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; | |
69fb2131 JA |
261 | struct mutex lock; |
262 | ||
263 | /* ctx's that are using this sqd */ | |
264 | struct list_head ctx_list; | |
265 | struct list_head ctx_new_list; | |
266 | struct mutex ctx_lock; | |
267 | ||
534ca6d6 JA |
268 | struct task_struct *thread; |
269 | struct wait_queue_head wait; | |
08369246 XW |
270 | |
271 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
272 | int sq_cpu; |
273 | pid_t task_pid; | |
274 | ||
275 | unsigned long state; | |
276 | struct completion startup; | |
277 | struct completion completion; | |
278 | struct completion exited; | |
534ca6d6 JA |
279 | }; |
280 | ||
258b29a9 | 281 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 282 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 283 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 284 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
285 | |
286 | struct io_comp_state { | |
6dd0be1e | 287 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 288 | unsigned int nr; |
c7dae4ba JA |
289 | unsigned int locked_free_nr; |
290 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 291 | struct list_head free_list; |
c7dae4ba JA |
292 | /* IRQ completion list, under ->completion_lock */ |
293 | struct list_head locked_free_list; | |
258b29a9 PB |
294 | }; |
295 | ||
a1ab7b35 PB |
296 | struct io_submit_link { |
297 | struct io_kiocb *head; | |
298 | struct io_kiocb *last; | |
299 | }; | |
300 | ||
258b29a9 PB |
301 | struct io_submit_state { |
302 | struct blk_plug plug; | |
a1ab7b35 | 303 | struct io_submit_link link; |
258b29a9 PB |
304 | |
305 | /* | |
306 | * io_kiocb alloc cache | |
307 | */ | |
bf019da7 | 308 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
309 | unsigned int free_reqs; |
310 | ||
311 | bool plug_started; | |
312 | ||
313 | /* | |
314 | * Batch completion logic | |
315 | */ | |
316 | struct io_comp_state comp; | |
317 | ||
318 | /* | |
319 | * File reference cache | |
320 | */ | |
321 | struct file *file; | |
322 | unsigned int fd; | |
323 | unsigned int file_refs; | |
324 | unsigned int ios_left; | |
325 | }; | |
326 | ||
2b188cc1 JA |
327 | struct io_ring_ctx { |
328 | struct { | |
329 | struct percpu_ref refs; | |
330 | } ____cacheline_aligned_in_smp; | |
331 | ||
332 | struct { | |
333 | unsigned int flags; | |
e1d85334 | 334 | unsigned int compat: 1; |
e1d85334 RD |
335 | unsigned int cq_overflow_flushed: 1; |
336 | unsigned int drain_next: 1; | |
337 | unsigned int eventfd_async: 1; | |
21b55dbc | 338 | unsigned int restricted: 1; |
5f3f26f9 | 339 | unsigned int sqo_exec: 1; |
2b188cc1 | 340 | |
75b28aff HV |
341 | /* |
342 | * Ring buffer of indices into array of io_uring_sqe, which is | |
343 | * mmapped by the application using the IORING_OFF_SQES offset. | |
344 | * | |
345 | * This indirection could e.g. be used to assign fixed | |
346 | * io_uring_sqe entries to operations and only submit them to | |
347 | * the queue when needed. | |
348 | * | |
349 | * The kernel modifies neither the indices array nor the entries | |
350 | * array. | |
351 | */ | |
352 | u32 *sq_array; | |
2b188cc1 JA |
353 | unsigned cached_sq_head; |
354 | unsigned sq_entries; | |
355 | unsigned sq_mask; | |
6c271ce2 | 356 | unsigned sq_thread_idle; |
498ccd9e | 357 | unsigned cached_sq_dropped; |
2c3bac6d | 358 | unsigned cached_cq_overflow; |
ad3eb2c8 | 359 | unsigned long sq_check_overflow; |
de0617e4 | 360 | |
e941894e JA |
361 | /* hashed buffered write serialization */ |
362 | struct io_wq_hash *hash_map; | |
363 | ||
de0617e4 | 364 | struct list_head defer_list; |
5262f567 | 365 | struct list_head timeout_list; |
1d7bb1d5 | 366 | struct list_head cq_overflow_list; |
fcb323cc | 367 | |
ad3eb2c8 | 368 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
369 | } ____cacheline_aligned_in_smp; |
370 | ||
3c1a2ead JA |
371 | struct { |
372 | struct mutex uring_lock; | |
373 | wait_queue_head_t wait; | |
374 | } ____cacheline_aligned_in_smp; | |
375 | ||
376 | struct io_submit_state submit_state; | |
377 | ||
206aefde JA |
378 | struct io_rings *rings; |
379 | ||
2aede0e4 JA |
380 | /* Only used for accounting purposes */ |
381 | struct mm_struct *mm_account; | |
382 | ||
534ca6d6 JA |
383 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
384 | ||
90554200 | 385 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 386 | struct list_head sqd_list; |
75b28aff | 387 | |
6b06314c JA |
388 | /* |
389 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
390 | * readers must ensure that ->refs is alive as long as the file* is | |
391 | * used. Only updated through io_uring_register(2). | |
392 | */ | |
269bbe5f | 393 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
394 | unsigned nr_user_files; |
395 | ||
edafccee JA |
396 | /* if used, fixed mapped user buffers */ |
397 | unsigned nr_user_bufs; | |
398 | struct io_mapped_ubuf *user_bufs; | |
399 | ||
2b188cc1 JA |
400 | struct user_struct *user; |
401 | ||
0f158b4c JA |
402 | struct completion ref_comp; |
403 | struct completion sq_thread_comp; | |
206aefde JA |
404 | |
405 | #if defined(CONFIG_UNIX) | |
406 | struct socket *ring_sock; | |
407 | #endif | |
408 | ||
5a2e745d JA |
409 | struct idr io_buffer_idr; |
410 | ||
071698e1 JA |
411 | struct idr personality_idr; |
412 | ||
206aefde JA |
413 | struct { |
414 | unsigned cached_cq_tail; | |
415 | unsigned cq_entries; | |
416 | unsigned cq_mask; | |
417 | atomic_t cq_timeouts; | |
f010505b | 418 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 419 | unsigned long cq_check_overflow; |
206aefde JA |
420 | struct wait_queue_head cq_wait; |
421 | struct fasync_struct *cq_fasync; | |
422 | struct eventfd_ctx *cq_ev_fd; | |
423 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 424 | |
2b188cc1 JA |
425 | struct { |
426 | spinlock_t completion_lock; | |
e94f141b | 427 | |
def596e9 | 428 | /* |
540e32a0 | 429 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
430 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
431 | * For SQPOLL, only the single threaded io_sq_thread() will | |
432 | * manipulate the list, hence no extra locking is needed there. | |
433 | */ | |
540e32a0 | 434 | struct list_head iopoll_list; |
78076bb6 JA |
435 | struct hlist_head *cancel_hash; |
436 | unsigned cancel_hash_bits; | |
e94f141b | 437 | bool poll_multi_file; |
31b51510 | 438 | |
fcb323cc JA |
439 | spinlock_t inflight_lock; |
440 | struct list_head inflight_list; | |
2b188cc1 | 441 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 442 | |
269bbe5f BM |
443 | struct delayed_work rsrc_put_work; |
444 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
445 | struct list_head rsrc_ref_list; |
446 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 447 | |
21b55dbc | 448 | struct io_restriction restrictions; |
3c1a2ead | 449 | |
7c25c0d1 JA |
450 | /* exit task_work */ |
451 | struct callback_head *exit_task_work; | |
452 | ||
e941894e JA |
453 | struct wait_queue_head hash_wait; |
454 | ||
3c1a2ead JA |
455 | /* Keep this last, we don't need it for the fast path */ |
456 | struct work_struct exit_work; | |
2b188cc1 JA |
457 | }; |
458 | ||
09bb8394 JA |
459 | /* |
460 | * First field must be the file pointer in all the | |
461 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
462 | */ | |
221c5eb2 JA |
463 | struct io_poll_iocb { |
464 | struct file *file; | |
018043be | 465 | struct wait_queue_head *head; |
221c5eb2 | 466 | __poll_t events; |
8c838788 | 467 | bool done; |
221c5eb2 | 468 | bool canceled; |
392edb45 | 469 | struct wait_queue_entry wait; |
221c5eb2 JA |
470 | }; |
471 | ||
018043be PB |
472 | struct io_poll_remove { |
473 | struct file *file; | |
474 | u64 addr; | |
475 | }; | |
476 | ||
b5dba59e JA |
477 | struct io_close { |
478 | struct file *file; | |
b5dba59e JA |
479 | int fd; |
480 | }; | |
481 | ||
ad8a48ac JA |
482 | struct io_timeout_data { |
483 | struct io_kiocb *req; | |
484 | struct hrtimer timer; | |
485 | struct timespec64 ts; | |
486 | enum hrtimer_mode mode; | |
487 | }; | |
488 | ||
8ed8d3c3 JA |
489 | struct io_accept { |
490 | struct file *file; | |
491 | struct sockaddr __user *addr; | |
492 | int __user *addr_len; | |
493 | int flags; | |
09952e3e | 494 | unsigned long nofile; |
8ed8d3c3 JA |
495 | }; |
496 | ||
497 | struct io_sync { | |
498 | struct file *file; | |
499 | loff_t len; | |
500 | loff_t off; | |
501 | int flags; | |
d63d1b5e | 502 | int mode; |
8ed8d3c3 JA |
503 | }; |
504 | ||
fbf23849 JA |
505 | struct io_cancel { |
506 | struct file *file; | |
507 | u64 addr; | |
508 | }; | |
509 | ||
b29472ee JA |
510 | struct io_timeout { |
511 | struct file *file; | |
bfe68a22 PB |
512 | u32 off; |
513 | u32 target_seq; | |
135fcde8 | 514 | struct list_head list; |
90cd7e42 PB |
515 | /* head of the link, used by linked timeouts only */ |
516 | struct io_kiocb *head; | |
b29472ee JA |
517 | }; |
518 | ||
0bdf7a2d PB |
519 | struct io_timeout_rem { |
520 | struct file *file; | |
521 | u64 addr; | |
9c8e11b3 PB |
522 | |
523 | /* timeout update */ | |
524 | struct timespec64 ts; | |
525 | u32 flags; | |
0bdf7a2d PB |
526 | }; |
527 | ||
9adbd45d JA |
528 | struct io_rw { |
529 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
530 | struct kiocb kiocb; | |
531 | u64 addr; | |
532 | u64 len; | |
533 | }; | |
534 | ||
3fbb51c1 JA |
535 | struct io_connect { |
536 | struct file *file; | |
537 | struct sockaddr __user *addr; | |
538 | int addr_len; | |
539 | }; | |
540 | ||
e47293fd JA |
541 | struct io_sr_msg { |
542 | struct file *file; | |
fddaface | 543 | union { |
270a5940 | 544 | struct user_msghdr __user *umsg; |
fddaface JA |
545 | void __user *buf; |
546 | }; | |
e47293fd | 547 | int msg_flags; |
bcda7baa | 548 | int bgid; |
fddaface | 549 | size_t len; |
bcda7baa | 550 | struct io_buffer *kbuf; |
e47293fd JA |
551 | }; |
552 | ||
15b71abe JA |
553 | struct io_open { |
554 | struct file *file; | |
555 | int dfd; | |
15b71abe | 556 | struct filename *filename; |
c12cedf2 | 557 | struct open_how how; |
4022e7af | 558 | unsigned long nofile; |
15b71abe JA |
559 | }; |
560 | ||
269bbe5f | 561 | struct io_rsrc_update { |
05f3fb3c JA |
562 | struct file *file; |
563 | u64 arg; | |
564 | u32 nr_args; | |
565 | u32 offset; | |
566 | }; | |
567 | ||
4840e418 JA |
568 | struct io_fadvise { |
569 | struct file *file; | |
570 | u64 offset; | |
571 | u32 len; | |
572 | u32 advice; | |
573 | }; | |
574 | ||
c1ca757b JA |
575 | struct io_madvise { |
576 | struct file *file; | |
577 | u64 addr; | |
578 | u32 len; | |
579 | u32 advice; | |
580 | }; | |
581 | ||
3e4827b0 JA |
582 | struct io_epoll { |
583 | struct file *file; | |
584 | int epfd; | |
585 | int op; | |
586 | int fd; | |
587 | struct epoll_event event; | |
e47293fd JA |
588 | }; |
589 | ||
7d67af2c PB |
590 | struct io_splice { |
591 | struct file *file_out; | |
592 | struct file *file_in; | |
593 | loff_t off_out; | |
594 | loff_t off_in; | |
595 | u64 len; | |
596 | unsigned int flags; | |
597 | }; | |
598 | ||
ddf0322d JA |
599 | struct io_provide_buf { |
600 | struct file *file; | |
601 | __u64 addr; | |
602 | __s32 len; | |
603 | __u32 bgid; | |
604 | __u16 nbufs; | |
605 | __u16 bid; | |
606 | }; | |
607 | ||
1d9e1288 BM |
608 | struct io_statx { |
609 | struct file *file; | |
610 | int dfd; | |
611 | unsigned int mask; | |
612 | unsigned int flags; | |
e62753e4 | 613 | const char __user *filename; |
1d9e1288 BM |
614 | struct statx __user *buffer; |
615 | }; | |
616 | ||
36f4fa68 JA |
617 | struct io_shutdown { |
618 | struct file *file; | |
619 | int how; | |
620 | }; | |
621 | ||
80a261fd JA |
622 | struct io_rename { |
623 | struct file *file; | |
624 | int old_dfd; | |
625 | int new_dfd; | |
626 | struct filename *oldpath; | |
627 | struct filename *newpath; | |
628 | int flags; | |
629 | }; | |
630 | ||
14a1143b JA |
631 | struct io_unlink { |
632 | struct file *file; | |
633 | int dfd; | |
634 | int flags; | |
635 | struct filename *filename; | |
636 | }; | |
637 | ||
3ca405eb PB |
638 | struct io_completion { |
639 | struct file *file; | |
640 | struct list_head list; | |
0f7e466b | 641 | int cflags; |
3ca405eb PB |
642 | }; |
643 | ||
f499a021 JA |
644 | struct io_async_connect { |
645 | struct sockaddr_storage address; | |
646 | }; | |
647 | ||
03b1230c JA |
648 | struct io_async_msghdr { |
649 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
650 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
651 | struct iovec *free_iov; | |
03b1230c JA |
652 | struct sockaddr __user *uaddr; |
653 | struct msghdr msg; | |
b537916c | 654 | struct sockaddr_storage addr; |
03b1230c JA |
655 | }; |
656 | ||
f67676d1 JA |
657 | struct io_async_rw { |
658 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
659 | const struct iovec *free_iovec; |
660 | struct iov_iter iter; | |
227c0c96 | 661 | size_t bytes_done; |
bcf5a063 | 662 | struct wait_page_queue wpq; |
f67676d1 JA |
663 | }; |
664 | ||
6b47ee6e PB |
665 | enum { |
666 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
667 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
668 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
669 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
670 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 671 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 672 | |
6b47ee6e PB |
673 | REQ_F_FAIL_LINK_BIT, |
674 | REQ_F_INFLIGHT_BIT, | |
675 | REQ_F_CUR_POS_BIT, | |
676 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 677 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 678 | REQ_F_ISREG_BIT, |
99bc4c38 | 679 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 680 | REQ_F_POLLED_BIT, |
bcda7baa | 681 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 682 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 683 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 684 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
685 | |
686 | /* not a real bit, just to check we're not overflowing the space */ | |
687 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
688 | }; |
689 | ||
690 | enum { | |
691 | /* ctx owns file */ | |
692 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
693 | /* drain existing IO first */ | |
694 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
695 | /* linked sqes */ | |
696 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
697 | /* doesn't sever on completion < 0 */ | |
698 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
699 | /* IOSQE_ASYNC */ | |
700 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
701 | /* IOSQE_BUFFER_SELECT */ |
702 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 703 | |
6b47ee6e PB |
704 | /* fail rest of links */ |
705 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
706 | /* on inflight list */ | |
707 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
708 | /* read/write uses file position */ | |
709 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
710 | /* must not punt to workers */ | |
711 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 712 | /* has or had linked timeout */ |
6b47ee6e | 713 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
714 | /* regular file */ |
715 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
716 | /* needs cleanup */ |
717 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
718 | /* already went through poll handler */ |
719 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
720 | /* buffer already selected */ |
721 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
722 | /* doesn't need file table for this request */ |
723 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
724 | /* linked timeout is active, i.e. prepared by link's head */ |
725 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
726 | /* completion is deferred through io_comp_state */ |
727 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
728 | }; |
729 | ||
730 | struct async_poll { | |
731 | struct io_poll_iocb poll; | |
807abcb0 | 732 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
733 | }; |
734 | ||
7cbf1722 JA |
735 | struct io_task_work { |
736 | struct io_wq_work_node node; | |
737 | task_work_func_t func; | |
738 | }; | |
739 | ||
09bb8394 JA |
740 | /* |
741 | * NOTE! Each of the iocb union members has the file pointer | |
742 | * as the first entry in their struct definition. So you can | |
743 | * access the file pointer through any of the sub-structs, | |
744 | * or directly as just 'ki_filp' in this struct. | |
745 | */ | |
2b188cc1 | 746 | struct io_kiocb { |
221c5eb2 | 747 | union { |
09bb8394 | 748 | struct file *file; |
9adbd45d | 749 | struct io_rw rw; |
221c5eb2 | 750 | struct io_poll_iocb poll; |
018043be | 751 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
752 | struct io_accept accept; |
753 | struct io_sync sync; | |
fbf23849 | 754 | struct io_cancel cancel; |
b29472ee | 755 | struct io_timeout timeout; |
0bdf7a2d | 756 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 757 | struct io_connect connect; |
e47293fd | 758 | struct io_sr_msg sr_msg; |
15b71abe | 759 | struct io_open open; |
b5dba59e | 760 | struct io_close close; |
269bbe5f | 761 | struct io_rsrc_update rsrc_update; |
4840e418 | 762 | struct io_fadvise fadvise; |
c1ca757b | 763 | struct io_madvise madvise; |
3e4827b0 | 764 | struct io_epoll epoll; |
7d67af2c | 765 | struct io_splice splice; |
ddf0322d | 766 | struct io_provide_buf pbuf; |
1d9e1288 | 767 | struct io_statx statx; |
36f4fa68 | 768 | struct io_shutdown shutdown; |
80a261fd | 769 | struct io_rename rename; |
14a1143b | 770 | struct io_unlink unlink; |
3ca405eb PB |
771 | /* use only after cleaning per-op data, see io_clean_op() */ |
772 | struct io_completion compl; | |
221c5eb2 | 773 | }; |
2b188cc1 | 774 | |
e8c2bc1f JA |
775 | /* opcode allocated if it needs to store data for async defer */ |
776 | void *async_data; | |
d625c6ee | 777 | u8 opcode; |
65a6543d XW |
778 | /* polled IO has completed */ |
779 | u8 iopoll_completed; | |
2b188cc1 | 780 | |
4f4eeba8 | 781 | u16 buf_index; |
9cf7c104 | 782 | u32 result; |
4f4eeba8 | 783 | |
010e8e6b PB |
784 | struct io_ring_ctx *ctx; |
785 | unsigned int flags; | |
786 | refcount_t refs; | |
787 | struct task_struct *task; | |
788 | u64 user_data; | |
d7718a9d | 789 | |
f2f87370 | 790 | struct io_kiocb *link; |
269bbe5f | 791 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 792 | |
d21ffe7e PB |
793 | /* |
794 | * 1. used with ctx->iopoll_list with reads/writes | |
795 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
796 | */ | |
010e8e6b | 797 | struct list_head inflight_entry; |
7cbf1722 JA |
798 | union { |
799 | struct io_task_work io_task_work; | |
800 | struct callback_head task_work; | |
801 | }; | |
010e8e6b PB |
802 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
803 | struct hlist_node hash_node; | |
804 | struct async_poll *apoll; | |
805 | struct io_wq_work work; | |
2b188cc1 | 806 | }; |
05589553 | 807 | |
27dc8338 PB |
808 | struct io_defer_entry { |
809 | struct list_head list; | |
810 | struct io_kiocb *req; | |
9cf7c104 | 811 | u32 seq; |
2b188cc1 JA |
812 | }; |
813 | ||
d3656344 | 814 | struct io_op_def { |
d3656344 JA |
815 | /* needs req->file assigned */ |
816 | unsigned needs_file : 1; | |
d3656344 JA |
817 | /* hash wq insertion if file is a regular file */ |
818 | unsigned hash_reg_file : 1; | |
819 | /* unbound wq insertion if file is a non-regular file */ | |
820 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
821 | /* opcode is not supported by this kernel */ |
822 | unsigned not_supported : 1; | |
8a72758c JA |
823 | /* set if opcode supports polled "wait" */ |
824 | unsigned pollin : 1; | |
825 | unsigned pollout : 1; | |
bcda7baa JA |
826 | /* op supports buffer selection */ |
827 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
828 | /* must always have async data allocated */ |
829 | unsigned needs_async_data : 1; | |
27926b68 JA |
830 | /* should block plug */ |
831 | unsigned plug : 1; | |
e8c2bc1f JA |
832 | /* size of async data needed, if any */ |
833 | unsigned short async_size; | |
d3656344 JA |
834 | }; |
835 | ||
0918682b | 836 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
837 | [IORING_OP_NOP] = {}, |
838 | [IORING_OP_READV] = { | |
d3656344 JA |
839 | .needs_file = 1, |
840 | .unbound_nonreg_file = 1, | |
8a72758c | 841 | .pollin = 1, |
4d954c25 | 842 | .buffer_select = 1, |
e8c2bc1f | 843 | .needs_async_data = 1, |
27926b68 | 844 | .plug = 1, |
e8c2bc1f | 845 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 846 | }, |
0463b6c5 | 847 | [IORING_OP_WRITEV] = { |
d3656344 JA |
848 | .needs_file = 1, |
849 | .hash_reg_file = 1, | |
850 | .unbound_nonreg_file = 1, | |
8a72758c | 851 | .pollout = 1, |
e8c2bc1f | 852 | .needs_async_data = 1, |
27926b68 | 853 | .plug = 1, |
e8c2bc1f | 854 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 855 | }, |
0463b6c5 | 856 | [IORING_OP_FSYNC] = { |
d3656344 JA |
857 | .needs_file = 1, |
858 | }, | |
0463b6c5 | 859 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
860 | .needs_file = 1, |
861 | .unbound_nonreg_file = 1, | |
8a72758c | 862 | .pollin = 1, |
27926b68 | 863 | .plug = 1, |
e8c2bc1f | 864 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 865 | }, |
0463b6c5 | 866 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
867 | .needs_file = 1, |
868 | .hash_reg_file = 1, | |
869 | .unbound_nonreg_file = 1, | |
8a72758c | 870 | .pollout = 1, |
27926b68 | 871 | .plug = 1, |
e8c2bc1f | 872 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 873 | }, |
0463b6c5 | 874 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
875 | .needs_file = 1, |
876 | .unbound_nonreg_file = 1, | |
877 | }, | |
0463b6c5 PB |
878 | [IORING_OP_POLL_REMOVE] = {}, |
879 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
880 | .needs_file = 1, |
881 | }, | |
0463b6c5 | 882 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
883 | .needs_file = 1, |
884 | .unbound_nonreg_file = 1, | |
8a72758c | 885 | .pollout = 1, |
e8c2bc1f JA |
886 | .needs_async_data = 1, |
887 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 888 | }, |
0463b6c5 | 889 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
890 | .needs_file = 1, |
891 | .unbound_nonreg_file = 1, | |
8a72758c | 892 | .pollin = 1, |
52de1fe1 | 893 | .buffer_select = 1, |
e8c2bc1f JA |
894 | .needs_async_data = 1, |
895 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 896 | }, |
0463b6c5 | 897 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
898 | .needs_async_data = 1, |
899 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 900 | }, |
9c8e11b3 PB |
901 | [IORING_OP_TIMEOUT_REMOVE] = { |
902 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 903 | }, |
0463b6c5 | 904 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
905 | .needs_file = 1, |
906 | .unbound_nonreg_file = 1, | |
8a72758c | 907 | .pollin = 1, |
d3656344 | 908 | }, |
0463b6c5 PB |
909 | [IORING_OP_ASYNC_CANCEL] = {}, |
910 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
911 | .needs_async_data = 1, |
912 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 913 | }, |
0463b6c5 | 914 | [IORING_OP_CONNECT] = { |
d3656344 JA |
915 | .needs_file = 1, |
916 | .unbound_nonreg_file = 1, | |
8a72758c | 917 | .pollout = 1, |
e8c2bc1f JA |
918 | .needs_async_data = 1, |
919 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 920 | }, |
0463b6c5 | 921 | [IORING_OP_FALLOCATE] = { |
d3656344 | 922 | .needs_file = 1, |
d3656344 | 923 | }, |
44526bed JA |
924 | [IORING_OP_OPENAT] = {}, |
925 | [IORING_OP_CLOSE] = {}, | |
926 | [IORING_OP_FILES_UPDATE] = {}, | |
927 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 928 | [IORING_OP_READ] = { |
3a6820f2 JA |
929 | .needs_file = 1, |
930 | .unbound_nonreg_file = 1, | |
8a72758c | 931 | .pollin = 1, |
bcda7baa | 932 | .buffer_select = 1, |
27926b68 | 933 | .plug = 1, |
e8c2bc1f | 934 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 935 | }, |
0463b6c5 | 936 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
937 | .needs_file = 1, |
938 | .unbound_nonreg_file = 1, | |
8a72758c | 939 | .pollout = 1, |
27926b68 | 940 | .plug = 1, |
e8c2bc1f | 941 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_FADVISE] = { |
4840e418 | 944 | .needs_file = 1, |
c1ca757b | 945 | }, |
44526bed | 946 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 947 | [IORING_OP_SEND] = { |
fddaface JA |
948 | .needs_file = 1, |
949 | .unbound_nonreg_file = 1, | |
8a72758c | 950 | .pollout = 1, |
fddaface | 951 | }, |
0463b6c5 | 952 | [IORING_OP_RECV] = { |
fddaface JA |
953 | .needs_file = 1, |
954 | .unbound_nonreg_file = 1, | |
8a72758c | 955 | .pollin = 1, |
bcda7baa | 956 | .buffer_select = 1, |
fddaface | 957 | }, |
0463b6c5 | 958 | [IORING_OP_OPENAT2] = { |
cebdb986 | 959 | }, |
3e4827b0 JA |
960 | [IORING_OP_EPOLL_CTL] = { |
961 | .unbound_nonreg_file = 1, | |
3e4827b0 | 962 | }, |
7d67af2c PB |
963 | [IORING_OP_SPLICE] = { |
964 | .needs_file = 1, | |
965 | .hash_reg_file = 1, | |
966 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
967 | }, |
968 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 969 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
970 | [IORING_OP_TEE] = { |
971 | .needs_file = 1, | |
972 | .hash_reg_file = 1, | |
973 | .unbound_nonreg_file = 1, | |
974 | }, | |
36f4fa68 JA |
975 | [IORING_OP_SHUTDOWN] = { |
976 | .needs_file = 1, | |
977 | }, | |
44526bed JA |
978 | [IORING_OP_RENAMEAT] = {}, |
979 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
980 | }; |
981 | ||
9936c7c2 PB |
982 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
983 | struct task_struct *task, | |
984 | struct files_struct *files); | |
37d1e2e3 | 985 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 986 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 987 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 988 | struct io_ring_ctx *ctx); |
f2303b1f | 989 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 990 | |
23faba36 | 991 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 992 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 993 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 994 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 995 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
996 | static void io_dismantle_req(struct io_kiocb *req); |
997 | static void io_put_task(struct task_struct *task, int nr); | |
998 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 999 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1000 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1001 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1002 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1003 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1004 | unsigned nr_args); |
3ca405eb | 1005 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1006 | static struct file *io_file_get(struct io_submit_state *state, |
1007 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1008 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1009 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1010 | |
847595de PB |
1011 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1012 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1013 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1014 | const struct iovec *fast_iov, | |
227c0c96 | 1015 | struct iov_iter *iter, bool force); |
907d1df3 | 1016 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1017 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1018 | struct io_ring_ctx *ctx); | |
de0617e4 | 1019 | |
2b188cc1 JA |
1020 | static struct kmem_cache *req_cachep; |
1021 | ||
0918682b | 1022 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1023 | |
1024 | struct sock *io_uring_get_socket(struct file *file) | |
1025 | { | |
1026 | #if defined(CONFIG_UNIX) | |
1027 | if (file->f_op == &io_uring_fops) { | |
1028 | struct io_ring_ctx *ctx = file->private_data; | |
1029 | ||
1030 | return ctx->ring_sock->sk; | |
1031 | } | |
1032 | #endif | |
1033 | return NULL; | |
1034 | } | |
1035 | EXPORT_SYMBOL(io_uring_get_socket); | |
1036 | ||
f2f87370 PB |
1037 | #define io_for_each_link(pos, head) \ |
1038 | for (pos = (head); pos; pos = pos->link) | |
1039 | ||
3ca405eb PB |
1040 | static inline void io_clean_op(struct io_kiocb *req) |
1041 | { | |
9d5c8190 | 1042 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1043 | __io_clean_op(req); |
1044 | } | |
1045 | ||
36f72fe2 PB |
1046 | static inline void io_set_resource_node(struct io_kiocb *req) |
1047 | { | |
1048 | struct io_ring_ctx *ctx = req->ctx; | |
1049 | ||
269bbe5f BM |
1050 | if (!req->fixed_rsrc_refs) { |
1051 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1052 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1053 | } |
1054 | } | |
1055 | ||
08d23634 PB |
1056 | static bool io_match_task(struct io_kiocb *head, |
1057 | struct task_struct *task, | |
1058 | struct files_struct *files) | |
1059 | { | |
1060 | struct io_kiocb *req; | |
1061 | ||
84965ff8 JA |
1062 | if (task && head->task != task) { |
1063 | /* in terms of cancelation, always match if req task is dead */ | |
1064 | if (head->task->flags & PF_EXITING) | |
1065 | return true; | |
08d23634 | 1066 | return false; |
84965ff8 | 1067 | } |
08d23634 PB |
1068 | if (!files) |
1069 | return true; | |
1070 | ||
1071 | io_for_each_link(req, head) { | |
02a13674 JA |
1072 | if (req->file && req->file->f_op == &io_uring_fops) |
1073 | return true; | |
4379bf8b | 1074 | if (req->task->files == files) |
08d23634 PB |
1075 | return true; |
1076 | } | |
1077 | return false; | |
1078 | } | |
1079 | ||
c40f6379 JA |
1080 | static inline void req_set_fail_links(struct io_kiocb *req) |
1081 | { | |
1082 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1083 | req->flags |= REQ_F_FAIL_LINK; | |
1084 | } | |
4a38aed2 | 1085 | |
2b188cc1 JA |
1086 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1087 | { | |
1088 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1089 | ||
0f158b4c | 1090 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1091 | } |
1092 | ||
8eb7e2d0 PB |
1093 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1094 | { | |
1095 | return !req->timeout.off; | |
1096 | } | |
1097 | ||
2b188cc1 JA |
1098 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1099 | { | |
1100 | struct io_ring_ctx *ctx; | |
78076bb6 | 1101 | int hash_bits; |
2b188cc1 JA |
1102 | |
1103 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1104 | if (!ctx) | |
1105 | return NULL; | |
1106 | ||
78076bb6 JA |
1107 | /* |
1108 | * Use 5 bits less than the max cq entries, that should give us around | |
1109 | * 32 entries per hash list if totally full and uniformly spread. | |
1110 | */ | |
1111 | hash_bits = ilog2(p->cq_entries); | |
1112 | hash_bits -= 5; | |
1113 | if (hash_bits <= 0) | |
1114 | hash_bits = 1; | |
1115 | ctx->cancel_hash_bits = hash_bits; | |
1116 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1117 | GFP_KERNEL); | |
1118 | if (!ctx->cancel_hash) | |
1119 | goto err; | |
1120 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1121 | ||
21482896 | 1122 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1123 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1124 | goto err; | |
2b188cc1 JA |
1125 | |
1126 | ctx->flags = p->flags; | |
90554200 | 1127 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1128 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1129 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1130 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1131 | init_completion(&ctx->ref_comp); |
1132 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1133 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1134 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1135 | mutex_init(&ctx->uring_lock); |
1136 | init_waitqueue_head(&ctx->wait); | |
1137 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1138 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1139 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1140 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1141 | spin_lock_init(&ctx->inflight_lock); |
1142 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1143 | spin_lock_init(&ctx->rsrc_ref_lock); |
1144 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1145 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1146 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1147 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1148 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1149 | return ctx; |
206aefde | 1150 | err: |
78076bb6 | 1151 | kfree(ctx->cancel_hash); |
206aefde JA |
1152 | kfree(ctx); |
1153 | return NULL; | |
2b188cc1 JA |
1154 | } |
1155 | ||
9cf7c104 | 1156 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1157 | { |
2bc9930e JA |
1158 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1159 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1160 | |
9cf7c104 | 1161 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1162 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1163 | } |
de0617e4 | 1164 | |
9d858b21 | 1165 | return false; |
de0617e4 JA |
1166 | } |
1167 | ||
ce3d5aae PB |
1168 | static void io_req_track_inflight(struct io_kiocb *req) |
1169 | { | |
1170 | struct io_ring_ctx *ctx = req->ctx; | |
1171 | ||
1172 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1173 | req->flags |= REQ_F_INFLIGHT; |
1174 | ||
1175 | spin_lock_irq(&ctx->inflight_lock); | |
1176 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1177 | spin_unlock_irq(&ctx->inflight_lock); | |
1178 | } | |
1179 | } | |
1180 | ||
1e6fa521 JA |
1181 | static void io_prep_async_work(struct io_kiocb *req) |
1182 | { | |
1183 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1184 | struct io_ring_ctx *ctx = req->ctx; |
1185 | ||
feaadc4f PB |
1186 | if (req->flags & REQ_F_FORCE_ASYNC) |
1187 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1188 | ||
1e6fa521 JA |
1189 | if (req->flags & REQ_F_ISREG) { |
1190 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1191 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1192 | } else { | |
1193 | if (def->unbound_nonreg_file) | |
1194 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1195 | } | |
561fb04a | 1196 | } |
cccf0ee8 | 1197 | |
cbdcb435 | 1198 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1199 | { |
cbdcb435 | 1200 | struct io_kiocb *cur; |
54a91f3b | 1201 | |
f2f87370 PB |
1202 | io_for_each_link(cur, req) |
1203 | io_prep_async_work(cur); | |
561fb04a JA |
1204 | } |
1205 | ||
ebf93667 | 1206 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1207 | { |
a197f664 | 1208 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1209 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1210 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1211 | |
3bfe6106 JA |
1212 | BUG_ON(!tctx); |
1213 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1214 | |
8766dd51 PB |
1215 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1216 | &req->work, req->flags); | |
cbdcb435 PB |
1217 | /* init ->work of the whole link before punting */ |
1218 | io_prep_async_link(req); | |
ebf93667 | 1219 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1220 | if (link) |
1221 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1222 | } |
1223 | ||
5262f567 JA |
1224 | static void io_kill_timeout(struct io_kiocb *req) |
1225 | { | |
e8c2bc1f | 1226 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1227 | int ret; |
1228 | ||
e8c2bc1f | 1229 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1230 | if (ret != -1) { |
01cec8c1 PB |
1231 | atomic_set(&req->ctx->cq_timeouts, |
1232 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1233 | list_del_init(&req->timeout.list); |
78e19bbe | 1234 | io_cqring_fill_event(req, 0); |
216578e5 | 1235 | io_put_req_deferred(req, 1); |
5262f567 JA |
1236 | } |
1237 | } | |
1238 | ||
76e1b642 JA |
1239 | /* |
1240 | * Returns true if we found and killed one or more timeouts | |
1241 | */ | |
6b81928d PB |
1242 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1243 | struct files_struct *files) | |
5262f567 JA |
1244 | { |
1245 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1246 | int canceled = 0; |
5262f567 JA |
1247 | |
1248 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1249 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1250 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1251 | io_kill_timeout(req); |
76e1b642 JA |
1252 | canceled++; |
1253 | } | |
f3606e3a | 1254 | } |
5262f567 | 1255 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1256 | return canceled != 0; |
5262f567 JA |
1257 | } |
1258 | ||
04518945 | 1259 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1260 | { |
04518945 | 1261 | do { |
27dc8338 PB |
1262 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1263 | struct io_defer_entry, list); | |
de0617e4 | 1264 | |
9cf7c104 | 1265 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1266 | break; |
27dc8338 | 1267 | list_del_init(&de->list); |
907d1df3 | 1268 | io_req_task_queue(de->req); |
27dc8338 | 1269 | kfree(de); |
04518945 PB |
1270 | } while (!list_empty(&ctx->defer_list)); |
1271 | } | |
1272 | ||
360428f8 | 1273 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1274 | { |
f010505b MDG |
1275 | u32 seq; |
1276 | ||
1277 | if (list_empty(&ctx->timeout_list)) | |
1278 | return; | |
1279 | ||
1280 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1281 | ||
1282 | do { | |
1283 | u32 events_needed, events_got; | |
360428f8 | 1284 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1285 | struct io_kiocb, timeout.list); |
de0617e4 | 1286 | |
8eb7e2d0 | 1287 | if (io_is_timeout_noseq(req)) |
360428f8 | 1288 | break; |
f010505b MDG |
1289 | |
1290 | /* | |
1291 | * Since seq can easily wrap around over time, subtract | |
1292 | * the last seq at which timeouts were flushed before comparing. | |
1293 | * Assuming not more than 2^31-1 events have happened since, | |
1294 | * these subtractions won't have wrapped, so we can check if | |
1295 | * target is in [last_seq, current_seq] by comparing the two. | |
1296 | */ | |
1297 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1298 | events_got = seq - ctx->cq_last_tm_flush; | |
1299 | if (events_got < events_needed) | |
360428f8 | 1300 | break; |
bfe68a22 | 1301 | |
135fcde8 | 1302 | list_del_init(&req->timeout.list); |
5262f567 | 1303 | io_kill_timeout(req); |
f010505b MDG |
1304 | } while (!list_empty(&ctx->timeout_list)); |
1305 | ||
1306 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1307 | } |
5262f567 | 1308 | |
360428f8 PB |
1309 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1310 | { | |
1311 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1312 | |
1313 | /* order cqe stores with ring update */ | |
1314 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1315 | |
04518945 PB |
1316 | if (unlikely(!list_empty(&ctx->defer_list))) |
1317 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1318 | } |
1319 | ||
90554200 JA |
1320 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1321 | { | |
1322 | struct io_rings *r = ctx->rings; | |
1323 | ||
1324 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1325 | } | |
1326 | ||
888aae2e PB |
1327 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1328 | { | |
1329 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1330 | } | |
1331 | ||
2b188cc1 JA |
1332 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1333 | { | |
75b28aff | 1334 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1335 | unsigned tail; |
1336 | ||
115e12e5 SB |
1337 | /* |
1338 | * writes to the cq entry need to come after reading head; the | |
1339 | * control dependency is enough as we're using WRITE_ONCE to | |
1340 | * fill the cq entry | |
1341 | */ | |
888aae2e | 1342 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1343 | return NULL; |
1344 | ||
888aae2e | 1345 | tail = ctx->cached_cq_tail++; |
75b28aff | 1346 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1347 | } |
1348 | ||
f2842ab5 JA |
1349 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1350 | { | |
f0b493e6 JA |
1351 | if (!ctx->cq_ev_fd) |
1352 | return false; | |
7e55a19c SG |
1353 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1354 | return false; | |
f2842ab5 JA |
1355 | if (!ctx->eventfd_async) |
1356 | return true; | |
b41e9852 | 1357 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1358 | } |
1359 | ||
b41e9852 | 1360 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1361 | { |
b1445e59 PB |
1362 | /* see waitqueue_active() comment */ |
1363 | smp_mb(); | |
1364 | ||
1d7bb1d5 JA |
1365 | if (waitqueue_active(&ctx->wait)) |
1366 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1367 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1368 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1369 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1370 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1371 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1372 | wake_up_interruptible(&ctx->cq_wait); |
1373 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1374 | } | |
1d7bb1d5 JA |
1375 | } |
1376 | ||
80c18e4a PB |
1377 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1378 | { | |
b1445e59 PB |
1379 | /* see waitqueue_active() comment */ |
1380 | smp_mb(); | |
1381 | ||
80c18e4a PB |
1382 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1383 | if (waitqueue_active(&ctx->wait)) | |
1384 | wake_up(&ctx->wait); | |
1385 | } | |
1386 | if (io_should_trigger_evfd(ctx)) | |
1387 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1388 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1389 | wake_up_interruptible(&ctx->cq_wait); |
1390 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1391 | } | |
80c18e4a PB |
1392 | } |
1393 | ||
c4a2ed72 | 1394 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1395 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1396 | struct task_struct *tsk, | |
1397 | struct files_struct *files) | |
1d7bb1d5 JA |
1398 | { |
1399 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1400 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1401 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1402 | unsigned long flags; |
b18032bb | 1403 | bool all_flushed, posted; |
1d7bb1d5 JA |
1404 | LIST_HEAD(list); |
1405 | ||
e23de15f PB |
1406 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1407 | return false; | |
1d7bb1d5 | 1408 | |
b18032bb | 1409 | posted = false; |
1d7bb1d5 | 1410 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1411 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1412 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1413 | continue; |
1414 | ||
1d7bb1d5 JA |
1415 | cqe = io_get_cqring(ctx); |
1416 | if (!cqe && !force) | |
1417 | break; | |
1418 | ||
40d8ddd4 | 1419 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1420 | if (cqe) { |
1421 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1422 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1423 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1424 | } else { |
2c3bac6d | 1425 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1426 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1427 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1428 | } |
b18032bb | 1429 | posted = true; |
1d7bb1d5 JA |
1430 | } |
1431 | ||
09e88404 PB |
1432 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1433 | if (all_flushed) { | |
1434 | clear_bit(0, &ctx->sq_check_overflow); | |
1435 | clear_bit(0, &ctx->cq_check_overflow); | |
1436 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1437 | } | |
46930143 | 1438 | |
b18032bb JA |
1439 | if (posted) |
1440 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1441 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1442 | if (posted) |
1443 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1444 | |
1445 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1446 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1447 | list_del(&req->compl.list); | |
ec9c02ad | 1448 | io_put_req(req); |
1d7bb1d5 | 1449 | } |
c4a2ed72 | 1450 | |
09e88404 | 1451 | return all_flushed; |
1d7bb1d5 JA |
1452 | } |
1453 | ||
6c503150 PB |
1454 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1455 | struct task_struct *tsk, | |
1456 | struct files_struct *files) | |
1457 | { | |
1458 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1459 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1460 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1461 | mutex_lock(&ctx->uring_lock); | |
1462 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1463 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1464 | mutex_unlock(&ctx->uring_lock); | |
1465 | } | |
1466 | } | |
1467 | ||
bcda7baa | 1468 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1469 | { |
78e19bbe | 1470 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1471 | struct io_uring_cqe *cqe; |
1472 | ||
78e19bbe | 1473 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1474 | |
2b188cc1 JA |
1475 | /* |
1476 | * If we can't get a cq entry, userspace overflowed the | |
1477 | * submission (by quite a lot). Increment the overflow count in | |
1478 | * the ring. | |
1479 | */ | |
1480 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1481 | if (likely(cqe)) { |
78e19bbe | 1482 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1483 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1484 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1485 | } else if (ctx->cq_overflow_flushed || |
1486 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1487 | /* |
1488 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1489 | * then we cannot store the request for later flushing, we need | |
1490 | * to drop it on the floor. | |
1491 | */ | |
2c3bac6d PB |
1492 | ctx->cached_cq_overflow++; |
1493 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1494 | } else { |
ad3eb2c8 JA |
1495 | if (list_empty(&ctx->cq_overflow_list)) { |
1496 | set_bit(0, &ctx->sq_check_overflow); | |
1497 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1498 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1499 | } |
40d8ddd4 | 1500 | io_clean_op(req); |
1d7bb1d5 | 1501 | req->result = res; |
0f7e466b | 1502 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1503 | refcount_inc(&req->refs); |
1504 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1505 | } |
1506 | } | |
1507 | ||
bcda7baa JA |
1508 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1509 | { | |
1510 | __io_cqring_fill_event(req, res, 0); | |
1511 | } | |
1512 | ||
c7dae4ba JA |
1513 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1514 | unsigned int cflags) | |
2b188cc1 | 1515 | { |
78e19bbe | 1516 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1517 | unsigned long flags; |
1518 | ||
1519 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1520 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1521 | io_commit_cqring(ctx); |
c7dae4ba JA |
1522 | /* |
1523 | * If we're the last reference to this request, add to our locked | |
1524 | * free_list cache. | |
1525 | */ | |
1526 | if (refcount_dec_and_test(&req->refs)) { | |
1527 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1528 | ||
1529 | io_dismantle_req(req); | |
1530 | io_put_task(req->task, 1); | |
1531 | list_add(&req->compl.list, &cs->locked_free_list); | |
1532 | cs->locked_free_nr++; | |
1533 | } else | |
1534 | req = NULL; | |
2b188cc1 JA |
1535 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1536 | ||
8c838788 | 1537 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1538 | if (req) { |
1539 | io_queue_next(req); | |
1540 | percpu_ref_put(&ctx->refs); | |
229a7b63 | 1541 | } |
229a7b63 JA |
1542 | } |
1543 | ||
a38d68db | 1544 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1545 | unsigned int cflags) |
229a7b63 | 1546 | { |
a38d68db PB |
1547 | io_clean_op(req); |
1548 | req->result = res; | |
1549 | req->compl.cflags = cflags; | |
e342c807 | 1550 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1551 | } |
1552 | ||
889fca73 PB |
1553 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1554 | long res, unsigned cflags) | |
bcda7baa | 1555 | { |
889fca73 PB |
1556 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1557 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1558 | else |
c7dae4ba | 1559 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1560 | } |
1561 | ||
a38d68db | 1562 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1563 | { |
889fca73 | 1564 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1565 | } |
1566 | ||
c7dae4ba | 1567 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1568 | { |
c7dae4ba JA |
1569 | struct io_submit_state *state = &ctx->submit_state; |
1570 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1571 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1572 | |
c7dae4ba JA |
1573 | /* |
1574 | * If we have more than a batch's worth of requests in our IRQ side | |
1575 | * locked cache, grab the lock and move them over to our submission | |
1576 | * side cache. | |
1577 | */ | |
1578 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1579 | spin_lock_irq(&ctx->completion_lock); | |
1580 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1581 | cs->locked_free_nr = 0; | |
1582 | spin_unlock_irq(&ctx->completion_lock); | |
1583 | } | |
0ddf92e8 | 1584 | |
c7dae4ba JA |
1585 | while (!list_empty(&cs->free_list)) { |
1586 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1587 | compl.list); |
1588 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1589 | state->reqs[state->free_reqs++] = req; |
1590 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1591 | break; | |
1b4c351f JA |
1592 | } |
1593 | ||
e5d1bc0a | 1594 | return req != NULL; |
0ddf92e8 JA |
1595 | } |
1596 | ||
e5d1bc0a | 1597 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1598 | { |
e5d1bc0a PB |
1599 | struct io_submit_state *state = &ctx->submit_state; |
1600 | ||
1601 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1602 | ||
f6b6c7d6 | 1603 | if (!state->free_reqs) { |
291b2821 | 1604 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1605 | int ret; |
1606 | ||
c7dae4ba | 1607 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1608 | goto got_req; |
1609 | ||
bf019da7 PB |
1610 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1611 | state->reqs); | |
fd6fab2c JA |
1612 | |
1613 | /* | |
1614 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1615 | * retry single alloc to be on the safe side. | |
1616 | */ | |
1617 | if (unlikely(ret <= 0)) { | |
1618 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1619 | if (!state->reqs[0]) | |
3893f39f | 1620 | return NULL; |
fd6fab2c JA |
1621 | ret = 1; |
1622 | } | |
291b2821 | 1623 | state->free_reqs = ret; |
2b188cc1 | 1624 | } |
e5d1bc0a | 1625 | got_req: |
291b2821 PB |
1626 | state->free_reqs--; |
1627 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1628 | } |
1629 | ||
8da11c19 PB |
1630 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1631 | bool fixed) | |
1632 | { | |
36f72fe2 | 1633 | if (!fixed) |
8da11c19 PB |
1634 | fput(file); |
1635 | } | |
1636 | ||
4edf20f9 | 1637 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1638 | { |
3ca405eb | 1639 | io_clean_op(req); |
929a3af9 | 1640 | |
e8c2bc1f JA |
1641 | if (req->async_data) |
1642 | kfree(req->async_data); | |
8da11c19 PB |
1643 | if (req->file) |
1644 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1645 | if (req->fixed_rsrc_refs) |
1646 | percpu_ref_put(req->fixed_rsrc_refs); | |
f85c310a PB |
1647 | |
1648 | if (req->flags & REQ_F_INFLIGHT) { | |
1649 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1650 | unsigned long flags; |
1651 | ||
1652 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1653 | list_del(&req->inflight_entry); | |
1654 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1655 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1656 | } |
e65ef56d JA |
1657 | } |
1658 | ||
b23fcf47 | 1659 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1660 | static inline void io_put_task(struct task_struct *task, int nr) |
1661 | { | |
1662 | struct io_uring_task *tctx = task->io_uring; | |
1663 | ||
1664 | percpu_counter_sub(&tctx->inflight, nr); | |
1665 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1666 | wake_up(&tctx->wait); | |
1667 | put_task_struct_many(task, nr); | |
1668 | } | |
1669 | ||
216578e5 | 1670 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1671 | { |
51a4cc11 | 1672 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1673 | |
216578e5 | 1674 | io_dismantle_req(req); |
7c660731 | 1675 | io_put_task(req->task, 1); |
c6ca97b3 | 1676 | |
3893f39f | 1677 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1678 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1679 | } |
1680 | ||
f2f87370 PB |
1681 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1682 | { | |
1683 | struct io_kiocb *nxt = req->link; | |
1684 | ||
1685 | req->link = nxt->link; | |
1686 | nxt->link = NULL; | |
1687 | } | |
1688 | ||
c9abd7ad | 1689 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1690 | { |
a197f664 | 1691 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1692 | struct io_kiocb *link; |
c9abd7ad PB |
1693 | bool cancelled = false; |
1694 | unsigned long flags; | |
7c86ffee | 1695 | |
c9abd7ad | 1696 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1697 | link = req->link; |
1698 | ||
900fad45 PB |
1699 | /* |
1700 | * Can happen if a linked timeout fired and link had been like | |
1701 | * req -> link t-out -> link t-out [-> ...] | |
1702 | */ | |
c9abd7ad PB |
1703 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1704 | struct io_timeout_data *io = link->async_data; | |
1705 | int ret; | |
7c86ffee | 1706 | |
f2f87370 | 1707 | io_remove_next_linked(req); |
90cd7e42 | 1708 | link->timeout.head = NULL; |
c9abd7ad PB |
1709 | ret = hrtimer_try_to_cancel(&io->timer); |
1710 | if (ret != -1) { | |
1711 | io_cqring_fill_event(link, -ECANCELED); | |
1712 | io_commit_cqring(ctx); | |
1713 | cancelled = true; | |
1714 | } | |
1715 | } | |
7c86ffee | 1716 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1717 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1718 | |
c9abd7ad | 1719 | if (cancelled) { |
7c86ffee | 1720 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1721 | io_put_req(link); |
1722 | } | |
7c86ffee PB |
1723 | } |
1724 | ||
9e645e11 | 1725 | |
d148ca4b | 1726 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1727 | { |
f2f87370 | 1728 | struct io_kiocb *link, *nxt; |
2665abfd | 1729 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1730 | unsigned long flags; |
9e645e11 | 1731 | |
d148ca4b | 1732 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1733 | link = req->link; |
1734 | req->link = NULL; | |
9e645e11 | 1735 | |
f2f87370 PB |
1736 | while (link) { |
1737 | nxt = link->link; | |
1738 | link->link = NULL; | |
2665abfd | 1739 | |
f2f87370 | 1740 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1741 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 | 1742 | |
1575f21a | 1743 | io_put_req_deferred(link, 2); |
f2f87370 | 1744 | link = nxt; |
9e645e11 | 1745 | } |
2665abfd | 1746 | io_commit_cqring(ctx); |
216578e5 | 1747 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1748 | |
2665abfd | 1749 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1750 | } |
1751 | ||
3fa5e0f3 | 1752 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1753 | { |
7c86ffee PB |
1754 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1755 | io_kill_linked_timeout(req); | |
944e58bf | 1756 | |
9e645e11 JA |
1757 | /* |
1758 | * If LINK is set, we have dependent requests in this chain. If we | |
1759 | * didn't fail this request, queue the first one up, moving any other | |
1760 | * dependencies to the next request. In case of failure, fail the rest | |
1761 | * of the chain. | |
1762 | */ | |
f2f87370 PB |
1763 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
1764 | struct io_kiocb *nxt = req->link; | |
1765 | ||
1766 | req->link = NULL; | |
1767 | return nxt; | |
1768 | } | |
9b5f7bd9 PB |
1769 | io_fail_links(req); |
1770 | return NULL; | |
4d7dd462 | 1771 | } |
9e645e11 | 1772 | |
f2f87370 | 1773 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1774 | { |
cdbff982 | 1775 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1776 | return NULL; |
1777 | return __io_req_find_next(req); | |
1778 | } | |
1779 | ||
2c32395d PB |
1780 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1781 | { | |
1782 | if (!ctx) | |
1783 | return; | |
1784 | if (ctx->submit_state.comp.nr) { | |
1785 | mutex_lock(&ctx->uring_lock); | |
1786 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1787 | mutex_unlock(&ctx->uring_lock); | |
1788 | } | |
1789 | percpu_ref_put(&ctx->refs); | |
1790 | } | |
1791 | ||
7cbf1722 | 1792 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1793 | { |
65453d1e | 1794 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1795 | struct io_wq_work_list list; |
1796 | struct io_wq_work_node *node; | |
c2c4c83c | 1797 | |
7cbf1722 JA |
1798 | if (wq_list_empty(&tctx->task_list)) |
1799 | return false; | |
6200b0ae | 1800 | |
0b81e80c | 1801 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1802 | list = tctx->task_list; |
1803 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1804 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1805 | |
7cbf1722 JA |
1806 | node = list.first; |
1807 | while (node) { | |
1808 | struct io_wq_work_node *next = node->next; | |
1809 | struct io_kiocb *req; | |
0ba9c9ed | 1810 | |
7cbf1722 | 1811 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1812 | if (req->ctx != ctx) { |
1813 | ctx_flush_and_put(ctx); | |
1814 | ctx = req->ctx; | |
1815 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1816 | } |
65453d1e | 1817 | |
2c32395d PB |
1818 | req->task_work.func(&req->task_work); |
1819 | node = next; | |
7cbf1722 JA |
1820 | } |
1821 | ||
2c32395d | 1822 | ctx_flush_and_put(ctx); |
7cbf1722 | 1823 | return list.first != NULL; |
c2c4c83c JA |
1824 | } |
1825 | ||
7cbf1722 | 1826 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1827 | { |
7cbf1722 | 1828 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1829 | |
1d5f360d JA |
1830 | clear_bit(0, &tctx->task_state); |
1831 | ||
7cbf1722 JA |
1832 | while (__tctx_task_work(tctx)) |
1833 | cond_resched(); | |
7cbf1722 JA |
1834 | } |
1835 | ||
1836 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1837 | enum task_work_notify_mode notify) | |
1838 | { | |
1839 | struct io_uring_task *tctx = tsk->io_uring; | |
1840 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1841 | unsigned long flags; |
7cbf1722 JA |
1842 | int ret; |
1843 | ||
1844 | WARN_ON_ONCE(!tctx); | |
1845 | ||
0b81e80c | 1846 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1847 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1848 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1849 | |
1850 | /* task_work already pending, we're done */ | |
1851 | if (test_bit(0, &tctx->task_state) || | |
1852 | test_and_set_bit(0, &tctx->task_state)) | |
1853 | return 0; | |
1854 | ||
1855 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1856 | return 0; | |
1857 | ||
1858 | /* | |
1859 | * Slow path - we failed, find and delete work. if the work is not | |
1860 | * in the list, it got run and we're fine. | |
1861 | */ | |
1862 | ret = 0; | |
0b81e80c | 1863 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1864 | wq_list_for_each(node, prev, &tctx->task_list) { |
1865 | if (&req->io_task_work.node == node) { | |
1866 | wq_list_del(&tctx->task_list, node, prev); | |
1867 | ret = 1; | |
1868 | break; | |
1869 | } | |
1870 | } | |
0b81e80c | 1871 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1872 | clear_bit(0, &tctx->task_state); |
1873 | return ret; | |
1874 | } | |
1875 | ||
355fb9e2 | 1876 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1877 | { |
1878 | struct task_struct *tsk = req->task; | |
1879 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1880 | enum task_work_notify_mode notify; |
1881 | int ret; | |
c2c4c83c | 1882 | |
6200b0ae JA |
1883 | if (tsk->flags & PF_EXITING) |
1884 | return -ESRCH; | |
1885 | ||
c2c4c83c | 1886 | /* |
0ba9c9ed JA |
1887 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1888 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1889 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1890 | * will do the job. | |
c2c4c83c | 1891 | */ |
91989c70 | 1892 | notify = TWA_NONE; |
355fb9e2 | 1893 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1894 | notify = TWA_SIGNAL; |
1895 | ||
7cbf1722 | 1896 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1897 | if (!ret) |
1898 | wake_up_process(tsk); | |
0ba9c9ed | 1899 | |
c2c4c83c JA |
1900 | return ret; |
1901 | } | |
1902 | ||
eab30c4d | 1903 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1904 | task_work_func_t cb) |
eab30c4d | 1905 | { |
7c25c0d1 JA |
1906 | struct io_ring_ctx *ctx = req->ctx; |
1907 | struct callback_head *head; | |
eab30c4d PB |
1908 | |
1909 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1910 | do { |
1911 | head = READ_ONCE(ctx->exit_task_work); | |
1912 | req->task_work.next = head; | |
1913 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1914 | } |
1915 | ||
c40f6379 JA |
1916 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1917 | { | |
1918 | struct io_ring_ctx *ctx = req->ctx; | |
1919 | ||
1920 | spin_lock_irq(&ctx->completion_lock); | |
1921 | io_cqring_fill_event(req, error); | |
1922 | io_commit_cqring(ctx); | |
1923 | spin_unlock_irq(&ctx->completion_lock); | |
1924 | ||
1925 | io_cqring_ev_posted(ctx); | |
1926 | req_set_fail_links(req); | |
1927 | io_double_put_req(req); | |
1928 | } | |
1929 | ||
1930 | static void io_req_task_cancel(struct callback_head *cb) | |
1931 | { | |
1932 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1933 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1934 | |
792bb6eb | 1935 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1936 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1937 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1938 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1939 | } |
1940 | ||
1941 | static void __io_req_task_submit(struct io_kiocb *req) | |
1942 | { | |
1943 | struct io_ring_ctx *ctx = req->ctx; | |
1944 | ||
04fc6c80 | 1945 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1946 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 1947 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1948 | __io_queue_sqe(req); |
81b6d05c | 1949 | else |
c40f6379 | 1950 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1951 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1952 | } |
1953 | ||
1954 | static void io_req_task_submit(struct callback_head *cb) | |
1955 | { | |
1956 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1957 | ||
1958 | __io_req_task_submit(req); | |
1959 | } | |
1960 | ||
1961 | static void io_req_task_queue(struct io_kiocb *req) | |
1962 | { | |
c40f6379 JA |
1963 | int ret; |
1964 | ||
7cbf1722 | 1965 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1966 | ret = io_req_task_work_add(req); |
c40f6379 | 1967 | if (unlikely(ret)) { |
a3df7698 | 1968 | req->result = -ECANCELED; |
04fc6c80 | 1969 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1970 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1971 | } |
c40f6379 JA |
1972 | } |
1973 | ||
a3df7698 PB |
1974 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
1975 | { | |
1976 | percpu_ref_get(&req->ctx->refs); | |
1977 | req->result = ret; | |
1978 | req->task_work.func = io_req_task_cancel; | |
1979 | ||
1980 | if (unlikely(io_req_task_work_add(req))) | |
1981 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
1982 | } | |
1983 | ||
f2f87370 | 1984 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 1985 | { |
9b5f7bd9 | 1986 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
1987 | |
1988 | if (nxt) | |
906a8c3f | 1989 | io_req_task_queue(nxt); |
c69f8dbe JL |
1990 | } |
1991 | ||
c3524383 | 1992 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 1993 | { |
c3524383 PB |
1994 | io_queue_next(req); |
1995 | __io_free_req(req); | |
1996 | } | |
8766dd51 | 1997 | |
2d6500d4 | 1998 | struct req_batch { |
5af1d13e PB |
1999 | struct task_struct *task; |
2000 | int task_refs; | |
1b4c351f | 2001 | int ctx_refs; |
2d6500d4 PB |
2002 | }; |
2003 | ||
5af1d13e PB |
2004 | static inline void io_init_req_batch(struct req_batch *rb) |
2005 | { | |
5af1d13e | 2006 | rb->task_refs = 0; |
9ae72463 | 2007 | rb->ctx_refs = 0; |
5af1d13e PB |
2008 | rb->task = NULL; |
2009 | } | |
2010 | ||
2d6500d4 PB |
2011 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2012 | struct req_batch *rb) | |
2013 | { | |
6e833d53 | 2014 | if (rb->task) |
7c660731 | 2015 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2016 | if (rb->ctx_refs) |
2017 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2018 | } |
2019 | ||
6ff119a6 PB |
2020 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2021 | struct io_submit_state *state) | |
2d6500d4 | 2022 | { |
f2f87370 | 2023 | io_queue_next(req); |
2d6500d4 | 2024 | |
e3bc8e9d | 2025 | if (req->task != rb->task) { |
7c660731 PB |
2026 | if (rb->task) |
2027 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2028 | rb->task = req->task; |
2029 | rb->task_refs = 0; | |
5af1d13e | 2030 | } |
e3bc8e9d | 2031 | rb->task_refs++; |
9ae72463 | 2032 | rb->ctx_refs++; |
5af1d13e | 2033 | |
4edf20f9 | 2034 | io_dismantle_req(req); |
bd759045 | 2035 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2036 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2037 | else |
2038 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2039 | } |
2040 | ||
905c172f PB |
2041 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2042 | struct io_ring_ctx *ctx) | |
2043 | { | |
2044 | int i, nr = cs->nr; | |
2045 | struct io_kiocb *req; | |
2046 | struct req_batch rb; | |
2047 | ||
2048 | io_init_req_batch(&rb); | |
2049 | spin_lock_irq(&ctx->completion_lock); | |
2050 | for (i = 0; i < nr; i++) { | |
2051 | req = cs->reqs[i]; | |
2052 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2053 | } | |
2054 | io_commit_cqring(ctx); | |
2055 | spin_unlock_irq(&ctx->completion_lock); | |
2056 | ||
2057 | io_cqring_ev_posted(ctx); | |
2058 | for (i = 0; i < nr; i++) { | |
2059 | req = cs->reqs[i]; | |
2060 | ||
2061 | /* submission and completion refs */ | |
2062 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2063 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2064 | } |
2065 | ||
2066 | io_req_free_batch_finish(ctx, &rb); | |
2067 | cs->nr = 0; | |
7a743e22 PB |
2068 | } |
2069 | ||
ba816ad6 JA |
2070 | /* |
2071 | * Drop reference to request, return next in chain (if there is one) if this | |
2072 | * was the last reference to this request. | |
2073 | */ | |
9b5f7bd9 | 2074 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2075 | { |
9b5f7bd9 PB |
2076 | struct io_kiocb *nxt = NULL; |
2077 | ||
2a44f467 | 2078 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2079 | nxt = io_req_find_next(req); |
4d7dd462 | 2080 | __io_free_req(req); |
2a44f467 | 2081 | } |
9b5f7bd9 | 2082 | return nxt; |
2b188cc1 JA |
2083 | } |
2084 | ||
e65ef56d JA |
2085 | static void io_put_req(struct io_kiocb *req) |
2086 | { | |
2087 | if (refcount_dec_and_test(&req->refs)) | |
2088 | io_free_req(req); | |
2b188cc1 JA |
2089 | } |
2090 | ||
216578e5 PB |
2091 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2092 | { | |
2093 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2094 | ||
2095 | io_free_req(req); | |
2096 | } | |
2097 | ||
2098 | static void io_free_req_deferred(struct io_kiocb *req) | |
2099 | { | |
2100 | int ret; | |
2101 | ||
7cbf1722 | 2102 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2103 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2104 | if (unlikely(ret)) |
2105 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2106 | } |
2107 | ||
2108 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2109 | { | |
2110 | if (refcount_sub_and_test(refs, &req->refs)) | |
2111 | io_free_req_deferred(req); | |
2112 | } | |
2113 | ||
978db57e JA |
2114 | static void io_double_put_req(struct io_kiocb *req) |
2115 | { | |
2116 | /* drop both submit and complete references */ | |
2117 | if (refcount_sub_and_test(2, &req->refs)) | |
2118 | io_free_req(req); | |
2119 | } | |
2120 | ||
6c503150 | 2121 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2122 | { |
2123 | /* See comment at the top of this file */ | |
2124 | smp_rmb(); | |
e23de15f | 2125 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2126 | } |
2127 | ||
fb5ccc98 PB |
2128 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2129 | { | |
2130 | struct io_rings *rings = ctx->rings; | |
2131 | ||
2132 | /* make sure SQ entry isn't read before tail */ | |
2133 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2134 | } | |
2135 | ||
8ff069bf | 2136 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2137 | { |
8ff069bf | 2138 | unsigned int cflags; |
e94f141b | 2139 | |
bcda7baa JA |
2140 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2141 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2142 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2143 | kfree(kbuf); |
2144 | return cflags; | |
e94f141b JA |
2145 | } |
2146 | ||
8ff069bf | 2147 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2148 | { |
4d954c25 | 2149 | struct io_buffer *kbuf; |
bcda7baa | 2150 | |
4d954c25 | 2151 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2152 | return io_put_kbuf(req, kbuf); |
2153 | } | |
2154 | ||
4c6e277c JA |
2155 | static inline bool io_run_task_work(void) |
2156 | { | |
6200b0ae JA |
2157 | /* |
2158 | * Not safe to run on exiting task, and the task_work handling will | |
2159 | * not add work to such a task. | |
2160 | */ | |
2161 | if (unlikely(current->flags & PF_EXITING)) | |
2162 | return false; | |
4c6e277c JA |
2163 | if (current->task_works) { |
2164 | __set_current_state(TASK_RUNNING); | |
2165 | task_work_run(); | |
2166 | return true; | |
2167 | } | |
2168 | ||
2169 | return false; | |
bcda7baa JA |
2170 | } |
2171 | ||
def596e9 JA |
2172 | /* |
2173 | * Find and free completed poll iocbs | |
2174 | */ | |
2175 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2176 | struct list_head *done) | |
2177 | { | |
8237e045 | 2178 | struct req_batch rb; |
def596e9 | 2179 | struct io_kiocb *req; |
bbde017a XW |
2180 | |
2181 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2182 | smp_rmb(); | |
def596e9 | 2183 | |
5af1d13e | 2184 | io_init_req_batch(&rb); |
def596e9 | 2185 | while (!list_empty(done)) { |
bcda7baa JA |
2186 | int cflags = 0; |
2187 | ||
d21ffe7e | 2188 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2189 | list_del(&req->inflight_entry); |
2190 | ||
bbde017a XW |
2191 | if (READ_ONCE(req->result) == -EAGAIN) { |
2192 | req->iopoll_completed = 0; | |
23faba36 | 2193 | if (io_rw_reissue(req)) |
f161340d | 2194 | continue; |
bbde017a | 2195 | } |
def596e9 | 2196 | |
bcda7baa | 2197 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2198 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2199 | |
2200 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2201 | (*nr_events)++; |
2202 | ||
c3524383 | 2203 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2204 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2205 | } |
def596e9 | 2206 | |
09bb8394 | 2207 | io_commit_cqring(ctx); |
80c18e4a | 2208 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2209 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2210 | } |
2211 | ||
def596e9 JA |
2212 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2213 | long min) | |
2214 | { | |
2215 | struct io_kiocb *req, *tmp; | |
2216 | LIST_HEAD(done); | |
2217 | bool spin; | |
2218 | int ret; | |
2219 | ||
2220 | /* | |
2221 | * Only spin for completions if we don't have multiple devices hanging | |
2222 | * off our complete list, and we're under the requested amount. | |
2223 | */ | |
2224 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2225 | ||
2226 | ret = 0; | |
d21ffe7e | 2227 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2228 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2229 | |
2230 | /* | |
581f9810 BM |
2231 | * Move completed and retryable entries to our local lists. |
2232 | * If we find a request that requires polling, break out | |
2233 | * and complete those lists first, if we have entries there. | |
def596e9 | 2234 | */ |
65a6543d | 2235 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2236 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2237 | continue; |
2238 | } | |
2239 | if (!list_empty(&done)) | |
2240 | break; | |
2241 | ||
2242 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2243 | if (ret < 0) | |
2244 | break; | |
2245 | ||
3aadc23e PB |
2246 | /* iopoll may have completed current req */ |
2247 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2248 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2249 | |
def596e9 JA |
2250 | if (ret && spin) |
2251 | spin = false; | |
2252 | ret = 0; | |
2253 | } | |
2254 | ||
2255 | if (!list_empty(&done)) | |
2256 | io_iopoll_complete(ctx, nr_events, &done); | |
2257 | ||
2258 | return ret; | |
2259 | } | |
2260 | ||
2261 | /* | |
d195a66e | 2262 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2263 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2264 | * as a non-spinning completion check. | |
2265 | */ | |
2266 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2267 | long min) | |
2268 | { | |
540e32a0 | 2269 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2270 | int ret; |
2271 | ||
2272 | ret = io_do_iopoll(ctx, nr_events, min); | |
2273 | if (ret < 0) | |
2274 | return ret; | |
eba0a4dd | 2275 | if (*nr_events >= min) |
def596e9 JA |
2276 | return 0; |
2277 | } | |
2278 | ||
2279 | return 1; | |
2280 | } | |
2281 | ||
2282 | /* | |
2283 | * We can't just wait for polled events to come to us, we have to actively | |
2284 | * find and complete them. | |
2285 | */ | |
b2edc0a7 | 2286 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2287 | { |
2288 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2289 | return; | |
2290 | ||
2291 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2292 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2293 | unsigned int nr_events = 0; |
2294 | ||
b2edc0a7 | 2295 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2296 | |
b2edc0a7 PB |
2297 | /* let it sleep and repeat later if can't complete a request */ |
2298 | if (nr_events == 0) | |
2299 | break; | |
08f5439f JA |
2300 | /* |
2301 | * Ensure we allow local-to-the-cpu processing to take place, | |
2302 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2303 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2304 | */ |
3fcee5a6 PB |
2305 | if (need_resched()) { |
2306 | mutex_unlock(&ctx->uring_lock); | |
2307 | cond_resched(); | |
2308 | mutex_lock(&ctx->uring_lock); | |
2309 | } | |
def596e9 JA |
2310 | } |
2311 | mutex_unlock(&ctx->uring_lock); | |
2312 | } | |
2313 | ||
7668b92a | 2314 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2315 | { |
7668b92a | 2316 | unsigned int nr_events = 0; |
2b2ed975 | 2317 | int iters = 0, ret = 0; |
500f9fba | 2318 | |
c7849be9 XW |
2319 | /* |
2320 | * We disallow the app entering submit/complete with polling, but we | |
2321 | * still need to lock the ring to prevent racing with polled issue | |
2322 | * that got punted to a workqueue. | |
2323 | */ | |
2324 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2325 | do { |
a3a0e43f JA |
2326 | /* |
2327 | * Don't enter poll loop if we already have events pending. | |
2328 | * If we do, we can potentially be spinning for commands that | |
2329 | * already triggered a CQE (eg in error). | |
2330 | */ | |
6c503150 PB |
2331 | if (test_bit(0, &ctx->cq_check_overflow)) |
2332 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2333 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2334 | break; |
2335 | ||
500f9fba JA |
2336 | /* |
2337 | * If a submit got punted to a workqueue, we can have the | |
2338 | * application entering polling for a command before it gets | |
2339 | * issued. That app will hold the uring_lock for the duration | |
2340 | * of the poll right here, so we need to take a breather every | |
2341 | * now and then to ensure that the issue has a chance to add | |
2342 | * the poll to the issued list. Otherwise we can spin here | |
2343 | * forever, while the workqueue is stuck trying to acquire the | |
2344 | * very same mutex. | |
2345 | */ | |
2346 | if (!(++iters & 7)) { | |
2347 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2348 | io_run_task_work(); |
500f9fba JA |
2349 | mutex_lock(&ctx->uring_lock); |
2350 | } | |
2351 | ||
7668b92a | 2352 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2353 | if (ret <= 0) |
2354 | break; | |
2355 | ret = 0; | |
7668b92a | 2356 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2357 | |
500f9fba | 2358 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2359 | return ret; |
2360 | } | |
2361 | ||
491381ce | 2362 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2363 | { |
491381ce JA |
2364 | /* |
2365 | * Tell lockdep we inherited freeze protection from submission | |
2366 | * thread. | |
2367 | */ | |
2368 | if (req->flags & REQ_F_ISREG) { | |
2369 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2370 | |
491381ce | 2371 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2372 | } |
491381ce | 2373 | file_end_write(req->file); |
2b188cc1 JA |
2374 | } |
2375 | ||
b63534c4 | 2376 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2377 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2378 | { |
2379 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2380 | int rw, ret; |
b63534c4 | 2381 | struct iov_iter iter; |
b63534c4 | 2382 | |
dc2a6e9a PB |
2383 | /* already prepared */ |
2384 | if (req->async_data) | |
2385 | return true; | |
b63534c4 JA |
2386 | |
2387 | switch (req->opcode) { | |
2388 | case IORING_OP_READV: | |
2389 | case IORING_OP_READ_FIXED: | |
2390 | case IORING_OP_READ: | |
2391 | rw = READ; | |
2392 | break; | |
2393 | case IORING_OP_WRITEV: | |
2394 | case IORING_OP_WRITE_FIXED: | |
2395 | case IORING_OP_WRITE: | |
2396 | rw = WRITE; | |
2397 | break; | |
2398 | default: | |
2399 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2400 | req->opcode); | |
dc2a6e9a | 2401 | return false; |
b63534c4 JA |
2402 | } |
2403 | ||
dc2a6e9a PB |
2404 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2405 | if (ret < 0) | |
2406 | return false; | |
6bf985dc | 2407 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2408 | } |
b63534c4 | 2409 | |
3e6a0d3c | 2410 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2411 | { |
355afaeb | 2412 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2413 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2414 | |
355afaeb JA |
2415 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2416 | return false; | |
3e6a0d3c JA |
2417 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2418 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2419 | return false; |
7c977a58 JA |
2420 | /* |
2421 | * If ref is dying, we might be running poll reap from the exit work. | |
2422 | * Don't attempt to reissue from that path, just let it fail with | |
2423 | * -EAGAIN. | |
2424 | */ | |
3e6a0d3c JA |
2425 | if (percpu_ref_is_dying(&ctx->refs)) |
2426 | return false; | |
2427 | return true; | |
2428 | } | |
2429 | #endif | |
2430 | ||
2431 | static bool io_rw_reissue(struct io_kiocb *req) | |
2432 | { | |
2433 | #ifdef CONFIG_BLOCK | |
2434 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2435 | return false; |
b63534c4 | 2436 | |
55e6ac1e PB |
2437 | lockdep_assert_held(&req->ctx->uring_lock); |
2438 | ||
37d1e2e3 | 2439 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2440 | refcount_inc(&req->refs); |
2441 | io_queue_async_work(req); | |
b63534c4 | 2442 | return true; |
fdee946d | 2443 | } |
dc2a6e9a | 2444 | req_set_fail_links(req); |
b63534c4 JA |
2445 | #endif |
2446 | return false; | |
2447 | } | |
2448 | ||
a1d7c393 | 2449 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2450 | unsigned int issue_flags) |
a1d7c393 | 2451 | { |
2f8e45f1 PB |
2452 | int cflags = 0; |
2453 | ||
23faba36 PB |
2454 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2455 | return; | |
2f8e45f1 PB |
2456 | if (res != req->result) |
2457 | req_set_fail_links(req); | |
23faba36 | 2458 | |
2f8e45f1 PB |
2459 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2460 | kiocb_end_write(req); | |
2461 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2462 | cflags = io_put_rw_kbuf(req); | |
2463 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2464 | } |
2465 | ||
2466 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2467 | { | |
9adbd45d | 2468 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2469 | |
889fca73 | 2470 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2471 | } |
2472 | ||
def596e9 JA |
2473 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2474 | { | |
9adbd45d | 2475 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2476 | |
3e6a0d3c JA |
2477 | #ifdef CONFIG_BLOCK |
2478 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2479 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2480 | struct io_async_rw *rw = req->async_data; | |
2481 | ||
2482 | if (rw) | |
2483 | iov_iter_revert(&rw->iter, | |
2484 | req->result - iov_iter_count(&rw->iter)); | |
2485 | else if (!io_resubmit_prep(req)) | |
2486 | res = -EIO; | |
2487 | } | |
2488 | #endif | |
2489 | ||
491381ce JA |
2490 | if (kiocb->ki_flags & IOCB_WRITE) |
2491 | kiocb_end_write(req); | |
def596e9 | 2492 | |
2d7d6792 | 2493 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2494 | req_set_fail_links(req); |
bbde017a XW |
2495 | |
2496 | WRITE_ONCE(req->result, res); | |
2497 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2498 | smp_wmb(); |
2499 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2500 | } |
2501 | ||
2502 | /* | |
2503 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2504 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2505 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2506 | * accessing the kiocb cookie. | |
2507 | */ | |
2e9dbe90 | 2508 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2509 | { |
2510 | struct io_ring_ctx *ctx = req->ctx; | |
2511 | ||
2512 | /* | |
2513 | * Track whether we have multiple files in our lists. This will impact | |
2514 | * how we do polling eventually, not spinning if we're on potentially | |
2515 | * different devices. | |
2516 | */ | |
540e32a0 | 2517 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2518 | ctx->poll_multi_file = false; |
2519 | } else if (!ctx->poll_multi_file) { | |
2520 | struct io_kiocb *list_req; | |
2521 | ||
540e32a0 | 2522 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2523 | inflight_entry); |
9adbd45d | 2524 | if (list_req->file != req->file) |
def596e9 JA |
2525 | ctx->poll_multi_file = true; |
2526 | } | |
2527 | ||
2528 | /* | |
2529 | * For fast devices, IO may have already completed. If it has, add | |
2530 | * it to the front so we find it first. | |
2531 | */ | |
65a6543d | 2532 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2533 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2534 | else |
d21ffe7e | 2535 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2536 | |
2e9dbe90 XW |
2537 | /* |
2538 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2539 | * task context or in io worker task context. If current task context is | |
2540 | * sq thread, we don't need to check whether should wake up sq thread. | |
2541 | */ | |
2542 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2543 | wq_has_sleeper(&ctx->sq_data->wait)) |
2544 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2545 | } |
2546 | ||
9f13c35b PB |
2547 | static inline void io_state_file_put(struct io_submit_state *state) |
2548 | { | |
02b23a9a PB |
2549 | if (state->file_refs) { |
2550 | fput_many(state->file, state->file_refs); | |
2551 | state->file_refs = 0; | |
2552 | } | |
9a56a232 JA |
2553 | } |
2554 | ||
2555 | /* | |
2556 | * Get as many references to a file as we have IOs left in this submission, | |
2557 | * assuming most submissions are for one file, or at least that each file | |
2558 | * has more than one submission. | |
2559 | */ | |
8da11c19 | 2560 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2561 | { |
2562 | if (!state) | |
2563 | return fget(fd); | |
2564 | ||
6e1271e6 | 2565 | if (state->file_refs) { |
9a56a232 | 2566 | if (state->fd == fd) { |
6e1271e6 | 2567 | state->file_refs--; |
9a56a232 JA |
2568 | return state->file; |
2569 | } | |
02b23a9a | 2570 | io_state_file_put(state); |
9a56a232 JA |
2571 | } |
2572 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2573 | if (unlikely(!state->file)) |
9a56a232 JA |
2574 | return NULL; |
2575 | ||
2576 | state->fd = fd; | |
6e1271e6 | 2577 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2578 | return state->file; |
2579 | } | |
2580 | ||
4503b767 JA |
2581 | static bool io_bdev_nowait(struct block_device *bdev) |
2582 | { | |
9ba0d0c8 | 2583 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2584 | } |
2585 | ||
2b188cc1 JA |
2586 | /* |
2587 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2588 | * any file. For now, just ensure that anything potentially problematic is done | |
2589 | * inline. | |
2590 | */ | |
af197f50 | 2591 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2592 | { |
2593 | umode_t mode = file_inode(file)->i_mode; | |
2594 | ||
4503b767 | 2595 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2596 | if (IS_ENABLED(CONFIG_BLOCK) && |
2597 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2598 | return true; |
2599 | return false; | |
2600 | } | |
2601 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2602 | return true; |
4503b767 | 2603 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2604 | if (IS_ENABLED(CONFIG_BLOCK) && |
2605 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2606 | file->f_op != &io_uring_fops) |
2607 | return true; | |
2608 | return false; | |
2609 | } | |
2b188cc1 | 2610 | |
c5b85625 JA |
2611 | /* any ->read/write should understand O_NONBLOCK */ |
2612 | if (file->f_flags & O_NONBLOCK) | |
2613 | return true; | |
2614 | ||
af197f50 JA |
2615 | if (!(file->f_mode & FMODE_NOWAIT)) |
2616 | return false; | |
2617 | ||
2618 | if (rw == READ) | |
2619 | return file->f_op->read_iter != NULL; | |
2620 | ||
2621 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2622 | } |
2623 | ||
a88fc400 | 2624 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2625 | { |
def596e9 | 2626 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2627 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2628 | struct file *file = req->file; |
09bb8394 JA |
2629 | unsigned ioprio; |
2630 | int ret; | |
2b188cc1 | 2631 | |
75c668cd | 2632 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2633 | req->flags |= REQ_F_ISREG; |
2634 | ||
2b188cc1 | 2635 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2636 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2637 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2638 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2639 | } |
2b188cc1 | 2640 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2641 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2642 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2643 | if (unlikely(ret)) | |
2644 | return ret; | |
2b188cc1 | 2645 | |
75c668cd PB |
2646 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2647 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2648 | req->flags |= REQ_F_NOWAIT; | |
2649 | ||
2b188cc1 JA |
2650 | ioprio = READ_ONCE(sqe->ioprio); |
2651 | if (ioprio) { | |
2652 | ret = ioprio_check_cap(ioprio); | |
2653 | if (ret) | |
09bb8394 | 2654 | return ret; |
2b188cc1 JA |
2655 | |
2656 | kiocb->ki_ioprio = ioprio; | |
2657 | } else | |
2658 | kiocb->ki_ioprio = get_current_ioprio(); | |
2659 | ||
def596e9 | 2660 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2661 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2662 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2663 | return -EOPNOTSUPP; |
2b188cc1 | 2664 | |
def596e9 JA |
2665 | kiocb->ki_flags |= IOCB_HIPRI; |
2666 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2667 | req->iopoll_completed = 0; |
def596e9 | 2668 | } else { |
09bb8394 JA |
2669 | if (kiocb->ki_flags & IOCB_HIPRI) |
2670 | return -EINVAL; | |
def596e9 JA |
2671 | kiocb->ki_complete = io_complete_rw; |
2672 | } | |
9adbd45d | 2673 | |
3529d8c2 JA |
2674 | req->rw.addr = READ_ONCE(sqe->addr); |
2675 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2676 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2677 | return 0; |
2b188cc1 JA |
2678 | } |
2679 | ||
2680 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2681 | { | |
2682 | switch (ret) { | |
2683 | case -EIOCBQUEUED: | |
2684 | break; | |
2685 | case -ERESTARTSYS: | |
2686 | case -ERESTARTNOINTR: | |
2687 | case -ERESTARTNOHAND: | |
2688 | case -ERESTART_RESTARTBLOCK: | |
2689 | /* | |
2690 | * We can't just restart the syscall, since previously | |
2691 | * submitted sqes may already be in progress. Just fail this | |
2692 | * IO with EINTR. | |
2693 | */ | |
2694 | ret = -EINTR; | |
df561f66 | 2695 | fallthrough; |
2b188cc1 JA |
2696 | default: |
2697 | kiocb->ki_complete(kiocb, ret, 0); | |
2698 | } | |
2699 | } | |
2700 | ||
a1d7c393 | 2701 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2702 | unsigned int issue_flags) |
ba816ad6 | 2703 | { |
ba04291e | 2704 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2705 | struct io_async_rw *io = req->async_data; |
ba04291e | 2706 | |
227c0c96 | 2707 | /* add previously done IO, if any */ |
e8c2bc1f | 2708 | if (io && io->bytes_done > 0) { |
227c0c96 | 2709 | if (ret < 0) |
e8c2bc1f | 2710 | ret = io->bytes_done; |
227c0c96 | 2711 | else |
e8c2bc1f | 2712 | ret += io->bytes_done; |
227c0c96 JA |
2713 | } |
2714 | ||
ba04291e JA |
2715 | if (req->flags & REQ_F_CUR_POS) |
2716 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2717 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2718 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2719 | else |
2720 | io_rw_done(kiocb, ret); | |
2721 | } | |
2722 | ||
847595de | 2723 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2724 | { |
9adbd45d JA |
2725 | struct io_ring_ctx *ctx = req->ctx; |
2726 | size_t len = req->rw.len; | |
edafccee | 2727 | struct io_mapped_ubuf *imu; |
4be1c615 | 2728 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2729 | size_t offset; |
2730 | u64 buf_addr; | |
2731 | ||
edafccee JA |
2732 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2733 | return -EFAULT; | |
edafccee JA |
2734 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2735 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2736 | buf_addr = req->rw.addr; |
edafccee JA |
2737 | |
2738 | /* overflow */ | |
2739 | if (buf_addr + len < buf_addr) | |
2740 | return -EFAULT; | |
2741 | /* not inside the mapped region */ | |
2742 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2743 | return -EFAULT; | |
2744 | ||
2745 | /* | |
2746 | * May not be a start of buffer, set size appropriately | |
2747 | * and advance us to the beginning. | |
2748 | */ | |
2749 | offset = buf_addr - imu->ubuf; | |
2750 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2751 | |
2752 | if (offset) { | |
2753 | /* | |
2754 | * Don't use iov_iter_advance() here, as it's really slow for | |
2755 | * using the latter parts of a big fixed buffer - it iterates | |
2756 | * over each segment manually. We can cheat a bit here, because | |
2757 | * we know that: | |
2758 | * | |
2759 | * 1) it's a BVEC iter, we set it up | |
2760 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2761 | * first and last bvec | |
2762 | * | |
2763 | * So just find our index, and adjust the iterator afterwards. | |
2764 | * If the offset is within the first bvec (or the whole first | |
2765 | * bvec, just use iov_iter_advance(). This makes it easier | |
2766 | * since we can just skip the first segment, which may not | |
2767 | * be PAGE_SIZE aligned. | |
2768 | */ | |
2769 | const struct bio_vec *bvec = imu->bvec; | |
2770 | ||
2771 | if (offset <= bvec->bv_len) { | |
2772 | iov_iter_advance(iter, offset); | |
2773 | } else { | |
2774 | unsigned long seg_skip; | |
2775 | ||
2776 | /* skip first vec */ | |
2777 | offset -= bvec->bv_len; | |
2778 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2779 | ||
2780 | iter->bvec = bvec + seg_skip; | |
2781 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2782 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2783 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2784 | } |
2785 | } | |
2786 | ||
847595de | 2787 | return 0; |
edafccee JA |
2788 | } |
2789 | ||
bcda7baa JA |
2790 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2791 | { | |
2792 | if (needs_lock) | |
2793 | mutex_unlock(&ctx->uring_lock); | |
2794 | } | |
2795 | ||
2796 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2797 | { | |
2798 | /* | |
2799 | * "Normal" inline submissions always hold the uring_lock, since we | |
2800 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2801 | * The only exception is when we've detached the request and issue it | |
2802 | * from an async worker thread, grab the lock for that case. | |
2803 | */ | |
2804 | if (needs_lock) | |
2805 | mutex_lock(&ctx->uring_lock); | |
2806 | } | |
2807 | ||
2808 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2809 | int bgid, struct io_buffer *kbuf, | |
2810 | bool needs_lock) | |
2811 | { | |
2812 | struct io_buffer *head; | |
2813 | ||
2814 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2815 | return kbuf; | |
2816 | ||
2817 | io_ring_submit_lock(req->ctx, needs_lock); | |
2818 | ||
2819 | lockdep_assert_held(&req->ctx->uring_lock); | |
2820 | ||
2821 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2822 | if (head) { | |
2823 | if (!list_empty(&head->list)) { | |
2824 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2825 | list); | |
2826 | list_del(&kbuf->list); | |
2827 | } else { | |
2828 | kbuf = head; | |
2829 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2830 | } | |
2831 | if (*len > kbuf->len) | |
2832 | *len = kbuf->len; | |
2833 | } else { | |
2834 | kbuf = ERR_PTR(-ENOBUFS); | |
2835 | } | |
2836 | ||
2837 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2838 | ||
2839 | return kbuf; | |
2840 | } | |
2841 | ||
4d954c25 JA |
2842 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2843 | bool needs_lock) | |
2844 | { | |
2845 | struct io_buffer *kbuf; | |
4f4eeba8 | 2846 | u16 bgid; |
4d954c25 JA |
2847 | |
2848 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2849 | bgid = req->buf_index; |
4d954c25 JA |
2850 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2851 | if (IS_ERR(kbuf)) | |
2852 | return kbuf; | |
2853 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2854 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2855 | return u64_to_user_ptr(kbuf->addr); | |
2856 | } | |
2857 | ||
2858 | #ifdef CONFIG_COMPAT | |
2859 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2860 | bool needs_lock) | |
2861 | { | |
2862 | struct compat_iovec __user *uiov; | |
2863 | compat_ssize_t clen; | |
2864 | void __user *buf; | |
2865 | ssize_t len; | |
2866 | ||
2867 | uiov = u64_to_user_ptr(req->rw.addr); | |
2868 | if (!access_ok(uiov, sizeof(*uiov))) | |
2869 | return -EFAULT; | |
2870 | if (__get_user(clen, &uiov->iov_len)) | |
2871 | return -EFAULT; | |
2872 | if (clen < 0) | |
2873 | return -EINVAL; | |
2874 | ||
2875 | len = clen; | |
2876 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2877 | if (IS_ERR(buf)) | |
2878 | return PTR_ERR(buf); | |
2879 | iov[0].iov_base = buf; | |
2880 | iov[0].iov_len = (compat_size_t) len; | |
2881 | return 0; | |
2882 | } | |
2883 | #endif | |
2884 | ||
2885 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2886 | bool needs_lock) | |
2887 | { | |
2888 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2889 | void __user *buf; | |
2890 | ssize_t len; | |
2891 | ||
2892 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2893 | return -EFAULT; | |
2894 | ||
2895 | len = iov[0].iov_len; | |
2896 | if (len < 0) | |
2897 | return -EINVAL; | |
2898 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2899 | if (IS_ERR(buf)) | |
2900 | return PTR_ERR(buf); | |
2901 | iov[0].iov_base = buf; | |
2902 | iov[0].iov_len = len; | |
2903 | return 0; | |
2904 | } | |
2905 | ||
2906 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2907 | bool needs_lock) | |
2908 | { | |
dddb3e26 JA |
2909 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2910 | struct io_buffer *kbuf; | |
2911 | ||
2912 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2913 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2914 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2915 | return 0; |
dddb3e26 | 2916 | } |
dd201662 | 2917 | if (req->rw.len != 1) |
4d954c25 JA |
2918 | return -EINVAL; |
2919 | ||
2920 | #ifdef CONFIG_COMPAT | |
2921 | if (req->ctx->compat) | |
2922 | return io_compat_import(req, iov, needs_lock); | |
2923 | #endif | |
2924 | ||
2925 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2926 | } | |
2927 | ||
847595de PB |
2928 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2929 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2930 | { |
9adbd45d JA |
2931 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2932 | size_t sqe_len = req->rw.len; | |
847595de | 2933 | u8 opcode = req->opcode; |
4d954c25 | 2934 | ssize_t ret; |
edafccee | 2935 | |
7d009165 | 2936 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2937 | *iovec = NULL; |
9adbd45d | 2938 | return io_import_fixed(req, rw, iter); |
edafccee | 2939 | } |
2b188cc1 | 2940 | |
bcda7baa | 2941 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2942 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2943 | return -EINVAL; |
2944 | ||
3a6820f2 | 2945 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2946 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2947 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2948 | if (IS_ERR(buf)) |
4d954c25 | 2949 | return PTR_ERR(buf); |
3f9d6441 | 2950 | req->rw.len = sqe_len; |
bcda7baa JA |
2951 | } |
2952 | ||
3a6820f2 JA |
2953 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2954 | *iovec = NULL; | |
10fc72e4 | 2955 | return ret; |
3a6820f2 JA |
2956 | } |
2957 | ||
4d954c25 JA |
2958 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2959 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2960 | if (!ret) |
2961 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2962 | *iovec = NULL; |
2963 | return ret; | |
2964 | } | |
2965 | ||
89cd35c5 CH |
2966 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2967 | req->ctx->compat); | |
2b188cc1 JA |
2968 | } |
2969 | ||
0fef9483 JA |
2970 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2971 | { | |
5b09e37e | 2972 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
2973 | } |
2974 | ||
31b51510 | 2975 | /* |
32960613 JA |
2976 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2977 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2978 | */ |
4017eb91 | 2979 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 2980 | { |
4017eb91 JA |
2981 | struct kiocb *kiocb = &req->rw.kiocb; |
2982 | struct file *file = req->file; | |
32960613 JA |
2983 | ssize_t ret = 0; |
2984 | ||
2985 | /* | |
2986 | * Don't support polled IO through this interface, and we can't | |
2987 | * support non-blocking either. For the latter, this just causes | |
2988 | * the kiocb to be handled from an async context. | |
2989 | */ | |
2990 | if (kiocb->ki_flags & IOCB_HIPRI) | |
2991 | return -EOPNOTSUPP; | |
2992 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
2993 | return -EAGAIN; | |
2994 | ||
2995 | while (iov_iter_count(iter)) { | |
311ae9e1 | 2996 | struct iovec iovec; |
32960613 JA |
2997 | ssize_t nr; |
2998 | ||
311ae9e1 PB |
2999 | if (!iov_iter_is_bvec(iter)) { |
3000 | iovec = iov_iter_iovec(iter); | |
3001 | } else { | |
4017eb91 JA |
3002 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3003 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3004 | } |
3005 | ||
32960613 JA |
3006 | if (rw == READ) { |
3007 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3008 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3009 | } else { |
3010 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3011 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3012 | } |
3013 | ||
3014 | if (nr < 0) { | |
3015 | if (!ret) | |
3016 | ret = nr; | |
3017 | break; | |
3018 | } | |
3019 | ret += nr; | |
3020 | if (nr != iovec.iov_len) | |
3021 | break; | |
4017eb91 JA |
3022 | req->rw.len -= nr; |
3023 | req->rw.addr += nr; | |
32960613 JA |
3024 | iov_iter_advance(iter, nr); |
3025 | } | |
3026 | ||
3027 | return ret; | |
3028 | } | |
3029 | ||
ff6165b2 JA |
3030 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3031 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3032 | { |
e8c2bc1f | 3033 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3034 | |
ff6165b2 | 3035 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3036 | rw->free_iovec = iovec; |
227c0c96 | 3037 | rw->bytes_done = 0; |
ff6165b2 | 3038 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3039 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3040 | return; |
b64e3444 | 3041 | if (!iovec) { |
ff6165b2 JA |
3042 | unsigned iov_off = 0; |
3043 | ||
3044 | rw->iter.iov = rw->fast_iov; | |
3045 | if (iter->iov != fast_iov) { | |
3046 | iov_off = iter->iov - fast_iov; | |
3047 | rw->iter.iov += iov_off; | |
3048 | } | |
3049 | if (rw->fast_iov != fast_iov) | |
3050 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3051 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3052 | } else { |
3053 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3054 | } |
3055 | } | |
3056 | ||
e8c2bc1f | 3057 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3058 | { |
e8c2bc1f JA |
3059 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3060 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3061 | return req->async_data == NULL; | |
3d9932a8 XW |
3062 | } |
3063 | ||
e8c2bc1f | 3064 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3065 | { |
e8c2bc1f | 3066 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3067 | return 0; |
3d9932a8 | 3068 | |
e8c2bc1f | 3069 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3070 | } |
3071 | ||
ff6165b2 JA |
3072 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3073 | const struct iovec *fast_iov, | |
227c0c96 | 3074 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3075 | { |
e8c2bc1f | 3076 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3077 | return 0; |
e8c2bc1f | 3078 | if (!req->async_data) { |
6bf985dc PB |
3079 | if (__io_alloc_async_data(req)) { |
3080 | kfree(iovec); | |
5d204bcf | 3081 | return -ENOMEM; |
6bf985dc | 3082 | } |
b7bb4f7d | 3083 | |
ff6165b2 | 3084 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3085 | } |
b7bb4f7d | 3086 | return 0; |
f67676d1 JA |
3087 | } |
3088 | ||
73debe68 | 3089 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3090 | { |
e8c2bc1f | 3091 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3092 | struct iovec *iov = iorw->fast_iov; |
847595de | 3093 | int ret; |
c3e330a4 | 3094 | |
2846c481 | 3095 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3096 | if (unlikely(ret < 0)) |
3097 | return ret; | |
3098 | ||
ab0b196c PB |
3099 | iorw->bytes_done = 0; |
3100 | iorw->free_iovec = iov; | |
3101 | if (iov) | |
3102 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3103 | return 0; |
3104 | } | |
3105 | ||
73debe68 | 3106 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3107 | { |
3529d8c2 JA |
3108 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3109 | return -EBADF; | |
93642ef8 | 3110 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3111 | } |
3112 | ||
c1dd91d1 JA |
3113 | /* |
3114 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3115 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3116 | * This gets called when the page is unlocked, and we generally expect that to | |
3117 | * happen when the page IO is completed and the page is now uptodate. This will | |
3118 | * queue a task_work based retry of the operation, attempting to copy the data | |
3119 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3120 | * do a thread based blocking retry of the operation. That's the unexpected | |
3121 | * slow path. | |
3122 | */ | |
bcf5a063 JA |
3123 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3124 | int sync, void *arg) | |
3125 | { | |
3126 | struct wait_page_queue *wpq; | |
3127 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3128 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3129 | |
3130 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3131 | ||
cdc8fcb4 LT |
3132 | if (!wake_page_match(wpq, key)) |
3133 | return 0; | |
3134 | ||
c8d317aa | 3135 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3136 | list_del_init(&wait->entry); |
3137 | ||
bcf5a063 JA |
3138 | /* submit ref gets dropped, acquire a new one */ |
3139 | refcount_inc(&req->refs); | |
921b9054 | 3140 | io_req_task_queue(req); |
bcf5a063 JA |
3141 | return 1; |
3142 | } | |
3143 | ||
c1dd91d1 JA |
3144 | /* |
3145 | * This controls whether a given IO request should be armed for async page | |
3146 | * based retry. If we return false here, the request is handed to the async | |
3147 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3148 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3149 | * will either succeed because the page is now uptodate and unlocked, or it | |
3150 | * will register a callback when the page is unlocked at IO completion. Through | |
3151 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3152 | * That retry will attempt the buffered read again. The retry will generally | |
3153 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3154 | * async worker threads for a blocking retry. | |
3155 | */ | |
227c0c96 | 3156 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3157 | { |
e8c2bc1f JA |
3158 | struct io_async_rw *rw = req->async_data; |
3159 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3160 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3161 | |
bcf5a063 JA |
3162 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3163 | if (req->flags & REQ_F_NOWAIT) | |
3164 | return false; | |
f67676d1 | 3165 | |
227c0c96 | 3166 | /* Only for buffered IO */ |
3b2a4439 | 3167 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3168 | return false; |
3b2a4439 | 3169 | |
bcf5a063 JA |
3170 | /* |
3171 | * just use poll if we can, and don't attempt if the fs doesn't | |
3172 | * support callback based unlocks | |
3173 | */ | |
3174 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3175 | return false; | |
f67676d1 | 3176 | |
3b2a4439 JA |
3177 | wait->wait.func = io_async_buf_func; |
3178 | wait->wait.private = req; | |
3179 | wait->wait.flags = 0; | |
3180 | INIT_LIST_HEAD(&wait->wait.entry); | |
3181 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3182 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3183 | kiocb->ki_waitq = wait; |
3b2a4439 | 3184 | return true; |
bcf5a063 JA |
3185 | } |
3186 | ||
3187 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3188 | { | |
3189 | if (req->file->f_op->read_iter) | |
3190 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3191 | else if (req->file->f_op->read) |
4017eb91 | 3192 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3193 | else |
3194 | return -EINVAL; | |
f67676d1 JA |
3195 | } |
3196 | ||
889fca73 | 3197 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3198 | { |
3199 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3200 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3201 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3202 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3203 | ssize_t io_size, ret, ret2; |
45d189c6 | 3204 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3205 | |
2846c481 | 3206 | if (rw) { |
e8c2bc1f | 3207 | iter = &rw->iter; |
2846c481 PB |
3208 | iovec = NULL; |
3209 | } else { | |
3210 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3211 | if (ret < 0) | |
3212 | return ret; | |
3213 | } | |
632546c4 | 3214 | io_size = iov_iter_count(iter); |
fa15bafb | 3215 | req->result = io_size; |
2b188cc1 | 3216 | |
fd6c2e4c JA |
3217 | /* Ensure we clear previously set non-block flag */ |
3218 | if (!force_nonblock) | |
29de5f6a | 3219 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3220 | else |
3221 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3222 | ||
24c74678 | 3223 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3224 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3225 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3226 | return ret ?: -EAGAIN; |
6713e7a6 | 3227 | } |
9e645e11 | 3228 | |
632546c4 | 3229 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3230 | if (unlikely(ret)) { |
3231 | kfree(iovec); | |
3232 | return ret; | |
3233 | } | |
2b188cc1 | 3234 | |
227c0c96 | 3235 | ret = io_iter_do_read(req, iter); |
32960613 | 3236 | |
57cd657b | 3237 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3238 | if (req->async_data) |
3239 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3240 | goto out_free; |
227c0c96 | 3241 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3242 | /* IOPOLL retry should happen for io-wq threads */ |
3243 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3244 | goto done; |
75c668cd PB |
3245 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3246 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3247 | goto done; |
84216315 | 3248 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3249 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3250 | ret = 0; |
7335e3bf | 3251 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3252 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3253 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3254 | goto done; |
227c0c96 JA |
3255 | } |
3256 | ||
227c0c96 | 3257 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3258 | if (ret2) |
3259 | return ret2; | |
3260 | ||
fe1cdd55 | 3261 | iovec = NULL; |
e8c2bc1f | 3262 | rw = req->async_data; |
227c0c96 | 3263 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3264 | iter = &rw->iter; |
227c0c96 | 3265 | |
b23df91b PB |
3266 | do { |
3267 | io_size -= ret; | |
3268 | rw->bytes_done += ret; | |
3269 | /* if we can retry, do so with the callbacks armed */ | |
3270 | if (!io_rw_should_retry(req)) { | |
3271 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3272 | return -EAGAIN; | |
3273 | } | |
3274 | ||
3275 | /* | |
3276 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3277 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3278 | * desired page gets unlocked. We can also get a partial read | |
3279 | * here, and if we do, then just retry at the new offset. | |
3280 | */ | |
3281 | ret = io_iter_do_read(req, iter); | |
3282 | if (ret == -EIOCBQUEUED) | |
3283 | return 0; | |
227c0c96 | 3284 | /* we got some bytes, but not all. retry. */ |
b23df91b | 3285 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3286 | done: |
889fca73 | 3287 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3288 | out_free: |
3289 | /* it's faster to check here then delegate to kfree */ | |
3290 | if (iovec) | |
3291 | kfree(iovec); | |
5ea5dd45 | 3292 | return 0; |
2b188cc1 JA |
3293 | } |
3294 | ||
73debe68 | 3295 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3296 | { |
3529d8c2 JA |
3297 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3298 | return -EBADF; | |
93642ef8 | 3299 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3300 | } |
3301 | ||
889fca73 | 3302 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3303 | { |
3304 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3305 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3306 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3307 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3308 | ssize_t ret, ret2, io_size; |
45d189c6 | 3309 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3310 | |
2846c481 | 3311 | if (rw) { |
e8c2bc1f | 3312 | iter = &rw->iter; |
2846c481 PB |
3313 | iovec = NULL; |
3314 | } else { | |
3315 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3316 | if (ret < 0) | |
3317 | return ret; | |
3318 | } | |
632546c4 | 3319 | io_size = iov_iter_count(iter); |
fa15bafb | 3320 | req->result = io_size; |
2b188cc1 | 3321 | |
fd6c2e4c JA |
3322 | /* Ensure we clear previously set non-block flag */ |
3323 | if (!force_nonblock) | |
a88fc400 PB |
3324 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3325 | else | |
3326 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3327 | |
24c74678 | 3328 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3329 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3330 | goto copy_iov; |
31b51510 | 3331 | |
10d59345 JA |
3332 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3333 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3334 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3335 | goto copy_iov; |
31b51510 | 3336 | |
632546c4 | 3337 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3338 | if (unlikely(ret)) |
3339 | goto out_free; | |
4ed734b0 | 3340 | |
fa15bafb PB |
3341 | /* |
3342 | * Open-code file_start_write here to grab freeze protection, | |
3343 | * which will be released by another thread in | |
3344 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3345 | * released so that it doesn't complain about the held lock when | |
3346 | * we return to userspace. | |
3347 | */ | |
3348 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3349 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3350 | __sb_writers_release(file_inode(req->file)->i_sb, |
3351 | SB_FREEZE_WRITE); | |
3352 | } | |
3353 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3354 | |
fa15bafb | 3355 | if (req->file->f_op->write_iter) |
ff6165b2 | 3356 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3357 | else if (req->file->f_op->write) |
4017eb91 | 3358 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3359 | else |
3360 | ret2 = -EINVAL; | |
4ed734b0 | 3361 | |
fa15bafb PB |
3362 | /* |
3363 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3364 | * retry them without IOCB_NOWAIT. | |
3365 | */ | |
3366 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3367 | ret2 = -EAGAIN; | |
75c668cd PB |
3368 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3369 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3370 | goto done; |
3e6a0d3c JA |
3371 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3372 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3373 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3374 | /* IOPOLL retry should happen for io-wq threads */ |
3375 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3376 | goto copy_iov; | |
355afaeb | 3377 | done: |
889fca73 | 3378 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3379 | } else { |
f67676d1 | 3380 | copy_iov: |
84216315 | 3381 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3382 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3383 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3384 | return ret ?: -EAGAIN; |
2b188cc1 | 3385 | } |
31b51510 | 3386 | out_free: |
f261c168 | 3387 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3388 | if (iovec) |
6f2cc166 | 3389 | kfree(iovec); |
2b188cc1 JA |
3390 | return ret; |
3391 | } | |
3392 | ||
80a261fd JA |
3393 | static int io_renameat_prep(struct io_kiocb *req, |
3394 | const struct io_uring_sqe *sqe) | |
3395 | { | |
3396 | struct io_rename *ren = &req->rename; | |
3397 | const char __user *oldf, *newf; | |
3398 | ||
3399 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3400 | return -EBADF; | |
3401 | ||
3402 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3403 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3404 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3405 | ren->new_dfd = READ_ONCE(sqe->len); | |
3406 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3407 | ||
3408 | ren->oldpath = getname(oldf); | |
3409 | if (IS_ERR(ren->oldpath)) | |
3410 | return PTR_ERR(ren->oldpath); | |
3411 | ||
3412 | ren->newpath = getname(newf); | |
3413 | if (IS_ERR(ren->newpath)) { | |
3414 | putname(ren->oldpath); | |
3415 | return PTR_ERR(ren->newpath); | |
3416 | } | |
3417 | ||
3418 | req->flags |= REQ_F_NEED_CLEANUP; | |
3419 | return 0; | |
3420 | } | |
3421 | ||
45d189c6 | 3422 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3423 | { |
3424 | struct io_rename *ren = &req->rename; | |
3425 | int ret; | |
3426 | ||
45d189c6 | 3427 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3428 | return -EAGAIN; |
3429 | ||
3430 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3431 | ren->newpath, ren->flags); | |
3432 | ||
3433 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3434 | if (ret < 0) | |
3435 | req_set_fail_links(req); | |
3436 | io_req_complete(req, ret); | |
3437 | return 0; | |
3438 | } | |
3439 | ||
14a1143b JA |
3440 | static int io_unlinkat_prep(struct io_kiocb *req, |
3441 | const struct io_uring_sqe *sqe) | |
3442 | { | |
3443 | struct io_unlink *un = &req->unlink; | |
3444 | const char __user *fname; | |
3445 | ||
3446 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3447 | return -EBADF; | |
3448 | ||
3449 | un->dfd = READ_ONCE(sqe->fd); | |
3450 | ||
3451 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3452 | if (un->flags & ~AT_REMOVEDIR) | |
3453 | return -EINVAL; | |
3454 | ||
3455 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3456 | un->filename = getname(fname); | |
3457 | if (IS_ERR(un->filename)) | |
3458 | return PTR_ERR(un->filename); | |
3459 | ||
3460 | req->flags |= REQ_F_NEED_CLEANUP; | |
3461 | return 0; | |
3462 | } | |
3463 | ||
45d189c6 | 3464 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3465 | { |
3466 | struct io_unlink *un = &req->unlink; | |
3467 | int ret; | |
3468 | ||
45d189c6 | 3469 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3470 | return -EAGAIN; |
3471 | ||
3472 | if (un->flags & AT_REMOVEDIR) | |
3473 | ret = do_rmdir(un->dfd, un->filename); | |
3474 | else | |
3475 | ret = do_unlinkat(un->dfd, un->filename); | |
3476 | ||
3477 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3478 | if (ret < 0) | |
3479 | req_set_fail_links(req); | |
3480 | io_req_complete(req, ret); | |
3481 | return 0; | |
3482 | } | |
3483 | ||
36f4fa68 JA |
3484 | static int io_shutdown_prep(struct io_kiocb *req, |
3485 | const struct io_uring_sqe *sqe) | |
3486 | { | |
3487 | #if defined(CONFIG_NET) | |
3488 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3489 | return -EINVAL; | |
3490 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3491 | sqe->buf_index) | |
3492 | return -EINVAL; | |
3493 | ||
3494 | req->shutdown.how = READ_ONCE(sqe->len); | |
3495 | return 0; | |
3496 | #else | |
3497 | return -EOPNOTSUPP; | |
3498 | #endif | |
3499 | } | |
3500 | ||
45d189c6 | 3501 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3502 | { |
3503 | #if defined(CONFIG_NET) | |
3504 | struct socket *sock; | |
3505 | int ret; | |
3506 | ||
45d189c6 | 3507 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3508 | return -EAGAIN; |
3509 | ||
48aba79b | 3510 | sock = sock_from_file(req->file); |
36f4fa68 | 3511 | if (unlikely(!sock)) |
48aba79b | 3512 | return -ENOTSOCK; |
36f4fa68 JA |
3513 | |
3514 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3515 | if (ret < 0) |
3516 | req_set_fail_links(req); | |
36f4fa68 JA |
3517 | io_req_complete(req, ret); |
3518 | return 0; | |
3519 | #else | |
3520 | return -EOPNOTSUPP; | |
3521 | #endif | |
3522 | } | |
3523 | ||
f2a8d5c7 PB |
3524 | static int __io_splice_prep(struct io_kiocb *req, |
3525 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3526 | { |
3527 | struct io_splice* sp = &req->splice; | |
3528 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3529 | |
3232dd02 PB |
3530 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3531 | return -EINVAL; | |
7d67af2c PB |
3532 | |
3533 | sp->file_in = NULL; | |
7d67af2c PB |
3534 | sp->len = READ_ONCE(sqe->len); |
3535 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3536 | ||
3537 | if (unlikely(sp->flags & ~valid_flags)) | |
3538 | return -EINVAL; | |
3539 | ||
8371adf5 PB |
3540 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3541 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3542 | if (!sp->file_in) | |
3543 | return -EBADF; | |
7d67af2c PB |
3544 | req->flags |= REQ_F_NEED_CLEANUP; |
3545 | ||
7cdaf587 XW |
3546 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3547 | /* | |
3548 | * Splice operation will be punted aync, and here need to | |
3549 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3550 | */ | |
7d67af2c | 3551 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3552 | } |
7d67af2c PB |
3553 | |
3554 | return 0; | |
3555 | } | |
3556 | ||
f2a8d5c7 PB |
3557 | static int io_tee_prep(struct io_kiocb *req, |
3558 | const struct io_uring_sqe *sqe) | |
3559 | { | |
3560 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3561 | return -EINVAL; | |
3562 | return __io_splice_prep(req, sqe); | |
3563 | } | |
3564 | ||
45d189c6 | 3565 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3566 | { |
3567 | struct io_splice *sp = &req->splice; | |
3568 | struct file *in = sp->file_in; | |
3569 | struct file *out = sp->file_out; | |
3570 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3571 | long ret = 0; | |
3572 | ||
45d189c6 | 3573 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3574 | return -EAGAIN; |
3575 | if (sp->len) | |
3576 | ret = do_tee(in, out, sp->len, flags); | |
3577 | ||
3578 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3579 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3580 | ||
f2a8d5c7 PB |
3581 | if (ret != sp->len) |
3582 | req_set_fail_links(req); | |
e1e16097 | 3583 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3584 | return 0; |
3585 | } | |
3586 | ||
3587 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3588 | { | |
3589 | struct io_splice* sp = &req->splice; | |
3590 | ||
3591 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3592 | sp->off_out = READ_ONCE(sqe->off); | |
3593 | return __io_splice_prep(req, sqe); | |
3594 | } | |
3595 | ||
45d189c6 | 3596 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3597 | { |
3598 | struct io_splice *sp = &req->splice; | |
3599 | struct file *in = sp->file_in; | |
3600 | struct file *out = sp->file_out; | |
3601 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3602 | loff_t *poff_in, *poff_out; | |
c9687426 | 3603 | long ret = 0; |
7d67af2c | 3604 | |
45d189c6 | 3605 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3606 | return -EAGAIN; |
7d67af2c PB |
3607 | |
3608 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3609 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3610 | |
948a7749 | 3611 | if (sp->len) |
c9687426 | 3612 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3613 | |
3614 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3615 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3616 | ||
7d67af2c PB |
3617 | if (ret != sp->len) |
3618 | req_set_fail_links(req); | |
e1e16097 | 3619 | io_req_complete(req, ret); |
7d67af2c PB |
3620 | return 0; |
3621 | } | |
3622 | ||
2b188cc1 JA |
3623 | /* |
3624 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3625 | */ | |
889fca73 | 3626 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3627 | { |
3628 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3629 | |
def596e9 JA |
3630 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3631 | return -EINVAL; | |
3632 | ||
889fca73 | 3633 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3634 | return 0; |
3635 | } | |
3636 | ||
1155c76a | 3637 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3638 | { |
6b06314c | 3639 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3640 | |
09bb8394 JA |
3641 | if (!req->file) |
3642 | return -EBADF; | |
c992fe29 | 3643 | |
6b06314c | 3644 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3645 | return -EINVAL; |
edafccee | 3646 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3647 | return -EINVAL; |
3648 | ||
8ed8d3c3 JA |
3649 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3650 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3651 | return -EINVAL; | |
3652 | ||
3653 | req->sync.off = READ_ONCE(sqe->off); | |
3654 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3655 | return 0; |
3656 | } | |
3657 | ||
45d189c6 | 3658 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3659 | { |
8ed8d3c3 | 3660 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3661 | int ret; |
3662 | ||
ac45abc0 | 3663 | /* fsync always requires a blocking context */ |
45d189c6 | 3664 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3665 | return -EAGAIN; |
3666 | ||
9adbd45d | 3667 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3668 | end > 0 ? end : LLONG_MAX, |
3669 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3670 | if (ret < 0) | |
3671 | req_set_fail_links(req); | |
e1e16097 | 3672 | io_req_complete(req, ret); |
c992fe29 CH |
3673 | return 0; |
3674 | } | |
3675 | ||
d63d1b5e JA |
3676 | static int io_fallocate_prep(struct io_kiocb *req, |
3677 | const struct io_uring_sqe *sqe) | |
3678 | { | |
3679 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3680 | return -EINVAL; | |
3232dd02 PB |
3681 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3682 | return -EINVAL; | |
d63d1b5e JA |
3683 | |
3684 | req->sync.off = READ_ONCE(sqe->off); | |
3685 | req->sync.len = READ_ONCE(sqe->addr); | |
3686 | req->sync.mode = READ_ONCE(sqe->len); | |
3687 | return 0; | |
3688 | } | |
3689 | ||
45d189c6 | 3690 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3691 | { |
ac45abc0 PB |
3692 | int ret; |
3693 | ||
d63d1b5e | 3694 | /* fallocate always requiring blocking context */ |
45d189c6 | 3695 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3696 | return -EAGAIN; |
ac45abc0 PB |
3697 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3698 | req->sync.len); | |
ac45abc0 PB |
3699 | if (ret < 0) |
3700 | req_set_fail_links(req); | |
e1e16097 | 3701 | io_req_complete(req, ret); |
5d17b4a4 JA |
3702 | return 0; |
3703 | } | |
3704 | ||
ec65fea5 | 3705 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3706 | { |
f8748881 | 3707 | const char __user *fname; |
15b71abe | 3708 | int ret; |
b7bb4f7d | 3709 | |
ec65fea5 | 3710 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3711 | return -EINVAL; |
ec65fea5 | 3712 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3713 | return -EBADF; |
03b1230c | 3714 | |
ec65fea5 PB |
3715 | /* open.how should be already initialised */ |
3716 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3717 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3718 | |
25e72d10 PB |
3719 | req->open.dfd = READ_ONCE(sqe->fd); |
3720 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3721 | req->open.filename = getname(fname); |
15b71abe JA |
3722 | if (IS_ERR(req->open.filename)) { |
3723 | ret = PTR_ERR(req->open.filename); | |
3724 | req->open.filename = NULL; | |
3725 | return ret; | |
3726 | } | |
4022e7af | 3727 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3728 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3729 | return 0; |
03b1230c JA |
3730 | } |
3731 | ||
ec65fea5 PB |
3732 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3733 | { | |
3734 | u64 flags, mode; | |
3735 | ||
14587a46 | 3736 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3737 | return -EINVAL; |
ec65fea5 PB |
3738 | mode = READ_ONCE(sqe->len); |
3739 | flags = READ_ONCE(sqe->open_flags); | |
3740 | req->open.how = build_open_how(flags, mode); | |
3741 | return __io_openat_prep(req, sqe); | |
3742 | } | |
3743 | ||
cebdb986 | 3744 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3745 | { |
cebdb986 | 3746 | struct open_how __user *how; |
cebdb986 | 3747 | size_t len; |
0fa03c62 JA |
3748 | int ret; |
3749 | ||
14587a46 | 3750 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3751 | return -EINVAL; |
cebdb986 JA |
3752 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3753 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3754 | if (len < OPEN_HOW_SIZE_VER0) |
3755 | return -EINVAL; | |
3529d8c2 | 3756 | |
cebdb986 JA |
3757 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3758 | len); | |
3759 | if (ret) | |
3760 | return ret; | |
3529d8c2 | 3761 | |
ec65fea5 | 3762 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3763 | } |
3764 | ||
45d189c6 | 3765 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3766 | { |
3767 | struct open_flags op; | |
15b71abe | 3768 | struct file *file; |
3a81fd02 JA |
3769 | bool nonblock_set; |
3770 | bool resolve_nonblock; | |
15b71abe JA |
3771 | int ret; |
3772 | ||
cebdb986 | 3773 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3774 | if (ret) |
3775 | goto err; | |
3a81fd02 JA |
3776 | nonblock_set = op.open_flag & O_NONBLOCK; |
3777 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3778 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3779 | /* |
3780 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3781 | * it'll always -EAGAIN | |
3782 | */ | |
3783 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3784 | return -EAGAIN; | |
3785 | op.lookup_flags |= LOOKUP_CACHED; | |
3786 | op.open_flag |= O_NONBLOCK; | |
3787 | } | |
15b71abe | 3788 | |
4022e7af | 3789 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3790 | if (ret < 0) |
3791 | goto err; | |
3792 | ||
3793 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3794 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3795 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3796 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3797 | /* |
3a81fd02 JA |
3798 | * We could hang on to this 'fd', but seems like marginal |
3799 | * gain for something that is now known to be a slower path. | |
3800 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3801 | */ |
3a81fd02 JA |
3802 | put_unused_fd(ret); |
3803 | return -EAGAIN; | |
3804 | } | |
3805 | ||
15b71abe JA |
3806 | if (IS_ERR(file)) { |
3807 | put_unused_fd(ret); | |
3808 | ret = PTR_ERR(file); | |
3809 | } else { | |
45d189c6 | 3810 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3811 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3812 | fsnotify_open(file); |
3813 | fd_install(ret, file); | |
3814 | } | |
3815 | err: | |
3816 | putname(req->open.filename); | |
8fef80bf | 3817 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3818 | if (ret < 0) |
3819 | req_set_fail_links(req); | |
e1e16097 | 3820 | io_req_complete(req, ret); |
15b71abe JA |
3821 | return 0; |
3822 | } | |
3823 | ||
45d189c6 | 3824 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3825 | { |
45d189c6 | 3826 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
3827 | } |
3828 | ||
067524e9 JA |
3829 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3830 | const struct io_uring_sqe *sqe) | |
3831 | { | |
3832 | struct io_provide_buf *p = &req->pbuf; | |
3833 | u64 tmp; | |
3834 | ||
3835 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3836 | return -EINVAL; | |
3837 | ||
3838 | tmp = READ_ONCE(sqe->fd); | |
3839 | if (!tmp || tmp > USHRT_MAX) | |
3840 | return -EINVAL; | |
3841 | ||
3842 | memset(p, 0, sizeof(*p)); | |
3843 | p->nbufs = tmp; | |
3844 | p->bgid = READ_ONCE(sqe->buf_group); | |
3845 | return 0; | |
3846 | } | |
3847 | ||
3848 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3849 | int bgid, unsigned nbufs) | |
3850 | { | |
3851 | unsigned i = 0; | |
3852 | ||
3853 | /* shouldn't happen */ | |
3854 | if (!nbufs) | |
3855 | return 0; | |
3856 | ||
3857 | /* the head kbuf is the list itself */ | |
3858 | while (!list_empty(&buf->list)) { | |
3859 | struct io_buffer *nxt; | |
3860 | ||
3861 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3862 | list_del(&nxt->list); | |
3863 | kfree(nxt); | |
3864 | if (++i == nbufs) | |
3865 | return i; | |
3866 | } | |
3867 | i++; | |
3868 | kfree(buf); | |
3869 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3870 | ||
3871 | return i; | |
3872 | } | |
3873 | ||
889fca73 | 3874 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3875 | { |
3876 | struct io_provide_buf *p = &req->pbuf; | |
3877 | struct io_ring_ctx *ctx = req->ctx; | |
3878 | struct io_buffer *head; | |
3879 | int ret = 0; | |
45d189c6 | 3880 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3881 | |
3882 | io_ring_submit_lock(ctx, !force_nonblock); | |
3883 | ||
3884 | lockdep_assert_held(&ctx->uring_lock); | |
3885 | ||
3886 | ret = -ENOENT; | |
3887 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3888 | if (head) | |
3889 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3890 | if (ret < 0) |
3891 | req_set_fail_links(req); | |
067524e9 | 3892 | |
31bff9a5 PB |
3893 | /* need to hold the lock to complete IOPOLL requests */ |
3894 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3895 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3896 | io_ring_submit_unlock(ctx, !force_nonblock); |
3897 | } else { | |
3898 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3899 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3900 | } |
067524e9 JA |
3901 | return 0; |
3902 | } | |
3903 | ||
ddf0322d JA |
3904 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3905 | const struct io_uring_sqe *sqe) | |
3906 | { | |
3907 | struct io_provide_buf *p = &req->pbuf; | |
3908 | u64 tmp; | |
3909 | ||
3910 | if (sqe->ioprio || sqe->rw_flags) | |
3911 | return -EINVAL; | |
3912 | ||
3913 | tmp = READ_ONCE(sqe->fd); | |
3914 | if (!tmp || tmp > USHRT_MAX) | |
3915 | return -E2BIG; | |
3916 | p->nbufs = tmp; | |
3917 | p->addr = READ_ONCE(sqe->addr); | |
3918 | p->len = READ_ONCE(sqe->len); | |
3919 | ||
efe68c1c | 3920 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3921 | return -EFAULT; |
3922 | ||
3923 | p->bgid = READ_ONCE(sqe->buf_group); | |
3924 | tmp = READ_ONCE(sqe->off); | |
3925 | if (tmp > USHRT_MAX) | |
3926 | return -E2BIG; | |
3927 | p->bid = tmp; | |
3928 | return 0; | |
3929 | } | |
3930 | ||
3931 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3932 | { | |
3933 | struct io_buffer *buf; | |
3934 | u64 addr = pbuf->addr; | |
3935 | int i, bid = pbuf->bid; | |
3936 | ||
3937 | for (i = 0; i < pbuf->nbufs; i++) { | |
3938 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3939 | if (!buf) | |
3940 | break; | |
3941 | ||
3942 | buf->addr = addr; | |
3943 | buf->len = pbuf->len; | |
3944 | buf->bid = bid; | |
3945 | addr += pbuf->len; | |
3946 | bid++; | |
3947 | if (!*head) { | |
3948 | INIT_LIST_HEAD(&buf->list); | |
3949 | *head = buf; | |
3950 | } else { | |
3951 | list_add_tail(&buf->list, &(*head)->list); | |
3952 | } | |
3953 | } | |
3954 | ||
3955 | return i ? i : -ENOMEM; | |
3956 | } | |
3957 | ||
889fca73 | 3958 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3959 | { |
3960 | struct io_provide_buf *p = &req->pbuf; | |
3961 | struct io_ring_ctx *ctx = req->ctx; | |
3962 | struct io_buffer *head, *list; | |
3963 | int ret = 0; | |
45d189c6 | 3964 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3965 | |
3966 | io_ring_submit_lock(ctx, !force_nonblock); | |
3967 | ||
3968 | lockdep_assert_held(&ctx->uring_lock); | |
3969 | ||
3970 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3971 | ||
3972 | ret = io_add_buffers(p, &head); | |
3973 | if (ret < 0) | |
3974 | goto out; | |
3975 | ||
3976 | if (!list) { | |
3977 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3978 | GFP_KERNEL); | |
3979 | if (ret < 0) { | |
067524e9 | 3980 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
3981 | goto out; |
3982 | } | |
3983 | } | |
3984 | out: | |
ddf0322d JA |
3985 | if (ret < 0) |
3986 | req_set_fail_links(req); | |
31bff9a5 PB |
3987 | |
3988 | /* need to hold the lock to complete IOPOLL requests */ | |
3989 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3990 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3991 | io_ring_submit_unlock(ctx, !force_nonblock); |
3992 | } else { | |
3993 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3994 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3995 | } |
ddf0322d | 3996 | return 0; |
cebdb986 JA |
3997 | } |
3998 | ||
3e4827b0 JA |
3999 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4000 | const struct io_uring_sqe *sqe) | |
4001 | { | |
4002 | #if defined(CONFIG_EPOLL) | |
4003 | if (sqe->ioprio || sqe->buf_index) | |
4004 | return -EINVAL; | |
6ca56f84 | 4005 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4006 | return -EINVAL; |
3e4827b0 JA |
4007 | |
4008 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4009 | req->epoll.op = READ_ONCE(sqe->len); | |
4010 | req->epoll.fd = READ_ONCE(sqe->off); | |
4011 | ||
4012 | if (ep_op_has_event(req->epoll.op)) { | |
4013 | struct epoll_event __user *ev; | |
4014 | ||
4015 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4016 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4017 | return -EFAULT; | |
4018 | } | |
4019 | ||
4020 | return 0; | |
4021 | #else | |
4022 | return -EOPNOTSUPP; | |
4023 | #endif | |
4024 | } | |
4025 | ||
889fca73 | 4026 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4027 | { |
4028 | #if defined(CONFIG_EPOLL) | |
4029 | struct io_epoll *ie = &req->epoll; | |
4030 | int ret; | |
45d189c6 | 4031 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4032 | |
4033 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4034 | if (force_nonblock && ret == -EAGAIN) | |
4035 | return -EAGAIN; | |
4036 | ||
4037 | if (ret < 0) | |
4038 | req_set_fail_links(req); | |
889fca73 | 4039 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4040 | return 0; |
4041 | #else | |
4042 | return -EOPNOTSUPP; | |
4043 | #endif | |
4044 | } | |
4045 | ||
c1ca757b JA |
4046 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4047 | { | |
4048 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4049 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4050 | return -EINVAL; | |
3232dd02 PB |
4051 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4052 | return -EINVAL; | |
c1ca757b JA |
4053 | |
4054 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4055 | req->madvise.len = READ_ONCE(sqe->len); | |
4056 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4057 | return 0; | |
4058 | #else | |
4059 | return -EOPNOTSUPP; | |
4060 | #endif | |
4061 | } | |
4062 | ||
45d189c6 | 4063 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4064 | { |
4065 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4066 | struct io_madvise *ma = &req->madvise; | |
4067 | int ret; | |
4068 | ||
45d189c6 | 4069 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4070 | return -EAGAIN; |
4071 | ||
0726b01e | 4072 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4073 | if (ret < 0) |
4074 | req_set_fail_links(req); | |
e1e16097 | 4075 | io_req_complete(req, ret); |
c1ca757b JA |
4076 | return 0; |
4077 | #else | |
4078 | return -EOPNOTSUPP; | |
4079 | #endif | |
4080 | } | |
4081 | ||
4840e418 JA |
4082 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4083 | { | |
4084 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4085 | return -EINVAL; | |
3232dd02 PB |
4086 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4087 | return -EINVAL; | |
4840e418 JA |
4088 | |
4089 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4090 | req->fadvise.len = READ_ONCE(sqe->len); | |
4091 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4092 | return 0; | |
4093 | } | |
4094 | ||
45d189c6 | 4095 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4096 | { |
4097 | struct io_fadvise *fa = &req->fadvise; | |
4098 | int ret; | |
4099 | ||
45d189c6 | 4100 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4101 | switch (fa->advice) { |
4102 | case POSIX_FADV_NORMAL: | |
4103 | case POSIX_FADV_RANDOM: | |
4104 | case POSIX_FADV_SEQUENTIAL: | |
4105 | break; | |
4106 | default: | |
4107 | return -EAGAIN; | |
4108 | } | |
4109 | } | |
4840e418 JA |
4110 | |
4111 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4112 | if (ret < 0) | |
4113 | req_set_fail_links(req); | |
e1e16097 | 4114 | io_req_complete(req, ret); |
4840e418 JA |
4115 | return 0; |
4116 | } | |
4117 | ||
eddc7ef5 JA |
4118 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4119 | { | |
6ca56f84 | 4120 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4121 | return -EINVAL; |
eddc7ef5 JA |
4122 | if (sqe->ioprio || sqe->buf_index) |
4123 | return -EINVAL; | |
9c280f90 | 4124 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4125 | return -EBADF; |
eddc7ef5 | 4126 | |
1d9e1288 BM |
4127 | req->statx.dfd = READ_ONCE(sqe->fd); |
4128 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4129 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4130 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4131 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4132 | |
4133 | return 0; | |
4134 | } | |
4135 | ||
45d189c6 | 4136 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4137 | { |
1d9e1288 | 4138 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4139 | int ret; |
4140 | ||
45d189c6 | 4141 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4142 | /* only need file table for an actual valid fd */ |
4143 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4144 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4145 | return -EAGAIN; |
5b0bbee4 | 4146 | } |
eddc7ef5 | 4147 | |
e62753e4 BM |
4148 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4149 | ctx->buffer); | |
eddc7ef5 | 4150 | |
eddc7ef5 JA |
4151 | if (ret < 0) |
4152 | req_set_fail_links(req); | |
e1e16097 | 4153 | io_req_complete(req, ret); |
eddc7ef5 JA |
4154 | return 0; |
4155 | } | |
4156 | ||
b5dba59e JA |
4157 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4158 | { | |
14587a46 | 4159 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4160 | return -EINVAL; |
b5dba59e JA |
4161 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4162 | sqe->rw_flags || sqe->buf_index) | |
4163 | return -EINVAL; | |
9c280f90 | 4164 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4165 | return -EBADF; |
b5dba59e JA |
4166 | |
4167 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4168 | return 0; |
b5dba59e JA |
4169 | } |
4170 | ||
889fca73 | 4171 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4172 | { |
9eac1904 | 4173 | struct files_struct *files = current->files; |
3af73b28 | 4174 | struct io_close *close = &req->close; |
9eac1904 JA |
4175 | struct fdtable *fdt; |
4176 | struct file *file; | |
b5dba59e JA |
4177 | int ret; |
4178 | ||
9eac1904 JA |
4179 | file = NULL; |
4180 | ret = -EBADF; | |
4181 | spin_lock(&files->file_lock); | |
4182 | fdt = files_fdtable(files); | |
4183 | if (close->fd >= fdt->max_fds) { | |
4184 | spin_unlock(&files->file_lock); | |
4185 | goto err; | |
4186 | } | |
4187 | file = fdt->fd[close->fd]; | |
4188 | if (!file) { | |
4189 | spin_unlock(&files->file_lock); | |
4190 | goto err; | |
4191 | } | |
4192 | ||
4193 | if (file->f_op == &io_uring_fops) { | |
4194 | spin_unlock(&files->file_lock); | |
4195 | file = NULL; | |
4196 | goto err; | |
3af73b28 | 4197 | } |
b5dba59e JA |
4198 | |
4199 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4200 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4201 | spin_unlock(&files->file_lock); |
0bf0eefd | 4202 | return -EAGAIN; |
a2100672 | 4203 | } |
b5dba59e | 4204 | |
9eac1904 JA |
4205 | ret = __close_fd_get_file(close->fd, &file); |
4206 | spin_unlock(&files->file_lock); | |
4207 | if (ret < 0) { | |
4208 | if (ret == -ENOENT) | |
4209 | ret = -EBADF; | |
4210 | goto err; | |
4211 | } | |
4212 | ||
3af73b28 | 4213 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4214 | ret = filp_close(file, current->files); |
4215 | err: | |
3af73b28 PB |
4216 | if (ret < 0) |
4217 | req_set_fail_links(req); | |
9eac1904 JA |
4218 | if (file) |
4219 | fput(file); | |
889fca73 | 4220 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4221 | return 0; |
b5dba59e JA |
4222 | } |
4223 | ||
1155c76a | 4224 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4225 | { |
4226 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4227 | |
5d17b4a4 JA |
4228 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4229 | return -EINVAL; | |
4230 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4231 | return -EINVAL; | |
4232 | ||
8ed8d3c3 JA |
4233 | req->sync.off = READ_ONCE(sqe->off); |
4234 | req->sync.len = READ_ONCE(sqe->len); | |
4235 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4236 | return 0; |
4237 | } | |
4238 | ||
45d189c6 | 4239 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4240 | { |
8ed8d3c3 JA |
4241 | int ret; |
4242 | ||
ac45abc0 | 4243 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4244 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4245 | return -EAGAIN; |
4246 | ||
9adbd45d | 4247 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4248 | req->sync.flags); |
4249 | if (ret < 0) | |
4250 | req_set_fail_links(req); | |
e1e16097 | 4251 | io_req_complete(req, ret); |
5d17b4a4 JA |
4252 | return 0; |
4253 | } | |
4254 | ||
469956e8 | 4255 | #if defined(CONFIG_NET) |
02d27d89 PB |
4256 | static int io_setup_async_msg(struct io_kiocb *req, |
4257 | struct io_async_msghdr *kmsg) | |
4258 | { | |
e8c2bc1f JA |
4259 | struct io_async_msghdr *async_msg = req->async_data; |
4260 | ||
4261 | if (async_msg) | |
02d27d89 | 4262 | return -EAGAIN; |
e8c2bc1f | 4263 | if (io_alloc_async_data(req)) { |
257e84a5 | 4264 | kfree(kmsg->free_iov); |
02d27d89 PB |
4265 | return -ENOMEM; |
4266 | } | |
e8c2bc1f | 4267 | async_msg = req->async_data; |
02d27d89 | 4268 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4269 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4270 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4271 | /* if were using fast_iov, set it to the new one */ |
4272 | if (!async_msg->free_iov) | |
4273 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4274 | ||
02d27d89 PB |
4275 | return -EAGAIN; |
4276 | } | |
4277 | ||
2ae523ed PB |
4278 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4279 | struct io_async_msghdr *iomsg) | |
4280 | { | |
2ae523ed | 4281 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4282 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4283 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4284 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4285 | } |
4286 | ||
93642ef8 PB |
4287 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4288 | { | |
4289 | int ret; | |
4290 | ||
4291 | if (!io_op_defs[req->opcode].needs_async_data) | |
4292 | return 0; | |
4293 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4294 | if (!ret) | |
4295 | req->flags |= REQ_F_NEED_CLEANUP; | |
4296 | return ret; | |
4297 | } | |
4298 | ||
3529d8c2 | 4299 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4300 | { |
e47293fd | 4301 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4302 | |
d2b6f48b PB |
4303 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4304 | return -EINVAL; | |
4305 | ||
e47293fd | 4306 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4307 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4308 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4309 | |
d8768362 JA |
4310 | #ifdef CONFIG_COMPAT |
4311 | if (req->ctx->compat) | |
4312 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4313 | #endif | |
93642ef8 | 4314 | return 0; |
03b1230c JA |
4315 | } |
4316 | ||
889fca73 | 4317 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4318 | { |
6b754c8b | 4319 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4320 | struct socket *sock; |
7a7cacba | 4321 | unsigned flags; |
0fa03c62 JA |
4322 | int ret; |
4323 | ||
dba4a925 | 4324 | sock = sock_from_file(req->file); |
7a7cacba | 4325 | if (unlikely(!sock)) |
dba4a925 | 4326 | return -ENOTSOCK; |
3529d8c2 | 4327 | |
257e84a5 PB |
4328 | kmsg = req->async_data; |
4329 | if (!kmsg) { | |
7a7cacba PB |
4330 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4331 | if (ret) | |
4332 | return ret; | |
4333 | kmsg = &iomsg; | |
0fa03c62 | 4334 | } |
0fa03c62 | 4335 | |
7a7cacba PB |
4336 | flags = req->sr_msg.msg_flags; |
4337 | if (flags & MSG_DONTWAIT) | |
4338 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4339 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4340 | flags |= MSG_DONTWAIT; |
e47293fd | 4341 | |
7a7cacba | 4342 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4343 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4344 | return io_setup_async_msg(req, kmsg); |
4345 | if (ret == -ERESTARTSYS) | |
4346 | ret = -EINTR; | |
0fa03c62 | 4347 | |
257e84a5 PB |
4348 | /* fast path, check for non-NULL to avoid function call */ |
4349 | if (kmsg->free_iov) | |
4350 | kfree(kmsg->free_iov); | |
99bc4c38 | 4351 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4352 | if (ret < 0) |
4353 | req_set_fail_links(req); | |
889fca73 | 4354 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4355 | return 0; |
03b1230c | 4356 | } |
aa1fa28f | 4357 | |
889fca73 | 4358 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4359 | { |
7a7cacba PB |
4360 | struct io_sr_msg *sr = &req->sr_msg; |
4361 | struct msghdr msg; | |
4362 | struct iovec iov; | |
fddaface | 4363 | struct socket *sock; |
7a7cacba | 4364 | unsigned flags; |
fddaface JA |
4365 | int ret; |
4366 | ||
dba4a925 | 4367 | sock = sock_from_file(req->file); |
7a7cacba | 4368 | if (unlikely(!sock)) |
dba4a925 | 4369 | return -ENOTSOCK; |
fddaface | 4370 | |
7a7cacba PB |
4371 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4372 | if (unlikely(ret)) | |
14db8411 | 4373 | return ret; |
fddaface | 4374 | |
7a7cacba PB |
4375 | msg.msg_name = NULL; |
4376 | msg.msg_control = NULL; | |
4377 | msg.msg_controllen = 0; | |
4378 | msg.msg_namelen = 0; | |
fddaface | 4379 | |
7a7cacba PB |
4380 | flags = req->sr_msg.msg_flags; |
4381 | if (flags & MSG_DONTWAIT) | |
4382 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4383 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4384 | flags |= MSG_DONTWAIT; |
fddaface | 4385 | |
7a7cacba PB |
4386 | msg.msg_flags = flags; |
4387 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4388 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4389 | return -EAGAIN; |
4390 | if (ret == -ERESTARTSYS) | |
4391 | ret = -EINTR; | |
fddaface | 4392 | |
fddaface JA |
4393 | if (ret < 0) |
4394 | req_set_fail_links(req); | |
889fca73 | 4395 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4396 | return 0; |
fddaface JA |
4397 | } |
4398 | ||
1400e697 PB |
4399 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4400 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4401 | { |
4402 | struct io_sr_msg *sr = &req->sr_msg; | |
4403 | struct iovec __user *uiov; | |
4404 | size_t iov_len; | |
4405 | int ret; | |
4406 | ||
1400e697 PB |
4407 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4408 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4409 | if (ret) |
4410 | return ret; | |
4411 | ||
4412 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4413 | if (iov_len > 1) | |
4414 | return -EINVAL; | |
5476dfed | 4415 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4416 | return -EFAULT; |
5476dfed | 4417 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4418 | iomsg->free_iov = NULL; |
52de1fe1 | 4419 | } else { |
257e84a5 | 4420 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4421 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4422 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4423 | false); |
52de1fe1 JA |
4424 | if (ret > 0) |
4425 | ret = 0; | |
4426 | } | |
4427 | ||
4428 | return ret; | |
4429 | } | |
4430 | ||
4431 | #ifdef CONFIG_COMPAT | |
4432 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4433 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4434 | { |
4435 | struct compat_msghdr __user *msg_compat; | |
4436 | struct io_sr_msg *sr = &req->sr_msg; | |
4437 | struct compat_iovec __user *uiov; | |
4438 | compat_uptr_t ptr; | |
4439 | compat_size_t len; | |
4440 | int ret; | |
4441 | ||
270a5940 | 4442 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4443 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4444 | &ptr, &len); |
4445 | if (ret) | |
4446 | return ret; | |
4447 | ||
4448 | uiov = compat_ptr(ptr); | |
4449 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4450 | compat_ssize_t clen; | |
4451 | ||
4452 | if (len > 1) | |
4453 | return -EINVAL; | |
4454 | if (!access_ok(uiov, sizeof(*uiov))) | |
4455 | return -EFAULT; | |
4456 | if (__get_user(clen, &uiov->iov_len)) | |
4457 | return -EFAULT; | |
4458 | if (clen < 0) | |
4459 | return -EINVAL; | |
2d280bc8 | 4460 | sr->len = clen; |
257e84a5 | 4461 | iomsg->free_iov = NULL; |
52de1fe1 | 4462 | } else { |
257e84a5 | 4463 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4464 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4465 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4466 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4467 | if (ret < 0) |
4468 | return ret; | |
4469 | } | |
4470 | ||
4471 | return 0; | |
4472 | } | |
4473 | #endif | |
4474 | ||
1400e697 PB |
4475 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4476 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4477 | { |
1400e697 | 4478 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4479 | |
4480 | #ifdef CONFIG_COMPAT | |
4481 | if (req->ctx->compat) | |
1400e697 | 4482 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4483 | #endif |
52de1fe1 | 4484 | |
1400e697 | 4485 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4486 | } |
4487 | ||
bcda7baa | 4488 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4489 | bool needs_lock) |
bcda7baa JA |
4490 | { |
4491 | struct io_sr_msg *sr = &req->sr_msg; | |
4492 | struct io_buffer *kbuf; | |
4493 | ||
bcda7baa JA |
4494 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4495 | if (IS_ERR(kbuf)) | |
4496 | return kbuf; | |
4497 | ||
4498 | sr->kbuf = kbuf; | |
4499 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4500 | return kbuf; |
fddaface JA |
4501 | } |
4502 | ||
7fbb1b54 PB |
4503 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4504 | { | |
4505 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4506 | } | |
4507 | ||
93642ef8 | 4508 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4509 | { |
99bc4c38 | 4510 | int ret; |
3529d8c2 | 4511 | |
93642ef8 PB |
4512 | if (!io_op_defs[req->opcode].needs_async_data) |
4513 | return 0; | |
4514 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4515 | if (!ret) | |
4516 | req->flags |= REQ_F_NEED_CLEANUP; | |
4517 | return ret; | |
4518 | } | |
4519 | ||
4520 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4521 | { | |
4522 | struct io_sr_msg *sr = &req->sr_msg; | |
4523 | ||
d2b6f48b PB |
4524 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4525 | return -EINVAL; | |
4526 | ||
3529d8c2 | 4527 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4528 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4529 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4530 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4531 | |
d8768362 JA |
4532 | #ifdef CONFIG_COMPAT |
4533 | if (req->ctx->compat) | |
4534 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4535 | #endif | |
93642ef8 | 4536 | return 0; |
aa1fa28f JA |
4537 | } |
4538 | ||
889fca73 | 4539 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4540 | { |
6b754c8b | 4541 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4542 | struct socket *sock; |
7fbb1b54 | 4543 | struct io_buffer *kbuf; |
7a7cacba | 4544 | unsigned flags; |
52de1fe1 | 4545 | int ret, cflags = 0; |
45d189c6 | 4546 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4547 | |
dba4a925 | 4548 | sock = sock_from_file(req->file); |
7a7cacba | 4549 | if (unlikely(!sock)) |
dba4a925 | 4550 | return -ENOTSOCK; |
3529d8c2 | 4551 | |
257e84a5 PB |
4552 | kmsg = req->async_data; |
4553 | if (!kmsg) { | |
7a7cacba PB |
4554 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4555 | if (ret) | |
681fda8d | 4556 | return ret; |
7a7cacba PB |
4557 | kmsg = &iomsg; |
4558 | } | |
03b1230c | 4559 | |
bc02ef33 | 4560 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4561 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4562 | if (IS_ERR(kbuf)) |
52de1fe1 | 4563 | return PTR_ERR(kbuf); |
7a7cacba | 4564 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4565 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4566 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4567 | 1, req->sr_msg.len); |
4568 | } | |
52de1fe1 | 4569 | |
7a7cacba PB |
4570 | flags = req->sr_msg.msg_flags; |
4571 | if (flags & MSG_DONTWAIT) | |
4572 | req->flags |= REQ_F_NOWAIT; | |
4573 | else if (force_nonblock) | |
4574 | flags |= MSG_DONTWAIT; | |
e47293fd | 4575 | |
7a7cacba PB |
4576 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4577 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4578 | if (force_nonblock && ret == -EAGAIN) |
4579 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4580 | if (ret == -ERESTARTSYS) |
4581 | ret = -EINTR; | |
03b1230c | 4582 | |
7fbb1b54 PB |
4583 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4584 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4585 | /* fast path, check for non-NULL to avoid function call */ |
4586 | if (kmsg->free_iov) | |
4587 | kfree(kmsg->free_iov); | |
99bc4c38 | 4588 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4589 | if (ret < 0) |
4590 | req_set_fail_links(req); | |
889fca73 | 4591 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4592 | return 0; |
0fa03c62 | 4593 | } |
5d17b4a4 | 4594 | |
889fca73 | 4595 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4596 | { |
6b754c8b | 4597 | struct io_buffer *kbuf; |
7a7cacba PB |
4598 | struct io_sr_msg *sr = &req->sr_msg; |
4599 | struct msghdr msg; | |
4600 | void __user *buf = sr->buf; | |
fddaface | 4601 | struct socket *sock; |
7a7cacba PB |
4602 | struct iovec iov; |
4603 | unsigned flags; | |
bcda7baa | 4604 | int ret, cflags = 0; |
45d189c6 | 4605 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4606 | |
dba4a925 | 4607 | sock = sock_from_file(req->file); |
7a7cacba | 4608 | if (unlikely(!sock)) |
dba4a925 | 4609 | return -ENOTSOCK; |
fddaface | 4610 | |
bc02ef33 | 4611 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4612 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4613 | if (IS_ERR(kbuf)) |
4614 | return PTR_ERR(kbuf); | |
7a7cacba | 4615 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4616 | } |
bcda7baa | 4617 | |
7a7cacba | 4618 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4619 | if (unlikely(ret)) |
4620 | goto out_free; | |
fddaface | 4621 | |
7a7cacba PB |
4622 | msg.msg_name = NULL; |
4623 | msg.msg_control = NULL; | |
4624 | msg.msg_controllen = 0; | |
4625 | msg.msg_namelen = 0; | |
4626 | msg.msg_iocb = NULL; | |
4627 | msg.msg_flags = 0; | |
fddaface | 4628 | |
7a7cacba PB |
4629 | flags = req->sr_msg.msg_flags; |
4630 | if (flags & MSG_DONTWAIT) | |
4631 | req->flags |= REQ_F_NOWAIT; | |
4632 | else if (force_nonblock) | |
4633 | flags |= MSG_DONTWAIT; | |
4634 | ||
4635 | ret = sock_recvmsg(sock, &msg, flags); | |
4636 | if (force_nonblock && ret == -EAGAIN) | |
4637 | return -EAGAIN; | |
4638 | if (ret == -ERESTARTSYS) | |
4639 | ret = -EINTR; | |
14c32eee | 4640 | out_free: |
7fbb1b54 PB |
4641 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4642 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4643 | if (ret < 0) |
4644 | req_set_fail_links(req); | |
889fca73 | 4645 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4646 | return 0; |
fddaface JA |
4647 | } |
4648 | ||
3529d8c2 | 4649 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4650 | { |
8ed8d3c3 JA |
4651 | struct io_accept *accept = &req->accept; |
4652 | ||
14587a46 | 4653 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4654 | return -EINVAL; |
8042d6ce | 4655 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4656 | return -EINVAL; |
4657 | ||
d55e5f5b JA |
4658 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4659 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4660 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4661 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4662 | return 0; |
8ed8d3c3 | 4663 | } |
17f2fe35 | 4664 | |
889fca73 | 4665 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4666 | { |
4667 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4668 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4669 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4670 | int ret; |
4671 | ||
e697deed JX |
4672 | if (req->file->f_flags & O_NONBLOCK) |
4673 | req->flags |= REQ_F_NOWAIT; | |
4674 | ||
8ed8d3c3 | 4675 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4676 | accept->addr_len, accept->flags, |
4677 | accept->nofile); | |
8ed8d3c3 | 4678 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4679 | return -EAGAIN; |
ac45abc0 PB |
4680 | if (ret < 0) { |
4681 | if (ret == -ERESTARTSYS) | |
4682 | ret = -EINTR; | |
4e88d6e7 | 4683 | req_set_fail_links(req); |
ac45abc0 | 4684 | } |
889fca73 | 4685 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4686 | return 0; |
8ed8d3c3 JA |
4687 | } |
4688 | ||
93642ef8 PB |
4689 | static int io_connect_prep_async(struct io_kiocb *req) |
4690 | { | |
4691 | struct io_async_connect *io = req->async_data; | |
4692 | struct io_connect *conn = &req->connect; | |
4693 | ||
4694 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4695 | } | |
4696 | ||
3529d8c2 | 4697 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4698 | { |
3529d8c2 | 4699 | struct io_connect *conn = &req->connect; |
f499a021 | 4700 | |
14587a46 | 4701 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4702 | return -EINVAL; |
4703 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4704 | return -EINVAL; | |
4705 | ||
3529d8c2 JA |
4706 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4707 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4708 | return 0; |
f499a021 JA |
4709 | } |
4710 | ||
889fca73 | 4711 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4712 | { |
e8c2bc1f | 4713 | struct io_async_connect __io, *io; |
f8e85cf2 | 4714 | unsigned file_flags; |
3fbb51c1 | 4715 | int ret; |
45d189c6 | 4716 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4717 | |
e8c2bc1f JA |
4718 | if (req->async_data) { |
4719 | io = req->async_data; | |
f499a021 | 4720 | } else { |
3529d8c2 JA |
4721 | ret = move_addr_to_kernel(req->connect.addr, |
4722 | req->connect.addr_len, | |
e8c2bc1f | 4723 | &__io.address); |
f499a021 JA |
4724 | if (ret) |
4725 | goto out; | |
4726 | io = &__io; | |
4727 | } | |
4728 | ||
3fbb51c1 JA |
4729 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4730 | ||
e8c2bc1f | 4731 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4732 | req->connect.addr_len, file_flags); |
87f80d62 | 4733 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4734 | if (req->async_data) |
b7bb4f7d | 4735 | return -EAGAIN; |
e8c2bc1f | 4736 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4737 | ret = -ENOMEM; |
4738 | goto out; | |
4739 | } | |
e8c2bc1f JA |
4740 | io = req->async_data; |
4741 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4742 | return -EAGAIN; |
f499a021 | 4743 | } |
f8e85cf2 JA |
4744 | if (ret == -ERESTARTSYS) |
4745 | ret = -EINTR; | |
f499a021 | 4746 | out: |
4e88d6e7 JA |
4747 | if (ret < 0) |
4748 | req_set_fail_links(req); | |
889fca73 | 4749 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4750 | return 0; |
469956e8 Y |
4751 | } |
4752 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4753 | #define IO_NETOP_FN(op) \ |
4754 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4755 | { \ | |
4756 | return -EOPNOTSUPP; \ | |
4757 | } | |
4758 | ||
4759 | #define IO_NETOP_PREP(op) \ | |
4760 | IO_NETOP_FN(op) \ | |
4761 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4762 | { \ | |
4763 | return -EOPNOTSUPP; \ | |
4764 | } \ | |
4765 | ||
4766 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4767 | IO_NETOP_PREP(op) \ | |
4768 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4769 | { \ | |
4770 | return -EOPNOTSUPP; \ | |
4771 | } | |
4772 | ||
4773 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4774 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4775 | IO_NETOP_PREP_ASYNC(connect); | |
4776 | IO_NETOP_PREP(accept); | |
4777 | IO_NETOP_FN(send); | |
4778 | IO_NETOP_FN(recv); | |
469956e8 | 4779 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4780 | |
d7718a9d JA |
4781 | struct io_poll_table { |
4782 | struct poll_table_struct pt; | |
4783 | struct io_kiocb *req; | |
4784 | int error; | |
4785 | }; | |
ce593a6c | 4786 | |
d7718a9d JA |
4787 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4788 | __poll_t mask, task_work_func_t func) | |
4789 | { | |
aa96bf8a | 4790 | int ret; |
d7718a9d JA |
4791 | |
4792 | /* for instances that support it check for an event match first: */ | |
4793 | if (mask && !(mask & poll->events)) | |
4794 | return 0; | |
4795 | ||
4796 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4797 | ||
4798 | list_del_init(&poll->wait.entry); | |
4799 | ||
d7718a9d | 4800 | req->result = mask; |
7cbf1722 | 4801 | req->task_work.func = func; |
6d816e08 JA |
4802 | percpu_ref_get(&req->ctx->refs); |
4803 | ||
d7718a9d | 4804 | /* |
e3aabf95 JA |
4805 | * If this fails, then the task is exiting. When a task exits, the |
4806 | * work gets canceled, so just cancel this request as well instead | |
4807 | * of executing it. We can't safely execute it anyway, as we may not | |
4808 | * have the needed state needed for it anyway. | |
d7718a9d | 4809 | */ |
355fb9e2 | 4810 | ret = io_req_task_work_add(req); |
aa96bf8a | 4811 | if (unlikely(ret)) { |
e3aabf95 | 4812 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4813 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4814 | } |
d7718a9d JA |
4815 | return 1; |
4816 | } | |
4817 | ||
74ce6ce4 JA |
4818 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4819 | __acquires(&req->ctx->completion_lock) | |
4820 | { | |
4821 | struct io_ring_ctx *ctx = req->ctx; | |
4822 | ||
4823 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4824 | struct poll_table_struct pt = { ._key = poll->events }; | |
4825 | ||
4826 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4827 | } | |
4828 | ||
4829 | spin_lock_irq(&ctx->completion_lock); | |
4830 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4831 | add_wait_queue(poll->head, &poll->wait); | |
4832 | return true; | |
4833 | } | |
4834 | ||
4835 | return false; | |
4836 | } | |
4837 | ||
d4e7cd36 | 4838 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4839 | { |
e8c2bc1f | 4840 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4841 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4842 | return req->async_data; |
d4e7cd36 JA |
4843 | return req->apoll->double_poll; |
4844 | } | |
4845 | ||
4846 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4847 | { | |
4848 | if (req->opcode == IORING_OP_POLL_ADD) | |
4849 | return &req->poll; | |
4850 | return &req->apoll->poll; | |
4851 | } | |
4852 | ||
4853 | static void io_poll_remove_double(struct io_kiocb *req) | |
4854 | { | |
4855 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4856 | |
4857 | lockdep_assert_held(&req->ctx->completion_lock); | |
4858 | ||
4859 | if (poll && poll->head) { | |
4860 | struct wait_queue_head *head = poll->head; | |
4861 | ||
4862 | spin_lock(&head->lock); | |
4863 | list_del_init(&poll->wait.entry); | |
4864 | if (poll->wait.private) | |
4865 | refcount_dec(&req->refs); | |
4866 | poll->head = NULL; | |
4867 | spin_unlock(&head->lock); | |
4868 | } | |
4869 | } | |
4870 | ||
4871 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4872 | { | |
4873 | struct io_ring_ctx *ctx = req->ctx; | |
4874 | ||
d4e7cd36 | 4875 | io_poll_remove_double(req); |
18bceab1 JA |
4876 | req->poll.done = true; |
4877 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4878 | io_commit_cqring(ctx); | |
4879 | } | |
4880 | ||
dd221f46 | 4881 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4882 | { |
dd221f46 | 4883 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4884 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4885 | struct io_kiocb *nxt; |
18bceab1 JA |
4886 | |
4887 | if (io_poll_rewait(req, &req->poll)) { | |
4888 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4889 | } else { |
4890 | hash_del(&req->hash_node); | |
4891 | io_poll_complete(req, req->result, 0); | |
4892 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4893 | |
dd221f46 PB |
4894 | nxt = io_put_req_find_next(req); |
4895 | io_cqring_ev_posted(ctx); | |
4896 | if (nxt) | |
4897 | __io_req_task_submit(nxt); | |
4898 | } | |
18bceab1 | 4899 | |
6d816e08 | 4900 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4901 | } |
4902 | ||
4903 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4904 | int sync, void *key) | |
4905 | { | |
4906 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4907 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4908 | __poll_t mask = key_to_poll(key); |
4909 | ||
4910 | /* for instances that support it check for an event match first: */ | |
4911 | if (mask && !(mask & poll->events)) | |
4912 | return 0; | |
4913 | ||
8706e04e JA |
4914 | list_del_init(&wait->entry); |
4915 | ||
807abcb0 | 4916 | if (poll && poll->head) { |
18bceab1 JA |
4917 | bool done; |
4918 | ||
807abcb0 JA |
4919 | spin_lock(&poll->head->lock); |
4920 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4921 | if (!done) |
807abcb0 | 4922 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4923 | /* make sure double remove sees this as being gone */ |
4924 | wait->private = NULL; | |
807abcb0 | 4925 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4926 | if (!done) { |
4927 | /* use wait func handler, so it matches the rq type */ | |
4928 | poll->wait.func(&poll->wait, mode, sync, key); | |
4929 | } | |
18bceab1 JA |
4930 | } |
4931 | refcount_dec(&req->refs); | |
4932 | return 1; | |
4933 | } | |
4934 | ||
4935 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4936 | wait_queue_func_t wake_func) | |
4937 | { | |
4938 | poll->head = NULL; | |
4939 | poll->done = false; | |
4940 | poll->canceled = false; | |
4941 | poll->events = events; | |
4942 | INIT_LIST_HEAD(&poll->wait.entry); | |
4943 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4944 | } | |
4945 | ||
4946 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4947 | struct wait_queue_head *head, |
4948 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4949 | { |
4950 | struct io_kiocb *req = pt->req; | |
4951 | ||
4952 | /* | |
4953 | * If poll->head is already set, it's because the file being polled | |
4954 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4955 | * for write). Setup a separate io_poll_iocb if this happens. | |
4956 | */ | |
4957 | if (unlikely(poll->head)) { | |
58852d4d PB |
4958 | struct io_poll_iocb *poll_one = poll; |
4959 | ||
18bceab1 | 4960 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4961 | if (*poll_ptr) { |
18bceab1 JA |
4962 | pt->error = -EINVAL; |
4963 | return; | |
4964 | } | |
1c3b3e65 JA |
4965 | /* double add on the same waitqueue head, ignore */ |
4966 | if (poll->head == head) | |
4967 | return; | |
18bceab1 JA |
4968 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4969 | if (!poll) { | |
4970 | pt->error = -ENOMEM; | |
4971 | return; | |
4972 | } | |
58852d4d | 4973 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4974 | refcount_inc(&req->refs); |
4975 | poll->wait.private = req; | |
807abcb0 | 4976 | *poll_ptr = poll; |
18bceab1 JA |
4977 | } |
4978 | ||
4979 | pt->error = 0; | |
4980 | poll->head = head; | |
a31eb4a2 JX |
4981 | |
4982 | if (poll->events & EPOLLEXCLUSIVE) | |
4983 | add_wait_queue_exclusive(head, &poll->wait); | |
4984 | else | |
4985 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
4986 | } |
4987 | ||
4988 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
4989 | struct poll_table_struct *p) | |
4990 | { | |
4991 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 4992 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 4993 | |
807abcb0 | 4994 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
4995 | } |
4996 | ||
d7718a9d JA |
4997 | static void io_async_task_func(struct callback_head *cb) |
4998 | { | |
4999 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5000 | struct async_poll *apoll = req->apoll; | |
5001 | struct io_ring_ctx *ctx = req->ctx; | |
5002 | ||
5003 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5004 | ||
74ce6ce4 | 5005 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5006 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5007 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5008 | return; |
d7718a9d JA |
5009 | } |
5010 | ||
31067255 | 5011 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5012 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5013 | hash_del(&req->hash_node); |
2bae047e | 5014 | |
d4e7cd36 | 5015 | io_poll_remove_double(req); |
74ce6ce4 JA |
5016 | spin_unlock_irq(&ctx->completion_lock); |
5017 | ||
0be0b0e3 PB |
5018 | if (!READ_ONCE(apoll->poll.canceled)) |
5019 | __io_req_task_submit(req); | |
5020 | else | |
5021 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5022 | |
6d816e08 | 5023 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5024 | kfree(apoll->double_poll); |
31067255 | 5025 | kfree(apoll); |
d7718a9d JA |
5026 | } |
5027 | ||
5028 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5029 | void *key) | |
5030 | { | |
5031 | struct io_kiocb *req = wait->private; | |
5032 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5033 | ||
5034 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5035 | key_to_poll(key)); | |
5036 | ||
5037 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5038 | } | |
5039 | ||
5040 | static void io_poll_req_insert(struct io_kiocb *req) | |
5041 | { | |
5042 | struct io_ring_ctx *ctx = req->ctx; | |
5043 | struct hlist_head *list; | |
5044 | ||
5045 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5046 | hlist_add_head(&req->hash_node, list); | |
5047 | } | |
5048 | ||
5049 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5050 | struct io_poll_iocb *poll, | |
5051 | struct io_poll_table *ipt, __poll_t mask, | |
5052 | wait_queue_func_t wake_func) | |
5053 | __acquires(&ctx->completion_lock) | |
5054 | { | |
5055 | struct io_ring_ctx *ctx = req->ctx; | |
5056 | bool cancel = false; | |
5057 | ||
4d52f338 | 5058 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5059 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5060 | poll->file = req->file; |
18bceab1 | 5061 | poll->wait.private = req; |
d7718a9d JA |
5062 | |
5063 | ipt->pt._key = mask; | |
5064 | ipt->req = req; | |
5065 | ipt->error = -EINVAL; | |
5066 | ||
d7718a9d JA |
5067 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5068 | ||
5069 | spin_lock_irq(&ctx->completion_lock); | |
5070 | if (likely(poll->head)) { | |
5071 | spin_lock(&poll->head->lock); | |
5072 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5073 | if (ipt->error) | |
5074 | cancel = true; | |
5075 | ipt->error = 0; | |
5076 | mask = 0; | |
5077 | } | |
5078 | if (mask || ipt->error) | |
5079 | list_del_init(&poll->wait.entry); | |
5080 | else if (cancel) | |
5081 | WRITE_ONCE(poll->canceled, true); | |
5082 | else if (!poll->done) /* actually waiting for an event */ | |
5083 | io_poll_req_insert(req); | |
5084 | spin_unlock(&poll->head->lock); | |
5085 | } | |
5086 | ||
5087 | return mask; | |
5088 | } | |
5089 | ||
5090 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5091 | { | |
5092 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5093 | struct io_ring_ctx *ctx = req->ctx; | |
5094 | struct async_poll *apoll; | |
5095 | struct io_poll_table ipt; | |
5096 | __poll_t mask, ret; | |
9dab14b8 | 5097 | int rw; |
d7718a9d JA |
5098 | |
5099 | if (!req->file || !file_can_poll(req->file)) | |
5100 | return false; | |
24c74678 | 5101 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5102 | return false; |
9dab14b8 JA |
5103 | if (def->pollin) |
5104 | rw = READ; | |
5105 | else if (def->pollout) | |
5106 | rw = WRITE; | |
5107 | else | |
5108 | return false; | |
5109 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5110 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5111 | return false; |
5112 | ||
5113 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5114 | if (unlikely(!apoll)) | |
5115 | return false; | |
807abcb0 | 5116 | apoll->double_poll = NULL; |
d7718a9d JA |
5117 | |
5118 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5119 | req->apoll = apoll; |
d7718a9d | 5120 | |
8755d97a | 5121 | mask = 0; |
d7718a9d | 5122 | if (def->pollin) |
8755d97a | 5123 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5124 | if (def->pollout) |
5125 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5126 | |
5127 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5128 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5129 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5130 | mask &= ~POLLIN; | |
5131 | ||
d7718a9d JA |
5132 | mask |= POLLERR | POLLPRI; |
5133 | ||
5134 | ipt.pt._qproc = io_async_queue_proc; | |
5135 | ||
5136 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5137 | io_async_wake); | |
a36da65c | 5138 | if (ret || ipt.error) { |
d4e7cd36 | 5139 | io_poll_remove_double(req); |
d7718a9d | 5140 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5141 | kfree(apoll->double_poll); |
d7718a9d JA |
5142 | kfree(apoll); |
5143 | return false; | |
5144 | } | |
5145 | spin_unlock_irq(&ctx->completion_lock); | |
5146 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5147 | apoll->poll.events); | |
5148 | return true; | |
5149 | } | |
5150 | ||
5151 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5152 | struct io_poll_iocb *poll) | |
221c5eb2 | 5153 | { |
b41e9852 | 5154 | bool do_complete = false; |
221c5eb2 JA |
5155 | |
5156 | spin_lock(&poll->head->lock); | |
5157 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5158 | if (!list_empty(&poll->wait.entry)) { |
5159 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5160 | do_complete = true; |
221c5eb2 JA |
5161 | } |
5162 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5163 | hash_del(&req->hash_node); |
d7718a9d JA |
5164 | return do_complete; |
5165 | } | |
5166 | ||
5167 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5168 | { | |
5169 | bool do_complete; | |
5170 | ||
d4e7cd36 JA |
5171 | io_poll_remove_double(req); |
5172 | ||
d7718a9d JA |
5173 | if (req->opcode == IORING_OP_POLL_ADD) { |
5174 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5175 | } else { | |
3bfa5bcb JA |
5176 | struct async_poll *apoll = req->apoll; |
5177 | ||
d7718a9d | 5178 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5179 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5180 | if (do_complete) { | |
d7718a9d | 5181 | io_put_req(req); |
807abcb0 | 5182 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5183 | kfree(apoll); |
5184 | } | |
b1f573bd XW |
5185 | } |
5186 | ||
b41e9852 JA |
5187 | if (do_complete) { |
5188 | io_cqring_fill_event(req, -ECANCELED); | |
5189 | io_commit_cqring(req->ctx); | |
f254ac04 | 5190 | req_set_fail_links(req); |
216578e5 | 5191 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5192 | } |
5193 | ||
5194 | return do_complete; | |
221c5eb2 JA |
5195 | } |
5196 | ||
76e1b642 JA |
5197 | /* |
5198 | * Returns true if we found and killed one or more poll requests | |
5199 | */ | |
6b81928d PB |
5200 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5201 | struct files_struct *files) | |
221c5eb2 | 5202 | { |
78076bb6 | 5203 | struct hlist_node *tmp; |
221c5eb2 | 5204 | struct io_kiocb *req; |
8e2e1faf | 5205 | int posted = 0, i; |
221c5eb2 JA |
5206 | |
5207 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5208 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5209 | struct hlist_head *list; | |
5210 | ||
5211 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5212 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5213 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5214 | posted += io_poll_remove_one(req); |
5215 | } | |
221c5eb2 JA |
5216 | } |
5217 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5218 | |
8e2e1faf JA |
5219 | if (posted) |
5220 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5221 | |
5222 | return posted != 0; | |
221c5eb2 JA |
5223 | } |
5224 | ||
47f46768 JA |
5225 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5226 | { | |
78076bb6 | 5227 | struct hlist_head *list; |
47f46768 JA |
5228 | struct io_kiocb *req; |
5229 | ||
78076bb6 JA |
5230 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5231 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5232 | if (sqe_addr != req->user_data) |
5233 | continue; | |
5234 | if (io_poll_remove_one(req)) | |
eac406c6 | 5235 | return 0; |
b41e9852 | 5236 | return -EALREADY; |
47f46768 JA |
5237 | } |
5238 | ||
5239 | return -ENOENT; | |
5240 | } | |
5241 | ||
3529d8c2 JA |
5242 | static int io_poll_remove_prep(struct io_kiocb *req, |
5243 | const struct io_uring_sqe *sqe) | |
0969e783 | 5244 | { |
0969e783 JA |
5245 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5246 | return -EINVAL; | |
5247 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5248 | sqe->poll_events) | |
5249 | return -EINVAL; | |
5250 | ||
018043be | 5251 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5252 | return 0; |
5253 | } | |
5254 | ||
221c5eb2 JA |
5255 | /* |
5256 | * Find a running poll command that matches one specified in sqe->addr, | |
5257 | * and remove it if found. | |
5258 | */ | |
61e98203 | 5259 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5260 | { |
5261 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5262 | int ret; |
221c5eb2 | 5263 | |
221c5eb2 | 5264 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5265 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5266 | spin_unlock_irq(&ctx->completion_lock); |
5267 | ||
4e88d6e7 JA |
5268 | if (ret < 0) |
5269 | req_set_fail_links(req); | |
e1e16097 | 5270 | io_req_complete(req, ret); |
221c5eb2 JA |
5271 | return 0; |
5272 | } | |
5273 | ||
221c5eb2 JA |
5274 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5275 | void *key) | |
5276 | { | |
c2f2eb7d JA |
5277 | struct io_kiocb *req = wait->private; |
5278 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5279 | |
d7718a9d | 5280 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5281 | } |
5282 | ||
221c5eb2 JA |
5283 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5284 | struct poll_table_struct *p) | |
5285 | { | |
5286 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5287 | ||
e8c2bc1f | 5288 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5289 | } |
5290 | ||
3529d8c2 | 5291 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5292 | { |
5293 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5294 | u32 events; |
221c5eb2 JA |
5295 | |
5296 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5297 | return -EINVAL; | |
5298 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5299 | return -EINVAL; | |
5300 | ||
5769a351 JX |
5301 | events = READ_ONCE(sqe->poll32_events); |
5302 | #ifdef __BIG_ENDIAN | |
5303 | events = swahw32(events); | |
5304 | #endif | |
a31eb4a2 JX |
5305 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5306 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5307 | return 0; |
5308 | } | |
5309 | ||
61e98203 | 5310 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5311 | { |
5312 | struct io_poll_iocb *poll = &req->poll; | |
5313 | struct io_ring_ctx *ctx = req->ctx; | |
5314 | struct io_poll_table ipt; | |
0969e783 | 5315 | __poll_t mask; |
0969e783 | 5316 | |
d7718a9d | 5317 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5318 | |
d7718a9d JA |
5319 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5320 | io_poll_wake); | |
221c5eb2 | 5321 | |
8c838788 | 5322 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5323 | ipt.error = 0; |
b0dd8a41 | 5324 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5325 | } |
221c5eb2 JA |
5326 | spin_unlock_irq(&ctx->completion_lock); |
5327 | ||
8c838788 JA |
5328 | if (mask) { |
5329 | io_cqring_ev_posted(ctx); | |
014db007 | 5330 | io_put_req(req); |
221c5eb2 | 5331 | } |
8c838788 | 5332 | return ipt.error; |
221c5eb2 JA |
5333 | } |
5334 | ||
5262f567 JA |
5335 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5336 | { | |
ad8a48ac JA |
5337 | struct io_timeout_data *data = container_of(timer, |
5338 | struct io_timeout_data, timer); | |
5339 | struct io_kiocb *req = data->req; | |
5340 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5341 | unsigned long flags; |
5342 | ||
5262f567 | 5343 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5344 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5345 | atomic_set(&req->ctx->cq_timeouts, |
5346 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5347 | ||
78e19bbe | 5348 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5349 | io_commit_cqring(ctx); |
5350 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5351 | ||
5352 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5353 | req_set_fail_links(req); |
5262f567 JA |
5354 | io_put_req(req); |
5355 | return HRTIMER_NORESTART; | |
5356 | } | |
5357 | ||
fbd15848 PB |
5358 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5359 | __u64 user_data) | |
f254ac04 | 5360 | { |
fbd15848 | 5361 | struct io_timeout_data *io; |
47f46768 JA |
5362 | struct io_kiocb *req; |
5363 | int ret = -ENOENT; | |
f254ac04 | 5364 | |
135fcde8 | 5365 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5366 | if (user_data == req->user_data) { |
47f46768 JA |
5367 | ret = 0; |
5368 | break; | |
5369 | } | |
5370 | } | |
5371 | ||
5372 | if (ret == -ENOENT) | |
fbd15848 PB |
5373 | return ERR_PTR(ret); |
5374 | ||
5375 | io = req->async_data; | |
e8c2bc1f | 5376 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5377 | if (ret == -1) |
fbd15848 | 5378 | return ERR_PTR(-EALREADY); |
a71976f3 | 5379 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5380 | return req; |
5381 | } | |
47f46768 | 5382 | |
fbd15848 PB |
5383 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5384 | { | |
5385 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5386 | ||
5387 | if (IS_ERR(req)) | |
5388 | return PTR_ERR(req); | |
f254ac04 JA |
5389 | |
5390 | req_set_fail_links(req); | |
f254ac04 | 5391 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5392 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5393 | return 0; |
5394 | } | |
5395 | ||
9c8e11b3 PB |
5396 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5397 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5398 | { |
9c8e11b3 PB |
5399 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5400 | struct io_timeout_data *data; | |
47f46768 | 5401 | |
9c8e11b3 PB |
5402 | if (IS_ERR(req)) |
5403 | return PTR_ERR(req); | |
47f46768 | 5404 | |
9c8e11b3 PB |
5405 | req->timeout.off = 0; /* noseq */ |
5406 | data = req->async_data; | |
5407 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5408 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5409 | data->timer.function = io_timeout_fn; | |
5410 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5411 | return 0; | |
47f46768 JA |
5412 | } |
5413 | ||
3529d8c2 JA |
5414 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5415 | const struct io_uring_sqe *sqe) | |
b29472ee | 5416 | { |
9c8e11b3 PB |
5417 | struct io_timeout_rem *tr = &req->timeout_rem; |
5418 | ||
b29472ee JA |
5419 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5420 | return -EINVAL; | |
61710e43 DA |
5421 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5422 | return -EINVAL; | |
9c8e11b3 | 5423 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5424 | return -EINVAL; |
5425 | ||
9c8e11b3 PB |
5426 | tr->addr = READ_ONCE(sqe->addr); |
5427 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5428 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5429 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5430 | return -EINVAL; | |
5431 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5432 | return -EFAULT; | |
5433 | } else if (tr->flags) { | |
5434 | /* timeout removal doesn't support flags */ | |
b29472ee | 5435 | return -EINVAL; |
9c8e11b3 | 5436 | } |
b29472ee | 5437 | |
b29472ee JA |
5438 | return 0; |
5439 | } | |
5440 | ||
8662daec PB |
5441 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5442 | { | |
5443 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5444 | : HRTIMER_MODE_REL; | |
5445 | } | |
5446 | ||
11365043 JA |
5447 | /* |
5448 | * Remove or update an existing timeout command | |
5449 | */ | |
61e98203 | 5450 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5451 | { |
9c8e11b3 | 5452 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5453 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5454 | int ret; |
11365043 | 5455 | |
11365043 | 5456 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5457 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5458 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5459 | else |
5460 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5461 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5462 | |
47f46768 | 5463 | io_cqring_fill_event(req, ret); |
11365043 JA |
5464 | io_commit_cqring(ctx); |
5465 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5466 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5467 | if (ret < 0) |
5468 | req_set_fail_links(req); | |
ec9c02ad | 5469 | io_put_req(req); |
11365043 | 5470 | return 0; |
5262f567 JA |
5471 | } |
5472 | ||
3529d8c2 | 5473 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5474 | bool is_timeout_link) |
5262f567 | 5475 | { |
ad8a48ac | 5476 | struct io_timeout_data *data; |
a41525ab | 5477 | unsigned flags; |
56080b02 | 5478 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5479 | |
ad8a48ac | 5480 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5481 | return -EINVAL; |
ad8a48ac | 5482 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5483 | return -EINVAL; |
56080b02 | 5484 | if (off && is_timeout_link) |
2d28390a | 5485 | return -EINVAL; |
a41525ab JA |
5486 | flags = READ_ONCE(sqe->timeout_flags); |
5487 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5488 | return -EINVAL; |
bdf20073 | 5489 | |
bfe68a22 | 5490 | req->timeout.off = off; |
26a61679 | 5491 | |
e8c2bc1f | 5492 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5493 | return -ENOMEM; |
5494 | ||
e8c2bc1f | 5495 | data = req->async_data; |
ad8a48ac | 5496 | data->req = req; |
ad8a48ac JA |
5497 | |
5498 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5499 | return -EFAULT; |
5500 | ||
8662daec | 5501 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5502 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5503 | return 0; | |
5504 | } | |
5505 | ||
61e98203 | 5506 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5507 | { |
ad8a48ac | 5508 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5509 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5510 | struct list_head *entry; |
bfe68a22 | 5511 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5512 | |
733f5c95 | 5513 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5514 | |
5262f567 JA |
5515 | /* |
5516 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5517 | * timeout event to be satisfied. If it isn't set, then this is |
5518 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5519 | */ |
8eb7e2d0 | 5520 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5521 | entry = ctx->timeout_list.prev; |
5522 | goto add; | |
5523 | } | |
5262f567 | 5524 | |
bfe68a22 PB |
5525 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5526 | req->timeout.target_seq = tail + off; | |
5262f567 | 5527 | |
f010505b MDG |
5528 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5529 | * This is safe because ->completion_lock is held, and submissions | |
5530 | * and completions are never mixed in the same ->completion_lock section. | |
5531 | */ | |
5532 | ctx->cq_last_tm_flush = tail; | |
5533 | ||
5262f567 JA |
5534 | /* |
5535 | * Insertion sort, ensuring the first entry in the list is always | |
5536 | * the one we need first. | |
5537 | */ | |
5262f567 | 5538 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5539 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5540 | timeout.list); | |
5262f567 | 5541 | |
8eb7e2d0 | 5542 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5543 | continue; |
bfe68a22 PB |
5544 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5545 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5546 | break; |
5547 | } | |
93bd25bb | 5548 | add: |
135fcde8 | 5549 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5550 | data->timer.function = io_timeout_fn; |
5551 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5552 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5553 | return 0; |
5554 | } | |
5262f567 | 5555 | |
62755e35 JA |
5556 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5557 | { | |
5558 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5559 | ||
5560 | return req->user_data == (unsigned long) data; | |
5561 | } | |
5562 | ||
5aa75ed5 | 5563 | static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr) |
62755e35 | 5564 | { |
62755e35 | 5565 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5566 | int ret = 0; |
5567 | ||
5aa75ed5 JA |
5568 | if (!tctx->io_wq) |
5569 | return -ENOENT; | |
5570 | ||
5571 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false); | |
62755e35 JA |
5572 | switch (cancel_ret) { |
5573 | case IO_WQ_CANCEL_OK: | |
5574 | ret = 0; | |
5575 | break; | |
5576 | case IO_WQ_CANCEL_RUNNING: | |
5577 | ret = -EALREADY; | |
5578 | break; | |
5579 | case IO_WQ_CANCEL_NOTFOUND: | |
5580 | ret = -ENOENT; | |
5581 | break; | |
5582 | } | |
5583 | ||
e977d6d3 JA |
5584 | return ret; |
5585 | } | |
5586 | ||
47f46768 JA |
5587 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5588 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5589 | int success_ret) |
47f46768 JA |
5590 | { |
5591 | unsigned long flags; | |
5592 | int ret; | |
5593 | ||
5aa75ed5 JA |
5594 | ret = io_async_cancel_one(req->task->io_uring, |
5595 | (void *) (unsigned long) sqe_addr); | |
47f46768 JA |
5596 | if (ret != -ENOENT) { |
5597 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5598 | goto done; | |
5599 | } | |
5600 | ||
5601 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5602 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5603 | if (ret != -ENOENT) | |
5604 | goto done; | |
5605 | ret = io_poll_cancel(ctx, sqe_addr); | |
5606 | done: | |
b0dd8a41 JA |
5607 | if (!ret) |
5608 | ret = success_ret; | |
47f46768 JA |
5609 | io_cqring_fill_event(req, ret); |
5610 | io_commit_cqring(ctx); | |
5611 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5612 | io_cqring_ev_posted(ctx); | |
5613 | ||
4e88d6e7 JA |
5614 | if (ret < 0) |
5615 | req_set_fail_links(req); | |
014db007 | 5616 | io_put_req(req); |
47f46768 JA |
5617 | } |
5618 | ||
3529d8c2 JA |
5619 | static int io_async_cancel_prep(struct io_kiocb *req, |
5620 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5621 | { |
fbf23849 | 5622 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5623 | return -EINVAL; |
61710e43 DA |
5624 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5625 | return -EINVAL; | |
5626 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5627 | return -EINVAL; |
5628 | ||
fbf23849 JA |
5629 | req->cancel.addr = READ_ONCE(sqe->addr); |
5630 | return 0; | |
5631 | } | |
5632 | ||
61e98203 | 5633 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5634 | { |
5635 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5636 | |
014db007 | 5637 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5638 | return 0; |
5639 | } | |
5640 | ||
269bbe5f | 5641 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5642 | const struct io_uring_sqe *sqe) |
5643 | { | |
6ca56f84 JA |
5644 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5645 | return -EINVAL; | |
61710e43 DA |
5646 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5647 | return -EINVAL; | |
5648 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5649 | return -EINVAL; |
5650 | ||
269bbe5f BM |
5651 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5652 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5653 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5654 | return -EINVAL; |
269bbe5f | 5655 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5656 | return 0; |
5657 | } | |
5658 | ||
889fca73 | 5659 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5660 | { |
5661 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5662 | struct io_uring_rsrc_update up; |
05f3fb3c | 5663 | int ret; |
fbf23849 | 5664 | |
45d189c6 | 5665 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5666 | return -EAGAIN; |
05f3fb3c | 5667 | |
269bbe5f BM |
5668 | up.offset = req->rsrc_update.offset; |
5669 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5670 | |
5671 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5672 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5673 | mutex_unlock(&ctx->uring_lock); |
5674 | ||
5675 | if (ret < 0) | |
5676 | req_set_fail_links(req); | |
889fca73 | 5677 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5678 | return 0; |
5679 | } | |
5680 | ||
bfe76559 | 5681 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5682 | { |
d625c6ee | 5683 | switch (req->opcode) { |
e781573e | 5684 | case IORING_OP_NOP: |
bfe76559 | 5685 | return 0; |
f67676d1 JA |
5686 | case IORING_OP_READV: |
5687 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5688 | case IORING_OP_READ: |
bfe76559 | 5689 | return io_read_prep(req, sqe); |
f67676d1 JA |
5690 | case IORING_OP_WRITEV: |
5691 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5692 | case IORING_OP_WRITE: |
bfe76559 | 5693 | return io_write_prep(req, sqe); |
0969e783 | 5694 | case IORING_OP_POLL_ADD: |
bfe76559 | 5695 | return io_poll_add_prep(req, sqe); |
0969e783 | 5696 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5697 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5698 | case IORING_OP_FSYNC: |
1155c76a | 5699 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5700 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5701 | return io_sfr_prep(req, sqe); |
03b1230c | 5702 | case IORING_OP_SENDMSG: |
fddaface | 5703 | case IORING_OP_SEND: |
bfe76559 | 5704 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5705 | case IORING_OP_RECVMSG: |
fddaface | 5706 | case IORING_OP_RECV: |
bfe76559 | 5707 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5708 | case IORING_OP_CONNECT: |
bfe76559 | 5709 | return io_connect_prep(req, sqe); |
2d28390a | 5710 | case IORING_OP_TIMEOUT: |
bfe76559 | 5711 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5712 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5713 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5714 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5715 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5716 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5717 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5718 | case IORING_OP_ACCEPT: |
bfe76559 | 5719 | return io_accept_prep(req, sqe); |
d63d1b5e | 5720 | case IORING_OP_FALLOCATE: |
bfe76559 | 5721 | return io_fallocate_prep(req, sqe); |
15b71abe | 5722 | case IORING_OP_OPENAT: |
bfe76559 | 5723 | return io_openat_prep(req, sqe); |
b5dba59e | 5724 | case IORING_OP_CLOSE: |
bfe76559 | 5725 | return io_close_prep(req, sqe); |
05f3fb3c | 5726 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5727 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5728 | case IORING_OP_STATX: |
bfe76559 | 5729 | return io_statx_prep(req, sqe); |
4840e418 | 5730 | case IORING_OP_FADVISE: |
bfe76559 | 5731 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5732 | case IORING_OP_MADVISE: |
bfe76559 | 5733 | return io_madvise_prep(req, sqe); |
cebdb986 | 5734 | case IORING_OP_OPENAT2: |
bfe76559 | 5735 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5736 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5737 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5738 | case IORING_OP_SPLICE: |
bfe76559 | 5739 | return io_splice_prep(req, sqe); |
ddf0322d | 5740 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5741 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5742 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5743 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5744 | case IORING_OP_TEE: |
bfe76559 | 5745 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5746 | case IORING_OP_SHUTDOWN: |
5747 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5748 | case IORING_OP_RENAMEAT: |
5749 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5750 | case IORING_OP_UNLINKAT: |
5751 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5752 | } |
5753 | ||
bfe76559 PB |
5754 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5755 | req->opcode); | |
5756 | return-EINVAL; | |
5757 | } | |
5758 | ||
93642ef8 | 5759 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5760 | { |
93642ef8 PB |
5761 | switch (req->opcode) { |
5762 | case IORING_OP_READV: | |
5763 | case IORING_OP_READ_FIXED: | |
5764 | case IORING_OP_READ: | |
5765 | return io_rw_prep_async(req, READ); | |
5766 | case IORING_OP_WRITEV: | |
5767 | case IORING_OP_WRITE_FIXED: | |
5768 | case IORING_OP_WRITE: | |
5769 | return io_rw_prep_async(req, WRITE); | |
5770 | case IORING_OP_SENDMSG: | |
5771 | case IORING_OP_SEND: | |
5772 | return io_sendmsg_prep_async(req); | |
5773 | case IORING_OP_RECVMSG: | |
5774 | case IORING_OP_RECV: | |
5775 | return io_recvmsg_prep_async(req); | |
5776 | case IORING_OP_CONNECT: | |
5777 | return io_connect_prep_async(req); | |
5778 | } | |
5779 | return 0; | |
5780 | } | |
5781 | ||
be7053b7 | 5782 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5783 | { |
be7053b7 | 5784 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5785 | return 0; |
be7053b7 | 5786 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5787 | if (req->async_data) |
be7053b7 PB |
5788 | return 0; |
5789 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5790 | return -EAGAIN; |
be7053b7 | 5791 | return io_req_prep_async(req); |
f67676d1 JA |
5792 | } |
5793 | ||
9cf7c104 PB |
5794 | static u32 io_get_sequence(struct io_kiocb *req) |
5795 | { | |
5796 | struct io_kiocb *pos; | |
5797 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5798 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5799 | |
f2f87370 PB |
5800 | io_for_each_link(pos, req) |
5801 | nr_reqs++; | |
9cf7c104 PB |
5802 | |
5803 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5804 | return total_submitted - nr_reqs; | |
5805 | } | |
5806 | ||
be7053b7 | 5807 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5808 | { |
a197f664 | 5809 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5810 | struct io_defer_entry *de; |
f67676d1 | 5811 | int ret; |
9cf7c104 | 5812 | u32 seq; |
de0617e4 | 5813 | |
9d858b21 | 5814 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5815 | if (likely(list_empty_careful(&ctx->defer_list) && |
5816 | !(req->flags & REQ_F_IO_DRAIN))) | |
5817 | return 0; | |
5818 | ||
5819 | seq = io_get_sequence(req); | |
5820 | /* Still a chance to pass the sequence check */ | |
5821 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5822 | return 0; |
5823 | ||
be7053b7 PB |
5824 | ret = io_req_defer_prep(req); |
5825 | if (ret) | |
5826 | return ret; | |
cbdcb435 | 5827 | io_prep_async_link(req); |
27dc8338 PB |
5828 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5829 | if (!de) | |
5830 | return -ENOMEM; | |
2d28390a | 5831 | |
de0617e4 | 5832 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5833 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5834 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5835 | kfree(de); |
ae34817b PB |
5836 | io_queue_async_work(req); |
5837 | return -EIOCBQUEUED; | |
de0617e4 JA |
5838 | } |
5839 | ||
915967f6 | 5840 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5841 | de->req = req; |
9cf7c104 | 5842 | de->seq = seq; |
27dc8338 | 5843 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5844 | spin_unlock_irq(&ctx->completion_lock); |
5845 | return -EIOCBQUEUED; | |
5846 | } | |
5847 | ||
3ca405eb | 5848 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5849 | { |
0e1b6fe3 PB |
5850 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5851 | switch (req->opcode) { | |
5852 | case IORING_OP_READV: | |
5853 | case IORING_OP_READ_FIXED: | |
5854 | case IORING_OP_READ: | |
bcda7baa | 5855 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5856 | break; |
5857 | case IORING_OP_RECVMSG: | |
5858 | case IORING_OP_RECV: | |
bcda7baa | 5859 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5860 | break; |
5861 | } | |
5862 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5863 | } |
5864 | ||
0e1b6fe3 PB |
5865 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5866 | switch (req->opcode) { | |
5867 | case IORING_OP_READV: | |
5868 | case IORING_OP_READ_FIXED: | |
5869 | case IORING_OP_READ: | |
5870 | case IORING_OP_WRITEV: | |
5871 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5872 | case IORING_OP_WRITE: { |
5873 | struct io_async_rw *io = req->async_data; | |
5874 | if (io->free_iovec) | |
5875 | kfree(io->free_iovec); | |
0e1b6fe3 | 5876 | break; |
e8c2bc1f | 5877 | } |
0e1b6fe3 | 5878 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5879 | case IORING_OP_SENDMSG: { |
5880 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5881 | |
5882 | kfree(io->free_iov); | |
0e1b6fe3 | 5883 | break; |
e8c2bc1f | 5884 | } |
0e1b6fe3 PB |
5885 | case IORING_OP_SPLICE: |
5886 | case IORING_OP_TEE: | |
5887 | io_put_file(req, req->splice.file_in, | |
5888 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5889 | break; | |
f3cd4850 JA |
5890 | case IORING_OP_OPENAT: |
5891 | case IORING_OP_OPENAT2: | |
5892 | if (req->open.filename) | |
5893 | putname(req->open.filename); | |
5894 | break; | |
80a261fd JA |
5895 | case IORING_OP_RENAMEAT: |
5896 | putname(req->rename.oldpath); | |
5897 | putname(req->rename.newpath); | |
5898 | break; | |
14a1143b JA |
5899 | case IORING_OP_UNLINKAT: |
5900 | putname(req->unlink.filename); | |
5901 | break; | |
0e1b6fe3 PB |
5902 | } |
5903 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5904 | } |
99bc4c38 PB |
5905 | } |
5906 | ||
889fca73 | 5907 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5908 | { |
a197f664 | 5909 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5910 | const struct cred *creds = NULL; |
d625c6ee | 5911 | int ret; |
2b188cc1 | 5912 | |
5730b27e JA |
5913 | if (req->work.personality) { |
5914 | const struct cred *new_creds; | |
5915 | ||
5916 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5917 | mutex_lock(&ctx->uring_lock); | |
5918 | new_creds = idr_find(&ctx->personality_idr, req->work.personality); | |
5919 | if (!(issue_flags & IO_URING_F_NONBLOCK)) | |
5920 | mutex_unlock(&ctx->uring_lock); | |
5921 | if (!new_creds) | |
5922 | return -EINVAL; | |
5923 | creds = override_creds(new_creds); | |
5924 | } | |
5925 | ||
d625c6ee | 5926 | switch (req->opcode) { |
2b188cc1 | 5927 | case IORING_OP_NOP: |
889fca73 | 5928 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5929 | break; |
5930 | case IORING_OP_READV: | |
edafccee | 5931 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5932 | case IORING_OP_READ: |
889fca73 | 5933 | ret = io_read(req, issue_flags); |
edafccee | 5934 | break; |
3529d8c2 | 5935 | case IORING_OP_WRITEV: |
edafccee | 5936 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5937 | case IORING_OP_WRITE: |
889fca73 | 5938 | ret = io_write(req, issue_flags); |
2b188cc1 | 5939 | break; |
c992fe29 | 5940 | case IORING_OP_FSYNC: |
45d189c6 | 5941 | ret = io_fsync(req, issue_flags); |
c992fe29 | 5942 | break; |
221c5eb2 | 5943 | case IORING_OP_POLL_ADD: |
61e98203 | 5944 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
5945 | break; |
5946 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 5947 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 5948 | break; |
5d17b4a4 | 5949 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 5950 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 5951 | break; |
0fa03c62 | 5952 | case IORING_OP_SENDMSG: |
889fca73 | 5953 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 5954 | break; |
fddaface | 5955 | case IORING_OP_SEND: |
889fca73 | 5956 | ret = io_send(req, issue_flags); |
0fa03c62 | 5957 | break; |
aa1fa28f | 5958 | case IORING_OP_RECVMSG: |
889fca73 | 5959 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 5960 | break; |
fddaface | 5961 | case IORING_OP_RECV: |
889fca73 | 5962 | ret = io_recv(req, issue_flags); |
aa1fa28f | 5963 | break; |
5262f567 | 5964 | case IORING_OP_TIMEOUT: |
61e98203 | 5965 | ret = io_timeout(req, issue_flags); |
5262f567 | 5966 | break; |
11365043 | 5967 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 5968 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 5969 | break; |
17f2fe35 | 5970 | case IORING_OP_ACCEPT: |
889fca73 | 5971 | ret = io_accept(req, issue_flags); |
17f2fe35 | 5972 | break; |
f8e85cf2 | 5973 | case IORING_OP_CONNECT: |
889fca73 | 5974 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 5975 | break; |
62755e35 | 5976 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 5977 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 5978 | break; |
d63d1b5e | 5979 | case IORING_OP_FALLOCATE: |
45d189c6 | 5980 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 5981 | break; |
15b71abe | 5982 | case IORING_OP_OPENAT: |
45d189c6 | 5983 | ret = io_openat(req, issue_flags); |
15b71abe | 5984 | break; |
b5dba59e | 5985 | case IORING_OP_CLOSE: |
889fca73 | 5986 | ret = io_close(req, issue_flags); |
b5dba59e | 5987 | break; |
05f3fb3c | 5988 | case IORING_OP_FILES_UPDATE: |
889fca73 | 5989 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 5990 | break; |
eddc7ef5 | 5991 | case IORING_OP_STATX: |
45d189c6 | 5992 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 5993 | break; |
4840e418 | 5994 | case IORING_OP_FADVISE: |
45d189c6 | 5995 | ret = io_fadvise(req, issue_flags); |
4840e418 | 5996 | break; |
c1ca757b | 5997 | case IORING_OP_MADVISE: |
45d189c6 | 5998 | ret = io_madvise(req, issue_flags); |
c1ca757b | 5999 | break; |
cebdb986 | 6000 | case IORING_OP_OPENAT2: |
45d189c6 | 6001 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6002 | break; |
3e4827b0 | 6003 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6004 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6005 | break; |
7d67af2c | 6006 | case IORING_OP_SPLICE: |
45d189c6 | 6007 | ret = io_splice(req, issue_flags); |
7d67af2c | 6008 | break; |
ddf0322d | 6009 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6010 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6011 | break; |
067524e9 | 6012 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6013 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6014 | break; |
f2a8d5c7 | 6015 | case IORING_OP_TEE: |
45d189c6 | 6016 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6017 | break; |
36f4fa68 | 6018 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6019 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6020 | break; |
80a261fd | 6021 | case IORING_OP_RENAMEAT: |
45d189c6 | 6022 | ret = io_renameat(req, issue_flags); |
80a261fd | 6023 | break; |
14a1143b | 6024 | case IORING_OP_UNLINKAT: |
45d189c6 | 6025 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6026 | break; |
2b188cc1 JA |
6027 | default: |
6028 | ret = -EINVAL; | |
6029 | break; | |
6030 | } | |
6031 | ||
5730b27e JA |
6032 | if (creds) |
6033 | revert_creds(creds); | |
6034 | ||
def596e9 JA |
6035 | if (ret) |
6036 | return ret; | |
6037 | ||
b532576e JA |
6038 | /* If the op doesn't have a file, we're not polling for it */ |
6039 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6040 | const bool in_async = io_wq_current_is_worker(); |
6041 | ||
11ba820b JA |
6042 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6043 | if (in_async) | |
6044 | mutex_lock(&ctx->uring_lock); | |
6045 | ||
2e9dbe90 | 6046 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6047 | |
6048 | if (in_async) | |
6049 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6050 | } |
6051 | ||
6052 | return 0; | |
2b188cc1 JA |
6053 | } |
6054 | ||
5280f7e5 | 6055 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6056 | { |
6057 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6058 | struct io_kiocb *timeout; |
561fb04a | 6059 | int ret = 0; |
2b188cc1 | 6060 | |
6df1db6b PB |
6061 | timeout = io_prep_linked_timeout(req); |
6062 | if (timeout) | |
6063 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6064 | |
4014d943 | 6065 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6066 | ret = -ECANCELED; |
31b51510 | 6067 | |
561fb04a | 6068 | if (!ret) { |
561fb04a | 6069 | do { |
889fca73 | 6070 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6071 | /* |
6072 | * We can get EAGAIN for polled IO even though we're | |
6073 | * forcing a sync submission from here, since we can't | |
6074 | * wait for request slots on the block side. | |
6075 | */ | |
6076 | if (ret != -EAGAIN) | |
6077 | break; | |
6078 | cond_resched(); | |
6079 | } while (1); | |
6080 | } | |
31b51510 | 6081 | |
a3df7698 | 6082 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6083 | if (ret) { |
a3df7698 PB |
6084 | /* io-wq is going to take one down */ |
6085 | refcount_inc(&req->refs); | |
6086 | io_req_task_queue_fail(req, ret); | |
edafccee | 6087 | } |
2b188cc1 JA |
6088 | } |
6089 | ||
65e19f54 JA |
6090 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6091 | int index) | |
6092 | { | |
269bbe5f | 6093 | struct fixed_rsrc_table *table; |
65e19f54 | 6094 | |
05f3fb3c | 6095 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6096 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6097 | } |
6098 | ||
8371adf5 PB |
6099 | static struct file *io_file_get(struct io_submit_state *state, |
6100 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6101 | { |
a197f664 | 6102 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6103 | struct file *file; |
09bb8394 | 6104 | |
8da11c19 | 6105 | if (fixed) { |
479f517b | 6106 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6107 | return NULL; |
b7620121 | 6108 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6109 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6110 | io_set_resource_node(req); |
09bb8394 | 6111 | } else { |
c826bd7a | 6112 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6113 | file = __io_file_get(state, fd); |
09bb8394 JA |
6114 | } |
6115 | ||
ce3d5aae PB |
6116 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6117 | io_req_track_inflight(req); | |
8371adf5 | 6118 | return file; |
09bb8394 JA |
6119 | } |
6120 | ||
2665abfd | 6121 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6122 | { |
ad8a48ac JA |
6123 | struct io_timeout_data *data = container_of(timer, |
6124 | struct io_timeout_data, timer); | |
90cd7e42 | 6125 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6126 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6127 | unsigned long flags; |
2665abfd JA |
6128 | |
6129 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6130 | prev = req->timeout.head; |
6131 | req->timeout.head = NULL; | |
2665abfd JA |
6132 | |
6133 | /* | |
6134 | * We don't expect the list to be empty, that will only happen if we | |
6135 | * race with the completion of the linked work. | |
6136 | */ | |
90cd7e42 | 6137 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6138 | io_remove_next_linked(prev); |
90cd7e42 PB |
6139 | else |
6140 | prev = NULL; | |
2665abfd JA |
6141 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6142 | ||
6143 | if (prev) { | |
4e88d6e7 | 6144 | req_set_fail_links(prev); |
014db007 | 6145 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6146 | io_put_req_deferred(prev, 1); |
47f46768 | 6147 | } else { |
9ae1f8dd PB |
6148 | io_req_complete_post(req, -ETIME, 0); |
6149 | io_put_req_deferred(req, 1); | |
2665abfd | 6150 | } |
2665abfd JA |
6151 | return HRTIMER_NORESTART; |
6152 | } | |
6153 | ||
7271ef3a | 6154 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6155 | { |
76a46e06 | 6156 | /* |
f2f87370 PB |
6157 | * If the back reference is NULL, then our linked request finished |
6158 | * before we got a chance to setup the timer | |
76a46e06 | 6159 | */ |
90cd7e42 | 6160 | if (req->timeout.head) { |
e8c2bc1f | 6161 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6162 | |
ad8a48ac JA |
6163 | data->timer.function = io_link_timeout_fn; |
6164 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6165 | data->mode); | |
2665abfd | 6166 | } |
7271ef3a JA |
6167 | } |
6168 | ||
6169 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6170 | { | |
6171 | struct io_ring_ctx *ctx = req->ctx; | |
6172 | ||
6173 | spin_lock_irq(&ctx->completion_lock); | |
6174 | __io_queue_linked_timeout(req); | |
76a46e06 | 6175 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6176 | |
2665abfd | 6177 | /* drop submission reference */ |
76a46e06 JA |
6178 | io_put_req(req); |
6179 | } | |
2665abfd | 6180 | |
ad8a48ac | 6181 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6182 | { |
f2f87370 | 6183 | struct io_kiocb *nxt = req->link; |
2665abfd | 6184 | |
f2f87370 PB |
6185 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6186 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6187 | return NULL; |
2665abfd | 6188 | |
90cd7e42 | 6189 | nxt->timeout.head = req; |
900fad45 | 6190 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6191 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6192 | return nxt; |
2665abfd JA |
6193 | } |
6194 | ||
c5eef2b9 | 6195 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6196 | { |
d3d7298d | 6197 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6198 | int ret; |
2b188cc1 | 6199 | |
c5eef2b9 | 6200 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6201 | |
491381ce JA |
6202 | /* |
6203 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6204 | * doesn't support non-blocking read/write attempts | |
6205 | */ | |
24c74678 | 6206 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6207 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6208 | /* |
6209 | * Queued up for async execution, worker will release | |
6210 | * submit reference when the iocb is actually submitted. | |
6211 | */ | |
6212 | io_queue_async_work(req); | |
2b188cc1 | 6213 | } |
0d63c148 PB |
6214 | } else if (likely(!ret)) { |
6215 | /* drop submission reference */ | |
e342c807 | 6216 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6217 | struct io_ring_ctx *ctx = req->ctx; |
6218 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6219 | |
6dd0be1e | 6220 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6221 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6222 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6223 | } else { |
d3d7298d | 6224 | io_put_req(req); |
0d63c148 PB |
6225 | } |
6226 | } else { | |
4e88d6e7 | 6227 | req_set_fail_links(req); |
e65ef56d | 6228 | io_put_req(req); |
e1e16097 | 6229 | io_req_complete(req, ret); |
9e645e11 | 6230 | } |
d3d7298d PB |
6231 | if (linked_timeout) |
6232 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6233 | } |
6234 | ||
be7053b7 | 6235 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6236 | { |
6237 | int ret; | |
6238 | ||
be7053b7 | 6239 | ret = io_req_defer(req); |
4fe2c963 JL |
6240 | if (ret) { |
6241 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6242 | fail_req: |
4e88d6e7 | 6243 | req_set_fail_links(req); |
e1e16097 JA |
6244 | io_put_req(req); |
6245 | io_req_complete(req, ret); | |
4fe2c963 | 6246 | } |
2550878f | 6247 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6248 | ret = io_req_defer_prep(req); |
6249 | if (unlikely(ret)) | |
6250 | goto fail_req; | |
ce35a47a JA |
6251 | io_queue_async_work(req); |
6252 | } else { | |
c5eef2b9 | 6253 | __io_queue_sqe(req); |
ce35a47a | 6254 | } |
4fe2c963 JL |
6255 | } |
6256 | ||
b16fed66 PB |
6257 | /* |
6258 | * Check SQE restrictions (opcode and flags). | |
6259 | * | |
6260 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6261 | */ | |
6262 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6263 | struct io_kiocb *req, | |
6264 | unsigned int sqe_flags) | |
4fe2c963 | 6265 | { |
b16fed66 PB |
6266 | if (!ctx->restricted) |
6267 | return true; | |
6268 | ||
6269 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6270 | return false; | |
6271 | ||
6272 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6273 | ctx->restrictions.sqe_flags_required) | |
6274 | return false; | |
6275 | ||
6276 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6277 | ctx->restrictions.sqe_flags_required)) | |
6278 | return false; | |
6279 | ||
6280 | return true; | |
4fe2c963 JL |
6281 | } |
6282 | ||
b16fed66 PB |
6283 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6284 | const struct io_uring_sqe *sqe) | |
6285 | { | |
6286 | struct io_submit_state *state; | |
6287 | unsigned int sqe_flags; | |
5730b27e | 6288 | int ret = 0; |
b16fed66 PB |
6289 | |
6290 | req->opcode = READ_ONCE(sqe->opcode); | |
6291 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6292 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6293 | req->user_data = READ_ONCE(sqe->user_data); | |
6294 | req->async_data = NULL; | |
6295 | req->file = NULL; | |
6296 | req->ctx = ctx; | |
6297 | req->link = NULL; | |
6298 | req->fixed_rsrc_refs = NULL; | |
6299 | /* one is dropped after submission, the other at completion */ | |
6300 | refcount_set(&req->refs, 2); | |
6301 | req->task = current; | |
6302 | req->result = 0; | |
6303 | ||
6304 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6305 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6306 | req->flags = 0; | |
b16fed66 | 6307 | return -EINVAL; |
ebf4a5db | 6308 | } |
b16fed66 PB |
6309 | |
6310 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6311 | return -EINVAL; | |
6312 | ||
b16fed66 PB |
6313 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6314 | return -EACCES; | |
6315 | ||
6316 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6317 | !io_op_defs[req->opcode].buffer_select) | |
6318 | return -EOPNOTSUPP; | |
863e0560 | 6319 | |
5730b27e JA |
6320 | req->work.list.next = NULL; |
6321 | req->work.flags = 0; | |
6322 | req->work.personality = READ_ONCE(sqe->personality); | |
b16fed66 PB |
6323 | state = &ctx->submit_state; |
6324 | ||
6325 | /* | |
6326 | * Plug now if we have more than 1 IO left after this, and the target | |
6327 | * is potentially a read/write to block based storage. | |
6328 | */ | |
6329 | if (!state->plug_started && state->ios_left > 1 && | |
6330 | io_op_defs[req->opcode].plug) { | |
6331 | blk_start_plug(&state->plug); | |
6332 | state->plug_started = true; | |
6333 | } | |
6334 | ||
6335 | if (io_op_defs[req->opcode].needs_file) { | |
6336 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6337 | ||
6338 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6339 | if (unlikely(!req->file)) | |
6340 | ret = -EBADF; | |
6341 | } | |
6342 | ||
6343 | state->ios_left--; | |
6344 | return ret; | |
6345 | } | |
6346 | ||
a6b8cadc | 6347 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6348 | const struct io_uring_sqe *sqe) |
9e645e11 | 6349 | { |
a1ab7b35 | 6350 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6351 | int ret; |
9e645e11 | 6352 | |
a6b8cadc PB |
6353 | ret = io_init_req(ctx, req, sqe); |
6354 | if (unlikely(ret)) { | |
6355 | fail_req: | |
6356 | io_put_req(req); | |
6357 | io_req_complete(req, ret); | |
de59bc10 PB |
6358 | if (link->head) { |
6359 | /* fail even hard links since we don't submit */ | |
cf109604 | 6360 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6361 | io_put_req(link->head); |
6362 | io_req_complete(link->head, -ECANCELED); | |
6363 | link->head = NULL; | |
6364 | } | |
a6b8cadc PB |
6365 | return ret; |
6366 | } | |
be7053b7 PB |
6367 | ret = io_req_prep(req, sqe); |
6368 | if (unlikely(ret)) | |
6369 | goto fail_req; | |
a6b8cadc | 6370 | |
be7053b7 | 6371 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6372 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6373 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6374 | ||
9e645e11 JA |
6375 | /* |
6376 | * If we already have a head request, queue this one for async | |
6377 | * submittal once the head completes. If we don't have a head but | |
6378 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6379 | * submitted sync once the chain is complete. If none of those | |
6380 | * conditions are true (normal request), then just queue it. | |
6381 | */ | |
863e0560 PB |
6382 | if (link->head) { |
6383 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6384 | |
8cdf2193 PB |
6385 | /* |
6386 | * Taking sequential execution of a link, draining both sides | |
6387 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6388 | * requests in the link. So, it drains the head and the | |
6389 | * next after the link request. The last one is done via | |
6390 | * drain_next flag to persist the effect across calls. | |
6391 | */ | |
ef4ff581 | 6392 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6393 | head->flags |= REQ_F_IO_DRAIN; |
6394 | ctx->drain_next = 1; | |
6395 | } | |
be7053b7 | 6396 | ret = io_req_defer_prep(req); |
cf109604 | 6397 | if (unlikely(ret)) |
a6b8cadc | 6398 | goto fail_req; |
9d76377f | 6399 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6400 | link->last->link = req; |
863e0560 | 6401 | link->last = req; |
32fe525b PB |
6402 | |
6403 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6404 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6405 | io_queue_sqe(head); |
863e0560 | 6406 | link->head = NULL; |
32fe525b | 6407 | } |
9e645e11 | 6408 | } else { |
711be031 PB |
6409 | if (unlikely(ctx->drain_next)) { |
6410 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6411 | ctx->drain_next = 0; |
711be031 | 6412 | } |
ef4ff581 | 6413 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6414 | link->head = req; |
6415 | link->last = req; | |
711be031 | 6416 | } else { |
be7053b7 | 6417 | io_queue_sqe(req); |
711be031 | 6418 | } |
9e645e11 | 6419 | } |
2e6e1fde | 6420 | |
1d4240cc | 6421 | return 0; |
9e645e11 JA |
6422 | } |
6423 | ||
9a56a232 JA |
6424 | /* |
6425 | * Batched submission is done, ensure local IO is flushed out. | |
6426 | */ | |
ba88ff11 PB |
6427 | static void io_submit_state_end(struct io_submit_state *state, |
6428 | struct io_ring_ctx *ctx) | |
9a56a232 | 6429 | { |
a1ab7b35 | 6430 | if (state->link.head) |
de59bc10 | 6431 | io_queue_sqe(state->link.head); |
6dd0be1e | 6432 | if (state->comp.nr) |
ba88ff11 | 6433 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6434 | if (state->plug_started) |
6435 | blk_finish_plug(&state->plug); | |
9f13c35b | 6436 | io_state_file_put(state); |
9a56a232 JA |
6437 | } |
6438 | ||
6439 | /* | |
6440 | * Start submission side cache. | |
6441 | */ | |
6442 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6443 | unsigned int max_ios) |
9a56a232 | 6444 | { |
27926b68 | 6445 | state->plug_started = false; |
9a56a232 | 6446 | state->ios_left = max_ios; |
a1ab7b35 PB |
6447 | /* set only head, no need to init link_last in advance */ |
6448 | state->link.head = NULL; | |
9a56a232 JA |
6449 | } |
6450 | ||
2b188cc1 JA |
6451 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6452 | { | |
75b28aff | 6453 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6454 | |
caf582c6 PB |
6455 | /* |
6456 | * Ensure any loads from the SQEs are done at this point, | |
6457 | * since once we write the new head, the application could | |
6458 | * write new data to them. | |
6459 | */ | |
6460 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6461 | } |
6462 | ||
2b188cc1 | 6463 | /* |
3529d8c2 | 6464 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6465 | * that is mapped by userspace. This means that care needs to be taken to |
6466 | * ensure that reads are stable, as we cannot rely on userspace always | |
6467 | * being a good citizen. If members of the sqe are validated and then later | |
6468 | * used, it's important that those reads are done through READ_ONCE() to | |
6469 | * prevent a re-load down the line. | |
6470 | */ | |
709b302f | 6471 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6472 | { |
75b28aff | 6473 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6474 | unsigned head; |
6475 | ||
6476 | /* | |
6477 | * The cached sq head (or cq tail) serves two purposes: | |
6478 | * | |
6479 | * 1) allows us to batch the cost of updating the user visible | |
6480 | * head updates. | |
6481 | * 2) allows the kernel side to track the head on its own, even | |
6482 | * though the application is the one updating it. | |
6483 | */ | |
4fccfcbb | 6484 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6485 | if (likely(head < ctx->sq_entries)) |
6486 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6487 | |
6488 | /* drop invalid entries */ | |
498ccd9e | 6489 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6490 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6491 | return NULL; |
6492 | } | |
6493 | ||
0f212204 | 6494 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6495 | { |
46c4e16a | 6496 | int submitted = 0; |
6c271ce2 | 6497 | |
c4a2ed72 | 6498 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6499 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6500 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6501 | return -EBUSY; |
6502 | } | |
6c271ce2 | 6503 | |
ee7d46d9 PB |
6504 | /* make sure SQ entry isn't read before tail */ |
6505 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6506 | |
2b85edfc PB |
6507 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6508 | return -EAGAIN; | |
6c271ce2 | 6509 | |
d8a6df10 | 6510 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6511 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6512 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6513 | |
46c4e16a | 6514 | while (submitted < nr) { |
3529d8c2 | 6515 | const struct io_uring_sqe *sqe; |
196be95c | 6516 | struct io_kiocb *req; |
fb5ccc98 | 6517 | |
258b29a9 | 6518 | req = io_alloc_req(ctx); |
196be95c PB |
6519 | if (unlikely(!req)) { |
6520 | if (!submitted) | |
6521 | submitted = -EAGAIN; | |
fb5ccc98 | 6522 | break; |
196be95c | 6523 | } |
4fccfcbb PB |
6524 | sqe = io_get_sqe(ctx); |
6525 | if (unlikely(!sqe)) { | |
6526 | kmem_cache_free(req_cachep, req); | |
6527 | break; | |
6528 | } | |
d3656344 JA |
6529 | /* will complete beyond this point, count as submitted */ |
6530 | submitted++; | |
a1ab7b35 | 6531 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6532 | break; |
6c271ce2 JA |
6533 | } |
6534 | ||
9466f437 PB |
6535 | if (unlikely(submitted != nr)) { |
6536 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6537 | struct io_uring_task *tctx = current->io_uring; |
6538 | int unused = nr - ref_used; | |
9466f437 | 6539 | |
d8a6df10 JA |
6540 | percpu_ref_put_many(&ctx->refs, unused); |
6541 | percpu_counter_sub(&tctx->inflight, unused); | |
6542 | put_task_struct_many(current, unused); | |
9466f437 | 6543 | } |
6c271ce2 | 6544 | |
a1ab7b35 | 6545 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6546 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6547 | io_commit_sqring(ctx); | |
6548 | ||
6c271ce2 JA |
6549 | return submitted; |
6550 | } | |
6551 | ||
23b3628e XW |
6552 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6553 | { | |
6554 | /* Tell userspace we may need a wakeup call */ | |
6555 | spin_lock_irq(&ctx->completion_lock); | |
6556 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6557 | spin_unlock_irq(&ctx->completion_lock); | |
6558 | } | |
6559 | ||
6560 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6561 | { | |
6562 | spin_lock_irq(&ctx->completion_lock); | |
6563 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6564 | spin_unlock_irq(&ctx->completion_lock); | |
6565 | } | |
6566 | ||
08369246 | 6567 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6568 | { |
c8d1ba58 | 6569 | unsigned int to_submit; |
bdcd3eab | 6570 | int ret = 0; |
6c271ce2 | 6571 | |
c8d1ba58 | 6572 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6573 | /* if we're handling multiple rings, cap submit size for fairness */ |
6574 | if (cap_entries && to_submit > 8) | |
6575 | to_submit = 8; | |
6576 | ||
906a3c6f | 6577 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6578 | unsigned nr_events = 0; |
a4c0b3de | 6579 | |
c8d1ba58 | 6580 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6581 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6582 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6583 | |
70aacfe6 | 6584 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs))) |
08369246 | 6585 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6586 | mutex_unlock(&ctx->uring_lock); |
6587 | } | |
6c271ce2 | 6588 | |
90554200 JA |
6589 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6590 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6591 | |
08369246 XW |
6592 | return ret; |
6593 | } | |
6c271ce2 | 6594 | |
08369246 XW |
6595 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6596 | { | |
6597 | struct io_ring_ctx *ctx; | |
6598 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6599 | |
08369246 XW |
6600 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6601 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6602 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6603 | } |
c1edbf5f | 6604 | |
08369246 | 6605 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6606 | } |
6c271ce2 | 6607 | |
69fb2131 JA |
6608 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6609 | { | |
6610 | struct io_ring_ctx *ctx; | |
6611 | ||
6612 | while (!list_empty(&sqd->ctx_new_list)) { | |
6613 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
6614 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
6615 | complete(&ctx->sq_thread_comp); | |
6616 | } | |
08369246 XW |
6617 | |
6618 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
6619 | } |
6620 | ||
37d1e2e3 JA |
6621 | static bool io_sq_thread_should_stop(struct io_sq_data *sqd) |
6622 | { | |
6623 | return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
6624 | } | |
6625 | ||
6626 | static bool io_sq_thread_should_park(struct io_sq_data *sqd) | |
6627 | { | |
6628 | return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
6629 | } | |
6630 | ||
6631 | static void io_sq_thread_parkme(struct io_sq_data *sqd) | |
6632 | { | |
6633 | for (;;) { | |
6634 | /* | |
6635 | * TASK_PARKED is a special state; we must serialize against | |
6636 | * possible pending wakeups to avoid store-store collisions on | |
6637 | * task->state. | |
6638 | * | |
6639 | * Such a collision might possibly result in the task state | |
6640 | * changin from TASK_PARKED and us failing the | |
6641 | * wait_task_inactive() in kthread_park(). | |
6642 | */ | |
6643 | set_special_state(TASK_PARKED); | |
6644 | if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) | |
6645 | break; | |
6646 | ||
6647 | /* | |
6648 | * Thread is going to call schedule(), do not preempt it, | |
6649 | * or the caller of kthread_park() may spend more time in | |
6650 | * wait_task_inactive(). | |
6651 | */ | |
6652 | preempt_disable(); | |
6653 | complete(&sqd->completion); | |
6654 | schedule_preempt_disabled(); | |
6655 | preempt_enable(); | |
6656 | } | |
6657 | __set_current_state(TASK_RUNNING); | |
6658 | } | |
6659 | ||
c8d1ba58 JA |
6660 | static int io_sq_thread(void *data) |
6661 | { | |
69fb2131 JA |
6662 | struct io_sq_data *sqd = data; |
6663 | struct io_ring_ctx *ctx; | |
a0d9205f | 6664 | unsigned long timeout = 0; |
37d1e2e3 | 6665 | char buf[TASK_COMM_LEN]; |
08369246 | 6666 | DEFINE_WAIT(wait); |
6c271ce2 | 6667 | |
37d1e2e3 JA |
6668 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6669 | set_task_comm(current, buf); | |
6670 | sqd->thread = current; | |
6671 | current->pf_io_worker = NULL; | |
6672 | ||
6673 | if (sqd->sq_cpu != -1) | |
6674 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6675 | else | |
6676 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6677 | current->flags |= PF_NO_SETAFFINITY; | |
6678 | ||
6679 | complete(&sqd->completion); | |
6c271ce2 | 6680 | |
37d1e2e3 JA |
6681 | wait_for_completion(&sqd->startup); |
6682 | ||
6683 | while (!io_sq_thread_should_stop(sqd)) { | |
08369246 XW |
6684 | int ret; |
6685 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
6686 | |
6687 | /* | |
69fb2131 | 6688 | * Any changes to the sqd lists are synchronized through the |
37d1e2e3 | 6689 | * thread parking. This synchronizes the thread vs users, |
69fb2131 | 6690 | * the users are synchronized on the sqd->ctx_lock. |
c1edbf5f | 6691 | */ |
37d1e2e3 JA |
6692 | if (io_sq_thread_should_park(sqd)) { |
6693 | io_sq_thread_parkme(sqd); | |
6694 | continue; | |
65b2b213 | 6695 | } |
08369246 | 6696 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 6697 | io_sqd_init_new(sqd); |
08369246 XW |
6698 | timeout = jiffies + sqd->sq_thread_idle; |
6699 | } | |
37d1e2e3 JA |
6700 | if (fatal_signal_pending(current)) |
6701 | break; | |
08369246 | 6702 | sqt_spin = false; |
e95eee2d | 6703 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6704 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
08369246 XW |
6705 | ret = __io_sq_thread(ctx, cap_entries); |
6706 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
6707 | sqt_spin = true; | |
69fb2131 | 6708 | } |
6c271ce2 | 6709 | |
08369246 | 6710 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6711 | io_run_task_work(); |
6712 | cond_resched(); | |
08369246 XW |
6713 | if (sqt_spin) |
6714 | timeout = jiffies + sqd->sq_thread_idle; | |
6715 | continue; | |
6716 | } | |
6717 | ||
08369246 XW |
6718 | needs_sched = true; |
6719 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6720 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6721 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6722 | !list_empty_careful(&ctx->iopoll_list)) { | |
6723 | needs_sched = false; | |
6724 | break; | |
6725 | } | |
6726 | if (io_sqring_entries(ctx)) { | |
6727 | needs_sched = false; | |
6728 | break; | |
6729 | } | |
6730 | } | |
6731 | ||
37d1e2e3 | 6732 | if (needs_sched && !io_sq_thread_should_park(sqd)) { |
69fb2131 JA |
6733 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6734 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6735 | |
69fb2131 | 6736 | schedule(); |
e4b4a13f | 6737 | try_to_freeze(); |
69fb2131 JA |
6738 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6739 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6740 | } |
08369246 XW |
6741 | |
6742 | finish_wait(&sqd->wait, &wait); | |
6743 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
6744 | } |
6745 | ||
37d1e2e3 JA |
6746 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6747 | io_uring_cancel_sqpoll(ctx); | |
06058632 | 6748 | |
37d1e2e3 | 6749 | io_run_task_work(); |
28cea78a | 6750 | |
8629397e JA |
6751 | if (io_sq_thread_should_park(sqd)) |
6752 | io_sq_thread_parkme(sqd); | |
6753 | ||
37d1e2e3 JA |
6754 | /* |
6755 | * Clear thread under lock so that concurrent parks work correctly | |
6756 | */ | |
8629397e | 6757 | complete(&sqd->completion); |
37d1e2e3 JA |
6758 | mutex_lock(&sqd->lock); |
6759 | sqd->thread = NULL; | |
5f3f26f9 JA |
6760 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6761 | ctx->sqo_exec = 1; | |
6762 | io_ring_set_wakeup_flag(ctx); | |
6763 | } | |
06058632 | 6764 | |
37d1e2e3 | 6765 | complete(&sqd->exited); |
e54945ae | 6766 | mutex_unlock(&sqd->lock); |
37d1e2e3 | 6767 | do_exit(0); |
6c271ce2 JA |
6768 | } |
6769 | ||
bda52162 JA |
6770 | struct io_wait_queue { |
6771 | struct wait_queue_entry wq; | |
6772 | struct io_ring_ctx *ctx; | |
6773 | unsigned to_wait; | |
6774 | unsigned nr_timeouts; | |
6775 | }; | |
6776 | ||
6c503150 | 6777 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6778 | { |
6779 | struct io_ring_ctx *ctx = iowq->ctx; | |
6780 | ||
6781 | /* | |
d195a66e | 6782 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6783 | * started waiting. For timeouts, we always want to return to userspace, |
6784 | * regardless of event count. | |
6785 | */ | |
6c503150 | 6786 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6787 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6788 | } | |
6789 | ||
6790 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6791 | int wake_flags, void *key) | |
6792 | { | |
6793 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6794 | wq); | |
6795 | ||
6c503150 PB |
6796 | /* |
6797 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6798 | * the task, and the next invocation will do it. | |
6799 | */ | |
6800 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6801 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6802 | return -1; | |
bda52162 JA |
6803 | } |
6804 | ||
af9c1a44 JA |
6805 | static int io_run_task_work_sig(void) |
6806 | { | |
6807 | if (io_run_task_work()) | |
6808 | return 1; | |
6809 | if (!signal_pending(current)) | |
6810 | return 0; | |
792ee0f6 JA |
6811 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6812 | return -ERESTARTSYS; | |
af9c1a44 JA |
6813 | return -EINTR; |
6814 | } | |
6815 | ||
eeb60b9a PB |
6816 | /* when returns >0, the caller should retry */ |
6817 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6818 | struct io_wait_queue *iowq, | |
6819 | signed long *timeout) | |
6820 | { | |
6821 | int ret; | |
6822 | ||
6823 | /* make sure we run task_work before checking for signals */ | |
6824 | ret = io_run_task_work_sig(); | |
6825 | if (ret || io_should_wake(iowq)) | |
6826 | return ret; | |
6827 | /* let the caller flush overflows, retry */ | |
6828 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6829 | return 1; | |
6830 | ||
6831 | *timeout = schedule_timeout(*timeout); | |
6832 | return !*timeout ? -ETIME : 1; | |
6833 | } | |
6834 | ||
2b188cc1 JA |
6835 | /* |
6836 | * Wait until events become available, if we don't already have some. The | |
6837 | * application must reap them itself, as they reside on the shared cq ring. | |
6838 | */ | |
6839 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6840 | const sigset_t __user *sig, size_t sigsz, |
6841 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6842 | { |
bda52162 JA |
6843 | struct io_wait_queue iowq = { |
6844 | .wq = { | |
6845 | .private = current, | |
6846 | .func = io_wake_function, | |
6847 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6848 | }, | |
6849 | .ctx = ctx, | |
6850 | .to_wait = min_events, | |
6851 | }; | |
75b28aff | 6852 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6853 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6854 | int ret; | |
2b188cc1 | 6855 | |
b41e9852 | 6856 | do { |
6c503150 PB |
6857 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6858 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6859 | return 0; |
4c6e277c | 6860 | if (!io_run_task_work()) |
b41e9852 | 6861 | break; |
b41e9852 | 6862 | } while (1); |
2b188cc1 JA |
6863 | |
6864 | if (sig) { | |
9e75ad5d AB |
6865 | #ifdef CONFIG_COMPAT |
6866 | if (in_compat_syscall()) | |
6867 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6868 | sigsz); |
9e75ad5d AB |
6869 | else |
6870 | #endif | |
b772434b | 6871 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6872 | |
2b188cc1 JA |
6873 | if (ret) |
6874 | return ret; | |
6875 | } | |
6876 | ||
c73ebb68 | 6877 | if (uts) { |
c1d5a224 PB |
6878 | struct timespec64 ts; |
6879 | ||
c73ebb68 HX |
6880 | if (get_timespec64(&ts, uts)) |
6881 | return -EFAULT; | |
6882 | timeout = timespec64_to_jiffies(&ts); | |
6883 | } | |
6884 | ||
bda52162 | 6885 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6886 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6887 | do { |
6c503150 | 6888 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
6889 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6890 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6891 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6892 | finish_wait(&ctx->wait, &iowq.wq); | |
6893 | } while (ret > 0); | |
bda52162 | 6894 | |
b7db41c9 | 6895 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6896 | |
75b28aff | 6897 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6898 | } |
6899 | ||
6b06314c JA |
6900 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6901 | { | |
6902 | #if defined(CONFIG_UNIX) | |
6903 | if (ctx->ring_sock) { | |
6904 | struct sock *sock = ctx->ring_sock->sk; | |
6905 | struct sk_buff *skb; | |
6906 | ||
6907 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6908 | kfree_skb(skb); | |
6909 | } | |
6910 | #else | |
6911 | int i; | |
6912 | ||
65e19f54 JA |
6913 | for (i = 0; i < ctx->nr_user_files; i++) { |
6914 | struct file *file; | |
6915 | ||
6916 | file = io_file_from_index(ctx, i); | |
6917 | if (file) | |
6918 | fput(file); | |
6919 | } | |
6b06314c JA |
6920 | #endif |
6921 | } | |
6922 | ||
00835dce | 6923 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6924 | { |
269bbe5f | 6925 | struct fixed_rsrc_data *data; |
05f3fb3c | 6926 | |
269bbe5f | 6927 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6928 | complete(&data->done); |
6929 | } | |
6930 | ||
2a63b2d9 | 6931 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6932 | { |
2a63b2d9 | 6933 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6934 | } |
6935 | ||
2a63b2d9 | 6936 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6937 | { |
2a63b2d9 BM |
6938 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6939 | } | |
65e19f54 | 6940 | |
d67d2263 BM |
6941 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6942 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6943 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6944 | { |
2a63b2d9 | 6945 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6946 | rsrc_data->node = ref_node; |
d67d2263 | 6947 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6948 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6949 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6950 | } |
6951 | ||
8bad28d8 | 6952 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6953 | { |
8bad28d8 | 6954 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6955 | |
2a63b2d9 | 6956 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6957 | ref_node = data->node; |
e6cb007c | 6958 | data->node = NULL; |
2a63b2d9 | 6959 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6960 | if (ref_node) |
6961 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6962 | } |
6963 | ||
6964 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6965 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6966 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6967 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6968 | { |
f2303b1f | 6969 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6970 | int ret; |
05589553 | 6971 | |
8bad28d8 HX |
6972 | if (data->quiesce) |
6973 | return -ENXIO; | |
05589553 | 6974 | |
8bad28d8 | 6975 | data->quiesce = true; |
1ffc5422 | 6976 | do { |
f2303b1f PB |
6977 | ret = -ENOMEM; |
6978 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
6979 | if (!backup_node) | |
6980 | break; | |
6981 | backup_node->rsrc_data = data; | |
6982 | backup_node->rsrc_put = rsrc_put; | |
6983 | ||
8bad28d8 HX |
6984 | io_sqe_rsrc_kill_node(ctx, data); |
6985 | percpu_ref_kill(&data->refs); | |
6986 | flush_delayed_work(&ctx->rsrc_put_work); | |
6987 | ||
1ffc5422 PB |
6988 | ret = wait_for_completion_interruptible(&data->done); |
6989 | if (!ret) | |
6990 | break; | |
8bad28d8 | 6991 | |
cb5e1b81 | 6992 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
6993 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
6994 | backup_node = NULL; | |
cb5e1b81 | 6995 | reinit_completion(&data->done); |
8bad28d8 | 6996 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 6997 | ret = io_run_task_work_sig(); |
8bad28d8 | 6998 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 6999 | } while (ret >= 0); |
8bad28d8 | 7000 | data->quiesce = false; |
05f3fb3c | 7001 | |
8bad28d8 HX |
7002 | if (backup_node) |
7003 | destroy_fixed_rsrc_ref_node(backup_node); | |
7004 | return ret; | |
d7954b2b BM |
7005 | } |
7006 | ||
1ad555c6 BM |
7007 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7008 | { | |
7009 | struct fixed_rsrc_data *data; | |
7010 | ||
7011 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7012 | if (!data) | |
7013 | return NULL; | |
7014 | ||
00835dce | 7015 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7016 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7017 | kfree(data); | |
7018 | return NULL; | |
7019 | } | |
7020 | data->ctx = ctx; | |
7021 | init_completion(&data->done); | |
7022 | return data; | |
7023 | } | |
7024 | ||
7025 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7026 | { | |
7027 | percpu_ref_exit(&data->refs); | |
7028 | kfree(data->table); | |
7029 | kfree(data); | |
7030 | } | |
7031 | ||
d7954b2b BM |
7032 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7033 | { | |
7034 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7035 | unsigned nr_tables, i; |
7036 | int ret; | |
7037 | ||
8bad28d8 HX |
7038 | /* |
7039 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7040 | * Since we possibly drop uring lock later in this function to | |
7041 | * run task work. | |
7042 | */ | |
7043 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7044 | return -ENXIO; |
f2303b1f | 7045 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7046 | if (ret) |
7047 | return ret; | |
7048 | ||
6b06314c | 7049 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7050 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7051 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7052 | kfree(data->table[i].files); |
1ad555c6 | 7053 | free_fixed_rsrc_data(data); |
05f3fb3c | 7054 | ctx->file_data = NULL; |
6b06314c JA |
7055 | ctx->nr_user_files = 0; |
7056 | return 0; | |
7057 | } | |
7058 | ||
37d1e2e3 JA |
7059 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7060 | __releases(&sqd->lock) | |
7061 | { | |
7062 | if (!sqd->thread) | |
7063 | return; | |
7064 | if (sqd->thread == current) | |
7065 | return; | |
7066 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7067 | wake_up_state(sqd->thread, TASK_PARKED); | |
7068 | mutex_unlock(&sqd->lock); | |
7069 | } | |
7070 | ||
7071 | static bool io_sq_thread_park(struct io_sq_data *sqd) | |
7072 | __acquires(&sqd->lock) | |
7073 | { | |
7074 | if (sqd->thread == current) | |
7075 | return true; | |
7076 | mutex_lock(&sqd->lock); | |
7077 | if (!sqd->thread) { | |
7078 | mutex_unlock(&sqd->lock); | |
7079 | return false; | |
7080 | } | |
7081 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7082 | wake_up_process(sqd->thread); | |
7083 | wait_for_completion(&sqd->completion); | |
7084 | return true; | |
7085 | } | |
7086 | ||
7087 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7088 | { | |
e54945ae | 7089 | if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) |
37d1e2e3 | 7090 | return; |
e54945ae JA |
7091 | mutex_lock(&sqd->lock); |
7092 | if (sqd->thread) { | |
7093 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7094 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)); | |
7095 | wake_up_process(sqd->thread); | |
7096 | mutex_unlock(&sqd->lock); | |
7097 | wait_for_completion(&sqd->exited); | |
7098 | WARN_ON_ONCE(sqd->thread); | |
7099 | } else { | |
7100 | mutex_unlock(&sqd->lock); | |
7101 | } | |
37d1e2e3 JA |
7102 | } |
7103 | ||
534ca6d6 | 7104 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7105 | { |
534ca6d6 | 7106 | if (refcount_dec_and_test(&sqd->refs)) { |
37d1e2e3 JA |
7107 | io_sq_thread_stop(sqd); |
7108 | kfree(sqd); | |
7109 | } | |
7110 | } | |
7111 | ||
7112 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7113 | { | |
7114 | struct io_sq_data *sqd = ctx->sq_data; | |
7115 | ||
7116 | if (sqd) { | |
eb85890b | 7117 | complete(&sqd->startup); |
534ca6d6 | 7118 | if (sqd->thread) { |
37d1e2e3 JA |
7119 | wait_for_completion(&ctx->sq_thread_comp); |
7120 | io_sq_thread_park(sqd); | |
534ca6d6 JA |
7121 | } |
7122 | ||
37d1e2e3 JA |
7123 | mutex_lock(&sqd->ctx_lock); |
7124 | list_del(&ctx->sqd_list); | |
7125 | io_sqd_update_thread_idle(sqd); | |
7126 | mutex_unlock(&sqd->ctx_lock); | |
7127 | ||
7128 | if (sqd->thread) | |
7129 | io_sq_thread_unpark(sqd); | |
7130 | ||
7131 | io_put_sq_data(sqd); | |
7132 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7133 | } |
7134 | } | |
7135 | ||
aa06165d JA |
7136 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7137 | { | |
7138 | struct io_ring_ctx *ctx_attach; | |
7139 | struct io_sq_data *sqd; | |
7140 | struct fd f; | |
7141 | ||
7142 | f = fdget(p->wq_fd); | |
7143 | if (!f.file) | |
7144 | return ERR_PTR(-ENXIO); | |
7145 | if (f.file->f_op != &io_uring_fops) { | |
7146 | fdput(f); | |
7147 | return ERR_PTR(-EINVAL); | |
7148 | } | |
7149 | ||
7150 | ctx_attach = f.file->private_data; | |
7151 | sqd = ctx_attach->sq_data; | |
7152 | if (!sqd) { | |
7153 | fdput(f); | |
7154 | return ERR_PTR(-EINVAL); | |
7155 | } | |
7156 | ||
7157 | refcount_inc(&sqd->refs); | |
7158 | fdput(f); | |
7159 | return sqd; | |
7160 | } | |
7161 | ||
534ca6d6 JA |
7162 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7163 | { | |
7164 | struct io_sq_data *sqd; | |
7165 | ||
aa06165d JA |
7166 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7167 | return io_attach_sq_data(p); | |
7168 | ||
534ca6d6 JA |
7169 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7170 | if (!sqd) | |
7171 | return ERR_PTR(-ENOMEM); | |
7172 | ||
7173 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7174 | INIT_LIST_HEAD(&sqd->ctx_list); |
7175 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7176 | mutex_init(&sqd->ctx_lock); | |
7177 | mutex_init(&sqd->lock); | |
534ca6d6 | 7178 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 JA |
7179 | init_completion(&sqd->startup); |
7180 | init_completion(&sqd->completion); | |
7181 | init_completion(&sqd->exited); | |
534ca6d6 JA |
7182 | return sqd; |
7183 | } | |
7184 | ||
6b06314c | 7185 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7186 | /* |
7187 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7188 | * the io_uring can be safely unregistered on process exit, even if we have | |
7189 | * loops in the file referencing. | |
7190 | */ | |
7191 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7192 | { | |
7193 | struct sock *sk = ctx->ring_sock->sk; | |
7194 | struct scm_fp_list *fpl; | |
7195 | struct sk_buff *skb; | |
08a45173 | 7196 | int i, nr_files; |
6b06314c | 7197 | |
6b06314c JA |
7198 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7199 | if (!fpl) | |
7200 | return -ENOMEM; | |
7201 | ||
7202 | skb = alloc_skb(0, GFP_KERNEL); | |
7203 | if (!skb) { | |
7204 | kfree(fpl); | |
7205 | return -ENOMEM; | |
7206 | } | |
7207 | ||
7208 | skb->sk = sk; | |
6b06314c | 7209 | |
08a45173 | 7210 | nr_files = 0; |
62e398be | 7211 | fpl->user = get_uid(current_user()); |
6b06314c | 7212 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7213 | struct file *file = io_file_from_index(ctx, i + offset); |
7214 | ||
7215 | if (!file) | |
08a45173 | 7216 | continue; |
65e19f54 | 7217 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7218 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7219 | nr_files++; | |
6b06314c JA |
7220 | } |
7221 | ||
08a45173 JA |
7222 | if (nr_files) { |
7223 | fpl->max = SCM_MAX_FD; | |
7224 | fpl->count = nr_files; | |
7225 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7226 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7227 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7228 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7229 | |
08a45173 JA |
7230 | for (i = 0; i < nr_files; i++) |
7231 | fput(fpl->fp[i]); | |
7232 | } else { | |
7233 | kfree_skb(skb); | |
7234 | kfree(fpl); | |
7235 | } | |
6b06314c JA |
7236 | |
7237 | return 0; | |
7238 | } | |
7239 | ||
7240 | /* | |
7241 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7242 | * causes regular reference counting to break down. We rely on the UNIX | |
7243 | * garbage collection to take care of this problem for us. | |
7244 | */ | |
7245 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7246 | { | |
7247 | unsigned left, total; | |
7248 | int ret = 0; | |
7249 | ||
7250 | total = 0; | |
7251 | left = ctx->nr_user_files; | |
7252 | while (left) { | |
7253 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7254 | |
7255 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7256 | if (ret) | |
7257 | break; | |
7258 | left -= this_files; | |
7259 | total += this_files; | |
7260 | } | |
7261 | ||
7262 | if (!ret) | |
7263 | return 0; | |
7264 | ||
7265 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7266 | struct file *file = io_file_from_index(ctx, total); |
7267 | ||
7268 | if (file) | |
7269 | fput(file); | |
6b06314c JA |
7270 | total++; |
7271 | } | |
7272 | ||
7273 | return ret; | |
7274 | } | |
7275 | #else | |
7276 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7277 | { | |
7278 | return 0; | |
7279 | } | |
7280 | #endif | |
7281 | ||
269bbe5f | 7282 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7283 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7284 | { |
7285 | int i; | |
7286 | ||
7287 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7288 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7289 | unsigned this_files; |
7290 | ||
7291 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7292 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7293 | GFP_KERNEL); | |
7294 | if (!table->files) | |
7295 | break; | |
7296 | nr_files -= this_files; | |
7297 | } | |
7298 | ||
7299 | if (i == nr_tables) | |
7300 | return 0; | |
7301 | ||
7302 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7303 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7304 | kfree(table->files); |
7305 | } | |
7306 | return 1; | |
7307 | } | |
7308 | ||
50238531 | 7309 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7310 | { |
50238531 | 7311 | struct file *file = prsrc->file; |
05f3fb3c JA |
7312 | #if defined(CONFIG_UNIX) |
7313 | struct sock *sock = ctx->ring_sock->sk; | |
7314 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7315 | struct sk_buff *skb; | |
7316 | int i; | |
7317 | ||
7318 | __skb_queue_head_init(&list); | |
7319 | ||
7320 | /* | |
7321 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7322 | * remove this entry and rearrange the file array. | |
7323 | */ | |
7324 | skb = skb_dequeue(head); | |
7325 | while (skb) { | |
7326 | struct scm_fp_list *fp; | |
7327 | ||
7328 | fp = UNIXCB(skb).fp; | |
7329 | for (i = 0; i < fp->count; i++) { | |
7330 | int left; | |
7331 | ||
7332 | if (fp->fp[i] != file) | |
7333 | continue; | |
7334 | ||
7335 | unix_notinflight(fp->user, fp->fp[i]); | |
7336 | left = fp->count - 1 - i; | |
7337 | if (left) { | |
7338 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7339 | left * sizeof(struct file *)); | |
7340 | } | |
7341 | fp->count--; | |
7342 | if (!fp->count) { | |
7343 | kfree_skb(skb); | |
7344 | skb = NULL; | |
7345 | } else { | |
7346 | __skb_queue_tail(&list, skb); | |
7347 | } | |
7348 | fput(file); | |
7349 | file = NULL; | |
7350 | break; | |
7351 | } | |
7352 | ||
7353 | if (!file) | |
7354 | break; | |
7355 | ||
7356 | __skb_queue_tail(&list, skb); | |
7357 | ||
7358 | skb = skb_dequeue(head); | |
7359 | } | |
7360 | ||
7361 | if (skb_peek(&list)) { | |
7362 | spin_lock_irq(&head->lock); | |
7363 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7364 | __skb_queue_tail(head, skb); | |
7365 | spin_unlock_irq(&head->lock); | |
7366 | } | |
7367 | #else | |
7368 | fput(file); | |
7369 | #endif | |
7370 | } | |
7371 | ||
269bbe5f | 7372 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7373 | { |
269bbe5f BM |
7374 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7375 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7376 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7377 | |
269bbe5f BM |
7378 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7379 | list_del(&prsrc->list); | |
50238531 | 7380 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7381 | kfree(prsrc); |
65e19f54 | 7382 | } |
05589553 | 7383 | |
05589553 XW |
7384 | percpu_ref_exit(&ref_node->refs); |
7385 | kfree(ref_node); | |
269bbe5f | 7386 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7387 | } |
65e19f54 | 7388 | |
269bbe5f | 7389 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7390 | { |
7391 | struct io_ring_ctx *ctx; | |
7392 | struct llist_node *node; | |
7393 | ||
269bbe5f BM |
7394 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7395 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7396 | |
7397 | while (node) { | |
269bbe5f | 7398 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7399 | struct llist_node *next = node->next; |
7400 | ||
269bbe5f BM |
7401 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7402 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7403 | node = next; |
7404 | } | |
7405 | } | |
7406 | ||
ea64ec02 PB |
7407 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7408 | unsigned i) | |
2faf852d | 7409 | { |
ea64ec02 PB |
7410 | struct fixed_rsrc_table *table; |
7411 | ||
7412 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7413 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7414 | } | |
7415 | ||
00835dce | 7416 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7417 | { |
269bbe5f BM |
7418 | struct fixed_rsrc_ref_node *ref_node; |
7419 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7420 | struct io_ring_ctx *ctx; |
e297822b | 7421 | bool first_add = false; |
4a38aed2 | 7422 | int delay = HZ; |
65e19f54 | 7423 | |
269bbe5f BM |
7424 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7425 | data = ref_node->rsrc_data; | |
e297822b PB |
7426 | ctx = data->ctx; |
7427 | ||
2a63b2d9 | 7428 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7429 | ref_node->done = true; |
7430 | ||
d67d2263 BM |
7431 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7432 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7433 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7434 | /* recycle ref nodes in order */ |
7435 | if (!ref_node->done) | |
7436 | break; | |
7437 | list_del(&ref_node->node); | |
269bbe5f | 7438 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7439 | } |
2a63b2d9 | 7440 | io_rsrc_ref_unlock(ctx); |
05589553 | 7441 | |
e297822b | 7442 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7443 | delay = 0; |
05589553 | 7444 | |
4a38aed2 | 7445 | if (!delay) |
269bbe5f | 7446 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7447 | else if (first_add) |
269bbe5f | 7448 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7449 | } |
65e19f54 | 7450 | |
6802535d | 7451 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7452 | struct io_ring_ctx *ctx) |
05f3fb3c | 7453 | { |
269bbe5f | 7454 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7455 | |
05589553 XW |
7456 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7457 | if (!ref_node) | |
3e2224c5 | 7458 | return NULL; |
05f3fb3c | 7459 | |
00835dce | 7460 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7461 | 0, GFP_KERNEL)) { |
7462 | kfree(ref_node); | |
3e2224c5 | 7463 | return NULL; |
05589553 XW |
7464 | } |
7465 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7466 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7467 | ref_node->done = false; |
05589553 | 7468 | return ref_node; |
05589553 XW |
7469 | } |
7470 | ||
bc9744cd PB |
7471 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7472 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7473 | { |
269bbe5f | 7474 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7475 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7476 | } |
7477 | ||
269bbe5f | 7478 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7479 | { |
7480 | percpu_ref_exit(&ref_node->refs); | |
7481 | kfree(ref_node); | |
65e19f54 JA |
7482 | } |
7483 | ||
ea64ec02 | 7484 | |
6b06314c JA |
7485 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7486 | unsigned nr_args) | |
7487 | { | |
7488 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7489 | unsigned nr_tables, i; |
05f3fb3c | 7490 | struct file *file; |
600cf3f8 | 7491 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7492 | struct fixed_rsrc_ref_node *ref_node; |
7493 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7494 | |
05f3fb3c | 7495 | if (ctx->file_data) |
6b06314c JA |
7496 | return -EBUSY; |
7497 | if (!nr_args) | |
7498 | return -EINVAL; | |
7499 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7500 | return -EMFILE; | |
7501 | ||
1ad555c6 | 7502 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7503 | if (!file_data) |
05f3fb3c | 7504 | return -ENOMEM; |
13770a71 | 7505 | ctx->file_data = file_data; |
05f3fb3c | 7506 | |
65e19f54 | 7507 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7508 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7509 | GFP_KERNEL); |
600cf3f8 PB |
7510 | if (!file_data->table) |
7511 | goto out_free; | |
05f3fb3c | 7512 | |
600cf3f8 | 7513 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7514 | goto out_free; |
65e19f54 | 7515 | |
08a45173 | 7516 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7517 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7518 | ret = -EFAULT; | |
7519 | goto out_fput; | |
7520 | } | |
08a45173 | 7521 | /* allow sparse sets */ |
600cf3f8 | 7522 | if (fd == -1) |
08a45173 | 7523 | continue; |
6b06314c | 7524 | |
05f3fb3c | 7525 | file = fget(fd); |
6b06314c | 7526 | ret = -EBADF; |
05f3fb3c | 7527 | if (!file) |
600cf3f8 | 7528 | goto out_fput; |
05f3fb3c | 7529 | |
6b06314c JA |
7530 | /* |
7531 | * Don't allow io_uring instances to be registered. If UNIX | |
7532 | * isn't enabled, then this causes a reference cycle and this | |
7533 | * instance can never get freed. If UNIX is enabled we'll | |
7534 | * handle it just fine, but there's still no point in allowing | |
7535 | * a ring fd as it doesn't support regular read/write anyway. | |
7536 | */ | |
05f3fb3c JA |
7537 | if (file->f_op == &io_uring_fops) { |
7538 | fput(file); | |
600cf3f8 | 7539 | goto out_fput; |
6b06314c | 7540 | } |
ea64ec02 | 7541 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7542 | } |
7543 | ||
6b06314c | 7544 | ret = io_sqe_files_scm(ctx); |
05589553 | 7545 | if (ret) { |
6b06314c | 7546 | io_sqe_files_unregister(ctx); |
05589553 XW |
7547 | return ret; |
7548 | } | |
6b06314c | 7549 | |
bc9744cd | 7550 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7551 | if (!ref_node) { |
05589553 | 7552 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7553 | return -ENOMEM; |
05589553 | 7554 | } |
bc9744cd | 7555 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7556 | |
d67d2263 | 7557 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7558 | return ret; |
600cf3f8 PB |
7559 | out_fput: |
7560 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7561 | file = io_file_from_index(ctx, i); | |
7562 | if (file) | |
7563 | fput(file); | |
7564 | } | |
7565 | for (i = 0; i < nr_tables; i++) | |
7566 | kfree(file_data->table[i].files); | |
7567 | ctx->nr_user_files = 0; | |
600cf3f8 | 7568 | out_free: |
1ad555c6 | 7569 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7570 | ctx->file_data = NULL; |
6b06314c JA |
7571 | return ret; |
7572 | } | |
7573 | ||
c3a31e60 JA |
7574 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7575 | int index) | |
7576 | { | |
7577 | #if defined(CONFIG_UNIX) | |
7578 | struct sock *sock = ctx->ring_sock->sk; | |
7579 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7580 | struct sk_buff *skb; | |
7581 | ||
7582 | /* | |
7583 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7584 | * file set. If there's no room, fall back to allocating a new skb | |
7585 | * and filling it in. | |
7586 | */ | |
7587 | spin_lock_irq(&head->lock); | |
7588 | skb = skb_peek(head); | |
7589 | if (skb) { | |
7590 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7591 | ||
7592 | if (fpl->count < SCM_MAX_FD) { | |
7593 | __skb_unlink(skb, head); | |
7594 | spin_unlock_irq(&head->lock); | |
7595 | fpl->fp[fpl->count] = get_file(file); | |
7596 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7597 | fpl->count++; | |
7598 | spin_lock_irq(&head->lock); | |
7599 | __skb_queue_head(head, skb); | |
7600 | } else { | |
7601 | skb = NULL; | |
7602 | } | |
7603 | } | |
7604 | spin_unlock_irq(&head->lock); | |
7605 | ||
7606 | if (skb) { | |
7607 | fput(file); | |
7608 | return 0; | |
7609 | } | |
7610 | ||
7611 | return __io_sqe_files_scm(ctx, 1, index); | |
7612 | #else | |
7613 | return 0; | |
7614 | #endif | |
7615 | } | |
7616 | ||
50238531 | 7617 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7618 | { |
269bbe5f BM |
7619 | struct io_rsrc_put *prsrc; |
7620 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7621 | |
269bbe5f BM |
7622 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7623 | if (!prsrc) | |
a5318d3c | 7624 | return -ENOMEM; |
05f3fb3c | 7625 | |
50238531 | 7626 | prsrc->rsrc = rsrc; |
269bbe5f | 7627 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7628 | |
a5318d3c | 7629 | return 0; |
05f3fb3c JA |
7630 | } |
7631 | ||
269bbe5f BM |
7632 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7633 | struct file *file) | |
7634 | { | |
50238531 | 7635 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7636 | } |
7637 | ||
05f3fb3c | 7638 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7639 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7640 | unsigned nr_args) |
7641 | { | |
269bbe5f BM |
7642 | struct fixed_rsrc_data *data = ctx->file_data; |
7643 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7644 | struct file *file, **file_slot; |
c3a31e60 JA |
7645 | __s32 __user *fds; |
7646 | int fd, i, err; | |
7647 | __u32 done; | |
05589553 | 7648 | bool needs_switch = false; |
c3a31e60 | 7649 | |
05f3fb3c | 7650 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7651 | return -EOVERFLOW; |
7652 | if (done > ctx->nr_user_files) | |
7653 | return -EINVAL; | |
7654 | ||
bc9744cd | 7655 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7656 | if (!ref_node) |
7657 | return -ENOMEM; | |
bc9744cd | 7658 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7659 | |
269bbe5f | 7660 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7661 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7662 | err = 0; |
7663 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7664 | err = -EFAULT; | |
7665 | break; | |
7666 | } | |
4e0377a1 | 7667 | if (fd == IORING_REGISTER_FILES_SKIP) |
7668 | continue; | |
7669 | ||
67973b93 | 7670 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7671 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7672 | ||
7673 | if (*file_slot) { | |
7674 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7675 | if (err) |
7676 | break; | |
ea64ec02 | 7677 | *file_slot = NULL; |
05589553 | 7678 | needs_switch = true; |
c3a31e60 JA |
7679 | } |
7680 | if (fd != -1) { | |
c3a31e60 JA |
7681 | file = fget(fd); |
7682 | if (!file) { | |
7683 | err = -EBADF; | |
7684 | break; | |
7685 | } | |
7686 | /* | |
7687 | * Don't allow io_uring instances to be registered. If | |
7688 | * UNIX isn't enabled, then this causes a reference | |
7689 | * cycle and this instance can never get freed. If UNIX | |
7690 | * is enabled we'll handle it just fine, but there's | |
7691 | * still no point in allowing a ring fd as it doesn't | |
7692 | * support regular read/write anyway. | |
7693 | */ | |
7694 | if (file->f_op == &io_uring_fops) { | |
7695 | fput(file); | |
7696 | err = -EBADF; | |
7697 | break; | |
7698 | } | |
e68a3ff8 | 7699 | *file_slot = file; |
c3a31e60 | 7700 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7701 | if (err) { |
e68a3ff8 | 7702 | *file_slot = NULL; |
f3bd9dae | 7703 | fput(file); |
c3a31e60 | 7704 | break; |
f3bd9dae | 7705 | } |
c3a31e60 | 7706 | } |
05f3fb3c JA |
7707 | } |
7708 | ||
05589553 | 7709 | if (needs_switch) { |
b2e96852 | 7710 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7711 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7712 | } else |
269bbe5f | 7713 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7714 | |
7715 | return done ? done : err; | |
7716 | } | |
05589553 | 7717 | |
05f3fb3c JA |
7718 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7719 | unsigned nr_args) | |
7720 | { | |
269bbe5f | 7721 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7722 | |
7723 | if (!ctx->file_data) | |
7724 | return -ENXIO; | |
7725 | if (!nr_args) | |
7726 | return -EINVAL; | |
7727 | if (copy_from_user(&up, arg, sizeof(up))) | |
7728 | return -EFAULT; | |
7729 | if (up.resv) | |
7730 | return -EINVAL; | |
7731 | ||
7732 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7733 | } | |
c3a31e60 | 7734 | |
5280f7e5 | 7735 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7736 | { |
7737 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7738 | ||
5280f7e5 PB |
7739 | req = io_put_req_find_next(req); |
7740 | return req ? &req->work : NULL; | |
7d723065 JA |
7741 | } |
7742 | ||
5aa75ed5 | 7743 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7744 | { |
e941894e | 7745 | struct io_wq_hash *hash; |
24369c2e | 7746 | struct io_wq_data data; |
24369c2e | 7747 | unsigned int concurrency; |
24369c2e | 7748 | |
e941894e JA |
7749 | hash = ctx->hash_map; |
7750 | if (!hash) { | |
7751 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7752 | if (!hash) | |
7753 | return ERR_PTR(-ENOMEM); | |
7754 | refcount_set(&hash->refs, 1); | |
7755 | init_waitqueue_head(&hash->wait); | |
7756 | ctx->hash_map = hash; | |
24369c2e PB |
7757 | } |
7758 | ||
e941894e | 7759 | data.hash = hash; |
e9fd9396 | 7760 | data.free_work = io_free_work; |
f5fa38c5 | 7761 | data.do_work = io_wq_submit_work; |
24369c2e | 7762 | |
d25e3a3d JA |
7763 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7764 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7765 | |
5aa75ed5 | 7766 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7767 | } |
7768 | ||
5aa75ed5 JA |
7769 | static int io_uring_alloc_task_context(struct task_struct *task, |
7770 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7771 | { |
7772 | struct io_uring_task *tctx; | |
d8a6df10 | 7773 | int ret; |
0f212204 JA |
7774 | |
7775 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7776 | if (unlikely(!tctx)) | |
7777 | return -ENOMEM; | |
7778 | ||
d8a6df10 JA |
7779 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7780 | if (unlikely(ret)) { | |
7781 | kfree(tctx); | |
7782 | return ret; | |
7783 | } | |
7784 | ||
5aa75ed5 JA |
7785 | tctx->io_wq = io_init_wq_offload(ctx); |
7786 | if (IS_ERR(tctx->io_wq)) { | |
7787 | ret = PTR_ERR(tctx->io_wq); | |
7788 | percpu_counter_destroy(&tctx->inflight); | |
7789 | kfree(tctx); | |
7790 | return ret; | |
7791 | } | |
7792 | ||
0f212204 JA |
7793 | xa_init(&tctx->xa); |
7794 | init_waitqueue_head(&tctx->wait); | |
7795 | tctx->last = NULL; | |
fdaf083c JA |
7796 | atomic_set(&tctx->in_idle, 0); |
7797 | tctx->sqpoll = false; | |
0f212204 | 7798 | task->io_uring = tctx; |
7cbf1722 JA |
7799 | spin_lock_init(&tctx->task_lock); |
7800 | INIT_WQ_LIST(&tctx->task_list); | |
7801 | tctx->task_state = 0; | |
7802 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7803 | return 0; |
7804 | } | |
7805 | ||
7806 | void __io_uring_free(struct task_struct *tsk) | |
7807 | { | |
7808 | struct io_uring_task *tctx = tsk->io_uring; | |
7809 | ||
7810 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7811 | WARN_ON_ONCE(tctx->io_wq); |
7812 | ||
d8a6df10 | 7813 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7814 | kfree(tctx); |
7815 | tsk->io_uring = NULL; | |
7816 | } | |
7817 | ||
5f3f26f9 JA |
7818 | static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx) |
7819 | { | |
7820 | int ret; | |
7821 | ||
7822 | clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7823 | reinit_completion(&sqd->completion); | |
70aacfe6 | 7824 | ctx->sqo_exec = 0; |
5f3f26f9 JA |
7825 | sqd->task_pid = current->pid; |
7826 | current->flags |= PF_IO_WORKER; | |
7827 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7828 | current->flags &= ~PF_IO_WORKER; | |
7829 | if (ret < 0) { | |
7830 | sqd->thread = NULL; | |
7831 | return ret; | |
7832 | } | |
7833 | wait_for_completion(&sqd->completion); | |
7834 | return io_uring_alloc_task_context(sqd->thread, ctx); | |
7835 | } | |
7836 | ||
7e84e1c7 SG |
7837 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7838 | struct io_uring_params *p) | |
2b188cc1 JA |
7839 | { |
7840 | int ret; | |
7841 | ||
d25e3a3d JA |
7842 | /* Retain compatibility with failing for an invalid attach attempt */ |
7843 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7844 | IORING_SETUP_ATTACH_WQ) { | |
7845 | struct fd f; | |
7846 | ||
7847 | f = fdget(p->wq_fd); | |
7848 | if (!f.file) | |
7849 | return -ENXIO; | |
7850 | if (f.file->f_op != &io_uring_fops) { | |
7851 | fdput(f); | |
7852 | return -EINVAL; | |
7853 | } | |
7854 | fdput(f); | |
7855 | } | |
6c271ce2 | 7856 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
7857 | struct io_sq_data *sqd; |
7858 | ||
3ec482d1 | 7859 | ret = -EPERM; |
ce59fc69 | 7860 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7861 | goto err; |
7862 | ||
534ca6d6 JA |
7863 | sqd = io_get_sq_data(p); |
7864 | if (IS_ERR(sqd)) { | |
7865 | ret = PTR_ERR(sqd); | |
7866 | goto err; | |
7867 | } | |
69fb2131 | 7868 | |
534ca6d6 | 7869 | ctx->sq_data = sqd; |
69fb2131 JA |
7870 | io_sq_thread_park(sqd); |
7871 | mutex_lock(&sqd->ctx_lock); | |
7872 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7873 | mutex_unlock(&sqd->ctx_lock); | |
7874 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7875 | |
917257da JA |
7876 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7877 | if (!ctx->sq_thread_idle) | |
7878 | ctx->sq_thread_idle = HZ; | |
7879 | ||
aa06165d | 7880 | if (sqd->thread) |
5aa75ed5 | 7881 | return 0; |
aa06165d | 7882 | |
6c271ce2 | 7883 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7884 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7885 | |
917257da | 7886 | ret = -EINVAL; |
44a9bd18 JA |
7887 | if (cpu >= nr_cpu_ids) |
7888 | goto err; | |
7889f44d | 7889 | if (!cpu_online(cpu)) |
917257da JA |
7890 | goto err; |
7891 | ||
37d1e2e3 | 7892 | sqd->sq_cpu = cpu; |
6c271ce2 | 7893 | } else { |
37d1e2e3 | 7894 | sqd->sq_cpu = -1; |
6c271ce2 | 7895 | } |
37d1e2e3 JA |
7896 | |
7897 | sqd->task_pid = current->pid; | |
7898 | current->flags |= PF_IO_WORKER; | |
7899 | ret = io_wq_fork_thread(io_sq_thread, sqd); | |
7900 | current->flags &= ~PF_IO_WORKER; | |
7901 | if (ret < 0) { | |
534ca6d6 | 7902 | sqd->thread = NULL; |
6c271ce2 JA |
7903 | goto err; |
7904 | } | |
37d1e2e3 | 7905 | wait_for_completion(&sqd->completion); |
5aa75ed5 | 7906 | ret = io_uring_alloc_task_context(sqd->thread, ctx); |
0f212204 JA |
7907 | if (ret) |
7908 | goto err; | |
6c271ce2 JA |
7909 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7910 | /* Can't have SQ_AFF without SQPOLL */ | |
7911 | ret = -EINVAL; | |
7912 | goto err; | |
7913 | } | |
7914 | ||
2b188cc1 JA |
7915 | return 0; |
7916 | err: | |
37d1e2e3 | 7917 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
7918 | return ret; |
7919 | } | |
7920 | ||
7e84e1c7 SG |
7921 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7922 | { | |
534ca6d6 JA |
7923 | struct io_sq_data *sqd = ctx->sq_data; |
7924 | ||
3ebba796 | 7925 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
eb85890b | 7926 | if (ctx->flags & IORING_SETUP_SQPOLL) |
37d1e2e3 | 7927 | complete(&sqd->startup); |
7e84e1c7 SG |
7928 | } |
7929 | ||
a087e2b5 BM |
7930 | static inline void __io_unaccount_mem(struct user_struct *user, |
7931 | unsigned long nr_pages) | |
2b188cc1 JA |
7932 | { |
7933 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7934 | } | |
7935 | ||
a087e2b5 BM |
7936 | static inline int __io_account_mem(struct user_struct *user, |
7937 | unsigned long nr_pages) | |
2b188cc1 JA |
7938 | { |
7939 | unsigned long page_limit, cur_pages, new_pages; | |
7940 | ||
7941 | /* Don't allow more pages than we can safely lock */ | |
7942 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7943 | ||
7944 | do { | |
7945 | cur_pages = atomic_long_read(&user->locked_vm); | |
7946 | new_pages = cur_pages + nr_pages; | |
7947 | if (new_pages > page_limit) | |
7948 | return -ENOMEM; | |
7949 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7950 | new_pages) != cur_pages); | |
7951 | ||
7952 | return 0; | |
7953 | } | |
7954 | ||
26bfa89e | 7955 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7956 | { |
62e398be | 7957 | if (ctx->user) |
a087e2b5 | 7958 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7959 | |
26bfa89e JA |
7960 | if (ctx->mm_account) |
7961 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7962 | } |
7963 | ||
26bfa89e | 7964 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7965 | { |
30975825 BM |
7966 | int ret; |
7967 | ||
62e398be | 7968 | if (ctx->user) { |
30975825 BM |
7969 | ret = __io_account_mem(ctx->user, nr_pages); |
7970 | if (ret) | |
7971 | return ret; | |
7972 | } | |
7973 | ||
26bfa89e JA |
7974 | if (ctx->mm_account) |
7975 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7976 | |
7977 | return 0; | |
7978 | } | |
7979 | ||
2b188cc1 JA |
7980 | static void io_mem_free(void *ptr) |
7981 | { | |
52e04ef4 MR |
7982 | struct page *page; |
7983 | ||
7984 | if (!ptr) | |
7985 | return; | |
2b188cc1 | 7986 | |
52e04ef4 | 7987 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7988 | if (put_page_testzero(page)) |
7989 | free_compound_page(page); | |
7990 | } | |
7991 | ||
7992 | static void *io_mem_alloc(size_t size) | |
7993 | { | |
7994 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 7995 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
7996 | |
7997 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7998 | } | |
7999 | ||
75b28aff HV |
8000 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8001 | size_t *sq_offset) | |
8002 | { | |
8003 | struct io_rings *rings; | |
8004 | size_t off, sq_array_size; | |
8005 | ||
8006 | off = struct_size(rings, cqes, cq_entries); | |
8007 | if (off == SIZE_MAX) | |
8008 | return SIZE_MAX; | |
8009 | ||
8010 | #ifdef CONFIG_SMP | |
8011 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8012 | if (off == 0) | |
8013 | return SIZE_MAX; | |
8014 | #endif | |
8015 | ||
b36200f5 DV |
8016 | if (sq_offset) |
8017 | *sq_offset = off; | |
8018 | ||
75b28aff HV |
8019 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8020 | if (sq_array_size == SIZE_MAX) | |
8021 | return SIZE_MAX; | |
8022 | ||
8023 | if (check_add_overflow(off, sq_array_size, &off)) | |
8024 | return SIZE_MAX; | |
8025 | ||
75b28aff HV |
8026 | return off; |
8027 | } | |
8028 | ||
0a96bbe4 | 8029 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8030 | { |
8031 | int i, j; | |
8032 | ||
8033 | if (!ctx->user_bufs) | |
8034 | return -ENXIO; | |
8035 | ||
8036 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8037 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8038 | ||
8039 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8040 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8041 | |
de293938 | 8042 | if (imu->acct_pages) |
26bfa89e | 8043 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8044 | kvfree(imu->bvec); |
edafccee JA |
8045 | imu->nr_bvecs = 0; |
8046 | } | |
8047 | ||
8048 | kfree(ctx->user_bufs); | |
8049 | ctx->user_bufs = NULL; | |
8050 | ctx->nr_user_bufs = 0; | |
8051 | return 0; | |
8052 | } | |
8053 | ||
8054 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8055 | void __user *arg, unsigned index) | |
8056 | { | |
8057 | struct iovec __user *src; | |
8058 | ||
8059 | #ifdef CONFIG_COMPAT | |
8060 | if (ctx->compat) { | |
8061 | struct compat_iovec __user *ciovs; | |
8062 | struct compat_iovec ciov; | |
8063 | ||
8064 | ciovs = (struct compat_iovec __user *) arg; | |
8065 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8066 | return -EFAULT; | |
8067 | ||
d55e5f5b | 8068 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8069 | dst->iov_len = ciov.iov_len; |
8070 | return 0; | |
8071 | } | |
8072 | #endif | |
8073 | src = (struct iovec __user *) arg; | |
8074 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8075 | return -EFAULT; | |
8076 | return 0; | |
8077 | } | |
8078 | ||
de293938 JA |
8079 | /* |
8080 | * Not super efficient, but this is just a registration time. And we do cache | |
8081 | * the last compound head, so generally we'll only do a full search if we don't | |
8082 | * match that one. | |
8083 | * | |
8084 | * We check if the given compound head page has already been accounted, to | |
8085 | * avoid double accounting it. This allows us to account the full size of the | |
8086 | * page, not just the constituent pages of a huge page. | |
8087 | */ | |
8088 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8089 | int nr_pages, struct page *hpage) | |
8090 | { | |
8091 | int i, j; | |
8092 | ||
8093 | /* check current page array */ | |
8094 | for (i = 0; i < nr_pages; i++) { | |
8095 | if (!PageCompound(pages[i])) | |
8096 | continue; | |
8097 | if (compound_head(pages[i]) == hpage) | |
8098 | return true; | |
8099 | } | |
8100 | ||
8101 | /* check previously registered pages */ | |
8102 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8103 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8104 | ||
8105 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8106 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8107 | continue; | |
8108 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8109 | return true; | |
8110 | } | |
8111 | } | |
8112 | ||
8113 | return false; | |
8114 | } | |
8115 | ||
8116 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8117 | int nr_pages, struct io_mapped_ubuf *imu, | |
8118 | struct page **last_hpage) | |
8119 | { | |
8120 | int i, ret; | |
8121 | ||
8122 | for (i = 0; i < nr_pages; i++) { | |
8123 | if (!PageCompound(pages[i])) { | |
8124 | imu->acct_pages++; | |
8125 | } else { | |
8126 | struct page *hpage; | |
8127 | ||
8128 | hpage = compound_head(pages[i]); | |
8129 | if (hpage == *last_hpage) | |
8130 | continue; | |
8131 | *last_hpage = hpage; | |
8132 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8133 | continue; | |
8134 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8135 | } | |
8136 | } | |
8137 | ||
8138 | if (!imu->acct_pages) | |
8139 | return 0; | |
8140 | ||
26bfa89e | 8141 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8142 | if (ret) |
8143 | imu->acct_pages = 0; | |
8144 | return ret; | |
8145 | } | |
8146 | ||
0a96bbe4 BM |
8147 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8148 | struct io_mapped_ubuf *imu, | |
8149 | struct page **last_hpage) | |
edafccee JA |
8150 | { |
8151 | struct vm_area_struct **vmas = NULL; | |
8152 | struct page **pages = NULL; | |
0a96bbe4 BM |
8153 | unsigned long off, start, end, ubuf; |
8154 | size_t size; | |
8155 | int ret, pret, nr_pages, i; | |
8156 | ||
8157 | ubuf = (unsigned long) iov->iov_base; | |
8158 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8159 | start = ubuf >> PAGE_SHIFT; | |
8160 | nr_pages = end - start; | |
8161 | ||
8162 | ret = -ENOMEM; | |
8163 | ||
8164 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8165 | if (!pages) | |
8166 | goto done; | |
8167 | ||
8168 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8169 | GFP_KERNEL); | |
8170 | if (!vmas) | |
8171 | goto done; | |
edafccee | 8172 | |
0a96bbe4 BM |
8173 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8174 | GFP_KERNEL); | |
8175 | if (!imu->bvec) | |
8176 | goto done; | |
8177 | ||
8178 | ret = 0; | |
8179 | mmap_read_lock(current->mm); | |
8180 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8181 | pages, vmas); | |
8182 | if (pret == nr_pages) { | |
8183 | /* don't support file backed memory */ | |
8184 | for (i = 0; i < nr_pages; i++) { | |
8185 | struct vm_area_struct *vma = vmas[i]; | |
8186 | ||
8187 | if (vma->vm_file && | |
8188 | !is_file_hugepages(vma->vm_file)) { | |
8189 | ret = -EOPNOTSUPP; | |
8190 | break; | |
8191 | } | |
8192 | } | |
8193 | } else { | |
8194 | ret = pret < 0 ? pret : -EFAULT; | |
8195 | } | |
8196 | mmap_read_unlock(current->mm); | |
8197 | if (ret) { | |
8198 | /* | |
8199 | * if we did partial map, or found file backed vmas, | |
8200 | * release any pages we did get | |
8201 | */ | |
8202 | if (pret > 0) | |
8203 | unpin_user_pages(pages, pret); | |
8204 | kvfree(imu->bvec); | |
8205 | goto done; | |
8206 | } | |
8207 | ||
8208 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8209 | if (ret) { | |
8210 | unpin_user_pages(pages, pret); | |
8211 | kvfree(imu->bvec); | |
8212 | goto done; | |
8213 | } | |
8214 | ||
8215 | off = ubuf & ~PAGE_MASK; | |
8216 | size = iov->iov_len; | |
8217 | for (i = 0; i < nr_pages; i++) { | |
8218 | size_t vec_len; | |
8219 | ||
8220 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8221 | imu->bvec[i].bv_page = pages[i]; | |
8222 | imu->bvec[i].bv_len = vec_len; | |
8223 | imu->bvec[i].bv_offset = off; | |
8224 | off = 0; | |
8225 | size -= vec_len; | |
8226 | } | |
8227 | /* store original address for later verification */ | |
8228 | imu->ubuf = ubuf; | |
8229 | imu->len = iov->iov_len; | |
8230 | imu->nr_bvecs = nr_pages; | |
8231 | ret = 0; | |
8232 | done: | |
8233 | kvfree(pages); | |
8234 | kvfree(vmas); | |
8235 | return ret; | |
8236 | } | |
8237 | ||
2b358604 | 8238 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8239 | { |
edafccee JA |
8240 | if (ctx->user_bufs) |
8241 | return -EBUSY; | |
8242 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8243 | return -EINVAL; | |
8244 | ||
8245 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8246 | GFP_KERNEL); | |
8247 | if (!ctx->user_bufs) | |
8248 | return -ENOMEM; | |
8249 | ||
2b358604 BM |
8250 | return 0; |
8251 | } | |
edafccee | 8252 | |
2b358604 BM |
8253 | static int io_buffer_validate(struct iovec *iov) |
8254 | { | |
8255 | /* | |
8256 | * Don't impose further limits on the size and buffer | |
8257 | * constraints here, we'll -EINVAL later when IO is | |
8258 | * submitted if they are wrong. | |
8259 | */ | |
8260 | if (!iov->iov_base || !iov->iov_len) | |
8261 | return -EFAULT; | |
edafccee | 8262 | |
2b358604 BM |
8263 | /* arbitrary limit, but we need something */ |
8264 | if (iov->iov_len > SZ_1G) | |
8265 | return -EFAULT; | |
edafccee | 8266 | |
2b358604 BM |
8267 | return 0; |
8268 | } | |
edafccee | 8269 | |
2b358604 BM |
8270 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8271 | unsigned int nr_args) | |
8272 | { | |
8273 | int i, ret; | |
8274 | struct iovec iov; | |
8275 | struct page *last_hpage = NULL; | |
edafccee | 8276 | |
2b358604 BM |
8277 | ret = io_buffers_map_alloc(ctx, nr_args); |
8278 | if (ret) | |
8279 | return ret; | |
edafccee | 8280 | |
edafccee JA |
8281 | for (i = 0; i < nr_args; i++) { |
8282 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8283 | |
edafccee JA |
8284 | ret = io_copy_iov(ctx, &iov, arg, i); |
8285 | if (ret) | |
0a96bbe4 | 8286 | break; |
de293938 | 8287 | |
2b358604 BM |
8288 | ret = io_buffer_validate(&iov); |
8289 | if (ret) | |
0a96bbe4 | 8290 | break; |
edafccee | 8291 | |
0a96bbe4 BM |
8292 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8293 | if (ret) | |
8294 | break; | |
edafccee JA |
8295 | |
8296 | ctx->nr_user_bufs++; | |
8297 | } | |
0a96bbe4 BM |
8298 | |
8299 | if (ret) | |
8300 | io_sqe_buffers_unregister(ctx); | |
8301 | ||
edafccee JA |
8302 | return ret; |
8303 | } | |
8304 | ||
9b402849 JA |
8305 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8306 | { | |
8307 | __s32 __user *fds = arg; | |
8308 | int fd; | |
8309 | ||
8310 | if (ctx->cq_ev_fd) | |
8311 | return -EBUSY; | |
8312 | ||
8313 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8314 | return -EFAULT; | |
8315 | ||
8316 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8317 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8318 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8319 | ctx->cq_ev_fd = NULL; | |
8320 | return ret; | |
8321 | } | |
8322 | ||
8323 | return 0; | |
8324 | } | |
8325 | ||
8326 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8327 | { | |
8328 | if (ctx->cq_ev_fd) { | |
8329 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8330 | ctx->cq_ev_fd = NULL; | |
8331 | return 0; | |
8332 | } | |
8333 | ||
8334 | return -ENXIO; | |
8335 | } | |
8336 | ||
5a2e745d JA |
8337 | static int __io_destroy_buffers(int id, void *p, void *data) |
8338 | { | |
8339 | struct io_ring_ctx *ctx = data; | |
8340 | struct io_buffer *buf = p; | |
8341 | ||
067524e9 | 8342 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8343 | return 0; |
8344 | } | |
8345 | ||
8346 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8347 | { | |
8348 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8349 | idr_destroy(&ctx->io_buffer_idr); | |
8350 | } | |
8351 | ||
68e68ee6 | 8352 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8353 | { |
68e68ee6 | 8354 | struct io_kiocb *req, *nxt; |
1b4c351f | 8355 | |
68e68ee6 JA |
8356 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8357 | if (tsk && req->task != tsk) | |
8358 | continue; | |
1b4c351f JA |
8359 | list_del(&req->compl.list); |
8360 | kmem_cache_free(req_cachep, req); | |
8361 | } | |
8362 | } | |
8363 | ||
4010fec4 | 8364 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8365 | { |
bf019da7 | 8366 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8367 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8368 | |
9a4fdbd8 JA |
8369 | mutex_lock(&ctx->uring_lock); |
8370 | ||
8e5c66c4 | 8371 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8372 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8373 | submit_state->reqs); | |
8e5c66c4 PB |
8374 | submit_state->free_reqs = 0; |
8375 | } | |
9a4fdbd8 JA |
8376 | |
8377 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8378 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8379 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8380 | spin_unlock_irq(&ctx->completion_lock); |
8381 | ||
e5547d2c PB |
8382 | io_req_cache_free(&cs->free_list, NULL); |
8383 | ||
9a4fdbd8 JA |
8384 | mutex_unlock(&ctx->uring_lock); |
8385 | } | |
8386 | ||
2b188cc1 JA |
8387 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8388 | { | |
04fc6c80 PB |
8389 | /* |
8390 | * Some may use context even when all refs and requests have been put, | |
8391 | * and they are free to do so while still holding uring_lock, see | |
8392 | * __io_req_task_submit(). Wait for them to finish. | |
8393 | */ | |
8394 | mutex_lock(&ctx->uring_lock); | |
8395 | mutex_unlock(&ctx->uring_lock); | |
8396 | ||
37d1e2e3 | 8397 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8398 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8399 | |
37d1e2e3 | 8400 | if (ctx->mm_account) { |
2aede0e4 JA |
8401 | mmdrop(ctx->mm_account); |
8402 | ctx->mm_account = NULL; | |
30975825 | 8403 | } |
def596e9 | 8404 | |
8bad28d8 | 8405 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8406 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8407 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8408 | io_eventfd_unregister(ctx); |
5a2e745d | 8409 | io_destroy_buffers(ctx); |
41726c9a | 8410 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8411 | |
2b188cc1 | 8412 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8413 | if (ctx->ring_sock) { |
8414 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8415 | sock_release(ctx->ring_sock); |
355e8d26 | 8416 | } |
2b188cc1 JA |
8417 | #endif |
8418 | ||
75b28aff | 8419 | io_mem_free(ctx->rings); |
2b188cc1 | 8420 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8421 | |
8422 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8423 | free_uid(ctx->user); |
4010fec4 | 8424 | io_req_caches_free(ctx); |
e941894e JA |
8425 | if (ctx->hash_map) |
8426 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8427 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8428 | kfree(ctx); |
8429 | } | |
8430 | ||
8431 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8432 | { | |
8433 | struct io_ring_ctx *ctx = file->private_data; | |
8434 | __poll_t mask = 0; | |
8435 | ||
8436 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8437 | /* |
8438 | * synchronizes with barrier from wq_has_sleeper call in | |
8439 | * io_commit_cqring | |
8440 | */ | |
2b188cc1 | 8441 | smp_rmb(); |
90554200 | 8442 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8443 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8444 | |
8445 | /* | |
8446 | * Don't flush cqring overflow list here, just do a simple check. | |
8447 | * Otherwise there could possible be ABBA deadlock: | |
8448 | * CPU0 CPU1 | |
8449 | * ---- ---- | |
8450 | * lock(&ctx->uring_lock); | |
8451 | * lock(&ep->mtx); | |
8452 | * lock(&ctx->uring_lock); | |
8453 | * lock(&ep->mtx); | |
8454 | * | |
8455 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8456 | * pushs them to do the flush. | |
8457 | */ | |
8458 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8459 | mask |= EPOLLIN | EPOLLRDNORM; |
8460 | ||
8461 | return mask; | |
8462 | } | |
8463 | ||
8464 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8465 | { | |
8466 | struct io_ring_ctx *ctx = file->private_data; | |
8467 | ||
8468 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8469 | } | |
8470 | ||
0bead8cd | 8471 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8472 | { |
4379bf8b | 8473 | const struct cred *creds; |
071698e1 | 8474 | |
4379bf8b JA |
8475 | creds = idr_remove(&ctx->personality_idr, id); |
8476 | if (creds) { | |
8477 | put_cred(creds); | |
0bead8cd | 8478 | return 0; |
1e6fa521 | 8479 | } |
0bead8cd YD |
8480 | |
8481 | return -EINVAL; | |
8482 | } | |
8483 | ||
8484 | static int io_remove_personalities(int id, void *p, void *data) | |
8485 | { | |
8486 | struct io_ring_ctx *ctx = data; | |
8487 | ||
8488 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8489 | return 0; |
8490 | } | |
8491 | ||
ba50a036 | 8492 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8493 | { |
28c4721b | 8494 | struct callback_head *work, *next; |
ba50a036 | 8495 | bool executed = false; |
7c25c0d1 JA |
8496 | |
8497 | do { | |
28c4721b | 8498 | work = xchg(&ctx->exit_task_work, NULL); |
7c25c0d1 JA |
8499 | if (!work) |
8500 | break; | |
8501 | ||
8502 | do { | |
8503 | next = work->next; | |
8504 | work->func(work); | |
8505 | work = next; | |
8506 | cond_resched(); | |
8507 | } while (work); | |
ba50a036 | 8508 | executed = true; |
7c25c0d1 | 8509 | } while (1); |
ba50a036 PB |
8510 | |
8511 | return executed; | |
7c25c0d1 JA |
8512 | } |
8513 | ||
85faa7b8 JA |
8514 | static void io_ring_exit_work(struct work_struct *work) |
8515 | { | |
b2edc0a7 PB |
8516 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8517 | exit_work); | |
85faa7b8 | 8518 | |
56952e91 JA |
8519 | /* |
8520 | * If we're doing polled IO and end up having requests being | |
8521 | * submitted async (out-of-line), then completions can come in while | |
8522 | * we're waiting for refs to drop. We need to reap these manually, | |
8523 | * as nobody else will be looking for them. | |
8524 | */ | |
b2edc0a7 | 8525 | do { |
9936c7c2 | 8526 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8527 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8528 | io_ring_ctx_free(ctx); |
8529 | } | |
8530 | ||
2b188cc1 JA |
8531 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8532 | { | |
8533 | mutex_lock(&ctx->uring_lock); | |
8534 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8535 | /* if force is set, the ring is going away. always drop after that */ |
8536 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8537 | if (ctx->rings) |
6c503150 | 8538 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8539 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8540 | mutex_unlock(&ctx->uring_lock); |
8541 | ||
6b81928d PB |
8542 | io_kill_timeouts(ctx, NULL, NULL); |
8543 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8544 | |
15dff286 | 8545 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8546 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8547 | |
85faa7b8 | 8548 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8549 | /* |
8550 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8551 | * if we're exiting a ton of rings at the same time. It just adds | |
8552 | * noise and overhead, there's no discernable change in runtime | |
8553 | * over using system_wq. | |
8554 | */ | |
8555 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8556 | } |
8557 | ||
8558 | static int io_uring_release(struct inode *inode, struct file *file) | |
8559 | { | |
8560 | struct io_ring_ctx *ctx = file->private_data; | |
8561 | ||
8562 | file->private_data = NULL; | |
8563 | io_ring_ctx_wait_and_kill(ctx); | |
8564 | return 0; | |
8565 | } | |
8566 | ||
f6edbabb PB |
8567 | struct io_task_cancel { |
8568 | struct task_struct *task; | |
8569 | struct files_struct *files; | |
8570 | }; | |
f254ac04 | 8571 | |
f6edbabb | 8572 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8573 | { |
9a472ef7 | 8574 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8575 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8576 | bool ret; |
8577 | ||
f6edbabb | 8578 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8579 | unsigned long flags; |
8580 | struct io_ring_ctx *ctx = req->ctx; | |
8581 | ||
8582 | /* protect against races with linked timeouts */ | |
8583 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8584 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8585 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8586 | } else { | |
f6edbabb | 8587 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8588 | } |
8589 | return ret; | |
b711d4ea JA |
8590 | } |
8591 | ||
b7ddce3c | 8592 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8593 | struct task_struct *task, |
b7ddce3c PB |
8594 | struct files_struct *files) |
8595 | { | |
8596 | struct io_defer_entry *de = NULL; | |
8597 | LIST_HEAD(list); | |
8598 | ||
8599 | spin_lock_irq(&ctx->completion_lock); | |
8600 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8601 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8602 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8603 | break; | |
8604 | } | |
8605 | } | |
8606 | spin_unlock_irq(&ctx->completion_lock); | |
8607 | ||
8608 | while (!list_empty(&list)) { | |
8609 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8610 | list_del_init(&de->list); | |
8611 | req_set_fail_links(de->req); | |
8612 | io_put_req(de->req); | |
8613 | io_req_complete(de->req, -ECANCELED); | |
8614 | kfree(de); | |
8615 | } | |
8616 | } | |
8617 | ||
9936c7c2 PB |
8618 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8619 | struct task_struct *task, | |
8620 | struct files_struct *files) | |
8621 | { | |
8622 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
64c72123 PB |
8623 | struct task_struct *tctx_task = task ?: current; |
8624 | struct io_uring_task *tctx = tctx_task->io_uring; | |
9936c7c2 PB |
8625 | |
8626 | while (1) { | |
8627 | enum io_wq_cancel cret; | |
8628 | bool ret = false; | |
8629 | ||
5aa75ed5 JA |
8630 | if (tctx && tctx->io_wq) { |
8631 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, | |
9936c7c2 PB |
8632 | &cancel, true); |
8633 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8634 | } | |
8635 | ||
8636 | /* SQPOLL thread does its own polling */ | |
8637 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8638 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8639 | io_iopoll_try_reap_events(ctx); | |
8640 | ret = true; | |
8641 | } | |
8642 | } | |
8643 | ||
8644 | ret |= io_poll_remove_all(ctx, task, files); | |
8645 | ret |= io_kill_timeouts(ctx, task, files); | |
8646 | ret |= io_run_task_work(); | |
ba50a036 | 8647 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8648 | io_cqring_overflow_flush(ctx, true, task, files); |
8649 | if (!ret) | |
8650 | break; | |
8651 | cond_resched(); | |
8652 | } | |
8653 | } | |
8654 | ||
ca70f00b PB |
8655 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8656 | struct task_struct *task, | |
8657 | struct files_struct *files) | |
8658 | { | |
8659 | struct io_kiocb *req; | |
8660 | int cnt = 0; | |
8661 | ||
8662 | spin_lock_irq(&ctx->inflight_lock); | |
8663 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8664 | cnt += io_match_task(req, task, files); | |
8665 | spin_unlock_irq(&ctx->inflight_lock); | |
8666 | return cnt; | |
8667 | } | |
8668 | ||
b52fda00 | 8669 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8670 | struct task_struct *task, |
fcb323cc JA |
8671 | struct files_struct *files) |
8672 | { | |
fcb323cc | 8673 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8674 | DEFINE_WAIT(wait); |
ca70f00b | 8675 | int inflight; |
fcb323cc | 8676 | |
ca70f00b PB |
8677 | inflight = io_uring_count_inflight(ctx, task, files); |
8678 | if (!inflight) | |
fcb323cc | 8679 | break; |
f6edbabb | 8680 | |
9936c7c2 | 8681 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b | 8682 | |
34343786 PB |
8683 | if (ctx->sq_data) |
8684 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
8685 | prepare_to_wait(&task->io_uring->wait, &wait, |
8686 | TASK_UNINTERRUPTIBLE); | |
8687 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8688 | schedule(); | |
c98de08c | 8689 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
8690 | if (ctx->sq_data) |
8691 | io_sq_thread_park(ctx->sq_data); | |
0f212204 | 8692 | } |
0f212204 JA |
8693 | } |
8694 | ||
8695 | /* | |
8696 | * We need to iteratively cancel requests, in case a request has dependent | |
8697 | * hard links. These persist even for failure of cancelations, hence keep | |
8698 | * looping until none are found. | |
8699 | */ | |
8700 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8701 | struct files_struct *files) | |
8702 | { | |
8703 | struct task_struct *task = current; | |
37d1e2e3 | 8704 | bool did_park = false; |
0f212204 | 8705 | |
fdaf083c | 8706 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
70aacfe6 PB |
8707 | /* never started, nothing to cancel */ |
8708 | if (ctx->flags & IORING_SETUP_R_DISABLED) { | |
8709 | io_sq_offload_start(ctx); | |
8710 | return; | |
8711 | } | |
37d1e2e3 JA |
8712 | did_park = io_sq_thread_park(ctx->sq_data); |
8713 | if (did_park) { | |
8714 | task = ctx->sq_data->thread; | |
8715 | atomic_inc(&task->io_uring->in_idle); | |
8716 | } | |
fdaf083c | 8717 | } |
0f212204 | 8718 | |
df9923f9 | 8719 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 8720 | |
3a7efd1a | 8721 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 8722 | if (!files) |
9936c7c2 | 8723 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c | 8724 | |
37d1e2e3 | 8725 | if (did_park) { |
fdaf083c | 8726 | atomic_dec(&task->io_uring->in_idle); |
fdaf083c JA |
8727 | io_sq_thread_unpark(ctx->sq_data); |
8728 | } | |
0f212204 JA |
8729 | } |
8730 | ||
8731 | /* | |
8732 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8733 | */ | |
fdaf083c | 8734 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 8735 | { |
236434c3 | 8736 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 8737 | int ret; |
236434c3 MWO |
8738 | |
8739 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8740 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8741 | if (unlikely(ret)) |
8742 | return ret; | |
236434c3 | 8743 | tctx = current->io_uring; |
0f212204 | 8744 | } |
236434c3 MWO |
8745 | if (tctx->last != file) { |
8746 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 8747 | |
236434c3 | 8748 | if (!old) { |
0f212204 | 8749 | get_file(file); |
a528b04e PB |
8750 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
8751 | file, GFP_KERNEL)); | |
8752 | if (ret) { | |
8753 | fput(file); | |
8754 | return ret; | |
8755 | } | |
0f212204 | 8756 | } |
236434c3 | 8757 | tctx->last = file; |
0f212204 JA |
8758 | } |
8759 | ||
fdaf083c JA |
8760 | /* |
8761 | * This is race safe in that the task itself is doing this, hence it | |
8762 | * cannot be going through the exit/cancel paths at the same time. | |
8763 | * This cannot be modified while exit/cancel is running. | |
8764 | */ | |
8765 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
8766 | tctx->sqpoll = true; | |
8767 | ||
0f212204 JA |
8768 | return 0; |
8769 | } | |
8770 | ||
8771 | /* | |
8772 | * Remove this io_uring_file -> task mapping. | |
8773 | */ | |
8774 | static void io_uring_del_task_file(struct file *file) | |
8775 | { | |
8776 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
8777 | |
8778 | if (tctx->last == file) | |
8779 | tctx->last = NULL; | |
5e2ed8c4 | 8780 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
8781 | if (file) |
8782 | fput(file); | |
8783 | } | |
8784 | ||
8452d4a6 | 8785 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e PB |
8786 | { |
8787 | struct file *file; | |
8788 | unsigned long index; | |
8789 | ||
8790 | xa_for_each(&tctx->xa, index, file) | |
8791 | io_uring_del_task_file(file); | |
8452d4a6 PB |
8792 | if (tctx->io_wq) { |
8793 | io_wq_put_and_exit(tctx->io_wq); | |
8794 | tctx->io_wq = NULL; | |
8795 | } | |
de7f1d9e PB |
8796 | } |
8797 | ||
0f212204 JA |
8798 | void __io_uring_files_cancel(struct files_struct *files) |
8799 | { | |
8800 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
8801 | struct file *file; |
8802 | unsigned long index; | |
0f212204 JA |
8803 | |
8804 | /* make sure overflow events are dropped */ | |
fdaf083c | 8805 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
8806 | xa_for_each(&tctx->xa, index, file) |
8807 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 8808 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8809 | |
8452d4a6 PB |
8810 | if (files) |
8811 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8812 | } |
8813 | ||
8814 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
8815 | { | |
0e9ddb39 PB |
8816 | return percpu_counter_sum(&tctx->inflight); |
8817 | } | |
fdaf083c | 8818 | |
0e9ddb39 PB |
8819 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8820 | { | |
37d1e2e3 | 8821 | struct io_sq_data *sqd = ctx->sq_data; |
0e9ddb39 PB |
8822 | struct io_uring_task *tctx; |
8823 | s64 inflight; | |
8824 | DEFINE_WAIT(wait); | |
fdaf083c | 8825 | |
37d1e2e3 | 8826 | if (!sqd) |
0e9ddb39 | 8827 | return; |
37d1e2e3 JA |
8828 | if (!io_sq_thread_park(sqd)) |
8829 | return; | |
8830 | tctx = ctx->sq_data->thread->io_uring; | |
e54945ae JA |
8831 | /* can happen on fork/alloc failure, just ignore that state */ |
8832 | if (!tctx) { | |
8833 | io_sq_thread_unpark(sqd); | |
8834 | return; | |
8835 | } | |
fdaf083c | 8836 | |
0e9ddb39 PB |
8837 | atomic_inc(&tctx->in_idle); |
8838 | do { | |
8839 | /* read completions before cancelations */ | |
8840 | inflight = tctx_inflight(tctx); | |
8841 | if (!inflight) | |
8842 | break; | |
8843 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 8844 | |
0e9ddb39 PB |
8845 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8846 | /* | |
8847 | * If we've seen completions, retry without waiting. This | |
8848 | * avoids a race where a completion comes in before we did | |
8849 | * prepare_to_wait(). | |
8850 | */ | |
8851 | if (inflight == tctx_inflight(tctx)) | |
8852 | schedule(); | |
8853 | finish_wait(&tctx->wait, &wait); | |
8854 | } while (1); | |
8855 | atomic_dec(&tctx->in_idle); | |
37d1e2e3 | 8856 | io_sq_thread_unpark(sqd); |
0f212204 JA |
8857 | } |
8858 | ||
0f212204 JA |
8859 | /* |
8860 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8861 | * requests. | |
8862 | */ | |
8863 | void __io_uring_task_cancel(void) | |
8864 | { | |
8865 | struct io_uring_task *tctx = current->io_uring; | |
8866 | DEFINE_WAIT(wait); | |
d8a6df10 | 8867 | s64 inflight; |
0f212204 JA |
8868 | |
8869 | /* make sure overflow events are dropped */ | |
fdaf083c | 8870 | atomic_inc(&tctx->in_idle); |
0f212204 | 8871 | |
0e9ddb39 PB |
8872 | if (tctx->sqpoll) { |
8873 | struct file *file; | |
8874 | unsigned long index; | |
8875 | ||
8876 | xa_for_each(&tctx->xa, index, file) | |
8877 | io_uring_cancel_sqpoll(file->private_data); | |
8878 | } | |
0b5cd6c3 | 8879 | |
d8a6df10 | 8880 | do { |
0f212204 | 8881 | /* read completions before cancelations */ |
fdaf083c | 8882 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8883 | if (!inflight) |
8884 | break; | |
0f212204 JA |
8885 | __io_uring_files_cancel(NULL); |
8886 | ||
8887 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8888 | ||
8889 | /* | |
a1bb3cd5 PB |
8890 | * If we've seen completions, retry without waiting. This |
8891 | * avoids a race where a completion comes in before we did | |
8892 | * prepare_to_wait(). | |
0f212204 | 8893 | */ |
a1bb3cd5 PB |
8894 | if (inflight == tctx_inflight(tctx)) |
8895 | schedule(); | |
f57555ed | 8896 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8897 | } while (1); |
0f212204 | 8898 | |
fdaf083c | 8899 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8900 | |
8452d4a6 PB |
8901 | io_uring_clean_tctx(tctx); |
8902 | /* all current's requests should be gone, we can kill tctx */ | |
8903 | __io_uring_free(current); | |
44e728b8 PB |
8904 | } |
8905 | ||
6c5c240e RP |
8906 | static void *io_uring_validate_mmap_request(struct file *file, |
8907 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8908 | { |
2b188cc1 | 8909 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8910 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8911 | struct page *page; |
8912 | void *ptr; | |
8913 | ||
8914 | switch (offset) { | |
8915 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8916 | case IORING_OFF_CQ_RING: |
8917 | ptr = ctx->rings; | |
2b188cc1 JA |
8918 | break; |
8919 | case IORING_OFF_SQES: | |
8920 | ptr = ctx->sq_sqes; | |
8921 | break; | |
2b188cc1 | 8922 | default: |
6c5c240e | 8923 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8924 | } |
8925 | ||
8926 | page = virt_to_head_page(ptr); | |
a50b854e | 8927 | if (sz > page_size(page)) |
6c5c240e RP |
8928 | return ERR_PTR(-EINVAL); |
8929 | ||
8930 | return ptr; | |
8931 | } | |
8932 | ||
8933 | #ifdef CONFIG_MMU | |
8934 | ||
8935 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8936 | { | |
8937 | size_t sz = vma->vm_end - vma->vm_start; | |
8938 | unsigned long pfn; | |
8939 | void *ptr; | |
8940 | ||
8941 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
8942 | if (IS_ERR(ptr)) | |
8943 | return PTR_ERR(ptr); | |
2b188cc1 JA |
8944 | |
8945 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
8946 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
8947 | } | |
8948 | ||
6c5c240e RP |
8949 | #else /* !CONFIG_MMU */ |
8950 | ||
8951 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
8952 | { | |
8953 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
8954 | } | |
8955 | ||
8956 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
8957 | { | |
8958 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
8959 | } | |
8960 | ||
8961 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
8962 | unsigned long addr, unsigned long len, | |
8963 | unsigned long pgoff, unsigned long flags) | |
8964 | { | |
8965 | void *ptr; | |
8966 | ||
8967 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
8968 | if (IS_ERR(ptr)) | |
8969 | return PTR_ERR(ptr); | |
8970 | ||
8971 | return (unsigned long) ptr; | |
8972 | } | |
8973 | ||
8974 | #endif /* !CONFIG_MMU */ | |
8975 | ||
d9d05217 | 8976 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 8977 | { |
d9d05217 | 8978 | int ret = 0; |
90554200 JA |
8979 | DEFINE_WAIT(wait); |
8980 | ||
8981 | do { | |
8982 | if (!io_sqring_full(ctx)) | |
8983 | break; | |
90554200 JA |
8984 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
8985 | ||
8986 | if (!io_sqring_full(ctx)) | |
8987 | break; | |
90554200 JA |
8988 | schedule(); |
8989 | } while (!signal_pending(current)); | |
8990 | ||
8991 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 | 8992 | return ret; |
90554200 JA |
8993 | } |
8994 | ||
c73ebb68 HX |
8995 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
8996 | struct __kernel_timespec __user **ts, | |
8997 | const sigset_t __user **sig) | |
8998 | { | |
8999 | struct io_uring_getevents_arg arg; | |
9000 | ||
9001 | /* | |
9002 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9003 | * is just a pointer to the sigset_t. | |
9004 | */ | |
9005 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9006 | *sig = (const sigset_t __user *) argp; | |
9007 | *ts = NULL; | |
9008 | return 0; | |
9009 | } | |
9010 | ||
9011 | /* | |
9012 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9013 | * timespec and sigset_t pointers if good. | |
9014 | */ | |
9015 | if (*argsz != sizeof(arg)) | |
9016 | return -EINVAL; | |
9017 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9018 | return -EFAULT; | |
9019 | *sig = u64_to_user_ptr(arg.sigmask); | |
9020 | *argsz = arg.sigmask_sz; | |
9021 | *ts = u64_to_user_ptr(arg.ts); | |
9022 | return 0; | |
9023 | } | |
9024 | ||
2b188cc1 | 9025 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9026 | u32, min_complete, u32, flags, const void __user *, argp, |
9027 | size_t, argsz) | |
2b188cc1 JA |
9028 | { |
9029 | struct io_ring_ctx *ctx; | |
9030 | long ret = -EBADF; | |
9031 | int submitted = 0; | |
9032 | struct fd f; | |
9033 | ||
4c6e277c | 9034 | io_run_task_work(); |
b41e9852 | 9035 | |
90554200 | 9036 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9037 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9038 | return -EINVAL; |
9039 | ||
9040 | f = fdget(fd); | |
9041 | if (!f.file) | |
9042 | return -EBADF; | |
9043 | ||
9044 | ret = -EOPNOTSUPP; | |
9045 | if (f.file->f_op != &io_uring_fops) | |
9046 | goto out_fput; | |
9047 | ||
9048 | ret = -ENXIO; | |
9049 | ctx = f.file->private_data; | |
9050 | if (!percpu_ref_tryget(&ctx->refs)) | |
9051 | goto out_fput; | |
9052 | ||
7e84e1c7 SG |
9053 | ret = -EBADFD; |
9054 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9055 | goto out; | |
9056 | ||
6c271ce2 JA |
9057 | /* |
9058 | * For SQ polling, the thread will do all submissions and completions. | |
9059 | * Just return the requested submit count, and wake the thread if | |
9060 | * we were asked to. | |
9061 | */ | |
b2a9eada | 9062 | ret = 0; |
6c271ce2 | 9063 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9064 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9065 | |
5f3f26f9 JA |
9066 | if (unlikely(ctx->sqo_exec)) { |
9067 | ret = io_sq_thread_fork(ctx->sq_data, ctx); | |
9068 | if (ret) | |
9069 | goto out; | |
9070 | ctx->sqo_exec = 0; | |
9071 | } | |
d9d05217 | 9072 | ret = -EOWNERDEAD; |
6c271ce2 | 9073 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9074 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9075 | if (flags & IORING_ENTER_SQ_WAIT) { |
9076 | ret = io_sqpoll_wait_sq(ctx); | |
9077 | if (ret) | |
9078 | goto out; | |
9079 | } | |
6c271ce2 | 9080 | submitted = to_submit; |
b2a9eada | 9081 | } else if (to_submit) { |
fdaf083c | 9082 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9083 | if (unlikely(ret)) |
9084 | goto out; | |
2b188cc1 | 9085 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9086 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9087 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9088 | |
9089 | if (submitted != to_submit) | |
9090 | goto out; | |
2b188cc1 JA |
9091 | } |
9092 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9093 | const sigset_t __user *sig; |
9094 | struct __kernel_timespec __user *ts; | |
9095 | ||
9096 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9097 | if (unlikely(ret)) | |
9098 | goto out; | |
9099 | ||
2b188cc1 JA |
9100 | min_complete = min(min_complete, ctx->cq_entries); |
9101 | ||
32b2244a XW |
9102 | /* |
9103 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9104 | * space applications don't need to do io completion events | |
9105 | * polling again, they can rely on io_sq_thread to do polling | |
9106 | * work, which can reduce cpu usage and uring_lock contention. | |
9107 | */ | |
9108 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9109 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9110 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9111 | } else { |
c73ebb68 | 9112 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9113 | } |
2b188cc1 JA |
9114 | } |
9115 | ||
7c504e65 | 9116 | out: |
6805b32e | 9117 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9118 | out_fput: |
9119 | fdput(f); | |
9120 | return submitted ? submitted : ret; | |
9121 | } | |
9122 | ||
bebdb65e | 9123 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9124 | static int io_uring_show_cred(int id, void *p, void *data) |
9125 | { | |
4379bf8b | 9126 | const struct cred *cred = p; |
87ce955b JA |
9127 | struct seq_file *m = data; |
9128 | struct user_namespace *uns = seq_user_ns(m); | |
9129 | struct group_info *gi; | |
9130 | kernel_cap_t cap; | |
9131 | unsigned __capi; | |
9132 | int g; | |
9133 | ||
9134 | seq_printf(m, "%5d\n", id); | |
9135 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9136 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9137 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9138 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9139 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9140 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9141 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9142 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9143 | seq_puts(m, "\n\tGroups:\t"); | |
9144 | gi = cred->group_info; | |
9145 | for (g = 0; g < gi->ngroups; g++) { | |
9146 | seq_put_decimal_ull(m, g ? " " : "", | |
9147 | from_kgid_munged(uns, gi->gid[g])); | |
9148 | } | |
9149 | seq_puts(m, "\n\tCapEff:\t"); | |
9150 | cap = cred->cap_effective; | |
9151 | CAP_FOR_EACH_U32(__capi) | |
9152 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9153 | seq_putc(m, '\n'); | |
9154 | return 0; | |
9155 | } | |
9156 | ||
9157 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9158 | { | |
dbbe9c64 | 9159 | struct io_sq_data *sq = NULL; |
fad8e0de | 9160 | bool has_lock; |
87ce955b JA |
9161 | int i; |
9162 | ||
fad8e0de JA |
9163 | /* |
9164 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9165 | * since fdinfo case grabs it in the opposite direction of normal use | |
9166 | * cases. If we fail to get the lock, we just don't iterate any | |
9167 | * structures that could be going away outside the io_uring mutex. | |
9168 | */ | |
9169 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9170 | ||
5f3f26f9 | 9171 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9172 | sq = ctx->sq_data; |
5f3f26f9 JA |
9173 | if (!sq->thread) |
9174 | sq = NULL; | |
9175 | } | |
dbbe9c64 JQ |
9176 | |
9177 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9178 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9179 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9180 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9181 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9182 | |
87ce955b JA |
9183 | if (f) |
9184 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9185 | else | |
9186 | seq_printf(m, "%5u: <none>\n", i); | |
9187 | } | |
9188 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9189 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9190 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9191 | ||
9192 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9193 | (unsigned int) buf->len); | |
9194 | } | |
fad8e0de | 9195 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9196 | seq_printf(m, "Personalities:\n"); |
9197 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9198 | } | |
d7718a9d JA |
9199 | seq_printf(m, "PollList:\n"); |
9200 | spin_lock_irq(&ctx->completion_lock); | |
9201 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9202 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9203 | struct io_kiocb *req; | |
9204 | ||
9205 | hlist_for_each_entry(req, list, hash_node) | |
9206 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9207 | req->task->task_works != NULL); | |
9208 | } | |
9209 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9210 | if (has_lock) |
9211 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9212 | } |
9213 | ||
9214 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9215 | { | |
9216 | struct io_ring_ctx *ctx = f->private_data; | |
9217 | ||
9218 | if (percpu_ref_tryget(&ctx->refs)) { | |
9219 | __io_uring_show_fdinfo(ctx, m); | |
9220 | percpu_ref_put(&ctx->refs); | |
9221 | } | |
9222 | } | |
bebdb65e | 9223 | #endif |
87ce955b | 9224 | |
2b188cc1 JA |
9225 | static const struct file_operations io_uring_fops = { |
9226 | .release = io_uring_release, | |
9227 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9228 | #ifndef CONFIG_MMU |
9229 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9230 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9231 | #endif | |
2b188cc1 JA |
9232 | .poll = io_uring_poll, |
9233 | .fasync = io_uring_fasync, | |
bebdb65e | 9234 | #ifdef CONFIG_PROC_FS |
87ce955b | 9235 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9236 | #endif |
2b188cc1 JA |
9237 | }; |
9238 | ||
9239 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9240 | struct io_uring_params *p) | |
9241 | { | |
75b28aff HV |
9242 | struct io_rings *rings; |
9243 | size_t size, sq_array_offset; | |
2b188cc1 | 9244 | |
bd740481 JA |
9245 | /* make sure these are sane, as we already accounted them */ |
9246 | ctx->sq_entries = p->sq_entries; | |
9247 | ctx->cq_entries = p->cq_entries; | |
9248 | ||
75b28aff HV |
9249 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9250 | if (size == SIZE_MAX) | |
9251 | return -EOVERFLOW; | |
9252 | ||
9253 | rings = io_mem_alloc(size); | |
9254 | if (!rings) | |
2b188cc1 JA |
9255 | return -ENOMEM; |
9256 | ||
75b28aff HV |
9257 | ctx->rings = rings; |
9258 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9259 | rings->sq_ring_mask = p->sq_entries - 1; | |
9260 | rings->cq_ring_mask = p->cq_entries - 1; | |
9261 | rings->sq_ring_entries = p->sq_entries; | |
9262 | rings->cq_ring_entries = p->cq_entries; | |
9263 | ctx->sq_mask = rings->sq_ring_mask; | |
9264 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9265 | |
9266 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9267 | if (size == SIZE_MAX) { |
9268 | io_mem_free(ctx->rings); | |
9269 | ctx->rings = NULL; | |
2b188cc1 | 9270 | return -EOVERFLOW; |
eb065d30 | 9271 | } |
2b188cc1 JA |
9272 | |
9273 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9274 | if (!ctx->sq_sqes) { |
9275 | io_mem_free(ctx->rings); | |
9276 | ctx->rings = NULL; | |
2b188cc1 | 9277 | return -ENOMEM; |
eb065d30 | 9278 | } |
2b188cc1 | 9279 | |
2b188cc1 JA |
9280 | return 0; |
9281 | } | |
9282 | ||
9faadcc8 PB |
9283 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9284 | { | |
9285 | int ret, fd; | |
9286 | ||
9287 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9288 | if (fd < 0) | |
9289 | return fd; | |
9290 | ||
9291 | ret = io_uring_add_task_file(ctx, file); | |
9292 | if (ret) { | |
9293 | put_unused_fd(fd); | |
9294 | return ret; | |
9295 | } | |
9296 | fd_install(fd, file); | |
9297 | return fd; | |
9298 | } | |
9299 | ||
2b188cc1 JA |
9300 | /* |
9301 | * Allocate an anonymous fd, this is what constitutes the application | |
9302 | * visible backing of an io_uring instance. The application mmaps this | |
9303 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9304 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9305 | */ | |
9faadcc8 | 9306 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9307 | { |
9308 | struct file *file; | |
9faadcc8 | 9309 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9310 | int ret; |
9311 | ||
2b188cc1 JA |
9312 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9313 | &ctx->ring_sock); | |
9314 | if (ret) | |
9faadcc8 | 9315 | return ERR_PTR(ret); |
2b188cc1 JA |
9316 | #endif |
9317 | ||
2b188cc1 JA |
9318 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9319 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9320 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9321 | if (IS_ERR(file)) { |
9322 | sock_release(ctx->ring_sock); | |
9323 | ctx->ring_sock = NULL; | |
9324 | } else { | |
9325 | ctx->ring_sock->file = file; | |
0f212204 | 9326 | } |
2b188cc1 | 9327 | #endif |
9faadcc8 | 9328 | return file; |
2b188cc1 JA |
9329 | } |
9330 | ||
7f13657d XW |
9331 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9332 | struct io_uring_params __user *params) | |
2b188cc1 | 9333 | { |
2b188cc1 | 9334 | struct io_ring_ctx *ctx; |
9faadcc8 | 9335 | struct file *file; |
2b188cc1 JA |
9336 | int ret; |
9337 | ||
8110c1a6 | 9338 | if (!entries) |
2b188cc1 | 9339 | return -EINVAL; |
8110c1a6 JA |
9340 | if (entries > IORING_MAX_ENTRIES) { |
9341 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9342 | return -EINVAL; | |
9343 | entries = IORING_MAX_ENTRIES; | |
9344 | } | |
2b188cc1 JA |
9345 | |
9346 | /* | |
9347 | * Use twice as many entries for the CQ ring. It's possible for the | |
9348 | * application to drive a higher depth than the size of the SQ ring, | |
9349 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9350 | * some flexibility in overcommitting a bit. If the application has |
9351 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9352 | * of CQ ring entries manually. | |
2b188cc1 JA |
9353 | */ |
9354 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9355 | if (p->flags & IORING_SETUP_CQSIZE) { |
9356 | /* | |
9357 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9358 | * to a power-of-two, if it isn't already. We do NOT impose | |
9359 | * any cq vs sq ring sizing. | |
9360 | */ | |
eb2667b3 | 9361 | if (!p->cq_entries) |
33a107f0 | 9362 | return -EINVAL; |
8110c1a6 JA |
9363 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9364 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9365 | return -EINVAL; | |
9366 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9367 | } | |
eb2667b3 JQ |
9368 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9369 | if (p->cq_entries < p->sq_entries) | |
9370 | return -EINVAL; | |
33a107f0 JA |
9371 | } else { |
9372 | p->cq_entries = 2 * p->sq_entries; | |
9373 | } | |
2b188cc1 | 9374 | |
2b188cc1 | 9375 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9376 | if (!ctx) |
2b188cc1 | 9377 | return -ENOMEM; |
2b188cc1 | 9378 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9379 | if (!capable(CAP_IPC_LOCK)) |
9380 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9381 | |
9382 | /* | |
9383 | * This is just grabbed for accounting purposes. When a process exits, | |
9384 | * the mm is exited and dropped before the files, hence we need to hang | |
9385 | * on to this mm purely for the purposes of being able to unaccount | |
9386 | * memory (locked/pinned vm). It's not used for anything else. | |
9387 | */ | |
6b7898eb | 9388 | mmgrab(current->mm); |
2aede0e4 | 9389 | ctx->mm_account = current->mm; |
6b7898eb | 9390 | |
2b188cc1 JA |
9391 | ret = io_allocate_scq_urings(ctx, p); |
9392 | if (ret) | |
9393 | goto err; | |
9394 | ||
7e84e1c7 | 9395 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9396 | if (ret) |
9397 | goto err; | |
9398 | ||
7e84e1c7 SG |
9399 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9400 | io_sq_offload_start(ctx); | |
9401 | ||
2b188cc1 | 9402 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9403 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9404 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9405 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9406 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9407 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9408 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9409 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9410 | |
9411 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9412 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9413 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9414 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9415 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9416 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9417 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9418 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9419 | |
7f13657d XW |
9420 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9421 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9422 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9423 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9424 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9425 | |
9426 | if (copy_to_user(params, p, sizeof(*p))) { | |
9427 | ret = -EFAULT; | |
9428 | goto err; | |
9429 | } | |
d1719f70 | 9430 | |
9faadcc8 PB |
9431 | file = io_uring_get_file(ctx); |
9432 | if (IS_ERR(file)) { | |
9433 | ret = PTR_ERR(file); | |
9434 | goto err; | |
9435 | } | |
9436 | ||
044c1ab3 JA |
9437 | /* |
9438 | * Install ring fd as the very last thing, so we don't risk someone | |
9439 | * having closed it before we finish setup | |
9440 | */ | |
9faadcc8 PB |
9441 | ret = io_uring_install_fd(ctx, file); |
9442 | if (ret < 0) { | |
9443 | /* fput will clean it up */ | |
9444 | fput(file); | |
9445 | return ret; | |
9446 | } | |
044c1ab3 | 9447 | |
c826bd7a | 9448 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9449 | return ret; |
9450 | err: | |
9451 | io_ring_ctx_wait_and_kill(ctx); | |
9452 | return ret; | |
9453 | } | |
9454 | ||
9455 | /* | |
9456 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9457 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9458 | * params structure passed in. | |
9459 | */ | |
9460 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9461 | { | |
9462 | struct io_uring_params p; | |
2b188cc1 JA |
9463 | int i; |
9464 | ||
9465 | if (copy_from_user(&p, params, sizeof(p))) | |
9466 | return -EFAULT; | |
9467 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9468 | if (p.resv[i]) | |
9469 | return -EINVAL; | |
9470 | } | |
9471 | ||
6c271ce2 | 9472 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9473 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9474 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9475 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9476 | return -EINVAL; |
9477 | ||
7f13657d | 9478 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9479 | } |
9480 | ||
9481 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9482 | struct io_uring_params __user *, params) | |
9483 | { | |
9484 | return io_uring_setup(entries, params); | |
9485 | } | |
9486 | ||
66f4af93 JA |
9487 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9488 | { | |
9489 | struct io_uring_probe *p; | |
9490 | size_t size; | |
9491 | int i, ret; | |
9492 | ||
9493 | size = struct_size(p, ops, nr_args); | |
9494 | if (size == SIZE_MAX) | |
9495 | return -EOVERFLOW; | |
9496 | p = kzalloc(size, GFP_KERNEL); | |
9497 | if (!p) | |
9498 | return -ENOMEM; | |
9499 | ||
9500 | ret = -EFAULT; | |
9501 | if (copy_from_user(p, arg, size)) | |
9502 | goto out; | |
9503 | ret = -EINVAL; | |
9504 | if (memchr_inv(p, 0, size)) | |
9505 | goto out; | |
9506 | ||
9507 | p->last_op = IORING_OP_LAST - 1; | |
9508 | if (nr_args > IORING_OP_LAST) | |
9509 | nr_args = IORING_OP_LAST; | |
9510 | ||
9511 | for (i = 0; i < nr_args; i++) { | |
9512 | p->ops[i].op = i; | |
9513 | if (!io_op_defs[i].not_supported) | |
9514 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9515 | } | |
9516 | p->ops_len = i; | |
9517 | ||
9518 | ret = 0; | |
9519 | if (copy_to_user(arg, p, size)) | |
9520 | ret = -EFAULT; | |
9521 | out: | |
9522 | kfree(p); | |
9523 | return ret; | |
9524 | } | |
9525 | ||
071698e1 JA |
9526 | static int io_register_personality(struct io_ring_ctx *ctx) |
9527 | { | |
4379bf8b | 9528 | const struct cred *creds; |
1e6fa521 | 9529 | int ret; |
071698e1 | 9530 | |
4379bf8b | 9531 | creds = get_current_cred(); |
1e6fa521 | 9532 | |
4379bf8b JA |
9533 | ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, |
9534 | USHRT_MAX, GFP_KERNEL); | |
9535 | if (ret < 0) | |
9536 | put_cred(creds); | |
1e6fa521 | 9537 | return ret; |
071698e1 JA |
9538 | } |
9539 | ||
21b55dbc SG |
9540 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9541 | unsigned int nr_args) | |
9542 | { | |
9543 | struct io_uring_restriction *res; | |
9544 | size_t size; | |
9545 | int i, ret; | |
9546 | ||
7e84e1c7 SG |
9547 | /* Restrictions allowed only if rings started disabled */ |
9548 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9549 | return -EBADFD; | |
9550 | ||
21b55dbc | 9551 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9552 | if (ctx->restrictions.registered) |
21b55dbc SG |
9553 | return -EBUSY; |
9554 | ||
9555 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9556 | return -EINVAL; | |
9557 | ||
9558 | size = array_size(nr_args, sizeof(*res)); | |
9559 | if (size == SIZE_MAX) | |
9560 | return -EOVERFLOW; | |
9561 | ||
9562 | res = memdup_user(arg, size); | |
9563 | if (IS_ERR(res)) | |
9564 | return PTR_ERR(res); | |
9565 | ||
9566 | ret = 0; | |
9567 | ||
9568 | for (i = 0; i < nr_args; i++) { | |
9569 | switch (res[i].opcode) { | |
9570 | case IORING_RESTRICTION_REGISTER_OP: | |
9571 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9572 | ret = -EINVAL; | |
9573 | goto out; | |
9574 | } | |
9575 | ||
9576 | __set_bit(res[i].register_op, | |
9577 | ctx->restrictions.register_op); | |
9578 | break; | |
9579 | case IORING_RESTRICTION_SQE_OP: | |
9580 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9581 | ret = -EINVAL; | |
9582 | goto out; | |
9583 | } | |
9584 | ||
9585 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9586 | break; | |
9587 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9588 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9589 | break; | |
9590 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9591 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9592 | break; | |
9593 | default: | |
9594 | ret = -EINVAL; | |
9595 | goto out; | |
9596 | } | |
9597 | } | |
9598 | ||
9599 | out: | |
9600 | /* Reset all restrictions if an error happened */ | |
9601 | if (ret != 0) | |
9602 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9603 | else | |
7e84e1c7 | 9604 | ctx->restrictions.registered = true; |
21b55dbc SG |
9605 | |
9606 | kfree(res); | |
9607 | return ret; | |
9608 | } | |
9609 | ||
7e84e1c7 SG |
9610 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9611 | { | |
9612 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9613 | return -EBADFD; | |
9614 | ||
9615 | if (ctx->restrictions.registered) | |
9616 | ctx->restricted = 1; | |
9617 | ||
7e84e1c7 | 9618 | io_sq_offload_start(ctx); |
7e84e1c7 SG |
9619 | return 0; |
9620 | } | |
9621 | ||
071698e1 JA |
9622 | static bool io_register_op_must_quiesce(int op) |
9623 | { | |
9624 | switch (op) { | |
9625 | case IORING_UNREGISTER_FILES: | |
9626 | case IORING_REGISTER_FILES_UPDATE: | |
9627 | case IORING_REGISTER_PROBE: | |
9628 | case IORING_REGISTER_PERSONALITY: | |
9629 | case IORING_UNREGISTER_PERSONALITY: | |
9630 | return false; | |
9631 | default: | |
9632 | return true; | |
9633 | } | |
9634 | } | |
9635 | ||
edafccee JA |
9636 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9637 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9638 | __releases(ctx->uring_lock) |
9639 | __acquires(ctx->uring_lock) | |
edafccee JA |
9640 | { |
9641 | int ret; | |
9642 | ||
35fa71a0 JA |
9643 | /* |
9644 | * We're inside the ring mutex, if the ref is already dying, then | |
9645 | * someone else killed the ctx or is already going through | |
9646 | * io_uring_register(). | |
9647 | */ | |
9648 | if (percpu_ref_is_dying(&ctx->refs)) | |
9649 | return -ENXIO; | |
9650 | ||
071698e1 | 9651 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9652 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9653 | |
05f3fb3c JA |
9654 | /* |
9655 | * Drop uring mutex before waiting for references to exit. If | |
9656 | * another thread is currently inside io_uring_enter() it might | |
9657 | * need to grab the uring_lock to make progress. If we hold it | |
9658 | * here across the drain wait, then we can deadlock. It's safe | |
9659 | * to drop the mutex here, since no new references will come in | |
9660 | * after we've killed the percpu ref. | |
9661 | */ | |
9662 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9663 | do { |
9664 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9665 | if (!ret) | |
9666 | break; | |
ed6930c9 JA |
9667 | ret = io_run_task_work_sig(); |
9668 | if (ret < 0) | |
9669 | break; | |
af9c1a44 JA |
9670 | } while (1); |
9671 | ||
05f3fb3c | 9672 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9673 | |
c150368b JA |
9674 | if (ret) { |
9675 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9676 | goto out_quiesce; |
9677 | } | |
9678 | } | |
9679 | ||
9680 | if (ctx->restricted) { | |
9681 | if (opcode >= IORING_REGISTER_LAST) { | |
9682 | ret = -EINVAL; | |
9683 | goto out; | |
9684 | } | |
9685 | ||
9686 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9687 | ret = -EACCES; | |
c150368b JA |
9688 | goto out; |
9689 | } | |
05f3fb3c | 9690 | } |
edafccee JA |
9691 | |
9692 | switch (opcode) { | |
9693 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9694 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9695 | break; |
9696 | case IORING_UNREGISTER_BUFFERS: | |
9697 | ret = -EINVAL; | |
9698 | if (arg || nr_args) | |
9699 | break; | |
0a96bbe4 | 9700 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9701 | break; |
6b06314c JA |
9702 | case IORING_REGISTER_FILES: |
9703 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9704 | break; | |
9705 | case IORING_UNREGISTER_FILES: | |
9706 | ret = -EINVAL; | |
9707 | if (arg || nr_args) | |
9708 | break; | |
9709 | ret = io_sqe_files_unregister(ctx); | |
9710 | break; | |
c3a31e60 JA |
9711 | case IORING_REGISTER_FILES_UPDATE: |
9712 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9713 | break; | |
9b402849 | 9714 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9715 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9716 | ret = -EINVAL; |
9717 | if (nr_args != 1) | |
9718 | break; | |
9719 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9720 | if (ret) |
9721 | break; | |
9722 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9723 | ctx->eventfd_async = 1; | |
9724 | else | |
9725 | ctx->eventfd_async = 0; | |
9b402849 JA |
9726 | break; |
9727 | case IORING_UNREGISTER_EVENTFD: | |
9728 | ret = -EINVAL; | |
9729 | if (arg || nr_args) | |
9730 | break; | |
9731 | ret = io_eventfd_unregister(ctx); | |
9732 | break; | |
66f4af93 JA |
9733 | case IORING_REGISTER_PROBE: |
9734 | ret = -EINVAL; | |
9735 | if (!arg || nr_args > 256) | |
9736 | break; | |
9737 | ret = io_probe(ctx, arg, nr_args); | |
9738 | break; | |
071698e1 JA |
9739 | case IORING_REGISTER_PERSONALITY: |
9740 | ret = -EINVAL; | |
9741 | if (arg || nr_args) | |
9742 | break; | |
9743 | ret = io_register_personality(ctx); | |
9744 | break; | |
9745 | case IORING_UNREGISTER_PERSONALITY: | |
9746 | ret = -EINVAL; | |
9747 | if (arg) | |
9748 | break; | |
9749 | ret = io_unregister_personality(ctx, nr_args); | |
9750 | break; | |
7e84e1c7 SG |
9751 | case IORING_REGISTER_ENABLE_RINGS: |
9752 | ret = -EINVAL; | |
9753 | if (arg || nr_args) | |
9754 | break; | |
9755 | ret = io_register_enable_rings(ctx); | |
9756 | break; | |
21b55dbc SG |
9757 | case IORING_REGISTER_RESTRICTIONS: |
9758 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9759 | break; | |
edafccee JA |
9760 | default: |
9761 | ret = -EINVAL; | |
9762 | break; | |
9763 | } | |
9764 | ||
21b55dbc | 9765 | out: |
071698e1 | 9766 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9767 | /* bring the ctx back to life */ |
05f3fb3c | 9768 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9769 | out_quiesce: |
0f158b4c | 9770 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9771 | } |
edafccee JA |
9772 | return ret; |
9773 | } | |
9774 | ||
9775 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9776 | void __user *, arg, unsigned int, nr_args) | |
9777 | { | |
9778 | struct io_ring_ctx *ctx; | |
9779 | long ret = -EBADF; | |
9780 | struct fd f; | |
9781 | ||
9782 | f = fdget(fd); | |
9783 | if (!f.file) | |
9784 | return -EBADF; | |
9785 | ||
9786 | ret = -EOPNOTSUPP; | |
9787 | if (f.file->f_op != &io_uring_fops) | |
9788 | goto out_fput; | |
9789 | ||
9790 | ctx = f.file->private_data; | |
9791 | ||
b6c23dd5 PB |
9792 | io_run_task_work(); |
9793 | ||
edafccee JA |
9794 | mutex_lock(&ctx->uring_lock); |
9795 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9796 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9797 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9798 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9799 | out_fput: |
9800 | fdput(f); | |
9801 | return ret; | |
9802 | } | |
9803 | ||
2b188cc1 JA |
9804 | static int __init io_uring_init(void) |
9805 | { | |
d7f62e82 SM |
9806 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9807 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9808 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9809 | } while (0) | |
9810 | ||
9811 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9812 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9813 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9814 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9815 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9816 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9817 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9818 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9819 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9820 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9821 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9822 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9823 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9824 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9825 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9826 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9827 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9828 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9829 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9830 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9831 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9832 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9833 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9834 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9835 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9836 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9837 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9838 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9839 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9840 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9841 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9842 | |
d3656344 | 9843 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9844 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9845 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9846 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9847 | return 0; |
9848 | }; | |
9849 | __initcall(io_uring_init); |