76559 | 6086 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6087 | case IORING_OP_RECVMSG: |
fddaface | 6088 | case IORING_OP_RECV: |
bfe76559 | 6089 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6090 | case IORING_OP_CONNECT: |
bfe76559 | 6091 | return io_connect_prep(req, sqe); |
2d28390a | 6092 | case IORING_OP_TIMEOUT: |
bfe76559 | 6093 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6094 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6095 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6096 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6097 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6098 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6099 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6100 | case IORING_OP_ACCEPT: |
bfe76559 | 6101 | return io_accept_prep(req, sqe); |
d63d1b5e | 6102 | case IORING_OP_FALLOCATE: |
bfe76559 | 6103 | return io_fallocate_prep(req, sqe); |
15b71abe | 6104 | case IORING_OP_OPENAT: |
bfe76559 | 6105 | return io_openat_prep(req, sqe); |
b5dba59e | 6106 | case IORING_OP_CLOSE: |
bfe76559 | 6107 | return io_close_prep(req, sqe); |
05f3fb3c | 6108 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6109 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6110 | case IORING_OP_STATX: |
bfe76559 | 6111 | return io_statx_prep(req, sqe); |
4840e418 | 6112 | case IORING_OP_FADVISE: |
bfe76559 | 6113 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6114 | case IORING_OP_MADVISE: |
bfe76559 | 6115 | return io_madvise_prep(req, sqe); |
cebdb986 | 6116 | case IORING_OP_OPENAT2: |
bfe76559 | 6117 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6118 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6119 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6120 | case IORING_OP_SPLICE: |
bfe76559 | 6121 | return io_splice_prep(req, sqe); |
ddf0322d | 6122 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6123 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6124 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6125 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6126 | case IORING_OP_TEE: |
bfe76559 | 6127 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6128 | case IORING_OP_SHUTDOWN: |
6129 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6130 | case IORING_OP_RENAMEAT: |
6131 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6132 | case IORING_OP_UNLINKAT: |
6133 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6134 | } |
6135 | ||
bfe76559 PB |
6136 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6137 | req->opcode); | |
6138 | return-EINVAL; | |
6139 | } | |
6140 | ||
6141 | static int io_req_defer_prep(struct io_kiocb *req, | |
6142 | const struct io_uring_sqe *sqe) | |
6143 | { | |
bfe76559 PB |
6144 | if (!sqe) |
6145 | return 0; | |
6146 | if (io_alloc_async_data(req)) | |
6147 | return -EAGAIN; | |
bfe76559 | 6148 | return io_req_prep(req, sqe); |
f67676d1 JA |
6149 | } |
6150 | ||
9cf7c104 PB |
6151 | static u32 io_get_sequence(struct io_kiocb *req) |
6152 | { | |
6153 | struct io_kiocb *pos; | |
6154 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 6155 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 6156 | |
f2f87370 PB |
6157 | io_for_each_link(pos, req) |
6158 | nr_reqs++; | |
9cf7c104 PB |
6159 | |
6160 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6161 | return total_submitted - nr_reqs; | |
6162 | } | |
6163 | ||
3529d8c2 | 6164 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 6165 | { |
a197f664 | 6166 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6167 | struct io_defer_entry *de; |
f67676d1 | 6168 | int ret; |
9cf7c104 | 6169 | u32 seq; |
de0617e4 | 6170 | |
9d858b21 | 6171 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6172 | if (likely(list_empty_careful(&ctx->defer_list) && |
6173 | !(req->flags & REQ_F_IO_DRAIN))) | |
6174 | return 0; | |
6175 | ||
6176 | seq = io_get_sequence(req); | |
6177 | /* Still a chance to pass the sequence check */ | |
6178 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6179 | return 0; |
6180 | ||
e8c2bc1f | 6181 | if (!req->async_data) { |
650b5481 | 6182 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6183 | if (ret) |
650b5481 PB |
6184 | return ret; |
6185 | } | |
cbdcb435 | 6186 | io_prep_async_link(req); |
27dc8338 PB |
6187 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6188 | if (!de) | |
6189 | return -ENOMEM; | |
2d28390a | 6190 | |
de0617e4 | 6191 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6192 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6193 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6194 | kfree(de); |
ae34817b PB |
6195 | io_queue_async_work(req); |
6196 | return -EIOCBQUEUED; | |
de0617e4 JA |
6197 | } |
6198 | ||
915967f6 | 6199 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6200 | de->req = req; |
9cf7c104 | 6201 | de->seq = seq; |
27dc8338 | 6202 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6203 | spin_unlock_irq(&ctx->completion_lock); |
6204 | return -EIOCBQUEUED; | |
6205 | } | |
6206 | ||
3ca405eb | 6207 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6208 | { |
0e1b6fe3 PB |
6209 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6210 | switch (req->opcode) { | |
6211 | case IORING_OP_READV: | |
6212 | case IORING_OP_READ_FIXED: | |
6213 | case IORING_OP_READ: | |
bcda7baa | 6214 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6215 | break; |
6216 | case IORING_OP_RECVMSG: | |
6217 | case IORING_OP_RECV: | |
bcda7baa | 6218 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6219 | break; |
6220 | } | |
6221 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6222 | } |
6223 | ||
0e1b6fe3 PB |
6224 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6225 | switch (req->opcode) { | |
6226 | case IORING_OP_READV: | |
6227 | case IORING_OP_READ_FIXED: | |
6228 | case IORING_OP_READ: | |
6229 | case IORING_OP_WRITEV: | |
6230 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6231 | case IORING_OP_WRITE: { |
6232 | struct io_async_rw *io = req->async_data; | |
6233 | if (io->free_iovec) | |
6234 | kfree(io->free_iovec); | |
0e1b6fe3 | 6235 | break; |
e8c2bc1f | 6236 | } |
0e1b6fe3 | 6237 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6238 | case IORING_OP_SENDMSG: { |
6239 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6240 | |
6241 | kfree(io->free_iov); | |
0e1b6fe3 | 6242 | break; |
e8c2bc1f | 6243 | } |
0e1b6fe3 PB |
6244 | case IORING_OP_SPLICE: |
6245 | case IORING_OP_TEE: | |
6246 | io_put_file(req, req->splice.file_in, | |
6247 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6248 | break; | |
f3cd4850 JA |
6249 | case IORING_OP_OPENAT: |
6250 | case IORING_OP_OPENAT2: | |
6251 | if (req->open.filename) | |
6252 | putname(req->open.filename); | |
6253 | break; | |
80a261fd JA |
6254 | case IORING_OP_RENAMEAT: |
6255 | putname(req->rename.oldpath); | |
6256 | putname(req->rename.newpath); | |
6257 | break; | |
14a1143b JA |
6258 | case IORING_OP_UNLINKAT: |
6259 | putname(req->unlink.filename); | |
6260 | break; | |
0e1b6fe3 PB |
6261 | } |
6262 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6263 | } |
99bc4c38 PB |
6264 | } |
6265 | ||
889fca73 | 6266 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6267 | { |
a197f664 | 6268 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 6269 | int ret; |
2b188cc1 | 6270 | |
d625c6ee | 6271 | switch (req->opcode) { |
2b188cc1 | 6272 | case IORING_OP_NOP: |
889fca73 | 6273 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6274 | break; |
6275 | case IORING_OP_READV: | |
edafccee | 6276 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6277 | case IORING_OP_READ: |
889fca73 | 6278 | ret = io_read(req, issue_flags); |
edafccee | 6279 | break; |
3529d8c2 | 6280 | case IORING_OP_WRITEV: |
edafccee | 6281 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6282 | case IORING_OP_WRITE: |
889fca73 | 6283 | ret = io_write(req, issue_flags); |
2b188cc1 | 6284 | break; |
c992fe29 | 6285 | case IORING_OP_FSYNC: |
45d189c6 | 6286 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6287 | break; |
221c5eb2 | 6288 | case IORING_OP_POLL_ADD: |
61e98203 | 6289 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6290 | break; |
6291 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6292 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6293 | break; |
5d17b4a4 | 6294 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6295 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6296 | break; |
0fa03c62 | 6297 | case IORING_OP_SENDMSG: |
889fca73 | 6298 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6299 | break; |
fddaface | 6300 | case IORING_OP_SEND: |
889fca73 | 6301 | ret = io_send(req, issue_flags); |
0fa03c62 | 6302 | break; |
aa1fa28f | 6303 | case IORING_OP_RECVMSG: |
889fca73 | 6304 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6305 | break; |
fddaface | 6306 | case IORING_OP_RECV: |
889fca73 | 6307 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6308 | break; |
5262f567 | 6309 | case IORING_OP_TIMEOUT: |
61e98203 | 6310 | ret = io_timeout(req, issue_flags); |
5262f567 | 6311 | break; |
11365043 | 6312 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6313 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6314 | break; |
17f2fe35 | 6315 | case IORING_OP_ACCEPT: |
889fca73 | 6316 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6317 | break; |
f8e85cf2 | 6318 | case IORING_OP_CONNECT: |
889fca73 | 6319 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6320 | break; |
62755e35 | 6321 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6322 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6323 | break; |
d63d1b5e | 6324 | case IORING_OP_FALLOCATE: |
45d189c6 | 6325 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6326 | break; |
15b71abe | 6327 | case IORING_OP_OPENAT: |
45d189c6 | 6328 | ret = io_openat(req, issue_flags); |
15b71abe | 6329 | break; |
b5dba59e | 6330 | case IORING_OP_CLOSE: |
889fca73 | 6331 | ret = io_close(req, issue_flags); |
b5dba59e | 6332 | break; |
05f3fb3c | 6333 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6334 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6335 | break; |
eddc7ef5 | 6336 | case IORING_OP_STATX: |
45d189c6 | 6337 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6338 | break; |
4840e418 | 6339 | case IORING_OP_FADVISE: |
45d189c6 | 6340 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6341 | break; |
c1ca757b | 6342 | case IORING_OP_MADVISE: |
45d189c6 | 6343 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6344 | break; |
cebdb986 | 6345 | case IORING_OP_OPENAT2: |
45d189c6 | 6346 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6347 | break; |
3e4827b0 | 6348 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6349 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6350 | break; |
7d67af2c | 6351 | case IORING_OP_SPLICE: |
45d189c6 | 6352 | ret = io_splice(req, issue_flags); |
7d67af2c | 6353 | break; |
ddf0322d | 6354 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6355 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6356 | break; |
067524e9 | 6357 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6358 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6359 | break; |
f2a8d5c7 | 6360 | case IORING_OP_TEE: |
45d189c6 | 6361 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6362 | break; |
36f4fa68 | 6363 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6364 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6365 | break; |
80a261fd | 6366 | case IORING_OP_RENAMEAT: |
45d189c6 | 6367 | ret = io_renameat(req, issue_flags); |
80a261fd | 6368 | break; |
14a1143b | 6369 | case IORING_OP_UNLINKAT: |
45d189c6 | 6370 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6371 | break; |
2b188cc1 JA |
6372 | default: |
6373 | ret = -EINVAL; | |
6374 | break; | |
6375 | } | |
6376 | ||
def596e9 JA |
6377 | if (ret) |
6378 | return ret; | |
6379 | ||
b532576e JA |
6380 | /* If the op doesn't have a file, we're not polling for it */ |
6381 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6382 | const bool in_async = io_wq_current_is_worker(); |
6383 | ||
11ba820b JA |
6384 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6385 | if (in_async) | |
6386 | mutex_lock(&ctx->uring_lock); | |
6387 | ||
2e9dbe90 | 6388 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6389 | |
6390 | if (in_async) | |
6391 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6392 | } |
6393 | ||
6394 | return 0; | |
2b188cc1 JA |
6395 | } |
6396 | ||
5280f7e5 | 6397 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6398 | { |
6399 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6400 | struct io_kiocb *timeout; |
561fb04a | 6401 | int ret = 0; |
2b188cc1 | 6402 | |
6df1db6b PB |
6403 | timeout = io_prep_linked_timeout(req); |
6404 | if (timeout) | |
6405 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6406 | |
4014d943 | 6407 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6408 | ret = -ECANCELED; |
31b51510 | 6409 | |
561fb04a | 6410 | if (!ret) { |
561fb04a | 6411 | do { |
889fca73 | 6412 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6413 | /* |
6414 | * We can get EAGAIN for polled IO even though we're | |
6415 | * forcing a sync submission from here, since we can't | |
6416 | * wait for request slots on the block side. | |
6417 | */ | |
6418 | if (ret != -EAGAIN) | |
6419 | break; | |
6420 | cond_resched(); | |
6421 | } while (1); | |
6422 | } | |
31b51510 | 6423 | |
561fb04a | 6424 | if (ret) { |
c07e6719 XW |
6425 | struct io_ring_ctx *lock_ctx = NULL; |
6426 | ||
6427 | if (req->ctx->flags & IORING_SETUP_IOPOLL) | |
6428 | lock_ctx = req->ctx; | |
6429 | ||
dad1b124 | 6430 | /* |
c07e6719 XW |
6431 | * io_iopoll_complete() does not hold completion_lock to |
6432 | * complete polled io, so here for polled io, we can not call | |
6433 | * io_req_complete() directly, otherwise there maybe concurrent | |
6434 | * access to cqring, defer_list, etc, which is not safe. Given | |
6435 | * that io_iopoll_complete() is always called under uring_lock, | |
6436 | * so here for polled io, we also get uring_lock to complete | |
6437 | * it. | |
dad1b124 | 6438 | */ |
c07e6719 XW |
6439 | if (lock_ctx) |
6440 | mutex_lock(&lock_ctx->uring_lock); | |
dad1b124 | 6441 | |
c07e6719 XW |
6442 | req_set_fail_links(req); |
6443 | io_req_complete(req, ret); | |
6444 | ||
6445 | if (lock_ctx) | |
6446 | mutex_unlock(&lock_ctx->uring_lock); | |
edafccee | 6447 | } |
2b188cc1 JA |
6448 | } |
6449 | ||
65e19f54 JA |
6450 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6451 | int index) | |
6452 | { | |
269bbe5f | 6453 | struct fixed_rsrc_table *table; |
65e19f54 | 6454 | |
05f3fb3c | 6455 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6456 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6457 | } |
6458 | ||
8371adf5 PB |
6459 | static struct file *io_file_get(struct io_submit_state *state, |
6460 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6461 | { |
a197f664 | 6462 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6463 | struct file *file; |
09bb8394 | 6464 | |
8da11c19 | 6465 | if (fixed) { |
479f517b | 6466 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6467 | return NULL; |
b7620121 | 6468 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6469 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6470 | io_set_resource_node(req); |
09bb8394 | 6471 | } else { |
c826bd7a | 6472 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6473 | file = __io_file_get(state, fd); |
09bb8394 JA |
6474 | } |
6475 | ||
ce3d5aae PB |
6476 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6477 | io_req_track_inflight(req); | |
8371adf5 | 6478 | return file; |
09bb8394 JA |
6479 | } |
6480 | ||
2665abfd | 6481 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6482 | { |
ad8a48ac JA |
6483 | struct io_timeout_data *data = container_of(timer, |
6484 | struct io_timeout_data, timer); | |
90cd7e42 | 6485 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6486 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6487 | unsigned long flags; |
2665abfd JA |
6488 | |
6489 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6490 | prev = req->timeout.head; |
6491 | req->timeout.head = NULL; | |
2665abfd JA |
6492 | |
6493 | /* | |
6494 | * We don't expect the list to be empty, that will only happen if we | |
6495 | * race with the completion of the linked work. | |
6496 | */ | |
90cd7e42 | 6497 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6498 | io_remove_next_linked(prev); |
90cd7e42 PB |
6499 | else |
6500 | prev = NULL; | |
2665abfd JA |
6501 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6502 | ||
6503 | if (prev) { | |
4e88d6e7 | 6504 | req_set_fail_links(prev); |
014db007 | 6505 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6506 | io_put_req_deferred(prev, 1); |
47f46768 | 6507 | } else { |
9ae1f8dd PB |
6508 | io_req_complete_post(req, -ETIME, 0); |
6509 | io_put_req_deferred(req, 1); | |
2665abfd | 6510 | } |
2665abfd JA |
6511 | return HRTIMER_NORESTART; |
6512 | } | |
6513 | ||
7271ef3a | 6514 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6515 | { |
76a46e06 | 6516 | /* |
f2f87370 PB |
6517 | * If the back reference is NULL, then our linked request finished |
6518 | * before we got a chance to setup the timer | |
76a46e06 | 6519 | */ |
90cd7e42 | 6520 | if (req->timeout.head) { |
e8c2bc1f | 6521 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6522 | |
ad8a48ac JA |
6523 | data->timer.function = io_link_timeout_fn; |
6524 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6525 | data->mode); | |
2665abfd | 6526 | } |
7271ef3a JA |
6527 | } |
6528 | ||
6529 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6530 | { | |
6531 | struct io_ring_ctx *ctx = req->ctx; | |
6532 | ||
6533 | spin_lock_irq(&ctx->completion_lock); | |
6534 | __io_queue_linked_timeout(req); | |
76a46e06 | 6535 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6536 | |
2665abfd | 6537 | /* drop submission reference */ |
76a46e06 JA |
6538 | io_put_req(req); |
6539 | } | |
2665abfd | 6540 | |
ad8a48ac | 6541 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6542 | { |
f2f87370 | 6543 | struct io_kiocb *nxt = req->link; |
2665abfd | 6544 | |
f2f87370 PB |
6545 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6546 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6547 | return NULL; |
2665abfd | 6548 | |
90cd7e42 | 6549 | nxt->timeout.head = req; |
900fad45 | 6550 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6551 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6552 | return nxt; |
2665abfd JA |
6553 | } |
6554 | ||
c5eef2b9 | 6555 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6556 | { |
d3d7298d | 6557 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
193155c8 | 6558 | const struct cred *old_creds = NULL; |
c5eef2b9 | 6559 | int ret; |
2b188cc1 | 6560 | |
2e5aa6cb PB |
6561 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6562 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
d3d7298d PB |
6563 | req->work.identity->creds != current_cred()) |
6564 | old_creds = override_creds(req->work.identity->creds); | |
193155c8 | 6565 | |
c5eef2b9 | 6566 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
491381ce | 6567 | |
d3d7298d PB |
6568 | if (old_creds) |
6569 | revert_creds(old_creds); | |
6570 | ||
491381ce JA |
6571 | /* |
6572 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6573 | * doesn't support non-blocking read/write attempts | |
6574 | */ | |
24c74678 | 6575 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6576 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6577 | /* |
6578 | * Queued up for async execution, worker will release | |
6579 | * submit reference when the iocb is actually submitted. | |
6580 | */ | |
6581 | io_queue_async_work(req); | |
2b188cc1 | 6582 | } |
0d63c148 PB |
6583 | } else if (likely(!ret)) { |
6584 | /* drop submission reference */ | |
e342c807 | 6585 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6586 | struct io_ring_ctx *ctx = req->ctx; |
6587 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
6588 | ||
6dd0be1e | 6589 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6590 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6591 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6592 | } else { |
d3d7298d | 6593 | io_put_req(req); |
0d63c148 PB |
6594 | } |
6595 | } else { | |
4e88d6e7 | 6596 | req_set_fail_links(req); |
e65ef56d | 6597 | io_put_req(req); |
e1e16097 | 6598 | io_req_complete(req, ret); |
9e645e11 | 6599 | } |
d3d7298d PB |
6600 | if (linked_timeout) |
6601 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6602 | } |
6603 | ||
c5eef2b9 | 6604 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
6605 | { |
6606 | int ret; | |
6607 | ||
3529d8c2 | 6608 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6609 | if (ret) { |
6610 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6611 | fail_req: |
4e88d6e7 | 6612 | req_set_fail_links(req); |
e1e16097 JA |
6613 | io_put_req(req); |
6614 | io_req_complete(req, ret); | |
4fe2c963 | 6615 | } |
2550878f | 6616 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6617 | if (!req->async_data) { |
bd2ab18a | 6618 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6619 | if (unlikely(ret)) |
bd2ab18a PB |
6620 | goto fail_req; |
6621 | } | |
ce35a47a JA |
6622 | io_queue_async_work(req); |
6623 | } else { | |
c1379e24 PB |
6624 | if (sqe) { |
6625 | ret = io_req_prep(req, sqe); | |
6626 | if (unlikely(ret)) | |
6627 | goto fail_req; | |
6628 | } | |
c5eef2b9 | 6629 | __io_queue_sqe(req); |
ce35a47a | 6630 | } |
4fe2c963 JL |
6631 | } |
6632 | ||
c5eef2b9 | 6633 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 6634 | { |
94ae5e77 | 6635 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6636 | io_put_req(req); |
6637 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6638 | } else |
c5eef2b9 | 6639 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
6640 | } |
6641 | ||
863e0560 PB |
6642 | struct io_submit_link { |
6643 | struct io_kiocb *head; | |
6644 | struct io_kiocb *last; | |
6645 | }; | |
6646 | ||
1d4240cc | 6647 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
c5eef2b9 | 6648 | struct io_submit_link *link) |
9e645e11 | 6649 | { |
a197f664 | 6650 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6651 | int ret; |
9e645e11 | 6652 | |
9e645e11 JA |
6653 | /* |
6654 | * If we already have a head request, queue this one for async | |
6655 | * submittal once the head completes. If we don't have a head but | |
6656 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6657 | * submitted sync once the chain is complete. If none of those | |
6658 | * conditions are true (normal request), then just queue it. | |
6659 | */ | |
863e0560 PB |
6660 | if (link->head) { |
6661 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6662 | |
8cdf2193 PB |
6663 | /* |
6664 | * Taking sequential execution of a link, draining both sides | |
6665 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6666 | * requests in the link. So, it drains the head and the | |
6667 | * next after the link request. The last one is done via | |
6668 | * drain_next flag to persist the effect across calls. | |
6669 | */ | |
ef4ff581 | 6670 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6671 | head->flags |= REQ_F_IO_DRAIN; |
6672 | ctx->drain_next = 1; | |
6673 | } | |
3529d8c2 | 6674 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6675 | if (unlikely(ret)) { |
4e88d6e7 | 6676 | /* fail even hard links since we don't submit */ |
9d76377f | 6677 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6678 | return ret; |
2d28390a | 6679 | } |
9d76377f | 6680 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6681 | link->last->link = req; |
863e0560 | 6682 | link->last = req; |
32fe525b PB |
6683 | |
6684 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6685 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
c5eef2b9 | 6686 | io_queue_link_head(head); |
863e0560 | 6687 | link->head = NULL; |
32fe525b | 6688 | } |
9e645e11 | 6689 | } else { |
711be031 PB |
6690 | if (unlikely(ctx->drain_next)) { |
6691 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6692 | ctx->drain_next = 0; |
711be031 | 6693 | } |
ef4ff581 | 6694 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
711be031 | 6695 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6696 | if (unlikely(ret)) |
711be031 | 6697 | req->flags |= REQ_F_FAIL_LINK; |
863e0560 PB |
6698 | link->head = req; |
6699 | link->last = req; | |
711be031 | 6700 | } else { |
c5eef2b9 | 6701 | io_queue_sqe(req, sqe); |
711be031 | 6702 | } |
9e645e11 | 6703 | } |
2e6e1fde | 6704 | |
1d4240cc | 6705 | return 0; |
9e645e11 JA |
6706 | } |
6707 | ||
9a56a232 JA |
6708 | /* |
6709 | * Batched submission is done, ensure local IO is flushed out. | |
6710 | */ | |
ba88ff11 PB |
6711 | static void io_submit_state_end(struct io_submit_state *state, |
6712 | struct io_ring_ctx *ctx) | |
9a56a232 | 6713 | { |
6dd0be1e | 6714 | if (state->comp.nr) |
ba88ff11 | 6715 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6716 | if (state->plug_started) |
6717 | blk_finish_plug(&state->plug); | |
9f13c35b | 6718 | io_state_file_put(state); |
9a56a232 JA |
6719 | } |
6720 | ||
6721 | /* | |
6722 | * Start submission side cache. | |
6723 | */ | |
6724 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6725 | unsigned int max_ios) |
9a56a232 | 6726 | { |
27926b68 | 6727 | state->plug_started = false; |
9a56a232 JA |
6728 | state->ios_left = max_ios; |
6729 | } | |
6730 | ||
2b188cc1 JA |
6731 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6732 | { | |
75b28aff | 6733 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6734 | |
caf582c6 PB |
6735 | /* |
6736 | * Ensure any loads from the SQEs are done at this point, | |
6737 | * since once we write the new head, the application could | |
6738 | * write new data to them. | |
6739 | */ | |
6740 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6741 | } |
6742 | ||
2b188cc1 | 6743 | /* |
3529d8c2 | 6744 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6745 | * that is mapped by userspace. This means that care needs to be taken to |
6746 | * ensure that reads are stable, as we cannot rely on userspace always | |
6747 | * being a good citizen. If members of the sqe are validated and then later | |
6748 | * used, it's important that those reads are done through READ_ONCE() to | |
6749 | * prevent a re-load down the line. | |
6750 | */ | |
709b302f | 6751 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6752 | { |
75b28aff | 6753 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6754 | unsigned head; |
6755 | ||
6756 | /* | |
6757 | * The cached sq head (or cq tail) serves two purposes: | |
6758 | * | |
6759 | * 1) allows us to batch the cost of updating the user visible | |
6760 | * head updates. | |
6761 | * 2) allows the kernel side to track the head on its own, even | |
6762 | * though the application is the one updating it. | |
6763 | */ | |
4fccfcbb | 6764 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6765 | if (likely(head < ctx->sq_entries)) |
6766 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6767 | |
6768 | /* drop invalid entries */ | |
498ccd9e | 6769 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6770 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6771 | return NULL; |
6772 | } | |
6773 | ||
21b55dbc SG |
6774 | /* |
6775 | * Check SQE restrictions (opcode and flags). | |
6776 | * | |
6777 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6778 | */ | |
6779 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6780 | struct io_kiocb *req, | |
6781 | unsigned int sqe_flags) | |
6782 | { | |
6783 | if (!ctx->restricted) | |
6784 | return true; | |
6785 | ||
6786 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6787 | return false; | |
6788 | ||
6789 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6790 | ctx->restrictions.sqe_flags_required) | |
6791 | return false; | |
6792 | ||
6793 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6794 | ctx->restrictions.sqe_flags_required)) | |
6795 | return false; | |
6796 | ||
6797 | return true; | |
6798 | } | |
6799 | ||
ef4ff581 PB |
6800 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
6801 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
6802 | IOSQE_BUFFER_SELECT) | |
6803 | ||
6804 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
258b29a9 | 6805 | const struct io_uring_sqe *sqe) |
0553b8bd | 6806 | { |
258b29a9 | 6807 | struct io_submit_state *state; |
ef4ff581 | 6808 | unsigned int sqe_flags; |
71b547c0 | 6809 | int id, ret; |
ef4ff581 | 6810 | |
0553b8bd PB |
6811 | req->opcode = READ_ONCE(sqe->opcode); |
6812 | req->user_data = READ_ONCE(sqe->user_data); | |
e8c2bc1f | 6813 | req->async_data = NULL; |
0553b8bd PB |
6814 | req->file = NULL; |
6815 | req->ctx = ctx; | |
6816 | req->flags = 0; | |
f2f87370 | 6817 | req->link = NULL; |
269bbe5f | 6818 | req->fixed_rsrc_refs = NULL; |
0553b8bd PB |
6819 | /* one is dropped after submission, the other at completion */ |
6820 | refcount_set(&req->refs, 2); | |
4dd2824d | 6821 | req->task = current; |
0553b8bd | 6822 | req->result = 0; |
ef4ff581 PB |
6823 | |
6824 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6825 | return -EINVAL; | |
6826 | ||
28cea78a | 6827 | if (unlikely(io_sq_thread_acquire_mm_files(ctx, req))) |
9d8426a0 | 6828 | return -EFAULT; |
ef4ff581 PB |
6829 | |
6830 | sqe_flags = READ_ONCE(sqe->flags); | |
6831 | /* enforce forwards compatibility on users */ | |
6832 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6833 | return -EINVAL; | |
6834 | ||
21b55dbc SG |
6835 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6836 | return -EACCES; | |
6837 | ||
ef4ff581 PB |
6838 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && |
6839 | !io_op_defs[req->opcode].buffer_select) | |
6840 | return -EOPNOTSUPP; | |
6841 | ||
6842 | id = READ_ONCE(sqe->personality); | |
6843 | if (id) { | |
1e6fa521 JA |
6844 | struct io_identity *iod; |
6845 | ||
1e6fa521 JA |
6846 | iod = idr_find(&ctx->personality_idr, id); |
6847 | if (unlikely(!iod)) | |
ef4ff581 | 6848 | return -EINVAL; |
1e6fa521 | 6849 | refcount_inc(&iod->count); |
ec99ca6c PB |
6850 | |
6851 | __io_req_init_async(req); | |
1e6fa521 JA |
6852 | get_cred(iod->creds); |
6853 | req->work.identity = iod; | |
dfead8a8 | 6854 | req->work.flags |= IO_WQ_WORK_CREDS; |
ef4ff581 PB |
6855 | } |
6856 | ||
6857 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
c11368a5 | 6858 | req->flags |= sqe_flags; |
258b29a9 | 6859 | state = &ctx->submit_state; |
ef4ff581 | 6860 | |
27926b68 JA |
6861 | /* |
6862 | * Plug now if we have more than 1 IO left after this, and the target | |
6863 | * is potentially a read/write to block based storage. | |
6864 | */ | |
6865 | if (!state->plug_started && state->ios_left > 1 && | |
6866 | io_op_defs[req->opcode].plug) { | |
6867 | blk_start_plug(&state->plug); | |
6868 | state->plug_started = true; | |
6869 | } | |
6870 | ||
bd5bbda7 PB |
6871 | ret = 0; |
6872 | if (io_op_defs[req->opcode].needs_file) { | |
6873 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6874 | ||
6875 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
ba13e23f | 6876 | if (unlikely(!req->file)) |
bd5bbda7 PB |
6877 | ret = -EBADF; |
6878 | } | |
63ff8223 | 6879 | |
71b547c0 PB |
6880 | state->ios_left--; |
6881 | return ret; | |
0553b8bd PB |
6882 | } |
6883 | ||
0f212204 | 6884 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6885 | { |
863e0560 | 6886 | struct io_submit_link link; |
9e645e11 | 6887 | int i, submitted = 0; |
6c271ce2 | 6888 | |
c4a2ed72 | 6889 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6890 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6891 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6892 | return -EBUSY; |
6893 | } | |
6c271ce2 | 6894 | |
ee7d46d9 PB |
6895 | /* make sure SQ entry isn't read before tail */ |
6896 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6897 | |
2b85edfc PB |
6898 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6899 | return -EAGAIN; | |
6c271ce2 | 6900 | |
d8a6df10 | 6901 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6902 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6903 | |
ba88ff11 | 6904 | io_submit_state_start(&ctx->submit_state, nr); |
863e0560 | 6905 | link.head = NULL; |
b14cca0c | 6906 | |
6c271ce2 | 6907 | for (i = 0; i < nr; i++) { |
3529d8c2 | 6908 | const struct io_uring_sqe *sqe; |
196be95c | 6909 | struct io_kiocb *req; |
1cb1edb2 | 6910 | int err; |
fb5ccc98 | 6911 | |
258b29a9 | 6912 | req = io_alloc_req(ctx); |
196be95c PB |
6913 | if (unlikely(!req)) { |
6914 | if (!submitted) | |
6915 | submitted = -EAGAIN; | |
fb5ccc98 | 6916 | break; |
196be95c | 6917 | } |
4fccfcbb PB |
6918 | sqe = io_get_sqe(ctx); |
6919 | if (unlikely(!sqe)) { | |
6920 | kmem_cache_free(req_cachep, req); | |
6921 | break; | |
6922 | } | |
d3656344 JA |
6923 | /* will complete beyond this point, count as submitted */ |
6924 | submitted++; | |
6925 | ||
258b29a9 | 6926 | err = io_init_req(ctx, req, sqe); |
ef4ff581 | 6927 | if (unlikely(err)) { |
1cb1edb2 | 6928 | fail_req: |
e1e16097 JA |
6929 | io_put_req(req); |
6930 | io_req_complete(req, err); | |
196be95c PB |
6931 | break; |
6932 | } | |
fb5ccc98 | 6933 | |
354420f7 | 6934 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
2d7e9358 | 6935 | true, ctx->flags & IORING_SETUP_SQPOLL); |
c5eef2b9 | 6936 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
6937 | if (err) |
6938 | goto fail_req; | |
6c271ce2 JA |
6939 | } |
6940 | ||
9466f437 PB |
6941 | if (unlikely(submitted != nr)) { |
6942 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6943 | struct io_uring_task *tctx = current->io_uring; |
6944 | int unused = nr - ref_used; | |
9466f437 | 6945 | |
d8a6df10 JA |
6946 | percpu_ref_put_many(&ctx->refs, unused); |
6947 | percpu_counter_sub(&tctx->inflight, unused); | |
6948 | put_task_struct_many(current, unused); | |
9466f437 | 6949 | } |
863e0560 | 6950 | if (link.head) |
c5eef2b9 | 6951 | io_queue_link_head(link.head); |
ba88ff11 | 6952 | io_submit_state_end(&ctx->submit_state, ctx); |
6c271ce2 | 6953 | |
ae9428ca PB |
6954 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6955 | io_commit_sqring(ctx); | |
6956 | ||
6c271ce2 JA |
6957 | return submitted; |
6958 | } | |
6959 | ||
23b3628e XW |
6960 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6961 | { | |
6962 | /* Tell userspace we may need a wakeup call */ | |
6963 | spin_lock_irq(&ctx->completion_lock); | |
6964 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6965 | spin_unlock_irq(&ctx->completion_lock); | |
6966 | } | |
6967 | ||
6968 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6969 | { | |
6970 | spin_lock_irq(&ctx->completion_lock); | |
6971 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6972 | spin_unlock_irq(&ctx->completion_lock); | |
6973 | } | |
6974 | ||
08369246 | 6975 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6976 | { |
c8d1ba58 | 6977 | unsigned int to_submit; |
bdcd3eab | 6978 | int ret = 0; |
6c271ce2 | 6979 | |
c8d1ba58 | 6980 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6981 | /* if we're handling multiple rings, cap submit size for fairness */ |
6982 | if (cap_entries && to_submit > 8) | |
6983 | to_submit = 8; | |
6984 | ||
906a3c6f | 6985 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6986 | unsigned nr_events = 0; |
a4c0b3de | 6987 | |
c8d1ba58 | 6988 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6989 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6990 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6991 | |
d9d05217 PB |
6992 | if (to_submit && !ctx->sqo_dead && |
6993 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 6994 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6995 | mutex_unlock(&ctx->uring_lock); |
6996 | } | |
6c271ce2 | 6997 | |
90554200 JA |
6998 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6999 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 7000 | |
08369246 XW |
7001 | return ret; |
7002 | } | |
6c271ce2 | 7003 | |
08369246 XW |
7004 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7005 | { | |
7006 | struct io_ring_ctx *ctx; | |
7007 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7008 | |
08369246 XW |
7009 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7010 | if (sq_thread_idle < ctx->sq_thread_idle) | |
7011 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 7012 | } |
c1edbf5f | 7013 | |
08369246 | 7014 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7015 | } |
6c271ce2 | 7016 | |
69fb2131 JA |
7017 | static void io_sqd_init_new(struct io_sq_data *sqd) |
7018 | { | |
7019 | struct io_ring_ctx *ctx; | |
7020 | ||
7021 | while (!list_empty(&sqd->ctx_new_list)) { | |
7022 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
7023 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
7024 | complete(&ctx->sq_thread_comp); | |
7025 | } | |
08369246 XW |
7026 | |
7027 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
7028 | } |
7029 | ||
c8d1ba58 JA |
7030 | static int io_sq_thread(void *data) |
7031 | { | |
91d8f519 | 7032 | struct cgroup_subsys_state *cur_css = NULL; |
28cea78a JA |
7033 | struct files_struct *old_files = current->files; |
7034 | struct nsproxy *old_nsproxy = current->nsproxy; | |
69fb2131 JA |
7035 | const struct cred *old_cred = NULL; |
7036 | struct io_sq_data *sqd = data; | |
7037 | struct io_ring_ctx *ctx; | |
a0d9205f | 7038 | unsigned long timeout = 0; |
08369246 | 7039 | DEFINE_WAIT(wait); |
6c271ce2 | 7040 | |
28cea78a JA |
7041 | task_lock(current); |
7042 | current->files = NULL; | |
7043 | current->nsproxy = NULL; | |
7044 | task_unlock(current); | |
6c271ce2 | 7045 | |
69fb2131 | 7046 | while (!kthread_should_stop()) { |
08369246 XW |
7047 | int ret; |
7048 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
7049 | |
7050 | /* | |
69fb2131 JA |
7051 | * Any changes to the sqd lists are synchronized through the |
7052 | * kthread parking. This synchronizes the thread vs users, | |
7053 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 7054 | */ |
65b2b213 | 7055 | if (kthread_should_park()) { |
69fb2131 | 7056 | kthread_parkme(); |
65b2b213 XW |
7057 | /* |
7058 | * When sq thread is unparked, in case the previous park operation | |
7059 | * comes from io_put_sq_data(), which means that sq thread is going | |
7060 | * to be stopped, so here needs to have a check. | |
7061 | */ | |
7062 | if (kthread_should_stop()) | |
7063 | break; | |
7064 | } | |
7143b5ac | 7065 | |
08369246 | 7066 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 7067 | io_sqd_init_new(sqd); |
08369246 XW |
7068 | timeout = jiffies + sqd->sq_thread_idle; |
7069 | } | |
6c271ce2 | 7070 | |
08369246 | 7071 | sqt_spin = false; |
e95eee2d | 7072 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 JA |
7073 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7074 | if (current->cred != ctx->creds) { | |
7075 | if (old_cred) | |
7076 | revert_creds(old_cred); | |
7077 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 7078 | } |
91d8f519 | 7079 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
7080 | #ifdef CONFIG_AUDIT |
7081 | current->loginuid = ctx->loginuid; | |
7082 | current->sessionid = ctx->sessionid; | |
7083 | #endif | |
bdcd3eab | 7084 | |
08369246 XW |
7085 | ret = __io_sq_thread(ctx, cap_entries); |
7086 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
7087 | sqt_spin = true; | |
6c271ce2 | 7088 | |
28cea78a | 7089 | io_sq_thread_drop_mm_files(); |
69fb2131 | 7090 | } |
6c271ce2 | 7091 | |
08369246 | 7092 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7093 | io_run_task_work(); |
d434ab6d | 7094 | io_sq_thread_drop_mm_files(); |
c8d1ba58 | 7095 | cond_resched(); |
08369246 XW |
7096 | if (sqt_spin) |
7097 | timeout = jiffies + sqd->sq_thread_idle; | |
7098 | continue; | |
7099 | } | |
7100 | ||
08369246 XW |
7101 | needs_sched = true; |
7102 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
7103 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
7104 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
7105 | !list_empty_careful(&ctx->iopoll_list)) { | |
7106 | needs_sched = false; | |
7107 | break; | |
7108 | } | |
7109 | if (io_sqring_entries(ctx)) { | |
7110 | needs_sched = false; | |
7111 | break; | |
7112 | } | |
7113 | } | |
7114 | ||
8b28fdf2 | 7115 | if (needs_sched && !kthread_should_park()) { |
69fb2131 JA |
7116 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7117 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 7118 | |
69fb2131 | 7119 | schedule(); |
69fb2131 JA |
7120 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7121 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7122 | } |
08369246 XW |
7123 | |
7124 | finish_wait(&sqd->wait, &wait); | |
7125 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
7126 | } |
7127 | ||
4c6e277c | 7128 | io_run_task_work(); |
d434ab6d | 7129 | io_sq_thread_drop_mm_files(); |
b41e9852 | 7130 | |
91d8f519 DZ |
7131 | if (cur_css) |
7132 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
7133 | if (old_cred) |
7134 | revert_creds(old_cred); | |
06058632 | 7135 | |
28cea78a JA |
7136 | task_lock(current); |
7137 | current->files = old_files; | |
7138 | current->nsproxy = old_nsproxy; | |
7139 | task_unlock(current); | |
7140 | ||
2bbcd6d3 | 7141 | kthread_parkme(); |
06058632 | 7142 | |
6c271ce2 JA |
7143 | return 0; |
7144 | } | |
7145 | ||
bda52162 JA |
7146 | struct io_wait_queue { |
7147 | struct wait_queue_entry wq; | |
7148 | struct io_ring_ctx *ctx; | |
7149 | unsigned to_wait; | |
7150 | unsigned nr_timeouts; | |
7151 | }; | |
7152 | ||
6c503150 | 7153 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7154 | { |
7155 | struct io_ring_ctx *ctx = iowq->ctx; | |
7156 | ||
7157 | /* | |
d195a66e | 7158 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7159 | * started waiting. For timeouts, we always want to return to userspace, |
7160 | * regardless of event count. | |
7161 | */ | |
6c503150 | 7162 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
7163 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
7164 | } | |
7165 | ||
7166 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7167 | int wake_flags, void *key) | |
7168 | { | |
7169 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7170 | wq); | |
7171 | ||
6c503150 PB |
7172 | /* |
7173 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7174 | * the task, and the next invocation will do it. | |
7175 | */ | |
7176 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
7177 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
7178 | return -1; | |
bda52162 JA |
7179 | } |
7180 | ||
af9c1a44 JA |
7181 | static int io_run_task_work_sig(void) |
7182 | { | |
7183 | if (io_run_task_work()) | |
7184 | return 1; | |
7185 | if (!signal_pending(current)) | |
7186 | return 0; | |
792ee0f6 JA |
7187 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
7188 | return -ERESTARTSYS; | |
af9c1a44 JA |
7189 | return -EINTR; |
7190 | } | |
7191 | ||
eeb60b9a PB |
7192 | /* when returns >0, the caller should retry */ |
7193 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7194 | struct io_wait_queue *iowq, | |
7195 | signed long *timeout) | |
7196 | { | |
7197 | int ret; | |
7198 | ||
7199 | /* make sure we run task_work before checking for signals */ | |
7200 | ret = io_run_task_work_sig(); | |
7201 | if (ret || io_should_wake(iowq)) | |
7202 | return ret; | |
7203 | /* let the caller flush overflows, retry */ | |
7204 | if (test_bit(0, &ctx->cq_check_overflow)) | |
7205 | return 1; | |
7206 | ||
7207 | *timeout = schedule_timeout(*timeout); | |
7208 | return !*timeout ? -ETIME : 1; | |
7209 | } | |
7210 | ||
2b188cc1 JA |
7211 | /* |
7212 | * Wait until events become available, if we don't already have some. The | |
7213 | * application must reap them itself, as they reside on the shared cq ring. | |
7214 | */ | |
7215 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7216 | const sigset_t __user *sig, size_t sigsz, |
7217 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7218 | { |
bda52162 JA |
7219 | struct io_wait_queue iowq = { |
7220 | .wq = { | |
7221 | .private = current, | |
7222 | .func = io_wake_function, | |
7223 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7224 | }, | |
7225 | .ctx = ctx, | |
7226 | .to_wait = min_events, | |
7227 | }; | |
75b28aff | 7228 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7229 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7230 | int ret; | |
2b188cc1 | 7231 | |
b41e9852 | 7232 | do { |
6c503150 PB |
7233 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
7234 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 7235 | return 0; |
4c6e277c | 7236 | if (!io_run_task_work()) |
b41e9852 | 7237 | break; |
b41e9852 | 7238 | } while (1); |
2b188cc1 JA |
7239 | |
7240 | if (sig) { | |
9e75ad5d AB |
7241 | #ifdef CONFIG_COMPAT |
7242 | if (in_compat_syscall()) | |
7243 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7244 | sigsz); |
9e75ad5d AB |
7245 | else |
7246 | #endif | |
b772434b | 7247 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7248 | |
2b188cc1 JA |
7249 | if (ret) |
7250 | return ret; | |
7251 | } | |
7252 | ||
c73ebb68 | 7253 | if (uts) { |
c1d5a224 PB |
7254 | struct timespec64 ts; |
7255 | ||
c73ebb68 HX |
7256 | if (get_timespec64(&ts, uts)) |
7257 | return -EFAULT; | |
7258 | timeout = timespec64_to_jiffies(&ts); | |
7259 | } | |
7260 | ||
bda52162 | 7261 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7262 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7263 | do { |
6c503150 | 7264 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
7265 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7266 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7267 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7268 | finish_wait(&ctx->wait, &iowq.wq); | |
7269 | } while (ret > 0); | |
bda52162 | 7270 | |
b7db41c9 | 7271 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7272 | |
75b28aff | 7273 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7274 | } |
7275 | ||
6b06314c JA |
7276 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7277 | { | |
7278 | #if defined(CONFIG_UNIX) | |
7279 | if (ctx->ring_sock) { | |
7280 | struct sock *sock = ctx->ring_sock->sk; | |
7281 | struct sk_buff *skb; | |
7282 | ||
7283 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7284 | kfree_skb(skb); | |
7285 | } | |
7286 | #else | |
7287 | int i; | |
7288 | ||
65e19f54 JA |
7289 | for (i = 0; i < ctx->nr_user_files; i++) { |
7290 | struct file *file; | |
7291 | ||
7292 | file = io_file_from_index(ctx, i); | |
7293 | if (file) | |
7294 | fput(file); | |
7295 | } | |
6b06314c JA |
7296 | #endif |
7297 | } | |
7298 | ||
00835dce | 7299 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7300 | { |
269bbe5f | 7301 | struct fixed_rsrc_data *data; |
05f3fb3c | 7302 | |
269bbe5f | 7303 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7304 | complete(&data->done); |
7305 | } | |
7306 | ||
2a63b2d9 BM |
7307 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
7308 | { | |
7309 | spin_lock_bh(&ctx->rsrc_ref_lock); | |
7310 | } | |
7311 | ||
7312 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) | |
7313 | { | |
7314 | spin_unlock_bh(&ctx->rsrc_ref_lock); | |
7315 | } | |
7316 | ||
d67d2263 BM |
7317 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7318 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7319 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7320 | { |
2a63b2d9 | 7321 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7322 | rsrc_data->node = ref_node; |
d67d2263 | 7323 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7324 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7325 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7326 | } |
7327 | ||
d7954b2b BM |
7328 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, |
7329 | struct io_ring_ctx *ctx, | |
7330 | struct fixed_rsrc_ref_node *backup_node) | |
6b06314c | 7331 | { |
d7954b2b | 7332 | struct fixed_rsrc_ref_node *ref_node; |
1ffc5422 | 7333 | int ret; |
65e19f54 | 7334 | |
2a63b2d9 | 7335 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7336 | ref_node = data->node; |
2a63b2d9 | 7337 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7338 | if (ref_node) |
7339 | percpu_ref_kill(&ref_node->refs); | |
7340 | ||
7341 | percpu_ref_kill(&data->refs); | |
7342 | ||
7343 | /* wait for all refs nodes to complete */ | |
269bbe5f | 7344 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7345 | do { |
7346 | ret = wait_for_completion_interruptible(&data->done); | |
7347 | if (!ret) | |
7348 | break; | |
7349 | ret = io_run_task_work_sig(); | |
7350 | if (ret < 0) { | |
7351 | percpu_ref_resurrect(&data->refs); | |
7352 | reinit_completion(&data->done); | |
d67d2263 | 7353 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
1ffc5422 PB |
7354 | return ret; |
7355 | } | |
7356 | } while (1); | |
05f3fb3c | 7357 | |
d7954b2b BM |
7358 | destroy_fixed_rsrc_ref_node(backup_node); |
7359 | return 0; | |
7360 | } | |
7361 | ||
1ad555c6 BM |
7362 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7363 | { | |
7364 | struct fixed_rsrc_data *data; | |
7365 | ||
7366 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7367 | if (!data) | |
7368 | return NULL; | |
7369 | ||
00835dce | 7370 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7371 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7372 | kfree(data); | |
7373 | return NULL; | |
7374 | } | |
7375 | data->ctx = ctx; | |
7376 | init_completion(&data->done); | |
7377 | return data; | |
7378 | } | |
7379 | ||
7380 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7381 | { | |
7382 | percpu_ref_exit(&data->refs); | |
7383 | kfree(data->table); | |
7384 | kfree(data); | |
7385 | } | |
7386 | ||
d7954b2b BM |
7387 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7388 | { | |
7389 | struct fixed_rsrc_data *data = ctx->file_data; | |
7390 | struct fixed_rsrc_ref_node *backup_node; | |
7391 | unsigned nr_tables, i; | |
7392 | int ret; | |
7393 | ||
7394 | if (!data) | |
7395 | return -ENXIO; | |
7396 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7397 | if (!backup_node) | |
7398 | return -ENOMEM; | |
7399 | init_fixed_file_ref_node(ctx, backup_node); | |
7400 | ||
7401 | ret = io_rsrc_ref_quiesce(data, ctx, backup_node); | |
7402 | if (ret) | |
7403 | return ret; | |
7404 | ||
6b06314c | 7405 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7406 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7407 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7408 | kfree(data->table[i].files); |
1ad555c6 | 7409 | free_fixed_rsrc_data(data); |
05f3fb3c | 7410 | ctx->file_data = NULL; |
6b06314c JA |
7411 | ctx->nr_user_files = 0; |
7412 | return 0; | |
7413 | } | |
7414 | ||
534ca6d6 | 7415 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7416 | { |
534ca6d6 | 7417 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7418 | /* |
7419 | * The park is a bit of a work-around, without it we get | |
7420 | * warning spews on shutdown with SQPOLL set and affinity | |
7421 | * set to a single CPU. | |
7422 | */ | |
534ca6d6 JA |
7423 | if (sqd->thread) { |
7424 | kthread_park(sqd->thread); | |
7425 | kthread_stop(sqd->thread); | |
7426 | } | |
7427 | ||
7428 | kfree(sqd); | |
7429 | } | |
7430 | } | |
7431 | ||
aa06165d JA |
7432 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7433 | { | |
7434 | struct io_ring_ctx *ctx_attach; | |
7435 | struct io_sq_data *sqd; | |
7436 | struct fd f; | |
7437 | ||
7438 | f = fdget(p->wq_fd); | |
7439 | if (!f.file) | |
7440 | return ERR_PTR(-ENXIO); | |
7441 | if (f.file->f_op != &io_uring_fops) { | |
7442 | fdput(f); | |
7443 | return ERR_PTR(-EINVAL); | |
7444 | } | |
7445 | ||
7446 | ctx_attach = f.file->private_data; | |
7447 | sqd = ctx_attach->sq_data; | |
7448 | if (!sqd) { | |
7449 | fdput(f); | |
7450 | return ERR_PTR(-EINVAL); | |
7451 | } | |
7452 | ||
7453 | refcount_inc(&sqd->refs); | |
7454 | fdput(f); | |
7455 | return sqd; | |
7456 | } | |
7457 | ||
534ca6d6 JA |
7458 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7459 | { | |
7460 | struct io_sq_data *sqd; | |
7461 | ||
aa06165d JA |
7462 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7463 | return io_attach_sq_data(p); | |
7464 | ||
534ca6d6 JA |
7465 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7466 | if (!sqd) | |
7467 | return ERR_PTR(-ENOMEM); | |
7468 | ||
7469 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7470 | INIT_LIST_HEAD(&sqd->ctx_list); |
7471 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7472 | mutex_init(&sqd->ctx_lock); | |
7473 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7474 | init_waitqueue_head(&sqd->wait); |
7475 | return sqd; | |
7476 | } | |
7477 | ||
69fb2131 JA |
7478 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7479 | __releases(&sqd->lock) | |
7480 | { | |
7481 | if (!sqd->thread) | |
7482 | return; | |
7483 | kthread_unpark(sqd->thread); | |
7484 | mutex_unlock(&sqd->lock); | |
7485 | } | |
7486 | ||
7487 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7488 | __acquires(&sqd->lock) | |
7489 | { | |
7490 | if (!sqd->thread) | |
7491 | return; | |
7492 | mutex_lock(&sqd->lock); | |
7493 | kthread_park(sqd->thread); | |
7494 | } | |
7495 | ||
534ca6d6 JA |
7496 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7497 | { | |
7498 | struct io_sq_data *sqd = ctx->sq_data; | |
7499 | ||
7500 | if (sqd) { | |
7501 | if (sqd->thread) { | |
7502 | /* | |
7503 | * We may arrive here from the error branch in | |
7504 | * io_sq_offload_create() where the kthread is created | |
7505 | * without being waked up, thus wake it up now to make | |
7506 | * sure the wait will complete. | |
7507 | */ | |
7508 | wake_up_process(sqd->thread); | |
7509 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7510 | |
7511 | io_sq_thread_park(sqd); | |
7512 | } | |
7513 | ||
7514 | mutex_lock(&sqd->ctx_lock); | |
7515 | list_del(&ctx->sqd_list); | |
08369246 | 7516 | io_sqd_update_thread_idle(sqd); |
69fb2131 JA |
7517 | mutex_unlock(&sqd->ctx_lock); |
7518 | ||
08369246 | 7519 | if (sqd->thread) |
69fb2131 | 7520 | io_sq_thread_unpark(sqd); |
534ca6d6 JA |
7521 | |
7522 | io_put_sq_data(sqd); | |
7523 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7524 | } |
7525 | } | |
7526 | ||
6b06314c JA |
7527 | static void io_finish_async(struct io_ring_ctx *ctx) |
7528 | { | |
6c271ce2 JA |
7529 | io_sq_thread_stop(ctx); |
7530 | ||
561fb04a JA |
7531 | if (ctx->io_wq) { |
7532 | io_wq_destroy(ctx->io_wq); | |
7533 | ctx->io_wq = NULL; | |
6b06314c JA |
7534 | } |
7535 | } | |
7536 | ||
7537 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7538 | /* |
7539 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7540 | * the io_uring can be safely unregistered on process exit, even if we have | |
7541 | * loops in the file referencing. | |
7542 | */ | |
7543 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7544 | { | |
7545 | struct sock *sk = ctx->ring_sock->sk; | |
7546 | struct scm_fp_list *fpl; | |
7547 | struct sk_buff *skb; | |
08a45173 | 7548 | int i, nr_files; |
6b06314c | 7549 | |
6b06314c JA |
7550 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7551 | if (!fpl) | |
7552 | return -ENOMEM; | |
7553 | ||
7554 | skb = alloc_skb(0, GFP_KERNEL); | |
7555 | if (!skb) { | |
7556 | kfree(fpl); | |
7557 | return -ENOMEM; | |
7558 | } | |
7559 | ||
7560 | skb->sk = sk; | |
6b06314c | 7561 | |
08a45173 | 7562 | nr_files = 0; |
6b06314c JA |
7563 | fpl->user = get_uid(ctx->user); |
7564 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7565 | struct file *file = io_file_from_index(ctx, i + offset); |
7566 | ||
7567 | if (!file) | |
08a45173 | 7568 | continue; |
65e19f54 | 7569 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7570 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7571 | nr_files++; | |
6b06314c JA |
7572 | } |
7573 | ||
08a45173 JA |
7574 | if (nr_files) { |
7575 | fpl->max = SCM_MAX_FD; | |
7576 | fpl->count = nr_files; | |
7577 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7578 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7579 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7580 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7581 | |
08a45173 JA |
7582 | for (i = 0; i < nr_files; i++) |
7583 | fput(fpl->fp[i]); | |
7584 | } else { | |
7585 | kfree_skb(skb); | |
7586 | kfree(fpl); | |
7587 | } | |
6b06314c JA |
7588 | |
7589 | return 0; | |
7590 | } | |
7591 | ||
7592 | /* | |
7593 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7594 | * causes regular reference counting to break down. We rely on the UNIX | |
7595 | * garbage collection to take care of this problem for us. | |
7596 | */ | |
7597 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7598 | { | |
7599 | unsigned left, total; | |
7600 | int ret = 0; | |
7601 | ||
7602 | total = 0; | |
7603 | left = ctx->nr_user_files; | |
7604 | while (left) { | |
7605 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7606 | |
7607 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7608 | if (ret) | |
7609 | break; | |
7610 | left -= this_files; | |
7611 | total += this_files; | |
7612 | } | |
7613 | ||
7614 | if (!ret) | |
7615 | return 0; | |
7616 | ||
7617 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7618 | struct file *file = io_file_from_index(ctx, total); |
7619 | ||
7620 | if (file) | |
7621 | fput(file); | |
6b06314c JA |
7622 | total++; |
7623 | } | |
7624 | ||
7625 | return ret; | |
7626 | } | |
7627 | #else | |
7628 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7629 | { | |
7630 | return 0; | |
7631 | } | |
7632 | #endif | |
7633 | ||
269bbe5f | 7634 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7635 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7636 | { |
7637 | int i; | |
7638 | ||
7639 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7640 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7641 | unsigned this_files; |
7642 | ||
7643 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7644 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7645 | GFP_KERNEL); | |
7646 | if (!table->files) | |
7647 | break; | |
7648 | nr_files -= this_files; | |
7649 | } | |
7650 | ||
7651 | if (i == nr_tables) | |
7652 | return 0; | |
7653 | ||
7654 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7655 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7656 | kfree(table->files); |
7657 | } | |
7658 | return 1; | |
7659 | } | |
7660 | ||
50238531 | 7661 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7662 | { |
50238531 | 7663 | struct file *file = prsrc->file; |
05f3fb3c JA |
7664 | #if defined(CONFIG_UNIX) |
7665 | struct sock *sock = ctx->ring_sock->sk; | |
7666 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7667 | struct sk_buff *skb; | |
7668 | int i; | |
7669 | ||
7670 | __skb_queue_head_init(&list); | |
7671 | ||
7672 | /* | |
7673 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7674 | * remove this entry and rearrange the file array. | |
7675 | */ | |
7676 | skb = skb_dequeue(head); | |
7677 | while (skb) { | |
7678 | struct scm_fp_list *fp; | |
7679 | ||
7680 | fp = UNIXCB(skb).fp; | |
7681 | for (i = 0; i < fp->count; i++) { | |
7682 | int left; | |
7683 | ||
7684 | if (fp->fp[i] != file) | |
7685 | continue; | |
7686 | ||
7687 | unix_notinflight(fp->user, fp->fp[i]); | |
7688 | left = fp->count - 1 - i; | |
7689 | if (left) { | |
7690 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7691 | left * sizeof(struct file *)); | |
7692 | } | |
7693 | fp->count--; | |
7694 | if (!fp->count) { | |
7695 | kfree_skb(skb); | |
7696 | skb = NULL; | |
7697 | } else { | |
7698 | __skb_queue_tail(&list, skb); | |
7699 | } | |
7700 | fput(file); | |
7701 | file = NULL; | |
7702 | break; | |
7703 | } | |
7704 | ||
7705 | if (!file) | |
7706 | break; | |
7707 | ||
7708 | __skb_queue_tail(&list, skb); | |
7709 | ||
7710 | skb = skb_dequeue(head); | |
7711 | } | |
7712 | ||
7713 | if (skb_peek(&list)) { | |
7714 | spin_lock_irq(&head->lock); | |
7715 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7716 | __skb_queue_tail(head, skb); | |
7717 | spin_unlock_irq(&head->lock); | |
7718 | } | |
7719 | #else | |
7720 | fput(file); | |
7721 | #endif | |
7722 | } | |
7723 | ||
269bbe5f | 7724 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7725 | { |
269bbe5f BM |
7726 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7727 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7728 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7729 | |
269bbe5f BM |
7730 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7731 | list_del(&prsrc->list); | |
50238531 | 7732 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7733 | kfree(prsrc); |
65e19f54 | 7734 | } |
05589553 | 7735 | |
05589553 XW |
7736 | percpu_ref_exit(&ref_node->refs); |
7737 | kfree(ref_node); | |
269bbe5f | 7738 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7739 | } |
65e19f54 | 7740 | |
269bbe5f | 7741 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7742 | { |
7743 | struct io_ring_ctx *ctx; | |
7744 | struct llist_node *node; | |
7745 | ||
269bbe5f BM |
7746 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7747 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7748 | |
7749 | while (node) { | |
269bbe5f | 7750 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7751 | struct llist_node *next = node->next; |
7752 | ||
269bbe5f BM |
7753 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7754 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7755 | node = next; |
7756 | } | |
7757 | } | |
7758 | ||
ea64ec02 PB |
7759 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7760 | unsigned i) | |
7761 | { | |
7762 | struct fixed_rsrc_table *table; | |
7763 | ||
7764 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7765 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7766 | } | |
7767 | ||
00835dce | 7768 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7769 | { |
269bbe5f BM |
7770 | struct fixed_rsrc_ref_node *ref_node; |
7771 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7772 | struct io_ring_ctx *ctx; |
e297822b | 7773 | bool first_add = false; |
4a38aed2 | 7774 | int delay = HZ; |
65e19f54 | 7775 | |
269bbe5f BM |
7776 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7777 | data = ref_node->rsrc_data; | |
e297822b PB |
7778 | ctx = data->ctx; |
7779 | ||
2a63b2d9 | 7780 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7781 | ref_node->done = true; |
7782 | ||
d67d2263 BM |
7783 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7784 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7785 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7786 | /* recycle ref nodes in order */ |
7787 | if (!ref_node->done) | |
7788 | break; | |
7789 | list_del(&ref_node->node); | |
269bbe5f | 7790 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7791 | } |
2a63b2d9 | 7792 | io_rsrc_ref_unlock(ctx); |
05589553 | 7793 | |
e297822b | 7794 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7795 | delay = 0; |
05589553 | 7796 | |
4a38aed2 | 7797 | if (!delay) |
269bbe5f | 7798 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7799 | else if (first_add) |
269bbe5f | 7800 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7801 | } |
65e19f54 | 7802 | |
6802535d | 7803 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7804 | struct io_ring_ctx *ctx) |
05f3fb3c | 7805 | { |
269bbe5f | 7806 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7807 | |
05589553 XW |
7808 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7809 | if (!ref_node) | |
3e2224c5 | 7810 | return NULL; |
05f3fb3c | 7811 | |
00835dce | 7812 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7813 | 0, GFP_KERNEL)) { |
7814 | kfree(ref_node); | |
3e2224c5 | 7815 | return NULL; |
05589553 XW |
7816 | } |
7817 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7818 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
6802535d BM |
7819 | ref_node->done = false; |
7820 | return ref_node; | |
7821 | } | |
7822 | ||
bc9744cd PB |
7823 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7824 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7825 | { |
269bbe5f | 7826 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7827 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7828 | } |
7829 | ||
269bbe5f | 7830 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7831 | { |
7832 | percpu_ref_exit(&ref_node->refs); | |
7833 | kfree(ref_node); | |
65e19f54 JA |
7834 | } |
7835 | ||
ea64ec02 | 7836 | |
6b06314c JA |
7837 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7838 | unsigned nr_args) | |
7839 | { | |
7840 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7841 | unsigned nr_tables, i; |
05f3fb3c | 7842 | struct file *file; |
600cf3f8 | 7843 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7844 | struct fixed_rsrc_ref_node *ref_node; |
7845 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7846 | |
05f3fb3c | 7847 | if (ctx->file_data) |
6b06314c JA |
7848 | return -EBUSY; |
7849 | if (!nr_args) | |
7850 | return -EINVAL; | |
7851 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7852 | return -EMFILE; | |
7853 | ||
1ad555c6 | 7854 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7855 | if (!file_data) |
05f3fb3c | 7856 | return -ENOMEM; |
13770a71 | 7857 | ctx->file_data = file_data; |
05f3fb3c | 7858 | |
65e19f54 | 7859 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7860 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7861 | GFP_KERNEL); |
600cf3f8 PB |
7862 | if (!file_data->table) |
7863 | goto out_free; | |
05f3fb3c | 7864 | |
600cf3f8 | 7865 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7866 | goto out_free; |
65e19f54 | 7867 | |
08a45173 | 7868 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7869 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7870 | ret = -EFAULT; | |
7871 | goto out_fput; | |
7872 | } | |
08a45173 | 7873 | /* allow sparse sets */ |
600cf3f8 | 7874 | if (fd == -1) |
08a45173 | 7875 | continue; |
6b06314c | 7876 | |
05f3fb3c | 7877 | file = fget(fd); |
6b06314c | 7878 | ret = -EBADF; |
05f3fb3c | 7879 | if (!file) |
600cf3f8 | 7880 | goto out_fput; |
05f3fb3c | 7881 | |
6b06314c JA |
7882 | /* |
7883 | * Don't allow io_uring instances to be registered. If UNIX | |
7884 | * isn't enabled, then this causes a reference cycle and this | |
7885 | * instance can never get freed. If UNIX is enabled we'll | |
7886 | * handle it just fine, but there's still no point in allowing | |
7887 | * a ring fd as it doesn't support regular read/write anyway. | |
7888 | */ | |
05f3fb3c JA |
7889 | if (file->f_op == &io_uring_fops) { |
7890 | fput(file); | |
600cf3f8 | 7891 | goto out_fput; |
6b06314c | 7892 | } |
ea64ec02 | 7893 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7894 | } |
7895 | ||
6b06314c | 7896 | ret = io_sqe_files_scm(ctx); |
05589553 | 7897 | if (ret) { |
6b06314c | 7898 | io_sqe_files_unregister(ctx); |
05589553 XW |
7899 | return ret; |
7900 | } | |
6b06314c | 7901 | |
bc9744cd | 7902 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7903 | if (!ref_node) { |
05589553 | 7904 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7905 | return -ENOMEM; |
05589553 | 7906 | } |
bc9744cd | 7907 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7908 | |
d67d2263 | 7909 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7910 | return ret; |
600cf3f8 PB |
7911 | out_fput: |
7912 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7913 | file = io_file_from_index(ctx, i); | |
7914 | if (file) | |
7915 | fput(file); | |
7916 | } | |
7917 | for (i = 0; i < nr_tables; i++) | |
7918 | kfree(file_data->table[i].files); | |
7919 | ctx->nr_user_files = 0; | |
600cf3f8 | 7920 | out_free: |
1ad555c6 | 7921 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7922 | ctx->file_data = NULL; |
6b06314c JA |
7923 | return ret; |
7924 | } | |
7925 | ||
c3a31e60 JA |
7926 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7927 | int index) | |
7928 | { | |
7929 | #if defined(CONFIG_UNIX) | |
7930 | struct sock *sock = ctx->ring_sock->sk; | |
7931 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7932 | struct sk_buff *skb; | |
7933 | ||
7934 | /* | |
7935 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7936 | * file set. If there's no room, fall back to allocating a new skb | |
7937 | * and filling it in. | |
7938 | */ | |
7939 | spin_lock_irq(&head->lock); | |
7940 | skb = skb_peek(head); | |
7941 | if (skb) { | |
7942 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7943 | ||
7944 | if (fpl->count < SCM_MAX_FD) { | |
7945 | __skb_unlink(skb, head); | |
7946 | spin_unlock_irq(&head->lock); | |
7947 | fpl->fp[fpl->count] = get_file(file); | |
7948 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7949 | fpl->count++; | |
7950 | spin_lock_irq(&head->lock); | |
7951 | __skb_queue_head(head, skb); | |
7952 | } else { | |
7953 | skb = NULL; | |
7954 | } | |
7955 | } | |
7956 | spin_unlock_irq(&head->lock); | |
7957 | ||
7958 | if (skb) { | |
7959 | fput(file); | |
7960 | return 0; | |
7961 | } | |
7962 | ||
7963 | return __io_sqe_files_scm(ctx, 1, index); | |
7964 | #else | |
7965 | return 0; | |
7966 | #endif | |
7967 | } | |
7968 | ||
50238531 | 7969 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7970 | { |
269bbe5f BM |
7971 | struct io_rsrc_put *prsrc; |
7972 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7973 | |
269bbe5f BM |
7974 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7975 | if (!prsrc) | |
a5318d3c | 7976 | return -ENOMEM; |
05f3fb3c | 7977 | |
50238531 | 7978 | prsrc->rsrc = rsrc; |
269bbe5f | 7979 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7980 | |
a5318d3c | 7981 | return 0; |
05f3fb3c JA |
7982 | } |
7983 | ||
269bbe5f BM |
7984 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7985 | struct file *file) | |
7986 | { | |
50238531 | 7987 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7988 | } |
7989 | ||
05f3fb3c | 7990 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7991 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7992 | unsigned nr_args) |
7993 | { | |
269bbe5f BM |
7994 | struct fixed_rsrc_data *data = ctx->file_data; |
7995 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7996 | struct file *file, **file_slot; |
c3a31e60 JA |
7997 | __s32 __user *fds; |
7998 | int fd, i, err; | |
7999 | __u32 done; | |
05589553 | 8000 | bool needs_switch = false; |
c3a31e60 | 8001 | |
05f3fb3c | 8002 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
8003 | return -EOVERFLOW; |
8004 | if (done > ctx->nr_user_files) | |
8005 | return -EINVAL; | |
8006 | ||
bc9744cd | 8007 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
8008 | if (!ref_node) |
8009 | return -ENOMEM; | |
bc9744cd | 8010 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 8011 | |
269bbe5f | 8012 | fds = u64_to_user_ptr(up->data); |
67973b93 | 8013 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
8014 | err = 0; |
8015 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
8016 | err = -EFAULT; | |
8017 | break; | |
8018 | } | |
4e0377a1 | 8019 | if (fd == IORING_REGISTER_FILES_SKIP) |
8020 | continue; | |
8021 | ||
67973b93 | 8022 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
8023 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
8024 | ||
8025 | if (*file_slot) { | |
8026 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
8027 | if (err) |
8028 | break; | |
ea64ec02 | 8029 | *file_slot = NULL; |
05589553 | 8030 | needs_switch = true; |
c3a31e60 JA |
8031 | } |
8032 | if (fd != -1) { | |
c3a31e60 JA |
8033 | file = fget(fd); |
8034 | if (!file) { | |
8035 | err = -EBADF; | |
8036 | break; | |
8037 | } | |
8038 | /* | |
8039 | * Don't allow io_uring instances to be registered. If | |
8040 | * UNIX isn't enabled, then this causes a reference | |
8041 | * cycle and this instance can never get freed. If UNIX | |
8042 | * is enabled we'll handle it just fine, but there's | |
8043 | * still no point in allowing a ring fd as it doesn't | |
8044 | * support regular read/write anyway. | |
8045 | */ | |
8046 | if (file->f_op == &io_uring_fops) { | |
8047 | fput(file); | |
8048 | err = -EBADF; | |
8049 | break; | |
8050 | } | |
e68a3ff8 | 8051 | *file_slot = file; |
c3a31e60 | 8052 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8053 | if (err) { |
e68a3ff8 | 8054 | *file_slot = NULL; |
f3bd9dae | 8055 | fput(file); |
c3a31e60 | 8056 | break; |
f3bd9dae | 8057 | } |
c3a31e60 | 8058 | } |
05f3fb3c JA |
8059 | } |
8060 | ||
05589553 | 8061 | if (needs_switch) { |
b2e96852 | 8062 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 8063 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 8064 | } else |
269bbe5f | 8065 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
8066 | |
8067 | return done ? done : err; | |
8068 | } | |
05589553 | 8069 | |
05f3fb3c JA |
8070 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
8071 | unsigned nr_args) | |
8072 | { | |
269bbe5f | 8073 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
8074 | |
8075 | if (!ctx->file_data) | |
8076 | return -ENXIO; | |
8077 | if (!nr_args) | |
8078 | return -EINVAL; | |
8079 | if (copy_from_user(&up, arg, sizeof(up))) | |
8080 | return -EFAULT; | |
8081 | if (up.resv) | |
8082 | return -EINVAL; | |
8083 | ||
8084 | return __io_sqe_files_update(ctx, &up, nr_args); | |
8085 | } | |
c3a31e60 | 8086 | |
5280f7e5 | 8087 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
8088 | { |
8089 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8090 | ||
5280f7e5 PB |
8091 | req = io_put_req_find_next(req); |
8092 | return req ? &req->work : NULL; | |
7d723065 JA |
8093 | } |
8094 | ||
24369c2e PB |
8095 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
8096 | struct io_uring_params *p) | |
8097 | { | |
8098 | struct io_wq_data data; | |
8099 | struct fd f; | |
8100 | struct io_ring_ctx *ctx_attach; | |
8101 | unsigned int concurrency; | |
8102 | int ret = 0; | |
8103 | ||
8104 | data.user = ctx->user; | |
e9fd9396 | 8105 | data.free_work = io_free_work; |
f5fa38c5 | 8106 | data.do_work = io_wq_submit_work; |
24369c2e PB |
8107 | |
8108 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
8109 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
8110 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
8111 | ||
8112 | ctx->io_wq = io_wq_create(concurrency, &data); | |
8113 | if (IS_ERR(ctx->io_wq)) { | |
8114 | ret = PTR_ERR(ctx->io_wq); | |
8115 | ctx->io_wq = NULL; | |
8116 | } | |
8117 | return ret; | |
8118 | } | |
8119 | ||
8120 | f = fdget(p->wq_fd); | |
8121 | if (!f.file) | |
8122 | return -EBADF; | |
8123 | ||
8124 | if (f.file->f_op != &io_uring_fops) { | |
8125 | ret = -EINVAL; | |
8126 | goto out_fput; | |
8127 | } | |
8128 | ||
8129 | ctx_attach = f.file->private_data; | |
8130 | /* @io_wq is protected by holding the fd */ | |
8131 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
8132 | ret = -EINVAL; | |
8133 | goto out_fput; | |
8134 | } | |
8135 | ||
8136 | ctx->io_wq = ctx_attach->io_wq; | |
8137 | out_fput: | |
8138 | fdput(f); | |
8139 | return ret; | |
8140 | } | |
8141 | ||
0f212204 JA |
8142 | static int io_uring_alloc_task_context(struct task_struct *task) |
8143 | { | |
8144 | struct io_uring_task *tctx; | |
d8a6df10 | 8145 | int ret; |
0f212204 JA |
8146 | |
8147 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
8148 | if (unlikely(!tctx)) | |
8149 | return -ENOMEM; | |
8150 | ||
d8a6df10 JA |
8151 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8152 | if (unlikely(ret)) { | |
8153 | kfree(tctx); | |
8154 | return ret; | |
8155 | } | |
8156 | ||
0f212204 JA |
8157 | xa_init(&tctx->xa); |
8158 | init_waitqueue_head(&tctx->wait); | |
8159 | tctx->last = NULL; | |
fdaf083c JA |
8160 | atomic_set(&tctx->in_idle, 0); |
8161 | tctx->sqpoll = false; | |
500a373d JA |
8162 | io_init_identity(&tctx->__identity); |
8163 | tctx->identity = &tctx->__identity; | |
0f212204 | 8164 | task->io_uring = tctx; |
7cbf1722 JA |
8165 | spin_lock_init(&tctx->task_lock); |
8166 | INIT_WQ_LIST(&tctx->task_list); | |
8167 | tctx->task_state = 0; | |
8168 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
8169 | return 0; |
8170 | } | |
8171 | ||
8172 | void __io_uring_free(struct task_struct *tsk) | |
8173 | { | |
8174 | struct io_uring_task *tctx = tsk->io_uring; | |
8175 | ||
8176 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
8177 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
8178 | if (tctx->identity != &tctx->__identity) | |
8179 | kfree(tctx->identity); | |
d8a6df10 | 8180 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8181 | kfree(tctx); |
8182 | tsk->io_uring = NULL; | |
8183 | } | |
8184 | ||
7e84e1c7 SG |
8185 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8186 | struct io_uring_params *p) | |
2b188cc1 JA |
8187 | { |
8188 | int ret; | |
8189 | ||
6c271ce2 | 8190 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
8191 | struct io_sq_data *sqd; |
8192 | ||
3ec482d1 | 8193 | ret = -EPERM; |
ce59fc69 | 8194 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
8195 | goto err; |
8196 | ||
534ca6d6 JA |
8197 | sqd = io_get_sq_data(p); |
8198 | if (IS_ERR(sqd)) { | |
8199 | ret = PTR_ERR(sqd); | |
8200 | goto err; | |
8201 | } | |
69fb2131 | 8202 | |
534ca6d6 | 8203 | ctx->sq_data = sqd; |
69fb2131 JA |
8204 | io_sq_thread_park(sqd); |
8205 | mutex_lock(&sqd->ctx_lock); | |
8206 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
8207 | mutex_unlock(&sqd->ctx_lock); | |
8208 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 8209 | |
917257da JA |
8210 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8211 | if (!ctx->sq_thread_idle) | |
8212 | ctx->sq_thread_idle = HZ; | |
8213 | ||
aa06165d JA |
8214 | if (sqd->thread) |
8215 | goto done; | |
8216 | ||
6c271ce2 | 8217 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8218 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8219 | |
917257da | 8220 | ret = -EINVAL; |
44a9bd18 JA |
8221 | if (cpu >= nr_cpu_ids) |
8222 | goto err; | |
7889f44d | 8223 | if (!cpu_online(cpu)) |
917257da JA |
8224 | goto err; |
8225 | ||
69fb2131 | 8226 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 8227 | cpu, "io_uring-sq"); |
6c271ce2 | 8228 | } else { |
69fb2131 | 8229 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
8230 | "io_uring-sq"); |
8231 | } | |
534ca6d6 JA |
8232 | if (IS_ERR(sqd->thread)) { |
8233 | ret = PTR_ERR(sqd->thread); | |
8234 | sqd->thread = NULL; | |
6c271ce2 JA |
8235 | goto err; |
8236 | } | |
534ca6d6 | 8237 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
8238 | if (ret) |
8239 | goto err; | |
6c271ce2 JA |
8240 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8241 | /* Can't have SQ_AFF without SQPOLL */ | |
8242 | ret = -EINVAL; | |
8243 | goto err; | |
8244 | } | |
8245 | ||
aa06165d | 8246 | done: |
24369c2e PB |
8247 | ret = io_init_wq_offload(ctx, p); |
8248 | if (ret) | |
2b188cc1 | 8249 | goto err; |
2b188cc1 JA |
8250 | |
8251 | return 0; | |
8252 | err: | |
54a91f3b | 8253 | io_finish_async(ctx); |
2b188cc1 JA |
8254 | return ret; |
8255 | } | |
8256 | ||
7e84e1c7 SG |
8257 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
8258 | { | |
534ca6d6 JA |
8259 | struct io_sq_data *sqd = ctx->sq_data; |
8260 | ||
8261 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
8262 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
8263 | } |
8264 | ||
a087e2b5 BM |
8265 | static inline void __io_unaccount_mem(struct user_struct *user, |
8266 | unsigned long nr_pages) | |
2b188cc1 JA |
8267 | { |
8268 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8269 | } | |
8270 | ||
a087e2b5 BM |
8271 | static inline int __io_account_mem(struct user_struct *user, |
8272 | unsigned long nr_pages) | |
2b188cc1 JA |
8273 | { |
8274 | unsigned long page_limit, cur_pages, new_pages; | |
8275 | ||
8276 | /* Don't allow more pages than we can safely lock */ | |
8277 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8278 | ||
8279 | do { | |
8280 | cur_pages = atomic_long_read(&user->locked_vm); | |
8281 | new_pages = cur_pages + nr_pages; | |
8282 | if (new_pages > page_limit) | |
8283 | return -ENOMEM; | |
8284 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8285 | new_pages) != cur_pages); | |
8286 | ||
8287 | return 0; | |
8288 | } | |
8289 | ||
26bfa89e | 8290 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8291 | { |
aad5d8da | 8292 | if (ctx->limit_mem) |
a087e2b5 | 8293 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8294 | |
26bfa89e JA |
8295 | if (ctx->mm_account) |
8296 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8297 | } |
8298 | ||
26bfa89e | 8299 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8300 | { |
30975825 BM |
8301 | int ret; |
8302 | ||
8303 | if (ctx->limit_mem) { | |
8304 | ret = __io_account_mem(ctx->user, nr_pages); | |
8305 | if (ret) | |
8306 | return ret; | |
8307 | } | |
8308 | ||
26bfa89e JA |
8309 | if (ctx->mm_account) |
8310 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8311 | |
8312 | return 0; | |
8313 | } | |
8314 | ||
2b188cc1 JA |
8315 | static void io_mem_free(void *ptr) |
8316 | { | |
52e04ef4 MR |
8317 | struct page *page; |
8318 | ||
8319 | if (!ptr) | |
8320 | return; | |
2b188cc1 | 8321 | |
52e04ef4 | 8322 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8323 | if (put_page_testzero(page)) |
8324 | free_compound_page(page); | |
8325 | } | |
8326 | ||
8327 | static void *io_mem_alloc(size_t size) | |
8328 | { | |
8329 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8330 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8331 | |
8332 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8333 | } | |
8334 | ||
75b28aff HV |
8335 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8336 | size_t *sq_offset) | |
8337 | { | |
8338 | struct io_rings *rings; | |
8339 | size_t off, sq_array_size; | |
8340 | ||
8341 | off = struct_size(rings, cqes, cq_entries); | |
8342 | if (off == SIZE_MAX) | |
8343 | return SIZE_MAX; | |
8344 | ||
8345 | #ifdef CONFIG_SMP | |
8346 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8347 | if (off == 0) | |
8348 | return SIZE_MAX; | |
8349 | #endif | |
8350 | ||
b36200f5 DV |
8351 | if (sq_offset) |
8352 | *sq_offset = off; | |
8353 | ||
75b28aff HV |
8354 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8355 | if (sq_array_size == SIZE_MAX) | |
8356 | return SIZE_MAX; | |
8357 | ||
8358 | if (check_add_overflow(off, sq_array_size, &off)) | |
8359 | return SIZE_MAX; | |
8360 | ||
75b28aff HV |
8361 | return off; |
8362 | } | |
8363 | ||
0a96bbe4 | 8364 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8365 | { |
8366 | int i, j; | |
8367 | ||
8368 | if (!ctx->user_bufs) | |
8369 | return -ENXIO; | |
8370 | ||
8371 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8372 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8373 | ||
8374 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8375 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8376 | |
de293938 | 8377 | if (imu->acct_pages) |
26bfa89e | 8378 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8379 | kvfree(imu->bvec); |
edafccee JA |
8380 | imu->nr_bvecs = 0; |
8381 | } | |
8382 | ||
8383 | kfree(ctx->user_bufs); | |
8384 | ctx->user_bufs = NULL; | |
8385 | ctx->nr_user_bufs = 0; | |
8386 | return 0; | |
8387 | } | |
8388 | ||
8389 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8390 | void __user *arg, unsigned index) | |
8391 | { | |
8392 | struct iovec __user *src; | |
8393 | ||
8394 | #ifdef CONFIG_COMPAT | |
8395 | if (ctx->compat) { | |
8396 | struct compat_iovec __user *ciovs; | |
8397 | struct compat_iovec ciov; | |
8398 | ||
8399 | ciovs = (struct compat_iovec __user *) arg; | |
8400 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8401 | return -EFAULT; | |
8402 | ||
d55e5f5b | 8403 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8404 | dst->iov_len = ciov.iov_len; |
8405 | return 0; | |
8406 | } | |
8407 | #endif | |
8408 | src = (struct iovec __user *) arg; | |
8409 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8410 | return -EFAULT; | |
8411 | return 0; | |
8412 | } | |
8413 | ||
de293938 JA |
8414 | /* |
8415 | * Not super efficient, but this is just a registration time. And we do cache | |
8416 | * the last compound head, so generally we'll only do a full search if we don't | |
8417 | * match that one. | |
8418 | * | |
8419 | * We check if the given compound head page has already been accounted, to | |
8420 | * avoid double accounting it. This allows us to account the full size of the | |
8421 | * page, not just the constituent pages of a huge page. | |
8422 | */ | |
8423 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8424 | int nr_pages, struct page *hpage) | |
8425 | { | |
8426 | int i, j; | |
8427 | ||
8428 | /* check current page array */ | |
8429 | for (i = 0; i < nr_pages; i++) { | |
8430 | if (!PageCompound(pages[i])) | |
8431 | continue; | |
8432 | if (compound_head(pages[i]) == hpage) | |
8433 | return true; | |
8434 | } | |
8435 | ||
8436 | /* check previously registered pages */ | |
8437 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8438 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8439 | ||
8440 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8441 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8442 | continue; | |
8443 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8444 | return true; | |
8445 | } | |
8446 | } | |
8447 | ||
8448 | return false; | |
8449 | } | |
8450 | ||
8451 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8452 | int nr_pages, struct io_mapped_ubuf *imu, | |
8453 | struct page **last_hpage) | |
8454 | { | |
8455 | int i, ret; | |
8456 | ||
8457 | for (i = 0; i < nr_pages; i++) { | |
8458 | if (!PageCompound(pages[i])) { | |
8459 | imu->acct_pages++; | |
8460 | } else { | |
8461 | struct page *hpage; | |
8462 | ||
8463 | hpage = compound_head(pages[i]); | |
8464 | if (hpage == *last_hpage) | |
8465 | continue; | |
8466 | *last_hpage = hpage; | |
8467 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8468 | continue; | |
8469 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8470 | } | |
8471 | } | |
8472 | ||
8473 | if (!imu->acct_pages) | |
8474 | return 0; | |
8475 | ||
26bfa89e | 8476 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8477 | if (ret) |
8478 | imu->acct_pages = 0; | |
8479 | return ret; | |
8480 | } | |
8481 | ||
0a96bbe4 BM |
8482 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8483 | struct io_mapped_ubuf *imu, | |
8484 | struct page **last_hpage) | |
edafccee JA |
8485 | { |
8486 | struct vm_area_struct **vmas = NULL; | |
8487 | struct page **pages = NULL; | |
0a96bbe4 BM |
8488 | unsigned long off, start, end, ubuf; |
8489 | size_t size; | |
8490 | int ret, pret, nr_pages, i; | |
8491 | ||
8492 | ubuf = (unsigned long) iov->iov_base; | |
8493 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8494 | start = ubuf >> PAGE_SHIFT; | |
8495 | nr_pages = end - start; | |
8496 | ||
8497 | ret = -ENOMEM; | |
8498 | ||
8499 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8500 | if (!pages) | |
8501 | goto done; | |
8502 | ||
8503 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8504 | GFP_KERNEL); | |
8505 | if (!vmas) | |
8506 | goto done; | |
8507 | ||
8508 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), | |
8509 | GFP_KERNEL); | |
8510 | if (!imu->bvec) | |
8511 | goto done; | |
8512 | ||
8513 | ret = 0; | |
8514 | mmap_read_lock(current->mm); | |
8515 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8516 | pages, vmas); | |
8517 | if (pret == nr_pages) { | |
8518 | /* don't support file backed memory */ | |
8519 | for (i = 0; i < nr_pages; i++) { | |
8520 | struct vm_area_struct *vma = vmas[i]; | |
8521 | ||
8522 | if (vma->vm_file && | |
8523 | !is_file_hugepages(vma->vm_file)) { | |
8524 | ret = -EOPNOTSUPP; | |
8525 | break; | |
8526 | } | |
8527 | } | |
8528 | } else { | |
8529 | ret = pret < 0 ? pret : -EFAULT; | |
8530 | } | |
8531 | mmap_read_unlock(current->mm); | |
8532 | if (ret) { | |
8533 | /* | |
8534 | * if we did partial map, or found file backed vmas, | |
8535 | * release any pages we did get | |
8536 | */ | |
8537 | if (pret > 0) | |
8538 | unpin_user_pages(pages, pret); | |
8539 | kvfree(imu->bvec); | |
8540 | goto done; | |
8541 | } | |
8542 | ||
8543 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8544 | if (ret) { | |
8545 | unpin_user_pages(pages, pret); | |
8546 | kvfree(imu->bvec); | |
8547 | goto done; | |
8548 | } | |
8549 | ||
8550 | off = ubuf & ~PAGE_MASK; | |
8551 | size = iov->iov_len; | |
8552 | for (i = 0; i < nr_pages; i++) { | |
8553 | size_t vec_len; | |
8554 | ||
8555 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8556 | imu->bvec[i].bv_page = pages[i]; | |
8557 | imu->bvec[i].bv_len = vec_len; | |
8558 | imu->bvec[i].bv_offset = off; | |
8559 | off = 0; | |
8560 | size -= vec_len; | |
8561 | } | |
8562 | /* store original address for later verification */ | |
8563 | imu->ubuf = ubuf; | |
8564 | imu->len = iov->iov_len; | |
8565 | imu->nr_bvecs = nr_pages; | |
8566 | ret = 0; | |
8567 | done: | |
8568 | kvfree(pages); | |
8569 | kvfree(vmas); | |
8570 | return ret; | |
8571 | } | |
8572 | ||
2b358604 | 8573 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8574 | { |
edafccee JA |
8575 | if (ctx->user_bufs) |
8576 | return -EBUSY; | |
8577 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8578 | return -EINVAL; | |
8579 | ||
8580 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8581 | GFP_KERNEL); | |
8582 | if (!ctx->user_bufs) | |
8583 | return -ENOMEM; | |
8584 | ||
2b358604 BM |
8585 | return 0; |
8586 | } | |
8587 | ||
8588 | static int io_buffer_validate(struct iovec *iov) | |
8589 | { | |
8590 | /* | |
8591 | * Don't impose further limits on the size and buffer | |
8592 | * constraints here, we'll -EINVAL later when IO is | |
8593 | * submitted if they are wrong. | |
8594 | */ | |
8595 | if (!iov->iov_base || !iov->iov_len) | |
8596 | return -EFAULT; | |
8597 | ||
8598 | /* arbitrary limit, but we need something */ | |
8599 | if (iov->iov_len > SZ_1G) | |
8600 | return -EFAULT; | |
8601 | ||
8602 | return 0; | |
8603 | } | |
8604 | ||
8605 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, | |
8606 | unsigned int nr_args) | |
8607 | { | |
8608 | int i, ret; | |
8609 | struct iovec iov; | |
8610 | struct page *last_hpage = NULL; | |
8611 | ||
8612 | ret = io_buffers_map_alloc(ctx, nr_args); | |
8613 | if (ret) | |
8614 | return ret; | |
8615 | ||
edafccee JA |
8616 | for (i = 0; i < nr_args; i++) { |
8617 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee JA |
8618 | |
8619 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8620 | if (ret) | |
0a96bbe4 | 8621 | break; |
edafccee | 8622 | |
2b358604 BM |
8623 | ret = io_buffer_validate(&iov); |
8624 | if (ret) | |
0a96bbe4 | 8625 | break; |
edafccee | 8626 | |
0a96bbe4 BM |
8627 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8628 | if (ret) | |
8629 | break; | |
edafccee JA |
8630 | |
8631 | ctx->nr_user_bufs++; | |
8632 | } | |
0a96bbe4 BM |
8633 | |
8634 | if (ret) | |
8635 | io_sqe_buffers_unregister(ctx); | |
8636 | ||
edafccee JA |
8637 | return ret; |
8638 | } | |
8639 | ||
9b402849 JA |
8640 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8641 | { | |
8642 | __s32 __user *fds = arg; | |
8643 | int fd; | |
8644 | ||
8645 | if (ctx->cq_ev_fd) | |
8646 | return -EBUSY; | |
8647 | ||
8648 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8649 | return -EFAULT; | |
8650 | ||
8651 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8652 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8653 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8654 | ctx->cq_ev_fd = NULL; | |
8655 | return ret; | |
8656 | } | |
8657 | ||
8658 | return 0; | |
8659 | } | |
8660 | ||
8661 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8662 | { | |
8663 | if (ctx->cq_ev_fd) { | |
8664 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8665 | ctx->cq_ev_fd = NULL; | |
8666 | return 0; | |
8667 | } | |
8668 | ||
8669 | return -ENXIO; | |
8670 | } | |
8671 | ||
5a2e745d JA |
8672 | static int __io_destroy_buffers(int id, void *p, void *data) |
8673 | { | |
8674 | struct io_ring_ctx *ctx = data; | |
8675 | struct io_buffer *buf = p; | |
8676 | ||
067524e9 | 8677 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8678 | return 0; |
8679 | } | |
8680 | ||
8681 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8682 | { | |
8683 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8684 | idr_destroy(&ctx->io_buffer_idr); | |
8685 | } | |
8686 | ||
c7dae4ba | 8687 | static void io_req_cache_free(struct list_head *list) |
1b4c351f | 8688 | { |
c7dae4ba | 8689 | while (!list_empty(list)) { |
1b4c351f JA |
8690 | struct io_kiocb *req; |
8691 | ||
c7dae4ba | 8692 | req = list_first_entry(list, struct io_kiocb, compl.list); |
1b4c351f JA |
8693 | list_del(&req->compl.list); |
8694 | kmem_cache_free(req_cachep, req); | |
8695 | } | |
8696 | } | |
8697 | ||
2b188cc1 JA |
8698 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8699 | { | |
bf019da7 PB |
8700 | struct io_submit_state *submit_state = &ctx->submit_state; |
8701 | ||
04fc6c80 PB |
8702 | /* |
8703 | * Some may use context even when all refs and requests have been put, | |
8704 | * and they are free to do so while still holding uring_lock, see | |
8705 | * __io_req_task_submit(). Wait for them to finish. | |
8706 | */ | |
8707 | mutex_lock(&ctx->uring_lock); | |
8708 | mutex_unlock(&ctx->uring_lock); | |
8709 | ||
6b06314c | 8710 | io_finish_async(ctx); |
0a96bbe4 | 8711 | io_sqe_buffers_unregister(ctx); |
2aede0e4 JA |
8712 | |
8713 | if (ctx->sqo_task) { | |
8714 | put_task_struct(ctx->sqo_task); | |
8715 | ctx->sqo_task = NULL; | |
8716 | mmdrop(ctx->mm_account); | |
8717 | ctx->mm_account = NULL; | |
30975825 | 8718 | } |
def596e9 | 8719 | |
bf019da7 PB |
8720 | if (submit_state->free_reqs) |
8721 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, | |
8722 | submit_state->reqs); | |
8723 | ||
91d8f519 DZ |
8724 | #ifdef CONFIG_BLK_CGROUP |
8725 | if (ctx->sqo_blkcg_css) | |
8726 | css_put(ctx->sqo_blkcg_css); | |
8727 | #endif | |
8728 | ||
6b06314c | 8729 | io_sqe_files_unregister(ctx); |
9b402849 | 8730 | io_eventfd_unregister(ctx); |
5a2e745d | 8731 | io_destroy_buffers(ctx); |
41726c9a | 8732 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8733 | |
2b188cc1 | 8734 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8735 | if (ctx->ring_sock) { |
8736 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8737 | sock_release(ctx->ring_sock); |
355e8d26 | 8738 | } |
2b188cc1 JA |
8739 | #endif |
8740 | ||
75b28aff | 8741 | io_mem_free(ctx->rings); |
2b188cc1 | 8742 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8743 | |
8744 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8745 | free_uid(ctx->user); |
181e448d | 8746 | put_cred(ctx->creds); |
78076bb6 | 8747 | kfree(ctx->cancel_hash); |
c7dae4ba JA |
8748 | io_req_cache_free(&ctx->submit_state.comp.free_list); |
8749 | io_req_cache_free(&ctx->submit_state.comp.locked_free_list); | |
2b188cc1 JA |
8750 | kfree(ctx); |
8751 | } | |
8752 | ||
8753 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8754 | { | |
8755 | struct io_ring_ctx *ctx = file->private_data; | |
8756 | __poll_t mask = 0; | |
8757 | ||
8758 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8759 | /* |
8760 | * synchronizes with barrier from wq_has_sleeper call in | |
8761 | * io_commit_cqring | |
8762 | */ | |
2b188cc1 | 8763 | smp_rmb(); |
90554200 | 8764 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8765 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8766 | |
8767 | /* | |
8768 | * Don't flush cqring overflow list here, just do a simple check. | |
8769 | * Otherwise there could possible be ABBA deadlock: | |
8770 | * CPU0 CPU1 | |
8771 | * ---- ---- | |
8772 | * lock(&ctx->uring_lock); | |
8773 | * lock(&ep->mtx); | |
8774 | * lock(&ctx->uring_lock); | |
8775 | * lock(&ep->mtx); | |
8776 | * | |
8777 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8778 | * pushs them to do the flush. | |
8779 | */ | |
8780 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8781 | mask |= EPOLLIN | EPOLLRDNORM; |
8782 | ||
8783 | return mask; | |
8784 | } | |
8785 | ||
8786 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8787 | { | |
8788 | struct io_ring_ctx *ctx = file->private_data; | |
8789 | ||
8790 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8791 | } | |
8792 | ||
0bead8cd | 8793 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8794 | { |
1e6fa521 | 8795 | struct io_identity *iod; |
071698e1 | 8796 | |
1e6fa521 JA |
8797 | iod = idr_remove(&ctx->personality_idr, id); |
8798 | if (iod) { | |
8799 | put_cred(iod->creds); | |
8800 | if (refcount_dec_and_test(&iod->count)) | |
8801 | kfree(iod); | |
0bead8cd | 8802 | return 0; |
1e6fa521 | 8803 | } |
0bead8cd YD |
8804 | |
8805 | return -EINVAL; | |
8806 | } | |
8807 | ||
8808 | static int io_remove_personalities(int id, void *p, void *data) | |
8809 | { | |
8810 | struct io_ring_ctx *ctx = data; | |
8811 | ||
8812 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8813 | return 0; |
8814 | } | |
8815 | ||
85faa7b8 JA |
8816 | static void io_ring_exit_work(struct work_struct *work) |
8817 | { | |
b2edc0a7 PB |
8818 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8819 | exit_work); | |
85faa7b8 | 8820 | |
56952e91 JA |
8821 | /* |
8822 | * If we're doing polled IO and end up having requests being | |
8823 | * submitted async (out-of-line), then completions can come in while | |
8824 | * we're waiting for refs to drop. We need to reap these manually, | |
8825 | * as nobody else will be looking for them. | |
8826 | */ | |
b2edc0a7 | 8827 | do { |
9936c7c2 | 8828 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8829 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8830 | io_ring_ctx_free(ctx); |
8831 | } | |
8832 | ||
00c18640 JA |
8833 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8834 | { | |
8835 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8836 | ||
8837 | return req->ctx == data; | |
8838 | } | |
8839 | ||
2b188cc1 JA |
8840 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8841 | { | |
8842 | mutex_lock(&ctx->uring_lock); | |
8843 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8844 | |
8845 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8846 | ctx->sqo_dead = 1; | |
8847 | ||
cda286f0 PB |
8848 | /* if force is set, the ring is going away. always drop after that */ |
8849 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8850 | if (ctx->rings) |
6c503150 | 8851 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8852 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8853 | mutex_unlock(&ctx->uring_lock); |
8854 | ||
6b81928d PB |
8855 | io_kill_timeouts(ctx, NULL, NULL); |
8856 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a JA |
8857 | |
8858 | if (ctx->io_wq) | |
00c18640 | 8859 | io_wq_cancel_cb(ctx->io_wq, io_cancel_ctx_cb, ctx, true); |
561fb04a | 8860 | |
15dff286 | 8861 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8862 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8863 | |
85faa7b8 | 8864 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8865 | /* |
8866 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8867 | * if we're exiting a ton of rings at the same time. It just adds | |
8868 | * noise and overhead, there's no discernable change in runtime | |
8869 | * over using system_wq. | |
8870 | */ | |
8871 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8872 | } |
8873 | ||
8874 | static int io_uring_release(struct inode *inode, struct file *file) | |
8875 | { | |
8876 | struct io_ring_ctx *ctx = file->private_data; | |
8877 | ||
8878 | file->private_data = NULL; | |
8879 | io_ring_ctx_wait_and_kill(ctx); | |
8880 | return 0; | |
8881 | } | |
8882 | ||
f6edbabb PB |
8883 | struct io_task_cancel { |
8884 | struct task_struct *task; | |
8885 | struct files_struct *files; | |
8886 | }; | |
f254ac04 | 8887 | |
f6edbabb | 8888 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8889 | { |
9a472ef7 | 8890 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8891 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8892 | bool ret; |
8893 | ||
f6edbabb | 8894 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8895 | unsigned long flags; |
8896 | struct io_ring_ctx *ctx = req->ctx; | |
8897 | ||
8898 | /* protect against races with linked timeouts */ | |
8899 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8900 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8901 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8902 | } else { | |
f6edbabb | 8903 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8904 | } |
8905 | return ret; | |
b711d4ea JA |
8906 | } |
8907 | ||
b7ddce3c | 8908 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8909 | struct task_struct *task, |
b7ddce3c PB |
8910 | struct files_struct *files) |
8911 | { | |
8912 | struct io_defer_entry *de = NULL; | |
8913 | LIST_HEAD(list); | |
8914 | ||
8915 | spin_lock_irq(&ctx->completion_lock); | |
8916 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8917 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8918 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8919 | break; | |
8920 | } | |
8921 | } | |
8922 | spin_unlock_irq(&ctx->completion_lock); | |
8923 | ||
8924 | while (!list_empty(&list)) { | |
8925 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8926 | list_del_init(&de->list); | |
8927 | req_set_fail_links(de->req); | |
8928 | io_put_req(de->req); | |
8929 | io_req_complete(de->req, -ECANCELED); | |
8930 | kfree(de); | |
8931 | } | |
8932 | } | |
8933 | ||
9936c7c2 PB |
8934 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8935 | struct task_struct *task, | |
8936 | struct files_struct *files) | |
8937 | { | |
8938 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
8939 | ||
8940 | while (1) { | |
8941 | enum io_wq_cancel cret; | |
8942 | bool ret = false; | |
8943 | ||
8944 | if (ctx->io_wq) { | |
8945 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, | |
8946 | &cancel, true); | |
8947 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8948 | } | |
8949 | ||
8950 | /* SQPOLL thread does its own polling */ | |
8951 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8952 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8953 | io_iopoll_try_reap_events(ctx); | |
8954 | ret = true; | |
8955 | } | |
8956 | } | |
8957 | ||
8958 | ret |= io_poll_remove_all(ctx, task, files); | |
8959 | ret |= io_kill_timeouts(ctx, task, files); | |
8960 | ret |= io_run_task_work(); | |
8961 | io_cqring_overflow_flush(ctx, true, task, files); | |
8962 | if (!ret) | |
8963 | break; | |
8964 | cond_resched(); | |
8965 | } | |
8966 | } | |
8967 | ||
ca70f00b PB |
8968 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8969 | struct task_struct *task, | |
8970 | struct files_struct *files) | |
8971 | { | |
8972 | struct io_kiocb *req; | |
8973 | int cnt = 0; | |
8974 | ||
8975 | spin_lock_irq(&ctx->inflight_lock); | |
8976 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8977 | cnt += io_match_task(req, task, files); | |
8978 | spin_unlock_irq(&ctx->inflight_lock); | |
8979 | return cnt; | |
8980 | } | |
8981 | ||
b52fda00 | 8982 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8983 | struct task_struct *task, |
fcb323cc JA |
8984 | struct files_struct *files) |
8985 | { | |
fcb323cc | 8986 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8987 | DEFINE_WAIT(wait); |
ca70f00b | 8988 | int inflight; |
fcb323cc | 8989 | |
ca70f00b PB |
8990 | inflight = io_uring_count_inflight(ctx, task, files); |
8991 | if (!inflight) | |
fcb323cc | 8992 | break; |
f6edbabb | 8993 | |
9936c7c2 | 8994 | io_uring_try_cancel_requests(ctx, task, files); |
34343786 PB |
8995 | |
8996 | if (ctx->sq_data) | |
8997 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
8998 | prepare_to_wait(&task->io_uring->wait, &wait, |
8999 | TASK_UNINTERRUPTIBLE); | |
9000 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
9001 | schedule(); | |
c98de08c | 9002 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
9003 | if (ctx->sq_data) |
9004 | io_sq_thread_park(ctx->sq_data); | |
fcb323cc | 9005 | } |
44e728b8 PB |
9006 | } |
9007 | ||
d9d05217 PB |
9008 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
9009 | { | |
d9d05217 PB |
9010 | mutex_lock(&ctx->uring_lock); |
9011 | ctx->sqo_dead = 1; | |
9012 | mutex_unlock(&ctx->uring_lock); | |
9013 | ||
9014 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
9015 | if (ctx->rings) |
9016 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
9017 | } |
9018 | ||
0f212204 JA |
9019 | /* |
9020 | * We need to iteratively cancel requests, in case a request has dependent | |
9021 | * hard links. These persist even for failure of cancelations, hence keep | |
9022 | * looping until none are found. | |
9023 | */ | |
9024 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
9025 | struct files_struct *files) | |
9026 | { | |
9027 | struct task_struct *task = current; | |
9028 | ||
fdaf083c | 9029 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 | 9030 | io_disable_sqo_submit(ctx); |
534ca6d6 | 9031 | task = ctx->sq_data->thread; |
fdaf083c JA |
9032 | atomic_inc(&task->io_uring->in_idle); |
9033 | io_sq_thread_park(ctx->sq_data); | |
9034 | } | |
0f212204 | 9035 | |
df9923f9 | 9036 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 9037 | |
3a7efd1a | 9038 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 9039 | if (!files) |
9936c7c2 | 9040 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c JA |
9041 | |
9042 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
9043 | atomic_dec(&task->io_uring->in_idle); | |
9044 | /* | |
9045 | * If the files that are going away are the ones in the thread | |
9046 | * identity, clear them out. | |
9047 | */ | |
9048 | if (task->io_uring->identity->files == files) | |
9049 | task->io_uring->identity->files = NULL; | |
9050 | io_sq_thread_unpark(ctx->sq_data); | |
9051 | } | |
0f212204 JA |
9052 | } |
9053 | ||
9054 | /* | |
9055 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9056 | */ | |
fdaf083c | 9057 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 9058 | { |
236434c3 | 9059 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 9060 | int ret; |
236434c3 MWO |
9061 | |
9062 | if (unlikely(!tctx)) { | |
0f212204 JA |
9063 | ret = io_uring_alloc_task_context(current); |
9064 | if (unlikely(ret)) | |
9065 | return ret; | |
236434c3 | 9066 | tctx = current->io_uring; |
0f212204 | 9067 | } |
236434c3 MWO |
9068 | if (tctx->last != file) { |
9069 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 9070 | |
236434c3 | 9071 | if (!old) { |
0f212204 | 9072 | get_file(file); |
a528b04e PB |
9073 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
9074 | file, GFP_KERNEL)); | |
9075 | if (ret) { | |
9076 | fput(file); | |
9077 | return ret; | |
9078 | } | |
ecfc8492 PB |
9079 | |
9080 | /* one and only SQPOLL file note, held by sqo_task */ | |
9081 | WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && | |
9082 | current != ctx->sqo_task); | |
0f212204 | 9083 | } |
236434c3 | 9084 | tctx->last = file; |
0f212204 JA |
9085 | } |
9086 | ||
fdaf083c JA |
9087 | /* |
9088 | * This is race safe in that the task itself is doing this, hence it | |
9089 | * cannot be going through the exit/cancel paths at the same time. | |
9090 | * This cannot be modified while exit/cancel is running. | |
9091 | */ | |
9092 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
9093 | tctx->sqpoll = true; | |
9094 | ||
0f212204 JA |
9095 | return 0; |
9096 | } | |
9097 | ||
9098 | /* | |
9099 | * Remove this io_uring_file -> task mapping. | |
9100 | */ | |
9101 | static void io_uring_del_task_file(struct file *file) | |
9102 | { | |
9103 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
9104 | |
9105 | if (tctx->last == file) | |
9106 | tctx->last = NULL; | |
5e2ed8c4 | 9107 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
9108 | if (file) |
9109 | fput(file); | |
9110 | } | |
9111 | ||
de7f1d9e PB |
9112 | static void io_uring_remove_task_files(struct io_uring_task *tctx) |
9113 | { | |
9114 | struct file *file; | |
9115 | unsigned long index; | |
9116 | ||
9117 | xa_for_each(&tctx->xa, index, file) | |
9118 | io_uring_del_task_file(file); | |
9119 | } | |
9120 | ||
0f212204 JA |
9121 | void __io_uring_files_cancel(struct files_struct *files) |
9122 | { | |
9123 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
9124 | struct file *file; |
9125 | unsigned long index; | |
0f212204 JA |
9126 | |
9127 | /* make sure overflow events are dropped */ | |
fdaf083c | 9128 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
9129 | xa_for_each(&tctx->xa, index, file) |
9130 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 9131 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9132 | |
9133 | if (files) | |
9134 | io_uring_remove_task_files(tctx); | |
fdaf083c JA |
9135 | } |
9136 | ||
9137 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
9138 | { | |
0e9ddb39 PB |
9139 | return percpu_counter_sum(&tctx->inflight); |
9140 | } | |
fdaf083c | 9141 | |
0e9ddb39 PB |
9142 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
9143 | { | |
9144 | struct io_uring_task *tctx; | |
9145 | s64 inflight; | |
9146 | DEFINE_WAIT(wait); | |
fdaf083c | 9147 | |
0e9ddb39 PB |
9148 | if (!ctx->sq_data) |
9149 | return; | |
9150 | tctx = ctx->sq_data->thread->io_uring; | |
9151 | io_disable_sqo_submit(ctx); | |
fdaf083c | 9152 | |
0e9ddb39 PB |
9153 | atomic_inc(&tctx->in_idle); |
9154 | do { | |
9155 | /* read completions before cancelations */ | |
9156 | inflight = tctx_inflight(tctx); | |
9157 | if (!inflight) | |
9158 | break; | |
9159 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 9160 | |
0e9ddb39 PB |
9161 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9162 | /* | |
9163 | * If we've seen completions, retry without waiting. This | |
9164 | * avoids a race where a completion comes in before we did | |
9165 | * prepare_to_wait(). | |
9166 | */ | |
9167 | if (inflight == tctx_inflight(tctx)) | |
9168 | schedule(); | |
9169 | finish_wait(&tctx->wait, &wait); | |
9170 | } while (1); | |
9171 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9172 | } |
9173 | ||
0f212204 JA |
9174 | /* |
9175 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9176 | * requests. | |
9177 | */ | |
9178 | void __io_uring_task_cancel(void) | |
9179 | { | |
9180 | struct io_uring_task *tctx = current->io_uring; | |
9181 | DEFINE_WAIT(wait); | |
d8a6df10 | 9182 | s64 inflight; |
0f212204 JA |
9183 | |
9184 | /* make sure overflow events are dropped */ | |
fdaf083c | 9185 | atomic_inc(&tctx->in_idle); |
0f212204 | 9186 | |
0b5cd6c3 | 9187 | /* trigger io_disable_sqo_submit() */ |
0e9ddb39 PB |
9188 | if (tctx->sqpoll) { |
9189 | struct file *file; | |
9190 | unsigned long index; | |
9191 | ||
9192 | xa_for_each(&tctx->xa, index, file) | |
9193 | io_uring_cancel_sqpoll(file->private_data); | |
9194 | } | |
0b5cd6c3 | 9195 | |
d8a6df10 | 9196 | do { |
0f212204 | 9197 | /* read completions before cancelations */ |
fdaf083c | 9198 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9199 | if (!inflight) |
9200 | break; | |
0f212204 JA |
9201 | __io_uring_files_cancel(NULL); |
9202 | ||
9203 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9204 | ||
9205 | /* | |
a1bb3cd5 PB |
9206 | * If we've seen completions, retry without waiting. This |
9207 | * avoids a race where a completion comes in before we did | |
9208 | * prepare_to_wait(). | |
0f212204 | 9209 | */ |
a1bb3cd5 PB |
9210 | if (inflight == tctx_inflight(tctx)) |
9211 | schedule(); | |
f57555ed | 9212 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9213 | } while (1); |
0f212204 | 9214 | |
fdaf083c | 9215 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9216 | |
9217 | io_uring_remove_task_files(tctx); | |
44e728b8 PB |
9218 | } |
9219 | ||
fcb323cc JA |
9220 | static int io_uring_flush(struct file *file, void *data) |
9221 | { | |
6b5733eb | 9222 | struct io_uring_task *tctx = current->io_uring; |
d9d05217 | 9223 | struct io_ring_ctx *ctx = file->private_data; |
6b5733eb | 9224 | |
84965ff8 JA |
9225 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) |
9226 | io_uring_cancel_task_requests(ctx, NULL); | |
9227 | ||
6b5733eb | 9228 | if (!tctx) |
4f793dc4 PB |
9229 | return 0; |
9230 | ||
6b5733eb PB |
9231 | /* we should have cancelled and erased it before PF_EXITING */ |
9232 | WARN_ON_ONCE((current->flags & PF_EXITING) && | |
9233 | xa_load(&tctx->xa, (unsigned long)file)); | |
9234 | ||
4f793dc4 PB |
9235 | /* |
9236 | * fput() is pending, will be 2 if the only other ref is our potential | |
9237 | * task file note. If the task is exiting, drop regardless of count. | |
9238 | */ | |
6b5733eb PB |
9239 | if (atomic_long_read(&file->f_count) != 2) |
9240 | return 0; | |
4f793dc4 | 9241 | |
d9d05217 PB |
9242 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
9243 | /* there is only one file note, which is owned by sqo_task */ | |
4325cb49 PB |
9244 | WARN_ON_ONCE(ctx->sqo_task != current && |
9245 | xa_load(&tctx->xa, (unsigned long)file)); | |
9246 | /* sqo_dead check is for when this happens after cancellation */ | |
9247 | WARN_ON_ONCE(ctx->sqo_task == current && !ctx->sqo_dead && | |
d9d05217 PB |
9248 | !xa_load(&tctx->xa, (unsigned long)file)); |
9249 | ||
9250 | io_disable_sqo_submit(ctx); | |
9251 | } | |
9252 | ||
9253 | if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current) | |
9254 | io_uring_del_task_file(file); | |
fcb323cc JA |
9255 | return 0; |
9256 | } | |
9257 | ||
6c5c240e RP |
9258 | static void *io_uring_validate_mmap_request(struct file *file, |
9259 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9260 | { |
2b188cc1 | 9261 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9262 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9263 | struct page *page; |
9264 | void *ptr; | |
9265 | ||
9266 | switch (offset) { | |
9267 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9268 | case IORING_OFF_CQ_RING: |
9269 | ptr = ctx->rings; | |
2b188cc1 JA |
9270 | break; |
9271 | case IORING_OFF_SQES: | |
9272 | ptr = ctx->sq_sqes; | |
9273 | break; | |
2b188cc1 | 9274 | default: |
6c5c240e | 9275 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9276 | } |
9277 | ||
9278 | page = virt_to_head_page(ptr); | |
a50b854e | 9279 | if (sz > page_size(page)) |
6c5c240e RP |
9280 | return ERR_PTR(-EINVAL); |
9281 | ||
9282 | return ptr; | |
9283 | } | |
9284 | ||
9285 | #ifdef CONFIG_MMU | |
9286 | ||
9287 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9288 | { | |
9289 | size_t sz = vma->vm_end - vma->vm_start; | |
9290 | unsigned long pfn; | |
9291 | void *ptr; | |
9292 | ||
9293 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9294 | if (IS_ERR(ptr)) | |
9295 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9296 | |
9297 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9298 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9299 | } | |
9300 | ||
6c5c240e RP |
9301 | #else /* !CONFIG_MMU */ |
9302 | ||
9303 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9304 | { | |
9305 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9306 | } | |
9307 | ||
9308 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9309 | { | |
9310 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9311 | } | |
9312 | ||
9313 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9314 | unsigned long addr, unsigned long len, | |
9315 | unsigned long pgoff, unsigned long flags) | |
9316 | { | |
9317 | void *ptr; | |
9318 | ||
9319 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9320 | if (IS_ERR(ptr)) | |
9321 | return PTR_ERR(ptr); | |
9322 | ||
9323 | return (unsigned long) ptr; | |
9324 | } | |
9325 | ||
9326 | #endif /* !CONFIG_MMU */ | |
9327 | ||
d9d05217 | 9328 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9329 | { |
d9d05217 | 9330 | int ret = 0; |
90554200 JA |
9331 | DEFINE_WAIT(wait); |
9332 | ||
9333 | do { | |
9334 | if (!io_sqring_full(ctx)) | |
9335 | break; | |
9336 | ||
9337 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
9338 | ||
d9d05217 PB |
9339 | if (unlikely(ctx->sqo_dead)) { |
9340 | ret = -EOWNERDEAD; | |
9341 | goto out; | |
9342 | } | |
9343 | ||
90554200 JA |
9344 | if (!io_sqring_full(ctx)) |
9345 | break; | |
9346 | ||
9347 | schedule(); | |
9348 | } while (!signal_pending(current)); | |
9349 | ||
9350 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9351 | out: |
9352 | return ret; | |
90554200 JA |
9353 | } |
9354 | ||
c73ebb68 HX |
9355 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9356 | struct __kernel_timespec __user **ts, | |
9357 | const sigset_t __user **sig) | |
9358 | { | |
9359 | struct io_uring_getevents_arg arg; | |
9360 | ||
9361 | /* | |
9362 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9363 | * is just a pointer to the sigset_t. | |
9364 | */ | |
9365 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9366 | *sig = (const sigset_t __user *) argp; | |
9367 | *ts = NULL; | |
9368 | return 0; | |
9369 | } | |
9370 | ||
9371 | /* | |
9372 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9373 | * timespec and sigset_t pointers if good. | |
9374 | */ | |
9375 | if (*argsz != sizeof(arg)) | |
9376 | return -EINVAL; | |
9377 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9378 | return -EFAULT; | |
9379 | *sig = u64_to_user_ptr(arg.sigmask); | |
9380 | *argsz = arg.sigmask_sz; | |
9381 | *ts = u64_to_user_ptr(arg.ts); | |
9382 | return 0; | |
9383 | } | |
9384 | ||
2b188cc1 | 9385 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9386 | u32, min_complete, u32, flags, const void __user *, argp, |
9387 | size_t, argsz) | |
2b188cc1 JA |
9388 | { |
9389 | struct io_ring_ctx *ctx; | |
9390 | long ret = -EBADF; | |
9391 | int submitted = 0; | |
9392 | struct fd f; | |
9393 | ||
4c6e277c | 9394 | io_run_task_work(); |
b41e9852 | 9395 | |
90554200 | 9396 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9397 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9398 | return -EINVAL; |
9399 | ||
9400 | f = fdget(fd); | |
9401 | if (!f.file) | |
9402 | return -EBADF; | |
9403 | ||
9404 | ret = -EOPNOTSUPP; | |
9405 | if (f.file->f_op != &io_uring_fops) | |
9406 | goto out_fput; | |
9407 | ||
9408 | ret = -ENXIO; | |
9409 | ctx = f.file->private_data; | |
9410 | if (!percpu_ref_tryget(&ctx->refs)) | |
9411 | goto out_fput; | |
9412 | ||
7e84e1c7 SG |
9413 | ret = -EBADFD; |
9414 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9415 | goto out; | |
9416 | ||
6c271ce2 JA |
9417 | /* |
9418 | * For SQ polling, the thread will do all submissions and completions. | |
9419 | * Just return the requested submit count, and wake the thread if | |
9420 | * we were asked to. | |
9421 | */ | |
b2a9eada | 9422 | ret = 0; |
6c271ce2 | 9423 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9424 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9425 | |
d9d05217 PB |
9426 | ret = -EOWNERDEAD; |
9427 | if (unlikely(ctx->sqo_dead)) | |
9428 | goto out; | |
6c271ce2 | 9429 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9430 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9431 | if (flags & IORING_ENTER_SQ_WAIT) { |
9432 | ret = io_sqpoll_wait_sq(ctx); | |
9433 | if (ret) | |
9434 | goto out; | |
9435 | } | |
6c271ce2 | 9436 | submitted = to_submit; |
b2a9eada | 9437 | } else if (to_submit) { |
fdaf083c | 9438 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9439 | if (unlikely(ret)) |
9440 | goto out; | |
2b188cc1 | 9441 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9442 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9443 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9444 | |
9445 | if (submitted != to_submit) | |
9446 | goto out; | |
2b188cc1 JA |
9447 | } |
9448 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9449 | const sigset_t __user *sig; |
9450 | struct __kernel_timespec __user *ts; | |
9451 | ||
9452 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9453 | if (unlikely(ret)) | |
9454 | goto out; | |
9455 | ||
2b188cc1 JA |
9456 | min_complete = min(min_complete, ctx->cq_entries); |
9457 | ||
32b2244a XW |
9458 | /* |
9459 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9460 | * space applications don't need to do io completion events | |
9461 | * polling again, they can rely on io_sq_thread to do polling | |
9462 | * work, which can reduce cpu usage and uring_lock contention. | |
9463 | */ | |
9464 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9465 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9466 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9467 | } else { |
c73ebb68 | 9468 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9469 | } |
2b188cc1 JA |
9470 | } |
9471 | ||
7c504e65 | 9472 | out: |
6805b32e | 9473 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9474 | out_fput: |
9475 | fdput(f); | |
9476 | return submitted ? submitted : ret; | |
9477 | } | |
9478 | ||
bebdb65e | 9479 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9480 | static int io_uring_show_cred(int id, void *p, void *data) |
9481 | { | |
6b47ab81 JA |
9482 | struct io_identity *iod = p; |
9483 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9484 | struct seq_file *m = data; |
9485 | struct user_namespace *uns = seq_user_ns(m); | |
9486 | struct group_info *gi; | |
9487 | kernel_cap_t cap; | |
9488 | unsigned __capi; | |
9489 | int g; | |
9490 | ||
9491 | seq_printf(m, "%5d\n", id); | |
9492 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9493 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9494 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9495 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9496 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9497 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9498 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9499 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9500 | seq_puts(m, "\n\tGroups:\t"); | |
9501 | gi = cred->group_info; | |
9502 | for (g = 0; g < gi->ngroups; g++) { | |
9503 | seq_put_decimal_ull(m, g ? " " : "", | |
9504 | from_kgid_munged(uns, gi->gid[g])); | |
9505 | } | |
9506 | seq_puts(m, "\n\tCapEff:\t"); | |
9507 | cap = cred->cap_effective; | |
9508 | CAP_FOR_EACH_U32(__capi) | |
9509 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9510 | seq_putc(m, '\n'); | |
9511 | return 0; | |
9512 | } | |
9513 | ||
9514 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9515 | { | |
dbbe9c64 | 9516 | struct io_sq_data *sq = NULL; |
fad8e0de | 9517 | bool has_lock; |
87ce955b JA |
9518 | int i; |
9519 | ||
fad8e0de JA |
9520 | /* |
9521 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9522 | * since fdinfo case grabs it in the opposite direction of normal use | |
9523 | * cases. If we fail to get the lock, we just don't iterate any | |
9524 | * structures that could be going away outside the io_uring mutex. | |
9525 | */ | |
9526 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9527 | ||
dbbe9c64 JQ |
9528 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9529 | sq = ctx->sq_data; | |
9530 | ||
9531 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9532 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9533 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9534 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9535 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9536 | |
87ce955b JA |
9537 | if (f) |
9538 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9539 | else | |
9540 | seq_printf(m, "%5u: <none>\n", i); | |
9541 | } | |
9542 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9543 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9544 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9545 | ||
9546 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9547 | (unsigned int) buf->len); | |
9548 | } | |
fad8e0de | 9549 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9550 | seq_printf(m, "Personalities:\n"); |
9551 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9552 | } | |
d7718a9d JA |
9553 | seq_printf(m, "PollList:\n"); |
9554 | spin_lock_irq(&ctx->completion_lock); | |
9555 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9556 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9557 | struct io_kiocb *req; | |
9558 | ||
9559 | hlist_for_each_entry(req, list, hash_node) | |
9560 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9561 | req->task->task_works != NULL); | |
9562 | } | |
9563 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9564 | if (has_lock) |
9565 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9566 | } |
9567 | ||
9568 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9569 | { | |
9570 | struct io_ring_ctx *ctx = f->private_data; | |
9571 | ||
9572 | if (percpu_ref_tryget(&ctx->refs)) { | |
9573 | __io_uring_show_fdinfo(ctx, m); | |
9574 | percpu_ref_put(&ctx->refs); | |
9575 | } | |
9576 | } | |
bebdb65e | 9577 | #endif |
87ce955b | 9578 | |
2b188cc1 JA |
9579 | static const struct file_operations io_uring_fops = { |
9580 | .release = io_uring_release, | |
fcb323cc | 9581 | .flush = io_uring_flush, |
2b188cc1 | 9582 | .mmap = io_uring_mmap, |
6c5c240e RP |
9583 | #ifndef CONFIG_MMU |
9584 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9585 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9586 | #endif | |
2b188cc1 JA |
9587 | .poll = io_uring_poll, |
9588 | .fasync = io_uring_fasync, | |
bebdb65e | 9589 | #ifdef CONFIG_PROC_FS |
87ce955b | 9590 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9591 | #endif |
2b188cc1 JA |
9592 | }; |
9593 | ||
9594 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9595 | struct io_uring_params *p) | |
9596 | { | |
75b28aff HV |
9597 | struct io_rings *rings; |
9598 | size_t size, sq_array_offset; | |
2b188cc1 | 9599 | |
bd740481 JA |
9600 | /* make sure these are sane, as we already accounted them */ |
9601 | ctx->sq_entries = p->sq_entries; | |
9602 | ctx->cq_entries = p->cq_entries; | |
9603 | ||
75b28aff HV |
9604 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9605 | if (size == SIZE_MAX) | |
9606 | return -EOVERFLOW; | |
9607 | ||
9608 | rings = io_mem_alloc(size); | |
9609 | if (!rings) | |
2b188cc1 JA |
9610 | return -ENOMEM; |
9611 | ||
75b28aff HV |
9612 | ctx->rings = rings; |
9613 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9614 | rings->sq_ring_mask = p->sq_entries - 1; | |
9615 | rings->cq_ring_mask = p->cq_entries - 1; | |
9616 | rings->sq_ring_entries = p->sq_entries; | |
9617 | rings->cq_ring_entries = p->cq_entries; | |
9618 | ctx->sq_mask = rings->sq_ring_mask; | |
9619 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9620 | |
9621 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9622 | if (size == SIZE_MAX) { |
9623 | io_mem_free(ctx->rings); | |
9624 | ctx->rings = NULL; | |
2b188cc1 | 9625 | return -EOVERFLOW; |
eb065d30 | 9626 | } |
2b188cc1 JA |
9627 | |
9628 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9629 | if (!ctx->sq_sqes) { |
9630 | io_mem_free(ctx->rings); | |
9631 | ctx->rings = NULL; | |
2b188cc1 | 9632 | return -ENOMEM; |
eb065d30 | 9633 | } |
2b188cc1 | 9634 | |
2b188cc1 JA |
9635 | return 0; |
9636 | } | |
9637 | ||
9faadcc8 PB |
9638 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9639 | { | |
9640 | int ret, fd; | |
9641 | ||
9642 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9643 | if (fd < 0) | |
9644 | return fd; | |
9645 | ||
9646 | ret = io_uring_add_task_file(ctx, file); | |
9647 | if (ret) { | |
9648 | put_unused_fd(fd); | |
9649 | return ret; | |
9650 | } | |
9651 | fd_install(fd, file); | |
9652 | return fd; | |
9653 | } | |
9654 | ||
2b188cc1 JA |
9655 | /* |
9656 | * Allocate an anonymous fd, this is what constitutes the application | |
9657 | * visible backing of an io_uring instance. The application mmaps this | |
9658 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9659 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9660 | */ | |
9faadcc8 | 9661 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9662 | { |
9663 | struct file *file; | |
9faadcc8 | 9664 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9665 | int ret; |
9666 | ||
2b188cc1 JA |
9667 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9668 | &ctx->ring_sock); | |
9669 | if (ret) | |
9faadcc8 | 9670 | return ERR_PTR(ret); |
2b188cc1 JA |
9671 | #endif |
9672 | ||
2b188cc1 JA |
9673 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9674 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9675 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9676 | if (IS_ERR(file)) { |
9677 | sock_release(ctx->ring_sock); | |
9678 | ctx->ring_sock = NULL; | |
9679 | } else { | |
9680 | ctx->ring_sock->file = file; | |
0f212204 | 9681 | } |
2b188cc1 | 9682 | #endif |
9faadcc8 | 9683 | return file; |
2b188cc1 JA |
9684 | } |
9685 | ||
7f13657d XW |
9686 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9687 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9688 | { |
9689 | struct user_struct *user = NULL; | |
9690 | struct io_ring_ctx *ctx; | |
9faadcc8 | 9691 | struct file *file; |
2b188cc1 JA |
9692 | int ret; |
9693 | ||
8110c1a6 | 9694 | if (!entries) |
2b188cc1 | 9695 | return -EINVAL; |
8110c1a6 JA |
9696 | if (entries > IORING_MAX_ENTRIES) { |
9697 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9698 | return -EINVAL; | |
9699 | entries = IORING_MAX_ENTRIES; | |
9700 | } | |
2b188cc1 JA |
9701 | |
9702 | /* | |
9703 | * Use twice as many entries for the CQ ring. It's possible for the | |
9704 | * application to drive a higher depth than the size of the SQ ring, | |
9705 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9706 | * some flexibility in overcommitting a bit. If the application has |
9707 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9708 | * of CQ ring entries manually. | |
2b188cc1 JA |
9709 | */ |
9710 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9711 | if (p->flags & IORING_SETUP_CQSIZE) { |
9712 | /* | |
9713 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9714 | * to a power-of-two, if it isn't already. We do NOT impose | |
9715 | * any cq vs sq ring sizing. | |
9716 | */ | |
eb2667b3 | 9717 | if (!p->cq_entries) |
33a107f0 | 9718 | return -EINVAL; |
8110c1a6 JA |
9719 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9720 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9721 | return -EINVAL; | |
9722 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9723 | } | |
eb2667b3 JQ |
9724 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9725 | if (p->cq_entries < p->sq_entries) | |
9726 | return -EINVAL; | |
33a107f0 JA |
9727 | } else { |
9728 | p->cq_entries = 2 * p->sq_entries; | |
9729 | } | |
2b188cc1 JA |
9730 | |
9731 | user = get_uid(current_user()); | |
2b188cc1 JA |
9732 | |
9733 | ctx = io_ring_ctx_alloc(p); | |
9734 | if (!ctx) { | |
2b188cc1 JA |
9735 | free_uid(user); |
9736 | return -ENOMEM; | |
9737 | } | |
9738 | ctx->compat = in_compat_syscall(); | |
26bfa89e | 9739 | ctx->limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9740 | ctx->user = user; |
0b8c0ec7 | 9741 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9742 | #ifdef CONFIG_AUDIT |
9743 | ctx->loginuid = current->loginuid; | |
9744 | ctx->sessionid = current->sessionid; | |
9745 | #endif | |
2aede0e4 JA |
9746 | ctx->sqo_task = get_task_struct(current); |
9747 | ||
9748 | /* | |
9749 | * This is just grabbed for accounting purposes. When a process exits, | |
9750 | * the mm is exited and dropped before the files, hence we need to hang | |
9751 | * on to this mm purely for the purposes of being able to unaccount | |
9752 | * memory (locked/pinned vm). It's not used for anything else. | |
9753 | */ | |
6b7898eb | 9754 | mmgrab(current->mm); |
2aede0e4 | 9755 | ctx->mm_account = current->mm; |
6b7898eb | 9756 | |
91d8f519 DZ |
9757 | #ifdef CONFIG_BLK_CGROUP |
9758 | /* | |
9759 | * The sq thread will belong to the original cgroup it was inited in. | |
9760 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9761 | * issued bios will be associated with the closest cgroup later in the | |
9762 | * block layer. | |
9763 | */ | |
9764 | rcu_read_lock(); | |
9765 | ctx->sqo_blkcg_css = blkcg_css(); | |
9766 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9767 | rcu_read_unlock(); | |
9768 | if (!ret) { | |
9769 | /* don't init against a dying cgroup, have the user try again */ | |
9770 | ctx->sqo_blkcg_css = NULL; | |
9771 | ret = -ENODEV; | |
9772 | goto err; | |
9773 | } | |
9774 | #endif | |
2b188cc1 JA |
9775 | ret = io_allocate_scq_urings(ctx, p); |
9776 | if (ret) | |
9777 | goto err; | |
9778 | ||
7e84e1c7 | 9779 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9780 | if (ret) |
9781 | goto err; | |
9782 | ||
7e84e1c7 SG |
9783 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9784 | io_sq_offload_start(ctx); | |
9785 | ||
2b188cc1 | 9786 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9787 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9788 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9789 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9790 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9791 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9792 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9793 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9794 | |
9795 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9796 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9797 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9798 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9799 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9800 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9801 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9802 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9803 | |
7f13657d XW |
9804 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9805 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9806 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 HX |
9807 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9808 | IORING_FEAT_EXT_ARG; | |
7f13657d XW |
9809 | |
9810 | if (copy_to_user(params, p, sizeof(*p))) { | |
9811 | ret = -EFAULT; | |
9812 | goto err; | |
9813 | } | |
d1719f70 | 9814 | |
9faadcc8 PB |
9815 | file = io_uring_get_file(ctx); |
9816 | if (IS_ERR(file)) { | |
9817 | ret = PTR_ERR(file); | |
9818 | goto err; | |
9819 | } | |
9820 | ||
044c1ab3 JA |
9821 | /* |
9822 | * Install ring fd as the very last thing, so we don't risk someone | |
9823 | * having closed it before we finish setup | |
9824 | */ | |
9faadcc8 PB |
9825 | ret = io_uring_install_fd(ctx, file); |
9826 | if (ret < 0) { | |
06585c49 | 9827 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9828 | /* fput will clean it up */ |
9829 | fput(file); | |
9830 | return ret; | |
9831 | } | |
044c1ab3 | 9832 | |
c826bd7a | 9833 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9834 | return ret; |
9835 | err: | |
d9d05217 | 9836 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9837 | io_ring_ctx_wait_and_kill(ctx); |
9838 | return ret; | |
9839 | } | |
9840 | ||
9841 | /* | |
9842 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9843 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9844 | * params structure passed in. | |
9845 | */ | |
9846 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9847 | { | |
9848 | struct io_uring_params p; | |
2b188cc1 JA |
9849 | int i; |
9850 | ||
9851 | if (copy_from_user(&p, params, sizeof(p))) | |
9852 | return -EFAULT; | |
9853 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9854 | if (p.resv[i]) | |
9855 | return -EINVAL; | |
9856 | } | |
9857 | ||
6c271ce2 | 9858 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9859 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9860 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9861 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9862 | return -EINVAL; |
9863 | ||
7f13657d | 9864 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9865 | } |
9866 | ||
9867 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9868 | struct io_uring_params __user *, params) | |
9869 | { | |
9870 | return io_uring_setup(entries, params); | |
9871 | } | |
9872 | ||
66f4af93 JA |
9873 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9874 | { | |
9875 | struct io_uring_probe *p; | |
9876 | size_t size; | |
9877 | int i, ret; | |
9878 | ||
9879 | size = struct_size(p, ops, nr_args); | |
9880 | if (size == SIZE_MAX) | |
9881 | return -EOVERFLOW; | |
9882 | p = kzalloc(size, GFP_KERNEL); | |
9883 | if (!p) | |
9884 | return -ENOMEM; | |
9885 | ||
9886 | ret = -EFAULT; | |
9887 | if (copy_from_user(p, arg, size)) | |
9888 | goto out; | |
9889 | ret = -EINVAL; | |
9890 | if (memchr_inv(p, 0, size)) | |
9891 | goto out; | |
9892 | ||
9893 | p->last_op = IORING_OP_LAST - 1; | |
9894 | if (nr_args > IORING_OP_LAST) | |
9895 | nr_args = IORING_OP_LAST; | |
9896 | ||
9897 | for (i = 0; i < nr_args; i++) { | |
9898 | p->ops[i].op = i; | |
9899 | if (!io_op_defs[i].not_supported) | |
9900 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9901 | } | |
9902 | p->ops_len = i; | |
9903 | ||
9904 | ret = 0; | |
9905 | if (copy_to_user(arg, p, size)) | |
9906 | ret = -EFAULT; | |
9907 | out: | |
9908 | kfree(p); | |
9909 | return ret; | |
9910 | } | |
9911 | ||
071698e1 JA |
9912 | static int io_register_personality(struct io_ring_ctx *ctx) |
9913 | { | |
1e6fa521 JA |
9914 | struct io_identity *id; |
9915 | int ret; | |
071698e1 | 9916 | |
1e6fa521 JA |
9917 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9918 | if (unlikely(!id)) | |
9919 | return -ENOMEM; | |
9920 | ||
9921 | io_init_identity(id); | |
9922 | id->creds = get_current_cred(); | |
9923 | ||
9924 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9925 | if (ret < 0) { | |
9926 | put_cred(id->creds); | |
9927 | kfree(id); | |
9928 | } | |
9929 | return ret; | |
071698e1 JA |
9930 | } |
9931 | ||
21b55dbc SG |
9932 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9933 | unsigned int nr_args) | |
9934 | { | |
9935 | struct io_uring_restriction *res; | |
9936 | size_t size; | |
9937 | int i, ret; | |
9938 | ||
7e84e1c7 SG |
9939 | /* Restrictions allowed only if rings started disabled */ |
9940 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9941 | return -EBADFD; | |
9942 | ||
21b55dbc | 9943 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9944 | if (ctx->restrictions.registered) |
21b55dbc SG |
9945 | return -EBUSY; |
9946 | ||
9947 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9948 | return -EINVAL; | |
9949 | ||
9950 | size = array_size(nr_args, sizeof(*res)); | |
9951 | if (size == SIZE_MAX) | |
9952 | return -EOVERFLOW; | |
9953 | ||
9954 | res = memdup_user(arg, size); | |
9955 | if (IS_ERR(res)) | |
9956 | return PTR_ERR(res); | |
9957 | ||
9958 | ret = 0; | |
9959 | ||
9960 | for (i = 0; i < nr_args; i++) { | |
9961 | switch (res[i].opcode) { | |
9962 | case IORING_RESTRICTION_REGISTER_OP: | |
9963 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9964 | ret = -EINVAL; | |
9965 | goto out; | |
9966 | } | |
9967 | ||
9968 | __set_bit(res[i].register_op, | |
9969 | ctx->restrictions.register_op); | |
9970 | break; | |
9971 | case IORING_RESTRICTION_SQE_OP: | |
9972 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9973 | ret = -EINVAL; | |
9974 | goto out; | |
9975 | } | |
9976 | ||
9977 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9978 | break; | |
9979 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9980 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9981 | break; | |
9982 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9983 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9984 | break; | |
9985 | default: | |
9986 | ret = -EINVAL; | |
9987 | goto out; | |
9988 | } | |
9989 | } | |
9990 | ||
9991 | out: | |
9992 | /* Reset all restrictions if an error happened */ | |
9993 | if (ret != 0) | |
9994 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9995 | else | |
7e84e1c7 | 9996 | ctx->restrictions.registered = true; |
21b55dbc SG |
9997 | |
9998 | kfree(res); | |
9999 | return ret; | |
10000 | } | |
10001 | ||
7e84e1c7 SG |
10002 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10003 | { | |
10004 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10005 | return -EBADFD; | |
10006 | ||
10007 | if (ctx->restrictions.registered) | |
10008 | ctx->restricted = 1; | |
10009 | ||
10010 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
10011 | ||
10012 | io_sq_offload_start(ctx); | |
10013 | ||
10014 | return 0; | |
10015 | } | |
10016 | ||
071698e1 JA |
10017 | static bool io_register_op_must_quiesce(int op) |
10018 | { | |
10019 | switch (op) { | |
10020 | case IORING_UNREGISTER_FILES: | |
10021 | case IORING_REGISTER_FILES_UPDATE: | |
10022 | case IORING_REGISTER_PROBE: | |
10023 | case IORING_REGISTER_PERSONALITY: | |
10024 | case IORING_UNREGISTER_PERSONALITY: | |
10025 | return false; | |
10026 | default: | |
10027 | return true; | |
10028 | } | |
10029 | } | |
10030 | ||
edafccee JA |
10031 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10032 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10033 | __releases(ctx->uring_lock) |
10034 | __acquires(ctx->uring_lock) | |
edafccee JA |
10035 | { |
10036 | int ret; | |
10037 | ||
35fa71a0 JA |
10038 | /* |
10039 | * We're inside the ring mutex, if the ref is already dying, then | |
10040 | * someone else killed the ctx or is already going through | |
10041 | * io_uring_register(). | |
10042 | */ | |
10043 | if (percpu_ref_is_dying(&ctx->refs)) | |
10044 | return -ENXIO; | |
10045 | ||
071698e1 | 10046 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10047 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10048 | |
05f3fb3c JA |
10049 | /* |
10050 | * Drop uring mutex before waiting for references to exit. If | |
10051 | * another thread is currently inside io_uring_enter() it might | |
10052 | * need to grab the uring_lock to make progress. If we hold it | |
10053 | * here across the drain wait, then we can deadlock. It's safe | |
10054 | * to drop the mutex here, since no new references will come in | |
10055 | * after we've killed the percpu ref. | |
10056 | */ | |
10057 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10058 | do { |
10059 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10060 | if (!ret) | |
10061 | break; | |
ed6930c9 JA |
10062 | ret = io_run_task_work_sig(); |
10063 | if (ret < 0) | |
10064 | break; | |
af9c1a44 JA |
10065 | } while (1); |
10066 | ||
05f3fb3c | 10067 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10068 | |
c150368b JA |
10069 | if (ret) { |
10070 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
10071 | goto out_quiesce; |
10072 | } | |
10073 | } | |
10074 | ||
10075 | if (ctx->restricted) { | |
10076 | if (opcode >= IORING_REGISTER_LAST) { | |
10077 | ret = -EINVAL; | |
10078 | goto out; | |
10079 | } | |
10080 | ||
10081 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
10082 | ret = -EACCES; | |
c150368b JA |
10083 | goto out; |
10084 | } | |
05f3fb3c | 10085 | } |
edafccee JA |
10086 | |
10087 | switch (opcode) { | |
10088 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 10089 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
10090 | break; |
10091 | case IORING_UNREGISTER_BUFFERS: | |
10092 | ret = -EINVAL; | |
10093 | if (arg || nr_args) | |
10094 | break; | |
0a96bbe4 | 10095 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10096 | break; |
6b06314c JA |
10097 | case IORING_REGISTER_FILES: |
10098 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
10099 | break; | |
10100 | case IORING_UNREGISTER_FILES: | |
10101 | ret = -EINVAL; | |
10102 | if (arg || nr_args) | |
10103 | break; | |
10104 | ret = io_sqe_files_unregister(ctx); | |
10105 | break; | |
c3a31e60 JA |
10106 | case IORING_REGISTER_FILES_UPDATE: |
10107 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
10108 | break; | |
9b402849 | 10109 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10110 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10111 | ret = -EINVAL; |
10112 | if (nr_args != 1) | |
10113 | break; | |
10114 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10115 | if (ret) |
10116 | break; | |
10117 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10118 | ctx->eventfd_async = 1; | |
10119 | else | |
10120 | ctx->eventfd_async = 0; | |
9b402849 JA |
10121 | break; |
10122 | case IORING_UNREGISTER_EVENTFD: | |
10123 | ret = -EINVAL; | |
10124 | if (arg || nr_args) | |
10125 | break; | |
10126 | ret = io_eventfd_unregister(ctx); | |
10127 | break; | |
66f4af93 JA |
10128 | case IORING_REGISTER_PROBE: |
10129 | ret = -EINVAL; | |
10130 | if (!arg || nr_args > 256) | |
10131 | break; | |
10132 | ret = io_probe(ctx, arg, nr_args); | |
10133 | break; | |
071698e1 JA |
10134 | case IORING_REGISTER_PERSONALITY: |
10135 | ret = -EINVAL; | |
10136 | if (arg || nr_args) | |
10137 | break; | |
10138 | ret = io_register_personality(ctx); | |
10139 | break; | |
10140 | case IORING_UNREGISTER_PERSONALITY: | |
10141 | ret = -EINVAL; | |
10142 | if (arg) | |
10143 | break; | |
10144 | ret = io_unregister_personality(ctx, nr_args); | |
10145 | break; | |
7e84e1c7 SG |
10146 | case IORING_REGISTER_ENABLE_RINGS: |
10147 | ret = -EINVAL; | |
10148 | if (arg || nr_args) | |
10149 | break; | |
10150 | ret = io_register_enable_rings(ctx); | |
10151 | break; | |
21b55dbc SG |
10152 | case IORING_REGISTER_RESTRICTIONS: |
10153 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10154 | break; | |
edafccee JA |
10155 | default: |
10156 | ret = -EINVAL; | |
10157 | break; | |
10158 | } | |
10159 | ||
21b55dbc | 10160 | out: |
071698e1 | 10161 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10162 | /* bring the ctx back to life */ |
05f3fb3c | 10163 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 10164 | out_quiesce: |
0f158b4c | 10165 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10166 | } |
edafccee JA |
10167 | return ret; |
10168 | } | |
10169 | ||
10170 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10171 | void __user *, arg, unsigned int, nr_args) | |
10172 | { | |
10173 | struct io_ring_ctx *ctx; | |
10174 | long ret = -EBADF; | |
10175 | struct fd f; | |
10176 | ||
10177 | f = fdget(fd); | |
10178 | if (!f.file) | |
10179 | return -EBADF; | |
10180 | ||
10181 | ret = -EOPNOTSUPP; | |
10182 | if (f.file->f_op != &io_uring_fops) | |
10183 | goto out_fput; | |
10184 | ||
10185 | ctx = f.file->private_data; | |
10186 | ||
10187 | mutex_lock(&ctx->uring_lock); | |
10188 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10189 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10190 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10191 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10192 | out_fput: |
10193 | fdput(f); | |
10194 | return ret; | |
10195 | } | |
10196 | ||
2b188cc1 JA |
10197 | static int __init io_uring_init(void) |
10198 | { | |
d7f62e82 SM |
10199 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10200 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10201 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10202 | } while (0) | |
10203 | ||
10204 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10205 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10206 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10207 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10208 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10209 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10210 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10211 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10212 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10213 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10214 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10215 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10216 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10217 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10218 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10219 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10220 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10221 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10222 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10223 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10224 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10225 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10226 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10227 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10228 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10229 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10230 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10231 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10232 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10233 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10234 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10235 | |
d3656344 | 10236 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10237 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10238 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10239 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10240 | return 0; |
10241 | }; | |
10242 | __initcall(io_uring_init); |