]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
ff002b30 | 77 | #include <linux/fs_struct.h> |
7d67af2c | 78 | #include <linux/splice.h> |
b41e9852 | 79 | #include <linux/task_work.h> |
bcf5a063 | 80 | #include <linux/pagemap.h> |
0f212204 | 81 | #include <linux/io_uring.h> |
91d8f519 | 82 | #include <linux/blk-cgroup.h> |
4ea33a97 | 83 | #include <linux/audit.h> |
2b188cc1 | 84 | |
c826bd7a DD |
85 | #define CREATE_TRACE_POINTS |
86 | #include <trace/events/io_uring.h> | |
87 | ||
2b188cc1 JA |
88 | #include <uapi/linux/io_uring.h> |
89 | ||
90 | #include "internal.h" | |
561fb04a | 91 | #include "io-wq.h" |
2b188cc1 | 92 | |
5277deaa | 93 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 94 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
95 | |
96 | /* | |
97 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
98 | */ | |
99 | #define IORING_FILE_TABLE_SHIFT 9 | |
100 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
101 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
102 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
103 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
104 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 105 | |
b16fed66 PB |
106 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
107 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
108 | IOSQE_BUFFER_SELECT) | |
109 | ||
2b188cc1 JA |
110 | struct io_uring { |
111 | u32 head ____cacheline_aligned_in_smp; | |
112 | u32 tail ____cacheline_aligned_in_smp; | |
113 | }; | |
114 | ||
1e84b97b | 115 | /* |
75b28aff HV |
116 | * This data is shared with the application through the mmap at offsets |
117 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
118 | * |
119 | * The offsets to the member fields are published through struct | |
120 | * io_sqring_offsets when calling io_uring_setup. | |
121 | */ | |
75b28aff | 122 | struct io_rings { |
1e84b97b SB |
123 | /* |
124 | * Head and tail offsets into the ring; the offsets need to be | |
125 | * masked to get valid indices. | |
126 | * | |
75b28aff HV |
127 | * The kernel controls head of the sq ring and the tail of the cq ring, |
128 | * and the application controls tail of the sq ring and the head of the | |
129 | * cq ring. | |
1e84b97b | 130 | */ |
75b28aff | 131 | struct io_uring sq, cq; |
1e84b97b | 132 | /* |
75b28aff | 133 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
134 | * ring_entries - 1) |
135 | */ | |
75b28aff HV |
136 | u32 sq_ring_mask, cq_ring_mask; |
137 | /* Ring sizes (constant, power of 2) */ | |
138 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
139 | /* |
140 | * Number of invalid entries dropped by the kernel due to | |
141 | * invalid index stored in array | |
142 | * | |
143 | * Written by the kernel, shouldn't be modified by the | |
144 | * application (i.e. get number of "new events" by comparing to | |
145 | * cached value). | |
146 | * | |
147 | * After a new SQ head value was read by the application this | |
148 | * counter includes all submissions that were dropped reaching | |
149 | * the new SQ head (and possibly more). | |
150 | */ | |
75b28aff | 151 | u32 sq_dropped; |
1e84b97b | 152 | /* |
0d9b5b3a | 153 | * Runtime SQ flags |
1e84b97b SB |
154 | * |
155 | * Written by the kernel, shouldn't be modified by the | |
156 | * application. | |
157 | * | |
158 | * The application needs a full memory barrier before checking | |
159 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
160 | */ | |
75b28aff | 161 | u32 sq_flags; |
0d9b5b3a SG |
162 | /* |
163 | * Runtime CQ flags | |
164 | * | |
165 | * Written by the application, shouldn't be modified by the | |
166 | * kernel. | |
167 | */ | |
168 | u32 cq_flags; | |
1e84b97b SB |
169 | /* |
170 | * Number of completion events lost because the queue was full; | |
171 | * this should be avoided by the application by making sure | |
0b4295b5 | 172 | * there are not more requests pending than there is space in |
1e84b97b SB |
173 | * the completion queue. |
174 | * | |
175 | * Written by the kernel, shouldn't be modified by the | |
176 | * application (i.e. get number of "new events" by comparing to | |
177 | * cached value). | |
178 | * | |
179 | * As completion events come in out of order this counter is not | |
180 | * ordered with any other data. | |
181 | */ | |
75b28aff | 182 | u32 cq_overflow; |
1e84b97b SB |
183 | /* |
184 | * Ring buffer of completion events. | |
185 | * | |
186 | * The kernel writes completion events fresh every time they are | |
187 | * produced, so the application is allowed to modify pending | |
188 | * entries. | |
189 | */ | |
75b28aff | 190 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
191 | }; |
192 | ||
45d189c6 PB |
193 | enum io_uring_cmd_flags { |
194 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 195 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
196 | }; |
197 | ||
edafccee JA |
198 | struct io_mapped_ubuf { |
199 | u64 ubuf; | |
200 | size_t len; | |
201 | struct bio_vec *bvec; | |
202 | unsigned int nr_bvecs; | |
de293938 | 203 | unsigned long acct_pages; |
edafccee JA |
204 | }; |
205 | ||
50238531 BM |
206 | struct io_ring_ctx; |
207 | ||
269bbe5f BM |
208 | struct io_rsrc_put { |
209 | struct list_head list; | |
50238531 BM |
210 | union { |
211 | void *rsrc; | |
212 | struct file *file; | |
213 | }; | |
269bbe5f BM |
214 | }; |
215 | ||
216 | struct fixed_rsrc_table { | |
65e19f54 | 217 | struct file **files; |
31b51510 JA |
218 | }; |
219 | ||
269bbe5f | 220 | struct fixed_rsrc_ref_node { |
05589553 XW |
221 | struct percpu_ref refs; |
222 | struct list_head node; | |
269bbe5f BM |
223 | struct list_head rsrc_list; |
224 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
225 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
226 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 227 | struct llist_node llist; |
e297822b | 228 | bool done; |
05589553 XW |
229 | }; |
230 | ||
269bbe5f BM |
231 | struct fixed_rsrc_data { |
232 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
233 | struct io_ring_ctx *ctx; |
234 | ||
269bbe5f | 235 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 236 | struct percpu_ref refs; |
05f3fb3c | 237 | struct completion done; |
8bad28d8 | 238 | bool quiesce; |
05f3fb3c JA |
239 | }; |
240 | ||
5a2e745d JA |
241 | struct io_buffer { |
242 | struct list_head list; | |
243 | __u64 addr; | |
244 | __s32 len; | |
245 | __u16 bid; | |
246 | }; | |
247 | ||
21b55dbc SG |
248 | struct io_restriction { |
249 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
250 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
251 | u8 sqe_flags_allowed; | |
252 | u8 sqe_flags_required; | |
7e84e1c7 | 253 | bool registered; |
21b55dbc SG |
254 | }; |
255 | ||
37d1e2e3 JA |
256 | enum { |
257 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
258 | IO_SQ_THREAD_SHOULD_PARK, | |
259 | }; | |
260 | ||
534ca6d6 JA |
261 | struct io_sq_data { |
262 | refcount_t refs; | |
69fb2131 JA |
263 | struct mutex lock; |
264 | ||
265 | /* ctx's that are using this sqd */ | |
266 | struct list_head ctx_list; | |
267 | struct list_head ctx_new_list; | |
268 | struct mutex ctx_lock; | |
269 | ||
534ca6d6 JA |
270 | struct task_struct *thread; |
271 | struct wait_queue_head wait; | |
08369246 XW |
272 | |
273 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
274 | int sq_cpu; |
275 | pid_t task_pid; | |
276 | ||
277 | unsigned long state; | |
278 | struct completion startup; | |
279 | struct completion completion; | |
280 | struct completion exited; | |
534ca6d6 JA |
281 | }; |
282 | ||
258b29a9 | 283 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 284 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 285 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 286 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
287 | |
288 | struct io_comp_state { | |
6dd0be1e | 289 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 290 | unsigned int nr; |
c7dae4ba JA |
291 | unsigned int locked_free_nr; |
292 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 293 | struct list_head free_list; |
c7dae4ba JA |
294 | /* IRQ completion list, under ->completion_lock */ |
295 | struct list_head locked_free_list; | |
258b29a9 PB |
296 | }; |
297 | ||
a1ab7b35 PB |
298 | struct io_submit_link { |
299 | struct io_kiocb *head; | |
300 | struct io_kiocb *last; | |
301 | }; | |
302 | ||
258b29a9 PB |
303 | struct io_submit_state { |
304 | struct blk_plug plug; | |
a1ab7b35 | 305 | struct io_submit_link link; |
258b29a9 PB |
306 | |
307 | /* | |
308 | * io_kiocb alloc cache | |
309 | */ | |
bf019da7 | 310 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
311 | unsigned int free_reqs; |
312 | ||
313 | bool plug_started; | |
314 | ||
315 | /* | |
316 | * Batch completion logic | |
317 | */ | |
318 | struct io_comp_state comp; | |
319 | ||
320 | /* | |
321 | * File reference cache | |
322 | */ | |
323 | struct file *file; | |
324 | unsigned int fd; | |
325 | unsigned int file_refs; | |
326 | unsigned int ios_left; | |
327 | }; | |
328 | ||
2b188cc1 JA |
329 | struct io_ring_ctx { |
330 | struct { | |
331 | struct percpu_ref refs; | |
332 | } ____cacheline_aligned_in_smp; | |
333 | ||
334 | struct { | |
335 | unsigned int flags; | |
e1d85334 | 336 | unsigned int compat: 1; |
e1d85334 RD |
337 | unsigned int cq_overflow_flushed: 1; |
338 | unsigned int drain_next: 1; | |
339 | unsigned int eventfd_async: 1; | |
21b55dbc | 340 | unsigned int restricted: 1; |
5f3f26f9 | 341 | unsigned int sqo_exec: 1; |
2b188cc1 | 342 | |
75b28aff HV |
343 | /* |
344 | * Ring buffer of indices into array of io_uring_sqe, which is | |
345 | * mmapped by the application using the IORING_OFF_SQES offset. | |
346 | * | |
347 | * This indirection could e.g. be used to assign fixed | |
348 | * io_uring_sqe entries to operations and only submit them to | |
349 | * the queue when needed. | |
350 | * | |
351 | * The kernel modifies neither the indices array nor the entries | |
352 | * array. | |
353 | */ | |
354 | u32 *sq_array; | |
2b188cc1 JA |
355 | unsigned cached_sq_head; |
356 | unsigned sq_entries; | |
357 | unsigned sq_mask; | |
6c271ce2 | 358 | unsigned sq_thread_idle; |
498ccd9e | 359 | unsigned cached_sq_dropped; |
2c3bac6d | 360 | unsigned cached_cq_overflow; |
ad3eb2c8 | 361 | unsigned long sq_check_overflow; |
de0617e4 | 362 | |
e941894e JA |
363 | /* hashed buffered write serialization */ |
364 | struct io_wq_hash *hash_map; | |
365 | ||
de0617e4 | 366 | struct list_head defer_list; |
5262f567 | 367 | struct list_head timeout_list; |
1d7bb1d5 | 368 | struct list_head cq_overflow_list; |
fcb323cc | 369 | |
ad3eb2c8 | 370 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
371 | } ____cacheline_aligned_in_smp; |
372 | ||
3c1a2ead JA |
373 | struct { |
374 | struct mutex uring_lock; | |
375 | wait_queue_head_t wait; | |
376 | } ____cacheline_aligned_in_smp; | |
377 | ||
378 | struct io_submit_state submit_state; | |
379 | ||
206aefde JA |
380 | struct io_rings *rings; |
381 | ||
2aede0e4 JA |
382 | /* Only used for accounting purposes */ |
383 | struct mm_struct *mm_account; | |
384 | ||
534ca6d6 JA |
385 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
386 | ||
90554200 | 387 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 388 | struct list_head sqd_list; |
75b28aff | 389 | |
6b06314c JA |
390 | /* |
391 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
392 | * readers must ensure that ->refs is alive as long as the file* is | |
393 | * used. Only updated through io_uring_register(2). | |
394 | */ | |
269bbe5f | 395 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
396 | unsigned nr_user_files; |
397 | ||
edafccee JA |
398 | /* if used, fixed mapped user buffers */ |
399 | unsigned nr_user_bufs; | |
400 | struct io_mapped_ubuf *user_bufs; | |
401 | ||
2b188cc1 JA |
402 | struct user_struct *user; |
403 | ||
0f158b4c JA |
404 | struct completion ref_comp; |
405 | struct completion sq_thread_comp; | |
206aefde JA |
406 | |
407 | #if defined(CONFIG_UNIX) | |
408 | struct socket *ring_sock; | |
409 | #endif | |
410 | ||
5a2e745d JA |
411 | struct idr io_buffer_idr; |
412 | ||
071698e1 JA |
413 | struct idr personality_idr; |
414 | ||
206aefde JA |
415 | struct { |
416 | unsigned cached_cq_tail; | |
417 | unsigned cq_entries; | |
418 | unsigned cq_mask; | |
419 | atomic_t cq_timeouts; | |
f010505b | 420 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 421 | unsigned long cq_check_overflow; |
206aefde JA |
422 | struct wait_queue_head cq_wait; |
423 | struct fasync_struct *cq_fasync; | |
424 | struct eventfd_ctx *cq_ev_fd; | |
425 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 426 | |
2b188cc1 JA |
427 | struct { |
428 | spinlock_t completion_lock; | |
e94f141b | 429 | |
def596e9 | 430 | /* |
540e32a0 | 431 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
432 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
433 | * For SQPOLL, only the single threaded io_sq_thread() will | |
434 | * manipulate the list, hence no extra locking is needed there. | |
435 | */ | |
540e32a0 | 436 | struct list_head iopoll_list; |
78076bb6 JA |
437 | struct hlist_head *cancel_hash; |
438 | unsigned cancel_hash_bits; | |
e94f141b | 439 | bool poll_multi_file; |
31b51510 | 440 | |
fcb323cc JA |
441 | spinlock_t inflight_lock; |
442 | struct list_head inflight_list; | |
2b188cc1 | 443 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 444 | |
269bbe5f BM |
445 | struct delayed_work rsrc_put_work; |
446 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
447 | struct list_head rsrc_ref_list; |
448 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 449 | |
21b55dbc | 450 | struct io_restriction restrictions; |
3c1a2ead | 451 | |
7c25c0d1 JA |
452 | /* exit task_work */ |
453 | struct callback_head *exit_task_work; | |
454 | ||
e941894e JA |
455 | struct wait_queue_head hash_wait; |
456 | ||
3c1a2ead JA |
457 | /* Keep this last, we don't need it for the fast path */ |
458 | struct work_struct exit_work; | |
2b188cc1 JA |
459 | }; |
460 | ||
09bb8394 JA |
461 | /* |
462 | * First field must be the file pointer in all the | |
463 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
464 | */ | |
221c5eb2 JA |
465 | struct io_poll_iocb { |
466 | struct file *file; | |
018043be | 467 | struct wait_queue_head *head; |
221c5eb2 | 468 | __poll_t events; |
8c838788 | 469 | bool done; |
221c5eb2 | 470 | bool canceled; |
392edb45 | 471 | struct wait_queue_entry wait; |
221c5eb2 JA |
472 | }; |
473 | ||
018043be PB |
474 | struct io_poll_remove { |
475 | struct file *file; | |
476 | u64 addr; | |
477 | }; | |
478 | ||
b5dba59e JA |
479 | struct io_close { |
480 | struct file *file; | |
b5dba59e JA |
481 | int fd; |
482 | }; | |
483 | ||
ad8a48ac JA |
484 | struct io_timeout_data { |
485 | struct io_kiocb *req; | |
486 | struct hrtimer timer; | |
487 | struct timespec64 ts; | |
488 | enum hrtimer_mode mode; | |
489 | }; | |
490 | ||
8ed8d3c3 JA |
491 | struct io_accept { |
492 | struct file *file; | |
493 | struct sockaddr __user *addr; | |
494 | int __user *addr_len; | |
495 | int flags; | |
09952e3e | 496 | unsigned long nofile; |
8ed8d3c3 JA |
497 | }; |
498 | ||
499 | struct io_sync { | |
500 | struct file *file; | |
501 | loff_t len; | |
502 | loff_t off; | |
503 | int flags; | |
d63d1b5e | 504 | int mode; |
8ed8d3c3 JA |
505 | }; |
506 | ||
fbf23849 JA |
507 | struct io_cancel { |
508 | struct file *file; | |
509 | u64 addr; | |
510 | }; | |
511 | ||
b29472ee JA |
512 | struct io_timeout { |
513 | struct file *file; | |
bfe68a22 PB |
514 | u32 off; |
515 | u32 target_seq; | |
135fcde8 | 516 | struct list_head list; |
90cd7e42 PB |
517 | /* head of the link, used by linked timeouts only */ |
518 | struct io_kiocb *head; | |
b29472ee JA |
519 | }; |
520 | ||
0bdf7a2d PB |
521 | struct io_timeout_rem { |
522 | struct file *file; | |
523 | u64 addr; | |
9c8e11b3 PB |
524 | |
525 | /* timeout update */ | |
526 | struct timespec64 ts; | |
527 | u32 flags; | |
0bdf7a2d PB |
528 | }; |
529 | ||
9adbd45d JA |
530 | struct io_rw { |
531 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
532 | struct kiocb kiocb; | |
533 | u64 addr; | |
534 | u64 len; | |
535 | }; | |
536 | ||
3fbb51c1 JA |
537 | struct io_connect { |
538 | struct file *file; | |
539 | struct sockaddr __user *addr; | |
540 | int addr_len; | |
541 | }; | |
542 | ||
e47293fd JA |
543 | struct io_sr_msg { |
544 | struct file *file; | |
fddaface | 545 | union { |
270a5940 | 546 | struct user_msghdr __user *umsg; |
fddaface JA |
547 | void __user *buf; |
548 | }; | |
e47293fd | 549 | int msg_flags; |
bcda7baa | 550 | int bgid; |
fddaface | 551 | size_t len; |
bcda7baa | 552 | struct io_buffer *kbuf; |
e47293fd JA |
553 | }; |
554 | ||
15b71abe JA |
555 | struct io_open { |
556 | struct file *file; | |
557 | int dfd; | |
15b71abe | 558 | struct filename *filename; |
c12cedf2 | 559 | struct open_how how; |
4022e7af | 560 | unsigned long nofile; |
15b71abe JA |
561 | }; |
562 | ||
269bbe5f | 563 | struct io_rsrc_update { |
05f3fb3c JA |
564 | struct file *file; |
565 | u64 arg; | |
566 | u32 nr_args; | |
567 | u32 offset; | |
568 | }; | |
569 | ||
4840e418 JA |
570 | struct io_fadvise { |
571 | struct file *file; | |
572 | u64 offset; | |
573 | u32 len; | |
574 | u32 advice; | |
575 | }; | |
576 | ||
c1ca757b JA |
577 | struct io_madvise { |
578 | struct file *file; | |
579 | u64 addr; | |
580 | u32 len; | |
581 | u32 advice; | |
582 | }; | |
583 | ||
3e4827b0 JA |
584 | struct io_epoll { |
585 | struct file *file; | |
586 | int epfd; | |
587 | int op; | |
588 | int fd; | |
589 | struct epoll_event event; | |
e47293fd JA |
590 | }; |
591 | ||
7d67af2c PB |
592 | struct io_splice { |
593 | struct file *file_out; | |
594 | struct file *file_in; | |
595 | loff_t off_out; | |
596 | loff_t off_in; | |
597 | u64 len; | |
598 | unsigned int flags; | |
599 | }; | |
600 | ||
ddf0322d JA |
601 | struct io_provide_buf { |
602 | struct file *file; | |
603 | __u64 addr; | |
604 | __s32 len; | |
605 | __u32 bgid; | |
606 | __u16 nbufs; | |
607 | __u16 bid; | |
608 | }; | |
609 | ||
1d9e1288 BM |
610 | struct io_statx { |
611 | struct file *file; | |
612 | int dfd; | |
613 | unsigned int mask; | |
614 | unsigned int flags; | |
e62753e4 | 615 | const char __user *filename; |
1d9e1288 BM |
616 | struct statx __user *buffer; |
617 | }; | |
618 | ||
36f4fa68 JA |
619 | struct io_shutdown { |
620 | struct file *file; | |
621 | int how; | |
622 | }; | |
623 | ||
80a261fd JA |
624 | struct io_rename { |
625 | struct file *file; | |
626 | int old_dfd; | |
627 | int new_dfd; | |
628 | struct filename *oldpath; | |
629 | struct filename *newpath; | |
630 | int flags; | |
631 | }; | |
632 | ||
14a1143b JA |
633 | struct io_unlink { |
634 | struct file *file; | |
635 | int dfd; | |
636 | int flags; | |
637 | struct filename *filename; | |
638 | }; | |
639 | ||
3ca405eb PB |
640 | struct io_completion { |
641 | struct file *file; | |
642 | struct list_head list; | |
0f7e466b | 643 | int cflags; |
3ca405eb PB |
644 | }; |
645 | ||
f499a021 JA |
646 | struct io_async_connect { |
647 | struct sockaddr_storage address; | |
648 | }; | |
649 | ||
03b1230c JA |
650 | struct io_async_msghdr { |
651 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
652 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
653 | struct iovec *free_iov; | |
03b1230c JA |
654 | struct sockaddr __user *uaddr; |
655 | struct msghdr msg; | |
b537916c | 656 | struct sockaddr_storage addr; |
03b1230c JA |
657 | }; |
658 | ||
f67676d1 JA |
659 | struct io_async_rw { |
660 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
661 | const struct iovec *free_iovec; |
662 | struct iov_iter iter; | |
227c0c96 | 663 | size_t bytes_done; |
bcf5a063 | 664 | struct wait_page_queue wpq; |
f67676d1 JA |
665 | }; |
666 | ||
6b47ee6e PB |
667 | enum { |
668 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
669 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
670 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
671 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
672 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 673 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 674 | |
6b47ee6e PB |
675 | REQ_F_FAIL_LINK_BIT, |
676 | REQ_F_INFLIGHT_BIT, | |
677 | REQ_F_CUR_POS_BIT, | |
678 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 679 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 680 | REQ_F_ISREG_BIT, |
99bc4c38 | 681 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 682 | REQ_F_POLLED_BIT, |
bcda7baa | 683 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 684 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 685 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 686 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
687 | |
688 | /* not a real bit, just to check we're not overflowing the space */ | |
689 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
690 | }; |
691 | ||
692 | enum { | |
693 | /* ctx owns file */ | |
694 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
695 | /* drain existing IO first */ | |
696 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
697 | /* linked sqes */ | |
698 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
699 | /* doesn't sever on completion < 0 */ | |
700 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
701 | /* IOSQE_ASYNC */ | |
702 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
703 | /* IOSQE_BUFFER_SELECT */ |
704 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 705 | |
6b47ee6e PB |
706 | /* fail rest of links */ |
707 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
708 | /* on inflight list */ | |
709 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
710 | /* read/write uses file position */ | |
711 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
712 | /* must not punt to workers */ | |
713 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 714 | /* has or had linked timeout */ |
6b47ee6e | 715 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
716 | /* regular file */ |
717 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
718 | /* needs cleanup */ |
719 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
720 | /* already went through poll handler */ |
721 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
722 | /* buffer already selected */ |
723 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
724 | /* doesn't need file table for this request */ |
725 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
726 | /* linked timeout is active, i.e. prepared by link's head */ |
727 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
728 | /* completion is deferred through io_comp_state */ |
729 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
730 | }; |
731 | ||
732 | struct async_poll { | |
733 | struct io_poll_iocb poll; | |
807abcb0 | 734 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
735 | }; |
736 | ||
7cbf1722 JA |
737 | struct io_task_work { |
738 | struct io_wq_work_node node; | |
739 | task_work_func_t func; | |
740 | }; | |
741 | ||
09bb8394 JA |
742 | /* |
743 | * NOTE! Each of the iocb union members has the file pointer | |
744 | * as the first entry in their struct definition. So you can | |
745 | * access the file pointer through any of the sub-structs, | |
746 | * or directly as just 'ki_filp' in this struct. | |
747 | */ | |
2b188cc1 | 748 | struct io_kiocb { |
221c5eb2 | 749 | union { |
09bb8394 | 750 | struct file *file; |
9adbd45d | 751 | struct io_rw rw; |
221c5eb2 | 752 | struct io_poll_iocb poll; |
018043be | 753 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
754 | struct io_accept accept; |
755 | struct io_sync sync; | |
fbf23849 | 756 | struct io_cancel cancel; |
b29472ee | 757 | struct io_timeout timeout; |
0bdf7a2d | 758 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 759 | struct io_connect connect; |
e47293fd | 760 | struct io_sr_msg sr_msg; |
15b71abe | 761 | struct io_open open; |
b5dba59e | 762 | struct io_close close; |
269bbe5f | 763 | struct io_rsrc_update rsrc_update; |
4840e418 | 764 | struct io_fadvise fadvise; |
c1ca757b | 765 | struct io_madvise madvise; |
3e4827b0 | 766 | struct io_epoll epoll; |
7d67af2c | 767 | struct io_splice splice; |
ddf0322d | 768 | struct io_provide_buf pbuf; |
1d9e1288 | 769 | struct io_statx statx; |
36f4fa68 | 770 | struct io_shutdown shutdown; |
80a261fd | 771 | struct io_rename rename; |
14a1143b | 772 | struct io_unlink unlink; |
3ca405eb PB |
773 | /* use only after cleaning per-op data, see io_clean_op() */ |
774 | struct io_completion compl; | |
221c5eb2 | 775 | }; |
2b188cc1 | 776 | |
e8c2bc1f JA |
777 | /* opcode allocated if it needs to store data for async defer */ |
778 | void *async_data; | |
d625c6ee | 779 | u8 opcode; |
65a6543d XW |
780 | /* polled IO has completed */ |
781 | u8 iopoll_completed; | |
2b188cc1 | 782 | |
4f4eeba8 | 783 | u16 buf_index; |
9cf7c104 | 784 | u32 result; |
4f4eeba8 | 785 | |
010e8e6b PB |
786 | struct io_ring_ctx *ctx; |
787 | unsigned int flags; | |
788 | refcount_t refs; | |
789 | struct task_struct *task; | |
790 | u64 user_data; | |
d7718a9d | 791 | |
f2f87370 | 792 | struct io_kiocb *link; |
269bbe5f | 793 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 794 | |
d21ffe7e PB |
795 | /* |
796 | * 1. used with ctx->iopoll_list with reads/writes | |
797 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
798 | */ | |
010e8e6b | 799 | struct list_head inflight_entry; |
7cbf1722 JA |
800 | union { |
801 | struct io_task_work io_task_work; | |
802 | struct callback_head task_work; | |
803 | }; | |
010e8e6b PB |
804 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
805 | struct hlist_node hash_node; | |
806 | struct async_poll *apoll; | |
807 | struct io_wq_work work; | |
2b188cc1 | 808 | }; |
05589553 | 809 | |
27dc8338 PB |
810 | struct io_defer_entry { |
811 | struct list_head list; | |
812 | struct io_kiocb *req; | |
9cf7c104 | 813 | u32 seq; |
2b188cc1 JA |
814 | }; |
815 | ||
d3656344 | 816 | struct io_op_def { |
d3656344 JA |
817 | /* needs req->file assigned */ |
818 | unsigned needs_file : 1; | |
d3656344 JA |
819 | /* hash wq insertion if file is a regular file */ |
820 | unsigned hash_reg_file : 1; | |
821 | /* unbound wq insertion if file is a non-regular file */ | |
822 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
823 | /* opcode is not supported by this kernel */ |
824 | unsigned not_supported : 1; | |
8a72758c JA |
825 | /* set if opcode supports polled "wait" */ |
826 | unsigned pollin : 1; | |
827 | unsigned pollout : 1; | |
bcda7baa JA |
828 | /* op supports buffer selection */ |
829 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
830 | /* must always have async data allocated */ |
831 | unsigned needs_async_data : 1; | |
27926b68 JA |
832 | /* should block plug */ |
833 | unsigned plug : 1; | |
e8c2bc1f JA |
834 | /* size of async data needed, if any */ |
835 | unsigned short async_size; | |
d3656344 JA |
836 | }; |
837 | ||
0918682b | 838 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
839 | [IORING_OP_NOP] = {}, |
840 | [IORING_OP_READV] = { | |
d3656344 JA |
841 | .needs_file = 1, |
842 | .unbound_nonreg_file = 1, | |
8a72758c | 843 | .pollin = 1, |
4d954c25 | 844 | .buffer_select = 1, |
e8c2bc1f | 845 | .needs_async_data = 1, |
27926b68 | 846 | .plug = 1, |
e8c2bc1f | 847 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 848 | }, |
0463b6c5 | 849 | [IORING_OP_WRITEV] = { |
d3656344 JA |
850 | .needs_file = 1, |
851 | .hash_reg_file = 1, | |
852 | .unbound_nonreg_file = 1, | |
8a72758c | 853 | .pollout = 1, |
e8c2bc1f | 854 | .needs_async_data = 1, |
27926b68 | 855 | .plug = 1, |
e8c2bc1f | 856 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 857 | }, |
0463b6c5 | 858 | [IORING_OP_FSYNC] = { |
d3656344 JA |
859 | .needs_file = 1, |
860 | }, | |
0463b6c5 | 861 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
862 | .needs_file = 1, |
863 | .unbound_nonreg_file = 1, | |
8a72758c | 864 | .pollin = 1, |
27926b68 | 865 | .plug = 1, |
e8c2bc1f | 866 | .async_size = sizeof(struct io_async_rw), |
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), |
d3656344 | 875 | }, |
0463b6c5 | 876 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
877 | .needs_file = 1, |
878 | .unbound_nonreg_file = 1, | |
879 | }, | |
0463b6c5 PB |
880 | [IORING_OP_POLL_REMOVE] = {}, |
881 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
882 | .needs_file = 1, |
883 | }, | |
0463b6c5 | 884 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
885 | .needs_file = 1, |
886 | .unbound_nonreg_file = 1, | |
8a72758c | 887 | .pollout = 1, |
e8c2bc1f JA |
888 | .needs_async_data = 1, |
889 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 890 | }, |
0463b6c5 | 891 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
892 | .needs_file = 1, |
893 | .unbound_nonreg_file = 1, | |
8a72758c | 894 | .pollin = 1, |
52de1fe1 | 895 | .buffer_select = 1, |
e8c2bc1f JA |
896 | .needs_async_data = 1, |
897 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 898 | }, |
0463b6c5 | 899 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
900 | .needs_async_data = 1, |
901 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 902 | }, |
9c8e11b3 PB |
903 | [IORING_OP_TIMEOUT_REMOVE] = { |
904 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 905 | }, |
0463b6c5 | 906 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
907 | .needs_file = 1, |
908 | .unbound_nonreg_file = 1, | |
8a72758c | 909 | .pollin = 1, |
d3656344 | 910 | }, |
0463b6c5 PB |
911 | [IORING_OP_ASYNC_CANCEL] = {}, |
912 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
913 | .needs_async_data = 1, |
914 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 915 | }, |
0463b6c5 | 916 | [IORING_OP_CONNECT] = { |
d3656344 JA |
917 | .needs_file = 1, |
918 | .unbound_nonreg_file = 1, | |
8a72758c | 919 | .pollout = 1, |
e8c2bc1f JA |
920 | .needs_async_data = 1, |
921 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 922 | }, |
0463b6c5 | 923 | [IORING_OP_FALLOCATE] = { |
d3656344 | 924 | .needs_file = 1, |
d3656344 | 925 | }, |
44526bed JA |
926 | [IORING_OP_OPENAT] = {}, |
927 | [IORING_OP_CLOSE] = {}, | |
928 | [IORING_OP_FILES_UPDATE] = {}, | |
929 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 930 | [IORING_OP_READ] = { |
3a6820f2 JA |
931 | .needs_file = 1, |
932 | .unbound_nonreg_file = 1, | |
8a72758c | 933 | .pollin = 1, |
bcda7baa | 934 | .buffer_select = 1, |
27926b68 | 935 | .plug = 1, |
e8c2bc1f | 936 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
8a72758c | 941 | .pollout = 1, |
27926b68 | 942 | .plug = 1, |
e8c2bc1f | 943 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 944 | }, |
0463b6c5 | 945 | [IORING_OP_FADVISE] = { |
4840e418 | 946 | .needs_file = 1, |
c1ca757b | 947 | }, |
44526bed | 948 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 949 | [IORING_OP_SEND] = { |
fddaface JA |
950 | .needs_file = 1, |
951 | .unbound_nonreg_file = 1, | |
8a72758c | 952 | .pollout = 1, |
fddaface | 953 | }, |
0463b6c5 | 954 | [IORING_OP_RECV] = { |
fddaface JA |
955 | .needs_file = 1, |
956 | .unbound_nonreg_file = 1, | |
8a72758c | 957 | .pollin = 1, |
bcda7baa | 958 | .buffer_select = 1, |
fddaface | 959 | }, |
0463b6c5 | 960 | [IORING_OP_OPENAT2] = { |
cebdb986 | 961 | }, |
3e4827b0 JA |
962 | [IORING_OP_EPOLL_CTL] = { |
963 | .unbound_nonreg_file = 1, | |
3e4827b0 | 964 | }, |
7d67af2c PB |
965 | [IORING_OP_SPLICE] = { |
966 | .needs_file = 1, | |
967 | .hash_reg_file = 1, | |
968 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
969 | }, |
970 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 971 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
972 | [IORING_OP_TEE] = { |
973 | .needs_file = 1, | |
974 | .hash_reg_file = 1, | |
975 | .unbound_nonreg_file = 1, | |
976 | }, | |
36f4fa68 JA |
977 | [IORING_OP_SHUTDOWN] = { |
978 | .needs_file = 1, | |
979 | }, | |
44526bed JA |
980 | [IORING_OP_RENAMEAT] = {}, |
981 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
982 | }; |
983 | ||
9936c7c2 PB |
984 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
985 | struct task_struct *task, | |
986 | struct files_struct *files); | |
37d1e2e3 | 987 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 988 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 989 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 990 | struct io_ring_ctx *ctx); |
f2303b1f | 991 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 992 | |
23faba36 | 993 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 994 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 995 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 996 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 997 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
998 | static void io_dismantle_req(struct io_kiocb *req); |
999 | static void io_put_task(struct task_struct *task, int nr); | |
1000 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1001 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1002 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1003 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1004 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1005 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1006 | unsigned nr_args); |
3ca405eb | 1007 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1008 | static struct file *io_file_get(struct io_submit_state *state, |
1009 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1010 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1011 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1012 | |
847595de PB |
1013 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1014 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1015 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1016 | const struct iovec *fast_iov, | |
227c0c96 | 1017 | struct iov_iter *iter, bool force); |
907d1df3 | 1018 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1019 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1020 | struct io_ring_ctx *ctx); | |
de0617e4 | 1021 | |
2b188cc1 JA |
1022 | static struct kmem_cache *req_cachep; |
1023 | ||
0918682b | 1024 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1025 | |
1026 | struct sock *io_uring_get_socket(struct file *file) | |
1027 | { | |
1028 | #if defined(CONFIG_UNIX) | |
1029 | if (file->f_op == &io_uring_fops) { | |
1030 | struct io_ring_ctx *ctx = file->private_data; | |
1031 | ||
1032 | return ctx->ring_sock->sk; | |
1033 | } | |
1034 | #endif | |
1035 | return NULL; | |
1036 | } | |
1037 | EXPORT_SYMBOL(io_uring_get_socket); | |
1038 | ||
f2f87370 PB |
1039 | #define io_for_each_link(pos, head) \ |
1040 | for (pos = (head); pos; pos = pos->link) | |
1041 | ||
3ca405eb PB |
1042 | static inline void io_clean_op(struct io_kiocb *req) |
1043 | { | |
9d5c8190 | 1044 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1045 | __io_clean_op(req); |
1046 | } | |
1047 | ||
36f72fe2 PB |
1048 | static inline void io_set_resource_node(struct io_kiocb *req) |
1049 | { | |
1050 | struct io_ring_ctx *ctx = req->ctx; | |
1051 | ||
269bbe5f BM |
1052 | if (!req->fixed_rsrc_refs) { |
1053 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1054 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1055 | } |
1056 | } | |
1057 | ||
08d23634 PB |
1058 | static bool io_match_task(struct io_kiocb *head, |
1059 | struct task_struct *task, | |
1060 | struct files_struct *files) | |
1061 | { | |
1062 | struct io_kiocb *req; | |
1063 | ||
84965ff8 JA |
1064 | if (task && head->task != task) { |
1065 | /* in terms of cancelation, always match if req task is dead */ | |
1066 | if (head->task->flags & PF_EXITING) | |
1067 | return true; | |
08d23634 | 1068 | return false; |
84965ff8 | 1069 | } |
08d23634 PB |
1070 | if (!files) |
1071 | return true; | |
1072 | ||
1073 | io_for_each_link(req, head) { | |
02a13674 JA |
1074 | if (req->file && req->file->f_op == &io_uring_fops) |
1075 | return true; | |
4379bf8b | 1076 | if (req->task->files == files) |
08d23634 PB |
1077 | return true; |
1078 | } | |
1079 | return false; | |
1080 | } | |
1081 | ||
c40f6379 JA |
1082 | static inline void req_set_fail_links(struct io_kiocb *req) |
1083 | { | |
1084 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1085 | req->flags |= REQ_F_FAIL_LINK; | |
1086 | } | |
4a38aed2 | 1087 | |
2b188cc1 JA |
1088 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1089 | { | |
1090 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1091 | ||
0f158b4c | 1092 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1093 | } |
1094 | ||
8eb7e2d0 PB |
1095 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1096 | { | |
1097 | return !req->timeout.off; | |
1098 | } | |
1099 | ||
2b188cc1 JA |
1100 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1101 | { | |
1102 | struct io_ring_ctx *ctx; | |
78076bb6 | 1103 | int hash_bits; |
2b188cc1 JA |
1104 | |
1105 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1106 | if (!ctx) | |
1107 | return NULL; | |
1108 | ||
78076bb6 JA |
1109 | /* |
1110 | * Use 5 bits less than the max cq entries, that should give us around | |
1111 | * 32 entries per hash list if totally full and uniformly spread. | |
1112 | */ | |
1113 | hash_bits = ilog2(p->cq_entries); | |
1114 | hash_bits -= 5; | |
1115 | if (hash_bits <= 0) | |
1116 | hash_bits = 1; | |
1117 | ctx->cancel_hash_bits = hash_bits; | |
1118 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1119 | GFP_KERNEL); | |
1120 | if (!ctx->cancel_hash) | |
1121 | goto err; | |
1122 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1123 | ||
21482896 | 1124 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1125 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1126 | goto err; | |
2b188cc1 JA |
1127 | |
1128 | ctx->flags = p->flags; | |
90554200 | 1129 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1130 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1131 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1132 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1133 | init_completion(&ctx->ref_comp); |
1134 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1135 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1136 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1137 | mutex_init(&ctx->uring_lock); |
1138 | init_waitqueue_head(&ctx->wait); | |
1139 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1140 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1141 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1142 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1143 | spin_lock_init(&ctx->inflight_lock); |
1144 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1145 | spin_lock_init(&ctx->rsrc_ref_lock); |
1146 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1147 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1148 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1149 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1150 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1151 | return ctx; |
206aefde | 1152 | err: |
78076bb6 | 1153 | kfree(ctx->cancel_hash); |
206aefde JA |
1154 | kfree(ctx); |
1155 | return NULL; | |
2b188cc1 JA |
1156 | } |
1157 | ||
9cf7c104 | 1158 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1159 | { |
2bc9930e JA |
1160 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1161 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1162 | |
9cf7c104 | 1163 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1164 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1165 | } |
de0617e4 | 1166 | |
9d858b21 | 1167 | return false; |
de0617e4 JA |
1168 | } |
1169 | ||
4edf20f9 | 1170 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1171 | { |
34e08fed PB |
1172 | if (req->flags & REQ_F_INFLIGHT) { |
1173 | struct io_ring_ctx *ctx = req->ctx; | |
1174 | struct io_uring_task *tctx = req->task->io_uring; | |
1175 | unsigned long flags; | |
1176 | ||
1177 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1178 | list_del(&req->inflight_entry); | |
1179 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1180 | req->flags &= ~REQ_F_INFLIGHT; | |
1181 | if (atomic_read(&tctx->in_idle)) | |
1182 | wake_up(&tctx->wait); | |
1183 | } | |
1e6fa521 JA |
1184 | } |
1185 | ||
ce3d5aae PB |
1186 | static void io_req_track_inflight(struct io_kiocb *req) |
1187 | { | |
1188 | struct io_ring_ctx *ctx = req->ctx; | |
1189 | ||
1190 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1191 | req->flags |= REQ_F_INFLIGHT; |
1192 | ||
1193 | spin_lock_irq(&ctx->inflight_lock); | |
1194 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1195 | spin_unlock_irq(&ctx->inflight_lock); | |
1196 | } | |
1197 | } | |
1198 | ||
1e6fa521 JA |
1199 | static void io_prep_async_work(struct io_kiocb *req) |
1200 | { | |
1201 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1202 | struct io_ring_ctx *ctx = req->ctx; |
1203 | ||
feaadc4f PB |
1204 | if (req->flags & REQ_F_FORCE_ASYNC) |
1205 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1206 | ||
1e6fa521 JA |
1207 | if (req->flags & REQ_F_ISREG) { |
1208 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1209 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1210 | } else { | |
1211 | if (def->unbound_nonreg_file) | |
1212 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1213 | } | |
561fb04a | 1214 | } |
cccf0ee8 | 1215 | |
cbdcb435 | 1216 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1217 | { |
cbdcb435 | 1218 | struct io_kiocb *cur; |
54a91f3b | 1219 | |
f2f87370 PB |
1220 | io_for_each_link(cur, req) |
1221 | io_prep_async_work(cur); | |
561fb04a JA |
1222 | } |
1223 | ||
7271ef3a | 1224 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1225 | { |
a197f664 | 1226 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1227 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1228 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1229 | |
3bfe6106 JA |
1230 | BUG_ON(!tctx); |
1231 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1232 | |
8766dd51 PB |
1233 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1234 | &req->work, req->flags); | |
5aa75ed5 | 1235 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a | 1236 | return link; |
18d9be1a JA |
1237 | } |
1238 | ||
cbdcb435 PB |
1239 | static void io_queue_async_work(struct io_kiocb *req) |
1240 | { | |
7271ef3a JA |
1241 | struct io_kiocb *link; |
1242 | ||
cbdcb435 PB |
1243 | /* init ->work of the whole link before punting */ |
1244 | io_prep_async_link(req); | |
7271ef3a JA |
1245 | link = __io_queue_async_work(req); |
1246 | ||
1247 | if (link) | |
1248 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1249 | } |
1250 | ||
5262f567 JA |
1251 | static void io_kill_timeout(struct io_kiocb *req) |
1252 | { | |
e8c2bc1f | 1253 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1254 | int ret; |
1255 | ||
e8c2bc1f | 1256 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1257 | if (ret != -1) { |
01cec8c1 PB |
1258 | atomic_set(&req->ctx->cq_timeouts, |
1259 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1260 | list_del_init(&req->timeout.list); |
78e19bbe | 1261 | io_cqring_fill_event(req, 0); |
216578e5 | 1262 | io_put_req_deferred(req, 1); |
5262f567 JA |
1263 | } |
1264 | } | |
1265 | ||
76e1b642 JA |
1266 | /* |
1267 | * Returns true if we found and killed one or more timeouts | |
1268 | */ | |
6b81928d PB |
1269 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1270 | struct files_struct *files) | |
5262f567 JA |
1271 | { |
1272 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1273 | int canceled = 0; |
5262f567 JA |
1274 | |
1275 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1276 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1277 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1278 | io_kill_timeout(req); |
76e1b642 JA |
1279 | canceled++; |
1280 | } | |
f3606e3a | 1281 | } |
5262f567 | 1282 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1283 | return canceled != 0; |
5262f567 JA |
1284 | } |
1285 | ||
04518945 | 1286 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1287 | { |
04518945 | 1288 | do { |
27dc8338 PB |
1289 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1290 | struct io_defer_entry, list); | |
de0617e4 | 1291 | |
9cf7c104 | 1292 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1293 | break; |
27dc8338 | 1294 | list_del_init(&de->list); |
907d1df3 | 1295 | io_req_task_queue(de->req); |
27dc8338 | 1296 | kfree(de); |
04518945 PB |
1297 | } while (!list_empty(&ctx->defer_list)); |
1298 | } | |
1299 | ||
360428f8 | 1300 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1301 | { |
f010505b MDG |
1302 | u32 seq; |
1303 | ||
1304 | if (list_empty(&ctx->timeout_list)) | |
1305 | return; | |
1306 | ||
1307 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1308 | ||
1309 | do { | |
1310 | u32 events_needed, events_got; | |
360428f8 | 1311 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1312 | struct io_kiocb, timeout.list); |
de0617e4 | 1313 | |
8eb7e2d0 | 1314 | if (io_is_timeout_noseq(req)) |
360428f8 | 1315 | break; |
f010505b MDG |
1316 | |
1317 | /* | |
1318 | * Since seq can easily wrap around over time, subtract | |
1319 | * the last seq at which timeouts were flushed before comparing. | |
1320 | * Assuming not more than 2^31-1 events have happened since, | |
1321 | * these subtractions won't have wrapped, so we can check if | |
1322 | * target is in [last_seq, current_seq] by comparing the two. | |
1323 | */ | |
1324 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1325 | events_got = seq - ctx->cq_last_tm_flush; | |
1326 | if (events_got < events_needed) | |
360428f8 | 1327 | break; |
bfe68a22 | 1328 | |
135fcde8 | 1329 | list_del_init(&req->timeout.list); |
5262f567 | 1330 | io_kill_timeout(req); |
f010505b MDG |
1331 | } while (!list_empty(&ctx->timeout_list)); |
1332 | ||
1333 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1334 | } |
5262f567 | 1335 | |
360428f8 PB |
1336 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1337 | { | |
1338 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1339 | |
1340 | /* order cqe stores with ring update */ | |
1341 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1342 | |
04518945 PB |
1343 | if (unlikely(!list_empty(&ctx->defer_list))) |
1344 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1345 | } |
1346 | ||
90554200 JA |
1347 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1348 | { | |
1349 | struct io_rings *r = ctx->rings; | |
1350 | ||
1351 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1352 | } | |
1353 | ||
888aae2e PB |
1354 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1355 | { | |
1356 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1357 | } | |
1358 | ||
2b188cc1 JA |
1359 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1360 | { | |
75b28aff | 1361 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1362 | unsigned tail; |
1363 | ||
115e12e5 SB |
1364 | /* |
1365 | * writes to the cq entry need to come after reading head; the | |
1366 | * control dependency is enough as we're using WRITE_ONCE to | |
1367 | * fill the cq entry | |
1368 | */ | |
888aae2e | 1369 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1370 | return NULL; |
1371 | ||
888aae2e | 1372 | tail = ctx->cached_cq_tail++; |
75b28aff | 1373 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1374 | } |
1375 | ||
f2842ab5 JA |
1376 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1377 | { | |
f0b493e6 JA |
1378 | if (!ctx->cq_ev_fd) |
1379 | return false; | |
7e55a19c SG |
1380 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1381 | return false; | |
f2842ab5 JA |
1382 | if (!ctx->eventfd_async) |
1383 | return true; | |
b41e9852 | 1384 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1385 | } |
1386 | ||
b41e9852 | 1387 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1388 | { |
b1445e59 PB |
1389 | /* see waitqueue_active() comment */ |
1390 | smp_mb(); | |
1391 | ||
1d7bb1d5 JA |
1392 | if (waitqueue_active(&ctx->wait)) |
1393 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1394 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1395 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1396 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1397 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1398 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1399 | wake_up_interruptible(&ctx->cq_wait); |
1400 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1401 | } | |
1d7bb1d5 JA |
1402 | } |
1403 | ||
80c18e4a PB |
1404 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1405 | { | |
b1445e59 PB |
1406 | /* see waitqueue_active() comment */ |
1407 | smp_mb(); | |
1408 | ||
80c18e4a PB |
1409 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1410 | if (waitqueue_active(&ctx->wait)) | |
1411 | wake_up(&ctx->wait); | |
1412 | } | |
1413 | if (io_should_trigger_evfd(ctx)) | |
1414 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1415 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1416 | wake_up_interruptible(&ctx->cq_wait); |
1417 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1418 | } | |
80c18e4a PB |
1419 | } |
1420 | ||
c4a2ed72 | 1421 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1422 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1423 | struct task_struct *tsk, | |
1424 | struct files_struct *files) | |
1d7bb1d5 JA |
1425 | { |
1426 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1427 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1428 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1429 | unsigned long flags; |
b18032bb | 1430 | bool all_flushed, posted; |
1d7bb1d5 JA |
1431 | LIST_HEAD(list); |
1432 | ||
e23de15f PB |
1433 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1434 | return false; | |
1d7bb1d5 | 1435 | |
b18032bb | 1436 | posted = false; |
1d7bb1d5 | 1437 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1438 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1439 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1440 | continue; |
1441 | ||
1d7bb1d5 JA |
1442 | cqe = io_get_cqring(ctx); |
1443 | if (!cqe && !force) | |
1444 | break; | |
1445 | ||
40d8ddd4 | 1446 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1447 | if (cqe) { |
1448 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1449 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1450 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1451 | } else { |
2c3bac6d | 1452 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1453 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1454 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1455 | } |
b18032bb | 1456 | posted = true; |
1d7bb1d5 JA |
1457 | } |
1458 | ||
09e88404 PB |
1459 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1460 | if (all_flushed) { | |
1461 | clear_bit(0, &ctx->sq_check_overflow); | |
1462 | clear_bit(0, &ctx->cq_check_overflow); | |
1463 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1464 | } | |
46930143 | 1465 | |
b18032bb JA |
1466 | if (posted) |
1467 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1468 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1469 | if (posted) |
1470 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1471 | |
1472 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1473 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1474 | list_del(&req->compl.list); | |
ec9c02ad | 1475 | io_put_req(req); |
1d7bb1d5 | 1476 | } |
c4a2ed72 | 1477 | |
09e88404 | 1478 | return all_flushed; |
1d7bb1d5 JA |
1479 | } |
1480 | ||
6c503150 PB |
1481 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1482 | struct task_struct *tsk, | |
1483 | struct files_struct *files) | |
1484 | { | |
1485 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1486 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1487 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1488 | mutex_lock(&ctx->uring_lock); | |
1489 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1490 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1491 | mutex_unlock(&ctx->uring_lock); | |
1492 | } | |
1493 | } | |
1494 | ||
bcda7baa | 1495 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1496 | { |
78e19bbe | 1497 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1498 | struct io_uring_cqe *cqe; |
1499 | ||
78e19bbe | 1500 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1501 | |
2b188cc1 JA |
1502 | /* |
1503 | * If we can't get a cq entry, userspace overflowed the | |
1504 | * submission (by quite a lot). Increment the overflow count in | |
1505 | * the ring. | |
1506 | */ | |
1507 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1508 | if (likely(cqe)) { |
78e19bbe | 1509 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1510 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1511 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1512 | } else if (ctx->cq_overflow_flushed || |
1513 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1514 | /* |
1515 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1516 | * then we cannot store the request for later flushing, we need | |
1517 | * to drop it on the floor. | |
1518 | */ | |
2c3bac6d PB |
1519 | ctx->cached_cq_overflow++; |
1520 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1521 | } else { |
ad3eb2c8 JA |
1522 | if (list_empty(&ctx->cq_overflow_list)) { |
1523 | set_bit(0, &ctx->sq_check_overflow); | |
1524 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1525 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1526 | } |
40d8ddd4 | 1527 | io_clean_op(req); |
1d7bb1d5 | 1528 | req->result = res; |
0f7e466b | 1529 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1530 | refcount_inc(&req->refs); |
1531 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1532 | } |
1533 | } | |
1534 | ||
bcda7baa JA |
1535 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1536 | { | |
1537 | __io_cqring_fill_event(req, res, 0); | |
1538 | } | |
1539 | ||
c7dae4ba JA |
1540 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1541 | unsigned int cflags) | |
2b188cc1 | 1542 | { |
78e19bbe | 1543 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1544 | unsigned long flags; |
1545 | ||
1546 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1547 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1548 | io_commit_cqring(ctx); |
c7dae4ba JA |
1549 | /* |
1550 | * If we're the last reference to this request, add to our locked | |
1551 | * free_list cache. | |
1552 | */ | |
1553 | if (refcount_dec_and_test(&req->refs)) { | |
1554 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1555 | ||
1556 | io_dismantle_req(req); | |
1557 | io_put_task(req->task, 1); | |
1558 | list_add(&req->compl.list, &cs->locked_free_list); | |
1559 | cs->locked_free_nr++; | |
1560 | } else | |
1561 | req = NULL; | |
2b188cc1 JA |
1562 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1563 | ||
8c838788 | 1564 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1565 | if (req) { |
1566 | io_queue_next(req); | |
1567 | percpu_ref_put(&ctx->refs); | |
229a7b63 | 1568 | } |
229a7b63 JA |
1569 | } |
1570 | ||
a38d68db | 1571 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1572 | unsigned int cflags) |
229a7b63 | 1573 | { |
a38d68db PB |
1574 | io_clean_op(req); |
1575 | req->result = res; | |
1576 | req->compl.cflags = cflags; | |
e342c807 | 1577 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1578 | } |
1579 | ||
889fca73 PB |
1580 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1581 | long res, unsigned cflags) | |
bcda7baa | 1582 | { |
889fca73 PB |
1583 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1584 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1585 | else |
c7dae4ba | 1586 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1587 | } |
1588 | ||
a38d68db | 1589 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1590 | { |
889fca73 | 1591 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1592 | } |
1593 | ||
c7dae4ba | 1594 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1595 | { |
c7dae4ba JA |
1596 | struct io_submit_state *state = &ctx->submit_state; |
1597 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1598 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1599 | |
c7dae4ba JA |
1600 | /* |
1601 | * If we have more than a batch's worth of requests in our IRQ side | |
1602 | * locked cache, grab the lock and move them over to our submission | |
1603 | * side cache. | |
1604 | */ | |
1605 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1606 | spin_lock_irq(&ctx->completion_lock); | |
1607 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1608 | cs->locked_free_nr = 0; | |
1609 | spin_unlock_irq(&ctx->completion_lock); | |
1610 | } | |
0ddf92e8 | 1611 | |
c7dae4ba JA |
1612 | while (!list_empty(&cs->free_list)) { |
1613 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1614 | compl.list); |
1615 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1616 | state->reqs[state->free_reqs++] = req; |
1617 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1618 | break; | |
1b4c351f JA |
1619 | } |
1620 | ||
e5d1bc0a | 1621 | return req != NULL; |
0ddf92e8 JA |
1622 | } |
1623 | ||
e5d1bc0a | 1624 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1625 | { |
e5d1bc0a PB |
1626 | struct io_submit_state *state = &ctx->submit_state; |
1627 | ||
1628 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1629 | ||
f6b6c7d6 | 1630 | if (!state->free_reqs) { |
291b2821 | 1631 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1632 | int ret; |
1633 | ||
c7dae4ba | 1634 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1635 | goto got_req; |
1636 | ||
bf019da7 PB |
1637 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1638 | state->reqs); | |
fd6fab2c JA |
1639 | |
1640 | /* | |
1641 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1642 | * retry single alloc to be on the safe side. | |
1643 | */ | |
1644 | if (unlikely(ret <= 0)) { | |
1645 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1646 | if (!state->reqs[0]) | |
3893f39f | 1647 | return NULL; |
fd6fab2c JA |
1648 | ret = 1; |
1649 | } | |
291b2821 | 1650 | state->free_reqs = ret; |
2b188cc1 | 1651 | } |
e5d1bc0a | 1652 | got_req: |
291b2821 PB |
1653 | state->free_reqs--; |
1654 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1655 | } |
1656 | ||
8da11c19 PB |
1657 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1658 | bool fixed) | |
1659 | { | |
36f72fe2 | 1660 | if (!fixed) |
8da11c19 PB |
1661 | fput(file); |
1662 | } | |
1663 | ||
4edf20f9 | 1664 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1665 | { |
3ca405eb | 1666 | io_clean_op(req); |
929a3af9 | 1667 | |
e8c2bc1f JA |
1668 | if (req->async_data) |
1669 | kfree(req->async_data); | |
8da11c19 PB |
1670 | if (req->file) |
1671 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1672 | if (req->fixed_rsrc_refs) |
1673 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 1674 | io_req_clean_work(req); |
e65ef56d JA |
1675 | } |
1676 | ||
7c660731 PB |
1677 | static inline void io_put_task(struct task_struct *task, int nr) |
1678 | { | |
1679 | struct io_uring_task *tctx = task->io_uring; | |
1680 | ||
1681 | percpu_counter_sub(&tctx->inflight, nr); | |
1682 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1683 | wake_up(&tctx->wait); | |
1684 | put_task_struct_many(task, nr); | |
1685 | } | |
1686 | ||
216578e5 | 1687 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1688 | { |
51a4cc11 | 1689 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1690 | |
216578e5 | 1691 | io_dismantle_req(req); |
7c660731 | 1692 | io_put_task(req->task, 1); |
c6ca97b3 | 1693 | |
3893f39f | 1694 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1695 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1696 | } |
1697 | ||
f2f87370 PB |
1698 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1699 | { | |
1700 | struct io_kiocb *nxt = req->link; | |
1701 | ||
1702 | req->link = nxt->link; | |
1703 | nxt->link = NULL; | |
1704 | } | |
1705 | ||
c9abd7ad | 1706 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1707 | { |
a197f664 | 1708 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1709 | struct io_kiocb *link; |
c9abd7ad PB |
1710 | bool cancelled = false; |
1711 | unsigned long flags; | |
7c86ffee | 1712 | |
c9abd7ad | 1713 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1714 | link = req->link; |
1715 | ||
900fad45 PB |
1716 | /* |
1717 | * Can happen if a linked timeout fired and link had been like | |
1718 | * req -> link t-out -> link t-out [-> ...] | |
1719 | */ | |
c9abd7ad PB |
1720 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1721 | struct io_timeout_data *io = link->async_data; | |
1722 | int ret; | |
7c86ffee | 1723 | |
f2f87370 | 1724 | io_remove_next_linked(req); |
90cd7e42 | 1725 | link->timeout.head = NULL; |
c9abd7ad PB |
1726 | ret = hrtimer_try_to_cancel(&io->timer); |
1727 | if (ret != -1) { | |
1728 | io_cqring_fill_event(link, -ECANCELED); | |
1729 | io_commit_cqring(ctx); | |
1730 | cancelled = true; | |
1731 | } | |
1732 | } | |
7c86ffee | 1733 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1734 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1735 | |
c9abd7ad | 1736 | if (cancelled) { |
7c86ffee | 1737 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1738 | io_put_req(link); |
1739 | } | |
7c86ffee PB |
1740 | } |
1741 | ||
9e645e11 | 1742 | |
d148ca4b | 1743 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1744 | { |
f2f87370 | 1745 | struct io_kiocb *link, *nxt; |
2665abfd | 1746 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1747 | unsigned long flags; |
9e645e11 | 1748 | |
d148ca4b | 1749 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1750 | link = req->link; |
1751 | req->link = NULL; | |
9e645e11 | 1752 | |
f2f87370 PB |
1753 | while (link) { |
1754 | nxt = link->link; | |
1755 | link->link = NULL; | |
2665abfd | 1756 | |
f2f87370 | 1757 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1758 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 | 1759 | |
1575f21a | 1760 | io_put_req_deferred(link, 2); |
f2f87370 | 1761 | link = nxt; |
9e645e11 | 1762 | } |
2665abfd | 1763 | io_commit_cqring(ctx); |
216578e5 | 1764 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1765 | |
2665abfd | 1766 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1767 | } |
1768 | ||
3fa5e0f3 | 1769 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1770 | { |
7c86ffee PB |
1771 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1772 | io_kill_linked_timeout(req); | |
944e58bf | 1773 | |
9e645e11 JA |
1774 | /* |
1775 | * If LINK is set, we have dependent requests in this chain. If we | |
1776 | * didn't fail this request, queue the first one up, moving any other | |
1777 | * dependencies to the next request. In case of failure, fail the rest | |
1778 | * of the chain. | |
1779 | */ | |
f2f87370 PB |
1780 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
1781 | struct io_kiocb *nxt = req->link; | |
1782 | ||
1783 | req->link = NULL; | |
1784 | return nxt; | |
1785 | } | |
9b5f7bd9 PB |
1786 | io_fail_links(req); |
1787 | return NULL; | |
4d7dd462 | 1788 | } |
9e645e11 | 1789 | |
f2f87370 | 1790 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1791 | { |
cdbff982 | 1792 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1793 | return NULL; |
1794 | return __io_req_find_next(req); | |
1795 | } | |
1796 | ||
2c32395d PB |
1797 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1798 | { | |
1799 | if (!ctx) | |
1800 | return; | |
1801 | if (ctx->submit_state.comp.nr) { | |
1802 | mutex_lock(&ctx->uring_lock); | |
1803 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1804 | mutex_unlock(&ctx->uring_lock); | |
1805 | } | |
1806 | percpu_ref_put(&ctx->refs); | |
1807 | } | |
1808 | ||
7cbf1722 | 1809 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1810 | { |
65453d1e | 1811 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1812 | struct io_wq_work_list list; |
1813 | struct io_wq_work_node *node; | |
c2c4c83c | 1814 | |
7cbf1722 JA |
1815 | if (wq_list_empty(&tctx->task_list)) |
1816 | return false; | |
6200b0ae | 1817 | |
0b81e80c | 1818 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1819 | list = tctx->task_list; |
1820 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1821 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1822 | |
7cbf1722 JA |
1823 | node = list.first; |
1824 | while (node) { | |
1825 | struct io_wq_work_node *next = node->next; | |
1826 | struct io_kiocb *req; | |
0ba9c9ed | 1827 | |
7cbf1722 | 1828 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1829 | if (req->ctx != ctx) { |
1830 | ctx_flush_and_put(ctx); | |
1831 | ctx = req->ctx; | |
1832 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1833 | } |
65453d1e | 1834 | |
2c32395d PB |
1835 | req->task_work.func(&req->task_work); |
1836 | node = next; | |
7cbf1722 JA |
1837 | } |
1838 | ||
2c32395d | 1839 | ctx_flush_and_put(ctx); |
7cbf1722 | 1840 | return list.first != NULL; |
c2c4c83c JA |
1841 | } |
1842 | ||
7cbf1722 | 1843 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1844 | { |
7cbf1722 | 1845 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1846 | |
1d5f360d JA |
1847 | clear_bit(0, &tctx->task_state); |
1848 | ||
7cbf1722 JA |
1849 | while (__tctx_task_work(tctx)) |
1850 | cond_resched(); | |
7cbf1722 JA |
1851 | } |
1852 | ||
1853 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1854 | enum task_work_notify_mode notify) | |
1855 | { | |
1856 | struct io_uring_task *tctx = tsk->io_uring; | |
1857 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1858 | unsigned long flags; |
7cbf1722 JA |
1859 | int ret; |
1860 | ||
1861 | WARN_ON_ONCE(!tctx); | |
1862 | ||
0b81e80c | 1863 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1864 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1865 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1866 | |
1867 | /* task_work already pending, we're done */ | |
1868 | if (test_bit(0, &tctx->task_state) || | |
1869 | test_and_set_bit(0, &tctx->task_state)) | |
1870 | return 0; | |
1871 | ||
1872 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1873 | return 0; | |
1874 | ||
1875 | /* | |
1876 | * Slow path - we failed, find and delete work. if the work is not | |
1877 | * in the list, it got run and we're fine. | |
1878 | */ | |
1879 | ret = 0; | |
0b81e80c | 1880 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1881 | wq_list_for_each(node, prev, &tctx->task_list) { |
1882 | if (&req->io_task_work.node == node) { | |
1883 | wq_list_del(&tctx->task_list, node, prev); | |
1884 | ret = 1; | |
1885 | break; | |
1886 | } | |
1887 | } | |
0b81e80c | 1888 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1889 | clear_bit(0, &tctx->task_state); |
1890 | return ret; | |
1891 | } | |
1892 | ||
355fb9e2 | 1893 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1894 | { |
1895 | struct task_struct *tsk = req->task; | |
1896 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1897 | enum task_work_notify_mode notify; |
1898 | int ret; | |
c2c4c83c | 1899 | |
6200b0ae JA |
1900 | if (tsk->flags & PF_EXITING) |
1901 | return -ESRCH; | |
1902 | ||
c2c4c83c | 1903 | /* |
0ba9c9ed JA |
1904 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1905 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1906 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1907 | * will do the job. | |
c2c4c83c | 1908 | */ |
91989c70 | 1909 | notify = TWA_NONE; |
355fb9e2 | 1910 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1911 | notify = TWA_SIGNAL; |
1912 | ||
7cbf1722 | 1913 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1914 | if (!ret) |
1915 | wake_up_process(tsk); | |
0ba9c9ed | 1916 | |
c2c4c83c JA |
1917 | return ret; |
1918 | } | |
1919 | ||
eab30c4d | 1920 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1921 | task_work_func_t cb) |
eab30c4d | 1922 | { |
7c25c0d1 JA |
1923 | struct io_ring_ctx *ctx = req->ctx; |
1924 | struct callback_head *head; | |
eab30c4d PB |
1925 | |
1926 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1927 | do { |
1928 | head = READ_ONCE(ctx->exit_task_work); | |
1929 | req->task_work.next = head; | |
1930 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1931 | } |
1932 | ||
c40f6379 JA |
1933 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1934 | { | |
1935 | struct io_ring_ctx *ctx = req->ctx; | |
1936 | ||
1937 | spin_lock_irq(&ctx->completion_lock); | |
1938 | io_cqring_fill_event(req, error); | |
1939 | io_commit_cqring(ctx); | |
1940 | spin_unlock_irq(&ctx->completion_lock); | |
1941 | ||
1942 | io_cqring_ev_posted(ctx); | |
1943 | req_set_fail_links(req); | |
1944 | io_double_put_req(req); | |
1945 | } | |
1946 | ||
1947 | static void io_req_task_cancel(struct callback_head *cb) | |
1948 | { | |
1949 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1950 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1951 | |
792bb6eb | 1952 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1953 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1954 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1955 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1956 | } |
1957 | ||
1958 | static void __io_req_task_submit(struct io_kiocb *req) | |
1959 | { | |
1960 | struct io_ring_ctx *ctx = req->ctx; | |
1961 | ||
04fc6c80 | 1962 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1963 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 1964 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1965 | __io_queue_sqe(req); |
81b6d05c | 1966 | else |
c40f6379 | 1967 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1968 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1969 | } |
1970 | ||
1971 | static void io_req_task_submit(struct callback_head *cb) | |
1972 | { | |
1973 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1974 | ||
1975 | __io_req_task_submit(req); | |
1976 | } | |
1977 | ||
1978 | static void io_req_task_queue(struct io_kiocb *req) | |
1979 | { | |
c40f6379 JA |
1980 | int ret; |
1981 | ||
7cbf1722 | 1982 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1983 | ret = io_req_task_work_add(req); |
c40f6379 | 1984 | if (unlikely(ret)) { |
a3df7698 | 1985 | req->result = -ECANCELED; |
04fc6c80 | 1986 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1987 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1988 | } |
c40f6379 JA |
1989 | } |
1990 | ||
a3df7698 PB |
1991 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
1992 | { | |
1993 | percpu_ref_get(&req->ctx->refs); | |
1994 | req->result = ret; | |
1995 | req->task_work.func = io_req_task_cancel; | |
1996 | ||
1997 | if (unlikely(io_req_task_work_add(req))) | |
1998 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
1999 | } | |
2000 | ||
f2f87370 | 2001 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2002 | { |
9b5f7bd9 | 2003 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2004 | |
2005 | if (nxt) | |
906a8c3f | 2006 | io_req_task_queue(nxt); |
c69f8dbe JL |
2007 | } |
2008 | ||
c3524383 | 2009 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2010 | { |
c3524383 PB |
2011 | io_queue_next(req); |
2012 | __io_free_req(req); | |
2013 | } | |
8766dd51 | 2014 | |
2d6500d4 | 2015 | struct req_batch { |
5af1d13e PB |
2016 | struct task_struct *task; |
2017 | int task_refs; | |
1b4c351f | 2018 | int ctx_refs; |
2d6500d4 PB |
2019 | }; |
2020 | ||
5af1d13e PB |
2021 | static inline void io_init_req_batch(struct req_batch *rb) |
2022 | { | |
5af1d13e | 2023 | rb->task_refs = 0; |
9ae72463 | 2024 | rb->ctx_refs = 0; |
5af1d13e PB |
2025 | rb->task = NULL; |
2026 | } | |
2027 | ||
2d6500d4 PB |
2028 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2029 | struct req_batch *rb) | |
2030 | { | |
6e833d53 | 2031 | if (rb->task) |
7c660731 | 2032 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2033 | if (rb->ctx_refs) |
2034 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2035 | } |
2036 | ||
6ff119a6 PB |
2037 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2038 | struct io_submit_state *state) | |
2d6500d4 | 2039 | { |
f2f87370 | 2040 | io_queue_next(req); |
2d6500d4 | 2041 | |
e3bc8e9d | 2042 | if (req->task != rb->task) { |
7c660731 PB |
2043 | if (rb->task) |
2044 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2045 | rb->task = req->task; |
2046 | rb->task_refs = 0; | |
5af1d13e | 2047 | } |
e3bc8e9d | 2048 | rb->task_refs++; |
9ae72463 | 2049 | rb->ctx_refs++; |
5af1d13e | 2050 | |
4edf20f9 | 2051 | io_dismantle_req(req); |
bd759045 | 2052 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2053 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2054 | else |
2055 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2056 | } |
2057 | ||
905c172f PB |
2058 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2059 | struct io_ring_ctx *ctx) | |
2060 | { | |
2061 | int i, nr = cs->nr; | |
2062 | struct io_kiocb *req; | |
2063 | struct req_batch rb; | |
2064 | ||
2065 | io_init_req_batch(&rb); | |
2066 | spin_lock_irq(&ctx->completion_lock); | |
2067 | for (i = 0; i < nr; i++) { | |
2068 | req = cs->reqs[i]; | |
2069 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2070 | } | |
2071 | io_commit_cqring(ctx); | |
2072 | spin_unlock_irq(&ctx->completion_lock); | |
2073 | ||
2074 | io_cqring_ev_posted(ctx); | |
2075 | for (i = 0; i < nr; i++) { | |
2076 | req = cs->reqs[i]; | |
2077 | ||
2078 | /* submission and completion refs */ | |
2079 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2080 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2081 | } |
2082 | ||
2083 | io_req_free_batch_finish(ctx, &rb); | |
2084 | cs->nr = 0; | |
7a743e22 PB |
2085 | } |
2086 | ||
ba816ad6 JA |
2087 | /* |
2088 | * Drop reference to request, return next in chain (if there is one) if this | |
2089 | * was the last reference to this request. | |
2090 | */ | |
9b5f7bd9 | 2091 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2092 | { |
9b5f7bd9 PB |
2093 | struct io_kiocb *nxt = NULL; |
2094 | ||
2a44f467 | 2095 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2096 | nxt = io_req_find_next(req); |
4d7dd462 | 2097 | __io_free_req(req); |
2a44f467 | 2098 | } |
9b5f7bd9 | 2099 | return nxt; |
2b188cc1 JA |
2100 | } |
2101 | ||
e65ef56d JA |
2102 | static void io_put_req(struct io_kiocb *req) |
2103 | { | |
2104 | if (refcount_dec_and_test(&req->refs)) | |
2105 | io_free_req(req); | |
2b188cc1 JA |
2106 | } |
2107 | ||
216578e5 PB |
2108 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2109 | { | |
2110 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2111 | ||
2112 | io_free_req(req); | |
2113 | } | |
2114 | ||
2115 | static void io_free_req_deferred(struct io_kiocb *req) | |
2116 | { | |
2117 | int ret; | |
2118 | ||
7cbf1722 | 2119 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2120 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2121 | if (unlikely(ret)) |
2122 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2123 | } |
2124 | ||
2125 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2126 | { | |
2127 | if (refcount_sub_and_test(refs, &req->refs)) | |
2128 | io_free_req_deferred(req); | |
2129 | } | |
2130 | ||
978db57e JA |
2131 | static void io_double_put_req(struct io_kiocb *req) |
2132 | { | |
2133 | /* drop both submit and complete references */ | |
2134 | if (refcount_sub_and_test(2, &req->refs)) | |
2135 | io_free_req(req); | |
2136 | } | |
2137 | ||
6c503150 | 2138 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2139 | { |
2140 | /* See comment at the top of this file */ | |
2141 | smp_rmb(); | |
e23de15f | 2142 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2143 | } |
2144 | ||
fb5ccc98 PB |
2145 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2146 | { | |
2147 | struct io_rings *rings = ctx->rings; | |
2148 | ||
2149 | /* make sure SQ entry isn't read before tail */ | |
2150 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2151 | } | |
2152 | ||
8ff069bf | 2153 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2154 | { |
8ff069bf | 2155 | unsigned int cflags; |
e94f141b | 2156 | |
bcda7baa JA |
2157 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2158 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2159 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2160 | kfree(kbuf); |
2161 | return cflags; | |
e94f141b JA |
2162 | } |
2163 | ||
8ff069bf | 2164 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2165 | { |
4d954c25 | 2166 | struct io_buffer *kbuf; |
bcda7baa | 2167 | |
4d954c25 | 2168 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2169 | return io_put_kbuf(req, kbuf); |
2170 | } | |
2171 | ||
4c6e277c JA |
2172 | static inline bool io_run_task_work(void) |
2173 | { | |
6200b0ae JA |
2174 | /* |
2175 | * Not safe to run on exiting task, and the task_work handling will | |
2176 | * not add work to such a task. | |
2177 | */ | |
2178 | if (unlikely(current->flags & PF_EXITING)) | |
2179 | return false; | |
4c6e277c JA |
2180 | if (current->task_works) { |
2181 | __set_current_state(TASK_RUNNING); | |
2182 | task_work_run(); | |
2183 | return true; | |
2184 | } | |
2185 | ||
2186 | return false; | |
bcda7baa JA |
2187 | } |
2188 | ||
def596e9 JA |
2189 | /* |
2190 | * Find and free completed poll iocbs | |
2191 | */ | |
2192 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2193 | struct list_head *done) | |
2194 | { | |
8237e045 | 2195 | struct req_batch rb; |
def596e9 | 2196 | struct io_kiocb *req; |
bbde017a XW |
2197 | |
2198 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2199 | smp_rmb(); | |
def596e9 | 2200 | |
5af1d13e | 2201 | io_init_req_batch(&rb); |
def596e9 | 2202 | while (!list_empty(done)) { |
bcda7baa JA |
2203 | int cflags = 0; |
2204 | ||
d21ffe7e | 2205 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2206 | list_del(&req->inflight_entry); |
2207 | ||
bbde017a XW |
2208 | if (READ_ONCE(req->result) == -EAGAIN) { |
2209 | req->iopoll_completed = 0; | |
23faba36 | 2210 | if (io_rw_reissue(req)) |
f161340d | 2211 | continue; |
bbde017a | 2212 | } |
def596e9 | 2213 | |
bcda7baa | 2214 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2215 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2216 | |
2217 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2218 | (*nr_events)++; |
2219 | ||
c3524383 | 2220 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2221 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2222 | } |
def596e9 | 2223 | |
09bb8394 | 2224 | io_commit_cqring(ctx); |
80c18e4a | 2225 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2226 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2227 | } |
2228 | ||
def596e9 JA |
2229 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2230 | long min) | |
2231 | { | |
2232 | struct io_kiocb *req, *tmp; | |
2233 | LIST_HEAD(done); | |
2234 | bool spin; | |
2235 | int ret; | |
2236 | ||
2237 | /* | |
2238 | * Only spin for completions if we don't have multiple devices hanging | |
2239 | * off our complete list, and we're under the requested amount. | |
2240 | */ | |
2241 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2242 | ||
2243 | ret = 0; | |
d21ffe7e | 2244 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2245 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2246 | |
2247 | /* | |
581f9810 BM |
2248 | * Move completed and retryable entries to our local lists. |
2249 | * If we find a request that requires polling, break out | |
2250 | * and complete those lists first, if we have entries there. | |
def596e9 | 2251 | */ |
65a6543d | 2252 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2253 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2254 | continue; |
2255 | } | |
2256 | if (!list_empty(&done)) | |
2257 | break; | |
2258 | ||
2259 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2260 | if (ret < 0) | |
2261 | break; | |
2262 | ||
3aadc23e PB |
2263 | /* iopoll may have completed current req */ |
2264 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2265 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2266 | |
def596e9 JA |
2267 | if (ret && spin) |
2268 | spin = false; | |
2269 | ret = 0; | |
2270 | } | |
2271 | ||
2272 | if (!list_empty(&done)) | |
2273 | io_iopoll_complete(ctx, nr_events, &done); | |
2274 | ||
2275 | return ret; | |
2276 | } | |
2277 | ||
2278 | /* | |
d195a66e | 2279 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2280 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2281 | * as a non-spinning completion check. | |
2282 | */ | |
2283 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2284 | long min) | |
2285 | { | |
540e32a0 | 2286 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2287 | int ret; |
2288 | ||
2289 | ret = io_do_iopoll(ctx, nr_events, min); | |
2290 | if (ret < 0) | |
2291 | return ret; | |
eba0a4dd | 2292 | if (*nr_events >= min) |
def596e9 JA |
2293 | return 0; |
2294 | } | |
2295 | ||
2296 | return 1; | |
2297 | } | |
2298 | ||
2299 | /* | |
2300 | * We can't just wait for polled events to come to us, we have to actively | |
2301 | * find and complete them. | |
2302 | */ | |
b2edc0a7 | 2303 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2304 | { |
2305 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2306 | return; | |
2307 | ||
2308 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2309 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2310 | unsigned int nr_events = 0; |
2311 | ||
b2edc0a7 | 2312 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2313 | |
b2edc0a7 PB |
2314 | /* let it sleep and repeat later if can't complete a request */ |
2315 | if (nr_events == 0) | |
2316 | break; | |
08f5439f JA |
2317 | /* |
2318 | * Ensure we allow local-to-the-cpu processing to take place, | |
2319 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2320 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2321 | */ |
3fcee5a6 PB |
2322 | if (need_resched()) { |
2323 | mutex_unlock(&ctx->uring_lock); | |
2324 | cond_resched(); | |
2325 | mutex_lock(&ctx->uring_lock); | |
2326 | } | |
def596e9 JA |
2327 | } |
2328 | mutex_unlock(&ctx->uring_lock); | |
2329 | } | |
2330 | ||
7668b92a | 2331 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2332 | { |
7668b92a | 2333 | unsigned int nr_events = 0; |
2b2ed975 | 2334 | int iters = 0, ret = 0; |
500f9fba | 2335 | |
c7849be9 XW |
2336 | /* |
2337 | * We disallow the app entering submit/complete with polling, but we | |
2338 | * still need to lock the ring to prevent racing with polled issue | |
2339 | * that got punted to a workqueue. | |
2340 | */ | |
2341 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2342 | do { |
a3a0e43f JA |
2343 | /* |
2344 | * Don't enter poll loop if we already have events pending. | |
2345 | * If we do, we can potentially be spinning for commands that | |
2346 | * already triggered a CQE (eg in error). | |
2347 | */ | |
6c503150 PB |
2348 | if (test_bit(0, &ctx->cq_check_overflow)) |
2349 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2350 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2351 | break; |
2352 | ||
500f9fba JA |
2353 | /* |
2354 | * If a submit got punted to a workqueue, we can have the | |
2355 | * application entering polling for a command before it gets | |
2356 | * issued. That app will hold the uring_lock for the duration | |
2357 | * of the poll right here, so we need to take a breather every | |
2358 | * now and then to ensure that the issue has a chance to add | |
2359 | * the poll to the issued list. Otherwise we can spin here | |
2360 | * forever, while the workqueue is stuck trying to acquire the | |
2361 | * very same mutex. | |
2362 | */ | |
2363 | if (!(++iters & 7)) { | |
2364 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2365 | io_run_task_work(); |
500f9fba JA |
2366 | mutex_lock(&ctx->uring_lock); |
2367 | } | |
2368 | ||
7668b92a | 2369 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2370 | if (ret <= 0) |
2371 | break; | |
2372 | ret = 0; | |
7668b92a | 2373 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2374 | |
500f9fba | 2375 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2376 | return ret; |
2377 | } | |
2378 | ||
491381ce | 2379 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2380 | { |
491381ce JA |
2381 | /* |
2382 | * Tell lockdep we inherited freeze protection from submission | |
2383 | * thread. | |
2384 | */ | |
2385 | if (req->flags & REQ_F_ISREG) { | |
2386 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2387 | |
491381ce | 2388 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2389 | } |
491381ce | 2390 | file_end_write(req->file); |
2b188cc1 JA |
2391 | } |
2392 | ||
b63534c4 | 2393 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2394 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2395 | { |
2396 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2397 | int rw, ret; |
b63534c4 | 2398 | struct iov_iter iter; |
b63534c4 | 2399 | |
dc2a6e9a PB |
2400 | /* already prepared */ |
2401 | if (req->async_data) | |
2402 | return true; | |
b63534c4 JA |
2403 | |
2404 | switch (req->opcode) { | |
2405 | case IORING_OP_READV: | |
2406 | case IORING_OP_READ_FIXED: | |
2407 | case IORING_OP_READ: | |
2408 | rw = READ; | |
2409 | break; | |
2410 | case IORING_OP_WRITEV: | |
2411 | case IORING_OP_WRITE_FIXED: | |
2412 | case IORING_OP_WRITE: | |
2413 | rw = WRITE; | |
2414 | break; | |
2415 | default: | |
2416 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2417 | req->opcode); | |
dc2a6e9a | 2418 | return false; |
b63534c4 JA |
2419 | } |
2420 | ||
dc2a6e9a PB |
2421 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2422 | if (ret < 0) | |
2423 | return false; | |
6bf985dc | 2424 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2425 | } |
b63534c4 | 2426 | |
3e6a0d3c | 2427 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2428 | { |
355afaeb | 2429 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2430 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2431 | |
355afaeb JA |
2432 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2433 | return false; | |
3e6a0d3c JA |
2434 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2435 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2436 | return false; |
7c977a58 JA |
2437 | /* |
2438 | * If ref is dying, we might be running poll reap from the exit work. | |
2439 | * Don't attempt to reissue from that path, just let it fail with | |
2440 | * -EAGAIN. | |
2441 | */ | |
3e6a0d3c JA |
2442 | if (percpu_ref_is_dying(&ctx->refs)) |
2443 | return false; | |
2444 | return true; | |
2445 | } | |
2446 | #endif | |
2447 | ||
2448 | static bool io_rw_reissue(struct io_kiocb *req) | |
2449 | { | |
2450 | #ifdef CONFIG_BLOCK | |
2451 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2452 | return false; |
b63534c4 | 2453 | |
55e6ac1e PB |
2454 | lockdep_assert_held(&req->ctx->uring_lock); |
2455 | ||
37d1e2e3 | 2456 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2457 | refcount_inc(&req->refs); |
2458 | io_queue_async_work(req); | |
b63534c4 | 2459 | return true; |
fdee946d | 2460 | } |
dc2a6e9a | 2461 | req_set_fail_links(req); |
b63534c4 JA |
2462 | #endif |
2463 | return false; | |
2464 | } | |
2465 | ||
a1d7c393 | 2466 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2467 | unsigned int issue_flags) |
a1d7c393 | 2468 | { |
2f8e45f1 PB |
2469 | int cflags = 0; |
2470 | ||
23faba36 PB |
2471 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2472 | return; | |
2f8e45f1 PB |
2473 | if (res != req->result) |
2474 | req_set_fail_links(req); | |
23faba36 | 2475 | |
2f8e45f1 PB |
2476 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2477 | kiocb_end_write(req); | |
2478 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2479 | cflags = io_put_rw_kbuf(req); | |
2480 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2481 | } |
2482 | ||
2483 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2484 | { | |
9adbd45d | 2485 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2486 | |
889fca73 | 2487 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2488 | } |
2489 | ||
def596e9 JA |
2490 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2491 | { | |
9adbd45d | 2492 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2493 | |
3e6a0d3c JA |
2494 | #ifdef CONFIG_BLOCK |
2495 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2496 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2497 | struct io_async_rw *rw = req->async_data; | |
2498 | ||
2499 | if (rw) | |
2500 | iov_iter_revert(&rw->iter, | |
2501 | req->result - iov_iter_count(&rw->iter)); | |
2502 | else if (!io_resubmit_prep(req)) | |
2503 | res = -EIO; | |
2504 | } | |
2505 | #endif | |
2506 | ||
491381ce JA |
2507 | if (kiocb->ki_flags & IOCB_WRITE) |
2508 | kiocb_end_write(req); | |
def596e9 | 2509 | |
2d7d6792 | 2510 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2511 | req_set_fail_links(req); |
bbde017a XW |
2512 | |
2513 | WRITE_ONCE(req->result, res); | |
2514 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2515 | smp_wmb(); |
2516 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2517 | } |
2518 | ||
2519 | /* | |
2520 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2521 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2522 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2523 | * accessing the kiocb cookie. | |
2524 | */ | |
2e9dbe90 | 2525 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2526 | { |
2527 | struct io_ring_ctx *ctx = req->ctx; | |
2528 | ||
2529 | /* | |
2530 | * Track whether we have multiple files in our lists. This will impact | |
2531 | * how we do polling eventually, not spinning if we're on potentially | |
2532 | * different devices. | |
2533 | */ | |
540e32a0 | 2534 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2535 | ctx->poll_multi_file = false; |
2536 | } else if (!ctx->poll_multi_file) { | |
2537 | struct io_kiocb *list_req; | |
2538 | ||
540e32a0 | 2539 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2540 | inflight_entry); |
9adbd45d | 2541 | if (list_req->file != req->file) |
def596e9 JA |
2542 | ctx->poll_multi_file = true; |
2543 | } | |
2544 | ||
2545 | /* | |
2546 | * For fast devices, IO may have already completed. If it has, add | |
2547 | * it to the front so we find it first. | |
2548 | */ | |
65a6543d | 2549 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2550 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2551 | else |
d21ffe7e | 2552 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2553 | |
2e9dbe90 XW |
2554 | /* |
2555 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2556 | * task context or in io worker task context. If current task context is | |
2557 | * sq thread, we don't need to check whether should wake up sq thread. | |
2558 | */ | |
2559 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2560 | wq_has_sleeper(&ctx->sq_data->wait)) |
2561 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2562 | } |
2563 | ||
9f13c35b PB |
2564 | static inline void io_state_file_put(struct io_submit_state *state) |
2565 | { | |
02b23a9a PB |
2566 | if (state->file_refs) { |
2567 | fput_many(state->file, state->file_refs); | |
2568 | state->file_refs = 0; | |
2569 | } | |
9a56a232 JA |
2570 | } |
2571 | ||
2572 | /* | |
2573 | * Get as many references to a file as we have IOs left in this submission, | |
2574 | * assuming most submissions are for one file, or at least that each file | |
2575 | * has more than one submission. | |
2576 | */ | |
8da11c19 | 2577 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2578 | { |
2579 | if (!state) | |
2580 | return fget(fd); | |
2581 | ||
6e1271e6 | 2582 | if (state->file_refs) { |
9a56a232 | 2583 | if (state->fd == fd) { |
6e1271e6 | 2584 | state->file_refs--; |
9a56a232 JA |
2585 | return state->file; |
2586 | } | |
02b23a9a | 2587 | io_state_file_put(state); |
9a56a232 JA |
2588 | } |
2589 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2590 | if (unlikely(!state->file)) |
9a56a232 JA |
2591 | return NULL; |
2592 | ||
2593 | state->fd = fd; | |
6e1271e6 | 2594 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2595 | return state->file; |
2596 | } | |
2597 | ||
4503b767 JA |
2598 | static bool io_bdev_nowait(struct block_device *bdev) |
2599 | { | |
9ba0d0c8 | 2600 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2601 | } |
2602 | ||
2b188cc1 JA |
2603 | /* |
2604 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2605 | * any file. For now, just ensure that anything potentially problematic is done | |
2606 | * inline. | |
2607 | */ | |
af197f50 | 2608 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2609 | { |
2610 | umode_t mode = file_inode(file)->i_mode; | |
2611 | ||
4503b767 | 2612 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2613 | if (IS_ENABLED(CONFIG_BLOCK) && |
2614 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2615 | return true; |
2616 | return false; | |
2617 | } | |
2618 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2619 | return true; |
4503b767 | 2620 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2621 | if (IS_ENABLED(CONFIG_BLOCK) && |
2622 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2623 | file->f_op != &io_uring_fops) |
2624 | return true; | |
2625 | return false; | |
2626 | } | |
2b188cc1 | 2627 | |
c5b85625 JA |
2628 | /* any ->read/write should understand O_NONBLOCK */ |
2629 | if (file->f_flags & O_NONBLOCK) | |
2630 | return true; | |
2631 | ||
af197f50 JA |
2632 | if (!(file->f_mode & FMODE_NOWAIT)) |
2633 | return false; | |
2634 | ||
2635 | if (rw == READ) | |
2636 | return file->f_op->read_iter != NULL; | |
2637 | ||
2638 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2639 | } |
2640 | ||
a88fc400 | 2641 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2642 | { |
def596e9 | 2643 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2644 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2645 | struct file *file = req->file; |
09bb8394 JA |
2646 | unsigned ioprio; |
2647 | int ret; | |
2b188cc1 | 2648 | |
75c668cd | 2649 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2650 | req->flags |= REQ_F_ISREG; |
2651 | ||
2b188cc1 | 2652 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2653 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2654 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2655 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2656 | } |
2b188cc1 | 2657 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2658 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2659 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2660 | if (unlikely(ret)) | |
2661 | return ret; | |
2b188cc1 | 2662 | |
75c668cd PB |
2663 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2664 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2665 | req->flags |= REQ_F_NOWAIT; | |
2666 | ||
2b188cc1 JA |
2667 | ioprio = READ_ONCE(sqe->ioprio); |
2668 | if (ioprio) { | |
2669 | ret = ioprio_check_cap(ioprio); | |
2670 | if (ret) | |
09bb8394 | 2671 | return ret; |
2b188cc1 JA |
2672 | |
2673 | kiocb->ki_ioprio = ioprio; | |
2674 | } else | |
2675 | kiocb->ki_ioprio = get_current_ioprio(); | |
2676 | ||
def596e9 | 2677 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2678 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2679 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2680 | return -EOPNOTSUPP; |
2b188cc1 | 2681 | |
def596e9 JA |
2682 | kiocb->ki_flags |= IOCB_HIPRI; |
2683 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2684 | req->iopoll_completed = 0; |
def596e9 | 2685 | } else { |
09bb8394 JA |
2686 | if (kiocb->ki_flags & IOCB_HIPRI) |
2687 | return -EINVAL; | |
def596e9 JA |
2688 | kiocb->ki_complete = io_complete_rw; |
2689 | } | |
9adbd45d | 2690 | |
3529d8c2 JA |
2691 | req->rw.addr = READ_ONCE(sqe->addr); |
2692 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2693 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2694 | return 0; |
2b188cc1 JA |
2695 | } |
2696 | ||
2697 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2698 | { | |
2699 | switch (ret) { | |
2700 | case -EIOCBQUEUED: | |
2701 | break; | |
2702 | case -ERESTARTSYS: | |
2703 | case -ERESTARTNOINTR: | |
2704 | case -ERESTARTNOHAND: | |
2705 | case -ERESTART_RESTARTBLOCK: | |
2706 | /* | |
2707 | * We can't just restart the syscall, since previously | |
2708 | * submitted sqes may already be in progress. Just fail this | |
2709 | * IO with EINTR. | |
2710 | */ | |
2711 | ret = -EINTR; | |
df561f66 | 2712 | fallthrough; |
2b188cc1 JA |
2713 | default: |
2714 | kiocb->ki_complete(kiocb, ret, 0); | |
2715 | } | |
2716 | } | |
2717 | ||
a1d7c393 | 2718 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2719 | unsigned int issue_flags) |
ba816ad6 | 2720 | { |
ba04291e | 2721 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2722 | struct io_async_rw *io = req->async_data; |
ba04291e | 2723 | |
227c0c96 | 2724 | /* add previously done IO, if any */ |
e8c2bc1f | 2725 | if (io && io->bytes_done > 0) { |
227c0c96 | 2726 | if (ret < 0) |
e8c2bc1f | 2727 | ret = io->bytes_done; |
227c0c96 | 2728 | else |
e8c2bc1f | 2729 | ret += io->bytes_done; |
227c0c96 JA |
2730 | } |
2731 | ||
ba04291e JA |
2732 | if (req->flags & REQ_F_CUR_POS) |
2733 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2734 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2735 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2736 | else |
2737 | io_rw_done(kiocb, ret); | |
2738 | } | |
2739 | ||
847595de | 2740 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2741 | { |
9adbd45d JA |
2742 | struct io_ring_ctx *ctx = req->ctx; |
2743 | size_t len = req->rw.len; | |
edafccee | 2744 | struct io_mapped_ubuf *imu; |
4be1c615 | 2745 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2746 | size_t offset; |
2747 | u64 buf_addr; | |
2748 | ||
edafccee JA |
2749 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2750 | return -EFAULT; | |
edafccee JA |
2751 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2752 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2753 | buf_addr = req->rw.addr; |
edafccee JA |
2754 | |
2755 | /* overflow */ | |
2756 | if (buf_addr + len < buf_addr) | |
2757 | return -EFAULT; | |
2758 | /* not inside the mapped region */ | |
2759 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2760 | return -EFAULT; | |
2761 | ||
2762 | /* | |
2763 | * May not be a start of buffer, set size appropriately | |
2764 | * and advance us to the beginning. | |
2765 | */ | |
2766 | offset = buf_addr - imu->ubuf; | |
2767 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2768 | |
2769 | if (offset) { | |
2770 | /* | |
2771 | * Don't use iov_iter_advance() here, as it's really slow for | |
2772 | * using the latter parts of a big fixed buffer - it iterates | |
2773 | * over each segment manually. We can cheat a bit here, because | |
2774 | * we know that: | |
2775 | * | |
2776 | * 1) it's a BVEC iter, we set it up | |
2777 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2778 | * first and last bvec | |
2779 | * | |
2780 | * So just find our index, and adjust the iterator afterwards. | |
2781 | * If the offset is within the first bvec (or the whole first | |
2782 | * bvec, just use iov_iter_advance(). This makes it easier | |
2783 | * since we can just skip the first segment, which may not | |
2784 | * be PAGE_SIZE aligned. | |
2785 | */ | |
2786 | const struct bio_vec *bvec = imu->bvec; | |
2787 | ||
2788 | if (offset <= bvec->bv_len) { | |
2789 | iov_iter_advance(iter, offset); | |
2790 | } else { | |
2791 | unsigned long seg_skip; | |
2792 | ||
2793 | /* skip first vec */ | |
2794 | offset -= bvec->bv_len; | |
2795 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2796 | ||
2797 | iter->bvec = bvec + seg_skip; | |
2798 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2799 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2800 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2801 | } |
2802 | } | |
2803 | ||
847595de | 2804 | return 0; |
edafccee JA |
2805 | } |
2806 | ||
bcda7baa JA |
2807 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2808 | { | |
2809 | if (needs_lock) | |
2810 | mutex_unlock(&ctx->uring_lock); | |
2811 | } | |
2812 | ||
2813 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2814 | { | |
2815 | /* | |
2816 | * "Normal" inline submissions always hold the uring_lock, since we | |
2817 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2818 | * The only exception is when we've detached the request and issue it | |
2819 | * from an async worker thread, grab the lock for that case. | |
2820 | */ | |
2821 | if (needs_lock) | |
2822 | mutex_lock(&ctx->uring_lock); | |
2823 | } | |
2824 | ||
2825 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2826 | int bgid, struct io_buffer *kbuf, | |
2827 | bool needs_lock) | |
2828 | { | |
2829 | struct io_buffer *head; | |
2830 | ||
2831 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2832 | return kbuf; | |
2833 | ||
2834 | io_ring_submit_lock(req->ctx, needs_lock); | |
2835 | ||
2836 | lockdep_assert_held(&req->ctx->uring_lock); | |
2837 | ||
2838 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2839 | if (head) { | |
2840 | if (!list_empty(&head->list)) { | |
2841 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2842 | list); | |
2843 | list_del(&kbuf->list); | |
2844 | } else { | |
2845 | kbuf = head; | |
2846 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2847 | } | |
2848 | if (*len > kbuf->len) | |
2849 | *len = kbuf->len; | |
2850 | } else { | |
2851 | kbuf = ERR_PTR(-ENOBUFS); | |
2852 | } | |
2853 | ||
2854 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2855 | ||
2856 | return kbuf; | |
2857 | } | |
2858 | ||
4d954c25 JA |
2859 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2860 | bool needs_lock) | |
2861 | { | |
2862 | struct io_buffer *kbuf; | |
4f4eeba8 | 2863 | u16 bgid; |
4d954c25 JA |
2864 | |
2865 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2866 | bgid = req->buf_index; |
4d954c25 JA |
2867 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2868 | if (IS_ERR(kbuf)) | |
2869 | return kbuf; | |
2870 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2871 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2872 | return u64_to_user_ptr(kbuf->addr); | |
2873 | } | |
2874 | ||
2875 | #ifdef CONFIG_COMPAT | |
2876 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2877 | bool needs_lock) | |
2878 | { | |
2879 | struct compat_iovec __user *uiov; | |
2880 | compat_ssize_t clen; | |
2881 | void __user *buf; | |
2882 | ssize_t len; | |
2883 | ||
2884 | uiov = u64_to_user_ptr(req->rw.addr); | |
2885 | if (!access_ok(uiov, sizeof(*uiov))) | |
2886 | return -EFAULT; | |
2887 | if (__get_user(clen, &uiov->iov_len)) | |
2888 | return -EFAULT; | |
2889 | if (clen < 0) | |
2890 | return -EINVAL; | |
2891 | ||
2892 | len = clen; | |
2893 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2894 | if (IS_ERR(buf)) | |
2895 | return PTR_ERR(buf); | |
2896 | iov[0].iov_base = buf; | |
2897 | iov[0].iov_len = (compat_size_t) len; | |
2898 | return 0; | |
2899 | } | |
2900 | #endif | |
2901 | ||
2902 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2903 | bool needs_lock) | |
2904 | { | |
2905 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2906 | void __user *buf; | |
2907 | ssize_t len; | |
2908 | ||
2909 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2910 | return -EFAULT; | |
2911 | ||
2912 | len = iov[0].iov_len; | |
2913 | if (len < 0) | |
2914 | return -EINVAL; | |
2915 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2916 | if (IS_ERR(buf)) | |
2917 | return PTR_ERR(buf); | |
2918 | iov[0].iov_base = buf; | |
2919 | iov[0].iov_len = len; | |
2920 | return 0; | |
2921 | } | |
2922 | ||
2923 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2924 | bool needs_lock) | |
2925 | { | |
dddb3e26 JA |
2926 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2927 | struct io_buffer *kbuf; | |
2928 | ||
2929 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2930 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2931 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2932 | return 0; |
dddb3e26 | 2933 | } |
dd201662 | 2934 | if (req->rw.len != 1) |
4d954c25 JA |
2935 | return -EINVAL; |
2936 | ||
2937 | #ifdef CONFIG_COMPAT | |
2938 | if (req->ctx->compat) | |
2939 | return io_compat_import(req, iov, needs_lock); | |
2940 | #endif | |
2941 | ||
2942 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2943 | } | |
2944 | ||
847595de PB |
2945 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2946 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2947 | { |
9adbd45d JA |
2948 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2949 | size_t sqe_len = req->rw.len; | |
847595de | 2950 | u8 opcode = req->opcode; |
4d954c25 | 2951 | ssize_t ret; |
edafccee | 2952 | |
7d009165 | 2953 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2954 | *iovec = NULL; |
9adbd45d | 2955 | return io_import_fixed(req, rw, iter); |
edafccee | 2956 | } |
2b188cc1 | 2957 | |
bcda7baa | 2958 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2959 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2960 | return -EINVAL; |
2961 | ||
3a6820f2 | 2962 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2963 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2964 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2965 | if (IS_ERR(buf)) |
4d954c25 | 2966 | return PTR_ERR(buf); |
3f9d6441 | 2967 | req->rw.len = sqe_len; |
bcda7baa JA |
2968 | } |
2969 | ||
3a6820f2 JA |
2970 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2971 | *iovec = NULL; | |
10fc72e4 | 2972 | return ret; |
3a6820f2 JA |
2973 | } |
2974 | ||
4d954c25 JA |
2975 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2976 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2977 | if (!ret) |
2978 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2979 | *iovec = NULL; |
2980 | return ret; | |
2981 | } | |
2982 | ||
89cd35c5 CH |
2983 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2984 | req->ctx->compat); | |
2b188cc1 JA |
2985 | } |
2986 | ||
0fef9483 JA |
2987 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2988 | { | |
5b09e37e | 2989 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
2990 | } |
2991 | ||
31b51510 | 2992 | /* |
32960613 JA |
2993 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2994 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2995 | */ |
4017eb91 | 2996 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 2997 | { |
4017eb91 JA |
2998 | struct kiocb *kiocb = &req->rw.kiocb; |
2999 | struct file *file = req->file; | |
32960613 JA |
3000 | ssize_t ret = 0; |
3001 | ||
3002 | /* | |
3003 | * Don't support polled IO through this interface, and we can't | |
3004 | * support non-blocking either. For the latter, this just causes | |
3005 | * the kiocb to be handled from an async context. | |
3006 | */ | |
3007 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3008 | return -EOPNOTSUPP; | |
3009 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3010 | return -EAGAIN; | |
3011 | ||
3012 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3013 | struct iovec iovec; |
32960613 JA |
3014 | ssize_t nr; |
3015 | ||
311ae9e1 PB |
3016 | if (!iov_iter_is_bvec(iter)) { |
3017 | iovec = iov_iter_iovec(iter); | |
3018 | } else { | |
4017eb91 JA |
3019 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3020 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3021 | } |
3022 | ||
32960613 JA |
3023 | if (rw == READ) { |
3024 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3025 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3026 | } else { |
3027 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3028 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3029 | } |
3030 | ||
3031 | if (nr < 0) { | |
3032 | if (!ret) | |
3033 | ret = nr; | |
3034 | break; | |
3035 | } | |
3036 | ret += nr; | |
3037 | if (nr != iovec.iov_len) | |
3038 | break; | |
4017eb91 JA |
3039 | req->rw.len -= nr; |
3040 | req->rw.addr += nr; | |
32960613 JA |
3041 | iov_iter_advance(iter, nr); |
3042 | } | |
3043 | ||
3044 | return ret; | |
3045 | } | |
3046 | ||
ff6165b2 JA |
3047 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3048 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3049 | { |
e8c2bc1f | 3050 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3051 | |
ff6165b2 | 3052 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3053 | rw->free_iovec = iovec; |
227c0c96 | 3054 | rw->bytes_done = 0; |
ff6165b2 | 3055 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3056 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3057 | return; |
b64e3444 | 3058 | if (!iovec) { |
ff6165b2 JA |
3059 | unsigned iov_off = 0; |
3060 | ||
3061 | rw->iter.iov = rw->fast_iov; | |
3062 | if (iter->iov != fast_iov) { | |
3063 | iov_off = iter->iov - fast_iov; | |
3064 | rw->iter.iov += iov_off; | |
3065 | } | |
3066 | if (rw->fast_iov != fast_iov) | |
3067 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3068 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3069 | } else { |
3070 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3071 | } |
3072 | } | |
3073 | ||
e8c2bc1f | 3074 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3075 | { |
e8c2bc1f JA |
3076 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3077 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3078 | return req->async_data == NULL; | |
3d9932a8 XW |
3079 | } |
3080 | ||
e8c2bc1f | 3081 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3082 | { |
e8c2bc1f | 3083 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3084 | return 0; |
3d9932a8 | 3085 | |
e8c2bc1f | 3086 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3087 | } |
3088 | ||
ff6165b2 JA |
3089 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3090 | const struct iovec *fast_iov, | |
227c0c96 | 3091 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3092 | { |
e8c2bc1f | 3093 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3094 | return 0; |
e8c2bc1f | 3095 | if (!req->async_data) { |
6bf985dc PB |
3096 | if (__io_alloc_async_data(req)) { |
3097 | kfree(iovec); | |
5d204bcf | 3098 | return -ENOMEM; |
6bf985dc | 3099 | } |
b7bb4f7d | 3100 | |
ff6165b2 | 3101 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3102 | } |
b7bb4f7d | 3103 | return 0; |
f67676d1 JA |
3104 | } |
3105 | ||
73debe68 | 3106 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3107 | { |
e8c2bc1f | 3108 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3109 | struct iovec *iov = iorw->fast_iov; |
847595de | 3110 | int ret; |
c3e330a4 | 3111 | |
2846c481 | 3112 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3113 | if (unlikely(ret < 0)) |
3114 | return ret; | |
3115 | ||
ab0b196c PB |
3116 | iorw->bytes_done = 0; |
3117 | iorw->free_iovec = iov; | |
3118 | if (iov) | |
3119 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3120 | return 0; |
3121 | } | |
3122 | ||
73debe68 | 3123 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3124 | { |
3529d8c2 JA |
3125 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3126 | return -EBADF; | |
93642ef8 | 3127 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3128 | } |
3129 | ||
c1dd91d1 JA |
3130 | /* |
3131 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3132 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3133 | * This gets called when the page is unlocked, and we generally expect that to | |
3134 | * happen when the page IO is completed and the page is now uptodate. This will | |
3135 | * queue a task_work based retry of the operation, attempting to copy the data | |
3136 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3137 | * do a thread based blocking retry of the operation. That's the unexpected | |
3138 | * slow path. | |
3139 | */ | |
bcf5a063 JA |
3140 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3141 | int sync, void *arg) | |
3142 | { | |
3143 | struct wait_page_queue *wpq; | |
3144 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3145 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3146 | |
3147 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3148 | ||
cdc8fcb4 LT |
3149 | if (!wake_page_match(wpq, key)) |
3150 | return 0; | |
3151 | ||
c8d317aa | 3152 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3153 | list_del_init(&wait->entry); |
3154 | ||
bcf5a063 JA |
3155 | /* submit ref gets dropped, acquire a new one */ |
3156 | refcount_inc(&req->refs); | |
921b9054 | 3157 | io_req_task_queue(req); |
bcf5a063 JA |
3158 | return 1; |
3159 | } | |
3160 | ||
c1dd91d1 JA |
3161 | /* |
3162 | * This controls whether a given IO request should be armed for async page | |
3163 | * based retry. If we return false here, the request is handed to the async | |
3164 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3165 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3166 | * will either succeed because the page is now uptodate and unlocked, or it | |
3167 | * will register a callback when the page is unlocked at IO completion. Through | |
3168 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3169 | * That retry will attempt the buffered read again. The retry will generally | |
3170 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3171 | * async worker threads for a blocking retry. | |
3172 | */ | |
227c0c96 | 3173 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3174 | { |
e8c2bc1f JA |
3175 | struct io_async_rw *rw = req->async_data; |
3176 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3177 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3178 | |
bcf5a063 JA |
3179 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3180 | if (req->flags & REQ_F_NOWAIT) | |
3181 | return false; | |
f67676d1 | 3182 | |
227c0c96 | 3183 | /* Only for buffered IO */ |
3b2a4439 | 3184 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3185 | return false; |
3b2a4439 | 3186 | |
bcf5a063 JA |
3187 | /* |
3188 | * just use poll if we can, and don't attempt if the fs doesn't | |
3189 | * support callback based unlocks | |
3190 | */ | |
3191 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3192 | return false; | |
f67676d1 | 3193 | |
3b2a4439 JA |
3194 | wait->wait.func = io_async_buf_func; |
3195 | wait->wait.private = req; | |
3196 | wait->wait.flags = 0; | |
3197 | INIT_LIST_HEAD(&wait->wait.entry); | |
3198 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3199 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3200 | kiocb->ki_waitq = wait; |
3b2a4439 | 3201 | return true; |
bcf5a063 JA |
3202 | } |
3203 | ||
3204 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3205 | { | |
3206 | if (req->file->f_op->read_iter) | |
3207 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3208 | else if (req->file->f_op->read) |
4017eb91 | 3209 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3210 | else |
3211 | return -EINVAL; | |
f67676d1 JA |
3212 | } |
3213 | ||
889fca73 | 3214 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3215 | { |
3216 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3217 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3218 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3219 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3220 | ssize_t io_size, ret, ret2; |
45d189c6 | 3221 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3222 | |
2846c481 | 3223 | if (rw) { |
e8c2bc1f | 3224 | iter = &rw->iter; |
2846c481 PB |
3225 | iovec = NULL; |
3226 | } else { | |
3227 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3228 | if (ret < 0) | |
3229 | return ret; | |
3230 | } | |
632546c4 | 3231 | io_size = iov_iter_count(iter); |
fa15bafb | 3232 | req->result = io_size; |
2b188cc1 | 3233 | |
fd6c2e4c JA |
3234 | /* Ensure we clear previously set non-block flag */ |
3235 | if (!force_nonblock) | |
29de5f6a | 3236 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3237 | else |
3238 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3239 | ||
24c74678 | 3240 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3241 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3242 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3243 | return ret ?: -EAGAIN; |
6713e7a6 | 3244 | } |
9e645e11 | 3245 | |
632546c4 | 3246 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3247 | if (unlikely(ret)) { |
3248 | kfree(iovec); | |
3249 | return ret; | |
3250 | } | |
2b188cc1 | 3251 | |
227c0c96 | 3252 | ret = io_iter_do_read(req, iter); |
32960613 | 3253 | |
57cd657b | 3254 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3255 | if (req->async_data) |
3256 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3257 | goto out_free; |
227c0c96 | 3258 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3259 | /* IOPOLL retry should happen for io-wq threads */ |
3260 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3261 | goto done; |
75c668cd PB |
3262 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3263 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3264 | goto done; |
84216315 | 3265 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3266 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3267 | ret = 0; |
7335e3bf | 3268 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3269 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3270 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3271 | goto done; |
227c0c96 JA |
3272 | } |
3273 | ||
227c0c96 | 3274 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3275 | if (ret2) |
3276 | return ret2; | |
3277 | ||
fe1cdd55 | 3278 | iovec = NULL; |
e8c2bc1f | 3279 | rw = req->async_data; |
227c0c96 | 3280 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3281 | iter = &rw->iter; |
227c0c96 | 3282 | |
b23df91b PB |
3283 | do { |
3284 | io_size -= ret; | |
3285 | rw->bytes_done += ret; | |
3286 | /* if we can retry, do so with the callbacks armed */ | |
3287 | if (!io_rw_should_retry(req)) { | |
3288 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3289 | return -EAGAIN; | |
3290 | } | |
3291 | ||
3292 | /* | |
3293 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3294 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3295 | * desired page gets unlocked. We can also get a partial read | |
3296 | * here, and if we do, then just retry at the new offset. | |
3297 | */ | |
3298 | ret = io_iter_do_read(req, iter); | |
3299 | if (ret == -EIOCBQUEUED) | |
3300 | return 0; | |
227c0c96 | 3301 | /* we got some bytes, but not all. retry. */ |
b23df91b | 3302 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3303 | done: |
889fca73 | 3304 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3305 | out_free: |
3306 | /* it's faster to check here then delegate to kfree */ | |
3307 | if (iovec) | |
3308 | kfree(iovec); | |
5ea5dd45 | 3309 | return 0; |
2b188cc1 JA |
3310 | } |
3311 | ||
73debe68 | 3312 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3313 | { |
3529d8c2 JA |
3314 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3315 | return -EBADF; | |
93642ef8 | 3316 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3317 | } |
3318 | ||
889fca73 | 3319 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3320 | { |
3321 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3322 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3323 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3324 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3325 | ssize_t ret, ret2, io_size; |
45d189c6 | 3326 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3327 | |
2846c481 | 3328 | if (rw) { |
e8c2bc1f | 3329 | iter = &rw->iter; |
2846c481 PB |
3330 | iovec = NULL; |
3331 | } else { | |
3332 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3333 | if (ret < 0) | |
3334 | return ret; | |
3335 | } | |
632546c4 | 3336 | io_size = iov_iter_count(iter); |
fa15bafb | 3337 | req->result = io_size; |
2b188cc1 | 3338 | |
fd6c2e4c JA |
3339 | /* Ensure we clear previously set non-block flag */ |
3340 | if (!force_nonblock) | |
a88fc400 PB |
3341 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3342 | else | |
3343 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3344 | |
24c74678 | 3345 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3346 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3347 | goto copy_iov; |
31b51510 | 3348 | |
10d59345 JA |
3349 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3350 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3351 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3352 | goto copy_iov; |
31b51510 | 3353 | |
632546c4 | 3354 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3355 | if (unlikely(ret)) |
3356 | goto out_free; | |
4ed734b0 | 3357 | |
fa15bafb PB |
3358 | /* |
3359 | * Open-code file_start_write here to grab freeze protection, | |
3360 | * which will be released by another thread in | |
3361 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3362 | * released so that it doesn't complain about the held lock when | |
3363 | * we return to userspace. | |
3364 | */ | |
3365 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3366 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3367 | __sb_writers_release(file_inode(req->file)->i_sb, |
3368 | SB_FREEZE_WRITE); | |
3369 | } | |
3370 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3371 | |
fa15bafb | 3372 | if (req->file->f_op->write_iter) |
ff6165b2 | 3373 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3374 | else if (req->file->f_op->write) |
4017eb91 | 3375 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3376 | else |
3377 | ret2 = -EINVAL; | |
4ed734b0 | 3378 | |
fa15bafb PB |
3379 | /* |
3380 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3381 | * retry them without IOCB_NOWAIT. | |
3382 | */ | |
3383 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3384 | ret2 = -EAGAIN; | |
75c668cd PB |
3385 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3386 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3387 | goto done; |
3e6a0d3c JA |
3388 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3389 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3390 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3391 | /* IOPOLL retry should happen for io-wq threads */ |
3392 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3393 | goto copy_iov; | |
355afaeb | 3394 | done: |
889fca73 | 3395 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3396 | } else { |
f67676d1 | 3397 | copy_iov: |
84216315 | 3398 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3399 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3400 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3401 | return ret ?: -EAGAIN; |
2b188cc1 | 3402 | } |
31b51510 | 3403 | out_free: |
f261c168 | 3404 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3405 | if (iovec) |
6f2cc166 | 3406 | kfree(iovec); |
2b188cc1 JA |
3407 | return ret; |
3408 | } | |
3409 | ||
80a261fd JA |
3410 | static int io_renameat_prep(struct io_kiocb *req, |
3411 | const struct io_uring_sqe *sqe) | |
3412 | { | |
3413 | struct io_rename *ren = &req->rename; | |
3414 | const char __user *oldf, *newf; | |
3415 | ||
3416 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3417 | return -EBADF; | |
3418 | ||
3419 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3420 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3421 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3422 | ren->new_dfd = READ_ONCE(sqe->len); | |
3423 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3424 | ||
3425 | ren->oldpath = getname(oldf); | |
3426 | if (IS_ERR(ren->oldpath)) | |
3427 | return PTR_ERR(ren->oldpath); | |
3428 | ||
3429 | ren->newpath = getname(newf); | |
3430 | if (IS_ERR(ren->newpath)) { | |
3431 | putname(ren->oldpath); | |
3432 | return PTR_ERR(ren->newpath); | |
3433 | } | |
3434 | ||
3435 | req->flags |= REQ_F_NEED_CLEANUP; | |
3436 | return 0; | |
3437 | } | |
3438 | ||
45d189c6 | 3439 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3440 | { |
3441 | struct io_rename *ren = &req->rename; | |
3442 | int ret; | |
3443 | ||
45d189c6 | 3444 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3445 | return -EAGAIN; |
3446 | ||
3447 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3448 | ren->newpath, ren->flags); | |
3449 | ||
3450 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3451 | if (ret < 0) | |
3452 | req_set_fail_links(req); | |
3453 | io_req_complete(req, ret); | |
3454 | return 0; | |
3455 | } | |
3456 | ||
14a1143b JA |
3457 | static int io_unlinkat_prep(struct io_kiocb *req, |
3458 | const struct io_uring_sqe *sqe) | |
3459 | { | |
3460 | struct io_unlink *un = &req->unlink; | |
3461 | const char __user *fname; | |
3462 | ||
3463 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3464 | return -EBADF; | |
3465 | ||
3466 | un->dfd = READ_ONCE(sqe->fd); | |
3467 | ||
3468 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3469 | if (un->flags & ~AT_REMOVEDIR) | |
3470 | return -EINVAL; | |
3471 | ||
3472 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3473 | un->filename = getname(fname); | |
3474 | if (IS_ERR(un->filename)) | |
3475 | return PTR_ERR(un->filename); | |
3476 | ||
3477 | req->flags |= REQ_F_NEED_CLEANUP; | |
3478 | return 0; | |
3479 | } | |
3480 | ||
45d189c6 | 3481 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3482 | { |
3483 | struct io_unlink *un = &req->unlink; | |
3484 | int ret; | |
3485 | ||
45d189c6 | 3486 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3487 | return -EAGAIN; |
3488 | ||
3489 | if (un->flags & AT_REMOVEDIR) | |
3490 | ret = do_rmdir(un->dfd, un->filename); | |
3491 | else | |
3492 | ret = do_unlinkat(un->dfd, un->filename); | |
3493 | ||
3494 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3495 | if (ret < 0) | |
3496 | req_set_fail_links(req); | |
3497 | io_req_complete(req, ret); | |
3498 | return 0; | |
3499 | } | |
3500 | ||
36f4fa68 JA |
3501 | static int io_shutdown_prep(struct io_kiocb *req, |
3502 | const struct io_uring_sqe *sqe) | |
3503 | { | |
3504 | #if defined(CONFIG_NET) | |
3505 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3506 | return -EINVAL; | |
3507 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3508 | sqe->buf_index) | |
3509 | return -EINVAL; | |
3510 | ||
3511 | req->shutdown.how = READ_ONCE(sqe->len); | |
3512 | return 0; | |
3513 | #else | |
3514 | return -EOPNOTSUPP; | |
3515 | #endif | |
3516 | } | |
3517 | ||
45d189c6 | 3518 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3519 | { |
3520 | #if defined(CONFIG_NET) | |
3521 | struct socket *sock; | |
3522 | int ret; | |
3523 | ||
45d189c6 | 3524 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3525 | return -EAGAIN; |
3526 | ||
48aba79b | 3527 | sock = sock_from_file(req->file); |
36f4fa68 | 3528 | if (unlikely(!sock)) |
48aba79b | 3529 | return -ENOTSOCK; |
36f4fa68 JA |
3530 | |
3531 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3532 | if (ret < 0) |
3533 | req_set_fail_links(req); | |
36f4fa68 JA |
3534 | io_req_complete(req, ret); |
3535 | return 0; | |
3536 | #else | |
3537 | return -EOPNOTSUPP; | |
3538 | #endif | |
3539 | } | |
3540 | ||
f2a8d5c7 PB |
3541 | static int __io_splice_prep(struct io_kiocb *req, |
3542 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3543 | { |
3544 | struct io_splice* sp = &req->splice; | |
3545 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3546 | |
3232dd02 PB |
3547 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3548 | return -EINVAL; | |
7d67af2c PB |
3549 | |
3550 | sp->file_in = NULL; | |
7d67af2c PB |
3551 | sp->len = READ_ONCE(sqe->len); |
3552 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3553 | ||
3554 | if (unlikely(sp->flags & ~valid_flags)) | |
3555 | return -EINVAL; | |
3556 | ||
8371adf5 PB |
3557 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3558 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3559 | if (!sp->file_in) | |
3560 | return -EBADF; | |
7d67af2c PB |
3561 | req->flags |= REQ_F_NEED_CLEANUP; |
3562 | ||
7cdaf587 XW |
3563 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3564 | /* | |
3565 | * Splice operation will be punted aync, and here need to | |
3566 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3567 | */ | |
7d67af2c | 3568 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3569 | } |
7d67af2c PB |
3570 | |
3571 | return 0; | |
3572 | } | |
3573 | ||
f2a8d5c7 PB |
3574 | static int io_tee_prep(struct io_kiocb *req, |
3575 | const struct io_uring_sqe *sqe) | |
3576 | { | |
3577 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3578 | return -EINVAL; | |
3579 | return __io_splice_prep(req, sqe); | |
3580 | } | |
3581 | ||
45d189c6 | 3582 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3583 | { |
3584 | struct io_splice *sp = &req->splice; | |
3585 | struct file *in = sp->file_in; | |
3586 | struct file *out = sp->file_out; | |
3587 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3588 | long ret = 0; | |
3589 | ||
45d189c6 | 3590 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3591 | return -EAGAIN; |
3592 | if (sp->len) | |
3593 | ret = do_tee(in, out, sp->len, flags); | |
3594 | ||
3595 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3596 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3597 | ||
f2a8d5c7 PB |
3598 | if (ret != sp->len) |
3599 | req_set_fail_links(req); | |
e1e16097 | 3600 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3601 | return 0; |
3602 | } | |
3603 | ||
3604 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3605 | { | |
3606 | struct io_splice* sp = &req->splice; | |
3607 | ||
3608 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3609 | sp->off_out = READ_ONCE(sqe->off); | |
3610 | return __io_splice_prep(req, sqe); | |
3611 | } | |
3612 | ||
45d189c6 | 3613 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3614 | { |
3615 | struct io_splice *sp = &req->splice; | |
3616 | struct file *in = sp->file_in; | |
3617 | struct file *out = sp->file_out; | |
3618 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3619 | loff_t *poff_in, *poff_out; | |
c9687426 | 3620 | long ret = 0; |
7d67af2c | 3621 | |
45d189c6 | 3622 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3623 | return -EAGAIN; |
7d67af2c PB |
3624 | |
3625 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3626 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3627 | |
948a7749 | 3628 | if (sp->len) |
c9687426 | 3629 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3630 | |
3631 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3632 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3633 | ||
7d67af2c PB |
3634 | if (ret != sp->len) |
3635 | req_set_fail_links(req); | |
e1e16097 | 3636 | io_req_complete(req, ret); |
7d67af2c PB |
3637 | return 0; |
3638 | } | |
3639 | ||
2b188cc1 JA |
3640 | /* |
3641 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3642 | */ | |
889fca73 | 3643 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3644 | { |
3645 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3646 | |
def596e9 JA |
3647 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3648 | return -EINVAL; | |
3649 | ||
889fca73 | 3650 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3651 | return 0; |
3652 | } | |
3653 | ||
1155c76a | 3654 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3655 | { |
6b06314c | 3656 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3657 | |
09bb8394 JA |
3658 | if (!req->file) |
3659 | return -EBADF; | |
c992fe29 | 3660 | |
6b06314c | 3661 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3662 | return -EINVAL; |
edafccee | 3663 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3664 | return -EINVAL; |
3665 | ||
8ed8d3c3 JA |
3666 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3667 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3668 | return -EINVAL; | |
3669 | ||
3670 | req->sync.off = READ_ONCE(sqe->off); | |
3671 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3672 | return 0; |
3673 | } | |
3674 | ||
45d189c6 | 3675 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3676 | { |
8ed8d3c3 | 3677 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3678 | int ret; |
3679 | ||
ac45abc0 | 3680 | /* fsync always requires a blocking context */ |
45d189c6 | 3681 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3682 | return -EAGAIN; |
3683 | ||
9adbd45d | 3684 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3685 | end > 0 ? end : LLONG_MAX, |
3686 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3687 | if (ret < 0) | |
3688 | req_set_fail_links(req); | |
e1e16097 | 3689 | io_req_complete(req, ret); |
c992fe29 CH |
3690 | return 0; |
3691 | } | |
3692 | ||
d63d1b5e JA |
3693 | static int io_fallocate_prep(struct io_kiocb *req, |
3694 | const struct io_uring_sqe *sqe) | |
3695 | { | |
3696 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3697 | return -EINVAL; | |
3232dd02 PB |
3698 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3699 | return -EINVAL; | |
d63d1b5e JA |
3700 | |
3701 | req->sync.off = READ_ONCE(sqe->off); | |
3702 | req->sync.len = READ_ONCE(sqe->addr); | |
3703 | req->sync.mode = READ_ONCE(sqe->len); | |
3704 | return 0; | |
3705 | } | |
3706 | ||
45d189c6 | 3707 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3708 | { |
ac45abc0 PB |
3709 | int ret; |
3710 | ||
d63d1b5e | 3711 | /* fallocate always requiring blocking context */ |
45d189c6 | 3712 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3713 | return -EAGAIN; |
ac45abc0 PB |
3714 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3715 | req->sync.len); | |
ac45abc0 PB |
3716 | if (ret < 0) |
3717 | req_set_fail_links(req); | |
e1e16097 | 3718 | io_req_complete(req, ret); |
5d17b4a4 JA |
3719 | return 0; |
3720 | } | |
3721 | ||
ec65fea5 | 3722 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3723 | { |
f8748881 | 3724 | const char __user *fname; |
15b71abe | 3725 | int ret; |
b7bb4f7d | 3726 | |
ec65fea5 | 3727 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3728 | return -EINVAL; |
ec65fea5 | 3729 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3730 | return -EBADF; |
03b1230c | 3731 | |
ec65fea5 PB |
3732 | /* open.how should be already initialised */ |
3733 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3734 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3735 | |
25e72d10 PB |
3736 | req->open.dfd = READ_ONCE(sqe->fd); |
3737 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3738 | req->open.filename = getname(fname); |
15b71abe JA |
3739 | if (IS_ERR(req->open.filename)) { |
3740 | ret = PTR_ERR(req->open.filename); | |
3741 | req->open.filename = NULL; | |
3742 | return ret; | |
3743 | } | |
4022e7af | 3744 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3745 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3746 | return 0; |
03b1230c JA |
3747 | } |
3748 | ||
ec65fea5 PB |
3749 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3750 | { | |
3751 | u64 flags, mode; | |
3752 | ||
14587a46 | 3753 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3754 | return -EINVAL; |
ec65fea5 PB |
3755 | mode = READ_ONCE(sqe->len); |
3756 | flags = READ_ONCE(sqe->open_flags); | |
3757 | req->open.how = build_open_how(flags, mode); | |
3758 | return __io_openat_prep(req, sqe); | |
3759 | } | |
3760 | ||
cebdb986 | 3761 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3762 | { |
cebdb986 | 3763 | struct open_how __user *how; |
cebdb986 | 3764 | size_t len; |
0fa03c62 JA |
3765 | int ret; |
3766 | ||
14587a46 | 3767 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3768 | return -EINVAL; |
cebdb986 JA |
3769 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3770 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3771 | if (len < OPEN_HOW_SIZE_VER0) |
3772 | return -EINVAL; | |
3529d8c2 | 3773 | |
cebdb986 JA |
3774 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3775 | len); | |
3776 | if (ret) | |
3777 | return ret; | |
3529d8c2 | 3778 | |
ec65fea5 | 3779 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3780 | } |
3781 | ||
45d189c6 | 3782 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3783 | { |
3784 | struct open_flags op; | |
15b71abe | 3785 | struct file *file; |
3a81fd02 JA |
3786 | bool nonblock_set; |
3787 | bool resolve_nonblock; | |
15b71abe JA |
3788 | int ret; |
3789 | ||
cebdb986 | 3790 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3791 | if (ret) |
3792 | goto err; | |
3a81fd02 JA |
3793 | nonblock_set = op.open_flag & O_NONBLOCK; |
3794 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3795 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3796 | /* |
3797 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3798 | * it'll always -EAGAIN | |
3799 | */ | |
3800 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3801 | return -EAGAIN; | |
3802 | op.lookup_flags |= LOOKUP_CACHED; | |
3803 | op.open_flag |= O_NONBLOCK; | |
3804 | } | |
15b71abe | 3805 | |
4022e7af | 3806 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3807 | if (ret < 0) |
3808 | goto err; | |
3809 | ||
3810 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3811 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3812 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3813 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3814 | /* |
3a81fd02 JA |
3815 | * We could hang on to this 'fd', but seems like marginal |
3816 | * gain for something that is now known to be a slower path. | |
3817 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3818 | */ |
3a81fd02 JA |
3819 | put_unused_fd(ret); |
3820 | return -EAGAIN; | |
3821 | } | |
3822 | ||
15b71abe JA |
3823 | if (IS_ERR(file)) { |
3824 | put_unused_fd(ret); | |
3825 | ret = PTR_ERR(file); | |
3826 | } else { | |
45d189c6 | 3827 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3828 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3829 | fsnotify_open(file); |
3830 | fd_install(ret, file); | |
3831 | } | |
3832 | err: | |
3833 | putname(req->open.filename); | |
8fef80bf | 3834 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3835 | if (ret < 0) |
3836 | req_set_fail_links(req); | |
e1e16097 | 3837 | io_req_complete(req, ret); |
15b71abe JA |
3838 | return 0; |
3839 | } | |
3840 | ||
45d189c6 | 3841 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3842 | { |
45d189c6 | 3843 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
3844 | } |
3845 | ||
067524e9 JA |
3846 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3847 | const struct io_uring_sqe *sqe) | |
3848 | { | |
3849 | struct io_provide_buf *p = &req->pbuf; | |
3850 | u64 tmp; | |
3851 | ||
3852 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3853 | return -EINVAL; | |
3854 | ||
3855 | tmp = READ_ONCE(sqe->fd); | |
3856 | if (!tmp || tmp > USHRT_MAX) | |
3857 | return -EINVAL; | |
3858 | ||
3859 | memset(p, 0, sizeof(*p)); | |
3860 | p->nbufs = tmp; | |
3861 | p->bgid = READ_ONCE(sqe->buf_group); | |
3862 | return 0; | |
3863 | } | |
3864 | ||
3865 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3866 | int bgid, unsigned nbufs) | |
3867 | { | |
3868 | unsigned i = 0; | |
3869 | ||
3870 | /* shouldn't happen */ | |
3871 | if (!nbufs) | |
3872 | return 0; | |
3873 | ||
3874 | /* the head kbuf is the list itself */ | |
3875 | while (!list_empty(&buf->list)) { | |
3876 | struct io_buffer *nxt; | |
3877 | ||
3878 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3879 | list_del(&nxt->list); | |
3880 | kfree(nxt); | |
3881 | if (++i == nbufs) | |
3882 | return i; | |
3883 | } | |
3884 | i++; | |
3885 | kfree(buf); | |
3886 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3887 | ||
3888 | return i; | |
3889 | } | |
3890 | ||
889fca73 | 3891 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3892 | { |
3893 | struct io_provide_buf *p = &req->pbuf; | |
3894 | struct io_ring_ctx *ctx = req->ctx; | |
3895 | struct io_buffer *head; | |
3896 | int ret = 0; | |
45d189c6 | 3897 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3898 | |
3899 | io_ring_submit_lock(ctx, !force_nonblock); | |
3900 | ||
3901 | lockdep_assert_held(&ctx->uring_lock); | |
3902 | ||
3903 | ret = -ENOENT; | |
3904 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3905 | if (head) | |
3906 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3907 | if (ret < 0) |
3908 | req_set_fail_links(req); | |
067524e9 | 3909 | |
31bff9a5 PB |
3910 | /* need to hold the lock to complete IOPOLL requests */ |
3911 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3912 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3913 | io_ring_submit_unlock(ctx, !force_nonblock); |
3914 | } else { | |
3915 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3916 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3917 | } |
067524e9 JA |
3918 | return 0; |
3919 | } | |
3920 | ||
ddf0322d JA |
3921 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3922 | const struct io_uring_sqe *sqe) | |
3923 | { | |
3924 | struct io_provide_buf *p = &req->pbuf; | |
3925 | u64 tmp; | |
3926 | ||
3927 | if (sqe->ioprio || sqe->rw_flags) | |
3928 | return -EINVAL; | |
3929 | ||
3930 | tmp = READ_ONCE(sqe->fd); | |
3931 | if (!tmp || tmp > USHRT_MAX) | |
3932 | return -E2BIG; | |
3933 | p->nbufs = tmp; | |
3934 | p->addr = READ_ONCE(sqe->addr); | |
3935 | p->len = READ_ONCE(sqe->len); | |
3936 | ||
efe68c1c | 3937 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3938 | return -EFAULT; |
3939 | ||
3940 | p->bgid = READ_ONCE(sqe->buf_group); | |
3941 | tmp = READ_ONCE(sqe->off); | |
3942 | if (tmp > USHRT_MAX) | |
3943 | return -E2BIG; | |
3944 | p->bid = tmp; | |
3945 | return 0; | |
3946 | } | |
3947 | ||
3948 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3949 | { | |
3950 | struct io_buffer *buf; | |
3951 | u64 addr = pbuf->addr; | |
3952 | int i, bid = pbuf->bid; | |
3953 | ||
3954 | for (i = 0; i < pbuf->nbufs; i++) { | |
3955 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3956 | if (!buf) | |
3957 | break; | |
3958 | ||
3959 | buf->addr = addr; | |
3960 | buf->len = pbuf->len; | |
3961 | buf->bid = bid; | |
3962 | addr += pbuf->len; | |
3963 | bid++; | |
3964 | if (!*head) { | |
3965 | INIT_LIST_HEAD(&buf->list); | |
3966 | *head = buf; | |
3967 | } else { | |
3968 | list_add_tail(&buf->list, &(*head)->list); | |
3969 | } | |
3970 | } | |
3971 | ||
3972 | return i ? i : -ENOMEM; | |
3973 | } | |
3974 | ||
889fca73 | 3975 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3976 | { |
3977 | struct io_provide_buf *p = &req->pbuf; | |
3978 | struct io_ring_ctx *ctx = req->ctx; | |
3979 | struct io_buffer *head, *list; | |
3980 | int ret = 0; | |
45d189c6 | 3981 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3982 | |
3983 | io_ring_submit_lock(ctx, !force_nonblock); | |
3984 | ||
3985 | lockdep_assert_held(&ctx->uring_lock); | |
3986 | ||
3987 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3988 | ||
3989 | ret = io_add_buffers(p, &head); | |
3990 | if (ret < 0) | |
3991 | goto out; | |
3992 | ||
3993 | if (!list) { | |
3994 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3995 | GFP_KERNEL); | |
3996 | if (ret < 0) { | |
067524e9 | 3997 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3998 | goto out; |
3999 | } | |
4000 | } | |
4001 | out: | |
ddf0322d JA |
4002 | if (ret < 0) |
4003 | req_set_fail_links(req); | |
31bff9a5 PB |
4004 | |
4005 | /* need to hold the lock to complete IOPOLL requests */ | |
4006 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4007 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4008 | io_ring_submit_unlock(ctx, !force_nonblock); |
4009 | } else { | |
4010 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4011 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4012 | } |
ddf0322d | 4013 | return 0; |
cebdb986 JA |
4014 | } |
4015 | ||
3e4827b0 JA |
4016 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4017 | const struct io_uring_sqe *sqe) | |
4018 | { | |
4019 | #if defined(CONFIG_EPOLL) | |
4020 | if (sqe->ioprio || sqe->buf_index) | |
4021 | return -EINVAL; | |
6ca56f84 | 4022 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4023 | return -EINVAL; |
3e4827b0 JA |
4024 | |
4025 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4026 | req->epoll.op = READ_ONCE(sqe->len); | |
4027 | req->epoll.fd = READ_ONCE(sqe->off); | |
4028 | ||
4029 | if (ep_op_has_event(req->epoll.op)) { | |
4030 | struct epoll_event __user *ev; | |
4031 | ||
4032 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4033 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4034 | return -EFAULT; | |
4035 | } | |
4036 | ||
4037 | return 0; | |
4038 | #else | |
4039 | return -EOPNOTSUPP; | |
4040 | #endif | |
4041 | } | |
4042 | ||
889fca73 | 4043 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4044 | { |
4045 | #if defined(CONFIG_EPOLL) | |
4046 | struct io_epoll *ie = &req->epoll; | |
4047 | int ret; | |
45d189c6 | 4048 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4049 | |
4050 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4051 | if (force_nonblock && ret == -EAGAIN) | |
4052 | return -EAGAIN; | |
4053 | ||
4054 | if (ret < 0) | |
4055 | req_set_fail_links(req); | |
889fca73 | 4056 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4057 | return 0; |
4058 | #else | |
4059 | return -EOPNOTSUPP; | |
4060 | #endif | |
4061 | } | |
4062 | ||
c1ca757b JA |
4063 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4064 | { | |
4065 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4066 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4067 | return -EINVAL; | |
3232dd02 PB |
4068 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4069 | return -EINVAL; | |
c1ca757b JA |
4070 | |
4071 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4072 | req->madvise.len = READ_ONCE(sqe->len); | |
4073 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4074 | return 0; | |
4075 | #else | |
4076 | return -EOPNOTSUPP; | |
4077 | #endif | |
4078 | } | |
4079 | ||
45d189c6 | 4080 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4081 | { |
4082 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4083 | struct io_madvise *ma = &req->madvise; | |
4084 | int ret; | |
4085 | ||
45d189c6 | 4086 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4087 | return -EAGAIN; |
4088 | ||
0726b01e | 4089 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4090 | if (ret < 0) |
4091 | req_set_fail_links(req); | |
e1e16097 | 4092 | io_req_complete(req, ret); |
c1ca757b JA |
4093 | return 0; |
4094 | #else | |
4095 | return -EOPNOTSUPP; | |
4096 | #endif | |
4097 | } | |
4098 | ||
4840e418 JA |
4099 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4100 | { | |
4101 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4102 | return -EINVAL; | |
3232dd02 PB |
4103 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4104 | return -EINVAL; | |
4840e418 JA |
4105 | |
4106 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4107 | req->fadvise.len = READ_ONCE(sqe->len); | |
4108 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4109 | return 0; | |
4110 | } | |
4111 | ||
45d189c6 | 4112 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4113 | { |
4114 | struct io_fadvise *fa = &req->fadvise; | |
4115 | int ret; | |
4116 | ||
45d189c6 | 4117 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4118 | switch (fa->advice) { |
4119 | case POSIX_FADV_NORMAL: | |
4120 | case POSIX_FADV_RANDOM: | |
4121 | case POSIX_FADV_SEQUENTIAL: | |
4122 | break; | |
4123 | default: | |
4124 | return -EAGAIN; | |
4125 | } | |
4126 | } | |
4840e418 JA |
4127 | |
4128 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4129 | if (ret < 0) | |
4130 | req_set_fail_links(req); | |
e1e16097 | 4131 | io_req_complete(req, ret); |
4840e418 JA |
4132 | return 0; |
4133 | } | |
4134 | ||
eddc7ef5 JA |
4135 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4136 | { | |
6ca56f84 | 4137 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4138 | return -EINVAL; |
eddc7ef5 JA |
4139 | if (sqe->ioprio || sqe->buf_index) |
4140 | return -EINVAL; | |
9c280f90 | 4141 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4142 | return -EBADF; |
eddc7ef5 | 4143 | |
1d9e1288 BM |
4144 | req->statx.dfd = READ_ONCE(sqe->fd); |
4145 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4146 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4147 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4148 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4149 | |
4150 | return 0; | |
4151 | } | |
4152 | ||
45d189c6 | 4153 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4154 | { |
1d9e1288 | 4155 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4156 | int ret; |
4157 | ||
45d189c6 | 4158 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4159 | /* only need file table for an actual valid fd */ |
4160 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4161 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4162 | return -EAGAIN; |
5b0bbee4 | 4163 | } |
eddc7ef5 | 4164 | |
e62753e4 BM |
4165 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4166 | ctx->buffer); | |
eddc7ef5 | 4167 | |
eddc7ef5 JA |
4168 | if (ret < 0) |
4169 | req_set_fail_links(req); | |
e1e16097 | 4170 | io_req_complete(req, ret); |
eddc7ef5 JA |
4171 | return 0; |
4172 | } | |
4173 | ||
b5dba59e JA |
4174 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4175 | { | |
14587a46 | 4176 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4177 | return -EINVAL; |
b5dba59e JA |
4178 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4179 | sqe->rw_flags || sqe->buf_index) | |
4180 | return -EINVAL; | |
9c280f90 | 4181 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4182 | return -EBADF; |
b5dba59e JA |
4183 | |
4184 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4185 | return 0; |
b5dba59e JA |
4186 | } |
4187 | ||
889fca73 | 4188 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4189 | { |
9eac1904 | 4190 | struct files_struct *files = current->files; |
3af73b28 | 4191 | struct io_close *close = &req->close; |
9eac1904 JA |
4192 | struct fdtable *fdt; |
4193 | struct file *file; | |
b5dba59e JA |
4194 | int ret; |
4195 | ||
9eac1904 JA |
4196 | file = NULL; |
4197 | ret = -EBADF; | |
4198 | spin_lock(&files->file_lock); | |
4199 | fdt = files_fdtable(files); | |
4200 | if (close->fd >= fdt->max_fds) { | |
4201 | spin_unlock(&files->file_lock); | |
4202 | goto err; | |
4203 | } | |
4204 | file = fdt->fd[close->fd]; | |
4205 | if (!file) { | |
4206 | spin_unlock(&files->file_lock); | |
4207 | goto err; | |
4208 | } | |
4209 | ||
4210 | if (file->f_op == &io_uring_fops) { | |
4211 | spin_unlock(&files->file_lock); | |
4212 | file = NULL; | |
4213 | goto err; | |
3af73b28 | 4214 | } |
b5dba59e JA |
4215 | |
4216 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4217 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4218 | spin_unlock(&files->file_lock); |
0bf0eefd | 4219 | return -EAGAIN; |
a2100672 | 4220 | } |
b5dba59e | 4221 | |
9eac1904 JA |
4222 | ret = __close_fd_get_file(close->fd, &file); |
4223 | spin_unlock(&files->file_lock); | |
4224 | if (ret < 0) { | |
4225 | if (ret == -ENOENT) | |
4226 | ret = -EBADF; | |
4227 | goto err; | |
4228 | } | |
4229 | ||
3af73b28 | 4230 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4231 | ret = filp_close(file, current->files); |
4232 | err: | |
3af73b28 PB |
4233 | if (ret < 0) |
4234 | req_set_fail_links(req); | |
9eac1904 JA |
4235 | if (file) |
4236 | fput(file); | |
889fca73 | 4237 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4238 | return 0; |
b5dba59e JA |
4239 | } |
4240 | ||
1155c76a | 4241 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4242 | { |
4243 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4244 | |
5d17b4a4 JA |
4245 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4246 | return -EINVAL; | |
4247 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4248 | return -EINVAL; | |
4249 | ||
8ed8d3c3 JA |
4250 | req->sync.off = READ_ONCE(sqe->off); |
4251 | req->sync.len = READ_ONCE(sqe->len); | |
4252 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4253 | return 0; |
4254 | } | |
4255 | ||
45d189c6 | 4256 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4257 | { |
8ed8d3c3 JA |
4258 | int ret; |
4259 | ||
ac45abc0 | 4260 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4261 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4262 | return -EAGAIN; |
4263 | ||
9adbd45d | 4264 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4265 | req->sync.flags); |
4266 | if (ret < 0) | |
4267 | req_set_fail_links(req); | |
e1e16097 | 4268 | io_req_complete(req, ret); |
5d17b4a4 JA |
4269 | return 0; |
4270 | } | |
4271 | ||
469956e8 | 4272 | #if defined(CONFIG_NET) |
02d27d89 PB |
4273 | static int io_setup_async_msg(struct io_kiocb *req, |
4274 | struct io_async_msghdr *kmsg) | |
4275 | { | |
e8c2bc1f JA |
4276 | struct io_async_msghdr *async_msg = req->async_data; |
4277 | ||
4278 | if (async_msg) | |
02d27d89 | 4279 | return -EAGAIN; |
e8c2bc1f | 4280 | if (io_alloc_async_data(req)) { |
257e84a5 | 4281 | kfree(kmsg->free_iov); |
02d27d89 PB |
4282 | return -ENOMEM; |
4283 | } | |
e8c2bc1f | 4284 | async_msg = req->async_data; |
02d27d89 | 4285 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4286 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4287 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4288 | /* if were using fast_iov, set it to the new one */ |
4289 | if (!async_msg->free_iov) | |
4290 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4291 | ||
02d27d89 PB |
4292 | return -EAGAIN; |
4293 | } | |
4294 | ||
2ae523ed PB |
4295 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4296 | struct io_async_msghdr *iomsg) | |
4297 | { | |
2ae523ed | 4298 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4299 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4300 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4301 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4302 | } |
4303 | ||
93642ef8 PB |
4304 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4305 | { | |
4306 | int ret; | |
4307 | ||
4308 | if (!io_op_defs[req->opcode].needs_async_data) | |
4309 | return 0; | |
4310 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4311 | if (!ret) | |
4312 | req->flags |= REQ_F_NEED_CLEANUP; | |
4313 | return ret; | |
4314 | } | |
4315 | ||
3529d8c2 | 4316 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4317 | { |
e47293fd | 4318 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4319 | |
d2b6f48b PB |
4320 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4321 | return -EINVAL; | |
4322 | ||
e47293fd | 4323 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4324 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4325 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4326 | |
d8768362 JA |
4327 | #ifdef CONFIG_COMPAT |
4328 | if (req->ctx->compat) | |
4329 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4330 | #endif | |
93642ef8 | 4331 | return 0; |
03b1230c JA |
4332 | } |
4333 | ||
889fca73 | 4334 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4335 | { |
6b754c8b | 4336 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4337 | struct socket *sock; |
7a7cacba | 4338 | unsigned flags; |
0fa03c62 JA |
4339 | int ret; |
4340 | ||
dba4a925 | 4341 | sock = sock_from_file(req->file); |
7a7cacba | 4342 | if (unlikely(!sock)) |
dba4a925 | 4343 | return -ENOTSOCK; |
3529d8c2 | 4344 | |
257e84a5 PB |
4345 | kmsg = req->async_data; |
4346 | if (!kmsg) { | |
7a7cacba PB |
4347 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4348 | if (ret) | |
4349 | return ret; | |
4350 | kmsg = &iomsg; | |
0fa03c62 | 4351 | } |
0fa03c62 | 4352 | |
7a7cacba PB |
4353 | flags = req->sr_msg.msg_flags; |
4354 | if (flags & MSG_DONTWAIT) | |
4355 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4356 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4357 | flags |= MSG_DONTWAIT; |
e47293fd | 4358 | |
7a7cacba | 4359 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4360 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4361 | return io_setup_async_msg(req, kmsg); |
4362 | if (ret == -ERESTARTSYS) | |
4363 | ret = -EINTR; | |
0fa03c62 | 4364 | |
257e84a5 PB |
4365 | /* fast path, check for non-NULL to avoid function call */ |
4366 | if (kmsg->free_iov) | |
4367 | kfree(kmsg->free_iov); | |
99bc4c38 | 4368 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4369 | if (ret < 0) |
4370 | req_set_fail_links(req); | |
889fca73 | 4371 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4372 | return 0; |
03b1230c | 4373 | } |
aa1fa28f | 4374 | |
889fca73 | 4375 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4376 | { |
7a7cacba PB |
4377 | struct io_sr_msg *sr = &req->sr_msg; |
4378 | struct msghdr msg; | |
4379 | struct iovec iov; | |
fddaface | 4380 | struct socket *sock; |
7a7cacba | 4381 | unsigned flags; |
fddaface JA |
4382 | int ret; |
4383 | ||
dba4a925 | 4384 | sock = sock_from_file(req->file); |
7a7cacba | 4385 | if (unlikely(!sock)) |
dba4a925 | 4386 | return -ENOTSOCK; |
fddaface | 4387 | |
7a7cacba PB |
4388 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4389 | if (unlikely(ret)) | |
14db8411 | 4390 | return ret; |
fddaface | 4391 | |
7a7cacba PB |
4392 | msg.msg_name = NULL; |
4393 | msg.msg_control = NULL; | |
4394 | msg.msg_controllen = 0; | |
4395 | msg.msg_namelen = 0; | |
fddaface | 4396 | |
7a7cacba PB |
4397 | flags = req->sr_msg.msg_flags; |
4398 | if (flags & MSG_DONTWAIT) | |
4399 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4400 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4401 | flags |= MSG_DONTWAIT; |
fddaface | 4402 | |
7a7cacba PB |
4403 | msg.msg_flags = flags; |
4404 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4405 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4406 | return -EAGAIN; |
4407 | if (ret == -ERESTARTSYS) | |
4408 | ret = -EINTR; | |
fddaface | 4409 | |
fddaface JA |
4410 | if (ret < 0) |
4411 | req_set_fail_links(req); | |
889fca73 | 4412 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4413 | return 0; |
fddaface JA |
4414 | } |
4415 | ||
1400e697 PB |
4416 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4417 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4418 | { |
4419 | struct io_sr_msg *sr = &req->sr_msg; | |
4420 | struct iovec __user *uiov; | |
4421 | size_t iov_len; | |
4422 | int ret; | |
4423 | ||
1400e697 PB |
4424 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4425 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4426 | if (ret) |
4427 | return ret; | |
4428 | ||
4429 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4430 | if (iov_len > 1) | |
4431 | return -EINVAL; | |
5476dfed | 4432 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4433 | return -EFAULT; |
5476dfed | 4434 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4435 | iomsg->free_iov = NULL; |
52de1fe1 | 4436 | } else { |
257e84a5 | 4437 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4438 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4439 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4440 | false); |
52de1fe1 JA |
4441 | if (ret > 0) |
4442 | ret = 0; | |
4443 | } | |
4444 | ||
4445 | return ret; | |
4446 | } | |
4447 | ||
4448 | #ifdef CONFIG_COMPAT | |
4449 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4450 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4451 | { |
4452 | struct compat_msghdr __user *msg_compat; | |
4453 | struct io_sr_msg *sr = &req->sr_msg; | |
4454 | struct compat_iovec __user *uiov; | |
4455 | compat_uptr_t ptr; | |
4456 | compat_size_t len; | |
4457 | int ret; | |
4458 | ||
270a5940 | 4459 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4460 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4461 | &ptr, &len); |
4462 | if (ret) | |
4463 | return ret; | |
4464 | ||
4465 | uiov = compat_ptr(ptr); | |
4466 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4467 | compat_ssize_t clen; | |
4468 | ||
4469 | if (len > 1) | |
4470 | return -EINVAL; | |
4471 | if (!access_ok(uiov, sizeof(*uiov))) | |
4472 | return -EFAULT; | |
4473 | if (__get_user(clen, &uiov->iov_len)) | |
4474 | return -EFAULT; | |
4475 | if (clen < 0) | |
4476 | return -EINVAL; | |
2d280bc8 | 4477 | sr->len = clen; |
257e84a5 | 4478 | iomsg->free_iov = NULL; |
52de1fe1 | 4479 | } else { |
257e84a5 | 4480 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4481 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4482 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4483 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4484 | if (ret < 0) |
4485 | return ret; | |
4486 | } | |
4487 | ||
4488 | return 0; | |
4489 | } | |
4490 | #endif | |
4491 | ||
1400e697 PB |
4492 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4493 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4494 | { |
1400e697 | 4495 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4496 | |
4497 | #ifdef CONFIG_COMPAT | |
4498 | if (req->ctx->compat) | |
1400e697 | 4499 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4500 | #endif |
52de1fe1 | 4501 | |
1400e697 | 4502 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4503 | } |
4504 | ||
bcda7baa | 4505 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4506 | bool needs_lock) |
bcda7baa JA |
4507 | { |
4508 | struct io_sr_msg *sr = &req->sr_msg; | |
4509 | struct io_buffer *kbuf; | |
4510 | ||
bcda7baa JA |
4511 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4512 | if (IS_ERR(kbuf)) | |
4513 | return kbuf; | |
4514 | ||
4515 | sr->kbuf = kbuf; | |
4516 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4517 | return kbuf; |
fddaface JA |
4518 | } |
4519 | ||
7fbb1b54 PB |
4520 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4521 | { | |
4522 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4523 | } | |
4524 | ||
93642ef8 | 4525 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4526 | { |
99bc4c38 | 4527 | int ret; |
3529d8c2 | 4528 | |
93642ef8 PB |
4529 | if (!io_op_defs[req->opcode].needs_async_data) |
4530 | return 0; | |
4531 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4532 | if (!ret) | |
4533 | req->flags |= REQ_F_NEED_CLEANUP; | |
4534 | return ret; | |
4535 | } | |
4536 | ||
4537 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4538 | { | |
4539 | struct io_sr_msg *sr = &req->sr_msg; | |
4540 | ||
d2b6f48b PB |
4541 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4542 | return -EINVAL; | |
4543 | ||
3529d8c2 | 4544 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4545 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4546 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4547 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4548 | |
d8768362 JA |
4549 | #ifdef CONFIG_COMPAT |
4550 | if (req->ctx->compat) | |
4551 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4552 | #endif | |
93642ef8 | 4553 | return 0; |
aa1fa28f JA |
4554 | } |
4555 | ||
889fca73 | 4556 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4557 | { |
6b754c8b | 4558 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4559 | struct socket *sock; |
7fbb1b54 | 4560 | struct io_buffer *kbuf; |
7a7cacba | 4561 | unsigned flags; |
52de1fe1 | 4562 | int ret, cflags = 0; |
45d189c6 | 4563 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4564 | |
dba4a925 | 4565 | sock = sock_from_file(req->file); |
7a7cacba | 4566 | if (unlikely(!sock)) |
dba4a925 | 4567 | return -ENOTSOCK; |
3529d8c2 | 4568 | |
257e84a5 PB |
4569 | kmsg = req->async_data; |
4570 | if (!kmsg) { | |
7a7cacba PB |
4571 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4572 | if (ret) | |
681fda8d | 4573 | return ret; |
7a7cacba PB |
4574 | kmsg = &iomsg; |
4575 | } | |
03b1230c | 4576 | |
bc02ef33 | 4577 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4578 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4579 | if (IS_ERR(kbuf)) |
52de1fe1 | 4580 | return PTR_ERR(kbuf); |
7a7cacba | 4581 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4582 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4583 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4584 | 1, req->sr_msg.len); |
4585 | } | |
52de1fe1 | 4586 | |
7a7cacba PB |
4587 | flags = req->sr_msg.msg_flags; |
4588 | if (flags & MSG_DONTWAIT) | |
4589 | req->flags |= REQ_F_NOWAIT; | |
4590 | else if (force_nonblock) | |
4591 | flags |= MSG_DONTWAIT; | |
e47293fd | 4592 | |
7a7cacba PB |
4593 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4594 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4595 | if (force_nonblock && ret == -EAGAIN) |
4596 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4597 | if (ret == -ERESTARTSYS) |
4598 | ret = -EINTR; | |
03b1230c | 4599 | |
7fbb1b54 PB |
4600 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4601 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4602 | /* fast path, check for non-NULL to avoid function call */ |
4603 | if (kmsg->free_iov) | |
4604 | kfree(kmsg->free_iov); | |
99bc4c38 | 4605 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4606 | if (ret < 0) |
4607 | req_set_fail_links(req); | |
889fca73 | 4608 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4609 | return 0; |
0fa03c62 | 4610 | } |
5d17b4a4 | 4611 | |
889fca73 | 4612 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4613 | { |
6b754c8b | 4614 | struct io_buffer *kbuf; |
7a7cacba PB |
4615 | struct io_sr_msg *sr = &req->sr_msg; |
4616 | struct msghdr msg; | |
4617 | void __user *buf = sr->buf; | |
fddaface | 4618 | struct socket *sock; |
7a7cacba PB |
4619 | struct iovec iov; |
4620 | unsigned flags; | |
bcda7baa | 4621 | int ret, cflags = 0; |
45d189c6 | 4622 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4623 | |
dba4a925 | 4624 | sock = sock_from_file(req->file); |
7a7cacba | 4625 | if (unlikely(!sock)) |
dba4a925 | 4626 | return -ENOTSOCK; |
fddaface | 4627 | |
bc02ef33 | 4628 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4629 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4630 | if (IS_ERR(kbuf)) |
4631 | return PTR_ERR(kbuf); | |
7a7cacba | 4632 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4633 | } |
bcda7baa | 4634 | |
7a7cacba | 4635 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4636 | if (unlikely(ret)) |
4637 | goto out_free; | |
fddaface | 4638 | |
7a7cacba PB |
4639 | msg.msg_name = NULL; |
4640 | msg.msg_control = NULL; | |
4641 | msg.msg_controllen = 0; | |
4642 | msg.msg_namelen = 0; | |
4643 | msg.msg_iocb = NULL; | |
4644 | msg.msg_flags = 0; | |
fddaface | 4645 | |
7a7cacba PB |
4646 | flags = req->sr_msg.msg_flags; |
4647 | if (flags & MSG_DONTWAIT) | |
4648 | req->flags |= REQ_F_NOWAIT; | |
4649 | else if (force_nonblock) | |
4650 | flags |= MSG_DONTWAIT; | |
4651 | ||
4652 | ret = sock_recvmsg(sock, &msg, flags); | |
4653 | if (force_nonblock && ret == -EAGAIN) | |
4654 | return -EAGAIN; | |
4655 | if (ret == -ERESTARTSYS) | |
4656 | ret = -EINTR; | |
14c32eee | 4657 | out_free: |
7fbb1b54 PB |
4658 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4659 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4660 | if (ret < 0) |
4661 | req_set_fail_links(req); | |
889fca73 | 4662 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4663 | return 0; |
fddaface JA |
4664 | } |
4665 | ||
3529d8c2 | 4666 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4667 | { |
8ed8d3c3 JA |
4668 | struct io_accept *accept = &req->accept; |
4669 | ||
14587a46 | 4670 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4671 | return -EINVAL; |
8042d6ce | 4672 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4673 | return -EINVAL; |
4674 | ||
d55e5f5b JA |
4675 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4676 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4677 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4678 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4679 | return 0; |
8ed8d3c3 | 4680 | } |
17f2fe35 | 4681 | |
889fca73 | 4682 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4683 | { |
4684 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4685 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4686 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4687 | int ret; |
4688 | ||
e697deed JX |
4689 | if (req->file->f_flags & O_NONBLOCK) |
4690 | req->flags |= REQ_F_NOWAIT; | |
4691 | ||
8ed8d3c3 | 4692 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4693 | accept->addr_len, accept->flags, |
4694 | accept->nofile); | |
8ed8d3c3 | 4695 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4696 | return -EAGAIN; |
ac45abc0 PB |
4697 | if (ret < 0) { |
4698 | if (ret == -ERESTARTSYS) | |
4699 | ret = -EINTR; | |
4e88d6e7 | 4700 | req_set_fail_links(req); |
ac45abc0 | 4701 | } |
889fca73 | 4702 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4703 | return 0; |
8ed8d3c3 JA |
4704 | } |
4705 | ||
93642ef8 PB |
4706 | static int io_connect_prep_async(struct io_kiocb *req) |
4707 | { | |
4708 | struct io_async_connect *io = req->async_data; | |
4709 | struct io_connect *conn = &req->connect; | |
4710 | ||
4711 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4712 | } | |
4713 | ||
3529d8c2 | 4714 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4715 | { |
3529d8c2 | 4716 | struct io_connect *conn = &req->connect; |
f499a021 | 4717 | |
14587a46 | 4718 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4719 | return -EINVAL; |
4720 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4721 | return -EINVAL; | |
4722 | ||
3529d8c2 JA |
4723 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4724 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4725 | return 0; |
f499a021 JA |
4726 | } |
4727 | ||
889fca73 | 4728 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4729 | { |
e8c2bc1f | 4730 | struct io_async_connect __io, *io; |
f8e85cf2 | 4731 | unsigned file_flags; |
3fbb51c1 | 4732 | int ret; |
45d189c6 | 4733 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4734 | |
e8c2bc1f JA |
4735 | if (req->async_data) { |
4736 | io = req->async_data; | |
f499a021 | 4737 | } else { |
3529d8c2 JA |
4738 | ret = move_addr_to_kernel(req->connect.addr, |
4739 | req->connect.addr_len, | |
e8c2bc1f | 4740 | &__io.address); |
f499a021 JA |
4741 | if (ret) |
4742 | goto out; | |
4743 | io = &__io; | |
4744 | } | |
4745 | ||
3fbb51c1 JA |
4746 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4747 | ||
e8c2bc1f | 4748 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4749 | req->connect.addr_len, file_flags); |
87f80d62 | 4750 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4751 | if (req->async_data) |
b7bb4f7d | 4752 | return -EAGAIN; |
e8c2bc1f | 4753 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4754 | ret = -ENOMEM; |
4755 | goto out; | |
4756 | } | |
e8c2bc1f JA |
4757 | io = req->async_data; |
4758 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4759 | return -EAGAIN; |
f499a021 | 4760 | } |
f8e85cf2 JA |
4761 | if (ret == -ERESTARTSYS) |
4762 | ret = -EINTR; | |
f499a021 | 4763 | out: |
4e88d6e7 JA |
4764 | if (ret < 0) |
4765 | req_set_fail_links(req); | |
889fca73 | 4766 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4767 | return 0; |
469956e8 Y |
4768 | } |
4769 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4770 | #define IO_NETOP_FN(op) \ |
4771 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4772 | { \ | |
4773 | return -EOPNOTSUPP; \ | |
4774 | } | |
4775 | ||
4776 | #define IO_NETOP_PREP(op) \ | |
4777 | IO_NETOP_FN(op) \ | |
4778 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4779 | { \ | |
4780 | return -EOPNOTSUPP; \ | |
4781 | } \ | |
4782 | ||
4783 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4784 | IO_NETOP_PREP(op) \ | |
4785 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4786 | { \ | |
4787 | return -EOPNOTSUPP; \ | |
4788 | } | |
4789 | ||
4790 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4791 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4792 | IO_NETOP_PREP_ASYNC(connect); | |
4793 | IO_NETOP_PREP(accept); | |
4794 | IO_NETOP_FN(send); | |
4795 | IO_NETOP_FN(recv); | |
469956e8 | 4796 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4797 | |
d7718a9d JA |
4798 | struct io_poll_table { |
4799 | struct poll_table_struct pt; | |
4800 | struct io_kiocb *req; | |
4801 | int error; | |
4802 | }; | |
ce593a6c | 4803 | |
d7718a9d JA |
4804 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4805 | __poll_t mask, task_work_func_t func) | |
4806 | { | |
aa96bf8a | 4807 | int ret; |
d7718a9d JA |
4808 | |
4809 | /* for instances that support it check for an event match first: */ | |
4810 | if (mask && !(mask & poll->events)) | |
4811 | return 0; | |
4812 | ||
4813 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4814 | ||
4815 | list_del_init(&poll->wait.entry); | |
4816 | ||
d7718a9d | 4817 | req->result = mask; |
7cbf1722 | 4818 | req->task_work.func = func; |
6d816e08 JA |
4819 | percpu_ref_get(&req->ctx->refs); |
4820 | ||
d7718a9d | 4821 | /* |
e3aabf95 JA |
4822 | * If this fails, then the task is exiting. When a task exits, the |
4823 | * work gets canceled, so just cancel this request as well instead | |
4824 | * of executing it. We can't safely execute it anyway, as we may not | |
4825 | * have the needed state needed for it anyway. | |
d7718a9d | 4826 | */ |
355fb9e2 | 4827 | ret = io_req_task_work_add(req); |
aa96bf8a | 4828 | if (unlikely(ret)) { |
e3aabf95 | 4829 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4830 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4831 | } |
d7718a9d JA |
4832 | return 1; |
4833 | } | |
4834 | ||
74ce6ce4 JA |
4835 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4836 | __acquires(&req->ctx->completion_lock) | |
4837 | { | |
4838 | struct io_ring_ctx *ctx = req->ctx; | |
4839 | ||
4840 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4841 | struct poll_table_struct pt = { ._key = poll->events }; | |
4842 | ||
4843 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4844 | } | |
4845 | ||
4846 | spin_lock_irq(&ctx->completion_lock); | |
4847 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4848 | add_wait_queue(poll->head, &poll->wait); | |
4849 | return true; | |
4850 | } | |
4851 | ||
4852 | return false; | |
4853 | } | |
4854 | ||
d4e7cd36 | 4855 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4856 | { |
e8c2bc1f | 4857 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4858 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4859 | return req->async_data; |
d4e7cd36 JA |
4860 | return req->apoll->double_poll; |
4861 | } | |
4862 | ||
4863 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4864 | { | |
4865 | if (req->opcode == IORING_OP_POLL_ADD) | |
4866 | return &req->poll; | |
4867 | return &req->apoll->poll; | |
4868 | } | |
4869 | ||
4870 | static void io_poll_remove_double(struct io_kiocb *req) | |
4871 | { | |
4872 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4873 | |
4874 | lockdep_assert_held(&req->ctx->completion_lock); | |
4875 | ||
4876 | if (poll && poll->head) { | |
4877 | struct wait_queue_head *head = poll->head; | |
4878 | ||
4879 | spin_lock(&head->lock); | |
4880 | list_del_init(&poll->wait.entry); | |
4881 | if (poll->wait.private) | |
4882 | refcount_dec(&req->refs); | |
4883 | poll->head = NULL; | |
4884 | spin_unlock(&head->lock); | |
4885 | } | |
4886 | } | |
4887 | ||
4888 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4889 | { | |
4890 | struct io_ring_ctx *ctx = req->ctx; | |
4891 | ||
d4e7cd36 | 4892 | io_poll_remove_double(req); |
18bceab1 JA |
4893 | req->poll.done = true; |
4894 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4895 | io_commit_cqring(ctx); | |
4896 | } | |
4897 | ||
dd221f46 | 4898 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4899 | { |
dd221f46 | 4900 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4901 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4902 | struct io_kiocb *nxt; |
18bceab1 JA |
4903 | |
4904 | if (io_poll_rewait(req, &req->poll)) { | |
4905 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4906 | } else { |
4907 | hash_del(&req->hash_node); | |
4908 | io_poll_complete(req, req->result, 0); | |
4909 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4910 | |
dd221f46 PB |
4911 | nxt = io_put_req_find_next(req); |
4912 | io_cqring_ev_posted(ctx); | |
4913 | if (nxt) | |
4914 | __io_req_task_submit(nxt); | |
4915 | } | |
18bceab1 | 4916 | |
6d816e08 | 4917 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4918 | } |
4919 | ||
4920 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4921 | int sync, void *key) | |
4922 | { | |
4923 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4924 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4925 | __poll_t mask = key_to_poll(key); |
4926 | ||
4927 | /* for instances that support it check for an event match first: */ | |
4928 | if (mask && !(mask & poll->events)) | |
4929 | return 0; | |
4930 | ||
8706e04e JA |
4931 | list_del_init(&wait->entry); |
4932 | ||
807abcb0 | 4933 | if (poll && poll->head) { |
18bceab1 JA |
4934 | bool done; |
4935 | ||
807abcb0 JA |
4936 | spin_lock(&poll->head->lock); |
4937 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4938 | if (!done) |
807abcb0 | 4939 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4940 | /* make sure double remove sees this as being gone */ |
4941 | wait->private = NULL; | |
807abcb0 | 4942 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4943 | if (!done) { |
4944 | /* use wait func handler, so it matches the rq type */ | |
4945 | poll->wait.func(&poll->wait, mode, sync, key); | |
4946 | } | |
18bceab1 JA |
4947 | } |
4948 | refcount_dec(&req->refs); | |
4949 | return 1; | |
4950 | } | |
4951 | ||
4952 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4953 | wait_queue_func_t wake_func) | |
4954 | { | |
4955 | poll->head = NULL; | |
4956 | poll->done = false; | |
4957 | poll->canceled = false; | |
4958 | poll->events = events; | |
4959 | INIT_LIST_HEAD(&poll->wait.entry); | |
4960 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4961 | } | |
4962 | ||
4963 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4964 | struct wait_queue_head *head, |
4965 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4966 | { |
4967 | struct io_kiocb *req = pt->req; | |
4968 | ||
4969 | /* | |
4970 | * If poll->head is already set, it's because the file being polled | |
4971 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4972 | * for write). Setup a separate io_poll_iocb if this happens. | |
4973 | */ | |
4974 | if (unlikely(poll->head)) { | |
58852d4d PB |
4975 | struct io_poll_iocb *poll_one = poll; |
4976 | ||
18bceab1 | 4977 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4978 | if (*poll_ptr) { |
18bceab1 JA |
4979 | pt->error = -EINVAL; |
4980 | return; | |
4981 | } | |
1c3b3e65 JA |
4982 | /* double add on the same waitqueue head, ignore */ |
4983 | if (poll->head == head) | |
4984 | return; | |
18bceab1 JA |
4985 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4986 | if (!poll) { | |
4987 | pt->error = -ENOMEM; | |
4988 | return; | |
4989 | } | |
58852d4d | 4990 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4991 | refcount_inc(&req->refs); |
4992 | poll->wait.private = req; | |
807abcb0 | 4993 | *poll_ptr = poll; |
18bceab1 JA |
4994 | } |
4995 | ||
4996 | pt->error = 0; | |
4997 | poll->head = head; | |
a31eb4a2 JX |
4998 | |
4999 | if (poll->events & EPOLLEXCLUSIVE) | |
5000 | add_wait_queue_exclusive(head, &poll->wait); | |
5001 | else | |
5002 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5003 | } |
5004 | ||
5005 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5006 | struct poll_table_struct *p) | |
5007 | { | |
5008 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5009 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5010 | |
807abcb0 | 5011 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5012 | } |
5013 | ||
d7718a9d JA |
5014 | static void io_async_task_func(struct callback_head *cb) |
5015 | { | |
5016 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5017 | struct async_poll *apoll = req->apoll; | |
5018 | struct io_ring_ctx *ctx = req->ctx; | |
5019 | ||
5020 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5021 | ||
74ce6ce4 | 5022 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5023 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5024 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5025 | return; |
d7718a9d JA |
5026 | } |
5027 | ||
31067255 | 5028 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5029 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5030 | hash_del(&req->hash_node); |
2bae047e | 5031 | |
d4e7cd36 | 5032 | io_poll_remove_double(req); |
74ce6ce4 JA |
5033 | spin_unlock_irq(&ctx->completion_lock); |
5034 | ||
0be0b0e3 PB |
5035 | if (!READ_ONCE(apoll->poll.canceled)) |
5036 | __io_req_task_submit(req); | |
5037 | else | |
5038 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5039 | |
6d816e08 | 5040 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5041 | kfree(apoll->double_poll); |
31067255 | 5042 | kfree(apoll); |
d7718a9d JA |
5043 | } |
5044 | ||
5045 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5046 | void *key) | |
5047 | { | |
5048 | struct io_kiocb *req = wait->private; | |
5049 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5050 | ||
5051 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5052 | key_to_poll(key)); | |
5053 | ||
5054 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5055 | } | |
5056 | ||
5057 | static void io_poll_req_insert(struct io_kiocb *req) | |
5058 | { | |
5059 | struct io_ring_ctx *ctx = req->ctx; | |
5060 | struct hlist_head *list; | |
5061 | ||
5062 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5063 | hlist_add_head(&req->hash_node, list); | |
5064 | } | |
5065 | ||
5066 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5067 | struct io_poll_iocb *poll, | |
5068 | struct io_poll_table *ipt, __poll_t mask, | |
5069 | wait_queue_func_t wake_func) | |
5070 | __acquires(&ctx->completion_lock) | |
5071 | { | |
5072 | struct io_ring_ctx *ctx = req->ctx; | |
5073 | bool cancel = false; | |
5074 | ||
4d52f338 | 5075 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5076 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5077 | poll->file = req->file; |
18bceab1 | 5078 | poll->wait.private = req; |
d7718a9d JA |
5079 | |
5080 | ipt->pt._key = mask; | |
5081 | ipt->req = req; | |
5082 | ipt->error = -EINVAL; | |
5083 | ||
d7718a9d JA |
5084 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5085 | ||
5086 | spin_lock_irq(&ctx->completion_lock); | |
5087 | if (likely(poll->head)) { | |
5088 | spin_lock(&poll->head->lock); | |
5089 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5090 | if (ipt->error) | |
5091 | cancel = true; | |
5092 | ipt->error = 0; | |
5093 | mask = 0; | |
5094 | } | |
5095 | if (mask || ipt->error) | |
5096 | list_del_init(&poll->wait.entry); | |
5097 | else if (cancel) | |
5098 | WRITE_ONCE(poll->canceled, true); | |
5099 | else if (!poll->done) /* actually waiting for an event */ | |
5100 | io_poll_req_insert(req); | |
5101 | spin_unlock(&poll->head->lock); | |
5102 | } | |
5103 | ||
5104 | return mask; | |
5105 | } | |
5106 | ||
5107 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5108 | { | |
5109 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5110 | struct io_ring_ctx *ctx = req->ctx; | |
5111 | struct async_poll *apoll; | |
5112 | struct io_poll_table ipt; | |
5113 | __poll_t mask, ret; | |
9dab14b8 | 5114 | int rw; |
d7718a9d JA |
5115 | |
5116 | if (!req->file || !file_can_poll(req->file)) | |
5117 | return false; | |
24c74678 | 5118 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5119 | return false; |
9dab14b8 JA |
5120 | if (def->pollin) |
5121 | rw = READ; | |
5122 | else if (def->pollout) | |
5123 | rw = WRITE; | |
5124 | else | |
5125 | return false; | |
5126 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5127 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5128 | return false; |
5129 | ||
5130 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5131 | if (unlikely(!apoll)) | |
5132 | return false; | |
807abcb0 | 5133 | apoll->double_poll = NULL; |
d7718a9d JA |
5134 | |
5135 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5136 | req->apoll = apoll; |
d7718a9d | 5137 | |
8755d97a | 5138 | mask = 0; |
d7718a9d | 5139 | if (def->pollin) |
8755d97a | 5140 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5141 | if (def->pollout) |
5142 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5143 | |
5144 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5145 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5146 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5147 | mask &= ~POLLIN; | |
5148 | ||
d7718a9d JA |
5149 | mask |= POLLERR | POLLPRI; |
5150 | ||
5151 | ipt.pt._qproc = io_async_queue_proc; | |
5152 | ||
5153 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5154 | io_async_wake); | |
a36da65c | 5155 | if (ret || ipt.error) { |
d4e7cd36 | 5156 | io_poll_remove_double(req); |
d7718a9d | 5157 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5158 | kfree(apoll->double_poll); |
d7718a9d JA |
5159 | kfree(apoll); |
5160 | return false; | |
5161 | } | |
5162 | spin_unlock_irq(&ctx->completion_lock); | |
5163 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5164 | apoll->poll.events); | |
5165 | return true; | |
5166 | } | |
5167 | ||
5168 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5169 | struct io_poll_iocb *poll) | |
221c5eb2 | 5170 | { |
b41e9852 | 5171 | bool do_complete = false; |
221c5eb2 JA |
5172 | |
5173 | spin_lock(&poll->head->lock); | |
5174 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5175 | if (!list_empty(&poll->wait.entry)) { |
5176 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5177 | do_complete = true; |
221c5eb2 JA |
5178 | } |
5179 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5180 | hash_del(&req->hash_node); |
d7718a9d JA |
5181 | return do_complete; |
5182 | } | |
5183 | ||
5184 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5185 | { | |
5186 | bool do_complete; | |
5187 | ||
d4e7cd36 JA |
5188 | io_poll_remove_double(req); |
5189 | ||
d7718a9d JA |
5190 | if (req->opcode == IORING_OP_POLL_ADD) { |
5191 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5192 | } else { | |
3bfa5bcb JA |
5193 | struct async_poll *apoll = req->apoll; |
5194 | ||
d7718a9d | 5195 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5196 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5197 | if (do_complete) { | |
d7718a9d | 5198 | io_put_req(req); |
807abcb0 | 5199 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5200 | kfree(apoll); |
5201 | } | |
b1f573bd XW |
5202 | } |
5203 | ||
b41e9852 JA |
5204 | if (do_complete) { |
5205 | io_cqring_fill_event(req, -ECANCELED); | |
5206 | io_commit_cqring(req->ctx); | |
f254ac04 | 5207 | req_set_fail_links(req); |
216578e5 | 5208 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5209 | } |
5210 | ||
5211 | return do_complete; | |
221c5eb2 JA |
5212 | } |
5213 | ||
76e1b642 JA |
5214 | /* |
5215 | * Returns true if we found and killed one or more poll requests | |
5216 | */ | |
6b81928d PB |
5217 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5218 | struct files_struct *files) | |
221c5eb2 | 5219 | { |
78076bb6 | 5220 | struct hlist_node *tmp; |
221c5eb2 | 5221 | struct io_kiocb *req; |
8e2e1faf | 5222 | int posted = 0, i; |
221c5eb2 JA |
5223 | |
5224 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5225 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5226 | struct hlist_head *list; | |
5227 | ||
5228 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5229 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5230 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5231 | posted += io_poll_remove_one(req); |
5232 | } | |
221c5eb2 JA |
5233 | } |
5234 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5235 | |
8e2e1faf JA |
5236 | if (posted) |
5237 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5238 | |
5239 | return posted != 0; | |
221c5eb2 JA |
5240 | } |
5241 | ||
47f46768 JA |
5242 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5243 | { | |
78076bb6 | 5244 | struct hlist_head *list; |
47f46768 JA |
5245 | struct io_kiocb *req; |
5246 | ||
78076bb6 JA |
5247 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5248 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5249 | if (sqe_addr != req->user_data) |
5250 | continue; | |
5251 | if (io_poll_remove_one(req)) | |
eac406c6 | 5252 | return 0; |
b41e9852 | 5253 | return -EALREADY; |
47f46768 JA |
5254 | } |
5255 | ||
5256 | return -ENOENT; | |
5257 | } | |
5258 | ||
3529d8c2 JA |
5259 | static int io_poll_remove_prep(struct io_kiocb *req, |
5260 | const struct io_uring_sqe *sqe) | |
0969e783 | 5261 | { |
0969e783 JA |
5262 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5263 | return -EINVAL; | |
5264 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5265 | sqe->poll_events) | |
5266 | return -EINVAL; | |
5267 | ||
018043be | 5268 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5269 | return 0; |
5270 | } | |
5271 | ||
221c5eb2 JA |
5272 | /* |
5273 | * Find a running poll command that matches one specified in sqe->addr, | |
5274 | * and remove it if found. | |
5275 | */ | |
61e98203 | 5276 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5277 | { |
5278 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5279 | int ret; |
221c5eb2 | 5280 | |
221c5eb2 | 5281 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5282 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5283 | spin_unlock_irq(&ctx->completion_lock); |
5284 | ||
4e88d6e7 JA |
5285 | if (ret < 0) |
5286 | req_set_fail_links(req); | |
e1e16097 | 5287 | io_req_complete(req, ret); |
221c5eb2 JA |
5288 | return 0; |
5289 | } | |
5290 | ||
221c5eb2 JA |
5291 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5292 | void *key) | |
5293 | { | |
c2f2eb7d JA |
5294 | struct io_kiocb *req = wait->private; |
5295 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5296 | |
d7718a9d | 5297 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5298 | } |
5299 | ||
221c5eb2 JA |
5300 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5301 | struct poll_table_struct *p) | |
5302 | { | |
5303 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5304 | ||
e8c2bc1f | 5305 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5306 | } |
5307 | ||
3529d8c2 | 5308 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5309 | { |
5310 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5311 | u32 events; |
221c5eb2 JA |
5312 | |
5313 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5314 | return -EINVAL; | |
5315 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5316 | return -EINVAL; | |
5317 | ||
5769a351 JX |
5318 | events = READ_ONCE(sqe->poll32_events); |
5319 | #ifdef __BIG_ENDIAN | |
5320 | events = swahw32(events); | |
5321 | #endif | |
a31eb4a2 JX |
5322 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5323 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5324 | return 0; |
5325 | } | |
5326 | ||
61e98203 | 5327 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5328 | { |
5329 | struct io_poll_iocb *poll = &req->poll; | |
5330 | struct io_ring_ctx *ctx = req->ctx; | |
5331 | struct io_poll_table ipt; | |
0969e783 | 5332 | __poll_t mask; |
0969e783 | 5333 | |
d7718a9d | 5334 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5335 | |
d7718a9d JA |
5336 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5337 | io_poll_wake); | |
221c5eb2 | 5338 | |
8c838788 | 5339 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5340 | ipt.error = 0; |
b0dd8a41 | 5341 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5342 | } |
221c5eb2 JA |
5343 | spin_unlock_irq(&ctx->completion_lock); |
5344 | ||
8c838788 JA |
5345 | if (mask) { |
5346 | io_cqring_ev_posted(ctx); | |
014db007 | 5347 | io_put_req(req); |
221c5eb2 | 5348 | } |
8c838788 | 5349 | return ipt.error; |
221c5eb2 JA |
5350 | } |
5351 | ||
5262f567 JA |
5352 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5353 | { | |
ad8a48ac JA |
5354 | struct io_timeout_data *data = container_of(timer, |
5355 | struct io_timeout_data, timer); | |
5356 | struct io_kiocb *req = data->req; | |
5357 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5358 | unsigned long flags; |
5359 | ||
5262f567 | 5360 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5361 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5362 | atomic_set(&req->ctx->cq_timeouts, |
5363 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5364 | ||
78e19bbe | 5365 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5366 | io_commit_cqring(ctx); |
5367 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5368 | ||
5369 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5370 | req_set_fail_links(req); |
5262f567 JA |
5371 | io_put_req(req); |
5372 | return HRTIMER_NORESTART; | |
5373 | } | |
5374 | ||
fbd15848 PB |
5375 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5376 | __u64 user_data) | |
f254ac04 | 5377 | { |
fbd15848 | 5378 | struct io_timeout_data *io; |
47f46768 JA |
5379 | struct io_kiocb *req; |
5380 | int ret = -ENOENT; | |
f254ac04 | 5381 | |
135fcde8 | 5382 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5383 | if (user_data == req->user_data) { |
47f46768 JA |
5384 | ret = 0; |
5385 | break; | |
5386 | } | |
5387 | } | |
5388 | ||
5389 | if (ret == -ENOENT) | |
fbd15848 PB |
5390 | return ERR_PTR(ret); |
5391 | ||
5392 | io = req->async_data; | |
e8c2bc1f | 5393 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5394 | if (ret == -1) |
fbd15848 | 5395 | return ERR_PTR(-EALREADY); |
a71976f3 | 5396 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5397 | return req; |
5398 | } | |
47f46768 | 5399 | |
fbd15848 PB |
5400 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5401 | { | |
5402 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5403 | ||
5404 | if (IS_ERR(req)) | |
5405 | return PTR_ERR(req); | |
f254ac04 JA |
5406 | |
5407 | req_set_fail_links(req); | |
f254ac04 | 5408 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5409 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5410 | return 0; |
5411 | } | |
5412 | ||
9c8e11b3 PB |
5413 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5414 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5415 | { |
9c8e11b3 PB |
5416 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5417 | struct io_timeout_data *data; | |
47f46768 | 5418 | |
9c8e11b3 PB |
5419 | if (IS_ERR(req)) |
5420 | return PTR_ERR(req); | |
47f46768 | 5421 | |
9c8e11b3 PB |
5422 | req->timeout.off = 0; /* noseq */ |
5423 | data = req->async_data; | |
5424 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5425 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5426 | data->timer.function = io_timeout_fn; | |
5427 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5428 | return 0; | |
47f46768 JA |
5429 | } |
5430 | ||
3529d8c2 JA |
5431 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5432 | const struct io_uring_sqe *sqe) | |
b29472ee | 5433 | { |
9c8e11b3 PB |
5434 | struct io_timeout_rem *tr = &req->timeout_rem; |
5435 | ||
b29472ee JA |
5436 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5437 | return -EINVAL; | |
61710e43 DA |
5438 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5439 | return -EINVAL; | |
9c8e11b3 | 5440 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5441 | return -EINVAL; |
5442 | ||
9c8e11b3 PB |
5443 | tr->addr = READ_ONCE(sqe->addr); |
5444 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5445 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5446 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5447 | return -EINVAL; | |
5448 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5449 | return -EFAULT; | |
5450 | } else if (tr->flags) { | |
5451 | /* timeout removal doesn't support flags */ | |
b29472ee | 5452 | return -EINVAL; |
9c8e11b3 | 5453 | } |
b29472ee | 5454 | |
b29472ee JA |
5455 | return 0; |
5456 | } | |
5457 | ||
8662daec PB |
5458 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5459 | { | |
5460 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5461 | : HRTIMER_MODE_REL; | |
5462 | } | |
5463 | ||
11365043 JA |
5464 | /* |
5465 | * Remove or update an existing timeout command | |
5466 | */ | |
61e98203 | 5467 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5468 | { |
9c8e11b3 | 5469 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5470 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5471 | int ret; |
11365043 | 5472 | |
11365043 | 5473 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5474 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5475 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5476 | else |
5477 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5478 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5479 | |
47f46768 | 5480 | io_cqring_fill_event(req, ret); |
11365043 JA |
5481 | io_commit_cqring(ctx); |
5482 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5483 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5484 | if (ret < 0) |
5485 | req_set_fail_links(req); | |
ec9c02ad | 5486 | io_put_req(req); |
11365043 | 5487 | return 0; |
5262f567 JA |
5488 | } |
5489 | ||
3529d8c2 | 5490 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5491 | bool is_timeout_link) |
5262f567 | 5492 | { |
ad8a48ac | 5493 | struct io_timeout_data *data; |
a41525ab | 5494 | unsigned flags; |
56080b02 | 5495 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5496 | |
ad8a48ac | 5497 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5498 | return -EINVAL; |
ad8a48ac | 5499 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5500 | return -EINVAL; |
56080b02 | 5501 | if (off && is_timeout_link) |
2d28390a | 5502 | return -EINVAL; |
a41525ab JA |
5503 | flags = READ_ONCE(sqe->timeout_flags); |
5504 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5505 | return -EINVAL; |
bdf20073 | 5506 | |
bfe68a22 | 5507 | req->timeout.off = off; |
26a61679 | 5508 | |
e8c2bc1f | 5509 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5510 | return -ENOMEM; |
5511 | ||
e8c2bc1f | 5512 | data = req->async_data; |
ad8a48ac | 5513 | data->req = req; |
ad8a48ac JA |
5514 | |
5515 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5516 | return -EFAULT; |
5517 | ||
8662daec | 5518 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5519 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5520 | return 0; | |
5521 | } | |
5522 | ||
61e98203 | 5523 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5524 | { |
ad8a48ac | 5525 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5526 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5527 | struct list_head *entry; |
bfe68a22 | 5528 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5529 | |
733f5c95 | 5530 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5531 | |
5262f567 JA |
5532 | /* |
5533 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5534 | * timeout event to be satisfied. If it isn't set, then this is |
5535 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5536 | */ |
8eb7e2d0 | 5537 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5538 | entry = ctx->timeout_list.prev; |
5539 | goto add; | |
5540 | } | |
5262f567 | 5541 | |
bfe68a22 PB |
5542 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5543 | req->timeout.target_seq = tail + off; | |
5262f567 | 5544 | |
f010505b MDG |
5545 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5546 | * This is safe because ->completion_lock is held, and submissions | |
5547 | * and completions are never mixed in the same ->completion_lock section. | |
5548 | */ | |
5549 | ctx->cq_last_tm_flush = tail; | |
5550 | ||
5262f567 JA |
5551 | /* |
5552 | * Insertion sort, ensuring the first entry in the list is always | |
5553 | * the one we need first. | |
5554 | */ | |
5262f567 | 5555 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5556 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5557 | timeout.list); | |
5262f567 | 5558 | |
8eb7e2d0 | 5559 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5560 | continue; |
bfe68a22 PB |
5561 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5562 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5563 | break; |
5564 | } | |
93bd25bb | 5565 | add: |
135fcde8 | 5566 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5567 | data->timer.function = io_timeout_fn; |
5568 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5569 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5570 | return 0; |
5571 | } | |
5262f567 | 5572 | |
62755e35 JA |
5573 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5574 | { | |
5575 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5576 | ||
5577 | return req->user_data == (unsigned long) data; | |
5578 | } | |
5579 | ||
5aa75ed5 | 5580 | static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr) |
62755e35 | 5581 | { |
62755e35 | 5582 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5583 | int ret = 0; |
5584 | ||
5aa75ed5 JA |
5585 | if (!tctx->io_wq) |
5586 | return -ENOENT; | |
5587 | ||
5588 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false); | |
62755e35 JA |
5589 | switch (cancel_ret) { |
5590 | case IO_WQ_CANCEL_OK: | |
5591 | ret = 0; | |
5592 | break; | |
5593 | case IO_WQ_CANCEL_RUNNING: | |
5594 | ret = -EALREADY; | |
5595 | break; | |
5596 | case IO_WQ_CANCEL_NOTFOUND: | |
5597 | ret = -ENOENT; | |
5598 | break; | |
5599 | } | |
5600 | ||
e977d6d3 JA |
5601 | return ret; |
5602 | } | |
5603 | ||
47f46768 JA |
5604 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5605 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5606 | int success_ret) |
47f46768 JA |
5607 | { |
5608 | unsigned long flags; | |
5609 | int ret; | |
5610 | ||
5aa75ed5 JA |
5611 | ret = io_async_cancel_one(req->task->io_uring, |
5612 | (void *) (unsigned long) sqe_addr); | |
47f46768 JA |
5613 | if (ret != -ENOENT) { |
5614 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5615 | goto done; | |
5616 | } | |
5617 | ||
5618 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5619 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5620 | if (ret != -ENOENT) | |
5621 | goto done; | |
5622 | ret = io_poll_cancel(ctx, sqe_addr); | |
5623 | done: | |
b0dd8a41 JA |
5624 | if (!ret) |
5625 | ret = success_ret; | |
47f46768 JA |
5626 | io_cqring_fill_event(req, ret); |
5627 | io_commit_cqring(ctx); | |
5628 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5629 | io_cqring_ev_posted(ctx); | |
5630 | ||
4e88d6e7 JA |
5631 | if (ret < 0) |
5632 | req_set_fail_links(req); | |
014db007 | 5633 | io_put_req(req); |
47f46768 JA |
5634 | } |
5635 | ||
3529d8c2 JA |
5636 | static int io_async_cancel_prep(struct io_kiocb *req, |
5637 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5638 | { |
fbf23849 | 5639 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5640 | return -EINVAL; |
61710e43 DA |
5641 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5642 | return -EINVAL; | |
5643 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5644 | return -EINVAL; |
5645 | ||
fbf23849 JA |
5646 | req->cancel.addr = READ_ONCE(sqe->addr); |
5647 | return 0; | |
5648 | } | |
5649 | ||
61e98203 | 5650 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5651 | { |
5652 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5653 | |
014db007 | 5654 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5655 | return 0; |
5656 | } | |
5657 | ||
269bbe5f | 5658 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5659 | const struct io_uring_sqe *sqe) |
5660 | { | |
6ca56f84 JA |
5661 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5662 | return -EINVAL; | |
61710e43 DA |
5663 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5664 | return -EINVAL; | |
5665 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5666 | return -EINVAL; |
5667 | ||
269bbe5f BM |
5668 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5669 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5670 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5671 | return -EINVAL; |
269bbe5f | 5672 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5673 | return 0; |
5674 | } | |
5675 | ||
889fca73 | 5676 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5677 | { |
5678 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5679 | struct io_uring_rsrc_update up; |
05f3fb3c | 5680 | int ret; |
fbf23849 | 5681 | |
45d189c6 | 5682 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5683 | return -EAGAIN; |
05f3fb3c | 5684 | |
269bbe5f BM |
5685 | up.offset = req->rsrc_update.offset; |
5686 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5687 | |
5688 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5689 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5690 | mutex_unlock(&ctx->uring_lock); |
5691 | ||
5692 | if (ret < 0) | |
5693 | req_set_fail_links(req); | |
889fca73 | 5694 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5695 | return 0; |
5696 | } | |
5697 | ||
bfe76559 | 5698 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5699 | { |
d625c6ee | 5700 | switch (req->opcode) { |
e781573e | 5701 | case IORING_OP_NOP: |
bfe76559 | 5702 | return 0; |
f67676d1 JA |
5703 | case IORING_OP_READV: |
5704 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5705 | case IORING_OP_READ: |
bfe76559 | 5706 | return io_read_prep(req, sqe); |
f67676d1 JA |
5707 | case IORING_OP_WRITEV: |
5708 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5709 | case IORING_OP_WRITE: |
bfe76559 | 5710 | return io_write_prep(req, sqe); |
0969e783 | 5711 | case IORING_OP_POLL_ADD: |
bfe76559 | 5712 | return io_poll_add_prep(req, sqe); |
0969e783 | 5713 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5714 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5715 | case IORING_OP_FSYNC: |
1155c76a | 5716 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5717 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5718 | return io_sfr_prep(req, sqe); |
03b1230c | 5719 | case IORING_OP_SENDMSG: |
fddaface | 5720 | case IORING_OP_SEND: |
bfe76559 | 5721 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5722 | case IORING_OP_RECVMSG: |
fddaface | 5723 | case IORING_OP_RECV: |
bfe76559 | 5724 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5725 | case IORING_OP_CONNECT: |
bfe76559 | 5726 | return io_connect_prep(req, sqe); |
2d28390a | 5727 | case IORING_OP_TIMEOUT: |
bfe76559 | 5728 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5729 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5730 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5731 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5732 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5733 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5734 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5735 | case IORING_OP_ACCEPT: |
bfe76559 | 5736 | return io_accept_prep(req, sqe); |
d63d1b5e | 5737 | case IORING_OP_FALLOCATE: |
bfe76559 | 5738 | return io_fallocate_prep(req, sqe); |
15b71abe | 5739 | case IORING_OP_OPENAT: |
bfe76559 | 5740 | return io_openat_prep(req, sqe); |
b5dba59e | 5741 | case IORING_OP_CLOSE: |
bfe76559 | 5742 | return io_close_prep(req, sqe); |
05f3fb3c | 5743 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5744 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5745 | case IORING_OP_STATX: |
bfe76559 | 5746 | return io_statx_prep(req, sqe); |
4840e418 | 5747 | case IORING_OP_FADVISE: |
bfe76559 | 5748 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5749 | case IORING_OP_MADVISE: |
bfe76559 | 5750 | return io_madvise_prep(req, sqe); |
cebdb986 | 5751 | case IORING_OP_OPENAT2: |
bfe76559 | 5752 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5753 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5754 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5755 | case IORING_OP_SPLICE: |
bfe76559 | 5756 | return io_splice_prep(req, sqe); |
ddf0322d | 5757 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5758 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5759 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5760 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5761 | case IORING_OP_TEE: |
bfe76559 | 5762 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5763 | case IORING_OP_SHUTDOWN: |
5764 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5765 | case IORING_OP_RENAMEAT: |
5766 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5767 | case IORING_OP_UNLINKAT: |
5768 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5769 | } |
5770 | ||
bfe76559 PB |
5771 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5772 | req->opcode); | |
5773 | return-EINVAL; | |
5774 | } | |
5775 | ||
93642ef8 | 5776 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5777 | { |
93642ef8 PB |
5778 | switch (req->opcode) { |
5779 | case IORING_OP_READV: | |
5780 | case IORING_OP_READ_FIXED: | |
5781 | case IORING_OP_READ: | |
5782 | return io_rw_prep_async(req, READ); | |
5783 | case IORING_OP_WRITEV: | |
5784 | case IORING_OP_WRITE_FIXED: | |
5785 | case IORING_OP_WRITE: | |
5786 | return io_rw_prep_async(req, WRITE); | |
5787 | case IORING_OP_SENDMSG: | |
5788 | case IORING_OP_SEND: | |
5789 | return io_sendmsg_prep_async(req); | |
5790 | case IORING_OP_RECVMSG: | |
5791 | case IORING_OP_RECV: | |
5792 | return io_recvmsg_prep_async(req); | |
5793 | case IORING_OP_CONNECT: | |
5794 | return io_connect_prep_async(req); | |
5795 | } | |
5796 | return 0; | |
5797 | } | |
5798 | ||
be7053b7 | 5799 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5800 | { |
be7053b7 | 5801 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5802 | return 0; |
be7053b7 | 5803 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5804 | if (req->async_data) |
be7053b7 PB |
5805 | return 0; |
5806 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5807 | return -EAGAIN; |
be7053b7 | 5808 | return io_req_prep_async(req); |
f67676d1 JA |
5809 | } |
5810 | ||
9cf7c104 PB |
5811 | static u32 io_get_sequence(struct io_kiocb *req) |
5812 | { | |
5813 | struct io_kiocb *pos; | |
5814 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5815 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5816 | |
f2f87370 PB |
5817 | io_for_each_link(pos, req) |
5818 | nr_reqs++; | |
9cf7c104 PB |
5819 | |
5820 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5821 | return total_submitted - nr_reqs; | |
5822 | } | |
5823 | ||
be7053b7 | 5824 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5825 | { |
a197f664 | 5826 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5827 | struct io_defer_entry *de; |
f67676d1 | 5828 | int ret; |
9cf7c104 | 5829 | u32 seq; |
de0617e4 | 5830 | |
9d858b21 | 5831 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5832 | if (likely(list_empty_careful(&ctx->defer_list) && |
5833 | !(req->flags & REQ_F_IO_DRAIN))) | |
5834 | return 0; | |
5835 | ||
5836 | seq = io_get_sequence(req); | |
5837 | /* Still a chance to pass the sequence check */ | |
5838 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5839 | return 0; |
5840 | ||
be7053b7 PB |
5841 | ret = io_req_defer_prep(req); |
5842 | if (ret) | |
5843 | return ret; | |
cbdcb435 | 5844 | io_prep_async_link(req); |
27dc8338 PB |
5845 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5846 | if (!de) | |
5847 | return -ENOMEM; | |
2d28390a | 5848 | |
de0617e4 | 5849 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5850 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5851 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5852 | kfree(de); |
ae34817b PB |
5853 | io_queue_async_work(req); |
5854 | return -EIOCBQUEUED; | |
de0617e4 JA |
5855 | } |
5856 | ||
915967f6 | 5857 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5858 | de->req = req; |
9cf7c104 | 5859 | de->seq = seq; |
27dc8338 | 5860 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5861 | spin_unlock_irq(&ctx->completion_lock); |
5862 | return -EIOCBQUEUED; | |
5863 | } | |
5864 | ||
3ca405eb | 5865 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5866 | { |
0e1b6fe3 PB |
5867 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5868 | switch (req->opcode) { | |
5869 | case IORING_OP_READV: | |
5870 | case IORING_OP_READ_FIXED: | |
5871 | case IORING_OP_READ: | |
bcda7baa | 5872 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5873 | break; |
5874 | case IORING_OP_RECVMSG: | |
5875 | case IORING_OP_RECV: | |
bcda7baa | 5876 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5877 | break; |
5878 | } | |
5879 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5880 | } |
5881 | ||
0e1b6fe3 PB |
5882 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5883 | switch (req->opcode) { | |
5884 | case IORING_OP_READV: | |
5885 | case IORING_OP_READ_FIXED: | |
5886 | case IORING_OP_READ: | |
5887 | case IORING_OP_WRITEV: | |
5888 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5889 | case IORING_OP_WRITE: { |
5890 | struct io_async_rw *io = req->async_data; | |
5891 | if (io->free_iovec) | |
5892 | kfree(io->free_iovec); | |
0e1b6fe3 | 5893 | break; |
e8c2bc1f | 5894 | } |
0e1b6fe3 | 5895 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5896 | case IORING_OP_SENDMSG: { |
5897 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5898 | |
5899 | kfree(io->free_iov); | |
0e1b6fe3 | 5900 | break; |
e8c2bc1f | 5901 | } |
0e1b6fe3 PB |
5902 | case IORING_OP_SPLICE: |
5903 | case IORING_OP_TEE: | |
5904 | io_put_file(req, req->splice.file_in, | |
5905 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5906 | break; | |
f3cd4850 JA |
5907 | case IORING_OP_OPENAT: |
5908 | case IORING_OP_OPENAT2: | |
5909 | if (req->open.filename) | |
5910 | putname(req->open.filename); | |
5911 | break; | |
80a261fd JA |
5912 | case IORING_OP_RENAMEAT: |
5913 | putname(req->rename.oldpath); | |
5914 | putname(req->rename.newpath); | |
5915 | break; | |
14a1143b JA |
5916 | case IORING_OP_UNLINKAT: |
5917 | putname(req->unlink.filename); | |
5918 | break; | |
0e1b6fe3 PB |
5919 | } |
5920 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5921 | } |
99bc4c38 PB |
5922 | } |
5923 | ||
889fca73 | 5924 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5925 | { |
a197f664 | 5926 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5927 | const struct cred *creds = NULL; |
d625c6ee | 5928 | int ret; |
2b188cc1 | 5929 | |
5730b27e JA |
5930 | if (req->work.personality) { |
5931 | const struct cred *new_creds; | |
5932 | ||
5933 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5934 | mutex_lock(&ctx->uring_lock); | |
5935 | new_creds = idr_find(&ctx->personality_idr, req->work.personality); | |
5936 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5937 | mutex_unlock(&ctx->uring_lock); | |
5938 | if (!new_creds) | |
5939 | return -EINVAL; | |
5940 | creds = override_creds(new_creds); | |
5941 | } | |
5942 | ||
d625c6ee | 5943 | switch (req->opcode) { |
2b188cc1 | 5944 | case IORING_OP_NOP: |
889fca73 | 5945 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5946 | break; |
5947 | case IORING_OP_READV: | |
edafccee | 5948 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5949 | case IORING_OP_READ: |
889fca73 | 5950 | ret = io_read(req, issue_flags); |
edafccee | 5951 | break; |
3529d8c2 | 5952 | case IORING_OP_WRITEV: |
edafccee | 5953 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5954 | case IORING_OP_WRITE: |
889fca73 | 5955 | ret = io_write(req, issue_flags); |
2b188cc1 | 5956 | break; |
c992fe29 | 5957 | case IORING_OP_FSYNC: |
45d189c6 | 5958 | ret = io_fsync(req, issue_flags); |
c992fe29 | 5959 | break; |
221c5eb2 | 5960 | case IORING_OP_POLL_ADD: |
61e98203 | 5961 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
5962 | break; |
5963 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 5964 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 5965 | break; |
5d17b4a4 | 5966 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 5967 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 5968 | break; |
0fa03c62 | 5969 | case IORING_OP_SENDMSG: |
889fca73 | 5970 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 5971 | break; |
fddaface | 5972 | case IORING_OP_SEND: |
889fca73 | 5973 | ret = io_send(req, issue_flags); |
0fa03c62 | 5974 | break; |
aa1fa28f | 5975 | case IORING_OP_RECVMSG: |
889fca73 | 5976 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 5977 | break; |
fddaface | 5978 | case IORING_OP_RECV: |
889fca73 | 5979 | ret = io_recv(req, issue_flags); |
aa1fa28f | 5980 | break; |
5262f567 | 5981 | case IORING_OP_TIMEOUT: |
61e98203 | 5982 | ret = io_timeout(req, issue_flags); |
5262f567 | 5983 | break; |
11365043 | 5984 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 5985 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 5986 | break; |
17f2fe35 | 5987 | case IORING_OP_ACCEPT: |
889fca73 | 5988 | ret = io_accept(req, issue_flags); |
17f2fe35 | 5989 | break; |
f8e85cf2 | 5990 | case IORING_OP_CONNECT: |
889fca73 | 5991 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 5992 | break; |
62755e35 | 5993 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 5994 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 5995 | break; |
d63d1b5e | 5996 | case IORING_OP_FALLOCATE: |
45d189c6 | 5997 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 5998 | break; |
15b71abe | 5999 | case IORING_OP_OPENAT: |
45d189c6 | 6000 | ret = io_openat(req, issue_flags); |
15b71abe | 6001 | break; |
b5dba59e | 6002 | case IORING_OP_CLOSE: |
889fca73 | 6003 | ret = io_close(req, issue_flags); |
b5dba59e | 6004 | break; |
05f3fb3c | 6005 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6006 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6007 | break; |
eddc7ef5 | 6008 | case IORING_OP_STATX: |
45d189c6 | 6009 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6010 | break; |
4840e418 | 6011 | case IORING_OP_FADVISE: |
45d189c6 | 6012 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6013 | break; |
c1ca757b | 6014 | case IORING_OP_MADVISE: |
45d189c6 | 6015 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6016 | break; |
cebdb986 | 6017 | case IORING_OP_OPENAT2: |
45d189c6 | 6018 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6019 | break; |
3e4827b0 | 6020 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6021 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6022 | break; |
7d67af2c | 6023 | case IORING_OP_SPLICE: |
45d189c6 | 6024 | ret = io_splice(req, issue_flags); |
7d67af2c | 6025 | break; |
ddf0322d | 6026 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6027 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6028 | break; |
067524e9 | 6029 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6030 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6031 | break; |
f2a8d5c7 | 6032 | case IORING_OP_TEE: |
45d189c6 | 6033 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6034 | break; |
36f4fa68 | 6035 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6036 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6037 | break; |
80a261fd | 6038 | case IORING_OP_RENAMEAT: |
45d189c6 | 6039 | ret = io_renameat(req, issue_flags); |
80a261fd | 6040 | break; |
14a1143b | 6041 | case IORING_OP_UNLINKAT: |
45d189c6 | 6042 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6043 | break; |
2b188cc1 JA |
6044 | default: |
6045 | ret = -EINVAL; | |
6046 | break; | |
6047 | } | |
6048 | ||
5730b27e JA |
6049 | if (creds) |
6050 | revert_creds(creds); | |
6051 | ||
def596e9 JA |
6052 | if (ret) |
6053 | return ret; | |
6054 | ||
b532576e JA |
6055 | /* If the op doesn't have a file, we're not polling for it */ |
6056 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6057 | const bool in_async = io_wq_current_is_worker(); |
6058 | ||
11ba820b JA |
6059 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6060 | if (in_async) | |
6061 | mutex_lock(&ctx->uring_lock); | |
6062 | ||
2e9dbe90 | 6063 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6064 | |
6065 | if (in_async) | |
6066 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6067 | } |
6068 | ||
6069 | return 0; | |
2b188cc1 JA |
6070 | } |
6071 | ||
5280f7e5 | 6072 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6073 | { |
6074 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6075 | struct io_kiocb *timeout; |
561fb04a | 6076 | int ret = 0; |
2b188cc1 | 6077 | |
6df1db6b PB |
6078 | timeout = io_prep_linked_timeout(req); |
6079 | if (timeout) | |
6080 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6081 | |
4014d943 | 6082 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6083 | ret = -ECANCELED; |
31b51510 | 6084 | |
561fb04a | 6085 | if (!ret) { |
561fb04a | 6086 | do { |
889fca73 | 6087 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6088 | /* |
6089 | * We can get EAGAIN for polled IO even though we're | |
6090 | * forcing a sync submission from here, since we can't | |
6091 | * wait for request slots on the block side. | |
6092 | */ | |
6093 | if (ret != -EAGAIN) | |
6094 | break; | |
6095 | cond_resched(); | |
6096 | } while (1); | |
6097 | } | |
31b51510 | 6098 | |
a3df7698 | 6099 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6100 | if (ret) { |
a3df7698 PB |
6101 | /* io-wq is going to take one down */ |
6102 | refcount_inc(&req->refs); | |
6103 | io_req_task_queue_fail(req, ret); | |
edafccee | 6104 | } |
2b188cc1 JA |
6105 | } |
6106 | ||
65e19f54 JA |
6107 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6108 | int index) | |
6109 | { | |
269bbe5f | 6110 | struct fixed_rsrc_table *table; |
65e19f54 | 6111 | |
05f3fb3c | 6112 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6113 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6114 | } |
6115 | ||
8371adf5 PB |
6116 | static struct file *io_file_get(struct io_submit_state *state, |
6117 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6118 | { |
a197f664 | 6119 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6120 | struct file *file; |
09bb8394 | 6121 | |
8da11c19 | 6122 | if (fixed) { |
479f517b | 6123 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6124 | return NULL; |
b7620121 | 6125 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6126 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6127 | io_set_resource_node(req); |
09bb8394 | 6128 | } else { |
c826bd7a | 6129 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6130 | file = __io_file_get(state, fd); |
09bb8394 JA |
6131 | } |
6132 | ||
ce3d5aae PB |
6133 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6134 | io_req_track_inflight(req); | |
8371adf5 | 6135 | return file; |
09bb8394 JA |
6136 | } |
6137 | ||
2665abfd | 6138 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6139 | { |
ad8a48ac JA |
6140 | struct io_timeout_data *data = container_of(timer, |
6141 | struct io_timeout_data, timer); | |
90cd7e42 | 6142 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6143 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6144 | unsigned long flags; |
2665abfd JA |
6145 | |
6146 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6147 | prev = req->timeout.head; |
6148 | req->timeout.head = NULL; | |
2665abfd JA |
6149 | |
6150 | /* | |
6151 | * We don't expect the list to be empty, that will only happen if we | |
6152 | * race with the completion of the linked work. | |
6153 | */ | |
90cd7e42 | 6154 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6155 | io_remove_next_linked(prev); |
90cd7e42 PB |
6156 | else |
6157 | prev = NULL; | |
2665abfd JA |
6158 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6159 | ||
6160 | if (prev) { | |
4e88d6e7 | 6161 | req_set_fail_links(prev); |
014db007 | 6162 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6163 | io_put_req_deferred(prev, 1); |
47f46768 | 6164 | } else { |
9ae1f8dd PB |
6165 | io_req_complete_post(req, -ETIME, 0); |
6166 | io_put_req_deferred(req, 1); | |
2665abfd | 6167 | } |
2665abfd JA |
6168 | return HRTIMER_NORESTART; |
6169 | } | |
6170 | ||
7271ef3a | 6171 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6172 | { |
76a46e06 | 6173 | /* |
f2f87370 PB |
6174 | * If the back reference is NULL, then our linked request finished |
6175 | * before we got a chance to setup the timer | |
76a46e06 | 6176 | */ |
90cd7e42 | 6177 | if (req->timeout.head) { |
e8c2bc1f | 6178 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6179 | |
ad8a48ac JA |
6180 | data->timer.function = io_link_timeout_fn; |
6181 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6182 | data->mode); | |
2665abfd | 6183 | } |
7271ef3a JA |
6184 | } |
6185 | ||
6186 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6187 | { | |
6188 | struct io_ring_ctx *ctx = req->ctx; | |
6189 | ||
6190 | spin_lock_irq(&ctx->completion_lock); | |
6191 | __io_queue_linked_timeout(req); | |
76a46e06 | 6192 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6193 | |
2665abfd | 6194 | /* drop submission reference */ |
76a46e06 JA |
6195 | io_put_req(req); |
6196 | } | |
2665abfd | 6197 | |
ad8a48ac | 6198 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6199 | { |
f2f87370 | 6200 | struct io_kiocb *nxt = req->link; |
2665abfd | 6201 | |
f2f87370 PB |
6202 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6203 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6204 | return NULL; |
2665abfd | 6205 | |
90cd7e42 | 6206 | nxt->timeout.head = req; |
900fad45 | 6207 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6208 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6209 | return nxt; |
2665abfd JA |
6210 | } |
6211 | ||
c5eef2b9 | 6212 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6213 | { |
d3d7298d | 6214 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6215 | int ret; |
2b188cc1 | 6216 | |
c5eef2b9 | 6217 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6218 | |
491381ce JA |
6219 | /* |
6220 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6221 | * doesn't support non-blocking read/write attempts | |
6222 | */ | |
24c74678 | 6223 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6224 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6225 | /* |
6226 | * Queued up for async execution, worker will release | |
6227 | * submit reference when the iocb is actually submitted. | |
6228 | */ | |
6229 | io_queue_async_work(req); | |
2b188cc1 | 6230 | } |
0d63c148 PB |
6231 | } else if (likely(!ret)) { |
6232 | /* drop submission reference */ | |
e342c807 | 6233 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6234 | struct io_ring_ctx *ctx = req->ctx; |
6235 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6236 | |
6dd0be1e | 6237 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6238 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6239 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6240 | } else { |
d3d7298d | 6241 | io_put_req(req); |
0d63c148 PB |
6242 | } |
6243 | } else { | |
4e88d6e7 | 6244 | req_set_fail_links(req); |
e65ef56d | 6245 | io_put_req(req); |
e1e16097 | 6246 | io_req_complete(req, ret); |
9e645e11 | 6247 | } |
d3d7298d PB |
6248 | if (linked_timeout) |
6249 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6250 | } |
6251 | ||
be7053b7 | 6252 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6253 | { |
6254 | int ret; | |
6255 | ||
be7053b7 | 6256 | ret = io_req_defer(req); |
4fe2c963 JL |
6257 | if (ret) { |
6258 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6259 | fail_req: |
4e88d6e7 | 6260 | req_set_fail_links(req); |
e1e16097 JA |
6261 | io_put_req(req); |
6262 | io_req_complete(req, ret); | |
4fe2c963 | 6263 | } |
2550878f | 6264 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6265 | ret = io_req_defer_prep(req); |
6266 | if (unlikely(ret)) | |
6267 | goto fail_req; | |
ce35a47a JA |
6268 | io_queue_async_work(req); |
6269 | } else { | |
c5eef2b9 | 6270 | __io_queue_sqe(req); |
ce35a47a | 6271 | } |
4fe2c963 JL |
6272 | } |
6273 | ||
b16fed66 PB |
6274 | /* |
6275 | * Check SQE restrictions (opcode and flags). | |
6276 | * | |
6277 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6278 | */ | |
6279 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6280 | struct io_kiocb *req, | |
6281 | unsigned int sqe_flags) | |
4fe2c963 | 6282 | { |
b16fed66 PB |
6283 | if (!ctx->restricted) |
6284 | return true; | |
6285 | ||
6286 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6287 | return false; | |
6288 | ||
6289 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6290 | ctx->restrictions.sqe_flags_required) | |
6291 | return false; | |
6292 | ||
6293 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6294 | ctx->restrictions.sqe_flags_required)) | |
6295 | return false; | |
6296 | ||
6297 | return true; | |
4fe2c963 JL |
6298 | } |
6299 | ||
b16fed66 PB |
6300 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6301 | const struct io_uring_sqe *sqe) | |
6302 | { | |
6303 | struct io_submit_state *state; | |
6304 | unsigned int sqe_flags; | |
5730b27e | 6305 | int ret = 0; |
b16fed66 PB |
6306 | |
6307 | req->opcode = READ_ONCE(sqe->opcode); | |
6308 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6309 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6310 | req->user_data = READ_ONCE(sqe->user_data); | |
6311 | req->async_data = NULL; | |
6312 | req->file = NULL; | |
6313 | req->ctx = ctx; | |
6314 | req->link = NULL; | |
6315 | req->fixed_rsrc_refs = NULL; | |
6316 | /* one is dropped after submission, the other at completion */ | |
6317 | refcount_set(&req->refs, 2); | |
6318 | req->task = current; | |
6319 | req->result = 0; | |
6320 | ||
6321 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6322 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6323 | req->flags = 0; | |
b16fed66 | 6324 | return -EINVAL; |
ebf4a5db | 6325 | } |
b16fed66 PB |
6326 | |
6327 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6328 | return -EINVAL; | |
6329 | ||
b16fed66 PB |
6330 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6331 | return -EACCES; | |
6332 | ||
6333 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6334 | !io_op_defs[req->opcode].buffer_select) | |
6335 | return -EOPNOTSUPP; | |
863e0560 | 6336 | |
5730b27e JA |
6337 | req->work.list.next = NULL; |
6338 | req->work.flags = 0; | |
6339 | req->work.personality = READ_ONCE(sqe->personality); | |
b16fed66 PB |
6340 | state = &ctx->submit_state; |
6341 | ||
6342 | /* | |
6343 | * Plug now if we have more than 1 IO left after this, and the target | |
6344 | * is potentially a read/write to block based storage. | |
6345 | */ | |
6346 | if (!state->plug_started && state->ios_left > 1 && | |
6347 | io_op_defs[req->opcode].plug) { | |
6348 | blk_start_plug(&state->plug); | |
6349 | state->plug_started = true; | |
6350 | } | |
6351 | ||
6352 | if (io_op_defs[req->opcode].needs_file) { | |
6353 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6354 | ||
6355 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6356 | if (unlikely(!req->file)) | |
6357 | ret = -EBADF; | |
6358 | } | |
6359 | ||
6360 | state->ios_left--; | |
6361 | return ret; | |
6362 | } | |
6363 | ||
a6b8cadc | 6364 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6365 | const struct io_uring_sqe *sqe) |
9e645e11 | 6366 | { |
a1ab7b35 | 6367 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6368 | int ret; |
9e645e11 | 6369 | |
a6b8cadc PB |
6370 | ret = io_init_req(ctx, req, sqe); |
6371 | if (unlikely(ret)) { | |
6372 | fail_req: | |
6373 | io_put_req(req); | |
6374 | io_req_complete(req, ret); | |
de59bc10 PB |
6375 | if (link->head) { |
6376 | /* fail even hard links since we don't submit */ | |
cf109604 | 6377 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6378 | io_put_req(link->head); |
6379 | io_req_complete(link->head, -ECANCELED); | |
6380 | link->head = NULL; | |
6381 | } | |
a6b8cadc PB |
6382 | return ret; |
6383 | } | |
be7053b7 PB |
6384 | ret = io_req_prep(req, sqe); |
6385 | if (unlikely(ret)) | |
6386 | goto fail_req; | |
a6b8cadc | 6387 | |
be7053b7 | 6388 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6389 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6390 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6391 | ||
9e645e11 JA |
6392 | /* |
6393 | * If we already have a head request, queue this one for async | |
6394 | * submittal once the head completes. If we don't have a head but | |
6395 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6396 | * submitted sync once the chain is complete. If none of those | |
6397 | * conditions are true (normal request), then just queue it. | |
6398 | */ | |
863e0560 PB |
6399 | if (link->head) { |
6400 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6401 | |
8cdf2193 PB |
6402 | /* |
6403 | * Taking sequential execution of a link, draining both sides | |
6404 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6405 | * requests in the link. So, it drains the head and the | |
6406 | * next after the link request. The last one is done via | |
6407 | * drain_next flag to persist the effect across calls. | |
6408 | */ | |
ef4ff581 | 6409 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6410 | head->flags |= REQ_F_IO_DRAIN; |
6411 | ctx->drain_next = 1; | |
6412 | } | |
be7053b7 | 6413 | ret = io_req_defer_prep(req); |
cf109604 | 6414 | if (unlikely(ret)) |
a6b8cadc | 6415 | goto fail_req; |
9d76377f | 6416 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6417 | link->last->link = req; |
863e0560 | 6418 | link->last = req; |
32fe525b PB |
6419 | |
6420 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6421 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6422 | io_queue_sqe(head); |
863e0560 | 6423 | link->head = NULL; |
32fe525b | 6424 | } |
9e645e11 | 6425 | } else { |
711be031 PB |
6426 | if (unlikely(ctx->drain_next)) { |
6427 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6428 | ctx->drain_next = 0; |
711be031 | 6429 | } |
ef4ff581 | 6430 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6431 | link->head = req; |
6432 | link->last = req; | |
711be031 | 6433 | } else { |
be7053b7 | 6434 | io_queue_sqe(req); |
711be031 | 6435 | } |
9e645e11 | 6436 | } |
2e6e1fde | 6437 | |
1d4240cc | 6438 | return 0; |
9e645e11 JA |
6439 | } |
6440 | ||
9a56a232 JA |
6441 | /* |
6442 | * Batched submission is done, ensure local IO is flushed out. | |
6443 | */ | |
ba88ff11 PB |
6444 | static void io_submit_state_end(struct io_submit_state *state, |
6445 | struct io_ring_ctx *ctx) | |
9a56a232 | 6446 | { |
a1ab7b35 | 6447 | if (state->link.head) |
de59bc10 | 6448 | io_queue_sqe(state->link.head); |
6dd0be1e | 6449 | if (state->comp.nr) |
ba88ff11 | 6450 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6451 | if (state->plug_started) |
6452 | blk_finish_plug(&state->plug); | |
9f13c35b | 6453 | io_state_file_put(state); |
9a56a232 JA |
6454 | } |
6455 | ||
6456 | /* | |
6457 | * Start submission side cache. | |
6458 | */ | |
6459 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6460 | unsigned int max_ios) |
9a56a232 | 6461 | { |
27926b68 | 6462 | state->plug_started = false; |
9a56a232 | 6463 | state->ios_left = max_ios; |
a1ab7b35 PB |
6464 | /* set only head, no need to init link_last in advance */ |
6465 | state->link.head = NULL; | |
9a56a232 JA |
6466 | } |
6467 | ||
2b188cc1 JA |
6468 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6469 | { | |
75b28aff | 6470 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6471 | |
caf582c6 PB |
6472 | /* |
6473 | * Ensure any loads from the SQEs are done at this point, | |
6474 | * since once we write the new head, the application could | |
6475 | * write new data to them. | |
6476 | */ | |
6477 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6478 | } |
6479 | ||
2b188cc1 | 6480 | /* |
3529d8c2 | 6481 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6482 | * that is mapped by userspace. This means that care needs to be taken to |
6483 | * ensure that reads are stable, as we cannot rely on userspace always | |
6484 | * being a good citizen. If members of the sqe are validated and then later | |
6485 | * used, it's important that those reads are done through READ_ONCE() to | |
6486 | * prevent a re-load down the line. | |
6487 | */ | |
709b302f | 6488 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6489 | { |
75b28aff | 6490 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6491 | unsigned head; |
6492 | ||
6493 | /* | |
6494 | * The cached sq head (or cq tail) serves two purposes: | |
6495 | * | |
6496 | * 1) allows us to batch the cost of updating the user visible | |
6497 | * head updates. | |
6498 | * 2) allows the kernel side to track the head on its own, even | |
6499 | * though the application is the one updating it. | |
6500 | */ | |
4fccfcbb | 6501 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6502 | if (likely(head < ctx->sq_entries)) |
6503 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6504 | |
6505 | /* drop invalid entries */ | |
498ccd9e | 6506 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6507 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6508 | return NULL; |
6509 | } | |
6510 | ||
0f212204 | 6511 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6512 | { |
46c4e16a | 6513 | int submitted = 0; |
6c271ce2 | 6514 | |
c4a2ed72 | 6515 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6516 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6517 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6518 | return -EBUSY; |
6519 | } | |
6c271ce2 | 6520 | |
ee7d46d9 PB |
6521 | /* make sure SQ entry isn't read before tail */ |
6522 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6523 | |
2b85edfc PB |
6524 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6525 | return -EAGAIN; | |
6c271ce2 | 6526 | |
d8a6df10 | 6527 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6528 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6529 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6530 | |
46c4e16a | 6531 | while (submitted < nr) { |
3529d8c2 | 6532 | const struct io_uring_sqe *sqe; |
196be95c | 6533 | struct io_kiocb *req; |
fb5ccc98 | 6534 | |
258b29a9 | 6535 | req = io_alloc_req(ctx); |
196be95c PB |
6536 | if (unlikely(!req)) { |
6537 | if (!submitted) | |
6538 | submitted = -EAGAIN; | |
fb5ccc98 | 6539 | break; |
196be95c | 6540 | } |
4fccfcbb PB |
6541 | sqe = io_get_sqe(ctx); |
6542 | if (unlikely(!sqe)) { | |
6543 | kmem_cache_free(req_cachep, req); | |
6544 | break; | |
6545 | } | |
d3656344 JA |
6546 | /* will complete beyond this point, count as submitted */ |
6547 | submitted++; | |
a1ab7b35 | 6548 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6549 | break; |
6c271ce2 JA |
6550 | } |
6551 | ||
9466f437 PB |
6552 | if (unlikely(submitted != nr)) { |
6553 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6554 | struct io_uring_task *tctx = current->io_uring; |
6555 | int unused = nr - ref_used; | |
9466f437 | 6556 | |
d8a6df10 JA |
6557 | percpu_ref_put_many(&ctx->refs, unused); |
6558 | percpu_counter_sub(&tctx->inflight, unused); | |
6559 | put_task_struct_many(current, unused); | |
9466f437 | 6560 | } |
6c271ce2 | 6561 | |
a1ab7b35 | 6562 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6563 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6564 | io_commit_sqring(ctx); | |
6565 | ||
6c271ce2 JA |
6566 | return submitted; |
6567 | } | |
6568 | ||
23b3628e XW |
6569 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6570 | { | |
6571 | /* Tell userspace we may need a wakeup call */ | |
6572 | spin_lock_irq(&ctx->completion_lock); | |
6573 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6574 | spin_unlock_irq(&ctx->completion_lock); | |
6575 | } | |
6576 | ||
6577 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6578 | { | |
6579 | spin_lock_irq(&ctx->completion_lock); | |
6580 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6581 | spin_unlock_irq(&ctx->completion_lock); | |
6582 | } | |
6583 | ||
08369246 | 6584 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6585 | { |
c8d1ba58 | 6586 | unsigned int to_submit; |
bdcd3eab | 6587 | int ret = 0; |
6c271ce2 | 6588 | |
c8d1ba58 | 6589 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6590 | /* if we're handling multiple rings, cap submit size for fairness */ |
6591 | if (cap_entries && to_submit > 8) | |
6592 | to_submit = 8; | |
6593 | ||
906a3c6f | 6594 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6595 | unsigned nr_events = 0; |
a4c0b3de | 6596 | |
c8d1ba58 | 6597 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6598 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6599 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6600 | |
70aacfe6 | 6601 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs))) |
08369246 | 6602 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6603 | mutex_unlock(&ctx->uring_lock); |
6604 | } | |
6c271ce2 | 6605 | |
90554200 JA |
6606 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6607 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6608 | |
08369246 XW |
6609 | return ret; |
6610 | } | |
6c271ce2 | 6611 | |
08369246 XW |
6612 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6613 | { | |
6614 | struct io_ring_ctx *ctx; | |
6615 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6616 | |
08369246 XW |
6617 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6618 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6619 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6620 | } |
c1edbf5f | 6621 | |
08369246 | 6622 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6623 | } |
6c271ce2 | 6624 | |
69fb2131 JA |
6625 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6626 | { | |
6627 | struct io_ring_ctx *ctx; | |
6628 | ||
6629 | while (!list_empty(&sqd->ctx_new_list)) { | |
6630 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
6631 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
6632 | complete(&ctx->sq_thread_comp); | |
6633 | } | |
08369246 XW |
6634 | |
6635 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
6636 | } |
6637 | ||
37d1e2e3 JA |
6638 | static bool io_sq_thread_should_stop(struct io_sq_data *sqd) |
6639 | { | |
6640 | return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
6641 | } | |
6642 | ||
6643 | static bool io_sq_thread_should_park(struct io_sq_data *sqd) | |
6644 | { | |
6645 | return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
6646 | } | |
6647 | ||
6648 | static void io_sq_thread_parkme(struct io_sq_data *sqd) | |
6649 | { | |
6650 | for (;;) { | |
6651 | /* | |
6652 | * TASK_PARKED is a special state; we must serialize against | |
6653 | * possible pending wakeups to avoid store-store collisions on | |
6654 | * task->state. | |
6655 | * | |
6656 | * Such a collision might possibly result in the task state | |
6657 | * changin from TASK_PARKED and us failing the | |
6658 | * wait_task_inactive() in kthread_park(). | |
6659 | */ | |
6660 | set_special_state(TASK_PARKED); | |
6661 | if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) | |
6662 | break; | |
6663 | ||
6664 | /* | |
6665 | * Thread is going to call schedule(), do not preempt it, | |
6666 | * or the caller of kthread_park() may spend more time in | |
6667 | * wait_task_inactive(). | |
6668 | */ | |
6669 | preempt_disable(); | |
6670 | complete(&sqd->completion); | |
6671 | schedule_preempt_disabled(); | |
6672 | preempt_enable(); | |
6673 | } | |
6674 | __set_current_state(TASK_RUNNING); | |
6675 | } | |
6676 | ||
c8d1ba58 JA |
6677 | static int io_sq_thread(void *data) |
6678 | { | |
69fb2131 JA |
6679 | struct io_sq_data *sqd = data; |
6680 | struct io_ring_ctx *ctx; | |
a0d9205f | 6681 | unsigned long timeout = 0; |
37d1e2e3 | 6682 | char buf[TASK_COMM_LEN]; |
08369246 | 6683 | DEFINE_WAIT(wait); |
6c271ce2 | 6684 | |
37d1e2e3 JA |
6685 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6686 | set_task_comm(current, buf); | |
6687 | sqd->thread = current; | |
6688 | current->pf_io_worker = NULL; | |
6689 | ||
6690 | if (sqd->sq_cpu != -1) | |
6691 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6692 | else | |
6693 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6694 | current->flags |= PF_NO_SETAFFINITY; | |
6695 | ||
6696 | complete(&sqd->completion); | |
6c271ce2 | 6697 | |
37d1e2e3 JA |
6698 | wait_for_completion(&sqd->startup); |
6699 | ||
6700 | while (!io_sq_thread_should_stop(sqd)) { | |
08369246 XW |
6701 | int ret; |
6702 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
6703 | |
6704 | /* | |
69fb2131 | 6705 | * Any changes to the sqd lists are synchronized through the |
37d1e2e3 | 6706 | * thread parking. This synchronizes the thread vs users, |
69fb2131 | 6707 | * the users are synchronized on the sqd->ctx_lock. |
c1edbf5f | 6708 | */ |
37d1e2e3 JA |
6709 | if (io_sq_thread_should_park(sqd)) { |
6710 | io_sq_thread_parkme(sqd); | |
6711 | continue; | |
65b2b213 | 6712 | } |
08369246 | 6713 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 6714 | io_sqd_init_new(sqd); |
08369246 XW |
6715 | timeout = jiffies + sqd->sq_thread_idle; |
6716 | } | |
37d1e2e3 JA |
6717 | if (fatal_signal_pending(current)) |
6718 | break; | |
08369246 | 6719 | sqt_spin = false; |
e95eee2d | 6720 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6721 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
08369246 XW |
6722 | ret = __io_sq_thread(ctx, cap_entries); |
6723 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
6724 | sqt_spin = true; | |
69fb2131 | 6725 | } |
6c271ce2 | 6726 | |
08369246 | 6727 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6728 | io_run_task_work(); |
6729 | cond_resched(); | |
08369246 XW |
6730 | if (sqt_spin) |
6731 | timeout = jiffies + sqd->sq_thread_idle; | |
6732 | continue; | |
6733 | } | |
6734 | ||
08369246 XW |
6735 | needs_sched = true; |
6736 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6737 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6738 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6739 | !list_empty_careful(&ctx->iopoll_list)) { | |
6740 | needs_sched = false; | |
6741 | break; | |
6742 | } | |
6743 | if (io_sqring_entries(ctx)) { | |
6744 | needs_sched = false; | |
6745 | break; | |
6746 | } | |
6747 | } | |
6748 | ||
37d1e2e3 | 6749 | if (needs_sched && !io_sq_thread_should_park(sqd)) { |
69fb2131 JA |
6750 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6751 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6752 | |
69fb2131 | 6753 | schedule(); |
69fb2131 JA |
6754 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6755 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6756 | } |
08369246 XW |
6757 | |
6758 | finish_wait(&sqd->wait, &wait); | |
6759 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
6760 | } |
6761 | ||
37d1e2e3 JA |
6762 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6763 | io_uring_cancel_sqpoll(ctx); | |
06058632 | 6764 | |
37d1e2e3 | 6765 | io_run_task_work(); |
28cea78a | 6766 | |
8629397e JA |
6767 | if (io_sq_thread_should_park(sqd)) |
6768 | io_sq_thread_parkme(sqd); | |
6769 | ||
37d1e2e3 JA |
6770 | /* |
6771 | * Clear thread under lock so that concurrent parks work correctly | |
6772 | */ | |
8629397e | 6773 | complete(&sqd->completion); |
37d1e2e3 JA |
6774 | mutex_lock(&sqd->lock); |
6775 | sqd->thread = NULL; | |
5f3f26f9 JA |
6776 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6777 | ctx->sqo_exec = 1; | |
6778 | io_ring_set_wakeup_flag(ctx); | |
6779 | } | |
06058632 | 6780 | |
37d1e2e3 | 6781 | complete(&sqd->exited); |
e54945ae | 6782 | mutex_unlock(&sqd->lock); |
37d1e2e3 | 6783 | do_exit(0); |
6c271ce2 JA |
6784 | } |
6785 | ||
bda52162 JA |
6786 | struct io_wait_queue { |
6787 | struct wait_queue_entry wq; | |
6788 | struct io_ring_ctx *ctx; | |
6789 | unsigned to_wait; | |
6790 | unsigned nr_timeouts; | |
6791 | }; | |
6792 | ||
6c503150 | 6793 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6794 | { |
6795 | struct io_ring_ctx *ctx = iowq->ctx; | |
6796 | ||
6797 | /* | |
d195a66e | 6798 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6799 | * started waiting. For timeouts, we always want to return to userspace, |
6800 | * regardless of event count. | |
6801 | */ | |
6c503150 | 6802 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6803 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6804 | } | |
6805 | ||
6806 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6807 | int wake_flags, void *key) | |
6808 | { | |
6809 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6810 | wq); | |
6811 | ||
6c503150 PB |
6812 | /* |
6813 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6814 | * the task, and the next invocation will do it. | |
6815 | */ | |
6816 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6817 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6818 | return -1; | |
bda52162 JA |
6819 | } |
6820 | ||
af9c1a44 JA |
6821 | static int io_run_task_work_sig(void) |
6822 | { | |
6823 | if (io_run_task_work()) | |
6824 | return 1; | |
6825 | if (!signal_pending(current)) | |
6826 | return 0; | |
792ee0f6 JA |
6827 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6828 | return -ERESTARTSYS; | |
af9c1a44 JA |
6829 | return -EINTR; |
6830 | } | |
6831 | ||
eeb60b9a PB |
6832 | /* when returns >0, the caller should retry */ |
6833 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6834 | struct io_wait_queue *iowq, | |
6835 | signed long *timeout) | |
6836 | { | |
6837 | int ret; | |
6838 | ||
6839 | /* make sure we run task_work before checking for signals */ | |
6840 | ret = io_run_task_work_sig(); | |
6841 | if (ret || io_should_wake(iowq)) | |
6842 | return ret; | |
6843 | /* let the caller flush overflows, retry */ | |
6844 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6845 | return 1; | |
6846 | ||
6847 | *timeout = schedule_timeout(*timeout); | |
6848 | return !*timeout ? -ETIME : 1; | |
6849 | } | |
6850 | ||
2b188cc1 JA |
6851 | /* |
6852 | * Wait until events become available, if we don't already have some. The | |
6853 | * application must reap them itself, as they reside on the shared cq ring. | |
6854 | */ | |
6855 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6856 | const sigset_t __user *sig, size_t sigsz, |
6857 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6858 | { |
bda52162 JA |
6859 | struct io_wait_queue iowq = { |
6860 | .wq = { | |
6861 | .private = current, | |
6862 | .func = io_wake_function, | |
6863 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6864 | }, | |
6865 | .ctx = ctx, | |
6866 | .to_wait = min_events, | |
6867 | }; | |
75b28aff | 6868 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6869 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6870 | int ret; | |
2b188cc1 | 6871 | |
b41e9852 | 6872 | do { |
6c503150 PB |
6873 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6874 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6875 | return 0; |
4c6e277c | 6876 | if (!io_run_task_work()) |
b41e9852 | 6877 | break; |
b41e9852 | 6878 | } while (1); |
2b188cc1 JA |
6879 | |
6880 | if (sig) { | |
9e75ad5d AB |
6881 | #ifdef CONFIG_COMPAT |
6882 | if (in_compat_syscall()) | |
6883 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6884 | sigsz); |
9e75ad5d AB |
6885 | else |
6886 | #endif | |
b772434b | 6887 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6888 | |
2b188cc1 JA |
6889 | if (ret) |
6890 | return ret; | |
6891 | } | |
6892 | ||
c73ebb68 | 6893 | if (uts) { |
c1d5a224 PB |
6894 | struct timespec64 ts; |
6895 | ||
c73ebb68 HX |
6896 | if (get_timespec64(&ts, uts)) |
6897 | return -EFAULT; | |
6898 | timeout = timespec64_to_jiffies(&ts); | |
6899 | } | |
6900 | ||
bda52162 | 6901 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6902 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6903 | do { |
6c503150 | 6904 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
6905 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6906 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6907 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6908 | finish_wait(&ctx->wait, &iowq.wq); | |
6909 | } while (ret > 0); | |
bda52162 | 6910 | |
b7db41c9 | 6911 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6912 | |
75b28aff | 6913 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6914 | } |
6915 | ||
6b06314c JA |
6916 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6917 | { | |
6918 | #if defined(CONFIG_UNIX) | |
6919 | if (ctx->ring_sock) { | |
6920 | struct sock *sock = ctx->ring_sock->sk; | |
6921 | struct sk_buff *skb; | |
6922 | ||
6923 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6924 | kfree_skb(skb); | |
6925 | } | |
6926 | #else | |
6927 | int i; | |
6928 | ||
65e19f54 JA |
6929 | for (i = 0; i < ctx->nr_user_files; i++) { |
6930 | struct file *file; | |
6931 | ||
6932 | file = io_file_from_index(ctx, i); | |
6933 | if (file) | |
6934 | fput(file); | |
6935 | } | |
6b06314c JA |
6936 | #endif |
6937 | } | |
6938 | ||
00835dce | 6939 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6940 | { |
269bbe5f | 6941 | struct fixed_rsrc_data *data; |
05f3fb3c | 6942 | |
269bbe5f | 6943 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6944 | complete(&data->done); |
6945 | } | |
6946 | ||
2a63b2d9 | 6947 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6948 | { |
2a63b2d9 | 6949 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6950 | } |
6951 | ||
2a63b2d9 | 6952 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6953 | { |
2a63b2d9 BM |
6954 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6955 | } | |
65e19f54 | 6956 | |
d67d2263 BM |
6957 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6958 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6959 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6960 | { |
2a63b2d9 | 6961 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6962 | rsrc_data->node = ref_node; |
d67d2263 | 6963 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6964 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6965 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6966 | } |
6967 | ||
8bad28d8 | 6968 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6969 | { |
8bad28d8 | 6970 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6971 | |
2a63b2d9 | 6972 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6973 | ref_node = data->node; |
e6cb007c | 6974 | data->node = NULL; |
2a63b2d9 | 6975 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6976 | if (ref_node) |
6977 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6978 | } |
6979 | ||
6980 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6981 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6982 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6983 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6984 | { |
f2303b1f | 6985 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6986 | int ret; |
05589553 | 6987 | |
8bad28d8 HX |
6988 | if (data->quiesce) |
6989 | return -ENXIO; | |
05589553 | 6990 | |
8bad28d8 | 6991 | data->quiesce = true; |
1ffc5422 | 6992 | do { |
f2303b1f PB |
6993 | ret = -ENOMEM; |
6994 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
6995 | if (!backup_node) | |
6996 | break; | |
6997 | backup_node->rsrc_data = data; | |
6998 | backup_node->rsrc_put = rsrc_put; | |
6999 | ||
8bad28d8 HX |
7000 | io_sqe_rsrc_kill_node(ctx, data); |
7001 | percpu_ref_kill(&data->refs); | |
7002 | flush_delayed_work(&ctx->rsrc_put_work); | |
7003 | ||
1ffc5422 PB |
7004 | ret = wait_for_completion_interruptible(&data->done); |
7005 | if (!ret) | |
7006 | break; | |
8bad28d8 | 7007 | |
cb5e1b81 | 7008 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7009 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7010 | backup_node = NULL; | |
cb5e1b81 | 7011 | reinit_completion(&data->done); |
8bad28d8 | 7012 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7013 | ret = io_run_task_work_sig(); |
8bad28d8 | 7014 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7015 | } while (ret >= 0); |
8bad28d8 | 7016 | data->quiesce = false; |
05f3fb3c | 7017 | |
8bad28d8 HX |
7018 | if (backup_node) |
7019 | destroy_fixed_rsrc_ref_node(backup_node); | |
7020 | return ret; | |
d7954b2b BM |
7021 | } |
7022 | ||
1ad555c6 BM |
7023 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7024 | { | |
7025 | struct fixed_rsrc_data *data; | |
7026 | ||
7027 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7028 | if (!data) | |
7029 | return NULL; | |
7030 | ||
00835dce | 7031 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7032 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7033 | kfree(data); | |
7034 | return NULL; | |
7035 | } | |
7036 | data->ctx = ctx; | |
7037 | init_completion(&data->done); | |
7038 | return data; | |
7039 | } | |
7040 | ||
7041 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7042 | { | |
7043 | percpu_ref_exit(&data->refs); | |
7044 | kfree(data->table); | |
7045 | kfree(data); | |
7046 | } | |
7047 | ||
d7954b2b BM |
7048 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7049 | { | |
7050 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7051 | unsigned nr_tables, i; |
7052 | int ret; | |
7053 | ||
8bad28d8 HX |
7054 | /* |
7055 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7056 | * Since we possibly drop uring lock later in this function to | |
7057 | * run task work. | |
7058 | */ | |
7059 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7060 | return -ENXIO; |
f2303b1f | 7061 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7062 | if (ret) |
7063 | return ret; | |
7064 | ||
6b06314c | 7065 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7066 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7067 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7068 | kfree(data->table[i].files); |
1ad555c6 | 7069 | free_fixed_rsrc_data(data); |
05f3fb3c | 7070 | ctx->file_data = NULL; |
6b06314c JA |
7071 | ctx->nr_user_files = 0; |
7072 | return 0; | |
7073 | } | |
7074 | ||
37d1e2e3 JA |
7075 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7076 | __releases(&sqd->lock) | |
7077 | { | |
7078 | if (!sqd->thread) | |
7079 | return; | |
7080 | if (sqd->thread == current) | |
7081 | return; | |
7082 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7083 | wake_up_state(sqd->thread, TASK_PARKED); | |
7084 | mutex_unlock(&sqd->lock); | |
7085 | } | |
7086 | ||
7087 | static bool io_sq_thread_park(struct io_sq_data *sqd) | |
7088 | __acquires(&sqd->lock) | |
7089 | { | |
7090 | if (sqd->thread == current) | |
7091 | return true; | |
7092 | mutex_lock(&sqd->lock); | |
7093 | if (!sqd->thread) { | |
7094 | mutex_unlock(&sqd->lock); | |
7095 | return false; | |
7096 | } | |
7097 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7098 | wake_up_process(sqd->thread); | |
7099 | wait_for_completion(&sqd->completion); | |
7100 | return true; | |
7101 | } | |
7102 | ||
7103 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7104 | { | |
e54945ae | 7105 | if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) |
37d1e2e3 | 7106 | return; |
e54945ae JA |
7107 | mutex_lock(&sqd->lock); |
7108 | if (sqd->thread) { | |
7109 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7110 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)); | |
7111 | wake_up_process(sqd->thread); | |
7112 | mutex_unlock(&sqd->lock); | |
7113 | wait_for_completion(&sqd->exited); | |
7114 | WARN_ON_ONCE(sqd->thread); | |
7115 | } else { | |
7116 | mutex_unlock(&sqd->lock); | |
7117 | } | |
37d1e2e3 JA |
7118 | } |
7119 | ||
534ca6d6 | 7120 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7121 | { |
534ca6d6 | 7122 | if (refcount_dec_and_test(&sqd->refs)) { |
37d1e2e3 JA |
7123 | io_sq_thread_stop(sqd); |
7124 | kfree(sqd); | |
7125 | } | |
7126 | } | |
7127 | ||
7128 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7129 | { | |
7130 | struct io_sq_data *sqd = ctx->sq_data; | |
7131 | ||
7132 | if (sqd) { | |
eb85890b | 7133 | complete(&sqd->startup); |
534ca6d6 | 7134 | if (sqd->thread) { |
37d1e2e3 JA |
7135 | wait_for_completion(&ctx->sq_thread_comp); |
7136 | io_sq_thread_park(sqd); | |
534ca6d6 JA |
7137 | } |
7138 | ||
37d1e2e3 JA |
7139 | mutex_lock(&sqd->ctx_lock); |
7140 | list_del(&ctx->sqd_list); | |
7141 | io_sqd_update_thread_idle(sqd); | |
7142 | mutex_unlock(&sqd->ctx_lock); | |
7143 | ||
7144 | if (sqd->thread) | |
7145 | io_sq_thread_unpark(sqd); | |
7146 | ||
7147 | io_put_sq_data(sqd); | |
7148 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7149 | } |
7150 | } | |
7151 | ||
aa06165d JA |
7152 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7153 | { | |
7154 | struct io_ring_ctx *ctx_attach; | |
7155 | struct io_sq_data *sqd; | |
7156 | struct fd f; | |
7157 | ||
7158 | f = fdget(p->wq_fd); | |
7159 | if (!f.file) | |
7160 | return ERR_PTR(-ENXIO); | |
7161 | if (f.file->f_op != &io_uring_fops) { | |
7162 | fdput(f); | |
7163 | return ERR_PTR(-EINVAL); | |
7164 | } | |
7165 | ||
7166 | ctx_attach = f.file->private_data; | |
7167 | sqd = ctx_attach->sq_data; | |
7168 | if (!sqd) { | |
7169 | fdput(f); | |
7170 | return ERR_PTR(-EINVAL); | |
7171 | } | |
7172 | ||
7173 | refcount_inc(&sqd->refs); | |
7174 | fdput(f); | |
7175 | return sqd; | |
7176 | } | |
7177 | ||
534ca6d6 JA |
7178 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7179 | { | |
7180 | struct io_sq_data *sqd; | |
7181 | ||
aa06165d JA |
7182 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7183 | return io_attach_sq_data(p); | |
7184 | ||
534ca6d6 JA |
7185 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7186 | if (!sqd) | |
7187 | return ERR_PTR(-ENOMEM); | |
7188 | ||
7189 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7190 | INIT_LIST_HEAD(&sqd->ctx_list); |
7191 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7192 | mutex_init(&sqd->ctx_lock); | |
7193 | mutex_init(&sqd->lock); | |
534ca6d6 | 7194 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 JA |
7195 | init_completion(&sqd->startup); |
7196 | init_completion(&sqd->completion); | |
7197 | init_completion(&sqd->exited); | |
534ca6d6 JA |
7198 | return sqd; |
7199 | } | |
7200 | ||
6b06314c | 7201 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7202 | /* |
7203 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7204 | * the io_uring can be safely unregistered on process exit, even if we have | |
7205 | * loops in the file referencing. | |
7206 | */ | |
7207 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7208 | { | |
7209 | struct sock *sk = ctx->ring_sock->sk; | |
7210 | struct scm_fp_list *fpl; | |
7211 | struct sk_buff *skb; | |
08a45173 | 7212 | int i, nr_files; |
6b06314c | 7213 | |
6b06314c JA |
7214 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7215 | if (!fpl) | |
7216 | return -ENOMEM; | |
7217 | ||
7218 | skb = alloc_skb(0, GFP_KERNEL); | |
7219 | if (!skb) { | |
7220 | kfree(fpl); | |
7221 | return -ENOMEM; | |
7222 | } | |
7223 | ||
7224 | skb->sk = sk; | |
6b06314c | 7225 | |
08a45173 | 7226 | nr_files = 0; |
62e398be | 7227 | fpl->user = get_uid(current_user()); |
6b06314c | 7228 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7229 | struct file *file = io_file_from_index(ctx, i + offset); |
7230 | ||
7231 | if (!file) | |
08a45173 | 7232 | continue; |
65e19f54 | 7233 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7234 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7235 | nr_files++; | |
6b06314c JA |
7236 | } |
7237 | ||
08a45173 JA |
7238 | if (nr_files) { |
7239 | fpl->max = SCM_MAX_FD; | |
7240 | fpl->count = nr_files; | |
7241 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7242 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7243 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7244 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7245 | |
08a45173 JA |
7246 | for (i = 0; i < nr_files; i++) |
7247 | fput(fpl->fp[i]); | |
7248 | } else { | |
7249 | kfree_skb(skb); | |
7250 | kfree(fpl); | |
7251 | } | |
6b06314c JA |
7252 | |
7253 | return 0; | |
7254 | } | |
7255 | ||
7256 | /* | |
7257 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7258 | * causes regular reference counting to break down. We rely on the UNIX | |
7259 | * garbage collection to take care of this problem for us. | |
7260 | */ | |
7261 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7262 | { | |
7263 | unsigned left, total; | |
7264 | int ret = 0; | |
7265 | ||
7266 | total = 0; | |
7267 | left = ctx->nr_user_files; | |
7268 | while (left) { | |
7269 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7270 | |
7271 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7272 | if (ret) | |
7273 | break; | |
7274 | left -= this_files; | |
7275 | total += this_files; | |
7276 | } | |
7277 | ||
7278 | if (!ret) | |
7279 | return 0; | |
7280 | ||
7281 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7282 | struct file *file = io_file_from_index(ctx, total); |
7283 | ||
7284 | if (file) | |
7285 | fput(file); | |
6b06314c JA |
7286 | total++; |
7287 | } | |
7288 | ||
7289 | return ret; | |
7290 | } | |
7291 | #else | |
7292 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7293 | { | |
7294 | return 0; | |
7295 | } | |
7296 | #endif | |
7297 | ||
269bbe5f | 7298 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7299 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7300 | { |
7301 | int i; | |
7302 | ||
7303 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7304 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7305 | unsigned this_files; |
7306 | ||
7307 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7308 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7309 | GFP_KERNEL); | |
7310 | if (!table->files) | |
7311 | break; | |
7312 | nr_files -= this_files; | |
7313 | } | |
7314 | ||
7315 | if (i == nr_tables) | |
7316 | return 0; | |
7317 | ||
7318 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7319 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7320 | kfree(table->files); |
7321 | } | |
7322 | return 1; | |
7323 | } | |
7324 | ||
50238531 | 7325 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7326 | { |
50238531 | 7327 | struct file *file = prsrc->file; |
05f3fb3c JA |
7328 | #if defined(CONFIG_UNIX) |
7329 | struct sock *sock = ctx->ring_sock->sk; | |
7330 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7331 | struct sk_buff *skb; | |
7332 | int i; | |
7333 | ||
7334 | __skb_queue_head_init(&list); | |
7335 | ||
7336 | /* | |
7337 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7338 | * remove this entry and rearrange the file array. | |
7339 | */ | |
7340 | skb = skb_dequeue(head); | |
7341 | while (skb) { | |
7342 | struct scm_fp_list *fp; | |
7343 | ||
7344 | fp = UNIXCB(skb).fp; | |
7345 | for (i = 0; i < fp->count; i++) { | |
7346 | int left; | |
7347 | ||
7348 | if (fp->fp[i] != file) | |
7349 | continue; | |
7350 | ||
7351 | unix_notinflight(fp->user, fp->fp[i]); | |
7352 | left = fp->count - 1 - i; | |
7353 | if (left) { | |
7354 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7355 | left * sizeof(struct file *)); | |
7356 | } | |
7357 | fp->count--; | |
7358 | if (!fp->count) { | |
7359 | kfree_skb(skb); | |
7360 | skb = NULL; | |
7361 | } else { | |
7362 | __skb_queue_tail(&list, skb); | |
7363 | } | |
7364 | fput(file); | |
7365 | file = NULL; | |
7366 | break; | |
7367 | } | |
7368 | ||
7369 | if (!file) | |
7370 | break; | |
7371 | ||
7372 | __skb_queue_tail(&list, skb); | |
7373 | ||
7374 | skb = skb_dequeue(head); | |
7375 | } | |
7376 | ||
7377 | if (skb_peek(&list)) { | |
7378 | spin_lock_irq(&head->lock); | |
7379 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7380 | __skb_queue_tail(head, skb); | |
7381 | spin_unlock_irq(&head->lock); | |
7382 | } | |
7383 | #else | |
7384 | fput(file); | |
7385 | #endif | |
7386 | } | |
7387 | ||
269bbe5f | 7388 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7389 | { |
269bbe5f BM |
7390 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7391 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7392 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7393 | |
269bbe5f BM |
7394 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7395 | list_del(&prsrc->list); | |
50238531 | 7396 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7397 | kfree(prsrc); |
65e19f54 | 7398 | } |
05589553 | 7399 | |
05589553 XW |
7400 | percpu_ref_exit(&ref_node->refs); |
7401 | kfree(ref_node); | |
269bbe5f | 7402 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7403 | } |
65e19f54 | 7404 | |
269bbe5f | 7405 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7406 | { |
7407 | struct io_ring_ctx *ctx; | |
7408 | struct llist_node *node; | |
7409 | ||
269bbe5f BM |
7410 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7411 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7412 | |
7413 | while (node) { | |
269bbe5f | 7414 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7415 | struct llist_node *next = node->next; |
7416 | ||
269bbe5f BM |
7417 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7418 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7419 | node = next; |
7420 | } | |
7421 | } | |
7422 | ||
ea64ec02 PB |
7423 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7424 | unsigned i) | |
2faf852d | 7425 | { |
ea64ec02 PB |
7426 | struct fixed_rsrc_table *table; |
7427 | ||
7428 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7429 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7430 | } | |
7431 | ||
00835dce | 7432 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7433 | { |
269bbe5f BM |
7434 | struct fixed_rsrc_ref_node *ref_node; |
7435 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7436 | struct io_ring_ctx *ctx; |
e297822b | 7437 | bool first_add = false; |
4a38aed2 | 7438 | int delay = HZ; |
65e19f54 | 7439 | |
269bbe5f BM |
7440 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7441 | data = ref_node->rsrc_data; | |
e297822b PB |
7442 | ctx = data->ctx; |
7443 | ||
2a63b2d9 | 7444 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7445 | ref_node->done = true; |
7446 | ||
d67d2263 BM |
7447 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7448 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7449 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7450 | /* recycle ref nodes in order */ |
7451 | if (!ref_node->done) | |
7452 | break; | |
7453 | list_del(&ref_node->node); | |
269bbe5f | 7454 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7455 | } |
2a63b2d9 | 7456 | io_rsrc_ref_unlock(ctx); |
05589553 | 7457 | |
e297822b | 7458 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7459 | delay = 0; |
05589553 | 7460 | |
4a38aed2 | 7461 | if (!delay) |
269bbe5f | 7462 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7463 | else if (first_add) |
269bbe5f | 7464 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7465 | } |
65e19f54 | 7466 | |
6802535d | 7467 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7468 | struct io_ring_ctx *ctx) |
05f3fb3c | 7469 | { |
269bbe5f | 7470 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7471 | |
05589553 XW |
7472 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7473 | if (!ref_node) | |
3e2224c5 | 7474 | return NULL; |
05f3fb3c | 7475 | |
00835dce | 7476 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7477 | 0, GFP_KERNEL)) { |
7478 | kfree(ref_node); | |
3e2224c5 | 7479 | return NULL; |
05589553 XW |
7480 | } |
7481 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7482 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7483 | ref_node->done = false; |
05589553 | 7484 | return ref_node; |
05589553 XW |
7485 | } |
7486 | ||
bc9744cd PB |
7487 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7488 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7489 | { |
269bbe5f | 7490 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7491 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7492 | } |
7493 | ||
269bbe5f | 7494 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7495 | { |
7496 | percpu_ref_exit(&ref_node->refs); | |
7497 | kfree(ref_node); | |
65e19f54 JA |
7498 | } |
7499 | ||
ea64ec02 | 7500 | |
6b06314c JA |
7501 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7502 | unsigned nr_args) | |
7503 | { | |
7504 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7505 | unsigned nr_tables, i; |
05f3fb3c | 7506 | struct file *file; |
600cf3f8 | 7507 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7508 | struct fixed_rsrc_ref_node *ref_node; |
7509 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7510 | |
05f3fb3c | 7511 | if (ctx->file_data) |
6b06314c JA |
7512 | return -EBUSY; |
7513 | if (!nr_args) | |
7514 | return -EINVAL; | |
7515 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7516 | return -EMFILE; | |
7517 | ||
1ad555c6 | 7518 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7519 | if (!file_data) |
05f3fb3c | 7520 | return -ENOMEM; |
13770a71 | 7521 | ctx->file_data = file_data; |
05f3fb3c | 7522 | |
65e19f54 | 7523 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7524 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7525 | GFP_KERNEL); |
600cf3f8 PB |
7526 | if (!file_data->table) |
7527 | goto out_free; | |
05f3fb3c | 7528 | |
600cf3f8 | 7529 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7530 | goto out_free; |
65e19f54 | 7531 | |
08a45173 | 7532 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7533 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7534 | ret = -EFAULT; | |
7535 | goto out_fput; | |
7536 | } | |
08a45173 | 7537 | /* allow sparse sets */ |
600cf3f8 | 7538 | if (fd == -1) |
08a45173 | 7539 | continue; |
6b06314c | 7540 | |
05f3fb3c | 7541 | file = fget(fd); |
6b06314c | 7542 | ret = -EBADF; |
05f3fb3c | 7543 | if (!file) |
600cf3f8 | 7544 | goto out_fput; |
05f3fb3c | 7545 | |
6b06314c JA |
7546 | /* |
7547 | * Don't allow io_uring instances to be registered. If UNIX | |
7548 | * isn't enabled, then this causes a reference cycle and this | |
7549 | * instance can never get freed. If UNIX is enabled we'll | |
7550 | * handle it just fine, but there's still no point in allowing | |
7551 | * a ring fd as it doesn't support regular read/write anyway. | |
7552 | */ | |
05f3fb3c JA |
7553 | if (file->f_op == &io_uring_fops) { |
7554 | fput(file); | |
600cf3f8 | 7555 | goto out_fput; |
6b06314c | 7556 | } |
ea64ec02 | 7557 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7558 | } |
7559 | ||
6b06314c | 7560 | ret = io_sqe_files_scm(ctx); |
05589553 | 7561 | if (ret) { |
6b06314c | 7562 | io_sqe_files_unregister(ctx); |
05589553 XW |
7563 | return ret; |
7564 | } | |
6b06314c | 7565 | |
bc9744cd | 7566 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7567 | if (!ref_node) { |
05589553 | 7568 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7569 | return -ENOMEM; |
05589553 | 7570 | } |
bc9744cd | 7571 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7572 | |
d67d2263 | 7573 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7574 | return ret; |
600cf3f8 PB |
7575 | out_fput: |
7576 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7577 | file = io_file_from_index(ctx, i); | |
7578 | if (file) | |
7579 | fput(file); | |
7580 | } | |
7581 | for (i = 0; i < nr_tables; i++) | |
7582 | kfree(file_data->table[i].files); | |
7583 | ctx->nr_user_files = 0; | |
600cf3f8 | 7584 | out_free: |
1ad555c6 | 7585 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7586 | ctx->file_data = NULL; |
6b06314c JA |
7587 | return ret; |
7588 | } | |
7589 | ||
c3a31e60 JA |
7590 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7591 | int index) | |
7592 | { | |
7593 | #if defined(CONFIG_UNIX) | |
7594 | struct sock *sock = ctx->ring_sock->sk; | |
7595 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7596 | struct sk_buff *skb; | |
7597 | ||
7598 | /* | |
7599 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7600 | * file set. If there's no room, fall back to allocating a new skb | |
7601 | * and filling it in. | |
7602 | */ | |
7603 | spin_lock_irq(&head->lock); | |
7604 | skb = skb_peek(head); | |
7605 | if (skb) { | |
7606 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7607 | ||
7608 | if (fpl->count < SCM_MAX_FD) { | |
7609 | __skb_unlink(skb, head); | |
7610 | spin_unlock_irq(&head->lock); | |
7611 | fpl->fp[fpl->count] = get_file(file); | |
7612 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7613 | fpl->count++; | |
7614 | spin_lock_irq(&head->lock); | |
7615 | __skb_queue_head(head, skb); | |
7616 | } else { | |
7617 | skb = NULL; | |
7618 | } | |
7619 | } | |
7620 | spin_unlock_irq(&head->lock); | |
7621 | ||
7622 | if (skb) { | |
7623 | fput(file); | |
7624 | return 0; | |
7625 | } | |
7626 | ||
7627 | return __io_sqe_files_scm(ctx, 1, index); | |
7628 | #else | |
7629 | return 0; | |
7630 | #endif | |
7631 | } | |
7632 | ||
50238531 | 7633 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7634 | { |
269bbe5f BM |
7635 | struct io_rsrc_put *prsrc; |
7636 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7637 | |
269bbe5f BM |
7638 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7639 | if (!prsrc) | |
a5318d3c | 7640 | return -ENOMEM; |
05f3fb3c | 7641 | |
50238531 | 7642 | prsrc->rsrc = rsrc; |
269bbe5f | 7643 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7644 | |
a5318d3c | 7645 | return 0; |
05f3fb3c JA |
7646 | } |
7647 | ||
269bbe5f BM |
7648 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7649 | struct file *file) | |
7650 | { | |
50238531 | 7651 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7652 | } |
7653 | ||
05f3fb3c | 7654 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7655 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7656 | unsigned nr_args) |
7657 | { | |
269bbe5f BM |
7658 | struct fixed_rsrc_data *data = ctx->file_data; |
7659 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7660 | struct file *file, **file_slot; |
c3a31e60 JA |
7661 | __s32 __user *fds; |
7662 | int fd, i, err; | |
7663 | __u32 done; | |
05589553 | 7664 | bool needs_switch = false; |
c3a31e60 | 7665 | |
05f3fb3c | 7666 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7667 | return -EOVERFLOW; |
7668 | if (done > ctx->nr_user_files) | |
7669 | return -EINVAL; | |
7670 | ||
bc9744cd | 7671 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7672 | if (!ref_node) |
7673 | return -ENOMEM; | |
bc9744cd | 7674 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7675 | |
269bbe5f | 7676 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7677 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7678 | err = 0; |
7679 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7680 | err = -EFAULT; | |
7681 | break; | |
7682 | } | |
4e0377a1 | 7683 | if (fd == IORING_REGISTER_FILES_SKIP) |
7684 | continue; | |
7685 | ||
67973b93 | 7686 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7687 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7688 | ||
7689 | if (*file_slot) { | |
7690 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7691 | if (err) |
7692 | break; | |
ea64ec02 | 7693 | *file_slot = NULL; |
05589553 | 7694 | needs_switch = true; |
c3a31e60 JA |
7695 | } |
7696 | if (fd != -1) { | |
c3a31e60 JA |
7697 | file = fget(fd); |
7698 | if (!file) { | |
7699 | err = -EBADF; | |
7700 | break; | |
7701 | } | |
7702 | /* | |
7703 | * Don't allow io_uring instances to be registered. If | |
7704 | * UNIX isn't enabled, then this causes a reference | |
7705 | * cycle and this instance can never get freed. If UNIX | |
7706 | * is enabled we'll handle it just fine, but there's | |
7707 | * still no point in allowing a ring fd as it doesn't | |
7708 | * support regular read/write anyway. | |
7709 | */ | |
7710 | if (file->f_op == &io_uring_fops) { | |
7711 | fput(file); | |
7712 | err = -EBADF; | |
7713 | break; | |
7714 | } | |
e68a3ff8 | 7715 | *file_slot = file; |
c3a31e60 | 7716 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7717 | if (err) { |
e68a3ff8 | 7718 | *file_slot = NULL; |
f3bd9dae | 7719 | fput(file); |
c3a31e60 | 7720 | break; |
f3bd9dae | 7721 | } |
c3a31e60 | 7722 | } |
05f3fb3c JA |
7723 | } |
7724 | ||
05589553 | 7725 | if (needs_switch) { |
b2e96852 | 7726 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7727 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7728 | } else |
269bbe5f | 7729 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7730 | |
7731 | return done ? done : err; | |
7732 | } | |
05589553 | 7733 | |
05f3fb3c JA |
7734 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7735 | unsigned nr_args) | |
7736 | { | |
269bbe5f | 7737 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7738 | |
7739 | if (!ctx->file_data) | |
7740 | return -ENXIO; | |
7741 | if (!nr_args) | |
7742 | return -EINVAL; | |
7743 | if (copy_from_user(&up, arg, sizeof(up))) | |
7744 | return -EFAULT; | |
7745 | if (up.resv) | |
7746 | return -EINVAL; | |
7747 | ||
7748 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7749 | } | |
c3a31e60 | 7750 | |
5280f7e5 | 7751 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7752 | { |
7753 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7754 | ||
5280f7e5 PB |
7755 | req = io_put_req_find_next(req); |
7756 | return req ? &req->work : NULL; | |
7d723065 JA |
7757 | } |
7758 | ||
5aa75ed5 | 7759 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7760 | { |
e941894e | 7761 | struct io_wq_hash *hash; |
24369c2e | 7762 | struct io_wq_data data; |
24369c2e | 7763 | unsigned int concurrency; |
24369c2e | 7764 | |
e941894e JA |
7765 | hash = ctx->hash_map; |
7766 | if (!hash) { | |
7767 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7768 | if (!hash) | |
7769 | return ERR_PTR(-ENOMEM); | |
7770 | refcount_set(&hash->refs, 1); | |
7771 | init_waitqueue_head(&hash->wait); | |
7772 | ctx->hash_map = hash; | |
24369c2e PB |
7773 | } |
7774 | ||
e941894e | 7775 | data.hash = hash; |
e9fd9396 | 7776 | data.free_work = io_free_work; |
f5fa38c5 | 7777 | data.do_work = io_wq_submit_work; |
24369c2e | 7778 | |
d25e3a3d JA |
7779 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7780 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7781 | |
5aa75ed5 | 7782 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7783 | } |
7784 | ||
5aa75ed5 JA |
7785 | static int io_uring_alloc_task_context(struct task_struct *task, |
7786 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7787 | { |
7788 | struct io_uring_task *tctx; | |
d8a6df10 | 7789 | int ret; |
0f212204 JA |
7790 | |
7791 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7792 | if (unlikely(!tctx)) | |
7793 | return -ENOMEM; | |
7794 | ||
d8a6df10 JA |
7795 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7796 | if (unlikely(ret)) { | |
7797 | kfree(tctx); | |
7798 | return ret; | |
7799 | } | |
7800 | ||
5aa75ed5 JA |
7801 | tctx->io_wq = io_init_wq_offload(ctx); |
7802 | if (IS_ERR(tctx->io_wq)) { | |
7803 | ret = PTR_ERR(tctx->io_wq); | |
7804 | percpu_counter_destroy(&tctx->inflight); | |
7805 | kfree(tctx); | |
7806 | return ret; | |
7807 | } | |
7808 | ||
0f212204 JA |
7809 | xa_init(&tctx->xa); |
7810 | init_waitqueue_head(&tctx->wait); | |
7811 | tctx->last = NULL; | |
fdaf083c JA |
7812 | atomic_set(&tctx->in_idle, 0); |
7813 | tctx->sqpoll = false; | |
0f212204 | 7814 | task->io_uring = tctx; |
7cbf1722 JA |
7815 | spin_lock_init(&tctx->task_lock); |
7816 | INIT_WQ_LIST(&tctx->task_list); | |
7817 | tctx->task_state = 0; | |
7818 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7819 | return 0; |
7820 | } | |
7821 | ||
7822 | void __io_uring_free(struct task_struct *tsk) | |
7823 | { | |
7824 | struct io_uring_task *tctx = tsk->io_uring; | |
7825 | ||
7826 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7827 | WARN_ON_ONCE(tctx->io_wq); |
7828 | ||
d8a6df10 | 7829 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7830 | kfree(tctx); |
7831 | tsk->io_uring = NULL; | |
7832 | } | |
7833 | ||
5f3f26f9 JA |
7834 | static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx) |
7835 | { | |
7836 | int ret; | |
7837 | ||
7838 | clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7839 | reinit_completion(&sqd->completion); | |
70aacfe6 | 7840 | ctx->sqo_exec = 0; |
5f3f26f9 JA |
7841 | sqd->task_pid = current->pid; |
7842 | current->flags |= PF_IO_WORKER; | |
7843 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7844 | current->flags &= ~PF_IO_WORKER; | |
7845 | if (ret < 0) { | |
7846 | sqd->thread = NULL; | |
7847 | return ret; | |
7848 | } | |
7849 | wait_for_completion(&sqd->completion); | |
7850 | return io_uring_alloc_task_context(sqd->thread, ctx); | |
7851 | } | |
7852 | ||
7e84e1c7 SG |
7853 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7854 | struct io_uring_params *p) | |
2b188cc1 JA |
7855 | { |
7856 | int ret; | |
7857 | ||
d25e3a3d JA |
7858 | /* Retain compatibility with failing for an invalid attach attempt */ |
7859 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7860 | IORING_SETUP_ATTACH_WQ) { | |
7861 | struct fd f; | |
7862 | ||
7863 | f = fdget(p->wq_fd); | |
7864 | if (!f.file) | |
7865 | return -ENXIO; | |
7866 | if (f.file->f_op != &io_uring_fops) { | |
7867 | fdput(f); | |
7868 | return -EINVAL; | |
7869 | } | |
7870 | fdput(f); | |
7871 | } | |
6c271ce2 | 7872 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7873 | struct io_sq_data *sqd; |
7874 | ||
3ec482d1 | 7875 | ret = -EPERM; |
ce59fc69 | 7876 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7877 | goto err; |
7878 | ||
534ca6d6 JA |
7879 | sqd = io_get_sq_data(p); |
7880 | if (IS_ERR(sqd)) { | |
7881 | ret = PTR_ERR(sqd); | |
7882 | goto err; | |
7883 | } | |
69fb2131 | 7884 | |
534ca6d6 | 7885 | ctx->sq_data = sqd; |
69fb2131 JA |
7886 | io_sq_thread_park(sqd); |
7887 | mutex_lock(&sqd->ctx_lock); | |
7888 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7889 | mutex_unlock(&sqd->ctx_lock); | |
7890 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7891 | |
917257da JA |
7892 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7893 | if (!ctx->sq_thread_idle) | |
7894 | ctx->sq_thread_idle = HZ; | |
7895 | ||
aa06165d | 7896 | if (sqd->thread) |
5aa75ed5 | 7897 | return 0; |
aa06165d | 7898 | |
6c271ce2 | 7899 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7900 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7901 | |
917257da | 7902 | ret = -EINVAL; |
44a9bd18 JA |
7903 | if (cpu >= nr_cpu_ids) |
7904 | goto err; | |
7889f44d | 7905 | if (!cpu_online(cpu)) |
917257da JA |
7906 | goto err; |
7907 | ||
37d1e2e3 | 7908 | sqd->sq_cpu = cpu; |
6c271ce2 | 7909 | } else { |
37d1e2e3 | 7910 | sqd->sq_cpu = -1; |
6c271ce2 | 7911 | } |
37d1e2e3 JA |
7912 | |
7913 | sqd->task_pid = current->pid; | |
7914 | current->flags |= PF_IO_WORKER; | |
7915 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7916 | current->flags &= ~PF_IO_WORKER; | |
7917 | if (ret < 0) { | |
534ca6d6 | 7918 | sqd->thread = NULL; |
6c271ce2 JA |
7919 | goto err; |
7920 | } | |
37d1e2e3 | 7921 | wait_for_completion(&sqd->completion); |
5aa75ed5 | 7922 | ret = io_uring_alloc_task_context(sqd->thread, ctx); |
0f212204 JA |
7923 | if (ret) |
7924 | goto err; | |
6c271ce2 JA |
7925 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7926 | /* Can't have SQ_AFF without SQPOLL */ | |
7927 | ret = -EINVAL; | |
7928 | goto err; | |
7929 | } | |
7930 | ||
2b188cc1 JA |
7931 | return 0; |
7932 | err: | |
37d1e2e3 | 7933 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
7934 | return ret; |
7935 | } | |
7936 | ||
7e84e1c7 SG |
7937 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7938 | { | |
534ca6d6 JA |
7939 | struct io_sq_data *sqd = ctx->sq_data; |
7940 | ||
3ebba796 | 7941 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
eb85890b | 7942 | if (ctx->flags & IORING_SETUP_SQPOLL) |
37d1e2e3 | 7943 | complete(&sqd->startup); |
7e84e1c7 SG |
7944 | } |
7945 | ||
a087e2b5 BM |
7946 | static inline void __io_unaccount_mem(struct user_struct *user, |
7947 | unsigned long nr_pages) | |
2b188cc1 JA |
7948 | { |
7949 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7950 | } | |
7951 | ||
a087e2b5 BM |
7952 | static inline int __io_account_mem(struct user_struct *user, |
7953 | unsigned long nr_pages) | |
2b188cc1 JA |
7954 | { |
7955 | unsigned long page_limit, cur_pages, new_pages; | |
7956 | ||
7957 | /* Don't allow more pages than we can safely lock */ | |
7958 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7959 | ||
7960 | do { | |
7961 | cur_pages = atomic_long_read(&user->locked_vm); | |
7962 | new_pages = cur_pages + nr_pages; | |
7963 | if (new_pages > page_limit) | |
7964 | return -ENOMEM; | |
7965 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7966 | new_pages) != cur_pages); | |
7967 | ||
7968 | return 0; | |
7969 | } | |
7970 | ||
26bfa89e | 7971 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7972 | { |
62e398be | 7973 | if (ctx->user) |
a087e2b5 | 7974 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7975 | |
26bfa89e JA |
7976 | if (ctx->mm_account) |
7977 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7978 | } |
7979 | ||
26bfa89e | 7980 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7981 | { |
30975825 BM |
7982 | int ret; |
7983 | ||
62e398be | 7984 | if (ctx->user) { |
30975825 BM |
7985 | ret = __io_account_mem(ctx->user, nr_pages); |
7986 | if (ret) | |
7987 | return ret; | |
7988 | } | |
7989 | ||
26bfa89e JA |
7990 | if (ctx->mm_account) |
7991 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7992 | |
7993 | return 0; | |
7994 | } | |
7995 | ||
2b188cc1 JA |
7996 | static void io_mem_free(void *ptr) |
7997 | { | |
52e04ef4 MR |
7998 | struct page *page; |
7999 | ||
8000 | if (!ptr) | |
8001 | return; | |
2b188cc1 | 8002 | |
52e04ef4 | 8003 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8004 | if (put_page_testzero(page)) |
8005 | free_compound_page(page); | |
8006 | } | |
8007 | ||
8008 | static void *io_mem_alloc(size_t size) | |
8009 | { | |
8010 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8011 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8012 | |
8013 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8014 | } | |
8015 | ||
75b28aff HV |
8016 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8017 | size_t *sq_offset) | |
8018 | { | |
8019 | struct io_rings *rings; | |
8020 | size_t off, sq_array_size; | |
8021 | ||
8022 | off = struct_size(rings, cqes, cq_entries); | |
8023 | if (off == SIZE_MAX) | |
8024 | return SIZE_MAX; | |
8025 | ||
8026 | #ifdef CONFIG_SMP | |
8027 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8028 | if (off == 0) | |
8029 | return SIZE_MAX; | |
8030 | #endif | |
8031 | ||
b36200f5 DV |
8032 | if (sq_offset) |
8033 | *sq_offset = off; | |
8034 | ||
75b28aff HV |
8035 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8036 | if (sq_array_size == SIZE_MAX) | |
8037 | return SIZE_MAX; | |
8038 | ||
8039 | if (check_add_overflow(off, sq_array_size, &off)) | |
8040 | return SIZE_MAX; | |
8041 | ||
75b28aff HV |
8042 | return off; |
8043 | } | |
8044 | ||
0a96bbe4 | 8045 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8046 | { |
8047 | int i, j; | |
8048 | ||
8049 | if (!ctx->user_bufs) | |
8050 | return -ENXIO; | |
8051 | ||
8052 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8053 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8054 | ||
8055 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8056 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8057 | |
de293938 | 8058 | if (imu->acct_pages) |
26bfa89e | 8059 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8060 | kvfree(imu->bvec); |
edafccee JA |
8061 | imu->nr_bvecs = 0; |
8062 | } | |
8063 | ||
8064 | kfree(ctx->user_bufs); | |
8065 | ctx->user_bufs = NULL; | |
8066 | ctx->nr_user_bufs = 0; | |
8067 | return 0; | |
8068 | } | |
8069 | ||
8070 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8071 | void __user *arg, unsigned index) | |
8072 | { | |
8073 | struct iovec __user *src; | |
8074 | ||
8075 | #ifdef CONFIG_COMPAT | |
8076 | if (ctx->compat) { | |
8077 | struct compat_iovec __user *ciovs; | |
8078 | struct compat_iovec ciov; | |
8079 | ||
8080 | ciovs = (struct compat_iovec __user *) arg; | |
8081 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8082 | return -EFAULT; | |
8083 | ||
d55e5f5b | 8084 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8085 | dst->iov_len = ciov.iov_len; |
8086 | return 0; | |
8087 | } | |
8088 | #endif | |
8089 | src = (struct iovec __user *) arg; | |
8090 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8091 | return -EFAULT; | |
8092 | return 0; | |
8093 | } | |
8094 | ||
de293938 JA |
8095 | /* |
8096 | * Not super efficient, but this is just a registration time. And we do cache | |
8097 | * the last compound head, so generally we'll only do a full search if we don't | |
8098 | * match that one. | |
8099 | * | |
8100 | * We check if the given compound head page has already been accounted, to | |
8101 | * avoid double accounting it. This allows us to account the full size of the | |
8102 | * page, not just the constituent pages of a huge page. | |
8103 | */ | |
8104 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8105 | int nr_pages, struct page *hpage) | |
8106 | { | |
8107 | int i, j; | |
8108 | ||
8109 | /* check current page array */ | |
8110 | for (i = 0; i < nr_pages; i++) { | |
8111 | if (!PageCompound(pages[i])) | |
8112 | continue; | |
8113 | if (compound_head(pages[i]) == hpage) | |
8114 | return true; | |
8115 | } | |
8116 | ||
8117 | /* check previously registered pages */ | |
8118 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8119 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8120 | ||
8121 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8122 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8123 | continue; | |
8124 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8125 | return true; | |
8126 | } | |
8127 | } | |
8128 | ||
8129 | return false; | |
8130 | } | |
8131 | ||
8132 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8133 | int nr_pages, struct io_mapped_ubuf *imu, | |
8134 | struct page **last_hpage) | |
8135 | { | |
8136 | int i, ret; | |
8137 | ||
8138 | for (i = 0; i < nr_pages; i++) { | |
8139 | if (!PageCompound(pages[i])) { | |
8140 | imu->acct_pages++; | |
8141 | } else { | |
8142 | struct page *hpage; | |
8143 | ||
8144 | hpage = compound_head(pages[i]); | |
8145 | if (hpage == *last_hpage) | |
8146 | continue; | |
8147 | *last_hpage = hpage; | |
8148 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8149 | continue; | |
8150 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8151 | } | |
8152 | } | |
8153 | ||
8154 | if (!imu->acct_pages) | |
8155 | return 0; | |
8156 | ||
26bfa89e | 8157 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8158 | if (ret) |
8159 | imu->acct_pages = 0; | |
8160 | return ret; | |
8161 | } | |
8162 | ||
0a96bbe4 BM |
8163 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8164 | struct io_mapped_ubuf *imu, | |
8165 | struct page **last_hpage) | |
edafccee JA |
8166 | { |
8167 | struct vm_area_struct **vmas = NULL; | |
8168 | struct page **pages = NULL; | |
0a96bbe4 BM |
8169 | unsigned long off, start, end, ubuf; |
8170 | size_t size; | |
8171 | int ret, pret, nr_pages, i; | |
8172 | ||
8173 | ubuf = (unsigned long) iov->iov_base; | |
8174 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8175 | start = ubuf >> PAGE_SHIFT; | |
8176 | nr_pages = end - start; | |
8177 | ||
8178 | ret = -ENOMEM; | |
8179 | ||
8180 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8181 | if (!pages) | |
8182 | goto done; | |
8183 | ||
8184 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8185 | GFP_KERNEL); | |
8186 | if (!vmas) | |
8187 | goto done; | |
edafccee | 8188 | |
0a96bbe4 BM |
8189 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8190 | GFP_KERNEL); | |
8191 | if (!imu->bvec) | |
8192 | goto done; | |
8193 | ||
8194 | ret = 0; | |
8195 | mmap_read_lock(current->mm); | |
8196 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8197 | pages, vmas); | |
8198 | if (pret == nr_pages) { | |
8199 | /* don't support file backed memory */ | |
8200 | for (i = 0; i < nr_pages; i++) { | |
8201 | struct vm_area_struct *vma = vmas[i]; | |
8202 | ||
8203 | if (vma->vm_file && | |
8204 | !is_file_hugepages(vma->vm_file)) { | |
8205 | ret = -EOPNOTSUPP; | |
8206 | break; | |
8207 | } | |
8208 | } | |
8209 | } else { | |
8210 | ret = pret < 0 ? pret : -EFAULT; | |
8211 | } | |
8212 | mmap_read_unlock(current->mm); | |
8213 | if (ret) { | |
8214 | /* | |
8215 | * if we did partial map, or found file backed vmas, | |
8216 | * release any pages we did get | |
8217 | */ | |
8218 | if (pret > 0) | |
8219 | unpin_user_pages(pages, pret); | |
8220 | kvfree(imu->bvec); | |
8221 | goto done; | |
8222 | } | |
8223 | ||
8224 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8225 | if (ret) { | |
8226 | unpin_user_pages(pages, pret); | |
8227 | kvfree(imu->bvec); | |
8228 | goto done; | |
8229 | } | |
8230 | ||
8231 | off = ubuf & ~PAGE_MASK; | |
8232 | size = iov->iov_len; | |
8233 | for (i = 0; i < nr_pages; i++) { | |
8234 | size_t vec_len; | |
8235 | ||
8236 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8237 | imu->bvec[i].bv_page = pages[i]; | |
8238 | imu->bvec[i].bv_len = vec_len; | |
8239 | imu->bvec[i].bv_offset = off; | |
8240 | off = 0; | |
8241 | size -= vec_len; | |
8242 | } | |
8243 | /* store original address for later verification */ | |
8244 | imu->ubuf = ubuf; | |
8245 | imu->len = iov->iov_len; | |
8246 | imu->nr_bvecs = nr_pages; | |
8247 | ret = 0; | |
8248 | done: | |
8249 | kvfree(pages); | |
8250 | kvfree(vmas); | |
8251 | return ret; | |
8252 | } | |
8253 | ||
2b358604 | 8254 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8255 | { |
edafccee JA |
8256 | if (ctx->user_bufs) |
8257 | return -EBUSY; | |
8258 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8259 | return -EINVAL; | |
8260 | ||
8261 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8262 | GFP_KERNEL); | |
8263 | if (!ctx->user_bufs) | |
8264 | return -ENOMEM; | |
8265 | ||
2b358604 BM |
8266 | return 0; |
8267 | } | |
edafccee | 8268 | |
2b358604 BM |
8269 | static int io_buffer_validate(struct iovec *iov) |
8270 | { | |
8271 | /* | |
8272 | * Don't impose further limits on the size and buffer | |
8273 | * constraints here, we'll -EINVAL later when IO is | |
8274 | * submitted if they are wrong. | |
8275 | */ | |
8276 | if (!iov->iov_base || !iov->iov_len) | |
8277 | return -EFAULT; | |
edafccee | 8278 | |
2b358604 BM |
8279 | /* arbitrary limit, but we need something */ |
8280 | if (iov->iov_len > SZ_1G) | |
8281 | return -EFAULT; | |
edafccee | 8282 | |
2b358604 BM |
8283 | return 0; |
8284 | } | |
edafccee | 8285 | |
2b358604 BM |
8286 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8287 | unsigned int nr_args) | |
8288 | { | |
8289 | int i, ret; | |
8290 | struct iovec iov; | |
8291 | struct page *last_hpage = NULL; | |
edafccee | 8292 | |
2b358604 BM |
8293 | ret = io_buffers_map_alloc(ctx, nr_args); |
8294 | if (ret) | |
8295 | return ret; | |
edafccee | 8296 | |
edafccee JA |
8297 | for (i = 0; i < nr_args; i++) { |
8298 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8299 | |
edafccee JA |
8300 | ret = io_copy_iov(ctx, &iov, arg, i); |
8301 | if (ret) | |
0a96bbe4 | 8302 | break; |
de293938 | 8303 | |
2b358604 BM |
8304 | ret = io_buffer_validate(&iov); |
8305 | if (ret) | |
0a96bbe4 | 8306 | break; |
edafccee | 8307 | |
0a96bbe4 BM |
8308 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8309 | if (ret) | |
8310 | break; | |
edafccee JA |
8311 | |
8312 | ctx->nr_user_bufs++; | |
8313 | } | |
0a96bbe4 BM |
8314 | |
8315 | if (ret) | |
8316 | io_sqe_buffers_unregister(ctx); | |
8317 | ||
edafccee JA |
8318 | return ret; |
8319 | } | |
8320 | ||
9b402849 JA |
8321 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8322 | { | |
8323 | __s32 __user *fds = arg; | |
8324 | int fd; | |
8325 | ||
8326 | if (ctx->cq_ev_fd) | |
8327 | return -EBUSY; | |
8328 | ||
8329 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8330 | return -EFAULT; | |
8331 | ||
8332 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8333 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8334 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8335 | ctx->cq_ev_fd = NULL; | |
8336 | return ret; | |
8337 | } | |
8338 | ||
8339 | return 0; | |
8340 | } | |
8341 | ||
8342 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8343 | { | |
8344 | if (ctx->cq_ev_fd) { | |
8345 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8346 | ctx->cq_ev_fd = NULL; | |
8347 | return 0; | |
8348 | } | |
8349 | ||
8350 | return -ENXIO; | |
8351 | } | |
8352 | ||
5a2e745d JA |
8353 | static int __io_destroy_buffers(int id, void *p, void *data) |
8354 | { | |
8355 | struct io_ring_ctx *ctx = data; | |
8356 | struct io_buffer *buf = p; | |
8357 | ||
067524e9 | 8358 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8359 | return 0; |
8360 | } | |
8361 | ||
8362 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8363 | { | |
8364 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8365 | idr_destroy(&ctx->io_buffer_idr); | |
8366 | } | |
8367 | ||
68e68ee6 | 8368 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8369 | { |
68e68ee6 | 8370 | struct io_kiocb *req, *nxt; |
1b4c351f | 8371 | |
68e68ee6 JA |
8372 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8373 | if (tsk && req->task != tsk) | |
8374 | continue; | |
1b4c351f JA |
8375 | list_del(&req->compl.list); |
8376 | kmem_cache_free(req_cachep, req); | |
8377 | } | |
8378 | } | |
8379 | ||
4010fec4 | 8380 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8381 | { |
bf019da7 | 8382 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8383 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8384 | |
9a4fdbd8 JA |
8385 | mutex_lock(&ctx->uring_lock); |
8386 | ||
8e5c66c4 | 8387 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8388 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8389 | submit_state->reqs); | |
8e5c66c4 PB |
8390 | submit_state->free_reqs = 0; |
8391 | } | |
9a4fdbd8 JA |
8392 | |
8393 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8394 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8395 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8396 | spin_unlock_irq(&ctx->completion_lock); |
8397 | ||
e5547d2c PB |
8398 | io_req_cache_free(&cs->free_list, NULL); |
8399 | ||
9a4fdbd8 JA |
8400 | mutex_unlock(&ctx->uring_lock); |
8401 | } | |
8402 | ||
2b188cc1 JA |
8403 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8404 | { | |
04fc6c80 PB |
8405 | /* |
8406 | * Some may use context even when all refs and requests have been put, | |
8407 | * and they are free to do so while still holding uring_lock, see | |
8408 | * __io_req_task_submit(). Wait for them to finish. | |
8409 | */ | |
8410 | mutex_lock(&ctx->uring_lock); | |
8411 | mutex_unlock(&ctx->uring_lock); | |
8412 | ||
37d1e2e3 | 8413 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8414 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8415 | |
37d1e2e3 | 8416 | if (ctx->mm_account) { |
2aede0e4 JA |
8417 | mmdrop(ctx->mm_account); |
8418 | ctx->mm_account = NULL; | |
30975825 | 8419 | } |
def596e9 | 8420 | |
8bad28d8 | 8421 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8422 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8423 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8424 | io_eventfd_unregister(ctx); |
5a2e745d | 8425 | io_destroy_buffers(ctx); |
41726c9a | 8426 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8427 | |
2b188cc1 | 8428 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8429 | if (ctx->ring_sock) { |
8430 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8431 | sock_release(ctx->ring_sock); |
355e8d26 | 8432 | } |
2b188cc1 JA |
8433 | #endif |
8434 | ||
75b28aff | 8435 | io_mem_free(ctx->rings); |
2b188cc1 | 8436 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8437 | |
8438 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8439 | free_uid(ctx->user); |
4010fec4 | 8440 | io_req_caches_free(ctx); |
e941894e JA |
8441 | if (ctx->hash_map) |
8442 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8443 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8444 | kfree(ctx); |
8445 | } | |
8446 | ||
8447 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8448 | { | |
8449 | struct io_ring_ctx *ctx = file->private_data; | |
8450 | __poll_t mask = 0; | |
8451 | ||
8452 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8453 | /* |
8454 | * synchronizes with barrier from wq_has_sleeper call in | |
8455 | * io_commit_cqring | |
8456 | */ | |
2b188cc1 | 8457 | smp_rmb(); |
90554200 | 8458 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8459 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8460 | |
8461 | /* | |
8462 | * Don't flush cqring overflow list here, just do a simple check. | |
8463 | * Otherwise there could possible be ABBA deadlock: | |
8464 | * CPU0 CPU1 | |
8465 | * ---- ---- | |
8466 | * lock(&ctx->uring_lock); | |
8467 | * lock(&ep->mtx); | |
8468 | * lock(&ctx->uring_lock); | |
8469 | * lock(&ep->mtx); | |
8470 | * | |
8471 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8472 | * pushs them to do the flush. | |
8473 | */ | |
8474 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8475 | mask |= EPOLLIN | EPOLLRDNORM; |
8476 | ||
8477 | return mask; | |
8478 | } | |
8479 | ||
8480 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8481 | { | |
8482 | struct io_ring_ctx *ctx = file->private_data; | |
8483 | ||
8484 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8485 | } | |
8486 | ||
0bead8cd | 8487 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8488 | { |
4379bf8b | 8489 | const struct cred *creds; |
071698e1 | 8490 | |
4379bf8b JA |
8491 | creds = idr_remove(&ctx->personality_idr, id); |
8492 | if (creds) { | |
8493 | put_cred(creds); | |
0bead8cd | 8494 | return 0; |
1e6fa521 | 8495 | } |
0bead8cd YD |
8496 | |
8497 | return -EINVAL; | |
8498 | } | |
8499 | ||
8500 | static int io_remove_personalities(int id, void *p, void *data) | |
8501 | { | |
8502 | struct io_ring_ctx *ctx = data; | |
8503 | ||
8504 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8505 | return 0; |
8506 | } | |
8507 | ||
ba50a036 | 8508 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8509 | { |
28c4721b | 8510 | struct callback_head *work, *next; |
ba50a036 | 8511 | bool executed = false; |
7c25c0d1 JA |
8512 | |
8513 | do { | |
28c4721b | 8514 | work = xchg(&ctx->exit_task_work, NULL); |
7c25c0d1 JA |
8515 | if (!work) |
8516 | break; | |
8517 | ||
8518 | do { | |
8519 | next = work->next; | |
8520 | work->func(work); | |
8521 | work = next; | |
8522 | cond_resched(); | |
8523 | } while (work); | |
ba50a036 | 8524 | executed = true; |
7c25c0d1 | 8525 | } while (1); |
ba50a036 PB |
8526 | |
8527 | return executed; | |
7c25c0d1 JA |
8528 | } |
8529 | ||
85faa7b8 JA |
8530 | static void io_ring_exit_work(struct work_struct *work) |
8531 | { | |
b2edc0a7 PB |
8532 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8533 | exit_work); | |
85faa7b8 | 8534 | |
56952e91 JA |
8535 | /* |
8536 | * If we're doing polled IO and end up having requests being | |
8537 | * submitted async (out-of-line), then completions can come in while | |
8538 | * we're waiting for refs to drop. We need to reap these manually, | |
8539 | * as nobody else will be looking for them. | |
8540 | */ | |
b2edc0a7 | 8541 | do { |
9936c7c2 | 8542 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8543 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8544 | io_ring_ctx_free(ctx); |
8545 | } | |
8546 | ||
2b188cc1 JA |
8547 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8548 | { | |
8549 | mutex_lock(&ctx->uring_lock); | |
8550 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8551 | /* if force is set, the ring is going away. always drop after that */ |
8552 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8553 | if (ctx->rings) |
6c503150 | 8554 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8555 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8556 | mutex_unlock(&ctx->uring_lock); |
8557 | ||
6b81928d PB |
8558 | io_kill_timeouts(ctx, NULL, NULL); |
8559 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8560 | |
15dff286 | 8561 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8562 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8563 | |
85faa7b8 | 8564 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8565 | /* |
8566 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8567 | * if we're exiting a ton of rings at the same time. It just adds | |
8568 | * noise and overhead, there's no discernable change in runtime | |
8569 | * over using system_wq. | |
8570 | */ | |
8571 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8572 | } |
8573 | ||
8574 | static int io_uring_release(struct inode *inode, struct file *file) | |
8575 | { | |
8576 | struct io_ring_ctx *ctx = file->private_data; | |
8577 | ||
8578 | file->private_data = NULL; | |
8579 | io_ring_ctx_wait_and_kill(ctx); | |
8580 | return 0; | |
8581 | } | |
8582 | ||
f6edbabb PB |
8583 | struct io_task_cancel { |
8584 | struct task_struct *task; | |
8585 | struct files_struct *files; | |
8586 | }; | |
f254ac04 | 8587 | |
f6edbabb | 8588 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8589 | { |
9a472ef7 | 8590 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8591 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8592 | bool ret; |
8593 | ||
f6edbabb | 8594 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8595 | unsigned long flags; |
8596 | struct io_ring_ctx *ctx = req->ctx; | |
8597 | ||
8598 | /* protect against races with linked timeouts */ | |
8599 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8600 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8601 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8602 | } else { | |
f6edbabb | 8603 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8604 | } |
8605 | return ret; | |
b711d4ea JA |
8606 | } |
8607 | ||
b7ddce3c | 8608 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8609 | struct task_struct *task, |
b7ddce3c PB |
8610 | struct files_struct *files) |
8611 | { | |
8612 | struct io_defer_entry *de = NULL; | |
8613 | LIST_HEAD(list); | |
8614 | ||
8615 | spin_lock_irq(&ctx->completion_lock); | |
8616 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8617 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8618 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8619 | break; | |
8620 | } | |
8621 | } | |
8622 | spin_unlock_irq(&ctx->completion_lock); | |
8623 | ||
8624 | while (!list_empty(&list)) { | |
8625 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8626 | list_del_init(&de->list); | |
8627 | req_set_fail_links(de->req); | |
8628 | io_put_req(de->req); | |
8629 | io_req_complete(de->req, -ECANCELED); | |
8630 | kfree(de); | |
8631 | } | |
8632 | } | |
8633 | ||
9936c7c2 PB |
8634 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8635 | struct task_struct *task, | |
8636 | struct files_struct *files) | |
8637 | { | |
8638 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
64c72123 PB |
8639 | struct task_struct *tctx_task = task ?: current; |
8640 | struct io_uring_task *tctx = tctx_task->io_uring; | |
9936c7c2 PB |
8641 | |
8642 | while (1) { | |
8643 | enum io_wq_cancel cret; | |
8644 | bool ret = false; | |
8645 | ||
5aa75ed5 JA |
8646 | if (tctx && tctx->io_wq) { |
8647 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, | |
9936c7c2 PB |
8648 | &cancel, true); |
8649 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8650 | } | |
8651 | ||
8652 | /* SQPOLL thread does its own polling */ | |
8653 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8654 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8655 | io_iopoll_try_reap_events(ctx); | |
8656 | ret = true; | |
8657 | } | |
8658 | } | |
8659 | ||
8660 | ret |= io_poll_remove_all(ctx, task, files); | |
8661 | ret |= io_kill_timeouts(ctx, task, files); | |
8662 | ret |= io_run_task_work(); | |
ba50a036 | 8663 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8664 | io_cqring_overflow_flush(ctx, true, task, files); |
8665 | if (!ret) | |
8666 | break; | |
8667 | cond_resched(); | |
8668 | } | |
8669 | } | |
8670 | ||
ca70f00b PB |
8671 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8672 | struct task_struct *task, | |
8673 | struct files_struct *files) | |
8674 | { | |
8675 | struct io_kiocb *req; | |
8676 | int cnt = 0; | |
8677 | ||
8678 | spin_lock_irq(&ctx->inflight_lock); | |
8679 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8680 | cnt += io_match_task(req, task, files); | |
8681 | spin_unlock_irq(&ctx->inflight_lock); | |
8682 | return cnt; | |
8683 | } | |
8684 | ||
b52fda00 | 8685 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8686 | struct task_struct *task, |
fcb323cc JA |
8687 | struct files_struct *files) |
8688 | { | |
fcb323cc | 8689 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8690 | DEFINE_WAIT(wait); |
ca70f00b | 8691 | int inflight; |
fcb323cc | 8692 | |
ca70f00b PB |
8693 | inflight = io_uring_count_inflight(ctx, task, files); |
8694 | if (!inflight) | |
fcb323cc | 8695 | break; |
f6edbabb | 8696 | |
9936c7c2 | 8697 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b | 8698 | |
34343786 PB |
8699 | if (ctx->sq_data) |
8700 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
8701 | prepare_to_wait(&task->io_uring->wait, &wait, |
8702 | TASK_UNINTERRUPTIBLE); | |
8703 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8704 | schedule(); | |
c98de08c | 8705 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
8706 | if (ctx->sq_data) |
8707 | io_sq_thread_park(ctx->sq_data); | |
0f212204 | 8708 | } |
0f212204 JA |
8709 | } |
8710 | ||
8711 | /* | |
8712 | * We need to iteratively cancel requests, in case a request has dependent | |
8713 | * hard links. These persist even for failure of cancelations, hence keep | |
8714 | * looping until none are found. | |
8715 | */ | |
8716 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8717 | struct files_struct *files) | |
8718 | { | |
8719 | struct task_struct *task = current; | |
37d1e2e3 | 8720 | bool did_park = false; |
0f212204 | 8721 | |
fdaf083c | 8722 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
70aacfe6 PB |
8723 | /* never started, nothing to cancel */ |
8724 | if (ctx->flags & IORING_SETUP_R_DISABLED) { | |
8725 | io_sq_offload_start(ctx); | |
8726 | return; | |
8727 | } | |
37d1e2e3 JA |
8728 | did_park = io_sq_thread_park(ctx->sq_data); |
8729 | if (did_park) { | |
8730 | task = ctx->sq_data->thread; | |
8731 | atomic_inc(&task->io_uring->in_idle); | |
8732 | } | |
fdaf083c | 8733 | } |
0f212204 | 8734 | |
df9923f9 | 8735 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 8736 | |
3a7efd1a | 8737 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 8738 | if (!files) |
9936c7c2 | 8739 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c | 8740 | |
37d1e2e3 | 8741 | if (did_park) { |
fdaf083c | 8742 | atomic_dec(&task->io_uring->in_idle); |
fdaf083c JA |
8743 | io_sq_thread_unpark(ctx->sq_data); |
8744 | } | |
0f212204 JA |
8745 | } |
8746 | ||
8747 | /* | |
8748 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8749 | */ | |
fdaf083c | 8750 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 8751 | { |
236434c3 | 8752 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 8753 | int ret; |
236434c3 MWO |
8754 | |
8755 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8756 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8757 | if (unlikely(ret)) |
8758 | return ret; | |
236434c3 | 8759 | tctx = current->io_uring; |
0f212204 | 8760 | } |
236434c3 MWO |
8761 | if (tctx->last != file) { |
8762 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8763 | |
236434c3 | 8764 | if (!old) { |
0f212204 | 8765 | get_file(file); |
a528b04e PB |
8766 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
8767 | file, GFP_KERNEL)); | |
8768 | if (ret) { | |
8769 | fput(file); | |
8770 | return ret; | |
8771 | } | |
0f212204 | 8772 | } |
236434c3 | 8773 | tctx->last = file; |
0f212204 JA |
8774 | } |
8775 | ||
fdaf083c JA |
8776 | /* |
8777 | * This is race safe in that the task itself is doing this, hence it | |
8778 | * cannot be going through the exit/cancel paths at the same time. | |
8779 | * This cannot be modified while exit/cancel is running. | |
8780 | */ | |
8781 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
8782 | tctx->sqpoll = true; | |
8783 | ||
0f212204 JA |
8784 | return 0; |
8785 | } | |
8786 | ||
8787 | /* | |
8788 | * Remove this io_uring_file -> task mapping. | |
8789 | */ | |
8790 | static void io_uring_del_task_file(struct file *file) | |
8791 | { | |
8792 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8793 | |
8794 | if (tctx->last == file) | |
8795 | tctx->last = NULL; | |
5e2ed8c4 | 8796 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8797 | if (file) |
8798 | fput(file); | |
8799 | } | |
8800 | ||
8452d4a6 | 8801 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e PB |
8802 | { |
8803 | struct file *file; | |
8804 | unsigned long index; | |
8805 | ||
8806 | xa_for_each(&tctx->xa, index, file) | |
8807 | io_uring_del_task_file(file); | |
8452d4a6 PB |
8808 | if (tctx->io_wq) { |
8809 | io_wq_put_and_exit(tctx->io_wq); | |
8810 | tctx->io_wq = NULL; | |
8811 | } | |
de7f1d9e PB |
8812 | } |
8813 | ||
0f212204 JA |
8814 | void __io_uring_files_cancel(struct files_struct *files) |
8815 | { | |
8816 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8817 | struct file *file; |
8818 | unsigned long index; | |
0f212204 JA |
8819 | |
8820 | /* make sure overflow events are dropped */ | |
fdaf083c | 8821 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
8822 | xa_for_each(&tctx->xa, index, file) |
8823 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 8824 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8825 | |
8452d4a6 PB |
8826 | if (files) |
8827 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8828 | } |
8829 | ||
8830 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
8831 | { | |
0e9ddb39 PB |
8832 | return percpu_counter_sum(&tctx->inflight); |
8833 | } | |
fdaf083c | 8834 | |
0e9ddb39 PB |
8835 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8836 | { | |
37d1e2e3 | 8837 | struct io_sq_data *sqd = ctx->sq_data; |
0e9ddb39 PB |
8838 | struct io_uring_task *tctx; |
8839 | s64 inflight; | |
8840 | DEFINE_WAIT(wait); | |
fdaf083c | 8841 | |
37d1e2e3 | 8842 | if (!sqd) |
0e9ddb39 | 8843 | return; |
37d1e2e3 JA |
8844 | if (!io_sq_thread_park(sqd)) |
8845 | return; | |
8846 | tctx = ctx->sq_data->thread->io_uring; | |
e54945ae JA |
8847 | /* can happen on fork/alloc failure, just ignore that state */ |
8848 | if (!tctx) { | |
8849 | io_sq_thread_unpark(sqd); | |
8850 | return; | |
8851 | } | |
fdaf083c | 8852 | |
0e9ddb39 PB |
8853 | atomic_inc(&tctx->in_idle); |
8854 | do { | |
8855 | /* read completions before cancelations */ | |
8856 | inflight = tctx_inflight(tctx); | |
8857 | if (!inflight) | |
8858 | break; | |
8859 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 8860 | |
0e9ddb39 PB |
8861 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8862 | /* | |
8863 | * If we've seen completions, retry without waiting. This | |
8864 | * avoids a race where a completion comes in before we did | |
8865 | * prepare_to_wait(). | |
8866 | */ | |
8867 | if (inflight == tctx_inflight(tctx)) | |
8868 | schedule(); | |
8869 | finish_wait(&tctx->wait, &wait); | |
8870 | } while (1); | |
8871 | atomic_dec(&tctx->in_idle); | |
37d1e2e3 | 8872 | io_sq_thread_unpark(sqd); |
0f212204 JA |
8873 | } |
8874 | ||
0f212204 JA |
8875 | /* |
8876 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8877 | * requests. | |
8878 | */ | |
8879 | void __io_uring_task_cancel(void) | |
8880 | { | |
8881 | struct io_uring_task *tctx = current->io_uring; | |
8882 | DEFINE_WAIT(wait); | |
d8a6df10 | 8883 | s64 inflight; |
0f212204 JA |
8884 | |
8885 | /* make sure overflow events are dropped */ | |
fdaf083c | 8886 | atomic_inc(&tctx->in_idle); |
0f212204 | 8887 | |
0e9ddb39 PB |
8888 | if (tctx->sqpoll) { |
8889 | struct file *file; | |
8890 | unsigned long index; | |
8891 | ||
8892 | xa_for_each(&tctx->xa, index, file) | |
8893 | io_uring_cancel_sqpoll(file->private_data); | |
8894 | } | |
0b5cd6c3 | 8895 | |
d8a6df10 | 8896 | do { |
0f212204 | 8897 | /* read completions before cancelations */ |
fdaf083c | 8898 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8899 | if (!inflight) |
8900 | break; | |
0f212204 JA |
8901 | __io_uring_files_cancel(NULL); |
8902 | ||
8903 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8904 | ||
8905 | /* | |
a1bb3cd5 PB |
8906 | * If we've seen completions, retry without waiting. This |
8907 | * avoids a race where a completion comes in before we did | |
8908 | * prepare_to_wait(). | |
0f212204 | 8909 | */ |
a1bb3cd5 PB |
8910 | if (inflight == tctx_inflight(tctx)) |
8911 | schedule(); | |
f57555ed | 8912 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8913 | } while (1); |
0f212204 | 8914 | |
fdaf083c | 8915 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8916 | |
8452d4a6 PB |
8917 | io_uring_clean_tctx(tctx); |
8918 | /* all current's requests should be gone, we can kill tctx */ | |
8919 | __io_uring_free(current); | |
44e728b8 PB |
8920 | } |
8921 | ||
6c5c240e RP |
8922 | static void *io_uring_validate_mmap_request(struct file *file, |
8923 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8924 | { |
2b188cc1 | 8925 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8926 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8927 | struct page *page; |
8928 | void *ptr; | |
8929 | ||
8930 | switch (offset) { | |
8931 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8932 | case IORING_OFF_CQ_RING: |
8933 | ptr = ctx->rings; | |
2b188cc1 JA |
8934 | break; |
8935 | case IORING_OFF_SQES: | |
8936 | ptr = ctx->sq_sqes; | |
8937 | break; | |
2b188cc1 | 8938 | default: |
6c5c240e | 8939 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8940 | } |
8941 | ||
8942 | page = virt_to_head_page(ptr); | |
a50b854e | 8943 | if (sz > page_size(page)) |
6c5c240e RP |
8944 | return ERR_PTR(-EINVAL); |
8945 | ||
8946 | return ptr; | |
8947 | } | |
8948 | ||
8949 | #ifdef CONFIG_MMU | |
8950 | ||
8951 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8952 | { | |
8953 | size_t sz = vma->vm_end - vma->vm_start; | |
8954 | unsigned long pfn; | |
8955 | void *ptr; | |
8956 | ||
8957 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8958 | if (IS_ERR(ptr)) | |
8959 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8960 | |
8961 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8962 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8963 | } | |
8964 | ||
6c5c240e RP |
8965 | #else /* !CONFIG_MMU */ |
8966 | ||
8967 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8968 | { | |
8969 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8970 | } | |
8971 | ||
8972 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8973 | { | |
8974 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8975 | } | |
8976 | ||
8977 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8978 | unsigned long addr, unsigned long len, | |
8979 | unsigned long pgoff, unsigned long flags) | |
8980 | { | |
8981 | void *ptr; | |
8982 | ||
8983 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8984 | if (IS_ERR(ptr)) | |
8985 | return PTR_ERR(ptr); | |
8986 | ||
8987 | return (unsigned long) ptr; | |
8988 | } | |
8989 | ||
8990 | #endif /* !CONFIG_MMU */ | |
8991 | ||
d9d05217 | 8992 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 8993 | { |
d9d05217 | 8994 | int ret = 0; |
90554200 JA |
8995 | DEFINE_WAIT(wait); |
8996 | ||
8997 | do { | |
8998 | if (!io_sqring_full(ctx)) | |
8999 | break; | |
90554200 JA |
9000 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9001 | ||
9002 | if (!io_sqring_full(ctx)) | |
9003 | break; | |
90554200 JA |
9004 | schedule(); |
9005 | } while (!signal_pending(current)); | |
9006 | ||
9007 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 | 9008 | return ret; |
90554200 JA |
9009 | } |
9010 | ||
c73ebb68 HX |
9011 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9012 | struct __kernel_timespec __user **ts, | |
9013 | const sigset_t __user **sig) | |
9014 | { | |
9015 | struct io_uring_getevents_arg arg; | |
9016 | ||
9017 | /* | |
9018 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9019 | * is just a pointer to the sigset_t. | |
9020 | */ | |
9021 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9022 | *sig = (const sigset_t __user *) argp; | |
9023 | *ts = NULL; | |
9024 | return 0; | |
9025 | } | |
9026 | ||
9027 | /* | |
9028 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9029 | * timespec and sigset_t pointers if good. | |
9030 | */ | |
9031 | if (*argsz != sizeof(arg)) | |
9032 | return -EINVAL; | |
9033 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9034 | return -EFAULT; | |
9035 | *sig = u64_to_user_ptr(arg.sigmask); | |
9036 | *argsz = arg.sigmask_sz; | |
9037 | *ts = u64_to_user_ptr(arg.ts); | |
9038 | return 0; | |
9039 | } | |
9040 | ||
2b188cc1 | 9041 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9042 | u32, min_complete, u32, flags, const void __user *, argp, |
9043 | size_t, argsz) | |
2b188cc1 JA |
9044 | { |
9045 | struct io_ring_ctx *ctx; | |
9046 | long ret = -EBADF; | |
9047 | int submitted = 0; | |
9048 | struct fd f; | |
9049 | ||
4c6e277c | 9050 | io_run_task_work(); |
b41e9852 | 9051 | |
90554200 | 9052 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9053 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9054 | return -EINVAL; |
9055 | ||
9056 | f = fdget(fd); | |
9057 | if (!f.file) | |
9058 | return -EBADF; | |
9059 | ||
9060 | ret = -EOPNOTSUPP; | |
9061 | if (f.file->f_op != &io_uring_fops) | |
9062 | goto out_fput; | |
9063 | ||
9064 | ret = -ENXIO; | |
9065 | ctx = f.file->private_data; | |
9066 | if (!percpu_ref_tryget(&ctx->refs)) | |
9067 | goto out_fput; | |
9068 | ||
7e84e1c7 SG |
9069 | ret = -EBADFD; |
9070 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9071 | goto out; | |
9072 | ||
6c271ce2 JA |
9073 | /* |
9074 | * For SQ polling, the thread will do all submissions and completions. | |
9075 | * Just return the requested submit count, and wake the thread if | |
9076 | * we were asked to. | |
9077 | */ | |
b2a9eada | 9078 | ret = 0; |
6c271ce2 | 9079 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9080 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9081 | |
5f3f26f9 JA |
9082 | if (unlikely(ctx->sqo_exec)) { |
9083 | ret = io_sq_thread_fork(ctx->sq_data, ctx); | |
9084 | if (ret) | |
9085 | goto out; | |
9086 | ctx->sqo_exec = 0; | |
9087 | } | |
d9d05217 | 9088 | ret = -EOWNERDEAD; |
6c271ce2 | 9089 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9090 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9091 | if (flags & IORING_ENTER_SQ_WAIT) { |
9092 | ret = io_sqpoll_wait_sq(ctx); | |
9093 | if (ret) | |
9094 | goto out; | |
9095 | } | |
6c271ce2 | 9096 | submitted = to_submit; |
b2a9eada | 9097 | } else if (to_submit) { |
fdaf083c | 9098 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9099 | if (unlikely(ret)) |
9100 | goto out; | |
2b188cc1 | 9101 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9102 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9103 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9104 | |
9105 | if (submitted != to_submit) | |
9106 | goto out; | |
2b188cc1 JA |
9107 | } |
9108 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9109 | const sigset_t __user *sig; |
9110 | struct __kernel_timespec __user *ts; | |
9111 | ||
9112 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9113 | if (unlikely(ret)) | |
9114 | goto out; | |
9115 | ||
2b188cc1 JA |
9116 | min_complete = min(min_complete, ctx->cq_entries); |
9117 | ||
32b2244a XW |
9118 | /* |
9119 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9120 | * space applications don't need to do io completion events | |
9121 | * polling again, they can rely on io_sq_thread to do polling | |
9122 | * work, which can reduce cpu usage and uring_lock contention. | |
9123 | */ | |
9124 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9125 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9126 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9127 | } else { |
c73ebb68 | 9128 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9129 | } |
2b188cc1 JA |
9130 | } |
9131 | ||
7c504e65 | 9132 | out: |
6805b32e | 9133 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9134 | out_fput: |
9135 | fdput(f); | |
9136 | return submitted ? submitted : ret; | |
9137 | } | |
9138 | ||
bebdb65e | 9139 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9140 | static int io_uring_show_cred(int id, void *p, void *data) |
9141 | { | |
4379bf8b | 9142 | const struct cred *cred = p; |
87ce955b JA |
9143 | struct seq_file *m = data; |
9144 | struct user_namespace *uns = seq_user_ns(m); | |
9145 | struct group_info *gi; | |
9146 | kernel_cap_t cap; | |
9147 | unsigned __capi; | |
9148 | int g; | |
9149 | ||
9150 | seq_printf(m, "%5d\n", id); | |
9151 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9152 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9153 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9154 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9155 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9156 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9157 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9158 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9159 | seq_puts(m, "\n\tGroups:\t"); | |
9160 | gi = cred->group_info; | |
9161 | for (g = 0; g < gi->ngroups; g++) { | |
9162 | seq_put_decimal_ull(m, g ? " " : "", | |
9163 | from_kgid_munged(uns, gi->gid[g])); | |
9164 | } | |
9165 | seq_puts(m, "\n\tCapEff:\t"); | |
9166 | cap = cred->cap_effective; | |
9167 | CAP_FOR_EACH_U32(__capi) | |
9168 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9169 | seq_putc(m, '\n'); | |
9170 | return 0; | |
9171 | } | |
9172 | ||
9173 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9174 | { | |
dbbe9c64 | 9175 | struct io_sq_data *sq = NULL; |
fad8e0de | 9176 | bool has_lock; |
87ce955b JA |
9177 | int i; |
9178 | ||
fad8e0de JA |
9179 | /* |
9180 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9181 | * since fdinfo case grabs it in the opposite direction of normal use | |
9182 | * cases. If we fail to get the lock, we just don't iterate any | |
9183 | * structures that could be going away outside the io_uring mutex. | |
9184 | */ | |
9185 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9186 | ||
5f3f26f9 | 9187 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9188 | sq = ctx->sq_data; |
5f3f26f9 JA |
9189 | if (!sq->thread) |
9190 | sq = NULL; | |
9191 | } | |
dbbe9c64 JQ |
9192 | |
9193 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9194 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9195 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9196 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9197 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9198 | |
87ce955b JA |
9199 | if (f) |
9200 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9201 | else | |
9202 | seq_printf(m, "%5u: <none>\n", i); | |
9203 | } | |
9204 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9205 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9206 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9207 | ||
9208 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9209 | (unsigned int) buf->len); | |
9210 | } | |
fad8e0de | 9211 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9212 | seq_printf(m, "Personalities:\n"); |
9213 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9214 | } | |
d7718a9d JA |
9215 | seq_printf(m, "PollList:\n"); |
9216 | spin_lock_irq(&ctx->completion_lock); | |
9217 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9218 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9219 | struct io_kiocb *req; | |
9220 | ||
9221 | hlist_for_each_entry(req, list, hash_node) | |
9222 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9223 | req->task->task_works != NULL); | |
9224 | } | |
9225 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9226 | if (has_lock) |
9227 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9228 | } |
9229 | ||
9230 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9231 | { | |
9232 | struct io_ring_ctx *ctx = f->private_data; | |
9233 | ||
9234 | if (percpu_ref_tryget(&ctx->refs)) { | |
9235 | __io_uring_show_fdinfo(ctx, m); | |
9236 | percpu_ref_put(&ctx->refs); | |
9237 | } | |
9238 | } | |
bebdb65e | 9239 | #endif |
87ce955b | 9240 | |
2b188cc1 JA |
9241 | static const struct file_operations io_uring_fops = { |
9242 | .release = io_uring_release, | |
9243 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9244 | #ifndef CONFIG_MMU |
9245 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9246 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9247 | #endif | |
2b188cc1 JA |
9248 | .poll = io_uring_poll, |
9249 | .fasync = io_uring_fasync, | |
bebdb65e | 9250 | #ifdef CONFIG_PROC_FS |
87ce955b | 9251 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9252 | #endif |
2b188cc1 JA |
9253 | }; |
9254 | ||
9255 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9256 | struct io_uring_params *p) | |
9257 | { | |
75b28aff HV |
9258 | struct io_rings *rings; |
9259 | size_t size, sq_array_offset; | |
2b188cc1 | 9260 | |
bd740481 JA |
9261 | /* make sure these are sane, as we already accounted them */ |
9262 | ctx->sq_entries = p->sq_entries; | |
9263 | ctx->cq_entries = p->cq_entries; | |
9264 | ||
75b28aff HV |
9265 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9266 | if (size == SIZE_MAX) | |
9267 | return -EOVERFLOW; | |
9268 | ||
9269 | rings = io_mem_alloc(size); | |
9270 | if (!rings) | |
2b188cc1 JA |
9271 | return -ENOMEM; |
9272 | ||
75b28aff HV |
9273 | ctx->rings = rings; |
9274 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9275 | rings->sq_ring_mask = p->sq_entries - 1; | |
9276 | rings->cq_ring_mask = p->cq_entries - 1; | |
9277 | rings->sq_ring_entries = p->sq_entries; | |
9278 | rings->cq_ring_entries = p->cq_entries; | |
9279 | ctx->sq_mask = rings->sq_ring_mask; | |
9280 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9281 | |
9282 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9283 | if (size == SIZE_MAX) { |
9284 | io_mem_free(ctx->rings); | |
9285 | ctx->rings = NULL; | |
2b188cc1 | 9286 | return -EOVERFLOW; |
eb065d30 | 9287 | } |
2b188cc1 JA |
9288 | |
9289 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9290 | if (!ctx->sq_sqes) { |
9291 | io_mem_free(ctx->rings); | |
9292 | ctx->rings = NULL; | |
2b188cc1 | 9293 | return -ENOMEM; |
eb065d30 | 9294 | } |
2b188cc1 | 9295 | |
2b188cc1 JA |
9296 | return 0; |
9297 | } | |
9298 | ||
9faadcc8 PB |
9299 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9300 | { | |
9301 | int ret, fd; | |
9302 | ||
9303 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9304 | if (fd < 0) | |
9305 | return fd; | |
9306 | ||
9307 | ret = io_uring_add_task_file(ctx, file); | |
9308 | if (ret) { | |
9309 | put_unused_fd(fd); | |
9310 | return ret; | |
9311 | } | |
9312 | fd_install(fd, file); | |
9313 | return fd; | |
9314 | } | |
9315 | ||
2b188cc1 JA |
9316 | /* |
9317 | * Allocate an anonymous fd, this is what constitutes the application | |
9318 | * visible backing of an io_uring instance. The application mmaps this | |
9319 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9320 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9321 | */ | |
9faadcc8 | 9322 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9323 | { |
9324 | struct file *file; | |
9faadcc8 | 9325 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9326 | int ret; |
9327 | ||
2b188cc1 JA |
9328 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9329 | &ctx->ring_sock); | |
9330 | if (ret) | |
9faadcc8 | 9331 | return ERR_PTR(ret); |
2b188cc1 JA |
9332 | #endif |
9333 | ||
2b188cc1 JA |
9334 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9335 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9336 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9337 | if (IS_ERR(file)) { |
9338 | sock_release(ctx->ring_sock); | |
9339 | ctx->ring_sock = NULL; | |
9340 | } else { | |
9341 | ctx->ring_sock->file = file; | |
0f212204 | 9342 | } |
2b188cc1 | 9343 | #endif |
9faadcc8 | 9344 | return file; |
2b188cc1 JA |
9345 | } |
9346 | ||
7f13657d XW |
9347 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9348 | struct io_uring_params __user *params) | |
2b188cc1 | 9349 | { |
2b188cc1 | 9350 | struct io_ring_ctx *ctx; |
9faadcc8 | 9351 | struct file *file; |
2b188cc1 JA |
9352 | int ret; |
9353 | ||
8110c1a6 | 9354 | if (!entries) |
2b188cc1 | 9355 | return -EINVAL; |
8110c1a6 JA |
9356 | if (entries > IORING_MAX_ENTRIES) { |
9357 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9358 | return -EINVAL; | |
9359 | entries = IORING_MAX_ENTRIES; | |
9360 | } | |
2b188cc1 JA |
9361 | |
9362 | /* | |
9363 | * Use twice as many entries for the CQ ring. It's possible for the | |
9364 | * application to drive a higher depth than the size of the SQ ring, | |
9365 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9366 | * some flexibility in overcommitting a bit. If the application has |
9367 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9368 | * of CQ ring entries manually. | |
2b188cc1 JA |
9369 | */ |
9370 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9371 | if (p->flags & IORING_SETUP_CQSIZE) { |
9372 | /* | |
9373 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9374 | * to a power-of-two, if it isn't already. We do NOT impose | |
9375 | * any cq vs sq ring sizing. | |
9376 | */ | |
eb2667b3 | 9377 | if (!p->cq_entries) |
33a107f0 | 9378 | return -EINVAL; |
8110c1a6 JA |
9379 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9380 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9381 | return -EINVAL; | |
9382 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9383 | } | |
eb2667b3 JQ |
9384 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9385 | if (p->cq_entries < p->sq_entries) | |
9386 | return -EINVAL; | |
33a107f0 JA |
9387 | } else { |
9388 | p->cq_entries = 2 * p->sq_entries; | |
9389 | } | |
2b188cc1 | 9390 | |
2b188cc1 | 9391 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9392 | if (!ctx) |
2b188cc1 | 9393 | return -ENOMEM; |
2b188cc1 | 9394 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9395 | if (!capable(CAP_IPC_LOCK)) |
9396 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9397 | |
9398 | /* | |
9399 | * This is just grabbed for accounting purposes. When a process exits, | |
9400 | * the mm is exited and dropped before the files, hence we need to hang | |
9401 | * on to this mm purely for the purposes of being able to unaccount | |
9402 | * memory (locked/pinned vm). It's not used for anything else. | |
9403 | */ | |
6b7898eb | 9404 | mmgrab(current->mm); |
2aede0e4 | 9405 | ctx->mm_account = current->mm; |
6b7898eb | 9406 | |
2b188cc1 JA |
9407 | ret = io_allocate_scq_urings(ctx, p); |
9408 | if (ret) | |
9409 | goto err; | |
9410 | ||
7e84e1c7 | 9411 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9412 | if (ret) |
9413 | goto err; | |
9414 | ||
7e84e1c7 SG |
9415 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9416 | io_sq_offload_start(ctx); | |
9417 | ||
2b188cc1 | 9418 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9419 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9420 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9421 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9422 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9423 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9424 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9425 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9426 | |
9427 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9428 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9429 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9430 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9431 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9432 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9433 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9434 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9435 | |
7f13657d XW |
9436 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9437 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9438 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9439 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9440 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9441 | |
9442 | if (copy_to_user(params, p, sizeof(*p))) { | |
9443 | ret = -EFAULT; | |
9444 | goto err; | |
9445 | } | |
d1719f70 | 9446 | |
9faadcc8 PB |
9447 | file = io_uring_get_file(ctx); |
9448 | if (IS_ERR(file)) { | |
9449 | ret = PTR_ERR(file); | |
9450 | goto err; | |
9451 | } | |
9452 | ||
044c1ab3 JA |
9453 | /* |
9454 | * Install ring fd as the very last thing, so we don't risk someone | |
9455 | * having closed it before we finish setup | |
9456 | */ | |
9faadcc8 PB |
9457 | ret = io_uring_install_fd(ctx, file); |
9458 | if (ret < 0) { | |
9459 | /* fput will clean it up */ | |
9460 | fput(file); | |
9461 | return ret; | |
9462 | } | |
044c1ab3 | 9463 | |
c826bd7a | 9464 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9465 | return ret; |
9466 | err: | |
9467 | io_ring_ctx_wait_and_kill(ctx); | |
9468 | return ret; | |
9469 | } | |
9470 | ||
9471 | /* | |
9472 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9473 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9474 | * params structure passed in. | |
9475 | */ | |
9476 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9477 | { | |
9478 | struct io_uring_params p; | |
2b188cc1 JA |
9479 | int i; |
9480 | ||
9481 | if (copy_from_user(&p, params, sizeof(p))) | |
9482 | return -EFAULT; | |
9483 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9484 | if (p.resv[i]) | |
9485 | return -EINVAL; | |
9486 | } | |
9487 | ||
6c271ce2 | 9488 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9489 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9490 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9491 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9492 | return -EINVAL; |
9493 | ||
7f13657d | 9494 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9495 | } |
9496 | ||
9497 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9498 | struct io_uring_params __user *, params) | |
9499 | { | |
9500 | return io_uring_setup(entries, params); | |
9501 | } | |
9502 | ||
66f4af93 JA |
9503 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9504 | { | |
9505 | struct io_uring_probe *p; | |
9506 | size_t size; | |
9507 | int i, ret; | |
9508 | ||
9509 | size = struct_size(p, ops, nr_args); | |
9510 | if (size == SIZE_MAX) | |
9511 | return -EOVERFLOW; | |
9512 | p = kzalloc(size, GFP_KERNEL); | |
9513 | if (!p) | |
9514 | return -ENOMEM; | |
9515 | ||
9516 | ret = -EFAULT; | |
9517 | if (copy_from_user(p, arg, size)) | |
9518 | goto out; | |
9519 | ret = -EINVAL; | |
9520 | if (memchr_inv(p, 0, size)) | |
9521 | goto out; | |
9522 | ||
9523 | p->last_op = IORING_OP_LAST - 1; | |
9524 | if (nr_args > IORING_OP_LAST) | |
9525 | nr_args = IORING_OP_LAST; | |
9526 | ||
9527 | for (i = 0; i < nr_args; i++) { | |
9528 | p->ops[i].op = i; | |
9529 | if (!io_op_defs[i].not_supported) | |
9530 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9531 | } | |
9532 | p->ops_len = i; | |
9533 | ||
9534 | ret = 0; | |
9535 | if (copy_to_user(arg, p, size)) | |
9536 | ret = -EFAULT; | |
9537 | out: | |
9538 | kfree(p); | |
9539 | return ret; | |
9540 | } | |
9541 | ||
071698e1 JA |
9542 | static int io_register_personality(struct io_ring_ctx *ctx) |
9543 | { | |
4379bf8b | 9544 | const struct cred *creds; |
1e6fa521 | 9545 | int ret; |
071698e1 | 9546 | |
4379bf8b | 9547 | creds = get_current_cred(); |
1e6fa521 | 9548 | |
4379bf8b JA |
9549 | ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, |
9550 | USHRT_MAX, GFP_KERNEL); | |
9551 | if (ret < 0) | |
9552 | put_cred(creds); | |
1e6fa521 | 9553 | return ret; |
071698e1 JA |
9554 | } |
9555 | ||
21b55dbc SG |
9556 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9557 | unsigned int nr_args) | |
9558 | { | |
9559 | struct io_uring_restriction *res; | |
9560 | size_t size; | |
9561 | int i, ret; | |
9562 | ||
7e84e1c7 SG |
9563 | /* Restrictions allowed only if rings started disabled */ |
9564 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9565 | return -EBADFD; | |
9566 | ||
21b55dbc | 9567 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9568 | if (ctx->restrictions.registered) |
21b55dbc SG |
9569 | return -EBUSY; |
9570 | ||
9571 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9572 | return -EINVAL; | |
9573 | ||
9574 | size = array_size(nr_args, sizeof(*res)); | |
9575 | if (size == SIZE_MAX) | |
9576 | return -EOVERFLOW; | |
9577 | ||
9578 | res = memdup_user(arg, size); | |
9579 | if (IS_ERR(res)) | |
9580 | return PTR_ERR(res); | |
9581 | ||
9582 | ret = 0; | |
9583 | ||
9584 | for (i = 0; i < nr_args; i++) { | |
9585 | switch (res[i].opcode) { | |
9586 | case IORING_RESTRICTION_REGISTER_OP: | |
9587 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9588 | ret = -EINVAL; | |
9589 | goto out; | |
9590 | } | |
9591 | ||
9592 | __set_bit(res[i].register_op, | |
9593 | ctx->restrictions.register_op); | |
9594 | break; | |
9595 | case IORING_RESTRICTION_SQE_OP: | |
9596 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9597 | ret = -EINVAL; | |
9598 | goto out; | |
9599 | } | |
9600 | ||
9601 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9602 | break; | |
9603 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9604 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9605 | break; | |
9606 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9607 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9608 | break; | |
9609 | default: | |
9610 | ret = -EINVAL; | |
9611 | goto out; | |
9612 | } | |
9613 | } | |
9614 | ||
9615 | out: | |
9616 | /* Reset all restrictions if an error happened */ | |
9617 | if (ret != 0) | |
9618 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9619 | else | |
7e84e1c7 | 9620 | ctx->restrictions.registered = true; |
21b55dbc SG |
9621 | |
9622 | kfree(res); | |
9623 | return ret; | |
9624 | } | |
9625 | ||
7e84e1c7 SG |
9626 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9627 | { | |
9628 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9629 | return -EBADFD; | |
9630 | ||
9631 | if (ctx->restrictions.registered) | |
9632 | ctx->restricted = 1; | |
9633 | ||
7e84e1c7 | 9634 | io_sq_offload_start(ctx); |
7e84e1c7 SG |
9635 | return 0; |
9636 | } | |
9637 | ||
071698e1 JA |
9638 | static bool io_register_op_must_quiesce(int op) |
9639 | { | |
9640 | switch (op) { | |
9641 | case IORING_UNREGISTER_FILES: | |
9642 | case IORING_REGISTER_FILES_UPDATE: | |
9643 | case IORING_REGISTER_PROBE: | |
9644 | case IORING_REGISTER_PERSONALITY: | |
9645 | case IORING_UNREGISTER_PERSONALITY: | |
9646 | return false; | |
9647 | default: | |
9648 | return true; | |
9649 | } | |
9650 | } | |
9651 | ||
edafccee JA |
9652 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9653 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9654 | __releases(ctx->uring_lock) |
9655 | __acquires(ctx->uring_lock) | |
edafccee JA |
9656 | { |
9657 | int ret; | |
9658 | ||
35fa71a0 JA |
9659 | /* |
9660 | * We're inside the ring mutex, if the ref is already dying, then | |
9661 | * someone else killed the ctx or is already going through | |
9662 | * io_uring_register(). | |
9663 | */ | |
9664 | if (percpu_ref_is_dying(&ctx->refs)) | |
9665 | return -ENXIO; | |
9666 | ||
071698e1 | 9667 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9668 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9669 | |
05f3fb3c JA |
9670 | /* |
9671 | * Drop uring mutex before waiting for references to exit. If | |
9672 | * another thread is currently inside io_uring_enter() it might | |
9673 | * need to grab the uring_lock to make progress. If we hold it | |
9674 | * here across the drain wait, then we can deadlock. It's safe | |
9675 | * to drop the mutex here, since no new references will come in | |
9676 | * after we've killed the percpu ref. | |
9677 | */ | |
9678 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9679 | do { |
9680 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9681 | if (!ret) | |
9682 | break; | |
ed6930c9 JA |
9683 | ret = io_run_task_work_sig(); |
9684 | if (ret < 0) | |
9685 | break; | |
af9c1a44 JA |
9686 | } while (1); |
9687 | ||
05f3fb3c | 9688 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9689 | |
c150368b JA |
9690 | if (ret) { |
9691 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9692 | goto out_quiesce; |
9693 | } | |
9694 | } | |
9695 | ||
9696 | if (ctx->restricted) { | |
9697 | if (opcode >= IORING_REGISTER_LAST) { | |
9698 | ret = -EINVAL; | |
9699 | goto out; | |
9700 | } | |
9701 | ||
9702 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9703 | ret = -EACCES; | |
c150368b JA |
9704 | goto out; |
9705 | } | |
05f3fb3c | 9706 | } |
edafccee JA |
9707 | |
9708 | switch (opcode) { | |
9709 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9710 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9711 | break; |
9712 | case IORING_UNREGISTER_BUFFERS: | |
9713 | ret = -EINVAL; | |
9714 | if (arg || nr_args) | |
9715 | break; | |
0a96bbe4 | 9716 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9717 | break; |
6b06314c JA |
9718 | case IORING_REGISTER_FILES: |
9719 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9720 | break; | |
9721 | case IORING_UNREGISTER_FILES: | |
9722 | ret = -EINVAL; | |
9723 | if (arg || nr_args) | |
9724 | break; | |
9725 | ret = io_sqe_files_unregister(ctx); | |
9726 | break; | |
c3a31e60 JA |
9727 | case IORING_REGISTER_FILES_UPDATE: |
9728 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9729 | break; | |
9b402849 | 9730 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9731 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9732 | ret = -EINVAL; |
9733 | if (nr_args != 1) | |
9734 | break; | |
9735 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9736 | if (ret) |
9737 | break; | |
9738 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9739 | ctx->eventfd_async = 1; | |
9740 | else | |
9741 | ctx->eventfd_async = 0; | |
9b402849 JA |
9742 | break; |
9743 | case IORING_UNREGISTER_EVENTFD: | |
9744 | ret = -EINVAL; | |
9745 | if (arg || nr_args) | |
9746 | break; | |
9747 | ret = io_eventfd_unregister(ctx); | |
9748 | break; | |
66f4af93 JA |
9749 | case IORING_REGISTER_PROBE: |
9750 | ret = -EINVAL; | |
9751 | if (!arg || nr_args > 256) | |
9752 | break; | |
9753 | ret = io_probe(ctx, arg, nr_args); | |
9754 | break; | |
071698e1 JA |
9755 | case IORING_REGISTER_PERSONALITY: |
9756 | ret = -EINVAL; | |
9757 | if (arg || nr_args) | |
9758 | break; | |
9759 | ret = io_register_personality(ctx); | |
9760 | break; | |
9761 | case IORING_UNREGISTER_PERSONALITY: | |
9762 | ret = -EINVAL; | |
9763 | if (arg) | |
9764 | break; | |
9765 | ret = io_unregister_personality(ctx, nr_args); | |
9766 | break; | |
7e84e1c7 SG |
9767 | case IORING_REGISTER_ENABLE_RINGS: |
9768 | ret = -EINVAL; | |
9769 | if (arg || nr_args) | |
9770 | break; | |
9771 | ret = io_register_enable_rings(ctx); | |
9772 | break; | |
21b55dbc SG |
9773 | case IORING_REGISTER_RESTRICTIONS: |
9774 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9775 | break; | |
edafccee JA |
9776 | default: |
9777 | ret = -EINVAL; | |
9778 | break; | |
9779 | } | |
9780 | ||
21b55dbc | 9781 | out: |
071698e1 | 9782 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9783 | /* bring the ctx back to life */ |
05f3fb3c | 9784 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9785 | out_quiesce: |
0f158b4c | 9786 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9787 | } |
edafccee JA |
9788 | return ret; |
9789 | } | |
9790 | ||
9791 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9792 | void __user *, arg, unsigned int, nr_args) | |
9793 | { | |
9794 | struct io_ring_ctx *ctx; | |
9795 | long ret = -EBADF; | |
9796 | struct fd f; | |
9797 | ||
9798 | f = fdget(fd); | |
9799 | if (!f.file) | |
9800 | return -EBADF; | |
9801 | ||
9802 | ret = -EOPNOTSUPP; | |
9803 | if (f.file->f_op != &io_uring_fops) | |
9804 | goto out_fput; | |
9805 | ||
9806 | ctx = f.file->private_data; | |
9807 | ||
b6c23dd5 PB |
9808 | io_run_task_work(); |
9809 | ||
edafccee JA |
9810 | mutex_lock(&ctx->uring_lock); |
9811 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9812 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9813 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9814 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9815 | out_fput: |
9816 | fdput(f); | |
9817 | return ret; | |
9818 | } | |
9819 | ||
2b188cc1 JA |
9820 | static int __init io_uring_init(void) |
9821 | { | |
d7f62e82 SM |
9822 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9823 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9824 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9825 | } while (0) | |
9826 | ||
9827 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9828 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9829 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9830 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9831 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9832 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9833 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9834 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9835 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9836 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9837 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9838 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9839 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9840 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9841 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9842 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9843 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9844 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9845 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9846 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9847 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9848 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9849 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9850 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9851 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9852 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9853 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9854 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9855 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9856 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9857 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9858 | |
d3656344 | 9859 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9860 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9861 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9862 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9863 | return 0; |
9864 | }; | |
9865 | __initcall(io_uring_init); |