Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
e4b4a13f | 81 | #include <linux/freezer.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
101 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
102 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 103 | |
b16fed66 PB |
104 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
105 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
106 | IOSQE_BUFFER_SELECT) | |
107 | ||
2b188cc1 JA |
108 | struct io_uring { |
109 | u32 head ____cacheline_aligned_in_smp; | |
110 | u32 tail ____cacheline_aligned_in_smp; | |
111 | }; | |
112 | ||
1e84b97b | 113 | /* |
75b28aff HV |
114 | * This data is shared with the application through the mmap at offsets |
115 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
116 | * |
117 | * The offsets to the member fields are published through struct | |
118 | * io_sqring_offsets when calling io_uring_setup. | |
119 | */ | |
75b28aff | 120 | struct io_rings { |
1e84b97b SB |
121 | /* |
122 | * Head and tail offsets into the ring; the offsets need to be | |
123 | * masked to get valid indices. | |
124 | * | |
75b28aff HV |
125 | * The kernel controls head of the sq ring and the tail of the cq ring, |
126 | * and the application controls tail of the sq ring and the head of the | |
127 | * cq ring. | |
1e84b97b | 128 | */ |
75b28aff | 129 | struct io_uring sq, cq; |
1e84b97b | 130 | /* |
75b28aff | 131 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
132 | * ring_entries - 1) |
133 | */ | |
75b28aff HV |
134 | u32 sq_ring_mask, cq_ring_mask; |
135 | /* Ring sizes (constant, power of 2) */ | |
136 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
137 | /* |
138 | * Number of invalid entries dropped by the kernel due to | |
139 | * invalid index stored in array | |
140 | * | |
141 | * Written by the kernel, shouldn't be modified by the | |
142 | * application (i.e. get number of "new events" by comparing to | |
143 | * cached value). | |
144 | * | |
145 | * After a new SQ head value was read by the application this | |
146 | * counter includes all submissions that were dropped reaching | |
147 | * the new SQ head (and possibly more). | |
148 | */ | |
75b28aff | 149 | u32 sq_dropped; |
1e84b97b | 150 | /* |
0d9b5b3a | 151 | * Runtime SQ flags |
1e84b97b SB |
152 | * |
153 | * Written by the kernel, shouldn't be modified by the | |
154 | * application. | |
155 | * | |
156 | * The application needs a full memory barrier before checking | |
157 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
158 | */ | |
75b28aff | 159 | u32 sq_flags; |
0d9b5b3a SG |
160 | /* |
161 | * Runtime CQ flags | |
162 | * | |
163 | * Written by the application, shouldn't be modified by the | |
164 | * kernel. | |
165 | */ | |
166 | u32 cq_flags; | |
1e84b97b SB |
167 | /* |
168 | * Number of completion events lost because the queue was full; | |
169 | * this should be avoided by the application by making sure | |
0b4295b5 | 170 | * there are not more requests pending than there is space in |
1e84b97b SB |
171 | * the completion queue. |
172 | * | |
173 | * Written by the kernel, shouldn't be modified by the | |
174 | * application (i.e. get number of "new events" by comparing to | |
175 | * cached value). | |
176 | * | |
177 | * As completion events come in out of order this counter is not | |
178 | * ordered with any other data. | |
179 | */ | |
75b28aff | 180 | u32 cq_overflow; |
1e84b97b SB |
181 | /* |
182 | * Ring buffer of completion events. | |
183 | * | |
184 | * The kernel writes completion events fresh every time they are | |
185 | * produced, so the application is allowed to modify pending | |
186 | * entries. | |
187 | */ | |
75b28aff | 188 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
189 | }; |
190 | ||
45d189c6 PB |
191 | enum io_uring_cmd_flags { |
192 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 193 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
194 | }; |
195 | ||
edafccee JA |
196 | struct io_mapped_ubuf { |
197 | u64 ubuf; | |
198 | size_t len; | |
199 | struct bio_vec *bvec; | |
200 | unsigned int nr_bvecs; | |
de293938 | 201 | unsigned long acct_pages; |
edafccee JA |
202 | }; |
203 | ||
50238531 BM |
204 | struct io_ring_ctx; |
205 | ||
269bbe5f BM |
206 | struct io_rsrc_put { |
207 | struct list_head list; | |
50238531 BM |
208 | union { |
209 | void *rsrc; | |
210 | struct file *file; | |
211 | }; | |
269bbe5f BM |
212 | }; |
213 | ||
214 | struct fixed_rsrc_table { | |
65e19f54 | 215 | struct file **files; |
31b51510 JA |
216 | }; |
217 | ||
269bbe5f | 218 | struct fixed_rsrc_ref_node { |
05589553 XW |
219 | struct percpu_ref refs; |
220 | struct list_head node; | |
269bbe5f BM |
221 | struct list_head rsrc_list; |
222 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
223 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
224 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 225 | struct llist_node llist; |
e297822b | 226 | bool done; |
05589553 XW |
227 | }; |
228 | ||
269bbe5f BM |
229 | struct fixed_rsrc_data { |
230 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
231 | struct io_ring_ctx *ctx; |
232 | ||
269bbe5f | 233 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 234 | struct percpu_ref refs; |
05f3fb3c | 235 | struct completion done; |
8bad28d8 | 236 | bool quiesce; |
05f3fb3c JA |
237 | }; |
238 | ||
5a2e745d JA |
239 | struct io_buffer { |
240 | struct list_head list; | |
241 | __u64 addr; | |
242 | __s32 len; | |
243 | __u16 bid; | |
244 | }; | |
245 | ||
21b55dbc SG |
246 | struct io_restriction { |
247 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
248 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
249 | u8 sqe_flags_allowed; | |
250 | u8 sqe_flags_required; | |
7e84e1c7 | 251 | bool registered; |
21b55dbc SG |
252 | }; |
253 | ||
37d1e2e3 JA |
254 | enum { |
255 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
256 | IO_SQ_THREAD_SHOULD_PARK, | |
257 | }; | |
258 | ||
534ca6d6 JA |
259 | struct io_sq_data { |
260 | refcount_t refs; | |
09a6f4ef | 261 | struct mutex lock; |
69fb2131 JA |
262 | |
263 | /* ctx's that are using this sqd */ | |
264 | struct list_head ctx_list; | |
69fb2131 | 265 | |
534ca6d6 JA |
266 | struct task_struct *thread; |
267 | struct wait_queue_head wait; | |
08369246 XW |
268 | |
269 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
270 | int sq_cpu; |
271 | pid_t task_pid; | |
5c2469e0 | 272 | pid_t task_tgid; |
37d1e2e3 JA |
273 | |
274 | unsigned long state; | |
37d1e2e3 | 275 | struct completion exited; |
534ca6d6 JA |
276 | }; |
277 | ||
258b29a9 | 278 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 279 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 280 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 281 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
282 | |
283 | struct io_comp_state { | |
6dd0be1e | 284 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 285 | unsigned int nr; |
c7dae4ba JA |
286 | unsigned int locked_free_nr; |
287 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 288 | struct list_head free_list; |
c7dae4ba JA |
289 | /* IRQ completion list, under ->completion_lock */ |
290 | struct list_head locked_free_list; | |
258b29a9 PB |
291 | }; |
292 | ||
a1ab7b35 PB |
293 | struct io_submit_link { |
294 | struct io_kiocb *head; | |
295 | struct io_kiocb *last; | |
296 | }; | |
297 | ||
258b29a9 PB |
298 | struct io_submit_state { |
299 | struct blk_plug plug; | |
a1ab7b35 | 300 | struct io_submit_link link; |
258b29a9 PB |
301 | |
302 | /* | |
303 | * io_kiocb alloc cache | |
304 | */ | |
bf019da7 | 305 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
306 | unsigned int free_reqs; |
307 | ||
308 | bool plug_started; | |
309 | ||
310 | /* | |
311 | * Batch completion logic | |
312 | */ | |
313 | struct io_comp_state comp; | |
314 | ||
315 | /* | |
316 | * File reference cache | |
317 | */ | |
318 | struct file *file; | |
319 | unsigned int fd; | |
320 | unsigned int file_refs; | |
321 | unsigned int ios_left; | |
322 | }; | |
323 | ||
2b188cc1 JA |
324 | struct io_ring_ctx { |
325 | struct { | |
326 | struct percpu_ref refs; | |
327 | } ____cacheline_aligned_in_smp; | |
328 | ||
329 | struct { | |
330 | unsigned int flags; | |
e1d85334 | 331 | unsigned int compat: 1; |
e1d85334 RD |
332 | unsigned int cq_overflow_flushed: 1; |
333 | unsigned int drain_next: 1; | |
334 | unsigned int eventfd_async: 1; | |
21b55dbc | 335 | unsigned int restricted: 1; |
2b188cc1 | 336 | |
75b28aff HV |
337 | /* |
338 | * Ring buffer of indices into array of io_uring_sqe, which is | |
339 | * mmapped by the application using the IORING_OFF_SQES offset. | |
340 | * | |
341 | * This indirection could e.g. be used to assign fixed | |
342 | * io_uring_sqe entries to operations and only submit them to | |
343 | * the queue when needed. | |
344 | * | |
345 | * The kernel modifies neither the indices array nor the entries | |
346 | * array. | |
347 | */ | |
348 | u32 *sq_array; | |
2b188cc1 JA |
349 | unsigned cached_sq_head; |
350 | unsigned sq_entries; | |
351 | unsigned sq_mask; | |
6c271ce2 | 352 | unsigned sq_thread_idle; |
498ccd9e | 353 | unsigned cached_sq_dropped; |
2c3bac6d | 354 | unsigned cached_cq_overflow; |
ad3eb2c8 | 355 | unsigned long sq_check_overflow; |
de0617e4 | 356 | |
e941894e JA |
357 | /* hashed buffered write serialization */ |
358 | struct io_wq_hash *hash_map; | |
359 | ||
de0617e4 | 360 | struct list_head defer_list; |
5262f567 | 361 | struct list_head timeout_list; |
1d7bb1d5 | 362 | struct list_head cq_overflow_list; |
fcb323cc | 363 | |
ad3eb2c8 | 364 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
365 | } ____cacheline_aligned_in_smp; |
366 | ||
3c1a2ead JA |
367 | struct { |
368 | struct mutex uring_lock; | |
369 | wait_queue_head_t wait; | |
370 | } ____cacheline_aligned_in_smp; | |
371 | ||
372 | struct io_submit_state submit_state; | |
373 | ||
206aefde JA |
374 | struct io_rings *rings; |
375 | ||
2aede0e4 JA |
376 | /* Only used for accounting purposes */ |
377 | struct mm_struct *mm_account; | |
378 | ||
7c30f36a | 379 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
380 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
381 | ||
90554200 | 382 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 383 | struct list_head sqd_list; |
75b28aff | 384 | |
6b06314c JA |
385 | /* |
386 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
387 | * readers must ensure that ->refs is alive as long as the file* is | |
388 | * used. Only updated through io_uring_register(2). | |
389 | */ | |
269bbe5f | 390 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
391 | unsigned nr_user_files; |
392 | ||
edafccee JA |
393 | /* if used, fixed mapped user buffers */ |
394 | unsigned nr_user_bufs; | |
395 | struct io_mapped_ubuf *user_bufs; | |
396 | ||
2b188cc1 JA |
397 | struct user_struct *user; |
398 | ||
0f158b4c | 399 | struct completion ref_comp; |
206aefde JA |
400 | |
401 | #if defined(CONFIG_UNIX) | |
402 | struct socket *ring_sock; | |
403 | #endif | |
404 | ||
9e15c3a0 | 405 | struct xarray io_buffers; |
5a2e745d | 406 | |
61cf9370 MWO |
407 | struct xarray personalities; |
408 | u32 pers_next; | |
071698e1 | 409 | |
206aefde JA |
410 | struct { |
411 | unsigned cached_cq_tail; | |
412 | unsigned cq_entries; | |
413 | unsigned cq_mask; | |
414 | atomic_t cq_timeouts; | |
f010505b | 415 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 416 | unsigned long cq_check_overflow; |
206aefde JA |
417 | struct wait_queue_head cq_wait; |
418 | struct fasync_struct *cq_fasync; | |
419 | struct eventfd_ctx *cq_ev_fd; | |
420 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 421 | |
2b188cc1 JA |
422 | struct { |
423 | spinlock_t completion_lock; | |
e94f141b | 424 | |
def596e9 | 425 | /* |
540e32a0 | 426 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
427 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
428 | * For SQPOLL, only the single threaded io_sq_thread() will | |
429 | * manipulate the list, hence no extra locking is needed there. | |
430 | */ | |
540e32a0 | 431 | struct list_head iopoll_list; |
78076bb6 JA |
432 | struct hlist_head *cancel_hash; |
433 | unsigned cancel_hash_bits; | |
e94f141b | 434 | bool poll_multi_file; |
31b51510 | 435 | |
fcb323cc JA |
436 | spinlock_t inflight_lock; |
437 | struct list_head inflight_list; | |
2b188cc1 | 438 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 439 | |
269bbe5f BM |
440 | struct delayed_work rsrc_put_work; |
441 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
442 | struct list_head rsrc_ref_list; |
443 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 444 | |
21b55dbc | 445 | struct io_restriction restrictions; |
3c1a2ead | 446 | |
7c25c0d1 JA |
447 | /* exit task_work */ |
448 | struct callback_head *exit_task_work; | |
449 | ||
e941894e JA |
450 | struct wait_queue_head hash_wait; |
451 | ||
3c1a2ead JA |
452 | /* Keep this last, we don't need it for the fast path */ |
453 | struct work_struct exit_work; | |
13bf43f5 | 454 | struct list_head tctx_list; |
2b188cc1 JA |
455 | }; |
456 | ||
09bb8394 JA |
457 | /* |
458 | * First field must be the file pointer in all the | |
459 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
460 | */ | |
221c5eb2 JA |
461 | struct io_poll_iocb { |
462 | struct file *file; | |
018043be | 463 | struct wait_queue_head *head; |
221c5eb2 | 464 | __poll_t events; |
8c838788 | 465 | bool done; |
221c5eb2 | 466 | bool canceled; |
392edb45 | 467 | struct wait_queue_entry wait; |
221c5eb2 JA |
468 | }; |
469 | ||
018043be PB |
470 | struct io_poll_remove { |
471 | struct file *file; | |
472 | u64 addr; | |
473 | }; | |
474 | ||
b5dba59e JA |
475 | struct io_close { |
476 | struct file *file; | |
b5dba59e JA |
477 | int fd; |
478 | }; | |
479 | ||
ad8a48ac JA |
480 | struct io_timeout_data { |
481 | struct io_kiocb *req; | |
482 | struct hrtimer timer; | |
483 | struct timespec64 ts; | |
484 | enum hrtimer_mode mode; | |
485 | }; | |
486 | ||
8ed8d3c3 JA |
487 | struct io_accept { |
488 | struct file *file; | |
489 | struct sockaddr __user *addr; | |
490 | int __user *addr_len; | |
491 | int flags; | |
09952e3e | 492 | unsigned long nofile; |
8ed8d3c3 JA |
493 | }; |
494 | ||
495 | struct io_sync { | |
496 | struct file *file; | |
497 | loff_t len; | |
498 | loff_t off; | |
499 | int flags; | |
d63d1b5e | 500 | int mode; |
8ed8d3c3 JA |
501 | }; |
502 | ||
fbf23849 JA |
503 | struct io_cancel { |
504 | struct file *file; | |
505 | u64 addr; | |
506 | }; | |
507 | ||
b29472ee JA |
508 | struct io_timeout { |
509 | struct file *file; | |
bfe68a22 PB |
510 | u32 off; |
511 | u32 target_seq; | |
135fcde8 | 512 | struct list_head list; |
90cd7e42 PB |
513 | /* head of the link, used by linked timeouts only */ |
514 | struct io_kiocb *head; | |
b29472ee JA |
515 | }; |
516 | ||
0bdf7a2d PB |
517 | struct io_timeout_rem { |
518 | struct file *file; | |
519 | u64 addr; | |
9c8e11b3 PB |
520 | |
521 | /* timeout update */ | |
522 | struct timespec64 ts; | |
523 | u32 flags; | |
0bdf7a2d PB |
524 | }; |
525 | ||
9adbd45d JA |
526 | struct io_rw { |
527 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
528 | struct kiocb kiocb; | |
529 | u64 addr; | |
530 | u64 len; | |
531 | }; | |
532 | ||
3fbb51c1 JA |
533 | struct io_connect { |
534 | struct file *file; | |
535 | struct sockaddr __user *addr; | |
536 | int addr_len; | |
537 | }; | |
538 | ||
e47293fd JA |
539 | struct io_sr_msg { |
540 | struct file *file; | |
fddaface | 541 | union { |
270a5940 | 542 | struct user_msghdr __user *umsg; |
fddaface JA |
543 | void __user *buf; |
544 | }; | |
e47293fd | 545 | int msg_flags; |
bcda7baa | 546 | int bgid; |
fddaface | 547 | size_t len; |
bcda7baa | 548 | struct io_buffer *kbuf; |
e47293fd JA |
549 | }; |
550 | ||
15b71abe JA |
551 | struct io_open { |
552 | struct file *file; | |
553 | int dfd; | |
15b71abe | 554 | struct filename *filename; |
c12cedf2 | 555 | struct open_how how; |
4022e7af | 556 | unsigned long nofile; |
15b71abe JA |
557 | }; |
558 | ||
269bbe5f | 559 | struct io_rsrc_update { |
05f3fb3c JA |
560 | struct file *file; |
561 | u64 arg; | |
562 | u32 nr_args; | |
563 | u32 offset; | |
564 | }; | |
565 | ||
4840e418 JA |
566 | struct io_fadvise { |
567 | struct file *file; | |
568 | u64 offset; | |
569 | u32 len; | |
570 | u32 advice; | |
571 | }; | |
572 | ||
c1ca757b JA |
573 | struct io_madvise { |
574 | struct file *file; | |
575 | u64 addr; | |
576 | u32 len; | |
577 | u32 advice; | |
578 | }; | |
579 | ||
3e4827b0 JA |
580 | struct io_epoll { |
581 | struct file *file; | |
582 | int epfd; | |
583 | int op; | |
584 | int fd; | |
585 | struct epoll_event event; | |
e47293fd JA |
586 | }; |
587 | ||
7d67af2c PB |
588 | struct io_splice { |
589 | struct file *file_out; | |
590 | struct file *file_in; | |
591 | loff_t off_out; | |
592 | loff_t off_in; | |
593 | u64 len; | |
594 | unsigned int flags; | |
595 | }; | |
596 | ||
ddf0322d JA |
597 | struct io_provide_buf { |
598 | struct file *file; | |
599 | __u64 addr; | |
600 | __s32 len; | |
601 | __u32 bgid; | |
602 | __u16 nbufs; | |
603 | __u16 bid; | |
604 | }; | |
605 | ||
1d9e1288 BM |
606 | struct io_statx { |
607 | struct file *file; | |
608 | int dfd; | |
609 | unsigned int mask; | |
610 | unsigned int flags; | |
e62753e4 | 611 | const char __user *filename; |
1d9e1288 BM |
612 | struct statx __user *buffer; |
613 | }; | |
614 | ||
36f4fa68 JA |
615 | struct io_shutdown { |
616 | struct file *file; | |
617 | int how; | |
618 | }; | |
619 | ||
80a261fd JA |
620 | struct io_rename { |
621 | struct file *file; | |
622 | int old_dfd; | |
623 | int new_dfd; | |
624 | struct filename *oldpath; | |
625 | struct filename *newpath; | |
626 | int flags; | |
627 | }; | |
628 | ||
14a1143b JA |
629 | struct io_unlink { |
630 | struct file *file; | |
631 | int dfd; | |
632 | int flags; | |
633 | struct filename *filename; | |
634 | }; | |
635 | ||
3ca405eb PB |
636 | struct io_completion { |
637 | struct file *file; | |
638 | struct list_head list; | |
0f7e466b | 639 | int cflags; |
3ca405eb PB |
640 | }; |
641 | ||
f499a021 JA |
642 | struct io_async_connect { |
643 | struct sockaddr_storage address; | |
644 | }; | |
645 | ||
03b1230c JA |
646 | struct io_async_msghdr { |
647 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
648 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
649 | struct iovec *free_iov; | |
03b1230c JA |
650 | struct sockaddr __user *uaddr; |
651 | struct msghdr msg; | |
b537916c | 652 | struct sockaddr_storage addr; |
03b1230c JA |
653 | }; |
654 | ||
f67676d1 JA |
655 | struct io_async_rw { |
656 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
657 | const struct iovec *free_iovec; |
658 | struct iov_iter iter; | |
227c0c96 | 659 | size_t bytes_done; |
bcf5a063 | 660 | struct wait_page_queue wpq; |
f67676d1 JA |
661 | }; |
662 | ||
6b47ee6e PB |
663 | enum { |
664 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
665 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
666 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
667 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
668 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 669 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 670 | |
6b47ee6e PB |
671 | REQ_F_FAIL_LINK_BIT, |
672 | REQ_F_INFLIGHT_BIT, | |
673 | REQ_F_CUR_POS_BIT, | |
674 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 675 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 676 | REQ_F_ISREG_BIT, |
99bc4c38 | 677 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 678 | REQ_F_POLLED_BIT, |
bcda7baa | 679 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 680 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 681 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 682 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
683 | |
684 | /* not a real bit, just to check we're not overflowing the space */ | |
685 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
686 | }; |
687 | ||
688 | enum { | |
689 | /* ctx owns file */ | |
690 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
691 | /* drain existing IO first */ | |
692 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
693 | /* linked sqes */ | |
694 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
695 | /* doesn't sever on completion < 0 */ | |
696 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
697 | /* IOSQE_ASYNC */ | |
698 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
699 | /* IOSQE_BUFFER_SELECT */ |
700 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 701 | |
6b47ee6e PB |
702 | /* fail rest of links */ |
703 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 704 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
705 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
706 | /* read/write uses file position */ | |
707 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
708 | /* must not punt to workers */ | |
709 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 710 | /* has or had linked timeout */ |
6b47ee6e | 711 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
712 | /* regular file */ |
713 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
714 | /* needs cleanup */ |
715 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
716 | /* already went through poll handler */ |
717 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
718 | /* buffer already selected */ |
719 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
720 | /* doesn't need file table for this request */ |
721 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
722 | /* linked timeout is active, i.e. prepared by link's head */ |
723 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
724 | /* completion is deferred through io_comp_state */ |
725 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
726 | }; |
727 | ||
728 | struct async_poll { | |
729 | struct io_poll_iocb poll; | |
807abcb0 | 730 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
731 | }; |
732 | ||
7cbf1722 JA |
733 | struct io_task_work { |
734 | struct io_wq_work_node node; | |
735 | task_work_func_t func; | |
736 | }; | |
737 | ||
09bb8394 JA |
738 | /* |
739 | * NOTE! Each of the iocb union members has the file pointer | |
740 | * as the first entry in their struct definition. So you can | |
741 | * access the file pointer through any of the sub-structs, | |
742 | * or directly as just 'ki_filp' in this struct. | |
743 | */ | |
2b188cc1 | 744 | struct io_kiocb { |
221c5eb2 | 745 | union { |
09bb8394 | 746 | struct file *file; |
9adbd45d | 747 | struct io_rw rw; |
221c5eb2 | 748 | struct io_poll_iocb poll; |
018043be | 749 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
750 | struct io_accept accept; |
751 | struct io_sync sync; | |
fbf23849 | 752 | struct io_cancel cancel; |
b29472ee | 753 | struct io_timeout timeout; |
0bdf7a2d | 754 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 755 | struct io_connect connect; |
e47293fd | 756 | struct io_sr_msg sr_msg; |
15b71abe | 757 | struct io_open open; |
b5dba59e | 758 | struct io_close close; |
269bbe5f | 759 | struct io_rsrc_update rsrc_update; |
4840e418 | 760 | struct io_fadvise fadvise; |
c1ca757b | 761 | struct io_madvise madvise; |
3e4827b0 | 762 | struct io_epoll epoll; |
7d67af2c | 763 | struct io_splice splice; |
ddf0322d | 764 | struct io_provide_buf pbuf; |
1d9e1288 | 765 | struct io_statx statx; |
36f4fa68 | 766 | struct io_shutdown shutdown; |
80a261fd | 767 | struct io_rename rename; |
14a1143b | 768 | struct io_unlink unlink; |
3ca405eb PB |
769 | /* use only after cleaning per-op data, see io_clean_op() */ |
770 | struct io_completion compl; | |
221c5eb2 | 771 | }; |
2b188cc1 | 772 | |
e8c2bc1f JA |
773 | /* opcode allocated if it needs to store data for async defer */ |
774 | void *async_data; | |
d625c6ee | 775 | u8 opcode; |
65a6543d XW |
776 | /* polled IO has completed */ |
777 | u8 iopoll_completed; | |
2b188cc1 | 778 | |
4f4eeba8 | 779 | u16 buf_index; |
9cf7c104 | 780 | u32 result; |
4f4eeba8 | 781 | |
010e8e6b PB |
782 | struct io_ring_ctx *ctx; |
783 | unsigned int flags; | |
784 | refcount_t refs; | |
785 | struct task_struct *task; | |
786 | u64 user_data; | |
d7718a9d | 787 | |
f2f87370 | 788 | struct io_kiocb *link; |
269bbe5f | 789 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 790 | |
d21ffe7e PB |
791 | /* |
792 | * 1. used with ctx->iopoll_list with reads/writes | |
793 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
794 | */ | |
010e8e6b | 795 | struct list_head inflight_entry; |
7cbf1722 JA |
796 | union { |
797 | struct io_task_work io_task_work; | |
798 | struct callback_head task_work; | |
799 | }; | |
010e8e6b PB |
800 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
801 | struct hlist_node hash_node; | |
802 | struct async_poll *apoll; | |
803 | struct io_wq_work work; | |
2b188cc1 | 804 | }; |
05589553 | 805 | |
13bf43f5 PB |
806 | struct io_tctx_node { |
807 | struct list_head ctx_node; | |
808 | struct task_struct *task; | |
13bf43f5 PB |
809 | struct io_ring_ctx *ctx; |
810 | }; | |
811 | ||
27dc8338 PB |
812 | struct io_defer_entry { |
813 | struct list_head list; | |
814 | struct io_kiocb *req; | |
9cf7c104 | 815 | u32 seq; |
2b188cc1 JA |
816 | }; |
817 | ||
d3656344 | 818 | struct io_op_def { |
d3656344 JA |
819 | /* needs req->file assigned */ |
820 | unsigned needs_file : 1; | |
d3656344 JA |
821 | /* hash wq insertion if file is a regular file */ |
822 | unsigned hash_reg_file : 1; | |
823 | /* unbound wq insertion if file is a non-regular file */ | |
824 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
825 | /* opcode is not supported by this kernel */ |
826 | unsigned not_supported : 1; | |
8a72758c JA |
827 | /* set if opcode supports polled "wait" */ |
828 | unsigned pollin : 1; | |
829 | unsigned pollout : 1; | |
bcda7baa JA |
830 | /* op supports buffer selection */ |
831 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
832 | /* must always have async data allocated */ |
833 | unsigned needs_async_data : 1; | |
27926b68 JA |
834 | /* should block plug */ |
835 | unsigned plug : 1; | |
e8c2bc1f JA |
836 | /* size of async data needed, if any */ |
837 | unsigned short async_size; | |
d3656344 JA |
838 | }; |
839 | ||
0918682b | 840 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
841 | [IORING_OP_NOP] = {}, |
842 | [IORING_OP_READV] = { | |
d3656344 JA |
843 | .needs_file = 1, |
844 | .unbound_nonreg_file = 1, | |
8a72758c | 845 | .pollin = 1, |
4d954c25 | 846 | .buffer_select = 1, |
e8c2bc1f | 847 | .needs_async_data = 1, |
27926b68 | 848 | .plug = 1, |
e8c2bc1f | 849 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 850 | }, |
0463b6c5 | 851 | [IORING_OP_WRITEV] = { |
d3656344 JA |
852 | .needs_file = 1, |
853 | .hash_reg_file = 1, | |
854 | .unbound_nonreg_file = 1, | |
8a72758c | 855 | .pollout = 1, |
e8c2bc1f | 856 | .needs_async_data = 1, |
27926b68 | 857 | .plug = 1, |
e8c2bc1f | 858 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 859 | }, |
0463b6c5 | 860 | [IORING_OP_FSYNC] = { |
d3656344 JA |
861 | .needs_file = 1, |
862 | }, | |
0463b6c5 | 863 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
864 | .needs_file = 1, |
865 | .unbound_nonreg_file = 1, | |
8a72758c | 866 | .pollin = 1, |
27926b68 | 867 | .plug = 1, |
e8c2bc1f | 868 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 869 | }, |
0463b6c5 | 870 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
871 | .needs_file = 1, |
872 | .hash_reg_file = 1, | |
873 | .unbound_nonreg_file = 1, | |
8a72758c | 874 | .pollout = 1, |
27926b68 | 875 | .plug = 1, |
e8c2bc1f | 876 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | .unbound_nonreg_file = 1, | |
881 | }, | |
0463b6c5 PB |
882 | [IORING_OP_POLL_REMOVE] = {}, |
883 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
884 | .needs_file = 1, |
885 | }, | |
0463b6c5 | 886 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
887 | .needs_file = 1, |
888 | .unbound_nonreg_file = 1, | |
8a72758c | 889 | .pollout = 1, |
e8c2bc1f JA |
890 | .needs_async_data = 1, |
891 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 892 | }, |
0463b6c5 | 893 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
894 | .needs_file = 1, |
895 | .unbound_nonreg_file = 1, | |
8a72758c | 896 | .pollin = 1, |
52de1fe1 | 897 | .buffer_select = 1, |
e8c2bc1f JA |
898 | .needs_async_data = 1, |
899 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 900 | }, |
0463b6c5 | 901 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
902 | .needs_async_data = 1, |
903 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 904 | }, |
9c8e11b3 PB |
905 | [IORING_OP_TIMEOUT_REMOVE] = { |
906 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 907 | }, |
0463b6c5 | 908 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
909 | .needs_file = 1, |
910 | .unbound_nonreg_file = 1, | |
8a72758c | 911 | .pollin = 1, |
d3656344 | 912 | }, |
0463b6c5 PB |
913 | [IORING_OP_ASYNC_CANCEL] = {}, |
914 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
915 | .needs_async_data = 1, |
916 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 917 | }, |
0463b6c5 | 918 | [IORING_OP_CONNECT] = { |
d3656344 JA |
919 | .needs_file = 1, |
920 | .unbound_nonreg_file = 1, | |
8a72758c | 921 | .pollout = 1, |
e8c2bc1f JA |
922 | .needs_async_data = 1, |
923 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 924 | }, |
0463b6c5 | 925 | [IORING_OP_FALLOCATE] = { |
d3656344 | 926 | .needs_file = 1, |
d3656344 | 927 | }, |
44526bed JA |
928 | [IORING_OP_OPENAT] = {}, |
929 | [IORING_OP_CLOSE] = {}, | |
930 | [IORING_OP_FILES_UPDATE] = {}, | |
931 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 932 | [IORING_OP_READ] = { |
3a6820f2 JA |
933 | .needs_file = 1, |
934 | .unbound_nonreg_file = 1, | |
8a72758c | 935 | .pollin = 1, |
bcda7baa | 936 | .buffer_select = 1, |
27926b68 | 937 | .plug = 1, |
e8c2bc1f | 938 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 939 | }, |
0463b6c5 | 940 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
941 | .needs_file = 1, |
942 | .unbound_nonreg_file = 1, | |
8a72758c | 943 | .pollout = 1, |
27926b68 | 944 | .plug = 1, |
e8c2bc1f | 945 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 946 | }, |
0463b6c5 | 947 | [IORING_OP_FADVISE] = { |
4840e418 | 948 | .needs_file = 1, |
c1ca757b | 949 | }, |
44526bed | 950 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 951 | [IORING_OP_SEND] = { |
fddaface JA |
952 | .needs_file = 1, |
953 | .unbound_nonreg_file = 1, | |
8a72758c | 954 | .pollout = 1, |
fddaface | 955 | }, |
0463b6c5 | 956 | [IORING_OP_RECV] = { |
fddaface JA |
957 | .needs_file = 1, |
958 | .unbound_nonreg_file = 1, | |
8a72758c | 959 | .pollin = 1, |
bcda7baa | 960 | .buffer_select = 1, |
fddaface | 961 | }, |
0463b6c5 | 962 | [IORING_OP_OPENAT2] = { |
cebdb986 | 963 | }, |
3e4827b0 JA |
964 | [IORING_OP_EPOLL_CTL] = { |
965 | .unbound_nonreg_file = 1, | |
3e4827b0 | 966 | }, |
7d67af2c PB |
967 | [IORING_OP_SPLICE] = { |
968 | .needs_file = 1, | |
969 | .hash_reg_file = 1, | |
970 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
971 | }, |
972 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 973 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
974 | [IORING_OP_TEE] = { |
975 | .needs_file = 1, | |
976 | .hash_reg_file = 1, | |
977 | .unbound_nonreg_file = 1, | |
978 | }, | |
36f4fa68 JA |
979 | [IORING_OP_SHUTDOWN] = { |
980 | .needs_file = 1, | |
981 | }, | |
44526bed JA |
982 | [IORING_OP_RENAMEAT] = {}, |
983 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
984 | }; |
985 | ||
7a612350 | 986 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 987 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
988 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
989 | struct task_struct *task, | |
990 | struct files_struct *files); | |
37d1e2e3 | 991 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 992 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 993 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 994 | struct io_ring_ctx *ctx); |
f2303b1f | 995 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 996 | |
23faba36 | 997 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 998 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 999 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1000 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1001 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1002 | static void io_dismantle_req(struct io_kiocb *req); |
1003 | static void io_put_task(struct task_struct *task, int nr); | |
1004 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1005 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1006 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1007 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1008 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1009 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1010 | unsigned nr_args); |
3ca405eb | 1011 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1012 | static struct file *io_file_get(struct io_submit_state *state, |
1013 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1014 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1015 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1016 | |
847595de PB |
1017 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1018 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1019 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1020 | const struct iovec *fast_iov, | |
227c0c96 | 1021 | struct iov_iter *iter, bool force); |
907d1df3 | 1022 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1023 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1024 | struct io_ring_ctx *ctx); | |
de0617e4 | 1025 | |
2b188cc1 JA |
1026 | static struct kmem_cache *req_cachep; |
1027 | ||
0918682b | 1028 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1029 | |
1030 | struct sock *io_uring_get_socket(struct file *file) | |
1031 | { | |
1032 | #if defined(CONFIG_UNIX) | |
1033 | if (file->f_op == &io_uring_fops) { | |
1034 | struct io_ring_ctx *ctx = file->private_data; | |
1035 | ||
1036 | return ctx->ring_sock->sk; | |
1037 | } | |
1038 | #endif | |
1039 | return NULL; | |
1040 | } | |
1041 | EXPORT_SYMBOL(io_uring_get_socket); | |
1042 | ||
f2f87370 PB |
1043 | #define io_for_each_link(pos, head) \ |
1044 | for (pos = (head); pos; pos = pos->link) | |
1045 | ||
3ca405eb PB |
1046 | static inline void io_clean_op(struct io_kiocb *req) |
1047 | { | |
9d5c8190 | 1048 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1049 | __io_clean_op(req); |
1050 | } | |
1051 | ||
36f72fe2 PB |
1052 | static inline void io_set_resource_node(struct io_kiocb *req) |
1053 | { | |
1054 | struct io_ring_ctx *ctx = req->ctx; | |
1055 | ||
269bbe5f BM |
1056 | if (!req->fixed_rsrc_refs) { |
1057 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1058 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1059 | } |
1060 | } | |
1061 | ||
08d23634 PB |
1062 | static bool io_match_task(struct io_kiocb *head, |
1063 | struct task_struct *task, | |
1064 | struct files_struct *files) | |
1065 | { | |
1066 | struct io_kiocb *req; | |
1067 | ||
84965ff8 JA |
1068 | if (task && head->task != task) { |
1069 | /* in terms of cancelation, always match if req task is dead */ | |
1070 | if (head->task->flags & PF_EXITING) | |
1071 | return true; | |
08d23634 | 1072 | return false; |
84965ff8 | 1073 | } |
08d23634 PB |
1074 | if (!files) |
1075 | return true; | |
1076 | ||
1077 | io_for_each_link(req, head) { | |
b05a1bcd | 1078 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1079 | return true; |
4379bf8b | 1080 | if (req->task->files == files) |
08d23634 PB |
1081 | return true; |
1082 | } | |
1083 | return false; | |
1084 | } | |
1085 | ||
c40f6379 JA |
1086 | static inline void req_set_fail_links(struct io_kiocb *req) |
1087 | { | |
1088 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1089 | req->flags |= REQ_F_FAIL_LINK; | |
1090 | } | |
4a38aed2 | 1091 | |
2b188cc1 JA |
1092 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1093 | { | |
1094 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1095 | ||
0f158b4c | 1096 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1097 | } |
1098 | ||
8eb7e2d0 PB |
1099 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1100 | { | |
1101 | return !req->timeout.off; | |
1102 | } | |
1103 | ||
2b188cc1 JA |
1104 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1105 | { | |
1106 | struct io_ring_ctx *ctx; | |
78076bb6 | 1107 | int hash_bits; |
2b188cc1 JA |
1108 | |
1109 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1110 | if (!ctx) | |
1111 | return NULL; | |
1112 | ||
78076bb6 JA |
1113 | /* |
1114 | * Use 5 bits less than the max cq entries, that should give us around | |
1115 | * 32 entries per hash list if totally full and uniformly spread. | |
1116 | */ | |
1117 | hash_bits = ilog2(p->cq_entries); | |
1118 | hash_bits -= 5; | |
1119 | if (hash_bits <= 0) | |
1120 | hash_bits = 1; | |
1121 | ctx->cancel_hash_bits = hash_bits; | |
1122 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1123 | GFP_KERNEL); | |
1124 | if (!ctx->cancel_hash) | |
1125 | goto err; | |
1126 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1127 | ||
21482896 | 1128 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1129 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1130 | goto err; | |
2b188cc1 JA |
1131 | |
1132 | ctx->flags = p->flags; | |
90554200 | 1133 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1134 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1135 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1136 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1137 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1138 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1139 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1140 | mutex_init(&ctx->uring_lock); |
1141 | init_waitqueue_head(&ctx->wait); | |
1142 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1143 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1144 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1145 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1146 | spin_lock_init(&ctx->inflight_lock); |
1147 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1148 | spin_lock_init(&ctx->rsrc_ref_lock); |
1149 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1150 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1151 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1152 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1153 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1154 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1155 | return ctx; |
206aefde | 1156 | err: |
78076bb6 | 1157 | kfree(ctx->cancel_hash); |
206aefde JA |
1158 | kfree(ctx); |
1159 | return NULL; | |
2b188cc1 JA |
1160 | } |
1161 | ||
9cf7c104 | 1162 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1163 | { |
2bc9930e JA |
1164 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1165 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1166 | |
9cf7c104 | 1167 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1168 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1169 | } |
de0617e4 | 1170 | |
9d858b21 | 1171 | return false; |
de0617e4 JA |
1172 | } |
1173 | ||
ce3d5aae PB |
1174 | static void io_req_track_inflight(struct io_kiocb *req) |
1175 | { | |
1176 | struct io_ring_ctx *ctx = req->ctx; | |
1177 | ||
1178 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1179 | req->flags |= REQ_F_INFLIGHT; |
1180 | ||
1181 | spin_lock_irq(&ctx->inflight_lock); | |
1182 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1183 | spin_unlock_irq(&ctx->inflight_lock); | |
1184 | } | |
1185 | } | |
1186 | ||
1e6fa521 JA |
1187 | static void io_prep_async_work(struct io_kiocb *req) |
1188 | { | |
1189 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1190 | struct io_ring_ctx *ctx = req->ctx; |
1191 | ||
003e8dcc JA |
1192 | if (!req->work.creds) |
1193 | req->work.creds = get_current_cred(); | |
1194 | ||
feaadc4f PB |
1195 | if (req->flags & REQ_F_FORCE_ASYNC) |
1196 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1197 | ||
1e6fa521 JA |
1198 | if (req->flags & REQ_F_ISREG) { |
1199 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1200 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1201 | } else { | |
1202 | if (def->unbound_nonreg_file) | |
1203 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1204 | } | |
561fb04a | 1205 | } |
cccf0ee8 | 1206 | |
cbdcb435 | 1207 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1208 | { |
cbdcb435 | 1209 | struct io_kiocb *cur; |
54a91f3b | 1210 | |
f2f87370 PB |
1211 | io_for_each_link(cur, req) |
1212 | io_prep_async_work(cur); | |
561fb04a JA |
1213 | } |
1214 | ||
ebf93667 | 1215 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1216 | { |
a197f664 | 1217 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1218 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1219 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1220 | |
3bfe6106 JA |
1221 | BUG_ON(!tctx); |
1222 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1223 | |
8766dd51 PB |
1224 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1225 | &req->work, req->flags); | |
cbdcb435 PB |
1226 | /* init ->work of the whole link before punting */ |
1227 | io_prep_async_link(req); | |
ebf93667 | 1228 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1229 | if (link) |
1230 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1231 | } |
1232 | ||
5262f567 JA |
1233 | static void io_kill_timeout(struct io_kiocb *req) |
1234 | { | |
e8c2bc1f | 1235 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1236 | int ret; |
1237 | ||
e8c2bc1f | 1238 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1239 | if (ret != -1) { |
01cec8c1 PB |
1240 | atomic_set(&req->ctx->cq_timeouts, |
1241 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1242 | list_del_init(&req->timeout.list); |
78e19bbe | 1243 | io_cqring_fill_event(req, 0); |
216578e5 | 1244 | io_put_req_deferred(req, 1); |
5262f567 JA |
1245 | } |
1246 | } | |
1247 | ||
76e1b642 JA |
1248 | /* |
1249 | * Returns true if we found and killed one or more timeouts | |
1250 | */ | |
6b81928d PB |
1251 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1252 | struct files_struct *files) | |
5262f567 JA |
1253 | { |
1254 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1255 | int canceled = 0; |
5262f567 JA |
1256 | |
1257 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1258 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1259 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1260 | io_kill_timeout(req); |
76e1b642 JA |
1261 | canceled++; |
1262 | } | |
f3606e3a | 1263 | } |
5262f567 | 1264 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1265 | return canceled != 0; |
5262f567 JA |
1266 | } |
1267 | ||
04518945 | 1268 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1269 | { |
04518945 | 1270 | do { |
27dc8338 PB |
1271 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1272 | struct io_defer_entry, list); | |
de0617e4 | 1273 | |
9cf7c104 | 1274 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1275 | break; |
27dc8338 | 1276 | list_del_init(&de->list); |
907d1df3 | 1277 | io_req_task_queue(de->req); |
27dc8338 | 1278 | kfree(de); |
04518945 PB |
1279 | } while (!list_empty(&ctx->defer_list)); |
1280 | } | |
1281 | ||
360428f8 | 1282 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1283 | { |
f010505b MDG |
1284 | u32 seq; |
1285 | ||
1286 | if (list_empty(&ctx->timeout_list)) | |
1287 | return; | |
1288 | ||
1289 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1290 | ||
1291 | do { | |
1292 | u32 events_needed, events_got; | |
360428f8 | 1293 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1294 | struct io_kiocb, timeout.list); |
de0617e4 | 1295 | |
8eb7e2d0 | 1296 | if (io_is_timeout_noseq(req)) |
360428f8 | 1297 | break; |
f010505b MDG |
1298 | |
1299 | /* | |
1300 | * Since seq can easily wrap around over time, subtract | |
1301 | * the last seq at which timeouts were flushed before comparing. | |
1302 | * Assuming not more than 2^31-1 events have happened since, | |
1303 | * these subtractions won't have wrapped, so we can check if | |
1304 | * target is in [last_seq, current_seq] by comparing the two. | |
1305 | */ | |
1306 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1307 | events_got = seq - ctx->cq_last_tm_flush; | |
1308 | if (events_got < events_needed) | |
360428f8 | 1309 | break; |
bfe68a22 | 1310 | |
135fcde8 | 1311 | list_del_init(&req->timeout.list); |
5262f567 | 1312 | io_kill_timeout(req); |
f010505b MDG |
1313 | } while (!list_empty(&ctx->timeout_list)); |
1314 | ||
1315 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1316 | } |
5262f567 | 1317 | |
360428f8 PB |
1318 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1319 | { | |
1320 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1321 | |
1322 | /* order cqe stores with ring update */ | |
1323 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1324 | |
04518945 PB |
1325 | if (unlikely(!list_empty(&ctx->defer_list))) |
1326 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1327 | } |
1328 | ||
90554200 JA |
1329 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1330 | { | |
1331 | struct io_rings *r = ctx->rings; | |
1332 | ||
1333 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1334 | } | |
1335 | ||
888aae2e PB |
1336 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1337 | { | |
1338 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1339 | } | |
1340 | ||
2b188cc1 JA |
1341 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1342 | { | |
75b28aff | 1343 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1344 | unsigned tail; |
1345 | ||
115e12e5 SB |
1346 | /* |
1347 | * writes to the cq entry need to come after reading head; the | |
1348 | * control dependency is enough as we're using WRITE_ONCE to | |
1349 | * fill the cq entry | |
1350 | */ | |
888aae2e | 1351 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1352 | return NULL; |
1353 | ||
888aae2e | 1354 | tail = ctx->cached_cq_tail++; |
75b28aff | 1355 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1356 | } |
1357 | ||
f2842ab5 JA |
1358 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1359 | { | |
f0b493e6 JA |
1360 | if (!ctx->cq_ev_fd) |
1361 | return false; | |
7e55a19c SG |
1362 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1363 | return false; | |
f2842ab5 JA |
1364 | if (!ctx->eventfd_async) |
1365 | return true; | |
b41e9852 | 1366 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1367 | } |
1368 | ||
b41e9852 | 1369 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1370 | { |
b1445e59 PB |
1371 | /* see waitqueue_active() comment */ |
1372 | smp_mb(); | |
1373 | ||
1d7bb1d5 JA |
1374 | if (waitqueue_active(&ctx->wait)) |
1375 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1376 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1377 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1378 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1379 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1380 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1381 | wake_up_interruptible(&ctx->cq_wait); |
1382 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1383 | } | |
1d7bb1d5 JA |
1384 | } |
1385 | ||
80c18e4a PB |
1386 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1387 | { | |
b1445e59 PB |
1388 | /* see waitqueue_active() comment */ |
1389 | smp_mb(); | |
1390 | ||
80c18e4a PB |
1391 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1392 | if (waitqueue_active(&ctx->wait)) | |
1393 | wake_up(&ctx->wait); | |
1394 | } | |
1395 | if (io_should_trigger_evfd(ctx)) | |
1396 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1397 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1398 | wake_up_interruptible(&ctx->cq_wait); |
1399 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1400 | } | |
80c18e4a PB |
1401 | } |
1402 | ||
c4a2ed72 | 1403 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1404 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1405 | struct task_struct *tsk, | |
1406 | struct files_struct *files) | |
1d7bb1d5 JA |
1407 | { |
1408 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1409 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1410 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1411 | unsigned long flags; |
b18032bb | 1412 | bool all_flushed, posted; |
1d7bb1d5 JA |
1413 | LIST_HEAD(list); |
1414 | ||
e23de15f PB |
1415 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1416 | return false; | |
1d7bb1d5 | 1417 | |
b18032bb | 1418 | posted = false; |
1d7bb1d5 | 1419 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1420 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1421 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1422 | continue; |
1423 | ||
1d7bb1d5 JA |
1424 | cqe = io_get_cqring(ctx); |
1425 | if (!cqe && !force) | |
1426 | break; | |
1427 | ||
40d8ddd4 | 1428 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1429 | if (cqe) { |
1430 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1431 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1432 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1433 | } else { |
2c3bac6d | 1434 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1435 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1436 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1437 | } |
b18032bb | 1438 | posted = true; |
1d7bb1d5 JA |
1439 | } |
1440 | ||
09e88404 PB |
1441 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1442 | if (all_flushed) { | |
1443 | clear_bit(0, &ctx->sq_check_overflow); | |
1444 | clear_bit(0, &ctx->cq_check_overflow); | |
1445 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1446 | } | |
46930143 | 1447 | |
b18032bb JA |
1448 | if (posted) |
1449 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1450 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1451 | if (posted) |
1452 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1453 | |
1454 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1455 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1456 | list_del(&req->compl.list); | |
ec9c02ad | 1457 | io_put_req(req); |
1d7bb1d5 | 1458 | } |
c4a2ed72 | 1459 | |
09e88404 | 1460 | return all_flushed; |
1d7bb1d5 JA |
1461 | } |
1462 | ||
ca0a2651 | 1463 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1464 | struct task_struct *tsk, |
1465 | struct files_struct *files) | |
1466 | { | |
ca0a2651 JA |
1467 | bool ret = true; |
1468 | ||
6c503150 PB |
1469 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1470 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1471 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1472 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1473 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1474 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1475 | mutex_unlock(&ctx->uring_lock); | |
1476 | } | |
ca0a2651 JA |
1477 | |
1478 | return ret; | |
6c503150 PB |
1479 | } |
1480 | ||
bcda7baa | 1481 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1482 | { |
78e19bbe | 1483 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1484 | struct io_uring_cqe *cqe; |
1485 | ||
78e19bbe | 1486 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1487 | |
2b188cc1 JA |
1488 | /* |
1489 | * If we can't get a cq entry, userspace overflowed the | |
1490 | * submission (by quite a lot). Increment the overflow count in | |
1491 | * the ring. | |
1492 | */ | |
1493 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1494 | if (likely(cqe)) { |
78e19bbe | 1495 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1496 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1497 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1498 | } else if (ctx->cq_overflow_flushed || |
1499 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1500 | /* |
1501 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1502 | * then we cannot store the request for later flushing, we need | |
1503 | * to drop it on the floor. | |
1504 | */ | |
2c3bac6d PB |
1505 | ctx->cached_cq_overflow++; |
1506 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1507 | } else { |
ad3eb2c8 JA |
1508 | if (list_empty(&ctx->cq_overflow_list)) { |
1509 | set_bit(0, &ctx->sq_check_overflow); | |
1510 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1511 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1512 | } |
40d8ddd4 | 1513 | io_clean_op(req); |
1d7bb1d5 | 1514 | req->result = res; |
0f7e466b | 1515 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1516 | refcount_inc(&req->refs); |
1517 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1518 | } |
1519 | } | |
1520 | ||
bcda7baa JA |
1521 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1522 | { | |
1523 | __io_cqring_fill_event(req, res, 0); | |
1524 | } | |
1525 | ||
7a612350 PB |
1526 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1527 | unsigned int cflags) | |
2b188cc1 | 1528 | { |
78e19bbe | 1529 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1530 | unsigned long flags; |
1531 | ||
1532 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1533 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1534 | /* |
1535 | * If we're the last reference to this request, add to our locked | |
1536 | * free_list cache. | |
1537 | */ | |
1538 | if (refcount_dec_and_test(&req->refs)) { | |
1539 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1540 | ||
7a612350 PB |
1541 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1542 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1543 | io_disarm_next(req); | |
1544 | if (req->link) { | |
1545 | io_req_task_queue(req->link); | |
1546 | req->link = NULL; | |
1547 | } | |
1548 | } | |
c7dae4ba JA |
1549 | io_dismantle_req(req); |
1550 | io_put_task(req->task, 1); | |
1551 | list_add(&req->compl.list, &cs->locked_free_list); | |
1552 | cs->locked_free_nr++; | |
180f829f PB |
1553 | } else { |
1554 | if (!percpu_ref_tryget(&ctx->refs)) | |
1555 | req = NULL; | |
1556 | } | |
7a612350 | 1557 | io_commit_cqring(ctx); |
2b188cc1 | 1558 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1559 | |
180f829f PB |
1560 | if (req) { |
1561 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1562 | percpu_ref_put(&ctx->refs); |
180f829f | 1563 | } |
229a7b63 JA |
1564 | } |
1565 | ||
a38d68db | 1566 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1567 | unsigned int cflags) |
229a7b63 | 1568 | { |
a38d68db PB |
1569 | io_clean_op(req); |
1570 | req->result = res; | |
1571 | req->compl.cflags = cflags; | |
e342c807 | 1572 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1573 | } |
1574 | ||
889fca73 PB |
1575 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1576 | long res, unsigned cflags) | |
bcda7baa | 1577 | { |
889fca73 PB |
1578 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1579 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1580 | else |
c7dae4ba | 1581 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1582 | } |
1583 | ||
a38d68db | 1584 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1585 | { |
889fca73 | 1586 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1587 | } |
1588 | ||
c7dae4ba | 1589 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1590 | { |
c7dae4ba JA |
1591 | struct io_submit_state *state = &ctx->submit_state; |
1592 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1593 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1594 | |
c7dae4ba JA |
1595 | /* |
1596 | * If we have more than a batch's worth of requests in our IRQ side | |
1597 | * locked cache, grab the lock and move them over to our submission | |
1598 | * side cache. | |
1599 | */ | |
1600 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1601 | spin_lock_irq(&ctx->completion_lock); | |
1602 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1603 | cs->locked_free_nr = 0; | |
1604 | spin_unlock_irq(&ctx->completion_lock); | |
1605 | } | |
0ddf92e8 | 1606 | |
c7dae4ba JA |
1607 | while (!list_empty(&cs->free_list)) { |
1608 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1609 | compl.list); |
1610 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1611 | state->reqs[state->free_reqs++] = req; |
1612 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1613 | break; | |
1b4c351f JA |
1614 | } |
1615 | ||
e5d1bc0a | 1616 | return req != NULL; |
0ddf92e8 JA |
1617 | } |
1618 | ||
e5d1bc0a | 1619 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1620 | { |
e5d1bc0a PB |
1621 | struct io_submit_state *state = &ctx->submit_state; |
1622 | ||
1623 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1624 | ||
f6b6c7d6 | 1625 | if (!state->free_reqs) { |
291b2821 | 1626 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1627 | int ret; |
1628 | ||
c7dae4ba | 1629 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1630 | goto got_req; |
1631 | ||
bf019da7 PB |
1632 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1633 | state->reqs); | |
fd6fab2c JA |
1634 | |
1635 | /* | |
1636 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1637 | * retry single alloc to be on the safe side. | |
1638 | */ | |
1639 | if (unlikely(ret <= 0)) { | |
1640 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1641 | if (!state->reqs[0]) | |
3893f39f | 1642 | return NULL; |
fd6fab2c JA |
1643 | ret = 1; |
1644 | } | |
291b2821 | 1645 | state->free_reqs = ret; |
2b188cc1 | 1646 | } |
e5d1bc0a | 1647 | got_req: |
291b2821 PB |
1648 | state->free_reqs--; |
1649 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1650 | } |
1651 | ||
8da11c19 PB |
1652 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1653 | bool fixed) | |
1654 | { | |
36f72fe2 | 1655 | if (!fixed) |
8da11c19 PB |
1656 | fput(file); |
1657 | } | |
1658 | ||
4edf20f9 | 1659 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1660 | { |
3ca405eb | 1661 | io_clean_op(req); |
929a3af9 | 1662 | |
e8c2bc1f JA |
1663 | if (req->async_data) |
1664 | kfree(req->async_data); | |
8da11c19 PB |
1665 | if (req->file) |
1666 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1667 | if (req->fixed_rsrc_refs) |
1668 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1669 | if (req->work.creds) { |
1670 | put_cred(req->work.creds); | |
1671 | req->work.creds = NULL; | |
1672 | } | |
f85c310a PB |
1673 | |
1674 | if (req->flags & REQ_F_INFLIGHT) { | |
1675 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1676 | unsigned long flags; |
1677 | ||
1678 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1679 | list_del(&req->inflight_entry); | |
1680 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1681 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1682 | } |
e65ef56d JA |
1683 | } |
1684 | ||
b23fcf47 | 1685 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1686 | static inline void io_put_task(struct task_struct *task, int nr) |
1687 | { | |
1688 | struct io_uring_task *tctx = task->io_uring; | |
1689 | ||
1690 | percpu_counter_sub(&tctx->inflight, nr); | |
1691 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1692 | wake_up(&tctx->wait); | |
1693 | put_task_struct_many(task, nr); | |
1694 | } | |
1695 | ||
216578e5 | 1696 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1697 | { |
51a4cc11 | 1698 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1699 | |
216578e5 | 1700 | io_dismantle_req(req); |
7c660731 | 1701 | io_put_task(req->task, 1); |
c6ca97b3 | 1702 | |
3893f39f | 1703 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1704 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1705 | } |
1706 | ||
f2f87370 PB |
1707 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1708 | { | |
1709 | struct io_kiocb *nxt = req->link; | |
1710 | ||
1711 | req->link = nxt->link; | |
1712 | nxt->link = NULL; | |
1713 | } | |
1714 | ||
33cc89a9 PB |
1715 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1716 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1717 | { |
33cc89a9 | 1718 | struct io_kiocb *link = req->link; |
c9abd7ad | 1719 | bool cancelled = false; |
f2f87370 | 1720 | |
900fad45 PB |
1721 | /* |
1722 | * Can happen if a linked timeout fired and link had been like | |
1723 | * req -> link t-out -> link t-out [-> ...] | |
1724 | */ | |
c9abd7ad PB |
1725 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1726 | struct io_timeout_data *io = link->async_data; | |
1727 | int ret; | |
7c86ffee | 1728 | |
f2f87370 | 1729 | io_remove_next_linked(req); |
90cd7e42 | 1730 | link->timeout.head = NULL; |
c9abd7ad PB |
1731 | ret = hrtimer_try_to_cancel(&io->timer); |
1732 | if (ret != -1) { | |
1733 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1734 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1735 | cancelled = true; |
1736 | } | |
1737 | } | |
7c86ffee | 1738 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1739 | return cancelled; |
7c86ffee PB |
1740 | } |
1741 | ||
d148ca4b | 1742 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1743 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1744 | { |
33cc89a9 | 1745 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1746 | |
f2f87370 | 1747 | req->link = NULL; |
f2f87370 PB |
1748 | while (link) { |
1749 | nxt = link->link; | |
1750 | link->link = NULL; | |
2665abfd | 1751 | |
f2f87370 | 1752 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1753 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1754 | io_put_req_deferred(link, 2); |
f2f87370 | 1755 | link = nxt; |
9e645e11 | 1756 | } |
33cc89a9 | 1757 | } |
9e645e11 | 1758 | |
33cc89a9 PB |
1759 | static bool io_disarm_next(struct io_kiocb *req) |
1760 | __must_hold(&req->ctx->completion_lock) | |
1761 | { | |
1762 | bool posted = false; | |
1763 | ||
1764 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1765 | posted = io_kill_linked_timeout(req); | |
1766 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1767 | posted |= (req->link != NULL); | |
1768 | io_fail_links(req); | |
1769 | } | |
1770 | return posted; | |
9e645e11 JA |
1771 | } |
1772 | ||
3fa5e0f3 | 1773 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1774 | { |
33cc89a9 | 1775 | struct io_kiocb *nxt; |
944e58bf | 1776 | |
9e645e11 JA |
1777 | /* |
1778 | * If LINK is set, we have dependent requests in this chain. If we | |
1779 | * didn't fail this request, queue the first one up, moving any other | |
1780 | * dependencies to the next request. In case of failure, fail the rest | |
1781 | * of the chain. | |
1782 | */ | |
33cc89a9 PB |
1783 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1784 | struct io_ring_ctx *ctx = req->ctx; | |
1785 | unsigned long flags; | |
1786 | bool posted; | |
1787 | ||
1788 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1789 | posted = io_disarm_next(req); | |
1790 | if (posted) | |
1791 | io_commit_cqring(req->ctx); | |
1792 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1793 | if (posted) | |
1794 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1795 | } |
33cc89a9 PB |
1796 | nxt = req->link; |
1797 | req->link = NULL; | |
1798 | return nxt; | |
4d7dd462 | 1799 | } |
9e645e11 | 1800 | |
f2f87370 | 1801 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1802 | { |
cdbff982 | 1803 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1804 | return NULL; |
1805 | return __io_req_find_next(req); | |
1806 | } | |
1807 | ||
2c32395d PB |
1808 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1809 | { | |
1810 | if (!ctx) | |
1811 | return; | |
1812 | if (ctx->submit_state.comp.nr) { | |
1813 | mutex_lock(&ctx->uring_lock); | |
1814 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1815 | mutex_unlock(&ctx->uring_lock); | |
1816 | } | |
1817 | percpu_ref_put(&ctx->refs); | |
1818 | } | |
1819 | ||
7cbf1722 | 1820 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1821 | { |
65453d1e | 1822 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1823 | struct io_wq_work_list list; |
1824 | struct io_wq_work_node *node; | |
c2c4c83c | 1825 | |
7cbf1722 JA |
1826 | if (wq_list_empty(&tctx->task_list)) |
1827 | return false; | |
6200b0ae | 1828 | |
0b81e80c | 1829 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1830 | list = tctx->task_list; |
1831 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1832 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1833 | |
7cbf1722 JA |
1834 | node = list.first; |
1835 | while (node) { | |
1836 | struct io_wq_work_node *next = node->next; | |
1837 | struct io_kiocb *req; | |
0ba9c9ed | 1838 | |
7cbf1722 | 1839 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1840 | if (req->ctx != ctx) { |
1841 | ctx_flush_and_put(ctx); | |
1842 | ctx = req->ctx; | |
1843 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1844 | } |
65453d1e | 1845 | |
2c32395d PB |
1846 | req->task_work.func(&req->task_work); |
1847 | node = next; | |
7cbf1722 JA |
1848 | } |
1849 | ||
2c32395d | 1850 | ctx_flush_and_put(ctx); |
7cbf1722 | 1851 | return list.first != NULL; |
c2c4c83c JA |
1852 | } |
1853 | ||
7cbf1722 | 1854 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1855 | { |
7cbf1722 | 1856 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1857 | |
1d5f360d JA |
1858 | clear_bit(0, &tctx->task_state); |
1859 | ||
7cbf1722 JA |
1860 | while (__tctx_task_work(tctx)) |
1861 | cond_resched(); | |
7cbf1722 JA |
1862 | } |
1863 | ||
1864 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1865 | enum task_work_notify_mode notify) | |
1866 | { | |
1867 | struct io_uring_task *tctx = tsk->io_uring; | |
1868 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1869 | unsigned long flags; |
7cbf1722 JA |
1870 | int ret; |
1871 | ||
1872 | WARN_ON_ONCE(!tctx); | |
1873 | ||
0b81e80c | 1874 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1875 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1876 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1877 | |
1878 | /* task_work already pending, we're done */ | |
1879 | if (test_bit(0, &tctx->task_state) || | |
1880 | test_and_set_bit(0, &tctx->task_state)) | |
1881 | return 0; | |
1882 | ||
1883 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1884 | return 0; | |
1885 | ||
1886 | /* | |
1887 | * Slow path - we failed, find and delete work. if the work is not | |
1888 | * in the list, it got run and we're fine. | |
1889 | */ | |
1890 | ret = 0; | |
0b81e80c | 1891 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1892 | wq_list_for_each(node, prev, &tctx->task_list) { |
1893 | if (&req->io_task_work.node == node) { | |
1894 | wq_list_del(&tctx->task_list, node, prev); | |
1895 | ret = 1; | |
1896 | break; | |
1897 | } | |
1898 | } | |
0b81e80c | 1899 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1900 | clear_bit(0, &tctx->task_state); |
1901 | return ret; | |
1902 | } | |
1903 | ||
355fb9e2 | 1904 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1905 | { |
1906 | struct task_struct *tsk = req->task; | |
1907 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1908 | enum task_work_notify_mode notify; |
1909 | int ret; | |
c2c4c83c | 1910 | |
6200b0ae JA |
1911 | if (tsk->flags & PF_EXITING) |
1912 | return -ESRCH; | |
1913 | ||
c2c4c83c | 1914 | /* |
0ba9c9ed JA |
1915 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1916 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1917 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1918 | * will do the job. | |
c2c4c83c | 1919 | */ |
91989c70 | 1920 | notify = TWA_NONE; |
355fb9e2 | 1921 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1922 | notify = TWA_SIGNAL; |
1923 | ||
7cbf1722 | 1924 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1925 | if (!ret) |
1926 | wake_up_process(tsk); | |
0ba9c9ed | 1927 | |
c2c4c83c JA |
1928 | return ret; |
1929 | } | |
1930 | ||
eab30c4d | 1931 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1932 | task_work_func_t cb) |
eab30c4d | 1933 | { |
7c25c0d1 JA |
1934 | struct io_ring_ctx *ctx = req->ctx; |
1935 | struct callback_head *head; | |
eab30c4d PB |
1936 | |
1937 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1938 | do { |
1939 | head = READ_ONCE(ctx->exit_task_work); | |
1940 | req->task_work.next = head; | |
1941 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1942 | } |
1943 | ||
c40f6379 JA |
1944 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1945 | { | |
1946 | struct io_ring_ctx *ctx = req->ctx; | |
1947 | ||
1948 | spin_lock_irq(&ctx->completion_lock); | |
1949 | io_cqring_fill_event(req, error); | |
1950 | io_commit_cqring(ctx); | |
1951 | spin_unlock_irq(&ctx->completion_lock); | |
1952 | ||
1953 | io_cqring_ev_posted(ctx); | |
1954 | req_set_fail_links(req); | |
1955 | io_double_put_req(req); | |
1956 | } | |
1957 | ||
1958 | static void io_req_task_cancel(struct callback_head *cb) | |
1959 | { | |
1960 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1961 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1962 | |
792bb6eb | 1963 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1964 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1965 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1966 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1967 | } |
1968 | ||
1969 | static void __io_req_task_submit(struct io_kiocb *req) | |
1970 | { | |
1971 | struct io_ring_ctx *ctx = req->ctx; | |
1972 | ||
04fc6c80 | 1973 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1974 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 1975 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1976 | __io_queue_sqe(req); |
81b6d05c | 1977 | else |
c40f6379 | 1978 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1979 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1980 | } |
1981 | ||
1982 | static void io_req_task_submit(struct callback_head *cb) | |
1983 | { | |
1984 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1985 | ||
1986 | __io_req_task_submit(req); | |
1987 | } | |
1988 | ||
1989 | static void io_req_task_queue(struct io_kiocb *req) | |
1990 | { | |
c40f6379 JA |
1991 | int ret; |
1992 | ||
7cbf1722 | 1993 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1994 | ret = io_req_task_work_add(req); |
c40f6379 | 1995 | if (unlikely(ret)) { |
a3df7698 | 1996 | req->result = -ECANCELED; |
04fc6c80 | 1997 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1998 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1999 | } |
c40f6379 JA |
2000 | } |
2001 | ||
a3df7698 PB |
2002 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2003 | { | |
2004 | percpu_ref_get(&req->ctx->refs); | |
2005 | req->result = ret; | |
2006 | req->task_work.func = io_req_task_cancel; | |
2007 | ||
2008 | if (unlikely(io_req_task_work_add(req))) | |
2009 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2010 | } | |
2011 | ||
f2f87370 | 2012 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2013 | { |
9b5f7bd9 | 2014 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2015 | |
2016 | if (nxt) | |
906a8c3f | 2017 | io_req_task_queue(nxt); |
c69f8dbe JL |
2018 | } |
2019 | ||
c3524383 | 2020 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2021 | { |
c3524383 PB |
2022 | io_queue_next(req); |
2023 | __io_free_req(req); | |
2024 | } | |
8766dd51 | 2025 | |
2d6500d4 | 2026 | struct req_batch { |
5af1d13e PB |
2027 | struct task_struct *task; |
2028 | int task_refs; | |
1b4c351f | 2029 | int ctx_refs; |
2d6500d4 PB |
2030 | }; |
2031 | ||
5af1d13e PB |
2032 | static inline void io_init_req_batch(struct req_batch *rb) |
2033 | { | |
5af1d13e | 2034 | rb->task_refs = 0; |
9ae72463 | 2035 | rb->ctx_refs = 0; |
5af1d13e PB |
2036 | rb->task = NULL; |
2037 | } | |
2038 | ||
2d6500d4 PB |
2039 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2040 | struct req_batch *rb) | |
2041 | { | |
6e833d53 | 2042 | if (rb->task) |
7c660731 | 2043 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2044 | if (rb->ctx_refs) |
2045 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2046 | } |
2047 | ||
6ff119a6 PB |
2048 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2049 | struct io_submit_state *state) | |
2d6500d4 | 2050 | { |
f2f87370 | 2051 | io_queue_next(req); |
2d6500d4 | 2052 | |
e3bc8e9d | 2053 | if (req->task != rb->task) { |
7c660731 PB |
2054 | if (rb->task) |
2055 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2056 | rb->task = req->task; |
2057 | rb->task_refs = 0; | |
5af1d13e | 2058 | } |
e3bc8e9d | 2059 | rb->task_refs++; |
9ae72463 | 2060 | rb->ctx_refs++; |
5af1d13e | 2061 | |
4edf20f9 | 2062 | io_dismantle_req(req); |
bd759045 | 2063 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2064 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2065 | else |
2066 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2067 | } |
2068 | ||
905c172f PB |
2069 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2070 | struct io_ring_ctx *ctx) | |
2071 | { | |
2072 | int i, nr = cs->nr; | |
2073 | struct io_kiocb *req; | |
2074 | struct req_batch rb; | |
2075 | ||
2076 | io_init_req_batch(&rb); | |
2077 | spin_lock_irq(&ctx->completion_lock); | |
2078 | for (i = 0; i < nr; i++) { | |
2079 | req = cs->reqs[i]; | |
2080 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2081 | } | |
2082 | io_commit_cqring(ctx); | |
2083 | spin_unlock_irq(&ctx->completion_lock); | |
2084 | ||
2085 | io_cqring_ev_posted(ctx); | |
2086 | for (i = 0; i < nr; i++) { | |
2087 | req = cs->reqs[i]; | |
2088 | ||
2089 | /* submission and completion refs */ | |
2090 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2091 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2092 | } |
2093 | ||
2094 | io_req_free_batch_finish(ctx, &rb); | |
2095 | cs->nr = 0; | |
7a743e22 PB |
2096 | } |
2097 | ||
ba816ad6 JA |
2098 | /* |
2099 | * Drop reference to request, return next in chain (if there is one) if this | |
2100 | * was the last reference to this request. | |
2101 | */ | |
9b5f7bd9 | 2102 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2103 | { |
9b5f7bd9 PB |
2104 | struct io_kiocb *nxt = NULL; |
2105 | ||
2a44f467 | 2106 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2107 | nxt = io_req_find_next(req); |
4d7dd462 | 2108 | __io_free_req(req); |
2a44f467 | 2109 | } |
9b5f7bd9 | 2110 | return nxt; |
2b188cc1 JA |
2111 | } |
2112 | ||
e65ef56d JA |
2113 | static void io_put_req(struct io_kiocb *req) |
2114 | { | |
2115 | if (refcount_dec_and_test(&req->refs)) | |
2116 | io_free_req(req); | |
2b188cc1 JA |
2117 | } |
2118 | ||
216578e5 PB |
2119 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2120 | { | |
2121 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2122 | ||
2123 | io_free_req(req); | |
2124 | } | |
2125 | ||
2126 | static void io_free_req_deferred(struct io_kiocb *req) | |
2127 | { | |
2128 | int ret; | |
2129 | ||
7cbf1722 | 2130 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2131 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2132 | if (unlikely(ret)) |
2133 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2134 | } |
2135 | ||
2136 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2137 | { | |
2138 | if (refcount_sub_and_test(refs, &req->refs)) | |
2139 | io_free_req_deferred(req); | |
2140 | } | |
2141 | ||
978db57e JA |
2142 | static void io_double_put_req(struct io_kiocb *req) |
2143 | { | |
2144 | /* drop both submit and complete references */ | |
2145 | if (refcount_sub_and_test(2, &req->refs)) | |
2146 | io_free_req(req); | |
2147 | } | |
2148 | ||
6c503150 | 2149 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2150 | { |
2151 | /* See comment at the top of this file */ | |
2152 | smp_rmb(); | |
e23de15f | 2153 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2154 | } |
2155 | ||
fb5ccc98 PB |
2156 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2157 | { | |
2158 | struct io_rings *rings = ctx->rings; | |
2159 | ||
2160 | /* make sure SQ entry isn't read before tail */ | |
2161 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2162 | } | |
2163 | ||
8ff069bf | 2164 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2165 | { |
8ff069bf | 2166 | unsigned int cflags; |
e94f141b | 2167 | |
bcda7baa JA |
2168 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2169 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2170 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2171 | kfree(kbuf); |
2172 | return cflags; | |
e94f141b JA |
2173 | } |
2174 | ||
8ff069bf | 2175 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2176 | { |
4d954c25 | 2177 | struct io_buffer *kbuf; |
bcda7baa | 2178 | |
4d954c25 | 2179 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2180 | return io_put_kbuf(req, kbuf); |
2181 | } | |
2182 | ||
4c6e277c JA |
2183 | static inline bool io_run_task_work(void) |
2184 | { | |
6200b0ae JA |
2185 | /* |
2186 | * Not safe to run on exiting task, and the task_work handling will | |
2187 | * not add work to such a task. | |
2188 | */ | |
2189 | if (unlikely(current->flags & PF_EXITING)) | |
2190 | return false; | |
4c6e277c JA |
2191 | if (current->task_works) { |
2192 | __set_current_state(TASK_RUNNING); | |
2193 | task_work_run(); | |
2194 | return true; | |
2195 | } | |
2196 | ||
2197 | return false; | |
bcda7baa JA |
2198 | } |
2199 | ||
def596e9 JA |
2200 | /* |
2201 | * Find and free completed poll iocbs | |
2202 | */ | |
2203 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2204 | struct list_head *done) | |
2205 | { | |
8237e045 | 2206 | struct req_batch rb; |
def596e9 | 2207 | struct io_kiocb *req; |
bbde017a XW |
2208 | |
2209 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2210 | smp_rmb(); | |
def596e9 | 2211 | |
5af1d13e | 2212 | io_init_req_batch(&rb); |
def596e9 | 2213 | while (!list_empty(done)) { |
bcda7baa JA |
2214 | int cflags = 0; |
2215 | ||
d21ffe7e | 2216 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2217 | list_del(&req->inflight_entry); |
2218 | ||
bbde017a XW |
2219 | if (READ_ONCE(req->result) == -EAGAIN) { |
2220 | req->iopoll_completed = 0; | |
23faba36 | 2221 | if (io_rw_reissue(req)) |
f161340d | 2222 | continue; |
bbde017a | 2223 | } |
def596e9 | 2224 | |
bcda7baa | 2225 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2226 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2227 | |
2228 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2229 | (*nr_events)++; |
2230 | ||
c3524383 | 2231 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2232 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2233 | } |
def596e9 | 2234 | |
09bb8394 | 2235 | io_commit_cqring(ctx); |
80c18e4a | 2236 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2237 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2238 | } |
2239 | ||
def596e9 JA |
2240 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2241 | long min) | |
2242 | { | |
2243 | struct io_kiocb *req, *tmp; | |
2244 | LIST_HEAD(done); | |
2245 | bool spin; | |
2246 | int ret; | |
2247 | ||
2248 | /* | |
2249 | * Only spin for completions if we don't have multiple devices hanging | |
2250 | * off our complete list, and we're under the requested amount. | |
2251 | */ | |
2252 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2253 | ||
2254 | ret = 0; | |
d21ffe7e | 2255 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2256 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2257 | |
2258 | /* | |
581f9810 BM |
2259 | * Move completed and retryable entries to our local lists. |
2260 | * If we find a request that requires polling, break out | |
2261 | * and complete those lists first, if we have entries there. | |
def596e9 | 2262 | */ |
65a6543d | 2263 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2264 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2265 | continue; |
2266 | } | |
2267 | if (!list_empty(&done)) | |
2268 | break; | |
2269 | ||
2270 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2271 | if (ret < 0) | |
2272 | break; | |
2273 | ||
3aadc23e PB |
2274 | /* iopoll may have completed current req */ |
2275 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2276 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2277 | |
def596e9 JA |
2278 | if (ret && spin) |
2279 | spin = false; | |
2280 | ret = 0; | |
2281 | } | |
2282 | ||
2283 | if (!list_empty(&done)) | |
2284 | io_iopoll_complete(ctx, nr_events, &done); | |
2285 | ||
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | /* | |
d195a66e | 2290 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2291 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2292 | * as a non-spinning completion check. | |
2293 | */ | |
2294 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2295 | long min) | |
2296 | { | |
540e32a0 | 2297 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2298 | int ret; |
2299 | ||
2300 | ret = io_do_iopoll(ctx, nr_events, min); | |
2301 | if (ret < 0) | |
2302 | return ret; | |
eba0a4dd | 2303 | if (*nr_events >= min) |
def596e9 JA |
2304 | return 0; |
2305 | } | |
2306 | ||
2307 | return 1; | |
2308 | } | |
2309 | ||
2310 | /* | |
2311 | * We can't just wait for polled events to come to us, we have to actively | |
2312 | * find and complete them. | |
2313 | */ | |
b2edc0a7 | 2314 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2315 | { |
2316 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2317 | return; | |
2318 | ||
2319 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2320 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2321 | unsigned int nr_events = 0; |
2322 | ||
b2edc0a7 | 2323 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2324 | |
b2edc0a7 PB |
2325 | /* let it sleep and repeat later if can't complete a request */ |
2326 | if (nr_events == 0) | |
2327 | break; | |
08f5439f JA |
2328 | /* |
2329 | * Ensure we allow local-to-the-cpu processing to take place, | |
2330 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2331 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2332 | */ |
3fcee5a6 PB |
2333 | if (need_resched()) { |
2334 | mutex_unlock(&ctx->uring_lock); | |
2335 | cond_resched(); | |
2336 | mutex_lock(&ctx->uring_lock); | |
2337 | } | |
def596e9 JA |
2338 | } |
2339 | mutex_unlock(&ctx->uring_lock); | |
2340 | } | |
2341 | ||
7668b92a | 2342 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2343 | { |
7668b92a | 2344 | unsigned int nr_events = 0; |
2b2ed975 | 2345 | int iters = 0, ret = 0; |
500f9fba | 2346 | |
c7849be9 XW |
2347 | /* |
2348 | * We disallow the app entering submit/complete with polling, but we | |
2349 | * still need to lock the ring to prevent racing with polled issue | |
2350 | * that got punted to a workqueue. | |
2351 | */ | |
2352 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2353 | do { |
a3a0e43f JA |
2354 | /* |
2355 | * Don't enter poll loop if we already have events pending. | |
2356 | * If we do, we can potentially be spinning for commands that | |
2357 | * already triggered a CQE (eg in error). | |
2358 | */ | |
6c503150 PB |
2359 | if (test_bit(0, &ctx->cq_check_overflow)) |
2360 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2361 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2362 | break; |
2363 | ||
500f9fba JA |
2364 | /* |
2365 | * If a submit got punted to a workqueue, we can have the | |
2366 | * application entering polling for a command before it gets | |
2367 | * issued. That app will hold the uring_lock for the duration | |
2368 | * of the poll right here, so we need to take a breather every | |
2369 | * now and then to ensure that the issue has a chance to add | |
2370 | * the poll to the issued list. Otherwise we can spin here | |
2371 | * forever, while the workqueue is stuck trying to acquire the | |
2372 | * very same mutex. | |
2373 | */ | |
2374 | if (!(++iters & 7)) { | |
2375 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2376 | io_run_task_work(); |
500f9fba JA |
2377 | mutex_lock(&ctx->uring_lock); |
2378 | } | |
2379 | ||
7668b92a | 2380 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2381 | if (ret <= 0) |
2382 | break; | |
2383 | ret = 0; | |
7668b92a | 2384 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2385 | |
500f9fba | 2386 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2387 | return ret; |
2388 | } | |
2389 | ||
491381ce | 2390 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2391 | { |
491381ce JA |
2392 | /* |
2393 | * Tell lockdep we inherited freeze protection from submission | |
2394 | * thread. | |
2395 | */ | |
2396 | if (req->flags & REQ_F_ISREG) { | |
2397 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2398 | |
491381ce | 2399 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2400 | } |
491381ce | 2401 | file_end_write(req->file); |
2b188cc1 JA |
2402 | } |
2403 | ||
b63534c4 | 2404 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2405 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2406 | { |
2407 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2408 | int rw, ret; |
b63534c4 | 2409 | struct iov_iter iter; |
b63534c4 | 2410 | |
dc2a6e9a PB |
2411 | /* already prepared */ |
2412 | if (req->async_data) | |
2413 | return true; | |
b63534c4 JA |
2414 | |
2415 | switch (req->opcode) { | |
2416 | case IORING_OP_READV: | |
2417 | case IORING_OP_READ_FIXED: | |
2418 | case IORING_OP_READ: | |
2419 | rw = READ; | |
2420 | break; | |
2421 | case IORING_OP_WRITEV: | |
2422 | case IORING_OP_WRITE_FIXED: | |
2423 | case IORING_OP_WRITE: | |
2424 | rw = WRITE; | |
2425 | break; | |
2426 | default: | |
2427 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2428 | req->opcode); | |
dc2a6e9a | 2429 | return false; |
b63534c4 JA |
2430 | } |
2431 | ||
dc2a6e9a PB |
2432 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2433 | if (ret < 0) | |
2434 | return false; | |
6bf985dc | 2435 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2436 | } |
b63534c4 | 2437 | |
3e6a0d3c | 2438 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2439 | { |
355afaeb | 2440 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2441 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2442 | |
355afaeb JA |
2443 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2444 | return false; | |
3e6a0d3c JA |
2445 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2446 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2447 | return false; |
7c977a58 JA |
2448 | /* |
2449 | * If ref is dying, we might be running poll reap from the exit work. | |
2450 | * Don't attempt to reissue from that path, just let it fail with | |
2451 | * -EAGAIN. | |
2452 | */ | |
3e6a0d3c JA |
2453 | if (percpu_ref_is_dying(&ctx->refs)) |
2454 | return false; | |
2455 | return true; | |
2456 | } | |
2457 | #endif | |
2458 | ||
2459 | static bool io_rw_reissue(struct io_kiocb *req) | |
2460 | { | |
2461 | #ifdef CONFIG_BLOCK | |
2462 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2463 | return false; |
b63534c4 | 2464 | |
55e6ac1e PB |
2465 | lockdep_assert_held(&req->ctx->uring_lock); |
2466 | ||
37d1e2e3 | 2467 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2468 | refcount_inc(&req->refs); |
2469 | io_queue_async_work(req); | |
b63534c4 | 2470 | return true; |
fdee946d | 2471 | } |
dc2a6e9a | 2472 | req_set_fail_links(req); |
b63534c4 JA |
2473 | #endif |
2474 | return false; | |
2475 | } | |
2476 | ||
a1d7c393 | 2477 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2478 | unsigned int issue_flags) |
a1d7c393 | 2479 | { |
2f8e45f1 PB |
2480 | int cflags = 0; |
2481 | ||
23faba36 PB |
2482 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2483 | return; | |
2f8e45f1 PB |
2484 | if (res != req->result) |
2485 | req_set_fail_links(req); | |
23faba36 | 2486 | |
2f8e45f1 PB |
2487 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2488 | kiocb_end_write(req); | |
2489 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2490 | cflags = io_put_rw_kbuf(req); | |
2491 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2492 | } |
2493 | ||
2494 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2495 | { | |
9adbd45d | 2496 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2497 | |
889fca73 | 2498 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2499 | } |
2500 | ||
def596e9 JA |
2501 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2502 | { | |
9adbd45d | 2503 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2504 | |
3e6a0d3c JA |
2505 | #ifdef CONFIG_BLOCK |
2506 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2507 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2508 | struct io_async_rw *rw = req->async_data; | |
2509 | ||
2510 | if (rw) | |
2511 | iov_iter_revert(&rw->iter, | |
2512 | req->result - iov_iter_count(&rw->iter)); | |
2513 | else if (!io_resubmit_prep(req)) | |
2514 | res = -EIO; | |
2515 | } | |
2516 | #endif | |
2517 | ||
491381ce JA |
2518 | if (kiocb->ki_flags & IOCB_WRITE) |
2519 | kiocb_end_write(req); | |
def596e9 | 2520 | |
2d7d6792 | 2521 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2522 | req_set_fail_links(req); |
bbde017a XW |
2523 | |
2524 | WRITE_ONCE(req->result, res); | |
2525 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2526 | smp_wmb(); |
2527 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2528 | } |
2529 | ||
2530 | /* | |
2531 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2532 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2533 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2534 | * accessing the kiocb cookie. | |
2535 | */ | |
2e9dbe90 | 2536 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2537 | { |
2538 | struct io_ring_ctx *ctx = req->ctx; | |
2539 | ||
2540 | /* | |
2541 | * Track whether we have multiple files in our lists. This will impact | |
2542 | * how we do polling eventually, not spinning if we're on potentially | |
2543 | * different devices. | |
2544 | */ | |
540e32a0 | 2545 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2546 | ctx->poll_multi_file = false; |
2547 | } else if (!ctx->poll_multi_file) { | |
2548 | struct io_kiocb *list_req; | |
2549 | ||
540e32a0 | 2550 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2551 | inflight_entry); |
9adbd45d | 2552 | if (list_req->file != req->file) |
def596e9 JA |
2553 | ctx->poll_multi_file = true; |
2554 | } | |
2555 | ||
2556 | /* | |
2557 | * For fast devices, IO may have already completed. If it has, add | |
2558 | * it to the front so we find it first. | |
2559 | */ | |
65a6543d | 2560 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2561 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2562 | else |
d21ffe7e | 2563 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2564 | |
2e9dbe90 XW |
2565 | /* |
2566 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2567 | * task context or in io worker task context. If current task context is | |
2568 | * sq thread, we don't need to check whether should wake up sq thread. | |
2569 | */ | |
2570 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2571 | wq_has_sleeper(&ctx->sq_data->wait)) |
2572 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2573 | } |
2574 | ||
9f13c35b PB |
2575 | static inline void io_state_file_put(struct io_submit_state *state) |
2576 | { | |
02b23a9a PB |
2577 | if (state->file_refs) { |
2578 | fput_many(state->file, state->file_refs); | |
2579 | state->file_refs = 0; | |
2580 | } | |
9a56a232 JA |
2581 | } |
2582 | ||
2583 | /* | |
2584 | * Get as many references to a file as we have IOs left in this submission, | |
2585 | * assuming most submissions are for one file, or at least that each file | |
2586 | * has more than one submission. | |
2587 | */ | |
8da11c19 | 2588 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2589 | { |
2590 | if (!state) | |
2591 | return fget(fd); | |
2592 | ||
6e1271e6 | 2593 | if (state->file_refs) { |
9a56a232 | 2594 | if (state->fd == fd) { |
6e1271e6 | 2595 | state->file_refs--; |
9a56a232 JA |
2596 | return state->file; |
2597 | } | |
02b23a9a | 2598 | io_state_file_put(state); |
9a56a232 JA |
2599 | } |
2600 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2601 | if (unlikely(!state->file)) |
9a56a232 JA |
2602 | return NULL; |
2603 | ||
2604 | state->fd = fd; | |
6e1271e6 | 2605 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2606 | return state->file; |
2607 | } | |
2608 | ||
4503b767 JA |
2609 | static bool io_bdev_nowait(struct block_device *bdev) |
2610 | { | |
9ba0d0c8 | 2611 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2612 | } |
2613 | ||
2b188cc1 JA |
2614 | /* |
2615 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2616 | * any file. For now, just ensure that anything potentially problematic is done | |
2617 | * inline. | |
2618 | */ | |
af197f50 | 2619 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2620 | { |
2621 | umode_t mode = file_inode(file)->i_mode; | |
2622 | ||
4503b767 | 2623 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2624 | if (IS_ENABLED(CONFIG_BLOCK) && |
2625 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2626 | return true; |
2627 | return false; | |
2628 | } | |
2629 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2630 | return true; |
4503b767 | 2631 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2632 | if (IS_ENABLED(CONFIG_BLOCK) && |
2633 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2634 | file->f_op != &io_uring_fops) |
2635 | return true; | |
2636 | return false; | |
2637 | } | |
2b188cc1 | 2638 | |
c5b85625 JA |
2639 | /* any ->read/write should understand O_NONBLOCK */ |
2640 | if (file->f_flags & O_NONBLOCK) | |
2641 | return true; | |
2642 | ||
af197f50 JA |
2643 | if (!(file->f_mode & FMODE_NOWAIT)) |
2644 | return false; | |
2645 | ||
2646 | if (rw == READ) | |
2647 | return file->f_op->read_iter != NULL; | |
2648 | ||
2649 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2650 | } |
2651 | ||
a88fc400 | 2652 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2653 | { |
def596e9 | 2654 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2655 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2656 | struct file *file = req->file; |
09bb8394 JA |
2657 | unsigned ioprio; |
2658 | int ret; | |
2b188cc1 | 2659 | |
75c668cd | 2660 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2661 | req->flags |= REQ_F_ISREG; |
2662 | ||
2b188cc1 | 2663 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2664 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2665 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2666 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2667 | } |
2b188cc1 | 2668 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2669 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2670 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2671 | if (unlikely(ret)) | |
2672 | return ret; | |
2b188cc1 | 2673 | |
75c668cd PB |
2674 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2675 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2676 | req->flags |= REQ_F_NOWAIT; | |
2677 | ||
2b188cc1 JA |
2678 | ioprio = READ_ONCE(sqe->ioprio); |
2679 | if (ioprio) { | |
2680 | ret = ioprio_check_cap(ioprio); | |
2681 | if (ret) | |
09bb8394 | 2682 | return ret; |
2b188cc1 JA |
2683 | |
2684 | kiocb->ki_ioprio = ioprio; | |
2685 | } else | |
2686 | kiocb->ki_ioprio = get_current_ioprio(); | |
2687 | ||
def596e9 | 2688 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2689 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2690 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2691 | return -EOPNOTSUPP; |
2b188cc1 | 2692 | |
def596e9 JA |
2693 | kiocb->ki_flags |= IOCB_HIPRI; |
2694 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2695 | req->iopoll_completed = 0; |
def596e9 | 2696 | } else { |
09bb8394 JA |
2697 | if (kiocb->ki_flags & IOCB_HIPRI) |
2698 | return -EINVAL; | |
def596e9 JA |
2699 | kiocb->ki_complete = io_complete_rw; |
2700 | } | |
9adbd45d | 2701 | |
3529d8c2 JA |
2702 | req->rw.addr = READ_ONCE(sqe->addr); |
2703 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2704 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2705 | return 0; |
2b188cc1 JA |
2706 | } |
2707 | ||
2708 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2709 | { | |
2710 | switch (ret) { | |
2711 | case -EIOCBQUEUED: | |
2712 | break; | |
2713 | case -ERESTARTSYS: | |
2714 | case -ERESTARTNOINTR: | |
2715 | case -ERESTARTNOHAND: | |
2716 | case -ERESTART_RESTARTBLOCK: | |
2717 | /* | |
2718 | * We can't just restart the syscall, since previously | |
2719 | * submitted sqes may already be in progress. Just fail this | |
2720 | * IO with EINTR. | |
2721 | */ | |
2722 | ret = -EINTR; | |
df561f66 | 2723 | fallthrough; |
2b188cc1 JA |
2724 | default: |
2725 | kiocb->ki_complete(kiocb, ret, 0); | |
2726 | } | |
2727 | } | |
2728 | ||
a1d7c393 | 2729 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2730 | unsigned int issue_flags) |
ba816ad6 | 2731 | { |
ba04291e | 2732 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2733 | struct io_async_rw *io = req->async_data; |
ba04291e | 2734 | |
227c0c96 | 2735 | /* add previously done IO, if any */ |
e8c2bc1f | 2736 | if (io && io->bytes_done > 0) { |
227c0c96 | 2737 | if (ret < 0) |
e8c2bc1f | 2738 | ret = io->bytes_done; |
227c0c96 | 2739 | else |
e8c2bc1f | 2740 | ret += io->bytes_done; |
227c0c96 JA |
2741 | } |
2742 | ||
ba04291e JA |
2743 | if (req->flags & REQ_F_CUR_POS) |
2744 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2745 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2746 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2747 | else |
2748 | io_rw_done(kiocb, ret); | |
2749 | } | |
2750 | ||
847595de | 2751 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2752 | { |
9adbd45d JA |
2753 | struct io_ring_ctx *ctx = req->ctx; |
2754 | size_t len = req->rw.len; | |
edafccee | 2755 | struct io_mapped_ubuf *imu; |
4be1c615 | 2756 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2757 | size_t offset; |
2758 | u64 buf_addr; | |
2759 | ||
edafccee JA |
2760 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2761 | return -EFAULT; | |
edafccee JA |
2762 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2763 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2764 | buf_addr = req->rw.addr; |
edafccee JA |
2765 | |
2766 | /* overflow */ | |
2767 | if (buf_addr + len < buf_addr) | |
2768 | return -EFAULT; | |
2769 | /* not inside the mapped region */ | |
2770 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2771 | return -EFAULT; | |
2772 | ||
2773 | /* | |
2774 | * May not be a start of buffer, set size appropriately | |
2775 | * and advance us to the beginning. | |
2776 | */ | |
2777 | offset = buf_addr - imu->ubuf; | |
2778 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2779 | |
2780 | if (offset) { | |
2781 | /* | |
2782 | * Don't use iov_iter_advance() here, as it's really slow for | |
2783 | * using the latter parts of a big fixed buffer - it iterates | |
2784 | * over each segment manually. We can cheat a bit here, because | |
2785 | * we know that: | |
2786 | * | |
2787 | * 1) it's a BVEC iter, we set it up | |
2788 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2789 | * first and last bvec | |
2790 | * | |
2791 | * So just find our index, and adjust the iterator afterwards. | |
2792 | * If the offset is within the first bvec (or the whole first | |
2793 | * bvec, just use iov_iter_advance(). This makes it easier | |
2794 | * since we can just skip the first segment, which may not | |
2795 | * be PAGE_SIZE aligned. | |
2796 | */ | |
2797 | const struct bio_vec *bvec = imu->bvec; | |
2798 | ||
2799 | if (offset <= bvec->bv_len) { | |
2800 | iov_iter_advance(iter, offset); | |
2801 | } else { | |
2802 | unsigned long seg_skip; | |
2803 | ||
2804 | /* skip first vec */ | |
2805 | offset -= bvec->bv_len; | |
2806 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2807 | ||
2808 | iter->bvec = bvec + seg_skip; | |
2809 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2810 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2811 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2812 | } |
2813 | } | |
2814 | ||
847595de | 2815 | return 0; |
edafccee JA |
2816 | } |
2817 | ||
bcda7baa JA |
2818 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2819 | { | |
2820 | if (needs_lock) | |
2821 | mutex_unlock(&ctx->uring_lock); | |
2822 | } | |
2823 | ||
2824 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2825 | { | |
2826 | /* | |
2827 | * "Normal" inline submissions always hold the uring_lock, since we | |
2828 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2829 | * The only exception is when we've detached the request and issue it | |
2830 | * from an async worker thread, grab the lock for that case. | |
2831 | */ | |
2832 | if (needs_lock) | |
2833 | mutex_lock(&ctx->uring_lock); | |
2834 | } | |
2835 | ||
2836 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2837 | int bgid, struct io_buffer *kbuf, | |
2838 | bool needs_lock) | |
2839 | { | |
2840 | struct io_buffer *head; | |
2841 | ||
2842 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2843 | return kbuf; | |
2844 | ||
2845 | io_ring_submit_lock(req->ctx, needs_lock); | |
2846 | ||
2847 | lockdep_assert_held(&req->ctx->uring_lock); | |
2848 | ||
9e15c3a0 | 2849 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2850 | if (head) { |
2851 | if (!list_empty(&head->list)) { | |
2852 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2853 | list); | |
2854 | list_del(&kbuf->list); | |
2855 | } else { | |
2856 | kbuf = head; | |
9e15c3a0 | 2857 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2858 | } |
2859 | if (*len > kbuf->len) | |
2860 | *len = kbuf->len; | |
2861 | } else { | |
2862 | kbuf = ERR_PTR(-ENOBUFS); | |
2863 | } | |
2864 | ||
2865 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2866 | ||
2867 | return kbuf; | |
2868 | } | |
2869 | ||
4d954c25 JA |
2870 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2871 | bool needs_lock) | |
2872 | { | |
2873 | struct io_buffer *kbuf; | |
4f4eeba8 | 2874 | u16 bgid; |
4d954c25 JA |
2875 | |
2876 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2877 | bgid = req->buf_index; |
4d954c25 JA |
2878 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2879 | if (IS_ERR(kbuf)) | |
2880 | return kbuf; | |
2881 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2882 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2883 | return u64_to_user_ptr(kbuf->addr); | |
2884 | } | |
2885 | ||
2886 | #ifdef CONFIG_COMPAT | |
2887 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2888 | bool needs_lock) | |
2889 | { | |
2890 | struct compat_iovec __user *uiov; | |
2891 | compat_ssize_t clen; | |
2892 | void __user *buf; | |
2893 | ssize_t len; | |
2894 | ||
2895 | uiov = u64_to_user_ptr(req->rw.addr); | |
2896 | if (!access_ok(uiov, sizeof(*uiov))) | |
2897 | return -EFAULT; | |
2898 | if (__get_user(clen, &uiov->iov_len)) | |
2899 | return -EFAULT; | |
2900 | if (clen < 0) | |
2901 | return -EINVAL; | |
2902 | ||
2903 | len = clen; | |
2904 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2905 | if (IS_ERR(buf)) | |
2906 | return PTR_ERR(buf); | |
2907 | iov[0].iov_base = buf; | |
2908 | iov[0].iov_len = (compat_size_t) len; | |
2909 | return 0; | |
2910 | } | |
2911 | #endif | |
2912 | ||
2913 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2914 | bool needs_lock) | |
2915 | { | |
2916 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2917 | void __user *buf; | |
2918 | ssize_t len; | |
2919 | ||
2920 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2921 | return -EFAULT; | |
2922 | ||
2923 | len = iov[0].iov_len; | |
2924 | if (len < 0) | |
2925 | return -EINVAL; | |
2926 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2927 | if (IS_ERR(buf)) | |
2928 | return PTR_ERR(buf); | |
2929 | iov[0].iov_base = buf; | |
2930 | iov[0].iov_len = len; | |
2931 | return 0; | |
2932 | } | |
2933 | ||
2934 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2935 | bool needs_lock) | |
2936 | { | |
dddb3e26 JA |
2937 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2938 | struct io_buffer *kbuf; | |
2939 | ||
2940 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2941 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2942 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2943 | return 0; |
dddb3e26 | 2944 | } |
dd201662 | 2945 | if (req->rw.len != 1) |
4d954c25 JA |
2946 | return -EINVAL; |
2947 | ||
2948 | #ifdef CONFIG_COMPAT | |
2949 | if (req->ctx->compat) | |
2950 | return io_compat_import(req, iov, needs_lock); | |
2951 | #endif | |
2952 | ||
2953 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2954 | } | |
2955 | ||
847595de PB |
2956 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2957 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2958 | { |
9adbd45d JA |
2959 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2960 | size_t sqe_len = req->rw.len; | |
847595de | 2961 | u8 opcode = req->opcode; |
4d954c25 | 2962 | ssize_t ret; |
edafccee | 2963 | |
7d009165 | 2964 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2965 | *iovec = NULL; |
9adbd45d | 2966 | return io_import_fixed(req, rw, iter); |
edafccee | 2967 | } |
2b188cc1 | 2968 | |
bcda7baa | 2969 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2970 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2971 | return -EINVAL; |
2972 | ||
3a6820f2 | 2973 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2974 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2975 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2976 | if (IS_ERR(buf)) |
4d954c25 | 2977 | return PTR_ERR(buf); |
3f9d6441 | 2978 | req->rw.len = sqe_len; |
bcda7baa JA |
2979 | } |
2980 | ||
3a6820f2 JA |
2981 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2982 | *iovec = NULL; | |
10fc72e4 | 2983 | return ret; |
3a6820f2 JA |
2984 | } |
2985 | ||
4d954c25 JA |
2986 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2987 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2988 | if (!ret) |
2989 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2990 | *iovec = NULL; |
2991 | return ret; | |
2992 | } | |
2993 | ||
89cd35c5 CH |
2994 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2995 | req->ctx->compat); | |
2b188cc1 JA |
2996 | } |
2997 | ||
0fef9483 JA |
2998 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2999 | { | |
5b09e37e | 3000 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3001 | } |
3002 | ||
31b51510 | 3003 | /* |
32960613 JA |
3004 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3005 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3006 | */ |
4017eb91 | 3007 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3008 | { |
4017eb91 JA |
3009 | struct kiocb *kiocb = &req->rw.kiocb; |
3010 | struct file *file = req->file; | |
32960613 JA |
3011 | ssize_t ret = 0; |
3012 | ||
3013 | /* | |
3014 | * Don't support polled IO through this interface, and we can't | |
3015 | * support non-blocking either. For the latter, this just causes | |
3016 | * the kiocb to be handled from an async context. | |
3017 | */ | |
3018 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3019 | return -EOPNOTSUPP; | |
3020 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3021 | return -EAGAIN; | |
3022 | ||
3023 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3024 | struct iovec iovec; |
32960613 JA |
3025 | ssize_t nr; |
3026 | ||
311ae9e1 PB |
3027 | if (!iov_iter_is_bvec(iter)) { |
3028 | iovec = iov_iter_iovec(iter); | |
3029 | } else { | |
4017eb91 JA |
3030 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3031 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3032 | } |
3033 | ||
32960613 JA |
3034 | if (rw == READ) { |
3035 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3036 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3037 | } else { |
3038 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3039 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3040 | } |
3041 | ||
3042 | if (nr < 0) { | |
3043 | if (!ret) | |
3044 | ret = nr; | |
3045 | break; | |
3046 | } | |
3047 | ret += nr; | |
3048 | if (nr != iovec.iov_len) | |
3049 | break; | |
4017eb91 JA |
3050 | req->rw.len -= nr; |
3051 | req->rw.addr += nr; | |
32960613 JA |
3052 | iov_iter_advance(iter, nr); |
3053 | } | |
3054 | ||
3055 | return ret; | |
3056 | } | |
3057 | ||
ff6165b2 JA |
3058 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3059 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3060 | { |
e8c2bc1f | 3061 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3062 | |
ff6165b2 | 3063 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3064 | rw->free_iovec = iovec; |
227c0c96 | 3065 | rw->bytes_done = 0; |
ff6165b2 | 3066 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3067 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3068 | return; |
b64e3444 | 3069 | if (!iovec) { |
ff6165b2 JA |
3070 | unsigned iov_off = 0; |
3071 | ||
3072 | rw->iter.iov = rw->fast_iov; | |
3073 | if (iter->iov != fast_iov) { | |
3074 | iov_off = iter->iov - fast_iov; | |
3075 | rw->iter.iov += iov_off; | |
3076 | } | |
3077 | if (rw->fast_iov != fast_iov) | |
3078 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3079 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3080 | } else { |
3081 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3082 | } |
3083 | } | |
3084 | ||
e8c2bc1f | 3085 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3086 | { |
e8c2bc1f JA |
3087 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3088 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3089 | return req->async_data == NULL; | |
3d9932a8 XW |
3090 | } |
3091 | ||
e8c2bc1f | 3092 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3093 | { |
e8c2bc1f | 3094 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3095 | return 0; |
3d9932a8 | 3096 | |
e8c2bc1f | 3097 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3098 | } |
3099 | ||
ff6165b2 JA |
3100 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3101 | const struct iovec *fast_iov, | |
227c0c96 | 3102 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3103 | { |
e8c2bc1f | 3104 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3105 | return 0; |
e8c2bc1f | 3106 | if (!req->async_data) { |
6bf985dc PB |
3107 | if (__io_alloc_async_data(req)) { |
3108 | kfree(iovec); | |
5d204bcf | 3109 | return -ENOMEM; |
6bf985dc | 3110 | } |
b7bb4f7d | 3111 | |
ff6165b2 | 3112 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3113 | } |
b7bb4f7d | 3114 | return 0; |
f67676d1 JA |
3115 | } |
3116 | ||
73debe68 | 3117 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3118 | { |
e8c2bc1f | 3119 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3120 | struct iovec *iov = iorw->fast_iov; |
847595de | 3121 | int ret; |
c3e330a4 | 3122 | |
2846c481 | 3123 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3124 | if (unlikely(ret < 0)) |
3125 | return ret; | |
3126 | ||
ab0b196c PB |
3127 | iorw->bytes_done = 0; |
3128 | iorw->free_iovec = iov; | |
3129 | if (iov) | |
3130 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3131 | return 0; |
3132 | } | |
3133 | ||
73debe68 | 3134 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3135 | { |
3529d8c2 JA |
3136 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3137 | return -EBADF; | |
93642ef8 | 3138 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3139 | } |
3140 | ||
c1dd91d1 JA |
3141 | /* |
3142 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3143 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3144 | * This gets called when the page is unlocked, and we generally expect that to | |
3145 | * happen when the page IO is completed and the page is now uptodate. This will | |
3146 | * queue a task_work based retry of the operation, attempting to copy the data | |
3147 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3148 | * do a thread based blocking retry of the operation. That's the unexpected | |
3149 | * slow path. | |
3150 | */ | |
bcf5a063 JA |
3151 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3152 | int sync, void *arg) | |
3153 | { | |
3154 | struct wait_page_queue *wpq; | |
3155 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3156 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3157 | |
3158 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3159 | ||
cdc8fcb4 LT |
3160 | if (!wake_page_match(wpq, key)) |
3161 | return 0; | |
3162 | ||
c8d317aa | 3163 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3164 | list_del_init(&wait->entry); |
3165 | ||
bcf5a063 JA |
3166 | /* submit ref gets dropped, acquire a new one */ |
3167 | refcount_inc(&req->refs); | |
921b9054 | 3168 | io_req_task_queue(req); |
bcf5a063 JA |
3169 | return 1; |
3170 | } | |
3171 | ||
c1dd91d1 JA |
3172 | /* |
3173 | * This controls whether a given IO request should be armed for async page | |
3174 | * based retry. If we return false here, the request is handed to the async | |
3175 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3176 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3177 | * will either succeed because the page is now uptodate and unlocked, or it | |
3178 | * will register a callback when the page is unlocked at IO completion. Through | |
3179 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3180 | * That retry will attempt the buffered read again. The retry will generally | |
3181 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3182 | * async worker threads for a blocking retry. | |
3183 | */ | |
227c0c96 | 3184 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3185 | { |
e8c2bc1f JA |
3186 | struct io_async_rw *rw = req->async_data; |
3187 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3188 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3189 | |
bcf5a063 JA |
3190 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3191 | if (req->flags & REQ_F_NOWAIT) | |
3192 | return false; | |
f67676d1 | 3193 | |
227c0c96 | 3194 | /* Only for buffered IO */ |
3b2a4439 | 3195 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3196 | return false; |
3b2a4439 | 3197 | |
bcf5a063 JA |
3198 | /* |
3199 | * just use poll if we can, and don't attempt if the fs doesn't | |
3200 | * support callback based unlocks | |
3201 | */ | |
3202 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3203 | return false; | |
f67676d1 | 3204 | |
3b2a4439 JA |
3205 | wait->wait.func = io_async_buf_func; |
3206 | wait->wait.private = req; | |
3207 | wait->wait.flags = 0; | |
3208 | INIT_LIST_HEAD(&wait->wait.entry); | |
3209 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3210 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3211 | kiocb->ki_waitq = wait; |
3b2a4439 | 3212 | return true; |
bcf5a063 JA |
3213 | } |
3214 | ||
3215 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3216 | { | |
3217 | if (req->file->f_op->read_iter) | |
3218 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3219 | else if (req->file->f_op->read) |
4017eb91 | 3220 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3221 | else |
3222 | return -EINVAL; | |
f67676d1 JA |
3223 | } |
3224 | ||
889fca73 | 3225 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3226 | { |
3227 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3228 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3229 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3230 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3231 | ssize_t io_size, ret, ret2; |
45d189c6 | 3232 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3233 | |
2846c481 | 3234 | if (rw) { |
e8c2bc1f | 3235 | iter = &rw->iter; |
2846c481 PB |
3236 | iovec = NULL; |
3237 | } else { | |
3238 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3239 | if (ret < 0) | |
3240 | return ret; | |
3241 | } | |
632546c4 | 3242 | io_size = iov_iter_count(iter); |
fa15bafb | 3243 | req->result = io_size; |
2b188cc1 | 3244 | |
fd6c2e4c JA |
3245 | /* Ensure we clear previously set non-block flag */ |
3246 | if (!force_nonblock) | |
29de5f6a | 3247 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3248 | else |
3249 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3250 | ||
24c74678 | 3251 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3252 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3253 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3254 | return ret ?: -EAGAIN; |
6713e7a6 | 3255 | } |
9e645e11 | 3256 | |
632546c4 | 3257 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3258 | if (unlikely(ret)) { |
3259 | kfree(iovec); | |
3260 | return ret; | |
3261 | } | |
2b188cc1 | 3262 | |
227c0c96 | 3263 | ret = io_iter_do_read(req, iter); |
32960613 | 3264 | |
57cd657b | 3265 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3266 | if (req->async_data) |
3267 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3268 | goto out_free; |
227c0c96 | 3269 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3270 | /* IOPOLL retry should happen for io-wq threads */ |
3271 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3272 | goto done; |
75c668cd PB |
3273 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3274 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3275 | goto done; |
84216315 | 3276 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3277 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3278 | ret = 0; |
7335e3bf | 3279 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3280 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3281 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3282 | goto done; |
227c0c96 JA |
3283 | } |
3284 | ||
227c0c96 | 3285 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3286 | if (ret2) |
3287 | return ret2; | |
3288 | ||
fe1cdd55 | 3289 | iovec = NULL; |
e8c2bc1f | 3290 | rw = req->async_data; |
227c0c96 | 3291 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3292 | iter = &rw->iter; |
227c0c96 | 3293 | |
b23df91b PB |
3294 | do { |
3295 | io_size -= ret; | |
3296 | rw->bytes_done += ret; | |
3297 | /* if we can retry, do so with the callbacks armed */ | |
3298 | if (!io_rw_should_retry(req)) { | |
3299 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3300 | return -EAGAIN; | |
3301 | } | |
3302 | ||
3303 | /* | |
3304 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3305 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3306 | * desired page gets unlocked. We can also get a partial read | |
3307 | * here, and if we do, then just retry at the new offset. | |
3308 | */ | |
3309 | ret = io_iter_do_read(req, iter); | |
3310 | if (ret == -EIOCBQUEUED) | |
3311 | return 0; | |
227c0c96 | 3312 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3313 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3314 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3315 | done: |
889fca73 | 3316 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3317 | out_free: |
3318 | /* it's faster to check here then delegate to kfree */ | |
3319 | if (iovec) | |
3320 | kfree(iovec); | |
5ea5dd45 | 3321 | return 0; |
2b188cc1 JA |
3322 | } |
3323 | ||
73debe68 | 3324 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3325 | { |
3529d8c2 JA |
3326 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3327 | return -EBADF; | |
93642ef8 | 3328 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3329 | } |
3330 | ||
889fca73 | 3331 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3332 | { |
3333 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3334 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3335 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3336 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3337 | ssize_t ret, ret2, io_size; |
45d189c6 | 3338 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3339 | |
2846c481 | 3340 | if (rw) { |
e8c2bc1f | 3341 | iter = &rw->iter; |
2846c481 PB |
3342 | iovec = NULL; |
3343 | } else { | |
3344 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3345 | if (ret < 0) | |
3346 | return ret; | |
3347 | } | |
632546c4 | 3348 | io_size = iov_iter_count(iter); |
fa15bafb | 3349 | req->result = io_size; |
2b188cc1 | 3350 | |
fd6c2e4c JA |
3351 | /* Ensure we clear previously set non-block flag */ |
3352 | if (!force_nonblock) | |
a88fc400 PB |
3353 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3354 | else | |
3355 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3356 | |
24c74678 | 3357 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3358 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3359 | goto copy_iov; |
31b51510 | 3360 | |
10d59345 JA |
3361 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3362 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3363 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3364 | goto copy_iov; |
31b51510 | 3365 | |
632546c4 | 3366 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3367 | if (unlikely(ret)) |
3368 | goto out_free; | |
4ed734b0 | 3369 | |
fa15bafb PB |
3370 | /* |
3371 | * Open-code file_start_write here to grab freeze protection, | |
3372 | * which will be released by another thread in | |
3373 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3374 | * released so that it doesn't complain about the held lock when | |
3375 | * we return to userspace. | |
3376 | */ | |
3377 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3378 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3379 | __sb_writers_release(file_inode(req->file)->i_sb, |
3380 | SB_FREEZE_WRITE); | |
3381 | } | |
3382 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3383 | |
fa15bafb | 3384 | if (req->file->f_op->write_iter) |
ff6165b2 | 3385 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3386 | else if (req->file->f_op->write) |
4017eb91 | 3387 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3388 | else |
3389 | ret2 = -EINVAL; | |
4ed734b0 | 3390 | |
fa15bafb PB |
3391 | /* |
3392 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3393 | * retry them without IOCB_NOWAIT. | |
3394 | */ | |
3395 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3396 | ret2 = -EAGAIN; | |
75c668cd PB |
3397 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3398 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3399 | goto done; |
3e6a0d3c JA |
3400 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3401 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3402 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3403 | /* IOPOLL retry should happen for io-wq threads */ |
3404 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3405 | goto copy_iov; | |
355afaeb | 3406 | done: |
889fca73 | 3407 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3408 | } else { |
f67676d1 | 3409 | copy_iov: |
84216315 | 3410 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3411 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3412 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3413 | return ret ?: -EAGAIN; |
2b188cc1 | 3414 | } |
31b51510 | 3415 | out_free: |
f261c168 | 3416 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3417 | if (iovec) |
6f2cc166 | 3418 | kfree(iovec); |
2b188cc1 JA |
3419 | return ret; |
3420 | } | |
3421 | ||
80a261fd JA |
3422 | static int io_renameat_prep(struct io_kiocb *req, |
3423 | const struct io_uring_sqe *sqe) | |
3424 | { | |
3425 | struct io_rename *ren = &req->rename; | |
3426 | const char __user *oldf, *newf; | |
3427 | ||
3428 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3429 | return -EBADF; | |
3430 | ||
3431 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3432 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3433 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3434 | ren->new_dfd = READ_ONCE(sqe->len); | |
3435 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3436 | ||
3437 | ren->oldpath = getname(oldf); | |
3438 | if (IS_ERR(ren->oldpath)) | |
3439 | return PTR_ERR(ren->oldpath); | |
3440 | ||
3441 | ren->newpath = getname(newf); | |
3442 | if (IS_ERR(ren->newpath)) { | |
3443 | putname(ren->oldpath); | |
3444 | return PTR_ERR(ren->newpath); | |
3445 | } | |
3446 | ||
3447 | req->flags |= REQ_F_NEED_CLEANUP; | |
3448 | return 0; | |
3449 | } | |
3450 | ||
45d189c6 | 3451 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3452 | { |
3453 | struct io_rename *ren = &req->rename; | |
3454 | int ret; | |
3455 | ||
45d189c6 | 3456 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3457 | return -EAGAIN; |
3458 | ||
3459 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3460 | ren->newpath, ren->flags); | |
3461 | ||
3462 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3463 | if (ret < 0) | |
3464 | req_set_fail_links(req); | |
3465 | io_req_complete(req, ret); | |
3466 | return 0; | |
3467 | } | |
3468 | ||
14a1143b JA |
3469 | static int io_unlinkat_prep(struct io_kiocb *req, |
3470 | const struct io_uring_sqe *sqe) | |
3471 | { | |
3472 | struct io_unlink *un = &req->unlink; | |
3473 | const char __user *fname; | |
3474 | ||
3475 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3476 | return -EBADF; | |
3477 | ||
3478 | un->dfd = READ_ONCE(sqe->fd); | |
3479 | ||
3480 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3481 | if (un->flags & ~AT_REMOVEDIR) | |
3482 | return -EINVAL; | |
3483 | ||
3484 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3485 | un->filename = getname(fname); | |
3486 | if (IS_ERR(un->filename)) | |
3487 | return PTR_ERR(un->filename); | |
3488 | ||
3489 | req->flags |= REQ_F_NEED_CLEANUP; | |
3490 | return 0; | |
3491 | } | |
3492 | ||
45d189c6 | 3493 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3494 | { |
3495 | struct io_unlink *un = &req->unlink; | |
3496 | int ret; | |
3497 | ||
45d189c6 | 3498 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3499 | return -EAGAIN; |
3500 | ||
3501 | if (un->flags & AT_REMOVEDIR) | |
3502 | ret = do_rmdir(un->dfd, un->filename); | |
3503 | else | |
3504 | ret = do_unlinkat(un->dfd, un->filename); | |
3505 | ||
3506 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3507 | if (ret < 0) | |
3508 | req_set_fail_links(req); | |
3509 | io_req_complete(req, ret); | |
3510 | return 0; | |
3511 | } | |
3512 | ||
36f4fa68 JA |
3513 | static int io_shutdown_prep(struct io_kiocb *req, |
3514 | const struct io_uring_sqe *sqe) | |
3515 | { | |
3516 | #if defined(CONFIG_NET) | |
3517 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3518 | return -EINVAL; | |
3519 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3520 | sqe->buf_index) | |
3521 | return -EINVAL; | |
3522 | ||
3523 | req->shutdown.how = READ_ONCE(sqe->len); | |
3524 | return 0; | |
3525 | #else | |
3526 | return -EOPNOTSUPP; | |
3527 | #endif | |
3528 | } | |
3529 | ||
45d189c6 | 3530 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3531 | { |
3532 | #if defined(CONFIG_NET) | |
3533 | struct socket *sock; | |
3534 | int ret; | |
3535 | ||
45d189c6 | 3536 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3537 | return -EAGAIN; |
3538 | ||
48aba79b | 3539 | sock = sock_from_file(req->file); |
36f4fa68 | 3540 | if (unlikely(!sock)) |
48aba79b | 3541 | return -ENOTSOCK; |
36f4fa68 JA |
3542 | |
3543 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3544 | if (ret < 0) |
3545 | req_set_fail_links(req); | |
36f4fa68 JA |
3546 | io_req_complete(req, ret); |
3547 | return 0; | |
3548 | #else | |
3549 | return -EOPNOTSUPP; | |
3550 | #endif | |
3551 | } | |
3552 | ||
f2a8d5c7 PB |
3553 | static int __io_splice_prep(struct io_kiocb *req, |
3554 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3555 | { |
3556 | struct io_splice* sp = &req->splice; | |
3557 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3558 | |
3232dd02 PB |
3559 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3560 | return -EINVAL; | |
7d67af2c PB |
3561 | |
3562 | sp->file_in = NULL; | |
7d67af2c PB |
3563 | sp->len = READ_ONCE(sqe->len); |
3564 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3565 | ||
3566 | if (unlikely(sp->flags & ~valid_flags)) | |
3567 | return -EINVAL; | |
3568 | ||
8371adf5 PB |
3569 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3570 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3571 | if (!sp->file_in) | |
3572 | return -EBADF; | |
7d67af2c PB |
3573 | req->flags |= REQ_F_NEED_CLEANUP; |
3574 | ||
7cdaf587 XW |
3575 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3576 | /* | |
3577 | * Splice operation will be punted aync, and here need to | |
3578 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3579 | */ | |
7d67af2c | 3580 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3581 | } |
7d67af2c PB |
3582 | |
3583 | return 0; | |
3584 | } | |
3585 | ||
f2a8d5c7 PB |
3586 | static int io_tee_prep(struct io_kiocb *req, |
3587 | const struct io_uring_sqe *sqe) | |
3588 | { | |
3589 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3590 | return -EINVAL; | |
3591 | return __io_splice_prep(req, sqe); | |
3592 | } | |
3593 | ||
45d189c6 | 3594 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3595 | { |
3596 | struct io_splice *sp = &req->splice; | |
3597 | struct file *in = sp->file_in; | |
3598 | struct file *out = sp->file_out; | |
3599 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3600 | long ret = 0; | |
3601 | ||
45d189c6 | 3602 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3603 | return -EAGAIN; |
3604 | if (sp->len) | |
3605 | ret = do_tee(in, out, sp->len, flags); | |
3606 | ||
3607 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3608 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3609 | ||
f2a8d5c7 PB |
3610 | if (ret != sp->len) |
3611 | req_set_fail_links(req); | |
e1e16097 | 3612 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3613 | return 0; |
3614 | } | |
3615 | ||
3616 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3617 | { | |
3618 | struct io_splice* sp = &req->splice; | |
3619 | ||
3620 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3621 | sp->off_out = READ_ONCE(sqe->off); | |
3622 | return __io_splice_prep(req, sqe); | |
3623 | } | |
3624 | ||
45d189c6 | 3625 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3626 | { |
3627 | struct io_splice *sp = &req->splice; | |
3628 | struct file *in = sp->file_in; | |
3629 | struct file *out = sp->file_out; | |
3630 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3631 | loff_t *poff_in, *poff_out; | |
c9687426 | 3632 | long ret = 0; |
7d67af2c | 3633 | |
45d189c6 | 3634 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3635 | return -EAGAIN; |
7d67af2c PB |
3636 | |
3637 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3638 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3639 | |
948a7749 | 3640 | if (sp->len) |
c9687426 | 3641 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3642 | |
3643 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3644 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3645 | ||
7d67af2c PB |
3646 | if (ret != sp->len) |
3647 | req_set_fail_links(req); | |
e1e16097 | 3648 | io_req_complete(req, ret); |
7d67af2c PB |
3649 | return 0; |
3650 | } | |
3651 | ||
2b188cc1 JA |
3652 | /* |
3653 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3654 | */ | |
889fca73 | 3655 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3656 | { |
3657 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3658 | |
def596e9 JA |
3659 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3660 | return -EINVAL; | |
3661 | ||
889fca73 | 3662 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3663 | return 0; |
3664 | } | |
3665 | ||
1155c76a | 3666 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3667 | { |
6b06314c | 3668 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3669 | |
09bb8394 JA |
3670 | if (!req->file) |
3671 | return -EBADF; | |
c992fe29 | 3672 | |
6b06314c | 3673 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3674 | return -EINVAL; |
edafccee | 3675 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3676 | return -EINVAL; |
3677 | ||
8ed8d3c3 JA |
3678 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3679 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3680 | return -EINVAL; | |
3681 | ||
3682 | req->sync.off = READ_ONCE(sqe->off); | |
3683 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3684 | return 0; |
3685 | } | |
3686 | ||
45d189c6 | 3687 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3688 | { |
8ed8d3c3 | 3689 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3690 | int ret; |
3691 | ||
ac45abc0 | 3692 | /* fsync always requires a blocking context */ |
45d189c6 | 3693 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3694 | return -EAGAIN; |
3695 | ||
9adbd45d | 3696 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3697 | end > 0 ? end : LLONG_MAX, |
3698 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3699 | if (ret < 0) | |
3700 | req_set_fail_links(req); | |
e1e16097 | 3701 | io_req_complete(req, ret); |
c992fe29 CH |
3702 | return 0; |
3703 | } | |
3704 | ||
d63d1b5e JA |
3705 | static int io_fallocate_prep(struct io_kiocb *req, |
3706 | const struct io_uring_sqe *sqe) | |
3707 | { | |
3708 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3709 | return -EINVAL; | |
3232dd02 PB |
3710 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3711 | return -EINVAL; | |
d63d1b5e JA |
3712 | |
3713 | req->sync.off = READ_ONCE(sqe->off); | |
3714 | req->sync.len = READ_ONCE(sqe->addr); | |
3715 | req->sync.mode = READ_ONCE(sqe->len); | |
3716 | return 0; | |
3717 | } | |
3718 | ||
45d189c6 | 3719 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3720 | { |
ac45abc0 PB |
3721 | int ret; |
3722 | ||
d63d1b5e | 3723 | /* fallocate always requiring blocking context */ |
45d189c6 | 3724 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3725 | return -EAGAIN; |
ac45abc0 PB |
3726 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3727 | req->sync.len); | |
ac45abc0 PB |
3728 | if (ret < 0) |
3729 | req_set_fail_links(req); | |
e1e16097 | 3730 | io_req_complete(req, ret); |
5d17b4a4 JA |
3731 | return 0; |
3732 | } | |
3733 | ||
ec65fea5 | 3734 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3735 | { |
f8748881 | 3736 | const char __user *fname; |
15b71abe | 3737 | int ret; |
b7bb4f7d | 3738 | |
ec65fea5 | 3739 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3740 | return -EINVAL; |
ec65fea5 | 3741 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3742 | return -EBADF; |
03b1230c | 3743 | |
ec65fea5 PB |
3744 | /* open.how should be already initialised */ |
3745 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3746 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3747 | |
25e72d10 PB |
3748 | req->open.dfd = READ_ONCE(sqe->fd); |
3749 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3750 | req->open.filename = getname(fname); |
15b71abe JA |
3751 | if (IS_ERR(req->open.filename)) { |
3752 | ret = PTR_ERR(req->open.filename); | |
3753 | req->open.filename = NULL; | |
3754 | return ret; | |
3755 | } | |
4022e7af | 3756 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3757 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3758 | return 0; |
03b1230c JA |
3759 | } |
3760 | ||
ec65fea5 PB |
3761 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3762 | { | |
3763 | u64 flags, mode; | |
3764 | ||
14587a46 | 3765 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3766 | return -EINVAL; |
ec65fea5 PB |
3767 | mode = READ_ONCE(sqe->len); |
3768 | flags = READ_ONCE(sqe->open_flags); | |
3769 | req->open.how = build_open_how(flags, mode); | |
3770 | return __io_openat_prep(req, sqe); | |
3771 | } | |
3772 | ||
cebdb986 | 3773 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3774 | { |
cebdb986 | 3775 | struct open_how __user *how; |
cebdb986 | 3776 | size_t len; |
0fa03c62 JA |
3777 | int ret; |
3778 | ||
14587a46 | 3779 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3780 | return -EINVAL; |
cebdb986 JA |
3781 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3782 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3783 | if (len < OPEN_HOW_SIZE_VER0) |
3784 | return -EINVAL; | |
3529d8c2 | 3785 | |
cebdb986 JA |
3786 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3787 | len); | |
3788 | if (ret) | |
3789 | return ret; | |
3529d8c2 | 3790 | |
ec65fea5 | 3791 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3792 | } |
3793 | ||
45d189c6 | 3794 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3795 | { |
3796 | struct open_flags op; | |
15b71abe | 3797 | struct file *file; |
3a81fd02 JA |
3798 | bool nonblock_set; |
3799 | bool resolve_nonblock; | |
15b71abe JA |
3800 | int ret; |
3801 | ||
cebdb986 | 3802 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3803 | if (ret) |
3804 | goto err; | |
3a81fd02 JA |
3805 | nonblock_set = op.open_flag & O_NONBLOCK; |
3806 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3807 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3808 | /* |
3809 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3810 | * it'll always -EAGAIN | |
3811 | */ | |
3812 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3813 | return -EAGAIN; | |
3814 | op.lookup_flags |= LOOKUP_CACHED; | |
3815 | op.open_flag |= O_NONBLOCK; | |
3816 | } | |
15b71abe | 3817 | |
4022e7af | 3818 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3819 | if (ret < 0) |
3820 | goto err; | |
3821 | ||
3822 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3823 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3824 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3825 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3826 | /* |
3a81fd02 JA |
3827 | * We could hang on to this 'fd', but seems like marginal |
3828 | * gain for something that is now known to be a slower path. | |
3829 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3830 | */ |
3a81fd02 JA |
3831 | put_unused_fd(ret); |
3832 | return -EAGAIN; | |
3833 | } | |
3834 | ||
15b71abe JA |
3835 | if (IS_ERR(file)) { |
3836 | put_unused_fd(ret); | |
3837 | ret = PTR_ERR(file); | |
3838 | } else { | |
45d189c6 | 3839 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3840 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3841 | fsnotify_open(file); |
3842 | fd_install(ret, file); | |
3843 | } | |
3844 | err: | |
3845 | putname(req->open.filename); | |
8fef80bf | 3846 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3847 | if (ret < 0) |
3848 | req_set_fail_links(req); | |
e1e16097 | 3849 | io_req_complete(req, ret); |
15b71abe JA |
3850 | return 0; |
3851 | } | |
3852 | ||
45d189c6 | 3853 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3854 | { |
e45cff58 | 3855 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3856 | } |
3857 | ||
067524e9 JA |
3858 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3859 | const struct io_uring_sqe *sqe) | |
3860 | { | |
3861 | struct io_provide_buf *p = &req->pbuf; | |
3862 | u64 tmp; | |
3863 | ||
3864 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3865 | return -EINVAL; | |
3866 | ||
3867 | tmp = READ_ONCE(sqe->fd); | |
3868 | if (!tmp || tmp > USHRT_MAX) | |
3869 | return -EINVAL; | |
3870 | ||
3871 | memset(p, 0, sizeof(*p)); | |
3872 | p->nbufs = tmp; | |
3873 | p->bgid = READ_ONCE(sqe->buf_group); | |
3874 | return 0; | |
3875 | } | |
3876 | ||
3877 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3878 | int bgid, unsigned nbufs) | |
3879 | { | |
3880 | unsigned i = 0; | |
3881 | ||
3882 | /* shouldn't happen */ | |
3883 | if (!nbufs) | |
3884 | return 0; | |
3885 | ||
3886 | /* the head kbuf is the list itself */ | |
3887 | while (!list_empty(&buf->list)) { | |
3888 | struct io_buffer *nxt; | |
3889 | ||
3890 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3891 | list_del(&nxt->list); | |
3892 | kfree(nxt); | |
3893 | if (++i == nbufs) | |
3894 | return i; | |
3895 | } | |
3896 | i++; | |
3897 | kfree(buf); | |
9e15c3a0 | 3898 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3899 | |
3900 | return i; | |
3901 | } | |
3902 | ||
889fca73 | 3903 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3904 | { |
3905 | struct io_provide_buf *p = &req->pbuf; | |
3906 | struct io_ring_ctx *ctx = req->ctx; | |
3907 | struct io_buffer *head; | |
3908 | int ret = 0; | |
45d189c6 | 3909 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3910 | |
3911 | io_ring_submit_lock(ctx, !force_nonblock); | |
3912 | ||
3913 | lockdep_assert_held(&ctx->uring_lock); | |
3914 | ||
3915 | ret = -ENOENT; | |
9e15c3a0 | 3916 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3917 | if (head) |
3918 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3919 | if (ret < 0) |
3920 | req_set_fail_links(req); | |
067524e9 | 3921 | |
31bff9a5 PB |
3922 | /* need to hold the lock to complete IOPOLL requests */ |
3923 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3924 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3925 | io_ring_submit_unlock(ctx, !force_nonblock); |
3926 | } else { | |
3927 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3928 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3929 | } |
067524e9 JA |
3930 | return 0; |
3931 | } | |
3932 | ||
ddf0322d JA |
3933 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3934 | const struct io_uring_sqe *sqe) | |
3935 | { | |
3936 | struct io_provide_buf *p = &req->pbuf; | |
3937 | u64 tmp; | |
3938 | ||
3939 | if (sqe->ioprio || sqe->rw_flags) | |
3940 | return -EINVAL; | |
3941 | ||
3942 | tmp = READ_ONCE(sqe->fd); | |
3943 | if (!tmp || tmp > USHRT_MAX) | |
3944 | return -E2BIG; | |
3945 | p->nbufs = tmp; | |
3946 | p->addr = READ_ONCE(sqe->addr); | |
3947 | p->len = READ_ONCE(sqe->len); | |
3948 | ||
efe68c1c | 3949 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3950 | return -EFAULT; |
3951 | ||
3952 | p->bgid = READ_ONCE(sqe->buf_group); | |
3953 | tmp = READ_ONCE(sqe->off); | |
3954 | if (tmp > USHRT_MAX) | |
3955 | return -E2BIG; | |
3956 | p->bid = tmp; | |
3957 | return 0; | |
3958 | } | |
3959 | ||
3960 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3961 | { | |
3962 | struct io_buffer *buf; | |
3963 | u64 addr = pbuf->addr; | |
3964 | int i, bid = pbuf->bid; | |
3965 | ||
3966 | for (i = 0; i < pbuf->nbufs; i++) { | |
3967 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3968 | if (!buf) | |
3969 | break; | |
3970 | ||
3971 | buf->addr = addr; | |
3972 | buf->len = pbuf->len; | |
3973 | buf->bid = bid; | |
3974 | addr += pbuf->len; | |
3975 | bid++; | |
3976 | if (!*head) { | |
3977 | INIT_LIST_HEAD(&buf->list); | |
3978 | *head = buf; | |
3979 | } else { | |
3980 | list_add_tail(&buf->list, &(*head)->list); | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | return i ? i : -ENOMEM; | |
3985 | } | |
3986 | ||
889fca73 | 3987 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3988 | { |
3989 | struct io_provide_buf *p = &req->pbuf; | |
3990 | struct io_ring_ctx *ctx = req->ctx; | |
3991 | struct io_buffer *head, *list; | |
3992 | int ret = 0; | |
45d189c6 | 3993 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3994 | |
3995 | io_ring_submit_lock(ctx, !force_nonblock); | |
3996 | ||
3997 | lockdep_assert_held(&ctx->uring_lock); | |
3998 | ||
9e15c3a0 | 3999 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4000 | |
4001 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4002 | if (ret >= 0 && !list) { |
4003 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4004 | if (ret < 0) | |
067524e9 | 4005 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4006 | } |
ddf0322d JA |
4007 | if (ret < 0) |
4008 | req_set_fail_links(req); | |
31bff9a5 PB |
4009 | |
4010 | /* need to hold the lock to complete IOPOLL requests */ | |
4011 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4012 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4013 | io_ring_submit_unlock(ctx, !force_nonblock); |
4014 | } else { | |
4015 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4016 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4017 | } |
ddf0322d | 4018 | return 0; |
cebdb986 JA |
4019 | } |
4020 | ||
3e4827b0 JA |
4021 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4022 | const struct io_uring_sqe *sqe) | |
4023 | { | |
4024 | #if defined(CONFIG_EPOLL) | |
4025 | if (sqe->ioprio || sqe->buf_index) | |
4026 | return -EINVAL; | |
6ca56f84 | 4027 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4028 | return -EINVAL; |
3e4827b0 JA |
4029 | |
4030 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4031 | req->epoll.op = READ_ONCE(sqe->len); | |
4032 | req->epoll.fd = READ_ONCE(sqe->off); | |
4033 | ||
4034 | if (ep_op_has_event(req->epoll.op)) { | |
4035 | struct epoll_event __user *ev; | |
4036 | ||
4037 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4038 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4039 | return -EFAULT; | |
4040 | } | |
4041 | ||
4042 | return 0; | |
4043 | #else | |
4044 | return -EOPNOTSUPP; | |
4045 | #endif | |
4046 | } | |
4047 | ||
889fca73 | 4048 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4049 | { |
4050 | #if defined(CONFIG_EPOLL) | |
4051 | struct io_epoll *ie = &req->epoll; | |
4052 | int ret; | |
45d189c6 | 4053 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4054 | |
4055 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4056 | if (force_nonblock && ret == -EAGAIN) | |
4057 | return -EAGAIN; | |
4058 | ||
4059 | if (ret < 0) | |
4060 | req_set_fail_links(req); | |
889fca73 | 4061 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4062 | return 0; |
4063 | #else | |
4064 | return -EOPNOTSUPP; | |
4065 | #endif | |
4066 | } | |
4067 | ||
c1ca757b JA |
4068 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4069 | { | |
4070 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4071 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4072 | return -EINVAL; | |
3232dd02 PB |
4073 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4074 | return -EINVAL; | |
c1ca757b JA |
4075 | |
4076 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4077 | req->madvise.len = READ_ONCE(sqe->len); | |
4078 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4079 | return 0; | |
4080 | #else | |
4081 | return -EOPNOTSUPP; | |
4082 | #endif | |
4083 | } | |
4084 | ||
45d189c6 | 4085 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4086 | { |
4087 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4088 | struct io_madvise *ma = &req->madvise; | |
4089 | int ret; | |
4090 | ||
45d189c6 | 4091 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4092 | return -EAGAIN; |
4093 | ||
0726b01e | 4094 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4095 | if (ret < 0) |
4096 | req_set_fail_links(req); | |
e1e16097 | 4097 | io_req_complete(req, ret); |
c1ca757b JA |
4098 | return 0; |
4099 | #else | |
4100 | return -EOPNOTSUPP; | |
4101 | #endif | |
4102 | } | |
4103 | ||
4840e418 JA |
4104 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4105 | { | |
4106 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4107 | return -EINVAL; | |
3232dd02 PB |
4108 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4109 | return -EINVAL; | |
4840e418 JA |
4110 | |
4111 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4112 | req->fadvise.len = READ_ONCE(sqe->len); | |
4113 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4114 | return 0; | |
4115 | } | |
4116 | ||
45d189c6 | 4117 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4118 | { |
4119 | struct io_fadvise *fa = &req->fadvise; | |
4120 | int ret; | |
4121 | ||
45d189c6 | 4122 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4123 | switch (fa->advice) { |
4124 | case POSIX_FADV_NORMAL: | |
4125 | case POSIX_FADV_RANDOM: | |
4126 | case POSIX_FADV_SEQUENTIAL: | |
4127 | break; | |
4128 | default: | |
4129 | return -EAGAIN; | |
4130 | } | |
4131 | } | |
4840e418 JA |
4132 | |
4133 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4134 | if (ret < 0) | |
4135 | req_set_fail_links(req); | |
e1e16097 | 4136 | io_req_complete(req, ret); |
4840e418 JA |
4137 | return 0; |
4138 | } | |
4139 | ||
eddc7ef5 JA |
4140 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4141 | { | |
6ca56f84 | 4142 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4143 | return -EINVAL; |
eddc7ef5 JA |
4144 | if (sqe->ioprio || sqe->buf_index) |
4145 | return -EINVAL; | |
9c280f90 | 4146 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4147 | return -EBADF; |
eddc7ef5 | 4148 | |
1d9e1288 BM |
4149 | req->statx.dfd = READ_ONCE(sqe->fd); |
4150 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4151 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4152 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4153 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4154 | |
4155 | return 0; | |
4156 | } | |
4157 | ||
45d189c6 | 4158 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4159 | { |
1d9e1288 | 4160 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4161 | int ret; |
4162 | ||
45d189c6 | 4163 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4164 | /* only need file table for an actual valid fd */ |
4165 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4166 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4167 | return -EAGAIN; |
5b0bbee4 | 4168 | } |
eddc7ef5 | 4169 | |
e62753e4 BM |
4170 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4171 | ctx->buffer); | |
eddc7ef5 | 4172 | |
eddc7ef5 JA |
4173 | if (ret < 0) |
4174 | req_set_fail_links(req); | |
e1e16097 | 4175 | io_req_complete(req, ret); |
eddc7ef5 JA |
4176 | return 0; |
4177 | } | |
4178 | ||
b5dba59e JA |
4179 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4180 | { | |
14587a46 | 4181 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4182 | return -EINVAL; |
b5dba59e JA |
4183 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4184 | sqe->rw_flags || sqe->buf_index) | |
4185 | return -EINVAL; | |
9c280f90 | 4186 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4187 | return -EBADF; |
b5dba59e JA |
4188 | |
4189 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4190 | return 0; |
b5dba59e JA |
4191 | } |
4192 | ||
889fca73 | 4193 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4194 | { |
9eac1904 | 4195 | struct files_struct *files = current->files; |
3af73b28 | 4196 | struct io_close *close = &req->close; |
9eac1904 JA |
4197 | struct fdtable *fdt; |
4198 | struct file *file; | |
b5dba59e JA |
4199 | int ret; |
4200 | ||
9eac1904 JA |
4201 | file = NULL; |
4202 | ret = -EBADF; | |
4203 | spin_lock(&files->file_lock); | |
4204 | fdt = files_fdtable(files); | |
4205 | if (close->fd >= fdt->max_fds) { | |
4206 | spin_unlock(&files->file_lock); | |
4207 | goto err; | |
4208 | } | |
4209 | file = fdt->fd[close->fd]; | |
4210 | if (!file) { | |
4211 | spin_unlock(&files->file_lock); | |
4212 | goto err; | |
4213 | } | |
4214 | ||
4215 | if (file->f_op == &io_uring_fops) { | |
4216 | spin_unlock(&files->file_lock); | |
4217 | file = NULL; | |
4218 | goto err; | |
3af73b28 | 4219 | } |
b5dba59e JA |
4220 | |
4221 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4222 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4223 | spin_unlock(&files->file_lock); |
0bf0eefd | 4224 | return -EAGAIN; |
a2100672 | 4225 | } |
b5dba59e | 4226 | |
9eac1904 JA |
4227 | ret = __close_fd_get_file(close->fd, &file); |
4228 | spin_unlock(&files->file_lock); | |
4229 | if (ret < 0) { | |
4230 | if (ret == -ENOENT) | |
4231 | ret = -EBADF; | |
4232 | goto err; | |
4233 | } | |
4234 | ||
3af73b28 | 4235 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4236 | ret = filp_close(file, current->files); |
4237 | err: | |
3af73b28 PB |
4238 | if (ret < 0) |
4239 | req_set_fail_links(req); | |
9eac1904 JA |
4240 | if (file) |
4241 | fput(file); | |
889fca73 | 4242 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4243 | return 0; |
b5dba59e JA |
4244 | } |
4245 | ||
1155c76a | 4246 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4247 | { |
4248 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4249 | |
5d17b4a4 JA |
4250 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4251 | return -EINVAL; | |
4252 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4253 | return -EINVAL; | |
4254 | ||
8ed8d3c3 JA |
4255 | req->sync.off = READ_ONCE(sqe->off); |
4256 | req->sync.len = READ_ONCE(sqe->len); | |
4257 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4258 | return 0; |
4259 | } | |
4260 | ||
45d189c6 | 4261 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4262 | { |
8ed8d3c3 JA |
4263 | int ret; |
4264 | ||
ac45abc0 | 4265 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4266 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4267 | return -EAGAIN; |
4268 | ||
9adbd45d | 4269 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4270 | req->sync.flags); |
4271 | if (ret < 0) | |
4272 | req_set_fail_links(req); | |
e1e16097 | 4273 | io_req_complete(req, ret); |
5d17b4a4 JA |
4274 | return 0; |
4275 | } | |
4276 | ||
469956e8 | 4277 | #if defined(CONFIG_NET) |
02d27d89 PB |
4278 | static int io_setup_async_msg(struct io_kiocb *req, |
4279 | struct io_async_msghdr *kmsg) | |
4280 | { | |
e8c2bc1f JA |
4281 | struct io_async_msghdr *async_msg = req->async_data; |
4282 | ||
4283 | if (async_msg) | |
02d27d89 | 4284 | return -EAGAIN; |
e8c2bc1f | 4285 | if (io_alloc_async_data(req)) { |
257e84a5 | 4286 | kfree(kmsg->free_iov); |
02d27d89 PB |
4287 | return -ENOMEM; |
4288 | } | |
e8c2bc1f | 4289 | async_msg = req->async_data; |
02d27d89 | 4290 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4291 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4292 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4293 | /* if were using fast_iov, set it to the new one */ |
4294 | if (!async_msg->free_iov) | |
4295 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4296 | ||
02d27d89 PB |
4297 | return -EAGAIN; |
4298 | } | |
4299 | ||
2ae523ed PB |
4300 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4301 | struct io_async_msghdr *iomsg) | |
4302 | { | |
2ae523ed | 4303 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4304 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4305 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4306 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4307 | } |
4308 | ||
93642ef8 PB |
4309 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4310 | { | |
4311 | int ret; | |
4312 | ||
4313 | if (!io_op_defs[req->opcode].needs_async_data) | |
4314 | return 0; | |
4315 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4316 | if (!ret) | |
4317 | req->flags |= REQ_F_NEED_CLEANUP; | |
4318 | return ret; | |
4319 | } | |
4320 | ||
3529d8c2 | 4321 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4322 | { |
e47293fd | 4323 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4324 | |
d2b6f48b PB |
4325 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4326 | return -EINVAL; | |
4327 | ||
e47293fd | 4328 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4329 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4330 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4331 | |
d8768362 JA |
4332 | #ifdef CONFIG_COMPAT |
4333 | if (req->ctx->compat) | |
4334 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4335 | #endif | |
93642ef8 | 4336 | return 0; |
03b1230c JA |
4337 | } |
4338 | ||
889fca73 | 4339 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4340 | { |
6b754c8b | 4341 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4342 | struct socket *sock; |
7a7cacba | 4343 | unsigned flags; |
0fa03c62 JA |
4344 | int ret; |
4345 | ||
dba4a925 | 4346 | sock = sock_from_file(req->file); |
7a7cacba | 4347 | if (unlikely(!sock)) |
dba4a925 | 4348 | return -ENOTSOCK; |
3529d8c2 | 4349 | |
257e84a5 PB |
4350 | kmsg = req->async_data; |
4351 | if (!kmsg) { | |
7a7cacba PB |
4352 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4353 | if (ret) | |
4354 | return ret; | |
4355 | kmsg = &iomsg; | |
0fa03c62 | 4356 | } |
0fa03c62 | 4357 | |
7a7cacba PB |
4358 | flags = req->sr_msg.msg_flags; |
4359 | if (flags & MSG_DONTWAIT) | |
4360 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4361 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4362 | flags |= MSG_DONTWAIT; |
e47293fd | 4363 | |
7a7cacba | 4364 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4365 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4366 | return io_setup_async_msg(req, kmsg); |
4367 | if (ret == -ERESTARTSYS) | |
4368 | ret = -EINTR; | |
0fa03c62 | 4369 | |
257e84a5 PB |
4370 | /* fast path, check for non-NULL to avoid function call */ |
4371 | if (kmsg->free_iov) | |
4372 | kfree(kmsg->free_iov); | |
99bc4c38 | 4373 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4374 | if (ret < 0) |
4375 | req_set_fail_links(req); | |
889fca73 | 4376 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4377 | return 0; |
03b1230c | 4378 | } |
aa1fa28f | 4379 | |
889fca73 | 4380 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4381 | { |
7a7cacba PB |
4382 | struct io_sr_msg *sr = &req->sr_msg; |
4383 | struct msghdr msg; | |
4384 | struct iovec iov; | |
fddaface | 4385 | struct socket *sock; |
7a7cacba | 4386 | unsigned flags; |
fddaface JA |
4387 | int ret; |
4388 | ||
dba4a925 | 4389 | sock = sock_from_file(req->file); |
7a7cacba | 4390 | if (unlikely(!sock)) |
dba4a925 | 4391 | return -ENOTSOCK; |
fddaface | 4392 | |
7a7cacba PB |
4393 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4394 | if (unlikely(ret)) | |
14db8411 | 4395 | return ret; |
fddaface | 4396 | |
7a7cacba PB |
4397 | msg.msg_name = NULL; |
4398 | msg.msg_control = NULL; | |
4399 | msg.msg_controllen = 0; | |
4400 | msg.msg_namelen = 0; | |
fddaface | 4401 | |
7a7cacba PB |
4402 | flags = req->sr_msg.msg_flags; |
4403 | if (flags & MSG_DONTWAIT) | |
4404 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4405 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4406 | flags |= MSG_DONTWAIT; |
fddaface | 4407 | |
7a7cacba PB |
4408 | msg.msg_flags = flags; |
4409 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4410 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4411 | return -EAGAIN; |
4412 | if (ret == -ERESTARTSYS) | |
4413 | ret = -EINTR; | |
fddaface | 4414 | |
fddaface JA |
4415 | if (ret < 0) |
4416 | req_set_fail_links(req); | |
889fca73 | 4417 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4418 | return 0; |
fddaface JA |
4419 | } |
4420 | ||
1400e697 PB |
4421 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4422 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4423 | { |
4424 | struct io_sr_msg *sr = &req->sr_msg; | |
4425 | struct iovec __user *uiov; | |
4426 | size_t iov_len; | |
4427 | int ret; | |
4428 | ||
1400e697 PB |
4429 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4430 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4431 | if (ret) |
4432 | return ret; | |
4433 | ||
4434 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4435 | if (iov_len > 1) | |
4436 | return -EINVAL; | |
5476dfed | 4437 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4438 | return -EFAULT; |
5476dfed | 4439 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4440 | iomsg->free_iov = NULL; |
52de1fe1 | 4441 | } else { |
257e84a5 | 4442 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4443 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4444 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4445 | false); |
52de1fe1 JA |
4446 | if (ret > 0) |
4447 | ret = 0; | |
4448 | } | |
4449 | ||
4450 | return ret; | |
4451 | } | |
4452 | ||
4453 | #ifdef CONFIG_COMPAT | |
4454 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4455 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4456 | { |
4457 | struct compat_msghdr __user *msg_compat; | |
4458 | struct io_sr_msg *sr = &req->sr_msg; | |
4459 | struct compat_iovec __user *uiov; | |
4460 | compat_uptr_t ptr; | |
4461 | compat_size_t len; | |
4462 | int ret; | |
4463 | ||
270a5940 | 4464 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4465 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4466 | &ptr, &len); |
4467 | if (ret) | |
4468 | return ret; | |
4469 | ||
4470 | uiov = compat_ptr(ptr); | |
4471 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4472 | compat_ssize_t clen; | |
4473 | ||
4474 | if (len > 1) | |
4475 | return -EINVAL; | |
4476 | if (!access_ok(uiov, sizeof(*uiov))) | |
4477 | return -EFAULT; | |
4478 | if (__get_user(clen, &uiov->iov_len)) | |
4479 | return -EFAULT; | |
4480 | if (clen < 0) | |
4481 | return -EINVAL; | |
2d280bc8 | 4482 | sr->len = clen; |
257e84a5 | 4483 | iomsg->free_iov = NULL; |
52de1fe1 | 4484 | } else { |
257e84a5 | 4485 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4486 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4487 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4488 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4489 | if (ret < 0) |
4490 | return ret; | |
4491 | } | |
4492 | ||
4493 | return 0; | |
4494 | } | |
4495 | #endif | |
4496 | ||
1400e697 PB |
4497 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4498 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4499 | { |
1400e697 | 4500 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4501 | |
4502 | #ifdef CONFIG_COMPAT | |
4503 | if (req->ctx->compat) | |
1400e697 | 4504 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4505 | #endif |
52de1fe1 | 4506 | |
1400e697 | 4507 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4508 | } |
4509 | ||
bcda7baa | 4510 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4511 | bool needs_lock) |
bcda7baa JA |
4512 | { |
4513 | struct io_sr_msg *sr = &req->sr_msg; | |
4514 | struct io_buffer *kbuf; | |
4515 | ||
bcda7baa JA |
4516 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4517 | if (IS_ERR(kbuf)) | |
4518 | return kbuf; | |
4519 | ||
4520 | sr->kbuf = kbuf; | |
4521 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4522 | return kbuf; |
fddaface JA |
4523 | } |
4524 | ||
7fbb1b54 PB |
4525 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4526 | { | |
4527 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4528 | } | |
4529 | ||
93642ef8 | 4530 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4531 | { |
99bc4c38 | 4532 | int ret; |
3529d8c2 | 4533 | |
93642ef8 PB |
4534 | if (!io_op_defs[req->opcode].needs_async_data) |
4535 | return 0; | |
4536 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4537 | if (!ret) | |
4538 | req->flags |= REQ_F_NEED_CLEANUP; | |
4539 | return ret; | |
4540 | } | |
4541 | ||
4542 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4543 | { | |
4544 | struct io_sr_msg *sr = &req->sr_msg; | |
4545 | ||
d2b6f48b PB |
4546 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4547 | return -EINVAL; | |
4548 | ||
3529d8c2 | 4549 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4550 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4551 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4552 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4553 | |
d8768362 JA |
4554 | #ifdef CONFIG_COMPAT |
4555 | if (req->ctx->compat) | |
4556 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4557 | #endif | |
93642ef8 | 4558 | return 0; |
aa1fa28f JA |
4559 | } |
4560 | ||
889fca73 | 4561 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4562 | { |
6b754c8b | 4563 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4564 | struct socket *sock; |
7fbb1b54 | 4565 | struct io_buffer *kbuf; |
7a7cacba | 4566 | unsigned flags; |
52de1fe1 | 4567 | int ret, cflags = 0; |
45d189c6 | 4568 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4569 | |
dba4a925 | 4570 | sock = sock_from_file(req->file); |
7a7cacba | 4571 | if (unlikely(!sock)) |
dba4a925 | 4572 | return -ENOTSOCK; |
3529d8c2 | 4573 | |
257e84a5 PB |
4574 | kmsg = req->async_data; |
4575 | if (!kmsg) { | |
7a7cacba PB |
4576 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4577 | if (ret) | |
681fda8d | 4578 | return ret; |
7a7cacba PB |
4579 | kmsg = &iomsg; |
4580 | } | |
03b1230c | 4581 | |
bc02ef33 | 4582 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4583 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4584 | if (IS_ERR(kbuf)) |
52de1fe1 | 4585 | return PTR_ERR(kbuf); |
7a7cacba | 4586 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4587 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4588 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4589 | 1, req->sr_msg.len); |
4590 | } | |
52de1fe1 | 4591 | |
7a7cacba PB |
4592 | flags = req->sr_msg.msg_flags; |
4593 | if (flags & MSG_DONTWAIT) | |
4594 | req->flags |= REQ_F_NOWAIT; | |
4595 | else if (force_nonblock) | |
4596 | flags |= MSG_DONTWAIT; | |
e47293fd | 4597 | |
7a7cacba PB |
4598 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4599 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4600 | if (force_nonblock && ret == -EAGAIN) |
4601 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4602 | if (ret == -ERESTARTSYS) |
4603 | ret = -EINTR; | |
03b1230c | 4604 | |
7fbb1b54 PB |
4605 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4606 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4607 | /* fast path, check for non-NULL to avoid function call */ |
4608 | if (kmsg->free_iov) | |
4609 | kfree(kmsg->free_iov); | |
99bc4c38 | 4610 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4611 | if (ret < 0) |
4612 | req_set_fail_links(req); | |
889fca73 | 4613 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4614 | return 0; |
0fa03c62 | 4615 | } |
5d17b4a4 | 4616 | |
889fca73 | 4617 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4618 | { |
6b754c8b | 4619 | struct io_buffer *kbuf; |
7a7cacba PB |
4620 | struct io_sr_msg *sr = &req->sr_msg; |
4621 | struct msghdr msg; | |
4622 | void __user *buf = sr->buf; | |
fddaface | 4623 | struct socket *sock; |
7a7cacba PB |
4624 | struct iovec iov; |
4625 | unsigned flags; | |
bcda7baa | 4626 | int ret, cflags = 0; |
45d189c6 | 4627 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4628 | |
dba4a925 | 4629 | sock = sock_from_file(req->file); |
7a7cacba | 4630 | if (unlikely(!sock)) |
dba4a925 | 4631 | return -ENOTSOCK; |
fddaface | 4632 | |
bc02ef33 | 4633 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4634 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4635 | if (IS_ERR(kbuf)) |
4636 | return PTR_ERR(kbuf); | |
7a7cacba | 4637 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4638 | } |
bcda7baa | 4639 | |
7a7cacba | 4640 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4641 | if (unlikely(ret)) |
4642 | goto out_free; | |
fddaface | 4643 | |
7a7cacba PB |
4644 | msg.msg_name = NULL; |
4645 | msg.msg_control = NULL; | |
4646 | msg.msg_controllen = 0; | |
4647 | msg.msg_namelen = 0; | |
4648 | msg.msg_iocb = NULL; | |
4649 | msg.msg_flags = 0; | |
fddaface | 4650 | |
7a7cacba PB |
4651 | flags = req->sr_msg.msg_flags; |
4652 | if (flags & MSG_DONTWAIT) | |
4653 | req->flags |= REQ_F_NOWAIT; | |
4654 | else if (force_nonblock) | |
4655 | flags |= MSG_DONTWAIT; | |
4656 | ||
4657 | ret = sock_recvmsg(sock, &msg, flags); | |
4658 | if (force_nonblock && ret == -EAGAIN) | |
4659 | return -EAGAIN; | |
4660 | if (ret == -ERESTARTSYS) | |
4661 | ret = -EINTR; | |
14c32eee | 4662 | out_free: |
7fbb1b54 PB |
4663 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4664 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4665 | if (ret < 0) |
4666 | req_set_fail_links(req); | |
889fca73 | 4667 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4668 | return 0; |
fddaface JA |
4669 | } |
4670 | ||
3529d8c2 | 4671 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4672 | { |
8ed8d3c3 JA |
4673 | struct io_accept *accept = &req->accept; |
4674 | ||
14587a46 | 4675 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4676 | return -EINVAL; |
8042d6ce | 4677 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4678 | return -EINVAL; |
4679 | ||
d55e5f5b JA |
4680 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4681 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4682 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4683 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4684 | return 0; |
8ed8d3c3 | 4685 | } |
17f2fe35 | 4686 | |
889fca73 | 4687 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4688 | { |
4689 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4690 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4691 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4692 | int ret; |
4693 | ||
e697deed JX |
4694 | if (req->file->f_flags & O_NONBLOCK) |
4695 | req->flags |= REQ_F_NOWAIT; | |
4696 | ||
8ed8d3c3 | 4697 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4698 | accept->addr_len, accept->flags, |
4699 | accept->nofile); | |
8ed8d3c3 | 4700 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4701 | return -EAGAIN; |
ac45abc0 PB |
4702 | if (ret < 0) { |
4703 | if (ret == -ERESTARTSYS) | |
4704 | ret = -EINTR; | |
4e88d6e7 | 4705 | req_set_fail_links(req); |
ac45abc0 | 4706 | } |
889fca73 | 4707 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4708 | return 0; |
8ed8d3c3 JA |
4709 | } |
4710 | ||
93642ef8 PB |
4711 | static int io_connect_prep_async(struct io_kiocb *req) |
4712 | { | |
4713 | struct io_async_connect *io = req->async_data; | |
4714 | struct io_connect *conn = &req->connect; | |
4715 | ||
4716 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4717 | } | |
4718 | ||
3529d8c2 | 4719 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4720 | { |
3529d8c2 | 4721 | struct io_connect *conn = &req->connect; |
f499a021 | 4722 | |
14587a46 | 4723 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4724 | return -EINVAL; |
4725 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4726 | return -EINVAL; | |
4727 | ||
3529d8c2 JA |
4728 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4729 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4730 | return 0; |
f499a021 JA |
4731 | } |
4732 | ||
889fca73 | 4733 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4734 | { |
e8c2bc1f | 4735 | struct io_async_connect __io, *io; |
f8e85cf2 | 4736 | unsigned file_flags; |
3fbb51c1 | 4737 | int ret; |
45d189c6 | 4738 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4739 | |
e8c2bc1f JA |
4740 | if (req->async_data) { |
4741 | io = req->async_data; | |
f499a021 | 4742 | } else { |
3529d8c2 JA |
4743 | ret = move_addr_to_kernel(req->connect.addr, |
4744 | req->connect.addr_len, | |
e8c2bc1f | 4745 | &__io.address); |
f499a021 JA |
4746 | if (ret) |
4747 | goto out; | |
4748 | io = &__io; | |
4749 | } | |
4750 | ||
3fbb51c1 JA |
4751 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4752 | ||
e8c2bc1f | 4753 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4754 | req->connect.addr_len, file_flags); |
87f80d62 | 4755 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4756 | if (req->async_data) |
b7bb4f7d | 4757 | return -EAGAIN; |
e8c2bc1f | 4758 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4759 | ret = -ENOMEM; |
4760 | goto out; | |
4761 | } | |
e8c2bc1f JA |
4762 | io = req->async_data; |
4763 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4764 | return -EAGAIN; |
f499a021 | 4765 | } |
f8e85cf2 JA |
4766 | if (ret == -ERESTARTSYS) |
4767 | ret = -EINTR; | |
f499a021 | 4768 | out: |
4e88d6e7 JA |
4769 | if (ret < 0) |
4770 | req_set_fail_links(req); | |
889fca73 | 4771 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4772 | return 0; |
469956e8 Y |
4773 | } |
4774 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4775 | #define IO_NETOP_FN(op) \ |
4776 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4777 | { \ | |
4778 | return -EOPNOTSUPP; \ | |
4779 | } | |
4780 | ||
4781 | #define IO_NETOP_PREP(op) \ | |
4782 | IO_NETOP_FN(op) \ | |
4783 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4784 | { \ | |
4785 | return -EOPNOTSUPP; \ | |
4786 | } \ | |
4787 | ||
4788 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4789 | IO_NETOP_PREP(op) \ | |
4790 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4791 | { \ | |
4792 | return -EOPNOTSUPP; \ | |
4793 | } | |
4794 | ||
4795 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4796 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4797 | IO_NETOP_PREP_ASYNC(connect); | |
4798 | IO_NETOP_PREP(accept); | |
4799 | IO_NETOP_FN(send); | |
4800 | IO_NETOP_FN(recv); | |
469956e8 | 4801 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4802 | |
d7718a9d JA |
4803 | struct io_poll_table { |
4804 | struct poll_table_struct pt; | |
4805 | struct io_kiocb *req; | |
4806 | int error; | |
4807 | }; | |
ce593a6c | 4808 | |
d7718a9d JA |
4809 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4810 | __poll_t mask, task_work_func_t func) | |
4811 | { | |
aa96bf8a | 4812 | int ret; |
d7718a9d JA |
4813 | |
4814 | /* for instances that support it check for an event match first: */ | |
4815 | if (mask && !(mask & poll->events)) | |
4816 | return 0; | |
4817 | ||
4818 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4819 | ||
4820 | list_del_init(&poll->wait.entry); | |
4821 | ||
d7718a9d | 4822 | req->result = mask; |
7cbf1722 | 4823 | req->task_work.func = func; |
6d816e08 JA |
4824 | percpu_ref_get(&req->ctx->refs); |
4825 | ||
d7718a9d | 4826 | /* |
e3aabf95 JA |
4827 | * If this fails, then the task is exiting. When a task exits, the |
4828 | * work gets canceled, so just cancel this request as well instead | |
4829 | * of executing it. We can't safely execute it anyway, as we may not | |
4830 | * have the needed state needed for it anyway. | |
d7718a9d | 4831 | */ |
355fb9e2 | 4832 | ret = io_req_task_work_add(req); |
aa96bf8a | 4833 | if (unlikely(ret)) { |
e3aabf95 | 4834 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4835 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4836 | } |
d7718a9d JA |
4837 | return 1; |
4838 | } | |
4839 | ||
74ce6ce4 JA |
4840 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4841 | __acquires(&req->ctx->completion_lock) | |
4842 | { | |
4843 | struct io_ring_ctx *ctx = req->ctx; | |
4844 | ||
4845 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4846 | struct poll_table_struct pt = { ._key = poll->events }; | |
4847 | ||
4848 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4849 | } | |
4850 | ||
4851 | spin_lock_irq(&ctx->completion_lock); | |
4852 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4853 | add_wait_queue(poll->head, &poll->wait); | |
4854 | return true; | |
4855 | } | |
4856 | ||
4857 | return false; | |
4858 | } | |
4859 | ||
d4e7cd36 | 4860 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4861 | { |
e8c2bc1f | 4862 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4863 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4864 | return req->async_data; |
d4e7cd36 JA |
4865 | return req->apoll->double_poll; |
4866 | } | |
4867 | ||
4868 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4869 | { | |
4870 | if (req->opcode == IORING_OP_POLL_ADD) | |
4871 | return &req->poll; | |
4872 | return &req->apoll->poll; | |
4873 | } | |
4874 | ||
4875 | static void io_poll_remove_double(struct io_kiocb *req) | |
4876 | { | |
4877 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4878 | |
4879 | lockdep_assert_held(&req->ctx->completion_lock); | |
4880 | ||
4881 | if (poll && poll->head) { | |
4882 | struct wait_queue_head *head = poll->head; | |
4883 | ||
4884 | spin_lock(&head->lock); | |
4885 | list_del_init(&poll->wait.entry); | |
4886 | if (poll->wait.private) | |
4887 | refcount_dec(&req->refs); | |
4888 | poll->head = NULL; | |
4889 | spin_unlock(&head->lock); | |
4890 | } | |
4891 | } | |
4892 | ||
4893 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4894 | { | |
4895 | struct io_ring_ctx *ctx = req->ctx; | |
4896 | ||
d4e7cd36 | 4897 | io_poll_remove_double(req); |
18bceab1 JA |
4898 | req->poll.done = true; |
4899 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4900 | io_commit_cqring(ctx); | |
4901 | } | |
4902 | ||
dd221f46 | 4903 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4904 | { |
dd221f46 | 4905 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4906 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4907 | struct io_kiocb *nxt; |
18bceab1 JA |
4908 | |
4909 | if (io_poll_rewait(req, &req->poll)) { | |
4910 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4911 | } else { |
4912 | hash_del(&req->hash_node); | |
4913 | io_poll_complete(req, req->result, 0); | |
4914 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4915 | |
dd221f46 PB |
4916 | nxt = io_put_req_find_next(req); |
4917 | io_cqring_ev_posted(ctx); | |
4918 | if (nxt) | |
4919 | __io_req_task_submit(nxt); | |
4920 | } | |
18bceab1 | 4921 | |
6d816e08 | 4922 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4923 | } |
4924 | ||
4925 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4926 | int sync, void *key) | |
4927 | { | |
4928 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4929 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4930 | __poll_t mask = key_to_poll(key); |
4931 | ||
4932 | /* for instances that support it check for an event match first: */ | |
4933 | if (mask && !(mask & poll->events)) | |
4934 | return 0; | |
4935 | ||
8706e04e JA |
4936 | list_del_init(&wait->entry); |
4937 | ||
807abcb0 | 4938 | if (poll && poll->head) { |
18bceab1 JA |
4939 | bool done; |
4940 | ||
807abcb0 JA |
4941 | spin_lock(&poll->head->lock); |
4942 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4943 | if (!done) |
807abcb0 | 4944 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4945 | /* make sure double remove sees this as being gone */ |
4946 | wait->private = NULL; | |
807abcb0 | 4947 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4948 | if (!done) { |
4949 | /* use wait func handler, so it matches the rq type */ | |
4950 | poll->wait.func(&poll->wait, mode, sync, key); | |
4951 | } | |
18bceab1 JA |
4952 | } |
4953 | refcount_dec(&req->refs); | |
4954 | return 1; | |
4955 | } | |
4956 | ||
4957 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4958 | wait_queue_func_t wake_func) | |
4959 | { | |
4960 | poll->head = NULL; | |
4961 | poll->done = false; | |
4962 | poll->canceled = false; | |
4963 | poll->events = events; | |
4964 | INIT_LIST_HEAD(&poll->wait.entry); | |
4965 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4966 | } | |
4967 | ||
4968 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4969 | struct wait_queue_head *head, |
4970 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4971 | { |
4972 | struct io_kiocb *req = pt->req; | |
4973 | ||
4974 | /* | |
4975 | * If poll->head is already set, it's because the file being polled | |
4976 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4977 | * for write). Setup a separate io_poll_iocb if this happens. | |
4978 | */ | |
4979 | if (unlikely(poll->head)) { | |
58852d4d PB |
4980 | struct io_poll_iocb *poll_one = poll; |
4981 | ||
18bceab1 | 4982 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4983 | if (*poll_ptr) { |
18bceab1 JA |
4984 | pt->error = -EINVAL; |
4985 | return; | |
4986 | } | |
1c3b3e65 JA |
4987 | /* double add on the same waitqueue head, ignore */ |
4988 | if (poll->head == head) | |
4989 | return; | |
18bceab1 JA |
4990 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4991 | if (!poll) { | |
4992 | pt->error = -ENOMEM; | |
4993 | return; | |
4994 | } | |
58852d4d | 4995 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4996 | refcount_inc(&req->refs); |
4997 | poll->wait.private = req; | |
807abcb0 | 4998 | *poll_ptr = poll; |
18bceab1 JA |
4999 | } |
5000 | ||
5001 | pt->error = 0; | |
5002 | poll->head = head; | |
a31eb4a2 JX |
5003 | |
5004 | if (poll->events & EPOLLEXCLUSIVE) | |
5005 | add_wait_queue_exclusive(head, &poll->wait); | |
5006 | else | |
5007 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5008 | } |
5009 | ||
5010 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5011 | struct poll_table_struct *p) | |
5012 | { | |
5013 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5014 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5015 | |
807abcb0 | 5016 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5017 | } |
5018 | ||
d7718a9d JA |
5019 | static void io_async_task_func(struct callback_head *cb) |
5020 | { | |
5021 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5022 | struct async_poll *apoll = req->apoll; | |
5023 | struct io_ring_ctx *ctx = req->ctx; | |
5024 | ||
5025 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5026 | ||
74ce6ce4 | 5027 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5028 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5029 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5030 | return; |
d7718a9d JA |
5031 | } |
5032 | ||
31067255 | 5033 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5034 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5035 | hash_del(&req->hash_node); |
2bae047e | 5036 | |
d4e7cd36 | 5037 | io_poll_remove_double(req); |
74ce6ce4 JA |
5038 | spin_unlock_irq(&ctx->completion_lock); |
5039 | ||
0be0b0e3 PB |
5040 | if (!READ_ONCE(apoll->poll.canceled)) |
5041 | __io_req_task_submit(req); | |
5042 | else | |
5043 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5044 | |
6d816e08 | 5045 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5046 | kfree(apoll->double_poll); |
31067255 | 5047 | kfree(apoll); |
d7718a9d JA |
5048 | } |
5049 | ||
5050 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5051 | void *key) | |
5052 | { | |
5053 | struct io_kiocb *req = wait->private; | |
5054 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5055 | ||
5056 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5057 | key_to_poll(key)); | |
5058 | ||
5059 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5060 | } | |
5061 | ||
5062 | static void io_poll_req_insert(struct io_kiocb *req) | |
5063 | { | |
5064 | struct io_ring_ctx *ctx = req->ctx; | |
5065 | struct hlist_head *list; | |
5066 | ||
5067 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5068 | hlist_add_head(&req->hash_node, list); | |
5069 | } | |
5070 | ||
5071 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5072 | struct io_poll_iocb *poll, | |
5073 | struct io_poll_table *ipt, __poll_t mask, | |
5074 | wait_queue_func_t wake_func) | |
5075 | __acquires(&ctx->completion_lock) | |
5076 | { | |
5077 | struct io_ring_ctx *ctx = req->ctx; | |
5078 | bool cancel = false; | |
5079 | ||
4d52f338 | 5080 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5081 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5082 | poll->file = req->file; |
18bceab1 | 5083 | poll->wait.private = req; |
d7718a9d JA |
5084 | |
5085 | ipt->pt._key = mask; | |
5086 | ipt->req = req; | |
5087 | ipt->error = -EINVAL; | |
5088 | ||
d7718a9d JA |
5089 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5090 | ||
5091 | spin_lock_irq(&ctx->completion_lock); | |
5092 | if (likely(poll->head)) { | |
5093 | spin_lock(&poll->head->lock); | |
5094 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5095 | if (ipt->error) | |
5096 | cancel = true; | |
5097 | ipt->error = 0; | |
5098 | mask = 0; | |
5099 | } | |
5100 | if (mask || ipt->error) | |
5101 | list_del_init(&poll->wait.entry); | |
5102 | else if (cancel) | |
5103 | WRITE_ONCE(poll->canceled, true); | |
5104 | else if (!poll->done) /* actually waiting for an event */ | |
5105 | io_poll_req_insert(req); | |
5106 | spin_unlock(&poll->head->lock); | |
5107 | } | |
5108 | ||
5109 | return mask; | |
5110 | } | |
5111 | ||
5112 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5113 | { | |
5114 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5115 | struct io_ring_ctx *ctx = req->ctx; | |
5116 | struct async_poll *apoll; | |
5117 | struct io_poll_table ipt; | |
5118 | __poll_t mask, ret; | |
9dab14b8 | 5119 | int rw; |
d7718a9d JA |
5120 | |
5121 | if (!req->file || !file_can_poll(req->file)) | |
5122 | return false; | |
24c74678 | 5123 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5124 | return false; |
9dab14b8 JA |
5125 | if (def->pollin) |
5126 | rw = READ; | |
5127 | else if (def->pollout) | |
5128 | rw = WRITE; | |
5129 | else | |
5130 | return false; | |
5131 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5132 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5133 | return false; |
5134 | ||
5135 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5136 | if (unlikely(!apoll)) | |
5137 | return false; | |
807abcb0 | 5138 | apoll->double_poll = NULL; |
d7718a9d JA |
5139 | |
5140 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5141 | req->apoll = apoll; |
d7718a9d | 5142 | |
8755d97a | 5143 | mask = 0; |
d7718a9d | 5144 | if (def->pollin) |
8755d97a | 5145 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5146 | if (def->pollout) |
5147 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5148 | |
5149 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5150 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5151 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5152 | mask &= ~POLLIN; | |
5153 | ||
d7718a9d JA |
5154 | mask |= POLLERR | POLLPRI; |
5155 | ||
5156 | ipt.pt._qproc = io_async_queue_proc; | |
5157 | ||
5158 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5159 | io_async_wake); | |
a36da65c | 5160 | if (ret || ipt.error) { |
d4e7cd36 | 5161 | io_poll_remove_double(req); |
d7718a9d | 5162 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5163 | kfree(apoll->double_poll); |
d7718a9d JA |
5164 | kfree(apoll); |
5165 | return false; | |
5166 | } | |
5167 | spin_unlock_irq(&ctx->completion_lock); | |
5168 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5169 | apoll->poll.events); | |
5170 | return true; | |
5171 | } | |
5172 | ||
5173 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5174 | struct io_poll_iocb *poll) | |
221c5eb2 | 5175 | { |
b41e9852 | 5176 | bool do_complete = false; |
221c5eb2 JA |
5177 | |
5178 | spin_lock(&poll->head->lock); | |
5179 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5180 | if (!list_empty(&poll->wait.entry)) { |
5181 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5182 | do_complete = true; |
221c5eb2 JA |
5183 | } |
5184 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5185 | hash_del(&req->hash_node); |
d7718a9d JA |
5186 | return do_complete; |
5187 | } | |
5188 | ||
5189 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5190 | { | |
5191 | bool do_complete; | |
5192 | ||
d4e7cd36 JA |
5193 | io_poll_remove_double(req); |
5194 | ||
d7718a9d JA |
5195 | if (req->opcode == IORING_OP_POLL_ADD) { |
5196 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5197 | } else { | |
3bfa5bcb JA |
5198 | struct async_poll *apoll = req->apoll; |
5199 | ||
d7718a9d | 5200 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5201 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5202 | if (do_complete) { | |
d7718a9d | 5203 | io_put_req(req); |
807abcb0 | 5204 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5205 | kfree(apoll); |
5206 | } | |
b1f573bd XW |
5207 | } |
5208 | ||
b41e9852 JA |
5209 | if (do_complete) { |
5210 | io_cqring_fill_event(req, -ECANCELED); | |
5211 | io_commit_cqring(req->ctx); | |
f254ac04 | 5212 | req_set_fail_links(req); |
216578e5 | 5213 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5214 | } |
5215 | ||
5216 | return do_complete; | |
221c5eb2 JA |
5217 | } |
5218 | ||
76e1b642 JA |
5219 | /* |
5220 | * Returns true if we found and killed one or more poll requests | |
5221 | */ | |
6b81928d PB |
5222 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5223 | struct files_struct *files) | |
221c5eb2 | 5224 | { |
78076bb6 | 5225 | struct hlist_node *tmp; |
221c5eb2 | 5226 | struct io_kiocb *req; |
8e2e1faf | 5227 | int posted = 0, i; |
221c5eb2 JA |
5228 | |
5229 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5230 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5231 | struct hlist_head *list; | |
5232 | ||
5233 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5234 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5235 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5236 | posted += io_poll_remove_one(req); |
5237 | } | |
221c5eb2 JA |
5238 | } |
5239 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5240 | |
8e2e1faf JA |
5241 | if (posted) |
5242 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5243 | |
5244 | return posted != 0; | |
221c5eb2 JA |
5245 | } |
5246 | ||
47f46768 JA |
5247 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5248 | { | |
78076bb6 | 5249 | struct hlist_head *list; |
47f46768 JA |
5250 | struct io_kiocb *req; |
5251 | ||
78076bb6 JA |
5252 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5253 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5254 | if (sqe_addr != req->user_data) |
5255 | continue; | |
5256 | if (io_poll_remove_one(req)) | |
eac406c6 | 5257 | return 0; |
b41e9852 | 5258 | return -EALREADY; |
47f46768 JA |
5259 | } |
5260 | ||
5261 | return -ENOENT; | |
5262 | } | |
5263 | ||
3529d8c2 JA |
5264 | static int io_poll_remove_prep(struct io_kiocb *req, |
5265 | const struct io_uring_sqe *sqe) | |
0969e783 | 5266 | { |
0969e783 JA |
5267 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5268 | return -EINVAL; | |
5269 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5270 | sqe->poll_events) | |
5271 | return -EINVAL; | |
5272 | ||
018043be | 5273 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5274 | return 0; |
5275 | } | |
5276 | ||
221c5eb2 JA |
5277 | /* |
5278 | * Find a running poll command that matches one specified in sqe->addr, | |
5279 | * and remove it if found. | |
5280 | */ | |
61e98203 | 5281 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5282 | { |
5283 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5284 | int ret; |
221c5eb2 | 5285 | |
221c5eb2 | 5286 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5287 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5288 | spin_unlock_irq(&ctx->completion_lock); |
5289 | ||
4e88d6e7 JA |
5290 | if (ret < 0) |
5291 | req_set_fail_links(req); | |
e1e16097 | 5292 | io_req_complete(req, ret); |
221c5eb2 JA |
5293 | return 0; |
5294 | } | |
5295 | ||
221c5eb2 JA |
5296 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5297 | void *key) | |
5298 | { | |
c2f2eb7d JA |
5299 | struct io_kiocb *req = wait->private; |
5300 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5301 | |
d7718a9d | 5302 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5303 | } |
5304 | ||
221c5eb2 JA |
5305 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5306 | struct poll_table_struct *p) | |
5307 | { | |
5308 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5309 | ||
e8c2bc1f | 5310 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5311 | } |
5312 | ||
3529d8c2 | 5313 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5314 | { |
5315 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5316 | u32 events; |
221c5eb2 JA |
5317 | |
5318 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5319 | return -EINVAL; | |
5320 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5321 | return -EINVAL; | |
5322 | ||
5769a351 JX |
5323 | events = READ_ONCE(sqe->poll32_events); |
5324 | #ifdef __BIG_ENDIAN | |
5325 | events = swahw32(events); | |
5326 | #endif | |
a31eb4a2 JX |
5327 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5328 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5329 | return 0; |
5330 | } | |
5331 | ||
61e98203 | 5332 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5333 | { |
5334 | struct io_poll_iocb *poll = &req->poll; | |
5335 | struct io_ring_ctx *ctx = req->ctx; | |
5336 | struct io_poll_table ipt; | |
0969e783 | 5337 | __poll_t mask; |
0969e783 | 5338 | |
d7718a9d | 5339 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5340 | |
d7718a9d JA |
5341 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5342 | io_poll_wake); | |
221c5eb2 | 5343 | |
8c838788 | 5344 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5345 | ipt.error = 0; |
b0dd8a41 | 5346 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5347 | } |
221c5eb2 JA |
5348 | spin_unlock_irq(&ctx->completion_lock); |
5349 | ||
8c838788 JA |
5350 | if (mask) { |
5351 | io_cqring_ev_posted(ctx); | |
014db007 | 5352 | io_put_req(req); |
221c5eb2 | 5353 | } |
8c838788 | 5354 | return ipt.error; |
221c5eb2 JA |
5355 | } |
5356 | ||
5262f567 JA |
5357 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5358 | { | |
ad8a48ac JA |
5359 | struct io_timeout_data *data = container_of(timer, |
5360 | struct io_timeout_data, timer); | |
5361 | struct io_kiocb *req = data->req; | |
5362 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5363 | unsigned long flags; |
5364 | ||
5262f567 | 5365 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5366 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5367 | atomic_set(&req->ctx->cq_timeouts, |
5368 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5369 | ||
78e19bbe | 5370 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5371 | io_commit_cqring(ctx); |
5372 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5373 | ||
5374 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5375 | req_set_fail_links(req); |
5262f567 JA |
5376 | io_put_req(req); |
5377 | return HRTIMER_NORESTART; | |
5378 | } | |
5379 | ||
fbd15848 PB |
5380 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5381 | __u64 user_data) | |
f254ac04 | 5382 | { |
fbd15848 | 5383 | struct io_timeout_data *io; |
47f46768 JA |
5384 | struct io_kiocb *req; |
5385 | int ret = -ENOENT; | |
f254ac04 | 5386 | |
135fcde8 | 5387 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5388 | if (user_data == req->user_data) { |
47f46768 JA |
5389 | ret = 0; |
5390 | break; | |
5391 | } | |
5392 | } | |
5393 | ||
5394 | if (ret == -ENOENT) | |
fbd15848 PB |
5395 | return ERR_PTR(ret); |
5396 | ||
5397 | io = req->async_data; | |
e8c2bc1f | 5398 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5399 | if (ret == -1) |
fbd15848 | 5400 | return ERR_PTR(-EALREADY); |
a71976f3 | 5401 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5402 | return req; |
5403 | } | |
47f46768 | 5404 | |
fbd15848 PB |
5405 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5406 | { | |
5407 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5408 | ||
5409 | if (IS_ERR(req)) | |
5410 | return PTR_ERR(req); | |
f254ac04 JA |
5411 | |
5412 | req_set_fail_links(req); | |
f254ac04 | 5413 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5414 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5415 | return 0; |
5416 | } | |
5417 | ||
9c8e11b3 PB |
5418 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5419 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5420 | { |
9c8e11b3 PB |
5421 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5422 | struct io_timeout_data *data; | |
47f46768 | 5423 | |
9c8e11b3 PB |
5424 | if (IS_ERR(req)) |
5425 | return PTR_ERR(req); | |
47f46768 | 5426 | |
9c8e11b3 PB |
5427 | req->timeout.off = 0; /* noseq */ |
5428 | data = req->async_data; | |
5429 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5430 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5431 | data->timer.function = io_timeout_fn; | |
5432 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5433 | return 0; | |
47f46768 JA |
5434 | } |
5435 | ||
3529d8c2 JA |
5436 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5437 | const struct io_uring_sqe *sqe) | |
b29472ee | 5438 | { |
9c8e11b3 PB |
5439 | struct io_timeout_rem *tr = &req->timeout_rem; |
5440 | ||
b29472ee JA |
5441 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5442 | return -EINVAL; | |
61710e43 DA |
5443 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5444 | return -EINVAL; | |
9c8e11b3 | 5445 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5446 | return -EINVAL; |
5447 | ||
9c8e11b3 PB |
5448 | tr->addr = READ_ONCE(sqe->addr); |
5449 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5450 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5451 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5452 | return -EINVAL; | |
5453 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5454 | return -EFAULT; | |
5455 | } else if (tr->flags) { | |
5456 | /* timeout removal doesn't support flags */ | |
b29472ee | 5457 | return -EINVAL; |
9c8e11b3 | 5458 | } |
b29472ee | 5459 | |
b29472ee JA |
5460 | return 0; |
5461 | } | |
5462 | ||
8662daec PB |
5463 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5464 | { | |
5465 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5466 | : HRTIMER_MODE_REL; | |
5467 | } | |
5468 | ||
11365043 JA |
5469 | /* |
5470 | * Remove or update an existing timeout command | |
5471 | */ | |
61e98203 | 5472 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5473 | { |
9c8e11b3 | 5474 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5475 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5476 | int ret; |
11365043 | 5477 | |
11365043 | 5478 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5479 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5480 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5481 | else |
5482 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5483 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5484 | |
47f46768 | 5485 | io_cqring_fill_event(req, ret); |
11365043 JA |
5486 | io_commit_cqring(ctx); |
5487 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5488 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5489 | if (ret < 0) |
5490 | req_set_fail_links(req); | |
ec9c02ad | 5491 | io_put_req(req); |
11365043 | 5492 | return 0; |
5262f567 JA |
5493 | } |
5494 | ||
3529d8c2 | 5495 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5496 | bool is_timeout_link) |
5262f567 | 5497 | { |
ad8a48ac | 5498 | struct io_timeout_data *data; |
a41525ab | 5499 | unsigned flags; |
56080b02 | 5500 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5501 | |
ad8a48ac | 5502 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5503 | return -EINVAL; |
ad8a48ac | 5504 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5505 | return -EINVAL; |
56080b02 | 5506 | if (off && is_timeout_link) |
2d28390a | 5507 | return -EINVAL; |
a41525ab JA |
5508 | flags = READ_ONCE(sqe->timeout_flags); |
5509 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5510 | return -EINVAL; |
bdf20073 | 5511 | |
bfe68a22 | 5512 | req->timeout.off = off; |
26a61679 | 5513 | |
e8c2bc1f | 5514 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5515 | return -ENOMEM; |
5516 | ||
e8c2bc1f | 5517 | data = req->async_data; |
ad8a48ac | 5518 | data->req = req; |
ad8a48ac JA |
5519 | |
5520 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5521 | return -EFAULT; |
5522 | ||
8662daec | 5523 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5524 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
dd59a3d5 | 5525 | io_req_track_inflight(req); |
ad8a48ac JA |
5526 | return 0; |
5527 | } | |
5528 | ||
61e98203 | 5529 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5530 | { |
ad8a48ac | 5531 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5532 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5533 | struct list_head *entry; |
bfe68a22 | 5534 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5535 | |
733f5c95 | 5536 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5537 | |
5262f567 JA |
5538 | /* |
5539 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5540 | * timeout event to be satisfied. If it isn't set, then this is |
5541 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5542 | */ |
8eb7e2d0 | 5543 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5544 | entry = ctx->timeout_list.prev; |
5545 | goto add; | |
5546 | } | |
5262f567 | 5547 | |
bfe68a22 PB |
5548 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5549 | req->timeout.target_seq = tail + off; | |
5262f567 | 5550 | |
f010505b MDG |
5551 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5552 | * This is safe because ->completion_lock is held, and submissions | |
5553 | * and completions are never mixed in the same ->completion_lock section. | |
5554 | */ | |
5555 | ctx->cq_last_tm_flush = tail; | |
5556 | ||
5262f567 JA |
5557 | /* |
5558 | * Insertion sort, ensuring the first entry in the list is always | |
5559 | * the one we need first. | |
5560 | */ | |
5262f567 | 5561 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5562 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5563 | timeout.list); | |
5262f567 | 5564 | |
8eb7e2d0 | 5565 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5566 | continue; |
bfe68a22 PB |
5567 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5568 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5569 | break; |
5570 | } | |
93bd25bb | 5571 | add: |
135fcde8 | 5572 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5573 | data->timer.function = io_timeout_fn; |
5574 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5575 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5576 | return 0; |
5577 | } | |
5262f567 | 5578 | |
f458dd84 PB |
5579 | struct io_cancel_data { |
5580 | struct io_ring_ctx *ctx; | |
5581 | u64 user_data; | |
5582 | }; | |
5583 | ||
62755e35 JA |
5584 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5585 | { | |
5586 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5587 | struct io_cancel_data *cd = data; |
62755e35 | 5588 | |
f458dd84 | 5589 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5590 | } |
5591 | ||
f458dd84 PB |
5592 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5593 | struct io_ring_ctx *ctx) | |
62755e35 | 5594 | { |
f458dd84 | 5595 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5596 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5597 | int ret = 0; |
5598 | ||
f458dd84 | 5599 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5600 | return -ENOENT; |
5601 | ||
f458dd84 | 5602 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5603 | switch (cancel_ret) { |
5604 | case IO_WQ_CANCEL_OK: | |
5605 | ret = 0; | |
5606 | break; | |
5607 | case IO_WQ_CANCEL_RUNNING: | |
5608 | ret = -EALREADY; | |
5609 | break; | |
5610 | case IO_WQ_CANCEL_NOTFOUND: | |
5611 | ret = -ENOENT; | |
5612 | break; | |
5613 | } | |
5614 | ||
e977d6d3 JA |
5615 | return ret; |
5616 | } | |
5617 | ||
47f46768 JA |
5618 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5619 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5620 | int success_ret) |
47f46768 JA |
5621 | { |
5622 | unsigned long flags; | |
5623 | int ret; | |
5624 | ||
f458dd84 | 5625 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5626 | if (ret != -ENOENT) { |
5627 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5628 | goto done; | |
5629 | } | |
5630 | ||
5631 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5632 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5633 | if (ret != -ENOENT) | |
5634 | goto done; | |
5635 | ret = io_poll_cancel(ctx, sqe_addr); | |
5636 | done: | |
b0dd8a41 JA |
5637 | if (!ret) |
5638 | ret = success_ret; | |
47f46768 JA |
5639 | io_cqring_fill_event(req, ret); |
5640 | io_commit_cqring(ctx); | |
5641 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5642 | io_cqring_ev_posted(ctx); | |
5643 | ||
4e88d6e7 JA |
5644 | if (ret < 0) |
5645 | req_set_fail_links(req); | |
014db007 | 5646 | io_put_req(req); |
47f46768 JA |
5647 | } |
5648 | ||
3529d8c2 JA |
5649 | static int io_async_cancel_prep(struct io_kiocb *req, |
5650 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5651 | { |
fbf23849 | 5652 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5653 | return -EINVAL; |
61710e43 DA |
5654 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5655 | return -EINVAL; | |
5656 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5657 | return -EINVAL; |
5658 | ||
fbf23849 JA |
5659 | req->cancel.addr = READ_ONCE(sqe->addr); |
5660 | return 0; | |
5661 | } | |
5662 | ||
61e98203 | 5663 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5664 | { |
5665 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5666 | u64 sqe_addr = req->cancel.addr; |
5667 | struct io_tctx_node *node; | |
5668 | int ret; | |
5669 | ||
5670 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5671 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5672 | spin_lock_irq(&ctx->completion_lock); | |
5673 | if (ret != -ENOENT) | |
5674 | goto done; | |
5675 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5676 | if (ret != -ENOENT) | |
5677 | goto done; | |
5678 | ret = io_poll_cancel(ctx, sqe_addr); | |
5679 | if (ret != -ENOENT) | |
5680 | goto done; | |
5681 | spin_unlock_irq(&ctx->completion_lock); | |
5682 | ||
5683 | /* slow path, try all io-wq's */ | |
5684 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5685 | ret = -ENOENT; | |
5686 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5687 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5688 | |
58f99373 PB |
5689 | if (!tctx || !tctx->io_wq) |
5690 | continue; | |
5691 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5692 | if (ret != -ENOENT) | |
5693 | break; | |
5694 | } | |
5695 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5696 | ||
5697 | spin_lock_irq(&ctx->completion_lock); | |
5698 | done: | |
5699 | io_cqring_fill_event(req, ret); | |
5700 | io_commit_cqring(ctx); | |
5701 | spin_unlock_irq(&ctx->completion_lock); | |
5702 | io_cqring_ev_posted(ctx); | |
5703 | ||
5704 | if (ret < 0) | |
5705 | req_set_fail_links(req); | |
5706 | io_put_req(req); | |
5262f567 JA |
5707 | return 0; |
5708 | } | |
5709 | ||
269bbe5f | 5710 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5711 | const struct io_uring_sqe *sqe) |
5712 | { | |
6ca56f84 JA |
5713 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5714 | return -EINVAL; | |
61710e43 DA |
5715 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5716 | return -EINVAL; | |
5717 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5718 | return -EINVAL; |
5719 | ||
269bbe5f BM |
5720 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5721 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5722 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5723 | return -EINVAL; |
269bbe5f | 5724 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5725 | return 0; |
5726 | } | |
5727 | ||
889fca73 | 5728 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5729 | { |
5730 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5731 | struct io_uring_rsrc_update up; |
05f3fb3c | 5732 | int ret; |
fbf23849 | 5733 | |
45d189c6 | 5734 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5735 | return -EAGAIN; |
05f3fb3c | 5736 | |
269bbe5f BM |
5737 | up.offset = req->rsrc_update.offset; |
5738 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5739 | |
5740 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5741 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5742 | mutex_unlock(&ctx->uring_lock); |
5743 | ||
5744 | if (ret < 0) | |
5745 | req_set_fail_links(req); | |
889fca73 | 5746 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5747 | return 0; |
5748 | } | |
5749 | ||
bfe76559 | 5750 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5751 | { |
d625c6ee | 5752 | switch (req->opcode) { |
e781573e | 5753 | case IORING_OP_NOP: |
bfe76559 | 5754 | return 0; |
f67676d1 JA |
5755 | case IORING_OP_READV: |
5756 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5757 | case IORING_OP_READ: |
bfe76559 | 5758 | return io_read_prep(req, sqe); |
f67676d1 JA |
5759 | case IORING_OP_WRITEV: |
5760 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5761 | case IORING_OP_WRITE: |
bfe76559 | 5762 | return io_write_prep(req, sqe); |
0969e783 | 5763 | case IORING_OP_POLL_ADD: |
bfe76559 | 5764 | return io_poll_add_prep(req, sqe); |
0969e783 | 5765 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5766 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5767 | case IORING_OP_FSYNC: |
1155c76a | 5768 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5769 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5770 | return io_sfr_prep(req, sqe); |
03b1230c | 5771 | case IORING_OP_SENDMSG: |
fddaface | 5772 | case IORING_OP_SEND: |
bfe76559 | 5773 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5774 | case IORING_OP_RECVMSG: |
fddaface | 5775 | case IORING_OP_RECV: |
bfe76559 | 5776 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5777 | case IORING_OP_CONNECT: |
bfe76559 | 5778 | return io_connect_prep(req, sqe); |
2d28390a | 5779 | case IORING_OP_TIMEOUT: |
bfe76559 | 5780 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5781 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5782 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5783 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5784 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5785 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5786 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5787 | case IORING_OP_ACCEPT: |
bfe76559 | 5788 | return io_accept_prep(req, sqe); |
d63d1b5e | 5789 | case IORING_OP_FALLOCATE: |
bfe76559 | 5790 | return io_fallocate_prep(req, sqe); |
15b71abe | 5791 | case IORING_OP_OPENAT: |
bfe76559 | 5792 | return io_openat_prep(req, sqe); |
b5dba59e | 5793 | case IORING_OP_CLOSE: |
bfe76559 | 5794 | return io_close_prep(req, sqe); |
05f3fb3c | 5795 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5796 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5797 | case IORING_OP_STATX: |
bfe76559 | 5798 | return io_statx_prep(req, sqe); |
4840e418 | 5799 | case IORING_OP_FADVISE: |
bfe76559 | 5800 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5801 | case IORING_OP_MADVISE: |
bfe76559 | 5802 | return io_madvise_prep(req, sqe); |
cebdb986 | 5803 | case IORING_OP_OPENAT2: |
bfe76559 | 5804 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5805 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5806 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5807 | case IORING_OP_SPLICE: |
bfe76559 | 5808 | return io_splice_prep(req, sqe); |
ddf0322d | 5809 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5810 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5811 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5812 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5813 | case IORING_OP_TEE: |
bfe76559 | 5814 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5815 | case IORING_OP_SHUTDOWN: |
5816 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5817 | case IORING_OP_RENAMEAT: |
5818 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5819 | case IORING_OP_UNLINKAT: |
5820 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5821 | } |
5822 | ||
bfe76559 PB |
5823 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5824 | req->opcode); | |
5825 | return-EINVAL; | |
5826 | } | |
5827 | ||
93642ef8 | 5828 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5829 | { |
93642ef8 PB |
5830 | switch (req->opcode) { |
5831 | case IORING_OP_READV: | |
5832 | case IORING_OP_READ_FIXED: | |
5833 | case IORING_OP_READ: | |
5834 | return io_rw_prep_async(req, READ); | |
5835 | case IORING_OP_WRITEV: | |
5836 | case IORING_OP_WRITE_FIXED: | |
5837 | case IORING_OP_WRITE: | |
5838 | return io_rw_prep_async(req, WRITE); | |
5839 | case IORING_OP_SENDMSG: | |
5840 | case IORING_OP_SEND: | |
5841 | return io_sendmsg_prep_async(req); | |
5842 | case IORING_OP_RECVMSG: | |
5843 | case IORING_OP_RECV: | |
5844 | return io_recvmsg_prep_async(req); | |
5845 | case IORING_OP_CONNECT: | |
5846 | return io_connect_prep_async(req); | |
5847 | } | |
5848 | return 0; | |
5849 | } | |
5850 | ||
be7053b7 | 5851 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5852 | { |
be7053b7 | 5853 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5854 | return 0; |
be7053b7 | 5855 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5856 | if (req->async_data) |
be7053b7 PB |
5857 | return 0; |
5858 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5859 | return -EAGAIN; |
be7053b7 | 5860 | return io_req_prep_async(req); |
f67676d1 JA |
5861 | } |
5862 | ||
9cf7c104 PB |
5863 | static u32 io_get_sequence(struct io_kiocb *req) |
5864 | { | |
5865 | struct io_kiocb *pos; | |
5866 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5867 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5868 | |
f2f87370 PB |
5869 | io_for_each_link(pos, req) |
5870 | nr_reqs++; | |
9cf7c104 PB |
5871 | |
5872 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5873 | return total_submitted - nr_reqs; | |
5874 | } | |
5875 | ||
be7053b7 | 5876 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5877 | { |
a197f664 | 5878 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5879 | struct io_defer_entry *de; |
f67676d1 | 5880 | int ret; |
9cf7c104 | 5881 | u32 seq; |
de0617e4 | 5882 | |
9d858b21 | 5883 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5884 | if (likely(list_empty_careful(&ctx->defer_list) && |
5885 | !(req->flags & REQ_F_IO_DRAIN))) | |
5886 | return 0; | |
5887 | ||
5888 | seq = io_get_sequence(req); | |
5889 | /* Still a chance to pass the sequence check */ | |
5890 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5891 | return 0; |
5892 | ||
be7053b7 PB |
5893 | ret = io_req_defer_prep(req); |
5894 | if (ret) | |
5895 | return ret; | |
cbdcb435 | 5896 | io_prep_async_link(req); |
27dc8338 PB |
5897 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5898 | if (!de) | |
5899 | return -ENOMEM; | |
2d28390a | 5900 | |
de0617e4 | 5901 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5902 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5903 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5904 | kfree(de); |
ae34817b PB |
5905 | io_queue_async_work(req); |
5906 | return -EIOCBQUEUED; | |
de0617e4 JA |
5907 | } |
5908 | ||
915967f6 | 5909 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5910 | de->req = req; |
9cf7c104 | 5911 | de->seq = seq; |
27dc8338 | 5912 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5913 | spin_unlock_irq(&ctx->completion_lock); |
5914 | return -EIOCBQUEUED; | |
5915 | } | |
5916 | ||
3ca405eb | 5917 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5918 | { |
0e1b6fe3 PB |
5919 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5920 | switch (req->opcode) { | |
5921 | case IORING_OP_READV: | |
5922 | case IORING_OP_READ_FIXED: | |
5923 | case IORING_OP_READ: | |
bcda7baa | 5924 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5925 | break; |
5926 | case IORING_OP_RECVMSG: | |
5927 | case IORING_OP_RECV: | |
bcda7baa | 5928 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5929 | break; |
5930 | } | |
5931 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5932 | } |
5933 | ||
0e1b6fe3 PB |
5934 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5935 | switch (req->opcode) { | |
5936 | case IORING_OP_READV: | |
5937 | case IORING_OP_READ_FIXED: | |
5938 | case IORING_OP_READ: | |
5939 | case IORING_OP_WRITEV: | |
5940 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5941 | case IORING_OP_WRITE: { |
5942 | struct io_async_rw *io = req->async_data; | |
5943 | if (io->free_iovec) | |
5944 | kfree(io->free_iovec); | |
0e1b6fe3 | 5945 | break; |
e8c2bc1f | 5946 | } |
0e1b6fe3 | 5947 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5948 | case IORING_OP_SENDMSG: { |
5949 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5950 | |
5951 | kfree(io->free_iov); | |
0e1b6fe3 | 5952 | break; |
e8c2bc1f | 5953 | } |
0e1b6fe3 PB |
5954 | case IORING_OP_SPLICE: |
5955 | case IORING_OP_TEE: | |
5956 | io_put_file(req, req->splice.file_in, | |
5957 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5958 | break; | |
f3cd4850 JA |
5959 | case IORING_OP_OPENAT: |
5960 | case IORING_OP_OPENAT2: | |
5961 | if (req->open.filename) | |
5962 | putname(req->open.filename); | |
5963 | break; | |
80a261fd JA |
5964 | case IORING_OP_RENAMEAT: |
5965 | putname(req->rename.oldpath); | |
5966 | putname(req->rename.newpath); | |
5967 | break; | |
14a1143b JA |
5968 | case IORING_OP_UNLINKAT: |
5969 | putname(req->unlink.filename); | |
5970 | break; | |
0e1b6fe3 PB |
5971 | } |
5972 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5973 | } |
99bc4c38 PB |
5974 | } |
5975 | ||
889fca73 | 5976 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5977 | { |
a197f664 | 5978 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5979 | const struct cred *creds = NULL; |
d625c6ee | 5980 | int ret; |
2b188cc1 | 5981 | |
003e8dcc JA |
5982 | if (req->work.creds && req->work.creds != current_cred()) |
5983 | creds = override_creds(req->work.creds); | |
5730b27e | 5984 | |
d625c6ee | 5985 | switch (req->opcode) { |
2b188cc1 | 5986 | case IORING_OP_NOP: |
889fca73 | 5987 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5988 | break; |
5989 | case IORING_OP_READV: | |
edafccee | 5990 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5991 | case IORING_OP_READ: |
889fca73 | 5992 | ret = io_read(req, issue_flags); |
edafccee | 5993 | break; |
3529d8c2 | 5994 | case IORING_OP_WRITEV: |
edafccee | 5995 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5996 | case IORING_OP_WRITE: |
889fca73 | 5997 | ret = io_write(req, issue_flags); |
2b188cc1 | 5998 | break; |
c992fe29 | 5999 | case IORING_OP_FSYNC: |
45d189c6 | 6000 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6001 | break; |
221c5eb2 | 6002 | case IORING_OP_POLL_ADD: |
61e98203 | 6003 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6004 | break; |
6005 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6006 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6007 | break; |
5d17b4a4 | 6008 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6009 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6010 | break; |
0fa03c62 | 6011 | case IORING_OP_SENDMSG: |
889fca73 | 6012 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6013 | break; |
fddaface | 6014 | case IORING_OP_SEND: |
889fca73 | 6015 | ret = io_send(req, issue_flags); |
0fa03c62 | 6016 | break; |
aa1fa28f | 6017 | case IORING_OP_RECVMSG: |
889fca73 | 6018 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6019 | break; |
fddaface | 6020 | case IORING_OP_RECV: |
889fca73 | 6021 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6022 | break; |
5262f567 | 6023 | case IORING_OP_TIMEOUT: |
61e98203 | 6024 | ret = io_timeout(req, issue_flags); |
5262f567 | 6025 | break; |
11365043 | 6026 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6027 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6028 | break; |
17f2fe35 | 6029 | case IORING_OP_ACCEPT: |
889fca73 | 6030 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6031 | break; |
f8e85cf2 | 6032 | case IORING_OP_CONNECT: |
889fca73 | 6033 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6034 | break; |
62755e35 | 6035 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6036 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6037 | break; |
d63d1b5e | 6038 | case IORING_OP_FALLOCATE: |
45d189c6 | 6039 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6040 | break; |
15b71abe | 6041 | case IORING_OP_OPENAT: |
45d189c6 | 6042 | ret = io_openat(req, issue_flags); |
15b71abe | 6043 | break; |
b5dba59e | 6044 | case IORING_OP_CLOSE: |
889fca73 | 6045 | ret = io_close(req, issue_flags); |
b5dba59e | 6046 | break; |
05f3fb3c | 6047 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6048 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6049 | break; |
eddc7ef5 | 6050 | case IORING_OP_STATX: |
45d189c6 | 6051 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6052 | break; |
4840e418 | 6053 | case IORING_OP_FADVISE: |
45d189c6 | 6054 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6055 | break; |
c1ca757b | 6056 | case IORING_OP_MADVISE: |
45d189c6 | 6057 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6058 | break; |
cebdb986 | 6059 | case IORING_OP_OPENAT2: |
45d189c6 | 6060 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6061 | break; |
3e4827b0 | 6062 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6063 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6064 | break; |
7d67af2c | 6065 | case IORING_OP_SPLICE: |
45d189c6 | 6066 | ret = io_splice(req, issue_flags); |
7d67af2c | 6067 | break; |
ddf0322d | 6068 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6069 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6070 | break; |
067524e9 | 6071 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6072 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6073 | break; |
f2a8d5c7 | 6074 | case IORING_OP_TEE: |
45d189c6 | 6075 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6076 | break; |
36f4fa68 | 6077 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6078 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6079 | break; |
80a261fd | 6080 | case IORING_OP_RENAMEAT: |
45d189c6 | 6081 | ret = io_renameat(req, issue_flags); |
80a261fd | 6082 | break; |
14a1143b | 6083 | case IORING_OP_UNLINKAT: |
45d189c6 | 6084 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6085 | break; |
2b188cc1 JA |
6086 | default: |
6087 | ret = -EINVAL; | |
6088 | break; | |
6089 | } | |
6090 | ||
5730b27e JA |
6091 | if (creds) |
6092 | revert_creds(creds); | |
6093 | ||
def596e9 JA |
6094 | if (ret) |
6095 | return ret; | |
6096 | ||
b532576e JA |
6097 | /* If the op doesn't have a file, we're not polling for it */ |
6098 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6099 | const bool in_async = io_wq_current_is_worker(); |
6100 | ||
11ba820b JA |
6101 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6102 | if (in_async) | |
6103 | mutex_lock(&ctx->uring_lock); | |
6104 | ||
2e9dbe90 | 6105 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6106 | |
6107 | if (in_async) | |
6108 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6109 | } |
6110 | ||
6111 | return 0; | |
2b188cc1 JA |
6112 | } |
6113 | ||
5280f7e5 | 6114 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6115 | { |
6116 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6117 | struct io_kiocb *timeout; |
561fb04a | 6118 | int ret = 0; |
2b188cc1 | 6119 | |
6df1db6b PB |
6120 | timeout = io_prep_linked_timeout(req); |
6121 | if (timeout) | |
6122 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6123 | |
4014d943 | 6124 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6125 | ret = -ECANCELED; |
31b51510 | 6126 | |
561fb04a | 6127 | if (!ret) { |
561fb04a | 6128 | do { |
889fca73 | 6129 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6130 | /* |
6131 | * We can get EAGAIN for polled IO even though we're | |
6132 | * forcing a sync submission from here, since we can't | |
6133 | * wait for request slots on the block side. | |
6134 | */ | |
6135 | if (ret != -EAGAIN) | |
6136 | break; | |
6137 | cond_resched(); | |
6138 | } while (1); | |
6139 | } | |
31b51510 | 6140 | |
a3df7698 | 6141 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6142 | if (ret) { |
a3df7698 PB |
6143 | /* io-wq is going to take one down */ |
6144 | refcount_inc(&req->refs); | |
6145 | io_req_task_queue_fail(req, ret); | |
edafccee | 6146 | } |
2b188cc1 JA |
6147 | } |
6148 | ||
65e19f54 JA |
6149 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6150 | int index) | |
6151 | { | |
269bbe5f | 6152 | struct fixed_rsrc_table *table; |
65e19f54 | 6153 | |
05f3fb3c | 6154 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6155 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6156 | } |
6157 | ||
8371adf5 PB |
6158 | static struct file *io_file_get(struct io_submit_state *state, |
6159 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6160 | { |
a197f664 | 6161 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6162 | struct file *file; |
09bb8394 | 6163 | |
8da11c19 | 6164 | if (fixed) { |
479f517b | 6165 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6166 | return NULL; |
b7620121 | 6167 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6168 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6169 | io_set_resource_node(req); |
09bb8394 | 6170 | } else { |
c826bd7a | 6171 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6172 | file = __io_file_get(state, fd); |
09bb8394 JA |
6173 | } |
6174 | ||
ce3d5aae PB |
6175 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6176 | io_req_track_inflight(req); | |
8371adf5 | 6177 | return file; |
09bb8394 JA |
6178 | } |
6179 | ||
2665abfd | 6180 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6181 | { |
ad8a48ac JA |
6182 | struct io_timeout_data *data = container_of(timer, |
6183 | struct io_timeout_data, timer); | |
90cd7e42 | 6184 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6185 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6186 | unsigned long flags; |
2665abfd JA |
6187 | |
6188 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6189 | prev = req->timeout.head; |
6190 | req->timeout.head = NULL; | |
2665abfd JA |
6191 | |
6192 | /* | |
6193 | * We don't expect the list to be empty, that will only happen if we | |
6194 | * race with the completion of the linked work. | |
6195 | */ | |
90cd7e42 | 6196 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6197 | io_remove_next_linked(prev); |
90cd7e42 PB |
6198 | else |
6199 | prev = NULL; | |
2665abfd JA |
6200 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6201 | ||
6202 | if (prev) { | |
014db007 | 6203 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6204 | io_put_req_deferred(prev, 1); |
47f46768 | 6205 | } else { |
9ae1f8dd PB |
6206 | io_req_complete_post(req, -ETIME, 0); |
6207 | io_put_req_deferred(req, 1); | |
2665abfd | 6208 | } |
2665abfd JA |
6209 | return HRTIMER_NORESTART; |
6210 | } | |
6211 | ||
7271ef3a | 6212 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6213 | { |
76a46e06 | 6214 | /* |
f2f87370 PB |
6215 | * If the back reference is NULL, then our linked request finished |
6216 | * before we got a chance to setup the timer | |
76a46e06 | 6217 | */ |
90cd7e42 | 6218 | if (req->timeout.head) { |
e8c2bc1f | 6219 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6220 | |
ad8a48ac JA |
6221 | data->timer.function = io_link_timeout_fn; |
6222 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6223 | data->mode); | |
2665abfd | 6224 | } |
7271ef3a JA |
6225 | } |
6226 | ||
6227 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6228 | { | |
6229 | struct io_ring_ctx *ctx = req->ctx; | |
6230 | ||
6231 | spin_lock_irq(&ctx->completion_lock); | |
6232 | __io_queue_linked_timeout(req); | |
76a46e06 | 6233 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6234 | |
2665abfd | 6235 | /* drop submission reference */ |
76a46e06 JA |
6236 | io_put_req(req); |
6237 | } | |
2665abfd | 6238 | |
ad8a48ac | 6239 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6240 | { |
f2f87370 | 6241 | struct io_kiocb *nxt = req->link; |
2665abfd | 6242 | |
f2f87370 PB |
6243 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6244 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6245 | return NULL; |
2665abfd | 6246 | |
90cd7e42 | 6247 | nxt->timeout.head = req; |
900fad45 | 6248 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6249 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6250 | return nxt; |
2665abfd JA |
6251 | } |
6252 | ||
c5eef2b9 | 6253 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6254 | { |
d3d7298d | 6255 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6256 | int ret; |
2b188cc1 | 6257 | |
c5eef2b9 | 6258 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6259 | |
491381ce JA |
6260 | /* |
6261 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6262 | * doesn't support non-blocking read/write attempts | |
6263 | */ | |
24c74678 | 6264 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6265 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6266 | /* |
6267 | * Queued up for async execution, worker will release | |
6268 | * submit reference when the iocb is actually submitted. | |
6269 | */ | |
6270 | io_queue_async_work(req); | |
2b188cc1 | 6271 | } |
0d63c148 PB |
6272 | } else if (likely(!ret)) { |
6273 | /* drop submission reference */ | |
e342c807 | 6274 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6275 | struct io_ring_ctx *ctx = req->ctx; |
6276 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6277 | |
6dd0be1e | 6278 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6279 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6280 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6281 | } else { |
d3d7298d | 6282 | io_put_req(req); |
0d63c148 PB |
6283 | } |
6284 | } else { | |
4e88d6e7 | 6285 | req_set_fail_links(req); |
e65ef56d | 6286 | io_put_req(req); |
e1e16097 | 6287 | io_req_complete(req, ret); |
9e645e11 | 6288 | } |
d3d7298d PB |
6289 | if (linked_timeout) |
6290 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6291 | } |
6292 | ||
be7053b7 | 6293 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6294 | { |
6295 | int ret; | |
6296 | ||
be7053b7 | 6297 | ret = io_req_defer(req); |
4fe2c963 JL |
6298 | if (ret) { |
6299 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6300 | fail_req: |
4e88d6e7 | 6301 | req_set_fail_links(req); |
e1e16097 JA |
6302 | io_put_req(req); |
6303 | io_req_complete(req, ret); | |
4fe2c963 | 6304 | } |
2550878f | 6305 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6306 | ret = io_req_defer_prep(req); |
6307 | if (unlikely(ret)) | |
6308 | goto fail_req; | |
ce35a47a JA |
6309 | io_queue_async_work(req); |
6310 | } else { | |
c5eef2b9 | 6311 | __io_queue_sqe(req); |
ce35a47a | 6312 | } |
4fe2c963 JL |
6313 | } |
6314 | ||
b16fed66 PB |
6315 | /* |
6316 | * Check SQE restrictions (opcode and flags). | |
6317 | * | |
6318 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6319 | */ | |
6320 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6321 | struct io_kiocb *req, | |
6322 | unsigned int sqe_flags) | |
4fe2c963 | 6323 | { |
b16fed66 PB |
6324 | if (!ctx->restricted) |
6325 | return true; | |
6326 | ||
6327 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6328 | return false; | |
6329 | ||
6330 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6331 | ctx->restrictions.sqe_flags_required) | |
6332 | return false; | |
6333 | ||
6334 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6335 | ctx->restrictions.sqe_flags_required)) | |
6336 | return false; | |
6337 | ||
6338 | return true; | |
4fe2c963 JL |
6339 | } |
6340 | ||
b16fed66 PB |
6341 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6342 | const struct io_uring_sqe *sqe) | |
6343 | { | |
6344 | struct io_submit_state *state; | |
6345 | unsigned int sqe_flags; | |
003e8dcc | 6346 | int personality, ret = 0; |
b16fed66 PB |
6347 | |
6348 | req->opcode = READ_ONCE(sqe->opcode); | |
6349 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6350 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6351 | req->user_data = READ_ONCE(sqe->user_data); | |
6352 | req->async_data = NULL; | |
6353 | req->file = NULL; | |
6354 | req->ctx = ctx; | |
6355 | req->link = NULL; | |
6356 | req->fixed_rsrc_refs = NULL; | |
6357 | /* one is dropped after submission, the other at completion */ | |
6358 | refcount_set(&req->refs, 2); | |
6359 | req->task = current; | |
6360 | req->result = 0; | |
93e68e03 JA |
6361 | req->work.list.next = NULL; |
6362 | req->work.creds = NULL; | |
6363 | req->work.flags = 0; | |
b16fed66 PB |
6364 | |
6365 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6366 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6367 | req->flags = 0; | |
b16fed66 | 6368 | return -EINVAL; |
ebf4a5db | 6369 | } |
b16fed66 PB |
6370 | |
6371 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6372 | return -EINVAL; | |
6373 | ||
b16fed66 PB |
6374 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6375 | return -EACCES; | |
6376 | ||
6377 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6378 | !io_op_defs[req->opcode].buffer_select) | |
6379 | return -EOPNOTSUPP; | |
863e0560 | 6380 | |
003e8dcc JA |
6381 | personality = READ_ONCE(sqe->personality); |
6382 | if (personality) { | |
61cf9370 | 6383 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6384 | if (!req->work.creds) |
6385 | return -EINVAL; | |
6386 | get_cred(req->work.creds); | |
003e8dcc | 6387 | } |
b16fed66 PB |
6388 | state = &ctx->submit_state; |
6389 | ||
6390 | /* | |
6391 | * Plug now if we have more than 1 IO left after this, and the target | |
6392 | * is potentially a read/write to block based storage. | |
6393 | */ | |
6394 | if (!state->plug_started && state->ios_left > 1 && | |
6395 | io_op_defs[req->opcode].plug) { | |
6396 | blk_start_plug(&state->plug); | |
6397 | state->plug_started = true; | |
6398 | } | |
6399 | ||
6400 | if (io_op_defs[req->opcode].needs_file) { | |
6401 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6402 | ||
6403 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6404 | if (unlikely(!req->file)) | |
6405 | ret = -EBADF; | |
6406 | } | |
6407 | ||
6408 | state->ios_left--; | |
6409 | return ret; | |
6410 | } | |
6411 | ||
a6b8cadc | 6412 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6413 | const struct io_uring_sqe *sqe) |
9e645e11 | 6414 | { |
a1ab7b35 | 6415 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6416 | int ret; |
9e645e11 | 6417 | |
a6b8cadc PB |
6418 | ret = io_init_req(ctx, req, sqe); |
6419 | if (unlikely(ret)) { | |
6420 | fail_req: | |
6421 | io_put_req(req); | |
6422 | io_req_complete(req, ret); | |
de59bc10 PB |
6423 | if (link->head) { |
6424 | /* fail even hard links since we don't submit */ | |
cf109604 | 6425 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6426 | io_put_req(link->head); |
6427 | io_req_complete(link->head, -ECANCELED); | |
6428 | link->head = NULL; | |
6429 | } | |
a6b8cadc PB |
6430 | return ret; |
6431 | } | |
be7053b7 PB |
6432 | ret = io_req_prep(req, sqe); |
6433 | if (unlikely(ret)) | |
6434 | goto fail_req; | |
a6b8cadc | 6435 | |
be7053b7 | 6436 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6437 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6438 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6439 | ||
9e645e11 JA |
6440 | /* |
6441 | * If we already have a head request, queue this one for async | |
6442 | * submittal once the head completes. If we don't have a head but | |
6443 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6444 | * submitted sync once the chain is complete. If none of those | |
6445 | * conditions are true (normal request), then just queue it. | |
6446 | */ | |
863e0560 PB |
6447 | if (link->head) { |
6448 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6449 | |
8cdf2193 PB |
6450 | /* |
6451 | * Taking sequential execution of a link, draining both sides | |
6452 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6453 | * requests in the link. So, it drains the head and the | |
6454 | * next after the link request. The last one is done via | |
6455 | * drain_next flag to persist the effect across calls. | |
6456 | */ | |
ef4ff581 | 6457 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6458 | head->flags |= REQ_F_IO_DRAIN; |
6459 | ctx->drain_next = 1; | |
6460 | } | |
be7053b7 | 6461 | ret = io_req_defer_prep(req); |
cf109604 | 6462 | if (unlikely(ret)) |
a6b8cadc | 6463 | goto fail_req; |
9d76377f | 6464 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6465 | link->last->link = req; |
863e0560 | 6466 | link->last = req; |
32fe525b PB |
6467 | |
6468 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6469 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6470 | io_queue_sqe(head); |
863e0560 | 6471 | link->head = NULL; |
32fe525b | 6472 | } |
9e645e11 | 6473 | } else { |
711be031 PB |
6474 | if (unlikely(ctx->drain_next)) { |
6475 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6476 | ctx->drain_next = 0; |
711be031 | 6477 | } |
ef4ff581 | 6478 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6479 | link->head = req; |
6480 | link->last = req; | |
711be031 | 6481 | } else { |
be7053b7 | 6482 | io_queue_sqe(req); |
711be031 | 6483 | } |
9e645e11 | 6484 | } |
2e6e1fde | 6485 | |
1d4240cc | 6486 | return 0; |
9e645e11 JA |
6487 | } |
6488 | ||
9a56a232 JA |
6489 | /* |
6490 | * Batched submission is done, ensure local IO is flushed out. | |
6491 | */ | |
ba88ff11 PB |
6492 | static void io_submit_state_end(struct io_submit_state *state, |
6493 | struct io_ring_ctx *ctx) | |
9a56a232 | 6494 | { |
a1ab7b35 | 6495 | if (state->link.head) |
de59bc10 | 6496 | io_queue_sqe(state->link.head); |
6dd0be1e | 6497 | if (state->comp.nr) |
ba88ff11 | 6498 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6499 | if (state->plug_started) |
6500 | blk_finish_plug(&state->plug); | |
9f13c35b | 6501 | io_state_file_put(state); |
9a56a232 JA |
6502 | } |
6503 | ||
6504 | /* | |
6505 | * Start submission side cache. | |
6506 | */ | |
6507 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6508 | unsigned int max_ios) |
9a56a232 | 6509 | { |
27926b68 | 6510 | state->plug_started = false; |
9a56a232 | 6511 | state->ios_left = max_ios; |
a1ab7b35 PB |
6512 | /* set only head, no need to init link_last in advance */ |
6513 | state->link.head = NULL; | |
9a56a232 JA |
6514 | } |
6515 | ||
2b188cc1 JA |
6516 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6517 | { | |
75b28aff | 6518 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6519 | |
caf582c6 PB |
6520 | /* |
6521 | * Ensure any loads from the SQEs are done at this point, | |
6522 | * since once we write the new head, the application could | |
6523 | * write new data to them. | |
6524 | */ | |
6525 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6526 | } |
6527 | ||
2b188cc1 | 6528 | /* |
3529d8c2 | 6529 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6530 | * that is mapped by userspace. This means that care needs to be taken to |
6531 | * ensure that reads are stable, as we cannot rely on userspace always | |
6532 | * being a good citizen. If members of the sqe are validated and then later | |
6533 | * used, it's important that those reads are done through READ_ONCE() to | |
6534 | * prevent a re-load down the line. | |
6535 | */ | |
709b302f | 6536 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6537 | { |
75b28aff | 6538 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6539 | unsigned head; |
6540 | ||
6541 | /* | |
6542 | * The cached sq head (or cq tail) serves two purposes: | |
6543 | * | |
6544 | * 1) allows us to batch the cost of updating the user visible | |
6545 | * head updates. | |
6546 | * 2) allows the kernel side to track the head on its own, even | |
6547 | * though the application is the one updating it. | |
6548 | */ | |
4fccfcbb | 6549 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6550 | if (likely(head < ctx->sq_entries)) |
6551 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6552 | |
6553 | /* drop invalid entries */ | |
498ccd9e | 6554 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6555 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6556 | return NULL; |
6557 | } | |
6558 | ||
0f212204 | 6559 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6560 | { |
46c4e16a | 6561 | int submitted = 0; |
6c271ce2 | 6562 | |
c4a2ed72 | 6563 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6564 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6565 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6566 | return -EBUSY; |
6567 | } | |
6c271ce2 | 6568 | |
ee7d46d9 PB |
6569 | /* make sure SQ entry isn't read before tail */ |
6570 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6571 | |
2b85edfc PB |
6572 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6573 | return -EAGAIN; | |
6c271ce2 | 6574 | |
d8a6df10 | 6575 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6576 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6577 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6578 | |
46c4e16a | 6579 | while (submitted < nr) { |
3529d8c2 | 6580 | const struct io_uring_sqe *sqe; |
196be95c | 6581 | struct io_kiocb *req; |
fb5ccc98 | 6582 | |
258b29a9 | 6583 | req = io_alloc_req(ctx); |
196be95c PB |
6584 | if (unlikely(!req)) { |
6585 | if (!submitted) | |
6586 | submitted = -EAGAIN; | |
fb5ccc98 | 6587 | break; |
196be95c | 6588 | } |
4fccfcbb PB |
6589 | sqe = io_get_sqe(ctx); |
6590 | if (unlikely(!sqe)) { | |
6591 | kmem_cache_free(req_cachep, req); | |
6592 | break; | |
6593 | } | |
d3656344 JA |
6594 | /* will complete beyond this point, count as submitted */ |
6595 | submitted++; | |
a1ab7b35 | 6596 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6597 | break; |
6c271ce2 JA |
6598 | } |
6599 | ||
9466f437 PB |
6600 | if (unlikely(submitted != nr)) { |
6601 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6602 | struct io_uring_task *tctx = current->io_uring; |
6603 | int unused = nr - ref_used; | |
9466f437 | 6604 | |
d8a6df10 JA |
6605 | percpu_ref_put_many(&ctx->refs, unused); |
6606 | percpu_counter_sub(&tctx->inflight, unused); | |
6607 | put_task_struct_many(current, unused); | |
9466f437 | 6608 | } |
6c271ce2 | 6609 | |
a1ab7b35 | 6610 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6611 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6612 | io_commit_sqring(ctx); | |
6613 | ||
6c271ce2 JA |
6614 | return submitted; |
6615 | } | |
6616 | ||
23b3628e XW |
6617 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6618 | { | |
6619 | /* Tell userspace we may need a wakeup call */ | |
6620 | spin_lock_irq(&ctx->completion_lock); | |
6621 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6622 | spin_unlock_irq(&ctx->completion_lock); | |
6623 | } | |
6624 | ||
6625 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6626 | { | |
6627 | spin_lock_irq(&ctx->completion_lock); | |
6628 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6629 | spin_unlock_irq(&ctx->completion_lock); | |
6630 | } | |
6631 | ||
08369246 | 6632 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6633 | { |
c8d1ba58 | 6634 | unsigned int to_submit; |
bdcd3eab | 6635 | int ret = 0; |
6c271ce2 | 6636 | |
c8d1ba58 | 6637 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6638 | /* if we're handling multiple rings, cap submit size for fairness */ |
6639 | if (cap_entries && to_submit > 8) | |
6640 | to_submit = 8; | |
6641 | ||
906a3c6f | 6642 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6643 | unsigned nr_events = 0; |
a4c0b3de | 6644 | |
c8d1ba58 | 6645 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6646 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6647 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6648 | |
0298ef96 PB |
6649 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6650 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6651 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6652 | mutex_unlock(&ctx->uring_lock); |
6653 | } | |
6c271ce2 | 6654 | |
90554200 JA |
6655 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6656 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6657 | |
08369246 XW |
6658 | return ret; |
6659 | } | |
6c271ce2 | 6660 | |
08369246 XW |
6661 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6662 | { | |
6663 | struct io_ring_ctx *ctx; | |
6664 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6665 | |
08369246 XW |
6666 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6667 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6668 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6669 | } |
c1edbf5f | 6670 | |
08369246 | 6671 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6672 | } |
6c271ce2 | 6673 | |
c8d1ba58 JA |
6674 | static int io_sq_thread(void *data) |
6675 | { | |
69fb2131 JA |
6676 | struct io_sq_data *sqd = data; |
6677 | struct io_ring_ctx *ctx; | |
a0d9205f | 6678 | unsigned long timeout = 0; |
37d1e2e3 | 6679 | char buf[TASK_COMM_LEN]; |
08369246 | 6680 | DEFINE_WAIT(wait); |
6c271ce2 | 6681 | |
37d1e2e3 JA |
6682 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6683 | set_task_comm(current, buf); | |
37d1e2e3 JA |
6684 | current->pf_io_worker = NULL; |
6685 | ||
6686 | if (sqd->sq_cpu != -1) | |
6687 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6688 | else | |
6689 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6690 | current->flags |= PF_NO_SETAFFINITY; | |
6691 | ||
09a6f4ef | 6692 | mutex_lock(&sqd->lock); |
05962f95 | 6693 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6694 | int ret; |
6695 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6696 | |
05962f95 | 6697 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
09a6f4ef | 6698 | mutex_unlock(&sqd->lock); |
05962f95 | 6699 | cond_resched(); |
09a6f4ef | 6700 | mutex_lock(&sqd->lock); |
521d6a73 | 6701 | io_run_task_work(); |
08369246 | 6702 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6703 | continue; |
08369246 | 6704 | } |
37d1e2e3 JA |
6705 | if (fatal_signal_pending(current)) |
6706 | break; | |
08369246 | 6707 | sqt_spin = false; |
e95eee2d | 6708 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6709 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6710 | const struct cred *creds = NULL; |
6711 | ||
6712 | if (ctx->sq_creds != current_cred()) | |
6713 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6714 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6715 | if (creds) |
6716 | revert_creds(creds); | |
08369246 XW |
6717 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6718 | sqt_spin = true; | |
69fb2131 | 6719 | } |
6c271ce2 | 6720 | |
08369246 | 6721 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6722 | io_run_task_work(); |
6723 | cond_resched(); | |
08369246 XW |
6724 | if (sqt_spin) |
6725 | timeout = jiffies + sqd->sq_thread_idle; | |
6726 | continue; | |
6727 | } | |
6728 | ||
08369246 XW |
6729 | needs_sched = true; |
6730 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6731 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6732 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6733 | !list_empty_careful(&ctx->iopoll_list)) { | |
6734 | needs_sched = false; | |
6735 | break; | |
6736 | } | |
6737 | if (io_sqring_entries(ctx)) { | |
6738 | needs_sched = false; | |
6739 | break; | |
6740 | } | |
6741 | } | |
6742 | ||
05962f95 | 6743 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6744 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6745 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6746 | |
09a6f4ef | 6747 | mutex_unlock(&sqd->lock); |
69fb2131 | 6748 | schedule(); |
16efa4fc | 6749 | try_to_freeze(); |
09a6f4ef | 6750 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6751 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6752 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6753 | } |
08369246 XW |
6754 | |
6755 | finish_wait(&sqd->wait, &wait); | |
6756 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6757 | } |
28cea78a | 6758 | |
521d6a73 PB |
6759 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6760 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6761 | sqd->thread = NULL; |
05962f95 | 6762 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6763 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6764 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6765 | |
6766 | io_run_task_work(); | |
37d1e2e3 JA |
6767 | complete(&sqd->exited); |
6768 | do_exit(0); | |
6c271ce2 JA |
6769 | } |
6770 | ||
bda52162 JA |
6771 | struct io_wait_queue { |
6772 | struct wait_queue_entry wq; | |
6773 | struct io_ring_ctx *ctx; | |
6774 | unsigned to_wait; | |
6775 | unsigned nr_timeouts; | |
6776 | }; | |
6777 | ||
6c503150 | 6778 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6779 | { |
6780 | struct io_ring_ctx *ctx = iowq->ctx; | |
6781 | ||
6782 | /* | |
d195a66e | 6783 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6784 | * started waiting. For timeouts, we always want to return to userspace, |
6785 | * regardless of event count. | |
6786 | */ | |
6c503150 | 6787 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6788 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6789 | } | |
6790 | ||
6791 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6792 | int wake_flags, void *key) | |
6793 | { | |
6794 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6795 | wq); | |
6796 | ||
6c503150 PB |
6797 | /* |
6798 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6799 | * the task, and the next invocation will do it. | |
6800 | */ | |
6801 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6802 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6803 | return -1; | |
bda52162 JA |
6804 | } |
6805 | ||
af9c1a44 JA |
6806 | static int io_run_task_work_sig(void) |
6807 | { | |
6808 | if (io_run_task_work()) | |
6809 | return 1; | |
6810 | if (!signal_pending(current)) | |
6811 | return 0; | |
792ee0f6 JA |
6812 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6813 | return -ERESTARTSYS; | |
af9c1a44 JA |
6814 | return -EINTR; |
6815 | } | |
6816 | ||
eeb60b9a PB |
6817 | /* when returns >0, the caller should retry */ |
6818 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6819 | struct io_wait_queue *iowq, | |
6820 | signed long *timeout) | |
6821 | { | |
6822 | int ret; | |
6823 | ||
6824 | /* make sure we run task_work before checking for signals */ | |
6825 | ret = io_run_task_work_sig(); | |
6826 | if (ret || io_should_wake(iowq)) | |
6827 | return ret; | |
6828 | /* let the caller flush overflows, retry */ | |
6829 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6830 | return 1; | |
6831 | ||
6832 | *timeout = schedule_timeout(*timeout); | |
6833 | return !*timeout ? -ETIME : 1; | |
6834 | } | |
6835 | ||
2b188cc1 JA |
6836 | /* |
6837 | * Wait until events become available, if we don't already have some. The | |
6838 | * application must reap them itself, as they reside on the shared cq ring. | |
6839 | */ | |
6840 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6841 | const sigset_t __user *sig, size_t sigsz, |
6842 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6843 | { |
bda52162 JA |
6844 | struct io_wait_queue iowq = { |
6845 | .wq = { | |
6846 | .private = current, | |
6847 | .func = io_wake_function, | |
6848 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6849 | }, | |
6850 | .ctx = ctx, | |
6851 | .to_wait = min_events, | |
6852 | }; | |
75b28aff | 6853 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6854 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6855 | int ret; | |
2b188cc1 | 6856 | |
b41e9852 | 6857 | do { |
6c503150 PB |
6858 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6859 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6860 | return 0; |
4c6e277c | 6861 | if (!io_run_task_work()) |
b41e9852 | 6862 | break; |
b41e9852 | 6863 | } while (1); |
2b188cc1 JA |
6864 | |
6865 | if (sig) { | |
9e75ad5d AB |
6866 | #ifdef CONFIG_COMPAT |
6867 | if (in_compat_syscall()) | |
6868 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6869 | sigsz); |
9e75ad5d AB |
6870 | else |
6871 | #endif | |
b772434b | 6872 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6873 | |
2b188cc1 JA |
6874 | if (ret) |
6875 | return ret; | |
6876 | } | |
6877 | ||
c73ebb68 | 6878 | if (uts) { |
c1d5a224 PB |
6879 | struct timespec64 ts; |
6880 | ||
c73ebb68 HX |
6881 | if (get_timespec64(&ts, uts)) |
6882 | return -EFAULT; | |
6883 | timeout = timespec64_to_jiffies(&ts); | |
6884 | } | |
6885 | ||
bda52162 | 6886 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6887 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6888 | do { |
ca0a2651 JA |
6889 | /* if we can't even flush overflow, don't wait for more */ |
6890 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6891 | ret = -EBUSY; | |
6892 | break; | |
6893 | } | |
bda52162 JA |
6894 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6895 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6896 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6897 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6898 | cond_resched(); |
eeb60b9a | 6899 | } while (ret > 0); |
bda52162 | 6900 | |
b7db41c9 | 6901 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6902 | |
75b28aff | 6903 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6904 | } |
6905 | ||
6b06314c JA |
6906 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6907 | { | |
6908 | #if defined(CONFIG_UNIX) | |
6909 | if (ctx->ring_sock) { | |
6910 | struct sock *sock = ctx->ring_sock->sk; | |
6911 | struct sk_buff *skb; | |
6912 | ||
6913 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6914 | kfree_skb(skb); | |
6915 | } | |
6916 | #else | |
6917 | int i; | |
6918 | ||
65e19f54 JA |
6919 | for (i = 0; i < ctx->nr_user_files; i++) { |
6920 | struct file *file; | |
6921 | ||
6922 | file = io_file_from_index(ctx, i); | |
6923 | if (file) | |
6924 | fput(file); | |
6925 | } | |
6b06314c JA |
6926 | #endif |
6927 | } | |
6928 | ||
00835dce | 6929 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6930 | { |
269bbe5f | 6931 | struct fixed_rsrc_data *data; |
05f3fb3c | 6932 | |
269bbe5f | 6933 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6934 | complete(&data->done); |
6935 | } | |
6936 | ||
2a63b2d9 | 6937 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6938 | { |
2a63b2d9 | 6939 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6940 | } |
6941 | ||
2a63b2d9 | 6942 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6943 | { |
2a63b2d9 BM |
6944 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6945 | } | |
65e19f54 | 6946 | |
d67d2263 BM |
6947 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6948 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6949 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6950 | { |
2a63b2d9 | 6951 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6952 | rsrc_data->node = ref_node; |
d67d2263 | 6953 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6954 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6955 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6956 | } |
6957 | ||
8bad28d8 | 6958 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6959 | { |
8bad28d8 | 6960 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6961 | |
2a63b2d9 | 6962 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6963 | ref_node = data->node; |
e6cb007c | 6964 | data->node = NULL; |
2a63b2d9 | 6965 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6966 | if (ref_node) |
6967 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6968 | } |
6969 | ||
6970 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6971 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6972 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6973 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6974 | { |
f2303b1f | 6975 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6976 | int ret; |
05589553 | 6977 | |
8bad28d8 HX |
6978 | if (data->quiesce) |
6979 | return -ENXIO; | |
05589553 | 6980 | |
8bad28d8 | 6981 | data->quiesce = true; |
1ffc5422 | 6982 | do { |
f2303b1f PB |
6983 | ret = -ENOMEM; |
6984 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
6985 | if (!backup_node) | |
6986 | break; | |
6987 | backup_node->rsrc_data = data; | |
6988 | backup_node->rsrc_put = rsrc_put; | |
6989 | ||
8bad28d8 HX |
6990 | io_sqe_rsrc_kill_node(ctx, data); |
6991 | percpu_ref_kill(&data->refs); | |
6992 | flush_delayed_work(&ctx->rsrc_put_work); | |
6993 | ||
1ffc5422 PB |
6994 | ret = wait_for_completion_interruptible(&data->done); |
6995 | if (!ret) | |
6996 | break; | |
8bad28d8 | 6997 | |
cb5e1b81 | 6998 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
6999 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7000 | backup_node = NULL; | |
cb5e1b81 | 7001 | reinit_completion(&data->done); |
8bad28d8 | 7002 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7003 | ret = io_run_task_work_sig(); |
8bad28d8 | 7004 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7005 | } while (ret >= 0); |
8bad28d8 | 7006 | data->quiesce = false; |
05f3fb3c | 7007 | |
8bad28d8 HX |
7008 | if (backup_node) |
7009 | destroy_fixed_rsrc_ref_node(backup_node); | |
7010 | return ret; | |
d7954b2b BM |
7011 | } |
7012 | ||
1ad555c6 BM |
7013 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7014 | { | |
7015 | struct fixed_rsrc_data *data; | |
7016 | ||
7017 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7018 | if (!data) | |
7019 | return NULL; | |
7020 | ||
00835dce | 7021 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7022 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7023 | kfree(data); | |
7024 | return NULL; | |
7025 | } | |
7026 | data->ctx = ctx; | |
7027 | init_completion(&data->done); | |
7028 | return data; | |
7029 | } | |
7030 | ||
7031 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7032 | { | |
7033 | percpu_ref_exit(&data->refs); | |
7034 | kfree(data->table); | |
7035 | kfree(data); | |
7036 | } | |
7037 | ||
d7954b2b BM |
7038 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7039 | { | |
7040 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7041 | unsigned nr_tables, i; |
7042 | int ret; | |
7043 | ||
8bad28d8 HX |
7044 | /* |
7045 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7046 | * Since we possibly drop uring lock later in this function to | |
7047 | * run task work. | |
7048 | */ | |
7049 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7050 | return -ENXIO; |
f2303b1f | 7051 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7052 | if (ret) |
7053 | return ret; | |
7054 | ||
6b06314c | 7055 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7056 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7057 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7058 | kfree(data->table[i].files); |
1ad555c6 | 7059 | free_fixed_rsrc_data(data); |
05f3fb3c | 7060 | ctx->file_data = NULL; |
6b06314c JA |
7061 | ctx->nr_user_files = 0; |
7062 | return 0; | |
7063 | } | |
7064 | ||
37d1e2e3 | 7065 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7066 | __releases(&sqd->lock) |
37d1e2e3 | 7067 | { |
521d6a73 PB |
7068 | WARN_ON_ONCE(sqd->thread == current); |
7069 | ||
37d1e2e3 | 7070 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7071 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7072 | } |
7073 | ||
86e0d676 | 7074 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7075 | __acquires(&sqd->lock) |
37d1e2e3 | 7076 | { |
521d6a73 PB |
7077 | WARN_ON_ONCE(sqd->thread == current); |
7078 | ||
86e0d676 | 7079 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7080 | mutex_lock(&sqd->lock); |
05962f95 JA |
7081 | /* set again for consistency, in case concurrent parks are happening */ |
7082 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7083 | if (sqd->thread) | |
86e0d676 | 7084 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7085 | } |
7086 | ||
7087 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7088 | { | |
521d6a73 PB |
7089 | WARN_ON_ONCE(sqd->thread == current); |
7090 | ||
09a6f4ef | 7091 | mutex_lock(&sqd->lock); |
05962f95 | 7092 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7093 | if (sqd->thread) |
7094 | wake_up_process(sqd->thread); | |
09a6f4ef | 7095 | mutex_unlock(&sqd->lock); |
05962f95 | 7096 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7097 | } |
7098 | ||
534ca6d6 | 7099 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7100 | { |
534ca6d6 | 7101 | if (refcount_dec_and_test(&sqd->refs)) { |
37d1e2e3 JA |
7102 | io_sq_thread_stop(sqd); |
7103 | kfree(sqd); | |
7104 | } | |
7105 | } | |
7106 | ||
7107 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7108 | { | |
7109 | struct io_sq_data *sqd = ctx->sq_data; | |
7110 | ||
7111 | if (sqd) { | |
05962f95 | 7112 | io_sq_thread_park(sqd); |
521d6a73 | 7113 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7114 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7115 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7116 | |
7117 | io_put_sq_data(sqd); | |
7118 | ctx->sq_data = NULL; | |
7c30f36a SM |
7119 | if (ctx->sq_creds) |
7120 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7121 | } |
7122 | } | |
7123 | ||
aa06165d JA |
7124 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7125 | { | |
7126 | struct io_ring_ctx *ctx_attach; | |
7127 | struct io_sq_data *sqd; | |
7128 | struct fd f; | |
7129 | ||
7130 | f = fdget(p->wq_fd); | |
7131 | if (!f.file) | |
7132 | return ERR_PTR(-ENXIO); | |
7133 | if (f.file->f_op != &io_uring_fops) { | |
7134 | fdput(f); | |
7135 | return ERR_PTR(-EINVAL); | |
7136 | } | |
7137 | ||
7138 | ctx_attach = f.file->private_data; | |
7139 | sqd = ctx_attach->sq_data; | |
7140 | if (!sqd) { | |
7141 | fdput(f); | |
7142 | return ERR_PTR(-EINVAL); | |
7143 | } | |
5c2469e0 JA |
7144 | if (sqd->task_tgid != current->tgid) { |
7145 | fdput(f); | |
7146 | return ERR_PTR(-EPERM); | |
7147 | } | |
aa06165d JA |
7148 | |
7149 | refcount_inc(&sqd->refs); | |
7150 | fdput(f); | |
7151 | return sqd; | |
7152 | } | |
7153 | ||
26984fbf PB |
7154 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7155 | bool *attached) | |
534ca6d6 JA |
7156 | { |
7157 | struct io_sq_data *sqd; | |
7158 | ||
26984fbf | 7159 | *attached = false; |
5c2469e0 JA |
7160 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7161 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7162 | if (!IS_ERR(sqd)) { |
7163 | *attached = true; | |
5c2469e0 | 7164 | return sqd; |
26984fbf | 7165 | } |
5c2469e0 JA |
7166 | /* fall through for EPERM case, setup new sqd/task */ |
7167 | if (PTR_ERR(sqd) != -EPERM) | |
7168 | return sqd; | |
7169 | } | |
aa06165d | 7170 | |
534ca6d6 JA |
7171 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7172 | if (!sqd) | |
7173 | return ERR_PTR(-ENOMEM); | |
7174 | ||
7175 | refcount_set(&sqd->refs, 1); | |
69fb2131 | 7176 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7177 | mutex_init(&sqd->lock); |
534ca6d6 | 7178 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7179 | init_completion(&sqd->exited); |
534ca6d6 JA |
7180 | return sqd; |
7181 | } | |
7182 | ||
6b06314c | 7183 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7184 | /* |
7185 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7186 | * the io_uring can be safely unregistered on process exit, even if we have | |
7187 | * loops in the file referencing. | |
7188 | */ | |
7189 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7190 | { | |
7191 | struct sock *sk = ctx->ring_sock->sk; | |
7192 | struct scm_fp_list *fpl; | |
7193 | struct sk_buff *skb; | |
08a45173 | 7194 | int i, nr_files; |
6b06314c | 7195 | |
6b06314c JA |
7196 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7197 | if (!fpl) | |
7198 | return -ENOMEM; | |
7199 | ||
7200 | skb = alloc_skb(0, GFP_KERNEL); | |
7201 | if (!skb) { | |
7202 | kfree(fpl); | |
7203 | return -ENOMEM; | |
7204 | } | |
7205 | ||
7206 | skb->sk = sk; | |
6b06314c | 7207 | |
08a45173 | 7208 | nr_files = 0; |
62e398be | 7209 | fpl->user = get_uid(current_user()); |
6b06314c | 7210 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7211 | struct file *file = io_file_from_index(ctx, i + offset); |
7212 | ||
7213 | if (!file) | |
08a45173 | 7214 | continue; |
65e19f54 | 7215 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7216 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7217 | nr_files++; | |
6b06314c JA |
7218 | } |
7219 | ||
08a45173 JA |
7220 | if (nr_files) { |
7221 | fpl->max = SCM_MAX_FD; | |
7222 | fpl->count = nr_files; | |
7223 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7224 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7225 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7226 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7227 | |
08a45173 JA |
7228 | for (i = 0; i < nr_files; i++) |
7229 | fput(fpl->fp[i]); | |
7230 | } else { | |
7231 | kfree_skb(skb); | |
7232 | kfree(fpl); | |
7233 | } | |
6b06314c JA |
7234 | |
7235 | return 0; | |
7236 | } | |
7237 | ||
7238 | /* | |
7239 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7240 | * causes regular reference counting to break down. We rely on the UNIX | |
7241 | * garbage collection to take care of this problem for us. | |
7242 | */ | |
7243 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7244 | { | |
7245 | unsigned left, total; | |
7246 | int ret = 0; | |
7247 | ||
7248 | total = 0; | |
7249 | left = ctx->nr_user_files; | |
7250 | while (left) { | |
7251 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7252 | |
7253 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7254 | if (ret) | |
7255 | break; | |
7256 | left -= this_files; | |
7257 | total += this_files; | |
7258 | } | |
7259 | ||
7260 | if (!ret) | |
7261 | return 0; | |
7262 | ||
7263 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7264 | struct file *file = io_file_from_index(ctx, total); |
7265 | ||
7266 | if (file) | |
7267 | fput(file); | |
6b06314c JA |
7268 | total++; |
7269 | } | |
7270 | ||
7271 | return ret; | |
7272 | } | |
7273 | #else | |
7274 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7275 | { | |
7276 | return 0; | |
7277 | } | |
7278 | #endif | |
7279 | ||
269bbe5f | 7280 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7281 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7282 | { |
7283 | int i; | |
7284 | ||
7285 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7286 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7287 | unsigned this_files; |
7288 | ||
7289 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7290 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7291 | GFP_KERNEL); | |
7292 | if (!table->files) | |
7293 | break; | |
7294 | nr_files -= this_files; | |
7295 | } | |
7296 | ||
7297 | if (i == nr_tables) | |
7298 | return 0; | |
7299 | ||
7300 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7301 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7302 | kfree(table->files); |
7303 | } | |
7304 | return 1; | |
7305 | } | |
7306 | ||
50238531 | 7307 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7308 | { |
50238531 | 7309 | struct file *file = prsrc->file; |
05f3fb3c JA |
7310 | #if defined(CONFIG_UNIX) |
7311 | struct sock *sock = ctx->ring_sock->sk; | |
7312 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7313 | struct sk_buff *skb; | |
7314 | int i; | |
7315 | ||
7316 | __skb_queue_head_init(&list); | |
7317 | ||
7318 | /* | |
7319 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7320 | * remove this entry and rearrange the file array. | |
7321 | */ | |
7322 | skb = skb_dequeue(head); | |
7323 | while (skb) { | |
7324 | struct scm_fp_list *fp; | |
7325 | ||
7326 | fp = UNIXCB(skb).fp; | |
7327 | for (i = 0; i < fp->count; i++) { | |
7328 | int left; | |
7329 | ||
7330 | if (fp->fp[i] != file) | |
7331 | continue; | |
7332 | ||
7333 | unix_notinflight(fp->user, fp->fp[i]); | |
7334 | left = fp->count - 1 - i; | |
7335 | if (left) { | |
7336 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7337 | left * sizeof(struct file *)); | |
7338 | } | |
7339 | fp->count--; | |
7340 | if (!fp->count) { | |
7341 | kfree_skb(skb); | |
7342 | skb = NULL; | |
7343 | } else { | |
7344 | __skb_queue_tail(&list, skb); | |
7345 | } | |
7346 | fput(file); | |
7347 | file = NULL; | |
7348 | break; | |
7349 | } | |
7350 | ||
7351 | if (!file) | |
7352 | break; | |
7353 | ||
7354 | __skb_queue_tail(&list, skb); | |
7355 | ||
7356 | skb = skb_dequeue(head); | |
7357 | } | |
7358 | ||
7359 | if (skb_peek(&list)) { | |
7360 | spin_lock_irq(&head->lock); | |
7361 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7362 | __skb_queue_tail(head, skb); | |
7363 | spin_unlock_irq(&head->lock); | |
7364 | } | |
7365 | #else | |
7366 | fput(file); | |
7367 | #endif | |
7368 | } | |
7369 | ||
269bbe5f | 7370 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7371 | { |
269bbe5f BM |
7372 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7373 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7374 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7375 | |
269bbe5f BM |
7376 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7377 | list_del(&prsrc->list); | |
50238531 | 7378 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7379 | kfree(prsrc); |
65e19f54 | 7380 | } |
05589553 | 7381 | |
05589553 XW |
7382 | percpu_ref_exit(&ref_node->refs); |
7383 | kfree(ref_node); | |
269bbe5f | 7384 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7385 | } |
65e19f54 | 7386 | |
269bbe5f | 7387 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7388 | { |
7389 | struct io_ring_ctx *ctx; | |
7390 | struct llist_node *node; | |
7391 | ||
269bbe5f BM |
7392 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7393 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7394 | |
7395 | while (node) { | |
269bbe5f | 7396 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7397 | struct llist_node *next = node->next; |
7398 | ||
269bbe5f BM |
7399 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7400 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7401 | node = next; |
7402 | } | |
7403 | } | |
7404 | ||
ea64ec02 PB |
7405 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7406 | unsigned i) | |
2faf852d | 7407 | { |
ea64ec02 PB |
7408 | struct fixed_rsrc_table *table; |
7409 | ||
7410 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7411 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7412 | } | |
7413 | ||
00835dce | 7414 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7415 | { |
269bbe5f BM |
7416 | struct fixed_rsrc_ref_node *ref_node; |
7417 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7418 | struct io_ring_ctx *ctx; |
e297822b | 7419 | bool first_add = false; |
4a38aed2 | 7420 | int delay = HZ; |
65e19f54 | 7421 | |
269bbe5f BM |
7422 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7423 | data = ref_node->rsrc_data; | |
e297822b PB |
7424 | ctx = data->ctx; |
7425 | ||
2a63b2d9 | 7426 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7427 | ref_node->done = true; |
7428 | ||
d67d2263 BM |
7429 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7430 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7431 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7432 | /* recycle ref nodes in order */ |
7433 | if (!ref_node->done) | |
7434 | break; | |
7435 | list_del(&ref_node->node); | |
269bbe5f | 7436 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7437 | } |
2a63b2d9 | 7438 | io_rsrc_ref_unlock(ctx); |
05589553 | 7439 | |
e297822b | 7440 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7441 | delay = 0; |
05589553 | 7442 | |
4a38aed2 | 7443 | if (!delay) |
269bbe5f | 7444 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7445 | else if (first_add) |
269bbe5f | 7446 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7447 | } |
65e19f54 | 7448 | |
6802535d | 7449 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7450 | struct io_ring_ctx *ctx) |
05f3fb3c | 7451 | { |
269bbe5f | 7452 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7453 | |
05589553 XW |
7454 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7455 | if (!ref_node) | |
3e2224c5 | 7456 | return NULL; |
05f3fb3c | 7457 | |
00835dce | 7458 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7459 | 0, GFP_KERNEL)) { |
7460 | kfree(ref_node); | |
3e2224c5 | 7461 | return NULL; |
05589553 XW |
7462 | } |
7463 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7464 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7465 | ref_node->done = false; |
05589553 | 7466 | return ref_node; |
05589553 XW |
7467 | } |
7468 | ||
bc9744cd PB |
7469 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7470 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7471 | { |
269bbe5f | 7472 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7473 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7474 | } |
7475 | ||
269bbe5f | 7476 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7477 | { |
7478 | percpu_ref_exit(&ref_node->refs); | |
7479 | kfree(ref_node); | |
65e19f54 JA |
7480 | } |
7481 | ||
ea64ec02 | 7482 | |
6b06314c JA |
7483 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7484 | unsigned nr_args) | |
7485 | { | |
7486 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7487 | unsigned nr_tables, i; |
05f3fb3c | 7488 | struct file *file; |
600cf3f8 | 7489 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7490 | struct fixed_rsrc_ref_node *ref_node; |
7491 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7492 | |
05f3fb3c | 7493 | if (ctx->file_data) |
6b06314c JA |
7494 | return -EBUSY; |
7495 | if (!nr_args) | |
7496 | return -EINVAL; | |
7497 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7498 | return -EMFILE; | |
7499 | ||
1ad555c6 | 7500 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7501 | if (!file_data) |
05f3fb3c | 7502 | return -ENOMEM; |
13770a71 | 7503 | ctx->file_data = file_data; |
05f3fb3c | 7504 | |
65e19f54 | 7505 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7506 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7507 | GFP_KERNEL); |
600cf3f8 PB |
7508 | if (!file_data->table) |
7509 | goto out_free; | |
05f3fb3c | 7510 | |
600cf3f8 | 7511 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7512 | goto out_free; |
65e19f54 | 7513 | |
08a45173 | 7514 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7515 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7516 | ret = -EFAULT; | |
7517 | goto out_fput; | |
7518 | } | |
08a45173 | 7519 | /* allow sparse sets */ |
600cf3f8 | 7520 | if (fd == -1) |
08a45173 | 7521 | continue; |
6b06314c | 7522 | |
05f3fb3c | 7523 | file = fget(fd); |
6b06314c | 7524 | ret = -EBADF; |
05f3fb3c | 7525 | if (!file) |
600cf3f8 | 7526 | goto out_fput; |
05f3fb3c | 7527 | |
6b06314c JA |
7528 | /* |
7529 | * Don't allow io_uring instances to be registered. If UNIX | |
7530 | * isn't enabled, then this causes a reference cycle and this | |
7531 | * instance can never get freed. If UNIX is enabled we'll | |
7532 | * handle it just fine, but there's still no point in allowing | |
7533 | * a ring fd as it doesn't support regular read/write anyway. | |
7534 | */ | |
05f3fb3c JA |
7535 | if (file->f_op == &io_uring_fops) { |
7536 | fput(file); | |
600cf3f8 | 7537 | goto out_fput; |
6b06314c | 7538 | } |
ea64ec02 | 7539 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7540 | } |
7541 | ||
6b06314c | 7542 | ret = io_sqe_files_scm(ctx); |
05589553 | 7543 | if (ret) { |
6b06314c | 7544 | io_sqe_files_unregister(ctx); |
05589553 XW |
7545 | return ret; |
7546 | } | |
6b06314c | 7547 | |
bc9744cd | 7548 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7549 | if (!ref_node) { |
05589553 | 7550 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7551 | return -ENOMEM; |
05589553 | 7552 | } |
bc9744cd | 7553 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7554 | |
d67d2263 | 7555 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7556 | return ret; |
600cf3f8 PB |
7557 | out_fput: |
7558 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7559 | file = io_file_from_index(ctx, i); | |
7560 | if (file) | |
7561 | fput(file); | |
7562 | } | |
7563 | for (i = 0; i < nr_tables; i++) | |
7564 | kfree(file_data->table[i].files); | |
7565 | ctx->nr_user_files = 0; | |
600cf3f8 | 7566 | out_free: |
1ad555c6 | 7567 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7568 | ctx->file_data = NULL; |
6b06314c JA |
7569 | return ret; |
7570 | } | |
7571 | ||
c3a31e60 JA |
7572 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7573 | int index) | |
7574 | { | |
7575 | #if defined(CONFIG_UNIX) | |
7576 | struct sock *sock = ctx->ring_sock->sk; | |
7577 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7578 | struct sk_buff *skb; | |
7579 | ||
7580 | /* | |
7581 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7582 | * file set. If there's no room, fall back to allocating a new skb | |
7583 | * and filling it in. | |
7584 | */ | |
7585 | spin_lock_irq(&head->lock); | |
7586 | skb = skb_peek(head); | |
7587 | if (skb) { | |
7588 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7589 | ||
7590 | if (fpl->count < SCM_MAX_FD) { | |
7591 | __skb_unlink(skb, head); | |
7592 | spin_unlock_irq(&head->lock); | |
7593 | fpl->fp[fpl->count] = get_file(file); | |
7594 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7595 | fpl->count++; | |
7596 | spin_lock_irq(&head->lock); | |
7597 | __skb_queue_head(head, skb); | |
7598 | } else { | |
7599 | skb = NULL; | |
7600 | } | |
7601 | } | |
7602 | spin_unlock_irq(&head->lock); | |
7603 | ||
7604 | if (skb) { | |
7605 | fput(file); | |
7606 | return 0; | |
7607 | } | |
7608 | ||
7609 | return __io_sqe_files_scm(ctx, 1, index); | |
7610 | #else | |
7611 | return 0; | |
7612 | #endif | |
7613 | } | |
7614 | ||
50238531 | 7615 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7616 | { |
269bbe5f BM |
7617 | struct io_rsrc_put *prsrc; |
7618 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7619 | |
269bbe5f BM |
7620 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7621 | if (!prsrc) | |
a5318d3c | 7622 | return -ENOMEM; |
05f3fb3c | 7623 | |
50238531 | 7624 | prsrc->rsrc = rsrc; |
269bbe5f | 7625 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7626 | |
a5318d3c | 7627 | return 0; |
05f3fb3c JA |
7628 | } |
7629 | ||
269bbe5f BM |
7630 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7631 | struct file *file) | |
7632 | { | |
50238531 | 7633 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7634 | } |
7635 | ||
05f3fb3c | 7636 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7637 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7638 | unsigned nr_args) |
7639 | { | |
269bbe5f BM |
7640 | struct fixed_rsrc_data *data = ctx->file_data; |
7641 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7642 | struct file *file, **file_slot; |
c3a31e60 JA |
7643 | __s32 __user *fds; |
7644 | int fd, i, err; | |
7645 | __u32 done; | |
05589553 | 7646 | bool needs_switch = false; |
c3a31e60 | 7647 | |
05f3fb3c | 7648 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7649 | return -EOVERFLOW; |
7650 | if (done > ctx->nr_user_files) | |
7651 | return -EINVAL; | |
7652 | ||
bc9744cd | 7653 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7654 | if (!ref_node) |
7655 | return -ENOMEM; | |
bc9744cd | 7656 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7657 | |
269bbe5f | 7658 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7659 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7660 | err = 0; |
7661 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7662 | err = -EFAULT; | |
7663 | break; | |
7664 | } | |
4e0377a1 | 7665 | if (fd == IORING_REGISTER_FILES_SKIP) |
7666 | continue; | |
7667 | ||
67973b93 | 7668 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7669 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7670 | ||
7671 | if (*file_slot) { | |
7672 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7673 | if (err) |
7674 | break; | |
ea64ec02 | 7675 | *file_slot = NULL; |
05589553 | 7676 | needs_switch = true; |
c3a31e60 JA |
7677 | } |
7678 | if (fd != -1) { | |
c3a31e60 JA |
7679 | file = fget(fd); |
7680 | if (!file) { | |
7681 | err = -EBADF; | |
7682 | break; | |
7683 | } | |
7684 | /* | |
7685 | * Don't allow io_uring instances to be registered. If | |
7686 | * UNIX isn't enabled, then this causes a reference | |
7687 | * cycle and this instance can never get freed. If UNIX | |
7688 | * is enabled we'll handle it just fine, but there's | |
7689 | * still no point in allowing a ring fd as it doesn't | |
7690 | * support regular read/write anyway. | |
7691 | */ | |
7692 | if (file->f_op == &io_uring_fops) { | |
7693 | fput(file); | |
7694 | err = -EBADF; | |
7695 | break; | |
7696 | } | |
e68a3ff8 | 7697 | *file_slot = file; |
c3a31e60 | 7698 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7699 | if (err) { |
e68a3ff8 | 7700 | *file_slot = NULL; |
f3bd9dae | 7701 | fput(file); |
c3a31e60 | 7702 | break; |
f3bd9dae | 7703 | } |
c3a31e60 | 7704 | } |
05f3fb3c JA |
7705 | } |
7706 | ||
05589553 | 7707 | if (needs_switch) { |
b2e96852 | 7708 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7709 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7710 | } else |
269bbe5f | 7711 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7712 | |
7713 | return done ? done : err; | |
7714 | } | |
05589553 | 7715 | |
05f3fb3c JA |
7716 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7717 | unsigned nr_args) | |
7718 | { | |
269bbe5f | 7719 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7720 | |
7721 | if (!ctx->file_data) | |
7722 | return -ENXIO; | |
7723 | if (!nr_args) | |
7724 | return -EINVAL; | |
7725 | if (copy_from_user(&up, arg, sizeof(up))) | |
7726 | return -EFAULT; | |
7727 | if (up.resv) | |
7728 | return -EINVAL; | |
7729 | ||
7730 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7731 | } | |
c3a31e60 | 7732 | |
5280f7e5 | 7733 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7734 | { |
7735 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7736 | ||
5280f7e5 PB |
7737 | req = io_put_req_find_next(req); |
7738 | return req ? &req->work : NULL; | |
7d723065 JA |
7739 | } |
7740 | ||
5aa75ed5 | 7741 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7742 | { |
e941894e | 7743 | struct io_wq_hash *hash; |
24369c2e | 7744 | struct io_wq_data data; |
24369c2e | 7745 | unsigned int concurrency; |
24369c2e | 7746 | |
e941894e JA |
7747 | hash = ctx->hash_map; |
7748 | if (!hash) { | |
7749 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7750 | if (!hash) | |
7751 | return ERR_PTR(-ENOMEM); | |
7752 | refcount_set(&hash->refs, 1); | |
7753 | init_waitqueue_head(&hash->wait); | |
7754 | ctx->hash_map = hash; | |
24369c2e PB |
7755 | } |
7756 | ||
e941894e | 7757 | data.hash = hash; |
e9fd9396 | 7758 | data.free_work = io_free_work; |
f5fa38c5 | 7759 | data.do_work = io_wq_submit_work; |
24369c2e | 7760 | |
d25e3a3d JA |
7761 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7762 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7763 | |
5aa75ed5 | 7764 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7765 | } |
7766 | ||
5aa75ed5 JA |
7767 | static int io_uring_alloc_task_context(struct task_struct *task, |
7768 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7769 | { |
7770 | struct io_uring_task *tctx; | |
d8a6df10 | 7771 | int ret; |
0f212204 JA |
7772 | |
7773 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7774 | if (unlikely(!tctx)) | |
7775 | return -ENOMEM; | |
7776 | ||
d8a6df10 JA |
7777 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7778 | if (unlikely(ret)) { | |
7779 | kfree(tctx); | |
7780 | return ret; | |
7781 | } | |
7782 | ||
5aa75ed5 JA |
7783 | tctx->io_wq = io_init_wq_offload(ctx); |
7784 | if (IS_ERR(tctx->io_wq)) { | |
7785 | ret = PTR_ERR(tctx->io_wq); | |
7786 | percpu_counter_destroy(&tctx->inflight); | |
7787 | kfree(tctx); | |
7788 | return ret; | |
7789 | } | |
7790 | ||
0f212204 JA |
7791 | xa_init(&tctx->xa); |
7792 | init_waitqueue_head(&tctx->wait); | |
7793 | tctx->last = NULL; | |
fdaf083c | 7794 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7795 | task->io_uring = tctx; |
7cbf1722 JA |
7796 | spin_lock_init(&tctx->task_lock); |
7797 | INIT_WQ_LIST(&tctx->task_list); | |
7798 | tctx->task_state = 0; | |
7799 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7800 | return 0; |
7801 | } | |
7802 | ||
7803 | void __io_uring_free(struct task_struct *tsk) | |
7804 | { | |
7805 | struct io_uring_task *tctx = tsk->io_uring; | |
7806 | ||
7807 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7808 | WARN_ON_ONCE(tctx->io_wq); |
7809 | ||
d8a6df10 | 7810 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7811 | kfree(tctx); |
7812 | tsk->io_uring = NULL; | |
7813 | } | |
7814 | ||
7e84e1c7 SG |
7815 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7816 | struct io_uring_params *p) | |
2b188cc1 JA |
7817 | { |
7818 | int ret; | |
7819 | ||
d25e3a3d JA |
7820 | /* Retain compatibility with failing for an invalid attach attempt */ |
7821 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7822 | IORING_SETUP_ATTACH_WQ) { | |
7823 | struct fd f; | |
7824 | ||
7825 | f = fdget(p->wq_fd); | |
7826 | if (!f.file) | |
7827 | return -ENXIO; | |
7828 | if (f.file->f_op != &io_uring_fops) { | |
7829 | fdput(f); | |
7830 | return -EINVAL; | |
7831 | } | |
7832 | fdput(f); | |
7833 | } | |
6c271ce2 | 7834 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7835 | struct task_struct *tsk; |
534ca6d6 | 7836 | struct io_sq_data *sqd; |
26984fbf | 7837 | bool attached; |
534ca6d6 | 7838 | |
3ec482d1 | 7839 | ret = -EPERM; |
ce59fc69 | 7840 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7841 | goto err; |
7842 | ||
26984fbf | 7843 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7844 | if (IS_ERR(sqd)) { |
7845 | ret = PTR_ERR(sqd); | |
7846 | goto err; | |
7847 | } | |
69fb2131 | 7848 | |
7c30f36a | 7849 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7850 | ctx->sq_data = sqd; |
917257da JA |
7851 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7852 | if (!ctx->sq_thread_idle) | |
7853 | ctx->sq_thread_idle = HZ; | |
7854 | ||
26984fbf | 7855 | ret = 0; |
78d7f6ba | 7856 | io_sq_thread_park(sqd); |
26984fbf PB |
7857 | /* don't attach to a dying SQPOLL thread, would be racy */ |
7858 | if (attached && !sqd->thread) { | |
7859 | ret = -ENXIO; | |
7860 | } else { | |
7861 | list_add(&ctx->sqd_list, &sqd->ctx_list); | |
7862 | io_sqd_update_thread_idle(sqd); | |
7863 | } | |
78d7f6ba PB |
7864 | io_sq_thread_unpark(sqd); |
7865 | ||
26984fbf PB |
7866 | if (ret < 0) { |
7867 | io_put_sq_data(sqd); | |
7868 | ctx->sq_data = NULL; | |
7869 | return ret; | |
7870 | } else if (attached) { | |
5aa75ed5 | 7871 | return 0; |
26984fbf | 7872 | } |
aa06165d | 7873 | |
6c271ce2 | 7874 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7875 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7876 | |
917257da | 7877 | ret = -EINVAL; |
44a9bd18 | 7878 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7879 | goto err_sqpoll; |
7889f44d | 7880 | if (!cpu_online(cpu)) |
e8f98f24 | 7881 | goto err_sqpoll; |
917257da | 7882 | |
37d1e2e3 | 7883 | sqd->sq_cpu = cpu; |
6c271ce2 | 7884 | } else { |
37d1e2e3 | 7885 | sqd->sq_cpu = -1; |
6c271ce2 | 7886 | } |
37d1e2e3 JA |
7887 | |
7888 | sqd->task_pid = current->pid; | |
5c2469e0 | 7889 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7890 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7891 | if (IS_ERR(tsk)) { | |
7892 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7893 | goto err_sqpoll; |
6c271ce2 | 7894 | } |
97a73a0f | 7895 | |
46fe18b1 | 7896 | sqd->thread = tsk; |
97a73a0f | 7897 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7898 | wake_up_new_task(tsk); |
0f212204 JA |
7899 | if (ret) |
7900 | goto err; | |
6c271ce2 JA |
7901 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7902 | /* Can't have SQ_AFF without SQPOLL */ | |
7903 | ret = -EINVAL; | |
7904 | goto err; | |
7905 | } | |
7906 | ||
2b188cc1 JA |
7907 | return 0; |
7908 | err: | |
37d1e2e3 | 7909 | io_sq_thread_finish(ctx); |
2b188cc1 | 7910 | return ret; |
e8f98f24 JA |
7911 | err_sqpoll: |
7912 | complete(&ctx->sq_data->exited); | |
7913 | goto err; | |
2b188cc1 JA |
7914 | } |
7915 | ||
a087e2b5 BM |
7916 | static inline void __io_unaccount_mem(struct user_struct *user, |
7917 | unsigned long nr_pages) | |
2b188cc1 JA |
7918 | { |
7919 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7920 | } | |
7921 | ||
a087e2b5 BM |
7922 | static inline int __io_account_mem(struct user_struct *user, |
7923 | unsigned long nr_pages) | |
2b188cc1 JA |
7924 | { |
7925 | unsigned long page_limit, cur_pages, new_pages; | |
7926 | ||
7927 | /* Don't allow more pages than we can safely lock */ | |
7928 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7929 | ||
7930 | do { | |
7931 | cur_pages = atomic_long_read(&user->locked_vm); | |
7932 | new_pages = cur_pages + nr_pages; | |
7933 | if (new_pages > page_limit) | |
7934 | return -ENOMEM; | |
7935 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7936 | new_pages) != cur_pages); | |
7937 | ||
7938 | return 0; | |
7939 | } | |
7940 | ||
26bfa89e | 7941 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7942 | { |
62e398be | 7943 | if (ctx->user) |
a087e2b5 | 7944 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7945 | |
26bfa89e JA |
7946 | if (ctx->mm_account) |
7947 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7948 | } |
7949 | ||
26bfa89e | 7950 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7951 | { |
30975825 BM |
7952 | int ret; |
7953 | ||
62e398be | 7954 | if (ctx->user) { |
30975825 BM |
7955 | ret = __io_account_mem(ctx->user, nr_pages); |
7956 | if (ret) | |
7957 | return ret; | |
7958 | } | |
7959 | ||
26bfa89e JA |
7960 | if (ctx->mm_account) |
7961 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7962 | |
7963 | return 0; | |
7964 | } | |
7965 | ||
2b188cc1 JA |
7966 | static void io_mem_free(void *ptr) |
7967 | { | |
52e04ef4 MR |
7968 | struct page *page; |
7969 | ||
7970 | if (!ptr) | |
7971 | return; | |
2b188cc1 | 7972 | |
52e04ef4 | 7973 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7974 | if (put_page_testzero(page)) |
7975 | free_compound_page(page); | |
7976 | } | |
7977 | ||
7978 | static void *io_mem_alloc(size_t size) | |
7979 | { | |
7980 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 7981 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
7982 | |
7983 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7984 | } | |
7985 | ||
75b28aff HV |
7986 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7987 | size_t *sq_offset) | |
7988 | { | |
7989 | struct io_rings *rings; | |
7990 | size_t off, sq_array_size; | |
7991 | ||
7992 | off = struct_size(rings, cqes, cq_entries); | |
7993 | if (off == SIZE_MAX) | |
7994 | return SIZE_MAX; | |
7995 | ||
7996 | #ifdef CONFIG_SMP | |
7997 | off = ALIGN(off, SMP_CACHE_BYTES); | |
7998 | if (off == 0) | |
7999 | return SIZE_MAX; | |
8000 | #endif | |
8001 | ||
b36200f5 DV |
8002 | if (sq_offset) |
8003 | *sq_offset = off; | |
8004 | ||
75b28aff HV |
8005 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8006 | if (sq_array_size == SIZE_MAX) | |
8007 | return SIZE_MAX; | |
8008 | ||
8009 | if (check_add_overflow(off, sq_array_size, &off)) | |
8010 | return SIZE_MAX; | |
8011 | ||
75b28aff HV |
8012 | return off; |
8013 | } | |
8014 | ||
0a96bbe4 | 8015 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8016 | { |
8017 | int i, j; | |
8018 | ||
8019 | if (!ctx->user_bufs) | |
8020 | return -ENXIO; | |
8021 | ||
8022 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8023 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8024 | ||
8025 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8026 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8027 | |
de293938 | 8028 | if (imu->acct_pages) |
26bfa89e | 8029 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8030 | kvfree(imu->bvec); |
edafccee JA |
8031 | imu->nr_bvecs = 0; |
8032 | } | |
8033 | ||
8034 | kfree(ctx->user_bufs); | |
8035 | ctx->user_bufs = NULL; | |
8036 | ctx->nr_user_bufs = 0; | |
8037 | return 0; | |
8038 | } | |
8039 | ||
8040 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8041 | void __user *arg, unsigned index) | |
8042 | { | |
8043 | struct iovec __user *src; | |
8044 | ||
8045 | #ifdef CONFIG_COMPAT | |
8046 | if (ctx->compat) { | |
8047 | struct compat_iovec __user *ciovs; | |
8048 | struct compat_iovec ciov; | |
8049 | ||
8050 | ciovs = (struct compat_iovec __user *) arg; | |
8051 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8052 | return -EFAULT; | |
8053 | ||
d55e5f5b | 8054 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8055 | dst->iov_len = ciov.iov_len; |
8056 | return 0; | |
8057 | } | |
8058 | #endif | |
8059 | src = (struct iovec __user *) arg; | |
8060 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8061 | return -EFAULT; | |
8062 | return 0; | |
8063 | } | |
8064 | ||
de293938 JA |
8065 | /* |
8066 | * Not super efficient, but this is just a registration time. And we do cache | |
8067 | * the last compound head, so generally we'll only do a full search if we don't | |
8068 | * match that one. | |
8069 | * | |
8070 | * We check if the given compound head page has already been accounted, to | |
8071 | * avoid double accounting it. This allows us to account the full size of the | |
8072 | * page, not just the constituent pages of a huge page. | |
8073 | */ | |
8074 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8075 | int nr_pages, struct page *hpage) | |
8076 | { | |
8077 | int i, j; | |
8078 | ||
8079 | /* check current page array */ | |
8080 | for (i = 0; i < nr_pages; i++) { | |
8081 | if (!PageCompound(pages[i])) | |
8082 | continue; | |
8083 | if (compound_head(pages[i]) == hpage) | |
8084 | return true; | |
8085 | } | |
8086 | ||
8087 | /* check previously registered pages */ | |
8088 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8089 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8090 | ||
8091 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8092 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8093 | continue; | |
8094 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8095 | return true; | |
8096 | } | |
8097 | } | |
8098 | ||
8099 | return false; | |
8100 | } | |
8101 | ||
8102 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8103 | int nr_pages, struct io_mapped_ubuf *imu, | |
8104 | struct page **last_hpage) | |
8105 | { | |
8106 | int i, ret; | |
8107 | ||
8108 | for (i = 0; i < nr_pages; i++) { | |
8109 | if (!PageCompound(pages[i])) { | |
8110 | imu->acct_pages++; | |
8111 | } else { | |
8112 | struct page *hpage; | |
8113 | ||
8114 | hpage = compound_head(pages[i]); | |
8115 | if (hpage == *last_hpage) | |
8116 | continue; | |
8117 | *last_hpage = hpage; | |
8118 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8119 | continue; | |
8120 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8121 | } | |
8122 | } | |
8123 | ||
8124 | if (!imu->acct_pages) | |
8125 | return 0; | |
8126 | ||
26bfa89e | 8127 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8128 | if (ret) |
8129 | imu->acct_pages = 0; | |
8130 | return ret; | |
8131 | } | |
8132 | ||
0a96bbe4 BM |
8133 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8134 | struct io_mapped_ubuf *imu, | |
8135 | struct page **last_hpage) | |
edafccee JA |
8136 | { |
8137 | struct vm_area_struct **vmas = NULL; | |
8138 | struct page **pages = NULL; | |
0a96bbe4 BM |
8139 | unsigned long off, start, end, ubuf; |
8140 | size_t size; | |
8141 | int ret, pret, nr_pages, i; | |
8142 | ||
8143 | ubuf = (unsigned long) iov->iov_base; | |
8144 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8145 | start = ubuf >> PAGE_SHIFT; | |
8146 | nr_pages = end - start; | |
8147 | ||
8148 | ret = -ENOMEM; | |
8149 | ||
8150 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8151 | if (!pages) | |
8152 | goto done; | |
8153 | ||
8154 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8155 | GFP_KERNEL); | |
8156 | if (!vmas) | |
8157 | goto done; | |
edafccee | 8158 | |
0a96bbe4 BM |
8159 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8160 | GFP_KERNEL); | |
8161 | if (!imu->bvec) | |
8162 | goto done; | |
8163 | ||
8164 | ret = 0; | |
8165 | mmap_read_lock(current->mm); | |
8166 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8167 | pages, vmas); | |
8168 | if (pret == nr_pages) { | |
8169 | /* don't support file backed memory */ | |
8170 | for (i = 0; i < nr_pages; i++) { | |
8171 | struct vm_area_struct *vma = vmas[i]; | |
8172 | ||
8173 | if (vma->vm_file && | |
8174 | !is_file_hugepages(vma->vm_file)) { | |
8175 | ret = -EOPNOTSUPP; | |
8176 | break; | |
8177 | } | |
8178 | } | |
8179 | } else { | |
8180 | ret = pret < 0 ? pret : -EFAULT; | |
8181 | } | |
8182 | mmap_read_unlock(current->mm); | |
8183 | if (ret) { | |
8184 | /* | |
8185 | * if we did partial map, or found file backed vmas, | |
8186 | * release any pages we did get | |
8187 | */ | |
8188 | if (pret > 0) | |
8189 | unpin_user_pages(pages, pret); | |
8190 | kvfree(imu->bvec); | |
8191 | goto done; | |
8192 | } | |
8193 | ||
8194 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8195 | if (ret) { | |
8196 | unpin_user_pages(pages, pret); | |
8197 | kvfree(imu->bvec); | |
8198 | goto done; | |
8199 | } | |
8200 | ||
8201 | off = ubuf & ~PAGE_MASK; | |
8202 | size = iov->iov_len; | |
8203 | for (i = 0; i < nr_pages; i++) { | |
8204 | size_t vec_len; | |
8205 | ||
8206 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8207 | imu->bvec[i].bv_page = pages[i]; | |
8208 | imu->bvec[i].bv_len = vec_len; | |
8209 | imu->bvec[i].bv_offset = off; | |
8210 | off = 0; | |
8211 | size -= vec_len; | |
8212 | } | |
8213 | /* store original address for later verification */ | |
8214 | imu->ubuf = ubuf; | |
8215 | imu->len = iov->iov_len; | |
8216 | imu->nr_bvecs = nr_pages; | |
8217 | ret = 0; | |
8218 | done: | |
8219 | kvfree(pages); | |
8220 | kvfree(vmas); | |
8221 | return ret; | |
8222 | } | |
8223 | ||
2b358604 | 8224 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8225 | { |
edafccee JA |
8226 | if (ctx->user_bufs) |
8227 | return -EBUSY; | |
8228 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8229 | return -EINVAL; | |
8230 | ||
8231 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8232 | GFP_KERNEL); | |
8233 | if (!ctx->user_bufs) | |
8234 | return -ENOMEM; | |
8235 | ||
2b358604 BM |
8236 | return 0; |
8237 | } | |
edafccee | 8238 | |
2b358604 BM |
8239 | static int io_buffer_validate(struct iovec *iov) |
8240 | { | |
8241 | /* | |
8242 | * Don't impose further limits on the size and buffer | |
8243 | * constraints here, we'll -EINVAL later when IO is | |
8244 | * submitted if they are wrong. | |
8245 | */ | |
8246 | if (!iov->iov_base || !iov->iov_len) | |
8247 | return -EFAULT; | |
edafccee | 8248 | |
2b358604 BM |
8249 | /* arbitrary limit, but we need something */ |
8250 | if (iov->iov_len > SZ_1G) | |
8251 | return -EFAULT; | |
edafccee | 8252 | |
2b358604 BM |
8253 | return 0; |
8254 | } | |
edafccee | 8255 | |
2b358604 BM |
8256 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8257 | unsigned int nr_args) | |
8258 | { | |
8259 | int i, ret; | |
8260 | struct iovec iov; | |
8261 | struct page *last_hpage = NULL; | |
edafccee | 8262 | |
2b358604 BM |
8263 | ret = io_buffers_map_alloc(ctx, nr_args); |
8264 | if (ret) | |
8265 | return ret; | |
edafccee | 8266 | |
edafccee JA |
8267 | for (i = 0; i < nr_args; i++) { |
8268 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8269 | |
edafccee JA |
8270 | ret = io_copy_iov(ctx, &iov, arg, i); |
8271 | if (ret) | |
0a96bbe4 | 8272 | break; |
de293938 | 8273 | |
2b358604 BM |
8274 | ret = io_buffer_validate(&iov); |
8275 | if (ret) | |
0a96bbe4 | 8276 | break; |
edafccee | 8277 | |
0a96bbe4 BM |
8278 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8279 | if (ret) | |
8280 | break; | |
edafccee JA |
8281 | |
8282 | ctx->nr_user_bufs++; | |
8283 | } | |
0a96bbe4 BM |
8284 | |
8285 | if (ret) | |
8286 | io_sqe_buffers_unregister(ctx); | |
8287 | ||
edafccee JA |
8288 | return ret; |
8289 | } | |
8290 | ||
9b402849 JA |
8291 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8292 | { | |
8293 | __s32 __user *fds = arg; | |
8294 | int fd; | |
8295 | ||
8296 | if (ctx->cq_ev_fd) | |
8297 | return -EBUSY; | |
8298 | ||
8299 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8300 | return -EFAULT; | |
8301 | ||
8302 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8303 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8304 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8305 | ctx->cq_ev_fd = NULL; | |
8306 | return ret; | |
8307 | } | |
8308 | ||
8309 | return 0; | |
8310 | } | |
8311 | ||
8312 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8313 | { | |
8314 | if (ctx->cq_ev_fd) { | |
8315 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8316 | ctx->cq_ev_fd = NULL; | |
8317 | return 0; | |
8318 | } | |
8319 | ||
8320 | return -ENXIO; | |
8321 | } | |
8322 | ||
5a2e745d JA |
8323 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8324 | { | |
9e15c3a0 JA |
8325 | struct io_buffer *buf; |
8326 | unsigned long index; | |
8327 | ||
8328 | xa_for_each(&ctx->io_buffers, index, buf) | |
8329 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8330 | } |
8331 | ||
68e68ee6 | 8332 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8333 | { |
68e68ee6 | 8334 | struct io_kiocb *req, *nxt; |
1b4c351f | 8335 | |
68e68ee6 JA |
8336 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8337 | if (tsk && req->task != tsk) | |
8338 | continue; | |
1b4c351f JA |
8339 | list_del(&req->compl.list); |
8340 | kmem_cache_free(req_cachep, req); | |
8341 | } | |
8342 | } | |
8343 | ||
4010fec4 | 8344 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8345 | { |
bf019da7 | 8346 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8347 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8348 | |
9a4fdbd8 JA |
8349 | mutex_lock(&ctx->uring_lock); |
8350 | ||
8e5c66c4 | 8351 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8352 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8353 | submit_state->reqs); | |
8e5c66c4 PB |
8354 | submit_state->free_reqs = 0; |
8355 | } | |
9a4fdbd8 JA |
8356 | |
8357 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8358 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8359 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8360 | spin_unlock_irq(&ctx->completion_lock); |
8361 | ||
e5547d2c PB |
8362 | io_req_cache_free(&cs->free_list, NULL); |
8363 | ||
9a4fdbd8 JA |
8364 | mutex_unlock(&ctx->uring_lock); |
8365 | } | |
8366 | ||
2b188cc1 JA |
8367 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8368 | { | |
04fc6c80 PB |
8369 | /* |
8370 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8371 | * and they are free to do so while still holding uring_lock or |
8372 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8373 | */ |
8374 | mutex_lock(&ctx->uring_lock); | |
8375 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8376 | spin_lock_irq(&ctx->completion_lock); |
8377 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8378 | |
37d1e2e3 | 8379 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8380 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8381 | |
37d1e2e3 | 8382 | if (ctx->mm_account) { |
2aede0e4 JA |
8383 | mmdrop(ctx->mm_account); |
8384 | ctx->mm_account = NULL; | |
30975825 | 8385 | } |
def596e9 | 8386 | |
8bad28d8 | 8387 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8388 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8389 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8390 | io_eventfd_unregister(ctx); |
5a2e745d | 8391 | io_destroy_buffers(ctx); |
def596e9 | 8392 | |
2b188cc1 | 8393 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8394 | if (ctx->ring_sock) { |
8395 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8396 | sock_release(ctx->ring_sock); |
355e8d26 | 8397 | } |
2b188cc1 JA |
8398 | #endif |
8399 | ||
75b28aff | 8400 | io_mem_free(ctx->rings); |
2b188cc1 | 8401 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8402 | |
8403 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8404 | free_uid(ctx->user); |
4010fec4 | 8405 | io_req_caches_free(ctx); |
e941894e JA |
8406 | if (ctx->hash_map) |
8407 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8408 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8409 | kfree(ctx); |
8410 | } | |
8411 | ||
8412 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8413 | { | |
8414 | struct io_ring_ctx *ctx = file->private_data; | |
8415 | __poll_t mask = 0; | |
8416 | ||
8417 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8418 | /* |
8419 | * synchronizes with barrier from wq_has_sleeper call in | |
8420 | * io_commit_cqring | |
8421 | */ | |
2b188cc1 | 8422 | smp_rmb(); |
90554200 | 8423 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8424 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8425 | |
8426 | /* | |
8427 | * Don't flush cqring overflow list here, just do a simple check. | |
8428 | * Otherwise there could possible be ABBA deadlock: | |
8429 | * CPU0 CPU1 | |
8430 | * ---- ---- | |
8431 | * lock(&ctx->uring_lock); | |
8432 | * lock(&ep->mtx); | |
8433 | * lock(&ctx->uring_lock); | |
8434 | * lock(&ep->mtx); | |
8435 | * | |
8436 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8437 | * pushs them to do the flush. | |
8438 | */ | |
8439 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8440 | mask |= EPOLLIN | EPOLLRDNORM; |
8441 | ||
8442 | return mask; | |
8443 | } | |
8444 | ||
8445 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8446 | { | |
8447 | struct io_ring_ctx *ctx = file->private_data; | |
8448 | ||
8449 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8450 | } | |
8451 | ||
0bead8cd | 8452 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8453 | { |
4379bf8b | 8454 | const struct cred *creds; |
071698e1 | 8455 | |
61cf9370 | 8456 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8457 | if (creds) { |
8458 | put_cred(creds); | |
0bead8cd | 8459 | return 0; |
1e6fa521 | 8460 | } |
0bead8cd YD |
8461 | |
8462 | return -EINVAL; | |
8463 | } | |
8464 | ||
ba50a036 | 8465 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8466 | { |
28c4721b | 8467 | struct callback_head *work, *next; |
ba50a036 | 8468 | bool executed = false; |
7c25c0d1 JA |
8469 | |
8470 | do { | |
28c4721b | 8471 | work = xchg(&ctx->exit_task_work, NULL); |
7c25c0d1 JA |
8472 | if (!work) |
8473 | break; | |
8474 | ||
8475 | do { | |
8476 | next = work->next; | |
8477 | work->func(work); | |
8478 | work = next; | |
8479 | cond_resched(); | |
8480 | } while (work); | |
ba50a036 | 8481 | executed = true; |
7c25c0d1 | 8482 | } while (1); |
ba50a036 PB |
8483 | |
8484 | return executed; | |
7c25c0d1 JA |
8485 | } |
8486 | ||
d56d938b PB |
8487 | struct io_tctx_exit { |
8488 | struct callback_head task_work; | |
8489 | struct completion completion; | |
baf186c4 | 8490 | struct io_ring_ctx *ctx; |
d56d938b PB |
8491 | }; |
8492 | ||
8493 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8494 | { | |
8495 | struct io_uring_task *tctx = current->io_uring; | |
8496 | struct io_tctx_exit *work; | |
8497 | ||
8498 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8499 | /* | |
8500 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8501 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8502 | */ | |
8503 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8504 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8505 | complete(&work->completion); |
8506 | } | |
8507 | ||
85faa7b8 JA |
8508 | static void io_ring_exit_work(struct work_struct *work) |
8509 | { | |
d56d938b | 8510 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8511 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8512 | struct io_tctx_exit exit; |
8513 | struct io_tctx_node *node; | |
8514 | int ret; | |
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); |
b5bb3a24 PB |
8524 | |
8525 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8526 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8527 | |
8528 | mutex_lock(&ctx->uring_lock); | |
8529 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8530 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8531 | ||
d56d938b PB |
8532 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8533 | ctx_node); | |
baf186c4 | 8534 | exit.ctx = ctx; |
d56d938b PB |
8535 | init_completion(&exit.completion); |
8536 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8537 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8538 | if (WARN_ON_ONCE(ret)) | |
8539 | continue; | |
8540 | wake_up_process(node->task); | |
8541 | ||
8542 | mutex_unlock(&ctx->uring_lock); | |
8543 | wait_for_completion(&exit.completion); | |
8544 | cond_resched(); | |
8545 | mutex_lock(&ctx->uring_lock); | |
8546 | } | |
8547 | mutex_unlock(&ctx->uring_lock); | |
8548 | ||
85faa7b8 JA |
8549 | io_ring_ctx_free(ctx); |
8550 | } | |
8551 | ||
2b188cc1 JA |
8552 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8553 | { | |
61cf9370 MWO |
8554 | unsigned long index; |
8555 | struct creds *creds; | |
8556 | ||
2b188cc1 JA |
8557 | mutex_lock(&ctx->uring_lock); |
8558 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8559 | /* if force is set, the ring is going away. always drop after that */ |
8560 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8561 | if (ctx->rings) |
6c503150 | 8562 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8563 | xa_for_each(&ctx->personalities, index, creds) |
8564 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8565 | mutex_unlock(&ctx->uring_lock); |
8566 | ||
f6d54255 PB |
8567 | /* prevent SQPOLL from submitting new requests */ |
8568 | if (ctx->sq_data) { | |
8569 | io_sq_thread_park(ctx->sq_data); | |
8570 | list_del_init(&ctx->sqd_list); | |
8571 | io_sqd_update_thread_idle(ctx->sq_data); | |
8572 | io_sq_thread_unpark(ctx->sq_data); | |
8573 | } | |
8574 | ||
6b81928d PB |
8575 | io_kill_timeouts(ctx, NULL, NULL); |
8576 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8577 | |
15dff286 | 8578 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8579 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8580 | |
85faa7b8 | 8581 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8582 | /* |
8583 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8584 | * if we're exiting a ton of rings at the same time. It just adds | |
8585 | * noise and overhead, there's no discernable change in runtime | |
8586 | * over using system_wq. | |
8587 | */ | |
8588 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8589 | } |
8590 | ||
8591 | static int io_uring_release(struct inode *inode, struct file *file) | |
8592 | { | |
8593 | struct io_ring_ctx *ctx = file->private_data; | |
8594 | ||
8595 | file->private_data = NULL; | |
8596 | io_ring_ctx_wait_and_kill(ctx); | |
8597 | return 0; | |
8598 | } | |
8599 | ||
f6edbabb PB |
8600 | struct io_task_cancel { |
8601 | struct task_struct *task; | |
8602 | struct files_struct *files; | |
8603 | }; | |
f254ac04 | 8604 | |
f6edbabb | 8605 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8606 | { |
9a472ef7 | 8607 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8608 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8609 | bool ret; |
8610 | ||
f6edbabb | 8611 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8612 | unsigned long flags; |
8613 | struct io_ring_ctx *ctx = req->ctx; | |
8614 | ||
8615 | /* protect against races with linked timeouts */ | |
8616 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8617 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8618 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8619 | } else { | |
f6edbabb | 8620 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8621 | } |
8622 | return ret; | |
b711d4ea JA |
8623 | } |
8624 | ||
e1915f76 | 8625 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8626 | struct task_struct *task, |
b7ddce3c PB |
8627 | struct files_struct *files) |
8628 | { | |
e1915f76 | 8629 | struct io_defer_entry *de; |
b7ddce3c PB |
8630 | LIST_HEAD(list); |
8631 | ||
8632 | spin_lock_irq(&ctx->completion_lock); | |
8633 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8634 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8635 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8636 | break; | |
8637 | } | |
8638 | } | |
8639 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8640 | if (list_empty(&list)) |
8641 | return false; | |
b7ddce3c PB |
8642 | |
8643 | while (!list_empty(&list)) { | |
8644 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8645 | list_del_init(&de->list); | |
8646 | req_set_fail_links(de->req); | |
8647 | io_put_req(de->req); | |
8648 | io_req_complete(de->req, -ECANCELED); | |
8649 | kfree(de); | |
8650 | } | |
e1915f76 | 8651 | return true; |
b7ddce3c PB |
8652 | } |
8653 | ||
1b00764f PB |
8654 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8655 | { | |
8656 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8657 | ||
8658 | return req->ctx == data; | |
8659 | } | |
8660 | ||
8661 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8662 | { | |
8663 | struct io_tctx_node *node; | |
8664 | enum io_wq_cancel cret; | |
8665 | bool ret = false; | |
8666 | ||
8667 | mutex_lock(&ctx->uring_lock); | |
8668 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8669 | struct io_uring_task *tctx = node->task->io_uring; | |
8670 | ||
8671 | /* | |
8672 | * io_wq will stay alive while we hold uring_lock, because it's | |
8673 | * killed after ctx nodes, which requires to take the lock. | |
8674 | */ | |
8675 | if (!tctx || !tctx->io_wq) | |
8676 | continue; | |
8677 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8678 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8679 | } | |
8680 | mutex_unlock(&ctx->uring_lock); | |
8681 | ||
8682 | return ret; | |
8683 | } | |
8684 | ||
9936c7c2 PB |
8685 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8686 | struct task_struct *task, | |
8687 | struct files_struct *files) | |
8688 | { | |
8689 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8690 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8691 | |
8692 | while (1) { | |
8693 | enum io_wq_cancel cret; | |
8694 | bool ret = false; | |
8695 | ||
1b00764f PB |
8696 | if (!task) { |
8697 | ret |= io_uring_try_cancel_iowq(ctx); | |
8698 | } else if (tctx && tctx->io_wq) { | |
8699 | /* | |
8700 | * Cancels requests of all rings, not only @ctx, but | |
8701 | * it's fine as the task is in exit/exec. | |
8702 | */ | |
5aa75ed5 | 8703 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8704 | &cancel, true); |
8705 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8706 | } | |
8707 | ||
8708 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8709 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8710 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8711 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8712 | io_iopoll_try_reap_events(ctx); | |
8713 | ret = true; | |
8714 | } | |
8715 | } | |
8716 | ||
e1915f76 | 8717 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8718 | ret |= io_poll_remove_all(ctx, task, files); |
8719 | ret |= io_kill_timeouts(ctx, task, files); | |
8720 | ret |= io_run_task_work(); | |
ba50a036 | 8721 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8722 | io_cqring_overflow_flush(ctx, true, task, files); |
8723 | if (!ret) | |
8724 | break; | |
8725 | cond_resched(); | |
8726 | } | |
8727 | } | |
8728 | ||
ca70f00b PB |
8729 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8730 | struct task_struct *task, | |
8731 | struct files_struct *files) | |
8732 | { | |
8733 | struct io_kiocb *req; | |
8734 | int cnt = 0; | |
8735 | ||
8736 | spin_lock_irq(&ctx->inflight_lock); | |
8737 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8738 | cnt += io_match_task(req, task, files); | |
8739 | spin_unlock_irq(&ctx->inflight_lock); | |
8740 | return cnt; | |
8741 | } | |
8742 | ||
b52fda00 | 8743 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8744 | struct task_struct *task, |
fcb323cc JA |
8745 | struct files_struct *files) |
8746 | { | |
fcb323cc | 8747 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8748 | DEFINE_WAIT(wait); |
ca70f00b | 8749 | int inflight; |
fcb323cc | 8750 | |
ca70f00b PB |
8751 | inflight = io_uring_count_inflight(ctx, task, files); |
8752 | if (!inflight) | |
fcb323cc | 8753 | break; |
f6edbabb | 8754 | |
9936c7c2 | 8755 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8756 | |
8757 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8758 | TASK_UNINTERRUPTIBLE); | |
8759 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8760 | schedule(); | |
c98de08c | 8761 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8762 | } |
0f212204 JA |
8763 | } |
8764 | ||
0f212204 JA |
8765 | /* |
8766 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8767 | */ | |
baf186c4 | 8768 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8769 | { |
236434c3 | 8770 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8771 | struct io_tctx_node *node; |
a528b04e | 8772 | int ret; |
236434c3 MWO |
8773 | |
8774 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8775 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8776 | if (unlikely(ret)) |
8777 | return ret; | |
236434c3 | 8778 | tctx = current->io_uring; |
0f212204 | 8779 | } |
baf186c4 PB |
8780 | if (tctx->last != ctx) { |
8781 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8782 | |
236434c3 | 8783 | if (!old) { |
13bf43f5 PB |
8784 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8785 | if (!node) | |
8786 | return -ENOMEM; | |
8787 | node->ctx = ctx; | |
13bf43f5 PB |
8788 | node->task = current; |
8789 | ||
baf186c4 | 8790 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8791 | node, GFP_KERNEL)); |
a528b04e | 8792 | if (ret) { |
13bf43f5 | 8793 | kfree(node); |
a528b04e PB |
8794 | return ret; |
8795 | } | |
13bf43f5 PB |
8796 | |
8797 | mutex_lock(&ctx->uring_lock); | |
8798 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8799 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8800 | } |
baf186c4 | 8801 | tctx->last = ctx; |
0f212204 | 8802 | } |
0f212204 JA |
8803 | return 0; |
8804 | } | |
8805 | ||
8806 | /* | |
8807 | * Remove this io_uring_file -> task mapping. | |
8808 | */ | |
2941267b | 8809 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8810 | { |
8811 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8812 | struct io_tctx_node *node; |
2941267b | 8813 | |
eebd2e37 PB |
8814 | if (!tctx) |
8815 | return; | |
13bf43f5 PB |
8816 | node = xa_erase(&tctx->xa, index); |
8817 | if (!node) | |
2941267b | 8818 | return; |
0f212204 | 8819 | |
13bf43f5 PB |
8820 | WARN_ON_ONCE(current != node->task); |
8821 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8822 | ||
8823 | mutex_lock(&node->ctx->uring_lock); | |
8824 | list_del(&node->ctx_node); | |
8825 | mutex_unlock(&node->ctx->uring_lock); | |
8826 | ||
baf186c4 | 8827 | if (tctx->last == node->ctx) |
0f212204 | 8828 | tctx->last = NULL; |
13bf43f5 | 8829 | kfree(node); |
0f212204 JA |
8830 | } |
8831 | ||
8452d4a6 | 8832 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8833 | { |
13bf43f5 | 8834 | struct io_tctx_node *node; |
de7f1d9e PB |
8835 | unsigned long index; |
8836 | ||
13bf43f5 | 8837 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8838 | io_uring_del_task_file(index); |
8452d4a6 PB |
8839 | if (tctx->io_wq) { |
8840 | io_wq_put_and_exit(tctx->io_wq); | |
8841 | tctx->io_wq = NULL; | |
8842 | } | |
de7f1d9e PB |
8843 | } |
8844 | ||
521d6a73 PB |
8845 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8846 | { | |
8847 | return percpu_counter_sum(&tctx->inflight); | |
8848 | } | |
8849 | ||
8850 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8851 | { | |
8852 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8853 | struct io_ring_ctx *ctx = work->ctx; | |
8854 | struct io_sq_data *sqd = ctx->sq_data; | |
8855 | ||
8856 | if (sqd->thread) | |
8857 | io_uring_cancel_sqpoll(ctx); | |
8858 | complete(&work->completion); | |
8859 | } | |
8860 | ||
8861 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8862 | { | |
8863 | struct io_sq_data *sqd = ctx->sq_data; | |
8864 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8865 | struct task_struct *task; | |
8866 | ||
8867 | io_sq_thread_park(sqd); | |
8868 | list_del_init(&ctx->sqd_list); | |
8869 | io_sqd_update_thread_idle(sqd); | |
8870 | task = sqd->thread; | |
8871 | if (task) { | |
8872 | init_completion(&work.completion); | |
8873 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
8874 | WARN_ON_ONCE(task_work_add(task, &work.task_work, TWA_SIGNAL)); | |
8875 | wake_up_process(task); | |
8876 | } | |
8877 | io_sq_thread_unpark(sqd); | |
8878 | ||
8879 | if (task) | |
8880 | wait_for_completion(&work.completion); | |
8881 | } | |
8882 | ||
0f212204 JA |
8883 | void __io_uring_files_cancel(struct files_struct *files) |
8884 | { | |
8885 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8886 | struct io_tctx_node *node; |
ce765372 | 8887 | unsigned long index; |
0f212204 JA |
8888 | |
8889 | /* make sure overflow events are dropped */ | |
fdaf083c | 8890 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8891 | xa_for_each(&tctx->xa, index, node) { |
8892 | struct io_ring_ctx *ctx = node->ctx; | |
8893 | ||
8894 | if (ctx->sq_data) { | |
8895 | io_sqpoll_cancel_sync(ctx); | |
8896 | continue; | |
8897 | } | |
8898 | io_uring_cancel_files(ctx, current, files); | |
8899 | if (!files) | |
8900 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8901 | } | |
fdaf083c | 8902 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8903 | |
8452d4a6 PB |
8904 | if (files) |
8905 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8906 | } |
8907 | ||
521d6a73 | 8908 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8909 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8910 | { | |
37d1e2e3 | 8911 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8912 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8913 | s64 inflight; |
8914 | DEFINE_WAIT(wait); | |
fdaf083c | 8915 | |
521d6a73 PB |
8916 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8917 | ||
0e9ddb39 PB |
8918 | atomic_inc(&tctx->in_idle); |
8919 | do { | |
8920 | /* read completions before cancelations */ | |
8921 | inflight = tctx_inflight(tctx); | |
8922 | if (!inflight) | |
8923 | break; | |
521d6a73 | 8924 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8925 | |
0e9ddb39 PB |
8926 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8927 | /* | |
8928 | * If we've seen completions, retry without waiting. This | |
8929 | * avoids a race where a completion comes in before we did | |
8930 | * prepare_to_wait(). | |
8931 | */ | |
8932 | if (inflight == tctx_inflight(tctx)) | |
8933 | schedule(); | |
8934 | finish_wait(&tctx->wait, &wait); | |
8935 | } while (1); | |
8936 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8937 | } |
8938 | ||
0f212204 JA |
8939 | /* |
8940 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8941 | * requests. | |
8942 | */ | |
8943 | void __io_uring_task_cancel(void) | |
8944 | { | |
8945 | struct io_uring_task *tctx = current->io_uring; | |
8946 | DEFINE_WAIT(wait); | |
d8a6df10 | 8947 | s64 inflight; |
0f212204 JA |
8948 | |
8949 | /* make sure overflow events are dropped */ | |
fdaf083c | 8950 | atomic_inc(&tctx->in_idle); |
d8a6df10 | 8951 | do { |
0f212204 | 8952 | /* read completions before cancelations */ |
fdaf083c | 8953 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8954 | if (!inflight) |
8955 | break; | |
0f212204 JA |
8956 | __io_uring_files_cancel(NULL); |
8957 | ||
8958 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8959 | ||
8960 | /* | |
a1bb3cd5 PB |
8961 | * If we've seen completions, retry without waiting. This |
8962 | * avoids a race where a completion comes in before we did | |
8963 | * prepare_to_wait(). | |
0f212204 | 8964 | */ |
a1bb3cd5 PB |
8965 | if (inflight == tctx_inflight(tctx)) |
8966 | schedule(); | |
f57555ed | 8967 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8968 | } while (1); |
0f212204 | 8969 | |
fdaf083c | 8970 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8971 | |
8452d4a6 PB |
8972 | io_uring_clean_tctx(tctx); |
8973 | /* all current's requests should be gone, we can kill tctx */ | |
8974 | __io_uring_free(current); | |
44e728b8 PB |
8975 | } |
8976 | ||
6c5c240e RP |
8977 | static void *io_uring_validate_mmap_request(struct file *file, |
8978 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8979 | { |
2b188cc1 | 8980 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8981 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8982 | struct page *page; |
8983 | void *ptr; | |
8984 | ||
8985 | switch (offset) { | |
8986 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8987 | case IORING_OFF_CQ_RING: |
8988 | ptr = ctx->rings; | |
2b188cc1 JA |
8989 | break; |
8990 | case IORING_OFF_SQES: | |
8991 | ptr = ctx->sq_sqes; | |
8992 | break; | |
2b188cc1 | 8993 | default: |
6c5c240e | 8994 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8995 | } |
8996 | ||
8997 | page = virt_to_head_page(ptr); | |
a50b854e | 8998 | if (sz > page_size(page)) |
6c5c240e RP |
8999 | return ERR_PTR(-EINVAL); |
9000 | ||
9001 | return ptr; | |
9002 | } | |
9003 | ||
9004 | #ifdef CONFIG_MMU | |
9005 | ||
9006 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9007 | { | |
9008 | size_t sz = vma->vm_end - vma->vm_start; | |
9009 | unsigned long pfn; | |
9010 | void *ptr; | |
9011 | ||
9012 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9013 | if (IS_ERR(ptr)) | |
9014 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9015 | |
9016 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9017 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9018 | } | |
9019 | ||
6c5c240e RP |
9020 | #else /* !CONFIG_MMU */ |
9021 | ||
9022 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9023 | { | |
9024 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9025 | } | |
9026 | ||
9027 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9028 | { | |
9029 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9030 | } | |
9031 | ||
9032 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9033 | unsigned long addr, unsigned long len, | |
9034 | unsigned long pgoff, unsigned long flags) | |
9035 | { | |
9036 | void *ptr; | |
9037 | ||
9038 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9039 | if (IS_ERR(ptr)) | |
9040 | return PTR_ERR(ptr); | |
9041 | ||
9042 | return (unsigned long) ptr; | |
9043 | } | |
9044 | ||
9045 | #endif /* !CONFIG_MMU */ | |
9046 | ||
d9d05217 | 9047 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9048 | { |
9049 | DEFINE_WAIT(wait); | |
9050 | ||
9051 | do { | |
9052 | if (!io_sqring_full(ctx)) | |
9053 | break; | |
90554200 JA |
9054 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9055 | ||
9056 | if (!io_sqring_full(ctx)) | |
9057 | break; | |
90554200 JA |
9058 | schedule(); |
9059 | } while (!signal_pending(current)); | |
9060 | ||
9061 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9062 | return 0; |
90554200 JA |
9063 | } |
9064 | ||
c73ebb68 HX |
9065 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9066 | struct __kernel_timespec __user **ts, | |
9067 | const sigset_t __user **sig) | |
9068 | { | |
9069 | struct io_uring_getevents_arg arg; | |
9070 | ||
9071 | /* | |
9072 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9073 | * is just a pointer to the sigset_t. | |
9074 | */ | |
9075 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9076 | *sig = (const sigset_t __user *) argp; | |
9077 | *ts = NULL; | |
9078 | return 0; | |
9079 | } | |
9080 | ||
9081 | /* | |
9082 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9083 | * timespec and sigset_t pointers if good. | |
9084 | */ | |
9085 | if (*argsz != sizeof(arg)) | |
9086 | return -EINVAL; | |
9087 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9088 | return -EFAULT; | |
9089 | *sig = u64_to_user_ptr(arg.sigmask); | |
9090 | *argsz = arg.sigmask_sz; | |
9091 | *ts = u64_to_user_ptr(arg.ts); | |
9092 | return 0; | |
9093 | } | |
9094 | ||
2b188cc1 | 9095 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9096 | u32, min_complete, u32, flags, const void __user *, argp, |
9097 | size_t, argsz) | |
2b188cc1 JA |
9098 | { |
9099 | struct io_ring_ctx *ctx; | |
9100 | long ret = -EBADF; | |
9101 | int submitted = 0; | |
9102 | struct fd f; | |
9103 | ||
4c6e277c | 9104 | io_run_task_work(); |
b41e9852 | 9105 | |
90554200 | 9106 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9107 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9108 | return -EINVAL; |
9109 | ||
9110 | f = fdget(fd); | |
9111 | if (!f.file) | |
9112 | return -EBADF; | |
9113 | ||
9114 | ret = -EOPNOTSUPP; | |
9115 | if (f.file->f_op != &io_uring_fops) | |
9116 | goto out_fput; | |
9117 | ||
9118 | ret = -ENXIO; | |
9119 | ctx = f.file->private_data; | |
9120 | if (!percpu_ref_tryget(&ctx->refs)) | |
9121 | goto out_fput; | |
9122 | ||
7e84e1c7 SG |
9123 | ret = -EBADFD; |
9124 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9125 | goto out; | |
9126 | ||
6c271ce2 JA |
9127 | /* |
9128 | * For SQ polling, the thread will do all submissions and completions. | |
9129 | * Just return the requested submit count, and wake the thread if | |
9130 | * we were asked to. | |
9131 | */ | |
b2a9eada | 9132 | ret = 0; |
6c271ce2 | 9133 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9134 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9135 | |
d9d05217 | 9136 | ret = -EOWNERDEAD; |
04147488 SM |
9137 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9138 | goto out; | |
9139 | } | |
6c271ce2 | 9140 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9141 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9142 | if (flags & IORING_ENTER_SQ_WAIT) { |
9143 | ret = io_sqpoll_wait_sq(ctx); | |
9144 | if (ret) | |
9145 | goto out; | |
9146 | } | |
6c271ce2 | 9147 | submitted = to_submit; |
b2a9eada | 9148 | } else if (to_submit) { |
baf186c4 | 9149 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9150 | if (unlikely(ret)) |
9151 | goto out; | |
2b188cc1 | 9152 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9153 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9154 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9155 | |
9156 | if (submitted != to_submit) | |
9157 | goto out; | |
2b188cc1 JA |
9158 | } |
9159 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9160 | const sigset_t __user *sig; |
9161 | struct __kernel_timespec __user *ts; | |
9162 | ||
9163 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9164 | if (unlikely(ret)) | |
9165 | goto out; | |
9166 | ||
2b188cc1 JA |
9167 | min_complete = min(min_complete, ctx->cq_entries); |
9168 | ||
32b2244a XW |
9169 | /* |
9170 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9171 | * space applications don't need to do io completion events | |
9172 | * polling again, they can rely on io_sq_thread to do polling | |
9173 | * work, which can reduce cpu usage and uring_lock contention. | |
9174 | */ | |
9175 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9176 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9177 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9178 | } else { |
c73ebb68 | 9179 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9180 | } |
2b188cc1 JA |
9181 | } |
9182 | ||
7c504e65 | 9183 | out: |
6805b32e | 9184 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9185 | out_fput: |
9186 | fdput(f); | |
9187 | return submitted ? submitted : ret; | |
9188 | } | |
9189 | ||
bebdb65e | 9190 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9191 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9192 | const struct cred *cred) | |
87ce955b | 9193 | { |
87ce955b JA |
9194 | struct user_namespace *uns = seq_user_ns(m); |
9195 | struct group_info *gi; | |
9196 | kernel_cap_t cap; | |
9197 | unsigned __capi; | |
9198 | int g; | |
9199 | ||
9200 | seq_printf(m, "%5d\n", id); | |
9201 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9202 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9203 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9204 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9205 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9206 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9207 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9208 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9209 | seq_puts(m, "\n\tGroups:\t"); | |
9210 | gi = cred->group_info; | |
9211 | for (g = 0; g < gi->ngroups; g++) { | |
9212 | seq_put_decimal_ull(m, g ? " " : "", | |
9213 | from_kgid_munged(uns, gi->gid[g])); | |
9214 | } | |
9215 | seq_puts(m, "\n\tCapEff:\t"); | |
9216 | cap = cred->cap_effective; | |
9217 | CAP_FOR_EACH_U32(__capi) | |
9218 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9219 | seq_putc(m, '\n'); | |
9220 | return 0; | |
9221 | } | |
9222 | ||
9223 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9224 | { | |
dbbe9c64 | 9225 | struct io_sq_data *sq = NULL; |
fad8e0de | 9226 | bool has_lock; |
87ce955b JA |
9227 | int i; |
9228 | ||
fad8e0de JA |
9229 | /* |
9230 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9231 | * since fdinfo case grabs it in the opposite direction of normal use | |
9232 | * cases. If we fail to get the lock, we just don't iterate any | |
9233 | * structures that could be going away outside the io_uring mutex. | |
9234 | */ | |
9235 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9236 | ||
5f3f26f9 | 9237 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9238 | sq = ctx->sq_data; |
5f3f26f9 JA |
9239 | if (!sq->thread) |
9240 | sq = NULL; | |
9241 | } | |
dbbe9c64 JQ |
9242 | |
9243 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9244 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9245 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9246 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9247 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9248 | |
87ce955b JA |
9249 | if (f) |
9250 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9251 | else | |
9252 | seq_printf(m, "%5u: <none>\n", i); | |
9253 | } | |
9254 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9255 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9256 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9257 | ||
9258 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9259 | (unsigned int) buf->len); | |
9260 | } | |
61cf9370 MWO |
9261 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9262 | unsigned long index; | |
9263 | const struct cred *cred; | |
9264 | ||
87ce955b | 9265 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9266 | xa_for_each(&ctx->personalities, index, cred) |
9267 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9268 | } |
d7718a9d JA |
9269 | seq_printf(m, "PollList:\n"); |
9270 | spin_lock_irq(&ctx->completion_lock); | |
9271 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9272 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9273 | struct io_kiocb *req; | |
9274 | ||
9275 | hlist_for_each_entry(req, list, hash_node) | |
9276 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9277 | req->task->task_works != NULL); | |
9278 | } | |
9279 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9280 | if (has_lock) |
9281 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9282 | } |
9283 | ||
9284 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9285 | { | |
9286 | struct io_ring_ctx *ctx = f->private_data; | |
9287 | ||
9288 | if (percpu_ref_tryget(&ctx->refs)) { | |
9289 | __io_uring_show_fdinfo(ctx, m); | |
9290 | percpu_ref_put(&ctx->refs); | |
9291 | } | |
9292 | } | |
bebdb65e | 9293 | #endif |
87ce955b | 9294 | |
2b188cc1 JA |
9295 | static const struct file_operations io_uring_fops = { |
9296 | .release = io_uring_release, | |
9297 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9298 | #ifndef CONFIG_MMU |
9299 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9300 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9301 | #endif | |
2b188cc1 JA |
9302 | .poll = io_uring_poll, |
9303 | .fasync = io_uring_fasync, | |
bebdb65e | 9304 | #ifdef CONFIG_PROC_FS |
87ce955b | 9305 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9306 | #endif |
2b188cc1 JA |
9307 | }; |
9308 | ||
9309 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9310 | struct io_uring_params *p) | |
9311 | { | |
75b28aff HV |
9312 | struct io_rings *rings; |
9313 | size_t size, sq_array_offset; | |
2b188cc1 | 9314 | |
bd740481 JA |
9315 | /* make sure these are sane, as we already accounted them */ |
9316 | ctx->sq_entries = p->sq_entries; | |
9317 | ctx->cq_entries = p->cq_entries; | |
9318 | ||
75b28aff HV |
9319 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9320 | if (size == SIZE_MAX) | |
9321 | return -EOVERFLOW; | |
9322 | ||
9323 | rings = io_mem_alloc(size); | |
9324 | if (!rings) | |
2b188cc1 JA |
9325 | return -ENOMEM; |
9326 | ||
75b28aff HV |
9327 | ctx->rings = rings; |
9328 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9329 | rings->sq_ring_mask = p->sq_entries - 1; | |
9330 | rings->cq_ring_mask = p->cq_entries - 1; | |
9331 | rings->sq_ring_entries = p->sq_entries; | |
9332 | rings->cq_ring_entries = p->cq_entries; | |
9333 | ctx->sq_mask = rings->sq_ring_mask; | |
9334 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9335 | |
9336 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9337 | if (size == SIZE_MAX) { |
9338 | io_mem_free(ctx->rings); | |
9339 | ctx->rings = NULL; | |
2b188cc1 | 9340 | return -EOVERFLOW; |
eb065d30 | 9341 | } |
2b188cc1 JA |
9342 | |
9343 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9344 | if (!ctx->sq_sqes) { |
9345 | io_mem_free(ctx->rings); | |
9346 | ctx->rings = NULL; | |
2b188cc1 | 9347 | return -ENOMEM; |
eb065d30 | 9348 | } |
2b188cc1 | 9349 | |
2b188cc1 JA |
9350 | return 0; |
9351 | } | |
9352 | ||
9faadcc8 PB |
9353 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9354 | { | |
9355 | int ret, fd; | |
9356 | ||
9357 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9358 | if (fd < 0) | |
9359 | return fd; | |
9360 | ||
baf186c4 | 9361 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9362 | if (ret) { |
9363 | put_unused_fd(fd); | |
9364 | return ret; | |
9365 | } | |
9366 | fd_install(fd, file); | |
9367 | return fd; | |
9368 | } | |
9369 | ||
2b188cc1 JA |
9370 | /* |
9371 | * Allocate an anonymous fd, this is what constitutes the application | |
9372 | * visible backing of an io_uring instance. The application mmaps this | |
9373 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9374 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9375 | */ | |
9faadcc8 | 9376 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9377 | { |
9378 | struct file *file; | |
9faadcc8 | 9379 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9380 | int ret; |
9381 | ||
2b188cc1 JA |
9382 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9383 | &ctx->ring_sock); | |
9384 | if (ret) | |
9faadcc8 | 9385 | return ERR_PTR(ret); |
2b188cc1 JA |
9386 | #endif |
9387 | ||
2b188cc1 JA |
9388 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9389 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9390 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9391 | if (IS_ERR(file)) { |
9392 | sock_release(ctx->ring_sock); | |
9393 | ctx->ring_sock = NULL; | |
9394 | } else { | |
9395 | ctx->ring_sock->file = file; | |
0f212204 | 9396 | } |
2b188cc1 | 9397 | #endif |
9faadcc8 | 9398 | return file; |
2b188cc1 JA |
9399 | } |
9400 | ||
7f13657d XW |
9401 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9402 | struct io_uring_params __user *params) | |
2b188cc1 | 9403 | { |
2b188cc1 | 9404 | struct io_ring_ctx *ctx; |
9faadcc8 | 9405 | struct file *file; |
2b188cc1 JA |
9406 | int ret; |
9407 | ||
8110c1a6 | 9408 | if (!entries) |
2b188cc1 | 9409 | return -EINVAL; |
8110c1a6 JA |
9410 | if (entries > IORING_MAX_ENTRIES) { |
9411 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9412 | return -EINVAL; | |
9413 | entries = IORING_MAX_ENTRIES; | |
9414 | } | |
2b188cc1 JA |
9415 | |
9416 | /* | |
9417 | * Use twice as many entries for the CQ ring. It's possible for the | |
9418 | * application to drive a higher depth than the size of the SQ ring, | |
9419 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9420 | * some flexibility in overcommitting a bit. If the application has |
9421 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9422 | * of CQ ring entries manually. | |
2b188cc1 JA |
9423 | */ |
9424 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9425 | if (p->flags & IORING_SETUP_CQSIZE) { |
9426 | /* | |
9427 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9428 | * to a power-of-two, if it isn't already. We do NOT impose | |
9429 | * any cq vs sq ring sizing. | |
9430 | */ | |
eb2667b3 | 9431 | if (!p->cq_entries) |
33a107f0 | 9432 | return -EINVAL; |
8110c1a6 JA |
9433 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9434 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9435 | return -EINVAL; | |
9436 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9437 | } | |
eb2667b3 JQ |
9438 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9439 | if (p->cq_entries < p->sq_entries) | |
9440 | return -EINVAL; | |
33a107f0 JA |
9441 | } else { |
9442 | p->cq_entries = 2 * p->sq_entries; | |
9443 | } | |
2b188cc1 | 9444 | |
2b188cc1 | 9445 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9446 | if (!ctx) |
2b188cc1 | 9447 | return -ENOMEM; |
2b188cc1 | 9448 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9449 | if (!capable(CAP_IPC_LOCK)) |
9450 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9451 | |
9452 | /* | |
9453 | * This is just grabbed for accounting purposes. When a process exits, | |
9454 | * the mm is exited and dropped before the files, hence we need to hang | |
9455 | * on to this mm purely for the purposes of being able to unaccount | |
9456 | * memory (locked/pinned vm). It's not used for anything else. | |
9457 | */ | |
6b7898eb | 9458 | mmgrab(current->mm); |
2aede0e4 | 9459 | ctx->mm_account = current->mm; |
6b7898eb | 9460 | |
2b188cc1 JA |
9461 | ret = io_allocate_scq_urings(ctx, p); |
9462 | if (ret) | |
9463 | goto err; | |
9464 | ||
7e84e1c7 | 9465 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9466 | if (ret) |
9467 | goto err; | |
9468 | ||
2b188cc1 | 9469 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9470 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9471 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9472 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9473 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9474 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9475 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9476 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9477 | |
9478 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9479 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9480 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9481 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9482 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9483 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9484 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9485 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9486 | |
7f13657d XW |
9487 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9488 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9489 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9490 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9491 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9492 | |
9493 | if (copy_to_user(params, p, sizeof(*p))) { | |
9494 | ret = -EFAULT; | |
9495 | goto err; | |
9496 | } | |
d1719f70 | 9497 | |
9faadcc8 PB |
9498 | file = io_uring_get_file(ctx); |
9499 | if (IS_ERR(file)) { | |
9500 | ret = PTR_ERR(file); | |
9501 | goto err; | |
9502 | } | |
9503 | ||
044c1ab3 JA |
9504 | /* |
9505 | * Install ring fd as the very last thing, so we don't risk someone | |
9506 | * having closed it before we finish setup | |
9507 | */ | |
9faadcc8 PB |
9508 | ret = io_uring_install_fd(ctx, file); |
9509 | if (ret < 0) { | |
9510 | /* fput will clean it up */ | |
9511 | fput(file); | |
9512 | return ret; | |
9513 | } | |
044c1ab3 | 9514 | |
c826bd7a | 9515 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9516 | return ret; |
9517 | err: | |
9518 | io_ring_ctx_wait_and_kill(ctx); | |
9519 | return ret; | |
9520 | } | |
9521 | ||
9522 | /* | |
9523 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9524 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9525 | * params structure passed in. | |
9526 | */ | |
9527 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9528 | { | |
9529 | struct io_uring_params p; | |
2b188cc1 JA |
9530 | int i; |
9531 | ||
9532 | if (copy_from_user(&p, params, sizeof(p))) | |
9533 | return -EFAULT; | |
9534 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9535 | if (p.resv[i]) | |
9536 | return -EINVAL; | |
9537 | } | |
9538 | ||
6c271ce2 | 9539 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9540 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9541 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9542 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9543 | return -EINVAL; |
9544 | ||
7f13657d | 9545 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9546 | } |
9547 | ||
9548 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9549 | struct io_uring_params __user *, params) | |
9550 | { | |
9551 | return io_uring_setup(entries, params); | |
9552 | } | |
9553 | ||
66f4af93 JA |
9554 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9555 | { | |
9556 | struct io_uring_probe *p; | |
9557 | size_t size; | |
9558 | int i, ret; | |
9559 | ||
9560 | size = struct_size(p, ops, nr_args); | |
9561 | if (size == SIZE_MAX) | |
9562 | return -EOVERFLOW; | |
9563 | p = kzalloc(size, GFP_KERNEL); | |
9564 | if (!p) | |
9565 | return -ENOMEM; | |
9566 | ||
9567 | ret = -EFAULT; | |
9568 | if (copy_from_user(p, arg, size)) | |
9569 | goto out; | |
9570 | ret = -EINVAL; | |
9571 | if (memchr_inv(p, 0, size)) | |
9572 | goto out; | |
9573 | ||
9574 | p->last_op = IORING_OP_LAST - 1; | |
9575 | if (nr_args > IORING_OP_LAST) | |
9576 | nr_args = IORING_OP_LAST; | |
9577 | ||
9578 | for (i = 0; i < nr_args; i++) { | |
9579 | p->ops[i].op = i; | |
9580 | if (!io_op_defs[i].not_supported) | |
9581 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9582 | } | |
9583 | p->ops_len = i; | |
9584 | ||
9585 | ret = 0; | |
9586 | if (copy_to_user(arg, p, size)) | |
9587 | ret = -EFAULT; | |
9588 | out: | |
9589 | kfree(p); | |
9590 | return ret; | |
9591 | } | |
9592 | ||
071698e1 JA |
9593 | static int io_register_personality(struct io_ring_ctx *ctx) |
9594 | { | |
4379bf8b | 9595 | const struct cred *creds; |
61cf9370 | 9596 | u32 id; |
1e6fa521 | 9597 | int ret; |
071698e1 | 9598 | |
4379bf8b | 9599 | creds = get_current_cred(); |
1e6fa521 | 9600 | |
61cf9370 MWO |
9601 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9602 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9603 | if (!ret) | |
9604 | return id; | |
9605 | put_cred(creds); | |
1e6fa521 | 9606 | return ret; |
071698e1 JA |
9607 | } |
9608 | ||
21b55dbc SG |
9609 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9610 | unsigned int nr_args) | |
9611 | { | |
9612 | struct io_uring_restriction *res; | |
9613 | size_t size; | |
9614 | int i, ret; | |
9615 | ||
7e84e1c7 SG |
9616 | /* Restrictions allowed only if rings started disabled */ |
9617 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9618 | return -EBADFD; | |
9619 | ||
21b55dbc | 9620 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9621 | if (ctx->restrictions.registered) |
21b55dbc SG |
9622 | return -EBUSY; |
9623 | ||
9624 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9625 | return -EINVAL; | |
9626 | ||
9627 | size = array_size(nr_args, sizeof(*res)); | |
9628 | if (size == SIZE_MAX) | |
9629 | return -EOVERFLOW; | |
9630 | ||
9631 | res = memdup_user(arg, size); | |
9632 | if (IS_ERR(res)) | |
9633 | return PTR_ERR(res); | |
9634 | ||
9635 | ret = 0; | |
9636 | ||
9637 | for (i = 0; i < nr_args; i++) { | |
9638 | switch (res[i].opcode) { | |
9639 | case IORING_RESTRICTION_REGISTER_OP: | |
9640 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9641 | ret = -EINVAL; | |
9642 | goto out; | |
9643 | } | |
9644 | ||
9645 | __set_bit(res[i].register_op, | |
9646 | ctx->restrictions.register_op); | |
9647 | break; | |
9648 | case IORING_RESTRICTION_SQE_OP: | |
9649 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9650 | ret = -EINVAL; | |
9651 | goto out; | |
9652 | } | |
9653 | ||
9654 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9655 | break; | |
9656 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9657 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9658 | break; | |
9659 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9660 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9661 | break; | |
9662 | default: | |
9663 | ret = -EINVAL; | |
9664 | goto out; | |
9665 | } | |
9666 | } | |
9667 | ||
9668 | out: | |
9669 | /* Reset all restrictions if an error happened */ | |
9670 | if (ret != 0) | |
9671 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9672 | else | |
7e84e1c7 | 9673 | ctx->restrictions.registered = true; |
21b55dbc SG |
9674 | |
9675 | kfree(res); | |
9676 | return ret; | |
9677 | } | |
9678 | ||
7e84e1c7 SG |
9679 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9680 | { | |
9681 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9682 | return -EBADFD; | |
9683 | ||
9684 | if (ctx->restrictions.registered) | |
9685 | ctx->restricted = 1; | |
9686 | ||
0298ef96 PB |
9687 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9688 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9689 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9690 | return 0; |
9691 | } | |
9692 | ||
071698e1 JA |
9693 | static bool io_register_op_must_quiesce(int op) |
9694 | { | |
9695 | switch (op) { | |
9696 | case IORING_UNREGISTER_FILES: | |
9697 | case IORING_REGISTER_FILES_UPDATE: | |
9698 | case IORING_REGISTER_PROBE: | |
9699 | case IORING_REGISTER_PERSONALITY: | |
9700 | case IORING_UNREGISTER_PERSONALITY: | |
9701 | return false; | |
9702 | default: | |
9703 | return true; | |
9704 | } | |
9705 | } | |
9706 | ||
edafccee JA |
9707 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9708 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9709 | __releases(ctx->uring_lock) |
9710 | __acquires(ctx->uring_lock) | |
edafccee JA |
9711 | { |
9712 | int ret; | |
9713 | ||
35fa71a0 JA |
9714 | /* |
9715 | * We're inside the ring mutex, if the ref is already dying, then | |
9716 | * someone else killed the ctx or is already going through | |
9717 | * io_uring_register(). | |
9718 | */ | |
9719 | if (percpu_ref_is_dying(&ctx->refs)) | |
9720 | return -ENXIO; | |
9721 | ||
071698e1 | 9722 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9723 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9724 | |
05f3fb3c JA |
9725 | /* |
9726 | * Drop uring mutex before waiting for references to exit. If | |
9727 | * another thread is currently inside io_uring_enter() it might | |
9728 | * need to grab the uring_lock to make progress. If we hold it | |
9729 | * here across the drain wait, then we can deadlock. It's safe | |
9730 | * to drop the mutex here, since no new references will come in | |
9731 | * after we've killed the percpu ref. | |
9732 | */ | |
9733 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9734 | do { |
9735 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9736 | if (!ret) | |
9737 | break; | |
ed6930c9 JA |
9738 | ret = io_run_task_work_sig(); |
9739 | if (ret < 0) | |
9740 | break; | |
af9c1a44 JA |
9741 | } while (1); |
9742 | ||
05f3fb3c | 9743 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9744 | |
c150368b JA |
9745 | if (ret) { |
9746 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9747 | goto out_quiesce; |
9748 | } | |
9749 | } | |
9750 | ||
9751 | if (ctx->restricted) { | |
9752 | if (opcode >= IORING_REGISTER_LAST) { | |
9753 | ret = -EINVAL; | |
9754 | goto out; | |
9755 | } | |
9756 | ||
9757 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9758 | ret = -EACCES; | |
c150368b JA |
9759 | goto out; |
9760 | } | |
05f3fb3c | 9761 | } |
edafccee JA |
9762 | |
9763 | switch (opcode) { | |
9764 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9765 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9766 | break; |
9767 | case IORING_UNREGISTER_BUFFERS: | |
9768 | ret = -EINVAL; | |
9769 | if (arg || nr_args) | |
9770 | break; | |
0a96bbe4 | 9771 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9772 | break; |
6b06314c JA |
9773 | case IORING_REGISTER_FILES: |
9774 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9775 | break; | |
9776 | case IORING_UNREGISTER_FILES: | |
9777 | ret = -EINVAL; | |
9778 | if (arg || nr_args) | |
9779 | break; | |
9780 | ret = io_sqe_files_unregister(ctx); | |
9781 | break; | |
c3a31e60 JA |
9782 | case IORING_REGISTER_FILES_UPDATE: |
9783 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9784 | break; | |
9b402849 | 9785 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9786 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9787 | ret = -EINVAL; |
9788 | if (nr_args != 1) | |
9789 | break; | |
9790 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9791 | if (ret) |
9792 | break; | |
9793 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9794 | ctx->eventfd_async = 1; | |
9795 | else | |
9796 | ctx->eventfd_async = 0; | |
9b402849 JA |
9797 | break; |
9798 | case IORING_UNREGISTER_EVENTFD: | |
9799 | ret = -EINVAL; | |
9800 | if (arg || nr_args) | |
9801 | break; | |
9802 | ret = io_eventfd_unregister(ctx); | |
9803 | break; | |
66f4af93 JA |
9804 | case IORING_REGISTER_PROBE: |
9805 | ret = -EINVAL; | |
9806 | if (!arg || nr_args > 256) | |
9807 | break; | |
9808 | ret = io_probe(ctx, arg, nr_args); | |
9809 | break; | |
071698e1 JA |
9810 | case IORING_REGISTER_PERSONALITY: |
9811 | ret = -EINVAL; | |
9812 | if (arg || nr_args) | |
9813 | break; | |
9814 | ret = io_register_personality(ctx); | |
9815 | break; | |
9816 | case IORING_UNREGISTER_PERSONALITY: | |
9817 | ret = -EINVAL; | |
9818 | if (arg) | |
9819 | break; | |
9820 | ret = io_unregister_personality(ctx, nr_args); | |
9821 | break; | |
7e84e1c7 SG |
9822 | case IORING_REGISTER_ENABLE_RINGS: |
9823 | ret = -EINVAL; | |
9824 | if (arg || nr_args) | |
9825 | break; | |
9826 | ret = io_register_enable_rings(ctx); | |
9827 | break; | |
21b55dbc SG |
9828 | case IORING_REGISTER_RESTRICTIONS: |
9829 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9830 | break; | |
edafccee JA |
9831 | default: |
9832 | ret = -EINVAL; | |
9833 | break; | |
9834 | } | |
9835 | ||
21b55dbc | 9836 | out: |
071698e1 | 9837 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9838 | /* bring the ctx back to life */ |
05f3fb3c | 9839 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9840 | out_quiesce: |
0f158b4c | 9841 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9842 | } |
edafccee JA |
9843 | return ret; |
9844 | } | |
9845 | ||
9846 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9847 | void __user *, arg, unsigned int, nr_args) | |
9848 | { | |
9849 | struct io_ring_ctx *ctx; | |
9850 | long ret = -EBADF; | |
9851 | struct fd f; | |
9852 | ||
9853 | f = fdget(fd); | |
9854 | if (!f.file) | |
9855 | return -EBADF; | |
9856 | ||
9857 | ret = -EOPNOTSUPP; | |
9858 | if (f.file->f_op != &io_uring_fops) | |
9859 | goto out_fput; | |
9860 | ||
9861 | ctx = f.file->private_data; | |
9862 | ||
b6c23dd5 PB |
9863 | io_run_task_work(); |
9864 | ||
edafccee JA |
9865 | mutex_lock(&ctx->uring_lock); |
9866 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9867 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9868 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9869 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9870 | out_fput: |
9871 | fdput(f); | |
9872 | return ret; | |
9873 | } | |
9874 | ||
2b188cc1 JA |
9875 | static int __init io_uring_init(void) |
9876 | { | |
d7f62e82 SM |
9877 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9878 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9879 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9880 | } while (0) | |
9881 | ||
9882 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9883 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9884 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9885 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9886 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9887 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9888 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9889 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9890 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9891 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9892 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9893 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9894 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9895 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9896 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9897 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9898 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9899 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9900 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9901 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9902 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9903 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9904 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9905 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9906 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9907 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9908 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9909 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9910 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9911 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9912 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9913 | |
d3656344 | 9914 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9915 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9916 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9917 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9918 | return 0; |
9919 | }; | |
9920 | __initcall(io_uring_init); |