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; | |
05962f95 | 261 | struct rw_semaphore rw_lock; |
69fb2131 JA |
262 | |
263 | /* ctx's that are using this sqd */ | |
264 | struct list_head ctx_list; | |
69fb2131 | 265 | |
534ca6d6 JA |
266 | struct task_struct *thread; |
267 | struct wait_queue_head wait; | |
08369246 XW |
268 | |
269 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
270 | int sq_cpu; |
271 | pid_t task_pid; | |
5c2469e0 | 272 | pid_t task_tgid; |
37d1e2e3 JA |
273 | |
274 | unsigned long state; | |
37d1e2e3 | 275 | struct completion exited; |
534ca6d6 JA |
276 | }; |
277 | ||
258b29a9 | 278 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 279 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 280 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 281 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
282 | |
283 | struct io_comp_state { | |
6dd0be1e | 284 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 285 | unsigned int nr; |
c7dae4ba JA |
286 | unsigned int locked_free_nr; |
287 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 288 | struct list_head free_list; |
c7dae4ba JA |
289 | /* IRQ completion list, under ->completion_lock */ |
290 | struct list_head locked_free_list; | |
258b29a9 PB |
291 | }; |
292 | ||
a1ab7b35 PB |
293 | struct io_submit_link { |
294 | struct io_kiocb *head; | |
295 | struct io_kiocb *last; | |
296 | }; | |
297 | ||
258b29a9 PB |
298 | struct io_submit_state { |
299 | struct blk_plug plug; | |
a1ab7b35 | 300 | struct io_submit_link link; |
258b29a9 PB |
301 | |
302 | /* | |
303 | * io_kiocb alloc cache | |
304 | */ | |
bf019da7 | 305 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
306 | unsigned int free_reqs; |
307 | ||
308 | bool plug_started; | |
309 | ||
310 | /* | |
311 | * Batch completion logic | |
312 | */ | |
313 | struct io_comp_state comp; | |
314 | ||
315 | /* | |
316 | * File reference cache | |
317 | */ | |
318 | struct file *file; | |
319 | unsigned int fd; | |
320 | unsigned int file_refs; | |
321 | unsigned int ios_left; | |
322 | }; | |
323 | ||
2b188cc1 JA |
324 | struct io_ring_ctx { |
325 | struct { | |
326 | struct percpu_ref refs; | |
327 | } ____cacheline_aligned_in_smp; | |
328 | ||
329 | struct { | |
330 | unsigned int flags; | |
e1d85334 | 331 | unsigned int compat: 1; |
e1d85334 RD |
332 | unsigned int cq_overflow_flushed: 1; |
333 | unsigned int drain_next: 1; | |
334 | unsigned int eventfd_async: 1; | |
21b55dbc | 335 | unsigned int restricted: 1; |
2b188cc1 | 336 | |
75b28aff HV |
337 | /* |
338 | * Ring buffer of indices into array of io_uring_sqe, which is | |
339 | * mmapped by the application using the IORING_OFF_SQES offset. | |
340 | * | |
341 | * This indirection could e.g. be used to assign fixed | |
342 | * io_uring_sqe entries to operations and only submit them to | |
343 | * the queue when needed. | |
344 | * | |
345 | * The kernel modifies neither the indices array nor the entries | |
346 | * array. | |
347 | */ | |
348 | u32 *sq_array; | |
2b188cc1 JA |
349 | unsigned cached_sq_head; |
350 | unsigned sq_entries; | |
351 | unsigned sq_mask; | |
6c271ce2 | 352 | unsigned sq_thread_idle; |
498ccd9e | 353 | unsigned cached_sq_dropped; |
2c3bac6d | 354 | unsigned cached_cq_overflow; |
ad3eb2c8 | 355 | unsigned long sq_check_overflow; |
de0617e4 | 356 | |
e941894e JA |
357 | /* hashed buffered write serialization */ |
358 | struct io_wq_hash *hash_map; | |
359 | ||
de0617e4 | 360 | struct list_head defer_list; |
5262f567 | 361 | struct list_head timeout_list; |
1d7bb1d5 | 362 | struct list_head cq_overflow_list; |
fcb323cc | 363 | |
ad3eb2c8 | 364 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
365 | } ____cacheline_aligned_in_smp; |
366 | ||
3c1a2ead JA |
367 | struct { |
368 | struct mutex uring_lock; | |
369 | wait_queue_head_t wait; | |
370 | } ____cacheline_aligned_in_smp; | |
371 | ||
372 | struct io_submit_state submit_state; | |
373 | ||
206aefde JA |
374 | struct io_rings *rings; |
375 | ||
2aede0e4 JA |
376 | /* Only used for accounting purposes */ |
377 | struct mm_struct *mm_account; | |
378 | ||
7c30f36a | 379 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
380 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
381 | ||
90554200 | 382 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 383 | struct list_head sqd_list; |
75b28aff | 384 | |
6b06314c JA |
385 | /* |
386 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
387 | * readers must ensure that ->refs is alive as long as the file* is | |
388 | * used. Only updated through io_uring_register(2). | |
389 | */ | |
269bbe5f | 390 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
391 | unsigned nr_user_files; |
392 | ||
edafccee JA |
393 | /* if used, fixed mapped user buffers */ |
394 | unsigned nr_user_bufs; | |
395 | struct io_mapped_ubuf *user_bufs; | |
396 | ||
2b188cc1 JA |
397 | struct user_struct *user; |
398 | ||
0f158b4c | 399 | struct completion ref_comp; |
206aefde JA |
400 | |
401 | #if defined(CONFIG_UNIX) | |
402 | struct socket *ring_sock; | |
403 | #endif | |
404 | ||
9e15c3a0 | 405 | struct xarray io_buffers; |
5a2e745d | 406 | |
61cf9370 MWO |
407 | struct xarray personalities; |
408 | u32 pers_next; | |
071698e1 | 409 | |
206aefde JA |
410 | struct { |
411 | unsigned cached_cq_tail; | |
412 | unsigned cq_entries; | |
413 | unsigned cq_mask; | |
414 | atomic_t cq_timeouts; | |
f010505b | 415 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 416 | unsigned long cq_check_overflow; |
206aefde JA |
417 | struct wait_queue_head cq_wait; |
418 | struct fasync_struct *cq_fasync; | |
419 | struct eventfd_ctx *cq_ev_fd; | |
420 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 421 | |
2b188cc1 JA |
422 | struct { |
423 | spinlock_t completion_lock; | |
e94f141b | 424 | |
def596e9 | 425 | /* |
540e32a0 | 426 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
427 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
428 | * For SQPOLL, only the single threaded io_sq_thread() will | |
429 | * manipulate the list, hence no extra locking is needed there. | |
430 | */ | |
540e32a0 | 431 | struct list_head iopoll_list; |
78076bb6 JA |
432 | struct hlist_head *cancel_hash; |
433 | unsigned cancel_hash_bits; | |
e94f141b | 434 | bool poll_multi_file; |
31b51510 | 435 | |
fcb323cc JA |
436 | spinlock_t inflight_lock; |
437 | struct list_head inflight_list; | |
2b188cc1 | 438 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 439 | |
269bbe5f BM |
440 | struct delayed_work rsrc_put_work; |
441 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
442 | struct list_head rsrc_ref_list; |
443 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 444 | |
21b55dbc | 445 | struct io_restriction restrictions; |
3c1a2ead | 446 | |
7c25c0d1 JA |
447 | /* exit task_work */ |
448 | struct callback_head *exit_task_work; | |
449 | ||
e941894e JA |
450 | struct wait_queue_head hash_wait; |
451 | ||
3c1a2ead JA |
452 | /* Keep this last, we don't need it for the fast path */ |
453 | struct work_struct exit_work; | |
13bf43f5 | 454 | struct list_head tctx_list; |
2b188cc1 JA |
455 | }; |
456 | ||
09bb8394 JA |
457 | /* |
458 | * First field must be the file pointer in all the | |
459 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
460 | */ | |
221c5eb2 JA |
461 | struct io_poll_iocb { |
462 | struct file *file; | |
018043be | 463 | struct wait_queue_head *head; |
221c5eb2 | 464 | __poll_t events; |
8c838788 | 465 | bool done; |
221c5eb2 | 466 | bool canceled; |
392edb45 | 467 | struct wait_queue_entry wait; |
221c5eb2 JA |
468 | }; |
469 | ||
018043be PB |
470 | struct io_poll_remove { |
471 | struct file *file; | |
472 | u64 addr; | |
473 | }; | |
474 | ||
b5dba59e JA |
475 | struct io_close { |
476 | struct file *file; | |
b5dba59e JA |
477 | int fd; |
478 | }; | |
479 | ||
ad8a48ac JA |
480 | struct io_timeout_data { |
481 | struct io_kiocb *req; | |
482 | struct hrtimer timer; | |
483 | struct timespec64 ts; | |
484 | enum hrtimer_mode mode; | |
485 | }; | |
486 | ||
8ed8d3c3 JA |
487 | struct io_accept { |
488 | struct file *file; | |
489 | struct sockaddr __user *addr; | |
490 | int __user *addr_len; | |
491 | int flags; | |
09952e3e | 492 | unsigned long nofile; |
8ed8d3c3 JA |
493 | }; |
494 | ||
495 | struct io_sync { | |
496 | struct file *file; | |
497 | loff_t len; | |
498 | loff_t off; | |
499 | int flags; | |
d63d1b5e | 500 | int mode; |
8ed8d3c3 JA |
501 | }; |
502 | ||
fbf23849 JA |
503 | struct io_cancel { |
504 | struct file *file; | |
505 | u64 addr; | |
506 | }; | |
507 | ||
b29472ee JA |
508 | struct io_timeout { |
509 | struct file *file; | |
bfe68a22 PB |
510 | u32 off; |
511 | u32 target_seq; | |
135fcde8 | 512 | struct list_head list; |
90cd7e42 PB |
513 | /* head of the link, used by linked timeouts only */ |
514 | struct io_kiocb *head; | |
b29472ee JA |
515 | }; |
516 | ||
0bdf7a2d PB |
517 | struct io_timeout_rem { |
518 | struct file *file; | |
519 | u64 addr; | |
9c8e11b3 PB |
520 | |
521 | /* timeout update */ | |
522 | struct timespec64 ts; | |
523 | u32 flags; | |
0bdf7a2d PB |
524 | }; |
525 | ||
9adbd45d JA |
526 | struct io_rw { |
527 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
528 | struct kiocb kiocb; | |
529 | u64 addr; | |
530 | u64 len; | |
531 | }; | |
532 | ||
3fbb51c1 JA |
533 | struct io_connect { |
534 | struct file *file; | |
535 | struct sockaddr __user *addr; | |
536 | int addr_len; | |
537 | }; | |
538 | ||
e47293fd JA |
539 | struct io_sr_msg { |
540 | struct file *file; | |
fddaface | 541 | union { |
270a5940 | 542 | struct user_msghdr __user *umsg; |
fddaface JA |
543 | void __user *buf; |
544 | }; | |
e47293fd | 545 | int msg_flags; |
bcda7baa | 546 | int bgid; |
fddaface | 547 | size_t len; |
bcda7baa | 548 | struct io_buffer *kbuf; |
e47293fd JA |
549 | }; |
550 | ||
15b71abe JA |
551 | struct io_open { |
552 | struct file *file; | |
553 | int dfd; | |
15b71abe | 554 | struct filename *filename; |
c12cedf2 | 555 | struct open_how how; |
4022e7af | 556 | unsigned long nofile; |
15b71abe JA |
557 | }; |
558 | ||
269bbe5f | 559 | struct io_rsrc_update { |
05f3fb3c JA |
560 | struct file *file; |
561 | u64 arg; | |
562 | u32 nr_args; | |
563 | u32 offset; | |
564 | }; | |
565 | ||
4840e418 JA |
566 | struct io_fadvise { |
567 | struct file *file; | |
568 | u64 offset; | |
569 | u32 len; | |
570 | u32 advice; | |
571 | }; | |
572 | ||
c1ca757b JA |
573 | struct io_madvise { |
574 | struct file *file; | |
575 | u64 addr; | |
576 | u32 len; | |
577 | u32 advice; | |
578 | }; | |
579 | ||
3e4827b0 JA |
580 | struct io_epoll { |
581 | struct file *file; | |
582 | int epfd; | |
583 | int op; | |
584 | int fd; | |
585 | struct epoll_event event; | |
e47293fd JA |
586 | }; |
587 | ||
7d67af2c PB |
588 | struct io_splice { |
589 | struct file *file_out; | |
590 | struct file *file_in; | |
591 | loff_t off_out; | |
592 | loff_t off_in; | |
593 | u64 len; | |
594 | unsigned int flags; | |
595 | }; | |
596 | ||
ddf0322d JA |
597 | struct io_provide_buf { |
598 | struct file *file; | |
599 | __u64 addr; | |
600 | __s32 len; | |
601 | __u32 bgid; | |
602 | __u16 nbufs; | |
603 | __u16 bid; | |
604 | }; | |
605 | ||
1d9e1288 BM |
606 | struct io_statx { |
607 | struct file *file; | |
608 | int dfd; | |
609 | unsigned int mask; | |
610 | unsigned int flags; | |
e62753e4 | 611 | const char __user *filename; |
1d9e1288 BM |
612 | struct statx __user *buffer; |
613 | }; | |
614 | ||
36f4fa68 JA |
615 | struct io_shutdown { |
616 | struct file *file; | |
617 | int how; | |
618 | }; | |
619 | ||
80a261fd JA |
620 | struct io_rename { |
621 | struct file *file; | |
622 | int old_dfd; | |
623 | int new_dfd; | |
624 | struct filename *oldpath; | |
625 | struct filename *newpath; | |
626 | int flags; | |
627 | }; | |
628 | ||
14a1143b JA |
629 | struct io_unlink { |
630 | struct file *file; | |
631 | int dfd; | |
632 | int flags; | |
633 | struct filename *filename; | |
634 | }; | |
635 | ||
3ca405eb PB |
636 | struct io_completion { |
637 | struct file *file; | |
638 | struct list_head list; | |
0f7e466b | 639 | int cflags; |
3ca405eb PB |
640 | }; |
641 | ||
f499a021 JA |
642 | struct io_async_connect { |
643 | struct sockaddr_storage address; | |
644 | }; | |
645 | ||
03b1230c JA |
646 | struct io_async_msghdr { |
647 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
648 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
649 | struct iovec *free_iov; | |
03b1230c JA |
650 | struct sockaddr __user *uaddr; |
651 | struct msghdr msg; | |
b537916c | 652 | struct sockaddr_storage addr; |
03b1230c JA |
653 | }; |
654 | ||
f67676d1 JA |
655 | struct io_async_rw { |
656 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
657 | const struct iovec *free_iovec; |
658 | struct iov_iter iter; | |
227c0c96 | 659 | size_t bytes_done; |
bcf5a063 | 660 | struct wait_page_queue wpq; |
f67676d1 JA |
661 | }; |
662 | ||
6b47ee6e PB |
663 | enum { |
664 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
665 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
666 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
667 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
668 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 669 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 670 | |
6b47ee6e PB |
671 | REQ_F_FAIL_LINK_BIT, |
672 | REQ_F_INFLIGHT_BIT, | |
673 | REQ_F_CUR_POS_BIT, | |
674 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 675 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 676 | REQ_F_ISREG_BIT, |
99bc4c38 | 677 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 678 | REQ_F_POLLED_BIT, |
bcda7baa | 679 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 680 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 681 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 682 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
683 | |
684 | /* not a real bit, just to check we're not overflowing the space */ | |
685 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
686 | }; |
687 | ||
688 | enum { | |
689 | /* ctx owns file */ | |
690 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
691 | /* drain existing IO first */ | |
692 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
693 | /* linked sqes */ | |
694 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
695 | /* doesn't sever on completion < 0 */ | |
696 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
697 | /* IOSQE_ASYNC */ | |
698 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
699 | /* IOSQE_BUFFER_SELECT */ |
700 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 701 | |
6b47ee6e PB |
702 | /* fail rest of links */ |
703 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 704 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
705 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
706 | /* read/write uses file position */ | |
707 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
708 | /* must not punt to workers */ | |
709 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 710 | /* has or had linked timeout */ |
6b47ee6e | 711 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
712 | /* regular file */ |
713 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
714 | /* needs cleanup */ |
715 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
716 | /* already went through poll handler */ |
717 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
718 | /* buffer already selected */ |
719 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
720 | /* doesn't need file table for this request */ |
721 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
722 | /* linked timeout is active, i.e. prepared by link's head */ |
723 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
724 | /* completion is deferred through io_comp_state */ |
725 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
726 | }; |
727 | ||
728 | struct async_poll { | |
729 | struct io_poll_iocb poll; | |
807abcb0 | 730 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
731 | }; |
732 | ||
7cbf1722 JA |
733 | struct io_task_work { |
734 | struct io_wq_work_node node; | |
735 | task_work_func_t func; | |
736 | }; | |
737 | ||
09bb8394 JA |
738 | /* |
739 | * NOTE! Each of the iocb union members has the file pointer | |
740 | * as the first entry in their struct definition. So you can | |
741 | * access the file pointer through any of the sub-structs, | |
742 | * or directly as just 'ki_filp' in this struct. | |
743 | */ | |
2b188cc1 | 744 | struct io_kiocb { |
221c5eb2 | 745 | union { |
09bb8394 | 746 | struct file *file; |
9adbd45d | 747 | struct io_rw rw; |
221c5eb2 | 748 | struct io_poll_iocb poll; |
018043be | 749 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
750 | struct io_accept accept; |
751 | struct io_sync sync; | |
fbf23849 | 752 | struct io_cancel cancel; |
b29472ee | 753 | struct io_timeout timeout; |
0bdf7a2d | 754 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 755 | struct io_connect connect; |
e47293fd | 756 | struct io_sr_msg sr_msg; |
15b71abe | 757 | struct io_open open; |
b5dba59e | 758 | struct io_close close; |
269bbe5f | 759 | struct io_rsrc_update rsrc_update; |
4840e418 | 760 | struct io_fadvise fadvise; |
c1ca757b | 761 | struct io_madvise madvise; |
3e4827b0 | 762 | struct io_epoll epoll; |
7d67af2c | 763 | struct io_splice splice; |
ddf0322d | 764 | struct io_provide_buf pbuf; |
1d9e1288 | 765 | struct io_statx statx; |
36f4fa68 | 766 | struct io_shutdown shutdown; |
80a261fd | 767 | struct io_rename rename; |
14a1143b | 768 | struct io_unlink unlink; |
3ca405eb PB |
769 | /* use only after cleaning per-op data, see io_clean_op() */ |
770 | struct io_completion compl; | |
221c5eb2 | 771 | }; |
2b188cc1 | 772 | |
e8c2bc1f JA |
773 | /* opcode allocated if it needs to store data for async defer */ |
774 | void *async_data; | |
d625c6ee | 775 | u8 opcode; |
65a6543d XW |
776 | /* polled IO has completed */ |
777 | u8 iopoll_completed; | |
2b188cc1 | 778 | |
4f4eeba8 | 779 | u16 buf_index; |
9cf7c104 | 780 | u32 result; |
4f4eeba8 | 781 | |
010e8e6b PB |
782 | struct io_ring_ctx *ctx; |
783 | unsigned int flags; | |
784 | refcount_t refs; | |
785 | struct task_struct *task; | |
786 | u64 user_data; | |
d7718a9d | 787 | |
f2f87370 | 788 | struct io_kiocb *link; |
269bbe5f | 789 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 790 | |
d21ffe7e PB |
791 | /* |
792 | * 1. used with ctx->iopoll_list with reads/writes | |
793 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
794 | */ | |
010e8e6b | 795 | struct list_head inflight_entry; |
7cbf1722 JA |
796 | union { |
797 | struct io_task_work io_task_work; | |
798 | struct callback_head task_work; | |
799 | }; | |
010e8e6b PB |
800 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
801 | struct hlist_node hash_node; | |
802 | struct async_poll *apoll; | |
803 | struct io_wq_work work; | |
2b188cc1 | 804 | }; |
05589553 | 805 | |
13bf43f5 PB |
806 | struct io_tctx_node { |
807 | struct list_head ctx_node; | |
808 | struct task_struct *task; | |
13bf43f5 PB |
809 | struct io_ring_ctx *ctx; |
810 | }; | |
811 | ||
27dc8338 PB |
812 | struct io_defer_entry { |
813 | struct list_head list; | |
814 | struct io_kiocb *req; | |
9cf7c104 | 815 | u32 seq; |
2b188cc1 JA |
816 | }; |
817 | ||
d3656344 | 818 | struct io_op_def { |
d3656344 JA |
819 | /* needs req->file assigned */ |
820 | unsigned needs_file : 1; | |
d3656344 JA |
821 | /* hash wq insertion if file is a regular file */ |
822 | unsigned hash_reg_file : 1; | |
823 | /* unbound wq insertion if file is a non-regular file */ | |
824 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
825 | /* opcode is not supported by this kernel */ |
826 | unsigned not_supported : 1; | |
8a72758c JA |
827 | /* set if opcode supports polled "wait" */ |
828 | unsigned pollin : 1; | |
829 | unsigned pollout : 1; | |
bcda7baa JA |
830 | /* op supports buffer selection */ |
831 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
832 | /* must always have async data allocated */ |
833 | unsigned needs_async_data : 1; | |
27926b68 JA |
834 | /* should block plug */ |
835 | unsigned plug : 1; | |
e8c2bc1f JA |
836 | /* size of async data needed, if any */ |
837 | unsigned short async_size; | |
d3656344 JA |
838 | }; |
839 | ||
0918682b | 840 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
841 | [IORING_OP_NOP] = {}, |
842 | [IORING_OP_READV] = { | |
d3656344 JA |
843 | .needs_file = 1, |
844 | .unbound_nonreg_file = 1, | |
8a72758c | 845 | .pollin = 1, |
4d954c25 | 846 | .buffer_select = 1, |
e8c2bc1f | 847 | .needs_async_data = 1, |
27926b68 | 848 | .plug = 1, |
e8c2bc1f | 849 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 850 | }, |
0463b6c5 | 851 | [IORING_OP_WRITEV] = { |
d3656344 JA |
852 | .needs_file = 1, |
853 | .hash_reg_file = 1, | |
854 | .unbound_nonreg_file = 1, | |
8a72758c | 855 | .pollout = 1, |
e8c2bc1f | 856 | .needs_async_data = 1, |
27926b68 | 857 | .plug = 1, |
e8c2bc1f | 858 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 859 | }, |
0463b6c5 | 860 | [IORING_OP_FSYNC] = { |
d3656344 JA |
861 | .needs_file = 1, |
862 | }, | |
0463b6c5 | 863 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
864 | .needs_file = 1, |
865 | .unbound_nonreg_file = 1, | |
8a72758c | 866 | .pollin = 1, |
27926b68 | 867 | .plug = 1, |
e8c2bc1f | 868 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 869 | }, |
0463b6c5 | 870 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
871 | .needs_file = 1, |
872 | .hash_reg_file = 1, | |
873 | .unbound_nonreg_file = 1, | |
8a72758c | 874 | .pollout = 1, |
27926b68 | 875 | .plug = 1, |
e8c2bc1f | 876 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | .unbound_nonreg_file = 1, | |
881 | }, | |
0463b6c5 PB |
882 | [IORING_OP_POLL_REMOVE] = {}, |
883 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
884 | .needs_file = 1, |
885 | }, | |
0463b6c5 | 886 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
887 | .needs_file = 1, |
888 | .unbound_nonreg_file = 1, | |
8a72758c | 889 | .pollout = 1, |
e8c2bc1f JA |
890 | .needs_async_data = 1, |
891 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 892 | }, |
0463b6c5 | 893 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
894 | .needs_file = 1, |
895 | .unbound_nonreg_file = 1, | |
8a72758c | 896 | .pollin = 1, |
52de1fe1 | 897 | .buffer_select = 1, |
e8c2bc1f JA |
898 | .needs_async_data = 1, |
899 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 900 | }, |
0463b6c5 | 901 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
902 | .needs_async_data = 1, |
903 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 904 | }, |
9c8e11b3 PB |
905 | [IORING_OP_TIMEOUT_REMOVE] = { |
906 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 907 | }, |
0463b6c5 | 908 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
909 | .needs_file = 1, |
910 | .unbound_nonreg_file = 1, | |
8a72758c | 911 | .pollin = 1, |
d3656344 | 912 | }, |
0463b6c5 PB |
913 | [IORING_OP_ASYNC_CANCEL] = {}, |
914 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
915 | .needs_async_data = 1, |
916 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 917 | }, |
0463b6c5 | 918 | [IORING_OP_CONNECT] = { |
d3656344 JA |
919 | .needs_file = 1, |
920 | .unbound_nonreg_file = 1, | |
8a72758c | 921 | .pollout = 1, |
e8c2bc1f JA |
922 | .needs_async_data = 1, |
923 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 924 | }, |
0463b6c5 | 925 | [IORING_OP_FALLOCATE] = { |
d3656344 | 926 | .needs_file = 1, |
d3656344 | 927 | }, |
44526bed JA |
928 | [IORING_OP_OPENAT] = {}, |
929 | [IORING_OP_CLOSE] = {}, | |
930 | [IORING_OP_FILES_UPDATE] = {}, | |
931 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 932 | [IORING_OP_READ] = { |
3a6820f2 JA |
933 | .needs_file = 1, |
934 | .unbound_nonreg_file = 1, | |
8a72758c | 935 | .pollin = 1, |
bcda7baa | 936 | .buffer_select = 1, |
27926b68 | 937 | .plug = 1, |
e8c2bc1f | 938 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 939 | }, |
0463b6c5 | 940 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
941 | .needs_file = 1, |
942 | .unbound_nonreg_file = 1, | |
8a72758c | 943 | .pollout = 1, |
27926b68 | 944 | .plug = 1, |
e8c2bc1f | 945 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 946 | }, |
0463b6c5 | 947 | [IORING_OP_FADVISE] = { |
4840e418 | 948 | .needs_file = 1, |
c1ca757b | 949 | }, |
44526bed | 950 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 951 | [IORING_OP_SEND] = { |
fddaface JA |
952 | .needs_file = 1, |
953 | .unbound_nonreg_file = 1, | |
8a72758c | 954 | .pollout = 1, |
fddaface | 955 | }, |
0463b6c5 | 956 | [IORING_OP_RECV] = { |
fddaface JA |
957 | .needs_file = 1, |
958 | .unbound_nonreg_file = 1, | |
8a72758c | 959 | .pollin = 1, |
bcda7baa | 960 | .buffer_select = 1, |
fddaface | 961 | }, |
0463b6c5 | 962 | [IORING_OP_OPENAT2] = { |
cebdb986 | 963 | }, |
3e4827b0 JA |
964 | [IORING_OP_EPOLL_CTL] = { |
965 | .unbound_nonreg_file = 1, | |
3e4827b0 | 966 | }, |
7d67af2c PB |
967 | [IORING_OP_SPLICE] = { |
968 | .needs_file = 1, | |
969 | .hash_reg_file = 1, | |
970 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
971 | }, |
972 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 973 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
974 | [IORING_OP_TEE] = { |
975 | .needs_file = 1, | |
976 | .hash_reg_file = 1, | |
977 | .unbound_nonreg_file = 1, | |
978 | }, | |
36f4fa68 JA |
979 | [IORING_OP_SHUTDOWN] = { |
980 | .needs_file = 1, | |
981 | }, | |
44526bed JA |
982 | [IORING_OP_RENAMEAT] = {}, |
983 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
984 | }; |
985 | ||
7a612350 | 986 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 987 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
988 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
989 | struct task_struct *task, | |
990 | struct files_struct *files); | |
37d1e2e3 | 991 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 992 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 993 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 994 | struct io_ring_ctx *ctx); |
f2303b1f | 995 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 996 | |
23faba36 | 997 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 998 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 999 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1000 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1001 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1002 | static void io_dismantle_req(struct io_kiocb *req); |
1003 | static void io_put_task(struct task_struct *task, int nr); | |
1004 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1005 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1006 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1007 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1008 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1009 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1010 | unsigned nr_args); |
3ca405eb | 1011 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1012 | static struct file *io_file_get(struct io_submit_state *state, |
1013 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1014 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1015 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1016 | |
847595de PB |
1017 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1018 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1019 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1020 | const struct iovec *fast_iov, | |
227c0c96 | 1021 | struct iov_iter *iter, bool force); |
907d1df3 | 1022 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1023 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1024 | struct io_ring_ctx *ctx); | |
de0617e4 | 1025 | |
2b188cc1 JA |
1026 | static struct kmem_cache *req_cachep; |
1027 | ||
0918682b | 1028 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1029 | |
1030 | struct sock *io_uring_get_socket(struct file *file) | |
1031 | { | |
1032 | #if defined(CONFIG_UNIX) | |
1033 | if (file->f_op == &io_uring_fops) { | |
1034 | struct io_ring_ctx *ctx = file->private_data; | |
1035 | ||
1036 | return ctx->ring_sock->sk; | |
1037 | } | |
1038 | #endif | |
1039 | return NULL; | |
1040 | } | |
1041 | EXPORT_SYMBOL(io_uring_get_socket); | |
1042 | ||
f2f87370 PB |
1043 | #define io_for_each_link(pos, head) \ |
1044 | for (pos = (head); pos; pos = pos->link) | |
1045 | ||
3ca405eb PB |
1046 | static inline void io_clean_op(struct io_kiocb *req) |
1047 | { | |
9d5c8190 | 1048 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1049 | __io_clean_op(req); |
1050 | } | |
1051 | ||
36f72fe2 PB |
1052 | static inline void io_set_resource_node(struct io_kiocb *req) |
1053 | { | |
1054 | struct io_ring_ctx *ctx = req->ctx; | |
1055 | ||
269bbe5f BM |
1056 | if (!req->fixed_rsrc_refs) { |
1057 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1058 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1059 | } |
1060 | } | |
1061 | ||
08d23634 PB |
1062 | static bool io_match_task(struct io_kiocb *head, |
1063 | struct task_struct *task, | |
1064 | struct files_struct *files) | |
1065 | { | |
1066 | struct io_kiocb *req; | |
1067 | ||
84965ff8 JA |
1068 | if (task && head->task != task) { |
1069 | /* in terms of cancelation, always match if req task is dead */ | |
1070 | if (head->task->flags & PF_EXITING) | |
1071 | return true; | |
08d23634 | 1072 | return false; |
84965ff8 | 1073 | } |
08d23634 PB |
1074 | if (!files) |
1075 | return true; | |
1076 | ||
1077 | io_for_each_link(req, head) { | |
b05a1bcd | 1078 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1079 | return true; |
4379bf8b | 1080 | if (req->task->files == files) |
08d23634 PB |
1081 | return true; |
1082 | } | |
1083 | return false; | |
1084 | } | |
1085 | ||
c40f6379 JA |
1086 | static inline void req_set_fail_links(struct io_kiocb *req) |
1087 | { | |
1088 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1089 | req->flags |= REQ_F_FAIL_LINK; | |
1090 | } | |
4a38aed2 | 1091 | |
2b188cc1 JA |
1092 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1093 | { | |
1094 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1095 | ||
0f158b4c | 1096 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1097 | } |
1098 | ||
8eb7e2d0 PB |
1099 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1100 | { | |
1101 | return !req->timeout.off; | |
1102 | } | |
1103 | ||
2b188cc1 JA |
1104 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1105 | { | |
1106 | struct io_ring_ctx *ctx; | |
78076bb6 | 1107 | int hash_bits; |
2b188cc1 JA |
1108 | |
1109 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1110 | if (!ctx) | |
1111 | return NULL; | |
1112 | ||
78076bb6 JA |
1113 | /* |
1114 | * Use 5 bits less than the max cq entries, that should give us around | |
1115 | * 32 entries per hash list if totally full and uniformly spread. | |
1116 | */ | |
1117 | hash_bits = ilog2(p->cq_entries); | |
1118 | hash_bits -= 5; | |
1119 | if (hash_bits <= 0) | |
1120 | hash_bits = 1; | |
1121 | ctx->cancel_hash_bits = hash_bits; | |
1122 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1123 | GFP_KERNEL); | |
1124 | if (!ctx->cancel_hash) | |
1125 | goto err; | |
1126 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1127 | ||
21482896 | 1128 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1129 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1130 | goto err; | |
2b188cc1 JA |
1131 | |
1132 | ctx->flags = p->flags; | |
90554200 | 1133 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1134 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1135 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1136 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1137 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1138 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1139 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1140 | mutex_init(&ctx->uring_lock); |
1141 | init_waitqueue_head(&ctx->wait); | |
1142 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1143 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1144 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1145 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1146 | spin_lock_init(&ctx->inflight_lock); |
1147 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1148 | spin_lock_init(&ctx->rsrc_ref_lock); |
1149 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1150 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1151 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1152 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1153 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1154 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1155 | return ctx; |
206aefde | 1156 | err: |
78076bb6 | 1157 | kfree(ctx->cancel_hash); |
206aefde JA |
1158 | kfree(ctx); |
1159 | return NULL; | |
2b188cc1 JA |
1160 | } |
1161 | ||
9cf7c104 | 1162 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1163 | { |
2bc9930e JA |
1164 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1165 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1166 | |
9cf7c104 | 1167 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1168 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1169 | } |
de0617e4 | 1170 | |
9d858b21 | 1171 | return false; |
de0617e4 JA |
1172 | } |
1173 | ||
ce3d5aae PB |
1174 | static void io_req_track_inflight(struct io_kiocb *req) |
1175 | { | |
1176 | struct io_ring_ctx *ctx = req->ctx; | |
1177 | ||
1178 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1179 | req->flags |= REQ_F_INFLIGHT; |
1180 | ||
1181 | spin_lock_irq(&ctx->inflight_lock); | |
1182 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1183 | spin_unlock_irq(&ctx->inflight_lock); | |
1184 | } | |
1185 | } | |
1186 | ||
1e6fa521 JA |
1187 | static void io_prep_async_work(struct io_kiocb *req) |
1188 | { | |
1189 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1190 | struct io_ring_ctx *ctx = req->ctx; |
1191 | ||
003e8dcc JA |
1192 | if (!req->work.creds) |
1193 | req->work.creds = get_current_cred(); | |
1194 | ||
feaadc4f PB |
1195 | if (req->flags & REQ_F_FORCE_ASYNC) |
1196 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1197 | ||
1e6fa521 JA |
1198 | if (req->flags & REQ_F_ISREG) { |
1199 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1200 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1201 | } else { | |
1202 | if (def->unbound_nonreg_file) | |
1203 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1204 | } | |
561fb04a | 1205 | } |
cccf0ee8 | 1206 | |
cbdcb435 | 1207 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1208 | { |
cbdcb435 | 1209 | struct io_kiocb *cur; |
54a91f3b | 1210 | |
f2f87370 PB |
1211 | io_for_each_link(cur, req) |
1212 | io_prep_async_work(cur); | |
561fb04a JA |
1213 | } |
1214 | ||
ebf93667 | 1215 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1216 | { |
a197f664 | 1217 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1218 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1219 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1220 | |
3bfe6106 JA |
1221 | BUG_ON(!tctx); |
1222 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1223 | |
8766dd51 PB |
1224 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1225 | &req->work, req->flags); | |
cbdcb435 PB |
1226 | /* init ->work of the whole link before punting */ |
1227 | io_prep_async_link(req); | |
ebf93667 | 1228 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1229 | if (link) |
1230 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1231 | } |
1232 | ||
5262f567 JA |
1233 | static void io_kill_timeout(struct io_kiocb *req) |
1234 | { | |
e8c2bc1f | 1235 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1236 | int ret; |
1237 | ||
e8c2bc1f | 1238 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1239 | if (ret != -1) { |
01cec8c1 PB |
1240 | atomic_set(&req->ctx->cq_timeouts, |
1241 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1242 | list_del_init(&req->timeout.list); |
78e19bbe | 1243 | io_cqring_fill_event(req, 0); |
216578e5 | 1244 | io_put_req_deferred(req, 1); |
5262f567 JA |
1245 | } |
1246 | } | |
1247 | ||
76e1b642 JA |
1248 | /* |
1249 | * Returns true if we found and killed one or more timeouts | |
1250 | */ | |
6b81928d PB |
1251 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1252 | struct files_struct *files) | |
5262f567 JA |
1253 | { |
1254 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1255 | int canceled = 0; |
5262f567 JA |
1256 | |
1257 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1258 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1259 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1260 | io_kill_timeout(req); |
76e1b642 JA |
1261 | canceled++; |
1262 | } | |
f3606e3a | 1263 | } |
5262f567 | 1264 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1265 | return canceled != 0; |
5262f567 JA |
1266 | } |
1267 | ||
04518945 | 1268 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1269 | { |
04518945 | 1270 | do { |
27dc8338 PB |
1271 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1272 | struct io_defer_entry, list); | |
de0617e4 | 1273 | |
9cf7c104 | 1274 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1275 | break; |
27dc8338 | 1276 | list_del_init(&de->list); |
907d1df3 | 1277 | io_req_task_queue(de->req); |
27dc8338 | 1278 | kfree(de); |
04518945 PB |
1279 | } while (!list_empty(&ctx->defer_list)); |
1280 | } | |
1281 | ||
360428f8 | 1282 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1283 | { |
f010505b MDG |
1284 | u32 seq; |
1285 | ||
1286 | if (list_empty(&ctx->timeout_list)) | |
1287 | return; | |
1288 | ||
1289 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1290 | ||
1291 | do { | |
1292 | u32 events_needed, events_got; | |
360428f8 | 1293 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1294 | struct io_kiocb, timeout.list); |
de0617e4 | 1295 | |
8eb7e2d0 | 1296 | if (io_is_timeout_noseq(req)) |
360428f8 | 1297 | break; |
f010505b MDG |
1298 | |
1299 | /* | |
1300 | * Since seq can easily wrap around over time, subtract | |
1301 | * the last seq at which timeouts were flushed before comparing. | |
1302 | * Assuming not more than 2^31-1 events have happened since, | |
1303 | * these subtractions won't have wrapped, so we can check if | |
1304 | * target is in [last_seq, current_seq] by comparing the two. | |
1305 | */ | |
1306 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1307 | events_got = seq - ctx->cq_last_tm_flush; | |
1308 | if (events_got < events_needed) | |
360428f8 | 1309 | break; |
bfe68a22 | 1310 | |
135fcde8 | 1311 | list_del_init(&req->timeout.list); |
5262f567 | 1312 | io_kill_timeout(req); |
f010505b MDG |
1313 | } while (!list_empty(&ctx->timeout_list)); |
1314 | ||
1315 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1316 | } |
5262f567 | 1317 | |
360428f8 PB |
1318 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1319 | { | |
1320 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1321 | |
1322 | /* order cqe stores with ring update */ | |
1323 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1324 | |
04518945 PB |
1325 | if (unlikely(!list_empty(&ctx->defer_list))) |
1326 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1327 | } |
1328 | ||
90554200 JA |
1329 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1330 | { | |
1331 | struct io_rings *r = ctx->rings; | |
1332 | ||
1333 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1334 | } | |
1335 | ||
888aae2e PB |
1336 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1337 | { | |
1338 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1339 | } | |
1340 | ||
2b188cc1 JA |
1341 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1342 | { | |
75b28aff | 1343 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1344 | unsigned tail; |
1345 | ||
115e12e5 SB |
1346 | /* |
1347 | * writes to the cq entry need to come after reading head; the | |
1348 | * control dependency is enough as we're using WRITE_ONCE to | |
1349 | * fill the cq entry | |
1350 | */ | |
888aae2e | 1351 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1352 | return NULL; |
1353 | ||
888aae2e | 1354 | tail = ctx->cached_cq_tail++; |
75b28aff | 1355 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1356 | } |
1357 | ||
f2842ab5 JA |
1358 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1359 | { | |
f0b493e6 JA |
1360 | if (!ctx->cq_ev_fd) |
1361 | return false; | |
7e55a19c SG |
1362 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1363 | return false; | |
f2842ab5 JA |
1364 | if (!ctx->eventfd_async) |
1365 | return true; | |
b41e9852 | 1366 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1367 | } |
1368 | ||
b41e9852 | 1369 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1370 | { |
b1445e59 PB |
1371 | /* see waitqueue_active() comment */ |
1372 | smp_mb(); | |
1373 | ||
1d7bb1d5 JA |
1374 | if (waitqueue_active(&ctx->wait)) |
1375 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1376 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1377 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1378 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1379 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1380 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1381 | wake_up_interruptible(&ctx->cq_wait); |
1382 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1383 | } | |
1d7bb1d5 JA |
1384 | } |
1385 | ||
80c18e4a PB |
1386 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1387 | { | |
b1445e59 PB |
1388 | /* see waitqueue_active() comment */ |
1389 | smp_mb(); | |
1390 | ||
80c18e4a PB |
1391 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1392 | if (waitqueue_active(&ctx->wait)) | |
1393 | wake_up(&ctx->wait); | |
1394 | } | |
1395 | if (io_should_trigger_evfd(ctx)) | |
1396 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1397 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1398 | wake_up_interruptible(&ctx->cq_wait); |
1399 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1400 | } | |
80c18e4a PB |
1401 | } |
1402 | ||
c4a2ed72 | 1403 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1404 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1405 | struct task_struct *tsk, | |
1406 | struct files_struct *files) | |
1d7bb1d5 JA |
1407 | { |
1408 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1409 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1410 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1411 | unsigned long flags; |
b18032bb | 1412 | bool all_flushed, posted; |
1d7bb1d5 JA |
1413 | LIST_HEAD(list); |
1414 | ||
e23de15f PB |
1415 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1416 | return false; | |
1d7bb1d5 | 1417 | |
b18032bb | 1418 | posted = false; |
1d7bb1d5 | 1419 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1420 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1421 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1422 | continue; |
1423 | ||
1d7bb1d5 JA |
1424 | cqe = io_get_cqring(ctx); |
1425 | if (!cqe && !force) | |
1426 | break; | |
1427 | ||
40d8ddd4 | 1428 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1429 | if (cqe) { |
1430 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1431 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1432 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1433 | } else { |
2c3bac6d | 1434 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1435 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1436 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1437 | } |
b18032bb | 1438 | posted = true; |
1d7bb1d5 JA |
1439 | } |
1440 | ||
09e88404 PB |
1441 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1442 | if (all_flushed) { | |
1443 | clear_bit(0, &ctx->sq_check_overflow); | |
1444 | clear_bit(0, &ctx->cq_check_overflow); | |
1445 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1446 | } | |
46930143 | 1447 | |
b18032bb JA |
1448 | if (posted) |
1449 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1450 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1451 | if (posted) |
1452 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1453 | |
1454 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1455 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1456 | list_del(&req->compl.list); | |
ec9c02ad | 1457 | io_put_req(req); |
1d7bb1d5 | 1458 | } |
c4a2ed72 | 1459 | |
09e88404 | 1460 | return all_flushed; |
1d7bb1d5 JA |
1461 | } |
1462 | ||
ca0a2651 | 1463 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1464 | struct task_struct *tsk, |
1465 | struct files_struct *files) | |
1466 | { | |
ca0a2651 JA |
1467 | bool ret = true; |
1468 | ||
6c503150 PB |
1469 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1470 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1471 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1472 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1473 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1474 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1475 | mutex_unlock(&ctx->uring_lock); | |
1476 | } | |
ca0a2651 JA |
1477 | |
1478 | return ret; | |
6c503150 PB |
1479 | } |
1480 | ||
bcda7baa | 1481 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1482 | { |
78e19bbe | 1483 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1484 | struct io_uring_cqe *cqe; |
1485 | ||
78e19bbe | 1486 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1487 | |
2b188cc1 JA |
1488 | /* |
1489 | * If we can't get a cq entry, userspace overflowed the | |
1490 | * submission (by quite a lot). Increment the overflow count in | |
1491 | * the ring. | |
1492 | */ | |
1493 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1494 | if (likely(cqe)) { |
78e19bbe | 1495 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1496 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1497 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1498 | } else if (ctx->cq_overflow_flushed || |
1499 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1500 | /* |
1501 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1502 | * then we cannot store the request for later flushing, we need | |
1503 | * to drop it on the floor. | |
1504 | */ | |
2c3bac6d PB |
1505 | ctx->cached_cq_overflow++; |
1506 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1507 | } else { |
ad3eb2c8 JA |
1508 | if (list_empty(&ctx->cq_overflow_list)) { |
1509 | set_bit(0, &ctx->sq_check_overflow); | |
1510 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1511 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1512 | } |
40d8ddd4 | 1513 | io_clean_op(req); |
1d7bb1d5 | 1514 | req->result = res; |
0f7e466b | 1515 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1516 | refcount_inc(&req->refs); |
1517 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1518 | } |
1519 | } | |
1520 | ||
bcda7baa JA |
1521 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1522 | { | |
1523 | __io_cqring_fill_event(req, res, 0); | |
1524 | } | |
1525 | ||
7a612350 PB |
1526 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1527 | unsigned int cflags) | |
2b188cc1 | 1528 | { |
78e19bbe | 1529 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1530 | unsigned long flags; |
1531 | ||
1532 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1533 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1534 | /* |
1535 | * If we're the last reference to this request, add to our locked | |
1536 | * free_list cache. | |
1537 | */ | |
1538 | if (refcount_dec_and_test(&req->refs)) { | |
1539 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1540 | ||
7a612350 PB |
1541 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1542 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1543 | io_disarm_next(req); | |
1544 | if (req->link) { | |
1545 | io_req_task_queue(req->link); | |
1546 | req->link = NULL; | |
1547 | } | |
1548 | } | |
c7dae4ba JA |
1549 | io_dismantle_req(req); |
1550 | io_put_task(req->task, 1); | |
1551 | list_add(&req->compl.list, &cs->locked_free_list); | |
1552 | cs->locked_free_nr++; | |
1553 | } else | |
1554 | req = NULL; | |
7a612350 | 1555 | io_commit_cqring(ctx); |
2b188cc1 | 1556 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8c838788 | 1557 | io_cqring_ev_posted(ctx); |
7a612350 PB |
1558 | |
1559 | if (req) | |
c7dae4ba | 1560 | percpu_ref_put(&ctx->refs); |
229a7b63 JA |
1561 | } |
1562 | ||
a38d68db | 1563 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1564 | unsigned int cflags) |
229a7b63 | 1565 | { |
a38d68db PB |
1566 | io_clean_op(req); |
1567 | req->result = res; | |
1568 | req->compl.cflags = cflags; | |
e342c807 | 1569 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1570 | } |
1571 | ||
889fca73 PB |
1572 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1573 | long res, unsigned cflags) | |
bcda7baa | 1574 | { |
889fca73 PB |
1575 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1576 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1577 | else |
c7dae4ba | 1578 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1579 | } |
1580 | ||
a38d68db | 1581 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1582 | { |
889fca73 | 1583 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1584 | } |
1585 | ||
c7dae4ba | 1586 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1587 | { |
c7dae4ba JA |
1588 | struct io_submit_state *state = &ctx->submit_state; |
1589 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1590 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1591 | |
c7dae4ba JA |
1592 | /* |
1593 | * If we have more than a batch's worth of requests in our IRQ side | |
1594 | * locked cache, grab the lock and move them over to our submission | |
1595 | * side cache. | |
1596 | */ | |
1597 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1598 | spin_lock_irq(&ctx->completion_lock); | |
1599 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1600 | cs->locked_free_nr = 0; | |
1601 | spin_unlock_irq(&ctx->completion_lock); | |
1602 | } | |
0ddf92e8 | 1603 | |
c7dae4ba JA |
1604 | while (!list_empty(&cs->free_list)) { |
1605 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1606 | compl.list); |
1607 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1608 | state->reqs[state->free_reqs++] = req; |
1609 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1610 | break; | |
1b4c351f JA |
1611 | } |
1612 | ||
e5d1bc0a | 1613 | return req != NULL; |
0ddf92e8 JA |
1614 | } |
1615 | ||
e5d1bc0a | 1616 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1617 | { |
e5d1bc0a PB |
1618 | struct io_submit_state *state = &ctx->submit_state; |
1619 | ||
1620 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1621 | ||
f6b6c7d6 | 1622 | if (!state->free_reqs) { |
291b2821 | 1623 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1624 | int ret; |
1625 | ||
c7dae4ba | 1626 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1627 | goto got_req; |
1628 | ||
bf019da7 PB |
1629 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1630 | state->reqs); | |
fd6fab2c JA |
1631 | |
1632 | /* | |
1633 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1634 | * retry single alloc to be on the safe side. | |
1635 | */ | |
1636 | if (unlikely(ret <= 0)) { | |
1637 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1638 | if (!state->reqs[0]) | |
3893f39f | 1639 | return NULL; |
fd6fab2c JA |
1640 | ret = 1; |
1641 | } | |
291b2821 | 1642 | state->free_reqs = ret; |
2b188cc1 | 1643 | } |
e5d1bc0a | 1644 | got_req: |
291b2821 PB |
1645 | state->free_reqs--; |
1646 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1647 | } |
1648 | ||
8da11c19 PB |
1649 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1650 | bool fixed) | |
1651 | { | |
36f72fe2 | 1652 | if (!fixed) |
8da11c19 PB |
1653 | fput(file); |
1654 | } | |
1655 | ||
4edf20f9 | 1656 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1657 | { |
3ca405eb | 1658 | io_clean_op(req); |
929a3af9 | 1659 | |
e8c2bc1f JA |
1660 | if (req->async_data) |
1661 | kfree(req->async_data); | |
8da11c19 PB |
1662 | if (req->file) |
1663 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1664 | if (req->fixed_rsrc_refs) |
1665 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1666 | if (req->work.creds) { |
1667 | put_cred(req->work.creds); | |
1668 | req->work.creds = NULL; | |
1669 | } | |
f85c310a PB |
1670 | |
1671 | if (req->flags & REQ_F_INFLIGHT) { | |
1672 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1673 | unsigned long flags; |
1674 | ||
1675 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1676 | list_del(&req->inflight_entry); | |
1677 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1678 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1679 | } |
e65ef56d JA |
1680 | } |
1681 | ||
b23fcf47 | 1682 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1683 | static inline void io_put_task(struct task_struct *task, int nr) |
1684 | { | |
1685 | struct io_uring_task *tctx = task->io_uring; | |
1686 | ||
1687 | percpu_counter_sub(&tctx->inflight, nr); | |
1688 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1689 | wake_up(&tctx->wait); | |
1690 | put_task_struct_many(task, nr); | |
1691 | } | |
1692 | ||
216578e5 | 1693 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1694 | { |
51a4cc11 | 1695 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1696 | |
216578e5 | 1697 | io_dismantle_req(req); |
7c660731 | 1698 | io_put_task(req->task, 1); |
c6ca97b3 | 1699 | |
3893f39f | 1700 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1701 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1702 | } |
1703 | ||
f2f87370 PB |
1704 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1705 | { | |
1706 | struct io_kiocb *nxt = req->link; | |
1707 | ||
1708 | req->link = nxt->link; | |
1709 | nxt->link = NULL; | |
1710 | } | |
1711 | ||
33cc89a9 PB |
1712 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1713 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1714 | { |
33cc89a9 | 1715 | struct io_kiocb *link = req->link; |
c9abd7ad | 1716 | bool cancelled = false; |
f2f87370 | 1717 | |
900fad45 PB |
1718 | /* |
1719 | * Can happen if a linked timeout fired and link had been like | |
1720 | * req -> link t-out -> link t-out [-> ...] | |
1721 | */ | |
c9abd7ad PB |
1722 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1723 | struct io_timeout_data *io = link->async_data; | |
1724 | int ret; | |
7c86ffee | 1725 | |
f2f87370 | 1726 | io_remove_next_linked(req); |
90cd7e42 | 1727 | link->timeout.head = NULL; |
c9abd7ad PB |
1728 | ret = hrtimer_try_to_cancel(&io->timer); |
1729 | if (ret != -1) { | |
1730 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1731 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1732 | cancelled = true; |
1733 | } | |
1734 | } | |
7c86ffee | 1735 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1736 | return cancelled; |
7c86ffee PB |
1737 | } |
1738 | ||
d148ca4b | 1739 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1740 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1741 | { |
33cc89a9 | 1742 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1743 | |
f2f87370 | 1744 | req->link = NULL; |
f2f87370 PB |
1745 | while (link) { |
1746 | nxt = link->link; | |
1747 | link->link = NULL; | |
2665abfd | 1748 | |
f2f87370 | 1749 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1750 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1751 | io_put_req_deferred(link, 2); |
f2f87370 | 1752 | link = nxt; |
9e645e11 | 1753 | } |
33cc89a9 | 1754 | } |
9e645e11 | 1755 | |
33cc89a9 PB |
1756 | static bool io_disarm_next(struct io_kiocb *req) |
1757 | __must_hold(&req->ctx->completion_lock) | |
1758 | { | |
1759 | bool posted = false; | |
1760 | ||
1761 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1762 | posted = io_kill_linked_timeout(req); | |
1763 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1764 | posted |= (req->link != NULL); | |
1765 | io_fail_links(req); | |
1766 | } | |
1767 | return posted; | |
9e645e11 JA |
1768 | } |
1769 | ||
3fa5e0f3 | 1770 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1771 | { |
33cc89a9 | 1772 | struct io_kiocb *nxt; |
944e58bf | 1773 | |
9e645e11 JA |
1774 | /* |
1775 | * If LINK is set, we have dependent requests in this chain. If we | |
1776 | * didn't fail this request, queue the first one up, moving any other | |
1777 | * dependencies to the next request. In case of failure, fail the rest | |
1778 | * of the chain. | |
1779 | */ | |
33cc89a9 PB |
1780 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1781 | struct io_ring_ctx *ctx = req->ctx; | |
1782 | unsigned long flags; | |
1783 | bool posted; | |
1784 | ||
1785 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1786 | posted = io_disarm_next(req); | |
1787 | if (posted) | |
1788 | io_commit_cqring(req->ctx); | |
1789 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1790 | if (posted) | |
1791 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1792 | } |
33cc89a9 PB |
1793 | nxt = req->link; |
1794 | req->link = NULL; | |
1795 | return nxt; | |
4d7dd462 | 1796 | } |
9e645e11 | 1797 | |
f2f87370 | 1798 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1799 | { |
cdbff982 | 1800 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1801 | return NULL; |
1802 | return __io_req_find_next(req); | |
1803 | } | |
1804 | ||
2c32395d PB |
1805 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1806 | { | |
1807 | if (!ctx) | |
1808 | return; | |
1809 | if (ctx->submit_state.comp.nr) { | |
1810 | mutex_lock(&ctx->uring_lock); | |
1811 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1812 | mutex_unlock(&ctx->uring_lock); | |
1813 | } | |
1814 | percpu_ref_put(&ctx->refs); | |
1815 | } | |
1816 | ||
7cbf1722 | 1817 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1818 | { |
65453d1e | 1819 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1820 | struct io_wq_work_list list; |
1821 | struct io_wq_work_node *node; | |
c2c4c83c | 1822 | |
7cbf1722 JA |
1823 | if (wq_list_empty(&tctx->task_list)) |
1824 | return false; | |
6200b0ae | 1825 | |
0b81e80c | 1826 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1827 | list = tctx->task_list; |
1828 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1829 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1830 | |
7cbf1722 JA |
1831 | node = list.first; |
1832 | while (node) { | |
1833 | struct io_wq_work_node *next = node->next; | |
1834 | struct io_kiocb *req; | |
0ba9c9ed | 1835 | |
7cbf1722 | 1836 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1837 | if (req->ctx != ctx) { |
1838 | ctx_flush_and_put(ctx); | |
1839 | ctx = req->ctx; | |
1840 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1841 | } |
65453d1e | 1842 | |
2c32395d PB |
1843 | req->task_work.func(&req->task_work); |
1844 | node = next; | |
7cbf1722 JA |
1845 | } |
1846 | ||
2c32395d | 1847 | ctx_flush_and_put(ctx); |
7cbf1722 | 1848 | return list.first != NULL; |
c2c4c83c JA |
1849 | } |
1850 | ||
7cbf1722 | 1851 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1852 | { |
7cbf1722 | 1853 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1854 | |
1d5f360d JA |
1855 | clear_bit(0, &tctx->task_state); |
1856 | ||
7cbf1722 JA |
1857 | while (__tctx_task_work(tctx)) |
1858 | cond_resched(); | |
7cbf1722 JA |
1859 | } |
1860 | ||
1861 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1862 | enum task_work_notify_mode notify) | |
1863 | { | |
1864 | struct io_uring_task *tctx = tsk->io_uring; | |
1865 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1866 | unsigned long flags; |
7cbf1722 JA |
1867 | int ret; |
1868 | ||
1869 | WARN_ON_ONCE(!tctx); | |
1870 | ||
0b81e80c | 1871 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1872 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1873 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1874 | |
1875 | /* task_work already pending, we're done */ | |
1876 | if (test_bit(0, &tctx->task_state) || | |
1877 | test_and_set_bit(0, &tctx->task_state)) | |
1878 | return 0; | |
1879 | ||
1880 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1881 | return 0; | |
1882 | ||
1883 | /* | |
1884 | * Slow path - we failed, find and delete work. if the work is not | |
1885 | * in the list, it got run and we're fine. | |
1886 | */ | |
1887 | ret = 0; | |
0b81e80c | 1888 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1889 | wq_list_for_each(node, prev, &tctx->task_list) { |
1890 | if (&req->io_task_work.node == node) { | |
1891 | wq_list_del(&tctx->task_list, node, prev); | |
1892 | ret = 1; | |
1893 | break; | |
1894 | } | |
1895 | } | |
0b81e80c | 1896 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1897 | clear_bit(0, &tctx->task_state); |
1898 | return ret; | |
1899 | } | |
1900 | ||
355fb9e2 | 1901 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1902 | { |
1903 | struct task_struct *tsk = req->task; | |
1904 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1905 | enum task_work_notify_mode notify; |
1906 | int ret; | |
c2c4c83c | 1907 | |
6200b0ae JA |
1908 | if (tsk->flags & PF_EXITING) |
1909 | return -ESRCH; | |
1910 | ||
c2c4c83c | 1911 | /* |
0ba9c9ed JA |
1912 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1913 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1914 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1915 | * will do the job. | |
c2c4c83c | 1916 | */ |
91989c70 | 1917 | notify = TWA_NONE; |
355fb9e2 | 1918 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1919 | notify = TWA_SIGNAL; |
1920 | ||
7cbf1722 | 1921 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1922 | if (!ret) |
1923 | wake_up_process(tsk); | |
0ba9c9ed | 1924 | |
c2c4c83c JA |
1925 | return ret; |
1926 | } | |
1927 | ||
eab30c4d | 1928 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1929 | task_work_func_t cb) |
eab30c4d | 1930 | { |
7c25c0d1 JA |
1931 | struct io_ring_ctx *ctx = req->ctx; |
1932 | struct callback_head *head; | |
eab30c4d PB |
1933 | |
1934 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1935 | do { |
1936 | head = READ_ONCE(ctx->exit_task_work); | |
1937 | req->task_work.next = head; | |
1938 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1939 | } |
1940 | ||
c40f6379 JA |
1941 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1942 | { | |
1943 | struct io_ring_ctx *ctx = req->ctx; | |
1944 | ||
1945 | spin_lock_irq(&ctx->completion_lock); | |
1946 | io_cqring_fill_event(req, error); | |
1947 | io_commit_cqring(ctx); | |
1948 | spin_unlock_irq(&ctx->completion_lock); | |
1949 | ||
1950 | io_cqring_ev_posted(ctx); | |
1951 | req_set_fail_links(req); | |
1952 | io_double_put_req(req); | |
1953 | } | |
1954 | ||
1955 | static void io_req_task_cancel(struct callback_head *cb) | |
1956 | { | |
1957 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1958 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1959 | |
792bb6eb | 1960 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1961 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1962 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1963 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1964 | } |
1965 | ||
1966 | static void __io_req_task_submit(struct io_kiocb *req) | |
1967 | { | |
1968 | struct io_ring_ctx *ctx = req->ctx; | |
1969 | ||
04fc6c80 | 1970 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1971 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 1972 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1973 | __io_queue_sqe(req); |
81b6d05c | 1974 | else |
c40f6379 | 1975 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1976 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1977 | } |
1978 | ||
1979 | static void io_req_task_submit(struct callback_head *cb) | |
1980 | { | |
1981 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1982 | ||
1983 | __io_req_task_submit(req); | |
1984 | } | |
1985 | ||
1986 | static void io_req_task_queue(struct io_kiocb *req) | |
1987 | { | |
c40f6379 JA |
1988 | int ret; |
1989 | ||
7cbf1722 | 1990 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1991 | ret = io_req_task_work_add(req); |
c40f6379 | 1992 | if (unlikely(ret)) { |
a3df7698 | 1993 | req->result = -ECANCELED; |
04fc6c80 | 1994 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1995 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1996 | } |
c40f6379 JA |
1997 | } |
1998 | ||
a3df7698 PB |
1999 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2000 | { | |
2001 | percpu_ref_get(&req->ctx->refs); | |
2002 | req->result = ret; | |
2003 | req->task_work.func = io_req_task_cancel; | |
2004 | ||
2005 | if (unlikely(io_req_task_work_add(req))) | |
2006 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2007 | } | |
2008 | ||
f2f87370 | 2009 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2010 | { |
9b5f7bd9 | 2011 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2012 | |
2013 | if (nxt) | |
906a8c3f | 2014 | io_req_task_queue(nxt); |
c69f8dbe JL |
2015 | } |
2016 | ||
c3524383 | 2017 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2018 | { |
c3524383 PB |
2019 | io_queue_next(req); |
2020 | __io_free_req(req); | |
2021 | } | |
8766dd51 | 2022 | |
2d6500d4 | 2023 | struct req_batch { |
5af1d13e PB |
2024 | struct task_struct *task; |
2025 | int task_refs; | |
1b4c351f | 2026 | int ctx_refs; |
2d6500d4 PB |
2027 | }; |
2028 | ||
5af1d13e PB |
2029 | static inline void io_init_req_batch(struct req_batch *rb) |
2030 | { | |
5af1d13e | 2031 | rb->task_refs = 0; |
9ae72463 | 2032 | rb->ctx_refs = 0; |
5af1d13e PB |
2033 | rb->task = NULL; |
2034 | } | |
2035 | ||
2d6500d4 PB |
2036 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2037 | struct req_batch *rb) | |
2038 | { | |
6e833d53 | 2039 | if (rb->task) |
7c660731 | 2040 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2041 | if (rb->ctx_refs) |
2042 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2043 | } |
2044 | ||
6ff119a6 PB |
2045 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2046 | struct io_submit_state *state) | |
2d6500d4 | 2047 | { |
f2f87370 | 2048 | io_queue_next(req); |
2d6500d4 | 2049 | |
e3bc8e9d | 2050 | if (req->task != rb->task) { |
7c660731 PB |
2051 | if (rb->task) |
2052 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2053 | rb->task = req->task; |
2054 | rb->task_refs = 0; | |
5af1d13e | 2055 | } |
e3bc8e9d | 2056 | rb->task_refs++; |
9ae72463 | 2057 | rb->ctx_refs++; |
5af1d13e | 2058 | |
4edf20f9 | 2059 | io_dismantle_req(req); |
bd759045 | 2060 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2061 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2062 | else |
2063 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2064 | } |
2065 | ||
905c172f PB |
2066 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2067 | struct io_ring_ctx *ctx) | |
2068 | { | |
2069 | int i, nr = cs->nr; | |
2070 | struct io_kiocb *req; | |
2071 | struct req_batch rb; | |
2072 | ||
2073 | io_init_req_batch(&rb); | |
2074 | spin_lock_irq(&ctx->completion_lock); | |
2075 | for (i = 0; i < nr; i++) { | |
2076 | req = cs->reqs[i]; | |
2077 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2078 | } | |
2079 | io_commit_cqring(ctx); | |
2080 | spin_unlock_irq(&ctx->completion_lock); | |
2081 | ||
2082 | io_cqring_ev_posted(ctx); | |
2083 | for (i = 0; i < nr; i++) { | |
2084 | req = cs->reqs[i]; | |
2085 | ||
2086 | /* submission and completion refs */ | |
2087 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2088 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2089 | } |
2090 | ||
2091 | io_req_free_batch_finish(ctx, &rb); | |
2092 | cs->nr = 0; | |
7a743e22 PB |
2093 | } |
2094 | ||
ba816ad6 JA |
2095 | /* |
2096 | * Drop reference to request, return next in chain (if there is one) if this | |
2097 | * was the last reference to this request. | |
2098 | */ | |
9b5f7bd9 | 2099 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2100 | { |
9b5f7bd9 PB |
2101 | struct io_kiocb *nxt = NULL; |
2102 | ||
2a44f467 | 2103 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2104 | nxt = io_req_find_next(req); |
4d7dd462 | 2105 | __io_free_req(req); |
2a44f467 | 2106 | } |
9b5f7bd9 | 2107 | return nxt; |
2b188cc1 JA |
2108 | } |
2109 | ||
e65ef56d JA |
2110 | static void io_put_req(struct io_kiocb *req) |
2111 | { | |
2112 | if (refcount_dec_and_test(&req->refs)) | |
2113 | io_free_req(req); | |
2b188cc1 JA |
2114 | } |
2115 | ||
216578e5 PB |
2116 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2117 | { | |
2118 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2119 | ||
2120 | io_free_req(req); | |
2121 | } | |
2122 | ||
2123 | static void io_free_req_deferred(struct io_kiocb *req) | |
2124 | { | |
2125 | int ret; | |
2126 | ||
7cbf1722 | 2127 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2128 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2129 | if (unlikely(ret)) |
2130 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2131 | } |
2132 | ||
2133 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2134 | { | |
2135 | if (refcount_sub_and_test(refs, &req->refs)) | |
2136 | io_free_req_deferred(req); | |
2137 | } | |
2138 | ||
978db57e JA |
2139 | static void io_double_put_req(struct io_kiocb *req) |
2140 | { | |
2141 | /* drop both submit and complete references */ | |
2142 | if (refcount_sub_and_test(2, &req->refs)) | |
2143 | io_free_req(req); | |
2144 | } | |
2145 | ||
6c503150 | 2146 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2147 | { |
2148 | /* See comment at the top of this file */ | |
2149 | smp_rmb(); | |
e23de15f | 2150 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2151 | } |
2152 | ||
fb5ccc98 PB |
2153 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2154 | { | |
2155 | struct io_rings *rings = ctx->rings; | |
2156 | ||
2157 | /* make sure SQ entry isn't read before tail */ | |
2158 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2159 | } | |
2160 | ||
8ff069bf | 2161 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2162 | { |
8ff069bf | 2163 | unsigned int cflags; |
e94f141b | 2164 | |
bcda7baa JA |
2165 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2166 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2167 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2168 | kfree(kbuf); |
2169 | return cflags; | |
e94f141b JA |
2170 | } |
2171 | ||
8ff069bf | 2172 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2173 | { |
4d954c25 | 2174 | struct io_buffer *kbuf; |
bcda7baa | 2175 | |
4d954c25 | 2176 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2177 | return io_put_kbuf(req, kbuf); |
2178 | } | |
2179 | ||
4c6e277c JA |
2180 | static inline bool io_run_task_work(void) |
2181 | { | |
6200b0ae JA |
2182 | /* |
2183 | * Not safe to run on exiting task, and the task_work handling will | |
2184 | * not add work to such a task. | |
2185 | */ | |
2186 | if (unlikely(current->flags & PF_EXITING)) | |
2187 | return false; | |
4c6e277c JA |
2188 | if (current->task_works) { |
2189 | __set_current_state(TASK_RUNNING); | |
2190 | task_work_run(); | |
2191 | return true; | |
2192 | } | |
2193 | ||
2194 | return false; | |
bcda7baa JA |
2195 | } |
2196 | ||
def596e9 JA |
2197 | /* |
2198 | * Find and free completed poll iocbs | |
2199 | */ | |
2200 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2201 | struct list_head *done) | |
2202 | { | |
8237e045 | 2203 | struct req_batch rb; |
def596e9 | 2204 | struct io_kiocb *req; |
bbde017a XW |
2205 | |
2206 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2207 | smp_rmb(); | |
def596e9 | 2208 | |
5af1d13e | 2209 | io_init_req_batch(&rb); |
def596e9 | 2210 | while (!list_empty(done)) { |
bcda7baa JA |
2211 | int cflags = 0; |
2212 | ||
d21ffe7e | 2213 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2214 | list_del(&req->inflight_entry); |
2215 | ||
bbde017a XW |
2216 | if (READ_ONCE(req->result) == -EAGAIN) { |
2217 | req->iopoll_completed = 0; | |
23faba36 | 2218 | if (io_rw_reissue(req)) |
f161340d | 2219 | continue; |
bbde017a | 2220 | } |
def596e9 | 2221 | |
bcda7baa | 2222 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2223 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2224 | |
2225 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2226 | (*nr_events)++; |
2227 | ||
c3524383 | 2228 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2229 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2230 | } |
def596e9 | 2231 | |
09bb8394 | 2232 | io_commit_cqring(ctx); |
80c18e4a | 2233 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2234 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2235 | } |
2236 | ||
def596e9 JA |
2237 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2238 | long min) | |
2239 | { | |
2240 | struct io_kiocb *req, *tmp; | |
2241 | LIST_HEAD(done); | |
2242 | bool spin; | |
2243 | int ret; | |
2244 | ||
2245 | /* | |
2246 | * Only spin for completions if we don't have multiple devices hanging | |
2247 | * off our complete list, and we're under the requested amount. | |
2248 | */ | |
2249 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2250 | ||
2251 | ret = 0; | |
d21ffe7e | 2252 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2253 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2254 | |
2255 | /* | |
581f9810 BM |
2256 | * Move completed and retryable entries to our local lists. |
2257 | * If we find a request that requires polling, break out | |
2258 | * and complete those lists first, if we have entries there. | |
def596e9 | 2259 | */ |
65a6543d | 2260 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2261 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2262 | continue; |
2263 | } | |
2264 | if (!list_empty(&done)) | |
2265 | break; | |
2266 | ||
2267 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2268 | if (ret < 0) | |
2269 | break; | |
2270 | ||
3aadc23e PB |
2271 | /* iopoll may have completed current req */ |
2272 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2273 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2274 | |
def596e9 JA |
2275 | if (ret && spin) |
2276 | spin = false; | |
2277 | ret = 0; | |
2278 | } | |
2279 | ||
2280 | if (!list_empty(&done)) | |
2281 | io_iopoll_complete(ctx, nr_events, &done); | |
2282 | ||
2283 | return ret; | |
2284 | } | |
2285 | ||
2286 | /* | |
d195a66e | 2287 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2288 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2289 | * as a non-spinning completion check. | |
2290 | */ | |
2291 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2292 | long min) | |
2293 | { | |
540e32a0 | 2294 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2295 | int ret; |
2296 | ||
2297 | ret = io_do_iopoll(ctx, nr_events, min); | |
2298 | if (ret < 0) | |
2299 | return ret; | |
eba0a4dd | 2300 | if (*nr_events >= min) |
def596e9 JA |
2301 | return 0; |
2302 | } | |
2303 | ||
2304 | return 1; | |
2305 | } | |
2306 | ||
2307 | /* | |
2308 | * We can't just wait for polled events to come to us, we have to actively | |
2309 | * find and complete them. | |
2310 | */ | |
b2edc0a7 | 2311 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2312 | { |
2313 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2314 | return; | |
2315 | ||
2316 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2317 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2318 | unsigned int nr_events = 0; |
2319 | ||
b2edc0a7 | 2320 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2321 | |
b2edc0a7 PB |
2322 | /* let it sleep and repeat later if can't complete a request */ |
2323 | if (nr_events == 0) | |
2324 | break; | |
08f5439f JA |
2325 | /* |
2326 | * Ensure we allow local-to-the-cpu processing to take place, | |
2327 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2328 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2329 | */ |
3fcee5a6 PB |
2330 | if (need_resched()) { |
2331 | mutex_unlock(&ctx->uring_lock); | |
2332 | cond_resched(); | |
2333 | mutex_lock(&ctx->uring_lock); | |
2334 | } | |
def596e9 JA |
2335 | } |
2336 | mutex_unlock(&ctx->uring_lock); | |
2337 | } | |
2338 | ||
7668b92a | 2339 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2340 | { |
7668b92a | 2341 | unsigned int nr_events = 0; |
2b2ed975 | 2342 | int iters = 0, ret = 0; |
500f9fba | 2343 | |
c7849be9 XW |
2344 | /* |
2345 | * We disallow the app entering submit/complete with polling, but we | |
2346 | * still need to lock the ring to prevent racing with polled issue | |
2347 | * that got punted to a workqueue. | |
2348 | */ | |
2349 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2350 | do { |
a3a0e43f JA |
2351 | /* |
2352 | * Don't enter poll loop if we already have events pending. | |
2353 | * If we do, we can potentially be spinning for commands that | |
2354 | * already triggered a CQE (eg in error). | |
2355 | */ | |
6c503150 PB |
2356 | if (test_bit(0, &ctx->cq_check_overflow)) |
2357 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2358 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2359 | break; |
2360 | ||
500f9fba JA |
2361 | /* |
2362 | * If a submit got punted to a workqueue, we can have the | |
2363 | * application entering polling for a command before it gets | |
2364 | * issued. That app will hold the uring_lock for the duration | |
2365 | * of the poll right here, so we need to take a breather every | |
2366 | * now and then to ensure that the issue has a chance to add | |
2367 | * the poll to the issued list. Otherwise we can spin here | |
2368 | * forever, while the workqueue is stuck trying to acquire the | |
2369 | * very same mutex. | |
2370 | */ | |
2371 | if (!(++iters & 7)) { | |
2372 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2373 | io_run_task_work(); |
500f9fba JA |
2374 | mutex_lock(&ctx->uring_lock); |
2375 | } | |
2376 | ||
7668b92a | 2377 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2378 | if (ret <= 0) |
2379 | break; | |
2380 | ret = 0; | |
7668b92a | 2381 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2382 | |
500f9fba | 2383 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2384 | return ret; |
2385 | } | |
2386 | ||
491381ce | 2387 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2388 | { |
491381ce JA |
2389 | /* |
2390 | * Tell lockdep we inherited freeze protection from submission | |
2391 | * thread. | |
2392 | */ | |
2393 | if (req->flags & REQ_F_ISREG) { | |
2394 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2395 | |
491381ce | 2396 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2397 | } |
491381ce | 2398 | file_end_write(req->file); |
2b188cc1 JA |
2399 | } |
2400 | ||
b63534c4 | 2401 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2402 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2403 | { |
2404 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2405 | int rw, ret; |
b63534c4 | 2406 | struct iov_iter iter; |
b63534c4 | 2407 | |
dc2a6e9a PB |
2408 | /* already prepared */ |
2409 | if (req->async_data) | |
2410 | return true; | |
b63534c4 JA |
2411 | |
2412 | switch (req->opcode) { | |
2413 | case IORING_OP_READV: | |
2414 | case IORING_OP_READ_FIXED: | |
2415 | case IORING_OP_READ: | |
2416 | rw = READ; | |
2417 | break; | |
2418 | case IORING_OP_WRITEV: | |
2419 | case IORING_OP_WRITE_FIXED: | |
2420 | case IORING_OP_WRITE: | |
2421 | rw = WRITE; | |
2422 | break; | |
2423 | default: | |
2424 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2425 | req->opcode); | |
dc2a6e9a | 2426 | return false; |
b63534c4 JA |
2427 | } |
2428 | ||
dc2a6e9a PB |
2429 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2430 | if (ret < 0) | |
2431 | return false; | |
6bf985dc | 2432 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2433 | } |
b63534c4 | 2434 | |
3e6a0d3c | 2435 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2436 | { |
355afaeb | 2437 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2438 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2439 | |
355afaeb JA |
2440 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2441 | return false; | |
3e6a0d3c JA |
2442 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2443 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2444 | return false; |
7c977a58 JA |
2445 | /* |
2446 | * If ref is dying, we might be running poll reap from the exit work. | |
2447 | * Don't attempt to reissue from that path, just let it fail with | |
2448 | * -EAGAIN. | |
2449 | */ | |
3e6a0d3c JA |
2450 | if (percpu_ref_is_dying(&ctx->refs)) |
2451 | return false; | |
2452 | return true; | |
2453 | } | |
2454 | #endif | |
2455 | ||
2456 | static bool io_rw_reissue(struct io_kiocb *req) | |
2457 | { | |
2458 | #ifdef CONFIG_BLOCK | |
2459 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2460 | return false; |
b63534c4 | 2461 | |
55e6ac1e PB |
2462 | lockdep_assert_held(&req->ctx->uring_lock); |
2463 | ||
37d1e2e3 | 2464 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2465 | refcount_inc(&req->refs); |
2466 | io_queue_async_work(req); | |
b63534c4 | 2467 | return true; |
fdee946d | 2468 | } |
dc2a6e9a | 2469 | req_set_fail_links(req); |
b63534c4 JA |
2470 | #endif |
2471 | return false; | |
2472 | } | |
2473 | ||
a1d7c393 | 2474 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2475 | unsigned int issue_flags) |
a1d7c393 | 2476 | { |
2f8e45f1 PB |
2477 | int cflags = 0; |
2478 | ||
23faba36 PB |
2479 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2480 | return; | |
2f8e45f1 PB |
2481 | if (res != req->result) |
2482 | req_set_fail_links(req); | |
23faba36 | 2483 | |
2f8e45f1 PB |
2484 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2485 | kiocb_end_write(req); | |
2486 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2487 | cflags = io_put_rw_kbuf(req); | |
2488 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2489 | } |
2490 | ||
2491 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2492 | { | |
9adbd45d | 2493 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2494 | |
889fca73 | 2495 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2496 | } |
2497 | ||
def596e9 JA |
2498 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2499 | { | |
9adbd45d | 2500 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2501 | |
3e6a0d3c JA |
2502 | #ifdef CONFIG_BLOCK |
2503 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2504 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2505 | struct io_async_rw *rw = req->async_data; | |
2506 | ||
2507 | if (rw) | |
2508 | iov_iter_revert(&rw->iter, | |
2509 | req->result - iov_iter_count(&rw->iter)); | |
2510 | else if (!io_resubmit_prep(req)) | |
2511 | res = -EIO; | |
2512 | } | |
2513 | #endif | |
2514 | ||
491381ce JA |
2515 | if (kiocb->ki_flags & IOCB_WRITE) |
2516 | kiocb_end_write(req); | |
def596e9 | 2517 | |
2d7d6792 | 2518 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2519 | req_set_fail_links(req); |
bbde017a XW |
2520 | |
2521 | WRITE_ONCE(req->result, res); | |
2522 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2523 | smp_wmb(); |
2524 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2525 | } |
2526 | ||
2527 | /* | |
2528 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2529 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2530 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2531 | * accessing the kiocb cookie. | |
2532 | */ | |
2e9dbe90 | 2533 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2534 | { |
2535 | struct io_ring_ctx *ctx = req->ctx; | |
2536 | ||
2537 | /* | |
2538 | * Track whether we have multiple files in our lists. This will impact | |
2539 | * how we do polling eventually, not spinning if we're on potentially | |
2540 | * different devices. | |
2541 | */ | |
540e32a0 | 2542 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2543 | ctx->poll_multi_file = false; |
2544 | } else if (!ctx->poll_multi_file) { | |
2545 | struct io_kiocb *list_req; | |
2546 | ||
540e32a0 | 2547 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2548 | inflight_entry); |
9adbd45d | 2549 | if (list_req->file != req->file) |
def596e9 JA |
2550 | ctx->poll_multi_file = true; |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * For fast devices, IO may have already completed. If it has, add | |
2555 | * it to the front so we find it first. | |
2556 | */ | |
65a6543d | 2557 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2558 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2559 | else |
d21ffe7e | 2560 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2561 | |
2e9dbe90 XW |
2562 | /* |
2563 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2564 | * task context or in io worker task context. If current task context is | |
2565 | * sq thread, we don't need to check whether should wake up sq thread. | |
2566 | */ | |
2567 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2568 | wq_has_sleeper(&ctx->sq_data->wait)) |
2569 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2570 | } |
2571 | ||
9f13c35b PB |
2572 | static inline void io_state_file_put(struct io_submit_state *state) |
2573 | { | |
02b23a9a PB |
2574 | if (state->file_refs) { |
2575 | fput_many(state->file, state->file_refs); | |
2576 | state->file_refs = 0; | |
2577 | } | |
9a56a232 JA |
2578 | } |
2579 | ||
2580 | /* | |
2581 | * Get as many references to a file as we have IOs left in this submission, | |
2582 | * assuming most submissions are for one file, or at least that each file | |
2583 | * has more than one submission. | |
2584 | */ | |
8da11c19 | 2585 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2586 | { |
2587 | if (!state) | |
2588 | return fget(fd); | |
2589 | ||
6e1271e6 | 2590 | if (state->file_refs) { |
9a56a232 | 2591 | if (state->fd == fd) { |
6e1271e6 | 2592 | state->file_refs--; |
9a56a232 JA |
2593 | return state->file; |
2594 | } | |
02b23a9a | 2595 | io_state_file_put(state); |
9a56a232 JA |
2596 | } |
2597 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2598 | if (unlikely(!state->file)) |
9a56a232 JA |
2599 | return NULL; |
2600 | ||
2601 | state->fd = fd; | |
6e1271e6 | 2602 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2603 | return state->file; |
2604 | } | |
2605 | ||
4503b767 JA |
2606 | static bool io_bdev_nowait(struct block_device *bdev) |
2607 | { | |
9ba0d0c8 | 2608 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2609 | } |
2610 | ||
2b188cc1 JA |
2611 | /* |
2612 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2613 | * any file. For now, just ensure that anything potentially problematic is done | |
2614 | * inline. | |
2615 | */ | |
af197f50 | 2616 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2617 | { |
2618 | umode_t mode = file_inode(file)->i_mode; | |
2619 | ||
4503b767 | 2620 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2621 | if (IS_ENABLED(CONFIG_BLOCK) && |
2622 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2623 | return true; |
2624 | return false; | |
2625 | } | |
2626 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2627 | return true; |
4503b767 | 2628 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2629 | if (IS_ENABLED(CONFIG_BLOCK) && |
2630 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2631 | file->f_op != &io_uring_fops) |
2632 | return true; | |
2633 | return false; | |
2634 | } | |
2b188cc1 | 2635 | |
c5b85625 JA |
2636 | /* any ->read/write should understand O_NONBLOCK */ |
2637 | if (file->f_flags & O_NONBLOCK) | |
2638 | return true; | |
2639 | ||
af197f50 JA |
2640 | if (!(file->f_mode & FMODE_NOWAIT)) |
2641 | return false; | |
2642 | ||
2643 | if (rw == READ) | |
2644 | return file->f_op->read_iter != NULL; | |
2645 | ||
2646 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2647 | } |
2648 | ||
a88fc400 | 2649 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2650 | { |
def596e9 | 2651 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2652 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2653 | struct file *file = req->file; |
09bb8394 JA |
2654 | unsigned ioprio; |
2655 | int ret; | |
2b188cc1 | 2656 | |
75c668cd | 2657 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2658 | req->flags |= REQ_F_ISREG; |
2659 | ||
2b188cc1 | 2660 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2661 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2662 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2663 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2664 | } |
2b188cc1 | 2665 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2666 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2667 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2668 | if (unlikely(ret)) | |
2669 | return ret; | |
2b188cc1 | 2670 | |
75c668cd PB |
2671 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2672 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2673 | req->flags |= REQ_F_NOWAIT; | |
2674 | ||
2b188cc1 JA |
2675 | ioprio = READ_ONCE(sqe->ioprio); |
2676 | if (ioprio) { | |
2677 | ret = ioprio_check_cap(ioprio); | |
2678 | if (ret) | |
09bb8394 | 2679 | return ret; |
2b188cc1 JA |
2680 | |
2681 | kiocb->ki_ioprio = ioprio; | |
2682 | } else | |
2683 | kiocb->ki_ioprio = get_current_ioprio(); | |
2684 | ||
def596e9 | 2685 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2686 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2687 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2688 | return -EOPNOTSUPP; |
2b188cc1 | 2689 | |
def596e9 JA |
2690 | kiocb->ki_flags |= IOCB_HIPRI; |
2691 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2692 | req->iopoll_completed = 0; |
def596e9 | 2693 | } else { |
09bb8394 JA |
2694 | if (kiocb->ki_flags & IOCB_HIPRI) |
2695 | return -EINVAL; | |
def596e9 JA |
2696 | kiocb->ki_complete = io_complete_rw; |
2697 | } | |
9adbd45d | 2698 | |
3529d8c2 JA |
2699 | req->rw.addr = READ_ONCE(sqe->addr); |
2700 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2701 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2702 | return 0; |
2b188cc1 JA |
2703 | } |
2704 | ||
2705 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2706 | { | |
2707 | switch (ret) { | |
2708 | case -EIOCBQUEUED: | |
2709 | break; | |
2710 | case -ERESTARTSYS: | |
2711 | case -ERESTARTNOINTR: | |
2712 | case -ERESTARTNOHAND: | |
2713 | case -ERESTART_RESTARTBLOCK: | |
2714 | /* | |
2715 | * We can't just restart the syscall, since previously | |
2716 | * submitted sqes may already be in progress. Just fail this | |
2717 | * IO with EINTR. | |
2718 | */ | |
2719 | ret = -EINTR; | |
df561f66 | 2720 | fallthrough; |
2b188cc1 JA |
2721 | default: |
2722 | kiocb->ki_complete(kiocb, ret, 0); | |
2723 | } | |
2724 | } | |
2725 | ||
a1d7c393 | 2726 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2727 | unsigned int issue_flags) |
ba816ad6 | 2728 | { |
ba04291e | 2729 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2730 | struct io_async_rw *io = req->async_data; |
ba04291e | 2731 | |
227c0c96 | 2732 | /* add previously done IO, if any */ |
e8c2bc1f | 2733 | if (io && io->bytes_done > 0) { |
227c0c96 | 2734 | if (ret < 0) |
e8c2bc1f | 2735 | ret = io->bytes_done; |
227c0c96 | 2736 | else |
e8c2bc1f | 2737 | ret += io->bytes_done; |
227c0c96 JA |
2738 | } |
2739 | ||
ba04291e JA |
2740 | if (req->flags & REQ_F_CUR_POS) |
2741 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2742 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2743 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2744 | else |
2745 | io_rw_done(kiocb, ret); | |
2746 | } | |
2747 | ||
847595de | 2748 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2749 | { |
9adbd45d JA |
2750 | struct io_ring_ctx *ctx = req->ctx; |
2751 | size_t len = req->rw.len; | |
edafccee | 2752 | struct io_mapped_ubuf *imu; |
4be1c615 | 2753 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2754 | size_t offset; |
2755 | u64 buf_addr; | |
2756 | ||
edafccee JA |
2757 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2758 | return -EFAULT; | |
edafccee JA |
2759 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2760 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2761 | buf_addr = req->rw.addr; |
edafccee JA |
2762 | |
2763 | /* overflow */ | |
2764 | if (buf_addr + len < buf_addr) | |
2765 | return -EFAULT; | |
2766 | /* not inside the mapped region */ | |
2767 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2768 | return -EFAULT; | |
2769 | ||
2770 | /* | |
2771 | * May not be a start of buffer, set size appropriately | |
2772 | * and advance us to the beginning. | |
2773 | */ | |
2774 | offset = buf_addr - imu->ubuf; | |
2775 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2776 | |
2777 | if (offset) { | |
2778 | /* | |
2779 | * Don't use iov_iter_advance() here, as it's really slow for | |
2780 | * using the latter parts of a big fixed buffer - it iterates | |
2781 | * over each segment manually. We can cheat a bit here, because | |
2782 | * we know that: | |
2783 | * | |
2784 | * 1) it's a BVEC iter, we set it up | |
2785 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2786 | * first and last bvec | |
2787 | * | |
2788 | * So just find our index, and adjust the iterator afterwards. | |
2789 | * If the offset is within the first bvec (or the whole first | |
2790 | * bvec, just use iov_iter_advance(). This makes it easier | |
2791 | * since we can just skip the first segment, which may not | |
2792 | * be PAGE_SIZE aligned. | |
2793 | */ | |
2794 | const struct bio_vec *bvec = imu->bvec; | |
2795 | ||
2796 | if (offset <= bvec->bv_len) { | |
2797 | iov_iter_advance(iter, offset); | |
2798 | } else { | |
2799 | unsigned long seg_skip; | |
2800 | ||
2801 | /* skip first vec */ | |
2802 | offset -= bvec->bv_len; | |
2803 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2804 | ||
2805 | iter->bvec = bvec + seg_skip; | |
2806 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2807 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2808 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2809 | } |
2810 | } | |
2811 | ||
847595de | 2812 | return 0; |
edafccee JA |
2813 | } |
2814 | ||
bcda7baa JA |
2815 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2816 | { | |
2817 | if (needs_lock) | |
2818 | mutex_unlock(&ctx->uring_lock); | |
2819 | } | |
2820 | ||
2821 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2822 | { | |
2823 | /* | |
2824 | * "Normal" inline submissions always hold the uring_lock, since we | |
2825 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2826 | * The only exception is when we've detached the request and issue it | |
2827 | * from an async worker thread, grab the lock for that case. | |
2828 | */ | |
2829 | if (needs_lock) | |
2830 | mutex_lock(&ctx->uring_lock); | |
2831 | } | |
2832 | ||
2833 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2834 | int bgid, struct io_buffer *kbuf, | |
2835 | bool needs_lock) | |
2836 | { | |
2837 | struct io_buffer *head; | |
2838 | ||
2839 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2840 | return kbuf; | |
2841 | ||
2842 | io_ring_submit_lock(req->ctx, needs_lock); | |
2843 | ||
2844 | lockdep_assert_held(&req->ctx->uring_lock); | |
2845 | ||
9e15c3a0 | 2846 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2847 | if (head) { |
2848 | if (!list_empty(&head->list)) { | |
2849 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2850 | list); | |
2851 | list_del(&kbuf->list); | |
2852 | } else { | |
2853 | kbuf = head; | |
9e15c3a0 | 2854 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2855 | } |
2856 | if (*len > kbuf->len) | |
2857 | *len = kbuf->len; | |
2858 | } else { | |
2859 | kbuf = ERR_PTR(-ENOBUFS); | |
2860 | } | |
2861 | ||
2862 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2863 | ||
2864 | return kbuf; | |
2865 | } | |
2866 | ||
4d954c25 JA |
2867 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2868 | bool needs_lock) | |
2869 | { | |
2870 | struct io_buffer *kbuf; | |
4f4eeba8 | 2871 | u16 bgid; |
4d954c25 JA |
2872 | |
2873 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2874 | bgid = req->buf_index; |
4d954c25 JA |
2875 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2876 | if (IS_ERR(kbuf)) | |
2877 | return kbuf; | |
2878 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2879 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2880 | return u64_to_user_ptr(kbuf->addr); | |
2881 | } | |
2882 | ||
2883 | #ifdef CONFIG_COMPAT | |
2884 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2885 | bool needs_lock) | |
2886 | { | |
2887 | struct compat_iovec __user *uiov; | |
2888 | compat_ssize_t clen; | |
2889 | void __user *buf; | |
2890 | ssize_t len; | |
2891 | ||
2892 | uiov = u64_to_user_ptr(req->rw.addr); | |
2893 | if (!access_ok(uiov, sizeof(*uiov))) | |
2894 | return -EFAULT; | |
2895 | if (__get_user(clen, &uiov->iov_len)) | |
2896 | return -EFAULT; | |
2897 | if (clen < 0) | |
2898 | return -EINVAL; | |
2899 | ||
2900 | len = clen; | |
2901 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2902 | if (IS_ERR(buf)) | |
2903 | return PTR_ERR(buf); | |
2904 | iov[0].iov_base = buf; | |
2905 | iov[0].iov_len = (compat_size_t) len; | |
2906 | return 0; | |
2907 | } | |
2908 | #endif | |
2909 | ||
2910 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2911 | bool needs_lock) | |
2912 | { | |
2913 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2914 | void __user *buf; | |
2915 | ssize_t len; | |
2916 | ||
2917 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2918 | return -EFAULT; | |
2919 | ||
2920 | len = iov[0].iov_len; | |
2921 | if (len < 0) | |
2922 | return -EINVAL; | |
2923 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2924 | if (IS_ERR(buf)) | |
2925 | return PTR_ERR(buf); | |
2926 | iov[0].iov_base = buf; | |
2927 | iov[0].iov_len = len; | |
2928 | return 0; | |
2929 | } | |
2930 | ||
2931 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2932 | bool needs_lock) | |
2933 | { | |
dddb3e26 JA |
2934 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2935 | struct io_buffer *kbuf; | |
2936 | ||
2937 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2938 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2939 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2940 | return 0; |
dddb3e26 | 2941 | } |
dd201662 | 2942 | if (req->rw.len != 1) |
4d954c25 JA |
2943 | return -EINVAL; |
2944 | ||
2945 | #ifdef CONFIG_COMPAT | |
2946 | if (req->ctx->compat) | |
2947 | return io_compat_import(req, iov, needs_lock); | |
2948 | #endif | |
2949 | ||
2950 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2951 | } | |
2952 | ||
847595de PB |
2953 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2954 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2955 | { |
9adbd45d JA |
2956 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2957 | size_t sqe_len = req->rw.len; | |
847595de | 2958 | u8 opcode = req->opcode; |
4d954c25 | 2959 | ssize_t ret; |
edafccee | 2960 | |
7d009165 | 2961 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2962 | *iovec = NULL; |
9adbd45d | 2963 | return io_import_fixed(req, rw, iter); |
edafccee | 2964 | } |
2b188cc1 | 2965 | |
bcda7baa | 2966 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2967 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2968 | return -EINVAL; |
2969 | ||
3a6820f2 | 2970 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2971 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2972 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2973 | if (IS_ERR(buf)) |
4d954c25 | 2974 | return PTR_ERR(buf); |
3f9d6441 | 2975 | req->rw.len = sqe_len; |
bcda7baa JA |
2976 | } |
2977 | ||
3a6820f2 JA |
2978 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2979 | *iovec = NULL; | |
10fc72e4 | 2980 | return ret; |
3a6820f2 JA |
2981 | } |
2982 | ||
4d954c25 JA |
2983 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2984 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2985 | if (!ret) |
2986 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2987 | *iovec = NULL; |
2988 | return ret; | |
2989 | } | |
2990 | ||
89cd35c5 CH |
2991 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2992 | req->ctx->compat); | |
2b188cc1 JA |
2993 | } |
2994 | ||
0fef9483 JA |
2995 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2996 | { | |
5b09e37e | 2997 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
2998 | } |
2999 | ||
31b51510 | 3000 | /* |
32960613 JA |
3001 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3002 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3003 | */ |
4017eb91 | 3004 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3005 | { |
4017eb91 JA |
3006 | struct kiocb *kiocb = &req->rw.kiocb; |
3007 | struct file *file = req->file; | |
32960613 JA |
3008 | ssize_t ret = 0; |
3009 | ||
3010 | /* | |
3011 | * Don't support polled IO through this interface, and we can't | |
3012 | * support non-blocking either. For the latter, this just causes | |
3013 | * the kiocb to be handled from an async context. | |
3014 | */ | |
3015 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3016 | return -EOPNOTSUPP; | |
3017 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3018 | return -EAGAIN; | |
3019 | ||
3020 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3021 | struct iovec iovec; |
32960613 JA |
3022 | ssize_t nr; |
3023 | ||
311ae9e1 PB |
3024 | if (!iov_iter_is_bvec(iter)) { |
3025 | iovec = iov_iter_iovec(iter); | |
3026 | } else { | |
4017eb91 JA |
3027 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3028 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3029 | } |
3030 | ||
32960613 JA |
3031 | if (rw == READ) { |
3032 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3033 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3034 | } else { |
3035 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3036 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3037 | } |
3038 | ||
3039 | if (nr < 0) { | |
3040 | if (!ret) | |
3041 | ret = nr; | |
3042 | break; | |
3043 | } | |
3044 | ret += nr; | |
3045 | if (nr != iovec.iov_len) | |
3046 | break; | |
4017eb91 JA |
3047 | req->rw.len -= nr; |
3048 | req->rw.addr += nr; | |
32960613 JA |
3049 | iov_iter_advance(iter, nr); |
3050 | } | |
3051 | ||
3052 | return ret; | |
3053 | } | |
3054 | ||
ff6165b2 JA |
3055 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3056 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3057 | { |
e8c2bc1f | 3058 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3059 | |
ff6165b2 | 3060 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3061 | rw->free_iovec = iovec; |
227c0c96 | 3062 | rw->bytes_done = 0; |
ff6165b2 | 3063 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3064 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3065 | return; |
b64e3444 | 3066 | if (!iovec) { |
ff6165b2 JA |
3067 | unsigned iov_off = 0; |
3068 | ||
3069 | rw->iter.iov = rw->fast_iov; | |
3070 | if (iter->iov != fast_iov) { | |
3071 | iov_off = iter->iov - fast_iov; | |
3072 | rw->iter.iov += iov_off; | |
3073 | } | |
3074 | if (rw->fast_iov != fast_iov) | |
3075 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3076 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3077 | } else { |
3078 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3079 | } |
3080 | } | |
3081 | ||
e8c2bc1f | 3082 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3083 | { |
e8c2bc1f JA |
3084 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3085 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3086 | return req->async_data == NULL; | |
3d9932a8 XW |
3087 | } |
3088 | ||
e8c2bc1f | 3089 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3090 | { |
e8c2bc1f | 3091 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3092 | return 0; |
3d9932a8 | 3093 | |
e8c2bc1f | 3094 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3095 | } |
3096 | ||
ff6165b2 JA |
3097 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3098 | const struct iovec *fast_iov, | |
227c0c96 | 3099 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3100 | { |
e8c2bc1f | 3101 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3102 | return 0; |
e8c2bc1f | 3103 | if (!req->async_data) { |
6bf985dc PB |
3104 | if (__io_alloc_async_data(req)) { |
3105 | kfree(iovec); | |
5d204bcf | 3106 | return -ENOMEM; |
6bf985dc | 3107 | } |
b7bb4f7d | 3108 | |
ff6165b2 | 3109 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3110 | } |
b7bb4f7d | 3111 | return 0; |
f67676d1 JA |
3112 | } |
3113 | ||
73debe68 | 3114 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3115 | { |
e8c2bc1f | 3116 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3117 | struct iovec *iov = iorw->fast_iov; |
847595de | 3118 | int ret; |
c3e330a4 | 3119 | |
2846c481 | 3120 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3121 | if (unlikely(ret < 0)) |
3122 | return ret; | |
3123 | ||
ab0b196c PB |
3124 | iorw->bytes_done = 0; |
3125 | iorw->free_iovec = iov; | |
3126 | if (iov) | |
3127 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3128 | return 0; |
3129 | } | |
3130 | ||
73debe68 | 3131 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3132 | { |
3529d8c2 JA |
3133 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3134 | return -EBADF; | |
93642ef8 | 3135 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3136 | } |
3137 | ||
c1dd91d1 JA |
3138 | /* |
3139 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3140 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3141 | * This gets called when the page is unlocked, and we generally expect that to | |
3142 | * happen when the page IO is completed and the page is now uptodate. This will | |
3143 | * queue a task_work based retry of the operation, attempting to copy the data | |
3144 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3145 | * do a thread based blocking retry of the operation. That's the unexpected | |
3146 | * slow path. | |
3147 | */ | |
bcf5a063 JA |
3148 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3149 | int sync, void *arg) | |
3150 | { | |
3151 | struct wait_page_queue *wpq; | |
3152 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3153 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3154 | |
3155 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3156 | ||
cdc8fcb4 LT |
3157 | if (!wake_page_match(wpq, key)) |
3158 | return 0; | |
3159 | ||
c8d317aa | 3160 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3161 | list_del_init(&wait->entry); |
3162 | ||
bcf5a063 JA |
3163 | /* submit ref gets dropped, acquire a new one */ |
3164 | refcount_inc(&req->refs); | |
921b9054 | 3165 | io_req_task_queue(req); |
bcf5a063 JA |
3166 | return 1; |
3167 | } | |
3168 | ||
c1dd91d1 JA |
3169 | /* |
3170 | * This controls whether a given IO request should be armed for async page | |
3171 | * based retry. If we return false here, the request is handed to the async | |
3172 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3173 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3174 | * will either succeed because the page is now uptodate and unlocked, or it | |
3175 | * will register a callback when the page is unlocked at IO completion. Through | |
3176 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3177 | * That retry will attempt the buffered read again. The retry will generally | |
3178 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3179 | * async worker threads for a blocking retry. | |
3180 | */ | |
227c0c96 | 3181 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3182 | { |
e8c2bc1f JA |
3183 | struct io_async_rw *rw = req->async_data; |
3184 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3185 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3186 | |
bcf5a063 JA |
3187 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3188 | if (req->flags & REQ_F_NOWAIT) | |
3189 | return false; | |
f67676d1 | 3190 | |
227c0c96 | 3191 | /* Only for buffered IO */ |
3b2a4439 | 3192 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3193 | return false; |
3b2a4439 | 3194 | |
bcf5a063 JA |
3195 | /* |
3196 | * just use poll if we can, and don't attempt if the fs doesn't | |
3197 | * support callback based unlocks | |
3198 | */ | |
3199 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3200 | return false; | |
f67676d1 | 3201 | |
3b2a4439 JA |
3202 | wait->wait.func = io_async_buf_func; |
3203 | wait->wait.private = req; | |
3204 | wait->wait.flags = 0; | |
3205 | INIT_LIST_HEAD(&wait->wait.entry); | |
3206 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3207 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3208 | kiocb->ki_waitq = wait; |
3b2a4439 | 3209 | return true; |
bcf5a063 JA |
3210 | } |
3211 | ||
3212 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3213 | { | |
3214 | if (req->file->f_op->read_iter) | |
3215 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3216 | else if (req->file->f_op->read) |
4017eb91 | 3217 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3218 | else |
3219 | return -EINVAL; | |
f67676d1 JA |
3220 | } |
3221 | ||
889fca73 | 3222 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3223 | { |
3224 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3225 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3226 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3227 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3228 | ssize_t io_size, ret, ret2; |
45d189c6 | 3229 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3230 | |
2846c481 | 3231 | if (rw) { |
e8c2bc1f | 3232 | iter = &rw->iter; |
2846c481 PB |
3233 | iovec = NULL; |
3234 | } else { | |
3235 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3236 | if (ret < 0) | |
3237 | return ret; | |
3238 | } | |
632546c4 | 3239 | io_size = iov_iter_count(iter); |
fa15bafb | 3240 | req->result = io_size; |
2b188cc1 | 3241 | |
fd6c2e4c JA |
3242 | /* Ensure we clear previously set non-block flag */ |
3243 | if (!force_nonblock) | |
29de5f6a | 3244 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3245 | else |
3246 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3247 | ||
24c74678 | 3248 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3249 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3250 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3251 | return ret ?: -EAGAIN; |
6713e7a6 | 3252 | } |
9e645e11 | 3253 | |
632546c4 | 3254 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3255 | if (unlikely(ret)) { |
3256 | kfree(iovec); | |
3257 | return ret; | |
3258 | } | |
2b188cc1 | 3259 | |
227c0c96 | 3260 | ret = io_iter_do_read(req, iter); |
32960613 | 3261 | |
57cd657b | 3262 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3263 | if (req->async_data) |
3264 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3265 | goto out_free; |
227c0c96 | 3266 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3267 | /* IOPOLL retry should happen for io-wq threads */ |
3268 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3269 | goto done; |
75c668cd PB |
3270 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3271 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3272 | goto done; |
84216315 | 3273 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3274 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3275 | ret = 0; |
7335e3bf | 3276 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3277 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3278 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3279 | goto done; |
227c0c96 JA |
3280 | } |
3281 | ||
227c0c96 | 3282 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3283 | if (ret2) |
3284 | return ret2; | |
3285 | ||
fe1cdd55 | 3286 | iovec = NULL; |
e8c2bc1f | 3287 | rw = req->async_data; |
227c0c96 | 3288 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3289 | iter = &rw->iter; |
227c0c96 | 3290 | |
b23df91b PB |
3291 | do { |
3292 | io_size -= ret; | |
3293 | rw->bytes_done += ret; | |
3294 | /* if we can retry, do so with the callbacks armed */ | |
3295 | if (!io_rw_should_retry(req)) { | |
3296 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3297 | return -EAGAIN; | |
3298 | } | |
3299 | ||
3300 | /* | |
3301 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3302 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3303 | * desired page gets unlocked. We can also get a partial read | |
3304 | * here, and if we do, then just retry at the new offset. | |
3305 | */ | |
3306 | ret = io_iter_do_read(req, iter); | |
3307 | if (ret == -EIOCBQUEUED) | |
3308 | return 0; | |
227c0c96 | 3309 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3310 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3311 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3312 | done: |
889fca73 | 3313 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3314 | out_free: |
3315 | /* it's faster to check here then delegate to kfree */ | |
3316 | if (iovec) | |
3317 | kfree(iovec); | |
5ea5dd45 | 3318 | return 0; |
2b188cc1 JA |
3319 | } |
3320 | ||
73debe68 | 3321 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3322 | { |
3529d8c2 JA |
3323 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3324 | return -EBADF; | |
93642ef8 | 3325 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3326 | } |
3327 | ||
889fca73 | 3328 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3329 | { |
3330 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3331 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3332 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3333 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3334 | ssize_t ret, ret2, io_size; |
45d189c6 | 3335 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3336 | |
2846c481 | 3337 | if (rw) { |
e8c2bc1f | 3338 | iter = &rw->iter; |
2846c481 PB |
3339 | iovec = NULL; |
3340 | } else { | |
3341 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3342 | if (ret < 0) | |
3343 | return ret; | |
3344 | } | |
632546c4 | 3345 | io_size = iov_iter_count(iter); |
fa15bafb | 3346 | req->result = io_size; |
2b188cc1 | 3347 | |
fd6c2e4c JA |
3348 | /* Ensure we clear previously set non-block flag */ |
3349 | if (!force_nonblock) | |
a88fc400 PB |
3350 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3351 | else | |
3352 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3353 | |
24c74678 | 3354 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3355 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3356 | goto copy_iov; |
31b51510 | 3357 | |
10d59345 JA |
3358 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3359 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3360 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3361 | goto copy_iov; |
31b51510 | 3362 | |
632546c4 | 3363 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3364 | if (unlikely(ret)) |
3365 | goto out_free; | |
4ed734b0 | 3366 | |
fa15bafb PB |
3367 | /* |
3368 | * Open-code file_start_write here to grab freeze protection, | |
3369 | * which will be released by another thread in | |
3370 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3371 | * released so that it doesn't complain about the held lock when | |
3372 | * we return to userspace. | |
3373 | */ | |
3374 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3375 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3376 | __sb_writers_release(file_inode(req->file)->i_sb, |
3377 | SB_FREEZE_WRITE); | |
3378 | } | |
3379 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3380 | |
fa15bafb | 3381 | if (req->file->f_op->write_iter) |
ff6165b2 | 3382 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3383 | else if (req->file->f_op->write) |
4017eb91 | 3384 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3385 | else |
3386 | ret2 = -EINVAL; | |
4ed734b0 | 3387 | |
fa15bafb PB |
3388 | /* |
3389 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3390 | * retry them without IOCB_NOWAIT. | |
3391 | */ | |
3392 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3393 | ret2 = -EAGAIN; | |
75c668cd PB |
3394 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3395 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3396 | goto done; |
3e6a0d3c JA |
3397 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3398 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3399 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3400 | /* IOPOLL retry should happen for io-wq threads */ |
3401 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3402 | goto copy_iov; | |
355afaeb | 3403 | done: |
889fca73 | 3404 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3405 | } else { |
f67676d1 | 3406 | copy_iov: |
84216315 | 3407 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3408 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3409 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3410 | return ret ?: -EAGAIN; |
2b188cc1 | 3411 | } |
31b51510 | 3412 | out_free: |
f261c168 | 3413 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3414 | if (iovec) |
6f2cc166 | 3415 | kfree(iovec); |
2b188cc1 JA |
3416 | return ret; |
3417 | } | |
3418 | ||
80a261fd JA |
3419 | static int io_renameat_prep(struct io_kiocb *req, |
3420 | const struct io_uring_sqe *sqe) | |
3421 | { | |
3422 | struct io_rename *ren = &req->rename; | |
3423 | const char __user *oldf, *newf; | |
3424 | ||
3425 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3426 | return -EBADF; | |
3427 | ||
3428 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3429 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3430 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3431 | ren->new_dfd = READ_ONCE(sqe->len); | |
3432 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3433 | ||
3434 | ren->oldpath = getname(oldf); | |
3435 | if (IS_ERR(ren->oldpath)) | |
3436 | return PTR_ERR(ren->oldpath); | |
3437 | ||
3438 | ren->newpath = getname(newf); | |
3439 | if (IS_ERR(ren->newpath)) { | |
3440 | putname(ren->oldpath); | |
3441 | return PTR_ERR(ren->newpath); | |
3442 | } | |
3443 | ||
3444 | req->flags |= REQ_F_NEED_CLEANUP; | |
3445 | return 0; | |
3446 | } | |
3447 | ||
45d189c6 | 3448 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3449 | { |
3450 | struct io_rename *ren = &req->rename; | |
3451 | int ret; | |
3452 | ||
45d189c6 | 3453 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3454 | return -EAGAIN; |
3455 | ||
3456 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3457 | ren->newpath, ren->flags); | |
3458 | ||
3459 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3460 | if (ret < 0) | |
3461 | req_set_fail_links(req); | |
3462 | io_req_complete(req, ret); | |
3463 | return 0; | |
3464 | } | |
3465 | ||
14a1143b JA |
3466 | static int io_unlinkat_prep(struct io_kiocb *req, |
3467 | const struct io_uring_sqe *sqe) | |
3468 | { | |
3469 | struct io_unlink *un = &req->unlink; | |
3470 | const char __user *fname; | |
3471 | ||
3472 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3473 | return -EBADF; | |
3474 | ||
3475 | un->dfd = READ_ONCE(sqe->fd); | |
3476 | ||
3477 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3478 | if (un->flags & ~AT_REMOVEDIR) | |
3479 | return -EINVAL; | |
3480 | ||
3481 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3482 | un->filename = getname(fname); | |
3483 | if (IS_ERR(un->filename)) | |
3484 | return PTR_ERR(un->filename); | |
3485 | ||
3486 | req->flags |= REQ_F_NEED_CLEANUP; | |
3487 | return 0; | |
3488 | } | |
3489 | ||
45d189c6 | 3490 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3491 | { |
3492 | struct io_unlink *un = &req->unlink; | |
3493 | int ret; | |
3494 | ||
45d189c6 | 3495 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3496 | return -EAGAIN; |
3497 | ||
3498 | if (un->flags & AT_REMOVEDIR) | |
3499 | ret = do_rmdir(un->dfd, un->filename); | |
3500 | else | |
3501 | ret = do_unlinkat(un->dfd, un->filename); | |
3502 | ||
3503 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3504 | if (ret < 0) | |
3505 | req_set_fail_links(req); | |
3506 | io_req_complete(req, ret); | |
3507 | return 0; | |
3508 | } | |
3509 | ||
36f4fa68 JA |
3510 | static int io_shutdown_prep(struct io_kiocb *req, |
3511 | const struct io_uring_sqe *sqe) | |
3512 | { | |
3513 | #if defined(CONFIG_NET) | |
3514 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3515 | return -EINVAL; | |
3516 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3517 | sqe->buf_index) | |
3518 | return -EINVAL; | |
3519 | ||
3520 | req->shutdown.how = READ_ONCE(sqe->len); | |
3521 | return 0; | |
3522 | #else | |
3523 | return -EOPNOTSUPP; | |
3524 | #endif | |
3525 | } | |
3526 | ||
45d189c6 | 3527 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3528 | { |
3529 | #if defined(CONFIG_NET) | |
3530 | struct socket *sock; | |
3531 | int ret; | |
3532 | ||
45d189c6 | 3533 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3534 | return -EAGAIN; |
3535 | ||
48aba79b | 3536 | sock = sock_from_file(req->file); |
36f4fa68 | 3537 | if (unlikely(!sock)) |
48aba79b | 3538 | return -ENOTSOCK; |
36f4fa68 JA |
3539 | |
3540 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3541 | if (ret < 0) |
3542 | req_set_fail_links(req); | |
36f4fa68 JA |
3543 | io_req_complete(req, ret); |
3544 | return 0; | |
3545 | #else | |
3546 | return -EOPNOTSUPP; | |
3547 | #endif | |
3548 | } | |
3549 | ||
f2a8d5c7 PB |
3550 | static int __io_splice_prep(struct io_kiocb *req, |
3551 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3552 | { |
3553 | struct io_splice* sp = &req->splice; | |
3554 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3555 | |
3232dd02 PB |
3556 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3557 | return -EINVAL; | |
7d67af2c PB |
3558 | |
3559 | sp->file_in = NULL; | |
7d67af2c PB |
3560 | sp->len = READ_ONCE(sqe->len); |
3561 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3562 | ||
3563 | if (unlikely(sp->flags & ~valid_flags)) | |
3564 | return -EINVAL; | |
3565 | ||
8371adf5 PB |
3566 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3567 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3568 | if (!sp->file_in) | |
3569 | return -EBADF; | |
7d67af2c PB |
3570 | req->flags |= REQ_F_NEED_CLEANUP; |
3571 | ||
7cdaf587 XW |
3572 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3573 | /* | |
3574 | * Splice operation will be punted aync, and here need to | |
3575 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3576 | */ | |
7d67af2c | 3577 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3578 | } |
7d67af2c PB |
3579 | |
3580 | return 0; | |
3581 | } | |
3582 | ||
f2a8d5c7 PB |
3583 | static int io_tee_prep(struct io_kiocb *req, |
3584 | const struct io_uring_sqe *sqe) | |
3585 | { | |
3586 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3587 | return -EINVAL; | |
3588 | return __io_splice_prep(req, sqe); | |
3589 | } | |
3590 | ||
45d189c6 | 3591 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3592 | { |
3593 | struct io_splice *sp = &req->splice; | |
3594 | struct file *in = sp->file_in; | |
3595 | struct file *out = sp->file_out; | |
3596 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3597 | long ret = 0; | |
3598 | ||
45d189c6 | 3599 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3600 | return -EAGAIN; |
3601 | if (sp->len) | |
3602 | ret = do_tee(in, out, sp->len, flags); | |
3603 | ||
3604 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3605 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3606 | ||
f2a8d5c7 PB |
3607 | if (ret != sp->len) |
3608 | req_set_fail_links(req); | |
e1e16097 | 3609 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3610 | return 0; |
3611 | } | |
3612 | ||
3613 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3614 | { | |
3615 | struct io_splice* sp = &req->splice; | |
3616 | ||
3617 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3618 | sp->off_out = READ_ONCE(sqe->off); | |
3619 | return __io_splice_prep(req, sqe); | |
3620 | } | |
3621 | ||
45d189c6 | 3622 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3623 | { |
3624 | struct io_splice *sp = &req->splice; | |
3625 | struct file *in = sp->file_in; | |
3626 | struct file *out = sp->file_out; | |
3627 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3628 | loff_t *poff_in, *poff_out; | |
c9687426 | 3629 | long ret = 0; |
7d67af2c | 3630 | |
45d189c6 | 3631 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3632 | return -EAGAIN; |
7d67af2c PB |
3633 | |
3634 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3635 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3636 | |
948a7749 | 3637 | if (sp->len) |
c9687426 | 3638 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3639 | |
3640 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3641 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3642 | ||
7d67af2c PB |
3643 | if (ret != sp->len) |
3644 | req_set_fail_links(req); | |
e1e16097 | 3645 | io_req_complete(req, ret); |
7d67af2c PB |
3646 | return 0; |
3647 | } | |
3648 | ||
2b188cc1 JA |
3649 | /* |
3650 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3651 | */ | |
889fca73 | 3652 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3653 | { |
3654 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3655 | |
def596e9 JA |
3656 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3657 | return -EINVAL; | |
3658 | ||
889fca73 | 3659 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3660 | return 0; |
3661 | } | |
3662 | ||
1155c76a | 3663 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3664 | { |
6b06314c | 3665 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3666 | |
09bb8394 JA |
3667 | if (!req->file) |
3668 | return -EBADF; | |
c992fe29 | 3669 | |
6b06314c | 3670 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3671 | return -EINVAL; |
edafccee | 3672 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3673 | return -EINVAL; |
3674 | ||
8ed8d3c3 JA |
3675 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3676 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3677 | return -EINVAL; | |
3678 | ||
3679 | req->sync.off = READ_ONCE(sqe->off); | |
3680 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3681 | return 0; |
3682 | } | |
3683 | ||
45d189c6 | 3684 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3685 | { |
8ed8d3c3 | 3686 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3687 | int ret; |
3688 | ||
ac45abc0 | 3689 | /* fsync always requires a blocking context */ |
45d189c6 | 3690 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3691 | return -EAGAIN; |
3692 | ||
9adbd45d | 3693 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3694 | end > 0 ? end : LLONG_MAX, |
3695 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3696 | if (ret < 0) | |
3697 | req_set_fail_links(req); | |
e1e16097 | 3698 | io_req_complete(req, ret); |
c992fe29 CH |
3699 | return 0; |
3700 | } | |
3701 | ||
d63d1b5e JA |
3702 | static int io_fallocate_prep(struct io_kiocb *req, |
3703 | const struct io_uring_sqe *sqe) | |
3704 | { | |
3705 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3706 | return -EINVAL; | |
3232dd02 PB |
3707 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3708 | return -EINVAL; | |
d63d1b5e JA |
3709 | |
3710 | req->sync.off = READ_ONCE(sqe->off); | |
3711 | req->sync.len = READ_ONCE(sqe->addr); | |
3712 | req->sync.mode = READ_ONCE(sqe->len); | |
3713 | return 0; | |
3714 | } | |
3715 | ||
45d189c6 | 3716 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3717 | { |
ac45abc0 PB |
3718 | int ret; |
3719 | ||
d63d1b5e | 3720 | /* fallocate always requiring blocking context */ |
45d189c6 | 3721 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3722 | return -EAGAIN; |
ac45abc0 PB |
3723 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3724 | req->sync.len); | |
ac45abc0 PB |
3725 | if (ret < 0) |
3726 | req_set_fail_links(req); | |
e1e16097 | 3727 | io_req_complete(req, ret); |
5d17b4a4 JA |
3728 | return 0; |
3729 | } | |
3730 | ||
ec65fea5 | 3731 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3732 | { |
f8748881 | 3733 | const char __user *fname; |
15b71abe | 3734 | int ret; |
b7bb4f7d | 3735 | |
ec65fea5 | 3736 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3737 | return -EINVAL; |
ec65fea5 | 3738 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3739 | return -EBADF; |
03b1230c | 3740 | |
ec65fea5 PB |
3741 | /* open.how should be already initialised */ |
3742 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3743 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3744 | |
25e72d10 PB |
3745 | req->open.dfd = READ_ONCE(sqe->fd); |
3746 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3747 | req->open.filename = getname(fname); |
15b71abe JA |
3748 | if (IS_ERR(req->open.filename)) { |
3749 | ret = PTR_ERR(req->open.filename); | |
3750 | req->open.filename = NULL; | |
3751 | return ret; | |
3752 | } | |
4022e7af | 3753 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3754 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3755 | return 0; |
03b1230c JA |
3756 | } |
3757 | ||
ec65fea5 PB |
3758 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3759 | { | |
3760 | u64 flags, mode; | |
3761 | ||
14587a46 | 3762 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3763 | return -EINVAL; |
ec65fea5 PB |
3764 | mode = READ_ONCE(sqe->len); |
3765 | flags = READ_ONCE(sqe->open_flags); | |
3766 | req->open.how = build_open_how(flags, mode); | |
3767 | return __io_openat_prep(req, sqe); | |
3768 | } | |
3769 | ||
cebdb986 | 3770 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3771 | { |
cebdb986 | 3772 | struct open_how __user *how; |
cebdb986 | 3773 | size_t len; |
0fa03c62 JA |
3774 | int ret; |
3775 | ||
14587a46 | 3776 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3777 | return -EINVAL; |
cebdb986 JA |
3778 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3779 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3780 | if (len < OPEN_HOW_SIZE_VER0) |
3781 | return -EINVAL; | |
3529d8c2 | 3782 | |
cebdb986 JA |
3783 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3784 | len); | |
3785 | if (ret) | |
3786 | return ret; | |
3529d8c2 | 3787 | |
ec65fea5 | 3788 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3789 | } |
3790 | ||
45d189c6 | 3791 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3792 | { |
3793 | struct open_flags op; | |
15b71abe | 3794 | struct file *file; |
3a81fd02 JA |
3795 | bool nonblock_set; |
3796 | bool resolve_nonblock; | |
15b71abe JA |
3797 | int ret; |
3798 | ||
cebdb986 | 3799 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3800 | if (ret) |
3801 | goto err; | |
3a81fd02 JA |
3802 | nonblock_set = op.open_flag & O_NONBLOCK; |
3803 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3804 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3805 | /* |
3806 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3807 | * it'll always -EAGAIN | |
3808 | */ | |
3809 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3810 | return -EAGAIN; | |
3811 | op.lookup_flags |= LOOKUP_CACHED; | |
3812 | op.open_flag |= O_NONBLOCK; | |
3813 | } | |
15b71abe | 3814 | |
4022e7af | 3815 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3816 | if (ret < 0) |
3817 | goto err; | |
3818 | ||
3819 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3820 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3821 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3822 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3823 | /* |
3a81fd02 JA |
3824 | * We could hang on to this 'fd', but seems like marginal |
3825 | * gain for something that is now known to be a slower path. | |
3826 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3827 | */ |
3a81fd02 JA |
3828 | put_unused_fd(ret); |
3829 | return -EAGAIN; | |
3830 | } | |
3831 | ||
15b71abe JA |
3832 | if (IS_ERR(file)) { |
3833 | put_unused_fd(ret); | |
3834 | ret = PTR_ERR(file); | |
3835 | } else { | |
45d189c6 | 3836 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3837 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3838 | fsnotify_open(file); |
3839 | fd_install(ret, file); | |
3840 | } | |
3841 | err: | |
3842 | putname(req->open.filename); | |
8fef80bf | 3843 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3844 | if (ret < 0) |
3845 | req_set_fail_links(req); | |
e1e16097 | 3846 | io_req_complete(req, ret); |
15b71abe JA |
3847 | return 0; |
3848 | } | |
3849 | ||
45d189c6 | 3850 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3851 | { |
e45cff58 | 3852 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3853 | } |
3854 | ||
067524e9 JA |
3855 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3856 | const struct io_uring_sqe *sqe) | |
3857 | { | |
3858 | struct io_provide_buf *p = &req->pbuf; | |
3859 | u64 tmp; | |
3860 | ||
3861 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3862 | return -EINVAL; | |
3863 | ||
3864 | tmp = READ_ONCE(sqe->fd); | |
3865 | if (!tmp || tmp > USHRT_MAX) | |
3866 | return -EINVAL; | |
3867 | ||
3868 | memset(p, 0, sizeof(*p)); | |
3869 | p->nbufs = tmp; | |
3870 | p->bgid = READ_ONCE(sqe->buf_group); | |
3871 | return 0; | |
3872 | } | |
3873 | ||
3874 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3875 | int bgid, unsigned nbufs) | |
3876 | { | |
3877 | unsigned i = 0; | |
3878 | ||
3879 | /* shouldn't happen */ | |
3880 | if (!nbufs) | |
3881 | return 0; | |
3882 | ||
3883 | /* the head kbuf is the list itself */ | |
3884 | while (!list_empty(&buf->list)) { | |
3885 | struct io_buffer *nxt; | |
3886 | ||
3887 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3888 | list_del(&nxt->list); | |
3889 | kfree(nxt); | |
3890 | if (++i == nbufs) | |
3891 | return i; | |
3892 | } | |
3893 | i++; | |
3894 | kfree(buf); | |
9e15c3a0 | 3895 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3896 | |
3897 | return i; | |
3898 | } | |
3899 | ||
889fca73 | 3900 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3901 | { |
3902 | struct io_provide_buf *p = &req->pbuf; | |
3903 | struct io_ring_ctx *ctx = req->ctx; | |
3904 | struct io_buffer *head; | |
3905 | int ret = 0; | |
45d189c6 | 3906 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3907 | |
3908 | io_ring_submit_lock(ctx, !force_nonblock); | |
3909 | ||
3910 | lockdep_assert_held(&ctx->uring_lock); | |
3911 | ||
3912 | ret = -ENOENT; | |
9e15c3a0 | 3913 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3914 | if (head) |
3915 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3916 | if (ret < 0) |
3917 | req_set_fail_links(req); | |
067524e9 | 3918 | |
31bff9a5 PB |
3919 | /* need to hold the lock to complete IOPOLL requests */ |
3920 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3921 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3922 | io_ring_submit_unlock(ctx, !force_nonblock); |
3923 | } else { | |
3924 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3925 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3926 | } |
067524e9 JA |
3927 | return 0; |
3928 | } | |
3929 | ||
ddf0322d JA |
3930 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3931 | const struct io_uring_sqe *sqe) | |
3932 | { | |
3933 | struct io_provide_buf *p = &req->pbuf; | |
3934 | u64 tmp; | |
3935 | ||
3936 | if (sqe->ioprio || sqe->rw_flags) | |
3937 | return -EINVAL; | |
3938 | ||
3939 | tmp = READ_ONCE(sqe->fd); | |
3940 | if (!tmp || tmp > USHRT_MAX) | |
3941 | return -E2BIG; | |
3942 | p->nbufs = tmp; | |
3943 | p->addr = READ_ONCE(sqe->addr); | |
3944 | p->len = READ_ONCE(sqe->len); | |
3945 | ||
efe68c1c | 3946 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3947 | return -EFAULT; |
3948 | ||
3949 | p->bgid = READ_ONCE(sqe->buf_group); | |
3950 | tmp = READ_ONCE(sqe->off); | |
3951 | if (tmp > USHRT_MAX) | |
3952 | return -E2BIG; | |
3953 | p->bid = tmp; | |
3954 | return 0; | |
3955 | } | |
3956 | ||
3957 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3958 | { | |
3959 | struct io_buffer *buf; | |
3960 | u64 addr = pbuf->addr; | |
3961 | int i, bid = pbuf->bid; | |
3962 | ||
3963 | for (i = 0; i < pbuf->nbufs; i++) { | |
3964 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3965 | if (!buf) | |
3966 | break; | |
3967 | ||
3968 | buf->addr = addr; | |
3969 | buf->len = pbuf->len; | |
3970 | buf->bid = bid; | |
3971 | addr += pbuf->len; | |
3972 | bid++; | |
3973 | if (!*head) { | |
3974 | INIT_LIST_HEAD(&buf->list); | |
3975 | *head = buf; | |
3976 | } else { | |
3977 | list_add_tail(&buf->list, &(*head)->list); | |
3978 | } | |
3979 | } | |
3980 | ||
3981 | return i ? i : -ENOMEM; | |
3982 | } | |
3983 | ||
889fca73 | 3984 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3985 | { |
3986 | struct io_provide_buf *p = &req->pbuf; | |
3987 | struct io_ring_ctx *ctx = req->ctx; | |
3988 | struct io_buffer *head, *list; | |
3989 | int ret = 0; | |
45d189c6 | 3990 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3991 | |
3992 | io_ring_submit_lock(ctx, !force_nonblock); | |
3993 | ||
3994 | lockdep_assert_held(&ctx->uring_lock); | |
3995 | ||
9e15c3a0 | 3996 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
3997 | |
3998 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
3999 | if (ret >= 0 && !list) { |
4000 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4001 | if (ret < 0) | |
067524e9 | 4002 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4003 | } |
ddf0322d JA |
4004 | if (ret < 0) |
4005 | req_set_fail_links(req); | |
31bff9a5 PB |
4006 | |
4007 | /* need to hold the lock to complete IOPOLL requests */ | |
4008 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4009 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4010 | io_ring_submit_unlock(ctx, !force_nonblock); |
4011 | } else { | |
4012 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4013 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4014 | } |
ddf0322d | 4015 | return 0; |
cebdb986 JA |
4016 | } |
4017 | ||
3e4827b0 JA |
4018 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4019 | const struct io_uring_sqe *sqe) | |
4020 | { | |
4021 | #if defined(CONFIG_EPOLL) | |
4022 | if (sqe->ioprio || sqe->buf_index) | |
4023 | return -EINVAL; | |
6ca56f84 | 4024 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4025 | return -EINVAL; |
3e4827b0 JA |
4026 | |
4027 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4028 | req->epoll.op = READ_ONCE(sqe->len); | |
4029 | req->epoll.fd = READ_ONCE(sqe->off); | |
4030 | ||
4031 | if (ep_op_has_event(req->epoll.op)) { | |
4032 | struct epoll_event __user *ev; | |
4033 | ||
4034 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4035 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4036 | return -EFAULT; | |
4037 | } | |
4038 | ||
4039 | return 0; | |
4040 | #else | |
4041 | return -EOPNOTSUPP; | |
4042 | #endif | |
4043 | } | |
4044 | ||
889fca73 | 4045 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4046 | { |
4047 | #if defined(CONFIG_EPOLL) | |
4048 | struct io_epoll *ie = &req->epoll; | |
4049 | int ret; | |
45d189c6 | 4050 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4051 | |
4052 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4053 | if (force_nonblock && ret == -EAGAIN) | |
4054 | return -EAGAIN; | |
4055 | ||
4056 | if (ret < 0) | |
4057 | req_set_fail_links(req); | |
889fca73 | 4058 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4059 | return 0; |
4060 | #else | |
4061 | return -EOPNOTSUPP; | |
4062 | #endif | |
4063 | } | |
4064 | ||
c1ca757b JA |
4065 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4066 | { | |
4067 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4068 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4069 | return -EINVAL; | |
3232dd02 PB |
4070 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4071 | return -EINVAL; | |
c1ca757b JA |
4072 | |
4073 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4074 | req->madvise.len = READ_ONCE(sqe->len); | |
4075 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4076 | return 0; | |
4077 | #else | |
4078 | return -EOPNOTSUPP; | |
4079 | #endif | |
4080 | } | |
4081 | ||
45d189c6 | 4082 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4083 | { |
4084 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4085 | struct io_madvise *ma = &req->madvise; | |
4086 | int ret; | |
4087 | ||
45d189c6 | 4088 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4089 | return -EAGAIN; |
4090 | ||
0726b01e | 4091 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4092 | if (ret < 0) |
4093 | req_set_fail_links(req); | |
e1e16097 | 4094 | io_req_complete(req, ret); |
c1ca757b JA |
4095 | return 0; |
4096 | #else | |
4097 | return -EOPNOTSUPP; | |
4098 | #endif | |
4099 | } | |
4100 | ||
4840e418 JA |
4101 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4102 | { | |
4103 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4104 | return -EINVAL; | |
3232dd02 PB |
4105 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4106 | return -EINVAL; | |
4840e418 JA |
4107 | |
4108 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4109 | req->fadvise.len = READ_ONCE(sqe->len); | |
4110 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4111 | return 0; | |
4112 | } | |
4113 | ||
45d189c6 | 4114 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4115 | { |
4116 | struct io_fadvise *fa = &req->fadvise; | |
4117 | int ret; | |
4118 | ||
45d189c6 | 4119 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4120 | switch (fa->advice) { |
4121 | case POSIX_FADV_NORMAL: | |
4122 | case POSIX_FADV_RANDOM: | |
4123 | case POSIX_FADV_SEQUENTIAL: | |
4124 | break; | |
4125 | default: | |
4126 | return -EAGAIN; | |
4127 | } | |
4128 | } | |
4840e418 JA |
4129 | |
4130 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4131 | if (ret < 0) | |
4132 | req_set_fail_links(req); | |
e1e16097 | 4133 | io_req_complete(req, ret); |
4840e418 JA |
4134 | return 0; |
4135 | } | |
4136 | ||
eddc7ef5 JA |
4137 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4138 | { | |
6ca56f84 | 4139 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4140 | return -EINVAL; |
eddc7ef5 JA |
4141 | if (sqe->ioprio || sqe->buf_index) |
4142 | return -EINVAL; | |
9c280f90 | 4143 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4144 | return -EBADF; |
eddc7ef5 | 4145 | |
1d9e1288 BM |
4146 | req->statx.dfd = READ_ONCE(sqe->fd); |
4147 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4148 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4149 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4150 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4151 | |
4152 | return 0; | |
4153 | } | |
4154 | ||
45d189c6 | 4155 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4156 | { |
1d9e1288 | 4157 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4158 | int ret; |
4159 | ||
45d189c6 | 4160 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4161 | /* only need file table for an actual valid fd */ |
4162 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4163 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4164 | return -EAGAIN; |
5b0bbee4 | 4165 | } |
eddc7ef5 | 4166 | |
e62753e4 BM |
4167 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4168 | ctx->buffer); | |
eddc7ef5 | 4169 | |
eddc7ef5 JA |
4170 | if (ret < 0) |
4171 | req_set_fail_links(req); | |
e1e16097 | 4172 | io_req_complete(req, ret); |
eddc7ef5 JA |
4173 | return 0; |
4174 | } | |
4175 | ||
b5dba59e JA |
4176 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4177 | { | |
14587a46 | 4178 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4179 | return -EINVAL; |
b5dba59e JA |
4180 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4181 | sqe->rw_flags || sqe->buf_index) | |
4182 | return -EINVAL; | |
9c280f90 | 4183 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4184 | return -EBADF; |
b5dba59e JA |
4185 | |
4186 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4187 | return 0; |
b5dba59e JA |
4188 | } |
4189 | ||
889fca73 | 4190 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4191 | { |
9eac1904 | 4192 | struct files_struct *files = current->files; |
3af73b28 | 4193 | struct io_close *close = &req->close; |
9eac1904 JA |
4194 | struct fdtable *fdt; |
4195 | struct file *file; | |
b5dba59e JA |
4196 | int ret; |
4197 | ||
9eac1904 JA |
4198 | file = NULL; |
4199 | ret = -EBADF; | |
4200 | spin_lock(&files->file_lock); | |
4201 | fdt = files_fdtable(files); | |
4202 | if (close->fd >= fdt->max_fds) { | |
4203 | spin_unlock(&files->file_lock); | |
4204 | goto err; | |
4205 | } | |
4206 | file = fdt->fd[close->fd]; | |
4207 | if (!file) { | |
4208 | spin_unlock(&files->file_lock); | |
4209 | goto err; | |
4210 | } | |
4211 | ||
4212 | if (file->f_op == &io_uring_fops) { | |
4213 | spin_unlock(&files->file_lock); | |
4214 | file = NULL; | |
4215 | goto err; | |
3af73b28 | 4216 | } |
b5dba59e JA |
4217 | |
4218 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4219 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4220 | spin_unlock(&files->file_lock); |
0bf0eefd | 4221 | return -EAGAIN; |
a2100672 | 4222 | } |
b5dba59e | 4223 | |
9eac1904 JA |
4224 | ret = __close_fd_get_file(close->fd, &file); |
4225 | spin_unlock(&files->file_lock); | |
4226 | if (ret < 0) { | |
4227 | if (ret == -ENOENT) | |
4228 | ret = -EBADF; | |
4229 | goto err; | |
4230 | } | |
4231 | ||
3af73b28 | 4232 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4233 | ret = filp_close(file, current->files); |
4234 | err: | |
3af73b28 PB |
4235 | if (ret < 0) |
4236 | req_set_fail_links(req); | |
9eac1904 JA |
4237 | if (file) |
4238 | fput(file); | |
889fca73 | 4239 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4240 | return 0; |
b5dba59e JA |
4241 | } |
4242 | ||
1155c76a | 4243 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4244 | { |
4245 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4246 | |
5d17b4a4 JA |
4247 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4248 | return -EINVAL; | |
4249 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4250 | return -EINVAL; | |
4251 | ||
8ed8d3c3 JA |
4252 | req->sync.off = READ_ONCE(sqe->off); |
4253 | req->sync.len = READ_ONCE(sqe->len); | |
4254 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4255 | return 0; |
4256 | } | |
4257 | ||
45d189c6 | 4258 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4259 | { |
8ed8d3c3 JA |
4260 | int ret; |
4261 | ||
ac45abc0 | 4262 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4263 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4264 | return -EAGAIN; |
4265 | ||
9adbd45d | 4266 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4267 | req->sync.flags); |
4268 | if (ret < 0) | |
4269 | req_set_fail_links(req); | |
e1e16097 | 4270 | io_req_complete(req, ret); |
5d17b4a4 JA |
4271 | return 0; |
4272 | } | |
4273 | ||
469956e8 | 4274 | #if defined(CONFIG_NET) |
02d27d89 PB |
4275 | static int io_setup_async_msg(struct io_kiocb *req, |
4276 | struct io_async_msghdr *kmsg) | |
4277 | { | |
e8c2bc1f JA |
4278 | struct io_async_msghdr *async_msg = req->async_data; |
4279 | ||
4280 | if (async_msg) | |
02d27d89 | 4281 | return -EAGAIN; |
e8c2bc1f | 4282 | if (io_alloc_async_data(req)) { |
257e84a5 | 4283 | kfree(kmsg->free_iov); |
02d27d89 PB |
4284 | return -ENOMEM; |
4285 | } | |
e8c2bc1f | 4286 | async_msg = req->async_data; |
02d27d89 | 4287 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4288 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4289 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4290 | /* if were using fast_iov, set it to the new one */ |
4291 | if (!async_msg->free_iov) | |
4292 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4293 | ||
02d27d89 PB |
4294 | return -EAGAIN; |
4295 | } | |
4296 | ||
2ae523ed PB |
4297 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4298 | struct io_async_msghdr *iomsg) | |
4299 | { | |
2ae523ed | 4300 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4301 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4302 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4303 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4304 | } |
4305 | ||
93642ef8 PB |
4306 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4307 | { | |
4308 | int ret; | |
4309 | ||
4310 | if (!io_op_defs[req->opcode].needs_async_data) | |
4311 | return 0; | |
4312 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4313 | if (!ret) | |
4314 | req->flags |= REQ_F_NEED_CLEANUP; | |
4315 | return ret; | |
4316 | } | |
4317 | ||
3529d8c2 | 4318 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4319 | { |
e47293fd | 4320 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4321 | |
d2b6f48b PB |
4322 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4323 | return -EINVAL; | |
4324 | ||
e47293fd | 4325 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4326 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4327 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4328 | |
d8768362 JA |
4329 | #ifdef CONFIG_COMPAT |
4330 | if (req->ctx->compat) | |
4331 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4332 | #endif | |
93642ef8 | 4333 | return 0; |
03b1230c JA |
4334 | } |
4335 | ||
889fca73 | 4336 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4337 | { |
6b754c8b | 4338 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4339 | struct socket *sock; |
7a7cacba | 4340 | unsigned flags; |
0fa03c62 JA |
4341 | int ret; |
4342 | ||
dba4a925 | 4343 | sock = sock_from_file(req->file); |
7a7cacba | 4344 | if (unlikely(!sock)) |
dba4a925 | 4345 | return -ENOTSOCK; |
3529d8c2 | 4346 | |
257e84a5 PB |
4347 | kmsg = req->async_data; |
4348 | if (!kmsg) { | |
7a7cacba PB |
4349 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4350 | if (ret) | |
4351 | return ret; | |
4352 | kmsg = &iomsg; | |
0fa03c62 | 4353 | } |
0fa03c62 | 4354 | |
7a7cacba PB |
4355 | flags = req->sr_msg.msg_flags; |
4356 | if (flags & MSG_DONTWAIT) | |
4357 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4358 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4359 | flags |= MSG_DONTWAIT; |
e47293fd | 4360 | |
7a7cacba | 4361 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4362 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4363 | return io_setup_async_msg(req, kmsg); |
4364 | if (ret == -ERESTARTSYS) | |
4365 | ret = -EINTR; | |
0fa03c62 | 4366 | |
257e84a5 PB |
4367 | /* fast path, check for non-NULL to avoid function call */ |
4368 | if (kmsg->free_iov) | |
4369 | kfree(kmsg->free_iov); | |
99bc4c38 | 4370 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4371 | if (ret < 0) |
4372 | req_set_fail_links(req); | |
889fca73 | 4373 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4374 | return 0; |
03b1230c | 4375 | } |
aa1fa28f | 4376 | |
889fca73 | 4377 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4378 | { |
7a7cacba PB |
4379 | struct io_sr_msg *sr = &req->sr_msg; |
4380 | struct msghdr msg; | |
4381 | struct iovec iov; | |
fddaface | 4382 | struct socket *sock; |
7a7cacba | 4383 | unsigned flags; |
fddaface JA |
4384 | int ret; |
4385 | ||
dba4a925 | 4386 | sock = sock_from_file(req->file); |
7a7cacba | 4387 | if (unlikely(!sock)) |
dba4a925 | 4388 | return -ENOTSOCK; |
fddaface | 4389 | |
7a7cacba PB |
4390 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4391 | if (unlikely(ret)) | |
14db8411 | 4392 | return ret; |
fddaface | 4393 | |
7a7cacba PB |
4394 | msg.msg_name = NULL; |
4395 | msg.msg_control = NULL; | |
4396 | msg.msg_controllen = 0; | |
4397 | msg.msg_namelen = 0; | |
fddaface | 4398 | |
7a7cacba PB |
4399 | flags = req->sr_msg.msg_flags; |
4400 | if (flags & MSG_DONTWAIT) | |
4401 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4402 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4403 | flags |= MSG_DONTWAIT; |
fddaface | 4404 | |
7a7cacba PB |
4405 | msg.msg_flags = flags; |
4406 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4407 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4408 | return -EAGAIN; |
4409 | if (ret == -ERESTARTSYS) | |
4410 | ret = -EINTR; | |
fddaface | 4411 | |
fddaface JA |
4412 | if (ret < 0) |
4413 | req_set_fail_links(req); | |
889fca73 | 4414 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4415 | return 0; |
fddaface JA |
4416 | } |
4417 | ||
1400e697 PB |
4418 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4419 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4420 | { |
4421 | struct io_sr_msg *sr = &req->sr_msg; | |
4422 | struct iovec __user *uiov; | |
4423 | size_t iov_len; | |
4424 | int ret; | |
4425 | ||
1400e697 PB |
4426 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4427 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4428 | if (ret) |
4429 | return ret; | |
4430 | ||
4431 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4432 | if (iov_len > 1) | |
4433 | return -EINVAL; | |
5476dfed | 4434 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4435 | return -EFAULT; |
5476dfed | 4436 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4437 | iomsg->free_iov = NULL; |
52de1fe1 | 4438 | } else { |
257e84a5 | 4439 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4440 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4441 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4442 | false); |
52de1fe1 JA |
4443 | if (ret > 0) |
4444 | ret = 0; | |
4445 | } | |
4446 | ||
4447 | return ret; | |
4448 | } | |
4449 | ||
4450 | #ifdef CONFIG_COMPAT | |
4451 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4452 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4453 | { |
4454 | struct compat_msghdr __user *msg_compat; | |
4455 | struct io_sr_msg *sr = &req->sr_msg; | |
4456 | struct compat_iovec __user *uiov; | |
4457 | compat_uptr_t ptr; | |
4458 | compat_size_t len; | |
4459 | int ret; | |
4460 | ||
270a5940 | 4461 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4462 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4463 | &ptr, &len); |
4464 | if (ret) | |
4465 | return ret; | |
4466 | ||
4467 | uiov = compat_ptr(ptr); | |
4468 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4469 | compat_ssize_t clen; | |
4470 | ||
4471 | if (len > 1) | |
4472 | return -EINVAL; | |
4473 | if (!access_ok(uiov, sizeof(*uiov))) | |
4474 | return -EFAULT; | |
4475 | if (__get_user(clen, &uiov->iov_len)) | |
4476 | return -EFAULT; | |
4477 | if (clen < 0) | |
4478 | return -EINVAL; | |
2d280bc8 | 4479 | sr->len = clen; |
257e84a5 | 4480 | iomsg->free_iov = NULL; |
52de1fe1 | 4481 | } else { |
257e84a5 | 4482 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4483 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4484 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4485 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4486 | if (ret < 0) |
4487 | return ret; | |
4488 | } | |
4489 | ||
4490 | return 0; | |
4491 | } | |
4492 | #endif | |
4493 | ||
1400e697 PB |
4494 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4495 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4496 | { |
1400e697 | 4497 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4498 | |
4499 | #ifdef CONFIG_COMPAT | |
4500 | if (req->ctx->compat) | |
1400e697 | 4501 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4502 | #endif |
52de1fe1 | 4503 | |
1400e697 | 4504 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4505 | } |
4506 | ||
bcda7baa | 4507 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4508 | bool needs_lock) |
bcda7baa JA |
4509 | { |
4510 | struct io_sr_msg *sr = &req->sr_msg; | |
4511 | struct io_buffer *kbuf; | |
4512 | ||
bcda7baa JA |
4513 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4514 | if (IS_ERR(kbuf)) | |
4515 | return kbuf; | |
4516 | ||
4517 | sr->kbuf = kbuf; | |
4518 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4519 | return kbuf; |
fddaface JA |
4520 | } |
4521 | ||
7fbb1b54 PB |
4522 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4523 | { | |
4524 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4525 | } | |
4526 | ||
93642ef8 | 4527 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4528 | { |
99bc4c38 | 4529 | int ret; |
3529d8c2 | 4530 | |
93642ef8 PB |
4531 | if (!io_op_defs[req->opcode].needs_async_data) |
4532 | return 0; | |
4533 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4534 | if (!ret) | |
4535 | req->flags |= REQ_F_NEED_CLEANUP; | |
4536 | return ret; | |
4537 | } | |
4538 | ||
4539 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4540 | { | |
4541 | struct io_sr_msg *sr = &req->sr_msg; | |
4542 | ||
d2b6f48b PB |
4543 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4544 | return -EINVAL; | |
4545 | ||
3529d8c2 | 4546 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4547 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4548 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4549 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4550 | |
d8768362 JA |
4551 | #ifdef CONFIG_COMPAT |
4552 | if (req->ctx->compat) | |
4553 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4554 | #endif | |
93642ef8 | 4555 | return 0; |
aa1fa28f JA |
4556 | } |
4557 | ||
889fca73 | 4558 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4559 | { |
6b754c8b | 4560 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4561 | struct socket *sock; |
7fbb1b54 | 4562 | struct io_buffer *kbuf; |
7a7cacba | 4563 | unsigned flags; |
52de1fe1 | 4564 | int ret, cflags = 0; |
45d189c6 | 4565 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4566 | |
dba4a925 | 4567 | sock = sock_from_file(req->file); |
7a7cacba | 4568 | if (unlikely(!sock)) |
dba4a925 | 4569 | return -ENOTSOCK; |
3529d8c2 | 4570 | |
257e84a5 PB |
4571 | kmsg = req->async_data; |
4572 | if (!kmsg) { | |
7a7cacba PB |
4573 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4574 | if (ret) | |
681fda8d | 4575 | return ret; |
7a7cacba PB |
4576 | kmsg = &iomsg; |
4577 | } | |
03b1230c | 4578 | |
bc02ef33 | 4579 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4580 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4581 | if (IS_ERR(kbuf)) |
52de1fe1 | 4582 | return PTR_ERR(kbuf); |
7a7cacba | 4583 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4584 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4585 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4586 | 1, req->sr_msg.len); |
4587 | } | |
52de1fe1 | 4588 | |
7a7cacba PB |
4589 | flags = req->sr_msg.msg_flags; |
4590 | if (flags & MSG_DONTWAIT) | |
4591 | req->flags |= REQ_F_NOWAIT; | |
4592 | else if (force_nonblock) | |
4593 | flags |= MSG_DONTWAIT; | |
e47293fd | 4594 | |
7a7cacba PB |
4595 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4596 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4597 | if (force_nonblock && ret == -EAGAIN) |
4598 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4599 | if (ret == -ERESTARTSYS) |
4600 | ret = -EINTR; | |
03b1230c | 4601 | |
7fbb1b54 PB |
4602 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4603 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4604 | /* fast path, check for non-NULL to avoid function call */ |
4605 | if (kmsg->free_iov) | |
4606 | kfree(kmsg->free_iov); | |
99bc4c38 | 4607 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4608 | if (ret < 0) |
4609 | req_set_fail_links(req); | |
889fca73 | 4610 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4611 | return 0; |
0fa03c62 | 4612 | } |
5d17b4a4 | 4613 | |
889fca73 | 4614 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4615 | { |
6b754c8b | 4616 | struct io_buffer *kbuf; |
7a7cacba PB |
4617 | struct io_sr_msg *sr = &req->sr_msg; |
4618 | struct msghdr msg; | |
4619 | void __user *buf = sr->buf; | |
fddaface | 4620 | struct socket *sock; |
7a7cacba PB |
4621 | struct iovec iov; |
4622 | unsigned flags; | |
bcda7baa | 4623 | int ret, cflags = 0; |
45d189c6 | 4624 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4625 | |
dba4a925 | 4626 | sock = sock_from_file(req->file); |
7a7cacba | 4627 | if (unlikely(!sock)) |
dba4a925 | 4628 | return -ENOTSOCK; |
fddaface | 4629 | |
bc02ef33 | 4630 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4631 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4632 | if (IS_ERR(kbuf)) |
4633 | return PTR_ERR(kbuf); | |
7a7cacba | 4634 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4635 | } |
bcda7baa | 4636 | |
7a7cacba | 4637 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4638 | if (unlikely(ret)) |
4639 | goto out_free; | |
fddaface | 4640 | |
7a7cacba PB |
4641 | msg.msg_name = NULL; |
4642 | msg.msg_control = NULL; | |
4643 | msg.msg_controllen = 0; | |
4644 | msg.msg_namelen = 0; | |
4645 | msg.msg_iocb = NULL; | |
4646 | msg.msg_flags = 0; | |
fddaface | 4647 | |
7a7cacba PB |
4648 | flags = req->sr_msg.msg_flags; |
4649 | if (flags & MSG_DONTWAIT) | |
4650 | req->flags |= REQ_F_NOWAIT; | |
4651 | else if (force_nonblock) | |
4652 | flags |= MSG_DONTWAIT; | |
4653 | ||
4654 | ret = sock_recvmsg(sock, &msg, flags); | |
4655 | if (force_nonblock && ret == -EAGAIN) | |
4656 | return -EAGAIN; | |
4657 | if (ret == -ERESTARTSYS) | |
4658 | ret = -EINTR; | |
14c32eee | 4659 | out_free: |
7fbb1b54 PB |
4660 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4661 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4662 | if (ret < 0) |
4663 | req_set_fail_links(req); | |
889fca73 | 4664 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4665 | return 0; |
fddaface JA |
4666 | } |
4667 | ||
3529d8c2 | 4668 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4669 | { |
8ed8d3c3 JA |
4670 | struct io_accept *accept = &req->accept; |
4671 | ||
14587a46 | 4672 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4673 | return -EINVAL; |
8042d6ce | 4674 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4675 | return -EINVAL; |
4676 | ||
d55e5f5b JA |
4677 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4678 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4679 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4680 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4681 | return 0; |
8ed8d3c3 | 4682 | } |
17f2fe35 | 4683 | |
889fca73 | 4684 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4685 | { |
4686 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4687 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4688 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4689 | int ret; |
4690 | ||
e697deed JX |
4691 | if (req->file->f_flags & O_NONBLOCK) |
4692 | req->flags |= REQ_F_NOWAIT; | |
4693 | ||
8ed8d3c3 | 4694 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4695 | accept->addr_len, accept->flags, |
4696 | accept->nofile); | |
8ed8d3c3 | 4697 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4698 | return -EAGAIN; |
ac45abc0 PB |
4699 | if (ret < 0) { |
4700 | if (ret == -ERESTARTSYS) | |
4701 | ret = -EINTR; | |
4e88d6e7 | 4702 | req_set_fail_links(req); |
ac45abc0 | 4703 | } |
889fca73 | 4704 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4705 | return 0; |
8ed8d3c3 JA |
4706 | } |
4707 | ||
93642ef8 PB |
4708 | static int io_connect_prep_async(struct io_kiocb *req) |
4709 | { | |
4710 | struct io_async_connect *io = req->async_data; | |
4711 | struct io_connect *conn = &req->connect; | |
4712 | ||
4713 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4714 | } | |
4715 | ||
3529d8c2 | 4716 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4717 | { |
3529d8c2 | 4718 | struct io_connect *conn = &req->connect; |
f499a021 | 4719 | |
14587a46 | 4720 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4721 | return -EINVAL; |
4722 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4723 | return -EINVAL; | |
4724 | ||
3529d8c2 JA |
4725 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4726 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4727 | return 0; |
f499a021 JA |
4728 | } |
4729 | ||
889fca73 | 4730 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4731 | { |
e8c2bc1f | 4732 | struct io_async_connect __io, *io; |
f8e85cf2 | 4733 | unsigned file_flags; |
3fbb51c1 | 4734 | int ret; |
45d189c6 | 4735 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4736 | |
e8c2bc1f JA |
4737 | if (req->async_data) { |
4738 | io = req->async_data; | |
f499a021 | 4739 | } else { |
3529d8c2 JA |
4740 | ret = move_addr_to_kernel(req->connect.addr, |
4741 | req->connect.addr_len, | |
e8c2bc1f | 4742 | &__io.address); |
f499a021 JA |
4743 | if (ret) |
4744 | goto out; | |
4745 | io = &__io; | |
4746 | } | |
4747 | ||
3fbb51c1 JA |
4748 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4749 | ||
e8c2bc1f | 4750 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4751 | req->connect.addr_len, file_flags); |
87f80d62 | 4752 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4753 | if (req->async_data) |
b7bb4f7d | 4754 | return -EAGAIN; |
e8c2bc1f | 4755 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4756 | ret = -ENOMEM; |
4757 | goto out; | |
4758 | } | |
e8c2bc1f JA |
4759 | io = req->async_data; |
4760 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4761 | return -EAGAIN; |
f499a021 | 4762 | } |
f8e85cf2 JA |
4763 | if (ret == -ERESTARTSYS) |
4764 | ret = -EINTR; | |
f499a021 | 4765 | out: |
4e88d6e7 JA |
4766 | if (ret < 0) |
4767 | req_set_fail_links(req); | |
889fca73 | 4768 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4769 | return 0; |
469956e8 Y |
4770 | } |
4771 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4772 | #define IO_NETOP_FN(op) \ |
4773 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4774 | { \ | |
4775 | return -EOPNOTSUPP; \ | |
4776 | } | |
4777 | ||
4778 | #define IO_NETOP_PREP(op) \ | |
4779 | IO_NETOP_FN(op) \ | |
4780 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4781 | { \ | |
4782 | return -EOPNOTSUPP; \ | |
4783 | } \ | |
4784 | ||
4785 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4786 | IO_NETOP_PREP(op) \ | |
4787 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4788 | { \ | |
4789 | return -EOPNOTSUPP; \ | |
4790 | } | |
4791 | ||
4792 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4793 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4794 | IO_NETOP_PREP_ASYNC(connect); | |
4795 | IO_NETOP_PREP(accept); | |
4796 | IO_NETOP_FN(send); | |
4797 | IO_NETOP_FN(recv); | |
469956e8 | 4798 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4799 | |
d7718a9d JA |
4800 | struct io_poll_table { |
4801 | struct poll_table_struct pt; | |
4802 | struct io_kiocb *req; | |
4803 | int error; | |
4804 | }; | |
ce593a6c | 4805 | |
d7718a9d JA |
4806 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4807 | __poll_t mask, task_work_func_t func) | |
4808 | { | |
aa96bf8a | 4809 | int ret; |
d7718a9d JA |
4810 | |
4811 | /* for instances that support it check for an event match first: */ | |
4812 | if (mask && !(mask & poll->events)) | |
4813 | return 0; | |
4814 | ||
4815 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4816 | ||
4817 | list_del_init(&poll->wait.entry); | |
4818 | ||
d7718a9d | 4819 | req->result = mask; |
7cbf1722 | 4820 | req->task_work.func = func; |
6d816e08 JA |
4821 | percpu_ref_get(&req->ctx->refs); |
4822 | ||
d7718a9d | 4823 | /* |
e3aabf95 JA |
4824 | * If this fails, then the task is exiting. When a task exits, the |
4825 | * work gets canceled, so just cancel this request as well instead | |
4826 | * of executing it. We can't safely execute it anyway, as we may not | |
4827 | * have the needed state needed for it anyway. | |
d7718a9d | 4828 | */ |
355fb9e2 | 4829 | ret = io_req_task_work_add(req); |
aa96bf8a | 4830 | if (unlikely(ret)) { |
e3aabf95 | 4831 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4832 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4833 | } |
d7718a9d JA |
4834 | return 1; |
4835 | } | |
4836 | ||
74ce6ce4 JA |
4837 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4838 | __acquires(&req->ctx->completion_lock) | |
4839 | { | |
4840 | struct io_ring_ctx *ctx = req->ctx; | |
4841 | ||
4842 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4843 | struct poll_table_struct pt = { ._key = poll->events }; | |
4844 | ||
4845 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4846 | } | |
4847 | ||
4848 | spin_lock_irq(&ctx->completion_lock); | |
4849 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4850 | add_wait_queue(poll->head, &poll->wait); | |
4851 | return true; | |
4852 | } | |
4853 | ||
4854 | return false; | |
4855 | } | |
4856 | ||
d4e7cd36 | 4857 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4858 | { |
e8c2bc1f | 4859 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4860 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4861 | return req->async_data; |
d4e7cd36 JA |
4862 | return req->apoll->double_poll; |
4863 | } | |
4864 | ||
4865 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4866 | { | |
4867 | if (req->opcode == IORING_OP_POLL_ADD) | |
4868 | return &req->poll; | |
4869 | return &req->apoll->poll; | |
4870 | } | |
4871 | ||
4872 | static void io_poll_remove_double(struct io_kiocb *req) | |
4873 | { | |
4874 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4875 | |
4876 | lockdep_assert_held(&req->ctx->completion_lock); | |
4877 | ||
4878 | if (poll && poll->head) { | |
4879 | struct wait_queue_head *head = poll->head; | |
4880 | ||
4881 | spin_lock(&head->lock); | |
4882 | list_del_init(&poll->wait.entry); | |
4883 | if (poll->wait.private) | |
4884 | refcount_dec(&req->refs); | |
4885 | poll->head = NULL; | |
4886 | spin_unlock(&head->lock); | |
4887 | } | |
4888 | } | |
4889 | ||
4890 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4891 | { | |
4892 | struct io_ring_ctx *ctx = req->ctx; | |
4893 | ||
d4e7cd36 | 4894 | io_poll_remove_double(req); |
18bceab1 JA |
4895 | req->poll.done = true; |
4896 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4897 | io_commit_cqring(ctx); | |
4898 | } | |
4899 | ||
dd221f46 | 4900 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4901 | { |
dd221f46 | 4902 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4903 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4904 | struct io_kiocb *nxt; |
18bceab1 JA |
4905 | |
4906 | if (io_poll_rewait(req, &req->poll)) { | |
4907 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4908 | } else { |
4909 | hash_del(&req->hash_node); | |
4910 | io_poll_complete(req, req->result, 0); | |
4911 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4912 | |
dd221f46 PB |
4913 | nxt = io_put_req_find_next(req); |
4914 | io_cqring_ev_posted(ctx); | |
4915 | if (nxt) | |
4916 | __io_req_task_submit(nxt); | |
4917 | } | |
18bceab1 | 4918 | |
6d816e08 | 4919 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4920 | } |
4921 | ||
4922 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4923 | int sync, void *key) | |
4924 | { | |
4925 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4926 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4927 | __poll_t mask = key_to_poll(key); |
4928 | ||
4929 | /* for instances that support it check for an event match first: */ | |
4930 | if (mask && !(mask & poll->events)) | |
4931 | return 0; | |
4932 | ||
8706e04e JA |
4933 | list_del_init(&wait->entry); |
4934 | ||
807abcb0 | 4935 | if (poll && poll->head) { |
18bceab1 JA |
4936 | bool done; |
4937 | ||
807abcb0 JA |
4938 | spin_lock(&poll->head->lock); |
4939 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4940 | if (!done) |
807abcb0 | 4941 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4942 | /* make sure double remove sees this as being gone */ |
4943 | wait->private = NULL; | |
807abcb0 | 4944 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4945 | if (!done) { |
4946 | /* use wait func handler, so it matches the rq type */ | |
4947 | poll->wait.func(&poll->wait, mode, sync, key); | |
4948 | } | |
18bceab1 JA |
4949 | } |
4950 | refcount_dec(&req->refs); | |
4951 | return 1; | |
4952 | } | |
4953 | ||
4954 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4955 | wait_queue_func_t wake_func) | |
4956 | { | |
4957 | poll->head = NULL; | |
4958 | poll->done = false; | |
4959 | poll->canceled = false; | |
4960 | poll->events = events; | |
4961 | INIT_LIST_HEAD(&poll->wait.entry); | |
4962 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4963 | } | |
4964 | ||
4965 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4966 | struct wait_queue_head *head, |
4967 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4968 | { |
4969 | struct io_kiocb *req = pt->req; | |
4970 | ||
4971 | /* | |
4972 | * If poll->head is already set, it's because the file being polled | |
4973 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4974 | * for write). Setup a separate io_poll_iocb if this happens. | |
4975 | */ | |
4976 | if (unlikely(poll->head)) { | |
58852d4d PB |
4977 | struct io_poll_iocb *poll_one = poll; |
4978 | ||
18bceab1 | 4979 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4980 | if (*poll_ptr) { |
18bceab1 JA |
4981 | pt->error = -EINVAL; |
4982 | return; | |
4983 | } | |
1c3b3e65 JA |
4984 | /* double add on the same waitqueue head, ignore */ |
4985 | if (poll->head == head) | |
4986 | return; | |
18bceab1 JA |
4987 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4988 | if (!poll) { | |
4989 | pt->error = -ENOMEM; | |
4990 | return; | |
4991 | } | |
58852d4d | 4992 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4993 | refcount_inc(&req->refs); |
4994 | poll->wait.private = req; | |
807abcb0 | 4995 | *poll_ptr = poll; |
18bceab1 JA |
4996 | } |
4997 | ||
4998 | pt->error = 0; | |
4999 | poll->head = head; | |
a31eb4a2 JX |
5000 | |
5001 | if (poll->events & EPOLLEXCLUSIVE) | |
5002 | add_wait_queue_exclusive(head, &poll->wait); | |
5003 | else | |
5004 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5005 | } |
5006 | ||
5007 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5008 | struct poll_table_struct *p) | |
5009 | { | |
5010 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5011 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5012 | |
807abcb0 | 5013 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5014 | } |
5015 | ||
d7718a9d JA |
5016 | static void io_async_task_func(struct callback_head *cb) |
5017 | { | |
5018 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5019 | struct async_poll *apoll = req->apoll; | |
5020 | struct io_ring_ctx *ctx = req->ctx; | |
5021 | ||
5022 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5023 | ||
74ce6ce4 | 5024 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5025 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5026 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5027 | return; |
d7718a9d JA |
5028 | } |
5029 | ||
31067255 | 5030 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5031 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5032 | hash_del(&req->hash_node); |
2bae047e | 5033 | |
d4e7cd36 | 5034 | io_poll_remove_double(req); |
74ce6ce4 JA |
5035 | spin_unlock_irq(&ctx->completion_lock); |
5036 | ||
0be0b0e3 PB |
5037 | if (!READ_ONCE(apoll->poll.canceled)) |
5038 | __io_req_task_submit(req); | |
5039 | else | |
5040 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5041 | |
6d816e08 | 5042 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5043 | kfree(apoll->double_poll); |
31067255 | 5044 | kfree(apoll); |
d7718a9d JA |
5045 | } |
5046 | ||
5047 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5048 | void *key) | |
5049 | { | |
5050 | struct io_kiocb *req = wait->private; | |
5051 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5052 | ||
5053 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5054 | key_to_poll(key)); | |
5055 | ||
5056 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5057 | } | |
5058 | ||
5059 | static void io_poll_req_insert(struct io_kiocb *req) | |
5060 | { | |
5061 | struct io_ring_ctx *ctx = req->ctx; | |
5062 | struct hlist_head *list; | |
5063 | ||
5064 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5065 | hlist_add_head(&req->hash_node, list); | |
5066 | } | |
5067 | ||
5068 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5069 | struct io_poll_iocb *poll, | |
5070 | struct io_poll_table *ipt, __poll_t mask, | |
5071 | wait_queue_func_t wake_func) | |
5072 | __acquires(&ctx->completion_lock) | |
5073 | { | |
5074 | struct io_ring_ctx *ctx = req->ctx; | |
5075 | bool cancel = false; | |
5076 | ||
4d52f338 | 5077 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5078 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5079 | poll->file = req->file; |
18bceab1 | 5080 | poll->wait.private = req; |
d7718a9d JA |
5081 | |
5082 | ipt->pt._key = mask; | |
5083 | ipt->req = req; | |
5084 | ipt->error = -EINVAL; | |
5085 | ||
d7718a9d JA |
5086 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5087 | ||
5088 | spin_lock_irq(&ctx->completion_lock); | |
5089 | if (likely(poll->head)) { | |
5090 | spin_lock(&poll->head->lock); | |
5091 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5092 | if (ipt->error) | |
5093 | cancel = true; | |
5094 | ipt->error = 0; | |
5095 | mask = 0; | |
5096 | } | |
5097 | if (mask || ipt->error) | |
5098 | list_del_init(&poll->wait.entry); | |
5099 | else if (cancel) | |
5100 | WRITE_ONCE(poll->canceled, true); | |
5101 | else if (!poll->done) /* actually waiting for an event */ | |
5102 | io_poll_req_insert(req); | |
5103 | spin_unlock(&poll->head->lock); | |
5104 | } | |
5105 | ||
5106 | return mask; | |
5107 | } | |
5108 | ||
5109 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5110 | { | |
5111 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5112 | struct io_ring_ctx *ctx = req->ctx; | |
5113 | struct async_poll *apoll; | |
5114 | struct io_poll_table ipt; | |
5115 | __poll_t mask, ret; | |
9dab14b8 | 5116 | int rw; |
d7718a9d JA |
5117 | |
5118 | if (!req->file || !file_can_poll(req->file)) | |
5119 | return false; | |
24c74678 | 5120 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5121 | return false; |
9dab14b8 JA |
5122 | if (def->pollin) |
5123 | rw = READ; | |
5124 | else if (def->pollout) | |
5125 | rw = WRITE; | |
5126 | else | |
5127 | return false; | |
5128 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5129 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5130 | return false; |
5131 | ||
5132 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5133 | if (unlikely(!apoll)) | |
5134 | return false; | |
807abcb0 | 5135 | apoll->double_poll = NULL; |
d7718a9d JA |
5136 | |
5137 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5138 | req->apoll = apoll; |
d7718a9d | 5139 | |
8755d97a | 5140 | mask = 0; |
d7718a9d | 5141 | if (def->pollin) |
8755d97a | 5142 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5143 | if (def->pollout) |
5144 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5145 | |
5146 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5147 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5148 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5149 | mask &= ~POLLIN; | |
5150 | ||
d7718a9d JA |
5151 | mask |= POLLERR | POLLPRI; |
5152 | ||
5153 | ipt.pt._qproc = io_async_queue_proc; | |
5154 | ||
5155 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5156 | io_async_wake); | |
a36da65c | 5157 | if (ret || ipt.error) { |
d4e7cd36 | 5158 | io_poll_remove_double(req); |
d7718a9d | 5159 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5160 | kfree(apoll->double_poll); |
d7718a9d JA |
5161 | kfree(apoll); |
5162 | return false; | |
5163 | } | |
5164 | spin_unlock_irq(&ctx->completion_lock); | |
5165 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5166 | apoll->poll.events); | |
5167 | return true; | |
5168 | } | |
5169 | ||
5170 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5171 | struct io_poll_iocb *poll) | |
221c5eb2 | 5172 | { |
b41e9852 | 5173 | bool do_complete = false; |
221c5eb2 JA |
5174 | |
5175 | spin_lock(&poll->head->lock); | |
5176 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5177 | if (!list_empty(&poll->wait.entry)) { |
5178 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5179 | do_complete = true; |
221c5eb2 JA |
5180 | } |
5181 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5182 | hash_del(&req->hash_node); |
d7718a9d JA |
5183 | return do_complete; |
5184 | } | |
5185 | ||
5186 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5187 | { | |
5188 | bool do_complete; | |
5189 | ||
d4e7cd36 JA |
5190 | io_poll_remove_double(req); |
5191 | ||
d7718a9d JA |
5192 | if (req->opcode == IORING_OP_POLL_ADD) { |
5193 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5194 | } else { | |
3bfa5bcb JA |
5195 | struct async_poll *apoll = req->apoll; |
5196 | ||
d7718a9d | 5197 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5198 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5199 | if (do_complete) { | |
d7718a9d | 5200 | io_put_req(req); |
807abcb0 | 5201 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5202 | kfree(apoll); |
5203 | } | |
b1f573bd XW |
5204 | } |
5205 | ||
b41e9852 JA |
5206 | if (do_complete) { |
5207 | io_cqring_fill_event(req, -ECANCELED); | |
5208 | io_commit_cqring(req->ctx); | |
f254ac04 | 5209 | req_set_fail_links(req); |
216578e5 | 5210 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5211 | } |
5212 | ||
5213 | return do_complete; | |
221c5eb2 JA |
5214 | } |
5215 | ||
76e1b642 JA |
5216 | /* |
5217 | * Returns true if we found and killed one or more poll requests | |
5218 | */ | |
6b81928d PB |
5219 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5220 | struct files_struct *files) | |
221c5eb2 | 5221 | { |
78076bb6 | 5222 | struct hlist_node *tmp; |
221c5eb2 | 5223 | struct io_kiocb *req; |
8e2e1faf | 5224 | int posted = 0, i; |
221c5eb2 JA |
5225 | |
5226 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5227 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5228 | struct hlist_head *list; | |
5229 | ||
5230 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5231 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5232 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5233 | posted += io_poll_remove_one(req); |
5234 | } | |
221c5eb2 JA |
5235 | } |
5236 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5237 | |
8e2e1faf JA |
5238 | if (posted) |
5239 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5240 | |
5241 | return posted != 0; | |
221c5eb2 JA |
5242 | } |
5243 | ||
47f46768 JA |
5244 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5245 | { | |
78076bb6 | 5246 | struct hlist_head *list; |
47f46768 JA |
5247 | struct io_kiocb *req; |
5248 | ||
78076bb6 JA |
5249 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5250 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5251 | if (sqe_addr != req->user_data) |
5252 | continue; | |
5253 | if (io_poll_remove_one(req)) | |
eac406c6 | 5254 | return 0; |
b41e9852 | 5255 | return -EALREADY; |
47f46768 JA |
5256 | } |
5257 | ||
5258 | return -ENOENT; | |
5259 | } | |
5260 | ||
3529d8c2 JA |
5261 | static int io_poll_remove_prep(struct io_kiocb *req, |
5262 | const struct io_uring_sqe *sqe) | |
0969e783 | 5263 | { |
0969e783 JA |
5264 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5265 | return -EINVAL; | |
5266 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5267 | sqe->poll_events) | |
5268 | return -EINVAL; | |
5269 | ||
018043be | 5270 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5271 | return 0; |
5272 | } | |
5273 | ||
221c5eb2 JA |
5274 | /* |
5275 | * Find a running poll command that matches one specified in sqe->addr, | |
5276 | * and remove it if found. | |
5277 | */ | |
61e98203 | 5278 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5279 | { |
5280 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5281 | int ret; |
221c5eb2 | 5282 | |
221c5eb2 | 5283 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5284 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5285 | spin_unlock_irq(&ctx->completion_lock); |
5286 | ||
4e88d6e7 JA |
5287 | if (ret < 0) |
5288 | req_set_fail_links(req); | |
e1e16097 | 5289 | io_req_complete(req, ret); |
221c5eb2 JA |
5290 | return 0; |
5291 | } | |
5292 | ||
221c5eb2 JA |
5293 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5294 | void *key) | |
5295 | { | |
c2f2eb7d JA |
5296 | struct io_kiocb *req = wait->private; |
5297 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5298 | |
d7718a9d | 5299 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5300 | } |
5301 | ||
221c5eb2 JA |
5302 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5303 | struct poll_table_struct *p) | |
5304 | { | |
5305 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5306 | ||
e8c2bc1f | 5307 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5308 | } |
5309 | ||
3529d8c2 | 5310 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5311 | { |
5312 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5313 | u32 events; |
221c5eb2 JA |
5314 | |
5315 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5316 | return -EINVAL; | |
5317 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5318 | return -EINVAL; | |
5319 | ||
5769a351 JX |
5320 | events = READ_ONCE(sqe->poll32_events); |
5321 | #ifdef __BIG_ENDIAN | |
5322 | events = swahw32(events); | |
5323 | #endif | |
a31eb4a2 JX |
5324 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5325 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5326 | return 0; |
5327 | } | |
5328 | ||
61e98203 | 5329 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5330 | { |
5331 | struct io_poll_iocb *poll = &req->poll; | |
5332 | struct io_ring_ctx *ctx = req->ctx; | |
5333 | struct io_poll_table ipt; | |
0969e783 | 5334 | __poll_t mask; |
0969e783 | 5335 | |
d7718a9d | 5336 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5337 | |
d7718a9d JA |
5338 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5339 | io_poll_wake); | |
221c5eb2 | 5340 | |
8c838788 | 5341 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5342 | ipt.error = 0; |
b0dd8a41 | 5343 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5344 | } |
221c5eb2 JA |
5345 | spin_unlock_irq(&ctx->completion_lock); |
5346 | ||
8c838788 JA |
5347 | if (mask) { |
5348 | io_cqring_ev_posted(ctx); | |
014db007 | 5349 | io_put_req(req); |
221c5eb2 | 5350 | } |
8c838788 | 5351 | return ipt.error; |
221c5eb2 JA |
5352 | } |
5353 | ||
5262f567 JA |
5354 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5355 | { | |
ad8a48ac JA |
5356 | struct io_timeout_data *data = container_of(timer, |
5357 | struct io_timeout_data, timer); | |
5358 | struct io_kiocb *req = data->req; | |
5359 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5360 | unsigned long flags; |
5361 | ||
5262f567 | 5362 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5363 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5364 | atomic_set(&req->ctx->cq_timeouts, |
5365 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5366 | ||
78e19bbe | 5367 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5368 | io_commit_cqring(ctx); |
5369 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5370 | ||
5371 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5372 | req_set_fail_links(req); |
5262f567 JA |
5373 | io_put_req(req); |
5374 | return HRTIMER_NORESTART; | |
5375 | } | |
5376 | ||
fbd15848 PB |
5377 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5378 | __u64 user_data) | |
f254ac04 | 5379 | { |
fbd15848 | 5380 | struct io_timeout_data *io; |
47f46768 JA |
5381 | struct io_kiocb *req; |
5382 | int ret = -ENOENT; | |
f254ac04 | 5383 | |
135fcde8 | 5384 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5385 | if (user_data == req->user_data) { |
47f46768 JA |
5386 | ret = 0; |
5387 | break; | |
5388 | } | |
5389 | } | |
5390 | ||
5391 | if (ret == -ENOENT) | |
fbd15848 PB |
5392 | return ERR_PTR(ret); |
5393 | ||
5394 | io = req->async_data; | |
e8c2bc1f | 5395 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5396 | if (ret == -1) |
fbd15848 | 5397 | return ERR_PTR(-EALREADY); |
a71976f3 | 5398 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5399 | return req; |
5400 | } | |
47f46768 | 5401 | |
fbd15848 PB |
5402 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5403 | { | |
5404 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5405 | ||
5406 | if (IS_ERR(req)) | |
5407 | return PTR_ERR(req); | |
f254ac04 JA |
5408 | |
5409 | req_set_fail_links(req); | |
f254ac04 | 5410 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5411 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5412 | return 0; |
5413 | } | |
5414 | ||
9c8e11b3 PB |
5415 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5416 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5417 | { |
9c8e11b3 PB |
5418 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5419 | struct io_timeout_data *data; | |
47f46768 | 5420 | |
9c8e11b3 PB |
5421 | if (IS_ERR(req)) |
5422 | return PTR_ERR(req); | |
47f46768 | 5423 | |
9c8e11b3 PB |
5424 | req->timeout.off = 0; /* noseq */ |
5425 | data = req->async_data; | |
5426 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5427 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5428 | data->timer.function = io_timeout_fn; | |
5429 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5430 | return 0; | |
47f46768 JA |
5431 | } |
5432 | ||
3529d8c2 JA |
5433 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5434 | const struct io_uring_sqe *sqe) | |
b29472ee | 5435 | { |
9c8e11b3 PB |
5436 | struct io_timeout_rem *tr = &req->timeout_rem; |
5437 | ||
b29472ee JA |
5438 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5439 | return -EINVAL; | |
61710e43 DA |
5440 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5441 | return -EINVAL; | |
9c8e11b3 | 5442 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5443 | return -EINVAL; |
5444 | ||
9c8e11b3 PB |
5445 | tr->addr = READ_ONCE(sqe->addr); |
5446 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5447 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5448 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5449 | return -EINVAL; | |
5450 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5451 | return -EFAULT; | |
5452 | } else if (tr->flags) { | |
5453 | /* timeout removal doesn't support flags */ | |
b29472ee | 5454 | return -EINVAL; |
9c8e11b3 | 5455 | } |
b29472ee | 5456 | |
b29472ee JA |
5457 | return 0; |
5458 | } | |
5459 | ||
8662daec PB |
5460 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5461 | { | |
5462 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5463 | : HRTIMER_MODE_REL; | |
5464 | } | |
5465 | ||
11365043 JA |
5466 | /* |
5467 | * Remove or update an existing timeout command | |
5468 | */ | |
61e98203 | 5469 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5470 | { |
9c8e11b3 | 5471 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5472 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5473 | int ret; |
11365043 | 5474 | |
11365043 | 5475 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5476 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5477 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5478 | else |
5479 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5480 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5481 | |
47f46768 | 5482 | io_cqring_fill_event(req, ret); |
11365043 JA |
5483 | io_commit_cqring(ctx); |
5484 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5485 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5486 | if (ret < 0) |
5487 | req_set_fail_links(req); | |
ec9c02ad | 5488 | io_put_req(req); |
11365043 | 5489 | return 0; |
5262f567 JA |
5490 | } |
5491 | ||
3529d8c2 | 5492 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5493 | bool is_timeout_link) |
5262f567 | 5494 | { |
ad8a48ac | 5495 | struct io_timeout_data *data; |
a41525ab | 5496 | unsigned flags; |
56080b02 | 5497 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5498 | |
ad8a48ac | 5499 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5500 | return -EINVAL; |
ad8a48ac | 5501 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5502 | return -EINVAL; |
56080b02 | 5503 | if (off && is_timeout_link) |
2d28390a | 5504 | return -EINVAL; |
a41525ab JA |
5505 | flags = READ_ONCE(sqe->timeout_flags); |
5506 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5507 | return -EINVAL; |
bdf20073 | 5508 | |
bfe68a22 | 5509 | req->timeout.off = off; |
26a61679 | 5510 | |
e8c2bc1f | 5511 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5512 | return -ENOMEM; |
5513 | ||
e8c2bc1f | 5514 | data = req->async_data; |
ad8a48ac | 5515 | data->req = req; |
ad8a48ac JA |
5516 | |
5517 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5518 | return -EFAULT; |
5519 | ||
8662daec | 5520 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5521 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
dd59a3d5 | 5522 | io_req_track_inflight(req); |
ad8a48ac JA |
5523 | return 0; |
5524 | } | |
5525 | ||
61e98203 | 5526 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5527 | { |
ad8a48ac | 5528 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5529 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5530 | struct list_head *entry; |
bfe68a22 | 5531 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5532 | |
733f5c95 | 5533 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5534 | |
5262f567 JA |
5535 | /* |
5536 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5537 | * timeout event to be satisfied. If it isn't set, then this is |
5538 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5539 | */ |
8eb7e2d0 | 5540 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5541 | entry = ctx->timeout_list.prev; |
5542 | goto add; | |
5543 | } | |
5262f567 | 5544 | |
bfe68a22 PB |
5545 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5546 | req->timeout.target_seq = tail + off; | |
5262f567 | 5547 | |
f010505b MDG |
5548 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5549 | * This is safe because ->completion_lock is held, and submissions | |
5550 | * and completions are never mixed in the same ->completion_lock section. | |
5551 | */ | |
5552 | ctx->cq_last_tm_flush = tail; | |
5553 | ||
5262f567 JA |
5554 | /* |
5555 | * Insertion sort, ensuring the first entry in the list is always | |
5556 | * the one we need first. | |
5557 | */ | |
5262f567 | 5558 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5559 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5560 | timeout.list); | |
5262f567 | 5561 | |
8eb7e2d0 | 5562 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5563 | continue; |
bfe68a22 PB |
5564 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5565 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5566 | break; |
5567 | } | |
93bd25bb | 5568 | add: |
135fcde8 | 5569 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5570 | data->timer.function = io_timeout_fn; |
5571 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5572 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5573 | return 0; |
5574 | } | |
5262f567 | 5575 | |
f458dd84 PB |
5576 | struct io_cancel_data { |
5577 | struct io_ring_ctx *ctx; | |
5578 | u64 user_data; | |
5579 | }; | |
5580 | ||
62755e35 JA |
5581 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5582 | { | |
5583 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5584 | struct io_cancel_data *cd = data; |
62755e35 | 5585 | |
f458dd84 | 5586 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5587 | } |
5588 | ||
f458dd84 PB |
5589 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5590 | struct io_ring_ctx *ctx) | |
62755e35 | 5591 | { |
f458dd84 | 5592 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5593 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5594 | int ret = 0; |
5595 | ||
f458dd84 | 5596 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5597 | return -ENOENT; |
5598 | ||
f458dd84 | 5599 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5600 | switch (cancel_ret) { |
5601 | case IO_WQ_CANCEL_OK: | |
5602 | ret = 0; | |
5603 | break; | |
5604 | case IO_WQ_CANCEL_RUNNING: | |
5605 | ret = -EALREADY; | |
5606 | break; | |
5607 | case IO_WQ_CANCEL_NOTFOUND: | |
5608 | ret = -ENOENT; | |
5609 | break; | |
5610 | } | |
5611 | ||
e977d6d3 JA |
5612 | return ret; |
5613 | } | |
5614 | ||
47f46768 JA |
5615 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5616 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5617 | int success_ret) |
47f46768 JA |
5618 | { |
5619 | unsigned long flags; | |
5620 | int ret; | |
5621 | ||
f458dd84 | 5622 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5623 | if (ret != -ENOENT) { |
5624 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5625 | goto done; | |
5626 | } | |
5627 | ||
5628 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5629 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5630 | if (ret != -ENOENT) | |
5631 | goto done; | |
5632 | ret = io_poll_cancel(ctx, sqe_addr); | |
5633 | done: | |
b0dd8a41 JA |
5634 | if (!ret) |
5635 | ret = success_ret; | |
47f46768 JA |
5636 | io_cqring_fill_event(req, ret); |
5637 | io_commit_cqring(ctx); | |
5638 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5639 | io_cqring_ev_posted(ctx); | |
5640 | ||
4e88d6e7 JA |
5641 | if (ret < 0) |
5642 | req_set_fail_links(req); | |
014db007 | 5643 | io_put_req(req); |
47f46768 JA |
5644 | } |
5645 | ||
3529d8c2 JA |
5646 | static int io_async_cancel_prep(struct io_kiocb *req, |
5647 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5648 | { |
fbf23849 | 5649 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5650 | return -EINVAL; |
61710e43 DA |
5651 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5652 | return -EINVAL; | |
5653 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5654 | return -EINVAL; |
5655 | ||
fbf23849 JA |
5656 | req->cancel.addr = READ_ONCE(sqe->addr); |
5657 | return 0; | |
5658 | } | |
5659 | ||
61e98203 | 5660 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5661 | { |
5662 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5663 | u64 sqe_addr = req->cancel.addr; |
5664 | struct io_tctx_node *node; | |
5665 | int ret; | |
5666 | ||
5667 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5668 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5669 | spin_lock_irq(&ctx->completion_lock); | |
5670 | if (ret != -ENOENT) | |
5671 | goto done; | |
5672 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5673 | if (ret != -ENOENT) | |
5674 | goto done; | |
5675 | ret = io_poll_cancel(ctx, sqe_addr); | |
5676 | if (ret != -ENOENT) | |
5677 | goto done; | |
5678 | spin_unlock_irq(&ctx->completion_lock); | |
5679 | ||
5680 | /* slow path, try all io-wq's */ | |
5681 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5682 | ret = -ENOENT; | |
5683 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5684 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5685 | |
58f99373 PB |
5686 | if (!tctx || !tctx->io_wq) |
5687 | continue; | |
5688 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5689 | if (ret != -ENOENT) | |
5690 | break; | |
5691 | } | |
5692 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5693 | ||
5694 | spin_lock_irq(&ctx->completion_lock); | |
5695 | done: | |
5696 | io_cqring_fill_event(req, ret); | |
5697 | io_commit_cqring(ctx); | |
5698 | spin_unlock_irq(&ctx->completion_lock); | |
5699 | io_cqring_ev_posted(ctx); | |
5700 | ||
5701 | if (ret < 0) | |
5702 | req_set_fail_links(req); | |
5703 | io_put_req(req); | |
5262f567 JA |
5704 | return 0; |
5705 | } | |
5706 | ||
269bbe5f | 5707 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5708 | const struct io_uring_sqe *sqe) |
5709 | { | |
6ca56f84 JA |
5710 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5711 | return -EINVAL; | |
61710e43 DA |
5712 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5713 | return -EINVAL; | |
5714 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5715 | return -EINVAL; |
5716 | ||
269bbe5f BM |
5717 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5718 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5719 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5720 | return -EINVAL; |
269bbe5f | 5721 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5722 | return 0; |
5723 | } | |
5724 | ||
889fca73 | 5725 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5726 | { |
5727 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5728 | struct io_uring_rsrc_update up; |
05f3fb3c | 5729 | int ret; |
fbf23849 | 5730 | |
45d189c6 | 5731 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5732 | return -EAGAIN; |
05f3fb3c | 5733 | |
269bbe5f BM |
5734 | up.offset = req->rsrc_update.offset; |
5735 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5736 | |
5737 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5738 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5739 | mutex_unlock(&ctx->uring_lock); |
5740 | ||
5741 | if (ret < 0) | |
5742 | req_set_fail_links(req); | |
889fca73 | 5743 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5744 | return 0; |
5745 | } | |
5746 | ||
bfe76559 | 5747 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5748 | { |
d625c6ee | 5749 | switch (req->opcode) { |
e781573e | 5750 | case IORING_OP_NOP: |
bfe76559 | 5751 | return 0; |
f67676d1 JA |
5752 | case IORING_OP_READV: |
5753 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5754 | case IORING_OP_READ: |
bfe76559 | 5755 | return io_read_prep(req, sqe); |
f67676d1 JA |
5756 | case IORING_OP_WRITEV: |
5757 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5758 | case IORING_OP_WRITE: |
bfe76559 | 5759 | return io_write_prep(req, sqe); |
0969e783 | 5760 | case IORING_OP_POLL_ADD: |
bfe76559 | 5761 | return io_poll_add_prep(req, sqe); |
0969e783 | 5762 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5763 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5764 | case IORING_OP_FSYNC: |
1155c76a | 5765 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5766 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5767 | return io_sfr_prep(req, sqe); |
03b1230c | 5768 | case IORING_OP_SENDMSG: |
fddaface | 5769 | case IORING_OP_SEND: |
bfe76559 | 5770 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5771 | case IORING_OP_RECVMSG: |
fddaface | 5772 | case IORING_OP_RECV: |
bfe76559 | 5773 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5774 | case IORING_OP_CONNECT: |
bfe76559 | 5775 | return io_connect_prep(req, sqe); |
2d28390a | 5776 | case IORING_OP_TIMEOUT: |
bfe76559 | 5777 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5778 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5779 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5780 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5781 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5782 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5783 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5784 | case IORING_OP_ACCEPT: |
bfe76559 | 5785 | return io_accept_prep(req, sqe); |
d63d1b5e | 5786 | case IORING_OP_FALLOCATE: |
bfe76559 | 5787 | return io_fallocate_prep(req, sqe); |
15b71abe | 5788 | case IORING_OP_OPENAT: |
bfe76559 | 5789 | return io_openat_prep(req, sqe); |
b5dba59e | 5790 | case IORING_OP_CLOSE: |
bfe76559 | 5791 | return io_close_prep(req, sqe); |
05f3fb3c | 5792 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5793 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5794 | case IORING_OP_STATX: |
bfe76559 | 5795 | return io_statx_prep(req, sqe); |
4840e418 | 5796 | case IORING_OP_FADVISE: |
bfe76559 | 5797 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5798 | case IORING_OP_MADVISE: |
bfe76559 | 5799 | return io_madvise_prep(req, sqe); |
cebdb986 | 5800 | case IORING_OP_OPENAT2: |
bfe76559 | 5801 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5802 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5803 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5804 | case IORING_OP_SPLICE: |
bfe76559 | 5805 | return io_splice_prep(req, sqe); |
ddf0322d | 5806 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5807 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5808 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5809 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5810 | case IORING_OP_TEE: |
bfe76559 | 5811 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5812 | case IORING_OP_SHUTDOWN: |
5813 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5814 | case IORING_OP_RENAMEAT: |
5815 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5816 | case IORING_OP_UNLINKAT: |
5817 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5818 | } |
5819 | ||
bfe76559 PB |
5820 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5821 | req->opcode); | |
5822 | return-EINVAL; | |
5823 | } | |
5824 | ||
93642ef8 | 5825 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5826 | { |
93642ef8 PB |
5827 | switch (req->opcode) { |
5828 | case IORING_OP_READV: | |
5829 | case IORING_OP_READ_FIXED: | |
5830 | case IORING_OP_READ: | |
5831 | return io_rw_prep_async(req, READ); | |
5832 | case IORING_OP_WRITEV: | |
5833 | case IORING_OP_WRITE_FIXED: | |
5834 | case IORING_OP_WRITE: | |
5835 | return io_rw_prep_async(req, WRITE); | |
5836 | case IORING_OP_SENDMSG: | |
5837 | case IORING_OP_SEND: | |
5838 | return io_sendmsg_prep_async(req); | |
5839 | case IORING_OP_RECVMSG: | |
5840 | case IORING_OP_RECV: | |
5841 | return io_recvmsg_prep_async(req); | |
5842 | case IORING_OP_CONNECT: | |
5843 | return io_connect_prep_async(req); | |
5844 | } | |
5845 | return 0; | |
5846 | } | |
5847 | ||
be7053b7 | 5848 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5849 | { |
be7053b7 | 5850 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5851 | return 0; |
be7053b7 | 5852 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5853 | if (req->async_data) |
be7053b7 PB |
5854 | return 0; |
5855 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5856 | return -EAGAIN; |
be7053b7 | 5857 | return io_req_prep_async(req); |
f67676d1 JA |
5858 | } |
5859 | ||
9cf7c104 PB |
5860 | static u32 io_get_sequence(struct io_kiocb *req) |
5861 | { | |
5862 | struct io_kiocb *pos; | |
5863 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5864 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5865 | |
f2f87370 PB |
5866 | io_for_each_link(pos, req) |
5867 | nr_reqs++; | |
9cf7c104 PB |
5868 | |
5869 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5870 | return total_submitted - nr_reqs; | |
5871 | } | |
5872 | ||
be7053b7 | 5873 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5874 | { |
a197f664 | 5875 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5876 | struct io_defer_entry *de; |
f67676d1 | 5877 | int ret; |
9cf7c104 | 5878 | u32 seq; |
de0617e4 | 5879 | |
9d858b21 | 5880 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5881 | if (likely(list_empty_careful(&ctx->defer_list) && |
5882 | !(req->flags & REQ_F_IO_DRAIN))) | |
5883 | return 0; | |
5884 | ||
5885 | seq = io_get_sequence(req); | |
5886 | /* Still a chance to pass the sequence check */ | |
5887 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5888 | return 0; |
5889 | ||
be7053b7 PB |
5890 | ret = io_req_defer_prep(req); |
5891 | if (ret) | |
5892 | return ret; | |
cbdcb435 | 5893 | io_prep_async_link(req); |
27dc8338 PB |
5894 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5895 | if (!de) | |
5896 | return -ENOMEM; | |
2d28390a | 5897 | |
de0617e4 | 5898 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5899 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5900 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5901 | kfree(de); |
ae34817b PB |
5902 | io_queue_async_work(req); |
5903 | return -EIOCBQUEUED; | |
de0617e4 JA |
5904 | } |
5905 | ||
915967f6 | 5906 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5907 | de->req = req; |
9cf7c104 | 5908 | de->seq = seq; |
27dc8338 | 5909 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5910 | spin_unlock_irq(&ctx->completion_lock); |
5911 | return -EIOCBQUEUED; | |
5912 | } | |
5913 | ||
3ca405eb | 5914 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5915 | { |
0e1b6fe3 PB |
5916 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5917 | switch (req->opcode) { | |
5918 | case IORING_OP_READV: | |
5919 | case IORING_OP_READ_FIXED: | |
5920 | case IORING_OP_READ: | |
bcda7baa | 5921 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5922 | break; |
5923 | case IORING_OP_RECVMSG: | |
5924 | case IORING_OP_RECV: | |
bcda7baa | 5925 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5926 | break; |
5927 | } | |
5928 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5929 | } |
5930 | ||
0e1b6fe3 PB |
5931 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5932 | switch (req->opcode) { | |
5933 | case IORING_OP_READV: | |
5934 | case IORING_OP_READ_FIXED: | |
5935 | case IORING_OP_READ: | |
5936 | case IORING_OP_WRITEV: | |
5937 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5938 | case IORING_OP_WRITE: { |
5939 | struct io_async_rw *io = req->async_data; | |
5940 | if (io->free_iovec) | |
5941 | kfree(io->free_iovec); | |
0e1b6fe3 | 5942 | break; |
e8c2bc1f | 5943 | } |
0e1b6fe3 | 5944 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5945 | case IORING_OP_SENDMSG: { |
5946 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5947 | |
5948 | kfree(io->free_iov); | |
0e1b6fe3 | 5949 | break; |
e8c2bc1f | 5950 | } |
0e1b6fe3 PB |
5951 | case IORING_OP_SPLICE: |
5952 | case IORING_OP_TEE: | |
5953 | io_put_file(req, req->splice.file_in, | |
5954 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5955 | break; | |
f3cd4850 JA |
5956 | case IORING_OP_OPENAT: |
5957 | case IORING_OP_OPENAT2: | |
5958 | if (req->open.filename) | |
5959 | putname(req->open.filename); | |
5960 | break; | |
80a261fd JA |
5961 | case IORING_OP_RENAMEAT: |
5962 | putname(req->rename.oldpath); | |
5963 | putname(req->rename.newpath); | |
5964 | break; | |
14a1143b JA |
5965 | case IORING_OP_UNLINKAT: |
5966 | putname(req->unlink.filename); | |
5967 | break; | |
0e1b6fe3 PB |
5968 | } |
5969 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5970 | } |
99bc4c38 PB |
5971 | } |
5972 | ||
889fca73 | 5973 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5974 | { |
a197f664 | 5975 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5976 | const struct cred *creds = NULL; |
d625c6ee | 5977 | int ret; |
2b188cc1 | 5978 | |
003e8dcc JA |
5979 | if (req->work.creds && req->work.creds != current_cred()) |
5980 | creds = override_creds(req->work.creds); | |
5730b27e | 5981 | |
d625c6ee | 5982 | switch (req->opcode) { |
2b188cc1 | 5983 | case IORING_OP_NOP: |
889fca73 | 5984 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5985 | break; |
5986 | case IORING_OP_READV: | |
edafccee | 5987 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5988 | case IORING_OP_READ: |
889fca73 | 5989 | ret = io_read(req, issue_flags); |
edafccee | 5990 | break; |
3529d8c2 | 5991 | case IORING_OP_WRITEV: |
edafccee | 5992 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5993 | case IORING_OP_WRITE: |
889fca73 | 5994 | ret = io_write(req, issue_flags); |
2b188cc1 | 5995 | break; |
c992fe29 | 5996 | case IORING_OP_FSYNC: |
45d189c6 | 5997 | ret = io_fsync(req, issue_flags); |
c992fe29 | 5998 | break; |
221c5eb2 | 5999 | case IORING_OP_POLL_ADD: |
61e98203 | 6000 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6001 | break; |
6002 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6003 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6004 | break; |
5d17b4a4 | 6005 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6006 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6007 | break; |
0fa03c62 | 6008 | case IORING_OP_SENDMSG: |
889fca73 | 6009 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6010 | break; |
fddaface | 6011 | case IORING_OP_SEND: |
889fca73 | 6012 | ret = io_send(req, issue_flags); |
0fa03c62 | 6013 | break; |
aa1fa28f | 6014 | case IORING_OP_RECVMSG: |
889fca73 | 6015 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6016 | break; |
fddaface | 6017 | case IORING_OP_RECV: |
889fca73 | 6018 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6019 | break; |
5262f567 | 6020 | case IORING_OP_TIMEOUT: |
61e98203 | 6021 | ret = io_timeout(req, issue_flags); |
5262f567 | 6022 | break; |
11365043 | 6023 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6024 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6025 | break; |
17f2fe35 | 6026 | case IORING_OP_ACCEPT: |
889fca73 | 6027 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6028 | break; |
f8e85cf2 | 6029 | case IORING_OP_CONNECT: |
889fca73 | 6030 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6031 | break; |
62755e35 | 6032 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6033 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6034 | break; |
d63d1b5e | 6035 | case IORING_OP_FALLOCATE: |
45d189c6 | 6036 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6037 | break; |
15b71abe | 6038 | case IORING_OP_OPENAT: |
45d189c6 | 6039 | ret = io_openat(req, issue_flags); |
15b71abe | 6040 | break; |
b5dba59e | 6041 | case IORING_OP_CLOSE: |
889fca73 | 6042 | ret = io_close(req, issue_flags); |
b5dba59e | 6043 | break; |
05f3fb3c | 6044 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6045 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6046 | break; |
eddc7ef5 | 6047 | case IORING_OP_STATX: |
45d189c6 | 6048 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6049 | break; |
4840e418 | 6050 | case IORING_OP_FADVISE: |
45d189c6 | 6051 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6052 | break; |
c1ca757b | 6053 | case IORING_OP_MADVISE: |
45d189c6 | 6054 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6055 | break; |
cebdb986 | 6056 | case IORING_OP_OPENAT2: |
45d189c6 | 6057 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6058 | break; |
3e4827b0 | 6059 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6060 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6061 | break; |
7d67af2c | 6062 | case IORING_OP_SPLICE: |
45d189c6 | 6063 | ret = io_splice(req, issue_flags); |
7d67af2c | 6064 | break; |
ddf0322d | 6065 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6066 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6067 | break; |
067524e9 | 6068 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6069 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6070 | break; |
f2a8d5c7 | 6071 | case IORING_OP_TEE: |
45d189c6 | 6072 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6073 | break; |
36f4fa68 | 6074 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6075 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6076 | break; |
80a261fd | 6077 | case IORING_OP_RENAMEAT: |
45d189c6 | 6078 | ret = io_renameat(req, issue_flags); |
80a261fd | 6079 | break; |
14a1143b | 6080 | case IORING_OP_UNLINKAT: |
45d189c6 | 6081 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6082 | break; |
2b188cc1 JA |
6083 | default: |
6084 | ret = -EINVAL; | |
6085 | break; | |
6086 | } | |
6087 | ||
5730b27e JA |
6088 | if (creds) |
6089 | revert_creds(creds); | |
6090 | ||
def596e9 JA |
6091 | if (ret) |
6092 | return ret; | |
6093 | ||
b532576e JA |
6094 | /* If the op doesn't have a file, we're not polling for it */ |
6095 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6096 | const bool in_async = io_wq_current_is_worker(); |
6097 | ||
11ba820b JA |
6098 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6099 | if (in_async) | |
6100 | mutex_lock(&ctx->uring_lock); | |
6101 | ||
2e9dbe90 | 6102 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6103 | |
6104 | if (in_async) | |
6105 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6106 | } |
6107 | ||
6108 | return 0; | |
2b188cc1 JA |
6109 | } |
6110 | ||
5280f7e5 | 6111 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6112 | { |
6113 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6114 | struct io_kiocb *timeout; |
561fb04a | 6115 | int ret = 0; |
2b188cc1 | 6116 | |
6df1db6b PB |
6117 | timeout = io_prep_linked_timeout(req); |
6118 | if (timeout) | |
6119 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6120 | |
4014d943 | 6121 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6122 | ret = -ECANCELED; |
31b51510 | 6123 | |
561fb04a | 6124 | if (!ret) { |
561fb04a | 6125 | do { |
889fca73 | 6126 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6127 | /* |
6128 | * We can get EAGAIN for polled IO even though we're | |
6129 | * forcing a sync submission from here, since we can't | |
6130 | * wait for request slots on the block side. | |
6131 | */ | |
6132 | if (ret != -EAGAIN) | |
6133 | break; | |
6134 | cond_resched(); | |
6135 | } while (1); | |
6136 | } | |
31b51510 | 6137 | |
a3df7698 | 6138 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6139 | if (ret) { |
a3df7698 PB |
6140 | /* io-wq is going to take one down */ |
6141 | refcount_inc(&req->refs); | |
6142 | io_req_task_queue_fail(req, ret); | |
edafccee | 6143 | } |
2b188cc1 JA |
6144 | } |
6145 | ||
65e19f54 JA |
6146 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6147 | int index) | |
6148 | { | |
269bbe5f | 6149 | struct fixed_rsrc_table *table; |
65e19f54 | 6150 | |
05f3fb3c | 6151 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6152 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6153 | } |
6154 | ||
8371adf5 PB |
6155 | static struct file *io_file_get(struct io_submit_state *state, |
6156 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6157 | { |
a197f664 | 6158 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6159 | struct file *file; |
09bb8394 | 6160 | |
8da11c19 | 6161 | if (fixed) { |
479f517b | 6162 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6163 | return NULL; |
b7620121 | 6164 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6165 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6166 | io_set_resource_node(req); |
09bb8394 | 6167 | } else { |
c826bd7a | 6168 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6169 | file = __io_file_get(state, fd); |
09bb8394 JA |
6170 | } |
6171 | ||
ce3d5aae PB |
6172 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6173 | io_req_track_inflight(req); | |
8371adf5 | 6174 | return file; |
09bb8394 JA |
6175 | } |
6176 | ||
2665abfd | 6177 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6178 | { |
ad8a48ac JA |
6179 | struct io_timeout_data *data = container_of(timer, |
6180 | struct io_timeout_data, timer); | |
90cd7e42 | 6181 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6182 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6183 | unsigned long flags; |
2665abfd JA |
6184 | |
6185 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6186 | prev = req->timeout.head; |
6187 | req->timeout.head = NULL; | |
2665abfd JA |
6188 | |
6189 | /* | |
6190 | * We don't expect the list to be empty, that will only happen if we | |
6191 | * race with the completion of the linked work. | |
6192 | */ | |
90cd7e42 | 6193 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6194 | io_remove_next_linked(prev); |
90cd7e42 PB |
6195 | else |
6196 | prev = NULL; | |
2665abfd JA |
6197 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6198 | ||
6199 | if (prev) { | |
014db007 | 6200 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6201 | io_put_req_deferred(prev, 1); |
47f46768 | 6202 | } else { |
9ae1f8dd PB |
6203 | io_req_complete_post(req, -ETIME, 0); |
6204 | io_put_req_deferred(req, 1); | |
2665abfd | 6205 | } |
2665abfd JA |
6206 | return HRTIMER_NORESTART; |
6207 | } | |
6208 | ||
7271ef3a | 6209 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6210 | { |
76a46e06 | 6211 | /* |
f2f87370 PB |
6212 | * If the back reference is NULL, then our linked request finished |
6213 | * before we got a chance to setup the timer | |
76a46e06 | 6214 | */ |
90cd7e42 | 6215 | if (req->timeout.head) { |
e8c2bc1f | 6216 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6217 | |
ad8a48ac JA |
6218 | data->timer.function = io_link_timeout_fn; |
6219 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6220 | data->mode); | |
2665abfd | 6221 | } |
7271ef3a JA |
6222 | } |
6223 | ||
6224 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6225 | { | |
6226 | struct io_ring_ctx *ctx = req->ctx; | |
6227 | ||
6228 | spin_lock_irq(&ctx->completion_lock); | |
6229 | __io_queue_linked_timeout(req); | |
76a46e06 | 6230 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6231 | |
2665abfd | 6232 | /* drop submission reference */ |
76a46e06 JA |
6233 | io_put_req(req); |
6234 | } | |
2665abfd | 6235 | |
ad8a48ac | 6236 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6237 | { |
f2f87370 | 6238 | struct io_kiocb *nxt = req->link; |
2665abfd | 6239 | |
f2f87370 PB |
6240 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6241 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6242 | return NULL; |
2665abfd | 6243 | |
90cd7e42 | 6244 | nxt->timeout.head = req; |
900fad45 | 6245 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6246 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6247 | return nxt; |
2665abfd JA |
6248 | } |
6249 | ||
c5eef2b9 | 6250 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6251 | { |
d3d7298d | 6252 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6253 | int ret; |
2b188cc1 | 6254 | |
c5eef2b9 | 6255 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6256 | |
491381ce JA |
6257 | /* |
6258 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6259 | * doesn't support non-blocking read/write attempts | |
6260 | */ | |
24c74678 | 6261 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6262 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6263 | /* |
6264 | * Queued up for async execution, worker will release | |
6265 | * submit reference when the iocb is actually submitted. | |
6266 | */ | |
6267 | io_queue_async_work(req); | |
2b188cc1 | 6268 | } |
0d63c148 PB |
6269 | } else if (likely(!ret)) { |
6270 | /* drop submission reference */ | |
e342c807 | 6271 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6272 | struct io_ring_ctx *ctx = req->ctx; |
6273 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6274 | |
6dd0be1e | 6275 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6276 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6277 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6278 | } else { |
d3d7298d | 6279 | io_put_req(req); |
0d63c148 PB |
6280 | } |
6281 | } else { | |
4e88d6e7 | 6282 | req_set_fail_links(req); |
e65ef56d | 6283 | io_put_req(req); |
e1e16097 | 6284 | io_req_complete(req, ret); |
9e645e11 | 6285 | } |
d3d7298d PB |
6286 | if (linked_timeout) |
6287 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6288 | } |
6289 | ||
be7053b7 | 6290 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6291 | { |
6292 | int ret; | |
6293 | ||
be7053b7 | 6294 | ret = io_req_defer(req); |
4fe2c963 JL |
6295 | if (ret) { |
6296 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6297 | fail_req: |
4e88d6e7 | 6298 | req_set_fail_links(req); |
e1e16097 JA |
6299 | io_put_req(req); |
6300 | io_req_complete(req, ret); | |
4fe2c963 | 6301 | } |
2550878f | 6302 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6303 | ret = io_req_defer_prep(req); |
6304 | if (unlikely(ret)) | |
6305 | goto fail_req; | |
ce35a47a JA |
6306 | io_queue_async_work(req); |
6307 | } else { | |
c5eef2b9 | 6308 | __io_queue_sqe(req); |
ce35a47a | 6309 | } |
4fe2c963 JL |
6310 | } |
6311 | ||
b16fed66 PB |
6312 | /* |
6313 | * Check SQE restrictions (opcode and flags). | |
6314 | * | |
6315 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6316 | */ | |
6317 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6318 | struct io_kiocb *req, | |
6319 | unsigned int sqe_flags) | |
4fe2c963 | 6320 | { |
b16fed66 PB |
6321 | if (!ctx->restricted) |
6322 | return true; | |
6323 | ||
6324 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6325 | return false; | |
6326 | ||
6327 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6328 | ctx->restrictions.sqe_flags_required) | |
6329 | return false; | |
6330 | ||
6331 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6332 | ctx->restrictions.sqe_flags_required)) | |
6333 | return false; | |
6334 | ||
6335 | return true; | |
4fe2c963 JL |
6336 | } |
6337 | ||
b16fed66 PB |
6338 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6339 | const struct io_uring_sqe *sqe) | |
6340 | { | |
6341 | struct io_submit_state *state; | |
6342 | unsigned int sqe_flags; | |
003e8dcc | 6343 | int personality, ret = 0; |
b16fed66 PB |
6344 | |
6345 | req->opcode = READ_ONCE(sqe->opcode); | |
6346 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6347 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6348 | req->user_data = READ_ONCE(sqe->user_data); | |
6349 | req->async_data = NULL; | |
6350 | req->file = NULL; | |
6351 | req->ctx = ctx; | |
6352 | req->link = NULL; | |
6353 | req->fixed_rsrc_refs = NULL; | |
6354 | /* one is dropped after submission, the other at completion */ | |
6355 | refcount_set(&req->refs, 2); | |
6356 | req->task = current; | |
6357 | req->result = 0; | |
93e68e03 JA |
6358 | req->work.list.next = NULL; |
6359 | req->work.creds = NULL; | |
6360 | req->work.flags = 0; | |
b16fed66 PB |
6361 | |
6362 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6363 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6364 | req->flags = 0; | |
b16fed66 | 6365 | return -EINVAL; |
ebf4a5db | 6366 | } |
b16fed66 PB |
6367 | |
6368 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6369 | return -EINVAL; | |
6370 | ||
b16fed66 PB |
6371 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6372 | return -EACCES; | |
6373 | ||
6374 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6375 | !io_op_defs[req->opcode].buffer_select) | |
6376 | return -EOPNOTSUPP; | |
863e0560 | 6377 | |
003e8dcc JA |
6378 | personality = READ_ONCE(sqe->personality); |
6379 | if (personality) { | |
61cf9370 | 6380 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6381 | if (!req->work.creds) |
6382 | return -EINVAL; | |
6383 | get_cred(req->work.creds); | |
003e8dcc | 6384 | } |
b16fed66 PB |
6385 | state = &ctx->submit_state; |
6386 | ||
6387 | /* | |
6388 | * Plug now if we have more than 1 IO left after this, and the target | |
6389 | * is potentially a read/write to block based storage. | |
6390 | */ | |
6391 | if (!state->plug_started && state->ios_left > 1 && | |
6392 | io_op_defs[req->opcode].plug) { | |
6393 | blk_start_plug(&state->plug); | |
6394 | state->plug_started = true; | |
6395 | } | |
6396 | ||
6397 | if (io_op_defs[req->opcode].needs_file) { | |
6398 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6399 | ||
6400 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6401 | if (unlikely(!req->file)) | |
6402 | ret = -EBADF; | |
6403 | } | |
6404 | ||
6405 | state->ios_left--; | |
6406 | return ret; | |
6407 | } | |
6408 | ||
a6b8cadc | 6409 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6410 | const struct io_uring_sqe *sqe) |
9e645e11 | 6411 | { |
a1ab7b35 | 6412 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6413 | int ret; |
9e645e11 | 6414 | |
a6b8cadc PB |
6415 | ret = io_init_req(ctx, req, sqe); |
6416 | if (unlikely(ret)) { | |
6417 | fail_req: | |
6418 | io_put_req(req); | |
6419 | io_req_complete(req, ret); | |
de59bc10 PB |
6420 | if (link->head) { |
6421 | /* fail even hard links since we don't submit */ | |
cf109604 | 6422 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6423 | io_put_req(link->head); |
6424 | io_req_complete(link->head, -ECANCELED); | |
6425 | link->head = NULL; | |
6426 | } | |
a6b8cadc PB |
6427 | return ret; |
6428 | } | |
be7053b7 PB |
6429 | ret = io_req_prep(req, sqe); |
6430 | if (unlikely(ret)) | |
6431 | goto fail_req; | |
a6b8cadc | 6432 | |
be7053b7 | 6433 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6434 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6435 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6436 | ||
9e645e11 JA |
6437 | /* |
6438 | * If we already have a head request, queue this one for async | |
6439 | * submittal once the head completes. If we don't have a head but | |
6440 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6441 | * submitted sync once the chain is complete. If none of those | |
6442 | * conditions are true (normal request), then just queue it. | |
6443 | */ | |
863e0560 PB |
6444 | if (link->head) { |
6445 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6446 | |
8cdf2193 PB |
6447 | /* |
6448 | * Taking sequential execution of a link, draining both sides | |
6449 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6450 | * requests in the link. So, it drains the head and the | |
6451 | * next after the link request. The last one is done via | |
6452 | * drain_next flag to persist the effect across calls. | |
6453 | */ | |
ef4ff581 | 6454 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6455 | head->flags |= REQ_F_IO_DRAIN; |
6456 | ctx->drain_next = 1; | |
6457 | } | |
be7053b7 | 6458 | ret = io_req_defer_prep(req); |
cf109604 | 6459 | if (unlikely(ret)) |
a6b8cadc | 6460 | goto fail_req; |
9d76377f | 6461 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6462 | link->last->link = req; |
863e0560 | 6463 | link->last = req; |
32fe525b PB |
6464 | |
6465 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6466 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6467 | io_queue_sqe(head); |
863e0560 | 6468 | link->head = NULL; |
32fe525b | 6469 | } |
9e645e11 | 6470 | } else { |
711be031 PB |
6471 | if (unlikely(ctx->drain_next)) { |
6472 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6473 | ctx->drain_next = 0; |
711be031 | 6474 | } |
ef4ff581 | 6475 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6476 | link->head = req; |
6477 | link->last = req; | |
711be031 | 6478 | } else { |
be7053b7 | 6479 | io_queue_sqe(req); |
711be031 | 6480 | } |
9e645e11 | 6481 | } |
2e6e1fde | 6482 | |
1d4240cc | 6483 | return 0; |
9e645e11 JA |
6484 | } |
6485 | ||
9a56a232 JA |
6486 | /* |
6487 | * Batched submission is done, ensure local IO is flushed out. | |
6488 | */ | |
ba88ff11 PB |
6489 | static void io_submit_state_end(struct io_submit_state *state, |
6490 | struct io_ring_ctx *ctx) | |
9a56a232 | 6491 | { |
a1ab7b35 | 6492 | if (state->link.head) |
de59bc10 | 6493 | io_queue_sqe(state->link.head); |
6dd0be1e | 6494 | if (state->comp.nr) |
ba88ff11 | 6495 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6496 | if (state->plug_started) |
6497 | blk_finish_plug(&state->plug); | |
9f13c35b | 6498 | io_state_file_put(state); |
9a56a232 JA |
6499 | } |
6500 | ||
6501 | /* | |
6502 | * Start submission side cache. | |
6503 | */ | |
6504 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6505 | unsigned int max_ios) |
9a56a232 | 6506 | { |
27926b68 | 6507 | state->plug_started = false; |
9a56a232 | 6508 | state->ios_left = max_ios; |
a1ab7b35 PB |
6509 | /* set only head, no need to init link_last in advance */ |
6510 | state->link.head = NULL; | |
9a56a232 JA |
6511 | } |
6512 | ||
2b188cc1 JA |
6513 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6514 | { | |
75b28aff | 6515 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6516 | |
caf582c6 PB |
6517 | /* |
6518 | * Ensure any loads from the SQEs are done at this point, | |
6519 | * since once we write the new head, the application could | |
6520 | * write new data to them. | |
6521 | */ | |
6522 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6523 | } |
6524 | ||
2b188cc1 | 6525 | /* |
3529d8c2 | 6526 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6527 | * that is mapped by userspace. This means that care needs to be taken to |
6528 | * ensure that reads are stable, as we cannot rely on userspace always | |
6529 | * being a good citizen. If members of the sqe are validated and then later | |
6530 | * used, it's important that those reads are done through READ_ONCE() to | |
6531 | * prevent a re-load down the line. | |
6532 | */ | |
709b302f | 6533 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6534 | { |
75b28aff | 6535 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6536 | unsigned head; |
6537 | ||
6538 | /* | |
6539 | * The cached sq head (or cq tail) serves two purposes: | |
6540 | * | |
6541 | * 1) allows us to batch the cost of updating the user visible | |
6542 | * head updates. | |
6543 | * 2) allows the kernel side to track the head on its own, even | |
6544 | * though the application is the one updating it. | |
6545 | */ | |
4fccfcbb | 6546 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6547 | if (likely(head < ctx->sq_entries)) |
6548 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6549 | |
6550 | /* drop invalid entries */ | |
498ccd9e | 6551 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6552 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6553 | return NULL; |
6554 | } | |
6555 | ||
0f212204 | 6556 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6557 | { |
46c4e16a | 6558 | int submitted = 0; |
6c271ce2 | 6559 | |
c4a2ed72 | 6560 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6561 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6562 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6563 | return -EBUSY; |
6564 | } | |
6c271ce2 | 6565 | |
ee7d46d9 PB |
6566 | /* make sure SQ entry isn't read before tail */ |
6567 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6568 | |
2b85edfc PB |
6569 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6570 | return -EAGAIN; | |
6c271ce2 | 6571 | |
d8a6df10 | 6572 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6573 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6574 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6575 | |
46c4e16a | 6576 | while (submitted < nr) { |
3529d8c2 | 6577 | const struct io_uring_sqe *sqe; |
196be95c | 6578 | struct io_kiocb *req; |
fb5ccc98 | 6579 | |
258b29a9 | 6580 | req = io_alloc_req(ctx); |
196be95c PB |
6581 | if (unlikely(!req)) { |
6582 | if (!submitted) | |
6583 | submitted = -EAGAIN; | |
fb5ccc98 | 6584 | break; |
196be95c | 6585 | } |
4fccfcbb PB |
6586 | sqe = io_get_sqe(ctx); |
6587 | if (unlikely(!sqe)) { | |
6588 | kmem_cache_free(req_cachep, req); | |
6589 | break; | |
6590 | } | |
d3656344 JA |
6591 | /* will complete beyond this point, count as submitted */ |
6592 | submitted++; | |
a1ab7b35 | 6593 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6594 | break; |
6c271ce2 JA |
6595 | } |
6596 | ||
9466f437 PB |
6597 | if (unlikely(submitted != nr)) { |
6598 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6599 | struct io_uring_task *tctx = current->io_uring; |
6600 | int unused = nr - ref_used; | |
9466f437 | 6601 | |
d8a6df10 JA |
6602 | percpu_ref_put_many(&ctx->refs, unused); |
6603 | percpu_counter_sub(&tctx->inflight, unused); | |
6604 | put_task_struct_many(current, unused); | |
9466f437 | 6605 | } |
6c271ce2 | 6606 | |
a1ab7b35 | 6607 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6608 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6609 | io_commit_sqring(ctx); | |
6610 | ||
6c271ce2 JA |
6611 | return submitted; |
6612 | } | |
6613 | ||
23b3628e XW |
6614 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6615 | { | |
6616 | /* Tell userspace we may need a wakeup call */ | |
6617 | spin_lock_irq(&ctx->completion_lock); | |
6618 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6619 | spin_unlock_irq(&ctx->completion_lock); | |
6620 | } | |
6621 | ||
6622 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6623 | { | |
6624 | spin_lock_irq(&ctx->completion_lock); | |
6625 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6626 | spin_unlock_irq(&ctx->completion_lock); | |
6627 | } | |
6628 | ||
08369246 | 6629 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6630 | { |
c8d1ba58 | 6631 | unsigned int to_submit; |
bdcd3eab | 6632 | int ret = 0; |
6c271ce2 | 6633 | |
c8d1ba58 | 6634 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6635 | /* if we're handling multiple rings, cap submit size for fairness */ |
6636 | if (cap_entries && to_submit > 8) | |
6637 | to_submit = 8; | |
6638 | ||
906a3c6f | 6639 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6640 | unsigned nr_events = 0; |
a4c0b3de | 6641 | |
c8d1ba58 | 6642 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6643 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6644 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6645 | |
0298ef96 PB |
6646 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6647 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6648 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6649 | mutex_unlock(&ctx->uring_lock); |
6650 | } | |
6c271ce2 | 6651 | |
90554200 JA |
6652 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6653 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6654 | |
08369246 XW |
6655 | return ret; |
6656 | } | |
6c271ce2 | 6657 | |
08369246 XW |
6658 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6659 | { | |
6660 | struct io_ring_ctx *ctx; | |
6661 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6662 | |
08369246 XW |
6663 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6664 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6665 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6666 | } |
c1edbf5f | 6667 | |
08369246 | 6668 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6669 | } |
6c271ce2 | 6670 | |
c8d1ba58 JA |
6671 | static int io_sq_thread(void *data) |
6672 | { | |
69fb2131 JA |
6673 | struct io_sq_data *sqd = data; |
6674 | struct io_ring_ctx *ctx; | |
a0d9205f | 6675 | unsigned long timeout = 0; |
37d1e2e3 | 6676 | char buf[TASK_COMM_LEN]; |
08369246 | 6677 | DEFINE_WAIT(wait); |
6c271ce2 | 6678 | |
37d1e2e3 JA |
6679 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6680 | set_task_comm(current, buf); | |
37d1e2e3 JA |
6681 | current->pf_io_worker = NULL; |
6682 | ||
6683 | if (sqd->sq_cpu != -1) | |
6684 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6685 | else | |
6686 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6687 | current->flags |= PF_NO_SETAFFINITY; | |
6688 | ||
05962f95 JA |
6689 | down_read(&sqd->rw_lock); |
6690 | ||
6691 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { | |
08369246 XW |
6692 | int ret; |
6693 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6694 | |
05962f95 JA |
6695 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
6696 | up_read(&sqd->rw_lock); | |
6697 | cond_resched(); | |
6698 | down_read(&sqd->rw_lock); | |
521d6a73 | 6699 | io_run_task_work(); |
08369246 | 6700 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6701 | continue; |
08369246 | 6702 | } |
37d1e2e3 JA |
6703 | if (fatal_signal_pending(current)) |
6704 | break; | |
08369246 | 6705 | sqt_spin = false; |
e95eee2d | 6706 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6707 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6708 | const struct cred *creds = NULL; |
6709 | ||
6710 | if (ctx->sq_creds != current_cred()) | |
6711 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6712 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6713 | if (creds) |
6714 | revert_creds(creds); | |
08369246 XW |
6715 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6716 | sqt_spin = true; | |
69fb2131 | 6717 | } |
6c271ce2 | 6718 | |
08369246 | 6719 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6720 | io_run_task_work(); |
6721 | cond_resched(); | |
08369246 XW |
6722 | if (sqt_spin) |
6723 | timeout = jiffies + sqd->sq_thread_idle; | |
6724 | continue; | |
6725 | } | |
6726 | ||
08369246 XW |
6727 | needs_sched = true; |
6728 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6729 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6730 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6731 | !list_empty_careful(&ctx->iopoll_list)) { | |
6732 | needs_sched = false; | |
6733 | break; | |
6734 | } | |
6735 | if (io_sqring_entries(ctx)) { | |
6736 | needs_sched = false; | |
6737 | break; | |
6738 | } | |
6739 | } | |
6740 | ||
05962f95 | 6741 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6742 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6743 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6744 | |
05962f95 | 6745 | up_read(&sqd->rw_lock); |
69fb2131 | 6746 | schedule(); |
16efa4fc | 6747 | try_to_freeze(); |
05962f95 | 6748 | down_read(&sqd->rw_lock); |
69fb2131 JA |
6749 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6750 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6751 | } |
08369246 XW |
6752 | |
6753 | finish_wait(&sqd->wait, &wait); | |
6754 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 | 6755 | } |
05962f95 | 6756 | up_read(&sqd->rw_lock); |
521d6a73 PB |
6757 | down_write(&sqd->rw_lock); |
6758 | /* | |
6759 | * someone may have parked and added a cancellation task_work, run | |
6760 | * it first because we don't want it in io_uring_cancel_sqpoll() | |
6761 | */ | |
37d1e2e3 | 6762 | io_run_task_work(); |
28cea78a | 6763 | |
521d6a73 PB |
6764 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6765 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6766 | sqd->thread = NULL; |
05962f95 | 6767 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6768 | io_ring_set_wakeup_flag(ctx); |
05962f95 | 6769 | up_write(&sqd->rw_lock); |
521d6a73 PB |
6770 | |
6771 | io_run_task_work(); | |
37d1e2e3 JA |
6772 | complete(&sqd->exited); |
6773 | do_exit(0); | |
6c271ce2 JA |
6774 | } |
6775 | ||
bda52162 JA |
6776 | struct io_wait_queue { |
6777 | struct wait_queue_entry wq; | |
6778 | struct io_ring_ctx *ctx; | |
6779 | unsigned to_wait; | |
6780 | unsigned nr_timeouts; | |
6781 | }; | |
6782 | ||
6c503150 | 6783 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6784 | { |
6785 | struct io_ring_ctx *ctx = iowq->ctx; | |
6786 | ||
6787 | /* | |
d195a66e | 6788 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6789 | * started waiting. For timeouts, we always want to return to userspace, |
6790 | * regardless of event count. | |
6791 | */ | |
6c503150 | 6792 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6793 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6794 | } | |
6795 | ||
6796 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6797 | int wake_flags, void *key) | |
6798 | { | |
6799 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6800 | wq); | |
6801 | ||
6c503150 PB |
6802 | /* |
6803 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6804 | * the task, and the next invocation will do it. | |
6805 | */ | |
6806 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6807 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6808 | return -1; | |
bda52162 JA |
6809 | } |
6810 | ||
af9c1a44 JA |
6811 | static int io_run_task_work_sig(void) |
6812 | { | |
6813 | if (io_run_task_work()) | |
6814 | return 1; | |
6815 | if (!signal_pending(current)) | |
6816 | return 0; | |
792ee0f6 JA |
6817 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6818 | return -ERESTARTSYS; | |
af9c1a44 JA |
6819 | return -EINTR; |
6820 | } | |
6821 | ||
eeb60b9a PB |
6822 | /* when returns >0, the caller should retry */ |
6823 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6824 | struct io_wait_queue *iowq, | |
6825 | signed long *timeout) | |
6826 | { | |
6827 | int ret; | |
6828 | ||
6829 | /* make sure we run task_work before checking for signals */ | |
6830 | ret = io_run_task_work_sig(); | |
6831 | if (ret || io_should_wake(iowq)) | |
6832 | return ret; | |
6833 | /* let the caller flush overflows, retry */ | |
6834 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6835 | return 1; | |
6836 | ||
6837 | *timeout = schedule_timeout(*timeout); | |
6838 | return !*timeout ? -ETIME : 1; | |
6839 | } | |
6840 | ||
2b188cc1 JA |
6841 | /* |
6842 | * Wait until events become available, if we don't already have some. The | |
6843 | * application must reap them itself, as they reside on the shared cq ring. | |
6844 | */ | |
6845 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6846 | const sigset_t __user *sig, size_t sigsz, |
6847 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6848 | { |
bda52162 JA |
6849 | struct io_wait_queue iowq = { |
6850 | .wq = { | |
6851 | .private = current, | |
6852 | .func = io_wake_function, | |
6853 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6854 | }, | |
6855 | .ctx = ctx, | |
6856 | .to_wait = min_events, | |
6857 | }; | |
75b28aff | 6858 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6859 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6860 | int ret; | |
2b188cc1 | 6861 | |
b41e9852 | 6862 | do { |
6c503150 PB |
6863 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6864 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6865 | return 0; |
4c6e277c | 6866 | if (!io_run_task_work()) |
b41e9852 | 6867 | break; |
b41e9852 | 6868 | } while (1); |
2b188cc1 JA |
6869 | |
6870 | if (sig) { | |
9e75ad5d AB |
6871 | #ifdef CONFIG_COMPAT |
6872 | if (in_compat_syscall()) | |
6873 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6874 | sigsz); |
9e75ad5d AB |
6875 | else |
6876 | #endif | |
b772434b | 6877 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6878 | |
2b188cc1 JA |
6879 | if (ret) |
6880 | return ret; | |
6881 | } | |
6882 | ||
c73ebb68 | 6883 | if (uts) { |
c1d5a224 PB |
6884 | struct timespec64 ts; |
6885 | ||
c73ebb68 HX |
6886 | if (get_timespec64(&ts, uts)) |
6887 | return -EFAULT; | |
6888 | timeout = timespec64_to_jiffies(&ts); | |
6889 | } | |
6890 | ||
bda52162 | 6891 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6892 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6893 | do { |
ca0a2651 JA |
6894 | /* if we can't even flush overflow, don't wait for more */ |
6895 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6896 | ret = -EBUSY; | |
6897 | break; | |
6898 | } | |
bda52162 JA |
6899 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6900 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6901 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6902 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6903 | cond_resched(); |
eeb60b9a | 6904 | } while (ret > 0); |
bda52162 | 6905 | |
b7db41c9 | 6906 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6907 | |
75b28aff | 6908 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6909 | } |
6910 | ||
6b06314c JA |
6911 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6912 | { | |
6913 | #if defined(CONFIG_UNIX) | |
6914 | if (ctx->ring_sock) { | |
6915 | struct sock *sock = ctx->ring_sock->sk; | |
6916 | struct sk_buff *skb; | |
6917 | ||
6918 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6919 | kfree_skb(skb); | |
6920 | } | |
6921 | #else | |
6922 | int i; | |
6923 | ||
65e19f54 JA |
6924 | for (i = 0; i < ctx->nr_user_files; i++) { |
6925 | struct file *file; | |
6926 | ||
6927 | file = io_file_from_index(ctx, i); | |
6928 | if (file) | |
6929 | fput(file); | |
6930 | } | |
6b06314c JA |
6931 | #endif |
6932 | } | |
6933 | ||
00835dce | 6934 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6935 | { |
269bbe5f | 6936 | struct fixed_rsrc_data *data; |
05f3fb3c | 6937 | |
269bbe5f | 6938 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6939 | complete(&data->done); |
6940 | } | |
6941 | ||
2a63b2d9 | 6942 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6943 | { |
2a63b2d9 | 6944 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6945 | } |
6946 | ||
2a63b2d9 | 6947 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6948 | { |
2a63b2d9 BM |
6949 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6950 | } | |
65e19f54 | 6951 | |
d67d2263 BM |
6952 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6953 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6954 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6955 | { |
2a63b2d9 | 6956 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6957 | rsrc_data->node = ref_node; |
d67d2263 | 6958 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6959 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6960 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6961 | } |
6962 | ||
8bad28d8 | 6963 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6964 | { |
8bad28d8 | 6965 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6966 | |
2a63b2d9 | 6967 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6968 | ref_node = data->node; |
e6cb007c | 6969 | data->node = NULL; |
2a63b2d9 | 6970 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6971 | if (ref_node) |
6972 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6973 | } |
6974 | ||
6975 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6976 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6977 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6978 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6979 | { |
f2303b1f | 6980 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6981 | int ret; |
05589553 | 6982 | |
8bad28d8 HX |
6983 | if (data->quiesce) |
6984 | return -ENXIO; | |
05589553 | 6985 | |
8bad28d8 | 6986 | data->quiesce = true; |
1ffc5422 | 6987 | do { |
f2303b1f PB |
6988 | ret = -ENOMEM; |
6989 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
6990 | if (!backup_node) | |
6991 | break; | |
6992 | backup_node->rsrc_data = data; | |
6993 | backup_node->rsrc_put = rsrc_put; | |
6994 | ||
8bad28d8 HX |
6995 | io_sqe_rsrc_kill_node(ctx, data); |
6996 | percpu_ref_kill(&data->refs); | |
6997 | flush_delayed_work(&ctx->rsrc_put_work); | |
6998 | ||
1ffc5422 PB |
6999 | ret = wait_for_completion_interruptible(&data->done); |
7000 | if (!ret) | |
7001 | break; | |
8bad28d8 | 7002 | |
cb5e1b81 | 7003 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7004 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7005 | backup_node = NULL; | |
cb5e1b81 | 7006 | reinit_completion(&data->done); |
8bad28d8 | 7007 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7008 | ret = io_run_task_work_sig(); |
8bad28d8 | 7009 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7010 | } while (ret >= 0); |
8bad28d8 | 7011 | data->quiesce = false; |
05f3fb3c | 7012 | |
8bad28d8 HX |
7013 | if (backup_node) |
7014 | destroy_fixed_rsrc_ref_node(backup_node); | |
7015 | return ret; | |
d7954b2b BM |
7016 | } |
7017 | ||
1ad555c6 BM |
7018 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7019 | { | |
7020 | struct fixed_rsrc_data *data; | |
7021 | ||
7022 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7023 | if (!data) | |
7024 | return NULL; | |
7025 | ||
00835dce | 7026 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7027 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7028 | kfree(data); | |
7029 | return NULL; | |
7030 | } | |
7031 | data->ctx = ctx; | |
7032 | init_completion(&data->done); | |
7033 | return data; | |
7034 | } | |
7035 | ||
7036 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7037 | { | |
7038 | percpu_ref_exit(&data->refs); | |
7039 | kfree(data->table); | |
7040 | kfree(data); | |
7041 | } | |
7042 | ||
d7954b2b BM |
7043 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7044 | { | |
7045 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7046 | unsigned nr_tables, i; |
7047 | int ret; | |
7048 | ||
8bad28d8 HX |
7049 | /* |
7050 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7051 | * Since we possibly drop uring lock later in this function to | |
7052 | * run task work. | |
7053 | */ | |
7054 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7055 | return -ENXIO; |
f2303b1f | 7056 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7057 | if (ret) |
7058 | return ret; | |
7059 | ||
6b06314c | 7060 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7061 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7062 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7063 | kfree(data->table[i].files); |
1ad555c6 | 7064 | free_fixed_rsrc_data(data); |
05f3fb3c | 7065 | ctx->file_data = NULL; |
6b06314c JA |
7066 | ctx->nr_user_files = 0; |
7067 | return 0; | |
7068 | } | |
7069 | ||
37d1e2e3 | 7070 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
05962f95 | 7071 | __releases(&sqd->rw_lock) |
37d1e2e3 | 7072 | { |
521d6a73 PB |
7073 | WARN_ON_ONCE(sqd->thread == current); |
7074 | ||
37d1e2e3 | 7075 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
05962f95 | 7076 | up_write(&sqd->rw_lock); |
37d1e2e3 JA |
7077 | } |
7078 | ||
86e0d676 | 7079 | static void io_sq_thread_park(struct io_sq_data *sqd) |
05962f95 | 7080 | __acquires(&sqd->rw_lock) |
37d1e2e3 | 7081 | { |
521d6a73 PB |
7082 | WARN_ON_ONCE(sqd->thread == current); |
7083 | ||
86e0d676 | 7084 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
05962f95 JA |
7085 | down_write(&sqd->rw_lock); |
7086 | /* set again for consistency, in case concurrent parks are happening */ | |
7087 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
7088 | if (sqd->thread) | |
86e0d676 | 7089 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7090 | } |
7091 | ||
7092 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7093 | { | |
521d6a73 PB |
7094 | WARN_ON_ONCE(sqd->thread == current); |
7095 | ||
05962f95 | 7096 | down_write(&sqd->rw_lock); |
05962f95 | 7097 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7098 | if (sqd->thread) |
7099 | wake_up_process(sqd->thread); | |
05962f95 JA |
7100 | up_write(&sqd->rw_lock); |
7101 | wait_for_completion(&sqd->exited); | |
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) { | |
05962f95 | 7117 | io_sq_thread_park(sqd); |
521d6a73 | 7118 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7119 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7120 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7121 | |
7122 | io_put_sq_data(sqd); | |
7123 | ctx->sq_data = NULL; | |
7c30f36a SM |
7124 | if (ctx->sq_creds) |
7125 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7126 | } |
7127 | } | |
7128 | ||
aa06165d JA |
7129 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7130 | { | |
7131 | struct io_ring_ctx *ctx_attach; | |
7132 | struct io_sq_data *sqd; | |
7133 | struct fd f; | |
7134 | ||
7135 | f = fdget(p->wq_fd); | |
7136 | if (!f.file) | |
7137 | return ERR_PTR(-ENXIO); | |
7138 | if (f.file->f_op != &io_uring_fops) { | |
7139 | fdput(f); | |
7140 | return ERR_PTR(-EINVAL); | |
7141 | } | |
7142 | ||
7143 | ctx_attach = f.file->private_data; | |
7144 | sqd = ctx_attach->sq_data; | |
7145 | if (!sqd) { | |
7146 | fdput(f); | |
7147 | return ERR_PTR(-EINVAL); | |
7148 | } | |
5c2469e0 JA |
7149 | if (sqd->task_tgid != current->tgid) { |
7150 | fdput(f); | |
7151 | return ERR_PTR(-EPERM); | |
7152 | } | |
aa06165d JA |
7153 | |
7154 | refcount_inc(&sqd->refs); | |
7155 | fdput(f); | |
7156 | return sqd; | |
7157 | } | |
7158 | ||
26984fbf PB |
7159 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7160 | bool *attached) | |
534ca6d6 JA |
7161 | { |
7162 | struct io_sq_data *sqd; | |
7163 | ||
26984fbf | 7164 | *attached = false; |
5c2469e0 JA |
7165 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7166 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7167 | if (!IS_ERR(sqd)) { |
7168 | *attached = true; | |
5c2469e0 | 7169 | return sqd; |
26984fbf | 7170 | } |
5c2469e0 JA |
7171 | /* fall through for EPERM case, setup new sqd/task */ |
7172 | if (PTR_ERR(sqd) != -EPERM) | |
7173 | return sqd; | |
7174 | } | |
aa06165d | 7175 | |
534ca6d6 JA |
7176 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7177 | if (!sqd) | |
7178 | return ERR_PTR(-ENOMEM); | |
7179 | ||
7180 | refcount_set(&sqd->refs, 1); | |
69fb2131 | 7181 | INIT_LIST_HEAD(&sqd->ctx_list); |
05962f95 | 7182 | init_rwsem(&sqd->rw_lock); |
534ca6d6 | 7183 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7184 | init_completion(&sqd->exited); |
534ca6d6 JA |
7185 | return sqd; |
7186 | } | |
7187 | ||
6b06314c | 7188 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7189 | /* |
7190 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7191 | * the io_uring can be safely unregistered on process exit, even if we have | |
7192 | * loops in the file referencing. | |
7193 | */ | |
7194 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7195 | { | |
7196 | struct sock *sk = ctx->ring_sock->sk; | |
7197 | struct scm_fp_list *fpl; | |
7198 | struct sk_buff *skb; | |
08a45173 | 7199 | int i, nr_files; |
6b06314c | 7200 | |
6b06314c JA |
7201 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7202 | if (!fpl) | |
7203 | return -ENOMEM; | |
7204 | ||
7205 | skb = alloc_skb(0, GFP_KERNEL); | |
7206 | if (!skb) { | |
7207 | kfree(fpl); | |
7208 | return -ENOMEM; | |
7209 | } | |
7210 | ||
7211 | skb->sk = sk; | |
6b06314c | 7212 | |
08a45173 | 7213 | nr_files = 0; |
62e398be | 7214 | fpl->user = get_uid(current_user()); |
6b06314c | 7215 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7216 | struct file *file = io_file_from_index(ctx, i + offset); |
7217 | ||
7218 | if (!file) | |
08a45173 | 7219 | continue; |
65e19f54 | 7220 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7221 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7222 | nr_files++; | |
6b06314c JA |
7223 | } |
7224 | ||
08a45173 JA |
7225 | if (nr_files) { |
7226 | fpl->max = SCM_MAX_FD; | |
7227 | fpl->count = nr_files; | |
7228 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7229 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7230 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7231 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7232 | |
08a45173 JA |
7233 | for (i = 0; i < nr_files; i++) |
7234 | fput(fpl->fp[i]); | |
7235 | } else { | |
7236 | kfree_skb(skb); | |
7237 | kfree(fpl); | |
7238 | } | |
6b06314c JA |
7239 | |
7240 | return 0; | |
7241 | } | |
7242 | ||
7243 | /* | |
7244 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7245 | * causes regular reference counting to break down. We rely on the UNIX | |
7246 | * garbage collection to take care of this problem for us. | |
7247 | */ | |
7248 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7249 | { | |
7250 | unsigned left, total; | |
7251 | int ret = 0; | |
7252 | ||
7253 | total = 0; | |
7254 | left = ctx->nr_user_files; | |
7255 | while (left) { | |
7256 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7257 | |
7258 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7259 | if (ret) | |
7260 | break; | |
7261 | left -= this_files; | |
7262 | total += this_files; | |
7263 | } | |
7264 | ||
7265 | if (!ret) | |
7266 | return 0; | |
7267 | ||
7268 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7269 | struct file *file = io_file_from_index(ctx, total); |
7270 | ||
7271 | if (file) | |
7272 | fput(file); | |
6b06314c JA |
7273 | total++; |
7274 | } | |
7275 | ||
7276 | return ret; | |
7277 | } | |
7278 | #else | |
7279 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7280 | { | |
7281 | return 0; | |
7282 | } | |
7283 | #endif | |
7284 | ||
269bbe5f | 7285 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7286 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7287 | { |
7288 | int i; | |
7289 | ||
7290 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7291 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7292 | unsigned this_files; |
7293 | ||
7294 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7295 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7296 | GFP_KERNEL); | |
7297 | if (!table->files) | |
7298 | break; | |
7299 | nr_files -= this_files; | |
7300 | } | |
7301 | ||
7302 | if (i == nr_tables) | |
7303 | return 0; | |
7304 | ||
7305 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7306 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7307 | kfree(table->files); |
7308 | } | |
7309 | return 1; | |
7310 | } | |
7311 | ||
50238531 | 7312 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7313 | { |
50238531 | 7314 | struct file *file = prsrc->file; |
05f3fb3c JA |
7315 | #if defined(CONFIG_UNIX) |
7316 | struct sock *sock = ctx->ring_sock->sk; | |
7317 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7318 | struct sk_buff *skb; | |
7319 | int i; | |
7320 | ||
7321 | __skb_queue_head_init(&list); | |
7322 | ||
7323 | /* | |
7324 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7325 | * remove this entry and rearrange the file array. | |
7326 | */ | |
7327 | skb = skb_dequeue(head); | |
7328 | while (skb) { | |
7329 | struct scm_fp_list *fp; | |
7330 | ||
7331 | fp = UNIXCB(skb).fp; | |
7332 | for (i = 0; i < fp->count; i++) { | |
7333 | int left; | |
7334 | ||
7335 | if (fp->fp[i] != file) | |
7336 | continue; | |
7337 | ||
7338 | unix_notinflight(fp->user, fp->fp[i]); | |
7339 | left = fp->count - 1 - i; | |
7340 | if (left) { | |
7341 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7342 | left * sizeof(struct file *)); | |
7343 | } | |
7344 | fp->count--; | |
7345 | if (!fp->count) { | |
7346 | kfree_skb(skb); | |
7347 | skb = NULL; | |
7348 | } else { | |
7349 | __skb_queue_tail(&list, skb); | |
7350 | } | |
7351 | fput(file); | |
7352 | file = NULL; | |
7353 | break; | |
7354 | } | |
7355 | ||
7356 | if (!file) | |
7357 | break; | |
7358 | ||
7359 | __skb_queue_tail(&list, skb); | |
7360 | ||
7361 | skb = skb_dequeue(head); | |
7362 | } | |
7363 | ||
7364 | if (skb_peek(&list)) { | |
7365 | spin_lock_irq(&head->lock); | |
7366 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7367 | __skb_queue_tail(head, skb); | |
7368 | spin_unlock_irq(&head->lock); | |
7369 | } | |
7370 | #else | |
7371 | fput(file); | |
7372 | #endif | |
7373 | } | |
7374 | ||
269bbe5f | 7375 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7376 | { |
269bbe5f BM |
7377 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7378 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7379 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7380 | |
269bbe5f BM |
7381 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7382 | list_del(&prsrc->list); | |
50238531 | 7383 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7384 | kfree(prsrc); |
65e19f54 | 7385 | } |
05589553 | 7386 | |
05589553 XW |
7387 | percpu_ref_exit(&ref_node->refs); |
7388 | kfree(ref_node); | |
269bbe5f | 7389 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7390 | } |
65e19f54 | 7391 | |
269bbe5f | 7392 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7393 | { |
7394 | struct io_ring_ctx *ctx; | |
7395 | struct llist_node *node; | |
7396 | ||
269bbe5f BM |
7397 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7398 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7399 | |
7400 | while (node) { | |
269bbe5f | 7401 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7402 | struct llist_node *next = node->next; |
7403 | ||
269bbe5f BM |
7404 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7405 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7406 | node = next; |
7407 | } | |
7408 | } | |
7409 | ||
ea64ec02 PB |
7410 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7411 | unsigned i) | |
2faf852d | 7412 | { |
ea64ec02 PB |
7413 | struct fixed_rsrc_table *table; |
7414 | ||
7415 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7416 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7417 | } | |
7418 | ||
00835dce | 7419 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7420 | { |
269bbe5f BM |
7421 | struct fixed_rsrc_ref_node *ref_node; |
7422 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7423 | struct io_ring_ctx *ctx; |
e297822b | 7424 | bool first_add = false; |
4a38aed2 | 7425 | int delay = HZ; |
65e19f54 | 7426 | |
269bbe5f BM |
7427 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7428 | data = ref_node->rsrc_data; | |
e297822b PB |
7429 | ctx = data->ctx; |
7430 | ||
2a63b2d9 | 7431 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7432 | ref_node->done = true; |
7433 | ||
d67d2263 BM |
7434 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7435 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7436 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7437 | /* recycle ref nodes in order */ |
7438 | if (!ref_node->done) | |
7439 | break; | |
7440 | list_del(&ref_node->node); | |
269bbe5f | 7441 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7442 | } |
2a63b2d9 | 7443 | io_rsrc_ref_unlock(ctx); |
05589553 | 7444 | |
e297822b | 7445 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7446 | delay = 0; |
05589553 | 7447 | |
4a38aed2 | 7448 | if (!delay) |
269bbe5f | 7449 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7450 | else if (first_add) |
269bbe5f | 7451 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7452 | } |
65e19f54 | 7453 | |
6802535d | 7454 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7455 | struct io_ring_ctx *ctx) |
05f3fb3c | 7456 | { |
269bbe5f | 7457 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7458 | |
05589553 XW |
7459 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7460 | if (!ref_node) | |
3e2224c5 | 7461 | return NULL; |
05f3fb3c | 7462 | |
00835dce | 7463 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7464 | 0, GFP_KERNEL)) { |
7465 | kfree(ref_node); | |
3e2224c5 | 7466 | return NULL; |
05589553 XW |
7467 | } |
7468 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7469 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7470 | ref_node->done = false; |
05589553 | 7471 | return ref_node; |
05589553 XW |
7472 | } |
7473 | ||
bc9744cd PB |
7474 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7475 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7476 | { |
269bbe5f | 7477 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7478 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7479 | } |
7480 | ||
269bbe5f | 7481 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7482 | { |
7483 | percpu_ref_exit(&ref_node->refs); | |
7484 | kfree(ref_node); | |
65e19f54 JA |
7485 | } |
7486 | ||
ea64ec02 | 7487 | |
6b06314c JA |
7488 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7489 | unsigned nr_args) | |
7490 | { | |
7491 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7492 | unsigned nr_tables, i; |
05f3fb3c | 7493 | struct file *file; |
600cf3f8 | 7494 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7495 | struct fixed_rsrc_ref_node *ref_node; |
7496 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7497 | |
05f3fb3c | 7498 | if (ctx->file_data) |
6b06314c JA |
7499 | return -EBUSY; |
7500 | if (!nr_args) | |
7501 | return -EINVAL; | |
7502 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7503 | return -EMFILE; | |
7504 | ||
1ad555c6 | 7505 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7506 | if (!file_data) |
05f3fb3c | 7507 | return -ENOMEM; |
13770a71 | 7508 | ctx->file_data = file_data; |
05f3fb3c | 7509 | |
65e19f54 | 7510 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7511 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7512 | GFP_KERNEL); |
600cf3f8 PB |
7513 | if (!file_data->table) |
7514 | goto out_free; | |
05f3fb3c | 7515 | |
600cf3f8 | 7516 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7517 | goto out_free; |
65e19f54 | 7518 | |
08a45173 | 7519 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7520 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7521 | ret = -EFAULT; | |
7522 | goto out_fput; | |
7523 | } | |
08a45173 | 7524 | /* allow sparse sets */ |
600cf3f8 | 7525 | if (fd == -1) |
08a45173 | 7526 | continue; |
6b06314c | 7527 | |
05f3fb3c | 7528 | file = fget(fd); |
6b06314c | 7529 | ret = -EBADF; |
05f3fb3c | 7530 | if (!file) |
600cf3f8 | 7531 | goto out_fput; |
05f3fb3c | 7532 | |
6b06314c JA |
7533 | /* |
7534 | * Don't allow io_uring instances to be registered. If UNIX | |
7535 | * isn't enabled, then this causes a reference cycle and this | |
7536 | * instance can never get freed. If UNIX is enabled we'll | |
7537 | * handle it just fine, but there's still no point in allowing | |
7538 | * a ring fd as it doesn't support regular read/write anyway. | |
7539 | */ | |
05f3fb3c JA |
7540 | if (file->f_op == &io_uring_fops) { |
7541 | fput(file); | |
600cf3f8 | 7542 | goto out_fput; |
6b06314c | 7543 | } |
ea64ec02 | 7544 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7545 | } |
7546 | ||
6b06314c | 7547 | ret = io_sqe_files_scm(ctx); |
05589553 | 7548 | if (ret) { |
6b06314c | 7549 | io_sqe_files_unregister(ctx); |
05589553 XW |
7550 | return ret; |
7551 | } | |
6b06314c | 7552 | |
bc9744cd | 7553 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7554 | if (!ref_node) { |
05589553 | 7555 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7556 | return -ENOMEM; |
05589553 | 7557 | } |
bc9744cd | 7558 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7559 | |
d67d2263 | 7560 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7561 | return ret; |
600cf3f8 PB |
7562 | out_fput: |
7563 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7564 | file = io_file_from_index(ctx, i); | |
7565 | if (file) | |
7566 | fput(file); | |
7567 | } | |
7568 | for (i = 0; i < nr_tables; i++) | |
7569 | kfree(file_data->table[i].files); | |
7570 | ctx->nr_user_files = 0; | |
600cf3f8 | 7571 | out_free: |
1ad555c6 | 7572 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7573 | ctx->file_data = NULL; |
6b06314c JA |
7574 | return ret; |
7575 | } | |
7576 | ||
c3a31e60 JA |
7577 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7578 | int index) | |
7579 | { | |
7580 | #if defined(CONFIG_UNIX) | |
7581 | struct sock *sock = ctx->ring_sock->sk; | |
7582 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7583 | struct sk_buff *skb; | |
7584 | ||
7585 | /* | |
7586 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7587 | * file set. If there's no room, fall back to allocating a new skb | |
7588 | * and filling it in. | |
7589 | */ | |
7590 | spin_lock_irq(&head->lock); | |
7591 | skb = skb_peek(head); | |
7592 | if (skb) { | |
7593 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7594 | ||
7595 | if (fpl->count < SCM_MAX_FD) { | |
7596 | __skb_unlink(skb, head); | |
7597 | spin_unlock_irq(&head->lock); | |
7598 | fpl->fp[fpl->count] = get_file(file); | |
7599 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7600 | fpl->count++; | |
7601 | spin_lock_irq(&head->lock); | |
7602 | __skb_queue_head(head, skb); | |
7603 | } else { | |
7604 | skb = NULL; | |
7605 | } | |
7606 | } | |
7607 | spin_unlock_irq(&head->lock); | |
7608 | ||
7609 | if (skb) { | |
7610 | fput(file); | |
7611 | return 0; | |
7612 | } | |
7613 | ||
7614 | return __io_sqe_files_scm(ctx, 1, index); | |
7615 | #else | |
7616 | return 0; | |
7617 | #endif | |
7618 | } | |
7619 | ||
50238531 | 7620 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7621 | { |
269bbe5f BM |
7622 | struct io_rsrc_put *prsrc; |
7623 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7624 | |
269bbe5f BM |
7625 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7626 | if (!prsrc) | |
a5318d3c | 7627 | return -ENOMEM; |
05f3fb3c | 7628 | |
50238531 | 7629 | prsrc->rsrc = rsrc; |
269bbe5f | 7630 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7631 | |
a5318d3c | 7632 | return 0; |
05f3fb3c JA |
7633 | } |
7634 | ||
269bbe5f BM |
7635 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7636 | struct file *file) | |
7637 | { | |
50238531 | 7638 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7639 | } |
7640 | ||
05f3fb3c | 7641 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7642 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7643 | unsigned nr_args) |
7644 | { | |
269bbe5f BM |
7645 | struct fixed_rsrc_data *data = ctx->file_data; |
7646 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7647 | struct file *file, **file_slot; |
c3a31e60 JA |
7648 | __s32 __user *fds; |
7649 | int fd, i, err; | |
7650 | __u32 done; | |
05589553 | 7651 | bool needs_switch = false; |
c3a31e60 | 7652 | |
05f3fb3c | 7653 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7654 | return -EOVERFLOW; |
7655 | if (done > ctx->nr_user_files) | |
7656 | return -EINVAL; | |
7657 | ||
bc9744cd | 7658 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7659 | if (!ref_node) |
7660 | return -ENOMEM; | |
bc9744cd | 7661 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7662 | |
269bbe5f | 7663 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7664 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7665 | err = 0; |
7666 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7667 | err = -EFAULT; | |
7668 | break; | |
7669 | } | |
4e0377a1 | 7670 | if (fd == IORING_REGISTER_FILES_SKIP) |
7671 | continue; | |
7672 | ||
67973b93 | 7673 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7674 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7675 | ||
7676 | if (*file_slot) { | |
7677 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7678 | if (err) |
7679 | break; | |
ea64ec02 | 7680 | *file_slot = NULL; |
05589553 | 7681 | needs_switch = true; |
c3a31e60 JA |
7682 | } |
7683 | if (fd != -1) { | |
c3a31e60 JA |
7684 | file = fget(fd); |
7685 | if (!file) { | |
7686 | err = -EBADF; | |
7687 | break; | |
7688 | } | |
7689 | /* | |
7690 | * Don't allow io_uring instances to be registered. If | |
7691 | * UNIX isn't enabled, then this causes a reference | |
7692 | * cycle and this instance can never get freed. If UNIX | |
7693 | * is enabled we'll handle it just fine, but there's | |
7694 | * still no point in allowing a ring fd as it doesn't | |
7695 | * support regular read/write anyway. | |
7696 | */ | |
7697 | if (file->f_op == &io_uring_fops) { | |
7698 | fput(file); | |
7699 | err = -EBADF; | |
7700 | break; | |
7701 | } | |
e68a3ff8 | 7702 | *file_slot = file; |
c3a31e60 | 7703 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7704 | if (err) { |
e68a3ff8 | 7705 | *file_slot = NULL; |
f3bd9dae | 7706 | fput(file); |
c3a31e60 | 7707 | break; |
f3bd9dae | 7708 | } |
c3a31e60 | 7709 | } |
05f3fb3c JA |
7710 | } |
7711 | ||
05589553 | 7712 | if (needs_switch) { |
b2e96852 | 7713 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7714 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7715 | } else |
269bbe5f | 7716 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7717 | |
7718 | return done ? done : err; | |
7719 | } | |
05589553 | 7720 | |
05f3fb3c JA |
7721 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7722 | unsigned nr_args) | |
7723 | { | |
269bbe5f | 7724 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7725 | |
7726 | if (!ctx->file_data) | |
7727 | return -ENXIO; | |
7728 | if (!nr_args) | |
7729 | return -EINVAL; | |
7730 | if (copy_from_user(&up, arg, sizeof(up))) | |
7731 | return -EFAULT; | |
7732 | if (up.resv) | |
7733 | return -EINVAL; | |
7734 | ||
7735 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7736 | } | |
c3a31e60 | 7737 | |
5280f7e5 | 7738 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7739 | { |
7740 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7741 | ||
5280f7e5 PB |
7742 | req = io_put_req_find_next(req); |
7743 | return req ? &req->work : NULL; | |
7d723065 JA |
7744 | } |
7745 | ||
5aa75ed5 | 7746 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7747 | { |
e941894e | 7748 | struct io_wq_hash *hash; |
24369c2e | 7749 | struct io_wq_data data; |
24369c2e | 7750 | unsigned int concurrency; |
24369c2e | 7751 | |
e941894e JA |
7752 | hash = ctx->hash_map; |
7753 | if (!hash) { | |
7754 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7755 | if (!hash) | |
7756 | return ERR_PTR(-ENOMEM); | |
7757 | refcount_set(&hash->refs, 1); | |
7758 | init_waitqueue_head(&hash->wait); | |
7759 | ctx->hash_map = hash; | |
24369c2e PB |
7760 | } |
7761 | ||
e941894e | 7762 | data.hash = hash; |
e9fd9396 | 7763 | data.free_work = io_free_work; |
f5fa38c5 | 7764 | data.do_work = io_wq_submit_work; |
24369c2e | 7765 | |
d25e3a3d JA |
7766 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7767 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7768 | |
5aa75ed5 | 7769 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7770 | } |
7771 | ||
5aa75ed5 JA |
7772 | static int io_uring_alloc_task_context(struct task_struct *task, |
7773 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7774 | { |
7775 | struct io_uring_task *tctx; | |
d8a6df10 | 7776 | int ret; |
0f212204 JA |
7777 | |
7778 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7779 | if (unlikely(!tctx)) | |
7780 | return -ENOMEM; | |
7781 | ||
d8a6df10 JA |
7782 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7783 | if (unlikely(ret)) { | |
7784 | kfree(tctx); | |
7785 | return ret; | |
7786 | } | |
7787 | ||
5aa75ed5 JA |
7788 | tctx->io_wq = io_init_wq_offload(ctx); |
7789 | if (IS_ERR(tctx->io_wq)) { | |
7790 | ret = PTR_ERR(tctx->io_wq); | |
7791 | percpu_counter_destroy(&tctx->inflight); | |
7792 | kfree(tctx); | |
7793 | return ret; | |
7794 | } | |
7795 | ||
0f212204 JA |
7796 | xa_init(&tctx->xa); |
7797 | init_waitqueue_head(&tctx->wait); | |
7798 | tctx->last = NULL; | |
fdaf083c | 7799 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7800 | task->io_uring = tctx; |
7cbf1722 JA |
7801 | spin_lock_init(&tctx->task_lock); |
7802 | INIT_WQ_LIST(&tctx->task_list); | |
7803 | tctx->task_state = 0; | |
7804 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7805 | return 0; |
7806 | } | |
7807 | ||
7808 | void __io_uring_free(struct task_struct *tsk) | |
7809 | { | |
7810 | struct io_uring_task *tctx = tsk->io_uring; | |
7811 | ||
7812 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7813 | WARN_ON_ONCE(tctx->io_wq); |
7814 | ||
d8a6df10 | 7815 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7816 | kfree(tctx); |
7817 | tsk->io_uring = NULL; | |
7818 | } | |
7819 | ||
7e84e1c7 SG |
7820 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7821 | struct io_uring_params *p) | |
2b188cc1 JA |
7822 | { |
7823 | int ret; | |
7824 | ||
d25e3a3d JA |
7825 | /* Retain compatibility with failing for an invalid attach attempt */ |
7826 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7827 | IORING_SETUP_ATTACH_WQ) { | |
7828 | struct fd f; | |
7829 | ||
7830 | f = fdget(p->wq_fd); | |
7831 | if (!f.file) | |
7832 | return -ENXIO; | |
7833 | if (f.file->f_op != &io_uring_fops) { | |
7834 | fdput(f); | |
7835 | return -EINVAL; | |
7836 | } | |
7837 | fdput(f); | |
7838 | } | |
6c271ce2 | 7839 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7840 | struct task_struct *tsk; |
534ca6d6 | 7841 | struct io_sq_data *sqd; |
26984fbf | 7842 | bool attached; |
534ca6d6 | 7843 | |
3ec482d1 | 7844 | ret = -EPERM; |
ce59fc69 | 7845 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7846 | goto err; |
7847 | ||
26984fbf | 7848 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7849 | if (IS_ERR(sqd)) { |
7850 | ret = PTR_ERR(sqd); | |
7851 | goto err; | |
7852 | } | |
69fb2131 | 7853 | |
7c30f36a | 7854 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7855 | ctx->sq_data = sqd; |
917257da JA |
7856 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7857 | if (!ctx->sq_thread_idle) | |
7858 | ctx->sq_thread_idle = HZ; | |
7859 | ||
26984fbf | 7860 | ret = 0; |
78d7f6ba | 7861 | io_sq_thread_park(sqd); |
26984fbf PB |
7862 | /* don't attach to a dying SQPOLL thread, would be racy */ |
7863 | if (attached && !sqd->thread) { | |
7864 | ret = -ENXIO; | |
7865 | } else { | |
7866 | list_add(&ctx->sqd_list, &sqd->ctx_list); | |
7867 | io_sqd_update_thread_idle(sqd); | |
7868 | } | |
78d7f6ba PB |
7869 | io_sq_thread_unpark(sqd); |
7870 | ||
26984fbf PB |
7871 | if (ret < 0) { |
7872 | io_put_sq_data(sqd); | |
7873 | ctx->sq_data = NULL; | |
7874 | return ret; | |
7875 | } else if (attached) { | |
5aa75ed5 | 7876 | return 0; |
26984fbf | 7877 | } |
aa06165d | 7878 | |
6c271ce2 | 7879 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7880 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7881 | |
917257da | 7882 | ret = -EINVAL; |
44a9bd18 | 7883 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7884 | goto err_sqpoll; |
7889f44d | 7885 | if (!cpu_online(cpu)) |
e8f98f24 | 7886 | goto err_sqpoll; |
917257da | 7887 | |
37d1e2e3 | 7888 | sqd->sq_cpu = cpu; |
6c271ce2 | 7889 | } else { |
37d1e2e3 | 7890 | sqd->sq_cpu = -1; |
6c271ce2 | 7891 | } |
37d1e2e3 JA |
7892 | |
7893 | sqd->task_pid = current->pid; | |
5c2469e0 | 7894 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7895 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7896 | if (IS_ERR(tsk)) { | |
7897 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7898 | goto err_sqpoll; |
6c271ce2 | 7899 | } |
97a73a0f | 7900 | |
46fe18b1 | 7901 | sqd->thread = tsk; |
97a73a0f | 7902 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7903 | wake_up_new_task(tsk); |
0f212204 JA |
7904 | if (ret) |
7905 | goto err; | |
6c271ce2 JA |
7906 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7907 | /* Can't have SQ_AFF without SQPOLL */ | |
7908 | ret = -EINVAL; | |
7909 | goto err; | |
7910 | } | |
7911 | ||
2b188cc1 JA |
7912 | return 0; |
7913 | err: | |
37d1e2e3 | 7914 | io_sq_thread_finish(ctx); |
2b188cc1 | 7915 | return ret; |
e8f98f24 JA |
7916 | err_sqpoll: |
7917 | complete(&ctx->sq_data->exited); | |
7918 | goto err; | |
2b188cc1 JA |
7919 | } |
7920 | ||
a087e2b5 BM |
7921 | static inline void __io_unaccount_mem(struct user_struct *user, |
7922 | unsigned long nr_pages) | |
2b188cc1 JA |
7923 | { |
7924 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7925 | } | |
7926 | ||
a087e2b5 BM |
7927 | static inline int __io_account_mem(struct user_struct *user, |
7928 | unsigned long nr_pages) | |
2b188cc1 JA |
7929 | { |
7930 | unsigned long page_limit, cur_pages, new_pages; | |
7931 | ||
7932 | /* Don't allow more pages than we can safely lock */ | |
7933 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7934 | ||
7935 | do { | |
7936 | cur_pages = atomic_long_read(&user->locked_vm); | |
7937 | new_pages = cur_pages + nr_pages; | |
7938 | if (new_pages > page_limit) | |
7939 | return -ENOMEM; | |
7940 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7941 | new_pages) != cur_pages); | |
7942 | ||
7943 | return 0; | |
7944 | } | |
7945 | ||
26bfa89e | 7946 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7947 | { |
62e398be | 7948 | if (ctx->user) |
a087e2b5 | 7949 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7950 | |
26bfa89e JA |
7951 | if (ctx->mm_account) |
7952 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7953 | } |
7954 | ||
26bfa89e | 7955 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7956 | { |
30975825 BM |
7957 | int ret; |
7958 | ||
62e398be | 7959 | if (ctx->user) { |
30975825 BM |
7960 | ret = __io_account_mem(ctx->user, nr_pages); |
7961 | if (ret) | |
7962 | return ret; | |
7963 | } | |
7964 | ||
26bfa89e JA |
7965 | if (ctx->mm_account) |
7966 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7967 | |
7968 | return 0; | |
7969 | } | |
7970 | ||
2b188cc1 JA |
7971 | static void io_mem_free(void *ptr) |
7972 | { | |
52e04ef4 MR |
7973 | struct page *page; |
7974 | ||
7975 | if (!ptr) | |
7976 | return; | |
2b188cc1 | 7977 | |
52e04ef4 | 7978 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
7979 | if (put_page_testzero(page)) |
7980 | free_compound_page(page); | |
7981 | } | |
7982 | ||
7983 | static void *io_mem_alloc(size_t size) | |
7984 | { | |
7985 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 7986 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
7987 | |
7988 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
7989 | } | |
7990 | ||
75b28aff HV |
7991 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
7992 | size_t *sq_offset) | |
7993 | { | |
7994 | struct io_rings *rings; | |
7995 | size_t off, sq_array_size; | |
7996 | ||
7997 | off = struct_size(rings, cqes, cq_entries); | |
7998 | if (off == SIZE_MAX) | |
7999 | return SIZE_MAX; | |
8000 | ||
8001 | #ifdef CONFIG_SMP | |
8002 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8003 | if (off == 0) | |
8004 | return SIZE_MAX; | |
8005 | #endif | |
8006 | ||
b36200f5 DV |
8007 | if (sq_offset) |
8008 | *sq_offset = off; | |
8009 | ||
75b28aff HV |
8010 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8011 | if (sq_array_size == SIZE_MAX) | |
8012 | return SIZE_MAX; | |
8013 | ||
8014 | if (check_add_overflow(off, sq_array_size, &off)) | |
8015 | return SIZE_MAX; | |
8016 | ||
75b28aff HV |
8017 | return off; |
8018 | } | |
8019 | ||
0a96bbe4 | 8020 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8021 | { |
8022 | int i, j; | |
8023 | ||
8024 | if (!ctx->user_bufs) | |
8025 | return -ENXIO; | |
8026 | ||
8027 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8028 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8029 | ||
8030 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8031 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8032 | |
de293938 | 8033 | if (imu->acct_pages) |
26bfa89e | 8034 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8035 | kvfree(imu->bvec); |
edafccee JA |
8036 | imu->nr_bvecs = 0; |
8037 | } | |
8038 | ||
8039 | kfree(ctx->user_bufs); | |
8040 | ctx->user_bufs = NULL; | |
8041 | ctx->nr_user_bufs = 0; | |
8042 | return 0; | |
8043 | } | |
8044 | ||
8045 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8046 | void __user *arg, unsigned index) | |
8047 | { | |
8048 | struct iovec __user *src; | |
8049 | ||
8050 | #ifdef CONFIG_COMPAT | |
8051 | if (ctx->compat) { | |
8052 | struct compat_iovec __user *ciovs; | |
8053 | struct compat_iovec ciov; | |
8054 | ||
8055 | ciovs = (struct compat_iovec __user *) arg; | |
8056 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8057 | return -EFAULT; | |
8058 | ||
d55e5f5b | 8059 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8060 | dst->iov_len = ciov.iov_len; |
8061 | return 0; | |
8062 | } | |
8063 | #endif | |
8064 | src = (struct iovec __user *) arg; | |
8065 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8066 | return -EFAULT; | |
8067 | return 0; | |
8068 | } | |
8069 | ||
de293938 JA |
8070 | /* |
8071 | * Not super efficient, but this is just a registration time. And we do cache | |
8072 | * the last compound head, so generally we'll only do a full search if we don't | |
8073 | * match that one. | |
8074 | * | |
8075 | * We check if the given compound head page has already been accounted, to | |
8076 | * avoid double accounting it. This allows us to account the full size of the | |
8077 | * page, not just the constituent pages of a huge page. | |
8078 | */ | |
8079 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8080 | int nr_pages, struct page *hpage) | |
8081 | { | |
8082 | int i, j; | |
8083 | ||
8084 | /* check current page array */ | |
8085 | for (i = 0; i < nr_pages; i++) { | |
8086 | if (!PageCompound(pages[i])) | |
8087 | continue; | |
8088 | if (compound_head(pages[i]) == hpage) | |
8089 | return true; | |
8090 | } | |
8091 | ||
8092 | /* check previously registered pages */ | |
8093 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8094 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8095 | ||
8096 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8097 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8098 | continue; | |
8099 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8100 | return true; | |
8101 | } | |
8102 | } | |
8103 | ||
8104 | return false; | |
8105 | } | |
8106 | ||
8107 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8108 | int nr_pages, struct io_mapped_ubuf *imu, | |
8109 | struct page **last_hpage) | |
8110 | { | |
8111 | int i, ret; | |
8112 | ||
8113 | for (i = 0; i < nr_pages; i++) { | |
8114 | if (!PageCompound(pages[i])) { | |
8115 | imu->acct_pages++; | |
8116 | } else { | |
8117 | struct page *hpage; | |
8118 | ||
8119 | hpage = compound_head(pages[i]); | |
8120 | if (hpage == *last_hpage) | |
8121 | continue; | |
8122 | *last_hpage = hpage; | |
8123 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8124 | continue; | |
8125 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8126 | } | |
8127 | } | |
8128 | ||
8129 | if (!imu->acct_pages) | |
8130 | return 0; | |
8131 | ||
26bfa89e | 8132 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8133 | if (ret) |
8134 | imu->acct_pages = 0; | |
8135 | return ret; | |
8136 | } | |
8137 | ||
0a96bbe4 BM |
8138 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8139 | struct io_mapped_ubuf *imu, | |
8140 | struct page **last_hpage) | |
edafccee JA |
8141 | { |
8142 | struct vm_area_struct **vmas = NULL; | |
8143 | struct page **pages = NULL; | |
0a96bbe4 BM |
8144 | unsigned long off, start, end, ubuf; |
8145 | size_t size; | |
8146 | int ret, pret, nr_pages, i; | |
8147 | ||
8148 | ubuf = (unsigned long) iov->iov_base; | |
8149 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8150 | start = ubuf >> PAGE_SHIFT; | |
8151 | nr_pages = end - start; | |
8152 | ||
8153 | ret = -ENOMEM; | |
8154 | ||
8155 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8156 | if (!pages) | |
8157 | goto done; | |
8158 | ||
8159 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8160 | GFP_KERNEL); | |
8161 | if (!vmas) | |
8162 | goto done; | |
edafccee | 8163 | |
0a96bbe4 BM |
8164 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8165 | GFP_KERNEL); | |
8166 | if (!imu->bvec) | |
8167 | goto done; | |
8168 | ||
8169 | ret = 0; | |
8170 | mmap_read_lock(current->mm); | |
8171 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8172 | pages, vmas); | |
8173 | if (pret == nr_pages) { | |
8174 | /* don't support file backed memory */ | |
8175 | for (i = 0; i < nr_pages; i++) { | |
8176 | struct vm_area_struct *vma = vmas[i]; | |
8177 | ||
8178 | if (vma->vm_file && | |
8179 | !is_file_hugepages(vma->vm_file)) { | |
8180 | ret = -EOPNOTSUPP; | |
8181 | break; | |
8182 | } | |
8183 | } | |
8184 | } else { | |
8185 | ret = pret < 0 ? pret : -EFAULT; | |
8186 | } | |
8187 | mmap_read_unlock(current->mm); | |
8188 | if (ret) { | |
8189 | /* | |
8190 | * if we did partial map, or found file backed vmas, | |
8191 | * release any pages we did get | |
8192 | */ | |
8193 | if (pret > 0) | |
8194 | unpin_user_pages(pages, pret); | |
8195 | kvfree(imu->bvec); | |
8196 | goto done; | |
8197 | } | |
8198 | ||
8199 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8200 | if (ret) { | |
8201 | unpin_user_pages(pages, pret); | |
8202 | kvfree(imu->bvec); | |
8203 | goto done; | |
8204 | } | |
8205 | ||
8206 | off = ubuf & ~PAGE_MASK; | |
8207 | size = iov->iov_len; | |
8208 | for (i = 0; i < nr_pages; i++) { | |
8209 | size_t vec_len; | |
8210 | ||
8211 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8212 | imu->bvec[i].bv_page = pages[i]; | |
8213 | imu->bvec[i].bv_len = vec_len; | |
8214 | imu->bvec[i].bv_offset = off; | |
8215 | off = 0; | |
8216 | size -= vec_len; | |
8217 | } | |
8218 | /* store original address for later verification */ | |
8219 | imu->ubuf = ubuf; | |
8220 | imu->len = iov->iov_len; | |
8221 | imu->nr_bvecs = nr_pages; | |
8222 | ret = 0; | |
8223 | done: | |
8224 | kvfree(pages); | |
8225 | kvfree(vmas); | |
8226 | return ret; | |
8227 | } | |
8228 | ||
2b358604 | 8229 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8230 | { |
edafccee JA |
8231 | if (ctx->user_bufs) |
8232 | return -EBUSY; | |
8233 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8234 | return -EINVAL; | |
8235 | ||
8236 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8237 | GFP_KERNEL); | |
8238 | if (!ctx->user_bufs) | |
8239 | return -ENOMEM; | |
8240 | ||
2b358604 BM |
8241 | return 0; |
8242 | } | |
edafccee | 8243 | |
2b358604 BM |
8244 | static int io_buffer_validate(struct iovec *iov) |
8245 | { | |
8246 | /* | |
8247 | * Don't impose further limits on the size and buffer | |
8248 | * constraints here, we'll -EINVAL later when IO is | |
8249 | * submitted if they are wrong. | |
8250 | */ | |
8251 | if (!iov->iov_base || !iov->iov_len) | |
8252 | return -EFAULT; | |
edafccee | 8253 | |
2b358604 BM |
8254 | /* arbitrary limit, but we need something */ |
8255 | if (iov->iov_len > SZ_1G) | |
8256 | return -EFAULT; | |
edafccee | 8257 | |
2b358604 BM |
8258 | return 0; |
8259 | } | |
edafccee | 8260 | |
2b358604 BM |
8261 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8262 | unsigned int nr_args) | |
8263 | { | |
8264 | int i, ret; | |
8265 | struct iovec iov; | |
8266 | struct page *last_hpage = NULL; | |
edafccee | 8267 | |
2b358604 BM |
8268 | ret = io_buffers_map_alloc(ctx, nr_args); |
8269 | if (ret) | |
8270 | return ret; | |
edafccee | 8271 | |
edafccee JA |
8272 | for (i = 0; i < nr_args; i++) { |
8273 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8274 | |
edafccee JA |
8275 | ret = io_copy_iov(ctx, &iov, arg, i); |
8276 | if (ret) | |
0a96bbe4 | 8277 | break; |
de293938 | 8278 | |
2b358604 BM |
8279 | ret = io_buffer_validate(&iov); |
8280 | if (ret) | |
0a96bbe4 | 8281 | break; |
edafccee | 8282 | |
0a96bbe4 BM |
8283 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8284 | if (ret) | |
8285 | break; | |
edafccee JA |
8286 | |
8287 | ctx->nr_user_bufs++; | |
8288 | } | |
0a96bbe4 BM |
8289 | |
8290 | if (ret) | |
8291 | io_sqe_buffers_unregister(ctx); | |
8292 | ||
edafccee JA |
8293 | return ret; |
8294 | } | |
8295 | ||
9b402849 JA |
8296 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8297 | { | |
8298 | __s32 __user *fds = arg; | |
8299 | int fd; | |
8300 | ||
8301 | if (ctx->cq_ev_fd) | |
8302 | return -EBUSY; | |
8303 | ||
8304 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8305 | return -EFAULT; | |
8306 | ||
8307 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8308 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8309 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8310 | ctx->cq_ev_fd = NULL; | |
8311 | return ret; | |
8312 | } | |
8313 | ||
8314 | return 0; | |
8315 | } | |
8316 | ||
8317 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8318 | { | |
8319 | if (ctx->cq_ev_fd) { | |
8320 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8321 | ctx->cq_ev_fd = NULL; | |
8322 | return 0; | |
8323 | } | |
8324 | ||
8325 | return -ENXIO; | |
8326 | } | |
8327 | ||
5a2e745d JA |
8328 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8329 | { | |
9e15c3a0 JA |
8330 | struct io_buffer *buf; |
8331 | unsigned long index; | |
8332 | ||
8333 | xa_for_each(&ctx->io_buffers, index, buf) | |
8334 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8335 | } |
8336 | ||
68e68ee6 | 8337 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8338 | { |
68e68ee6 | 8339 | struct io_kiocb *req, *nxt; |
1b4c351f | 8340 | |
68e68ee6 JA |
8341 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8342 | if (tsk && req->task != tsk) | |
8343 | continue; | |
1b4c351f JA |
8344 | list_del(&req->compl.list); |
8345 | kmem_cache_free(req_cachep, req); | |
8346 | } | |
8347 | } | |
8348 | ||
4010fec4 | 8349 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8350 | { |
bf019da7 | 8351 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8352 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8353 | |
9a4fdbd8 JA |
8354 | mutex_lock(&ctx->uring_lock); |
8355 | ||
8e5c66c4 | 8356 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8357 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8358 | submit_state->reqs); | |
8e5c66c4 PB |
8359 | submit_state->free_reqs = 0; |
8360 | } | |
9a4fdbd8 JA |
8361 | |
8362 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8363 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8364 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8365 | spin_unlock_irq(&ctx->completion_lock); |
8366 | ||
e5547d2c PB |
8367 | io_req_cache_free(&cs->free_list, NULL); |
8368 | ||
9a4fdbd8 JA |
8369 | mutex_unlock(&ctx->uring_lock); |
8370 | } | |
8371 | ||
2b188cc1 JA |
8372 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8373 | { | |
04fc6c80 PB |
8374 | /* |
8375 | * Some may use context even when all refs and requests have been put, | |
8376 | * and they are free to do so while still holding uring_lock, see | |
8377 | * __io_req_task_submit(). Wait for them to finish. | |
8378 | */ | |
8379 | mutex_lock(&ctx->uring_lock); | |
8380 | mutex_unlock(&ctx->uring_lock); | |
8381 | ||
37d1e2e3 | 8382 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8383 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8384 | |
37d1e2e3 | 8385 | if (ctx->mm_account) { |
2aede0e4 JA |
8386 | mmdrop(ctx->mm_account); |
8387 | ctx->mm_account = NULL; | |
30975825 | 8388 | } |
def596e9 | 8389 | |
8bad28d8 | 8390 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8391 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8392 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8393 | io_eventfd_unregister(ctx); |
5a2e745d | 8394 | io_destroy_buffers(ctx); |
def596e9 | 8395 | |
2b188cc1 | 8396 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8397 | if (ctx->ring_sock) { |
8398 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8399 | sock_release(ctx->ring_sock); |
355e8d26 | 8400 | } |
2b188cc1 JA |
8401 | #endif |
8402 | ||
75b28aff | 8403 | io_mem_free(ctx->rings); |
2b188cc1 | 8404 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8405 | |
8406 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8407 | free_uid(ctx->user); |
4010fec4 | 8408 | io_req_caches_free(ctx); |
e941894e JA |
8409 | if (ctx->hash_map) |
8410 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8411 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8412 | kfree(ctx); |
8413 | } | |
8414 | ||
8415 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8416 | { | |
8417 | struct io_ring_ctx *ctx = file->private_data; | |
8418 | __poll_t mask = 0; | |
8419 | ||
8420 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8421 | /* |
8422 | * synchronizes with barrier from wq_has_sleeper call in | |
8423 | * io_commit_cqring | |
8424 | */ | |
2b188cc1 | 8425 | smp_rmb(); |
90554200 | 8426 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8427 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8428 | |
8429 | /* | |
8430 | * Don't flush cqring overflow list here, just do a simple check. | |
8431 | * Otherwise there could possible be ABBA deadlock: | |
8432 | * CPU0 CPU1 | |
8433 | * ---- ---- | |
8434 | * lock(&ctx->uring_lock); | |
8435 | * lock(&ep->mtx); | |
8436 | * lock(&ctx->uring_lock); | |
8437 | * lock(&ep->mtx); | |
8438 | * | |
8439 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8440 | * pushs them to do the flush. | |
8441 | */ | |
8442 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8443 | mask |= EPOLLIN | EPOLLRDNORM; |
8444 | ||
8445 | return mask; | |
8446 | } | |
8447 | ||
8448 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8449 | { | |
8450 | struct io_ring_ctx *ctx = file->private_data; | |
8451 | ||
8452 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8453 | } | |
8454 | ||
0bead8cd | 8455 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8456 | { |
4379bf8b | 8457 | const struct cred *creds; |
071698e1 | 8458 | |
61cf9370 | 8459 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8460 | if (creds) { |
8461 | put_cred(creds); | |
0bead8cd | 8462 | return 0; |
1e6fa521 | 8463 | } |
0bead8cd YD |
8464 | |
8465 | return -EINVAL; | |
8466 | } | |
8467 | ||
ba50a036 | 8468 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8469 | { |
28c4721b | 8470 | struct callback_head *work, *next; |
ba50a036 | 8471 | bool executed = false; |
7c25c0d1 JA |
8472 | |
8473 | do { | |
28c4721b | 8474 | work = xchg(&ctx->exit_task_work, NULL); |
7c25c0d1 JA |
8475 | if (!work) |
8476 | break; | |
8477 | ||
8478 | do { | |
8479 | next = work->next; | |
8480 | work->func(work); | |
8481 | work = next; | |
8482 | cond_resched(); | |
8483 | } while (work); | |
ba50a036 | 8484 | executed = true; |
7c25c0d1 | 8485 | } while (1); |
ba50a036 PB |
8486 | |
8487 | return executed; | |
7c25c0d1 JA |
8488 | } |
8489 | ||
d56d938b PB |
8490 | struct io_tctx_exit { |
8491 | struct callback_head task_work; | |
8492 | struct completion completion; | |
baf186c4 | 8493 | struct io_ring_ctx *ctx; |
d56d938b PB |
8494 | }; |
8495 | ||
8496 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8497 | { | |
8498 | struct io_uring_task *tctx = current->io_uring; | |
8499 | struct io_tctx_exit *work; | |
8500 | ||
8501 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8502 | /* | |
8503 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8504 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8505 | */ | |
8506 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8507 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8508 | complete(&work->completion); |
8509 | } | |
8510 | ||
85faa7b8 JA |
8511 | static void io_ring_exit_work(struct work_struct *work) |
8512 | { | |
d56d938b | 8513 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8514 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8515 | struct io_tctx_exit exit; |
8516 | struct io_tctx_node *node; | |
8517 | int ret; | |
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); |
b5bb3a24 PB |
8527 | |
8528 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8529 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8530 | |
8531 | mutex_lock(&ctx->uring_lock); | |
8532 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8533 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8534 | ||
d56d938b PB |
8535 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8536 | ctx_node); | |
baf186c4 | 8537 | exit.ctx = ctx; |
d56d938b PB |
8538 | init_completion(&exit.completion); |
8539 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8540 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8541 | if (WARN_ON_ONCE(ret)) | |
8542 | continue; | |
8543 | wake_up_process(node->task); | |
8544 | ||
8545 | mutex_unlock(&ctx->uring_lock); | |
8546 | wait_for_completion(&exit.completion); | |
8547 | cond_resched(); | |
8548 | mutex_lock(&ctx->uring_lock); | |
8549 | } | |
8550 | mutex_unlock(&ctx->uring_lock); | |
8551 | ||
85faa7b8 JA |
8552 | io_ring_ctx_free(ctx); |
8553 | } | |
8554 | ||
2b188cc1 JA |
8555 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8556 | { | |
61cf9370 MWO |
8557 | unsigned long index; |
8558 | struct creds *creds; | |
8559 | ||
2b188cc1 JA |
8560 | mutex_lock(&ctx->uring_lock); |
8561 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8562 | /* if force is set, the ring is going away. always drop after that */ |
8563 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8564 | if (ctx->rings) |
6c503150 | 8565 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8566 | xa_for_each(&ctx->personalities, index, creds) |
8567 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8568 | mutex_unlock(&ctx->uring_lock); |
8569 | ||
6b81928d PB |
8570 | io_kill_timeouts(ctx, NULL, NULL); |
8571 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8572 | |
15dff286 | 8573 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8574 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8575 | |
85faa7b8 | 8576 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8577 | /* |
8578 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8579 | * if we're exiting a ton of rings at the same time. It just adds | |
8580 | * noise and overhead, there's no discernable change in runtime | |
8581 | * over using system_wq. | |
8582 | */ | |
8583 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8584 | } |
8585 | ||
8586 | static int io_uring_release(struct inode *inode, struct file *file) | |
8587 | { | |
8588 | struct io_ring_ctx *ctx = file->private_data; | |
8589 | ||
8590 | file->private_data = NULL; | |
8591 | io_ring_ctx_wait_and_kill(ctx); | |
8592 | return 0; | |
8593 | } | |
8594 | ||
f6edbabb PB |
8595 | struct io_task_cancel { |
8596 | struct task_struct *task; | |
8597 | struct files_struct *files; | |
8598 | }; | |
f254ac04 | 8599 | |
f6edbabb | 8600 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8601 | { |
9a472ef7 | 8602 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8603 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8604 | bool ret; |
8605 | ||
f6edbabb | 8606 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8607 | unsigned long flags; |
8608 | struct io_ring_ctx *ctx = req->ctx; | |
8609 | ||
8610 | /* protect against races with linked timeouts */ | |
8611 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8612 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8613 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8614 | } else { | |
f6edbabb | 8615 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8616 | } |
8617 | return ret; | |
b711d4ea JA |
8618 | } |
8619 | ||
e1915f76 | 8620 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8621 | struct task_struct *task, |
b7ddce3c PB |
8622 | struct files_struct *files) |
8623 | { | |
e1915f76 | 8624 | struct io_defer_entry *de; |
b7ddce3c PB |
8625 | LIST_HEAD(list); |
8626 | ||
8627 | spin_lock_irq(&ctx->completion_lock); | |
8628 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8629 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8630 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8631 | break; | |
8632 | } | |
8633 | } | |
8634 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8635 | if (list_empty(&list)) |
8636 | return false; | |
b7ddce3c PB |
8637 | |
8638 | while (!list_empty(&list)) { | |
8639 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8640 | list_del_init(&de->list); | |
8641 | req_set_fail_links(de->req); | |
8642 | io_put_req(de->req); | |
8643 | io_req_complete(de->req, -ECANCELED); | |
8644 | kfree(de); | |
8645 | } | |
e1915f76 | 8646 | return true; |
b7ddce3c PB |
8647 | } |
8648 | ||
1b00764f PB |
8649 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8650 | { | |
8651 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8652 | ||
8653 | return req->ctx == data; | |
8654 | } | |
8655 | ||
8656 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8657 | { | |
8658 | struct io_tctx_node *node; | |
8659 | enum io_wq_cancel cret; | |
8660 | bool ret = false; | |
8661 | ||
8662 | mutex_lock(&ctx->uring_lock); | |
8663 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8664 | struct io_uring_task *tctx = node->task->io_uring; | |
8665 | ||
8666 | /* | |
8667 | * io_wq will stay alive while we hold uring_lock, because it's | |
8668 | * killed after ctx nodes, which requires to take the lock. | |
8669 | */ | |
8670 | if (!tctx || !tctx->io_wq) | |
8671 | continue; | |
8672 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8673 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8674 | } | |
8675 | mutex_unlock(&ctx->uring_lock); | |
8676 | ||
8677 | return ret; | |
8678 | } | |
8679 | ||
9936c7c2 PB |
8680 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8681 | struct task_struct *task, | |
8682 | struct files_struct *files) | |
8683 | { | |
8684 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8685 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8686 | |
8687 | while (1) { | |
8688 | enum io_wq_cancel cret; | |
8689 | bool ret = false; | |
8690 | ||
1b00764f PB |
8691 | if (!task) { |
8692 | ret |= io_uring_try_cancel_iowq(ctx); | |
8693 | } else if (tctx && tctx->io_wq) { | |
8694 | /* | |
8695 | * Cancels requests of all rings, not only @ctx, but | |
8696 | * it's fine as the task is in exit/exec. | |
8697 | */ | |
5aa75ed5 | 8698 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8699 | &cancel, true); |
8700 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8701 | } | |
8702 | ||
8703 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8704 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8705 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8706 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8707 | io_iopoll_try_reap_events(ctx); | |
8708 | ret = true; | |
8709 | } | |
8710 | } | |
8711 | ||
e1915f76 | 8712 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8713 | ret |= io_poll_remove_all(ctx, task, files); |
8714 | ret |= io_kill_timeouts(ctx, task, files); | |
8715 | ret |= io_run_task_work(); | |
ba50a036 | 8716 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8717 | io_cqring_overflow_flush(ctx, true, task, files); |
8718 | if (!ret) | |
8719 | break; | |
8720 | cond_resched(); | |
8721 | } | |
8722 | } | |
8723 | ||
ca70f00b PB |
8724 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8725 | struct task_struct *task, | |
8726 | struct files_struct *files) | |
8727 | { | |
8728 | struct io_kiocb *req; | |
8729 | int cnt = 0; | |
8730 | ||
8731 | spin_lock_irq(&ctx->inflight_lock); | |
8732 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8733 | cnt += io_match_task(req, task, files); | |
8734 | spin_unlock_irq(&ctx->inflight_lock); | |
8735 | return cnt; | |
8736 | } | |
8737 | ||
b52fda00 | 8738 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8739 | struct task_struct *task, |
fcb323cc JA |
8740 | struct files_struct *files) |
8741 | { | |
fcb323cc | 8742 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8743 | DEFINE_WAIT(wait); |
ca70f00b | 8744 | int inflight; |
fcb323cc | 8745 | |
ca70f00b PB |
8746 | inflight = io_uring_count_inflight(ctx, task, files); |
8747 | if (!inflight) | |
fcb323cc | 8748 | break; |
f6edbabb | 8749 | |
9936c7c2 | 8750 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8751 | |
8752 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8753 | TASK_UNINTERRUPTIBLE); | |
8754 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8755 | schedule(); | |
c98de08c | 8756 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8757 | } |
0f212204 JA |
8758 | } |
8759 | ||
0f212204 JA |
8760 | /* |
8761 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8762 | */ | |
baf186c4 | 8763 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8764 | { |
236434c3 | 8765 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8766 | struct io_tctx_node *node; |
a528b04e | 8767 | int ret; |
236434c3 MWO |
8768 | |
8769 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8770 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8771 | if (unlikely(ret)) |
8772 | return ret; | |
236434c3 | 8773 | tctx = current->io_uring; |
0f212204 | 8774 | } |
baf186c4 PB |
8775 | if (tctx->last != ctx) { |
8776 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8777 | |
236434c3 | 8778 | if (!old) { |
13bf43f5 PB |
8779 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8780 | if (!node) | |
8781 | return -ENOMEM; | |
8782 | node->ctx = ctx; | |
13bf43f5 PB |
8783 | node->task = current; |
8784 | ||
baf186c4 | 8785 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8786 | node, GFP_KERNEL)); |
a528b04e | 8787 | if (ret) { |
13bf43f5 | 8788 | kfree(node); |
a528b04e PB |
8789 | return ret; |
8790 | } | |
13bf43f5 PB |
8791 | |
8792 | mutex_lock(&ctx->uring_lock); | |
8793 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8794 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8795 | } |
baf186c4 | 8796 | tctx->last = ctx; |
0f212204 | 8797 | } |
0f212204 JA |
8798 | return 0; |
8799 | } | |
8800 | ||
8801 | /* | |
8802 | * Remove this io_uring_file -> task mapping. | |
8803 | */ | |
2941267b | 8804 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8805 | { |
8806 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8807 | struct io_tctx_node *node; |
2941267b | 8808 | |
eebd2e37 PB |
8809 | if (!tctx) |
8810 | return; | |
13bf43f5 PB |
8811 | node = xa_erase(&tctx->xa, index); |
8812 | if (!node) | |
2941267b | 8813 | return; |
0f212204 | 8814 | |
13bf43f5 PB |
8815 | WARN_ON_ONCE(current != node->task); |
8816 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8817 | ||
8818 | mutex_lock(&node->ctx->uring_lock); | |
8819 | list_del(&node->ctx_node); | |
8820 | mutex_unlock(&node->ctx->uring_lock); | |
8821 | ||
baf186c4 | 8822 | if (tctx->last == node->ctx) |
0f212204 | 8823 | tctx->last = NULL; |
13bf43f5 | 8824 | kfree(node); |
0f212204 JA |
8825 | } |
8826 | ||
8452d4a6 | 8827 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8828 | { |
13bf43f5 | 8829 | struct io_tctx_node *node; |
de7f1d9e PB |
8830 | unsigned long index; |
8831 | ||
13bf43f5 | 8832 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8833 | io_uring_del_task_file(index); |
8452d4a6 PB |
8834 | if (tctx->io_wq) { |
8835 | io_wq_put_and_exit(tctx->io_wq); | |
8836 | tctx->io_wq = NULL; | |
8837 | } | |
de7f1d9e PB |
8838 | } |
8839 | ||
521d6a73 PB |
8840 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8841 | { | |
8842 | return percpu_counter_sum(&tctx->inflight); | |
8843 | } | |
8844 | ||
8845 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8846 | { | |
8847 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8848 | struct io_ring_ctx *ctx = work->ctx; | |
8849 | struct io_sq_data *sqd = ctx->sq_data; | |
8850 | ||
8851 | if (sqd->thread) | |
8852 | io_uring_cancel_sqpoll(ctx); | |
8853 | complete(&work->completion); | |
8854 | } | |
8855 | ||
8856 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8857 | { | |
8858 | struct io_sq_data *sqd = ctx->sq_data; | |
8859 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8860 | struct task_struct *task; | |
8861 | ||
8862 | io_sq_thread_park(sqd); | |
8863 | list_del_init(&ctx->sqd_list); | |
8864 | io_sqd_update_thread_idle(sqd); | |
8865 | task = sqd->thread; | |
8866 | if (task) { | |
8867 | init_completion(&work.completion); | |
8868 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
8869 | WARN_ON_ONCE(task_work_add(task, &work.task_work, TWA_SIGNAL)); | |
8870 | wake_up_process(task); | |
8871 | } | |
8872 | io_sq_thread_unpark(sqd); | |
8873 | ||
8874 | if (task) | |
8875 | wait_for_completion(&work.completion); | |
8876 | } | |
8877 | ||
0f212204 JA |
8878 | void __io_uring_files_cancel(struct files_struct *files) |
8879 | { | |
8880 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8881 | struct io_tctx_node *node; |
ce765372 | 8882 | unsigned long index; |
0f212204 JA |
8883 | |
8884 | /* make sure overflow events are dropped */ | |
fdaf083c | 8885 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8886 | xa_for_each(&tctx->xa, index, node) { |
8887 | struct io_ring_ctx *ctx = node->ctx; | |
8888 | ||
8889 | if (ctx->sq_data) { | |
8890 | io_sqpoll_cancel_sync(ctx); | |
8891 | continue; | |
8892 | } | |
8893 | io_uring_cancel_files(ctx, current, files); | |
8894 | if (!files) | |
8895 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8896 | } | |
fdaf083c | 8897 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8898 | |
8452d4a6 PB |
8899 | if (files) |
8900 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8901 | } |
8902 | ||
521d6a73 | 8903 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8904 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8905 | { | |
37d1e2e3 | 8906 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8907 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8908 | s64 inflight; |
8909 | DEFINE_WAIT(wait); | |
fdaf083c | 8910 | |
521d6a73 PB |
8911 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8912 | ||
0e9ddb39 PB |
8913 | atomic_inc(&tctx->in_idle); |
8914 | do { | |
8915 | /* read completions before cancelations */ | |
8916 | inflight = tctx_inflight(tctx); | |
8917 | if (!inflight) | |
8918 | break; | |
521d6a73 | 8919 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8920 | |
0e9ddb39 PB |
8921 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8922 | /* | |
8923 | * If we've seen completions, retry without waiting. This | |
8924 | * avoids a race where a completion comes in before we did | |
8925 | * prepare_to_wait(). | |
8926 | */ | |
8927 | if (inflight == tctx_inflight(tctx)) | |
8928 | schedule(); | |
8929 | finish_wait(&tctx->wait, &wait); | |
8930 | } while (1); | |
8931 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8932 | } |
8933 | ||
0f212204 JA |
8934 | /* |
8935 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8936 | * requests. | |
8937 | */ | |
8938 | void __io_uring_task_cancel(void) | |
8939 | { | |
8940 | struct io_uring_task *tctx = current->io_uring; | |
8941 | DEFINE_WAIT(wait); | |
d8a6df10 | 8942 | s64 inflight; |
0f212204 JA |
8943 | |
8944 | /* make sure overflow events are dropped */ | |
fdaf083c | 8945 | atomic_inc(&tctx->in_idle); |
d8a6df10 | 8946 | do { |
0f212204 | 8947 | /* read completions before cancelations */ |
fdaf083c | 8948 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
8949 | if (!inflight) |
8950 | break; | |
0f212204 JA |
8951 | __io_uring_files_cancel(NULL); |
8952 | ||
8953 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
8954 | ||
8955 | /* | |
a1bb3cd5 PB |
8956 | * If we've seen completions, retry without waiting. This |
8957 | * avoids a race where a completion comes in before we did | |
8958 | * prepare_to_wait(). | |
0f212204 | 8959 | */ |
a1bb3cd5 PB |
8960 | if (inflight == tctx_inflight(tctx)) |
8961 | schedule(); | |
f57555ed | 8962 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 8963 | } while (1); |
0f212204 | 8964 | |
fdaf083c | 8965 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8966 | |
8452d4a6 PB |
8967 | io_uring_clean_tctx(tctx); |
8968 | /* all current's requests should be gone, we can kill tctx */ | |
8969 | __io_uring_free(current); | |
44e728b8 PB |
8970 | } |
8971 | ||
6c5c240e RP |
8972 | static void *io_uring_validate_mmap_request(struct file *file, |
8973 | loff_t pgoff, size_t sz) | |
2b188cc1 | 8974 | { |
2b188cc1 | 8975 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 8976 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
8977 | struct page *page; |
8978 | void *ptr; | |
8979 | ||
8980 | switch (offset) { | |
8981 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
8982 | case IORING_OFF_CQ_RING: |
8983 | ptr = ctx->rings; | |
2b188cc1 JA |
8984 | break; |
8985 | case IORING_OFF_SQES: | |
8986 | ptr = ctx->sq_sqes; | |
8987 | break; | |
2b188cc1 | 8988 | default: |
6c5c240e | 8989 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
8990 | } |
8991 | ||
8992 | page = virt_to_head_page(ptr); | |
a50b854e | 8993 | if (sz > page_size(page)) |
6c5c240e RP |
8994 | return ERR_PTR(-EINVAL); |
8995 | ||
8996 | return ptr; | |
8997 | } | |
8998 | ||
8999 | #ifdef CONFIG_MMU | |
9000 | ||
9001 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9002 | { | |
9003 | size_t sz = vma->vm_end - vma->vm_start; | |
9004 | unsigned long pfn; | |
9005 | void *ptr; | |
9006 | ||
9007 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9008 | if (IS_ERR(ptr)) | |
9009 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9010 | |
9011 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9012 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9013 | } | |
9014 | ||
6c5c240e RP |
9015 | #else /* !CONFIG_MMU */ |
9016 | ||
9017 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9018 | { | |
9019 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9020 | } | |
9021 | ||
9022 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9023 | { | |
9024 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9025 | } | |
9026 | ||
9027 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9028 | unsigned long addr, unsigned long len, | |
9029 | unsigned long pgoff, unsigned long flags) | |
9030 | { | |
9031 | void *ptr; | |
9032 | ||
9033 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9034 | if (IS_ERR(ptr)) | |
9035 | return PTR_ERR(ptr); | |
9036 | ||
9037 | return (unsigned long) ptr; | |
9038 | } | |
9039 | ||
9040 | #endif /* !CONFIG_MMU */ | |
9041 | ||
d9d05217 | 9042 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9043 | { |
9044 | DEFINE_WAIT(wait); | |
9045 | ||
9046 | do { | |
9047 | if (!io_sqring_full(ctx)) | |
9048 | break; | |
90554200 JA |
9049 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9050 | ||
9051 | if (!io_sqring_full(ctx)) | |
9052 | break; | |
90554200 JA |
9053 | schedule(); |
9054 | } while (!signal_pending(current)); | |
9055 | ||
9056 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9057 | return 0; |
90554200 JA |
9058 | } |
9059 | ||
c73ebb68 HX |
9060 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9061 | struct __kernel_timespec __user **ts, | |
9062 | const sigset_t __user **sig) | |
9063 | { | |
9064 | struct io_uring_getevents_arg arg; | |
9065 | ||
9066 | /* | |
9067 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9068 | * is just a pointer to the sigset_t. | |
9069 | */ | |
9070 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9071 | *sig = (const sigset_t __user *) argp; | |
9072 | *ts = NULL; | |
9073 | return 0; | |
9074 | } | |
9075 | ||
9076 | /* | |
9077 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9078 | * timespec and sigset_t pointers if good. | |
9079 | */ | |
9080 | if (*argsz != sizeof(arg)) | |
9081 | return -EINVAL; | |
9082 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9083 | return -EFAULT; | |
9084 | *sig = u64_to_user_ptr(arg.sigmask); | |
9085 | *argsz = arg.sigmask_sz; | |
9086 | *ts = u64_to_user_ptr(arg.ts); | |
9087 | return 0; | |
9088 | } | |
9089 | ||
2b188cc1 | 9090 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9091 | u32, min_complete, u32, flags, const void __user *, argp, |
9092 | size_t, argsz) | |
2b188cc1 JA |
9093 | { |
9094 | struct io_ring_ctx *ctx; | |
9095 | long ret = -EBADF; | |
9096 | int submitted = 0; | |
9097 | struct fd f; | |
9098 | ||
4c6e277c | 9099 | io_run_task_work(); |
b41e9852 | 9100 | |
90554200 | 9101 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9102 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9103 | return -EINVAL; |
9104 | ||
9105 | f = fdget(fd); | |
9106 | if (!f.file) | |
9107 | return -EBADF; | |
9108 | ||
9109 | ret = -EOPNOTSUPP; | |
9110 | if (f.file->f_op != &io_uring_fops) | |
9111 | goto out_fput; | |
9112 | ||
9113 | ret = -ENXIO; | |
9114 | ctx = f.file->private_data; | |
9115 | if (!percpu_ref_tryget(&ctx->refs)) | |
9116 | goto out_fput; | |
9117 | ||
7e84e1c7 SG |
9118 | ret = -EBADFD; |
9119 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9120 | goto out; | |
9121 | ||
6c271ce2 JA |
9122 | /* |
9123 | * For SQ polling, the thread will do all submissions and completions. | |
9124 | * Just return the requested submit count, and wake the thread if | |
9125 | * we were asked to. | |
9126 | */ | |
b2a9eada | 9127 | ret = 0; |
6c271ce2 | 9128 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9129 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9130 | |
d9d05217 | 9131 | ret = -EOWNERDEAD; |
04147488 SM |
9132 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9133 | goto out; | |
9134 | } | |
6c271ce2 | 9135 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9136 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9137 | if (flags & IORING_ENTER_SQ_WAIT) { |
9138 | ret = io_sqpoll_wait_sq(ctx); | |
9139 | if (ret) | |
9140 | goto out; | |
9141 | } | |
6c271ce2 | 9142 | submitted = to_submit; |
b2a9eada | 9143 | } else if (to_submit) { |
baf186c4 | 9144 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9145 | if (unlikely(ret)) |
9146 | goto out; | |
2b188cc1 | 9147 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9148 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9149 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9150 | |
9151 | if (submitted != to_submit) | |
9152 | goto out; | |
2b188cc1 JA |
9153 | } |
9154 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9155 | const sigset_t __user *sig; |
9156 | struct __kernel_timespec __user *ts; | |
9157 | ||
9158 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9159 | if (unlikely(ret)) | |
9160 | goto out; | |
9161 | ||
2b188cc1 JA |
9162 | min_complete = min(min_complete, ctx->cq_entries); |
9163 | ||
32b2244a XW |
9164 | /* |
9165 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9166 | * space applications don't need to do io completion events | |
9167 | * polling again, they can rely on io_sq_thread to do polling | |
9168 | * work, which can reduce cpu usage and uring_lock contention. | |
9169 | */ | |
9170 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9171 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9172 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9173 | } else { |
c73ebb68 | 9174 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9175 | } |
2b188cc1 JA |
9176 | } |
9177 | ||
7c504e65 | 9178 | out: |
6805b32e | 9179 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9180 | out_fput: |
9181 | fdput(f); | |
9182 | return submitted ? submitted : ret; | |
9183 | } | |
9184 | ||
bebdb65e | 9185 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9186 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9187 | const struct cred *cred) | |
87ce955b | 9188 | { |
87ce955b JA |
9189 | struct user_namespace *uns = seq_user_ns(m); |
9190 | struct group_info *gi; | |
9191 | kernel_cap_t cap; | |
9192 | unsigned __capi; | |
9193 | int g; | |
9194 | ||
9195 | seq_printf(m, "%5d\n", id); | |
9196 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9197 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9198 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9199 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9200 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9201 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9202 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9203 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9204 | seq_puts(m, "\n\tGroups:\t"); | |
9205 | gi = cred->group_info; | |
9206 | for (g = 0; g < gi->ngroups; g++) { | |
9207 | seq_put_decimal_ull(m, g ? " " : "", | |
9208 | from_kgid_munged(uns, gi->gid[g])); | |
9209 | } | |
9210 | seq_puts(m, "\n\tCapEff:\t"); | |
9211 | cap = cred->cap_effective; | |
9212 | CAP_FOR_EACH_U32(__capi) | |
9213 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9214 | seq_putc(m, '\n'); | |
9215 | return 0; | |
9216 | } | |
9217 | ||
9218 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9219 | { | |
dbbe9c64 | 9220 | struct io_sq_data *sq = NULL; |
fad8e0de | 9221 | bool has_lock; |
87ce955b JA |
9222 | int i; |
9223 | ||
fad8e0de JA |
9224 | /* |
9225 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9226 | * since fdinfo case grabs it in the opposite direction of normal use | |
9227 | * cases. If we fail to get the lock, we just don't iterate any | |
9228 | * structures that could be going away outside the io_uring mutex. | |
9229 | */ | |
9230 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9231 | ||
5f3f26f9 | 9232 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9233 | sq = ctx->sq_data; |
5f3f26f9 JA |
9234 | if (!sq->thread) |
9235 | sq = NULL; | |
9236 | } | |
dbbe9c64 JQ |
9237 | |
9238 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9239 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9240 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9241 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9242 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9243 | |
87ce955b JA |
9244 | if (f) |
9245 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9246 | else | |
9247 | seq_printf(m, "%5u: <none>\n", i); | |
9248 | } | |
9249 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9250 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9251 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9252 | ||
9253 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9254 | (unsigned int) buf->len); | |
9255 | } | |
61cf9370 MWO |
9256 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9257 | unsigned long index; | |
9258 | const struct cred *cred; | |
9259 | ||
87ce955b | 9260 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9261 | xa_for_each(&ctx->personalities, index, cred) |
9262 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9263 | } |
d7718a9d JA |
9264 | seq_printf(m, "PollList:\n"); |
9265 | spin_lock_irq(&ctx->completion_lock); | |
9266 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9267 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9268 | struct io_kiocb *req; | |
9269 | ||
9270 | hlist_for_each_entry(req, list, hash_node) | |
9271 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9272 | req->task->task_works != NULL); | |
9273 | } | |
9274 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9275 | if (has_lock) |
9276 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9277 | } |
9278 | ||
9279 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9280 | { | |
9281 | struct io_ring_ctx *ctx = f->private_data; | |
9282 | ||
9283 | if (percpu_ref_tryget(&ctx->refs)) { | |
9284 | __io_uring_show_fdinfo(ctx, m); | |
9285 | percpu_ref_put(&ctx->refs); | |
9286 | } | |
9287 | } | |
bebdb65e | 9288 | #endif |
87ce955b | 9289 | |
2b188cc1 JA |
9290 | static const struct file_operations io_uring_fops = { |
9291 | .release = io_uring_release, | |
9292 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9293 | #ifndef CONFIG_MMU |
9294 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9295 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9296 | #endif | |
2b188cc1 JA |
9297 | .poll = io_uring_poll, |
9298 | .fasync = io_uring_fasync, | |
bebdb65e | 9299 | #ifdef CONFIG_PROC_FS |
87ce955b | 9300 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9301 | #endif |
2b188cc1 JA |
9302 | }; |
9303 | ||
9304 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9305 | struct io_uring_params *p) | |
9306 | { | |
75b28aff HV |
9307 | struct io_rings *rings; |
9308 | size_t size, sq_array_offset; | |
2b188cc1 | 9309 | |
bd740481 JA |
9310 | /* make sure these are sane, as we already accounted them */ |
9311 | ctx->sq_entries = p->sq_entries; | |
9312 | ctx->cq_entries = p->cq_entries; | |
9313 | ||
75b28aff HV |
9314 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9315 | if (size == SIZE_MAX) | |
9316 | return -EOVERFLOW; | |
9317 | ||
9318 | rings = io_mem_alloc(size); | |
9319 | if (!rings) | |
2b188cc1 JA |
9320 | return -ENOMEM; |
9321 | ||
75b28aff HV |
9322 | ctx->rings = rings; |
9323 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9324 | rings->sq_ring_mask = p->sq_entries - 1; | |
9325 | rings->cq_ring_mask = p->cq_entries - 1; | |
9326 | rings->sq_ring_entries = p->sq_entries; | |
9327 | rings->cq_ring_entries = p->cq_entries; | |
9328 | ctx->sq_mask = rings->sq_ring_mask; | |
9329 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9330 | |
9331 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9332 | if (size == SIZE_MAX) { |
9333 | io_mem_free(ctx->rings); | |
9334 | ctx->rings = NULL; | |
2b188cc1 | 9335 | return -EOVERFLOW; |
eb065d30 | 9336 | } |
2b188cc1 JA |
9337 | |
9338 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9339 | if (!ctx->sq_sqes) { |
9340 | io_mem_free(ctx->rings); | |
9341 | ctx->rings = NULL; | |
2b188cc1 | 9342 | return -ENOMEM; |
eb065d30 | 9343 | } |
2b188cc1 | 9344 | |
2b188cc1 JA |
9345 | return 0; |
9346 | } | |
9347 | ||
9faadcc8 PB |
9348 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9349 | { | |
9350 | int ret, fd; | |
9351 | ||
9352 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9353 | if (fd < 0) | |
9354 | return fd; | |
9355 | ||
baf186c4 | 9356 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9357 | if (ret) { |
9358 | put_unused_fd(fd); | |
9359 | return ret; | |
9360 | } | |
9361 | fd_install(fd, file); | |
9362 | return fd; | |
9363 | } | |
9364 | ||
2b188cc1 JA |
9365 | /* |
9366 | * Allocate an anonymous fd, this is what constitutes the application | |
9367 | * visible backing of an io_uring instance. The application mmaps this | |
9368 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9369 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9370 | */ | |
9faadcc8 | 9371 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9372 | { |
9373 | struct file *file; | |
9faadcc8 | 9374 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9375 | int ret; |
9376 | ||
2b188cc1 JA |
9377 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9378 | &ctx->ring_sock); | |
9379 | if (ret) | |
9faadcc8 | 9380 | return ERR_PTR(ret); |
2b188cc1 JA |
9381 | #endif |
9382 | ||
2b188cc1 JA |
9383 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9384 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9385 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9386 | if (IS_ERR(file)) { |
9387 | sock_release(ctx->ring_sock); | |
9388 | ctx->ring_sock = NULL; | |
9389 | } else { | |
9390 | ctx->ring_sock->file = file; | |
0f212204 | 9391 | } |
2b188cc1 | 9392 | #endif |
9faadcc8 | 9393 | return file; |
2b188cc1 JA |
9394 | } |
9395 | ||
7f13657d XW |
9396 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9397 | struct io_uring_params __user *params) | |
2b188cc1 | 9398 | { |
2b188cc1 | 9399 | struct io_ring_ctx *ctx; |
9faadcc8 | 9400 | struct file *file; |
2b188cc1 JA |
9401 | int ret; |
9402 | ||
8110c1a6 | 9403 | if (!entries) |
2b188cc1 | 9404 | return -EINVAL; |
8110c1a6 JA |
9405 | if (entries > IORING_MAX_ENTRIES) { |
9406 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9407 | return -EINVAL; | |
9408 | entries = IORING_MAX_ENTRIES; | |
9409 | } | |
2b188cc1 JA |
9410 | |
9411 | /* | |
9412 | * Use twice as many entries for the CQ ring. It's possible for the | |
9413 | * application to drive a higher depth than the size of the SQ ring, | |
9414 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9415 | * some flexibility in overcommitting a bit. If the application has |
9416 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9417 | * of CQ ring entries manually. | |
2b188cc1 JA |
9418 | */ |
9419 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9420 | if (p->flags & IORING_SETUP_CQSIZE) { |
9421 | /* | |
9422 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9423 | * to a power-of-two, if it isn't already. We do NOT impose | |
9424 | * any cq vs sq ring sizing. | |
9425 | */ | |
eb2667b3 | 9426 | if (!p->cq_entries) |
33a107f0 | 9427 | return -EINVAL; |
8110c1a6 JA |
9428 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9429 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9430 | return -EINVAL; | |
9431 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9432 | } | |
eb2667b3 JQ |
9433 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9434 | if (p->cq_entries < p->sq_entries) | |
9435 | return -EINVAL; | |
33a107f0 JA |
9436 | } else { |
9437 | p->cq_entries = 2 * p->sq_entries; | |
9438 | } | |
2b188cc1 | 9439 | |
2b188cc1 | 9440 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9441 | if (!ctx) |
2b188cc1 | 9442 | return -ENOMEM; |
2b188cc1 | 9443 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9444 | if (!capable(CAP_IPC_LOCK)) |
9445 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9446 | |
9447 | /* | |
9448 | * This is just grabbed for accounting purposes. When a process exits, | |
9449 | * the mm is exited and dropped before the files, hence we need to hang | |
9450 | * on to this mm purely for the purposes of being able to unaccount | |
9451 | * memory (locked/pinned vm). It's not used for anything else. | |
9452 | */ | |
6b7898eb | 9453 | mmgrab(current->mm); |
2aede0e4 | 9454 | ctx->mm_account = current->mm; |
6b7898eb | 9455 | |
2b188cc1 JA |
9456 | ret = io_allocate_scq_urings(ctx, p); |
9457 | if (ret) | |
9458 | goto err; | |
9459 | ||
7e84e1c7 | 9460 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9461 | if (ret) |
9462 | goto err; | |
9463 | ||
2b188cc1 | 9464 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9465 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9466 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9467 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9468 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9469 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9470 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9471 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9472 | |
9473 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9474 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9475 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9476 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9477 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9478 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9479 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9480 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9481 | |
7f13657d XW |
9482 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9483 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9484 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9485 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9486 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9487 | |
9488 | if (copy_to_user(params, p, sizeof(*p))) { | |
9489 | ret = -EFAULT; | |
9490 | goto err; | |
9491 | } | |
d1719f70 | 9492 | |
9faadcc8 PB |
9493 | file = io_uring_get_file(ctx); |
9494 | if (IS_ERR(file)) { | |
9495 | ret = PTR_ERR(file); | |
9496 | goto err; | |
9497 | } | |
9498 | ||
044c1ab3 JA |
9499 | /* |
9500 | * Install ring fd as the very last thing, so we don't risk someone | |
9501 | * having closed it before we finish setup | |
9502 | */ | |
9faadcc8 PB |
9503 | ret = io_uring_install_fd(ctx, file); |
9504 | if (ret < 0) { | |
9505 | /* fput will clean it up */ | |
9506 | fput(file); | |
9507 | return ret; | |
9508 | } | |
044c1ab3 | 9509 | |
c826bd7a | 9510 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9511 | return ret; |
9512 | err: | |
9513 | io_ring_ctx_wait_and_kill(ctx); | |
9514 | return ret; | |
9515 | } | |
9516 | ||
9517 | /* | |
9518 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9519 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9520 | * params structure passed in. | |
9521 | */ | |
9522 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9523 | { | |
9524 | struct io_uring_params p; | |
2b188cc1 JA |
9525 | int i; |
9526 | ||
9527 | if (copy_from_user(&p, params, sizeof(p))) | |
9528 | return -EFAULT; | |
9529 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9530 | if (p.resv[i]) | |
9531 | return -EINVAL; | |
9532 | } | |
9533 | ||
6c271ce2 | 9534 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9535 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9536 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9537 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9538 | return -EINVAL; |
9539 | ||
7f13657d | 9540 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9541 | } |
9542 | ||
9543 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9544 | struct io_uring_params __user *, params) | |
9545 | { | |
9546 | return io_uring_setup(entries, params); | |
9547 | } | |
9548 | ||
66f4af93 JA |
9549 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9550 | { | |
9551 | struct io_uring_probe *p; | |
9552 | size_t size; | |
9553 | int i, ret; | |
9554 | ||
9555 | size = struct_size(p, ops, nr_args); | |
9556 | if (size == SIZE_MAX) | |
9557 | return -EOVERFLOW; | |
9558 | p = kzalloc(size, GFP_KERNEL); | |
9559 | if (!p) | |
9560 | return -ENOMEM; | |
9561 | ||
9562 | ret = -EFAULT; | |
9563 | if (copy_from_user(p, arg, size)) | |
9564 | goto out; | |
9565 | ret = -EINVAL; | |
9566 | if (memchr_inv(p, 0, size)) | |
9567 | goto out; | |
9568 | ||
9569 | p->last_op = IORING_OP_LAST - 1; | |
9570 | if (nr_args > IORING_OP_LAST) | |
9571 | nr_args = IORING_OP_LAST; | |
9572 | ||
9573 | for (i = 0; i < nr_args; i++) { | |
9574 | p->ops[i].op = i; | |
9575 | if (!io_op_defs[i].not_supported) | |
9576 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9577 | } | |
9578 | p->ops_len = i; | |
9579 | ||
9580 | ret = 0; | |
9581 | if (copy_to_user(arg, p, size)) | |
9582 | ret = -EFAULT; | |
9583 | out: | |
9584 | kfree(p); | |
9585 | return ret; | |
9586 | } | |
9587 | ||
071698e1 JA |
9588 | static int io_register_personality(struct io_ring_ctx *ctx) |
9589 | { | |
4379bf8b | 9590 | const struct cred *creds; |
61cf9370 | 9591 | u32 id; |
1e6fa521 | 9592 | int ret; |
071698e1 | 9593 | |
4379bf8b | 9594 | creds = get_current_cred(); |
1e6fa521 | 9595 | |
61cf9370 MWO |
9596 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9597 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9598 | if (!ret) | |
9599 | return id; | |
9600 | put_cred(creds); | |
1e6fa521 | 9601 | return ret; |
071698e1 JA |
9602 | } |
9603 | ||
21b55dbc SG |
9604 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9605 | unsigned int nr_args) | |
9606 | { | |
9607 | struct io_uring_restriction *res; | |
9608 | size_t size; | |
9609 | int i, ret; | |
9610 | ||
7e84e1c7 SG |
9611 | /* Restrictions allowed only if rings started disabled */ |
9612 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9613 | return -EBADFD; | |
9614 | ||
21b55dbc | 9615 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9616 | if (ctx->restrictions.registered) |
21b55dbc SG |
9617 | return -EBUSY; |
9618 | ||
9619 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9620 | return -EINVAL; | |
9621 | ||
9622 | size = array_size(nr_args, sizeof(*res)); | |
9623 | if (size == SIZE_MAX) | |
9624 | return -EOVERFLOW; | |
9625 | ||
9626 | res = memdup_user(arg, size); | |
9627 | if (IS_ERR(res)) | |
9628 | return PTR_ERR(res); | |
9629 | ||
9630 | ret = 0; | |
9631 | ||
9632 | for (i = 0; i < nr_args; i++) { | |
9633 | switch (res[i].opcode) { | |
9634 | case IORING_RESTRICTION_REGISTER_OP: | |
9635 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9636 | ret = -EINVAL; | |
9637 | goto out; | |
9638 | } | |
9639 | ||
9640 | __set_bit(res[i].register_op, | |
9641 | ctx->restrictions.register_op); | |
9642 | break; | |
9643 | case IORING_RESTRICTION_SQE_OP: | |
9644 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9645 | ret = -EINVAL; | |
9646 | goto out; | |
9647 | } | |
9648 | ||
9649 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9650 | break; | |
9651 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9652 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9653 | break; | |
9654 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9655 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9656 | break; | |
9657 | default: | |
9658 | ret = -EINVAL; | |
9659 | goto out; | |
9660 | } | |
9661 | } | |
9662 | ||
9663 | out: | |
9664 | /* Reset all restrictions if an error happened */ | |
9665 | if (ret != 0) | |
9666 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9667 | else | |
7e84e1c7 | 9668 | ctx->restrictions.registered = true; |
21b55dbc SG |
9669 | |
9670 | kfree(res); | |
9671 | return ret; | |
9672 | } | |
9673 | ||
7e84e1c7 SG |
9674 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9675 | { | |
9676 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9677 | return -EBADFD; | |
9678 | ||
9679 | if (ctx->restrictions.registered) | |
9680 | ctx->restricted = 1; | |
9681 | ||
0298ef96 PB |
9682 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9683 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9684 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9685 | return 0; |
9686 | } | |
9687 | ||
071698e1 JA |
9688 | static bool io_register_op_must_quiesce(int op) |
9689 | { | |
9690 | switch (op) { | |
9691 | case IORING_UNREGISTER_FILES: | |
9692 | case IORING_REGISTER_FILES_UPDATE: | |
9693 | case IORING_REGISTER_PROBE: | |
9694 | case IORING_REGISTER_PERSONALITY: | |
9695 | case IORING_UNREGISTER_PERSONALITY: | |
9696 | return false; | |
9697 | default: | |
9698 | return true; | |
9699 | } | |
9700 | } | |
9701 | ||
edafccee JA |
9702 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9703 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9704 | __releases(ctx->uring_lock) |
9705 | __acquires(ctx->uring_lock) | |
edafccee JA |
9706 | { |
9707 | int ret; | |
9708 | ||
35fa71a0 JA |
9709 | /* |
9710 | * We're inside the ring mutex, if the ref is already dying, then | |
9711 | * someone else killed the ctx or is already going through | |
9712 | * io_uring_register(). | |
9713 | */ | |
9714 | if (percpu_ref_is_dying(&ctx->refs)) | |
9715 | return -ENXIO; | |
9716 | ||
071698e1 | 9717 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9718 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9719 | |
05f3fb3c JA |
9720 | /* |
9721 | * Drop uring mutex before waiting for references to exit. If | |
9722 | * another thread is currently inside io_uring_enter() it might | |
9723 | * need to grab the uring_lock to make progress. If we hold it | |
9724 | * here across the drain wait, then we can deadlock. It's safe | |
9725 | * to drop the mutex here, since no new references will come in | |
9726 | * after we've killed the percpu ref. | |
9727 | */ | |
9728 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9729 | do { |
9730 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9731 | if (!ret) | |
9732 | break; | |
ed6930c9 JA |
9733 | ret = io_run_task_work_sig(); |
9734 | if (ret < 0) | |
9735 | break; | |
af9c1a44 JA |
9736 | } while (1); |
9737 | ||
05f3fb3c | 9738 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9739 | |
c150368b JA |
9740 | if (ret) { |
9741 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9742 | goto out_quiesce; |
9743 | } | |
9744 | } | |
9745 | ||
9746 | if (ctx->restricted) { | |
9747 | if (opcode >= IORING_REGISTER_LAST) { | |
9748 | ret = -EINVAL; | |
9749 | goto out; | |
9750 | } | |
9751 | ||
9752 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9753 | ret = -EACCES; | |
c150368b JA |
9754 | goto out; |
9755 | } | |
05f3fb3c | 9756 | } |
edafccee JA |
9757 | |
9758 | switch (opcode) { | |
9759 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9760 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9761 | break; |
9762 | case IORING_UNREGISTER_BUFFERS: | |
9763 | ret = -EINVAL; | |
9764 | if (arg || nr_args) | |
9765 | break; | |
0a96bbe4 | 9766 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9767 | break; |
6b06314c JA |
9768 | case IORING_REGISTER_FILES: |
9769 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9770 | break; | |
9771 | case IORING_UNREGISTER_FILES: | |
9772 | ret = -EINVAL; | |
9773 | if (arg || nr_args) | |
9774 | break; | |
9775 | ret = io_sqe_files_unregister(ctx); | |
9776 | break; | |
c3a31e60 JA |
9777 | case IORING_REGISTER_FILES_UPDATE: |
9778 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9779 | break; | |
9b402849 | 9780 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9781 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9782 | ret = -EINVAL; |
9783 | if (nr_args != 1) | |
9784 | break; | |
9785 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9786 | if (ret) |
9787 | break; | |
9788 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9789 | ctx->eventfd_async = 1; | |
9790 | else | |
9791 | ctx->eventfd_async = 0; | |
9b402849 JA |
9792 | break; |
9793 | case IORING_UNREGISTER_EVENTFD: | |
9794 | ret = -EINVAL; | |
9795 | if (arg || nr_args) | |
9796 | break; | |
9797 | ret = io_eventfd_unregister(ctx); | |
9798 | break; | |
66f4af93 JA |
9799 | case IORING_REGISTER_PROBE: |
9800 | ret = -EINVAL; | |
9801 | if (!arg || nr_args > 256) | |
9802 | break; | |
9803 | ret = io_probe(ctx, arg, nr_args); | |
9804 | break; | |
071698e1 JA |
9805 | case IORING_REGISTER_PERSONALITY: |
9806 | ret = -EINVAL; | |
9807 | if (arg || nr_args) | |
9808 | break; | |
9809 | ret = io_register_personality(ctx); | |
9810 | break; | |
9811 | case IORING_UNREGISTER_PERSONALITY: | |
9812 | ret = -EINVAL; | |
9813 | if (arg) | |
9814 | break; | |
9815 | ret = io_unregister_personality(ctx, nr_args); | |
9816 | break; | |
7e84e1c7 SG |
9817 | case IORING_REGISTER_ENABLE_RINGS: |
9818 | ret = -EINVAL; | |
9819 | if (arg || nr_args) | |
9820 | break; | |
9821 | ret = io_register_enable_rings(ctx); | |
9822 | break; | |
21b55dbc SG |
9823 | case IORING_REGISTER_RESTRICTIONS: |
9824 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9825 | break; | |
edafccee JA |
9826 | default: |
9827 | ret = -EINVAL; | |
9828 | break; | |
9829 | } | |
9830 | ||
21b55dbc | 9831 | out: |
071698e1 | 9832 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9833 | /* bring the ctx back to life */ |
05f3fb3c | 9834 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9835 | out_quiesce: |
0f158b4c | 9836 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9837 | } |
edafccee JA |
9838 | return ret; |
9839 | } | |
9840 | ||
9841 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9842 | void __user *, arg, unsigned int, nr_args) | |
9843 | { | |
9844 | struct io_ring_ctx *ctx; | |
9845 | long ret = -EBADF; | |
9846 | struct fd f; | |
9847 | ||
9848 | f = fdget(fd); | |
9849 | if (!f.file) | |
9850 | return -EBADF; | |
9851 | ||
9852 | ret = -EOPNOTSUPP; | |
9853 | if (f.file->f_op != &io_uring_fops) | |
9854 | goto out_fput; | |
9855 | ||
9856 | ctx = f.file->private_data; | |
9857 | ||
b6c23dd5 PB |
9858 | io_run_task_work(); |
9859 | ||
edafccee JA |
9860 | mutex_lock(&ctx->uring_lock); |
9861 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9862 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9863 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9864 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9865 | out_fput: |
9866 | fdput(f); | |
9867 | return ret; | |
9868 | } | |
9869 | ||
2b188cc1 JA |
9870 | static int __init io_uring_init(void) |
9871 | { | |
d7f62e82 SM |
9872 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9873 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9874 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9875 | } while (0) | |
9876 | ||
9877 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9878 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9879 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9880 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9881 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9882 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9883 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9884 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9885 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9886 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9887 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9888 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9889 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9890 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9891 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9892 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9893 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9894 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9895 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9896 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9897 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9898 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9899 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9900 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9901 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9902 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9903 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9904 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9905 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9906 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9907 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9908 | |
d3656344 | 9909 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9910 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9911 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9912 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9913 | return 0; |
9914 | }; | |
9915 | __initcall(io_uring_init); |