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; |
d9d05217 | 341 | unsigned int sqo_dead: 1; |
5f3f26f9 | 342 | unsigned int sqo_exec: 1; |
2b188cc1 | 343 | |
75b28aff HV |
344 | /* |
345 | * Ring buffer of indices into array of io_uring_sqe, which is | |
346 | * mmapped by the application using the IORING_OFF_SQES offset. | |
347 | * | |
348 | * This indirection could e.g. be used to assign fixed | |
349 | * io_uring_sqe entries to operations and only submit them to | |
350 | * the queue when needed. | |
351 | * | |
352 | * The kernel modifies neither the indices array nor the entries | |
353 | * array. | |
354 | */ | |
355 | u32 *sq_array; | |
2b188cc1 JA |
356 | unsigned cached_sq_head; |
357 | unsigned sq_entries; | |
358 | unsigned sq_mask; | |
6c271ce2 | 359 | unsigned sq_thread_idle; |
498ccd9e | 360 | unsigned cached_sq_dropped; |
2c3bac6d | 361 | unsigned cached_cq_overflow; |
ad3eb2c8 | 362 | unsigned long sq_check_overflow; |
de0617e4 | 363 | |
e941894e JA |
364 | /* hashed buffered write serialization */ |
365 | struct io_wq_hash *hash_map; | |
366 | ||
de0617e4 | 367 | struct list_head defer_list; |
5262f567 | 368 | struct list_head timeout_list; |
1d7bb1d5 | 369 | struct list_head cq_overflow_list; |
fcb323cc | 370 | |
ad3eb2c8 | 371 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
372 | } ____cacheline_aligned_in_smp; |
373 | ||
3c1a2ead JA |
374 | struct { |
375 | struct mutex uring_lock; | |
376 | wait_queue_head_t wait; | |
377 | } ____cacheline_aligned_in_smp; | |
378 | ||
379 | struct io_submit_state submit_state; | |
380 | ||
206aefde JA |
381 | struct io_rings *rings; |
382 | ||
2aede0e4 | 383 | /* |
37d1e2e3 | 384 | * For SQPOLL usage |
2aede0e4 JA |
385 | */ |
386 | struct task_struct *sqo_task; | |
387 | ||
388 | /* Only used for accounting purposes */ | |
389 | struct mm_struct *mm_account; | |
390 | ||
534ca6d6 JA |
391 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
392 | ||
90554200 | 393 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 394 | struct list_head sqd_list; |
75b28aff | 395 | |
6b06314c JA |
396 | /* |
397 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
398 | * readers must ensure that ->refs is alive as long as the file* is | |
399 | * used. Only updated through io_uring_register(2). | |
400 | */ | |
269bbe5f | 401 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
402 | unsigned nr_user_files; |
403 | ||
edafccee JA |
404 | /* if used, fixed mapped user buffers */ |
405 | unsigned nr_user_bufs; | |
406 | struct io_mapped_ubuf *user_bufs; | |
407 | ||
2b188cc1 JA |
408 | struct user_struct *user; |
409 | ||
0f158b4c JA |
410 | struct completion ref_comp; |
411 | struct completion sq_thread_comp; | |
206aefde JA |
412 | |
413 | #if defined(CONFIG_UNIX) | |
414 | struct socket *ring_sock; | |
415 | #endif | |
416 | ||
5a2e745d JA |
417 | struct idr io_buffer_idr; |
418 | ||
071698e1 JA |
419 | struct idr personality_idr; |
420 | ||
206aefde JA |
421 | struct { |
422 | unsigned cached_cq_tail; | |
423 | unsigned cq_entries; | |
424 | unsigned cq_mask; | |
425 | atomic_t cq_timeouts; | |
f010505b | 426 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 427 | unsigned long cq_check_overflow; |
206aefde JA |
428 | struct wait_queue_head cq_wait; |
429 | struct fasync_struct *cq_fasync; | |
430 | struct eventfd_ctx *cq_ev_fd; | |
431 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 432 | |
2b188cc1 JA |
433 | struct { |
434 | spinlock_t completion_lock; | |
e94f141b | 435 | |
def596e9 | 436 | /* |
540e32a0 | 437 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
438 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
439 | * For SQPOLL, only the single threaded io_sq_thread() will | |
440 | * manipulate the list, hence no extra locking is needed there. | |
441 | */ | |
540e32a0 | 442 | struct list_head iopoll_list; |
78076bb6 JA |
443 | struct hlist_head *cancel_hash; |
444 | unsigned cancel_hash_bits; | |
e94f141b | 445 | bool poll_multi_file; |
31b51510 | 446 | |
fcb323cc JA |
447 | spinlock_t inflight_lock; |
448 | struct list_head inflight_list; | |
2b188cc1 | 449 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 450 | |
269bbe5f BM |
451 | struct delayed_work rsrc_put_work; |
452 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
453 | struct list_head rsrc_ref_list; |
454 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 455 | |
21b55dbc | 456 | struct io_restriction restrictions; |
3c1a2ead | 457 | |
7c25c0d1 JA |
458 | /* exit task_work */ |
459 | struct callback_head *exit_task_work; | |
460 | ||
e941894e JA |
461 | struct wait_queue_head hash_wait; |
462 | ||
3c1a2ead JA |
463 | /* Keep this last, we don't need it for the fast path */ |
464 | struct work_struct exit_work; | |
2b188cc1 JA |
465 | }; |
466 | ||
09bb8394 JA |
467 | /* |
468 | * First field must be the file pointer in all the | |
469 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
470 | */ | |
221c5eb2 JA |
471 | struct io_poll_iocb { |
472 | struct file *file; | |
018043be | 473 | struct wait_queue_head *head; |
221c5eb2 | 474 | __poll_t events; |
8c838788 | 475 | bool done; |
221c5eb2 | 476 | bool canceled; |
392edb45 | 477 | struct wait_queue_entry wait; |
221c5eb2 JA |
478 | }; |
479 | ||
018043be PB |
480 | struct io_poll_remove { |
481 | struct file *file; | |
482 | u64 addr; | |
483 | }; | |
484 | ||
b5dba59e JA |
485 | struct io_close { |
486 | struct file *file; | |
b5dba59e JA |
487 | int fd; |
488 | }; | |
489 | ||
ad8a48ac JA |
490 | struct io_timeout_data { |
491 | struct io_kiocb *req; | |
492 | struct hrtimer timer; | |
493 | struct timespec64 ts; | |
494 | enum hrtimer_mode mode; | |
495 | }; | |
496 | ||
8ed8d3c3 JA |
497 | struct io_accept { |
498 | struct file *file; | |
499 | struct sockaddr __user *addr; | |
500 | int __user *addr_len; | |
501 | int flags; | |
09952e3e | 502 | unsigned long nofile; |
8ed8d3c3 JA |
503 | }; |
504 | ||
505 | struct io_sync { | |
506 | struct file *file; | |
507 | loff_t len; | |
508 | loff_t off; | |
509 | int flags; | |
d63d1b5e | 510 | int mode; |
8ed8d3c3 JA |
511 | }; |
512 | ||
fbf23849 JA |
513 | struct io_cancel { |
514 | struct file *file; | |
515 | u64 addr; | |
516 | }; | |
517 | ||
b29472ee JA |
518 | struct io_timeout { |
519 | struct file *file; | |
bfe68a22 PB |
520 | u32 off; |
521 | u32 target_seq; | |
135fcde8 | 522 | struct list_head list; |
90cd7e42 PB |
523 | /* head of the link, used by linked timeouts only */ |
524 | struct io_kiocb *head; | |
b29472ee JA |
525 | }; |
526 | ||
0bdf7a2d PB |
527 | struct io_timeout_rem { |
528 | struct file *file; | |
529 | u64 addr; | |
9c8e11b3 PB |
530 | |
531 | /* timeout update */ | |
532 | struct timespec64 ts; | |
533 | u32 flags; | |
0bdf7a2d PB |
534 | }; |
535 | ||
9adbd45d JA |
536 | struct io_rw { |
537 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
538 | struct kiocb kiocb; | |
539 | u64 addr; | |
540 | u64 len; | |
541 | }; | |
542 | ||
3fbb51c1 JA |
543 | struct io_connect { |
544 | struct file *file; | |
545 | struct sockaddr __user *addr; | |
546 | int addr_len; | |
547 | }; | |
548 | ||
e47293fd JA |
549 | struct io_sr_msg { |
550 | struct file *file; | |
fddaface | 551 | union { |
270a5940 | 552 | struct user_msghdr __user *umsg; |
fddaface JA |
553 | void __user *buf; |
554 | }; | |
e47293fd | 555 | int msg_flags; |
bcda7baa | 556 | int bgid; |
fddaface | 557 | size_t len; |
bcda7baa | 558 | struct io_buffer *kbuf; |
e47293fd JA |
559 | }; |
560 | ||
15b71abe JA |
561 | struct io_open { |
562 | struct file *file; | |
563 | int dfd; | |
15b71abe | 564 | struct filename *filename; |
c12cedf2 | 565 | struct open_how how; |
4022e7af | 566 | unsigned long nofile; |
15b71abe JA |
567 | }; |
568 | ||
269bbe5f | 569 | struct io_rsrc_update { |
05f3fb3c JA |
570 | struct file *file; |
571 | u64 arg; | |
572 | u32 nr_args; | |
573 | u32 offset; | |
574 | }; | |
575 | ||
4840e418 JA |
576 | struct io_fadvise { |
577 | struct file *file; | |
578 | u64 offset; | |
579 | u32 len; | |
580 | u32 advice; | |
581 | }; | |
582 | ||
c1ca757b JA |
583 | struct io_madvise { |
584 | struct file *file; | |
585 | u64 addr; | |
586 | u32 len; | |
587 | u32 advice; | |
588 | }; | |
589 | ||
3e4827b0 JA |
590 | struct io_epoll { |
591 | struct file *file; | |
592 | int epfd; | |
593 | int op; | |
594 | int fd; | |
595 | struct epoll_event event; | |
e47293fd JA |
596 | }; |
597 | ||
7d67af2c PB |
598 | struct io_splice { |
599 | struct file *file_out; | |
600 | struct file *file_in; | |
601 | loff_t off_out; | |
602 | loff_t off_in; | |
603 | u64 len; | |
604 | unsigned int flags; | |
605 | }; | |
606 | ||
ddf0322d JA |
607 | struct io_provide_buf { |
608 | struct file *file; | |
609 | __u64 addr; | |
610 | __s32 len; | |
611 | __u32 bgid; | |
612 | __u16 nbufs; | |
613 | __u16 bid; | |
614 | }; | |
615 | ||
1d9e1288 BM |
616 | struct io_statx { |
617 | struct file *file; | |
618 | int dfd; | |
619 | unsigned int mask; | |
620 | unsigned int flags; | |
e62753e4 | 621 | const char __user *filename; |
1d9e1288 BM |
622 | struct statx __user *buffer; |
623 | }; | |
624 | ||
36f4fa68 JA |
625 | struct io_shutdown { |
626 | struct file *file; | |
627 | int how; | |
628 | }; | |
629 | ||
80a261fd JA |
630 | struct io_rename { |
631 | struct file *file; | |
632 | int old_dfd; | |
633 | int new_dfd; | |
634 | struct filename *oldpath; | |
635 | struct filename *newpath; | |
636 | int flags; | |
637 | }; | |
638 | ||
14a1143b JA |
639 | struct io_unlink { |
640 | struct file *file; | |
641 | int dfd; | |
642 | int flags; | |
643 | struct filename *filename; | |
644 | }; | |
645 | ||
3ca405eb PB |
646 | struct io_completion { |
647 | struct file *file; | |
648 | struct list_head list; | |
0f7e466b | 649 | int cflags; |
3ca405eb PB |
650 | }; |
651 | ||
f499a021 JA |
652 | struct io_async_connect { |
653 | struct sockaddr_storage address; | |
654 | }; | |
655 | ||
03b1230c JA |
656 | struct io_async_msghdr { |
657 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
658 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
659 | struct iovec *free_iov; | |
03b1230c JA |
660 | struct sockaddr __user *uaddr; |
661 | struct msghdr msg; | |
b537916c | 662 | struct sockaddr_storage addr; |
03b1230c JA |
663 | }; |
664 | ||
f67676d1 JA |
665 | struct io_async_rw { |
666 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
667 | const struct iovec *free_iovec; |
668 | struct iov_iter iter; | |
227c0c96 | 669 | size_t bytes_done; |
bcf5a063 | 670 | struct wait_page_queue wpq; |
f67676d1 JA |
671 | }; |
672 | ||
6b47ee6e PB |
673 | enum { |
674 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
675 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
676 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
677 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
678 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 679 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 680 | |
6b47ee6e PB |
681 | REQ_F_FAIL_LINK_BIT, |
682 | REQ_F_INFLIGHT_BIT, | |
683 | REQ_F_CUR_POS_BIT, | |
684 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 685 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 686 | REQ_F_ISREG_BIT, |
99bc4c38 | 687 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 688 | REQ_F_POLLED_BIT, |
bcda7baa | 689 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 690 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 691 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 692 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
693 | |
694 | /* not a real bit, just to check we're not overflowing the space */ | |
695 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
696 | }; |
697 | ||
698 | enum { | |
699 | /* ctx owns file */ | |
700 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
701 | /* drain existing IO first */ | |
702 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
703 | /* linked sqes */ | |
704 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
705 | /* doesn't sever on completion < 0 */ | |
706 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
707 | /* IOSQE_ASYNC */ | |
708 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
709 | /* IOSQE_BUFFER_SELECT */ |
710 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 711 | |
6b47ee6e PB |
712 | /* fail rest of links */ |
713 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
714 | /* on inflight list */ | |
715 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
716 | /* read/write uses file position */ | |
717 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
718 | /* must not punt to workers */ | |
719 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 720 | /* has or had linked timeout */ |
6b47ee6e | 721 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
722 | /* regular file */ |
723 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
724 | /* needs cleanup */ |
725 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
726 | /* already went through poll handler */ |
727 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
728 | /* buffer already selected */ |
729 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
730 | /* doesn't need file table for this request */ |
731 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
732 | /* linked timeout is active, i.e. prepared by link's head */ |
733 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
734 | /* completion is deferred through io_comp_state */ |
735 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
736 | }; |
737 | ||
738 | struct async_poll { | |
739 | struct io_poll_iocb poll; | |
807abcb0 | 740 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
741 | }; |
742 | ||
7cbf1722 JA |
743 | struct io_task_work { |
744 | struct io_wq_work_node node; | |
745 | task_work_func_t func; | |
746 | }; | |
747 | ||
09bb8394 JA |
748 | /* |
749 | * NOTE! Each of the iocb union members has the file pointer | |
750 | * as the first entry in their struct definition. So you can | |
751 | * access the file pointer through any of the sub-structs, | |
752 | * or directly as just 'ki_filp' in this struct. | |
753 | */ | |
2b188cc1 | 754 | struct io_kiocb { |
221c5eb2 | 755 | union { |
09bb8394 | 756 | struct file *file; |
9adbd45d | 757 | struct io_rw rw; |
221c5eb2 | 758 | struct io_poll_iocb poll; |
018043be | 759 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
760 | struct io_accept accept; |
761 | struct io_sync sync; | |
fbf23849 | 762 | struct io_cancel cancel; |
b29472ee | 763 | struct io_timeout timeout; |
0bdf7a2d | 764 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 765 | struct io_connect connect; |
e47293fd | 766 | struct io_sr_msg sr_msg; |
15b71abe | 767 | struct io_open open; |
b5dba59e | 768 | struct io_close close; |
269bbe5f | 769 | struct io_rsrc_update rsrc_update; |
4840e418 | 770 | struct io_fadvise fadvise; |
c1ca757b | 771 | struct io_madvise madvise; |
3e4827b0 | 772 | struct io_epoll epoll; |
7d67af2c | 773 | struct io_splice splice; |
ddf0322d | 774 | struct io_provide_buf pbuf; |
1d9e1288 | 775 | struct io_statx statx; |
36f4fa68 | 776 | struct io_shutdown shutdown; |
80a261fd | 777 | struct io_rename rename; |
14a1143b | 778 | struct io_unlink unlink; |
3ca405eb PB |
779 | /* use only after cleaning per-op data, see io_clean_op() */ |
780 | struct io_completion compl; | |
221c5eb2 | 781 | }; |
2b188cc1 | 782 | |
e8c2bc1f JA |
783 | /* opcode allocated if it needs to store data for async defer */ |
784 | void *async_data; | |
d625c6ee | 785 | u8 opcode; |
65a6543d XW |
786 | /* polled IO has completed */ |
787 | u8 iopoll_completed; | |
2b188cc1 | 788 | |
4f4eeba8 | 789 | u16 buf_index; |
9cf7c104 | 790 | u32 result; |
4f4eeba8 | 791 | |
010e8e6b PB |
792 | struct io_ring_ctx *ctx; |
793 | unsigned int flags; | |
794 | refcount_t refs; | |
795 | struct task_struct *task; | |
796 | u64 user_data; | |
d7718a9d | 797 | |
f2f87370 | 798 | struct io_kiocb *link; |
269bbe5f | 799 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 800 | |
d21ffe7e PB |
801 | /* |
802 | * 1. used with ctx->iopoll_list with reads/writes | |
803 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
804 | */ | |
010e8e6b | 805 | struct list_head inflight_entry; |
7cbf1722 JA |
806 | union { |
807 | struct io_task_work io_task_work; | |
808 | struct callback_head task_work; | |
809 | }; | |
010e8e6b PB |
810 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
811 | struct hlist_node hash_node; | |
812 | struct async_poll *apoll; | |
813 | struct io_wq_work work; | |
2b188cc1 | 814 | }; |
05589553 | 815 | |
27dc8338 PB |
816 | struct io_defer_entry { |
817 | struct list_head list; | |
818 | struct io_kiocb *req; | |
9cf7c104 | 819 | u32 seq; |
2b188cc1 JA |
820 | }; |
821 | ||
d3656344 | 822 | struct io_op_def { |
d3656344 JA |
823 | /* needs req->file assigned */ |
824 | unsigned needs_file : 1; | |
d3656344 JA |
825 | /* hash wq insertion if file is a regular file */ |
826 | unsigned hash_reg_file : 1; | |
827 | /* unbound wq insertion if file is a non-regular file */ | |
828 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
829 | /* opcode is not supported by this kernel */ |
830 | unsigned not_supported : 1; | |
8a72758c JA |
831 | /* set if opcode supports polled "wait" */ |
832 | unsigned pollin : 1; | |
833 | unsigned pollout : 1; | |
bcda7baa JA |
834 | /* op supports buffer selection */ |
835 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
836 | /* must always have async data allocated */ |
837 | unsigned needs_async_data : 1; | |
27926b68 JA |
838 | /* should block plug */ |
839 | unsigned plug : 1; | |
e8c2bc1f JA |
840 | /* size of async data needed, if any */ |
841 | unsigned short async_size; | |
d3656344 JA |
842 | }; |
843 | ||
0918682b | 844 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
845 | [IORING_OP_NOP] = {}, |
846 | [IORING_OP_READV] = { | |
d3656344 JA |
847 | .needs_file = 1, |
848 | .unbound_nonreg_file = 1, | |
8a72758c | 849 | .pollin = 1, |
4d954c25 | 850 | .buffer_select = 1, |
e8c2bc1f | 851 | .needs_async_data = 1, |
27926b68 | 852 | .plug = 1, |
e8c2bc1f | 853 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 854 | }, |
0463b6c5 | 855 | [IORING_OP_WRITEV] = { |
d3656344 JA |
856 | .needs_file = 1, |
857 | .hash_reg_file = 1, | |
858 | .unbound_nonreg_file = 1, | |
8a72758c | 859 | .pollout = 1, |
e8c2bc1f | 860 | .needs_async_data = 1, |
27926b68 | 861 | .plug = 1, |
e8c2bc1f | 862 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 863 | }, |
0463b6c5 | 864 | [IORING_OP_FSYNC] = { |
d3656344 JA |
865 | .needs_file = 1, |
866 | }, | |
0463b6c5 | 867 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
868 | .needs_file = 1, |
869 | .unbound_nonreg_file = 1, | |
8a72758c | 870 | .pollin = 1, |
27926b68 | 871 | .plug = 1, |
e8c2bc1f | 872 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 873 | }, |
0463b6c5 | 874 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
875 | .needs_file = 1, |
876 | .hash_reg_file = 1, | |
877 | .unbound_nonreg_file = 1, | |
8a72758c | 878 | .pollout = 1, |
27926b68 | 879 | .plug = 1, |
e8c2bc1f | 880 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 881 | }, |
0463b6c5 | 882 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
883 | .needs_file = 1, |
884 | .unbound_nonreg_file = 1, | |
885 | }, | |
0463b6c5 PB |
886 | [IORING_OP_POLL_REMOVE] = {}, |
887 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
888 | .needs_file = 1, |
889 | }, | |
0463b6c5 | 890 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
891 | .needs_file = 1, |
892 | .unbound_nonreg_file = 1, | |
8a72758c | 893 | .pollout = 1, |
e8c2bc1f JA |
894 | .needs_async_data = 1, |
895 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 896 | }, |
0463b6c5 | 897 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
898 | .needs_file = 1, |
899 | .unbound_nonreg_file = 1, | |
8a72758c | 900 | .pollin = 1, |
52de1fe1 | 901 | .buffer_select = 1, |
e8c2bc1f JA |
902 | .needs_async_data = 1, |
903 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 904 | }, |
0463b6c5 | 905 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
906 | .needs_async_data = 1, |
907 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 908 | }, |
9c8e11b3 PB |
909 | [IORING_OP_TIMEOUT_REMOVE] = { |
910 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 911 | }, |
0463b6c5 | 912 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
913 | .needs_file = 1, |
914 | .unbound_nonreg_file = 1, | |
8a72758c | 915 | .pollin = 1, |
d3656344 | 916 | }, |
0463b6c5 PB |
917 | [IORING_OP_ASYNC_CANCEL] = {}, |
918 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
919 | .needs_async_data = 1, |
920 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 921 | }, |
0463b6c5 | 922 | [IORING_OP_CONNECT] = { |
d3656344 JA |
923 | .needs_file = 1, |
924 | .unbound_nonreg_file = 1, | |
8a72758c | 925 | .pollout = 1, |
e8c2bc1f JA |
926 | .needs_async_data = 1, |
927 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 928 | }, |
0463b6c5 | 929 | [IORING_OP_FALLOCATE] = { |
d3656344 | 930 | .needs_file = 1, |
d3656344 | 931 | }, |
44526bed JA |
932 | [IORING_OP_OPENAT] = {}, |
933 | [IORING_OP_CLOSE] = {}, | |
934 | [IORING_OP_FILES_UPDATE] = {}, | |
935 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 936 | [IORING_OP_READ] = { |
3a6820f2 JA |
937 | .needs_file = 1, |
938 | .unbound_nonreg_file = 1, | |
8a72758c | 939 | .pollin = 1, |
bcda7baa | 940 | .buffer_select = 1, |
27926b68 | 941 | .plug = 1, |
e8c2bc1f | 942 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 943 | }, |
0463b6c5 | 944 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
945 | .needs_file = 1, |
946 | .unbound_nonreg_file = 1, | |
8a72758c | 947 | .pollout = 1, |
27926b68 | 948 | .plug = 1, |
e8c2bc1f | 949 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 950 | }, |
0463b6c5 | 951 | [IORING_OP_FADVISE] = { |
4840e418 | 952 | .needs_file = 1, |
c1ca757b | 953 | }, |
44526bed | 954 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 955 | [IORING_OP_SEND] = { |
fddaface JA |
956 | .needs_file = 1, |
957 | .unbound_nonreg_file = 1, | |
8a72758c | 958 | .pollout = 1, |
fddaface | 959 | }, |
0463b6c5 | 960 | [IORING_OP_RECV] = { |
fddaface JA |
961 | .needs_file = 1, |
962 | .unbound_nonreg_file = 1, | |
8a72758c | 963 | .pollin = 1, |
bcda7baa | 964 | .buffer_select = 1, |
fddaface | 965 | }, |
0463b6c5 | 966 | [IORING_OP_OPENAT2] = { |
cebdb986 | 967 | }, |
3e4827b0 JA |
968 | [IORING_OP_EPOLL_CTL] = { |
969 | .unbound_nonreg_file = 1, | |
3e4827b0 | 970 | }, |
7d67af2c PB |
971 | [IORING_OP_SPLICE] = { |
972 | .needs_file = 1, | |
973 | .hash_reg_file = 1, | |
974 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
975 | }, |
976 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 977 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
978 | [IORING_OP_TEE] = { |
979 | .needs_file = 1, | |
980 | .hash_reg_file = 1, | |
981 | .unbound_nonreg_file = 1, | |
982 | }, | |
36f4fa68 JA |
983 | [IORING_OP_SHUTDOWN] = { |
984 | .needs_file = 1, | |
985 | }, | |
44526bed JA |
986 | [IORING_OP_RENAMEAT] = {}, |
987 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
988 | }; |
989 | ||
9936c7c2 PB |
990 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
991 | struct task_struct *task, | |
992 | struct files_struct *files); | |
37d1e2e3 | 993 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 994 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 995 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 996 | struct io_ring_ctx *ctx); |
f2303b1f | 997 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 998 | |
23faba36 | 999 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1000 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1001 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1002 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1003 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1004 | static void io_dismantle_req(struct io_kiocb *req); |
1005 | static void io_put_task(struct task_struct *task, int nr); | |
1006 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1007 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1008 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1009 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1010 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1011 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1012 | unsigned nr_args); |
3ca405eb | 1013 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1014 | static struct file *io_file_get(struct io_submit_state *state, |
1015 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1016 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1017 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1018 | |
847595de PB |
1019 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1020 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1021 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1022 | const struct iovec *fast_iov, | |
227c0c96 | 1023 | struct iov_iter *iter, bool force); |
907d1df3 | 1024 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1025 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1026 | struct io_ring_ctx *ctx); | |
de0617e4 | 1027 | |
2b188cc1 JA |
1028 | static struct kmem_cache *req_cachep; |
1029 | ||
0918682b | 1030 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1031 | |
1032 | struct sock *io_uring_get_socket(struct file *file) | |
1033 | { | |
1034 | #if defined(CONFIG_UNIX) | |
1035 | if (file->f_op == &io_uring_fops) { | |
1036 | struct io_ring_ctx *ctx = file->private_data; | |
1037 | ||
1038 | return ctx->ring_sock->sk; | |
1039 | } | |
1040 | #endif | |
1041 | return NULL; | |
1042 | } | |
1043 | EXPORT_SYMBOL(io_uring_get_socket); | |
1044 | ||
f2f87370 PB |
1045 | #define io_for_each_link(pos, head) \ |
1046 | for (pos = (head); pos; pos = pos->link) | |
1047 | ||
3ca405eb PB |
1048 | static inline void io_clean_op(struct io_kiocb *req) |
1049 | { | |
9d5c8190 | 1050 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1051 | __io_clean_op(req); |
1052 | } | |
1053 | ||
36f72fe2 PB |
1054 | static inline void io_set_resource_node(struct io_kiocb *req) |
1055 | { | |
1056 | struct io_ring_ctx *ctx = req->ctx; | |
1057 | ||
269bbe5f BM |
1058 | if (!req->fixed_rsrc_refs) { |
1059 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1060 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1061 | } |
1062 | } | |
1063 | ||
08d23634 PB |
1064 | static bool io_match_task(struct io_kiocb *head, |
1065 | struct task_struct *task, | |
1066 | struct files_struct *files) | |
1067 | { | |
1068 | struct io_kiocb *req; | |
1069 | ||
84965ff8 JA |
1070 | if (task && head->task != task) { |
1071 | /* in terms of cancelation, always match if req task is dead */ | |
1072 | if (head->task->flags & PF_EXITING) | |
1073 | return true; | |
08d23634 | 1074 | return false; |
84965ff8 | 1075 | } |
08d23634 PB |
1076 | if (!files) |
1077 | return true; | |
1078 | ||
1079 | io_for_each_link(req, head) { | |
02a13674 JA |
1080 | if (req->file && req->file->f_op == &io_uring_fops) |
1081 | return true; | |
4379bf8b | 1082 | if (req->task->files == files) |
08d23634 PB |
1083 | return true; |
1084 | } | |
1085 | return false; | |
1086 | } | |
1087 | ||
c40f6379 JA |
1088 | static inline void req_set_fail_links(struct io_kiocb *req) |
1089 | { | |
1090 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1091 | req->flags |= REQ_F_FAIL_LINK; | |
1092 | } | |
4a38aed2 | 1093 | |
2b188cc1 JA |
1094 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1095 | { | |
1096 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1097 | ||
0f158b4c | 1098 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1099 | } |
1100 | ||
8eb7e2d0 PB |
1101 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1102 | { | |
1103 | return !req->timeout.off; | |
1104 | } | |
1105 | ||
2b188cc1 JA |
1106 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1107 | { | |
1108 | struct io_ring_ctx *ctx; | |
78076bb6 | 1109 | int hash_bits; |
2b188cc1 JA |
1110 | |
1111 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1112 | if (!ctx) | |
1113 | return NULL; | |
1114 | ||
78076bb6 JA |
1115 | /* |
1116 | * Use 5 bits less than the max cq entries, that should give us around | |
1117 | * 32 entries per hash list if totally full and uniformly spread. | |
1118 | */ | |
1119 | hash_bits = ilog2(p->cq_entries); | |
1120 | hash_bits -= 5; | |
1121 | if (hash_bits <= 0) | |
1122 | hash_bits = 1; | |
1123 | ctx->cancel_hash_bits = hash_bits; | |
1124 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1125 | GFP_KERNEL); | |
1126 | if (!ctx->cancel_hash) | |
1127 | goto err; | |
1128 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1129 | ||
21482896 | 1130 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1131 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1132 | goto err; | |
2b188cc1 JA |
1133 | |
1134 | ctx->flags = p->flags; | |
90554200 | 1135 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1136 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1137 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1138 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1139 | init_completion(&ctx->ref_comp); |
1140 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1141 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1142 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1143 | mutex_init(&ctx->uring_lock); |
1144 | init_waitqueue_head(&ctx->wait); | |
1145 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1146 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1147 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1148 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1149 | spin_lock_init(&ctx->inflight_lock); |
1150 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1151 | spin_lock_init(&ctx->rsrc_ref_lock); |
1152 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1153 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1154 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1155 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1156 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1157 | return ctx; |
206aefde | 1158 | err: |
78076bb6 | 1159 | kfree(ctx->cancel_hash); |
206aefde JA |
1160 | kfree(ctx); |
1161 | return NULL; | |
2b188cc1 JA |
1162 | } |
1163 | ||
9cf7c104 | 1164 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1165 | { |
2bc9930e JA |
1166 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1167 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1168 | |
9cf7c104 | 1169 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1170 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1171 | } |
de0617e4 | 1172 | |
9d858b21 | 1173 | return false; |
de0617e4 JA |
1174 | } |
1175 | ||
4edf20f9 | 1176 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1177 | { |
34e08fed PB |
1178 | if (req->flags & REQ_F_INFLIGHT) { |
1179 | struct io_ring_ctx *ctx = req->ctx; | |
1180 | struct io_uring_task *tctx = req->task->io_uring; | |
1181 | unsigned long flags; | |
1182 | ||
1183 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1184 | list_del(&req->inflight_entry); | |
1185 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1186 | req->flags &= ~REQ_F_INFLIGHT; | |
1187 | if (atomic_read(&tctx->in_idle)) | |
1188 | wake_up(&tctx->wait); | |
1189 | } | |
1e6fa521 JA |
1190 | } |
1191 | ||
ce3d5aae PB |
1192 | static void io_req_track_inflight(struct io_kiocb *req) |
1193 | { | |
1194 | struct io_ring_ctx *ctx = req->ctx; | |
1195 | ||
1196 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1197 | req->flags |= REQ_F_INFLIGHT; |
1198 | ||
1199 | spin_lock_irq(&ctx->inflight_lock); | |
1200 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1201 | spin_unlock_irq(&ctx->inflight_lock); | |
1202 | } | |
1203 | } | |
1204 | ||
1e6fa521 JA |
1205 | static void io_prep_async_work(struct io_kiocb *req) |
1206 | { | |
1207 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1208 | struct io_ring_ctx *ctx = req->ctx; |
1209 | ||
feaadc4f PB |
1210 | if (req->flags & REQ_F_FORCE_ASYNC) |
1211 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1212 | ||
1e6fa521 JA |
1213 | if (req->flags & REQ_F_ISREG) { |
1214 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1215 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1216 | } else { | |
1217 | if (def->unbound_nonreg_file) | |
1218 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1219 | } | |
561fb04a | 1220 | } |
cccf0ee8 | 1221 | |
cbdcb435 | 1222 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1223 | { |
cbdcb435 | 1224 | struct io_kiocb *cur; |
54a91f3b | 1225 | |
f2f87370 PB |
1226 | io_for_each_link(cur, req) |
1227 | io_prep_async_work(cur); | |
561fb04a JA |
1228 | } |
1229 | ||
7271ef3a | 1230 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1231 | { |
a197f664 | 1232 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1233 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1234 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1235 | |
3bfe6106 JA |
1236 | BUG_ON(!tctx); |
1237 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1238 | |
8766dd51 PB |
1239 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1240 | &req->work, req->flags); | |
5aa75ed5 | 1241 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a | 1242 | return link; |
18d9be1a JA |
1243 | } |
1244 | ||
cbdcb435 PB |
1245 | static void io_queue_async_work(struct io_kiocb *req) |
1246 | { | |
7271ef3a JA |
1247 | struct io_kiocb *link; |
1248 | ||
cbdcb435 PB |
1249 | /* init ->work of the whole link before punting */ |
1250 | io_prep_async_link(req); | |
7271ef3a JA |
1251 | link = __io_queue_async_work(req); |
1252 | ||
1253 | if (link) | |
1254 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1255 | } |
1256 | ||
5262f567 JA |
1257 | static void io_kill_timeout(struct io_kiocb *req) |
1258 | { | |
e8c2bc1f | 1259 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1260 | int ret; |
1261 | ||
e8c2bc1f | 1262 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1263 | if (ret != -1) { |
01cec8c1 PB |
1264 | atomic_set(&req->ctx->cq_timeouts, |
1265 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1266 | list_del_init(&req->timeout.list); |
78e19bbe | 1267 | io_cqring_fill_event(req, 0); |
216578e5 | 1268 | io_put_req_deferred(req, 1); |
5262f567 JA |
1269 | } |
1270 | } | |
1271 | ||
76e1b642 JA |
1272 | /* |
1273 | * Returns true if we found and killed one or more timeouts | |
1274 | */ | |
6b81928d PB |
1275 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1276 | struct files_struct *files) | |
5262f567 JA |
1277 | { |
1278 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1279 | int canceled = 0; |
5262f567 JA |
1280 | |
1281 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1282 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1283 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1284 | io_kill_timeout(req); |
76e1b642 JA |
1285 | canceled++; |
1286 | } | |
f3606e3a | 1287 | } |
5262f567 | 1288 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1289 | return canceled != 0; |
5262f567 JA |
1290 | } |
1291 | ||
04518945 | 1292 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1293 | { |
04518945 | 1294 | do { |
27dc8338 PB |
1295 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1296 | struct io_defer_entry, list); | |
de0617e4 | 1297 | |
9cf7c104 | 1298 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1299 | break; |
27dc8338 | 1300 | list_del_init(&de->list); |
907d1df3 | 1301 | io_req_task_queue(de->req); |
27dc8338 | 1302 | kfree(de); |
04518945 PB |
1303 | } while (!list_empty(&ctx->defer_list)); |
1304 | } | |
1305 | ||
360428f8 | 1306 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1307 | { |
f010505b MDG |
1308 | u32 seq; |
1309 | ||
1310 | if (list_empty(&ctx->timeout_list)) | |
1311 | return; | |
1312 | ||
1313 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1314 | ||
1315 | do { | |
1316 | u32 events_needed, events_got; | |
360428f8 | 1317 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1318 | struct io_kiocb, timeout.list); |
de0617e4 | 1319 | |
8eb7e2d0 | 1320 | if (io_is_timeout_noseq(req)) |
360428f8 | 1321 | break; |
f010505b MDG |
1322 | |
1323 | /* | |
1324 | * Since seq can easily wrap around over time, subtract | |
1325 | * the last seq at which timeouts were flushed before comparing. | |
1326 | * Assuming not more than 2^31-1 events have happened since, | |
1327 | * these subtractions won't have wrapped, so we can check if | |
1328 | * target is in [last_seq, current_seq] by comparing the two. | |
1329 | */ | |
1330 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1331 | events_got = seq - ctx->cq_last_tm_flush; | |
1332 | if (events_got < events_needed) | |
360428f8 | 1333 | break; |
bfe68a22 | 1334 | |
135fcde8 | 1335 | list_del_init(&req->timeout.list); |
5262f567 | 1336 | io_kill_timeout(req); |
f010505b MDG |
1337 | } while (!list_empty(&ctx->timeout_list)); |
1338 | ||
1339 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1340 | } |
5262f567 | 1341 | |
360428f8 PB |
1342 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1343 | { | |
1344 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1345 | |
1346 | /* order cqe stores with ring update */ | |
1347 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1348 | |
04518945 PB |
1349 | if (unlikely(!list_empty(&ctx->defer_list))) |
1350 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1351 | } |
1352 | ||
90554200 JA |
1353 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1354 | { | |
1355 | struct io_rings *r = ctx->rings; | |
1356 | ||
1357 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1358 | } | |
1359 | ||
888aae2e PB |
1360 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1361 | { | |
1362 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1363 | } | |
1364 | ||
2b188cc1 JA |
1365 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1366 | { | |
75b28aff | 1367 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1368 | unsigned tail; |
1369 | ||
115e12e5 SB |
1370 | /* |
1371 | * writes to the cq entry need to come after reading head; the | |
1372 | * control dependency is enough as we're using WRITE_ONCE to | |
1373 | * fill the cq entry | |
1374 | */ | |
888aae2e | 1375 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1376 | return NULL; |
1377 | ||
888aae2e | 1378 | tail = ctx->cached_cq_tail++; |
75b28aff | 1379 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1380 | } |
1381 | ||
f2842ab5 JA |
1382 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1383 | { | |
f0b493e6 JA |
1384 | if (!ctx->cq_ev_fd) |
1385 | return false; | |
7e55a19c SG |
1386 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1387 | return false; | |
f2842ab5 JA |
1388 | if (!ctx->eventfd_async) |
1389 | return true; | |
b41e9852 | 1390 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1391 | } |
1392 | ||
b41e9852 | 1393 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1394 | { |
b1445e59 PB |
1395 | /* see waitqueue_active() comment */ |
1396 | smp_mb(); | |
1397 | ||
1d7bb1d5 JA |
1398 | if (waitqueue_active(&ctx->wait)) |
1399 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1400 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1401 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1402 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1403 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1404 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1405 | wake_up_interruptible(&ctx->cq_wait); |
1406 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1407 | } | |
1d7bb1d5 JA |
1408 | } |
1409 | ||
80c18e4a PB |
1410 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1411 | { | |
b1445e59 PB |
1412 | /* see waitqueue_active() comment */ |
1413 | smp_mb(); | |
1414 | ||
80c18e4a PB |
1415 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1416 | if (waitqueue_active(&ctx->wait)) | |
1417 | wake_up(&ctx->wait); | |
1418 | } | |
1419 | if (io_should_trigger_evfd(ctx)) | |
1420 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1421 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1422 | wake_up_interruptible(&ctx->cq_wait); |
1423 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1424 | } | |
80c18e4a PB |
1425 | } |
1426 | ||
c4a2ed72 | 1427 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1428 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1429 | struct task_struct *tsk, | |
1430 | struct files_struct *files) | |
1d7bb1d5 JA |
1431 | { |
1432 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1433 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1434 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1435 | unsigned long flags; |
b18032bb | 1436 | bool all_flushed, posted; |
1d7bb1d5 JA |
1437 | LIST_HEAD(list); |
1438 | ||
e23de15f PB |
1439 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1440 | return false; | |
1d7bb1d5 | 1441 | |
b18032bb | 1442 | posted = false; |
1d7bb1d5 | 1443 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1444 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1445 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1446 | continue; |
1447 | ||
1d7bb1d5 JA |
1448 | cqe = io_get_cqring(ctx); |
1449 | if (!cqe && !force) | |
1450 | break; | |
1451 | ||
40d8ddd4 | 1452 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1453 | if (cqe) { |
1454 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1455 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1456 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1457 | } else { |
2c3bac6d | 1458 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1459 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1460 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1461 | } |
b18032bb | 1462 | posted = true; |
1d7bb1d5 JA |
1463 | } |
1464 | ||
09e88404 PB |
1465 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1466 | if (all_flushed) { | |
1467 | clear_bit(0, &ctx->sq_check_overflow); | |
1468 | clear_bit(0, &ctx->cq_check_overflow); | |
1469 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1470 | } | |
46930143 | 1471 | |
b18032bb JA |
1472 | if (posted) |
1473 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1474 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1475 | if (posted) |
1476 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1477 | |
1478 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1479 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1480 | list_del(&req->compl.list); | |
ec9c02ad | 1481 | io_put_req(req); |
1d7bb1d5 | 1482 | } |
c4a2ed72 | 1483 | |
09e88404 | 1484 | return all_flushed; |
1d7bb1d5 JA |
1485 | } |
1486 | ||
6c503150 PB |
1487 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1488 | struct task_struct *tsk, | |
1489 | struct files_struct *files) | |
1490 | { | |
1491 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1492 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1493 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1494 | mutex_lock(&ctx->uring_lock); | |
1495 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1496 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1497 | mutex_unlock(&ctx->uring_lock); | |
1498 | } | |
1499 | } | |
1500 | ||
bcda7baa | 1501 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1502 | { |
78e19bbe | 1503 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1504 | struct io_uring_cqe *cqe; |
1505 | ||
78e19bbe | 1506 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1507 | |
2b188cc1 JA |
1508 | /* |
1509 | * If we can't get a cq entry, userspace overflowed the | |
1510 | * submission (by quite a lot). Increment the overflow count in | |
1511 | * the ring. | |
1512 | */ | |
1513 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1514 | if (likely(cqe)) { |
78e19bbe | 1515 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1516 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1517 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1518 | } else if (ctx->cq_overflow_flushed || |
1519 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1520 | /* |
1521 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1522 | * then we cannot store the request for later flushing, we need | |
1523 | * to drop it on the floor. | |
1524 | */ | |
2c3bac6d PB |
1525 | ctx->cached_cq_overflow++; |
1526 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1527 | } else { |
ad3eb2c8 JA |
1528 | if (list_empty(&ctx->cq_overflow_list)) { |
1529 | set_bit(0, &ctx->sq_check_overflow); | |
1530 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1531 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1532 | } |
40d8ddd4 | 1533 | io_clean_op(req); |
1d7bb1d5 | 1534 | req->result = res; |
0f7e466b | 1535 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1536 | refcount_inc(&req->refs); |
1537 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1538 | } |
1539 | } | |
1540 | ||
bcda7baa JA |
1541 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1542 | { | |
1543 | __io_cqring_fill_event(req, res, 0); | |
1544 | } | |
1545 | ||
c7dae4ba JA |
1546 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1547 | unsigned int cflags) | |
2b188cc1 | 1548 | { |
78e19bbe | 1549 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1550 | unsigned long flags; |
1551 | ||
1552 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1553 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1554 | io_commit_cqring(ctx); |
c7dae4ba JA |
1555 | /* |
1556 | * If we're the last reference to this request, add to our locked | |
1557 | * free_list cache. | |
1558 | */ | |
1559 | if (refcount_dec_and_test(&req->refs)) { | |
1560 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1561 | ||
1562 | io_dismantle_req(req); | |
1563 | io_put_task(req->task, 1); | |
1564 | list_add(&req->compl.list, &cs->locked_free_list); | |
1565 | cs->locked_free_nr++; | |
1566 | } else | |
1567 | req = NULL; | |
2b188cc1 JA |
1568 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1569 | ||
8c838788 | 1570 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1571 | if (req) { |
1572 | io_queue_next(req); | |
1573 | percpu_ref_put(&ctx->refs); | |
229a7b63 | 1574 | } |
229a7b63 JA |
1575 | } |
1576 | ||
a38d68db | 1577 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1578 | unsigned int cflags) |
229a7b63 | 1579 | { |
a38d68db PB |
1580 | io_clean_op(req); |
1581 | req->result = res; | |
1582 | req->compl.cflags = cflags; | |
e342c807 | 1583 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1584 | } |
1585 | ||
889fca73 PB |
1586 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1587 | long res, unsigned cflags) | |
bcda7baa | 1588 | { |
889fca73 PB |
1589 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1590 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1591 | else |
c7dae4ba | 1592 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1593 | } |
1594 | ||
a38d68db | 1595 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1596 | { |
889fca73 | 1597 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1598 | } |
1599 | ||
c7dae4ba | 1600 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1601 | { |
c7dae4ba JA |
1602 | struct io_submit_state *state = &ctx->submit_state; |
1603 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1604 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1605 | |
c7dae4ba JA |
1606 | /* |
1607 | * If we have more than a batch's worth of requests in our IRQ side | |
1608 | * locked cache, grab the lock and move them over to our submission | |
1609 | * side cache. | |
1610 | */ | |
1611 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1612 | spin_lock_irq(&ctx->completion_lock); | |
1613 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1614 | cs->locked_free_nr = 0; | |
1615 | spin_unlock_irq(&ctx->completion_lock); | |
1616 | } | |
0ddf92e8 | 1617 | |
c7dae4ba JA |
1618 | while (!list_empty(&cs->free_list)) { |
1619 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1620 | compl.list); |
1621 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1622 | state->reqs[state->free_reqs++] = req; |
1623 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1624 | break; | |
1b4c351f JA |
1625 | } |
1626 | ||
e5d1bc0a | 1627 | return req != NULL; |
0ddf92e8 JA |
1628 | } |
1629 | ||
e5d1bc0a | 1630 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1631 | { |
e5d1bc0a PB |
1632 | struct io_submit_state *state = &ctx->submit_state; |
1633 | ||
1634 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1635 | ||
f6b6c7d6 | 1636 | if (!state->free_reqs) { |
291b2821 | 1637 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1638 | int ret; |
1639 | ||
c7dae4ba | 1640 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1641 | goto got_req; |
1642 | ||
bf019da7 PB |
1643 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1644 | state->reqs); | |
fd6fab2c JA |
1645 | |
1646 | /* | |
1647 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1648 | * retry single alloc to be on the safe side. | |
1649 | */ | |
1650 | if (unlikely(ret <= 0)) { | |
1651 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1652 | if (!state->reqs[0]) | |
3893f39f | 1653 | return NULL; |
fd6fab2c JA |
1654 | ret = 1; |
1655 | } | |
291b2821 | 1656 | state->free_reqs = ret; |
2b188cc1 | 1657 | } |
e5d1bc0a | 1658 | got_req: |
291b2821 PB |
1659 | state->free_reqs--; |
1660 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1661 | } |
1662 | ||
8da11c19 PB |
1663 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1664 | bool fixed) | |
1665 | { | |
36f72fe2 | 1666 | if (!fixed) |
8da11c19 PB |
1667 | fput(file); |
1668 | } | |
1669 | ||
4edf20f9 | 1670 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1671 | { |
3ca405eb | 1672 | io_clean_op(req); |
929a3af9 | 1673 | |
e8c2bc1f JA |
1674 | if (req->async_data) |
1675 | kfree(req->async_data); | |
8da11c19 PB |
1676 | if (req->file) |
1677 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1678 | if (req->fixed_rsrc_refs) |
1679 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 1680 | io_req_clean_work(req); |
e65ef56d JA |
1681 | } |
1682 | ||
7c660731 PB |
1683 | static inline void io_put_task(struct task_struct *task, int nr) |
1684 | { | |
1685 | struct io_uring_task *tctx = task->io_uring; | |
1686 | ||
1687 | percpu_counter_sub(&tctx->inflight, nr); | |
1688 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1689 | wake_up(&tctx->wait); | |
1690 | put_task_struct_many(task, nr); | |
1691 | } | |
1692 | ||
216578e5 | 1693 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1694 | { |
51a4cc11 | 1695 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1696 | |
216578e5 | 1697 | io_dismantle_req(req); |
7c660731 | 1698 | io_put_task(req->task, 1); |
c6ca97b3 | 1699 | |
3893f39f | 1700 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1701 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1702 | } |
1703 | ||
f2f87370 PB |
1704 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1705 | { | |
1706 | struct io_kiocb *nxt = req->link; | |
1707 | ||
1708 | req->link = nxt->link; | |
1709 | nxt->link = NULL; | |
1710 | } | |
1711 | ||
c9abd7ad | 1712 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1713 | { |
a197f664 | 1714 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1715 | struct io_kiocb *link; |
c9abd7ad PB |
1716 | bool cancelled = false; |
1717 | unsigned long flags; | |
7c86ffee | 1718 | |
c9abd7ad | 1719 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1720 | link = req->link; |
1721 | ||
900fad45 PB |
1722 | /* |
1723 | * Can happen if a linked timeout fired and link had been like | |
1724 | * req -> link t-out -> link t-out [-> ...] | |
1725 | */ | |
c9abd7ad PB |
1726 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1727 | struct io_timeout_data *io = link->async_data; | |
1728 | int ret; | |
7c86ffee | 1729 | |
f2f87370 | 1730 | io_remove_next_linked(req); |
90cd7e42 | 1731 | link->timeout.head = NULL; |
c9abd7ad PB |
1732 | ret = hrtimer_try_to_cancel(&io->timer); |
1733 | if (ret != -1) { | |
1734 | io_cqring_fill_event(link, -ECANCELED); | |
1735 | io_commit_cqring(ctx); | |
1736 | cancelled = true; | |
1737 | } | |
1738 | } | |
7c86ffee | 1739 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1740 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1741 | |
c9abd7ad | 1742 | if (cancelled) { |
7c86ffee | 1743 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1744 | io_put_req(link); |
1745 | } | |
7c86ffee PB |
1746 | } |
1747 | ||
9e645e11 | 1748 | |
d148ca4b | 1749 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1750 | { |
f2f87370 | 1751 | struct io_kiocb *link, *nxt; |
2665abfd | 1752 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1753 | unsigned long flags; |
9e645e11 | 1754 | |
d148ca4b | 1755 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1756 | link = req->link; |
1757 | req->link = NULL; | |
9e645e11 | 1758 | |
f2f87370 PB |
1759 | while (link) { |
1760 | nxt = link->link; | |
1761 | link->link = NULL; | |
2665abfd | 1762 | |
f2f87370 | 1763 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1764 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 | 1765 | |
1575f21a | 1766 | io_put_req_deferred(link, 2); |
f2f87370 | 1767 | link = nxt; |
9e645e11 | 1768 | } |
2665abfd | 1769 | io_commit_cqring(ctx); |
216578e5 | 1770 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1771 | |
2665abfd | 1772 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1773 | } |
1774 | ||
3fa5e0f3 | 1775 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1776 | { |
7c86ffee PB |
1777 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1778 | io_kill_linked_timeout(req); | |
944e58bf | 1779 | |
9e645e11 JA |
1780 | /* |
1781 | * If LINK is set, we have dependent requests in this chain. If we | |
1782 | * didn't fail this request, queue the first one up, moving any other | |
1783 | * dependencies to the next request. In case of failure, fail the rest | |
1784 | * of the chain. | |
1785 | */ | |
f2f87370 PB |
1786 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
1787 | struct io_kiocb *nxt = req->link; | |
1788 | ||
1789 | req->link = NULL; | |
1790 | return nxt; | |
1791 | } | |
9b5f7bd9 PB |
1792 | io_fail_links(req); |
1793 | return NULL; | |
4d7dd462 | 1794 | } |
9e645e11 | 1795 | |
f2f87370 | 1796 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1797 | { |
cdbff982 | 1798 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1799 | return NULL; |
1800 | return __io_req_find_next(req); | |
1801 | } | |
1802 | ||
2c32395d PB |
1803 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1804 | { | |
1805 | if (!ctx) | |
1806 | return; | |
1807 | if (ctx->submit_state.comp.nr) { | |
1808 | mutex_lock(&ctx->uring_lock); | |
1809 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1810 | mutex_unlock(&ctx->uring_lock); | |
1811 | } | |
1812 | percpu_ref_put(&ctx->refs); | |
1813 | } | |
1814 | ||
7cbf1722 | 1815 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1816 | { |
65453d1e | 1817 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1818 | struct io_wq_work_list list; |
1819 | struct io_wq_work_node *node; | |
c2c4c83c | 1820 | |
7cbf1722 JA |
1821 | if (wq_list_empty(&tctx->task_list)) |
1822 | return false; | |
6200b0ae | 1823 | |
0b81e80c | 1824 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1825 | list = tctx->task_list; |
1826 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1827 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1828 | |
7cbf1722 JA |
1829 | node = list.first; |
1830 | while (node) { | |
1831 | struct io_wq_work_node *next = node->next; | |
1832 | struct io_kiocb *req; | |
0ba9c9ed | 1833 | |
7cbf1722 | 1834 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1835 | if (req->ctx != ctx) { |
1836 | ctx_flush_and_put(ctx); | |
1837 | ctx = req->ctx; | |
1838 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1839 | } |
65453d1e | 1840 | |
2c32395d PB |
1841 | req->task_work.func(&req->task_work); |
1842 | node = next; | |
7cbf1722 JA |
1843 | } |
1844 | ||
2c32395d | 1845 | ctx_flush_and_put(ctx); |
7cbf1722 | 1846 | return list.first != NULL; |
c2c4c83c JA |
1847 | } |
1848 | ||
7cbf1722 | 1849 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1850 | { |
7cbf1722 | 1851 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1852 | |
1d5f360d JA |
1853 | clear_bit(0, &tctx->task_state); |
1854 | ||
7cbf1722 JA |
1855 | while (__tctx_task_work(tctx)) |
1856 | cond_resched(); | |
7cbf1722 JA |
1857 | } |
1858 | ||
1859 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1860 | enum task_work_notify_mode notify) | |
1861 | { | |
1862 | struct io_uring_task *tctx = tsk->io_uring; | |
1863 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1864 | unsigned long flags; |
7cbf1722 JA |
1865 | int ret; |
1866 | ||
1867 | WARN_ON_ONCE(!tctx); | |
1868 | ||
0b81e80c | 1869 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1870 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1871 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1872 | |
1873 | /* task_work already pending, we're done */ | |
1874 | if (test_bit(0, &tctx->task_state) || | |
1875 | test_and_set_bit(0, &tctx->task_state)) | |
1876 | return 0; | |
1877 | ||
1878 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1879 | return 0; | |
1880 | ||
1881 | /* | |
1882 | * Slow path - we failed, find and delete work. if the work is not | |
1883 | * in the list, it got run and we're fine. | |
1884 | */ | |
1885 | ret = 0; | |
0b81e80c | 1886 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1887 | wq_list_for_each(node, prev, &tctx->task_list) { |
1888 | if (&req->io_task_work.node == node) { | |
1889 | wq_list_del(&tctx->task_list, node, prev); | |
1890 | ret = 1; | |
1891 | break; | |
1892 | } | |
1893 | } | |
0b81e80c | 1894 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1895 | clear_bit(0, &tctx->task_state); |
1896 | return ret; | |
1897 | } | |
1898 | ||
355fb9e2 | 1899 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1900 | { |
1901 | struct task_struct *tsk = req->task; | |
1902 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1903 | enum task_work_notify_mode notify; |
1904 | int ret; | |
c2c4c83c | 1905 | |
6200b0ae JA |
1906 | if (tsk->flags & PF_EXITING) |
1907 | return -ESRCH; | |
1908 | ||
c2c4c83c | 1909 | /* |
0ba9c9ed JA |
1910 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1911 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1912 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1913 | * will do the job. | |
c2c4c83c | 1914 | */ |
91989c70 | 1915 | notify = TWA_NONE; |
355fb9e2 | 1916 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1917 | notify = TWA_SIGNAL; |
1918 | ||
7cbf1722 | 1919 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1920 | if (!ret) |
1921 | wake_up_process(tsk); | |
0ba9c9ed | 1922 | |
c2c4c83c JA |
1923 | return ret; |
1924 | } | |
1925 | ||
eab30c4d | 1926 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1927 | task_work_func_t cb) |
eab30c4d | 1928 | { |
7c25c0d1 JA |
1929 | struct io_ring_ctx *ctx = req->ctx; |
1930 | struct callback_head *head; | |
eab30c4d PB |
1931 | |
1932 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1933 | do { |
1934 | head = READ_ONCE(ctx->exit_task_work); | |
1935 | req->task_work.next = head; | |
1936 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1937 | } |
1938 | ||
c40f6379 JA |
1939 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1940 | { | |
1941 | struct io_ring_ctx *ctx = req->ctx; | |
1942 | ||
1943 | spin_lock_irq(&ctx->completion_lock); | |
1944 | io_cqring_fill_event(req, error); | |
1945 | io_commit_cqring(ctx); | |
1946 | spin_unlock_irq(&ctx->completion_lock); | |
1947 | ||
1948 | io_cqring_ev_posted(ctx); | |
1949 | req_set_fail_links(req); | |
1950 | io_double_put_req(req); | |
1951 | } | |
1952 | ||
1953 | static void io_req_task_cancel(struct callback_head *cb) | |
1954 | { | |
1955 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1956 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1957 | |
792bb6eb | 1958 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1959 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1960 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1961 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1962 | } |
1963 | ||
1964 | static void __io_req_task_submit(struct io_kiocb *req) | |
1965 | { | |
1966 | struct io_ring_ctx *ctx = req->ctx; | |
1967 | ||
04fc6c80 | 1968 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1969 | mutex_lock(&ctx->uring_lock); |
4fb6ac32 | 1970 | if (!ctx->sqo_dead && !(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1971 | __io_queue_sqe(req); |
81b6d05c | 1972 | else |
c40f6379 | 1973 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1974 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1975 | } |
1976 | ||
1977 | static void io_req_task_submit(struct callback_head *cb) | |
1978 | { | |
1979 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1980 | ||
1981 | __io_req_task_submit(req); | |
1982 | } | |
1983 | ||
1984 | static void io_req_task_queue(struct io_kiocb *req) | |
1985 | { | |
c40f6379 JA |
1986 | int ret; |
1987 | ||
7cbf1722 | 1988 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1989 | ret = io_req_task_work_add(req); |
c40f6379 | 1990 | if (unlikely(ret)) { |
a3df7698 | 1991 | req->result = -ECANCELED; |
04fc6c80 | 1992 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1993 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1994 | } |
c40f6379 JA |
1995 | } |
1996 | ||
a3df7698 PB |
1997 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
1998 | { | |
1999 | percpu_ref_get(&req->ctx->refs); | |
2000 | req->result = ret; | |
2001 | req->task_work.func = io_req_task_cancel; | |
2002 | ||
2003 | if (unlikely(io_req_task_work_add(req))) | |
2004 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2005 | } | |
2006 | ||
f2f87370 | 2007 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2008 | { |
9b5f7bd9 | 2009 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2010 | |
2011 | if (nxt) | |
906a8c3f | 2012 | io_req_task_queue(nxt); |
c69f8dbe JL |
2013 | } |
2014 | ||
c3524383 | 2015 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2016 | { |
c3524383 PB |
2017 | io_queue_next(req); |
2018 | __io_free_req(req); | |
2019 | } | |
8766dd51 | 2020 | |
2d6500d4 | 2021 | struct req_batch { |
5af1d13e PB |
2022 | struct task_struct *task; |
2023 | int task_refs; | |
1b4c351f | 2024 | int ctx_refs; |
2d6500d4 PB |
2025 | }; |
2026 | ||
5af1d13e PB |
2027 | static inline void io_init_req_batch(struct req_batch *rb) |
2028 | { | |
5af1d13e | 2029 | rb->task_refs = 0; |
9ae72463 | 2030 | rb->ctx_refs = 0; |
5af1d13e PB |
2031 | rb->task = NULL; |
2032 | } | |
2033 | ||
2d6500d4 PB |
2034 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2035 | struct req_batch *rb) | |
2036 | { | |
6e833d53 | 2037 | if (rb->task) |
7c660731 | 2038 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2039 | if (rb->ctx_refs) |
2040 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2041 | } |
2042 | ||
6ff119a6 PB |
2043 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2044 | struct io_submit_state *state) | |
2d6500d4 | 2045 | { |
f2f87370 | 2046 | io_queue_next(req); |
2d6500d4 | 2047 | |
e3bc8e9d | 2048 | if (req->task != rb->task) { |
7c660731 PB |
2049 | if (rb->task) |
2050 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2051 | rb->task = req->task; |
2052 | rb->task_refs = 0; | |
5af1d13e | 2053 | } |
e3bc8e9d | 2054 | rb->task_refs++; |
9ae72463 | 2055 | rb->ctx_refs++; |
5af1d13e | 2056 | |
4edf20f9 | 2057 | io_dismantle_req(req); |
bd759045 | 2058 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2059 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2060 | else |
2061 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2062 | } |
2063 | ||
905c172f PB |
2064 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2065 | struct io_ring_ctx *ctx) | |
2066 | { | |
2067 | int i, nr = cs->nr; | |
2068 | struct io_kiocb *req; | |
2069 | struct req_batch rb; | |
2070 | ||
2071 | io_init_req_batch(&rb); | |
2072 | spin_lock_irq(&ctx->completion_lock); | |
2073 | for (i = 0; i < nr; i++) { | |
2074 | req = cs->reqs[i]; | |
2075 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2076 | } | |
2077 | io_commit_cqring(ctx); | |
2078 | spin_unlock_irq(&ctx->completion_lock); | |
2079 | ||
2080 | io_cqring_ev_posted(ctx); | |
2081 | for (i = 0; i < nr; i++) { | |
2082 | req = cs->reqs[i]; | |
2083 | ||
2084 | /* submission and completion refs */ | |
2085 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2086 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2087 | } |
2088 | ||
2089 | io_req_free_batch_finish(ctx, &rb); | |
2090 | cs->nr = 0; | |
7a743e22 PB |
2091 | } |
2092 | ||
ba816ad6 JA |
2093 | /* |
2094 | * Drop reference to request, return next in chain (if there is one) if this | |
2095 | * was the last reference to this request. | |
2096 | */ | |
9b5f7bd9 | 2097 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2098 | { |
9b5f7bd9 PB |
2099 | struct io_kiocb *nxt = NULL; |
2100 | ||
2a44f467 | 2101 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2102 | nxt = io_req_find_next(req); |
4d7dd462 | 2103 | __io_free_req(req); |
2a44f467 | 2104 | } |
9b5f7bd9 | 2105 | return nxt; |
2b188cc1 JA |
2106 | } |
2107 | ||
e65ef56d JA |
2108 | static void io_put_req(struct io_kiocb *req) |
2109 | { | |
2110 | if (refcount_dec_and_test(&req->refs)) | |
2111 | io_free_req(req); | |
2b188cc1 JA |
2112 | } |
2113 | ||
216578e5 PB |
2114 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2115 | { | |
2116 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2117 | ||
2118 | io_free_req(req); | |
2119 | } | |
2120 | ||
2121 | static void io_free_req_deferred(struct io_kiocb *req) | |
2122 | { | |
2123 | int ret; | |
2124 | ||
7cbf1722 | 2125 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2126 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2127 | if (unlikely(ret)) |
2128 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2129 | } |
2130 | ||
2131 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2132 | { | |
2133 | if (refcount_sub_and_test(refs, &req->refs)) | |
2134 | io_free_req_deferred(req); | |
2135 | } | |
2136 | ||
978db57e JA |
2137 | static void io_double_put_req(struct io_kiocb *req) |
2138 | { | |
2139 | /* drop both submit and complete references */ | |
2140 | if (refcount_sub_and_test(2, &req->refs)) | |
2141 | io_free_req(req); | |
2142 | } | |
2143 | ||
6c503150 | 2144 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2145 | { |
2146 | /* See comment at the top of this file */ | |
2147 | smp_rmb(); | |
e23de15f | 2148 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2149 | } |
2150 | ||
fb5ccc98 PB |
2151 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2152 | { | |
2153 | struct io_rings *rings = ctx->rings; | |
2154 | ||
2155 | /* make sure SQ entry isn't read before tail */ | |
2156 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2157 | } | |
2158 | ||
8ff069bf | 2159 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2160 | { |
8ff069bf | 2161 | unsigned int cflags; |
e94f141b | 2162 | |
bcda7baa JA |
2163 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2164 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2165 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2166 | kfree(kbuf); |
2167 | return cflags; | |
e94f141b JA |
2168 | } |
2169 | ||
8ff069bf | 2170 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2171 | { |
4d954c25 | 2172 | struct io_buffer *kbuf; |
bcda7baa | 2173 | |
4d954c25 | 2174 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2175 | return io_put_kbuf(req, kbuf); |
2176 | } | |
2177 | ||
4c6e277c JA |
2178 | static inline bool io_run_task_work(void) |
2179 | { | |
6200b0ae JA |
2180 | /* |
2181 | * Not safe to run on exiting task, and the task_work handling will | |
2182 | * not add work to such a task. | |
2183 | */ | |
2184 | if (unlikely(current->flags & PF_EXITING)) | |
2185 | return false; | |
4c6e277c JA |
2186 | if (current->task_works) { |
2187 | __set_current_state(TASK_RUNNING); | |
2188 | task_work_run(); | |
2189 | return true; | |
2190 | } | |
2191 | ||
2192 | return false; | |
bcda7baa JA |
2193 | } |
2194 | ||
def596e9 JA |
2195 | /* |
2196 | * Find and free completed poll iocbs | |
2197 | */ | |
2198 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2199 | struct list_head *done) | |
2200 | { | |
8237e045 | 2201 | struct req_batch rb; |
def596e9 | 2202 | struct io_kiocb *req; |
bbde017a XW |
2203 | |
2204 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2205 | smp_rmb(); | |
def596e9 | 2206 | |
5af1d13e | 2207 | io_init_req_batch(&rb); |
def596e9 | 2208 | while (!list_empty(done)) { |
bcda7baa JA |
2209 | int cflags = 0; |
2210 | ||
d21ffe7e | 2211 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2212 | list_del(&req->inflight_entry); |
2213 | ||
bbde017a XW |
2214 | if (READ_ONCE(req->result) == -EAGAIN) { |
2215 | req->iopoll_completed = 0; | |
23faba36 | 2216 | if (io_rw_reissue(req)) |
f161340d | 2217 | continue; |
bbde017a | 2218 | } |
def596e9 | 2219 | |
bcda7baa | 2220 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2221 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2222 | |
2223 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2224 | (*nr_events)++; |
2225 | ||
c3524383 | 2226 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2227 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2228 | } |
def596e9 | 2229 | |
09bb8394 | 2230 | io_commit_cqring(ctx); |
80c18e4a | 2231 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2232 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2233 | } |
2234 | ||
def596e9 JA |
2235 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2236 | long min) | |
2237 | { | |
2238 | struct io_kiocb *req, *tmp; | |
2239 | LIST_HEAD(done); | |
2240 | bool spin; | |
2241 | int ret; | |
2242 | ||
2243 | /* | |
2244 | * Only spin for completions if we don't have multiple devices hanging | |
2245 | * off our complete list, and we're under the requested amount. | |
2246 | */ | |
2247 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2248 | ||
2249 | ret = 0; | |
d21ffe7e | 2250 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2251 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2252 | |
2253 | /* | |
581f9810 BM |
2254 | * Move completed and retryable entries to our local lists. |
2255 | * If we find a request that requires polling, break out | |
2256 | * and complete those lists first, if we have entries there. | |
def596e9 | 2257 | */ |
65a6543d | 2258 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2259 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2260 | continue; |
2261 | } | |
2262 | if (!list_empty(&done)) | |
2263 | break; | |
2264 | ||
2265 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2266 | if (ret < 0) | |
2267 | break; | |
2268 | ||
3aadc23e PB |
2269 | /* iopoll may have completed current req */ |
2270 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2271 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2272 | |
def596e9 JA |
2273 | if (ret && spin) |
2274 | spin = false; | |
2275 | ret = 0; | |
2276 | } | |
2277 | ||
2278 | if (!list_empty(&done)) | |
2279 | io_iopoll_complete(ctx, nr_events, &done); | |
2280 | ||
2281 | return ret; | |
2282 | } | |
2283 | ||
2284 | /* | |
d195a66e | 2285 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2286 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2287 | * as a non-spinning completion check. | |
2288 | */ | |
2289 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2290 | long min) | |
2291 | { | |
540e32a0 | 2292 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2293 | int ret; |
2294 | ||
2295 | ret = io_do_iopoll(ctx, nr_events, min); | |
2296 | if (ret < 0) | |
2297 | return ret; | |
eba0a4dd | 2298 | if (*nr_events >= min) |
def596e9 JA |
2299 | return 0; |
2300 | } | |
2301 | ||
2302 | return 1; | |
2303 | } | |
2304 | ||
2305 | /* | |
2306 | * We can't just wait for polled events to come to us, we have to actively | |
2307 | * find and complete them. | |
2308 | */ | |
b2edc0a7 | 2309 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2310 | { |
2311 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2312 | return; | |
2313 | ||
2314 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2315 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2316 | unsigned int nr_events = 0; |
2317 | ||
b2edc0a7 | 2318 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2319 | |
b2edc0a7 PB |
2320 | /* let it sleep and repeat later if can't complete a request */ |
2321 | if (nr_events == 0) | |
2322 | break; | |
08f5439f JA |
2323 | /* |
2324 | * Ensure we allow local-to-the-cpu processing to take place, | |
2325 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2326 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2327 | */ |
3fcee5a6 PB |
2328 | if (need_resched()) { |
2329 | mutex_unlock(&ctx->uring_lock); | |
2330 | cond_resched(); | |
2331 | mutex_lock(&ctx->uring_lock); | |
2332 | } | |
def596e9 JA |
2333 | } |
2334 | mutex_unlock(&ctx->uring_lock); | |
2335 | } | |
2336 | ||
7668b92a | 2337 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2338 | { |
7668b92a | 2339 | unsigned int nr_events = 0; |
2b2ed975 | 2340 | int iters = 0, ret = 0; |
500f9fba | 2341 | |
c7849be9 XW |
2342 | /* |
2343 | * We disallow the app entering submit/complete with polling, but we | |
2344 | * still need to lock the ring to prevent racing with polled issue | |
2345 | * that got punted to a workqueue. | |
2346 | */ | |
2347 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2348 | do { |
a3a0e43f JA |
2349 | /* |
2350 | * Don't enter poll loop if we already have events pending. | |
2351 | * If we do, we can potentially be spinning for commands that | |
2352 | * already triggered a CQE (eg in error). | |
2353 | */ | |
6c503150 PB |
2354 | if (test_bit(0, &ctx->cq_check_overflow)) |
2355 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2356 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2357 | break; |
2358 | ||
500f9fba JA |
2359 | /* |
2360 | * If a submit got punted to a workqueue, we can have the | |
2361 | * application entering polling for a command before it gets | |
2362 | * issued. That app will hold the uring_lock for the duration | |
2363 | * of the poll right here, so we need to take a breather every | |
2364 | * now and then to ensure that the issue has a chance to add | |
2365 | * the poll to the issued list. Otherwise we can spin here | |
2366 | * forever, while the workqueue is stuck trying to acquire the | |
2367 | * very same mutex. | |
2368 | */ | |
2369 | if (!(++iters & 7)) { | |
2370 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2371 | io_run_task_work(); |
500f9fba JA |
2372 | mutex_lock(&ctx->uring_lock); |
2373 | } | |
2374 | ||
7668b92a | 2375 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2376 | if (ret <= 0) |
2377 | break; | |
2378 | ret = 0; | |
7668b92a | 2379 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2380 | |
500f9fba | 2381 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2382 | return ret; |
2383 | } | |
2384 | ||
491381ce | 2385 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2386 | { |
491381ce JA |
2387 | /* |
2388 | * Tell lockdep we inherited freeze protection from submission | |
2389 | * thread. | |
2390 | */ | |
2391 | if (req->flags & REQ_F_ISREG) { | |
2392 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2393 | |
491381ce | 2394 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2395 | } |
491381ce | 2396 | file_end_write(req->file); |
2b188cc1 JA |
2397 | } |
2398 | ||
b63534c4 | 2399 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2400 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2401 | { |
2402 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2403 | int rw, ret; |
b63534c4 | 2404 | struct iov_iter iter; |
b63534c4 | 2405 | |
dc2a6e9a PB |
2406 | /* already prepared */ |
2407 | if (req->async_data) | |
2408 | return true; | |
b63534c4 JA |
2409 | |
2410 | switch (req->opcode) { | |
2411 | case IORING_OP_READV: | |
2412 | case IORING_OP_READ_FIXED: | |
2413 | case IORING_OP_READ: | |
2414 | rw = READ; | |
2415 | break; | |
2416 | case IORING_OP_WRITEV: | |
2417 | case IORING_OP_WRITE_FIXED: | |
2418 | case IORING_OP_WRITE: | |
2419 | rw = WRITE; | |
2420 | break; | |
2421 | default: | |
2422 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2423 | req->opcode); | |
dc2a6e9a | 2424 | return false; |
b63534c4 JA |
2425 | } |
2426 | ||
dc2a6e9a PB |
2427 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2428 | if (ret < 0) | |
2429 | return false; | |
6bf985dc | 2430 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2431 | } |
b63534c4 JA |
2432 | #endif |
2433 | ||
23faba36 | 2434 | static bool io_rw_reissue(struct io_kiocb *req) |
b63534c4 JA |
2435 | { |
2436 | #ifdef CONFIG_BLOCK | |
355afaeb | 2437 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 | 2438 | |
355afaeb JA |
2439 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2440 | return false; | |
75c668cd | 2441 | if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker()) |
b63534c4 | 2442 | return false; |
7c977a58 JA |
2443 | /* |
2444 | * If ref is dying, we might be running poll reap from the exit work. | |
2445 | * Don't attempt to reissue from that path, just let it fail with | |
2446 | * -EAGAIN. | |
2447 | */ | |
2448 | if (percpu_ref_is_dying(&req->ctx->refs)) | |
2449 | return false; | |
b63534c4 | 2450 | |
55e6ac1e PB |
2451 | lockdep_assert_held(&req->ctx->uring_lock); |
2452 | ||
37d1e2e3 | 2453 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2454 | refcount_inc(&req->refs); |
2455 | io_queue_async_work(req); | |
b63534c4 | 2456 | return true; |
fdee946d | 2457 | } |
dc2a6e9a | 2458 | req_set_fail_links(req); |
b63534c4 JA |
2459 | #endif |
2460 | return false; | |
2461 | } | |
2462 | ||
a1d7c393 | 2463 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2464 | unsigned int issue_flags) |
a1d7c393 | 2465 | { |
2f8e45f1 PB |
2466 | int cflags = 0; |
2467 | ||
23faba36 PB |
2468 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2469 | return; | |
2f8e45f1 PB |
2470 | if (res != req->result) |
2471 | req_set_fail_links(req); | |
23faba36 | 2472 | |
2f8e45f1 PB |
2473 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2474 | kiocb_end_write(req); | |
2475 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2476 | cflags = io_put_rw_kbuf(req); | |
2477 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2478 | } |
2479 | ||
2480 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2481 | { | |
9adbd45d | 2482 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2483 | |
889fca73 | 2484 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2485 | } |
2486 | ||
def596e9 JA |
2487 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2488 | { | |
9adbd45d | 2489 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2490 | |
491381ce JA |
2491 | if (kiocb->ki_flags & IOCB_WRITE) |
2492 | kiocb_end_write(req); | |
def596e9 | 2493 | |
2d7d6792 | 2494 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2495 | req_set_fail_links(req); |
bbde017a XW |
2496 | |
2497 | WRITE_ONCE(req->result, res); | |
2498 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2499 | smp_wmb(); |
2500 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2501 | } |
2502 | ||
2503 | /* | |
2504 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2505 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2506 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2507 | * accessing the kiocb cookie. | |
2508 | */ | |
2e9dbe90 | 2509 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2510 | { |
2511 | struct io_ring_ctx *ctx = req->ctx; | |
2512 | ||
2513 | /* | |
2514 | * Track whether we have multiple files in our lists. This will impact | |
2515 | * how we do polling eventually, not spinning if we're on potentially | |
2516 | * different devices. | |
2517 | */ | |
540e32a0 | 2518 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2519 | ctx->poll_multi_file = false; |
2520 | } else if (!ctx->poll_multi_file) { | |
2521 | struct io_kiocb *list_req; | |
2522 | ||
540e32a0 | 2523 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2524 | inflight_entry); |
9adbd45d | 2525 | if (list_req->file != req->file) |
def596e9 JA |
2526 | ctx->poll_multi_file = true; |
2527 | } | |
2528 | ||
2529 | /* | |
2530 | * For fast devices, IO may have already completed. If it has, add | |
2531 | * it to the front so we find it first. | |
2532 | */ | |
65a6543d | 2533 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2534 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2535 | else |
d21ffe7e | 2536 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2537 | |
2e9dbe90 XW |
2538 | /* |
2539 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2540 | * task context or in io worker task context. If current task context is | |
2541 | * sq thread, we don't need to check whether should wake up sq thread. | |
2542 | */ | |
2543 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2544 | wq_has_sleeper(&ctx->sq_data->wait)) |
2545 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2546 | } |
2547 | ||
9f13c35b PB |
2548 | static inline void io_state_file_put(struct io_submit_state *state) |
2549 | { | |
02b23a9a PB |
2550 | if (state->file_refs) { |
2551 | fput_many(state->file, state->file_refs); | |
2552 | state->file_refs = 0; | |
2553 | } | |
9a56a232 JA |
2554 | } |
2555 | ||
2556 | /* | |
2557 | * Get as many references to a file as we have IOs left in this submission, | |
2558 | * assuming most submissions are for one file, or at least that each file | |
2559 | * has more than one submission. | |
2560 | */ | |
8da11c19 | 2561 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2562 | { |
2563 | if (!state) | |
2564 | return fget(fd); | |
2565 | ||
6e1271e6 | 2566 | if (state->file_refs) { |
9a56a232 | 2567 | if (state->fd == fd) { |
6e1271e6 | 2568 | state->file_refs--; |
9a56a232 JA |
2569 | return state->file; |
2570 | } | |
02b23a9a | 2571 | io_state_file_put(state); |
9a56a232 JA |
2572 | } |
2573 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2574 | if (unlikely(!state->file)) |
9a56a232 JA |
2575 | return NULL; |
2576 | ||
2577 | state->fd = fd; | |
6e1271e6 | 2578 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2579 | return state->file; |
2580 | } | |
2581 | ||
4503b767 JA |
2582 | static bool io_bdev_nowait(struct block_device *bdev) |
2583 | { | |
9ba0d0c8 | 2584 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2585 | } |
2586 | ||
2b188cc1 JA |
2587 | /* |
2588 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2589 | * any file. For now, just ensure that anything potentially problematic is done | |
2590 | * inline. | |
2591 | */ | |
af197f50 | 2592 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2593 | { |
2594 | umode_t mode = file_inode(file)->i_mode; | |
2595 | ||
4503b767 | 2596 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2597 | if (IS_ENABLED(CONFIG_BLOCK) && |
2598 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2599 | return true; |
2600 | return false; | |
2601 | } | |
2602 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2603 | return true; |
4503b767 | 2604 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2605 | if (IS_ENABLED(CONFIG_BLOCK) && |
2606 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2607 | file->f_op != &io_uring_fops) |
2608 | return true; | |
2609 | return false; | |
2610 | } | |
2b188cc1 | 2611 | |
c5b85625 JA |
2612 | /* any ->read/write should understand O_NONBLOCK */ |
2613 | if (file->f_flags & O_NONBLOCK) | |
2614 | return true; | |
2615 | ||
af197f50 JA |
2616 | if (!(file->f_mode & FMODE_NOWAIT)) |
2617 | return false; | |
2618 | ||
2619 | if (rw == READ) | |
2620 | return file->f_op->read_iter != NULL; | |
2621 | ||
2622 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2623 | } |
2624 | ||
a88fc400 | 2625 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2626 | { |
def596e9 | 2627 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2628 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2629 | struct file *file = req->file; |
09bb8394 JA |
2630 | unsigned ioprio; |
2631 | int ret; | |
2b188cc1 | 2632 | |
75c668cd | 2633 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2634 | req->flags |= REQ_F_ISREG; |
2635 | ||
2b188cc1 | 2636 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2637 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2638 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2639 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2640 | } |
2b188cc1 | 2641 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2642 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2643 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2644 | if (unlikely(ret)) | |
2645 | return ret; | |
2b188cc1 | 2646 | |
75c668cd PB |
2647 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2648 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2649 | req->flags |= REQ_F_NOWAIT; | |
2650 | ||
2b188cc1 JA |
2651 | ioprio = READ_ONCE(sqe->ioprio); |
2652 | if (ioprio) { | |
2653 | ret = ioprio_check_cap(ioprio); | |
2654 | if (ret) | |
09bb8394 | 2655 | return ret; |
2b188cc1 JA |
2656 | |
2657 | kiocb->ki_ioprio = ioprio; | |
2658 | } else | |
2659 | kiocb->ki_ioprio = get_current_ioprio(); | |
2660 | ||
def596e9 | 2661 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2662 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2663 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2664 | return -EOPNOTSUPP; |
2b188cc1 | 2665 | |
def596e9 JA |
2666 | kiocb->ki_flags |= IOCB_HIPRI; |
2667 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2668 | req->iopoll_completed = 0; |
def596e9 | 2669 | } else { |
09bb8394 JA |
2670 | if (kiocb->ki_flags & IOCB_HIPRI) |
2671 | return -EINVAL; | |
def596e9 JA |
2672 | kiocb->ki_complete = io_complete_rw; |
2673 | } | |
9adbd45d | 2674 | |
3529d8c2 JA |
2675 | req->rw.addr = READ_ONCE(sqe->addr); |
2676 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2677 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2678 | return 0; |
2b188cc1 JA |
2679 | } |
2680 | ||
2681 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2682 | { | |
2683 | switch (ret) { | |
2684 | case -EIOCBQUEUED: | |
2685 | break; | |
2686 | case -ERESTARTSYS: | |
2687 | case -ERESTARTNOINTR: | |
2688 | case -ERESTARTNOHAND: | |
2689 | case -ERESTART_RESTARTBLOCK: | |
2690 | /* | |
2691 | * We can't just restart the syscall, since previously | |
2692 | * submitted sqes may already be in progress. Just fail this | |
2693 | * IO with EINTR. | |
2694 | */ | |
2695 | ret = -EINTR; | |
df561f66 | 2696 | fallthrough; |
2b188cc1 JA |
2697 | default: |
2698 | kiocb->ki_complete(kiocb, ret, 0); | |
2699 | } | |
2700 | } | |
2701 | ||
a1d7c393 | 2702 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2703 | unsigned int issue_flags) |
ba816ad6 | 2704 | { |
ba04291e | 2705 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2706 | struct io_async_rw *io = req->async_data; |
ba04291e | 2707 | |
227c0c96 | 2708 | /* add previously done IO, if any */ |
e8c2bc1f | 2709 | if (io && io->bytes_done > 0) { |
227c0c96 | 2710 | if (ret < 0) |
e8c2bc1f | 2711 | ret = io->bytes_done; |
227c0c96 | 2712 | else |
e8c2bc1f | 2713 | ret += io->bytes_done; |
227c0c96 JA |
2714 | } |
2715 | ||
ba04291e JA |
2716 | if (req->flags & REQ_F_CUR_POS) |
2717 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2718 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2719 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2720 | else |
2721 | io_rw_done(kiocb, ret); | |
2722 | } | |
2723 | ||
847595de | 2724 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2725 | { |
9adbd45d JA |
2726 | struct io_ring_ctx *ctx = req->ctx; |
2727 | size_t len = req->rw.len; | |
edafccee | 2728 | struct io_mapped_ubuf *imu; |
4be1c615 | 2729 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2730 | size_t offset; |
2731 | u64 buf_addr; | |
2732 | ||
edafccee JA |
2733 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2734 | return -EFAULT; | |
edafccee JA |
2735 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2736 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2737 | buf_addr = req->rw.addr; |
edafccee JA |
2738 | |
2739 | /* overflow */ | |
2740 | if (buf_addr + len < buf_addr) | |
2741 | return -EFAULT; | |
2742 | /* not inside the mapped region */ | |
2743 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2744 | return -EFAULT; | |
2745 | ||
2746 | /* | |
2747 | * May not be a start of buffer, set size appropriately | |
2748 | * and advance us to the beginning. | |
2749 | */ | |
2750 | offset = buf_addr - imu->ubuf; | |
2751 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2752 | |
2753 | if (offset) { | |
2754 | /* | |
2755 | * Don't use iov_iter_advance() here, as it's really slow for | |
2756 | * using the latter parts of a big fixed buffer - it iterates | |
2757 | * over each segment manually. We can cheat a bit here, because | |
2758 | * we know that: | |
2759 | * | |
2760 | * 1) it's a BVEC iter, we set it up | |
2761 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2762 | * first and last bvec | |
2763 | * | |
2764 | * So just find our index, and adjust the iterator afterwards. | |
2765 | * If the offset is within the first bvec (or the whole first | |
2766 | * bvec, just use iov_iter_advance(). This makes it easier | |
2767 | * since we can just skip the first segment, which may not | |
2768 | * be PAGE_SIZE aligned. | |
2769 | */ | |
2770 | const struct bio_vec *bvec = imu->bvec; | |
2771 | ||
2772 | if (offset <= bvec->bv_len) { | |
2773 | iov_iter_advance(iter, offset); | |
2774 | } else { | |
2775 | unsigned long seg_skip; | |
2776 | ||
2777 | /* skip first vec */ | |
2778 | offset -= bvec->bv_len; | |
2779 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2780 | ||
2781 | iter->bvec = bvec + seg_skip; | |
2782 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2783 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2784 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2785 | } |
2786 | } | |
2787 | ||
847595de | 2788 | return 0; |
edafccee JA |
2789 | } |
2790 | ||
bcda7baa JA |
2791 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2792 | { | |
2793 | if (needs_lock) | |
2794 | mutex_unlock(&ctx->uring_lock); | |
2795 | } | |
2796 | ||
2797 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2798 | { | |
2799 | /* | |
2800 | * "Normal" inline submissions always hold the uring_lock, since we | |
2801 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2802 | * The only exception is when we've detached the request and issue it | |
2803 | * from an async worker thread, grab the lock for that case. | |
2804 | */ | |
2805 | if (needs_lock) | |
2806 | mutex_lock(&ctx->uring_lock); | |
2807 | } | |
2808 | ||
2809 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2810 | int bgid, struct io_buffer *kbuf, | |
2811 | bool needs_lock) | |
2812 | { | |
2813 | struct io_buffer *head; | |
2814 | ||
2815 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2816 | return kbuf; | |
2817 | ||
2818 | io_ring_submit_lock(req->ctx, needs_lock); | |
2819 | ||
2820 | lockdep_assert_held(&req->ctx->uring_lock); | |
2821 | ||
2822 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2823 | if (head) { | |
2824 | if (!list_empty(&head->list)) { | |
2825 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2826 | list); | |
2827 | list_del(&kbuf->list); | |
2828 | } else { | |
2829 | kbuf = head; | |
2830 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2831 | } | |
2832 | if (*len > kbuf->len) | |
2833 | *len = kbuf->len; | |
2834 | } else { | |
2835 | kbuf = ERR_PTR(-ENOBUFS); | |
2836 | } | |
2837 | ||
2838 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2839 | ||
2840 | return kbuf; | |
2841 | } | |
2842 | ||
4d954c25 JA |
2843 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2844 | bool needs_lock) | |
2845 | { | |
2846 | struct io_buffer *kbuf; | |
4f4eeba8 | 2847 | u16 bgid; |
4d954c25 JA |
2848 | |
2849 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2850 | bgid = req->buf_index; |
4d954c25 JA |
2851 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2852 | if (IS_ERR(kbuf)) | |
2853 | return kbuf; | |
2854 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2855 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2856 | return u64_to_user_ptr(kbuf->addr); | |
2857 | } | |
2858 | ||
2859 | #ifdef CONFIG_COMPAT | |
2860 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2861 | bool needs_lock) | |
2862 | { | |
2863 | struct compat_iovec __user *uiov; | |
2864 | compat_ssize_t clen; | |
2865 | void __user *buf; | |
2866 | ssize_t len; | |
2867 | ||
2868 | uiov = u64_to_user_ptr(req->rw.addr); | |
2869 | if (!access_ok(uiov, sizeof(*uiov))) | |
2870 | return -EFAULT; | |
2871 | if (__get_user(clen, &uiov->iov_len)) | |
2872 | return -EFAULT; | |
2873 | if (clen < 0) | |
2874 | return -EINVAL; | |
2875 | ||
2876 | len = clen; | |
2877 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2878 | if (IS_ERR(buf)) | |
2879 | return PTR_ERR(buf); | |
2880 | iov[0].iov_base = buf; | |
2881 | iov[0].iov_len = (compat_size_t) len; | |
2882 | return 0; | |
2883 | } | |
2884 | #endif | |
2885 | ||
2886 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2887 | bool needs_lock) | |
2888 | { | |
2889 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2890 | void __user *buf; | |
2891 | ssize_t len; | |
2892 | ||
2893 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2894 | return -EFAULT; | |
2895 | ||
2896 | len = iov[0].iov_len; | |
2897 | if (len < 0) | |
2898 | return -EINVAL; | |
2899 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2900 | if (IS_ERR(buf)) | |
2901 | return PTR_ERR(buf); | |
2902 | iov[0].iov_base = buf; | |
2903 | iov[0].iov_len = len; | |
2904 | return 0; | |
2905 | } | |
2906 | ||
2907 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2908 | bool needs_lock) | |
2909 | { | |
dddb3e26 JA |
2910 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2911 | struct io_buffer *kbuf; | |
2912 | ||
2913 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2914 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2915 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2916 | return 0; |
dddb3e26 | 2917 | } |
dd201662 | 2918 | if (req->rw.len != 1) |
4d954c25 JA |
2919 | return -EINVAL; |
2920 | ||
2921 | #ifdef CONFIG_COMPAT | |
2922 | if (req->ctx->compat) | |
2923 | return io_compat_import(req, iov, needs_lock); | |
2924 | #endif | |
2925 | ||
2926 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2927 | } | |
2928 | ||
847595de PB |
2929 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2930 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2931 | { |
9adbd45d JA |
2932 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2933 | size_t sqe_len = req->rw.len; | |
847595de | 2934 | u8 opcode = req->opcode; |
4d954c25 | 2935 | ssize_t ret; |
edafccee | 2936 | |
7d009165 | 2937 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2938 | *iovec = NULL; |
9adbd45d | 2939 | return io_import_fixed(req, rw, iter); |
edafccee | 2940 | } |
2b188cc1 | 2941 | |
bcda7baa | 2942 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2943 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2944 | return -EINVAL; |
2945 | ||
3a6820f2 | 2946 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2947 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2948 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2949 | if (IS_ERR(buf)) |
4d954c25 | 2950 | return PTR_ERR(buf); |
3f9d6441 | 2951 | req->rw.len = sqe_len; |
bcda7baa JA |
2952 | } |
2953 | ||
3a6820f2 JA |
2954 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2955 | *iovec = NULL; | |
10fc72e4 | 2956 | return ret; |
3a6820f2 JA |
2957 | } |
2958 | ||
4d954c25 JA |
2959 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2960 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2961 | if (!ret) |
2962 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2963 | *iovec = NULL; |
2964 | return ret; | |
2965 | } | |
2966 | ||
89cd35c5 CH |
2967 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2968 | req->ctx->compat); | |
2b188cc1 JA |
2969 | } |
2970 | ||
0fef9483 JA |
2971 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2972 | { | |
5b09e37e | 2973 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
2974 | } |
2975 | ||
31b51510 | 2976 | /* |
32960613 JA |
2977 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2978 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2979 | */ |
4017eb91 | 2980 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 2981 | { |
4017eb91 JA |
2982 | struct kiocb *kiocb = &req->rw.kiocb; |
2983 | struct file *file = req->file; | |
32960613 JA |
2984 | ssize_t ret = 0; |
2985 | ||
2986 | /* | |
2987 | * Don't support polled IO through this interface, and we can't | |
2988 | * support non-blocking either. For the latter, this just causes | |
2989 | * the kiocb to be handled from an async context. | |
2990 | */ | |
2991 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2992 | return -EOPNOTSUPP; | |
2993 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2994 | return -EAGAIN; | |
2995 | ||
2996 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2997 | struct iovec iovec; |
32960613 JA |
2998 | ssize_t nr; |
2999 | ||
311ae9e1 PB |
3000 | if (!iov_iter_is_bvec(iter)) { |
3001 | iovec = iov_iter_iovec(iter); | |
3002 | } else { | |
4017eb91 JA |
3003 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3004 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3005 | } |
3006 | ||
32960613 JA |
3007 | if (rw == READ) { |
3008 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3009 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3010 | } else { |
3011 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3012 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3013 | } |
3014 | ||
3015 | if (nr < 0) { | |
3016 | if (!ret) | |
3017 | ret = nr; | |
3018 | break; | |
3019 | } | |
3020 | ret += nr; | |
3021 | if (nr != iovec.iov_len) | |
3022 | break; | |
4017eb91 JA |
3023 | req->rw.len -= nr; |
3024 | req->rw.addr += nr; | |
32960613 JA |
3025 | iov_iter_advance(iter, nr); |
3026 | } | |
3027 | ||
3028 | return ret; | |
3029 | } | |
3030 | ||
ff6165b2 JA |
3031 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3032 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3033 | { |
e8c2bc1f | 3034 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3035 | |
ff6165b2 | 3036 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3037 | rw->free_iovec = iovec; |
227c0c96 | 3038 | rw->bytes_done = 0; |
ff6165b2 | 3039 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3040 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3041 | return; |
b64e3444 | 3042 | if (!iovec) { |
ff6165b2 JA |
3043 | unsigned iov_off = 0; |
3044 | ||
3045 | rw->iter.iov = rw->fast_iov; | |
3046 | if (iter->iov != fast_iov) { | |
3047 | iov_off = iter->iov - fast_iov; | |
3048 | rw->iter.iov += iov_off; | |
3049 | } | |
3050 | if (rw->fast_iov != fast_iov) | |
3051 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3052 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3053 | } else { |
3054 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3055 | } |
3056 | } | |
3057 | ||
e8c2bc1f | 3058 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3059 | { |
e8c2bc1f JA |
3060 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3061 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3062 | return req->async_data == NULL; | |
3d9932a8 XW |
3063 | } |
3064 | ||
e8c2bc1f | 3065 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3066 | { |
e8c2bc1f | 3067 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3068 | return 0; |
3d9932a8 | 3069 | |
e8c2bc1f | 3070 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3071 | } |
3072 | ||
ff6165b2 JA |
3073 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3074 | const struct iovec *fast_iov, | |
227c0c96 | 3075 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3076 | { |
e8c2bc1f | 3077 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3078 | return 0; |
e8c2bc1f | 3079 | if (!req->async_data) { |
6bf985dc PB |
3080 | if (__io_alloc_async_data(req)) { |
3081 | kfree(iovec); | |
5d204bcf | 3082 | return -ENOMEM; |
6bf985dc | 3083 | } |
b7bb4f7d | 3084 | |
ff6165b2 | 3085 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3086 | } |
b7bb4f7d | 3087 | return 0; |
f67676d1 JA |
3088 | } |
3089 | ||
73debe68 | 3090 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3091 | { |
e8c2bc1f | 3092 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3093 | struct iovec *iov = iorw->fast_iov; |
847595de | 3094 | int ret; |
c3e330a4 | 3095 | |
2846c481 | 3096 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3097 | if (unlikely(ret < 0)) |
3098 | return ret; | |
3099 | ||
ab0b196c PB |
3100 | iorw->bytes_done = 0; |
3101 | iorw->free_iovec = iov; | |
3102 | if (iov) | |
3103 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3104 | return 0; |
3105 | } | |
3106 | ||
73debe68 | 3107 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3108 | { |
3529d8c2 JA |
3109 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3110 | return -EBADF; | |
93642ef8 | 3111 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3112 | } |
3113 | ||
c1dd91d1 JA |
3114 | /* |
3115 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3116 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3117 | * This gets called when the page is unlocked, and we generally expect that to | |
3118 | * happen when the page IO is completed and the page is now uptodate. This will | |
3119 | * queue a task_work based retry of the operation, attempting to copy the data | |
3120 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3121 | * do a thread based blocking retry of the operation. That's the unexpected | |
3122 | * slow path. | |
3123 | */ | |
bcf5a063 JA |
3124 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3125 | int sync, void *arg) | |
3126 | { | |
3127 | struct wait_page_queue *wpq; | |
3128 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3129 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3130 | |
3131 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3132 | ||
cdc8fcb4 LT |
3133 | if (!wake_page_match(wpq, key)) |
3134 | return 0; | |
3135 | ||
c8d317aa | 3136 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3137 | list_del_init(&wait->entry); |
3138 | ||
bcf5a063 JA |
3139 | /* submit ref gets dropped, acquire a new one */ |
3140 | refcount_inc(&req->refs); | |
921b9054 | 3141 | io_req_task_queue(req); |
bcf5a063 JA |
3142 | return 1; |
3143 | } | |
3144 | ||
c1dd91d1 JA |
3145 | /* |
3146 | * This controls whether a given IO request should be armed for async page | |
3147 | * based retry. If we return false here, the request is handed to the async | |
3148 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3149 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3150 | * will either succeed because the page is now uptodate and unlocked, or it | |
3151 | * will register a callback when the page is unlocked at IO completion. Through | |
3152 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3153 | * That retry will attempt the buffered read again. The retry will generally | |
3154 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3155 | * async worker threads for a blocking retry. | |
3156 | */ | |
227c0c96 | 3157 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3158 | { |
e8c2bc1f JA |
3159 | struct io_async_rw *rw = req->async_data; |
3160 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3161 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3162 | |
bcf5a063 JA |
3163 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3164 | if (req->flags & REQ_F_NOWAIT) | |
3165 | return false; | |
f67676d1 | 3166 | |
227c0c96 | 3167 | /* Only for buffered IO */ |
3b2a4439 | 3168 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3169 | return false; |
3b2a4439 | 3170 | |
bcf5a063 JA |
3171 | /* |
3172 | * just use poll if we can, and don't attempt if the fs doesn't | |
3173 | * support callback based unlocks | |
3174 | */ | |
3175 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3176 | return false; | |
f67676d1 | 3177 | |
3b2a4439 JA |
3178 | wait->wait.func = io_async_buf_func; |
3179 | wait->wait.private = req; | |
3180 | wait->wait.flags = 0; | |
3181 | INIT_LIST_HEAD(&wait->wait.entry); | |
3182 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3183 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3184 | kiocb->ki_waitq = wait; |
3b2a4439 | 3185 | return true; |
bcf5a063 JA |
3186 | } |
3187 | ||
3188 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3189 | { | |
3190 | if (req->file->f_op->read_iter) | |
3191 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3192 | else if (req->file->f_op->read) |
4017eb91 | 3193 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3194 | else |
3195 | return -EINVAL; | |
f67676d1 JA |
3196 | } |
3197 | ||
889fca73 | 3198 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3199 | { |
3200 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3201 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3202 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3203 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3204 | ssize_t io_size, ret, ret2; |
45d189c6 | 3205 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3206 | |
2846c481 | 3207 | if (rw) { |
e8c2bc1f | 3208 | iter = &rw->iter; |
2846c481 PB |
3209 | iovec = NULL; |
3210 | } else { | |
3211 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3212 | if (ret < 0) | |
3213 | return ret; | |
3214 | } | |
632546c4 | 3215 | io_size = iov_iter_count(iter); |
fa15bafb | 3216 | req->result = io_size; |
2b188cc1 | 3217 | |
fd6c2e4c JA |
3218 | /* Ensure we clear previously set non-block flag */ |
3219 | if (!force_nonblock) | |
29de5f6a | 3220 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3221 | else |
3222 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3223 | ||
24c74678 | 3224 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3225 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3226 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3227 | return ret ?: -EAGAIN; |
6713e7a6 | 3228 | } |
9e645e11 | 3229 | |
632546c4 | 3230 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3231 | if (unlikely(ret)) { |
3232 | kfree(iovec); | |
3233 | return ret; | |
3234 | } | |
2b188cc1 | 3235 | |
227c0c96 | 3236 | ret = io_iter_do_read(req, iter); |
32960613 | 3237 | |
57cd657b | 3238 | if (ret == -EIOCBQUEUED) { |
fe1cdd55 | 3239 | goto out_free; |
227c0c96 | 3240 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3241 | /* IOPOLL retry should happen for io-wq threads */ |
3242 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3243 | goto done; |
75c668cd PB |
3244 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3245 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3246 | goto done; |
84216315 | 3247 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3248 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3249 | ret = 0; |
7335e3bf | 3250 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3251 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3252 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3253 | goto done; |
227c0c96 JA |
3254 | } |
3255 | ||
227c0c96 | 3256 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3257 | if (ret2) |
3258 | return ret2; | |
3259 | ||
fe1cdd55 | 3260 | iovec = NULL; |
e8c2bc1f | 3261 | rw = req->async_data; |
227c0c96 | 3262 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3263 | iter = &rw->iter; |
227c0c96 | 3264 | |
b23df91b PB |
3265 | do { |
3266 | io_size -= ret; | |
3267 | rw->bytes_done += ret; | |
3268 | /* if we can retry, do so with the callbacks armed */ | |
3269 | if (!io_rw_should_retry(req)) { | |
3270 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3271 | return -EAGAIN; | |
3272 | } | |
3273 | ||
3274 | /* | |
3275 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3276 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3277 | * desired page gets unlocked. We can also get a partial read | |
3278 | * here, and if we do, then just retry at the new offset. | |
3279 | */ | |
3280 | ret = io_iter_do_read(req, iter); | |
3281 | if (ret == -EIOCBQUEUED) | |
3282 | return 0; | |
227c0c96 | 3283 | /* we got some bytes, but not all. retry. */ |
b23df91b | 3284 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3285 | done: |
889fca73 | 3286 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3287 | out_free: |
3288 | /* it's faster to check here then delegate to kfree */ | |
3289 | if (iovec) | |
3290 | kfree(iovec); | |
5ea5dd45 | 3291 | return 0; |
2b188cc1 JA |
3292 | } |
3293 | ||
73debe68 | 3294 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3295 | { |
3529d8c2 JA |
3296 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3297 | return -EBADF; | |
93642ef8 | 3298 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3299 | } |
3300 | ||
889fca73 | 3301 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3302 | { |
3303 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3304 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3305 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3306 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3307 | ssize_t ret, ret2, io_size; |
45d189c6 | 3308 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3309 | |
2846c481 | 3310 | if (rw) { |
e8c2bc1f | 3311 | iter = &rw->iter; |
2846c481 PB |
3312 | iovec = NULL; |
3313 | } else { | |
3314 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3315 | if (ret < 0) | |
3316 | return ret; | |
3317 | } | |
632546c4 | 3318 | io_size = iov_iter_count(iter); |
fa15bafb | 3319 | req->result = io_size; |
2b188cc1 | 3320 | |
fd6c2e4c JA |
3321 | /* Ensure we clear previously set non-block flag */ |
3322 | if (!force_nonblock) | |
a88fc400 PB |
3323 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3324 | else | |
3325 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3326 | |
24c74678 | 3327 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3328 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3329 | goto copy_iov; |
31b51510 | 3330 | |
10d59345 JA |
3331 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3332 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3333 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3334 | goto copy_iov; |
31b51510 | 3335 | |
632546c4 | 3336 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3337 | if (unlikely(ret)) |
3338 | goto out_free; | |
4ed734b0 | 3339 | |
fa15bafb PB |
3340 | /* |
3341 | * Open-code file_start_write here to grab freeze protection, | |
3342 | * which will be released by another thread in | |
3343 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3344 | * released so that it doesn't complain about the held lock when | |
3345 | * we return to userspace. | |
3346 | */ | |
3347 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3348 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3349 | __sb_writers_release(file_inode(req->file)->i_sb, |
3350 | SB_FREEZE_WRITE); | |
3351 | } | |
3352 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3353 | |
fa15bafb | 3354 | if (req->file->f_op->write_iter) |
ff6165b2 | 3355 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3356 | else if (req->file->f_op->write) |
4017eb91 | 3357 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3358 | else |
3359 | ret2 = -EINVAL; | |
4ed734b0 | 3360 | |
fa15bafb PB |
3361 | /* |
3362 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3363 | * retry them without IOCB_NOWAIT. | |
3364 | */ | |
3365 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3366 | ret2 = -EAGAIN; | |
75c668cd PB |
3367 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3368 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3369 | goto done; |
fa15bafb | 3370 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3371 | /* IOPOLL retry should happen for io-wq threads */ |
3372 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3373 | goto copy_iov; | |
355afaeb | 3374 | done: |
889fca73 | 3375 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3376 | } else { |
f67676d1 | 3377 | copy_iov: |
84216315 | 3378 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3379 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3380 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3381 | return ret ?: -EAGAIN; |
2b188cc1 | 3382 | } |
31b51510 | 3383 | out_free: |
f261c168 | 3384 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3385 | if (iovec) |
6f2cc166 | 3386 | kfree(iovec); |
2b188cc1 JA |
3387 | return ret; |
3388 | } | |
3389 | ||
80a261fd JA |
3390 | static int io_renameat_prep(struct io_kiocb *req, |
3391 | const struct io_uring_sqe *sqe) | |
3392 | { | |
3393 | struct io_rename *ren = &req->rename; | |
3394 | const char __user *oldf, *newf; | |
3395 | ||
3396 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3397 | return -EBADF; | |
3398 | ||
3399 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3400 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3401 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3402 | ren->new_dfd = READ_ONCE(sqe->len); | |
3403 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3404 | ||
3405 | ren->oldpath = getname(oldf); | |
3406 | if (IS_ERR(ren->oldpath)) | |
3407 | return PTR_ERR(ren->oldpath); | |
3408 | ||
3409 | ren->newpath = getname(newf); | |
3410 | if (IS_ERR(ren->newpath)) { | |
3411 | putname(ren->oldpath); | |
3412 | return PTR_ERR(ren->newpath); | |
3413 | } | |
3414 | ||
3415 | req->flags |= REQ_F_NEED_CLEANUP; | |
3416 | return 0; | |
3417 | } | |
3418 | ||
45d189c6 | 3419 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3420 | { |
3421 | struct io_rename *ren = &req->rename; | |
3422 | int ret; | |
3423 | ||
45d189c6 | 3424 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3425 | return -EAGAIN; |
3426 | ||
3427 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3428 | ren->newpath, ren->flags); | |
3429 | ||
3430 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3431 | if (ret < 0) | |
3432 | req_set_fail_links(req); | |
3433 | io_req_complete(req, ret); | |
3434 | return 0; | |
3435 | } | |
3436 | ||
14a1143b JA |
3437 | static int io_unlinkat_prep(struct io_kiocb *req, |
3438 | const struct io_uring_sqe *sqe) | |
3439 | { | |
3440 | struct io_unlink *un = &req->unlink; | |
3441 | const char __user *fname; | |
3442 | ||
3443 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3444 | return -EBADF; | |
3445 | ||
3446 | un->dfd = READ_ONCE(sqe->fd); | |
3447 | ||
3448 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3449 | if (un->flags & ~AT_REMOVEDIR) | |
3450 | return -EINVAL; | |
3451 | ||
3452 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3453 | un->filename = getname(fname); | |
3454 | if (IS_ERR(un->filename)) | |
3455 | return PTR_ERR(un->filename); | |
3456 | ||
3457 | req->flags |= REQ_F_NEED_CLEANUP; | |
3458 | return 0; | |
3459 | } | |
3460 | ||
45d189c6 | 3461 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3462 | { |
3463 | struct io_unlink *un = &req->unlink; | |
3464 | int ret; | |
3465 | ||
45d189c6 | 3466 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3467 | return -EAGAIN; |
3468 | ||
3469 | if (un->flags & AT_REMOVEDIR) | |
3470 | ret = do_rmdir(un->dfd, un->filename); | |
3471 | else | |
3472 | ret = do_unlinkat(un->dfd, un->filename); | |
3473 | ||
3474 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3475 | if (ret < 0) | |
3476 | req_set_fail_links(req); | |
3477 | io_req_complete(req, ret); | |
3478 | return 0; | |
3479 | } | |
3480 | ||
36f4fa68 JA |
3481 | static int io_shutdown_prep(struct io_kiocb *req, |
3482 | const struct io_uring_sqe *sqe) | |
3483 | { | |
3484 | #if defined(CONFIG_NET) | |
3485 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3486 | return -EINVAL; | |
3487 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3488 | sqe->buf_index) | |
3489 | return -EINVAL; | |
3490 | ||
3491 | req->shutdown.how = READ_ONCE(sqe->len); | |
3492 | return 0; | |
3493 | #else | |
3494 | return -EOPNOTSUPP; | |
3495 | #endif | |
3496 | } | |
3497 | ||
45d189c6 | 3498 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3499 | { |
3500 | #if defined(CONFIG_NET) | |
3501 | struct socket *sock; | |
3502 | int ret; | |
3503 | ||
45d189c6 | 3504 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3505 | return -EAGAIN; |
3506 | ||
48aba79b | 3507 | sock = sock_from_file(req->file); |
36f4fa68 | 3508 | if (unlikely(!sock)) |
48aba79b | 3509 | return -ENOTSOCK; |
36f4fa68 JA |
3510 | |
3511 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3512 | if (ret < 0) |
3513 | req_set_fail_links(req); | |
36f4fa68 JA |
3514 | io_req_complete(req, ret); |
3515 | return 0; | |
3516 | #else | |
3517 | return -EOPNOTSUPP; | |
3518 | #endif | |
3519 | } | |
3520 | ||
f2a8d5c7 PB |
3521 | static int __io_splice_prep(struct io_kiocb *req, |
3522 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3523 | { |
3524 | struct io_splice* sp = &req->splice; | |
3525 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3526 | |
3232dd02 PB |
3527 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3528 | return -EINVAL; | |
7d67af2c PB |
3529 | |
3530 | sp->file_in = NULL; | |
7d67af2c PB |
3531 | sp->len = READ_ONCE(sqe->len); |
3532 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3533 | ||
3534 | if (unlikely(sp->flags & ~valid_flags)) | |
3535 | return -EINVAL; | |
3536 | ||
8371adf5 PB |
3537 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3538 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3539 | if (!sp->file_in) | |
3540 | return -EBADF; | |
7d67af2c PB |
3541 | req->flags |= REQ_F_NEED_CLEANUP; |
3542 | ||
7cdaf587 XW |
3543 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3544 | /* | |
3545 | * Splice operation will be punted aync, and here need to | |
3546 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3547 | */ | |
7d67af2c | 3548 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3549 | } |
7d67af2c PB |
3550 | |
3551 | return 0; | |
3552 | } | |
3553 | ||
f2a8d5c7 PB |
3554 | static int io_tee_prep(struct io_kiocb *req, |
3555 | const struct io_uring_sqe *sqe) | |
3556 | { | |
3557 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3558 | return -EINVAL; | |
3559 | return __io_splice_prep(req, sqe); | |
3560 | } | |
3561 | ||
45d189c6 | 3562 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3563 | { |
3564 | struct io_splice *sp = &req->splice; | |
3565 | struct file *in = sp->file_in; | |
3566 | struct file *out = sp->file_out; | |
3567 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3568 | long ret = 0; | |
3569 | ||
45d189c6 | 3570 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3571 | return -EAGAIN; |
3572 | if (sp->len) | |
3573 | ret = do_tee(in, out, sp->len, flags); | |
3574 | ||
3575 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3576 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3577 | ||
f2a8d5c7 PB |
3578 | if (ret != sp->len) |
3579 | req_set_fail_links(req); | |
e1e16097 | 3580 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3581 | return 0; |
3582 | } | |
3583 | ||
3584 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3585 | { | |
3586 | struct io_splice* sp = &req->splice; | |
3587 | ||
3588 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3589 | sp->off_out = READ_ONCE(sqe->off); | |
3590 | return __io_splice_prep(req, sqe); | |
3591 | } | |
3592 | ||
45d189c6 | 3593 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3594 | { |
3595 | struct io_splice *sp = &req->splice; | |
3596 | struct file *in = sp->file_in; | |
3597 | struct file *out = sp->file_out; | |
3598 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3599 | loff_t *poff_in, *poff_out; | |
c9687426 | 3600 | long ret = 0; |
7d67af2c | 3601 | |
45d189c6 | 3602 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3603 | return -EAGAIN; |
7d67af2c PB |
3604 | |
3605 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3606 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3607 | |
948a7749 | 3608 | if (sp->len) |
c9687426 | 3609 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3610 | |
3611 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3612 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3613 | ||
7d67af2c PB |
3614 | if (ret != sp->len) |
3615 | req_set_fail_links(req); | |
e1e16097 | 3616 | io_req_complete(req, ret); |
7d67af2c PB |
3617 | return 0; |
3618 | } | |
3619 | ||
2b188cc1 JA |
3620 | /* |
3621 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3622 | */ | |
889fca73 | 3623 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3624 | { |
3625 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3626 | |
def596e9 JA |
3627 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3628 | return -EINVAL; | |
3629 | ||
889fca73 | 3630 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3631 | return 0; |
3632 | } | |
3633 | ||
1155c76a | 3634 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3635 | { |
6b06314c | 3636 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3637 | |
09bb8394 JA |
3638 | if (!req->file) |
3639 | return -EBADF; | |
c992fe29 | 3640 | |
6b06314c | 3641 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3642 | return -EINVAL; |
edafccee | 3643 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3644 | return -EINVAL; |
3645 | ||
8ed8d3c3 JA |
3646 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3647 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3648 | return -EINVAL; | |
3649 | ||
3650 | req->sync.off = READ_ONCE(sqe->off); | |
3651 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3652 | return 0; |
3653 | } | |
3654 | ||
45d189c6 | 3655 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3656 | { |
8ed8d3c3 | 3657 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3658 | int ret; |
3659 | ||
ac45abc0 | 3660 | /* fsync always requires a blocking context */ |
45d189c6 | 3661 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3662 | return -EAGAIN; |
3663 | ||
9adbd45d | 3664 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3665 | end > 0 ? end : LLONG_MAX, |
3666 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3667 | if (ret < 0) | |
3668 | req_set_fail_links(req); | |
e1e16097 | 3669 | io_req_complete(req, ret); |
c992fe29 CH |
3670 | return 0; |
3671 | } | |
3672 | ||
d63d1b5e JA |
3673 | static int io_fallocate_prep(struct io_kiocb *req, |
3674 | const struct io_uring_sqe *sqe) | |
3675 | { | |
3676 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3677 | return -EINVAL; | |
3232dd02 PB |
3678 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3679 | return -EINVAL; | |
d63d1b5e JA |
3680 | |
3681 | req->sync.off = READ_ONCE(sqe->off); | |
3682 | req->sync.len = READ_ONCE(sqe->addr); | |
3683 | req->sync.mode = READ_ONCE(sqe->len); | |
3684 | return 0; | |
3685 | } | |
3686 | ||
45d189c6 | 3687 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3688 | { |
ac45abc0 PB |
3689 | int ret; |
3690 | ||
d63d1b5e | 3691 | /* fallocate always requiring blocking context */ |
45d189c6 | 3692 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3693 | return -EAGAIN; |
ac45abc0 PB |
3694 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3695 | req->sync.len); | |
ac45abc0 PB |
3696 | if (ret < 0) |
3697 | req_set_fail_links(req); | |
e1e16097 | 3698 | io_req_complete(req, ret); |
5d17b4a4 JA |
3699 | return 0; |
3700 | } | |
3701 | ||
ec65fea5 | 3702 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3703 | { |
f8748881 | 3704 | const char __user *fname; |
15b71abe | 3705 | int ret; |
b7bb4f7d | 3706 | |
ec65fea5 | 3707 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3708 | return -EINVAL; |
ec65fea5 | 3709 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3710 | return -EBADF; |
03b1230c | 3711 | |
ec65fea5 PB |
3712 | /* open.how should be already initialised */ |
3713 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3714 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3715 | |
25e72d10 PB |
3716 | req->open.dfd = READ_ONCE(sqe->fd); |
3717 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3718 | req->open.filename = getname(fname); |
15b71abe JA |
3719 | if (IS_ERR(req->open.filename)) { |
3720 | ret = PTR_ERR(req->open.filename); | |
3721 | req->open.filename = NULL; | |
3722 | return ret; | |
3723 | } | |
4022e7af | 3724 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3725 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3726 | return 0; |
03b1230c JA |
3727 | } |
3728 | ||
ec65fea5 PB |
3729 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3730 | { | |
3731 | u64 flags, mode; | |
3732 | ||
14587a46 | 3733 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3734 | return -EINVAL; |
ec65fea5 PB |
3735 | mode = READ_ONCE(sqe->len); |
3736 | flags = READ_ONCE(sqe->open_flags); | |
3737 | req->open.how = build_open_how(flags, mode); | |
3738 | return __io_openat_prep(req, sqe); | |
3739 | } | |
3740 | ||
cebdb986 | 3741 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3742 | { |
cebdb986 | 3743 | struct open_how __user *how; |
cebdb986 | 3744 | size_t len; |
0fa03c62 JA |
3745 | int ret; |
3746 | ||
14587a46 | 3747 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3748 | return -EINVAL; |
cebdb986 JA |
3749 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3750 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3751 | if (len < OPEN_HOW_SIZE_VER0) |
3752 | return -EINVAL; | |
3529d8c2 | 3753 | |
cebdb986 JA |
3754 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3755 | len); | |
3756 | if (ret) | |
3757 | return ret; | |
3529d8c2 | 3758 | |
ec65fea5 | 3759 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3760 | } |
3761 | ||
45d189c6 | 3762 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3763 | { |
3764 | struct open_flags op; | |
15b71abe | 3765 | struct file *file; |
3a81fd02 JA |
3766 | bool nonblock_set; |
3767 | bool resolve_nonblock; | |
15b71abe JA |
3768 | int ret; |
3769 | ||
cebdb986 | 3770 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3771 | if (ret) |
3772 | goto err; | |
3a81fd02 JA |
3773 | nonblock_set = op.open_flag & O_NONBLOCK; |
3774 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3775 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3776 | /* |
3777 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3778 | * it'll always -EAGAIN | |
3779 | */ | |
3780 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3781 | return -EAGAIN; | |
3782 | op.lookup_flags |= LOOKUP_CACHED; | |
3783 | op.open_flag |= O_NONBLOCK; | |
3784 | } | |
15b71abe | 3785 | |
4022e7af | 3786 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3787 | if (ret < 0) |
3788 | goto err; | |
3789 | ||
3790 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3791 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3792 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3793 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3794 | /* |
3a81fd02 JA |
3795 | * We could hang on to this 'fd', but seems like marginal |
3796 | * gain for something that is now known to be a slower path. | |
3797 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3798 | */ |
3a81fd02 JA |
3799 | put_unused_fd(ret); |
3800 | return -EAGAIN; | |
3801 | } | |
3802 | ||
15b71abe JA |
3803 | if (IS_ERR(file)) { |
3804 | put_unused_fd(ret); | |
3805 | ret = PTR_ERR(file); | |
3806 | } else { | |
45d189c6 | 3807 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3808 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3809 | fsnotify_open(file); |
3810 | fd_install(ret, file); | |
3811 | } | |
3812 | err: | |
3813 | putname(req->open.filename); | |
8fef80bf | 3814 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3815 | if (ret < 0) |
3816 | req_set_fail_links(req); | |
e1e16097 | 3817 | io_req_complete(req, ret); |
15b71abe JA |
3818 | return 0; |
3819 | } | |
3820 | ||
45d189c6 | 3821 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3822 | { |
45d189c6 | 3823 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
3824 | } |
3825 | ||
067524e9 JA |
3826 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3827 | const struct io_uring_sqe *sqe) | |
3828 | { | |
3829 | struct io_provide_buf *p = &req->pbuf; | |
3830 | u64 tmp; | |
3831 | ||
3832 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3833 | return -EINVAL; | |
3834 | ||
3835 | tmp = READ_ONCE(sqe->fd); | |
3836 | if (!tmp || tmp > USHRT_MAX) | |
3837 | return -EINVAL; | |
3838 | ||
3839 | memset(p, 0, sizeof(*p)); | |
3840 | p->nbufs = tmp; | |
3841 | p->bgid = READ_ONCE(sqe->buf_group); | |
3842 | return 0; | |
3843 | } | |
3844 | ||
3845 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3846 | int bgid, unsigned nbufs) | |
3847 | { | |
3848 | unsigned i = 0; | |
3849 | ||
3850 | /* shouldn't happen */ | |
3851 | if (!nbufs) | |
3852 | return 0; | |
3853 | ||
3854 | /* the head kbuf is the list itself */ | |
3855 | while (!list_empty(&buf->list)) { | |
3856 | struct io_buffer *nxt; | |
3857 | ||
3858 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3859 | list_del(&nxt->list); | |
3860 | kfree(nxt); | |
3861 | if (++i == nbufs) | |
3862 | return i; | |
3863 | } | |
3864 | i++; | |
3865 | kfree(buf); | |
3866 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3867 | ||
3868 | return i; | |
3869 | } | |
3870 | ||
889fca73 | 3871 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3872 | { |
3873 | struct io_provide_buf *p = &req->pbuf; | |
3874 | struct io_ring_ctx *ctx = req->ctx; | |
3875 | struct io_buffer *head; | |
3876 | int ret = 0; | |
45d189c6 | 3877 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3878 | |
3879 | io_ring_submit_lock(ctx, !force_nonblock); | |
3880 | ||
3881 | lockdep_assert_held(&ctx->uring_lock); | |
3882 | ||
3883 | ret = -ENOENT; | |
3884 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3885 | if (head) | |
3886 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3887 | if (ret < 0) |
3888 | req_set_fail_links(req); | |
067524e9 | 3889 | |
31bff9a5 PB |
3890 | /* need to hold the lock to complete IOPOLL requests */ |
3891 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3892 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3893 | io_ring_submit_unlock(ctx, !force_nonblock); |
3894 | } else { | |
3895 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3896 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3897 | } |
067524e9 JA |
3898 | return 0; |
3899 | } | |
3900 | ||
ddf0322d JA |
3901 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3902 | const struct io_uring_sqe *sqe) | |
3903 | { | |
3904 | struct io_provide_buf *p = &req->pbuf; | |
3905 | u64 tmp; | |
3906 | ||
3907 | if (sqe->ioprio || sqe->rw_flags) | |
3908 | return -EINVAL; | |
3909 | ||
3910 | tmp = READ_ONCE(sqe->fd); | |
3911 | if (!tmp || tmp > USHRT_MAX) | |
3912 | return -E2BIG; | |
3913 | p->nbufs = tmp; | |
3914 | p->addr = READ_ONCE(sqe->addr); | |
3915 | p->len = READ_ONCE(sqe->len); | |
3916 | ||
efe68c1c | 3917 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3918 | return -EFAULT; |
3919 | ||
3920 | p->bgid = READ_ONCE(sqe->buf_group); | |
3921 | tmp = READ_ONCE(sqe->off); | |
3922 | if (tmp > USHRT_MAX) | |
3923 | return -E2BIG; | |
3924 | p->bid = tmp; | |
3925 | return 0; | |
3926 | } | |
3927 | ||
3928 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3929 | { | |
3930 | struct io_buffer *buf; | |
3931 | u64 addr = pbuf->addr; | |
3932 | int i, bid = pbuf->bid; | |
3933 | ||
3934 | for (i = 0; i < pbuf->nbufs; i++) { | |
3935 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3936 | if (!buf) | |
3937 | break; | |
3938 | ||
3939 | buf->addr = addr; | |
3940 | buf->len = pbuf->len; | |
3941 | buf->bid = bid; | |
3942 | addr += pbuf->len; | |
3943 | bid++; | |
3944 | if (!*head) { | |
3945 | INIT_LIST_HEAD(&buf->list); | |
3946 | *head = buf; | |
3947 | } else { | |
3948 | list_add_tail(&buf->list, &(*head)->list); | |
3949 | } | |
3950 | } | |
3951 | ||
3952 | return i ? i : -ENOMEM; | |
3953 | } | |
3954 | ||
889fca73 | 3955 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3956 | { |
3957 | struct io_provide_buf *p = &req->pbuf; | |
3958 | struct io_ring_ctx *ctx = req->ctx; | |
3959 | struct io_buffer *head, *list; | |
3960 | int ret = 0; | |
45d189c6 | 3961 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3962 | |
3963 | io_ring_submit_lock(ctx, !force_nonblock); | |
3964 | ||
3965 | lockdep_assert_held(&ctx->uring_lock); | |
3966 | ||
3967 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3968 | ||
3969 | ret = io_add_buffers(p, &head); | |
3970 | if (ret < 0) | |
3971 | goto out; | |
3972 | ||
3973 | if (!list) { | |
3974 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3975 | GFP_KERNEL); | |
3976 | if (ret < 0) { | |
067524e9 | 3977 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3978 | goto out; |
3979 | } | |
3980 | } | |
3981 | out: | |
ddf0322d JA |
3982 | if (ret < 0) |
3983 | req_set_fail_links(req); | |
31bff9a5 PB |
3984 | |
3985 | /* need to hold the lock to complete IOPOLL requests */ | |
3986 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3987 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3988 | io_ring_submit_unlock(ctx, !force_nonblock); |
3989 | } else { | |
3990 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3991 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3992 | } |
ddf0322d | 3993 | return 0; |
cebdb986 JA |
3994 | } |
3995 | ||
3e4827b0 JA |
3996 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
3997 | const struct io_uring_sqe *sqe) | |
3998 | { | |
3999 | #if defined(CONFIG_EPOLL) | |
4000 | if (sqe->ioprio || sqe->buf_index) | |
4001 | return -EINVAL; | |
6ca56f84 | 4002 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4003 | return -EINVAL; |
3e4827b0 JA |
4004 | |
4005 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4006 | req->epoll.op = READ_ONCE(sqe->len); | |
4007 | req->epoll.fd = READ_ONCE(sqe->off); | |
4008 | ||
4009 | if (ep_op_has_event(req->epoll.op)) { | |
4010 | struct epoll_event __user *ev; | |
4011 | ||
4012 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4013 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4014 | return -EFAULT; | |
4015 | } | |
4016 | ||
4017 | return 0; | |
4018 | #else | |
4019 | return -EOPNOTSUPP; | |
4020 | #endif | |
4021 | } | |
4022 | ||
889fca73 | 4023 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4024 | { |
4025 | #if defined(CONFIG_EPOLL) | |
4026 | struct io_epoll *ie = &req->epoll; | |
4027 | int ret; | |
45d189c6 | 4028 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4029 | |
4030 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4031 | if (force_nonblock && ret == -EAGAIN) | |
4032 | return -EAGAIN; | |
4033 | ||
4034 | if (ret < 0) | |
4035 | req_set_fail_links(req); | |
889fca73 | 4036 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4037 | return 0; |
4038 | #else | |
4039 | return -EOPNOTSUPP; | |
4040 | #endif | |
4041 | } | |
4042 | ||
c1ca757b JA |
4043 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4044 | { | |
4045 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4046 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4047 | return -EINVAL; | |
3232dd02 PB |
4048 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4049 | return -EINVAL; | |
c1ca757b JA |
4050 | |
4051 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4052 | req->madvise.len = READ_ONCE(sqe->len); | |
4053 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4054 | return 0; | |
4055 | #else | |
4056 | return -EOPNOTSUPP; | |
4057 | #endif | |
4058 | } | |
4059 | ||
45d189c6 | 4060 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4061 | { |
4062 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4063 | struct io_madvise *ma = &req->madvise; | |
4064 | int ret; | |
4065 | ||
45d189c6 | 4066 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4067 | return -EAGAIN; |
4068 | ||
0726b01e | 4069 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4070 | if (ret < 0) |
4071 | req_set_fail_links(req); | |
e1e16097 | 4072 | io_req_complete(req, ret); |
c1ca757b JA |
4073 | return 0; |
4074 | #else | |
4075 | return -EOPNOTSUPP; | |
4076 | #endif | |
4077 | } | |
4078 | ||
4840e418 JA |
4079 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4080 | { | |
4081 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4082 | return -EINVAL; | |
3232dd02 PB |
4083 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4084 | return -EINVAL; | |
4840e418 JA |
4085 | |
4086 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4087 | req->fadvise.len = READ_ONCE(sqe->len); | |
4088 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4089 | return 0; | |
4090 | } | |
4091 | ||
45d189c6 | 4092 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4093 | { |
4094 | struct io_fadvise *fa = &req->fadvise; | |
4095 | int ret; | |
4096 | ||
45d189c6 | 4097 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4098 | switch (fa->advice) { |
4099 | case POSIX_FADV_NORMAL: | |
4100 | case POSIX_FADV_RANDOM: | |
4101 | case POSIX_FADV_SEQUENTIAL: | |
4102 | break; | |
4103 | default: | |
4104 | return -EAGAIN; | |
4105 | } | |
4106 | } | |
4840e418 JA |
4107 | |
4108 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4109 | if (ret < 0) | |
4110 | req_set_fail_links(req); | |
e1e16097 | 4111 | io_req_complete(req, ret); |
4840e418 JA |
4112 | return 0; |
4113 | } | |
4114 | ||
eddc7ef5 JA |
4115 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4116 | { | |
6ca56f84 | 4117 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4118 | return -EINVAL; |
eddc7ef5 JA |
4119 | if (sqe->ioprio || sqe->buf_index) |
4120 | return -EINVAL; | |
9c280f90 | 4121 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4122 | return -EBADF; |
eddc7ef5 | 4123 | |
1d9e1288 BM |
4124 | req->statx.dfd = READ_ONCE(sqe->fd); |
4125 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4126 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4127 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4128 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4129 | |
4130 | return 0; | |
4131 | } | |
4132 | ||
45d189c6 | 4133 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4134 | { |
1d9e1288 | 4135 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4136 | int ret; |
4137 | ||
45d189c6 | 4138 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4139 | /* only need file table for an actual valid fd */ |
4140 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4141 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4142 | return -EAGAIN; |
5b0bbee4 | 4143 | } |
eddc7ef5 | 4144 | |
e62753e4 BM |
4145 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4146 | ctx->buffer); | |
eddc7ef5 | 4147 | |
eddc7ef5 JA |
4148 | if (ret < 0) |
4149 | req_set_fail_links(req); | |
e1e16097 | 4150 | io_req_complete(req, ret); |
eddc7ef5 JA |
4151 | return 0; |
4152 | } | |
4153 | ||
b5dba59e JA |
4154 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4155 | { | |
14587a46 | 4156 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4157 | return -EINVAL; |
b5dba59e JA |
4158 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4159 | sqe->rw_flags || sqe->buf_index) | |
4160 | return -EINVAL; | |
9c280f90 | 4161 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4162 | return -EBADF; |
b5dba59e JA |
4163 | |
4164 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4165 | return 0; |
b5dba59e JA |
4166 | } |
4167 | ||
889fca73 | 4168 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4169 | { |
9eac1904 | 4170 | struct files_struct *files = current->files; |
3af73b28 | 4171 | struct io_close *close = &req->close; |
9eac1904 JA |
4172 | struct fdtable *fdt; |
4173 | struct file *file; | |
b5dba59e JA |
4174 | int ret; |
4175 | ||
9eac1904 JA |
4176 | file = NULL; |
4177 | ret = -EBADF; | |
4178 | spin_lock(&files->file_lock); | |
4179 | fdt = files_fdtable(files); | |
4180 | if (close->fd >= fdt->max_fds) { | |
4181 | spin_unlock(&files->file_lock); | |
4182 | goto err; | |
4183 | } | |
4184 | file = fdt->fd[close->fd]; | |
4185 | if (!file) { | |
4186 | spin_unlock(&files->file_lock); | |
4187 | goto err; | |
4188 | } | |
4189 | ||
4190 | if (file->f_op == &io_uring_fops) { | |
4191 | spin_unlock(&files->file_lock); | |
4192 | file = NULL; | |
4193 | goto err; | |
3af73b28 | 4194 | } |
b5dba59e JA |
4195 | |
4196 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4197 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4198 | spin_unlock(&files->file_lock); |
0bf0eefd | 4199 | return -EAGAIN; |
a2100672 | 4200 | } |
b5dba59e | 4201 | |
9eac1904 JA |
4202 | ret = __close_fd_get_file(close->fd, &file); |
4203 | spin_unlock(&files->file_lock); | |
4204 | if (ret < 0) { | |
4205 | if (ret == -ENOENT) | |
4206 | ret = -EBADF; | |
4207 | goto err; | |
4208 | } | |
4209 | ||
3af73b28 | 4210 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4211 | ret = filp_close(file, current->files); |
4212 | err: | |
3af73b28 PB |
4213 | if (ret < 0) |
4214 | req_set_fail_links(req); | |
9eac1904 JA |
4215 | if (file) |
4216 | fput(file); | |
889fca73 | 4217 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4218 | return 0; |
b5dba59e JA |
4219 | } |
4220 | ||
1155c76a | 4221 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4222 | { |
4223 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4224 | |
5d17b4a4 JA |
4225 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4226 | return -EINVAL; | |
4227 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4228 | return -EINVAL; | |
4229 | ||
8ed8d3c3 JA |
4230 | req->sync.off = READ_ONCE(sqe->off); |
4231 | req->sync.len = READ_ONCE(sqe->len); | |
4232 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4233 | return 0; |
4234 | } | |
4235 | ||
45d189c6 | 4236 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4237 | { |
8ed8d3c3 JA |
4238 | int ret; |
4239 | ||
ac45abc0 | 4240 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4241 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4242 | return -EAGAIN; |
4243 | ||
9adbd45d | 4244 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4245 | req->sync.flags); |
4246 | if (ret < 0) | |
4247 | req_set_fail_links(req); | |
e1e16097 | 4248 | io_req_complete(req, ret); |
5d17b4a4 JA |
4249 | return 0; |
4250 | } | |
4251 | ||
469956e8 | 4252 | #if defined(CONFIG_NET) |
02d27d89 PB |
4253 | static int io_setup_async_msg(struct io_kiocb *req, |
4254 | struct io_async_msghdr *kmsg) | |
4255 | { | |
e8c2bc1f JA |
4256 | struct io_async_msghdr *async_msg = req->async_data; |
4257 | ||
4258 | if (async_msg) | |
02d27d89 | 4259 | return -EAGAIN; |
e8c2bc1f | 4260 | if (io_alloc_async_data(req)) { |
257e84a5 | 4261 | kfree(kmsg->free_iov); |
02d27d89 PB |
4262 | return -ENOMEM; |
4263 | } | |
e8c2bc1f | 4264 | async_msg = req->async_data; |
02d27d89 | 4265 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4266 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4267 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4268 | /* if were using fast_iov, set it to the new one */ |
4269 | if (!async_msg->free_iov) | |
4270 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4271 | ||
02d27d89 PB |
4272 | return -EAGAIN; |
4273 | } | |
4274 | ||
2ae523ed PB |
4275 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4276 | struct io_async_msghdr *iomsg) | |
4277 | { | |
2ae523ed | 4278 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4279 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4280 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4281 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4282 | } |
4283 | ||
93642ef8 PB |
4284 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4285 | { | |
4286 | int ret; | |
4287 | ||
4288 | if (!io_op_defs[req->opcode].needs_async_data) | |
4289 | return 0; | |
4290 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4291 | if (!ret) | |
4292 | req->flags |= REQ_F_NEED_CLEANUP; | |
4293 | return ret; | |
4294 | } | |
4295 | ||
3529d8c2 | 4296 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4297 | { |
e47293fd | 4298 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4299 | |
d2b6f48b PB |
4300 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4301 | return -EINVAL; | |
4302 | ||
e47293fd | 4303 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4304 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4305 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4306 | |
d8768362 JA |
4307 | #ifdef CONFIG_COMPAT |
4308 | if (req->ctx->compat) | |
4309 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4310 | #endif | |
93642ef8 | 4311 | return 0; |
03b1230c JA |
4312 | } |
4313 | ||
889fca73 | 4314 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4315 | { |
6b754c8b | 4316 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4317 | struct socket *sock; |
7a7cacba | 4318 | unsigned flags; |
0fa03c62 JA |
4319 | int ret; |
4320 | ||
dba4a925 | 4321 | sock = sock_from_file(req->file); |
7a7cacba | 4322 | if (unlikely(!sock)) |
dba4a925 | 4323 | return -ENOTSOCK; |
3529d8c2 | 4324 | |
257e84a5 PB |
4325 | kmsg = req->async_data; |
4326 | if (!kmsg) { | |
7a7cacba PB |
4327 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4328 | if (ret) | |
4329 | return ret; | |
4330 | kmsg = &iomsg; | |
0fa03c62 | 4331 | } |
0fa03c62 | 4332 | |
7a7cacba PB |
4333 | flags = req->sr_msg.msg_flags; |
4334 | if (flags & MSG_DONTWAIT) | |
4335 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4336 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4337 | flags |= MSG_DONTWAIT; |
e47293fd | 4338 | |
7a7cacba | 4339 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4340 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4341 | return io_setup_async_msg(req, kmsg); |
4342 | if (ret == -ERESTARTSYS) | |
4343 | ret = -EINTR; | |
0fa03c62 | 4344 | |
257e84a5 PB |
4345 | /* fast path, check for non-NULL to avoid function call */ |
4346 | if (kmsg->free_iov) | |
4347 | kfree(kmsg->free_iov); | |
99bc4c38 | 4348 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4349 | if (ret < 0) |
4350 | req_set_fail_links(req); | |
889fca73 | 4351 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4352 | return 0; |
03b1230c | 4353 | } |
aa1fa28f | 4354 | |
889fca73 | 4355 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4356 | { |
7a7cacba PB |
4357 | struct io_sr_msg *sr = &req->sr_msg; |
4358 | struct msghdr msg; | |
4359 | struct iovec iov; | |
fddaface | 4360 | struct socket *sock; |
7a7cacba | 4361 | unsigned flags; |
fddaface JA |
4362 | int ret; |
4363 | ||
dba4a925 | 4364 | sock = sock_from_file(req->file); |
7a7cacba | 4365 | if (unlikely(!sock)) |
dba4a925 | 4366 | return -ENOTSOCK; |
fddaface | 4367 | |
7a7cacba PB |
4368 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4369 | if (unlikely(ret)) | |
14db8411 | 4370 | return ret; |
fddaface | 4371 | |
7a7cacba PB |
4372 | msg.msg_name = NULL; |
4373 | msg.msg_control = NULL; | |
4374 | msg.msg_controllen = 0; | |
4375 | msg.msg_namelen = 0; | |
fddaface | 4376 | |
7a7cacba PB |
4377 | flags = req->sr_msg.msg_flags; |
4378 | if (flags & MSG_DONTWAIT) | |
4379 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4380 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4381 | flags |= MSG_DONTWAIT; |
fddaface | 4382 | |
7a7cacba PB |
4383 | msg.msg_flags = flags; |
4384 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4385 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4386 | return -EAGAIN; |
4387 | if (ret == -ERESTARTSYS) | |
4388 | ret = -EINTR; | |
fddaface | 4389 | |
fddaface JA |
4390 | if (ret < 0) |
4391 | req_set_fail_links(req); | |
889fca73 | 4392 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4393 | return 0; |
fddaface JA |
4394 | } |
4395 | ||
1400e697 PB |
4396 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4397 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4398 | { |
4399 | struct io_sr_msg *sr = &req->sr_msg; | |
4400 | struct iovec __user *uiov; | |
4401 | size_t iov_len; | |
4402 | int ret; | |
4403 | ||
1400e697 PB |
4404 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4405 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4406 | if (ret) |
4407 | return ret; | |
4408 | ||
4409 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4410 | if (iov_len > 1) | |
4411 | return -EINVAL; | |
5476dfed | 4412 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4413 | return -EFAULT; |
5476dfed | 4414 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4415 | iomsg->free_iov = NULL; |
52de1fe1 | 4416 | } else { |
257e84a5 | 4417 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4418 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4419 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4420 | false); |
52de1fe1 JA |
4421 | if (ret > 0) |
4422 | ret = 0; | |
4423 | } | |
4424 | ||
4425 | return ret; | |
4426 | } | |
4427 | ||
4428 | #ifdef CONFIG_COMPAT | |
4429 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4430 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4431 | { |
4432 | struct compat_msghdr __user *msg_compat; | |
4433 | struct io_sr_msg *sr = &req->sr_msg; | |
4434 | struct compat_iovec __user *uiov; | |
4435 | compat_uptr_t ptr; | |
4436 | compat_size_t len; | |
4437 | int ret; | |
4438 | ||
270a5940 | 4439 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4440 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4441 | &ptr, &len); |
4442 | if (ret) | |
4443 | return ret; | |
4444 | ||
4445 | uiov = compat_ptr(ptr); | |
4446 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4447 | compat_ssize_t clen; | |
4448 | ||
4449 | if (len > 1) | |
4450 | return -EINVAL; | |
4451 | if (!access_ok(uiov, sizeof(*uiov))) | |
4452 | return -EFAULT; | |
4453 | if (__get_user(clen, &uiov->iov_len)) | |
4454 | return -EFAULT; | |
4455 | if (clen < 0) | |
4456 | return -EINVAL; | |
2d280bc8 | 4457 | sr->len = clen; |
257e84a5 | 4458 | iomsg->free_iov = NULL; |
52de1fe1 | 4459 | } else { |
257e84a5 | 4460 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4461 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4462 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4463 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4464 | if (ret < 0) |
4465 | return ret; | |
4466 | } | |
4467 | ||
4468 | return 0; | |
4469 | } | |
4470 | #endif | |
4471 | ||
1400e697 PB |
4472 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4473 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4474 | { |
1400e697 | 4475 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4476 | |
4477 | #ifdef CONFIG_COMPAT | |
4478 | if (req->ctx->compat) | |
1400e697 | 4479 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4480 | #endif |
52de1fe1 | 4481 | |
1400e697 | 4482 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4483 | } |
4484 | ||
bcda7baa | 4485 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4486 | bool needs_lock) |
bcda7baa JA |
4487 | { |
4488 | struct io_sr_msg *sr = &req->sr_msg; | |
4489 | struct io_buffer *kbuf; | |
4490 | ||
bcda7baa JA |
4491 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4492 | if (IS_ERR(kbuf)) | |
4493 | return kbuf; | |
4494 | ||
4495 | sr->kbuf = kbuf; | |
4496 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4497 | return kbuf; |
fddaface JA |
4498 | } |
4499 | ||
7fbb1b54 PB |
4500 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4501 | { | |
4502 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4503 | } | |
4504 | ||
93642ef8 | 4505 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4506 | { |
99bc4c38 | 4507 | int ret; |
3529d8c2 | 4508 | |
93642ef8 PB |
4509 | if (!io_op_defs[req->opcode].needs_async_data) |
4510 | return 0; | |
4511 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4512 | if (!ret) | |
4513 | req->flags |= REQ_F_NEED_CLEANUP; | |
4514 | return ret; | |
4515 | } | |
4516 | ||
4517 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4518 | { | |
4519 | struct io_sr_msg *sr = &req->sr_msg; | |
4520 | ||
d2b6f48b PB |
4521 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4522 | return -EINVAL; | |
4523 | ||
3529d8c2 | 4524 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4525 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4526 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4527 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4528 | |
d8768362 JA |
4529 | #ifdef CONFIG_COMPAT |
4530 | if (req->ctx->compat) | |
4531 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4532 | #endif | |
93642ef8 | 4533 | return 0; |
aa1fa28f JA |
4534 | } |
4535 | ||
889fca73 | 4536 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4537 | { |
6b754c8b | 4538 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4539 | struct socket *sock; |
7fbb1b54 | 4540 | struct io_buffer *kbuf; |
7a7cacba | 4541 | unsigned flags; |
52de1fe1 | 4542 | int ret, cflags = 0; |
45d189c6 | 4543 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4544 | |
dba4a925 | 4545 | sock = sock_from_file(req->file); |
7a7cacba | 4546 | if (unlikely(!sock)) |
dba4a925 | 4547 | return -ENOTSOCK; |
3529d8c2 | 4548 | |
257e84a5 PB |
4549 | kmsg = req->async_data; |
4550 | if (!kmsg) { | |
7a7cacba PB |
4551 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4552 | if (ret) | |
681fda8d | 4553 | return ret; |
7a7cacba PB |
4554 | kmsg = &iomsg; |
4555 | } | |
03b1230c | 4556 | |
bc02ef33 | 4557 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4558 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4559 | if (IS_ERR(kbuf)) |
52de1fe1 | 4560 | return PTR_ERR(kbuf); |
7a7cacba | 4561 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4562 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4563 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4564 | 1, req->sr_msg.len); |
4565 | } | |
52de1fe1 | 4566 | |
7a7cacba PB |
4567 | flags = req->sr_msg.msg_flags; |
4568 | if (flags & MSG_DONTWAIT) | |
4569 | req->flags |= REQ_F_NOWAIT; | |
4570 | else if (force_nonblock) | |
4571 | flags |= MSG_DONTWAIT; | |
e47293fd | 4572 | |
7a7cacba PB |
4573 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4574 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4575 | if (force_nonblock && ret == -EAGAIN) |
4576 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4577 | if (ret == -ERESTARTSYS) |
4578 | ret = -EINTR; | |
03b1230c | 4579 | |
7fbb1b54 PB |
4580 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4581 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4582 | /* fast path, check for non-NULL to avoid function call */ |
4583 | if (kmsg->free_iov) | |
4584 | kfree(kmsg->free_iov); | |
99bc4c38 | 4585 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4586 | if (ret < 0) |
4587 | req_set_fail_links(req); | |
889fca73 | 4588 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4589 | return 0; |
0fa03c62 | 4590 | } |
5d17b4a4 | 4591 | |
889fca73 | 4592 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4593 | { |
6b754c8b | 4594 | struct io_buffer *kbuf; |
7a7cacba PB |
4595 | struct io_sr_msg *sr = &req->sr_msg; |
4596 | struct msghdr msg; | |
4597 | void __user *buf = sr->buf; | |
fddaface | 4598 | struct socket *sock; |
7a7cacba PB |
4599 | struct iovec iov; |
4600 | unsigned flags; | |
bcda7baa | 4601 | int ret, cflags = 0; |
45d189c6 | 4602 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4603 | |
dba4a925 | 4604 | sock = sock_from_file(req->file); |
7a7cacba | 4605 | if (unlikely(!sock)) |
dba4a925 | 4606 | return -ENOTSOCK; |
fddaface | 4607 | |
bc02ef33 | 4608 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4609 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4610 | if (IS_ERR(kbuf)) |
4611 | return PTR_ERR(kbuf); | |
7a7cacba | 4612 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4613 | } |
bcda7baa | 4614 | |
7a7cacba | 4615 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4616 | if (unlikely(ret)) |
4617 | goto out_free; | |
fddaface | 4618 | |
7a7cacba PB |
4619 | msg.msg_name = NULL; |
4620 | msg.msg_control = NULL; | |
4621 | msg.msg_controllen = 0; | |
4622 | msg.msg_namelen = 0; | |
4623 | msg.msg_iocb = NULL; | |
4624 | msg.msg_flags = 0; | |
fddaface | 4625 | |
7a7cacba PB |
4626 | flags = req->sr_msg.msg_flags; |
4627 | if (flags & MSG_DONTWAIT) | |
4628 | req->flags |= REQ_F_NOWAIT; | |
4629 | else if (force_nonblock) | |
4630 | flags |= MSG_DONTWAIT; | |
4631 | ||
4632 | ret = sock_recvmsg(sock, &msg, flags); | |
4633 | if (force_nonblock && ret == -EAGAIN) | |
4634 | return -EAGAIN; | |
4635 | if (ret == -ERESTARTSYS) | |
4636 | ret = -EINTR; | |
14c32eee | 4637 | out_free: |
7fbb1b54 PB |
4638 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4639 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4640 | if (ret < 0) |
4641 | req_set_fail_links(req); | |
889fca73 | 4642 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4643 | return 0; |
fddaface JA |
4644 | } |
4645 | ||
3529d8c2 | 4646 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4647 | { |
8ed8d3c3 JA |
4648 | struct io_accept *accept = &req->accept; |
4649 | ||
14587a46 | 4650 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4651 | return -EINVAL; |
8042d6ce | 4652 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4653 | return -EINVAL; |
4654 | ||
d55e5f5b JA |
4655 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4656 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4657 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4658 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4659 | return 0; |
8ed8d3c3 | 4660 | } |
17f2fe35 | 4661 | |
889fca73 | 4662 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4663 | { |
4664 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4665 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4666 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4667 | int ret; |
4668 | ||
e697deed JX |
4669 | if (req->file->f_flags & O_NONBLOCK) |
4670 | req->flags |= REQ_F_NOWAIT; | |
4671 | ||
8ed8d3c3 | 4672 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4673 | accept->addr_len, accept->flags, |
4674 | accept->nofile); | |
8ed8d3c3 | 4675 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4676 | return -EAGAIN; |
ac45abc0 PB |
4677 | if (ret < 0) { |
4678 | if (ret == -ERESTARTSYS) | |
4679 | ret = -EINTR; | |
4e88d6e7 | 4680 | req_set_fail_links(req); |
ac45abc0 | 4681 | } |
889fca73 | 4682 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4683 | return 0; |
8ed8d3c3 JA |
4684 | } |
4685 | ||
93642ef8 PB |
4686 | static int io_connect_prep_async(struct io_kiocb *req) |
4687 | { | |
4688 | struct io_async_connect *io = req->async_data; | |
4689 | struct io_connect *conn = &req->connect; | |
4690 | ||
4691 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4692 | } | |
4693 | ||
3529d8c2 | 4694 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4695 | { |
3529d8c2 | 4696 | struct io_connect *conn = &req->connect; |
f499a021 | 4697 | |
14587a46 | 4698 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4699 | return -EINVAL; |
4700 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4701 | return -EINVAL; | |
4702 | ||
3529d8c2 JA |
4703 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4704 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4705 | return 0; |
f499a021 JA |
4706 | } |
4707 | ||
889fca73 | 4708 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4709 | { |
e8c2bc1f | 4710 | struct io_async_connect __io, *io; |
f8e85cf2 | 4711 | unsigned file_flags; |
3fbb51c1 | 4712 | int ret; |
45d189c6 | 4713 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4714 | |
e8c2bc1f JA |
4715 | if (req->async_data) { |
4716 | io = req->async_data; | |
f499a021 | 4717 | } else { |
3529d8c2 JA |
4718 | ret = move_addr_to_kernel(req->connect.addr, |
4719 | req->connect.addr_len, | |
e8c2bc1f | 4720 | &__io.address); |
f499a021 JA |
4721 | if (ret) |
4722 | goto out; | |
4723 | io = &__io; | |
4724 | } | |
4725 | ||
3fbb51c1 JA |
4726 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4727 | ||
e8c2bc1f | 4728 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4729 | req->connect.addr_len, file_flags); |
87f80d62 | 4730 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4731 | if (req->async_data) |
b7bb4f7d | 4732 | return -EAGAIN; |
e8c2bc1f | 4733 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4734 | ret = -ENOMEM; |
4735 | goto out; | |
4736 | } | |
e8c2bc1f JA |
4737 | io = req->async_data; |
4738 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4739 | return -EAGAIN; |
f499a021 | 4740 | } |
f8e85cf2 JA |
4741 | if (ret == -ERESTARTSYS) |
4742 | ret = -EINTR; | |
f499a021 | 4743 | out: |
4e88d6e7 JA |
4744 | if (ret < 0) |
4745 | req_set_fail_links(req); | |
889fca73 | 4746 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4747 | return 0; |
469956e8 Y |
4748 | } |
4749 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4750 | #define IO_NETOP_FN(op) \ |
4751 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4752 | { \ | |
4753 | return -EOPNOTSUPP; \ | |
4754 | } | |
4755 | ||
4756 | #define IO_NETOP_PREP(op) \ | |
4757 | IO_NETOP_FN(op) \ | |
4758 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4759 | { \ | |
4760 | return -EOPNOTSUPP; \ | |
4761 | } \ | |
4762 | ||
4763 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4764 | IO_NETOP_PREP(op) \ | |
4765 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4766 | { \ | |
4767 | return -EOPNOTSUPP; \ | |
4768 | } | |
4769 | ||
4770 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4771 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4772 | IO_NETOP_PREP_ASYNC(connect); | |
4773 | IO_NETOP_PREP(accept); | |
4774 | IO_NETOP_FN(send); | |
4775 | IO_NETOP_FN(recv); | |
469956e8 | 4776 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4777 | |
d7718a9d JA |
4778 | struct io_poll_table { |
4779 | struct poll_table_struct pt; | |
4780 | struct io_kiocb *req; | |
4781 | int error; | |
4782 | }; | |
ce593a6c | 4783 | |
d7718a9d JA |
4784 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4785 | __poll_t mask, task_work_func_t func) | |
4786 | { | |
aa96bf8a | 4787 | int ret; |
d7718a9d JA |
4788 | |
4789 | /* for instances that support it check for an event match first: */ | |
4790 | if (mask && !(mask & poll->events)) | |
4791 | return 0; | |
4792 | ||
4793 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4794 | ||
4795 | list_del_init(&poll->wait.entry); | |
4796 | ||
d7718a9d | 4797 | req->result = mask; |
7cbf1722 | 4798 | req->task_work.func = func; |
6d816e08 JA |
4799 | percpu_ref_get(&req->ctx->refs); |
4800 | ||
d7718a9d | 4801 | /* |
e3aabf95 JA |
4802 | * If this fails, then the task is exiting. When a task exits, the |
4803 | * work gets canceled, so just cancel this request as well instead | |
4804 | * of executing it. We can't safely execute it anyway, as we may not | |
4805 | * have the needed state needed for it anyway. | |
d7718a9d | 4806 | */ |
355fb9e2 | 4807 | ret = io_req_task_work_add(req); |
aa96bf8a | 4808 | if (unlikely(ret)) { |
e3aabf95 | 4809 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4810 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4811 | } |
d7718a9d JA |
4812 | return 1; |
4813 | } | |
4814 | ||
74ce6ce4 JA |
4815 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4816 | __acquires(&req->ctx->completion_lock) | |
4817 | { | |
4818 | struct io_ring_ctx *ctx = req->ctx; | |
4819 | ||
4820 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4821 | struct poll_table_struct pt = { ._key = poll->events }; | |
4822 | ||
4823 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4824 | } | |
4825 | ||
4826 | spin_lock_irq(&ctx->completion_lock); | |
4827 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4828 | add_wait_queue(poll->head, &poll->wait); | |
4829 | return true; | |
4830 | } | |
4831 | ||
4832 | return false; | |
4833 | } | |
4834 | ||
d4e7cd36 | 4835 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4836 | { |
e8c2bc1f | 4837 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4838 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4839 | return req->async_data; |
d4e7cd36 JA |
4840 | return req->apoll->double_poll; |
4841 | } | |
4842 | ||
4843 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4844 | { | |
4845 | if (req->opcode == IORING_OP_POLL_ADD) | |
4846 | return &req->poll; | |
4847 | return &req->apoll->poll; | |
4848 | } | |
4849 | ||
4850 | static void io_poll_remove_double(struct io_kiocb *req) | |
4851 | { | |
4852 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4853 | |
4854 | lockdep_assert_held(&req->ctx->completion_lock); | |
4855 | ||
4856 | if (poll && poll->head) { | |
4857 | struct wait_queue_head *head = poll->head; | |
4858 | ||
4859 | spin_lock(&head->lock); | |
4860 | list_del_init(&poll->wait.entry); | |
4861 | if (poll->wait.private) | |
4862 | refcount_dec(&req->refs); | |
4863 | poll->head = NULL; | |
4864 | spin_unlock(&head->lock); | |
4865 | } | |
4866 | } | |
4867 | ||
4868 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4869 | { | |
4870 | struct io_ring_ctx *ctx = req->ctx; | |
4871 | ||
d4e7cd36 | 4872 | io_poll_remove_double(req); |
18bceab1 JA |
4873 | req->poll.done = true; |
4874 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4875 | io_commit_cqring(ctx); | |
4876 | } | |
4877 | ||
dd221f46 | 4878 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4879 | { |
dd221f46 | 4880 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4881 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4882 | struct io_kiocb *nxt; |
18bceab1 JA |
4883 | |
4884 | if (io_poll_rewait(req, &req->poll)) { | |
4885 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4886 | } else { |
4887 | hash_del(&req->hash_node); | |
4888 | io_poll_complete(req, req->result, 0); | |
4889 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4890 | |
dd221f46 PB |
4891 | nxt = io_put_req_find_next(req); |
4892 | io_cqring_ev_posted(ctx); | |
4893 | if (nxt) | |
4894 | __io_req_task_submit(nxt); | |
4895 | } | |
18bceab1 | 4896 | |
6d816e08 | 4897 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4898 | } |
4899 | ||
4900 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4901 | int sync, void *key) | |
4902 | { | |
4903 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4904 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4905 | __poll_t mask = key_to_poll(key); |
4906 | ||
4907 | /* for instances that support it check for an event match first: */ | |
4908 | if (mask && !(mask & poll->events)) | |
4909 | return 0; | |
4910 | ||
8706e04e JA |
4911 | list_del_init(&wait->entry); |
4912 | ||
807abcb0 | 4913 | if (poll && poll->head) { |
18bceab1 JA |
4914 | bool done; |
4915 | ||
807abcb0 JA |
4916 | spin_lock(&poll->head->lock); |
4917 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4918 | if (!done) |
807abcb0 | 4919 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4920 | /* make sure double remove sees this as being gone */ |
4921 | wait->private = NULL; | |
807abcb0 | 4922 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4923 | if (!done) { |
4924 | /* use wait func handler, so it matches the rq type */ | |
4925 | poll->wait.func(&poll->wait, mode, sync, key); | |
4926 | } | |
18bceab1 JA |
4927 | } |
4928 | refcount_dec(&req->refs); | |
4929 | return 1; | |
4930 | } | |
4931 | ||
4932 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4933 | wait_queue_func_t wake_func) | |
4934 | { | |
4935 | poll->head = NULL; | |
4936 | poll->done = false; | |
4937 | poll->canceled = false; | |
4938 | poll->events = events; | |
4939 | INIT_LIST_HEAD(&poll->wait.entry); | |
4940 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4941 | } | |
4942 | ||
4943 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4944 | struct wait_queue_head *head, |
4945 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4946 | { |
4947 | struct io_kiocb *req = pt->req; | |
4948 | ||
4949 | /* | |
4950 | * If poll->head is already set, it's because the file being polled | |
4951 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4952 | * for write). Setup a separate io_poll_iocb if this happens. | |
4953 | */ | |
4954 | if (unlikely(poll->head)) { | |
58852d4d PB |
4955 | struct io_poll_iocb *poll_one = poll; |
4956 | ||
18bceab1 | 4957 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4958 | if (*poll_ptr) { |
18bceab1 JA |
4959 | pt->error = -EINVAL; |
4960 | return; | |
4961 | } | |
4962 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
4963 | if (!poll) { | |
4964 | pt->error = -ENOMEM; | |
4965 | return; | |
4966 | } | |
58852d4d | 4967 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4968 | refcount_inc(&req->refs); |
4969 | poll->wait.private = req; | |
807abcb0 | 4970 | *poll_ptr = poll; |
18bceab1 JA |
4971 | } |
4972 | ||
4973 | pt->error = 0; | |
4974 | poll->head = head; | |
a31eb4a2 JX |
4975 | |
4976 | if (poll->events & EPOLLEXCLUSIVE) | |
4977 | add_wait_queue_exclusive(head, &poll->wait); | |
4978 | else | |
4979 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4980 | } |
4981 | ||
4982 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4983 | struct poll_table_struct *p) | |
4984 | { | |
4985 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4986 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4987 | |
807abcb0 | 4988 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4989 | } |
4990 | ||
d7718a9d JA |
4991 | static void io_async_task_func(struct callback_head *cb) |
4992 | { | |
4993 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
4994 | struct async_poll *apoll = req->apoll; | |
4995 | struct io_ring_ctx *ctx = req->ctx; | |
4996 | ||
4997 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
4998 | ||
74ce6ce4 | 4999 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5000 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5001 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5002 | return; |
d7718a9d JA |
5003 | } |
5004 | ||
31067255 | 5005 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5006 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5007 | hash_del(&req->hash_node); |
2bae047e | 5008 | |
d4e7cd36 | 5009 | io_poll_remove_double(req); |
74ce6ce4 JA |
5010 | spin_unlock_irq(&ctx->completion_lock); |
5011 | ||
0be0b0e3 PB |
5012 | if (!READ_ONCE(apoll->poll.canceled)) |
5013 | __io_req_task_submit(req); | |
5014 | else | |
5015 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5016 | |
6d816e08 | 5017 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5018 | kfree(apoll->double_poll); |
31067255 | 5019 | kfree(apoll); |
d7718a9d JA |
5020 | } |
5021 | ||
5022 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5023 | void *key) | |
5024 | { | |
5025 | struct io_kiocb *req = wait->private; | |
5026 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5027 | ||
5028 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5029 | key_to_poll(key)); | |
5030 | ||
5031 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5032 | } | |
5033 | ||
5034 | static void io_poll_req_insert(struct io_kiocb *req) | |
5035 | { | |
5036 | struct io_ring_ctx *ctx = req->ctx; | |
5037 | struct hlist_head *list; | |
5038 | ||
5039 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5040 | hlist_add_head(&req->hash_node, list); | |
5041 | } | |
5042 | ||
5043 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5044 | struct io_poll_iocb *poll, | |
5045 | struct io_poll_table *ipt, __poll_t mask, | |
5046 | wait_queue_func_t wake_func) | |
5047 | __acquires(&ctx->completion_lock) | |
5048 | { | |
5049 | struct io_ring_ctx *ctx = req->ctx; | |
5050 | bool cancel = false; | |
5051 | ||
4d52f338 | 5052 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5053 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5054 | poll->file = req->file; |
18bceab1 | 5055 | poll->wait.private = req; |
d7718a9d JA |
5056 | |
5057 | ipt->pt._key = mask; | |
5058 | ipt->req = req; | |
5059 | ipt->error = -EINVAL; | |
5060 | ||
d7718a9d JA |
5061 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5062 | ||
5063 | spin_lock_irq(&ctx->completion_lock); | |
5064 | if (likely(poll->head)) { | |
5065 | spin_lock(&poll->head->lock); | |
5066 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5067 | if (ipt->error) | |
5068 | cancel = true; | |
5069 | ipt->error = 0; | |
5070 | mask = 0; | |
5071 | } | |
5072 | if (mask || ipt->error) | |
5073 | list_del_init(&poll->wait.entry); | |
5074 | else if (cancel) | |
5075 | WRITE_ONCE(poll->canceled, true); | |
5076 | else if (!poll->done) /* actually waiting for an event */ | |
5077 | io_poll_req_insert(req); | |
5078 | spin_unlock(&poll->head->lock); | |
5079 | } | |
5080 | ||
5081 | return mask; | |
5082 | } | |
5083 | ||
5084 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5085 | { | |
5086 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5087 | struct io_ring_ctx *ctx = req->ctx; | |
5088 | struct async_poll *apoll; | |
5089 | struct io_poll_table ipt; | |
5090 | __poll_t mask, ret; | |
9dab14b8 | 5091 | int rw; |
d7718a9d JA |
5092 | |
5093 | if (!req->file || !file_can_poll(req->file)) | |
5094 | return false; | |
24c74678 | 5095 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5096 | return false; |
9dab14b8 JA |
5097 | if (def->pollin) |
5098 | rw = READ; | |
5099 | else if (def->pollout) | |
5100 | rw = WRITE; | |
5101 | else | |
5102 | return false; | |
5103 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5104 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5105 | return false; |
5106 | ||
5107 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5108 | if (unlikely(!apoll)) | |
5109 | return false; | |
807abcb0 | 5110 | apoll->double_poll = NULL; |
d7718a9d JA |
5111 | |
5112 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5113 | req->apoll = apoll; |
d7718a9d | 5114 | |
8755d97a | 5115 | mask = 0; |
d7718a9d | 5116 | if (def->pollin) |
8755d97a | 5117 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5118 | if (def->pollout) |
5119 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5120 | |
5121 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5122 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5123 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5124 | mask &= ~POLLIN; | |
5125 | ||
d7718a9d JA |
5126 | mask |= POLLERR | POLLPRI; |
5127 | ||
5128 | ipt.pt._qproc = io_async_queue_proc; | |
5129 | ||
5130 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5131 | io_async_wake); | |
a36da65c | 5132 | if (ret || ipt.error) { |
d4e7cd36 | 5133 | io_poll_remove_double(req); |
d7718a9d | 5134 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5135 | kfree(apoll->double_poll); |
d7718a9d JA |
5136 | kfree(apoll); |
5137 | return false; | |
5138 | } | |
5139 | spin_unlock_irq(&ctx->completion_lock); | |
5140 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5141 | apoll->poll.events); | |
5142 | return true; | |
5143 | } | |
5144 | ||
5145 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5146 | struct io_poll_iocb *poll) | |
221c5eb2 | 5147 | { |
b41e9852 | 5148 | bool do_complete = false; |
221c5eb2 JA |
5149 | |
5150 | spin_lock(&poll->head->lock); | |
5151 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5152 | if (!list_empty(&poll->wait.entry)) { |
5153 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5154 | do_complete = true; |
221c5eb2 JA |
5155 | } |
5156 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5157 | hash_del(&req->hash_node); |
d7718a9d JA |
5158 | return do_complete; |
5159 | } | |
5160 | ||
5161 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5162 | { | |
5163 | bool do_complete; | |
5164 | ||
d4e7cd36 JA |
5165 | io_poll_remove_double(req); |
5166 | ||
d7718a9d JA |
5167 | if (req->opcode == IORING_OP_POLL_ADD) { |
5168 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5169 | } else { | |
3bfa5bcb JA |
5170 | struct async_poll *apoll = req->apoll; |
5171 | ||
d7718a9d | 5172 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5173 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5174 | if (do_complete) { | |
d7718a9d | 5175 | io_put_req(req); |
807abcb0 | 5176 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5177 | kfree(apoll); |
5178 | } | |
b1f573bd XW |
5179 | } |
5180 | ||
b41e9852 JA |
5181 | if (do_complete) { |
5182 | io_cqring_fill_event(req, -ECANCELED); | |
5183 | io_commit_cqring(req->ctx); | |
f254ac04 | 5184 | req_set_fail_links(req); |
216578e5 | 5185 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5186 | } |
5187 | ||
5188 | return do_complete; | |
221c5eb2 JA |
5189 | } |
5190 | ||
76e1b642 JA |
5191 | /* |
5192 | * Returns true if we found and killed one or more poll requests | |
5193 | */ | |
6b81928d PB |
5194 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5195 | struct files_struct *files) | |
221c5eb2 | 5196 | { |
78076bb6 | 5197 | struct hlist_node *tmp; |
221c5eb2 | 5198 | struct io_kiocb *req; |
8e2e1faf | 5199 | int posted = 0, i; |
221c5eb2 JA |
5200 | |
5201 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5202 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5203 | struct hlist_head *list; | |
5204 | ||
5205 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5206 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5207 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5208 | posted += io_poll_remove_one(req); |
5209 | } | |
221c5eb2 JA |
5210 | } |
5211 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5212 | |
8e2e1faf JA |
5213 | if (posted) |
5214 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5215 | |
5216 | return posted != 0; | |
221c5eb2 JA |
5217 | } |
5218 | ||
47f46768 JA |
5219 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5220 | { | |
78076bb6 | 5221 | struct hlist_head *list; |
47f46768 JA |
5222 | struct io_kiocb *req; |
5223 | ||
78076bb6 JA |
5224 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5225 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5226 | if (sqe_addr != req->user_data) |
5227 | continue; | |
5228 | if (io_poll_remove_one(req)) | |
eac406c6 | 5229 | return 0; |
b41e9852 | 5230 | return -EALREADY; |
47f46768 JA |
5231 | } |
5232 | ||
5233 | return -ENOENT; | |
5234 | } | |
5235 | ||
3529d8c2 JA |
5236 | static int io_poll_remove_prep(struct io_kiocb *req, |
5237 | const struct io_uring_sqe *sqe) | |
0969e783 | 5238 | { |
0969e783 JA |
5239 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5240 | return -EINVAL; | |
5241 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5242 | sqe->poll_events) | |
5243 | return -EINVAL; | |
5244 | ||
018043be | 5245 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5246 | return 0; |
5247 | } | |
5248 | ||
221c5eb2 JA |
5249 | /* |
5250 | * Find a running poll command that matches one specified in sqe->addr, | |
5251 | * and remove it if found. | |
5252 | */ | |
61e98203 | 5253 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5254 | { |
5255 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5256 | int ret; |
221c5eb2 | 5257 | |
221c5eb2 | 5258 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5259 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5260 | spin_unlock_irq(&ctx->completion_lock); |
5261 | ||
4e88d6e7 JA |
5262 | if (ret < 0) |
5263 | req_set_fail_links(req); | |
e1e16097 | 5264 | io_req_complete(req, ret); |
221c5eb2 JA |
5265 | return 0; |
5266 | } | |
5267 | ||
221c5eb2 JA |
5268 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5269 | void *key) | |
5270 | { | |
c2f2eb7d JA |
5271 | struct io_kiocb *req = wait->private; |
5272 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5273 | |
d7718a9d | 5274 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5275 | } |
5276 | ||
221c5eb2 JA |
5277 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5278 | struct poll_table_struct *p) | |
5279 | { | |
5280 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5281 | ||
e8c2bc1f | 5282 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5283 | } |
5284 | ||
3529d8c2 | 5285 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5286 | { |
5287 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5288 | u32 events; |
221c5eb2 JA |
5289 | |
5290 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5291 | return -EINVAL; | |
5292 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5293 | return -EINVAL; | |
5294 | ||
5769a351 JX |
5295 | events = READ_ONCE(sqe->poll32_events); |
5296 | #ifdef __BIG_ENDIAN | |
5297 | events = swahw32(events); | |
5298 | #endif | |
a31eb4a2 JX |
5299 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5300 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5301 | return 0; |
5302 | } | |
5303 | ||
61e98203 | 5304 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5305 | { |
5306 | struct io_poll_iocb *poll = &req->poll; | |
5307 | struct io_ring_ctx *ctx = req->ctx; | |
5308 | struct io_poll_table ipt; | |
0969e783 | 5309 | __poll_t mask; |
0969e783 | 5310 | |
d7718a9d | 5311 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5312 | |
d7718a9d JA |
5313 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5314 | io_poll_wake); | |
221c5eb2 | 5315 | |
8c838788 | 5316 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5317 | ipt.error = 0; |
b0dd8a41 | 5318 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5319 | } |
221c5eb2 JA |
5320 | spin_unlock_irq(&ctx->completion_lock); |
5321 | ||
8c838788 JA |
5322 | if (mask) { |
5323 | io_cqring_ev_posted(ctx); | |
014db007 | 5324 | io_put_req(req); |
221c5eb2 | 5325 | } |
8c838788 | 5326 | return ipt.error; |
221c5eb2 JA |
5327 | } |
5328 | ||
5262f567 JA |
5329 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5330 | { | |
ad8a48ac JA |
5331 | struct io_timeout_data *data = container_of(timer, |
5332 | struct io_timeout_data, timer); | |
5333 | struct io_kiocb *req = data->req; | |
5334 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5335 | unsigned long flags; |
5336 | ||
5262f567 | 5337 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5338 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5339 | atomic_set(&req->ctx->cq_timeouts, |
5340 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5341 | ||
78e19bbe | 5342 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5343 | io_commit_cqring(ctx); |
5344 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5345 | ||
5346 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5347 | req_set_fail_links(req); |
5262f567 JA |
5348 | io_put_req(req); |
5349 | return HRTIMER_NORESTART; | |
5350 | } | |
5351 | ||
fbd15848 PB |
5352 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5353 | __u64 user_data) | |
f254ac04 | 5354 | { |
fbd15848 | 5355 | struct io_timeout_data *io; |
47f46768 JA |
5356 | struct io_kiocb *req; |
5357 | int ret = -ENOENT; | |
f254ac04 | 5358 | |
135fcde8 | 5359 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5360 | if (user_data == req->user_data) { |
47f46768 JA |
5361 | ret = 0; |
5362 | break; | |
5363 | } | |
5364 | } | |
5365 | ||
5366 | if (ret == -ENOENT) | |
fbd15848 PB |
5367 | return ERR_PTR(ret); |
5368 | ||
5369 | io = req->async_data; | |
e8c2bc1f | 5370 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5371 | if (ret == -1) |
fbd15848 | 5372 | return ERR_PTR(-EALREADY); |
a71976f3 | 5373 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5374 | return req; |
5375 | } | |
47f46768 | 5376 | |
fbd15848 PB |
5377 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5378 | { | |
5379 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5380 | ||
5381 | if (IS_ERR(req)) | |
5382 | return PTR_ERR(req); | |
f254ac04 JA |
5383 | |
5384 | req_set_fail_links(req); | |
f254ac04 | 5385 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5386 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5387 | return 0; |
5388 | } | |
5389 | ||
9c8e11b3 PB |
5390 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5391 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5392 | { |
9c8e11b3 PB |
5393 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5394 | struct io_timeout_data *data; | |
47f46768 | 5395 | |
9c8e11b3 PB |
5396 | if (IS_ERR(req)) |
5397 | return PTR_ERR(req); | |
47f46768 | 5398 | |
9c8e11b3 PB |
5399 | req->timeout.off = 0; /* noseq */ |
5400 | data = req->async_data; | |
5401 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5402 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5403 | data->timer.function = io_timeout_fn; | |
5404 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5405 | return 0; | |
47f46768 JA |
5406 | } |
5407 | ||
3529d8c2 JA |
5408 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5409 | const struct io_uring_sqe *sqe) | |
b29472ee | 5410 | { |
9c8e11b3 PB |
5411 | struct io_timeout_rem *tr = &req->timeout_rem; |
5412 | ||
b29472ee JA |
5413 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5414 | return -EINVAL; | |
61710e43 DA |
5415 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5416 | return -EINVAL; | |
9c8e11b3 | 5417 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5418 | return -EINVAL; |
5419 | ||
9c8e11b3 PB |
5420 | tr->addr = READ_ONCE(sqe->addr); |
5421 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5422 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5423 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5424 | return -EINVAL; | |
5425 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5426 | return -EFAULT; | |
5427 | } else if (tr->flags) { | |
5428 | /* timeout removal doesn't support flags */ | |
b29472ee | 5429 | return -EINVAL; |
9c8e11b3 | 5430 | } |
b29472ee | 5431 | |
b29472ee JA |
5432 | return 0; |
5433 | } | |
5434 | ||
8662daec PB |
5435 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5436 | { | |
5437 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5438 | : HRTIMER_MODE_REL; | |
5439 | } | |
5440 | ||
11365043 JA |
5441 | /* |
5442 | * Remove or update an existing timeout command | |
5443 | */ | |
61e98203 | 5444 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5445 | { |
9c8e11b3 | 5446 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5447 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5448 | int ret; |
11365043 | 5449 | |
11365043 | 5450 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5451 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5452 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5453 | else |
5454 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5455 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5456 | |
47f46768 | 5457 | io_cqring_fill_event(req, ret); |
11365043 JA |
5458 | io_commit_cqring(ctx); |
5459 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5460 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5461 | if (ret < 0) |
5462 | req_set_fail_links(req); | |
ec9c02ad | 5463 | io_put_req(req); |
11365043 | 5464 | return 0; |
5262f567 JA |
5465 | } |
5466 | ||
3529d8c2 | 5467 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5468 | bool is_timeout_link) |
5262f567 | 5469 | { |
ad8a48ac | 5470 | struct io_timeout_data *data; |
a41525ab | 5471 | unsigned flags; |
56080b02 | 5472 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5473 | |
ad8a48ac | 5474 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5475 | return -EINVAL; |
ad8a48ac | 5476 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5477 | return -EINVAL; |
56080b02 | 5478 | if (off && is_timeout_link) |
2d28390a | 5479 | return -EINVAL; |
a41525ab JA |
5480 | flags = READ_ONCE(sqe->timeout_flags); |
5481 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5482 | return -EINVAL; |
bdf20073 | 5483 | |
bfe68a22 | 5484 | req->timeout.off = off; |
26a61679 | 5485 | |
e8c2bc1f | 5486 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5487 | return -ENOMEM; |
5488 | ||
e8c2bc1f | 5489 | data = req->async_data; |
ad8a48ac | 5490 | data->req = req; |
ad8a48ac JA |
5491 | |
5492 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5493 | return -EFAULT; |
5494 | ||
8662daec | 5495 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5496 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5497 | return 0; | |
5498 | } | |
5499 | ||
61e98203 | 5500 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5501 | { |
ad8a48ac | 5502 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5503 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5504 | struct list_head *entry; |
bfe68a22 | 5505 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5506 | |
733f5c95 | 5507 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5508 | |
5262f567 JA |
5509 | /* |
5510 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5511 | * timeout event to be satisfied. If it isn't set, then this is |
5512 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5513 | */ |
8eb7e2d0 | 5514 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5515 | entry = ctx->timeout_list.prev; |
5516 | goto add; | |
5517 | } | |
5262f567 | 5518 | |
bfe68a22 PB |
5519 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5520 | req->timeout.target_seq = tail + off; | |
5262f567 | 5521 | |
f010505b MDG |
5522 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5523 | * This is safe because ->completion_lock is held, and submissions | |
5524 | * and completions are never mixed in the same ->completion_lock section. | |
5525 | */ | |
5526 | ctx->cq_last_tm_flush = tail; | |
5527 | ||
5262f567 JA |
5528 | /* |
5529 | * Insertion sort, ensuring the first entry in the list is always | |
5530 | * the one we need first. | |
5531 | */ | |
5262f567 | 5532 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5533 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5534 | timeout.list); | |
5262f567 | 5535 | |
8eb7e2d0 | 5536 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5537 | continue; |
bfe68a22 PB |
5538 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5539 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5540 | break; |
5541 | } | |
93bd25bb | 5542 | add: |
135fcde8 | 5543 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5544 | data->timer.function = io_timeout_fn; |
5545 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5546 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5547 | return 0; |
5548 | } | |
5262f567 | 5549 | |
62755e35 JA |
5550 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5551 | { | |
5552 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5553 | ||
5554 | return req->user_data == (unsigned long) data; | |
5555 | } | |
5556 | ||
5aa75ed5 | 5557 | static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr) |
62755e35 | 5558 | { |
62755e35 | 5559 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5560 | int ret = 0; |
5561 | ||
5aa75ed5 JA |
5562 | if (!tctx->io_wq) |
5563 | return -ENOENT; | |
5564 | ||
5565 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false); | |
62755e35 JA |
5566 | switch (cancel_ret) { |
5567 | case IO_WQ_CANCEL_OK: | |
5568 | ret = 0; | |
5569 | break; | |
5570 | case IO_WQ_CANCEL_RUNNING: | |
5571 | ret = -EALREADY; | |
5572 | break; | |
5573 | case IO_WQ_CANCEL_NOTFOUND: | |
5574 | ret = -ENOENT; | |
5575 | break; | |
5576 | } | |
5577 | ||
e977d6d3 JA |
5578 | return ret; |
5579 | } | |
5580 | ||
47f46768 JA |
5581 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5582 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5583 | int success_ret) |
47f46768 JA |
5584 | { |
5585 | unsigned long flags; | |
5586 | int ret; | |
5587 | ||
5aa75ed5 JA |
5588 | ret = io_async_cancel_one(req->task->io_uring, |
5589 | (void *) (unsigned long) sqe_addr); | |
47f46768 JA |
5590 | if (ret != -ENOENT) { |
5591 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5592 | goto done; | |
5593 | } | |
5594 | ||
5595 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5596 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5597 | if (ret != -ENOENT) | |
5598 | goto done; | |
5599 | ret = io_poll_cancel(ctx, sqe_addr); | |
5600 | done: | |
b0dd8a41 JA |
5601 | if (!ret) |
5602 | ret = success_ret; | |
47f46768 JA |
5603 | io_cqring_fill_event(req, ret); |
5604 | io_commit_cqring(ctx); | |
5605 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5606 | io_cqring_ev_posted(ctx); | |
5607 | ||
4e88d6e7 JA |
5608 | if (ret < 0) |
5609 | req_set_fail_links(req); | |
014db007 | 5610 | io_put_req(req); |
47f46768 JA |
5611 | } |
5612 | ||
3529d8c2 JA |
5613 | static int io_async_cancel_prep(struct io_kiocb *req, |
5614 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5615 | { |
fbf23849 | 5616 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5617 | return -EINVAL; |
61710e43 DA |
5618 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5619 | return -EINVAL; | |
5620 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5621 | return -EINVAL; |
5622 | ||
fbf23849 JA |
5623 | req->cancel.addr = READ_ONCE(sqe->addr); |
5624 | return 0; | |
5625 | } | |
5626 | ||
61e98203 | 5627 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5628 | { |
5629 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5630 | |
014db007 | 5631 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5632 | return 0; |
5633 | } | |
5634 | ||
269bbe5f | 5635 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5636 | const struct io_uring_sqe *sqe) |
5637 | { | |
6ca56f84 JA |
5638 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5639 | return -EINVAL; | |
61710e43 DA |
5640 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5641 | return -EINVAL; | |
5642 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5643 | return -EINVAL; |
5644 | ||
269bbe5f BM |
5645 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5646 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5647 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5648 | return -EINVAL; |
269bbe5f | 5649 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5650 | return 0; |
5651 | } | |
5652 | ||
889fca73 | 5653 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5654 | { |
5655 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5656 | struct io_uring_rsrc_update up; |
05f3fb3c | 5657 | int ret; |
fbf23849 | 5658 | |
45d189c6 | 5659 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5660 | return -EAGAIN; |
05f3fb3c | 5661 | |
269bbe5f BM |
5662 | up.offset = req->rsrc_update.offset; |
5663 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5664 | |
5665 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5666 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5667 | mutex_unlock(&ctx->uring_lock); |
5668 | ||
5669 | if (ret < 0) | |
5670 | req_set_fail_links(req); | |
889fca73 | 5671 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5672 | return 0; |
5673 | } | |
5674 | ||
bfe76559 | 5675 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5676 | { |
d625c6ee | 5677 | switch (req->opcode) { |
e781573e | 5678 | case IORING_OP_NOP: |
bfe76559 | 5679 | return 0; |
f67676d1 JA |
5680 | case IORING_OP_READV: |
5681 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5682 | case IORING_OP_READ: |
bfe76559 | 5683 | return io_read_prep(req, sqe); |
f67676d1 JA |
5684 | case IORING_OP_WRITEV: |
5685 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5686 | case IORING_OP_WRITE: |
bfe76559 | 5687 | return io_write_prep(req, sqe); |
0969e783 | 5688 | case IORING_OP_POLL_ADD: |
bfe76559 | 5689 | return io_poll_add_prep(req, sqe); |
0969e783 | 5690 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5691 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5692 | case IORING_OP_FSYNC: |
1155c76a | 5693 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5694 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5695 | return io_sfr_prep(req, sqe); |
03b1230c | 5696 | case IORING_OP_SENDMSG: |
fddaface | 5697 | case IORING_OP_SEND: |
bfe76559 | 5698 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5699 | case IORING_OP_RECVMSG: |
fddaface | 5700 | case IORING_OP_RECV: |
bfe76559 | 5701 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5702 | case IORING_OP_CONNECT: |
bfe76559 | 5703 | return io_connect_prep(req, sqe); |
2d28390a | 5704 | case IORING_OP_TIMEOUT: |
bfe76559 | 5705 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5706 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5707 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5708 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5709 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5710 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5711 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5712 | case IORING_OP_ACCEPT: |
bfe76559 | 5713 | return io_accept_prep(req, sqe); |
d63d1b5e | 5714 | case IORING_OP_FALLOCATE: |
bfe76559 | 5715 | return io_fallocate_prep(req, sqe); |
15b71abe | 5716 | case IORING_OP_OPENAT: |
bfe76559 | 5717 | return io_openat_prep(req, sqe); |
b5dba59e | 5718 | case IORING_OP_CLOSE: |
bfe76559 | 5719 | return io_close_prep(req, sqe); |
05f3fb3c | 5720 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5721 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5722 | case IORING_OP_STATX: |
bfe76559 | 5723 | return io_statx_prep(req, sqe); |
4840e418 | 5724 | case IORING_OP_FADVISE: |
bfe76559 | 5725 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5726 | case IORING_OP_MADVISE: |
bfe76559 | 5727 | return io_madvise_prep(req, sqe); |
cebdb986 | 5728 | case IORING_OP_OPENAT2: |
bfe76559 | 5729 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5730 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5731 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5732 | case IORING_OP_SPLICE: |
bfe76559 | 5733 | return io_splice_prep(req, sqe); |
ddf0322d | 5734 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5735 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5736 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5737 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5738 | case IORING_OP_TEE: |
bfe76559 | 5739 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5740 | case IORING_OP_SHUTDOWN: |
5741 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5742 | case IORING_OP_RENAMEAT: |
5743 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5744 | case IORING_OP_UNLINKAT: |
5745 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5746 | } |
5747 | ||
bfe76559 PB |
5748 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5749 | req->opcode); | |
5750 | return-EINVAL; | |
5751 | } | |
5752 | ||
93642ef8 | 5753 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5754 | { |
93642ef8 PB |
5755 | switch (req->opcode) { |
5756 | case IORING_OP_READV: | |
5757 | case IORING_OP_READ_FIXED: | |
5758 | case IORING_OP_READ: | |
5759 | return io_rw_prep_async(req, READ); | |
5760 | case IORING_OP_WRITEV: | |
5761 | case IORING_OP_WRITE_FIXED: | |
5762 | case IORING_OP_WRITE: | |
5763 | return io_rw_prep_async(req, WRITE); | |
5764 | case IORING_OP_SENDMSG: | |
5765 | case IORING_OP_SEND: | |
5766 | return io_sendmsg_prep_async(req); | |
5767 | case IORING_OP_RECVMSG: | |
5768 | case IORING_OP_RECV: | |
5769 | return io_recvmsg_prep_async(req); | |
5770 | case IORING_OP_CONNECT: | |
5771 | return io_connect_prep_async(req); | |
5772 | } | |
5773 | return 0; | |
5774 | } | |
5775 | ||
be7053b7 | 5776 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5777 | { |
be7053b7 | 5778 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5779 | return 0; |
be7053b7 | 5780 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5781 | if (req->async_data) |
be7053b7 PB |
5782 | return 0; |
5783 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5784 | return -EAGAIN; |
be7053b7 | 5785 | return io_req_prep_async(req); |
f67676d1 JA |
5786 | } |
5787 | ||
9cf7c104 PB |
5788 | static u32 io_get_sequence(struct io_kiocb *req) |
5789 | { | |
5790 | struct io_kiocb *pos; | |
5791 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5792 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5793 | |
f2f87370 PB |
5794 | io_for_each_link(pos, req) |
5795 | nr_reqs++; | |
9cf7c104 PB |
5796 | |
5797 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5798 | return total_submitted - nr_reqs; | |
5799 | } | |
5800 | ||
be7053b7 | 5801 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5802 | { |
a197f664 | 5803 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5804 | struct io_defer_entry *de; |
f67676d1 | 5805 | int ret; |
9cf7c104 | 5806 | u32 seq; |
de0617e4 | 5807 | |
9d858b21 | 5808 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5809 | if (likely(list_empty_careful(&ctx->defer_list) && |
5810 | !(req->flags & REQ_F_IO_DRAIN))) | |
5811 | return 0; | |
5812 | ||
5813 | seq = io_get_sequence(req); | |
5814 | /* Still a chance to pass the sequence check */ | |
5815 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5816 | return 0; |
5817 | ||
be7053b7 PB |
5818 | ret = io_req_defer_prep(req); |
5819 | if (ret) | |
5820 | return ret; | |
cbdcb435 | 5821 | io_prep_async_link(req); |
27dc8338 PB |
5822 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5823 | if (!de) | |
5824 | return -ENOMEM; | |
2d28390a | 5825 | |
de0617e4 | 5826 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5827 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5828 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5829 | kfree(de); |
ae34817b PB |
5830 | io_queue_async_work(req); |
5831 | return -EIOCBQUEUED; | |
de0617e4 JA |
5832 | } |
5833 | ||
915967f6 | 5834 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5835 | de->req = req; |
9cf7c104 | 5836 | de->seq = seq; |
27dc8338 | 5837 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5838 | spin_unlock_irq(&ctx->completion_lock); |
5839 | return -EIOCBQUEUED; | |
5840 | } | |
5841 | ||
3ca405eb | 5842 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5843 | { |
0e1b6fe3 PB |
5844 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5845 | switch (req->opcode) { | |
5846 | case IORING_OP_READV: | |
5847 | case IORING_OP_READ_FIXED: | |
5848 | case IORING_OP_READ: | |
bcda7baa | 5849 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5850 | break; |
5851 | case IORING_OP_RECVMSG: | |
5852 | case IORING_OP_RECV: | |
bcda7baa | 5853 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5854 | break; |
5855 | } | |
5856 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5857 | } |
5858 | ||
0e1b6fe3 PB |
5859 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5860 | switch (req->opcode) { | |
5861 | case IORING_OP_READV: | |
5862 | case IORING_OP_READ_FIXED: | |
5863 | case IORING_OP_READ: | |
5864 | case IORING_OP_WRITEV: | |
5865 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5866 | case IORING_OP_WRITE: { |
5867 | struct io_async_rw *io = req->async_data; | |
5868 | if (io->free_iovec) | |
5869 | kfree(io->free_iovec); | |
0e1b6fe3 | 5870 | break; |
e8c2bc1f | 5871 | } |
0e1b6fe3 | 5872 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5873 | case IORING_OP_SENDMSG: { |
5874 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5875 | |
5876 | kfree(io->free_iov); | |
0e1b6fe3 | 5877 | break; |
e8c2bc1f | 5878 | } |
0e1b6fe3 PB |
5879 | case IORING_OP_SPLICE: |
5880 | case IORING_OP_TEE: | |
5881 | io_put_file(req, req->splice.file_in, | |
5882 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5883 | break; | |
f3cd4850 JA |
5884 | case IORING_OP_OPENAT: |
5885 | case IORING_OP_OPENAT2: | |
5886 | if (req->open.filename) | |
5887 | putname(req->open.filename); | |
5888 | break; | |
80a261fd JA |
5889 | case IORING_OP_RENAMEAT: |
5890 | putname(req->rename.oldpath); | |
5891 | putname(req->rename.newpath); | |
5892 | break; | |
14a1143b JA |
5893 | case IORING_OP_UNLINKAT: |
5894 | putname(req->unlink.filename); | |
5895 | break; | |
0e1b6fe3 PB |
5896 | } |
5897 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5898 | } |
99bc4c38 PB |
5899 | } |
5900 | ||
889fca73 | 5901 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5902 | { |
a197f664 | 5903 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5904 | const struct cred *creds = NULL; |
d625c6ee | 5905 | int ret; |
2b188cc1 | 5906 | |
5730b27e JA |
5907 | if (req->work.personality) { |
5908 | const struct cred *new_creds; | |
5909 | ||
5910 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5911 | mutex_lock(&ctx->uring_lock); | |
5912 | new_creds = idr_find(&ctx->personality_idr, req->work.personality); | |
5913 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5914 | mutex_unlock(&ctx->uring_lock); | |
5915 | if (!new_creds) | |
5916 | return -EINVAL; | |
5917 | creds = override_creds(new_creds); | |
5918 | } | |
5919 | ||
d625c6ee | 5920 | switch (req->opcode) { |
2b188cc1 | 5921 | case IORING_OP_NOP: |
889fca73 | 5922 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5923 | break; |
5924 | case IORING_OP_READV: | |
edafccee | 5925 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5926 | case IORING_OP_READ: |
889fca73 | 5927 | ret = io_read(req, issue_flags); |
edafccee | 5928 | break; |
3529d8c2 | 5929 | case IORING_OP_WRITEV: |
edafccee | 5930 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5931 | case IORING_OP_WRITE: |
889fca73 | 5932 | ret = io_write(req, issue_flags); |
2b188cc1 | 5933 | break; |
c992fe29 | 5934 | case IORING_OP_FSYNC: |
45d189c6 | 5935 | ret = io_fsync(req, issue_flags); |
c992fe29 | 5936 | break; |
221c5eb2 | 5937 | case IORING_OP_POLL_ADD: |
61e98203 | 5938 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
5939 | break; |
5940 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 5941 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 5942 | break; |
5d17b4a4 | 5943 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 5944 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 5945 | break; |
0fa03c62 | 5946 | case IORING_OP_SENDMSG: |
889fca73 | 5947 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 5948 | break; |
fddaface | 5949 | case IORING_OP_SEND: |
889fca73 | 5950 | ret = io_send(req, issue_flags); |
0fa03c62 | 5951 | break; |
aa1fa28f | 5952 | case IORING_OP_RECVMSG: |
889fca73 | 5953 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 5954 | break; |
fddaface | 5955 | case IORING_OP_RECV: |
889fca73 | 5956 | ret = io_recv(req, issue_flags); |
aa1fa28f | 5957 | break; |
5262f567 | 5958 | case IORING_OP_TIMEOUT: |
61e98203 | 5959 | ret = io_timeout(req, issue_flags); |
5262f567 | 5960 | break; |
11365043 | 5961 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 5962 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 5963 | break; |
17f2fe35 | 5964 | case IORING_OP_ACCEPT: |
889fca73 | 5965 | ret = io_accept(req, issue_flags); |
17f2fe35 | 5966 | break; |
f8e85cf2 | 5967 | case IORING_OP_CONNECT: |
889fca73 | 5968 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 5969 | break; |
62755e35 | 5970 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 5971 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 5972 | break; |
d63d1b5e | 5973 | case IORING_OP_FALLOCATE: |
45d189c6 | 5974 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 5975 | break; |
15b71abe | 5976 | case IORING_OP_OPENAT: |
45d189c6 | 5977 | ret = io_openat(req, issue_flags); |
15b71abe | 5978 | break; |
b5dba59e | 5979 | case IORING_OP_CLOSE: |
889fca73 | 5980 | ret = io_close(req, issue_flags); |
b5dba59e | 5981 | break; |
05f3fb3c | 5982 | case IORING_OP_FILES_UPDATE: |
889fca73 | 5983 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 5984 | break; |
eddc7ef5 | 5985 | case IORING_OP_STATX: |
45d189c6 | 5986 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 5987 | break; |
4840e418 | 5988 | case IORING_OP_FADVISE: |
45d189c6 | 5989 | ret = io_fadvise(req, issue_flags); |
4840e418 | 5990 | break; |
c1ca757b | 5991 | case IORING_OP_MADVISE: |
45d189c6 | 5992 | ret = io_madvise(req, issue_flags); |
c1ca757b | 5993 | break; |
cebdb986 | 5994 | case IORING_OP_OPENAT2: |
45d189c6 | 5995 | ret = io_openat2(req, issue_flags); |
cebdb986 | 5996 | break; |
3e4827b0 | 5997 | case IORING_OP_EPOLL_CTL: |
889fca73 | 5998 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 5999 | break; |
7d67af2c | 6000 | case IORING_OP_SPLICE: |
45d189c6 | 6001 | ret = io_splice(req, issue_flags); |
7d67af2c | 6002 | break; |
ddf0322d | 6003 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6004 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6005 | break; |
067524e9 | 6006 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6007 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6008 | break; |
f2a8d5c7 | 6009 | case IORING_OP_TEE: |
45d189c6 | 6010 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6011 | break; |
36f4fa68 | 6012 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6013 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6014 | break; |
80a261fd | 6015 | case IORING_OP_RENAMEAT: |
45d189c6 | 6016 | ret = io_renameat(req, issue_flags); |
80a261fd | 6017 | break; |
14a1143b | 6018 | case IORING_OP_UNLINKAT: |
45d189c6 | 6019 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6020 | break; |
2b188cc1 JA |
6021 | default: |
6022 | ret = -EINVAL; | |
6023 | break; | |
6024 | } | |
6025 | ||
5730b27e JA |
6026 | if (creds) |
6027 | revert_creds(creds); | |
6028 | ||
def596e9 JA |
6029 | if (ret) |
6030 | return ret; | |
6031 | ||
b532576e JA |
6032 | /* If the op doesn't have a file, we're not polling for it */ |
6033 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6034 | const bool in_async = io_wq_current_is_worker(); |
6035 | ||
11ba820b JA |
6036 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6037 | if (in_async) | |
6038 | mutex_lock(&ctx->uring_lock); | |
6039 | ||
2e9dbe90 | 6040 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6041 | |
6042 | if (in_async) | |
6043 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6044 | } |
6045 | ||
6046 | return 0; | |
2b188cc1 JA |
6047 | } |
6048 | ||
5280f7e5 | 6049 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6050 | { |
6051 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6052 | struct io_kiocb *timeout; |
561fb04a | 6053 | int ret = 0; |
2b188cc1 | 6054 | |
6df1db6b PB |
6055 | timeout = io_prep_linked_timeout(req); |
6056 | if (timeout) | |
6057 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6058 | |
4014d943 | 6059 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6060 | ret = -ECANCELED; |
31b51510 | 6061 | |
561fb04a | 6062 | if (!ret) { |
561fb04a | 6063 | do { |
889fca73 | 6064 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6065 | /* |
6066 | * We can get EAGAIN for polled IO even though we're | |
6067 | * forcing a sync submission from here, since we can't | |
6068 | * wait for request slots on the block side. | |
6069 | */ | |
6070 | if (ret != -EAGAIN) | |
6071 | break; | |
6072 | cond_resched(); | |
6073 | } while (1); | |
6074 | } | |
31b51510 | 6075 | |
a3df7698 | 6076 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6077 | if (ret) { |
a3df7698 PB |
6078 | /* io-wq is going to take one down */ |
6079 | refcount_inc(&req->refs); | |
6080 | io_req_task_queue_fail(req, ret); | |
edafccee | 6081 | } |
2b188cc1 JA |
6082 | } |
6083 | ||
65e19f54 JA |
6084 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6085 | int index) | |
6086 | { | |
269bbe5f | 6087 | struct fixed_rsrc_table *table; |
65e19f54 | 6088 | |
05f3fb3c | 6089 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6090 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6091 | } |
6092 | ||
8371adf5 PB |
6093 | static struct file *io_file_get(struct io_submit_state *state, |
6094 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6095 | { |
a197f664 | 6096 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6097 | struct file *file; |
09bb8394 | 6098 | |
8da11c19 | 6099 | if (fixed) { |
479f517b | 6100 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6101 | return NULL; |
b7620121 | 6102 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6103 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6104 | io_set_resource_node(req); |
09bb8394 | 6105 | } else { |
c826bd7a | 6106 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6107 | file = __io_file_get(state, fd); |
09bb8394 JA |
6108 | } |
6109 | ||
ce3d5aae PB |
6110 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6111 | io_req_track_inflight(req); | |
8371adf5 | 6112 | return file; |
09bb8394 JA |
6113 | } |
6114 | ||
2665abfd | 6115 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6116 | { |
ad8a48ac JA |
6117 | struct io_timeout_data *data = container_of(timer, |
6118 | struct io_timeout_data, timer); | |
90cd7e42 | 6119 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6120 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6121 | unsigned long flags; |
2665abfd JA |
6122 | |
6123 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6124 | prev = req->timeout.head; |
6125 | req->timeout.head = NULL; | |
2665abfd JA |
6126 | |
6127 | /* | |
6128 | * We don't expect the list to be empty, that will only happen if we | |
6129 | * race with the completion of the linked work. | |
6130 | */ | |
90cd7e42 | 6131 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6132 | io_remove_next_linked(prev); |
90cd7e42 PB |
6133 | else |
6134 | prev = NULL; | |
2665abfd JA |
6135 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6136 | ||
6137 | if (prev) { | |
4e88d6e7 | 6138 | req_set_fail_links(prev); |
014db007 | 6139 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6140 | io_put_req_deferred(prev, 1); |
47f46768 | 6141 | } else { |
9ae1f8dd PB |
6142 | io_req_complete_post(req, -ETIME, 0); |
6143 | io_put_req_deferred(req, 1); | |
2665abfd | 6144 | } |
2665abfd JA |
6145 | return HRTIMER_NORESTART; |
6146 | } | |
6147 | ||
7271ef3a | 6148 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6149 | { |
76a46e06 | 6150 | /* |
f2f87370 PB |
6151 | * If the back reference is NULL, then our linked request finished |
6152 | * before we got a chance to setup the timer | |
76a46e06 | 6153 | */ |
90cd7e42 | 6154 | if (req->timeout.head) { |
e8c2bc1f | 6155 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6156 | |
ad8a48ac JA |
6157 | data->timer.function = io_link_timeout_fn; |
6158 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6159 | data->mode); | |
2665abfd | 6160 | } |
7271ef3a JA |
6161 | } |
6162 | ||
6163 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6164 | { | |
6165 | struct io_ring_ctx *ctx = req->ctx; | |
6166 | ||
6167 | spin_lock_irq(&ctx->completion_lock); | |
6168 | __io_queue_linked_timeout(req); | |
76a46e06 | 6169 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6170 | |
2665abfd | 6171 | /* drop submission reference */ |
76a46e06 JA |
6172 | io_put_req(req); |
6173 | } | |
2665abfd | 6174 | |
ad8a48ac | 6175 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6176 | { |
f2f87370 | 6177 | struct io_kiocb *nxt = req->link; |
2665abfd | 6178 | |
f2f87370 PB |
6179 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6180 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6181 | return NULL; |
2665abfd | 6182 | |
90cd7e42 | 6183 | nxt->timeout.head = req; |
900fad45 | 6184 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6185 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6186 | return nxt; |
2665abfd JA |
6187 | } |
6188 | ||
c5eef2b9 | 6189 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6190 | { |
d3d7298d | 6191 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6192 | int ret; |
2b188cc1 | 6193 | |
c5eef2b9 | 6194 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6195 | |
491381ce JA |
6196 | /* |
6197 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6198 | * doesn't support non-blocking read/write attempts | |
6199 | */ | |
24c74678 | 6200 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6201 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6202 | /* |
6203 | * Queued up for async execution, worker will release | |
6204 | * submit reference when the iocb is actually submitted. | |
6205 | */ | |
6206 | io_queue_async_work(req); | |
2b188cc1 | 6207 | } |
0d63c148 PB |
6208 | } else if (likely(!ret)) { |
6209 | /* drop submission reference */ | |
e342c807 | 6210 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6211 | struct io_ring_ctx *ctx = req->ctx; |
6212 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6213 | |
6dd0be1e | 6214 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6215 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6216 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6217 | } else { |
d3d7298d | 6218 | io_put_req(req); |
0d63c148 PB |
6219 | } |
6220 | } else { | |
4e88d6e7 | 6221 | req_set_fail_links(req); |
e65ef56d | 6222 | io_put_req(req); |
e1e16097 | 6223 | io_req_complete(req, ret); |
9e645e11 | 6224 | } |
d3d7298d PB |
6225 | if (linked_timeout) |
6226 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6227 | } |
6228 | ||
be7053b7 | 6229 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6230 | { |
6231 | int ret; | |
6232 | ||
be7053b7 | 6233 | ret = io_req_defer(req); |
4fe2c963 JL |
6234 | if (ret) { |
6235 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6236 | fail_req: |
4e88d6e7 | 6237 | req_set_fail_links(req); |
e1e16097 JA |
6238 | io_put_req(req); |
6239 | io_req_complete(req, ret); | |
4fe2c963 | 6240 | } |
2550878f | 6241 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6242 | ret = io_req_defer_prep(req); |
6243 | if (unlikely(ret)) | |
6244 | goto fail_req; | |
ce35a47a JA |
6245 | io_queue_async_work(req); |
6246 | } else { | |
c5eef2b9 | 6247 | __io_queue_sqe(req); |
ce35a47a | 6248 | } |
4fe2c963 JL |
6249 | } |
6250 | ||
b16fed66 PB |
6251 | /* |
6252 | * Check SQE restrictions (opcode and flags). | |
6253 | * | |
6254 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6255 | */ | |
6256 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6257 | struct io_kiocb *req, | |
6258 | unsigned int sqe_flags) | |
4fe2c963 | 6259 | { |
b16fed66 PB |
6260 | if (!ctx->restricted) |
6261 | return true; | |
6262 | ||
6263 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6264 | return false; | |
6265 | ||
6266 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6267 | ctx->restrictions.sqe_flags_required) | |
6268 | return false; | |
6269 | ||
6270 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6271 | ctx->restrictions.sqe_flags_required)) | |
6272 | return false; | |
6273 | ||
6274 | return true; | |
4fe2c963 JL |
6275 | } |
6276 | ||
b16fed66 PB |
6277 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6278 | const struct io_uring_sqe *sqe) | |
6279 | { | |
6280 | struct io_submit_state *state; | |
6281 | unsigned int sqe_flags; | |
5730b27e | 6282 | int ret = 0; |
b16fed66 PB |
6283 | |
6284 | req->opcode = READ_ONCE(sqe->opcode); | |
6285 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6286 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6287 | req->user_data = READ_ONCE(sqe->user_data); | |
6288 | req->async_data = NULL; | |
6289 | req->file = NULL; | |
6290 | req->ctx = ctx; | |
6291 | req->link = NULL; | |
6292 | req->fixed_rsrc_refs = NULL; | |
6293 | /* one is dropped after submission, the other at completion */ | |
6294 | refcount_set(&req->refs, 2); | |
6295 | req->task = current; | |
6296 | req->result = 0; | |
6297 | ||
6298 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6299 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6300 | req->flags = 0; | |
b16fed66 | 6301 | return -EINVAL; |
ebf4a5db | 6302 | } |
b16fed66 PB |
6303 | |
6304 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6305 | return -EINVAL; | |
6306 | ||
b16fed66 PB |
6307 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6308 | return -EACCES; | |
6309 | ||
6310 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6311 | !io_op_defs[req->opcode].buffer_select) | |
6312 | return -EOPNOTSUPP; | |
863e0560 | 6313 | |
5730b27e JA |
6314 | req->work.list.next = NULL; |
6315 | req->work.flags = 0; | |
6316 | req->work.personality = READ_ONCE(sqe->personality); | |
b16fed66 PB |
6317 | state = &ctx->submit_state; |
6318 | ||
6319 | /* | |
6320 | * Plug now if we have more than 1 IO left after this, and the target | |
6321 | * is potentially a read/write to block based storage. | |
6322 | */ | |
6323 | if (!state->plug_started && state->ios_left > 1 && | |
6324 | io_op_defs[req->opcode].plug) { | |
6325 | blk_start_plug(&state->plug); | |
6326 | state->plug_started = true; | |
6327 | } | |
6328 | ||
6329 | if (io_op_defs[req->opcode].needs_file) { | |
6330 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6331 | ||
6332 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6333 | if (unlikely(!req->file)) | |
6334 | ret = -EBADF; | |
6335 | } | |
6336 | ||
6337 | state->ios_left--; | |
6338 | return ret; | |
6339 | } | |
6340 | ||
a6b8cadc | 6341 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6342 | const struct io_uring_sqe *sqe) |
9e645e11 | 6343 | { |
a1ab7b35 | 6344 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6345 | int ret; |
9e645e11 | 6346 | |
a6b8cadc PB |
6347 | ret = io_init_req(ctx, req, sqe); |
6348 | if (unlikely(ret)) { | |
6349 | fail_req: | |
6350 | io_put_req(req); | |
6351 | io_req_complete(req, ret); | |
de59bc10 PB |
6352 | if (link->head) { |
6353 | /* fail even hard links since we don't submit */ | |
cf109604 | 6354 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6355 | io_put_req(link->head); |
6356 | io_req_complete(link->head, -ECANCELED); | |
6357 | link->head = NULL; | |
6358 | } | |
a6b8cadc PB |
6359 | return ret; |
6360 | } | |
be7053b7 PB |
6361 | ret = io_req_prep(req, sqe); |
6362 | if (unlikely(ret)) | |
6363 | goto fail_req; | |
a6b8cadc | 6364 | |
be7053b7 | 6365 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6366 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6367 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6368 | ||
9e645e11 JA |
6369 | /* |
6370 | * If we already have a head request, queue this one for async | |
6371 | * submittal once the head completes. If we don't have a head but | |
6372 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6373 | * submitted sync once the chain is complete. If none of those | |
6374 | * conditions are true (normal request), then just queue it. | |
6375 | */ | |
863e0560 PB |
6376 | if (link->head) { |
6377 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6378 | |
8cdf2193 PB |
6379 | /* |
6380 | * Taking sequential execution of a link, draining both sides | |
6381 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6382 | * requests in the link. So, it drains the head and the | |
6383 | * next after the link request. The last one is done via | |
6384 | * drain_next flag to persist the effect across calls. | |
6385 | */ | |
ef4ff581 | 6386 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6387 | head->flags |= REQ_F_IO_DRAIN; |
6388 | ctx->drain_next = 1; | |
6389 | } | |
be7053b7 | 6390 | ret = io_req_defer_prep(req); |
cf109604 | 6391 | if (unlikely(ret)) |
a6b8cadc | 6392 | goto fail_req; |
9d76377f | 6393 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6394 | link->last->link = req; |
863e0560 | 6395 | link->last = req; |
32fe525b PB |
6396 | |
6397 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6398 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6399 | io_queue_sqe(head); |
863e0560 | 6400 | link->head = NULL; |
32fe525b | 6401 | } |
9e645e11 | 6402 | } else { |
711be031 PB |
6403 | if (unlikely(ctx->drain_next)) { |
6404 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6405 | ctx->drain_next = 0; |
711be031 | 6406 | } |
ef4ff581 | 6407 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6408 | link->head = req; |
6409 | link->last = req; | |
711be031 | 6410 | } else { |
be7053b7 | 6411 | io_queue_sqe(req); |
711be031 | 6412 | } |
9e645e11 | 6413 | } |
2e6e1fde | 6414 | |
1d4240cc | 6415 | return 0; |
9e645e11 JA |
6416 | } |
6417 | ||
9a56a232 JA |
6418 | /* |
6419 | * Batched submission is done, ensure local IO is flushed out. | |
6420 | */ | |
ba88ff11 PB |
6421 | static void io_submit_state_end(struct io_submit_state *state, |
6422 | struct io_ring_ctx *ctx) | |
9a56a232 | 6423 | { |
a1ab7b35 | 6424 | if (state->link.head) |
de59bc10 | 6425 | io_queue_sqe(state->link.head); |
6dd0be1e | 6426 | if (state->comp.nr) |
ba88ff11 | 6427 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6428 | if (state->plug_started) |
6429 | blk_finish_plug(&state->plug); | |
9f13c35b | 6430 | io_state_file_put(state); |
9a56a232 JA |
6431 | } |
6432 | ||
6433 | /* | |
6434 | * Start submission side cache. | |
6435 | */ | |
6436 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6437 | unsigned int max_ios) |
9a56a232 | 6438 | { |
27926b68 | 6439 | state->plug_started = false; |
9a56a232 | 6440 | state->ios_left = max_ios; |
a1ab7b35 PB |
6441 | /* set only head, no need to init link_last in advance */ |
6442 | state->link.head = NULL; | |
9a56a232 JA |
6443 | } |
6444 | ||
2b188cc1 JA |
6445 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6446 | { | |
75b28aff | 6447 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6448 | |
caf582c6 PB |
6449 | /* |
6450 | * Ensure any loads from the SQEs are done at this point, | |
6451 | * since once we write the new head, the application could | |
6452 | * write new data to them. | |
6453 | */ | |
6454 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6455 | } |
6456 | ||
2b188cc1 | 6457 | /* |
3529d8c2 | 6458 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6459 | * that is mapped by userspace. This means that care needs to be taken to |
6460 | * ensure that reads are stable, as we cannot rely on userspace always | |
6461 | * being a good citizen. If members of the sqe are validated and then later | |
6462 | * used, it's important that those reads are done through READ_ONCE() to | |
6463 | * prevent a re-load down the line. | |
6464 | */ | |
709b302f | 6465 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6466 | { |
75b28aff | 6467 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6468 | unsigned head; |
6469 | ||
6470 | /* | |
6471 | * The cached sq head (or cq tail) serves two purposes: | |
6472 | * | |
6473 | * 1) allows us to batch the cost of updating the user visible | |
6474 | * head updates. | |
6475 | * 2) allows the kernel side to track the head on its own, even | |
6476 | * though the application is the one updating it. | |
6477 | */ | |
4fccfcbb | 6478 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6479 | if (likely(head < ctx->sq_entries)) |
6480 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6481 | |
6482 | /* drop invalid entries */ | |
498ccd9e | 6483 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6484 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6485 | return NULL; |
6486 | } | |
6487 | ||
0f212204 | 6488 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6489 | { |
46c4e16a | 6490 | int submitted = 0; |
6c271ce2 | 6491 | |
c4a2ed72 | 6492 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6493 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6494 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6495 | return -EBUSY; |
6496 | } | |
6c271ce2 | 6497 | |
ee7d46d9 PB |
6498 | /* make sure SQ entry isn't read before tail */ |
6499 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6500 | |
2b85edfc PB |
6501 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6502 | return -EAGAIN; | |
6c271ce2 | 6503 | |
d8a6df10 | 6504 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6505 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6506 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6507 | |
46c4e16a | 6508 | while (submitted < nr) { |
3529d8c2 | 6509 | const struct io_uring_sqe *sqe; |
196be95c | 6510 | struct io_kiocb *req; |
fb5ccc98 | 6511 | |
258b29a9 | 6512 | req = io_alloc_req(ctx); |
196be95c PB |
6513 | if (unlikely(!req)) { |
6514 | if (!submitted) | |
6515 | submitted = -EAGAIN; | |
fb5ccc98 | 6516 | break; |
196be95c | 6517 | } |
4fccfcbb PB |
6518 | sqe = io_get_sqe(ctx); |
6519 | if (unlikely(!sqe)) { | |
6520 | kmem_cache_free(req_cachep, req); | |
6521 | break; | |
6522 | } | |
d3656344 JA |
6523 | /* will complete beyond this point, count as submitted */ |
6524 | submitted++; | |
a1ab7b35 | 6525 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6526 | break; |
6c271ce2 JA |
6527 | } |
6528 | ||
9466f437 PB |
6529 | if (unlikely(submitted != nr)) { |
6530 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6531 | struct io_uring_task *tctx = current->io_uring; |
6532 | int unused = nr - ref_used; | |
9466f437 | 6533 | |
d8a6df10 JA |
6534 | percpu_ref_put_many(&ctx->refs, unused); |
6535 | percpu_counter_sub(&tctx->inflight, unused); | |
6536 | put_task_struct_many(current, unused); | |
9466f437 | 6537 | } |
6c271ce2 | 6538 | |
a1ab7b35 | 6539 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6540 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6541 | io_commit_sqring(ctx); | |
6542 | ||
6c271ce2 JA |
6543 | return submitted; |
6544 | } | |
6545 | ||
23b3628e XW |
6546 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6547 | { | |
6548 | /* Tell userspace we may need a wakeup call */ | |
6549 | spin_lock_irq(&ctx->completion_lock); | |
6550 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6551 | spin_unlock_irq(&ctx->completion_lock); | |
6552 | } | |
6553 | ||
6554 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6555 | { | |
6556 | spin_lock_irq(&ctx->completion_lock); | |
6557 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6558 | spin_unlock_irq(&ctx->completion_lock); | |
6559 | } | |
6560 | ||
08369246 | 6561 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6562 | { |
c8d1ba58 | 6563 | unsigned int to_submit; |
bdcd3eab | 6564 | int ret = 0; |
6c271ce2 | 6565 | |
c8d1ba58 | 6566 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6567 | /* if we're handling multiple rings, cap submit size for fairness */ |
6568 | if (cap_entries && to_submit > 8) | |
6569 | to_submit = 8; | |
6570 | ||
906a3c6f | 6571 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6572 | unsigned nr_events = 0; |
a4c0b3de | 6573 | |
c8d1ba58 | 6574 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6575 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6576 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6577 | |
d9d05217 PB |
6578 | if (to_submit && !ctx->sqo_dead && |
6579 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 6580 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6581 | mutex_unlock(&ctx->uring_lock); |
6582 | } | |
6c271ce2 | 6583 | |
90554200 JA |
6584 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6585 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6586 | |
08369246 XW |
6587 | return ret; |
6588 | } | |
6c271ce2 | 6589 | |
08369246 XW |
6590 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6591 | { | |
6592 | struct io_ring_ctx *ctx; | |
6593 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6594 | |
08369246 XW |
6595 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6596 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6597 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6598 | } |
c1edbf5f | 6599 | |
08369246 | 6600 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6601 | } |
6c271ce2 | 6602 | |
69fb2131 JA |
6603 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6604 | { | |
6605 | struct io_ring_ctx *ctx; | |
6606 | ||
6607 | while (!list_empty(&sqd->ctx_new_list)) { | |
6608 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
6609 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
6610 | complete(&ctx->sq_thread_comp); | |
6611 | } | |
08369246 XW |
6612 | |
6613 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
6614 | } |
6615 | ||
37d1e2e3 JA |
6616 | static bool io_sq_thread_should_stop(struct io_sq_data *sqd) |
6617 | { | |
6618 | return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
6619 | } | |
6620 | ||
6621 | static bool io_sq_thread_should_park(struct io_sq_data *sqd) | |
6622 | { | |
6623 | return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
6624 | } | |
6625 | ||
6626 | static void io_sq_thread_parkme(struct io_sq_data *sqd) | |
6627 | { | |
6628 | for (;;) { | |
6629 | /* | |
6630 | * TASK_PARKED is a special state; we must serialize against | |
6631 | * possible pending wakeups to avoid store-store collisions on | |
6632 | * task->state. | |
6633 | * | |
6634 | * Such a collision might possibly result in the task state | |
6635 | * changin from TASK_PARKED and us failing the | |
6636 | * wait_task_inactive() in kthread_park(). | |
6637 | */ | |
6638 | set_special_state(TASK_PARKED); | |
6639 | if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) | |
6640 | break; | |
6641 | ||
6642 | /* | |
6643 | * Thread is going to call schedule(), do not preempt it, | |
6644 | * or the caller of kthread_park() may spend more time in | |
6645 | * wait_task_inactive(). | |
6646 | */ | |
6647 | preempt_disable(); | |
6648 | complete(&sqd->completion); | |
6649 | schedule_preempt_disabled(); | |
6650 | preempt_enable(); | |
6651 | } | |
6652 | __set_current_state(TASK_RUNNING); | |
6653 | } | |
6654 | ||
c8d1ba58 JA |
6655 | static int io_sq_thread(void *data) |
6656 | { | |
69fb2131 JA |
6657 | struct io_sq_data *sqd = data; |
6658 | struct io_ring_ctx *ctx; | |
a0d9205f | 6659 | unsigned long timeout = 0; |
37d1e2e3 | 6660 | char buf[TASK_COMM_LEN]; |
08369246 | 6661 | DEFINE_WAIT(wait); |
6c271ce2 | 6662 | |
37d1e2e3 JA |
6663 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6664 | set_task_comm(current, buf); | |
6665 | sqd->thread = current; | |
6666 | current->pf_io_worker = NULL; | |
6667 | ||
6668 | if (sqd->sq_cpu != -1) | |
6669 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6670 | else | |
6671 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6672 | current->flags |= PF_NO_SETAFFINITY; | |
6673 | ||
6674 | complete(&sqd->completion); | |
6c271ce2 | 6675 | |
37d1e2e3 JA |
6676 | wait_for_completion(&sqd->startup); |
6677 | ||
6678 | while (!io_sq_thread_should_stop(sqd)) { | |
08369246 XW |
6679 | int ret; |
6680 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
6681 | |
6682 | /* | |
69fb2131 | 6683 | * Any changes to the sqd lists are synchronized through the |
37d1e2e3 | 6684 | * thread parking. This synchronizes the thread vs users, |
69fb2131 | 6685 | * the users are synchronized on the sqd->ctx_lock. |
c1edbf5f | 6686 | */ |
37d1e2e3 JA |
6687 | if (io_sq_thread_should_park(sqd)) { |
6688 | io_sq_thread_parkme(sqd); | |
6689 | continue; | |
65b2b213 | 6690 | } |
08369246 | 6691 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 6692 | io_sqd_init_new(sqd); |
08369246 XW |
6693 | timeout = jiffies + sqd->sq_thread_idle; |
6694 | } | |
37d1e2e3 JA |
6695 | if (fatal_signal_pending(current)) |
6696 | break; | |
08369246 | 6697 | sqt_spin = false; |
e95eee2d | 6698 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6699 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
08369246 XW |
6700 | ret = __io_sq_thread(ctx, cap_entries); |
6701 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
6702 | sqt_spin = true; | |
69fb2131 | 6703 | } |
6c271ce2 | 6704 | |
08369246 | 6705 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6706 | io_run_task_work(); |
6707 | cond_resched(); | |
08369246 XW |
6708 | if (sqt_spin) |
6709 | timeout = jiffies + sqd->sq_thread_idle; | |
6710 | continue; | |
6711 | } | |
6712 | ||
08369246 XW |
6713 | needs_sched = true; |
6714 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6715 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6716 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6717 | !list_empty_careful(&ctx->iopoll_list)) { | |
6718 | needs_sched = false; | |
6719 | break; | |
6720 | } | |
6721 | if (io_sqring_entries(ctx)) { | |
6722 | needs_sched = false; | |
6723 | break; | |
6724 | } | |
6725 | } | |
6726 | ||
37d1e2e3 | 6727 | if (needs_sched && !io_sq_thread_should_park(sqd)) { |
69fb2131 JA |
6728 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6729 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6730 | |
69fb2131 | 6731 | schedule(); |
69fb2131 JA |
6732 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6733 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6734 | } |
08369246 XW |
6735 | |
6736 | finish_wait(&sqd->wait, &wait); | |
6737 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
6738 | } |
6739 | ||
37d1e2e3 JA |
6740 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6741 | io_uring_cancel_sqpoll(ctx); | |
06058632 | 6742 | |
37d1e2e3 | 6743 | io_run_task_work(); |
28cea78a | 6744 | |
8629397e JA |
6745 | if (io_sq_thread_should_park(sqd)) |
6746 | io_sq_thread_parkme(sqd); | |
6747 | ||
37d1e2e3 JA |
6748 | /* |
6749 | * Clear thread under lock so that concurrent parks work correctly | |
6750 | */ | |
8629397e | 6751 | complete(&sqd->completion); |
37d1e2e3 JA |
6752 | mutex_lock(&sqd->lock); |
6753 | sqd->thread = NULL; | |
5f3f26f9 JA |
6754 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6755 | ctx->sqo_exec = 1; | |
6756 | io_ring_set_wakeup_flag(ctx); | |
6757 | } | |
06058632 | 6758 | |
37d1e2e3 | 6759 | complete(&sqd->exited); |
e54945ae | 6760 | mutex_unlock(&sqd->lock); |
37d1e2e3 | 6761 | do_exit(0); |
6c271ce2 JA |
6762 | } |
6763 | ||
bda52162 JA |
6764 | struct io_wait_queue { |
6765 | struct wait_queue_entry wq; | |
6766 | struct io_ring_ctx *ctx; | |
6767 | unsigned to_wait; | |
6768 | unsigned nr_timeouts; | |
6769 | }; | |
6770 | ||
6c503150 | 6771 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6772 | { |
6773 | struct io_ring_ctx *ctx = iowq->ctx; | |
6774 | ||
6775 | /* | |
d195a66e | 6776 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6777 | * started waiting. For timeouts, we always want to return to userspace, |
6778 | * regardless of event count. | |
6779 | */ | |
6c503150 | 6780 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6781 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6782 | } | |
6783 | ||
6784 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6785 | int wake_flags, void *key) | |
6786 | { | |
6787 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6788 | wq); | |
6789 | ||
6c503150 PB |
6790 | /* |
6791 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6792 | * the task, and the next invocation will do it. | |
6793 | */ | |
6794 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6795 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6796 | return -1; | |
bda52162 JA |
6797 | } |
6798 | ||
af9c1a44 JA |
6799 | static int io_run_task_work_sig(void) |
6800 | { | |
6801 | if (io_run_task_work()) | |
6802 | return 1; | |
6803 | if (!signal_pending(current)) | |
6804 | return 0; | |
792ee0f6 JA |
6805 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6806 | return -ERESTARTSYS; | |
af9c1a44 JA |
6807 | return -EINTR; |
6808 | } | |
6809 | ||
eeb60b9a PB |
6810 | /* when returns >0, the caller should retry */ |
6811 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6812 | struct io_wait_queue *iowq, | |
6813 | signed long *timeout) | |
6814 | { | |
6815 | int ret; | |
6816 | ||
6817 | /* make sure we run task_work before checking for signals */ | |
6818 | ret = io_run_task_work_sig(); | |
6819 | if (ret || io_should_wake(iowq)) | |
6820 | return ret; | |
6821 | /* let the caller flush overflows, retry */ | |
6822 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6823 | return 1; | |
6824 | ||
6825 | *timeout = schedule_timeout(*timeout); | |
6826 | return !*timeout ? -ETIME : 1; | |
6827 | } | |
6828 | ||
2b188cc1 JA |
6829 | /* |
6830 | * Wait until events become available, if we don't already have some. The | |
6831 | * application must reap them itself, as they reside on the shared cq ring. | |
6832 | */ | |
6833 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6834 | const sigset_t __user *sig, size_t sigsz, |
6835 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6836 | { |
bda52162 JA |
6837 | struct io_wait_queue iowq = { |
6838 | .wq = { | |
6839 | .private = current, | |
6840 | .func = io_wake_function, | |
6841 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6842 | }, | |
6843 | .ctx = ctx, | |
6844 | .to_wait = min_events, | |
6845 | }; | |
75b28aff | 6846 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6847 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6848 | int ret; | |
2b188cc1 | 6849 | |
b41e9852 | 6850 | do { |
6c503150 PB |
6851 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6852 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6853 | return 0; |
4c6e277c | 6854 | if (!io_run_task_work()) |
b41e9852 | 6855 | break; |
b41e9852 | 6856 | } while (1); |
2b188cc1 JA |
6857 | |
6858 | if (sig) { | |
9e75ad5d AB |
6859 | #ifdef CONFIG_COMPAT |
6860 | if (in_compat_syscall()) | |
6861 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6862 | sigsz); |
9e75ad5d AB |
6863 | else |
6864 | #endif | |
b772434b | 6865 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6866 | |
2b188cc1 JA |
6867 | if (ret) |
6868 | return ret; | |
6869 | } | |
6870 | ||
c73ebb68 | 6871 | if (uts) { |
c1d5a224 PB |
6872 | struct timespec64 ts; |
6873 | ||
c73ebb68 HX |
6874 | if (get_timespec64(&ts, uts)) |
6875 | return -EFAULT; | |
6876 | timeout = timespec64_to_jiffies(&ts); | |
6877 | } | |
6878 | ||
bda52162 | 6879 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6880 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6881 | do { |
6c503150 | 6882 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
6883 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6884 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6885 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6886 | finish_wait(&ctx->wait, &iowq.wq); | |
6887 | } while (ret > 0); | |
bda52162 | 6888 | |
b7db41c9 | 6889 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6890 | |
75b28aff | 6891 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6892 | } |
6893 | ||
6b06314c JA |
6894 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6895 | { | |
6896 | #if defined(CONFIG_UNIX) | |
6897 | if (ctx->ring_sock) { | |
6898 | struct sock *sock = ctx->ring_sock->sk; | |
6899 | struct sk_buff *skb; | |
6900 | ||
6901 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6902 | kfree_skb(skb); | |
6903 | } | |
6904 | #else | |
6905 | int i; | |
6906 | ||
65e19f54 JA |
6907 | for (i = 0; i < ctx->nr_user_files; i++) { |
6908 | struct file *file; | |
6909 | ||
6910 | file = io_file_from_index(ctx, i); | |
6911 | if (file) | |
6912 | fput(file); | |
6913 | } | |
6b06314c JA |
6914 | #endif |
6915 | } | |
6916 | ||
00835dce | 6917 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6918 | { |
269bbe5f | 6919 | struct fixed_rsrc_data *data; |
05f3fb3c | 6920 | |
269bbe5f | 6921 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6922 | complete(&data->done); |
6923 | } | |
6924 | ||
2a63b2d9 | 6925 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6926 | { |
2a63b2d9 | 6927 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6928 | } |
6929 | ||
2a63b2d9 | 6930 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6931 | { |
2a63b2d9 BM |
6932 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6933 | } | |
65e19f54 | 6934 | |
d67d2263 BM |
6935 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6936 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6937 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6938 | { |
2a63b2d9 | 6939 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6940 | rsrc_data->node = ref_node; |
d67d2263 | 6941 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6942 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6943 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6944 | } |
6945 | ||
8bad28d8 | 6946 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6947 | { |
8bad28d8 | 6948 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6949 | |
2a63b2d9 | 6950 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6951 | ref_node = data->node; |
e6cb007c | 6952 | data->node = NULL; |
2a63b2d9 | 6953 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6954 | if (ref_node) |
6955 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6956 | } |
6957 | ||
6958 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6959 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6960 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6961 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6962 | { |
f2303b1f | 6963 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6964 | int ret; |
05589553 | 6965 | |
8bad28d8 HX |
6966 | if (data->quiesce) |
6967 | return -ENXIO; | |
05589553 | 6968 | |
8bad28d8 | 6969 | data->quiesce = true; |
1ffc5422 | 6970 | do { |
f2303b1f PB |
6971 | ret = -ENOMEM; |
6972 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
6973 | if (!backup_node) | |
6974 | break; | |
6975 | backup_node->rsrc_data = data; | |
6976 | backup_node->rsrc_put = rsrc_put; | |
6977 | ||
8bad28d8 HX |
6978 | io_sqe_rsrc_kill_node(ctx, data); |
6979 | percpu_ref_kill(&data->refs); | |
6980 | flush_delayed_work(&ctx->rsrc_put_work); | |
6981 | ||
1ffc5422 PB |
6982 | ret = wait_for_completion_interruptible(&data->done); |
6983 | if (!ret) | |
6984 | break; | |
8bad28d8 | 6985 | |
cb5e1b81 | 6986 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
6987 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
6988 | backup_node = NULL; | |
cb5e1b81 | 6989 | reinit_completion(&data->done); |
8bad28d8 | 6990 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 6991 | ret = io_run_task_work_sig(); |
8bad28d8 | 6992 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 6993 | } while (ret >= 0); |
8bad28d8 | 6994 | data->quiesce = false; |
05f3fb3c | 6995 | |
8bad28d8 HX |
6996 | if (backup_node) |
6997 | destroy_fixed_rsrc_ref_node(backup_node); | |
6998 | return ret; | |
d7954b2b BM |
6999 | } |
7000 | ||
1ad555c6 BM |
7001 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7002 | { | |
7003 | struct fixed_rsrc_data *data; | |
7004 | ||
7005 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7006 | if (!data) | |
7007 | return NULL; | |
7008 | ||
00835dce | 7009 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7010 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7011 | kfree(data); | |
7012 | return NULL; | |
7013 | } | |
7014 | data->ctx = ctx; | |
7015 | init_completion(&data->done); | |
7016 | return data; | |
7017 | } | |
7018 | ||
7019 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7020 | { | |
7021 | percpu_ref_exit(&data->refs); | |
7022 | kfree(data->table); | |
7023 | kfree(data); | |
7024 | } | |
7025 | ||
d7954b2b BM |
7026 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7027 | { | |
7028 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7029 | unsigned nr_tables, i; |
7030 | int ret; | |
7031 | ||
8bad28d8 HX |
7032 | /* |
7033 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7034 | * Since we possibly drop uring lock later in this function to | |
7035 | * run task work. | |
7036 | */ | |
7037 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7038 | return -ENXIO; |
f2303b1f | 7039 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7040 | if (ret) |
7041 | return ret; | |
7042 | ||
6b06314c | 7043 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7044 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7045 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7046 | kfree(data->table[i].files); |
1ad555c6 | 7047 | free_fixed_rsrc_data(data); |
05f3fb3c | 7048 | ctx->file_data = NULL; |
6b06314c JA |
7049 | ctx->nr_user_files = 0; |
7050 | return 0; | |
7051 | } | |
7052 | ||
37d1e2e3 JA |
7053 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7054 | __releases(&sqd->lock) | |
7055 | { | |
7056 | if (!sqd->thread) | |
7057 | return; | |
7058 | if (sqd->thread == current) | |
7059 | return; | |
7060 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7061 | wake_up_state(sqd->thread, TASK_PARKED); | |
7062 | mutex_unlock(&sqd->lock); | |
7063 | } | |
7064 | ||
7065 | static bool io_sq_thread_park(struct io_sq_data *sqd) | |
7066 | __acquires(&sqd->lock) | |
7067 | { | |
7068 | if (sqd->thread == current) | |
7069 | return true; | |
7070 | mutex_lock(&sqd->lock); | |
7071 | if (!sqd->thread) { | |
7072 | mutex_unlock(&sqd->lock); | |
7073 | return false; | |
7074 | } | |
7075 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7076 | wake_up_process(sqd->thread); | |
7077 | wait_for_completion(&sqd->completion); | |
7078 | return true; | |
7079 | } | |
7080 | ||
7081 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7082 | { | |
e54945ae | 7083 | if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) |
37d1e2e3 | 7084 | return; |
e54945ae JA |
7085 | mutex_lock(&sqd->lock); |
7086 | if (sqd->thread) { | |
7087 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7088 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)); | |
7089 | wake_up_process(sqd->thread); | |
7090 | mutex_unlock(&sqd->lock); | |
7091 | wait_for_completion(&sqd->exited); | |
7092 | WARN_ON_ONCE(sqd->thread); | |
7093 | } else { | |
7094 | mutex_unlock(&sqd->lock); | |
7095 | } | |
37d1e2e3 JA |
7096 | } |
7097 | ||
534ca6d6 | 7098 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7099 | { |
534ca6d6 | 7100 | if (refcount_dec_and_test(&sqd->refs)) { |
37d1e2e3 JA |
7101 | io_sq_thread_stop(sqd); |
7102 | kfree(sqd); | |
7103 | } | |
7104 | } | |
7105 | ||
7106 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7107 | { | |
7108 | struct io_sq_data *sqd = ctx->sq_data; | |
7109 | ||
7110 | if (sqd) { | |
eb85890b | 7111 | complete(&sqd->startup); |
534ca6d6 | 7112 | if (sqd->thread) { |
37d1e2e3 JA |
7113 | wait_for_completion(&ctx->sq_thread_comp); |
7114 | io_sq_thread_park(sqd); | |
534ca6d6 JA |
7115 | } |
7116 | ||
37d1e2e3 JA |
7117 | mutex_lock(&sqd->ctx_lock); |
7118 | list_del(&ctx->sqd_list); | |
7119 | io_sqd_update_thread_idle(sqd); | |
7120 | mutex_unlock(&sqd->ctx_lock); | |
7121 | ||
7122 | if (sqd->thread) | |
7123 | io_sq_thread_unpark(sqd); | |
7124 | ||
7125 | io_put_sq_data(sqd); | |
7126 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7127 | } |
7128 | } | |
7129 | ||
aa06165d JA |
7130 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7131 | { | |
7132 | struct io_ring_ctx *ctx_attach; | |
7133 | struct io_sq_data *sqd; | |
7134 | struct fd f; | |
7135 | ||
7136 | f = fdget(p->wq_fd); | |
7137 | if (!f.file) | |
7138 | return ERR_PTR(-ENXIO); | |
7139 | if (f.file->f_op != &io_uring_fops) { | |
7140 | fdput(f); | |
7141 | return ERR_PTR(-EINVAL); | |
7142 | } | |
7143 | ||
7144 | ctx_attach = f.file->private_data; | |
7145 | sqd = ctx_attach->sq_data; | |
7146 | if (!sqd) { | |
7147 | fdput(f); | |
7148 | return ERR_PTR(-EINVAL); | |
7149 | } | |
7150 | ||
7151 | refcount_inc(&sqd->refs); | |
7152 | fdput(f); | |
7153 | return sqd; | |
7154 | } | |
7155 | ||
534ca6d6 JA |
7156 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7157 | { | |
7158 | struct io_sq_data *sqd; | |
7159 | ||
aa06165d JA |
7160 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7161 | return io_attach_sq_data(p); | |
7162 | ||
534ca6d6 JA |
7163 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7164 | if (!sqd) | |
7165 | return ERR_PTR(-ENOMEM); | |
7166 | ||
7167 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7168 | INIT_LIST_HEAD(&sqd->ctx_list); |
7169 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7170 | mutex_init(&sqd->ctx_lock); | |
7171 | mutex_init(&sqd->lock); | |
534ca6d6 | 7172 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 JA |
7173 | init_completion(&sqd->startup); |
7174 | init_completion(&sqd->completion); | |
7175 | init_completion(&sqd->exited); | |
534ca6d6 JA |
7176 | return sqd; |
7177 | } | |
7178 | ||
6b06314c | 7179 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7180 | /* |
7181 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7182 | * the io_uring can be safely unregistered on process exit, even if we have | |
7183 | * loops in the file referencing. | |
7184 | */ | |
7185 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7186 | { | |
7187 | struct sock *sk = ctx->ring_sock->sk; | |
7188 | struct scm_fp_list *fpl; | |
7189 | struct sk_buff *skb; | |
08a45173 | 7190 | int i, nr_files; |
6b06314c | 7191 | |
6b06314c JA |
7192 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7193 | if (!fpl) | |
7194 | return -ENOMEM; | |
7195 | ||
7196 | skb = alloc_skb(0, GFP_KERNEL); | |
7197 | if (!skb) { | |
7198 | kfree(fpl); | |
7199 | return -ENOMEM; | |
7200 | } | |
7201 | ||
7202 | skb->sk = sk; | |
6b06314c | 7203 | |
08a45173 | 7204 | nr_files = 0; |
62e398be | 7205 | fpl->user = get_uid(current_user()); |
6b06314c | 7206 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7207 | struct file *file = io_file_from_index(ctx, i + offset); |
7208 | ||
7209 | if (!file) | |
08a45173 | 7210 | continue; |
65e19f54 | 7211 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7212 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7213 | nr_files++; | |
6b06314c JA |
7214 | } |
7215 | ||
08a45173 JA |
7216 | if (nr_files) { |
7217 | fpl->max = SCM_MAX_FD; | |
7218 | fpl->count = nr_files; | |
7219 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7220 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7221 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7222 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7223 | |
08a45173 JA |
7224 | for (i = 0; i < nr_files; i++) |
7225 | fput(fpl->fp[i]); | |
7226 | } else { | |
7227 | kfree_skb(skb); | |
7228 | kfree(fpl); | |
7229 | } | |
6b06314c JA |
7230 | |
7231 | return 0; | |
7232 | } | |
7233 | ||
7234 | /* | |
7235 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7236 | * causes regular reference counting to break down. We rely on the UNIX | |
7237 | * garbage collection to take care of this problem for us. | |
7238 | */ | |
7239 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7240 | { | |
7241 | unsigned left, total; | |
7242 | int ret = 0; | |
7243 | ||
7244 | total = 0; | |
7245 | left = ctx->nr_user_files; | |
7246 | while (left) { | |
7247 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7248 | |
7249 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7250 | if (ret) | |
7251 | break; | |
7252 | left -= this_files; | |
7253 | total += this_files; | |
7254 | } | |
7255 | ||
7256 | if (!ret) | |
7257 | return 0; | |
7258 | ||
7259 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7260 | struct file *file = io_file_from_index(ctx, total); |
7261 | ||
7262 | if (file) | |
7263 | fput(file); | |
6b06314c JA |
7264 | total++; |
7265 | } | |
7266 | ||
7267 | return ret; | |
7268 | } | |
7269 | #else | |
7270 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7271 | { | |
7272 | return 0; | |
7273 | } | |
7274 | #endif | |
7275 | ||
269bbe5f | 7276 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7277 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7278 | { |
7279 | int i; | |
7280 | ||
7281 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7282 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7283 | unsigned this_files; |
7284 | ||
7285 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7286 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7287 | GFP_KERNEL); | |
7288 | if (!table->files) | |
7289 | break; | |
7290 | nr_files -= this_files; | |
7291 | } | |
7292 | ||
7293 | if (i == nr_tables) | |
7294 | return 0; | |
7295 | ||
7296 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7297 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7298 | kfree(table->files); |
7299 | } | |
7300 | return 1; | |
7301 | } | |
7302 | ||
50238531 | 7303 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7304 | { |
50238531 | 7305 | struct file *file = prsrc->file; |
05f3fb3c JA |
7306 | #if defined(CONFIG_UNIX) |
7307 | struct sock *sock = ctx->ring_sock->sk; | |
7308 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7309 | struct sk_buff *skb; | |
7310 | int i; | |
7311 | ||
7312 | __skb_queue_head_init(&list); | |
7313 | ||
7314 | /* | |
7315 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7316 | * remove this entry and rearrange the file array. | |
7317 | */ | |
7318 | skb = skb_dequeue(head); | |
7319 | while (skb) { | |
7320 | struct scm_fp_list *fp; | |
7321 | ||
7322 | fp = UNIXCB(skb).fp; | |
7323 | for (i = 0; i < fp->count; i++) { | |
7324 | int left; | |
7325 | ||
7326 | if (fp->fp[i] != file) | |
7327 | continue; | |
7328 | ||
7329 | unix_notinflight(fp->user, fp->fp[i]); | |
7330 | left = fp->count - 1 - i; | |
7331 | if (left) { | |
7332 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7333 | left * sizeof(struct file *)); | |
7334 | } | |
7335 | fp->count--; | |
7336 | if (!fp->count) { | |
7337 | kfree_skb(skb); | |
7338 | skb = NULL; | |
7339 | } else { | |
7340 | __skb_queue_tail(&list, skb); | |
7341 | } | |
7342 | fput(file); | |
7343 | file = NULL; | |
7344 | break; | |
7345 | } | |
7346 | ||
7347 | if (!file) | |
7348 | break; | |
7349 | ||
7350 | __skb_queue_tail(&list, skb); | |
7351 | ||
7352 | skb = skb_dequeue(head); | |
7353 | } | |
7354 | ||
7355 | if (skb_peek(&list)) { | |
7356 | spin_lock_irq(&head->lock); | |
7357 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7358 | __skb_queue_tail(head, skb); | |
7359 | spin_unlock_irq(&head->lock); | |
7360 | } | |
7361 | #else | |
7362 | fput(file); | |
7363 | #endif | |
7364 | } | |
7365 | ||
269bbe5f | 7366 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7367 | { |
269bbe5f BM |
7368 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7369 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7370 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7371 | |
269bbe5f BM |
7372 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7373 | list_del(&prsrc->list); | |
50238531 | 7374 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7375 | kfree(prsrc); |
65e19f54 | 7376 | } |
05589553 | 7377 | |
05589553 XW |
7378 | percpu_ref_exit(&ref_node->refs); |
7379 | kfree(ref_node); | |
269bbe5f | 7380 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7381 | } |
65e19f54 | 7382 | |
269bbe5f | 7383 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7384 | { |
7385 | struct io_ring_ctx *ctx; | |
7386 | struct llist_node *node; | |
7387 | ||
269bbe5f BM |
7388 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7389 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7390 | |
7391 | while (node) { | |
269bbe5f | 7392 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7393 | struct llist_node *next = node->next; |
7394 | ||
269bbe5f BM |
7395 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7396 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7397 | node = next; |
7398 | } | |
7399 | } | |
7400 | ||
ea64ec02 PB |
7401 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7402 | unsigned i) | |
2faf852d | 7403 | { |
ea64ec02 PB |
7404 | struct fixed_rsrc_table *table; |
7405 | ||
7406 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7407 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7408 | } | |
7409 | ||
00835dce | 7410 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7411 | { |
269bbe5f BM |
7412 | struct fixed_rsrc_ref_node *ref_node; |
7413 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7414 | struct io_ring_ctx *ctx; |
e297822b | 7415 | bool first_add = false; |
4a38aed2 | 7416 | int delay = HZ; |
65e19f54 | 7417 | |
269bbe5f BM |
7418 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7419 | data = ref_node->rsrc_data; | |
e297822b PB |
7420 | ctx = data->ctx; |
7421 | ||
2a63b2d9 | 7422 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7423 | ref_node->done = true; |
7424 | ||
d67d2263 BM |
7425 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7426 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7427 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7428 | /* recycle ref nodes in order */ |
7429 | if (!ref_node->done) | |
7430 | break; | |
7431 | list_del(&ref_node->node); | |
269bbe5f | 7432 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7433 | } |
2a63b2d9 | 7434 | io_rsrc_ref_unlock(ctx); |
05589553 | 7435 | |
e297822b | 7436 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7437 | delay = 0; |
05589553 | 7438 | |
4a38aed2 | 7439 | if (!delay) |
269bbe5f | 7440 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7441 | else if (first_add) |
269bbe5f | 7442 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7443 | } |
65e19f54 | 7444 | |
6802535d | 7445 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7446 | struct io_ring_ctx *ctx) |
05f3fb3c | 7447 | { |
269bbe5f | 7448 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7449 | |
05589553 XW |
7450 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7451 | if (!ref_node) | |
3e2224c5 | 7452 | return NULL; |
05f3fb3c | 7453 | |
00835dce | 7454 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7455 | 0, GFP_KERNEL)) { |
7456 | kfree(ref_node); | |
3e2224c5 | 7457 | return NULL; |
05589553 XW |
7458 | } |
7459 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7460 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7461 | ref_node->done = false; |
05589553 | 7462 | return ref_node; |
05589553 XW |
7463 | } |
7464 | ||
bc9744cd PB |
7465 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7466 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7467 | { |
269bbe5f | 7468 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7469 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7470 | } |
7471 | ||
269bbe5f | 7472 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7473 | { |
7474 | percpu_ref_exit(&ref_node->refs); | |
7475 | kfree(ref_node); | |
65e19f54 JA |
7476 | } |
7477 | ||
ea64ec02 | 7478 | |
6b06314c JA |
7479 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7480 | unsigned nr_args) | |
7481 | { | |
7482 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7483 | unsigned nr_tables, i; |
05f3fb3c | 7484 | struct file *file; |
600cf3f8 | 7485 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7486 | struct fixed_rsrc_ref_node *ref_node; |
7487 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7488 | |
05f3fb3c | 7489 | if (ctx->file_data) |
6b06314c JA |
7490 | return -EBUSY; |
7491 | if (!nr_args) | |
7492 | return -EINVAL; | |
7493 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7494 | return -EMFILE; | |
7495 | ||
1ad555c6 | 7496 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7497 | if (!file_data) |
05f3fb3c | 7498 | return -ENOMEM; |
13770a71 | 7499 | ctx->file_data = file_data; |
05f3fb3c | 7500 | |
65e19f54 | 7501 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7502 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7503 | GFP_KERNEL); |
600cf3f8 PB |
7504 | if (!file_data->table) |
7505 | goto out_free; | |
05f3fb3c | 7506 | |
600cf3f8 | 7507 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7508 | goto out_free; |
65e19f54 | 7509 | |
08a45173 | 7510 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7511 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7512 | ret = -EFAULT; | |
7513 | goto out_fput; | |
7514 | } | |
08a45173 | 7515 | /* allow sparse sets */ |
600cf3f8 | 7516 | if (fd == -1) |
08a45173 | 7517 | continue; |
6b06314c | 7518 | |
05f3fb3c | 7519 | file = fget(fd); |
6b06314c | 7520 | ret = -EBADF; |
05f3fb3c | 7521 | if (!file) |
600cf3f8 | 7522 | goto out_fput; |
05f3fb3c | 7523 | |
6b06314c JA |
7524 | /* |
7525 | * Don't allow io_uring instances to be registered. If UNIX | |
7526 | * isn't enabled, then this causes a reference cycle and this | |
7527 | * instance can never get freed. If UNIX is enabled we'll | |
7528 | * handle it just fine, but there's still no point in allowing | |
7529 | * a ring fd as it doesn't support regular read/write anyway. | |
7530 | */ | |
05f3fb3c JA |
7531 | if (file->f_op == &io_uring_fops) { |
7532 | fput(file); | |
600cf3f8 | 7533 | goto out_fput; |
6b06314c | 7534 | } |
ea64ec02 | 7535 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7536 | } |
7537 | ||
6b06314c | 7538 | ret = io_sqe_files_scm(ctx); |
05589553 | 7539 | if (ret) { |
6b06314c | 7540 | io_sqe_files_unregister(ctx); |
05589553 XW |
7541 | return ret; |
7542 | } | |
6b06314c | 7543 | |
bc9744cd | 7544 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7545 | if (!ref_node) { |
05589553 | 7546 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7547 | return -ENOMEM; |
05589553 | 7548 | } |
bc9744cd | 7549 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7550 | |
d67d2263 | 7551 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7552 | return ret; |
600cf3f8 PB |
7553 | out_fput: |
7554 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7555 | file = io_file_from_index(ctx, i); | |
7556 | if (file) | |
7557 | fput(file); | |
7558 | } | |
7559 | for (i = 0; i < nr_tables; i++) | |
7560 | kfree(file_data->table[i].files); | |
7561 | ctx->nr_user_files = 0; | |
600cf3f8 | 7562 | out_free: |
1ad555c6 | 7563 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7564 | ctx->file_data = NULL; |
6b06314c JA |
7565 | return ret; |
7566 | } | |
7567 | ||
c3a31e60 JA |
7568 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7569 | int index) | |
7570 | { | |
7571 | #if defined(CONFIG_UNIX) | |
7572 | struct sock *sock = ctx->ring_sock->sk; | |
7573 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7574 | struct sk_buff *skb; | |
7575 | ||
7576 | /* | |
7577 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7578 | * file set. If there's no room, fall back to allocating a new skb | |
7579 | * and filling it in. | |
7580 | */ | |
7581 | spin_lock_irq(&head->lock); | |
7582 | skb = skb_peek(head); | |
7583 | if (skb) { | |
7584 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7585 | ||
7586 | if (fpl->count < SCM_MAX_FD) { | |
7587 | __skb_unlink(skb, head); | |
7588 | spin_unlock_irq(&head->lock); | |
7589 | fpl->fp[fpl->count] = get_file(file); | |
7590 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7591 | fpl->count++; | |
7592 | spin_lock_irq(&head->lock); | |
7593 | __skb_queue_head(head, skb); | |
7594 | } else { | |
7595 | skb = NULL; | |
7596 | } | |
7597 | } | |
7598 | spin_unlock_irq(&head->lock); | |
7599 | ||
7600 | if (skb) { | |
7601 | fput(file); | |
7602 | return 0; | |
7603 | } | |
7604 | ||
7605 | return __io_sqe_files_scm(ctx, 1, index); | |
7606 | #else | |
7607 | return 0; | |
7608 | #endif | |
7609 | } | |
7610 | ||
50238531 | 7611 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7612 | { |
269bbe5f BM |
7613 | struct io_rsrc_put *prsrc; |
7614 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7615 | |
269bbe5f BM |
7616 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7617 | if (!prsrc) | |
a5318d3c | 7618 | return -ENOMEM; |
05f3fb3c | 7619 | |
50238531 | 7620 | prsrc->rsrc = rsrc; |
269bbe5f | 7621 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7622 | |
a5318d3c | 7623 | return 0; |
05f3fb3c JA |
7624 | } |
7625 | ||
269bbe5f BM |
7626 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7627 | struct file *file) | |
7628 | { | |
50238531 | 7629 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7630 | } |
7631 | ||
05f3fb3c | 7632 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7633 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7634 | unsigned nr_args) |
7635 | { | |
269bbe5f BM |
7636 | struct fixed_rsrc_data *data = ctx->file_data; |
7637 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7638 | struct file *file, **file_slot; |
c3a31e60 JA |
7639 | __s32 __user *fds; |
7640 | int fd, i, err; | |
7641 | __u32 done; | |
05589553 | 7642 | bool needs_switch = false; |
c3a31e60 | 7643 | |
05f3fb3c | 7644 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7645 | return -EOVERFLOW; |
7646 | if (done > ctx->nr_user_files) | |
7647 | return -EINVAL; | |
7648 | ||
bc9744cd | 7649 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7650 | if (!ref_node) |
7651 | return -ENOMEM; | |
bc9744cd | 7652 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7653 | |
269bbe5f | 7654 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7655 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7656 | err = 0; |
7657 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7658 | err = -EFAULT; | |
7659 | break; | |
7660 | } | |
4e0377a1 | 7661 | if (fd == IORING_REGISTER_FILES_SKIP) |
7662 | continue; | |
7663 | ||
67973b93 | 7664 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7665 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7666 | ||
7667 | if (*file_slot) { | |
7668 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7669 | if (err) |
7670 | break; | |
ea64ec02 | 7671 | *file_slot = NULL; |
05589553 | 7672 | needs_switch = true; |
c3a31e60 JA |
7673 | } |
7674 | if (fd != -1) { | |
c3a31e60 JA |
7675 | file = fget(fd); |
7676 | if (!file) { | |
7677 | err = -EBADF; | |
7678 | break; | |
7679 | } | |
7680 | /* | |
7681 | * Don't allow io_uring instances to be registered. If | |
7682 | * UNIX isn't enabled, then this causes a reference | |
7683 | * cycle and this instance can never get freed. If UNIX | |
7684 | * is enabled we'll handle it just fine, but there's | |
7685 | * still no point in allowing a ring fd as it doesn't | |
7686 | * support regular read/write anyway. | |
7687 | */ | |
7688 | if (file->f_op == &io_uring_fops) { | |
7689 | fput(file); | |
7690 | err = -EBADF; | |
7691 | break; | |
7692 | } | |
e68a3ff8 | 7693 | *file_slot = file; |
c3a31e60 | 7694 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7695 | if (err) { |
e68a3ff8 | 7696 | *file_slot = NULL; |
f3bd9dae | 7697 | fput(file); |
c3a31e60 | 7698 | break; |
f3bd9dae | 7699 | } |
c3a31e60 | 7700 | } |
05f3fb3c JA |
7701 | } |
7702 | ||
05589553 | 7703 | if (needs_switch) { |
b2e96852 | 7704 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7705 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7706 | } else |
269bbe5f | 7707 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7708 | |
7709 | return done ? done : err; | |
7710 | } | |
05589553 | 7711 | |
05f3fb3c JA |
7712 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7713 | unsigned nr_args) | |
7714 | { | |
269bbe5f | 7715 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7716 | |
7717 | if (!ctx->file_data) | |
7718 | return -ENXIO; | |
7719 | if (!nr_args) | |
7720 | return -EINVAL; | |
7721 | if (copy_from_user(&up, arg, sizeof(up))) | |
7722 | return -EFAULT; | |
7723 | if (up.resv) | |
7724 | return -EINVAL; | |
7725 | ||
7726 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7727 | } | |
c3a31e60 | 7728 | |
5280f7e5 | 7729 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7730 | { |
7731 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7732 | ||
5280f7e5 PB |
7733 | req = io_put_req_find_next(req); |
7734 | return req ? &req->work : NULL; | |
7d723065 JA |
7735 | } |
7736 | ||
5aa75ed5 | 7737 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7738 | { |
e941894e | 7739 | struct io_wq_hash *hash; |
24369c2e | 7740 | struct io_wq_data data; |
24369c2e | 7741 | unsigned int concurrency; |
24369c2e | 7742 | |
e941894e JA |
7743 | hash = ctx->hash_map; |
7744 | if (!hash) { | |
7745 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7746 | if (!hash) | |
7747 | return ERR_PTR(-ENOMEM); | |
7748 | refcount_set(&hash->refs, 1); | |
7749 | init_waitqueue_head(&hash->wait); | |
7750 | ctx->hash_map = hash; | |
24369c2e PB |
7751 | } |
7752 | ||
e941894e | 7753 | data.hash = hash; |
e9fd9396 | 7754 | data.free_work = io_free_work; |
f5fa38c5 | 7755 | data.do_work = io_wq_submit_work; |
24369c2e | 7756 | |
d25e3a3d JA |
7757 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7758 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7759 | |
5aa75ed5 | 7760 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7761 | } |
7762 | ||
5aa75ed5 JA |
7763 | static int io_uring_alloc_task_context(struct task_struct *task, |
7764 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7765 | { |
7766 | struct io_uring_task *tctx; | |
d8a6df10 | 7767 | int ret; |
0f212204 JA |
7768 | |
7769 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7770 | if (unlikely(!tctx)) | |
7771 | return -ENOMEM; | |
7772 | ||
d8a6df10 JA |
7773 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7774 | if (unlikely(ret)) { | |
7775 | kfree(tctx); | |
7776 | return ret; | |
7777 | } | |
7778 | ||
5aa75ed5 JA |
7779 | tctx->io_wq = io_init_wq_offload(ctx); |
7780 | if (IS_ERR(tctx->io_wq)) { | |
7781 | ret = PTR_ERR(tctx->io_wq); | |
7782 | percpu_counter_destroy(&tctx->inflight); | |
7783 | kfree(tctx); | |
7784 | return ret; | |
7785 | } | |
7786 | ||
0f212204 JA |
7787 | xa_init(&tctx->xa); |
7788 | init_waitqueue_head(&tctx->wait); | |
7789 | tctx->last = NULL; | |
fdaf083c JA |
7790 | atomic_set(&tctx->in_idle, 0); |
7791 | tctx->sqpoll = false; | |
0f212204 | 7792 | task->io_uring = tctx; |
7cbf1722 JA |
7793 | spin_lock_init(&tctx->task_lock); |
7794 | INIT_WQ_LIST(&tctx->task_list); | |
7795 | tctx->task_state = 0; | |
7796 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7797 | return 0; |
7798 | } | |
7799 | ||
7800 | void __io_uring_free(struct task_struct *tsk) | |
7801 | { | |
7802 | struct io_uring_task *tctx = tsk->io_uring; | |
7803 | ||
7804 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7805 | WARN_ON_ONCE(tctx->io_wq); |
7806 | ||
d8a6df10 | 7807 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7808 | kfree(tctx); |
7809 | tsk->io_uring = NULL; | |
7810 | } | |
7811 | ||
5f3f26f9 JA |
7812 | static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx) |
7813 | { | |
7814 | int ret; | |
7815 | ||
7816 | clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7817 | reinit_completion(&sqd->completion); | |
7818 | ctx->sqo_dead = ctx->sqo_exec = 0; | |
7819 | sqd->task_pid = current->pid; | |
7820 | current->flags |= PF_IO_WORKER; | |
7821 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7822 | current->flags &= ~PF_IO_WORKER; | |
7823 | if (ret < 0) { | |
7824 | sqd->thread = NULL; | |
7825 | return ret; | |
7826 | } | |
7827 | wait_for_completion(&sqd->completion); | |
7828 | return io_uring_alloc_task_context(sqd->thread, ctx); | |
7829 | } | |
7830 | ||
7e84e1c7 SG |
7831 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7832 | struct io_uring_params *p) | |
2b188cc1 JA |
7833 | { |
7834 | int ret; | |
7835 | ||
d25e3a3d JA |
7836 | /* Retain compatibility with failing for an invalid attach attempt */ |
7837 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7838 | IORING_SETUP_ATTACH_WQ) { | |
7839 | struct fd f; | |
7840 | ||
7841 | f = fdget(p->wq_fd); | |
7842 | if (!f.file) | |
7843 | return -ENXIO; | |
7844 | if (f.file->f_op != &io_uring_fops) { | |
7845 | fdput(f); | |
7846 | return -EINVAL; | |
7847 | } | |
7848 | fdput(f); | |
7849 | } | |
6c271ce2 | 7850 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7851 | struct io_sq_data *sqd; |
7852 | ||
3ec482d1 | 7853 | ret = -EPERM; |
ce59fc69 | 7854 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7855 | goto err; |
7856 | ||
534ca6d6 JA |
7857 | sqd = io_get_sq_data(p); |
7858 | if (IS_ERR(sqd)) { | |
7859 | ret = PTR_ERR(sqd); | |
7860 | goto err; | |
7861 | } | |
69fb2131 | 7862 | |
534ca6d6 | 7863 | ctx->sq_data = sqd; |
69fb2131 JA |
7864 | io_sq_thread_park(sqd); |
7865 | mutex_lock(&sqd->ctx_lock); | |
7866 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7867 | mutex_unlock(&sqd->ctx_lock); | |
7868 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7869 | |
917257da JA |
7870 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7871 | if (!ctx->sq_thread_idle) | |
7872 | ctx->sq_thread_idle = HZ; | |
7873 | ||
aa06165d | 7874 | if (sqd->thread) |
5aa75ed5 | 7875 | return 0; |
aa06165d | 7876 | |
6c271ce2 | 7877 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7878 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7879 | |
917257da | 7880 | ret = -EINVAL; |
44a9bd18 JA |
7881 | if (cpu >= nr_cpu_ids) |
7882 | goto err; | |
7889f44d | 7883 | if (!cpu_online(cpu)) |
917257da JA |
7884 | goto err; |
7885 | ||
37d1e2e3 | 7886 | sqd->sq_cpu = cpu; |
6c271ce2 | 7887 | } else { |
37d1e2e3 | 7888 | sqd->sq_cpu = -1; |
6c271ce2 | 7889 | } |
37d1e2e3 JA |
7890 | |
7891 | sqd->task_pid = current->pid; | |
7892 | current->flags |= PF_IO_WORKER; | |
7893 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7894 | current->flags &= ~PF_IO_WORKER; | |
7895 | if (ret < 0) { | |
534ca6d6 | 7896 | sqd->thread = NULL; |
6c271ce2 JA |
7897 | goto err; |
7898 | } | |
37d1e2e3 | 7899 | wait_for_completion(&sqd->completion); |
5aa75ed5 | 7900 | ret = io_uring_alloc_task_context(sqd->thread, ctx); |
0f212204 JA |
7901 | if (ret) |
7902 | goto err; | |
6c271ce2 JA |
7903 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7904 | /* Can't have SQ_AFF without SQPOLL */ | |
7905 | ret = -EINVAL; | |
7906 | goto err; | |
7907 | } | |
7908 | ||
2b188cc1 JA |
7909 | return 0; |
7910 | err: | |
37d1e2e3 | 7911 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
7912 | return ret; |
7913 | } | |
7914 | ||
7e84e1c7 SG |
7915 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7916 | { | |
534ca6d6 JA |
7917 | struct io_sq_data *sqd = ctx->sq_data; |
7918 | ||
eb85890b | 7919 | if (ctx->flags & IORING_SETUP_SQPOLL) |
37d1e2e3 | 7920 | complete(&sqd->startup); |
7e84e1c7 SG |
7921 | } |
7922 | ||
a087e2b5 BM |
7923 | static inline void __io_unaccount_mem(struct user_struct *user, |
7924 | unsigned long nr_pages) | |
2b188cc1 JA |
7925 | { |
7926 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7927 | } | |
7928 | ||
a087e2b5 BM |
7929 | static inline int __io_account_mem(struct user_struct *user, |
7930 | unsigned long nr_pages) | |
2b188cc1 JA |
7931 | { |
7932 | unsigned long page_limit, cur_pages, new_pages; | |
7933 | ||
7934 | /* Don't allow more pages than we can safely lock */ | |
7935 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7936 | ||
7937 | do { | |
7938 | cur_pages = atomic_long_read(&user->locked_vm); | |
7939 | new_pages = cur_pages + nr_pages; | |
7940 | if (new_pages > page_limit) | |
7941 | return -ENOMEM; | |
7942 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7943 | new_pages) != cur_pages); | |
7944 | ||
7945 | return 0; | |
7946 | } | |
7947 | ||
26bfa89e | 7948 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7949 | { |
62e398be | 7950 | if (ctx->user) |
a087e2b5 | 7951 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7952 | |
26bfa89e JA |
7953 | if (ctx->mm_account) |
7954 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7955 | } |
7956 | ||
26bfa89e | 7957 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7958 | { |
30975825 BM |
7959 | int ret; |
7960 | ||
62e398be | 7961 | if (ctx->user) { |
30975825 BM |
7962 | ret = __io_account_mem(ctx->user, nr_pages); |
7963 | if (ret) | |
7964 | return ret; | |
7965 | } | |
7966 | ||
26bfa89e JA |
7967 | if (ctx->mm_account) |
7968 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7969 | |
7970 | return 0; | |
7971 | } | |
7972 | ||
2b188cc1 JA |
7973 | static void io_mem_free(void *ptr) |
7974 | { | |
52e04ef4 MR |
7975 | struct page *page; |
7976 | ||
7977 | if (!ptr) | |
7978 | return; | |
2b188cc1 | 7979 | |
52e04ef4 | 7980 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7981 | if (put_page_testzero(page)) |
7982 | free_compound_page(page); | |
7983 | } | |
7984 | ||
7985 | static void *io_mem_alloc(size_t size) | |
7986 | { | |
7987 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 7988 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
7989 | |
7990 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7991 | } | |
7992 | ||
75b28aff HV |
7993 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7994 | size_t *sq_offset) | |
7995 | { | |
7996 | struct io_rings *rings; | |
7997 | size_t off, sq_array_size; | |
7998 | ||
7999 | off = struct_size(rings, cqes, cq_entries); | |
8000 | if (off == SIZE_MAX) | |
8001 | return SIZE_MAX; | |
8002 | ||
8003 | #ifdef CONFIG_SMP | |
8004 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8005 | if (off == 0) | |
8006 | return SIZE_MAX; | |
8007 | #endif | |
8008 | ||
b36200f5 DV |
8009 | if (sq_offset) |
8010 | *sq_offset = off; | |
8011 | ||
75b28aff HV |
8012 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8013 | if (sq_array_size == SIZE_MAX) | |
8014 | return SIZE_MAX; | |
8015 | ||
8016 | if (check_add_overflow(off, sq_array_size, &off)) | |
8017 | return SIZE_MAX; | |
8018 | ||
75b28aff HV |
8019 | return off; |
8020 | } | |
8021 | ||
0a96bbe4 | 8022 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8023 | { |
8024 | int i, j; | |
8025 | ||
8026 | if (!ctx->user_bufs) | |
8027 | return -ENXIO; | |
8028 | ||
8029 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8030 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8031 | ||
8032 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8033 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8034 | |
de293938 | 8035 | if (imu->acct_pages) |
26bfa89e | 8036 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8037 | kvfree(imu->bvec); |
edafccee JA |
8038 | imu->nr_bvecs = 0; |
8039 | } | |
8040 | ||
8041 | kfree(ctx->user_bufs); | |
8042 | ctx->user_bufs = NULL; | |
8043 | ctx->nr_user_bufs = 0; | |
8044 | return 0; | |
8045 | } | |
8046 | ||
8047 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8048 | void __user *arg, unsigned index) | |
8049 | { | |
8050 | struct iovec __user *src; | |
8051 | ||
8052 | #ifdef CONFIG_COMPAT | |
8053 | if (ctx->compat) { | |
8054 | struct compat_iovec __user *ciovs; | |
8055 | struct compat_iovec ciov; | |
8056 | ||
8057 | ciovs = (struct compat_iovec __user *) arg; | |
8058 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8059 | return -EFAULT; | |
8060 | ||
d55e5f5b | 8061 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8062 | dst->iov_len = ciov.iov_len; |
8063 | return 0; | |
8064 | } | |
8065 | #endif | |
8066 | src = (struct iovec __user *) arg; | |
8067 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8068 | return -EFAULT; | |
8069 | return 0; | |
8070 | } | |
8071 | ||
de293938 JA |
8072 | /* |
8073 | * Not super efficient, but this is just a registration time. And we do cache | |
8074 | * the last compound head, so generally we'll only do a full search if we don't | |
8075 | * match that one. | |
8076 | * | |
8077 | * We check if the given compound head page has already been accounted, to | |
8078 | * avoid double accounting it. This allows us to account the full size of the | |
8079 | * page, not just the constituent pages of a huge page. | |
8080 | */ | |
8081 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8082 | int nr_pages, struct page *hpage) | |
8083 | { | |
8084 | int i, j; | |
8085 | ||
8086 | /* check current page array */ | |
8087 | for (i = 0; i < nr_pages; i++) { | |
8088 | if (!PageCompound(pages[i])) | |
8089 | continue; | |
8090 | if (compound_head(pages[i]) == hpage) | |
8091 | return true; | |
8092 | } | |
8093 | ||
8094 | /* check previously registered pages */ | |
8095 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8096 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8097 | ||
8098 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8099 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8100 | continue; | |
8101 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8102 | return true; | |
8103 | } | |
8104 | } | |
8105 | ||
8106 | return false; | |
8107 | } | |
8108 | ||
8109 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8110 | int nr_pages, struct io_mapped_ubuf *imu, | |
8111 | struct page **last_hpage) | |
8112 | { | |
8113 | int i, ret; | |
8114 | ||
8115 | for (i = 0; i < nr_pages; i++) { | |
8116 | if (!PageCompound(pages[i])) { | |
8117 | imu->acct_pages++; | |
8118 | } else { | |
8119 | struct page *hpage; | |
8120 | ||
8121 | hpage = compound_head(pages[i]); | |
8122 | if (hpage == *last_hpage) | |
8123 | continue; | |
8124 | *last_hpage = hpage; | |
8125 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8126 | continue; | |
8127 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8128 | } | |
8129 | } | |
8130 | ||
8131 | if (!imu->acct_pages) | |
8132 | return 0; | |
8133 | ||
26bfa89e | 8134 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8135 | if (ret) |
8136 | imu->acct_pages = 0; | |
8137 | return ret; | |
8138 | } | |
8139 | ||
0a96bbe4 BM |
8140 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8141 | struct io_mapped_ubuf *imu, | |
8142 | struct page **last_hpage) | |
edafccee JA |
8143 | { |
8144 | struct vm_area_struct **vmas = NULL; | |
8145 | struct page **pages = NULL; | |
0a96bbe4 BM |
8146 | unsigned long off, start, end, ubuf; |
8147 | size_t size; | |
8148 | int ret, pret, nr_pages, i; | |
8149 | ||
8150 | ubuf = (unsigned long) iov->iov_base; | |
8151 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8152 | start = ubuf >> PAGE_SHIFT; | |
8153 | nr_pages = end - start; | |
8154 | ||
8155 | ret = -ENOMEM; | |
8156 | ||
8157 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8158 | if (!pages) | |
8159 | goto done; | |
8160 | ||
8161 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8162 | GFP_KERNEL); | |
8163 | if (!vmas) | |
8164 | goto done; | |
edafccee | 8165 | |
0a96bbe4 BM |
8166 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8167 | GFP_KERNEL); | |
8168 | if (!imu->bvec) | |
8169 | goto done; | |
8170 | ||
8171 | ret = 0; | |
8172 | mmap_read_lock(current->mm); | |
8173 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8174 | pages, vmas); | |
8175 | if (pret == nr_pages) { | |
8176 | /* don't support file backed memory */ | |
8177 | for (i = 0; i < nr_pages; i++) { | |
8178 | struct vm_area_struct *vma = vmas[i]; | |
8179 | ||
8180 | if (vma->vm_file && | |
8181 | !is_file_hugepages(vma->vm_file)) { | |
8182 | ret = -EOPNOTSUPP; | |
8183 | break; | |
8184 | } | |
8185 | } | |
8186 | } else { | |
8187 | ret = pret < 0 ? pret : -EFAULT; | |
8188 | } | |
8189 | mmap_read_unlock(current->mm); | |
8190 | if (ret) { | |
8191 | /* | |
8192 | * if we did partial map, or found file backed vmas, | |
8193 | * release any pages we did get | |
8194 | */ | |
8195 | if (pret > 0) | |
8196 | unpin_user_pages(pages, pret); | |
8197 | kvfree(imu->bvec); | |
8198 | goto done; | |
8199 | } | |
8200 | ||
8201 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8202 | if (ret) { | |
8203 | unpin_user_pages(pages, pret); | |
8204 | kvfree(imu->bvec); | |
8205 | goto done; | |
8206 | } | |
8207 | ||
8208 | off = ubuf & ~PAGE_MASK; | |
8209 | size = iov->iov_len; | |
8210 | for (i = 0; i < nr_pages; i++) { | |
8211 | size_t vec_len; | |
8212 | ||
8213 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8214 | imu->bvec[i].bv_page = pages[i]; | |
8215 | imu->bvec[i].bv_len = vec_len; | |
8216 | imu->bvec[i].bv_offset = off; | |
8217 | off = 0; | |
8218 | size -= vec_len; | |
8219 | } | |
8220 | /* store original address for later verification */ | |
8221 | imu->ubuf = ubuf; | |
8222 | imu->len = iov->iov_len; | |
8223 | imu->nr_bvecs = nr_pages; | |
8224 | ret = 0; | |
8225 | done: | |
8226 | kvfree(pages); | |
8227 | kvfree(vmas); | |
8228 | return ret; | |
8229 | } | |
8230 | ||
2b358604 | 8231 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8232 | { |
edafccee JA |
8233 | if (ctx->user_bufs) |
8234 | return -EBUSY; | |
8235 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8236 | return -EINVAL; | |
8237 | ||
8238 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8239 | GFP_KERNEL); | |
8240 | if (!ctx->user_bufs) | |
8241 | return -ENOMEM; | |
8242 | ||
2b358604 BM |
8243 | return 0; |
8244 | } | |
edafccee | 8245 | |
2b358604 BM |
8246 | static int io_buffer_validate(struct iovec *iov) |
8247 | { | |
8248 | /* | |
8249 | * Don't impose further limits on the size and buffer | |
8250 | * constraints here, we'll -EINVAL later when IO is | |
8251 | * submitted if they are wrong. | |
8252 | */ | |
8253 | if (!iov->iov_base || !iov->iov_len) | |
8254 | return -EFAULT; | |
edafccee | 8255 | |
2b358604 BM |
8256 | /* arbitrary limit, but we need something */ |
8257 | if (iov->iov_len > SZ_1G) | |
8258 | return -EFAULT; | |
edafccee | 8259 | |
2b358604 BM |
8260 | return 0; |
8261 | } | |
edafccee | 8262 | |
2b358604 BM |
8263 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8264 | unsigned int nr_args) | |
8265 | { | |
8266 | int i, ret; | |
8267 | struct iovec iov; | |
8268 | struct page *last_hpage = NULL; | |
edafccee | 8269 | |
2b358604 BM |
8270 | ret = io_buffers_map_alloc(ctx, nr_args); |
8271 | if (ret) | |
8272 | return ret; | |
edafccee | 8273 | |
edafccee JA |
8274 | for (i = 0; i < nr_args; i++) { |
8275 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8276 | |
edafccee JA |
8277 | ret = io_copy_iov(ctx, &iov, arg, i); |
8278 | if (ret) | |
0a96bbe4 | 8279 | break; |
de293938 | 8280 | |
2b358604 BM |
8281 | ret = io_buffer_validate(&iov); |
8282 | if (ret) | |
0a96bbe4 | 8283 | break; |
edafccee | 8284 | |
0a96bbe4 BM |
8285 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8286 | if (ret) | |
8287 | break; | |
edafccee JA |
8288 | |
8289 | ctx->nr_user_bufs++; | |
8290 | } | |
0a96bbe4 BM |
8291 | |
8292 | if (ret) | |
8293 | io_sqe_buffers_unregister(ctx); | |
8294 | ||
edafccee JA |
8295 | return ret; |
8296 | } | |
8297 | ||
9b402849 JA |
8298 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8299 | { | |
8300 | __s32 __user *fds = arg; | |
8301 | int fd; | |
8302 | ||
8303 | if (ctx->cq_ev_fd) | |
8304 | return -EBUSY; | |
8305 | ||
8306 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8307 | return -EFAULT; | |
8308 | ||
8309 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8310 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8311 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8312 | ctx->cq_ev_fd = NULL; | |
8313 | return ret; | |
8314 | } | |
8315 | ||
8316 | return 0; | |
8317 | } | |
8318 | ||
8319 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8320 | { | |
8321 | if (ctx->cq_ev_fd) { | |
8322 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8323 | ctx->cq_ev_fd = NULL; | |
8324 | return 0; | |
8325 | } | |
8326 | ||
8327 | return -ENXIO; | |
8328 | } | |
8329 | ||
5a2e745d JA |
8330 | static int __io_destroy_buffers(int id, void *p, void *data) |
8331 | { | |
8332 | struct io_ring_ctx *ctx = data; | |
8333 | struct io_buffer *buf = p; | |
8334 | ||
067524e9 | 8335 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8336 | return 0; |
8337 | } | |
8338 | ||
8339 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8340 | { | |
8341 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8342 | idr_destroy(&ctx->io_buffer_idr); | |
8343 | } | |
8344 | ||
68e68ee6 | 8345 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8346 | { |
68e68ee6 | 8347 | struct io_kiocb *req, *nxt; |
1b4c351f | 8348 | |
68e68ee6 JA |
8349 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8350 | if (tsk && req->task != tsk) | |
8351 | continue; | |
1b4c351f JA |
8352 | list_del(&req->compl.list); |
8353 | kmem_cache_free(req_cachep, req); | |
8354 | } | |
8355 | } | |
8356 | ||
4010fec4 | 8357 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8358 | { |
bf019da7 | 8359 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8360 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8361 | |
9a4fdbd8 JA |
8362 | mutex_lock(&ctx->uring_lock); |
8363 | ||
8e5c66c4 | 8364 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8365 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8366 | submit_state->reqs); | |
8e5c66c4 PB |
8367 | submit_state->free_reqs = 0; |
8368 | } | |
9a4fdbd8 JA |
8369 | |
8370 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8371 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8372 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8373 | spin_unlock_irq(&ctx->completion_lock); |
8374 | ||
e5547d2c PB |
8375 | io_req_cache_free(&cs->free_list, NULL); |
8376 | ||
9a4fdbd8 JA |
8377 | mutex_unlock(&ctx->uring_lock); |
8378 | } | |
8379 | ||
2b188cc1 JA |
8380 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8381 | { | |
04fc6c80 PB |
8382 | /* |
8383 | * Some may use context even when all refs and requests have been put, | |
8384 | * and they are free to do so while still holding uring_lock, see | |
8385 | * __io_req_task_submit(). Wait for them to finish. | |
8386 | */ | |
8387 | mutex_lock(&ctx->uring_lock); | |
8388 | mutex_unlock(&ctx->uring_lock); | |
8389 | ||
37d1e2e3 | 8390 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8391 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8392 | |
37d1e2e3 | 8393 | if (ctx->mm_account) { |
2aede0e4 JA |
8394 | mmdrop(ctx->mm_account); |
8395 | ctx->mm_account = NULL; | |
30975825 | 8396 | } |
def596e9 | 8397 | |
8bad28d8 | 8398 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8399 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8400 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8401 | io_eventfd_unregister(ctx); |
5a2e745d | 8402 | io_destroy_buffers(ctx); |
41726c9a | 8403 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8404 | |
2b188cc1 | 8405 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8406 | if (ctx->ring_sock) { |
8407 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8408 | sock_release(ctx->ring_sock); |
355e8d26 | 8409 | } |
2b188cc1 JA |
8410 | #endif |
8411 | ||
75b28aff | 8412 | io_mem_free(ctx->rings); |
2b188cc1 | 8413 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8414 | |
8415 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8416 | free_uid(ctx->user); |
4010fec4 | 8417 | io_req_caches_free(ctx); |
e941894e JA |
8418 | if (ctx->hash_map) |
8419 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8420 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8421 | kfree(ctx); |
8422 | } | |
8423 | ||
8424 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8425 | { | |
8426 | struct io_ring_ctx *ctx = file->private_data; | |
8427 | __poll_t mask = 0; | |
8428 | ||
8429 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8430 | /* |
8431 | * synchronizes with barrier from wq_has_sleeper call in | |
8432 | * io_commit_cqring | |
8433 | */ | |
2b188cc1 | 8434 | smp_rmb(); |
90554200 | 8435 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8436 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8437 | |
8438 | /* | |
8439 | * Don't flush cqring overflow list here, just do a simple check. | |
8440 | * Otherwise there could possible be ABBA deadlock: | |
8441 | * CPU0 CPU1 | |
8442 | * ---- ---- | |
8443 | * lock(&ctx->uring_lock); | |
8444 | * lock(&ep->mtx); | |
8445 | * lock(&ctx->uring_lock); | |
8446 | * lock(&ep->mtx); | |
8447 | * | |
8448 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8449 | * pushs them to do the flush. | |
8450 | */ | |
8451 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8452 | mask |= EPOLLIN | EPOLLRDNORM; |
8453 | ||
8454 | return mask; | |
8455 | } | |
8456 | ||
8457 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8458 | { | |
8459 | struct io_ring_ctx *ctx = file->private_data; | |
8460 | ||
8461 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8462 | } | |
8463 | ||
0bead8cd | 8464 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8465 | { |
4379bf8b | 8466 | const struct cred *creds; |
071698e1 | 8467 | |
4379bf8b JA |
8468 | creds = idr_remove(&ctx->personality_idr, id); |
8469 | if (creds) { | |
8470 | put_cred(creds); | |
0bead8cd | 8471 | return 0; |
1e6fa521 | 8472 | } |
0bead8cd YD |
8473 | |
8474 | return -EINVAL; | |
8475 | } | |
8476 | ||
8477 | static int io_remove_personalities(int id, void *p, void *data) | |
8478 | { | |
8479 | struct io_ring_ctx *ctx = data; | |
8480 | ||
8481 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8482 | return 0; |
8483 | } | |
8484 | ||
ba50a036 | 8485 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 JA |
8486 | { |
8487 | struct callback_head *work, *head, *next; | |
ba50a036 | 8488 | bool executed = false; |
7c25c0d1 JA |
8489 | |
8490 | do { | |
8491 | do { | |
8492 | head = NULL; | |
8493 | work = READ_ONCE(ctx->exit_task_work); | |
8494 | } while (cmpxchg(&ctx->exit_task_work, work, head) != work); | |
8495 | ||
8496 | if (!work) | |
8497 | break; | |
8498 | ||
8499 | do { | |
8500 | next = work->next; | |
8501 | work->func(work); | |
8502 | work = next; | |
8503 | cond_resched(); | |
8504 | } while (work); | |
ba50a036 | 8505 | executed = true; |
7c25c0d1 | 8506 | } while (1); |
ba50a036 PB |
8507 | |
8508 | return executed; | |
7c25c0d1 JA |
8509 | } |
8510 | ||
85faa7b8 JA |
8511 | static void io_ring_exit_work(struct work_struct *work) |
8512 | { | |
b2edc0a7 PB |
8513 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8514 | exit_work); | |
85faa7b8 | 8515 | |
56952e91 JA |
8516 | /* |
8517 | * If we're doing polled IO and end up having requests being | |
8518 | * submitted async (out-of-line), then completions can come in while | |
8519 | * we're waiting for refs to drop. We need to reap these manually, | |
8520 | * as nobody else will be looking for them. | |
8521 | */ | |
b2edc0a7 | 8522 | do { |
9936c7c2 | 8523 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8524 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8525 | io_ring_ctx_free(ctx); |
8526 | } | |
8527 | ||
2b188cc1 JA |
8528 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8529 | { | |
8530 | mutex_lock(&ctx->uring_lock); | |
8531 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8532 | |
8533 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8534 | ctx->sqo_dead = 1; | |
8535 | ||
cda286f0 PB |
8536 | /* if force is set, the ring is going away. always drop after that */ |
8537 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8538 | if (ctx->rings) |
6c503150 | 8539 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8540 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8541 | mutex_unlock(&ctx->uring_lock); |
8542 | ||
6b81928d PB |
8543 | io_kill_timeouts(ctx, NULL, NULL); |
8544 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8545 | |
15dff286 | 8546 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8547 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8548 | |
85faa7b8 | 8549 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8550 | /* |
8551 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8552 | * if we're exiting a ton of rings at the same time. It just adds | |
8553 | * noise and overhead, there's no discernable change in runtime | |
8554 | * over using system_wq. | |
8555 | */ | |
8556 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8557 | } |
8558 | ||
8559 | static int io_uring_release(struct inode *inode, struct file *file) | |
8560 | { | |
8561 | struct io_ring_ctx *ctx = file->private_data; | |
8562 | ||
8563 | file->private_data = NULL; | |
8564 | io_ring_ctx_wait_and_kill(ctx); | |
8565 | return 0; | |
8566 | } | |
8567 | ||
f6edbabb PB |
8568 | struct io_task_cancel { |
8569 | struct task_struct *task; | |
8570 | struct files_struct *files; | |
8571 | }; | |
f254ac04 | 8572 | |
f6edbabb | 8573 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8574 | { |
9a472ef7 | 8575 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8576 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8577 | bool ret; |
8578 | ||
f6edbabb | 8579 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8580 | unsigned long flags; |
8581 | struct io_ring_ctx *ctx = req->ctx; | |
8582 | ||
8583 | /* protect against races with linked timeouts */ | |
8584 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8585 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8586 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8587 | } else { | |
f6edbabb | 8588 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8589 | } |
8590 | return ret; | |
b711d4ea JA |
8591 | } |
8592 | ||
b7ddce3c | 8593 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8594 | struct task_struct *task, |
b7ddce3c PB |
8595 | struct files_struct *files) |
8596 | { | |
8597 | struct io_defer_entry *de = NULL; | |
8598 | LIST_HEAD(list); | |
8599 | ||
8600 | spin_lock_irq(&ctx->completion_lock); | |
8601 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8602 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8603 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8604 | break; | |
8605 | } | |
8606 | } | |
8607 | spin_unlock_irq(&ctx->completion_lock); | |
8608 | ||
8609 | while (!list_empty(&list)) { | |
8610 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8611 | list_del_init(&de->list); | |
8612 | req_set_fail_links(de->req); | |
8613 | io_put_req(de->req); | |
8614 | io_req_complete(de->req, -ECANCELED); | |
8615 | kfree(de); | |
8616 | } | |
8617 | } | |
8618 | ||
9936c7c2 PB |
8619 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8620 | struct task_struct *task, | |
8621 | struct files_struct *files) | |
8622 | { | |
8623 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
5aa75ed5 | 8624 | struct io_uring_task *tctx = current->io_uring; |
9936c7c2 PB |
8625 | |
8626 | while (1) { | |
8627 | enum io_wq_cancel cret; | |
8628 | bool ret = false; | |
8629 | ||
5aa75ed5 JA |
8630 | if (tctx && tctx->io_wq) { |
8631 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, | |
9936c7c2 PB |
8632 | &cancel, true); |
8633 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8634 | } | |
8635 | ||
8636 | /* SQPOLL thread does its own polling */ | |
8637 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8638 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8639 | io_iopoll_try_reap_events(ctx); | |
8640 | ret = true; | |
8641 | } | |
8642 | } | |
8643 | ||
8644 | ret |= io_poll_remove_all(ctx, task, files); | |
8645 | ret |= io_kill_timeouts(ctx, task, files); | |
8646 | ret |= io_run_task_work(); | |
ba50a036 | 8647 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8648 | io_cqring_overflow_flush(ctx, true, task, files); |
8649 | if (!ret) | |
8650 | break; | |
8651 | cond_resched(); | |
8652 | } | |
8653 | } | |
8654 | ||
ca70f00b PB |
8655 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8656 | struct task_struct *task, | |
8657 | struct files_struct *files) | |
8658 | { | |
8659 | struct io_kiocb *req; | |
8660 | int cnt = 0; | |
8661 | ||
8662 | spin_lock_irq(&ctx->inflight_lock); | |
8663 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8664 | cnt += io_match_task(req, task, files); | |
8665 | spin_unlock_irq(&ctx->inflight_lock); | |
8666 | return cnt; | |
8667 | } | |
8668 | ||
b52fda00 | 8669 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8670 | struct task_struct *task, |
fcb323cc JA |
8671 | struct files_struct *files) |
8672 | { | |
fcb323cc | 8673 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8674 | DEFINE_WAIT(wait); |
ca70f00b | 8675 | int inflight; |
fcb323cc | 8676 | |
ca70f00b PB |
8677 | inflight = io_uring_count_inflight(ctx, task, files); |
8678 | if (!inflight) | |
fcb323cc | 8679 | break; |
f6edbabb | 8680 | |
9936c7c2 | 8681 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b | 8682 | |
34343786 PB |
8683 | if (ctx->sq_data) |
8684 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
8685 | prepare_to_wait(&task->io_uring->wait, &wait, |
8686 | TASK_UNINTERRUPTIBLE); | |
8687 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8688 | schedule(); | |
c98de08c | 8689 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
8690 | if (ctx->sq_data) |
8691 | io_sq_thread_park(ctx->sq_data); | |
0f212204 | 8692 | } |
0f212204 JA |
8693 | } |
8694 | ||
d9d05217 PB |
8695 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
8696 | { | |
d9d05217 PB |
8697 | mutex_lock(&ctx->uring_lock); |
8698 | ctx->sqo_dead = 1; | |
8699 | mutex_unlock(&ctx->uring_lock); | |
8700 | ||
8701 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
8702 | if (ctx->rings) |
8703 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
8704 | } |
8705 | ||
0f212204 JA |
8706 | /* |
8707 | * We need to iteratively cancel requests, in case a request has dependent | |
8708 | * hard links. These persist even for failure of cancelations, hence keep | |
8709 | * looping until none are found. | |
8710 | */ | |
8711 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8712 | struct files_struct *files) | |
8713 | { | |
8714 | struct task_struct *task = current; | |
37d1e2e3 | 8715 | bool did_park = false; |
0f212204 | 8716 | |
fdaf083c | 8717 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 | 8718 | io_disable_sqo_submit(ctx); |
37d1e2e3 JA |
8719 | did_park = io_sq_thread_park(ctx->sq_data); |
8720 | if (did_park) { | |
8721 | task = ctx->sq_data->thread; | |
8722 | atomic_inc(&task->io_uring->in_idle); | |
8723 | } | |
fdaf083c | 8724 | } |
0f212204 | 8725 | |
df9923f9 | 8726 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 8727 | |
3a7efd1a | 8728 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 8729 | if (!files) |
9936c7c2 | 8730 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c | 8731 | |
37d1e2e3 | 8732 | if (did_park) { |
fdaf083c | 8733 | atomic_dec(&task->io_uring->in_idle); |
fdaf083c JA |
8734 | io_sq_thread_unpark(ctx->sq_data); |
8735 | } | |
0f212204 JA |
8736 | } |
8737 | ||
8738 | /* | |
8739 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8740 | */ | |
fdaf083c | 8741 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 8742 | { |
236434c3 | 8743 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 8744 | int ret; |
236434c3 MWO |
8745 | |
8746 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8747 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8748 | if (unlikely(ret)) |
8749 | return ret; | |
236434c3 | 8750 | tctx = current->io_uring; |
0f212204 | 8751 | } |
236434c3 MWO |
8752 | if (tctx->last != file) { |
8753 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8754 | |
236434c3 | 8755 | if (!old) { |
0f212204 | 8756 | get_file(file); |
a528b04e PB |
8757 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
8758 | file, GFP_KERNEL)); | |
8759 | if (ret) { | |
8760 | fput(file); | |
8761 | return ret; | |
8762 | } | |
ecfc8492 PB |
8763 | |
8764 | /* one and only SQPOLL file note, held by sqo_task */ | |
8765 | WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && | |
8766 | current != ctx->sqo_task); | |
0f212204 | 8767 | } |
236434c3 | 8768 | tctx->last = file; |
0f212204 JA |
8769 | } |
8770 | ||
fdaf083c JA |
8771 | /* |
8772 | * This is race safe in that the task itself is doing this, hence it | |
8773 | * cannot be going through the exit/cancel paths at the same time. | |
8774 | * This cannot be modified while exit/cancel is running. | |
8775 | */ | |
8776 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
8777 | tctx->sqpoll = true; | |
8778 | ||
0f212204 JA |
8779 | return 0; |
8780 | } | |
8781 | ||
8782 | /* | |
8783 | * Remove this io_uring_file -> task mapping. | |
8784 | */ | |
8785 | static void io_uring_del_task_file(struct file *file) | |
8786 | { | |
8787 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8788 | |
8789 | if (tctx->last == file) | |
8790 | tctx->last = NULL; | |
5e2ed8c4 | 8791 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8792 | if (file) |
8793 | fput(file); | |
8794 | } | |
8795 | ||
8452d4a6 | 8796 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e PB |
8797 | { |
8798 | struct file *file; | |
8799 | unsigned long index; | |
8800 | ||
8801 | xa_for_each(&tctx->xa, index, file) | |
8802 | io_uring_del_task_file(file); | |
8452d4a6 PB |
8803 | if (tctx->io_wq) { |
8804 | io_wq_put_and_exit(tctx->io_wq); | |
8805 | tctx->io_wq = NULL; | |
8806 | } | |
de7f1d9e PB |
8807 | } |
8808 | ||
0f212204 JA |
8809 | void __io_uring_files_cancel(struct files_struct *files) |
8810 | { | |
8811 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8812 | struct file *file; |
8813 | unsigned long index; | |
0f212204 JA |
8814 | |
8815 | /* make sure overflow events are dropped */ | |
fdaf083c | 8816 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
8817 | xa_for_each(&tctx->xa, index, file) |
8818 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 8819 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8820 | |
8452d4a6 PB |
8821 | if (files) |
8822 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8823 | } |
8824 | ||
8825 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
8826 | { | |
0e9ddb39 PB |
8827 | return percpu_counter_sum(&tctx->inflight); |
8828 | } | |
fdaf083c | 8829 | |
0e9ddb39 PB |
8830 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8831 | { | |
37d1e2e3 | 8832 | struct io_sq_data *sqd = ctx->sq_data; |
0e9ddb39 PB |
8833 | struct io_uring_task *tctx; |
8834 | s64 inflight; | |
8835 | DEFINE_WAIT(wait); | |
fdaf083c | 8836 | |
37d1e2e3 | 8837 | if (!sqd) |
0e9ddb39 | 8838 | return; |
0e9ddb39 | 8839 | io_disable_sqo_submit(ctx); |
37d1e2e3 JA |
8840 | if (!io_sq_thread_park(sqd)) |
8841 | return; | |
8842 | tctx = ctx->sq_data->thread->io_uring; | |
e54945ae JA |
8843 | /* can happen on fork/alloc failure, just ignore that state */ |
8844 | if (!tctx) { | |
8845 | io_sq_thread_unpark(sqd); | |
8846 | return; | |
8847 | } | |
fdaf083c | 8848 | |
0e9ddb39 PB |
8849 | atomic_inc(&tctx->in_idle); |
8850 | do { | |
8851 | /* read completions before cancelations */ | |
8852 | inflight = tctx_inflight(tctx); | |
8853 | if (!inflight) | |
8854 | break; | |
8855 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 8856 | |
0e9ddb39 PB |
8857 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8858 | /* | |
8859 | * If we've seen completions, retry without waiting. This | |
8860 | * avoids a race where a completion comes in before we did | |
8861 | * prepare_to_wait(). | |
8862 | */ | |
8863 | if (inflight == tctx_inflight(tctx)) | |
8864 | schedule(); | |
8865 | finish_wait(&tctx->wait, &wait); | |
8866 | } while (1); | |
8867 | atomic_dec(&tctx->in_idle); | |
37d1e2e3 | 8868 | io_sq_thread_unpark(sqd); |
0f212204 JA |
8869 | } |
8870 | ||
0f212204 JA |
8871 | /* |
8872 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8873 | * requests. | |
8874 | */ | |
8875 | void __io_uring_task_cancel(void) | |
8876 | { | |
8877 | struct io_uring_task *tctx = current->io_uring; | |
8878 | DEFINE_WAIT(wait); | |
d8a6df10 | 8879 | s64 inflight; |
0f212204 JA |
8880 | |
8881 | /* make sure overflow events are dropped */ | |
fdaf083c | 8882 | atomic_inc(&tctx->in_idle); |
0f212204 | 8883 | |
0b5cd6c3 | 8884 | /* trigger io_disable_sqo_submit() */ |
0e9ddb39 PB |
8885 | if (tctx->sqpoll) { |
8886 | struct file *file; | |
8887 | unsigned long index; | |
8888 | ||
8889 | xa_for_each(&tctx->xa, index, file) | |
8890 | io_uring_cancel_sqpoll(file->private_data); | |
8891 | } | |
0b5cd6c3 | 8892 | |
d8a6df10 | 8893 | do { |
0f212204 | 8894 | /* read completions before cancelations */ |
fdaf083c | 8895 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8896 | if (!inflight) |
8897 | break; | |
0f212204 JA |
8898 | __io_uring_files_cancel(NULL); |
8899 | ||
8900 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8901 | ||
8902 | /* | |
a1bb3cd5 PB |
8903 | * If we've seen completions, retry without waiting. This |
8904 | * avoids a race where a completion comes in before we did | |
8905 | * prepare_to_wait(). | |
0f212204 | 8906 | */ |
a1bb3cd5 PB |
8907 | if (inflight == tctx_inflight(tctx)) |
8908 | schedule(); | |
f57555ed | 8909 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8910 | } while (1); |
0f212204 | 8911 | |
fdaf083c | 8912 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8913 | |
8452d4a6 PB |
8914 | io_uring_clean_tctx(tctx); |
8915 | /* all current's requests should be gone, we can kill tctx */ | |
8916 | __io_uring_free(current); | |
44e728b8 PB |
8917 | } |
8918 | ||
6c5c240e RP |
8919 | static void *io_uring_validate_mmap_request(struct file *file, |
8920 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8921 | { |
2b188cc1 | 8922 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8923 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8924 | struct page *page; |
8925 | void *ptr; | |
8926 | ||
8927 | switch (offset) { | |
8928 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8929 | case IORING_OFF_CQ_RING: |
8930 | ptr = ctx->rings; | |
2b188cc1 JA |
8931 | break; |
8932 | case IORING_OFF_SQES: | |
8933 | ptr = ctx->sq_sqes; | |
8934 | break; | |
2b188cc1 | 8935 | default: |
6c5c240e | 8936 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8937 | } |
8938 | ||
8939 | page = virt_to_head_page(ptr); | |
a50b854e | 8940 | if (sz > page_size(page)) |
6c5c240e RP |
8941 | return ERR_PTR(-EINVAL); |
8942 | ||
8943 | return ptr; | |
8944 | } | |
8945 | ||
8946 | #ifdef CONFIG_MMU | |
8947 | ||
8948 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8949 | { | |
8950 | size_t sz = vma->vm_end - vma->vm_start; | |
8951 | unsigned long pfn; | |
8952 | void *ptr; | |
8953 | ||
8954 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8955 | if (IS_ERR(ptr)) | |
8956 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8957 | |
8958 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8959 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8960 | } | |
8961 | ||
6c5c240e RP |
8962 | #else /* !CONFIG_MMU */ |
8963 | ||
8964 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8965 | { | |
8966 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8967 | } | |
8968 | ||
8969 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8970 | { | |
8971 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8972 | } | |
8973 | ||
8974 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8975 | unsigned long addr, unsigned long len, | |
8976 | unsigned long pgoff, unsigned long flags) | |
8977 | { | |
8978 | void *ptr; | |
8979 | ||
8980 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8981 | if (IS_ERR(ptr)) | |
8982 | return PTR_ERR(ptr); | |
8983 | ||
8984 | return (unsigned long) ptr; | |
8985 | } | |
8986 | ||
8987 | #endif /* !CONFIG_MMU */ | |
8988 | ||
d9d05217 | 8989 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 8990 | { |
d9d05217 | 8991 | int ret = 0; |
90554200 JA |
8992 | DEFINE_WAIT(wait); |
8993 | ||
8994 | do { | |
8995 | if (!io_sqring_full(ctx)) | |
8996 | break; | |
8997 | ||
8998 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
8999 | ||
d9d05217 PB |
9000 | if (unlikely(ctx->sqo_dead)) { |
9001 | ret = -EOWNERDEAD; | |
9002 | goto out; | |
9003 | } | |
9004 | ||
90554200 JA |
9005 | if (!io_sqring_full(ctx)) |
9006 | break; | |
9007 | ||
9008 | schedule(); | |
9009 | } while (!signal_pending(current)); | |
9010 | ||
9011 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9012 | out: |
9013 | return ret; | |
90554200 JA |
9014 | } |
9015 | ||
c73ebb68 HX |
9016 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9017 | struct __kernel_timespec __user **ts, | |
9018 | const sigset_t __user **sig) | |
9019 | { | |
9020 | struct io_uring_getevents_arg arg; | |
9021 | ||
9022 | /* | |
9023 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9024 | * is just a pointer to the sigset_t. | |
9025 | */ | |
9026 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9027 | *sig = (const sigset_t __user *) argp; | |
9028 | *ts = NULL; | |
9029 | return 0; | |
9030 | } | |
9031 | ||
9032 | /* | |
9033 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9034 | * timespec and sigset_t pointers if good. | |
9035 | */ | |
9036 | if (*argsz != sizeof(arg)) | |
9037 | return -EINVAL; | |
9038 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9039 | return -EFAULT; | |
9040 | *sig = u64_to_user_ptr(arg.sigmask); | |
9041 | *argsz = arg.sigmask_sz; | |
9042 | *ts = u64_to_user_ptr(arg.ts); | |
9043 | return 0; | |
9044 | } | |
9045 | ||
2b188cc1 | 9046 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9047 | u32, min_complete, u32, flags, const void __user *, argp, |
9048 | size_t, argsz) | |
2b188cc1 JA |
9049 | { |
9050 | struct io_ring_ctx *ctx; | |
9051 | long ret = -EBADF; | |
9052 | int submitted = 0; | |
9053 | struct fd f; | |
9054 | ||
4c6e277c | 9055 | io_run_task_work(); |
b41e9852 | 9056 | |
90554200 | 9057 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9058 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9059 | return -EINVAL; |
9060 | ||
9061 | f = fdget(fd); | |
9062 | if (!f.file) | |
9063 | return -EBADF; | |
9064 | ||
9065 | ret = -EOPNOTSUPP; | |
9066 | if (f.file->f_op != &io_uring_fops) | |
9067 | goto out_fput; | |
9068 | ||
9069 | ret = -ENXIO; | |
9070 | ctx = f.file->private_data; | |
9071 | if (!percpu_ref_tryget(&ctx->refs)) | |
9072 | goto out_fput; | |
9073 | ||
7e84e1c7 SG |
9074 | ret = -EBADFD; |
9075 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9076 | goto out; | |
9077 | ||
6c271ce2 JA |
9078 | /* |
9079 | * For SQ polling, the thread will do all submissions and completions. | |
9080 | * Just return the requested submit count, and wake the thread if | |
9081 | * we were asked to. | |
9082 | */ | |
b2a9eada | 9083 | ret = 0; |
6c271ce2 | 9084 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9085 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9086 | |
5f3f26f9 JA |
9087 | if (unlikely(ctx->sqo_exec)) { |
9088 | ret = io_sq_thread_fork(ctx->sq_data, ctx); | |
9089 | if (ret) | |
9090 | goto out; | |
9091 | ctx->sqo_exec = 0; | |
9092 | } | |
d9d05217 PB |
9093 | ret = -EOWNERDEAD; |
9094 | if (unlikely(ctx->sqo_dead)) | |
9095 | goto out; | |
6c271ce2 | 9096 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9097 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9098 | if (flags & IORING_ENTER_SQ_WAIT) { |
9099 | ret = io_sqpoll_wait_sq(ctx); | |
9100 | if (ret) | |
9101 | goto out; | |
9102 | } | |
6c271ce2 | 9103 | submitted = to_submit; |
b2a9eada | 9104 | } else if (to_submit) { |
fdaf083c | 9105 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9106 | if (unlikely(ret)) |
9107 | goto out; | |
2b188cc1 | 9108 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9109 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9110 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9111 | |
9112 | if (submitted != to_submit) | |
9113 | goto out; | |
2b188cc1 JA |
9114 | } |
9115 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9116 | const sigset_t __user *sig; |
9117 | struct __kernel_timespec __user *ts; | |
9118 | ||
9119 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9120 | if (unlikely(ret)) | |
9121 | goto out; | |
9122 | ||
2b188cc1 JA |
9123 | min_complete = min(min_complete, ctx->cq_entries); |
9124 | ||
32b2244a XW |
9125 | /* |
9126 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9127 | * space applications don't need to do io completion events | |
9128 | * polling again, they can rely on io_sq_thread to do polling | |
9129 | * work, which can reduce cpu usage and uring_lock contention. | |
9130 | */ | |
9131 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9132 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9133 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9134 | } else { |
c73ebb68 | 9135 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9136 | } |
2b188cc1 JA |
9137 | } |
9138 | ||
7c504e65 | 9139 | out: |
6805b32e | 9140 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9141 | out_fput: |
9142 | fdput(f); | |
9143 | return submitted ? submitted : ret; | |
9144 | } | |
9145 | ||
bebdb65e | 9146 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9147 | static int io_uring_show_cred(int id, void *p, void *data) |
9148 | { | |
4379bf8b | 9149 | const struct cred *cred = p; |
87ce955b JA |
9150 | struct seq_file *m = data; |
9151 | struct user_namespace *uns = seq_user_ns(m); | |
9152 | struct group_info *gi; | |
9153 | kernel_cap_t cap; | |
9154 | unsigned __capi; | |
9155 | int g; | |
9156 | ||
9157 | seq_printf(m, "%5d\n", id); | |
9158 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9159 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9160 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9161 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9162 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9163 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9164 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9165 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9166 | seq_puts(m, "\n\tGroups:\t"); | |
9167 | gi = cred->group_info; | |
9168 | for (g = 0; g < gi->ngroups; g++) { | |
9169 | seq_put_decimal_ull(m, g ? " " : "", | |
9170 | from_kgid_munged(uns, gi->gid[g])); | |
9171 | } | |
9172 | seq_puts(m, "\n\tCapEff:\t"); | |
9173 | cap = cred->cap_effective; | |
9174 | CAP_FOR_EACH_U32(__capi) | |
9175 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9176 | seq_putc(m, '\n'); | |
9177 | return 0; | |
9178 | } | |
9179 | ||
9180 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9181 | { | |
dbbe9c64 | 9182 | struct io_sq_data *sq = NULL; |
fad8e0de | 9183 | bool has_lock; |
87ce955b JA |
9184 | int i; |
9185 | ||
fad8e0de JA |
9186 | /* |
9187 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9188 | * since fdinfo case grabs it in the opposite direction of normal use | |
9189 | * cases. If we fail to get the lock, we just don't iterate any | |
9190 | * structures that could be going away outside the io_uring mutex. | |
9191 | */ | |
9192 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9193 | ||
5f3f26f9 | 9194 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9195 | sq = ctx->sq_data; |
5f3f26f9 JA |
9196 | if (!sq->thread) |
9197 | sq = NULL; | |
9198 | } | |
dbbe9c64 JQ |
9199 | |
9200 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9201 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9202 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9203 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9204 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9205 | |
87ce955b JA |
9206 | if (f) |
9207 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9208 | else | |
9209 | seq_printf(m, "%5u: <none>\n", i); | |
9210 | } | |
9211 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9212 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9213 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9214 | ||
9215 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9216 | (unsigned int) buf->len); | |
9217 | } | |
fad8e0de | 9218 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9219 | seq_printf(m, "Personalities:\n"); |
9220 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9221 | } | |
d7718a9d JA |
9222 | seq_printf(m, "PollList:\n"); |
9223 | spin_lock_irq(&ctx->completion_lock); | |
9224 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9225 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9226 | struct io_kiocb *req; | |
9227 | ||
9228 | hlist_for_each_entry(req, list, hash_node) | |
9229 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9230 | req->task->task_works != NULL); | |
9231 | } | |
9232 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9233 | if (has_lock) |
9234 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9235 | } |
9236 | ||
9237 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9238 | { | |
9239 | struct io_ring_ctx *ctx = f->private_data; | |
9240 | ||
9241 | if (percpu_ref_tryget(&ctx->refs)) { | |
9242 | __io_uring_show_fdinfo(ctx, m); | |
9243 | percpu_ref_put(&ctx->refs); | |
9244 | } | |
9245 | } | |
bebdb65e | 9246 | #endif |
87ce955b | 9247 | |
2b188cc1 JA |
9248 | static const struct file_operations io_uring_fops = { |
9249 | .release = io_uring_release, | |
9250 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9251 | #ifndef CONFIG_MMU |
9252 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9253 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9254 | #endif | |
2b188cc1 JA |
9255 | .poll = io_uring_poll, |
9256 | .fasync = io_uring_fasync, | |
bebdb65e | 9257 | #ifdef CONFIG_PROC_FS |
87ce955b | 9258 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9259 | #endif |
2b188cc1 JA |
9260 | }; |
9261 | ||
9262 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9263 | struct io_uring_params *p) | |
9264 | { | |
75b28aff HV |
9265 | struct io_rings *rings; |
9266 | size_t size, sq_array_offset; | |
2b188cc1 | 9267 | |
bd740481 JA |
9268 | /* make sure these are sane, as we already accounted them */ |
9269 | ctx->sq_entries = p->sq_entries; | |
9270 | ctx->cq_entries = p->cq_entries; | |
9271 | ||
75b28aff HV |
9272 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9273 | if (size == SIZE_MAX) | |
9274 | return -EOVERFLOW; | |
9275 | ||
9276 | rings = io_mem_alloc(size); | |
9277 | if (!rings) | |
2b188cc1 JA |
9278 | return -ENOMEM; |
9279 | ||
75b28aff HV |
9280 | ctx->rings = rings; |
9281 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9282 | rings->sq_ring_mask = p->sq_entries - 1; | |
9283 | rings->cq_ring_mask = p->cq_entries - 1; | |
9284 | rings->sq_ring_entries = p->sq_entries; | |
9285 | rings->cq_ring_entries = p->cq_entries; | |
9286 | ctx->sq_mask = rings->sq_ring_mask; | |
9287 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9288 | |
9289 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9290 | if (size == SIZE_MAX) { |
9291 | io_mem_free(ctx->rings); | |
9292 | ctx->rings = NULL; | |
2b188cc1 | 9293 | return -EOVERFLOW; |
eb065d30 | 9294 | } |
2b188cc1 JA |
9295 | |
9296 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9297 | if (!ctx->sq_sqes) { |
9298 | io_mem_free(ctx->rings); | |
9299 | ctx->rings = NULL; | |
2b188cc1 | 9300 | return -ENOMEM; |
eb065d30 | 9301 | } |
2b188cc1 | 9302 | |
2b188cc1 JA |
9303 | return 0; |
9304 | } | |
9305 | ||
9faadcc8 PB |
9306 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9307 | { | |
9308 | int ret, fd; | |
9309 | ||
9310 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9311 | if (fd < 0) | |
9312 | return fd; | |
9313 | ||
9314 | ret = io_uring_add_task_file(ctx, file); | |
9315 | if (ret) { | |
9316 | put_unused_fd(fd); | |
9317 | return ret; | |
9318 | } | |
9319 | fd_install(fd, file); | |
9320 | return fd; | |
9321 | } | |
9322 | ||
2b188cc1 JA |
9323 | /* |
9324 | * Allocate an anonymous fd, this is what constitutes the application | |
9325 | * visible backing of an io_uring instance. The application mmaps this | |
9326 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9327 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9328 | */ | |
9faadcc8 | 9329 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9330 | { |
9331 | struct file *file; | |
9faadcc8 | 9332 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9333 | int ret; |
9334 | ||
2b188cc1 JA |
9335 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9336 | &ctx->ring_sock); | |
9337 | if (ret) | |
9faadcc8 | 9338 | return ERR_PTR(ret); |
2b188cc1 JA |
9339 | #endif |
9340 | ||
2b188cc1 JA |
9341 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9342 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9343 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9344 | if (IS_ERR(file)) { |
9345 | sock_release(ctx->ring_sock); | |
9346 | ctx->ring_sock = NULL; | |
9347 | } else { | |
9348 | ctx->ring_sock->file = file; | |
0f212204 | 9349 | } |
2b188cc1 | 9350 | #endif |
9faadcc8 | 9351 | return file; |
2b188cc1 JA |
9352 | } |
9353 | ||
7f13657d XW |
9354 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9355 | struct io_uring_params __user *params) | |
2b188cc1 | 9356 | { |
2b188cc1 | 9357 | struct io_ring_ctx *ctx; |
9faadcc8 | 9358 | struct file *file; |
2b188cc1 JA |
9359 | int ret; |
9360 | ||
8110c1a6 | 9361 | if (!entries) |
2b188cc1 | 9362 | return -EINVAL; |
8110c1a6 JA |
9363 | if (entries > IORING_MAX_ENTRIES) { |
9364 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9365 | return -EINVAL; | |
9366 | entries = IORING_MAX_ENTRIES; | |
9367 | } | |
2b188cc1 JA |
9368 | |
9369 | /* | |
9370 | * Use twice as many entries for the CQ ring. It's possible for the | |
9371 | * application to drive a higher depth than the size of the SQ ring, | |
9372 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9373 | * some flexibility in overcommitting a bit. If the application has |
9374 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9375 | * of CQ ring entries manually. | |
2b188cc1 JA |
9376 | */ |
9377 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9378 | if (p->flags & IORING_SETUP_CQSIZE) { |
9379 | /* | |
9380 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9381 | * to a power-of-two, if it isn't already. We do NOT impose | |
9382 | * any cq vs sq ring sizing. | |
9383 | */ | |
eb2667b3 | 9384 | if (!p->cq_entries) |
33a107f0 | 9385 | return -EINVAL; |
8110c1a6 JA |
9386 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9387 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9388 | return -EINVAL; | |
9389 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9390 | } | |
eb2667b3 JQ |
9391 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9392 | if (p->cq_entries < p->sq_entries) | |
9393 | return -EINVAL; | |
33a107f0 JA |
9394 | } else { |
9395 | p->cq_entries = 2 * p->sq_entries; | |
9396 | } | |
2b188cc1 | 9397 | |
2b188cc1 | 9398 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9399 | if (!ctx) |
2b188cc1 | 9400 | return -ENOMEM; |
2b188cc1 | 9401 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9402 | if (!capable(CAP_IPC_LOCK)) |
9403 | ctx->user = get_uid(current_user()); | |
37d1e2e3 | 9404 | ctx->sqo_task = current; |
2aede0e4 JA |
9405 | |
9406 | /* | |
9407 | * This is just grabbed for accounting purposes. When a process exits, | |
9408 | * the mm is exited and dropped before the files, hence we need to hang | |
9409 | * on to this mm purely for the purposes of being able to unaccount | |
9410 | * memory (locked/pinned vm). It's not used for anything else. | |
9411 | */ | |
6b7898eb | 9412 | mmgrab(current->mm); |
2aede0e4 | 9413 | ctx->mm_account = current->mm; |
6b7898eb | 9414 | |
2b188cc1 JA |
9415 | ret = io_allocate_scq_urings(ctx, p); |
9416 | if (ret) | |
9417 | goto err; | |
9418 | ||
7e84e1c7 | 9419 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9420 | if (ret) |
9421 | goto err; | |
9422 | ||
7e84e1c7 SG |
9423 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9424 | io_sq_offload_start(ctx); | |
9425 | ||
2b188cc1 | 9426 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9427 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9428 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9429 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9430 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9431 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9432 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9433 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9434 | |
9435 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9436 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9437 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9438 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9439 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9440 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9441 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9442 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9443 | |
7f13657d XW |
9444 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9445 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9446 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9447 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9448 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9449 | |
9450 | if (copy_to_user(params, p, sizeof(*p))) { | |
9451 | ret = -EFAULT; | |
9452 | goto err; | |
9453 | } | |
d1719f70 | 9454 | |
9faadcc8 PB |
9455 | file = io_uring_get_file(ctx); |
9456 | if (IS_ERR(file)) { | |
9457 | ret = PTR_ERR(file); | |
9458 | goto err; | |
9459 | } | |
9460 | ||
044c1ab3 JA |
9461 | /* |
9462 | * Install ring fd as the very last thing, so we don't risk someone | |
9463 | * having closed it before we finish setup | |
9464 | */ | |
9faadcc8 PB |
9465 | ret = io_uring_install_fd(ctx, file); |
9466 | if (ret < 0) { | |
06585c49 | 9467 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9468 | /* fput will clean it up */ |
9469 | fput(file); | |
9470 | return ret; | |
9471 | } | |
044c1ab3 | 9472 | |
c826bd7a | 9473 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9474 | return ret; |
9475 | err: | |
d9d05217 | 9476 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9477 | io_ring_ctx_wait_and_kill(ctx); |
9478 | return ret; | |
9479 | } | |
9480 | ||
9481 | /* | |
9482 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9483 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9484 | * params structure passed in. | |
9485 | */ | |
9486 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9487 | { | |
9488 | struct io_uring_params p; | |
2b188cc1 JA |
9489 | int i; |
9490 | ||
9491 | if (copy_from_user(&p, params, sizeof(p))) | |
9492 | return -EFAULT; | |
9493 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9494 | if (p.resv[i]) | |
9495 | return -EINVAL; | |
9496 | } | |
9497 | ||
6c271ce2 | 9498 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9499 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9500 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9501 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9502 | return -EINVAL; |
9503 | ||
7f13657d | 9504 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9505 | } |
9506 | ||
9507 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9508 | struct io_uring_params __user *, params) | |
9509 | { | |
9510 | return io_uring_setup(entries, params); | |
9511 | } | |
9512 | ||
66f4af93 JA |
9513 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9514 | { | |
9515 | struct io_uring_probe *p; | |
9516 | size_t size; | |
9517 | int i, ret; | |
9518 | ||
9519 | size = struct_size(p, ops, nr_args); | |
9520 | if (size == SIZE_MAX) | |
9521 | return -EOVERFLOW; | |
9522 | p = kzalloc(size, GFP_KERNEL); | |
9523 | if (!p) | |
9524 | return -ENOMEM; | |
9525 | ||
9526 | ret = -EFAULT; | |
9527 | if (copy_from_user(p, arg, size)) | |
9528 | goto out; | |
9529 | ret = -EINVAL; | |
9530 | if (memchr_inv(p, 0, size)) | |
9531 | goto out; | |
9532 | ||
9533 | p->last_op = IORING_OP_LAST - 1; | |
9534 | if (nr_args > IORING_OP_LAST) | |
9535 | nr_args = IORING_OP_LAST; | |
9536 | ||
9537 | for (i = 0; i < nr_args; i++) { | |
9538 | p->ops[i].op = i; | |
9539 | if (!io_op_defs[i].not_supported) | |
9540 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9541 | } | |
9542 | p->ops_len = i; | |
9543 | ||
9544 | ret = 0; | |
9545 | if (copy_to_user(arg, p, size)) | |
9546 | ret = -EFAULT; | |
9547 | out: | |
9548 | kfree(p); | |
9549 | return ret; | |
9550 | } | |
9551 | ||
071698e1 JA |
9552 | static int io_register_personality(struct io_ring_ctx *ctx) |
9553 | { | |
4379bf8b | 9554 | const struct cred *creds; |
1e6fa521 | 9555 | int ret; |
071698e1 | 9556 | |
4379bf8b | 9557 | creds = get_current_cred(); |
1e6fa521 | 9558 | |
4379bf8b JA |
9559 | ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, |
9560 | USHRT_MAX, GFP_KERNEL); | |
9561 | if (ret < 0) | |
9562 | put_cred(creds); | |
1e6fa521 | 9563 | return ret; |
071698e1 JA |
9564 | } |
9565 | ||
21b55dbc SG |
9566 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9567 | unsigned int nr_args) | |
9568 | { | |
9569 | struct io_uring_restriction *res; | |
9570 | size_t size; | |
9571 | int i, ret; | |
9572 | ||
7e84e1c7 SG |
9573 | /* Restrictions allowed only if rings started disabled */ |
9574 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9575 | return -EBADFD; | |
9576 | ||
21b55dbc | 9577 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9578 | if (ctx->restrictions.registered) |
21b55dbc SG |
9579 | return -EBUSY; |
9580 | ||
9581 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9582 | return -EINVAL; | |
9583 | ||
9584 | size = array_size(nr_args, sizeof(*res)); | |
9585 | if (size == SIZE_MAX) | |
9586 | return -EOVERFLOW; | |
9587 | ||
9588 | res = memdup_user(arg, size); | |
9589 | if (IS_ERR(res)) | |
9590 | return PTR_ERR(res); | |
9591 | ||
9592 | ret = 0; | |
9593 | ||
9594 | for (i = 0; i < nr_args; i++) { | |
9595 | switch (res[i].opcode) { | |
9596 | case IORING_RESTRICTION_REGISTER_OP: | |
9597 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9598 | ret = -EINVAL; | |
9599 | goto out; | |
9600 | } | |
9601 | ||
9602 | __set_bit(res[i].register_op, | |
9603 | ctx->restrictions.register_op); | |
9604 | break; | |
9605 | case IORING_RESTRICTION_SQE_OP: | |
9606 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9607 | ret = -EINVAL; | |
9608 | goto out; | |
9609 | } | |
9610 | ||
9611 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9612 | break; | |
9613 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9614 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9615 | break; | |
9616 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9617 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9618 | break; | |
9619 | default: | |
9620 | ret = -EINVAL; | |
9621 | goto out; | |
9622 | } | |
9623 | } | |
9624 | ||
9625 | out: | |
9626 | /* Reset all restrictions if an error happened */ | |
9627 | if (ret != 0) | |
9628 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9629 | else | |
7e84e1c7 | 9630 | ctx->restrictions.registered = true; |
21b55dbc SG |
9631 | |
9632 | kfree(res); | |
9633 | return ret; | |
9634 | } | |
9635 | ||
7e84e1c7 SG |
9636 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9637 | { | |
9638 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9639 | return -EBADFD; | |
9640 | ||
9641 | if (ctx->restrictions.registered) | |
9642 | ctx->restricted = 1; | |
9643 | ||
9644 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
9645 | ||
9646 | io_sq_offload_start(ctx); | |
9647 | ||
9648 | return 0; | |
9649 | } | |
9650 | ||
071698e1 JA |
9651 | static bool io_register_op_must_quiesce(int op) |
9652 | { | |
9653 | switch (op) { | |
9654 | case IORING_UNREGISTER_FILES: | |
9655 | case IORING_REGISTER_FILES_UPDATE: | |
9656 | case IORING_REGISTER_PROBE: | |
9657 | case IORING_REGISTER_PERSONALITY: | |
9658 | case IORING_UNREGISTER_PERSONALITY: | |
9659 | return false; | |
9660 | default: | |
9661 | return true; | |
9662 | } | |
9663 | } | |
9664 | ||
edafccee JA |
9665 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9666 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9667 | __releases(ctx->uring_lock) |
9668 | __acquires(ctx->uring_lock) | |
edafccee JA |
9669 | { |
9670 | int ret; | |
9671 | ||
35fa71a0 JA |
9672 | /* |
9673 | * We're inside the ring mutex, if the ref is already dying, then | |
9674 | * someone else killed the ctx or is already going through | |
9675 | * io_uring_register(). | |
9676 | */ | |
9677 | if (percpu_ref_is_dying(&ctx->refs)) | |
9678 | return -ENXIO; | |
9679 | ||
071698e1 | 9680 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9681 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9682 | |
05f3fb3c JA |
9683 | /* |
9684 | * Drop uring mutex before waiting for references to exit. If | |
9685 | * another thread is currently inside io_uring_enter() it might | |
9686 | * need to grab the uring_lock to make progress. If we hold it | |
9687 | * here across the drain wait, then we can deadlock. It's safe | |
9688 | * to drop the mutex here, since no new references will come in | |
9689 | * after we've killed the percpu ref. | |
9690 | */ | |
9691 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9692 | do { |
9693 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9694 | if (!ret) | |
9695 | break; | |
ed6930c9 JA |
9696 | ret = io_run_task_work_sig(); |
9697 | if (ret < 0) | |
9698 | break; | |
af9c1a44 JA |
9699 | } while (1); |
9700 | ||
05f3fb3c | 9701 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9702 | |
c150368b JA |
9703 | if (ret) { |
9704 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9705 | goto out_quiesce; |
9706 | } | |
9707 | } | |
9708 | ||
9709 | if (ctx->restricted) { | |
9710 | if (opcode >= IORING_REGISTER_LAST) { | |
9711 | ret = -EINVAL; | |
9712 | goto out; | |
9713 | } | |
9714 | ||
9715 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9716 | ret = -EACCES; | |
c150368b JA |
9717 | goto out; |
9718 | } | |
05f3fb3c | 9719 | } |
edafccee JA |
9720 | |
9721 | switch (opcode) { | |
9722 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9723 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9724 | break; |
9725 | case IORING_UNREGISTER_BUFFERS: | |
9726 | ret = -EINVAL; | |
9727 | if (arg || nr_args) | |
9728 | break; | |
0a96bbe4 | 9729 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9730 | break; |
6b06314c JA |
9731 | case IORING_REGISTER_FILES: |
9732 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9733 | break; | |
9734 | case IORING_UNREGISTER_FILES: | |
9735 | ret = -EINVAL; | |
9736 | if (arg || nr_args) | |
9737 | break; | |
9738 | ret = io_sqe_files_unregister(ctx); | |
9739 | break; | |
c3a31e60 JA |
9740 | case IORING_REGISTER_FILES_UPDATE: |
9741 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9742 | break; | |
9b402849 | 9743 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9744 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9745 | ret = -EINVAL; |
9746 | if (nr_args != 1) | |
9747 | break; | |
9748 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9749 | if (ret) |
9750 | break; | |
9751 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9752 | ctx->eventfd_async = 1; | |
9753 | else | |
9754 | ctx->eventfd_async = 0; | |
9b402849 JA |
9755 | break; |
9756 | case IORING_UNREGISTER_EVENTFD: | |
9757 | ret = -EINVAL; | |
9758 | if (arg || nr_args) | |
9759 | break; | |
9760 | ret = io_eventfd_unregister(ctx); | |
9761 | break; | |
66f4af93 JA |
9762 | case IORING_REGISTER_PROBE: |
9763 | ret = -EINVAL; | |
9764 | if (!arg || nr_args > 256) | |
9765 | break; | |
9766 | ret = io_probe(ctx, arg, nr_args); | |
9767 | break; | |
071698e1 JA |
9768 | case IORING_REGISTER_PERSONALITY: |
9769 | ret = -EINVAL; | |
9770 | if (arg || nr_args) | |
9771 | break; | |
9772 | ret = io_register_personality(ctx); | |
9773 | break; | |
9774 | case IORING_UNREGISTER_PERSONALITY: | |
9775 | ret = -EINVAL; | |
9776 | if (arg) | |
9777 | break; | |
9778 | ret = io_unregister_personality(ctx, nr_args); | |
9779 | break; | |
7e84e1c7 SG |
9780 | case IORING_REGISTER_ENABLE_RINGS: |
9781 | ret = -EINVAL; | |
9782 | if (arg || nr_args) | |
9783 | break; | |
9784 | ret = io_register_enable_rings(ctx); | |
9785 | break; | |
21b55dbc SG |
9786 | case IORING_REGISTER_RESTRICTIONS: |
9787 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9788 | break; | |
edafccee JA |
9789 | default: |
9790 | ret = -EINVAL; | |
9791 | break; | |
9792 | } | |
9793 | ||
21b55dbc | 9794 | out: |
071698e1 | 9795 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9796 | /* bring the ctx back to life */ |
05f3fb3c | 9797 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9798 | out_quiesce: |
0f158b4c | 9799 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9800 | } |
edafccee JA |
9801 | return ret; |
9802 | } | |
9803 | ||
9804 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9805 | void __user *, arg, unsigned int, nr_args) | |
9806 | { | |
9807 | struct io_ring_ctx *ctx; | |
9808 | long ret = -EBADF; | |
9809 | struct fd f; | |
9810 | ||
9811 | f = fdget(fd); | |
9812 | if (!f.file) | |
9813 | return -EBADF; | |
9814 | ||
9815 | ret = -EOPNOTSUPP; | |
9816 | if (f.file->f_op != &io_uring_fops) | |
9817 | goto out_fput; | |
9818 | ||
9819 | ctx = f.file->private_data; | |
9820 | ||
b6c23dd5 PB |
9821 | io_run_task_work(); |
9822 | ||
edafccee JA |
9823 | mutex_lock(&ctx->uring_lock); |
9824 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9825 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9826 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9827 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9828 | out_fput: |
9829 | fdput(f); | |
9830 | return ret; | |
9831 | } | |
9832 | ||
2b188cc1 JA |
9833 | static int __init io_uring_init(void) |
9834 | { | |
d7f62e82 SM |
9835 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9836 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9837 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9838 | } while (0) | |
9839 | ||
9840 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9841 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9842 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9843 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9844 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9845 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9846 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9847 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9848 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9849 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9850 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9851 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9852 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9853 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9854 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9855 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9856 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9857 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9858 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9859 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9860 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9861 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9862 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9863 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9864 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9865 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9866 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9867 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9868 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9869 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9870 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9871 | |
d3656344 | 9872 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9873 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9874 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9875 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9876 | return 0; |
9877 | }; | |
9878 | __initcall(io_uring_init); |