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> | |
6c271ce2 | 60 | #include <linux/kthread.h> |
2b188cc1 | 61 | #include <linux/blkdev.h> |
edafccee | 62 | #include <linux/bvec.h> |
2b188cc1 JA |
63 | #include <linux/net.h> |
64 | #include <net/sock.h> | |
65 | #include <net/af_unix.h> | |
6b06314c | 66 | #include <net/scm.h> |
2b188cc1 JA |
67 | #include <linux/anon_inodes.h> |
68 | #include <linux/sched/mm.h> | |
69 | #include <linux/uaccess.h> | |
70 | #include <linux/nospec.h> | |
edafccee JA |
71 | #include <linux/sizes.h> |
72 | #include <linux/hugetlb.h> | |
aa4c3967 | 73 | #include <linux/highmem.h> |
15b71abe JA |
74 | #include <linux/namei.h> |
75 | #include <linux/fsnotify.h> | |
4840e418 | 76 | #include <linux/fadvise.h> |
3e4827b0 | 77 | #include <linux/eventpoll.h> |
ff002b30 | 78 | #include <linux/fs_struct.h> |
7d67af2c | 79 | #include <linux/splice.h> |
b41e9852 | 80 | #include <linux/task_work.h> |
bcf5a063 | 81 | #include <linux/pagemap.h> |
0f212204 | 82 | #include <linux/io_uring.h> |
91d8f519 | 83 | #include <linux/blk-cgroup.h> |
4ea33a97 | 84 | #include <linux/audit.h> |
2b188cc1 | 85 | |
c826bd7a DD |
86 | #define CREATE_TRACE_POINTS |
87 | #include <trace/events/io_uring.h> | |
88 | ||
2b188cc1 JA |
89 | #include <uapi/linux/io_uring.h> |
90 | ||
91 | #include "internal.h" | |
561fb04a | 92 | #include "io-wq.h" |
2b188cc1 | 93 | |
5277deaa | 94 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 95 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
96 | |
97 | /* | |
98 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
99 | */ | |
100 | #define IORING_FILE_TABLE_SHIFT 9 | |
101 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
102 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
103 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
104 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
105 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 106 | |
b16fed66 PB |
107 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
108 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
109 | IOSQE_BUFFER_SELECT) | |
110 | ||
2b188cc1 JA |
111 | struct io_uring { |
112 | u32 head ____cacheline_aligned_in_smp; | |
113 | u32 tail ____cacheline_aligned_in_smp; | |
114 | }; | |
115 | ||
1e84b97b | 116 | /* |
75b28aff HV |
117 | * This data is shared with the application through the mmap at offsets |
118 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
119 | * |
120 | * The offsets to the member fields are published through struct | |
121 | * io_sqring_offsets when calling io_uring_setup. | |
122 | */ | |
75b28aff | 123 | struct io_rings { |
1e84b97b SB |
124 | /* |
125 | * Head and tail offsets into the ring; the offsets need to be | |
126 | * masked to get valid indices. | |
127 | * | |
75b28aff HV |
128 | * The kernel controls head of the sq ring and the tail of the cq ring, |
129 | * and the application controls tail of the sq ring and the head of the | |
130 | * cq ring. | |
1e84b97b | 131 | */ |
75b28aff | 132 | struct io_uring sq, cq; |
1e84b97b | 133 | /* |
75b28aff | 134 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
135 | * ring_entries - 1) |
136 | */ | |
75b28aff HV |
137 | u32 sq_ring_mask, cq_ring_mask; |
138 | /* Ring sizes (constant, power of 2) */ | |
139 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
140 | /* |
141 | * Number of invalid entries dropped by the kernel due to | |
142 | * invalid index stored in array | |
143 | * | |
144 | * Written by the kernel, shouldn't be modified by the | |
145 | * application (i.e. get number of "new events" by comparing to | |
146 | * cached value). | |
147 | * | |
148 | * After a new SQ head value was read by the application this | |
149 | * counter includes all submissions that were dropped reaching | |
150 | * the new SQ head (and possibly more). | |
151 | */ | |
75b28aff | 152 | u32 sq_dropped; |
1e84b97b | 153 | /* |
0d9b5b3a | 154 | * Runtime SQ flags |
1e84b97b SB |
155 | * |
156 | * Written by the kernel, shouldn't be modified by the | |
157 | * application. | |
158 | * | |
159 | * The application needs a full memory barrier before checking | |
160 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
161 | */ | |
75b28aff | 162 | u32 sq_flags; |
0d9b5b3a SG |
163 | /* |
164 | * Runtime CQ flags | |
165 | * | |
166 | * Written by the application, shouldn't be modified by the | |
167 | * kernel. | |
168 | */ | |
169 | u32 cq_flags; | |
1e84b97b SB |
170 | /* |
171 | * Number of completion events lost because the queue was full; | |
172 | * this should be avoided by the application by making sure | |
0b4295b5 | 173 | * there are not more requests pending than there is space in |
1e84b97b SB |
174 | * the completion queue. |
175 | * | |
176 | * Written by the kernel, shouldn't be modified by the | |
177 | * application (i.e. get number of "new events" by comparing to | |
178 | * cached value). | |
179 | * | |
180 | * As completion events come in out of order this counter is not | |
181 | * ordered with any other data. | |
182 | */ | |
75b28aff | 183 | u32 cq_overflow; |
1e84b97b SB |
184 | /* |
185 | * Ring buffer of completion events. | |
186 | * | |
187 | * The kernel writes completion events fresh every time they are | |
188 | * produced, so the application is allowed to modify pending | |
189 | * entries. | |
190 | */ | |
75b28aff | 191 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
192 | }; |
193 | ||
45d189c6 PB |
194 | enum io_uring_cmd_flags { |
195 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 196 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
197 | }; |
198 | ||
edafccee JA |
199 | struct io_mapped_ubuf { |
200 | u64 ubuf; | |
201 | size_t len; | |
202 | struct bio_vec *bvec; | |
203 | unsigned int nr_bvecs; | |
de293938 | 204 | unsigned long acct_pages; |
edafccee JA |
205 | }; |
206 | ||
50238531 BM |
207 | struct io_ring_ctx; |
208 | ||
269bbe5f BM |
209 | struct io_rsrc_put { |
210 | struct list_head list; | |
50238531 BM |
211 | union { |
212 | void *rsrc; | |
213 | struct file *file; | |
214 | }; | |
269bbe5f BM |
215 | }; |
216 | ||
217 | struct fixed_rsrc_table { | |
65e19f54 | 218 | struct file **files; |
31b51510 JA |
219 | }; |
220 | ||
269bbe5f | 221 | struct fixed_rsrc_ref_node { |
05589553 XW |
222 | struct percpu_ref refs; |
223 | struct list_head node; | |
269bbe5f BM |
224 | struct list_head rsrc_list; |
225 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
226 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
227 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 228 | struct llist_node llist; |
e297822b | 229 | bool done; |
05589553 XW |
230 | }; |
231 | ||
269bbe5f BM |
232 | struct fixed_rsrc_data { |
233 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
234 | struct io_ring_ctx *ctx; |
235 | ||
269bbe5f | 236 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 237 | struct percpu_ref refs; |
05f3fb3c JA |
238 | struct completion done; |
239 | }; | |
240 | ||
5a2e745d JA |
241 | struct io_buffer { |
242 | struct list_head list; | |
243 | __u64 addr; | |
244 | __s32 len; | |
245 | __u16 bid; | |
246 | }; | |
247 | ||
21b55dbc SG |
248 | struct io_restriction { |
249 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
250 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
251 | u8 sqe_flags_allowed; | |
252 | u8 sqe_flags_required; | |
7e84e1c7 | 253 | bool registered; |
21b55dbc SG |
254 | }; |
255 | ||
534ca6d6 JA |
256 | struct io_sq_data { |
257 | refcount_t refs; | |
69fb2131 JA |
258 | struct mutex lock; |
259 | ||
260 | /* ctx's that are using this sqd */ | |
261 | struct list_head ctx_list; | |
262 | struct list_head ctx_new_list; | |
263 | struct mutex ctx_lock; | |
264 | ||
534ca6d6 JA |
265 | struct task_struct *thread; |
266 | struct wait_queue_head wait; | |
08369246 XW |
267 | |
268 | unsigned sq_thread_idle; | |
534ca6d6 JA |
269 | }; |
270 | ||
258b29a9 | 271 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 272 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 273 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 274 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
275 | |
276 | struct io_comp_state { | |
6dd0be1e | 277 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 278 | unsigned int nr; |
c7dae4ba JA |
279 | unsigned int locked_free_nr; |
280 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 281 | struct list_head free_list; |
c7dae4ba JA |
282 | /* IRQ completion list, under ->completion_lock */ |
283 | struct list_head locked_free_list; | |
258b29a9 PB |
284 | }; |
285 | ||
a1ab7b35 PB |
286 | struct io_submit_link { |
287 | struct io_kiocb *head; | |
288 | struct io_kiocb *last; | |
289 | }; | |
290 | ||
258b29a9 PB |
291 | struct io_submit_state { |
292 | struct blk_plug plug; | |
a1ab7b35 | 293 | struct io_submit_link link; |
258b29a9 PB |
294 | |
295 | /* | |
296 | * io_kiocb alloc cache | |
297 | */ | |
bf019da7 | 298 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
299 | unsigned int free_reqs; |
300 | ||
301 | bool plug_started; | |
302 | ||
303 | /* | |
304 | * Batch completion logic | |
305 | */ | |
306 | struct io_comp_state comp; | |
307 | ||
308 | /* | |
309 | * File reference cache | |
310 | */ | |
311 | struct file *file; | |
312 | unsigned int fd; | |
313 | unsigned int file_refs; | |
314 | unsigned int ios_left; | |
315 | }; | |
316 | ||
2b188cc1 JA |
317 | struct io_ring_ctx { |
318 | struct { | |
319 | struct percpu_ref refs; | |
320 | } ____cacheline_aligned_in_smp; | |
321 | ||
322 | struct { | |
323 | unsigned int flags; | |
e1d85334 | 324 | unsigned int compat: 1; |
aad5d8da | 325 | unsigned int limit_mem: 1; |
e1d85334 RD |
326 | unsigned int cq_overflow_flushed: 1; |
327 | unsigned int drain_next: 1; | |
328 | unsigned int eventfd_async: 1; | |
21b55dbc | 329 | unsigned int restricted: 1; |
d9d05217 | 330 | unsigned int sqo_dead: 1; |
2b188cc1 | 331 | |
75b28aff HV |
332 | /* |
333 | * Ring buffer of indices into array of io_uring_sqe, which is | |
334 | * mmapped by the application using the IORING_OFF_SQES offset. | |
335 | * | |
336 | * This indirection could e.g. be used to assign fixed | |
337 | * io_uring_sqe entries to operations and only submit them to | |
338 | * the queue when needed. | |
339 | * | |
340 | * The kernel modifies neither the indices array nor the entries | |
341 | * array. | |
342 | */ | |
343 | u32 *sq_array; | |
2b188cc1 JA |
344 | unsigned cached_sq_head; |
345 | unsigned sq_entries; | |
346 | unsigned sq_mask; | |
6c271ce2 | 347 | unsigned sq_thread_idle; |
498ccd9e | 348 | unsigned cached_sq_dropped; |
2c3bac6d | 349 | unsigned cached_cq_overflow; |
ad3eb2c8 | 350 | unsigned long sq_check_overflow; |
de0617e4 JA |
351 | |
352 | struct list_head defer_list; | |
5262f567 | 353 | struct list_head timeout_list; |
1d7bb1d5 | 354 | struct list_head cq_overflow_list; |
fcb323cc | 355 | |
ad3eb2c8 | 356 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
357 | } ____cacheline_aligned_in_smp; |
358 | ||
3c1a2ead JA |
359 | struct { |
360 | struct mutex uring_lock; | |
361 | wait_queue_head_t wait; | |
362 | } ____cacheline_aligned_in_smp; | |
363 | ||
364 | struct io_submit_state submit_state; | |
365 | ||
206aefde JA |
366 | struct io_rings *rings; |
367 | ||
2b188cc1 | 368 | /* IO offload */ |
561fb04a | 369 | struct io_wq *io_wq; |
2aede0e4 JA |
370 | |
371 | /* | |
372 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
373 | * have access to the ->files | |
374 | */ | |
375 | struct task_struct *sqo_task; | |
376 | ||
377 | /* Only used for accounting purposes */ | |
378 | struct mm_struct *mm_account; | |
379 | ||
91d8f519 DZ |
380 | #ifdef CONFIG_BLK_CGROUP |
381 | struct cgroup_subsys_state *sqo_blkcg_css; | |
382 | #endif | |
383 | ||
534ca6d6 JA |
384 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
385 | ||
90554200 | 386 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 387 | struct list_head sqd_list; |
75b28aff | 388 | |
6b06314c JA |
389 | /* |
390 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
391 | * readers must ensure that ->refs is alive as long as the file* is | |
392 | * used. Only updated through io_uring_register(2). | |
393 | */ | |
269bbe5f | 394 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
395 | unsigned nr_user_files; |
396 | ||
edafccee JA |
397 | /* if used, fixed mapped user buffers */ |
398 | unsigned nr_user_bufs; | |
399 | struct io_mapped_ubuf *user_bufs; | |
400 | ||
2b188cc1 JA |
401 | struct user_struct *user; |
402 | ||
0b8c0ec7 | 403 | const struct cred *creds; |
181e448d | 404 | |
4ea33a97 JA |
405 | #ifdef CONFIG_AUDIT |
406 | kuid_t loginuid; | |
407 | unsigned int sessionid; | |
408 | #endif | |
409 | ||
0f158b4c JA |
410 | struct completion ref_comp; |
411 | struct completion sq_thread_comp; | |
206aefde JA |
412 | |
413 | #if defined(CONFIG_UNIX) | |
414 | struct socket *ring_sock; | |
415 | #endif | |
416 | ||
5a2e745d JA |
417 | struct idr io_buffer_idr; |
418 | ||
071698e1 JA |
419 | struct idr personality_idr; |
420 | ||
206aefde JA |
421 | struct { |
422 | unsigned cached_cq_tail; | |
423 | unsigned cq_entries; | |
424 | unsigned cq_mask; | |
425 | atomic_t cq_timeouts; | |
f010505b | 426 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 427 | unsigned long cq_check_overflow; |
206aefde JA |
428 | struct wait_queue_head cq_wait; |
429 | struct fasync_struct *cq_fasync; | |
430 | struct eventfd_ctx *cq_ev_fd; | |
431 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 432 | |
2b188cc1 JA |
433 | struct { |
434 | spinlock_t completion_lock; | |
e94f141b | 435 | |
def596e9 | 436 | /* |
540e32a0 | 437 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
438 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
439 | * For SQPOLL, only the single threaded io_sq_thread() will | |
440 | * manipulate the list, hence no extra locking is needed there. | |
441 | */ | |
540e32a0 | 442 | struct list_head iopoll_list; |
78076bb6 JA |
443 | struct hlist_head *cancel_hash; |
444 | unsigned cancel_hash_bits; | |
e94f141b | 445 | bool poll_multi_file; |
31b51510 | 446 | |
fcb323cc JA |
447 | spinlock_t inflight_lock; |
448 | struct list_head inflight_list; | |
2b188cc1 | 449 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 450 | |
269bbe5f BM |
451 | struct delayed_work rsrc_put_work; |
452 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
453 | struct list_head rsrc_ref_list; |
454 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 455 | |
21b55dbc | 456 | struct io_restriction restrictions; |
3c1a2ead JA |
457 | |
458 | /* Keep this last, we don't need it for the fast path */ | |
459 | struct work_struct exit_work; | |
2b188cc1 JA |
460 | }; |
461 | ||
09bb8394 JA |
462 | /* |
463 | * First field must be the file pointer in all the | |
464 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
465 | */ | |
221c5eb2 JA |
466 | struct io_poll_iocb { |
467 | struct file *file; | |
018043be | 468 | struct wait_queue_head *head; |
221c5eb2 | 469 | __poll_t events; |
8c838788 | 470 | bool done; |
221c5eb2 | 471 | bool canceled; |
392edb45 | 472 | struct wait_queue_entry wait; |
221c5eb2 JA |
473 | }; |
474 | ||
018043be PB |
475 | struct io_poll_remove { |
476 | struct file *file; | |
477 | u64 addr; | |
478 | }; | |
479 | ||
b5dba59e JA |
480 | struct io_close { |
481 | struct file *file; | |
b5dba59e JA |
482 | int fd; |
483 | }; | |
484 | ||
ad8a48ac JA |
485 | struct io_timeout_data { |
486 | struct io_kiocb *req; | |
487 | struct hrtimer timer; | |
488 | struct timespec64 ts; | |
489 | enum hrtimer_mode mode; | |
490 | }; | |
491 | ||
8ed8d3c3 JA |
492 | struct io_accept { |
493 | struct file *file; | |
494 | struct sockaddr __user *addr; | |
495 | int __user *addr_len; | |
496 | int flags; | |
09952e3e | 497 | unsigned long nofile; |
8ed8d3c3 JA |
498 | }; |
499 | ||
500 | struct io_sync { | |
501 | struct file *file; | |
502 | loff_t len; | |
503 | loff_t off; | |
504 | int flags; | |
d63d1b5e | 505 | int mode; |
8ed8d3c3 JA |
506 | }; |
507 | ||
fbf23849 JA |
508 | struct io_cancel { |
509 | struct file *file; | |
510 | u64 addr; | |
511 | }; | |
512 | ||
b29472ee JA |
513 | struct io_timeout { |
514 | struct file *file; | |
bfe68a22 PB |
515 | u32 off; |
516 | u32 target_seq; | |
135fcde8 | 517 | struct list_head list; |
90cd7e42 PB |
518 | /* head of the link, used by linked timeouts only */ |
519 | struct io_kiocb *head; | |
b29472ee JA |
520 | }; |
521 | ||
0bdf7a2d PB |
522 | struct io_timeout_rem { |
523 | struct file *file; | |
524 | u64 addr; | |
9c8e11b3 PB |
525 | |
526 | /* timeout update */ | |
527 | struct timespec64 ts; | |
528 | u32 flags; | |
0bdf7a2d PB |
529 | }; |
530 | ||
9adbd45d JA |
531 | struct io_rw { |
532 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
533 | struct kiocb kiocb; | |
534 | u64 addr; | |
535 | u64 len; | |
536 | }; | |
537 | ||
3fbb51c1 JA |
538 | struct io_connect { |
539 | struct file *file; | |
540 | struct sockaddr __user *addr; | |
541 | int addr_len; | |
542 | }; | |
543 | ||
e47293fd JA |
544 | struct io_sr_msg { |
545 | struct file *file; | |
fddaface | 546 | union { |
270a5940 | 547 | struct user_msghdr __user *umsg; |
fddaface JA |
548 | void __user *buf; |
549 | }; | |
e47293fd | 550 | int msg_flags; |
bcda7baa | 551 | int bgid; |
fddaface | 552 | size_t len; |
bcda7baa | 553 | struct io_buffer *kbuf; |
e47293fd JA |
554 | }; |
555 | ||
15b71abe JA |
556 | struct io_open { |
557 | struct file *file; | |
558 | int dfd; | |
15b71abe | 559 | struct filename *filename; |
c12cedf2 | 560 | struct open_how how; |
4022e7af | 561 | unsigned long nofile; |
15b71abe JA |
562 | }; |
563 | ||
269bbe5f | 564 | struct io_rsrc_update { |
05f3fb3c JA |
565 | struct file *file; |
566 | u64 arg; | |
567 | u32 nr_args; | |
568 | u32 offset; | |
569 | }; | |
570 | ||
4840e418 JA |
571 | struct io_fadvise { |
572 | struct file *file; | |
573 | u64 offset; | |
574 | u32 len; | |
575 | u32 advice; | |
576 | }; | |
577 | ||
c1ca757b JA |
578 | struct io_madvise { |
579 | struct file *file; | |
580 | u64 addr; | |
581 | u32 len; | |
582 | u32 advice; | |
583 | }; | |
584 | ||
3e4827b0 JA |
585 | struct io_epoll { |
586 | struct file *file; | |
587 | int epfd; | |
588 | int op; | |
589 | int fd; | |
590 | struct epoll_event event; | |
e47293fd JA |
591 | }; |
592 | ||
7d67af2c PB |
593 | struct io_splice { |
594 | struct file *file_out; | |
595 | struct file *file_in; | |
596 | loff_t off_out; | |
597 | loff_t off_in; | |
598 | u64 len; | |
599 | unsigned int flags; | |
600 | }; | |
601 | ||
ddf0322d JA |
602 | struct io_provide_buf { |
603 | struct file *file; | |
604 | __u64 addr; | |
605 | __s32 len; | |
606 | __u32 bgid; | |
607 | __u16 nbufs; | |
608 | __u16 bid; | |
609 | }; | |
610 | ||
1d9e1288 BM |
611 | struct io_statx { |
612 | struct file *file; | |
613 | int dfd; | |
614 | unsigned int mask; | |
615 | unsigned int flags; | |
e62753e4 | 616 | const char __user *filename; |
1d9e1288 BM |
617 | struct statx __user *buffer; |
618 | }; | |
619 | ||
36f4fa68 JA |
620 | struct io_shutdown { |
621 | struct file *file; | |
622 | int how; | |
623 | }; | |
624 | ||
80a261fd JA |
625 | struct io_rename { |
626 | struct file *file; | |
627 | int old_dfd; | |
628 | int new_dfd; | |
629 | struct filename *oldpath; | |
630 | struct filename *newpath; | |
631 | int flags; | |
632 | }; | |
633 | ||
14a1143b JA |
634 | struct io_unlink { |
635 | struct file *file; | |
636 | int dfd; | |
637 | int flags; | |
638 | struct filename *filename; | |
639 | }; | |
640 | ||
3ca405eb PB |
641 | struct io_completion { |
642 | struct file *file; | |
643 | struct list_head list; | |
0f7e466b | 644 | int cflags; |
3ca405eb PB |
645 | }; |
646 | ||
f499a021 JA |
647 | struct io_async_connect { |
648 | struct sockaddr_storage address; | |
649 | }; | |
650 | ||
03b1230c JA |
651 | struct io_async_msghdr { |
652 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
653 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
654 | struct iovec *free_iov; | |
03b1230c JA |
655 | struct sockaddr __user *uaddr; |
656 | struct msghdr msg; | |
b537916c | 657 | struct sockaddr_storage addr; |
03b1230c JA |
658 | }; |
659 | ||
f67676d1 JA |
660 | struct io_async_rw { |
661 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
662 | const struct iovec *free_iovec; |
663 | struct iov_iter iter; | |
227c0c96 | 664 | size_t bytes_done; |
bcf5a063 | 665 | struct wait_page_queue wpq; |
f67676d1 JA |
666 | }; |
667 | ||
6b47ee6e PB |
668 | enum { |
669 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
670 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
671 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
672 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
673 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 674 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 675 | |
6b47ee6e PB |
676 | REQ_F_FAIL_LINK_BIT, |
677 | REQ_F_INFLIGHT_BIT, | |
678 | REQ_F_CUR_POS_BIT, | |
679 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 680 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 681 | REQ_F_ISREG_BIT, |
99bc4c38 | 682 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 683 | REQ_F_POLLED_BIT, |
bcda7baa | 684 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 685 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 686 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 687 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 688 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
689 | |
690 | /* not a real bit, just to check we're not overflowing the space */ | |
691 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
692 | }; |
693 | ||
694 | enum { | |
695 | /* ctx owns file */ | |
696 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
697 | /* drain existing IO first */ | |
698 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
699 | /* linked sqes */ | |
700 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
701 | /* doesn't sever on completion < 0 */ | |
702 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
703 | /* IOSQE_ASYNC */ | |
704 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
705 | /* IOSQE_BUFFER_SELECT */ |
706 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 707 | |
6b47ee6e PB |
708 | /* fail rest of links */ |
709 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
710 | /* on inflight list */ | |
711 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
712 | /* read/write uses file position */ | |
713 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
714 | /* must not punt to workers */ | |
715 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 716 | /* has or had linked timeout */ |
6b47ee6e | 717 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
718 | /* regular file */ |
719 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
720 | /* needs cleanup */ |
721 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
722 | /* already went through poll handler */ |
723 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
724 | /* buffer already selected */ |
725 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
726 | /* doesn't need file table for this request */ |
727 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
728 | /* io_wq_work is initialized */ |
729 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
730 | /* linked timeout is active, i.e. prepared by link's head */ |
731 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
732 | /* completion is deferred through io_comp_state */ |
733 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
734 | }; |
735 | ||
736 | struct async_poll { | |
737 | struct io_poll_iocb poll; | |
807abcb0 | 738 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
739 | }; |
740 | ||
7cbf1722 JA |
741 | struct io_task_work { |
742 | struct io_wq_work_node node; | |
743 | task_work_func_t func; | |
744 | }; | |
745 | ||
09bb8394 JA |
746 | /* |
747 | * NOTE! Each of the iocb union members has the file pointer | |
748 | * as the first entry in their struct definition. So you can | |
749 | * access the file pointer through any of the sub-structs, | |
750 | * or directly as just 'ki_filp' in this struct. | |
751 | */ | |
2b188cc1 | 752 | struct io_kiocb { |
221c5eb2 | 753 | union { |
09bb8394 | 754 | struct file *file; |
9adbd45d | 755 | struct io_rw rw; |
221c5eb2 | 756 | struct io_poll_iocb poll; |
018043be | 757 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
758 | struct io_accept accept; |
759 | struct io_sync sync; | |
fbf23849 | 760 | struct io_cancel cancel; |
b29472ee | 761 | struct io_timeout timeout; |
0bdf7a2d | 762 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 763 | struct io_connect connect; |
e47293fd | 764 | struct io_sr_msg sr_msg; |
15b71abe | 765 | struct io_open open; |
b5dba59e | 766 | struct io_close close; |
269bbe5f | 767 | struct io_rsrc_update rsrc_update; |
4840e418 | 768 | struct io_fadvise fadvise; |
c1ca757b | 769 | struct io_madvise madvise; |
3e4827b0 | 770 | struct io_epoll epoll; |
7d67af2c | 771 | struct io_splice splice; |
ddf0322d | 772 | struct io_provide_buf pbuf; |
1d9e1288 | 773 | struct io_statx statx; |
36f4fa68 | 774 | struct io_shutdown shutdown; |
80a261fd | 775 | struct io_rename rename; |
14a1143b | 776 | struct io_unlink unlink; |
3ca405eb PB |
777 | /* use only after cleaning per-op data, see io_clean_op() */ |
778 | struct io_completion compl; | |
221c5eb2 | 779 | }; |
2b188cc1 | 780 | |
e8c2bc1f JA |
781 | /* opcode allocated if it needs to store data for async defer */ |
782 | void *async_data; | |
d625c6ee | 783 | u8 opcode; |
65a6543d XW |
784 | /* polled IO has completed */ |
785 | u8 iopoll_completed; | |
2b188cc1 | 786 | |
4f4eeba8 | 787 | u16 buf_index; |
9cf7c104 | 788 | u32 result; |
4f4eeba8 | 789 | |
010e8e6b PB |
790 | struct io_ring_ctx *ctx; |
791 | unsigned int flags; | |
792 | refcount_t refs; | |
793 | struct task_struct *task; | |
794 | u64 user_data; | |
d7718a9d | 795 | |
f2f87370 | 796 | struct io_kiocb *link; |
269bbe5f | 797 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 798 | |
d21ffe7e PB |
799 | /* |
800 | * 1. used with ctx->iopoll_list with reads/writes | |
801 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
802 | */ | |
010e8e6b | 803 | struct list_head inflight_entry; |
7cbf1722 JA |
804 | union { |
805 | struct io_task_work io_task_work; | |
806 | struct callback_head task_work; | |
807 | }; | |
010e8e6b PB |
808 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
809 | struct hlist_node hash_node; | |
810 | struct async_poll *apoll; | |
811 | struct io_wq_work work; | |
2b188cc1 | 812 | }; |
05589553 | 813 | |
27dc8338 PB |
814 | struct io_defer_entry { |
815 | struct list_head list; | |
816 | struct io_kiocb *req; | |
9cf7c104 | 817 | u32 seq; |
2b188cc1 JA |
818 | }; |
819 | ||
d3656344 | 820 | struct io_op_def { |
d3656344 JA |
821 | /* needs req->file assigned */ |
822 | unsigned needs_file : 1; | |
d3656344 JA |
823 | /* hash wq insertion if file is a regular file */ |
824 | unsigned hash_reg_file : 1; | |
825 | /* unbound wq insertion if file is a non-regular file */ | |
826 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
827 | /* opcode is not supported by this kernel */ |
828 | unsigned not_supported : 1; | |
8a72758c JA |
829 | /* set if opcode supports polled "wait" */ |
830 | unsigned pollin : 1; | |
831 | unsigned pollout : 1; | |
bcda7baa JA |
832 | /* op supports buffer selection */ |
833 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
834 | /* must always have async data allocated */ |
835 | unsigned needs_async_data : 1; | |
27926b68 JA |
836 | /* should block plug */ |
837 | unsigned plug : 1; | |
e8c2bc1f JA |
838 | /* size of async data needed, if any */ |
839 | unsigned short async_size; | |
0f203765 | 840 | unsigned work_flags; |
d3656344 JA |
841 | }; |
842 | ||
0918682b | 843 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
844 | [IORING_OP_NOP] = {}, |
845 | [IORING_OP_READV] = { | |
d3656344 JA |
846 | .needs_file = 1, |
847 | .unbound_nonreg_file = 1, | |
8a72758c | 848 | .pollin = 1, |
4d954c25 | 849 | .buffer_select = 1, |
e8c2bc1f | 850 | .needs_async_data = 1, |
27926b68 | 851 | .plug = 1, |
e8c2bc1f | 852 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 853 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 854 | }, |
0463b6c5 | 855 | [IORING_OP_WRITEV] = { |
d3656344 JA |
856 | .needs_file = 1, |
857 | .hash_reg_file = 1, | |
858 | .unbound_nonreg_file = 1, | |
8a72758c | 859 | .pollout = 1, |
e8c2bc1f | 860 | .needs_async_data = 1, |
27926b68 | 861 | .plug = 1, |
e8c2bc1f | 862 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
863 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
864 | IO_WQ_WORK_FSIZE, | |
d3656344 | 865 | }, |
0463b6c5 | 866 | [IORING_OP_FSYNC] = { |
d3656344 | 867 | .needs_file = 1, |
0f203765 | 868 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 869 | }, |
0463b6c5 | 870 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
871 | .needs_file = 1, |
872 | .unbound_nonreg_file = 1, | |
8a72758c | 873 | .pollin = 1, |
27926b68 | 874 | .plug = 1, |
e8c2bc1f | 875 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 876 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | .hash_reg_file = 1, | |
881 | .unbound_nonreg_file = 1, | |
8a72758c | 882 | .pollout = 1, |
27926b68 | 883 | .plug = 1, |
e8c2bc1f | 884 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
885 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
886 | IO_WQ_WORK_MM, | |
d3656344 | 887 | }, |
0463b6c5 | 888 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
889 | .needs_file = 1, |
890 | .unbound_nonreg_file = 1, | |
891 | }, | |
0463b6c5 PB |
892 | [IORING_OP_POLL_REMOVE] = {}, |
893 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 894 | .needs_file = 1, |
0f203765 | 895 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 896 | }, |
0463b6c5 | 897 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
898 | .needs_file = 1, |
899 | .unbound_nonreg_file = 1, | |
8a72758c | 900 | .pollout = 1, |
e8c2bc1f JA |
901 | .needs_async_data = 1, |
902 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 903 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 904 | }, |
0463b6c5 | 905 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
906 | .needs_file = 1, |
907 | .unbound_nonreg_file = 1, | |
8a72758c | 908 | .pollin = 1, |
52de1fe1 | 909 | .buffer_select = 1, |
e8c2bc1f JA |
910 | .needs_async_data = 1, |
911 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 912 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 913 | }, |
0463b6c5 | 914 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
915 | .needs_async_data = 1, |
916 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 917 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 918 | }, |
9c8e11b3 PB |
919 | [IORING_OP_TIMEOUT_REMOVE] = { |
920 | /* used by timeout updates' prep() */ | |
921 | .work_flags = IO_WQ_WORK_MM, | |
922 | }, | |
0463b6c5 | 923 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
924 | .needs_file = 1, |
925 | .unbound_nonreg_file = 1, | |
8a72758c | 926 | .pollin = 1, |
0f203765 | 927 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 928 | }, |
0463b6c5 PB |
929 | [IORING_OP_ASYNC_CANCEL] = {}, |
930 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
931 | .needs_async_data = 1, |
932 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 933 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 934 | }, |
0463b6c5 | 935 | [IORING_OP_CONNECT] = { |
d3656344 JA |
936 | .needs_file = 1, |
937 | .unbound_nonreg_file = 1, | |
8a72758c | 938 | .pollout = 1, |
e8c2bc1f JA |
939 | .needs_async_data = 1, |
940 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 941 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_FALLOCATE] = { |
d3656344 | 944 | .needs_file = 1, |
69228338 | 945 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 946 | }, |
0463b6c5 | 947 | [IORING_OP_OPENAT] = { |
0f203765 | 948 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
14587a46 | 949 | IO_WQ_WORK_FS | IO_WQ_WORK_MM, |
d3656344 | 950 | }, |
0463b6c5 | 951 | [IORING_OP_CLOSE] = { |
0f203765 | 952 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 953 | }, |
0463b6c5 | 954 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 955 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 956 | }, |
0463b6c5 | 957 | [IORING_OP_STATX] = { |
0f203765 JA |
958 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
959 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 960 | }, |
0463b6c5 | 961 | [IORING_OP_READ] = { |
3a6820f2 JA |
962 | .needs_file = 1, |
963 | .unbound_nonreg_file = 1, | |
8a72758c | 964 | .pollin = 1, |
bcda7baa | 965 | .buffer_select = 1, |
27926b68 | 966 | .plug = 1, |
e8c2bc1f | 967 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 968 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 969 | }, |
0463b6c5 | 970 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
971 | .needs_file = 1, |
972 | .unbound_nonreg_file = 1, | |
8a72758c | 973 | .pollout = 1, |
27926b68 | 974 | .plug = 1, |
e8c2bc1f | 975 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
976 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
977 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 978 | }, |
0463b6c5 | 979 | [IORING_OP_FADVISE] = { |
4840e418 | 980 | .needs_file = 1, |
0f203765 | 981 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 982 | }, |
0463b6c5 | 983 | [IORING_OP_MADVISE] = { |
0f203765 | 984 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 985 | }, |
0463b6c5 | 986 | [IORING_OP_SEND] = { |
fddaface JA |
987 | .needs_file = 1, |
988 | .unbound_nonreg_file = 1, | |
8a72758c | 989 | .pollout = 1, |
0f203765 | 990 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 991 | }, |
0463b6c5 | 992 | [IORING_OP_RECV] = { |
fddaface JA |
993 | .needs_file = 1, |
994 | .unbound_nonreg_file = 1, | |
8a72758c | 995 | .pollin = 1, |
bcda7baa | 996 | .buffer_select = 1, |
0f203765 | 997 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 998 | }, |
0463b6c5 | 999 | [IORING_OP_OPENAT2] = { |
0f203765 | 1000 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
14587a46 | 1001 | IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
cebdb986 | 1002 | }, |
3e4827b0 JA |
1003 | [IORING_OP_EPOLL_CTL] = { |
1004 | .unbound_nonreg_file = 1, | |
0f203765 | 1005 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 1006 | }, |
7d67af2c PB |
1007 | [IORING_OP_SPLICE] = { |
1008 | .needs_file = 1, | |
1009 | .hash_reg_file = 1, | |
1010 | .unbound_nonreg_file = 1, | |
0f203765 | 1011 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
1012 | }, |
1013 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1014 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1015 | [IORING_OP_TEE] = { |
1016 | .needs_file = 1, | |
1017 | .hash_reg_file = 1, | |
1018 | .unbound_nonreg_file = 1, | |
1019 | }, | |
36f4fa68 JA |
1020 | [IORING_OP_SHUTDOWN] = { |
1021 | .needs_file = 1, | |
1022 | }, | |
80a261fd JA |
1023 | [IORING_OP_RENAMEAT] = { |
1024 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1025 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1026 | }, | |
14a1143b JA |
1027 | [IORING_OP_UNLINKAT] = { |
1028 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1029 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1030 | }, | |
d3656344 JA |
1031 | }; |
1032 | ||
9936c7c2 PB |
1033 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1034 | struct task_struct *task, | |
1035 | struct files_struct *files); | |
269bbe5f | 1036 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1037 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1038 | struct io_ring_ctx *ctx); |
bc9744cd PB |
1039 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
1040 | struct fixed_rsrc_ref_node *ref_node); | |
1ffc5422 | 1041 | |
23faba36 | 1042 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1043 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1044 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1045 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1046 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1047 | static void io_dismantle_req(struct io_kiocb *req); |
1048 | static void io_put_task(struct task_struct *task, int nr); | |
1049 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1050 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1051 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1052 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1053 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1054 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1055 | unsigned nr_args); |
3ca405eb | 1056 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1057 | static struct file *io_file_get(struct io_submit_state *state, |
1058 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1059 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1060 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1061 | |
847595de PB |
1062 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1063 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1064 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1065 | const struct iovec *fast_iov, | |
227c0c96 | 1066 | struct iov_iter *iter, bool force); |
907d1df3 | 1067 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1068 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1069 | struct io_ring_ctx *ctx); | |
de0617e4 | 1070 | |
2b188cc1 JA |
1071 | static struct kmem_cache *req_cachep; |
1072 | ||
0918682b | 1073 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1074 | |
1075 | struct sock *io_uring_get_socket(struct file *file) | |
1076 | { | |
1077 | #if defined(CONFIG_UNIX) | |
1078 | if (file->f_op == &io_uring_fops) { | |
1079 | struct io_ring_ctx *ctx = file->private_data; | |
1080 | ||
1081 | return ctx->ring_sock->sk; | |
1082 | } | |
1083 | #endif | |
1084 | return NULL; | |
1085 | } | |
1086 | EXPORT_SYMBOL(io_uring_get_socket); | |
1087 | ||
f2f87370 PB |
1088 | #define io_for_each_link(pos, head) \ |
1089 | for (pos = (head); pos; pos = pos->link) | |
1090 | ||
3ca405eb PB |
1091 | static inline void io_clean_op(struct io_kiocb *req) |
1092 | { | |
9d5c8190 | 1093 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1094 | __io_clean_op(req); |
1095 | } | |
1096 | ||
36f72fe2 PB |
1097 | static inline void io_set_resource_node(struct io_kiocb *req) |
1098 | { | |
1099 | struct io_ring_ctx *ctx = req->ctx; | |
1100 | ||
269bbe5f BM |
1101 | if (!req->fixed_rsrc_refs) { |
1102 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1103 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1104 | } |
1105 | } | |
1106 | ||
08d23634 PB |
1107 | static bool io_match_task(struct io_kiocb *head, |
1108 | struct task_struct *task, | |
1109 | struct files_struct *files) | |
1110 | { | |
1111 | struct io_kiocb *req; | |
1112 | ||
84965ff8 JA |
1113 | if (task && head->task != task) { |
1114 | /* in terms of cancelation, always match if req task is dead */ | |
1115 | if (head->task->flags & PF_EXITING) | |
1116 | return true; | |
08d23634 | 1117 | return false; |
84965ff8 | 1118 | } |
08d23634 PB |
1119 | if (!files) |
1120 | return true; | |
1121 | ||
1122 | io_for_each_link(req, head) { | |
02a13674 JA |
1123 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1124 | continue; | |
1125 | if (req->file && req->file->f_op == &io_uring_fops) | |
1126 | return true; | |
1127 | if ((req->work.flags & IO_WQ_WORK_FILES) && | |
08d23634 PB |
1128 | req->work.identity->files == files) |
1129 | return true; | |
1130 | } | |
1131 | return false; | |
1132 | } | |
1133 | ||
28cea78a | 1134 | static void io_sq_thread_drop_mm_files(void) |
c40f6379 | 1135 | { |
28cea78a | 1136 | struct files_struct *files = current->files; |
c40f6379 JA |
1137 | struct mm_struct *mm = current->mm; |
1138 | ||
1139 | if (mm) { | |
1140 | kthread_unuse_mm(mm); | |
1141 | mmput(mm); | |
4b70cf9d | 1142 | current->mm = NULL; |
c40f6379 | 1143 | } |
28cea78a JA |
1144 | if (files) { |
1145 | struct nsproxy *nsproxy = current->nsproxy; | |
1146 | ||
1147 | task_lock(current); | |
1148 | current->files = NULL; | |
1149 | current->nsproxy = NULL; | |
1150 | task_unlock(current); | |
1151 | put_files_struct(files); | |
1152 | put_nsproxy(nsproxy); | |
1153 | } | |
1154 | } | |
1155 | ||
1a38ffc9 | 1156 | static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx) |
28cea78a JA |
1157 | { |
1158 | if (!current->files) { | |
1159 | struct files_struct *files; | |
1160 | struct nsproxy *nsproxy; | |
1161 | ||
1162 | task_lock(ctx->sqo_task); | |
1163 | files = ctx->sqo_task->files; | |
1164 | if (!files) { | |
1165 | task_unlock(ctx->sqo_task); | |
1a38ffc9 | 1166 | return -EOWNERDEAD; |
28cea78a JA |
1167 | } |
1168 | atomic_inc(&files->count); | |
1169 | get_nsproxy(ctx->sqo_task->nsproxy); | |
1170 | nsproxy = ctx->sqo_task->nsproxy; | |
1171 | task_unlock(ctx->sqo_task); | |
1172 | ||
1173 | task_lock(current); | |
1174 | current->files = files; | |
1175 | current->nsproxy = nsproxy; | |
1176 | task_unlock(current); | |
1177 | } | |
1a38ffc9 | 1178 | return 0; |
c40f6379 JA |
1179 | } |
1180 | ||
1181 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1182 | { | |
4b70cf9d JA |
1183 | struct mm_struct *mm; |
1184 | ||
1185 | if (current->mm) | |
1186 | return 0; | |
1187 | ||
4b70cf9d JA |
1188 | task_lock(ctx->sqo_task); |
1189 | mm = ctx->sqo_task->mm; | |
1190 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1191 | mm = NULL; | |
1192 | task_unlock(ctx->sqo_task); | |
1193 | ||
1194 | if (mm) { | |
1195 | kthread_use_mm(mm); | |
1196 | return 0; | |
c40f6379 JA |
1197 | } |
1198 | ||
4b70cf9d | 1199 | return -EFAULT; |
c40f6379 JA |
1200 | } |
1201 | ||
4e326358 PB |
1202 | static int __io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1203 | struct io_kiocb *req) | |
c40f6379 | 1204 | { |
28cea78a | 1205 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1a38ffc9 | 1206 | int ret; |
28cea78a JA |
1207 | |
1208 | if (def->work_flags & IO_WQ_WORK_MM) { | |
1a38ffc9 | 1209 | ret = __io_sq_thread_acquire_mm(ctx); |
28cea78a JA |
1210 | if (unlikely(ret)) |
1211 | return ret; | |
1212 | } | |
1213 | ||
1a38ffc9 PB |
1214 | if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) { |
1215 | ret = __io_sq_thread_acquire_files(ctx); | |
1216 | if (unlikely(ret)) | |
1217 | return ret; | |
1218 | } | |
28cea78a JA |
1219 | |
1220 | return 0; | |
c40f6379 JA |
1221 | } |
1222 | ||
4e326358 PB |
1223 | static inline int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1224 | struct io_kiocb *req) | |
1225 | { | |
4e326358 PB |
1226 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
1227 | return 0; | |
1228 | return __io_sq_thread_acquire_mm_files(ctx, req); | |
1229 | } | |
1230 | ||
91d8f519 DZ |
1231 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1232 | struct cgroup_subsys_state **cur_css) | |
1233 | ||
1234 | { | |
1235 | #ifdef CONFIG_BLK_CGROUP | |
1236 | /* puts the old one when swapping */ | |
1237 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1238 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1239 | *cur_css = ctx->sqo_blkcg_css; | |
1240 | } | |
1241 | #endif | |
1242 | } | |
1243 | ||
1244 | static void io_sq_thread_unassociate_blkcg(void) | |
1245 | { | |
1246 | #ifdef CONFIG_BLK_CGROUP | |
1247 | kthread_associate_blkcg(NULL); | |
1248 | #endif | |
1249 | } | |
1250 | ||
c40f6379 JA |
1251 | static inline void req_set_fail_links(struct io_kiocb *req) |
1252 | { | |
1253 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1254 | req->flags |= REQ_F_FAIL_LINK; | |
1255 | } | |
4a38aed2 | 1256 | |
1e6fa521 JA |
1257 | /* |
1258 | * None of these are dereferenced, they are simply used to check if any of | |
1259 | * them have changed. If we're under current and check they are still the | |
1260 | * same, we're fine to grab references to them for actual out-of-line use. | |
1261 | */ | |
1262 | static void io_init_identity(struct io_identity *id) | |
1263 | { | |
1264 | id->files = current->files; | |
1265 | id->mm = current->mm; | |
1266 | #ifdef CONFIG_BLK_CGROUP | |
1267 | rcu_read_lock(); | |
1268 | id->blkcg_css = blkcg_css(); | |
1269 | rcu_read_unlock(); | |
1270 | #endif | |
1271 | id->creds = current_cred(); | |
1272 | id->nsproxy = current->nsproxy; | |
1273 | id->fs = current->fs; | |
1274 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1275 | #ifdef CONFIG_AUDIT |
1276 | id->loginuid = current->loginuid; | |
1277 | id->sessionid = current->sessionid; | |
1278 | #endif | |
1e6fa521 JA |
1279 | refcount_set(&id->count, 1); |
1280 | } | |
1281 | ||
ec99ca6c PB |
1282 | static inline void __io_req_init_async(struct io_kiocb *req) |
1283 | { | |
1284 | memset(&req->work, 0, sizeof(req->work)); | |
1285 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1286 | } | |
1287 | ||
7cdaf587 XW |
1288 | /* |
1289 | * Note: must call io_req_init_async() for the first time you | |
1290 | * touch any members of io_wq_work. | |
1291 | */ | |
1292 | static inline void io_req_init_async(struct io_kiocb *req) | |
1293 | { | |
500a373d JA |
1294 | struct io_uring_task *tctx = current->io_uring; |
1295 | ||
7cdaf587 XW |
1296 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1297 | return; | |
1298 | ||
ec99ca6c | 1299 | __io_req_init_async(req); |
500a373d JA |
1300 | |
1301 | /* Grab a ref if this isn't our static identity */ | |
1302 | req->work.identity = tctx->identity; | |
1303 | if (tctx->identity != &tctx->__identity) | |
1304 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1305 | } |
1306 | ||
2b188cc1 JA |
1307 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1308 | { | |
1309 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1310 | ||
0f158b4c | 1311 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1312 | } |
1313 | ||
8eb7e2d0 PB |
1314 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1315 | { | |
1316 | return !req->timeout.off; | |
1317 | } | |
1318 | ||
2b188cc1 JA |
1319 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1320 | { | |
1321 | struct io_ring_ctx *ctx; | |
78076bb6 | 1322 | int hash_bits; |
2b188cc1 JA |
1323 | |
1324 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1325 | if (!ctx) | |
1326 | return NULL; | |
1327 | ||
78076bb6 JA |
1328 | /* |
1329 | * Use 5 bits less than the max cq entries, that should give us around | |
1330 | * 32 entries per hash list if totally full and uniformly spread. | |
1331 | */ | |
1332 | hash_bits = ilog2(p->cq_entries); | |
1333 | hash_bits -= 5; | |
1334 | if (hash_bits <= 0) | |
1335 | hash_bits = 1; | |
1336 | ctx->cancel_hash_bits = hash_bits; | |
1337 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1338 | GFP_KERNEL); | |
1339 | if (!ctx->cancel_hash) | |
1340 | goto err; | |
1341 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1342 | ||
21482896 | 1343 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1344 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1345 | goto err; | |
2b188cc1 JA |
1346 | |
1347 | ctx->flags = p->flags; | |
90554200 | 1348 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1349 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1350 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1351 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1352 | init_completion(&ctx->ref_comp); |
1353 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1354 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1355 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1356 | mutex_init(&ctx->uring_lock); |
1357 | init_waitqueue_head(&ctx->wait); | |
1358 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1359 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1360 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1361 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1362 | spin_lock_init(&ctx->inflight_lock); |
1363 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1364 | spin_lock_init(&ctx->rsrc_ref_lock); |
1365 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1366 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1367 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1368 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1369 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1370 | return ctx; |
206aefde | 1371 | err: |
78076bb6 | 1372 | kfree(ctx->cancel_hash); |
206aefde JA |
1373 | kfree(ctx); |
1374 | return NULL; | |
2b188cc1 JA |
1375 | } |
1376 | ||
9cf7c104 | 1377 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1378 | { |
2bc9930e JA |
1379 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1380 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1381 | |
9cf7c104 | 1382 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1383 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1384 | } |
de0617e4 | 1385 | |
9d858b21 | 1386 | return false; |
de0617e4 JA |
1387 | } |
1388 | ||
5c3462cf | 1389 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1390 | { |
500a373d | 1391 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1392 | return; |
1393 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1394 | kfree(req->work.identity); | |
1395 | } | |
1396 | ||
4edf20f9 | 1397 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1398 | { |
7cdaf587 | 1399 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1400 | return; |
51a4cc11 | 1401 | |
e86d0047 | 1402 | if (req->work.flags & IO_WQ_WORK_MM) |
98447d65 | 1403 | mmdrop(req->work.identity->mm); |
91d8f519 | 1404 | #ifdef CONFIG_BLK_CGROUP |
e86d0047 | 1405 | if (req->work.flags & IO_WQ_WORK_BLKCG) |
98447d65 | 1406 | css_put(req->work.identity->blkcg_css); |
91d8f519 | 1407 | #endif |
e86d0047 | 1408 | if (req->work.flags & IO_WQ_WORK_CREDS) |
98447d65 | 1409 | put_cred(req->work.identity->creds); |
dfead8a8 | 1410 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1411 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1412 | |
98447d65 | 1413 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1414 | if (--fs->users) |
1415 | fs = NULL; | |
98447d65 | 1416 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1417 | if (fs) |
1418 | free_fs_struct(fs); | |
1419 | } | |
34e08fed PB |
1420 | if (req->work.flags & IO_WQ_WORK_FILES) { |
1421 | put_files_struct(req->work.identity->files); | |
1422 | put_nsproxy(req->work.identity->nsproxy); | |
34e08fed PB |
1423 | } |
1424 | if (req->flags & REQ_F_INFLIGHT) { | |
1425 | struct io_ring_ctx *ctx = req->ctx; | |
1426 | struct io_uring_task *tctx = req->task->io_uring; | |
1427 | unsigned long flags; | |
1428 | ||
1429 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1430 | list_del(&req->inflight_entry); | |
1431 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1432 | req->flags &= ~REQ_F_INFLIGHT; | |
1433 | if (atomic_read(&tctx->in_idle)) | |
1434 | wake_up(&tctx->wait); | |
1435 | } | |
51a4cc11 | 1436 | |
e86d0047 PB |
1437 | req->flags &= ~REQ_F_WORK_INITIALIZED; |
1438 | req->work.flags &= ~(IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | IO_WQ_WORK_FS | | |
1439 | IO_WQ_WORK_CREDS | IO_WQ_WORK_FILES); | |
5c3462cf | 1440 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1441 | } |
1442 | ||
1e6fa521 JA |
1443 | /* |
1444 | * Create a private copy of io_identity, since some fields don't match | |
1445 | * the current context. | |
1446 | */ | |
1447 | static bool io_identity_cow(struct io_kiocb *req) | |
1448 | { | |
5c3462cf | 1449 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1450 | const struct cred *creds = NULL; |
1451 | struct io_identity *id; | |
1452 | ||
1453 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1454 | creds = req->work.identity->creds; | |
1455 | ||
1456 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1457 | if (unlikely(!id)) { | |
1458 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1459 | return false; | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * We can safely just re-init the creds we copied Either the field | |
1464 | * matches the current one, or we haven't grabbed it yet. The only | |
1465 | * exception is ->creds, through registered personalities, so handle | |
1466 | * that one separately. | |
1467 | */ | |
1468 | io_init_identity(id); | |
1469 | if (creds) | |
e8c954df | 1470 | id->creds = creds; |
1e6fa521 JA |
1471 | |
1472 | /* add one for this request */ | |
1473 | refcount_inc(&id->count); | |
1474 | ||
cb8a8ae3 JA |
1475 | /* drop tctx and req identity references, if needed */ |
1476 | if (tctx->identity != &tctx->__identity && | |
1477 | refcount_dec_and_test(&tctx->identity->count)) | |
1478 | kfree(tctx->identity); | |
1479 | if (req->work.identity != &tctx->__identity && | |
1480 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1481 | kfree(req->work.identity); |
1482 | ||
1483 | req->work.identity = id; | |
500a373d | 1484 | tctx->identity = id; |
1e6fa521 JA |
1485 | return true; |
1486 | } | |
1487 | ||
ce3d5aae PB |
1488 | static void io_req_track_inflight(struct io_kiocb *req) |
1489 | { | |
1490 | struct io_ring_ctx *ctx = req->ctx; | |
1491 | ||
1492 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
1493 | io_req_init_async(req); | |
1494 | req->flags |= REQ_F_INFLIGHT; | |
1495 | ||
1496 | spin_lock_irq(&ctx->inflight_lock); | |
1497 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1498 | spin_unlock_irq(&ctx->inflight_lock); | |
1499 | } | |
1500 | } | |
1501 | ||
1e6fa521 | 1502 | static bool io_grab_identity(struct io_kiocb *req) |
18d9be1a | 1503 | { |
d3656344 | 1504 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1505 | struct io_identity *id = req->work.identity; |
54a91f3b | 1506 | |
69228338 JA |
1507 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1508 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1509 | return false; | |
1510 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1511 | } | |
91d8f519 | 1512 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1513 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1514 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1515 | rcu_read_lock(); |
1e6fa521 JA |
1516 | if (id->blkcg_css != blkcg_css()) { |
1517 | rcu_read_unlock(); | |
1518 | return false; | |
1519 | } | |
91d8f519 DZ |
1520 | /* |
1521 | * This should be rare, either the cgroup is dying or the task | |
1522 | * is moving cgroups. Just punt to root for the handful of ios. | |
1523 | */ | |
1e6fa521 | 1524 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1525 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1526 | rcu_read_unlock(); |
1527 | } | |
1528 | #endif | |
dfead8a8 | 1529 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1530 | if (id->creds != current_cred()) |
1531 | return false; | |
1532 | get_cred(id->creds); | |
dfead8a8 JA |
1533 | req->work.flags |= IO_WQ_WORK_CREDS; |
1534 | } | |
4ea33a97 JA |
1535 | #ifdef CONFIG_AUDIT |
1536 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1537 | current->sessionid != id->sessionid) | |
1538 | return false; | |
1539 | #endif | |
dfead8a8 JA |
1540 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1541 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1542 | if (current->fs != id->fs) |
1543 | return false; | |
1544 | spin_lock(&id->fs->lock); | |
1545 | if (!id->fs->in_exec) { | |
1546 | id->fs->users++; | |
dfead8a8 | 1547 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1548 | } else { |
1549 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1550 | } | |
1551 | spin_unlock(¤t->fs->lock); | |
1552 | } | |
af604703 PB |
1553 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1554 | (def->work_flags & IO_WQ_WORK_FILES) && | |
1555 | !(req->flags & REQ_F_NO_FILE_TABLE)) { | |
1556 | if (id->files != current->files || | |
1557 | id->nsproxy != current->nsproxy) | |
1558 | return false; | |
1559 | atomic_inc(&id->files->count); | |
1560 | get_nsproxy(id->nsproxy); | |
af604703 | 1561 | req->work.flags |= IO_WQ_WORK_FILES; |
ce3d5aae | 1562 | io_req_track_inflight(req); |
af604703 | 1563 | } |
77788775 JA |
1564 | if (!(req->work.flags & IO_WQ_WORK_MM) && |
1565 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1566 | if (id->mm != current->mm) | |
1567 | return false; | |
1568 | mmgrab(id->mm); | |
1569 | req->work.flags |= IO_WQ_WORK_MM; | |
1570 | } | |
1e6fa521 JA |
1571 | |
1572 | return true; | |
1573 | } | |
1574 | ||
1575 | static void io_prep_async_work(struct io_kiocb *req) | |
1576 | { | |
1577 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1578 | struct io_ring_ctx *ctx = req->ctx; |
1579 | ||
1580 | io_req_init_async(req); | |
1581 | ||
feaadc4f PB |
1582 | if (req->flags & REQ_F_FORCE_ASYNC) |
1583 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1584 | ||
1e6fa521 JA |
1585 | if (req->flags & REQ_F_ISREG) { |
1586 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1587 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1588 | } else { | |
1589 | if (def->unbound_nonreg_file) | |
1590 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1591 | } | |
1592 | ||
1e6fa521 JA |
1593 | /* if we fail grabbing identity, we must COW, regrab, and retry */ |
1594 | if (io_grab_identity(req)) | |
1595 | return; | |
1596 | ||
1597 | if (!io_identity_cow(req)) | |
1598 | return; | |
1599 | ||
1600 | /* can't fail at this point */ | |
1601 | if (!io_grab_identity(req)) | |
1602 | WARN_ON(1); | |
561fb04a | 1603 | } |
cccf0ee8 | 1604 | |
cbdcb435 | 1605 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1606 | { |
cbdcb435 | 1607 | struct io_kiocb *cur; |
54a91f3b | 1608 | |
f2f87370 PB |
1609 | io_for_each_link(cur, req) |
1610 | io_prep_async_work(cur); | |
561fb04a JA |
1611 | } |
1612 | ||
7271ef3a | 1613 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1614 | { |
a197f664 | 1615 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1616 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1617 | |
8766dd51 PB |
1618 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1619 | &req->work, req->flags); | |
1620 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1621 | return link; |
18d9be1a JA |
1622 | } |
1623 | ||
cbdcb435 PB |
1624 | static void io_queue_async_work(struct io_kiocb *req) |
1625 | { | |
7271ef3a JA |
1626 | struct io_kiocb *link; |
1627 | ||
cbdcb435 PB |
1628 | /* init ->work of the whole link before punting */ |
1629 | io_prep_async_link(req); | |
7271ef3a JA |
1630 | link = __io_queue_async_work(req); |
1631 | ||
1632 | if (link) | |
1633 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1634 | } |
1635 | ||
5262f567 JA |
1636 | static void io_kill_timeout(struct io_kiocb *req) |
1637 | { | |
e8c2bc1f | 1638 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1639 | int ret; |
1640 | ||
e8c2bc1f | 1641 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1642 | if (ret != -1) { |
01cec8c1 PB |
1643 | atomic_set(&req->ctx->cq_timeouts, |
1644 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1645 | list_del_init(&req->timeout.list); |
78e19bbe | 1646 | io_cqring_fill_event(req, 0); |
216578e5 | 1647 | io_put_req_deferred(req, 1); |
5262f567 JA |
1648 | } |
1649 | } | |
1650 | ||
76e1b642 JA |
1651 | /* |
1652 | * Returns true if we found and killed one or more timeouts | |
1653 | */ | |
6b81928d PB |
1654 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1655 | struct files_struct *files) | |
5262f567 JA |
1656 | { |
1657 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1658 | int canceled = 0; |
5262f567 JA |
1659 | |
1660 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1661 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1662 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1663 | io_kill_timeout(req); |
76e1b642 JA |
1664 | canceled++; |
1665 | } | |
f3606e3a | 1666 | } |
5262f567 | 1667 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1668 | return canceled != 0; |
5262f567 JA |
1669 | } |
1670 | ||
04518945 | 1671 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1672 | { |
04518945 | 1673 | do { |
27dc8338 PB |
1674 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1675 | struct io_defer_entry, list); | |
de0617e4 | 1676 | |
9cf7c104 | 1677 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1678 | break; |
27dc8338 | 1679 | list_del_init(&de->list); |
907d1df3 | 1680 | io_req_task_queue(de->req); |
27dc8338 | 1681 | kfree(de); |
04518945 PB |
1682 | } while (!list_empty(&ctx->defer_list)); |
1683 | } | |
1684 | ||
360428f8 | 1685 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1686 | { |
f010505b MDG |
1687 | u32 seq; |
1688 | ||
1689 | if (list_empty(&ctx->timeout_list)) | |
1690 | return; | |
1691 | ||
1692 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1693 | ||
1694 | do { | |
1695 | u32 events_needed, events_got; | |
360428f8 | 1696 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1697 | struct io_kiocb, timeout.list); |
de0617e4 | 1698 | |
8eb7e2d0 | 1699 | if (io_is_timeout_noseq(req)) |
360428f8 | 1700 | break; |
f010505b MDG |
1701 | |
1702 | /* | |
1703 | * Since seq can easily wrap around over time, subtract | |
1704 | * the last seq at which timeouts were flushed before comparing. | |
1705 | * Assuming not more than 2^31-1 events have happened since, | |
1706 | * these subtractions won't have wrapped, so we can check if | |
1707 | * target is in [last_seq, current_seq] by comparing the two. | |
1708 | */ | |
1709 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1710 | events_got = seq - ctx->cq_last_tm_flush; | |
1711 | if (events_got < events_needed) | |
360428f8 | 1712 | break; |
bfe68a22 | 1713 | |
135fcde8 | 1714 | list_del_init(&req->timeout.list); |
5262f567 | 1715 | io_kill_timeout(req); |
f010505b MDG |
1716 | } while (!list_empty(&ctx->timeout_list)); |
1717 | ||
1718 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1719 | } |
5262f567 | 1720 | |
360428f8 PB |
1721 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1722 | { | |
1723 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1724 | |
1725 | /* order cqe stores with ring update */ | |
1726 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1727 | |
04518945 PB |
1728 | if (unlikely(!list_empty(&ctx->defer_list))) |
1729 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1730 | } |
1731 | ||
90554200 JA |
1732 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1733 | { | |
1734 | struct io_rings *r = ctx->rings; | |
1735 | ||
1736 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1737 | } | |
1738 | ||
888aae2e PB |
1739 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1740 | { | |
1741 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1742 | } | |
1743 | ||
2b188cc1 JA |
1744 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1745 | { | |
75b28aff | 1746 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1747 | unsigned tail; |
1748 | ||
115e12e5 SB |
1749 | /* |
1750 | * writes to the cq entry need to come after reading head; the | |
1751 | * control dependency is enough as we're using WRITE_ONCE to | |
1752 | * fill the cq entry | |
1753 | */ | |
888aae2e | 1754 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1755 | return NULL; |
1756 | ||
888aae2e | 1757 | tail = ctx->cached_cq_tail++; |
75b28aff | 1758 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1759 | } |
1760 | ||
f2842ab5 JA |
1761 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1762 | { | |
f0b493e6 JA |
1763 | if (!ctx->cq_ev_fd) |
1764 | return false; | |
7e55a19c SG |
1765 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1766 | return false; | |
f2842ab5 JA |
1767 | if (!ctx->eventfd_async) |
1768 | return true; | |
b41e9852 | 1769 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1770 | } |
1771 | ||
b41e9852 | 1772 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1773 | { |
b1445e59 PB |
1774 | /* see waitqueue_active() comment */ |
1775 | smp_mb(); | |
1776 | ||
1d7bb1d5 JA |
1777 | if (waitqueue_active(&ctx->wait)) |
1778 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1779 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1780 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1781 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1782 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1783 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1784 | wake_up_interruptible(&ctx->cq_wait); |
1785 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1786 | } | |
1d7bb1d5 JA |
1787 | } |
1788 | ||
80c18e4a PB |
1789 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1790 | { | |
b1445e59 PB |
1791 | /* see waitqueue_active() comment */ |
1792 | smp_mb(); | |
1793 | ||
80c18e4a PB |
1794 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1795 | if (waitqueue_active(&ctx->wait)) | |
1796 | wake_up(&ctx->wait); | |
1797 | } | |
1798 | if (io_should_trigger_evfd(ctx)) | |
1799 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1800 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1801 | wake_up_interruptible(&ctx->cq_wait); |
1802 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1803 | } | |
80c18e4a PB |
1804 | } |
1805 | ||
c4a2ed72 | 1806 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1807 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1808 | struct task_struct *tsk, | |
1809 | struct files_struct *files) | |
1d7bb1d5 JA |
1810 | { |
1811 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1812 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1813 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1814 | unsigned long flags; |
b18032bb | 1815 | bool all_flushed, posted; |
1d7bb1d5 JA |
1816 | LIST_HEAD(list); |
1817 | ||
e23de15f PB |
1818 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1819 | return false; | |
1d7bb1d5 | 1820 | |
b18032bb | 1821 | posted = false; |
1d7bb1d5 | 1822 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1823 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1824 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1825 | continue; |
1826 | ||
1d7bb1d5 JA |
1827 | cqe = io_get_cqring(ctx); |
1828 | if (!cqe && !force) | |
1829 | break; | |
1830 | ||
40d8ddd4 | 1831 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1832 | if (cqe) { |
1833 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1834 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1835 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1836 | } else { |
2c3bac6d | 1837 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1838 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1839 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1840 | } |
b18032bb | 1841 | posted = true; |
1d7bb1d5 JA |
1842 | } |
1843 | ||
09e88404 PB |
1844 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1845 | if (all_flushed) { | |
1846 | clear_bit(0, &ctx->sq_check_overflow); | |
1847 | clear_bit(0, &ctx->cq_check_overflow); | |
1848 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1849 | } | |
46930143 | 1850 | |
b18032bb JA |
1851 | if (posted) |
1852 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1853 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1854 | if (posted) |
1855 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1856 | |
1857 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1858 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1859 | list_del(&req->compl.list); | |
ec9c02ad | 1860 | io_put_req(req); |
1d7bb1d5 | 1861 | } |
c4a2ed72 | 1862 | |
09e88404 | 1863 | return all_flushed; |
1d7bb1d5 JA |
1864 | } |
1865 | ||
6c503150 PB |
1866 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1867 | struct task_struct *tsk, | |
1868 | struct files_struct *files) | |
1869 | { | |
1870 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1871 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1872 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1873 | mutex_lock(&ctx->uring_lock); | |
1874 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1875 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1876 | mutex_unlock(&ctx->uring_lock); | |
1877 | } | |
1878 | } | |
1879 | ||
bcda7baa | 1880 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1881 | { |
78e19bbe | 1882 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1883 | struct io_uring_cqe *cqe; |
1884 | ||
78e19bbe | 1885 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1886 | |
2b188cc1 JA |
1887 | /* |
1888 | * If we can't get a cq entry, userspace overflowed the | |
1889 | * submission (by quite a lot). Increment the overflow count in | |
1890 | * the ring. | |
1891 | */ | |
1892 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1893 | if (likely(cqe)) { |
78e19bbe | 1894 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1895 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1896 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1897 | } else if (ctx->cq_overflow_flushed || |
1898 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1899 | /* |
1900 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1901 | * then we cannot store the request for later flushing, we need | |
1902 | * to drop it on the floor. | |
1903 | */ | |
2c3bac6d PB |
1904 | ctx->cached_cq_overflow++; |
1905 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1906 | } else { |
ad3eb2c8 JA |
1907 | if (list_empty(&ctx->cq_overflow_list)) { |
1908 | set_bit(0, &ctx->sq_check_overflow); | |
1909 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1910 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1911 | } |
40d8ddd4 | 1912 | io_clean_op(req); |
1d7bb1d5 | 1913 | req->result = res; |
0f7e466b | 1914 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1915 | refcount_inc(&req->refs); |
1916 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1917 | } |
1918 | } | |
1919 | ||
bcda7baa JA |
1920 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1921 | { | |
1922 | __io_cqring_fill_event(req, res, 0); | |
1923 | } | |
1924 | ||
c7dae4ba JA |
1925 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1926 | unsigned int cflags) | |
2b188cc1 | 1927 | { |
78e19bbe | 1928 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1929 | unsigned long flags; |
1930 | ||
1931 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1932 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1933 | io_commit_cqring(ctx); |
c7dae4ba JA |
1934 | /* |
1935 | * If we're the last reference to this request, add to our locked | |
1936 | * free_list cache. | |
1937 | */ | |
1938 | if (refcount_dec_and_test(&req->refs)) { | |
1939 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1940 | ||
1941 | io_dismantle_req(req); | |
1942 | io_put_task(req->task, 1); | |
1943 | list_add(&req->compl.list, &cs->locked_free_list); | |
1944 | cs->locked_free_nr++; | |
1945 | } else | |
1946 | req = NULL; | |
2b188cc1 JA |
1947 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1948 | ||
8c838788 | 1949 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1950 | if (req) { |
1951 | io_queue_next(req); | |
1952 | percpu_ref_put(&ctx->refs); | |
1953 | } | |
2b188cc1 JA |
1954 | } |
1955 | ||
a38d68db | 1956 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1957 | unsigned int cflags) |
229a7b63 | 1958 | { |
a38d68db PB |
1959 | io_clean_op(req); |
1960 | req->result = res; | |
1961 | req->compl.cflags = cflags; | |
e342c807 | 1962 | req->flags |= REQ_F_COMPLETE_INLINE; |
a38d68db PB |
1963 | } |
1964 | ||
889fca73 PB |
1965 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1966 | long res, unsigned cflags) | |
a38d68db | 1967 | { |
889fca73 PB |
1968 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1969 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1970 | else |
c7dae4ba | 1971 | io_req_complete_post(req, res, cflags); |
e1e16097 JA |
1972 | } |
1973 | ||
a38d68db | 1974 | static inline void io_req_complete(struct io_kiocb *req, long res) |
bcda7baa | 1975 | { |
889fca73 | 1976 | __io_req_complete(req, 0, res, 0); |
bcda7baa JA |
1977 | } |
1978 | ||
c7dae4ba | 1979 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
2b188cc1 | 1980 | { |
c7dae4ba JA |
1981 | struct io_submit_state *state = &ctx->submit_state; |
1982 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1983 | struct io_kiocb *req = NULL; |
1b4c351f | 1984 | |
c7dae4ba JA |
1985 | /* |
1986 | * If we have more than a batch's worth of requests in our IRQ side | |
1987 | * locked cache, grab the lock and move them over to our submission | |
1988 | * side cache. | |
1989 | */ | |
1990 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1991 | spin_lock_irq(&ctx->completion_lock); | |
1992 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1993 | cs->locked_free_nr = 0; | |
1994 | spin_unlock_irq(&ctx->completion_lock); | |
1995 | } | |
1996 | ||
1997 | while (!list_empty(&cs->free_list)) { | |
1998 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1999 | compl.list); |
2000 | list_del(&req->compl.list); | |
e5d1bc0a PB |
2001 | state->reqs[state->free_reqs++] = req; |
2002 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
2003 | break; | |
1b4c351f JA |
2004 | } |
2005 | ||
e5d1bc0a PB |
2006 | return req != NULL; |
2007 | } | |
2008 | ||
2009 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
2010 | { | |
2011 | struct io_submit_state *state = &ctx->submit_state; | |
2012 | ||
2013 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
2014 | ||
f6b6c7d6 | 2015 | if (!state->free_reqs) { |
291b2821 | 2016 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
2017 | int ret; |
2018 | ||
c7dae4ba | 2019 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
2020 | goto got_req; |
2021 | ||
bf019da7 PB |
2022 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
2023 | state->reqs); | |
fd6fab2c JA |
2024 | |
2025 | /* | |
2026 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2027 | * retry single alloc to be on the safe side. | |
2028 | */ | |
2029 | if (unlikely(ret <= 0)) { | |
2030 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
2031 | if (!state->reqs[0]) | |
3893f39f | 2032 | return NULL; |
fd6fab2c JA |
2033 | ret = 1; |
2034 | } | |
291b2821 | 2035 | state->free_reqs = ret; |
2b188cc1 | 2036 | } |
e5d1bc0a | 2037 | got_req: |
291b2821 PB |
2038 | state->free_reqs--; |
2039 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
2040 | } |
2041 | ||
8da11c19 PB |
2042 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
2043 | bool fixed) | |
2044 | { | |
36f72fe2 | 2045 | if (!fixed) |
8da11c19 PB |
2046 | fput(file); |
2047 | } | |
2048 | ||
4edf20f9 | 2049 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2050 | { |
3ca405eb | 2051 | io_clean_op(req); |
929a3af9 | 2052 | |
e8c2bc1f JA |
2053 | if (req->async_data) |
2054 | kfree(req->async_data); | |
8da11c19 PB |
2055 | if (req->file) |
2056 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
2057 | if (req->fixed_rsrc_refs) |
2058 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 2059 | io_req_clean_work(req); |
e65ef56d JA |
2060 | } |
2061 | ||
7c660731 PB |
2062 | static inline void io_put_task(struct task_struct *task, int nr) |
2063 | { | |
2064 | struct io_uring_task *tctx = task->io_uring; | |
2065 | ||
2066 | percpu_counter_sub(&tctx->inflight, nr); | |
2067 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2068 | wake_up(&tctx->wait); | |
2069 | put_task_struct_many(task, nr); | |
2070 | } | |
2071 | ||
216578e5 | 2072 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2073 | { |
51a4cc11 | 2074 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2075 | |
216578e5 | 2076 | io_dismantle_req(req); |
7c660731 | 2077 | io_put_task(req->task, 1); |
e3bc8e9d | 2078 | |
3893f39f | 2079 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 2080 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
2081 | } |
2082 | ||
f2f87370 PB |
2083 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2084 | { | |
2085 | struct io_kiocb *nxt = req->link; | |
2086 | ||
2087 | req->link = nxt->link; | |
2088 | nxt->link = NULL; | |
2089 | } | |
2090 | ||
c9abd7ad | 2091 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 2092 | { |
a197f664 | 2093 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 2094 | struct io_kiocb *link; |
c9abd7ad PB |
2095 | bool cancelled = false; |
2096 | unsigned long flags; | |
7c86ffee | 2097 | |
c9abd7ad | 2098 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2099 | link = req->link; |
2100 | ||
900fad45 PB |
2101 | /* |
2102 | * Can happen if a linked timeout fired and link had been like | |
2103 | * req -> link t-out -> link t-out [-> ...] | |
2104 | */ | |
c9abd7ad PB |
2105 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
2106 | struct io_timeout_data *io = link->async_data; | |
2107 | int ret; | |
7c86ffee | 2108 | |
f2f87370 | 2109 | io_remove_next_linked(req); |
90cd7e42 | 2110 | link->timeout.head = NULL; |
c9abd7ad PB |
2111 | ret = hrtimer_try_to_cancel(&io->timer); |
2112 | if (ret != -1) { | |
2113 | io_cqring_fill_event(link, -ECANCELED); | |
2114 | io_commit_cqring(ctx); | |
2115 | cancelled = true; | |
2116 | } | |
2117 | } | |
7c86ffee | 2118 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 2119 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 2120 | |
c9abd7ad | 2121 | if (cancelled) { |
7c86ffee | 2122 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
2123 | io_put_req(link); |
2124 | } | |
7c86ffee PB |
2125 | } |
2126 | ||
9e645e11 | 2127 | |
d148ca4b | 2128 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 2129 | { |
f2f87370 | 2130 | struct io_kiocb *link, *nxt; |
2665abfd | 2131 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 2132 | unsigned long flags; |
9e645e11 | 2133 | |
d148ca4b | 2134 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2135 | link = req->link; |
2136 | req->link = NULL; | |
9e645e11 | 2137 | |
f2f87370 PB |
2138 | while (link) { |
2139 | nxt = link->link; | |
2140 | link->link = NULL; | |
2665abfd | 2141 | |
f2f87370 | 2142 | trace_io_uring_fail_link(req, link); |
7c86ffee | 2143 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
2144 | |
2145 | /* | |
2146 | * It's ok to free under spinlock as they're not linked anymore, | |
2147 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
2148 | * work.fs->lock. | |
2149 | */ | |
2150 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
2151 | io_put_req_deferred(link, 2); | |
2152 | else | |
2153 | io_double_put_req(link); | |
f2f87370 | 2154 | link = nxt; |
9e645e11 | 2155 | } |
2665abfd | 2156 | io_commit_cqring(ctx); |
216578e5 | 2157 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 2158 | |
2665abfd | 2159 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
2160 | } |
2161 | ||
3fa5e0f3 | 2162 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2163 | { |
7c86ffee PB |
2164 | if (req->flags & REQ_F_LINK_TIMEOUT) |
2165 | io_kill_linked_timeout(req); | |
944e58bf | 2166 | |
9e645e11 JA |
2167 | /* |
2168 | * If LINK is set, we have dependent requests in this chain. If we | |
2169 | * didn't fail this request, queue the first one up, moving any other | |
2170 | * dependencies to the next request. In case of failure, fail the rest | |
2171 | * of the chain. | |
2172 | */ | |
f2f87370 PB |
2173 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
2174 | struct io_kiocb *nxt = req->link; | |
2175 | ||
2176 | req->link = NULL; | |
2177 | return nxt; | |
2178 | } | |
9b5f7bd9 PB |
2179 | io_fail_links(req); |
2180 | return NULL; | |
4d7dd462 | 2181 | } |
9e645e11 | 2182 | |
f2f87370 | 2183 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2184 | { |
cdbff982 | 2185 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2186 | return NULL; |
2187 | return __io_req_find_next(req); | |
2188 | } | |
2189 | ||
7cbf1722 JA |
2190 | static bool __tctx_task_work(struct io_uring_task *tctx) |
2191 | { | |
65453d1e | 2192 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
2193 | struct io_wq_work_list list; |
2194 | struct io_wq_work_node *node; | |
2195 | ||
2196 | if (wq_list_empty(&tctx->task_list)) | |
2197 | return false; | |
2198 | ||
0b81e80c | 2199 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
2200 | list = tctx->task_list; |
2201 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2202 | spin_unlock_irq(&tctx->task_lock); |
7cbf1722 JA |
2203 | |
2204 | node = list.first; | |
2205 | while (node) { | |
2206 | struct io_wq_work_node *next = node->next; | |
65453d1e | 2207 | struct io_ring_ctx *this_ctx; |
7cbf1722 JA |
2208 | struct io_kiocb *req; |
2209 | ||
2210 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
65453d1e | 2211 | this_ctx = req->ctx; |
7cbf1722 JA |
2212 | req->task_work.func(&req->task_work); |
2213 | node = next; | |
65453d1e JA |
2214 | |
2215 | if (!ctx) { | |
2216 | ctx = this_ctx; | |
2217 | } else if (ctx != this_ctx) { | |
2218 | mutex_lock(&ctx->uring_lock); | |
2219 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2220 | mutex_unlock(&ctx->uring_lock); | |
2221 | ctx = this_ctx; | |
2222 | } | |
2223 | } | |
2224 | ||
2225 | if (ctx && ctx->submit_state.comp.nr) { | |
2226 | mutex_lock(&ctx->uring_lock); | |
2227 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2228 | mutex_unlock(&ctx->uring_lock); | |
7cbf1722 JA |
2229 | } |
2230 | ||
2231 | return list.first != NULL; | |
2232 | } | |
2233 | ||
2234 | static void tctx_task_work(struct callback_head *cb) | |
2235 | { | |
2236 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); | |
2237 | ||
2238 | while (__tctx_task_work(tctx)) | |
2239 | cond_resched(); | |
2240 | ||
2241 | clear_bit(0, &tctx->task_state); | |
2242 | } | |
2243 | ||
2244 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
2245 | enum task_work_notify_mode notify) | |
2246 | { | |
2247 | struct io_uring_task *tctx = tsk->io_uring; | |
2248 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 2249 | unsigned long flags; |
7cbf1722 JA |
2250 | int ret; |
2251 | ||
2252 | WARN_ON_ONCE(!tctx); | |
2253 | ||
0b81e80c | 2254 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2255 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 2256 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2257 | |
2258 | /* task_work already pending, we're done */ | |
2259 | if (test_bit(0, &tctx->task_state) || | |
2260 | test_and_set_bit(0, &tctx->task_state)) | |
2261 | return 0; | |
2262 | ||
2263 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
2264 | return 0; | |
2265 | ||
2266 | /* | |
2267 | * Slow path - we failed, find and delete work. if the work is not | |
2268 | * in the list, it got run and we're fine. | |
2269 | */ | |
2270 | ret = 0; | |
0b81e80c | 2271 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
2272 | wq_list_for_each(node, prev, &tctx->task_list) { |
2273 | if (&req->io_task_work.node == node) { | |
2274 | wq_list_del(&tctx->task_list, node, prev); | |
2275 | ret = 1; | |
2276 | break; | |
2277 | } | |
2278 | } | |
0b81e80c | 2279 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2280 | clear_bit(0, &tctx->task_state); |
2281 | return ret; | |
2282 | } | |
2283 | ||
355fb9e2 | 2284 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
2285 | { |
2286 | struct task_struct *tsk = req->task; | |
2287 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2288 | enum task_work_notify_mode notify; |
2289 | int ret; | |
c2c4c83c | 2290 | |
6200b0ae JA |
2291 | if (tsk->flags & PF_EXITING) |
2292 | return -ESRCH; | |
2293 | ||
c2c4c83c | 2294 | /* |
0ba9c9ed JA |
2295 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2296 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2297 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2298 | * will do the job. | |
c2c4c83c | 2299 | */ |
91989c70 | 2300 | notify = TWA_NONE; |
355fb9e2 | 2301 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
2302 | notify = TWA_SIGNAL; |
2303 | ||
7cbf1722 | 2304 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
2305 | if (!ret) |
2306 | wake_up_process(tsk); | |
0ba9c9ed | 2307 | |
c2c4c83c JA |
2308 | return ret; |
2309 | } | |
2310 | ||
eab30c4d | 2311 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 2312 | task_work_func_t cb) |
eab30c4d PB |
2313 | { |
2314 | struct task_struct *tsk = io_wq_get_task(req->ctx->io_wq); | |
2315 | ||
2316 | init_task_work(&req->task_work, cb); | |
2317 | task_work_add(tsk, &req->task_work, TWA_NONE); | |
2318 | wake_up_process(tsk); | |
2319 | } | |
2320 | ||
c40f6379 JA |
2321 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2322 | { | |
2323 | struct io_ring_ctx *ctx = req->ctx; | |
2324 | ||
2325 | spin_lock_irq(&ctx->completion_lock); | |
2326 | io_cqring_fill_event(req, error); | |
2327 | io_commit_cqring(ctx); | |
2328 | spin_unlock_irq(&ctx->completion_lock); | |
2329 | ||
2330 | io_cqring_ev_posted(ctx); | |
2331 | req_set_fail_links(req); | |
2332 | io_double_put_req(req); | |
2333 | } | |
2334 | ||
2335 | static void io_req_task_cancel(struct callback_head *cb) | |
2336 | { | |
2337 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2338 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2339 | |
2340 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2341 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2342 | } |
2343 | ||
2344 | static void __io_req_task_submit(struct io_kiocb *req) | |
2345 | { | |
2346 | struct io_ring_ctx *ctx = req->ctx; | |
2347 | ||
04fc6c80 | 2348 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2349 | mutex_lock(&ctx->uring_lock); |
dc0eced5 PB |
2350 | if (!ctx->sqo_dead && !(current->flags & PF_EXITING) && |
2351 | !io_sq_thread_acquire_mm_files(ctx, req)) | |
c5eef2b9 | 2352 | __io_queue_sqe(req); |
81b6d05c | 2353 | else |
c40f6379 | 2354 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2355 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2356 | } |
2357 | ||
2358 | static void io_req_task_submit(struct callback_head *cb) | |
2359 | { | |
2360 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2361 | ||
2362 | __io_req_task_submit(req); | |
2363 | } | |
2364 | ||
2365 | static void io_req_task_queue(struct io_kiocb *req) | |
2366 | { | |
c40f6379 JA |
2367 | int ret; |
2368 | ||
7cbf1722 | 2369 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2370 | ret = io_req_task_work_add(req); |
04fc6c80 PB |
2371 | if (unlikely(ret)) { |
2372 | percpu_ref_get(&req->ctx->refs); | |
eab30c4d | 2373 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
04fc6c80 | 2374 | } |
c40f6379 JA |
2375 | } |
2376 | ||
f2f87370 | 2377 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2378 | { |
9b5f7bd9 | 2379 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2380 | |
2381 | if (nxt) | |
906a8c3f | 2382 | io_req_task_queue(nxt); |
c69f8dbe JL |
2383 | } |
2384 | ||
c3524383 | 2385 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2386 | { |
c3524383 PB |
2387 | io_queue_next(req); |
2388 | __io_free_req(req); | |
2389 | } | |
8766dd51 | 2390 | |
2d6500d4 | 2391 | struct req_batch { |
5af1d13e PB |
2392 | struct task_struct *task; |
2393 | int task_refs; | |
1b4c351f | 2394 | int ctx_refs; |
2d6500d4 PB |
2395 | }; |
2396 | ||
5af1d13e PB |
2397 | static inline void io_init_req_batch(struct req_batch *rb) |
2398 | { | |
5af1d13e | 2399 | rb->task_refs = 0; |
9ae72463 | 2400 | rb->ctx_refs = 0; |
5af1d13e PB |
2401 | rb->task = NULL; |
2402 | } | |
2403 | ||
2d6500d4 PB |
2404 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2405 | struct req_batch *rb) | |
2406 | { | |
6e833d53 | 2407 | if (rb->task) |
7c660731 | 2408 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2409 | if (rb->ctx_refs) |
2410 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2411 | } |
2412 | ||
6ff119a6 PB |
2413 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2414 | struct io_submit_state *state) | |
2d6500d4 | 2415 | { |
f2f87370 | 2416 | io_queue_next(req); |
2d6500d4 | 2417 | |
e3bc8e9d | 2418 | if (req->task != rb->task) { |
7c660731 PB |
2419 | if (rb->task) |
2420 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2421 | rb->task = req->task; |
2422 | rb->task_refs = 0; | |
5af1d13e | 2423 | } |
e3bc8e9d | 2424 | rb->task_refs++; |
9ae72463 | 2425 | rb->ctx_refs++; |
5af1d13e | 2426 | |
4edf20f9 | 2427 | io_dismantle_req(req); |
bd759045 | 2428 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2429 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2430 | else |
2431 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2432 | } |
2433 | ||
905c172f PB |
2434 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2435 | struct io_ring_ctx *ctx) | |
2436 | { | |
2437 | int i, nr = cs->nr; | |
2438 | struct io_kiocb *req; | |
2439 | struct req_batch rb; | |
2440 | ||
2441 | io_init_req_batch(&rb); | |
2442 | spin_lock_irq(&ctx->completion_lock); | |
2443 | for (i = 0; i < nr; i++) { | |
2444 | req = cs->reqs[i]; | |
2445 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2446 | } | |
2447 | io_commit_cqring(ctx); | |
2448 | spin_unlock_irq(&ctx->completion_lock); | |
2449 | ||
2450 | io_cqring_ev_posted(ctx); | |
2451 | for (i = 0; i < nr; i++) { | |
2452 | req = cs->reqs[i]; | |
2453 | ||
2454 | /* submission and completion refs */ | |
2455 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2456 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2457 | } |
2458 | ||
2459 | io_req_free_batch_finish(ctx, &rb); | |
2460 | cs->nr = 0; | |
2461 | } | |
2462 | ||
ba816ad6 JA |
2463 | /* |
2464 | * Drop reference to request, return next in chain (if there is one) if this | |
2465 | * was the last reference to this request. | |
2466 | */ | |
9b5f7bd9 | 2467 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2468 | { |
9b5f7bd9 PB |
2469 | struct io_kiocb *nxt = NULL; |
2470 | ||
2a44f467 | 2471 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2472 | nxt = io_req_find_next(req); |
4d7dd462 | 2473 | __io_free_req(req); |
2a44f467 | 2474 | } |
9b5f7bd9 | 2475 | return nxt; |
2b188cc1 JA |
2476 | } |
2477 | ||
e65ef56d JA |
2478 | static void io_put_req(struct io_kiocb *req) |
2479 | { | |
2480 | if (refcount_dec_and_test(&req->refs)) | |
2481 | io_free_req(req); | |
2b188cc1 JA |
2482 | } |
2483 | ||
216578e5 PB |
2484 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2485 | { | |
2486 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2487 | ||
2488 | io_free_req(req); | |
2489 | } | |
2490 | ||
2491 | static void io_free_req_deferred(struct io_kiocb *req) | |
2492 | { | |
2493 | int ret; | |
2494 | ||
7cbf1722 | 2495 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2496 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2497 | if (unlikely(ret)) |
2498 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2499 | } |
2500 | ||
2501 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2502 | { | |
2503 | if (refcount_sub_and_test(refs, &req->refs)) | |
2504 | io_free_req_deferred(req); | |
2505 | } | |
2506 | ||
978db57e JA |
2507 | static void io_double_put_req(struct io_kiocb *req) |
2508 | { | |
2509 | /* drop both submit and complete references */ | |
2510 | if (refcount_sub_and_test(2, &req->refs)) | |
2511 | io_free_req(req); | |
2512 | } | |
2513 | ||
6c503150 | 2514 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2515 | { |
2516 | /* See comment at the top of this file */ | |
2517 | smp_rmb(); | |
e23de15f | 2518 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2519 | } |
2520 | ||
fb5ccc98 PB |
2521 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2522 | { | |
2523 | struct io_rings *rings = ctx->rings; | |
2524 | ||
2525 | /* make sure SQ entry isn't read before tail */ | |
2526 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2527 | } | |
2528 | ||
8ff069bf | 2529 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2530 | { |
8ff069bf | 2531 | unsigned int cflags; |
e94f141b | 2532 | |
bcda7baa JA |
2533 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2534 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2535 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2536 | kfree(kbuf); |
2537 | return cflags; | |
e94f141b JA |
2538 | } |
2539 | ||
8ff069bf | 2540 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2541 | { |
4d954c25 | 2542 | struct io_buffer *kbuf; |
bcda7baa | 2543 | |
4d954c25 | 2544 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2545 | return io_put_kbuf(req, kbuf); |
2546 | } | |
2547 | ||
4c6e277c JA |
2548 | static inline bool io_run_task_work(void) |
2549 | { | |
6200b0ae JA |
2550 | /* |
2551 | * Not safe to run on exiting task, and the task_work handling will | |
2552 | * not add work to such a task. | |
2553 | */ | |
2554 | if (unlikely(current->flags & PF_EXITING)) | |
2555 | return false; | |
4c6e277c JA |
2556 | if (current->task_works) { |
2557 | __set_current_state(TASK_RUNNING); | |
2558 | task_work_run(); | |
2559 | return true; | |
2560 | } | |
2561 | ||
2562 | return false; | |
bcda7baa JA |
2563 | } |
2564 | ||
def596e9 JA |
2565 | /* |
2566 | * Find and free completed poll iocbs | |
2567 | */ | |
2568 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2569 | struct list_head *done) | |
2570 | { | |
8237e045 | 2571 | struct req_batch rb; |
def596e9 | 2572 | struct io_kiocb *req; |
bbde017a XW |
2573 | |
2574 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2575 | smp_rmb(); | |
def596e9 | 2576 | |
5af1d13e | 2577 | io_init_req_batch(&rb); |
def596e9 | 2578 | while (!list_empty(done)) { |
bcda7baa JA |
2579 | int cflags = 0; |
2580 | ||
d21ffe7e | 2581 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2582 | list_del(&req->inflight_entry); |
2583 | ||
bbde017a XW |
2584 | if (READ_ONCE(req->result) == -EAGAIN) { |
2585 | req->iopoll_completed = 0; | |
23faba36 | 2586 | if (io_rw_reissue(req)) |
f161340d | 2587 | continue; |
bbde017a | 2588 | } |
def596e9 | 2589 | |
bcda7baa | 2590 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2591 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2592 | |
2593 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2594 | (*nr_events)++; |
2595 | ||
c3524383 | 2596 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2597 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2598 | } |
def596e9 | 2599 | |
09bb8394 | 2600 | io_commit_cqring(ctx); |
80c18e4a | 2601 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2602 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2603 | } |
2604 | ||
def596e9 JA |
2605 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2606 | long min) | |
2607 | { | |
2608 | struct io_kiocb *req, *tmp; | |
2609 | LIST_HEAD(done); | |
2610 | bool spin; | |
2611 | int ret; | |
2612 | ||
2613 | /* | |
2614 | * Only spin for completions if we don't have multiple devices hanging | |
2615 | * off our complete list, and we're under the requested amount. | |
2616 | */ | |
2617 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2618 | ||
2619 | ret = 0; | |
d21ffe7e | 2620 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2621 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2622 | |
2623 | /* | |
581f9810 BM |
2624 | * Move completed and retryable entries to our local lists. |
2625 | * If we find a request that requires polling, break out | |
2626 | * and complete those lists first, if we have entries there. | |
def596e9 | 2627 | */ |
65a6543d | 2628 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2629 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2630 | continue; |
2631 | } | |
2632 | if (!list_empty(&done)) | |
2633 | break; | |
2634 | ||
2635 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2636 | if (ret < 0) | |
2637 | break; | |
2638 | ||
3aadc23e PB |
2639 | /* iopoll may have completed current req */ |
2640 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2641 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2642 | |
def596e9 JA |
2643 | if (ret && spin) |
2644 | spin = false; | |
2645 | ret = 0; | |
2646 | } | |
2647 | ||
2648 | if (!list_empty(&done)) | |
2649 | io_iopoll_complete(ctx, nr_events, &done); | |
2650 | ||
2651 | return ret; | |
2652 | } | |
2653 | ||
2654 | /* | |
d195a66e | 2655 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2656 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2657 | * as a non-spinning completion check. | |
2658 | */ | |
2659 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2660 | long min) | |
2661 | { | |
540e32a0 | 2662 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2663 | int ret; |
2664 | ||
2665 | ret = io_do_iopoll(ctx, nr_events, min); | |
2666 | if (ret < 0) | |
2667 | return ret; | |
eba0a4dd | 2668 | if (*nr_events >= min) |
def596e9 JA |
2669 | return 0; |
2670 | } | |
2671 | ||
2672 | return 1; | |
2673 | } | |
2674 | ||
2675 | /* | |
2676 | * We can't just wait for polled events to come to us, we have to actively | |
2677 | * find and complete them. | |
2678 | */ | |
b2edc0a7 | 2679 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2680 | { |
2681 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2682 | return; | |
2683 | ||
2684 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2685 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2686 | unsigned int nr_events = 0; |
2687 | ||
b2edc0a7 | 2688 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2689 | |
b2edc0a7 PB |
2690 | /* let it sleep and repeat later if can't complete a request */ |
2691 | if (nr_events == 0) | |
2692 | break; | |
08f5439f JA |
2693 | /* |
2694 | * Ensure we allow local-to-the-cpu processing to take place, | |
2695 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2696 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2697 | */ |
3fcee5a6 PB |
2698 | if (need_resched()) { |
2699 | mutex_unlock(&ctx->uring_lock); | |
2700 | cond_resched(); | |
2701 | mutex_lock(&ctx->uring_lock); | |
2702 | } | |
def596e9 JA |
2703 | } |
2704 | mutex_unlock(&ctx->uring_lock); | |
2705 | } | |
2706 | ||
7668b92a | 2707 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2708 | { |
7668b92a | 2709 | unsigned int nr_events = 0; |
2b2ed975 | 2710 | int iters = 0, ret = 0; |
500f9fba | 2711 | |
c7849be9 XW |
2712 | /* |
2713 | * We disallow the app entering submit/complete with polling, but we | |
2714 | * still need to lock the ring to prevent racing with polled issue | |
2715 | * that got punted to a workqueue. | |
2716 | */ | |
2717 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2718 | do { |
a3a0e43f JA |
2719 | /* |
2720 | * Don't enter poll loop if we already have events pending. | |
2721 | * If we do, we can potentially be spinning for commands that | |
2722 | * already triggered a CQE (eg in error). | |
2723 | */ | |
6c503150 PB |
2724 | if (test_bit(0, &ctx->cq_check_overflow)) |
2725 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2726 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2727 | break; |
2728 | ||
500f9fba JA |
2729 | /* |
2730 | * If a submit got punted to a workqueue, we can have the | |
2731 | * application entering polling for a command before it gets | |
2732 | * issued. That app will hold the uring_lock for the duration | |
2733 | * of the poll right here, so we need to take a breather every | |
2734 | * now and then to ensure that the issue has a chance to add | |
2735 | * the poll to the issued list. Otherwise we can spin here | |
2736 | * forever, while the workqueue is stuck trying to acquire the | |
2737 | * very same mutex. | |
2738 | */ | |
2739 | if (!(++iters & 7)) { | |
2740 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2741 | io_run_task_work(); |
500f9fba JA |
2742 | mutex_lock(&ctx->uring_lock); |
2743 | } | |
2744 | ||
7668b92a | 2745 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2746 | if (ret <= 0) |
2747 | break; | |
2748 | ret = 0; | |
7668b92a | 2749 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2750 | |
500f9fba | 2751 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2752 | return ret; |
2753 | } | |
2754 | ||
491381ce | 2755 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2756 | { |
491381ce JA |
2757 | /* |
2758 | * Tell lockdep we inherited freeze protection from submission | |
2759 | * thread. | |
2760 | */ | |
2761 | if (req->flags & REQ_F_ISREG) { | |
2762 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2763 | |
491381ce | 2764 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2765 | } |
491381ce | 2766 | file_end_write(req->file); |
2b188cc1 JA |
2767 | } |
2768 | ||
b63534c4 | 2769 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2770 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2771 | { |
2772 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2773 | int rw, ret; |
b63534c4 | 2774 | struct iov_iter iter; |
b63534c4 | 2775 | |
dc2a6e9a PB |
2776 | /* already prepared */ |
2777 | if (req->async_data) | |
2778 | return true; | |
b63534c4 JA |
2779 | |
2780 | switch (req->opcode) { | |
2781 | case IORING_OP_READV: | |
2782 | case IORING_OP_READ_FIXED: | |
2783 | case IORING_OP_READ: | |
2784 | rw = READ; | |
2785 | break; | |
2786 | case IORING_OP_WRITEV: | |
2787 | case IORING_OP_WRITE_FIXED: | |
2788 | case IORING_OP_WRITE: | |
2789 | rw = WRITE; | |
2790 | break; | |
2791 | default: | |
2792 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2793 | req->opcode); | |
dc2a6e9a | 2794 | return false; |
b63534c4 JA |
2795 | } |
2796 | ||
dc2a6e9a PB |
2797 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2798 | if (ret < 0) | |
2799 | return false; | |
6bf985dc | 2800 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2801 | } |
b63534c4 JA |
2802 | #endif |
2803 | ||
23faba36 | 2804 | static bool io_rw_reissue(struct io_kiocb *req) |
b63534c4 JA |
2805 | { |
2806 | #ifdef CONFIG_BLOCK | |
23faba36 | 2807 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2808 | int ret; |
2809 | ||
75c668cd PB |
2810 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2811 | return false; | |
2812 | if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker()) | |
b63534c4 JA |
2813 | return false; |
2814 | ||
55e6ac1e PB |
2815 | lockdep_assert_held(&req->ctx->uring_lock); |
2816 | ||
28cea78a | 2817 | ret = io_sq_thread_acquire_mm_files(req->ctx, req); |
6d816e08 | 2818 | |
dc2a6e9a | 2819 | if (!ret && io_resubmit_prep(req)) { |
fdee946d JA |
2820 | refcount_inc(&req->refs); |
2821 | io_queue_async_work(req); | |
b63534c4 | 2822 | return true; |
fdee946d | 2823 | } |
dc2a6e9a | 2824 | req_set_fail_links(req); |
b63534c4 JA |
2825 | #endif |
2826 | return false; | |
2827 | } | |
2828 | ||
a1d7c393 | 2829 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2830 | unsigned int issue_flags) |
a1d7c393 | 2831 | { |
2f8e45f1 PB |
2832 | int cflags = 0; |
2833 | ||
23faba36 PB |
2834 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2835 | return; | |
2f8e45f1 PB |
2836 | if (res != req->result) |
2837 | req_set_fail_links(req); | |
23faba36 | 2838 | |
2f8e45f1 PB |
2839 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2840 | kiocb_end_write(req); | |
2841 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2842 | cflags = io_put_rw_kbuf(req); | |
2843 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2844 | } |
2845 | ||
2846 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2847 | { | |
9adbd45d | 2848 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2849 | |
889fca73 | 2850 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2851 | } |
2852 | ||
def596e9 JA |
2853 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2854 | { | |
9adbd45d | 2855 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2856 | |
491381ce JA |
2857 | if (kiocb->ki_flags & IOCB_WRITE) |
2858 | kiocb_end_write(req); | |
def596e9 | 2859 | |
2d7d6792 | 2860 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2861 | req_set_fail_links(req); |
bbde017a XW |
2862 | |
2863 | WRITE_ONCE(req->result, res); | |
2864 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2865 | smp_wmb(); |
2866 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2867 | } |
2868 | ||
2869 | /* | |
2870 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2871 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2872 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2873 | * accessing the kiocb cookie. | |
2874 | */ | |
2e9dbe90 | 2875 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2876 | { |
2877 | struct io_ring_ctx *ctx = req->ctx; | |
2878 | ||
2879 | /* | |
2880 | * Track whether we have multiple files in our lists. This will impact | |
2881 | * how we do polling eventually, not spinning if we're on potentially | |
2882 | * different devices. | |
2883 | */ | |
540e32a0 | 2884 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2885 | ctx->poll_multi_file = false; |
2886 | } else if (!ctx->poll_multi_file) { | |
2887 | struct io_kiocb *list_req; | |
2888 | ||
540e32a0 | 2889 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2890 | inflight_entry); |
9adbd45d | 2891 | if (list_req->file != req->file) |
def596e9 JA |
2892 | ctx->poll_multi_file = true; |
2893 | } | |
2894 | ||
2895 | /* | |
2896 | * For fast devices, IO may have already completed. If it has, add | |
2897 | * it to the front so we find it first. | |
2898 | */ | |
65a6543d | 2899 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2900 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2901 | else |
d21ffe7e | 2902 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2903 | |
2e9dbe90 XW |
2904 | /* |
2905 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2906 | * task context or in io worker task context. If current task context is | |
2907 | * sq thread, we don't need to check whether should wake up sq thread. | |
2908 | */ | |
2909 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2910 | wq_has_sleeper(&ctx->sq_data->wait)) |
2911 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2912 | } |
2913 | ||
9f13c35b PB |
2914 | static inline void io_state_file_put(struct io_submit_state *state) |
2915 | { | |
02b23a9a PB |
2916 | if (state->file_refs) { |
2917 | fput_many(state->file, state->file_refs); | |
2918 | state->file_refs = 0; | |
2919 | } | |
9a56a232 JA |
2920 | } |
2921 | ||
2922 | /* | |
2923 | * Get as many references to a file as we have IOs left in this submission, | |
2924 | * assuming most submissions are for one file, or at least that each file | |
2925 | * has more than one submission. | |
2926 | */ | |
8da11c19 | 2927 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2928 | { |
2929 | if (!state) | |
2930 | return fget(fd); | |
2931 | ||
6e1271e6 | 2932 | if (state->file_refs) { |
9a56a232 | 2933 | if (state->fd == fd) { |
6e1271e6 | 2934 | state->file_refs--; |
9a56a232 JA |
2935 | return state->file; |
2936 | } | |
02b23a9a | 2937 | io_state_file_put(state); |
9a56a232 JA |
2938 | } |
2939 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2940 | if (unlikely(!state->file)) |
9a56a232 JA |
2941 | return NULL; |
2942 | ||
2943 | state->fd = fd; | |
6e1271e6 | 2944 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2945 | return state->file; |
2946 | } | |
2947 | ||
4503b767 JA |
2948 | static bool io_bdev_nowait(struct block_device *bdev) |
2949 | { | |
9ba0d0c8 | 2950 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2951 | } |
2952 | ||
2b188cc1 JA |
2953 | /* |
2954 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2955 | * any file. For now, just ensure that anything potentially problematic is done | |
2956 | * inline. | |
2957 | */ | |
af197f50 | 2958 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2959 | { |
2960 | umode_t mode = file_inode(file)->i_mode; | |
2961 | ||
4503b767 | 2962 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2963 | if (IS_ENABLED(CONFIG_BLOCK) && |
2964 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2965 | return true; |
2966 | return false; | |
2967 | } | |
2968 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2969 | return true; |
4503b767 | 2970 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2971 | if (IS_ENABLED(CONFIG_BLOCK) && |
2972 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2973 | file->f_op != &io_uring_fops) |
2974 | return true; | |
2975 | return false; | |
2976 | } | |
2b188cc1 | 2977 | |
c5b85625 JA |
2978 | /* any ->read/write should understand O_NONBLOCK */ |
2979 | if (file->f_flags & O_NONBLOCK) | |
2980 | return true; | |
2981 | ||
af197f50 JA |
2982 | if (!(file->f_mode & FMODE_NOWAIT)) |
2983 | return false; | |
2984 | ||
2985 | if (rw == READ) | |
2986 | return file->f_op->read_iter != NULL; | |
2987 | ||
2988 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2989 | } |
2990 | ||
a88fc400 | 2991 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2992 | { |
def596e9 | 2993 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2994 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2995 | struct file *file = req->file; |
09bb8394 JA |
2996 | unsigned ioprio; |
2997 | int ret; | |
2b188cc1 | 2998 | |
75c668cd | 2999 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
3000 | req->flags |= REQ_F_ISREG; |
3001 | ||
2b188cc1 | 3002 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 3003 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 3004 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 3005 | kiocb->ki_pos = file->f_pos; |
ba04291e | 3006 | } |
2b188cc1 | 3007 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
3008 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
3009 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
3010 | if (unlikely(ret)) | |
3011 | return ret; | |
2b188cc1 | 3012 | |
75c668cd PB |
3013 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
3014 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
3015 | req->flags |= REQ_F_NOWAIT; | |
3016 | ||
2b188cc1 JA |
3017 | ioprio = READ_ONCE(sqe->ioprio); |
3018 | if (ioprio) { | |
3019 | ret = ioprio_check_cap(ioprio); | |
3020 | if (ret) | |
09bb8394 | 3021 | return ret; |
2b188cc1 JA |
3022 | |
3023 | kiocb->ki_ioprio = ioprio; | |
3024 | } else | |
3025 | kiocb->ki_ioprio = get_current_ioprio(); | |
3026 | ||
def596e9 | 3027 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
3028 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
3029 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 3030 | return -EOPNOTSUPP; |
2b188cc1 | 3031 | |
def596e9 JA |
3032 | kiocb->ki_flags |= IOCB_HIPRI; |
3033 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 3034 | req->iopoll_completed = 0; |
def596e9 | 3035 | } else { |
09bb8394 JA |
3036 | if (kiocb->ki_flags & IOCB_HIPRI) |
3037 | return -EINVAL; | |
def596e9 JA |
3038 | kiocb->ki_complete = io_complete_rw; |
3039 | } | |
9adbd45d | 3040 | |
3529d8c2 JA |
3041 | req->rw.addr = READ_ONCE(sqe->addr); |
3042 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 3043 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 3044 | return 0; |
2b188cc1 JA |
3045 | } |
3046 | ||
3047 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3048 | { | |
3049 | switch (ret) { | |
3050 | case -EIOCBQUEUED: | |
3051 | break; | |
3052 | case -ERESTARTSYS: | |
3053 | case -ERESTARTNOINTR: | |
3054 | case -ERESTARTNOHAND: | |
3055 | case -ERESTART_RESTARTBLOCK: | |
3056 | /* | |
3057 | * We can't just restart the syscall, since previously | |
3058 | * submitted sqes may already be in progress. Just fail this | |
3059 | * IO with EINTR. | |
3060 | */ | |
3061 | ret = -EINTR; | |
df561f66 | 3062 | fallthrough; |
2b188cc1 JA |
3063 | default: |
3064 | kiocb->ki_complete(kiocb, ret, 0); | |
3065 | } | |
3066 | } | |
3067 | ||
a1d7c393 | 3068 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 3069 | unsigned int issue_flags) |
ba816ad6 | 3070 | { |
ba04291e | 3071 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 3072 | struct io_async_rw *io = req->async_data; |
ba04291e | 3073 | |
227c0c96 | 3074 | /* add previously done IO, if any */ |
e8c2bc1f | 3075 | if (io && io->bytes_done > 0) { |
227c0c96 | 3076 | if (ret < 0) |
e8c2bc1f | 3077 | ret = io->bytes_done; |
227c0c96 | 3078 | else |
e8c2bc1f | 3079 | ret += io->bytes_done; |
227c0c96 JA |
3080 | } |
3081 | ||
ba04291e JA |
3082 | if (req->flags & REQ_F_CUR_POS) |
3083 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 3084 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 3085 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
3086 | else |
3087 | io_rw_done(kiocb, ret); | |
3088 | } | |
3089 | ||
847595de | 3090 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 3091 | { |
9adbd45d JA |
3092 | struct io_ring_ctx *ctx = req->ctx; |
3093 | size_t len = req->rw.len; | |
edafccee | 3094 | struct io_mapped_ubuf *imu; |
4be1c615 | 3095 | u16 index, buf_index = req->buf_index; |
edafccee JA |
3096 | size_t offset; |
3097 | u64 buf_addr; | |
3098 | ||
edafccee JA |
3099 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3100 | return -EFAULT; | |
edafccee JA |
3101 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3102 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 3103 | buf_addr = req->rw.addr; |
edafccee JA |
3104 | |
3105 | /* overflow */ | |
3106 | if (buf_addr + len < buf_addr) | |
3107 | return -EFAULT; | |
3108 | /* not inside the mapped region */ | |
3109 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
3110 | return -EFAULT; | |
3111 | ||
3112 | /* | |
3113 | * May not be a start of buffer, set size appropriately | |
3114 | * and advance us to the beginning. | |
3115 | */ | |
3116 | offset = buf_addr - imu->ubuf; | |
3117 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3118 | |
3119 | if (offset) { | |
3120 | /* | |
3121 | * Don't use iov_iter_advance() here, as it's really slow for | |
3122 | * using the latter parts of a big fixed buffer - it iterates | |
3123 | * over each segment manually. We can cheat a bit here, because | |
3124 | * we know that: | |
3125 | * | |
3126 | * 1) it's a BVEC iter, we set it up | |
3127 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3128 | * first and last bvec | |
3129 | * | |
3130 | * So just find our index, and adjust the iterator afterwards. | |
3131 | * If the offset is within the first bvec (or the whole first | |
3132 | * bvec, just use iov_iter_advance(). This makes it easier | |
3133 | * since we can just skip the first segment, which may not | |
3134 | * be PAGE_SIZE aligned. | |
3135 | */ | |
3136 | const struct bio_vec *bvec = imu->bvec; | |
3137 | ||
3138 | if (offset <= bvec->bv_len) { | |
3139 | iov_iter_advance(iter, offset); | |
3140 | } else { | |
3141 | unsigned long seg_skip; | |
3142 | ||
3143 | /* skip first vec */ | |
3144 | offset -= bvec->bv_len; | |
3145 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3146 | ||
3147 | iter->bvec = bvec + seg_skip; | |
3148 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3149 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3150 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3151 | } |
3152 | } | |
3153 | ||
847595de | 3154 | return 0; |
edafccee JA |
3155 | } |
3156 | ||
bcda7baa JA |
3157 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3158 | { | |
3159 | if (needs_lock) | |
3160 | mutex_unlock(&ctx->uring_lock); | |
3161 | } | |
3162 | ||
3163 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3164 | { | |
3165 | /* | |
3166 | * "Normal" inline submissions always hold the uring_lock, since we | |
3167 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3168 | * The only exception is when we've detached the request and issue it | |
3169 | * from an async worker thread, grab the lock for that case. | |
3170 | */ | |
3171 | if (needs_lock) | |
3172 | mutex_lock(&ctx->uring_lock); | |
3173 | } | |
3174 | ||
3175 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3176 | int bgid, struct io_buffer *kbuf, | |
3177 | bool needs_lock) | |
3178 | { | |
3179 | struct io_buffer *head; | |
3180 | ||
3181 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3182 | return kbuf; | |
3183 | ||
3184 | io_ring_submit_lock(req->ctx, needs_lock); | |
3185 | ||
3186 | lockdep_assert_held(&req->ctx->uring_lock); | |
3187 | ||
3188 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
3189 | if (head) { | |
3190 | if (!list_empty(&head->list)) { | |
3191 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3192 | list); | |
3193 | list_del(&kbuf->list); | |
3194 | } else { | |
3195 | kbuf = head; | |
3196 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
3197 | } | |
3198 | if (*len > kbuf->len) | |
3199 | *len = kbuf->len; | |
3200 | } else { | |
3201 | kbuf = ERR_PTR(-ENOBUFS); | |
3202 | } | |
3203 | ||
3204 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3205 | ||
3206 | return kbuf; | |
3207 | } | |
3208 | ||
4d954c25 JA |
3209 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3210 | bool needs_lock) | |
3211 | { | |
3212 | struct io_buffer *kbuf; | |
4f4eeba8 | 3213 | u16 bgid; |
4d954c25 JA |
3214 | |
3215 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3216 | bgid = req->buf_index; |
4d954c25 JA |
3217 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3218 | if (IS_ERR(kbuf)) | |
3219 | return kbuf; | |
3220 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3221 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3222 | return u64_to_user_ptr(kbuf->addr); | |
3223 | } | |
3224 | ||
3225 | #ifdef CONFIG_COMPAT | |
3226 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3227 | bool needs_lock) | |
3228 | { | |
3229 | struct compat_iovec __user *uiov; | |
3230 | compat_ssize_t clen; | |
3231 | void __user *buf; | |
3232 | ssize_t len; | |
3233 | ||
3234 | uiov = u64_to_user_ptr(req->rw.addr); | |
3235 | if (!access_ok(uiov, sizeof(*uiov))) | |
3236 | return -EFAULT; | |
3237 | if (__get_user(clen, &uiov->iov_len)) | |
3238 | return -EFAULT; | |
3239 | if (clen < 0) | |
3240 | return -EINVAL; | |
3241 | ||
3242 | len = clen; | |
3243 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3244 | if (IS_ERR(buf)) | |
3245 | return PTR_ERR(buf); | |
3246 | iov[0].iov_base = buf; | |
3247 | iov[0].iov_len = (compat_size_t) len; | |
3248 | return 0; | |
3249 | } | |
3250 | #endif | |
3251 | ||
3252 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3253 | bool needs_lock) | |
3254 | { | |
3255 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3256 | void __user *buf; | |
3257 | ssize_t len; | |
3258 | ||
3259 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3260 | return -EFAULT; | |
3261 | ||
3262 | len = iov[0].iov_len; | |
3263 | if (len < 0) | |
3264 | return -EINVAL; | |
3265 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3266 | if (IS_ERR(buf)) | |
3267 | return PTR_ERR(buf); | |
3268 | iov[0].iov_base = buf; | |
3269 | iov[0].iov_len = len; | |
3270 | return 0; | |
3271 | } | |
3272 | ||
3273 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3274 | bool needs_lock) | |
3275 | { | |
dddb3e26 JA |
3276 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3277 | struct io_buffer *kbuf; | |
3278 | ||
3279 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3280 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3281 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3282 | return 0; |
dddb3e26 | 3283 | } |
dd201662 | 3284 | if (req->rw.len != 1) |
4d954c25 JA |
3285 | return -EINVAL; |
3286 | ||
3287 | #ifdef CONFIG_COMPAT | |
3288 | if (req->ctx->compat) | |
3289 | return io_compat_import(req, iov, needs_lock); | |
3290 | #endif | |
3291 | ||
3292 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3293 | } | |
3294 | ||
847595de PB |
3295 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3296 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3297 | { |
9adbd45d JA |
3298 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3299 | size_t sqe_len = req->rw.len; | |
847595de | 3300 | u8 opcode = req->opcode; |
4d954c25 | 3301 | ssize_t ret; |
edafccee | 3302 | |
7d009165 | 3303 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3304 | *iovec = NULL; |
9adbd45d | 3305 | return io_import_fixed(req, rw, iter); |
edafccee | 3306 | } |
2b188cc1 | 3307 | |
bcda7baa | 3308 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3309 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3310 | return -EINVAL; |
3311 | ||
3a6820f2 | 3312 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3313 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3314 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3315 | if (IS_ERR(buf)) |
4d954c25 | 3316 | return PTR_ERR(buf); |
3f9d6441 | 3317 | req->rw.len = sqe_len; |
bcda7baa JA |
3318 | } |
3319 | ||
3a6820f2 JA |
3320 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3321 | *iovec = NULL; | |
10fc72e4 | 3322 | return ret; |
3a6820f2 JA |
3323 | } |
3324 | ||
4d954c25 JA |
3325 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3326 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3327 | if (!ret) |
3328 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3329 | *iovec = NULL; |
3330 | return ret; | |
3331 | } | |
3332 | ||
89cd35c5 CH |
3333 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3334 | req->ctx->compat); | |
2b188cc1 JA |
3335 | } |
3336 | ||
0fef9483 JA |
3337 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3338 | { | |
5b09e37e | 3339 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3340 | } |
3341 | ||
31b51510 | 3342 | /* |
32960613 JA |
3343 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3344 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3345 | */ |
4017eb91 | 3346 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3347 | { |
4017eb91 JA |
3348 | struct kiocb *kiocb = &req->rw.kiocb; |
3349 | struct file *file = req->file; | |
32960613 JA |
3350 | ssize_t ret = 0; |
3351 | ||
3352 | /* | |
3353 | * Don't support polled IO through this interface, and we can't | |
3354 | * support non-blocking either. For the latter, this just causes | |
3355 | * the kiocb to be handled from an async context. | |
3356 | */ | |
3357 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3358 | return -EOPNOTSUPP; | |
3359 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3360 | return -EAGAIN; | |
3361 | ||
3362 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3363 | struct iovec iovec; |
32960613 JA |
3364 | ssize_t nr; |
3365 | ||
311ae9e1 PB |
3366 | if (!iov_iter_is_bvec(iter)) { |
3367 | iovec = iov_iter_iovec(iter); | |
3368 | } else { | |
4017eb91 JA |
3369 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3370 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3371 | } |
3372 | ||
32960613 JA |
3373 | if (rw == READ) { |
3374 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3375 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3376 | } else { |
3377 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3378 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3379 | } |
3380 | ||
3381 | if (nr < 0) { | |
3382 | if (!ret) | |
3383 | ret = nr; | |
3384 | break; | |
3385 | } | |
3386 | ret += nr; | |
3387 | if (nr != iovec.iov_len) | |
3388 | break; | |
4017eb91 JA |
3389 | req->rw.len -= nr; |
3390 | req->rw.addr += nr; | |
32960613 JA |
3391 | iov_iter_advance(iter, nr); |
3392 | } | |
3393 | ||
3394 | return ret; | |
3395 | } | |
3396 | ||
ff6165b2 JA |
3397 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3398 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3399 | { |
e8c2bc1f | 3400 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3401 | |
ff6165b2 | 3402 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3403 | rw->free_iovec = iovec; |
227c0c96 | 3404 | rw->bytes_done = 0; |
ff6165b2 | 3405 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3406 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3407 | return; |
b64e3444 | 3408 | if (!iovec) { |
ff6165b2 JA |
3409 | unsigned iov_off = 0; |
3410 | ||
3411 | rw->iter.iov = rw->fast_iov; | |
3412 | if (iter->iov != fast_iov) { | |
3413 | iov_off = iter->iov - fast_iov; | |
3414 | rw->iter.iov += iov_off; | |
3415 | } | |
3416 | if (rw->fast_iov != fast_iov) | |
3417 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3418 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3419 | } else { |
3420 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3421 | } |
3422 | } | |
3423 | ||
e8c2bc1f | 3424 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3425 | { |
e8c2bc1f JA |
3426 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3427 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3428 | return req->async_data == NULL; | |
3d9932a8 XW |
3429 | } |
3430 | ||
e8c2bc1f | 3431 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3432 | { |
e8c2bc1f | 3433 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3434 | return 0; |
3d9932a8 | 3435 | |
e8c2bc1f | 3436 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3437 | } |
3438 | ||
ff6165b2 JA |
3439 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3440 | const struct iovec *fast_iov, | |
227c0c96 | 3441 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3442 | { |
e8c2bc1f | 3443 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3444 | return 0; |
e8c2bc1f | 3445 | if (!req->async_data) { |
6bf985dc PB |
3446 | if (__io_alloc_async_data(req)) { |
3447 | kfree(iovec); | |
5d204bcf | 3448 | return -ENOMEM; |
6bf985dc | 3449 | } |
b7bb4f7d | 3450 | |
ff6165b2 | 3451 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3452 | } |
b7bb4f7d | 3453 | return 0; |
f67676d1 JA |
3454 | } |
3455 | ||
73debe68 | 3456 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3457 | { |
e8c2bc1f | 3458 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3459 | struct iovec *iov = iorw->fast_iov; |
847595de | 3460 | int ret; |
c3e330a4 | 3461 | |
2846c481 | 3462 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3463 | if (unlikely(ret < 0)) |
3464 | return ret; | |
3465 | ||
ab0b196c PB |
3466 | iorw->bytes_done = 0; |
3467 | iorw->free_iovec = iov; | |
3468 | if (iov) | |
3469 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3470 | return 0; |
3471 | } | |
3472 | ||
73debe68 | 3473 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3474 | { |
3529d8c2 JA |
3475 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3476 | return -EBADF; | |
93642ef8 | 3477 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3478 | } |
3479 | ||
c1dd91d1 JA |
3480 | /* |
3481 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3482 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3483 | * This gets called when the page is unlocked, and we generally expect that to | |
3484 | * happen when the page IO is completed and the page is now uptodate. This will | |
3485 | * queue a task_work based retry of the operation, attempting to copy the data | |
3486 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3487 | * do a thread based blocking retry of the operation. That's the unexpected | |
3488 | * slow path. | |
3489 | */ | |
bcf5a063 JA |
3490 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3491 | int sync, void *arg) | |
3492 | { | |
3493 | struct wait_page_queue *wpq; | |
3494 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3495 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3496 | |
3497 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3498 | ||
cdc8fcb4 LT |
3499 | if (!wake_page_match(wpq, key)) |
3500 | return 0; | |
3501 | ||
c8d317aa | 3502 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3503 | list_del_init(&wait->entry); |
3504 | ||
bcf5a063 JA |
3505 | /* submit ref gets dropped, acquire a new one */ |
3506 | refcount_inc(&req->refs); | |
921b9054 | 3507 | io_req_task_queue(req); |
bcf5a063 JA |
3508 | return 1; |
3509 | } | |
3510 | ||
c1dd91d1 JA |
3511 | /* |
3512 | * This controls whether a given IO request should be armed for async page | |
3513 | * based retry. If we return false here, the request is handed to the async | |
3514 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3515 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3516 | * will either succeed because the page is now uptodate and unlocked, or it | |
3517 | * will register a callback when the page is unlocked at IO completion. Through | |
3518 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3519 | * That retry will attempt the buffered read again. The retry will generally | |
3520 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3521 | * async worker threads for a blocking retry. | |
3522 | */ | |
227c0c96 | 3523 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3524 | { |
e8c2bc1f JA |
3525 | struct io_async_rw *rw = req->async_data; |
3526 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3527 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3528 | |
bcf5a063 JA |
3529 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3530 | if (req->flags & REQ_F_NOWAIT) | |
3531 | return false; | |
f67676d1 | 3532 | |
227c0c96 | 3533 | /* Only for buffered IO */ |
3b2a4439 | 3534 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3535 | return false; |
3b2a4439 | 3536 | |
bcf5a063 JA |
3537 | /* |
3538 | * just use poll if we can, and don't attempt if the fs doesn't | |
3539 | * support callback based unlocks | |
3540 | */ | |
3541 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3542 | return false; | |
f67676d1 | 3543 | |
3b2a4439 JA |
3544 | wait->wait.func = io_async_buf_func; |
3545 | wait->wait.private = req; | |
3546 | wait->wait.flags = 0; | |
3547 | INIT_LIST_HEAD(&wait->wait.entry); | |
3548 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3549 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3550 | kiocb->ki_waitq = wait; |
3b2a4439 | 3551 | return true; |
bcf5a063 JA |
3552 | } |
3553 | ||
3554 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3555 | { | |
3556 | if (req->file->f_op->read_iter) | |
3557 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3558 | else if (req->file->f_op->read) |
4017eb91 | 3559 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3560 | else |
3561 | return -EINVAL; | |
f67676d1 JA |
3562 | } |
3563 | ||
889fca73 | 3564 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3565 | { |
3566 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3567 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3568 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3569 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3570 | ssize_t io_size, ret, ret2; |
45d189c6 | 3571 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3572 | |
2846c481 | 3573 | if (rw) { |
e8c2bc1f | 3574 | iter = &rw->iter; |
2846c481 PB |
3575 | iovec = NULL; |
3576 | } else { | |
3577 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3578 | if (ret < 0) | |
3579 | return ret; | |
3580 | } | |
632546c4 | 3581 | io_size = iov_iter_count(iter); |
fa15bafb | 3582 | req->result = io_size; |
2b188cc1 | 3583 | |
fd6c2e4c JA |
3584 | /* Ensure we clear previously set non-block flag */ |
3585 | if (!force_nonblock) | |
29de5f6a | 3586 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3587 | else |
3588 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3589 | ||
24c74678 | 3590 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3591 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3592 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3593 | return ret ?: -EAGAIN; |
6713e7a6 | 3594 | } |
9e645e11 | 3595 | |
632546c4 | 3596 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3597 | if (unlikely(ret)) { |
3598 | kfree(iovec); | |
3599 | return ret; | |
3600 | } | |
2b188cc1 | 3601 | |
227c0c96 | 3602 | ret = io_iter_do_read(req, iter); |
32960613 | 3603 | |
57cd657b | 3604 | if (ret == -EIOCBQUEUED) { |
fe1cdd55 | 3605 | goto out_free; |
227c0c96 | 3606 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3607 | /* IOPOLL retry should happen for io-wq threads */ |
3608 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3609 | goto done; |
75c668cd PB |
3610 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3611 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3612 | goto done; |
84216315 | 3613 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3614 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3615 | ret = 0; |
7335e3bf | 3616 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3617 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3618 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3619 | goto done; |
1a2cc0ce | 3620 | } |
227c0c96 | 3621 | |
227c0c96 | 3622 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3623 | if (ret2) |
3624 | return ret2; | |
3625 | ||
fe1cdd55 | 3626 | iovec = NULL; |
e8c2bc1f | 3627 | rw = req->async_data; |
227c0c96 | 3628 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3629 | iter = &rw->iter; |
227c0c96 | 3630 | |
b23df91b PB |
3631 | do { |
3632 | io_size -= ret; | |
3633 | rw->bytes_done += ret; | |
3634 | /* if we can retry, do so with the callbacks armed */ | |
3635 | if (!io_rw_should_retry(req)) { | |
3636 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3637 | return -EAGAIN; | |
3638 | } | |
3639 | ||
3640 | /* | |
3641 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3642 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3643 | * desired page gets unlocked. We can also get a partial read | |
3644 | * here, and if we do, then just retry at the new offset. | |
3645 | */ | |
3646 | ret = io_iter_do_read(req, iter); | |
3647 | if (ret == -EIOCBQUEUED) | |
3648 | return 0; | |
3649 | /* we got some bytes, but not all. retry. */ | |
3650 | } while (ret > 0 && ret < io_size); | |
227c0c96 | 3651 | done: |
889fca73 | 3652 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3653 | out_free: |
3654 | /* it's faster to check here then delegate to kfree */ | |
3655 | if (iovec) | |
3656 | kfree(iovec); | |
5ea5dd45 | 3657 | return 0; |
2b188cc1 JA |
3658 | } |
3659 | ||
73debe68 | 3660 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3661 | { |
3529d8c2 JA |
3662 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3663 | return -EBADF; | |
93642ef8 | 3664 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3665 | } |
3666 | ||
889fca73 | 3667 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3668 | { |
3669 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3670 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3671 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3672 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3673 | ssize_t ret, ret2, io_size; |
45d189c6 | 3674 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3675 | |
2846c481 | 3676 | if (rw) { |
e8c2bc1f | 3677 | iter = &rw->iter; |
2846c481 PB |
3678 | iovec = NULL; |
3679 | } else { | |
3680 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3681 | if (ret < 0) | |
3682 | return ret; | |
3683 | } | |
632546c4 | 3684 | io_size = iov_iter_count(iter); |
fa15bafb | 3685 | req->result = io_size; |
2b188cc1 | 3686 | |
fd6c2e4c JA |
3687 | /* Ensure we clear previously set non-block flag */ |
3688 | if (!force_nonblock) | |
a88fc400 PB |
3689 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3690 | else | |
3691 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3692 | |
24c74678 | 3693 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3694 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3695 | goto copy_iov; |
31b51510 | 3696 | |
10d59345 JA |
3697 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3698 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3699 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3700 | goto copy_iov; |
31b51510 | 3701 | |
632546c4 | 3702 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3703 | if (unlikely(ret)) |
3704 | goto out_free; | |
4ed734b0 | 3705 | |
fa15bafb PB |
3706 | /* |
3707 | * Open-code file_start_write here to grab freeze protection, | |
3708 | * which will be released by another thread in | |
3709 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3710 | * released so that it doesn't complain about the held lock when | |
3711 | * we return to userspace. | |
3712 | */ | |
3713 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3714 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3715 | __sb_writers_release(file_inode(req->file)->i_sb, |
3716 | SB_FREEZE_WRITE); | |
3717 | } | |
3718 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3719 | |
fa15bafb | 3720 | if (req->file->f_op->write_iter) |
ff6165b2 | 3721 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3722 | else if (req->file->f_op->write) |
4017eb91 | 3723 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3724 | else |
3725 | ret2 = -EINVAL; | |
4ed734b0 | 3726 | |
fa15bafb PB |
3727 | /* |
3728 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3729 | * retry them without IOCB_NOWAIT. | |
3730 | */ | |
3731 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3732 | ret2 = -EAGAIN; | |
75c668cd PB |
3733 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3734 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3735 | goto done; |
fa15bafb | 3736 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3737 | /* IOPOLL retry should happen for io-wq threads */ |
3738 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3739 | goto copy_iov; | |
355afaeb | 3740 | done: |
889fca73 | 3741 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3742 | } else { |
f67676d1 | 3743 | copy_iov: |
84216315 | 3744 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3745 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3746 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3747 | return ret ?: -EAGAIN; |
2b188cc1 | 3748 | } |
31b51510 | 3749 | out_free: |
f261c168 | 3750 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3751 | if (iovec) |
6f2cc166 | 3752 | kfree(iovec); |
2b188cc1 JA |
3753 | return ret; |
3754 | } | |
3755 | ||
80a261fd JA |
3756 | static int io_renameat_prep(struct io_kiocb *req, |
3757 | const struct io_uring_sqe *sqe) | |
3758 | { | |
3759 | struct io_rename *ren = &req->rename; | |
3760 | const char __user *oldf, *newf; | |
3761 | ||
3762 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3763 | return -EBADF; | |
3764 | ||
3765 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3766 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3767 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3768 | ren->new_dfd = READ_ONCE(sqe->len); | |
3769 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3770 | ||
3771 | ren->oldpath = getname(oldf); | |
3772 | if (IS_ERR(ren->oldpath)) | |
3773 | return PTR_ERR(ren->oldpath); | |
3774 | ||
3775 | ren->newpath = getname(newf); | |
3776 | if (IS_ERR(ren->newpath)) { | |
3777 | putname(ren->oldpath); | |
3778 | return PTR_ERR(ren->newpath); | |
3779 | } | |
3780 | ||
3781 | req->flags |= REQ_F_NEED_CLEANUP; | |
3782 | return 0; | |
3783 | } | |
3784 | ||
45d189c6 | 3785 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3786 | { |
3787 | struct io_rename *ren = &req->rename; | |
3788 | int ret; | |
3789 | ||
45d189c6 | 3790 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3791 | return -EAGAIN; |
3792 | ||
3793 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3794 | ren->newpath, ren->flags); | |
3795 | ||
3796 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3797 | if (ret < 0) | |
3798 | req_set_fail_links(req); | |
3799 | io_req_complete(req, ret); | |
3800 | return 0; | |
3801 | } | |
3802 | ||
14a1143b JA |
3803 | static int io_unlinkat_prep(struct io_kiocb *req, |
3804 | const struct io_uring_sqe *sqe) | |
3805 | { | |
3806 | struct io_unlink *un = &req->unlink; | |
3807 | const char __user *fname; | |
3808 | ||
3809 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3810 | return -EBADF; | |
3811 | ||
3812 | un->dfd = READ_ONCE(sqe->fd); | |
3813 | ||
3814 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3815 | if (un->flags & ~AT_REMOVEDIR) | |
3816 | return -EINVAL; | |
3817 | ||
3818 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3819 | un->filename = getname(fname); | |
3820 | if (IS_ERR(un->filename)) | |
3821 | return PTR_ERR(un->filename); | |
3822 | ||
3823 | req->flags |= REQ_F_NEED_CLEANUP; | |
3824 | return 0; | |
3825 | } | |
3826 | ||
45d189c6 | 3827 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3828 | { |
3829 | struct io_unlink *un = &req->unlink; | |
3830 | int ret; | |
3831 | ||
45d189c6 | 3832 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3833 | return -EAGAIN; |
3834 | ||
3835 | if (un->flags & AT_REMOVEDIR) | |
3836 | ret = do_rmdir(un->dfd, un->filename); | |
3837 | else | |
3838 | ret = do_unlinkat(un->dfd, un->filename); | |
3839 | ||
3840 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3841 | if (ret < 0) | |
3842 | req_set_fail_links(req); | |
3843 | io_req_complete(req, ret); | |
3844 | return 0; | |
3845 | } | |
3846 | ||
36f4fa68 JA |
3847 | static int io_shutdown_prep(struct io_kiocb *req, |
3848 | const struct io_uring_sqe *sqe) | |
3849 | { | |
3850 | #if defined(CONFIG_NET) | |
3851 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3852 | return -EINVAL; | |
3853 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3854 | sqe->buf_index) | |
3855 | return -EINVAL; | |
3856 | ||
3857 | req->shutdown.how = READ_ONCE(sqe->len); | |
3858 | return 0; | |
3859 | #else | |
3860 | return -EOPNOTSUPP; | |
3861 | #endif | |
3862 | } | |
3863 | ||
45d189c6 | 3864 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3865 | { |
3866 | #if defined(CONFIG_NET) | |
3867 | struct socket *sock; | |
3868 | int ret; | |
3869 | ||
45d189c6 | 3870 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3871 | return -EAGAIN; |
3872 | ||
48aba79b | 3873 | sock = sock_from_file(req->file); |
36f4fa68 | 3874 | if (unlikely(!sock)) |
48aba79b | 3875 | return -ENOTSOCK; |
36f4fa68 JA |
3876 | |
3877 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3878 | if (ret < 0) |
3879 | req_set_fail_links(req); | |
36f4fa68 JA |
3880 | io_req_complete(req, ret); |
3881 | return 0; | |
3882 | #else | |
3883 | return -EOPNOTSUPP; | |
3884 | #endif | |
3885 | } | |
3886 | ||
f2a8d5c7 PB |
3887 | static int __io_splice_prep(struct io_kiocb *req, |
3888 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3889 | { |
3890 | struct io_splice* sp = &req->splice; | |
3891 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3892 | |
3232dd02 PB |
3893 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3894 | return -EINVAL; | |
7d67af2c PB |
3895 | |
3896 | sp->file_in = NULL; | |
7d67af2c PB |
3897 | sp->len = READ_ONCE(sqe->len); |
3898 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3899 | ||
3900 | if (unlikely(sp->flags & ~valid_flags)) | |
3901 | return -EINVAL; | |
3902 | ||
8371adf5 PB |
3903 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3904 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3905 | if (!sp->file_in) | |
3906 | return -EBADF; | |
7d67af2c PB |
3907 | req->flags |= REQ_F_NEED_CLEANUP; |
3908 | ||
7cdaf587 XW |
3909 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3910 | /* | |
3911 | * Splice operation will be punted aync, and here need to | |
3912 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3913 | */ | |
3914 | io_req_init_async(req); | |
7d67af2c | 3915 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3916 | } |
7d67af2c PB |
3917 | |
3918 | return 0; | |
3919 | } | |
3920 | ||
f2a8d5c7 PB |
3921 | static int io_tee_prep(struct io_kiocb *req, |
3922 | const struct io_uring_sqe *sqe) | |
3923 | { | |
3924 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3925 | return -EINVAL; | |
3926 | return __io_splice_prep(req, sqe); | |
3927 | } | |
3928 | ||
45d189c6 | 3929 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3930 | { |
3931 | struct io_splice *sp = &req->splice; | |
3932 | struct file *in = sp->file_in; | |
3933 | struct file *out = sp->file_out; | |
3934 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3935 | long ret = 0; | |
3936 | ||
45d189c6 | 3937 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3938 | return -EAGAIN; |
3939 | if (sp->len) | |
3940 | ret = do_tee(in, out, sp->len, flags); | |
3941 | ||
3942 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3943 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3944 | ||
f2a8d5c7 PB |
3945 | if (ret != sp->len) |
3946 | req_set_fail_links(req); | |
e1e16097 | 3947 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3948 | return 0; |
3949 | } | |
3950 | ||
3951 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3952 | { | |
3953 | struct io_splice* sp = &req->splice; | |
3954 | ||
3955 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3956 | sp->off_out = READ_ONCE(sqe->off); | |
3957 | return __io_splice_prep(req, sqe); | |
3958 | } | |
3959 | ||
45d189c6 | 3960 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3961 | { |
3962 | struct io_splice *sp = &req->splice; | |
3963 | struct file *in = sp->file_in; | |
3964 | struct file *out = sp->file_out; | |
3965 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3966 | loff_t *poff_in, *poff_out; | |
c9687426 | 3967 | long ret = 0; |
7d67af2c | 3968 | |
45d189c6 | 3969 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3970 | return -EAGAIN; |
7d67af2c PB |
3971 | |
3972 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3973 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3974 | |
948a7749 | 3975 | if (sp->len) |
c9687426 | 3976 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3977 | |
3978 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3979 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3980 | ||
7d67af2c PB |
3981 | if (ret != sp->len) |
3982 | req_set_fail_links(req); | |
e1e16097 | 3983 | io_req_complete(req, ret); |
7d67af2c PB |
3984 | return 0; |
3985 | } | |
3986 | ||
2b188cc1 JA |
3987 | /* |
3988 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3989 | */ | |
889fca73 | 3990 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3991 | { |
3992 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3993 | |
def596e9 JA |
3994 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3995 | return -EINVAL; | |
3996 | ||
889fca73 | 3997 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3998 | return 0; |
3999 | } | |
4000 | ||
1155c76a | 4001 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4002 | { |
6b06314c | 4003 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4004 | |
09bb8394 JA |
4005 | if (!req->file) |
4006 | return -EBADF; | |
c992fe29 | 4007 | |
6b06314c | 4008 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4009 | return -EINVAL; |
edafccee | 4010 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
4011 | return -EINVAL; |
4012 | ||
8ed8d3c3 JA |
4013 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4014 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4015 | return -EINVAL; | |
4016 | ||
4017 | req->sync.off = READ_ONCE(sqe->off); | |
4018 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4019 | return 0; |
4020 | } | |
4021 | ||
45d189c6 | 4022 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4023 | { |
8ed8d3c3 | 4024 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4025 | int ret; |
4026 | ||
ac45abc0 | 4027 | /* fsync always requires a blocking context */ |
45d189c6 | 4028 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4029 | return -EAGAIN; |
4030 | ||
9adbd45d | 4031 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4032 | end > 0 ? end : LLONG_MAX, |
4033 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4034 | if (ret < 0) | |
4035 | req_set_fail_links(req); | |
e1e16097 | 4036 | io_req_complete(req, ret); |
c992fe29 CH |
4037 | return 0; |
4038 | } | |
4039 | ||
d63d1b5e JA |
4040 | static int io_fallocate_prep(struct io_kiocb *req, |
4041 | const struct io_uring_sqe *sqe) | |
4042 | { | |
4043 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
4044 | return -EINVAL; | |
3232dd02 PB |
4045 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4046 | return -EINVAL; | |
d63d1b5e JA |
4047 | |
4048 | req->sync.off = READ_ONCE(sqe->off); | |
4049 | req->sync.len = READ_ONCE(sqe->addr); | |
4050 | req->sync.mode = READ_ONCE(sqe->len); | |
4051 | return 0; | |
4052 | } | |
4053 | ||
45d189c6 | 4054 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4055 | { |
ac45abc0 PB |
4056 | int ret; |
4057 | ||
d63d1b5e | 4058 | /* fallocate always requiring blocking context */ |
45d189c6 | 4059 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4060 | return -EAGAIN; |
ac45abc0 PB |
4061 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4062 | req->sync.len); | |
ac45abc0 PB |
4063 | if (ret < 0) |
4064 | req_set_fail_links(req); | |
e1e16097 | 4065 | io_req_complete(req, ret); |
5d17b4a4 JA |
4066 | return 0; |
4067 | } | |
4068 | ||
ec65fea5 | 4069 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4070 | { |
f8748881 | 4071 | const char __user *fname; |
15b71abe | 4072 | int ret; |
b7bb4f7d | 4073 | |
ec65fea5 | 4074 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4075 | return -EINVAL; |
ec65fea5 | 4076 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4077 | return -EBADF; |
03b1230c | 4078 | |
ec65fea5 PB |
4079 | /* open.how should be already initialised */ |
4080 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4081 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4082 | |
25e72d10 PB |
4083 | req->open.dfd = READ_ONCE(sqe->fd); |
4084 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4085 | req->open.filename = getname(fname); |
15b71abe JA |
4086 | if (IS_ERR(req->open.filename)) { |
4087 | ret = PTR_ERR(req->open.filename); | |
4088 | req->open.filename = NULL; | |
4089 | return ret; | |
4090 | } | |
4022e7af | 4091 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4092 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4093 | return 0; |
03b1230c JA |
4094 | } |
4095 | ||
ec65fea5 PB |
4096 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4097 | { | |
4098 | u64 flags, mode; | |
4099 | ||
14587a46 | 4100 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4101 | return -EINVAL; |
ec65fea5 PB |
4102 | mode = READ_ONCE(sqe->len); |
4103 | flags = READ_ONCE(sqe->open_flags); | |
4104 | req->open.how = build_open_how(flags, mode); | |
4105 | return __io_openat_prep(req, sqe); | |
4106 | } | |
4107 | ||
cebdb986 | 4108 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4109 | { |
cebdb986 | 4110 | struct open_how __user *how; |
cebdb986 | 4111 | size_t len; |
0fa03c62 JA |
4112 | int ret; |
4113 | ||
14587a46 | 4114 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4115 | return -EINVAL; |
cebdb986 JA |
4116 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4117 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4118 | if (len < OPEN_HOW_SIZE_VER0) |
4119 | return -EINVAL; | |
3529d8c2 | 4120 | |
cebdb986 JA |
4121 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4122 | len); | |
4123 | if (ret) | |
4124 | return ret; | |
3529d8c2 | 4125 | |
ec65fea5 | 4126 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4127 | } |
4128 | ||
45d189c6 | 4129 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4130 | { |
4131 | struct open_flags op; | |
15b71abe | 4132 | struct file *file; |
3a81fd02 JA |
4133 | bool nonblock_set; |
4134 | bool resolve_nonblock; | |
15b71abe JA |
4135 | int ret; |
4136 | ||
cebdb986 | 4137 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4138 | if (ret) |
4139 | goto err; | |
3a81fd02 JA |
4140 | nonblock_set = op.open_flag & O_NONBLOCK; |
4141 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4142 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4143 | /* |
4144 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4145 | * it'll always -EAGAIN | |
4146 | */ | |
4147 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4148 | return -EAGAIN; | |
4149 | op.lookup_flags |= LOOKUP_CACHED; | |
4150 | op.open_flag |= O_NONBLOCK; | |
4151 | } | |
15b71abe | 4152 | |
4022e7af | 4153 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
4154 | if (ret < 0) |
4155 | goto err; | |
4156 | ||
4157 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 4158 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
4159 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
4160 | file == ERR_PTR(-EAGAIN)) { | |
3a81fd02 JA |
4161 | /* |
4162 | * We could hang on to this 'fd', but seems like marginal | |
4163 | * gain for something that is now known to be a slower path. | |
4164 | * So just put it, and we'll get a new one when we retry. | |
4165 | */ | |
4166 | put_unused_fd(ret); | |
4167 | return -EAGAIN; | |
4168 | } | |
4169 | ||
15b71abe JA |
4170 | if (IS_ERR(file)) { |
4171 | put_unused_fd(ret); | |
4172 | ret = PTR_ERR(file); | |
4173 | } else { | |
45d189c6 | 4174 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 4175 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
4176 | fsnotify_open(file); |
4177 | fd_install(ret, file); | |
4178 | } | |
4179 | err: | |
4180 | putname(req->open.filename); | |
8fef80bf | 4181 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
4182 | if (ret < 0) |
4183 | req_set_fail_links(req); | |
e1e16097 | 4184 | io_req_complete(req, ret); |
15b71abe JA |
4185 | return 0; |
4186 | } | |
4187 | ||
45d189c6 | 4188 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4189 | { |
45d189c6 | 4190 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
4191 | } |
4192 | ||
067524e9 JA |
4193 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4194 | const struct io_uring_sqe *sqe) | |
4195 | { | |
4196 | struct io_provide_buf *p = &req->pbuf; | |
4197 | u64 tmp; | |
4198 | ||
4199 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
4200 | return -EINVAL; | |
4201 | ||
4202 | tmp = READ_ONCE(sqe->fd); | |
4203 | if (!tmp || tmp > USHRT_MAX) | |
4204 | return -EINVAL; | |
4205 | ||
4206 | memset(p, 0, sizeof(*p)); | |
4207 | p->nbufs = tmp; | |
4208 | p->bgid = READ_ONCE(sqe->buf_group); | |
4209 | return 0; | |
4210 | } | |
4211 | ||
4212 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4213 | int bgid, unsigned nbufs) | |
4214 | { | |
4215 | unsigned i = 0; | |
4216 | ||
4217 | /* shouldn't happen */ | |
4218 | if (!nbufs) | |
4219 | return 0; | |
4220 | ||
4221 | /* the head kbuf is the list itself */ | |
4222 | while (!list_empty(&buf->list)) { | |
4223 | struct io_buffer *nxt; | |
4224 | ||
4225 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4226 | list_del(&nxt->list); | |
4227 | kfree(nxt); | |
4228 | if (++i == nbufs) | |
4229 | return i; | |
4230 | } | |
4231 | i++; | |
4232 | kfree(buf); | |
4233 | idr_remove(&ctx->io_buffer_idr, bgid); | |
4234 | ||
4235 | return i; | |
4236 | } | |
4237 | ||
889fca73 | 4238 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4239 | { |
4240 | struct io_provide_buf *p = &req->pbuf; | |
4241 | struct io_ring_ctx *ctx = req->ctx; | |
4242 | struct io_buffer *head; | |
4243 | int ret = 0; | |
45d189c6 | 4244 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4245 | |
4246 | io_ring_submit_lock(ctx, !force_nonblock); | |
4247 | ||
4248 | lockdep_assert_held(&ctx->uring_lock); | |
4249 | ||
4250 | ret = -ENOENT; | |
4251 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4252 | if (head) | |
4253 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
4254 | if (ret < 0) |
4255 | req_set_fail_links(req); | |
067524e9 | 4256 | |
31bff9a5 PB |
4257 | /* need to hold the lock to complete IOPOLL requests */ |
4258 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4259 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4260 | io_ring_submit_unlock(ctx, !force_nonblock); |
4261 | } else { | |
4262 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4263 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4264 | } |
067524e9 JA |
4265 | return 0; |
4266 | } | |
4267 | ||
ddf0322d JA |
4268 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4269 | const struct io_uring_sqe *sqe) | |
4270 | { | |
4271 | struct io_provide_buf *p = &req->pbuf; | |
4272 | u64 tmp; | |
4273 | ||
4274 | if (sqe->ioprio || sqe->rw_flags) | |
4275 | return -EINVAL; | |
4276 | ||
4277 | tmp = READ_ONCE(sqe->fd); | |
4278 | if (!tmp || tmp > USHRT_MAX) | |
4279 | return -E2BIG; | |
4280 | p->nbufs = tmp; | |
4281 | p->addr = READ_ONCE(sqe->addr); | |
4282 | p->len = READ_ONCE(sqe->len); | |
4283 | ||
efe68c1c | 4284 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4285 | return -EFAULT; |
4286 | ||
4287 | p->bgid = READ_ONCE(sqe->buf_group); | |
4288 | tmp = READ_ONCE(sqe->off); | |
4289 | if (tmp > USHRT_MAX) | |
4290 | return -E2BIG; | |
4291 | p->bid = tmp; | |
4292 | return 0; | |
4293 | } | |
4294 | ||
4295 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4296 | { | |
4297 | struct io_buffer *buf; | |
4298 | u64 addr = pbuf->addr; | |
4299 | int i, bid = pbuf->bid; | |
4300 | ||
4301 | for (i = 0; i < pbuf->nbufs; i++) { | |
4302 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4303 | if (!buf) | |
4304 | break; | |
4305 | ||
4306 | buf->addr = addr; | |
4307 | buf->len = pbuf->len; | |
4308 | buf->bid = bid; | |
4309 | addr += pbuf->len; | |
4310 | bid++; | |
4311 | if (!*head) { | |
4312 | INIT_LIST_HEAD(&buf->list); | |
4313 | *head = buf; | |
4314 | } else { | |
4315 | list_add_tail(&buf->list, &(*head)->list); | |
4316 | } | |
4317 | } | |
4318 | ||
4319 | return i ? i : -ENOMEM; | |
4320 | } | |
4321 | ||
889fca73 | 4322 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4323 | { |
4324 | struct io_provide_buf *p = &req->pbuf; | |
4325 | struct io_ring_ctx *ctx = req->ctx; | |
4326 | struct io_buffer *head, *list; | |
4327 | int ret = 0; | |
45d189c6 | 4328 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4329 | |
4330 | io_ring_submit_lock(ctx, !force_nonblock); | |
4331 | ||
4332 | lockdep_assert_held(&ctx->uring_lock); | |
4333 | ||
4334 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4335 | ||
4336 | ret = io_add_buffers(p, &head); | |
4337 | if (ret < 0) | |
4338 | goto out; | |
4339 | ||
4340 | if (!list) { | |
4341 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4342 | GFP_KERNEL); | |
4343 | if (ret < 0) { | |
067524e9 | 4344 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4345 | goto out; |
4346 | } | |
4347 | } | |
4348 | out: | |
ddf0322d JA |
4349 | if (ret < 0) |
4350 | req_set_fail_links(req); | |
31bff9a5 PB |
4351 | |
4352 | /* need to hold the lock to complete IOPOLL requests */ | |
4353 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4354 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4355 | io_ring_submit_unlock(ctx, !force_nonblock); |
4356 | } else { | |
4357 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4358 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4359 | } |
ddf0322d | 4360 | return 0; |
cebdb986 JA |
4361 | } |
4362 | ||
3e4827b0 JA |
4363 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4364 | const struct io_uring_sqe *sqe) | |
4365 | { | |
4366 | #if defined(CONFIG_EPOLL) | |
4367 | if (sqe->ioprio || sqe->buf_index) | |
4368 | return -EINVAL; | |
6ca56f84 | 4369 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4370 | return -EINVAL; |
3e4827b0 JA |
4371 | |
4372 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4373 | req->epoll.op = READ_ONCE(sqe->len); | |
4374 | req->epoll.fd = READ_ONCE(sqe->off); | |
4375 | ||
4376 | if (ep_op_has_event(req->epoll.op)) { | |
4377 | struct epoll_event __user *ev; | |
4378 | ||
4379 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4380 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4381 | return -EFAULT; | |
4382 | } | |
4383 | ||
4384 | return 0; | |
4385 | #else | |
4386 | return -EOPNOTSUPP; | |
4387 | #endif | |
4388 | } | |
4389 | ||
889fca73 | 4390 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4391 | { |
4392 | #if defined(CONFIG_EPOLL) | |
4393 | struct io_epoll *ie = &req->epoll; | |
4394 | int ret; | |
45d189c6 | 4395 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4396 | |
4397 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4398 | if (force_nonblock && ret == -EAGAIN) | |
4399 | return -EAGAIN; | |
4400 | ||
4401 | if (ret < 0) | |
4402 | req_set_fail_links(req); | |
889fca73 | 4403 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4404 | return 0; |
4405 | #else | |
4406 | return -EOPNOTSUPP; | |
4407 | #endif | |
4408 | } | |
4409 | ||
c1ca757b JA |
4410 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4411 | { | |
4412 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4413 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4414 | return -EINVAL; | |
3232dd02 PB |
4415 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4416 | return -EINVAL; | |
c1ca757b JA |
4417 | |
4418 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4419 | req->madvise.len = READ_ONCE(sqe->len); | |
4420 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4421 | return 0; | |
4422 | #else | |
4423 | return -EOPNOTSUPP; | |
4424 | #endif | |
4425 | } | |
4426 | ||
45d189c6 | 4427 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4428 | { |
4429 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4430 | struct io_madvise *ma = &req->madvise; | |
4431 | int ret; | |
4432 | ||
45d189c6 | 4433 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4434 | return -EAGAIN; |
4435 | ||
0726b01e | 4436 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4437 | if (ret < 0) |
4438 | req_set_fail_links(req); | |
e1e16097 | 4439 | io_req_complete(req, ret); |
c1ca757b JA |
4440 | return 0; |
4441 | #else | |
4442 | return -EOPNOTSUPP; | |
4443 | #endif | |
4444 | } | |
4445 | ||
4840e418 JA |
4446 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4447 | { | |
4448 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4449 | return -EINVAL; | |
3232dd02 PB |
4450 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4451 | return -EINVAL; | |
4840e418 JA |
4452 | |
4453 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4454 | req->fadvise.len = READ_ONCE(sqe->len); | |
4455 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4456 | return 0; | |
4457 | } | |
4458 | ||
45d189c6 | 4459 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4460 | { |
4461 | struct io_fadvise *fa = &req->fadvise; | |
4462 | int ret; | |
4463 | ||
45d189c6 | 4464 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4465 | switch (fa->advice) { |
4466 | case POSIX_FADV_NORMAL: | |
4467 | case POSIX_FADV_RANDOM: | |
4468 | case POSIX_FADV_SEQUENTIAL: | |
4469 | break; | |
4470 | default: | |
4471 | return -EAGAIN; | |
4472 | } | |
4473 | } | |
4840e418 JA |
4474 | |
4475 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4476 | if (ret < 0) | |
4477 | req_set_fail_links(req); | |
e1e16097 | 4478 | io_req_complete(req, ret); |
4840e418 JA |
4479 | return 0; |
4480 | } | |
4481 | ||
eddc7ef5 JA |
4482 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4483 | { | |
6ca56f84 | 4484 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4485 | return -EINVAL; |
eddc7ef5 JA |
4486 | if (sqe->ioprio || sqe->buf_index) |
4487 | return -EINVAL; | |
9c280f90 | 4488 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4489 | return -EBADF; |
eddc7ef5 | 4490 | |
1d9e1288 BM |
4491 | req->statx.dfd = READ_ONCE(sqe->fd); |
4492 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4493 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4494 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4495 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4496 | |
4497 | return 0; | |
4498 | } | |
4499 | ||
45d189c6 | 4500 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4501 | { |
1d9e1288 | 4502 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4503 | int ret; |
4504 | ||
45d189c6 | 4505 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4506 | /* only need file table for an actual valid fd */ |
4507 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4508 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4509 | return -EAGAIN; |
5b0bbee4 | 4510 | } |
eddc7ef5 | 4511 | |
e62753e4 BM |
4512 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4513 | ctx->buffer); | |
eddc7ef5 | 4514 | |
eddc7ef5 JA |
4515 | if (ret < 0) |
4516 | req_set_fail_links(req); | |
e1e16097 | 4517 | io_req_complete(req, ret); |
eddc7ef5 JA |
4518 | return 0; |
4519 | } | |
4520 | ||
b5dba59e JA |
4521 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4522 | { | |
14587a46 | 4523 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4524 | return -EINVAL; |
b5dba59e JA |
4525 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4526 | sqe->rw_flags || sqe->buf_index) | |
4527 | return -EINVAL; | |
9c280f90 | 4528 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4529 | return -EBADF; |
b5dba59e JA |
4530 | |
4531 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4532 | return 0; |
b5dba59e JA |
4533 | } |
4534 | ||
889fca73 | 4535 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4536 | { |
9eac1904 | 4537 | struct files_struct *files = current->files; |
3af73b28 | 4538 | struct io_close *close = &req->close; |
9eac1904 JA |
4539 | struct fdtable *fdt; |
4540 | struct file *file; | |
b5dba59e JA |
4541 | int ret; |
4542 | ||
9eac1904 JA |
4543 | file = NULL; |
4544 | ret = -EBADF; | |
4545 | spin_lock(&files->file_lock); | |
4546 | fdt = files_fdtable(files); | |
4547 | if (close->fd >= fdt->max_fds) { | |
4548 | spin_unlock(&files->file_lock); | |
4549 | goto err; | |
4550 | } | |
4551 | file = fdt->fd[close->fd]; | |
4552 | if (!file) { | |
4553 | spin_unlock(&files->file_lock); | |
4554 | goto err; | |
4555 | } | |
4556 | ||
4557 | if (file->f_op == &io_uring_fops) { | |
4558 | spin_unlock(&files->file_lock); | |
4559 | file = NULL; | |
4560 | goto err; | |
3af73b28 | 4561 | } |
b5dba59e JA |
4562 | |
4563 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4564 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4565 | spin_unlock(&files->file_lock); |
0bf0eefd | 4566 | return -EAGAIN; |
a2100672 | 4567 | } |
b5dba59e | 4568 | |
9eac1904 JA |
4569 | ret = __close_fd_get_file(close->fd, &file); |
4570 | spin_unlock(&files->file_lock); | |
4571 | if (ret < 0) { | |
4572 | if (ret == -ENOENT) | |
4573 | ret = -EBADF; | |
4574 | goto err; | |
4575 | } | |
4576 | ||
3af73b28 | 4577 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4578 | ret = filp_close(file, current->files); |
4579 | err: | |
3af73b28 PB |
4580 | if (ret < 0) |
4581 | req_set_fail_links(req); | |
9eac1904 JA |
4582 | if (file) |
4583 | fput(file); | |
889fca73 | 4584 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4585 | return 0; |
b5dba59e JA |
4586 | } |
4587 | ||
1155c76a | 4588 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4589 | { |
4590 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4591 | |
5d17b4a4 JA |
4592 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4593 | return -EINVAL; | |
4594 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4595 | return -EINVAL; | |
4596 | ||
8ed8d3c3 JA |
4597 | req->sync.off = READ_ONCE(sqe->off); |
4598 | req->sync.len = READ_ONCE(sqe->len); | |
4599 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4600 | return 0; |
4601 | } | |
4602 | ||
45d189c6 | 4603 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4604 | { |
8ed8d3c3 JA |
4605 | int ret; |
4606 | ||
ac45abc0 | 4607 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4608 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4609 | return -EAGAIN; |
4610 | ||
9adbd45d | 4611 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4612 | req->sync.flags); |
4613 | if (ret < 0) | |
4614 | req_set_fail_links(req); | |
e1e16097 | 4615 | io_req_complete(req, ret); |
5d17b4a4 JA |
4616 | return 0; |
4617 | } | |
4618 | ||
469956e8 | 4619 | #if defined(CONFIG_NET) |
02d27d89 PB |
4620 | static int io_setup_async_msg(struct io_kiocb *req, |
4621 | struct io_async_msghdr *kmsg) | |
4622 | { | |
e8c2bc1f JA |
4623 | struct io_async_msghdr *async_msg = req->async_data; |
4624 | ||
4625 | if (async_msg) | |
02d27d89 | 4626 | return -EAGAIN; |
e8c2bc1f | 4627 | if (io_alloc_async_data(req)) { |
257e84a5 | 4628 | kfree(kmsg->free_iov); |
02d27d89 PB |
4629 | return -ENOMEM; |
4630 | } | |
e8c2bc1f | 4631 | async_msg = req->async_data; |
02d27d89 | 4632 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4633 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4634 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4635 | /* if were using fast_iov, set it to the new one */ |
4636 | if (!async_msg->free_iov) | |
4637 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4638 | ||
02d27d89 PB |
4639 | return -EAGAIN; |
4640 | } | |
4641 | ||
2ae523ed PB |
4642 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4643 | struct io_async_msghdr *iomsg) | |
4644 | { | |
2ae523ed | 4645 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4646 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4647 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4648 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4649 | } |
4650 | ||
93642ef8 PB |
4651 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4652 | { | |
4653 | int ret; | |
4654 | ||
4655 | if (!io_op_defs[req->opcode].needs_async_data) | |
4656 | return 0; | |
4657 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4658 | if (!ret) | |
4659 | req->flags |= REQ_F_NEED_CLEANUP; | |
4660 | return ret; | |
4661 | } | |
4662 | ||
3529d8c2 | 4663 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4664 | { |
e47293fd | 4665 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4666 | |
d2b6f48b PB |
4667 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4668 | return -EINVAL; | |
4669 | ||
e47293fd | 4670 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4671 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4672 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4673 | |
d8768362 JA |
4674 | #ifdef CONFIG_COMPAT |
4675 | if (req->ctx->compat) | |
4676 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4677 | #endif | |
93642ef8 | 4678 | return 0; |
03b1230c JA |
4679 | } |
4680 | ||
889fca73 | 4681 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4682 | { |
6b754c8b | 4683 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4684 | struct socket *sock; |
7a7cacba | 4685 | unsigned flags; |
0fa03c62 JA |
4686 | int ret; |
4687 | ||
dba4a925 | 4688 | sock = sock_from_file(req->file); |
7a7cacba | 4689 | if (unlikely(!sock)) |
dba4a925 | 4690 | return -ENOTSOCK; |
3529d8c2 | 4691 | |
257e84a5 PB |
4692 | kmsg = req->async_data; |
4693 | if (!kmsg) { | |
7a7cacba PB |
4694 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4695 | if (ret) | |
4696 | return ret; | |
4697 | kmsg = &iomsg; | |
0fa03c62 | 4698 | } |
0fa03c62 | 4699 | |
7a7cacba PB |
4700 | flags = req->sr_msg.msg_flags; |
4701 | if (flags & MSG_DONTWAIT) | |
4702 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4703 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4704 | flags |= MSG_DONTWAIT; |
e47293fd | 4705 | |
7a7cacba | 4706 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4707 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4708 | return io_setup_async_msg(req, kmsg); |
4709 | if (ret == -ERESTARTSYS) | |
4710 | ret = -EINTR; | |
0fa03c62 | 4711 | |
257e84a5 PB |
4712 | /* fast path, check for non-NULL to avoid function call */ |
4713 | if (kmsg->free_iov) | |
4714 | kfree(kmsg->free_iov); | |
99bc4c38 | 4715 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4716 | if (ret < 0) |
4717 | req_set_fail_links(req); | |
889fca73 | 4718 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4719 | return 0; |
03b1230c | 4720 | } |
aa1fa28f | 4721 | |
889fca73 | 4722 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4723 | { |
7a7cacba PB |
4724 | struct io_sr_msg *sr = &req->sr_msg; |
4725 | struct msghdr msg; | |
4726 | struct iovec iov; | |
fddaface | 4727 | struct socket *sock; |
7a7cacba | 4728 | unsigned flags; |
fddaface JA |
4729 | int ret; |
4730 | ||
dba4a925 | 4731 | sock = sock_from_file(req->file); |
7a7cacba | 4732 | if (unlikely(!sock)) |
dba4a925 | 4733 | return -ENOTSOCK; |
fddaface | 4734 | |
7a7cacba PB |
4735 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4736 | if (unlikely(ret)) | |
14db8411 | 4737 | return ret; |
fddaface | 4738 | |
7a7cacba PB |
4739 | msg.msg_name = NULL; |
4740 | msg.msg_control = NULL; | |
4741 | msg.msg_controllen = 0; | |
4742 | msg.msg_namelen = 0; | |
fddaface | 4743 | |
7a7cacba PB |
4744 | flags = req->sr_msg.msg_flags; |
4745 | if (flags & MSG_DONTWAIT) | |
4746 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4747 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4748 | flags |= MSG_DONTWAIT; |
fddaface | 4749 | |
7a7cacba PB |
4750 | msg.msg_flags = flags; |
4751 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4752 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4753 | return -EAGAIN; |
4754 | if (ret == -ERESTARTSYS) | |
4755 | ret = -EINTR; | |
fddaface | 4756 | |
fddaface JA |
4757 | if (ret < 0) |
4758 | req_set_fail_links(req); | |
889fca73 | 4759 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4760 | return 0; |
fddaface JA |
4761 | } |
4762 | ||
1400e697 PB |
4763 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4764 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4765 | { |
4766 | struct io_sr_msg *sr = &req->sr_msg; | |
4767 | struct iovec __user *uiov; | |
4768 | size_t iov_len; | |
4769 | int ret; | |
4770 | ||
1400e697 PB |
4771 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4772 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4773 | if (ret) |
4774 | return ret; | |
4775 | ||
4776 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4777 | if (iov_len > 1) | |
4778 | return -EINVAL; | |
5476dfed | 4779 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4780 | return -EFAULT; |
5476dfed | 4781 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4782 | iomsg->free_iov = NULL; |
52de1fe1 | 4783 | } else { |
257e84a5 | 4784 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4785 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4786 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4787 | false); |
52de1fe1 JA |
4788 | if (ret > 0) |
4789 | ret = 0; | |
4790 | } | |
4791 | ||
4792 | return ret; | |
4793 | } | |
4794 | ||
4795 | #ifdef CONFIG_COMPAT | |
4796 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4797 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4798 | { |
4799 | struct compat_msghdr __user *msg_compat; | |
4800 | struct io_sr_msg *sr = &req->sr_msg; | |
4801 | struct compat_iovec __user *uiov; | |
4802 | compat_uptr_t ptr; | |
4803 | compat_size_t len; | |
4804 | int ret; | |
4805 | ||
270a5940 | 4806 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4807 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4808 | &ptr, &len); |
4809 | if (ret) | |
4810 | return ret; | |
4811 | ||
4812 | uiov = compat_ptr(ptr); | |
4813 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4814 | compat_ssize_t clen; | |
4815 | ||
4816 | if (len > 1) | |
4817 | return -EINVAL; | |
4818 | if (!access_ok(uiov, sizeof(*uiov))) | |
4819 | return -EFAULT; | |
4820 | if (__get_user(clen, &uiov->iov_len)) | |
4821 | return -EFAULT; | |
4822 | if (clen < 0) | |
4823 | return -EINVAL; | |
2d280bc8 | 4824 | sr->len = clen; |
257e84a5 | 4825 | iomsg->free_iov = NULL; |
52de1fe1 | 4826 | } else { |
257e84a5 | 4827 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4828 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4829 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4830 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4831 | if (ret < 0) |
4832 | return ret; | |
4833 | } | |
4834 | ||
4835 | return 0; | |
4836 | } | |
4837 | #endif | |
4838 | ||
1400e697 PB |
4839 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4840 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4841 | { |
1400e697 | 4842 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4843 | |
4844 | #ifdef CONFIG_COMPAT | |
4845 | if (req->ctx->compat) | |
1400e697 | 4846 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4847 | #endif |
52de1fe1 | 4848 | |
1400e697 | 4849 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4850 | } |
4851 | ||
bcda7baa | 4852 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4853 | bool needs_lock) |
bcda7baa JA |
4854 | { |
4855 | struct io_sr_msg *sr = &req->sr_msg; | |
4856 | struct io_buffer *kbuf; | |
4857 | ||
bcda7baa JA |
4858 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4859 | if (IS_ERR(kbuf)) | |
4860 | return kbuf; | |
4861 | ||
4862 | sr->kbuf = kbuf; | |
4863 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4864 | return kbuf; |
fddaface JA |
4865 | } |
4866 | ||
7fbb1b54 PB |
4867 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4868 | { | |
4869 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4870 | } | |
4871 | ||
93642ef8 | 4872 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4873 | { |
99bc4c38 | 4874 | int ret; |
3529d8c2 | 4875 | |
93642ef8 PB |
4876 | if (!io_op_defs[req->opcode].needs_async_data) |
4877 | return 0; | |
4878 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4879 | if (!ret) | |
4880 | req->flags |= REQ_F_NEED_CLEANUP; | |
4881 | return ret; | |
4882 | } | |
4883 | ||
4884 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4885 | { | |
4886 | struct io_sr_msg *sr = &req->sr_msg; | |
4887 | ||
d2b6f48b PB |
4888 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4889 | return -EINVAL; | |
4890 | ||
3529d8c2 | 4891 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4892 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4893 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4894 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4895 | |
d8768362 JA |
4896 | #ifdef CONFIG_COMPAT |
4897 | if (req->ctx->compat) | |
4898 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4899 | #endif | |
93642ef8 | 4900 | return 0; |
aa1fa28f JA |
4901 | } |
4902 | ||
889fca73 | 4903 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4904 | { |
6b754c8b | 4905 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4906 | struct socket *sock; |
7fbb1b54 | 4907 | struct io_buffer *kbuf; |
7a7cacba | 4908 | unsigned flags; |
52de1fe1 | 4909 | int ret, cflags = 0; |
45d189c6 | 4910 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4911 | |
dba4a925 | 4912 | sock = sock_from_file(req->file); |
7a7cacba | 4913 | if (unlikely(!sock)) |
dba4a925 | 4914 | return -ENOTSOCK; |
3529d8c2 | 4915 | |
257e84a5 PB |
4916 | kmsg = req->async_data; |
4917 | if (!kmsg) { | |
7a7cacba PB |
4918 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4919 | if (ret) | |
681fda8d | 4920 | return ret; |
7a7cacba PB |
4921 | kmsg = &iomsg; |
4922 | } | |
03b1230c | 4923 | |
bc02ef33 | 4924 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4925 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4926 | if (IS_ERR(kbuf)) |
52de1fe1 | 4927 | return PTR_ERR(kbuf); |
7a7cacba | 4928 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4929 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4930 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4931 | 1, req->sr_msg.len); |
4932 | } | |
52de1fe1 | 4933 | |
7a7cacba PB |
4934 | flags = req->sr_msg.msg_flags; |
4935 | if (flags & MSG_DONTWAIT) | |
4936 | req->flags |= REQ_F_NOWAIT; | |
4937 | else if (force_nonblock) | |
4938 | flags |= MSG_DONTWAIT; | |
e47293fd | 4939 | |
7a7cacba PB |
4940 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4941 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4942 | if (force_nonblock && ret == -EAGAIN) |
4943 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4944 | if (ret == -ERESTARTSYS) |
4945 | ret = -EINTR; | |
03b1230c | 4946 | |
7fbb1b54 PB |
4947 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4948 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4949 | /* fast path, check for non-NULL to avoid function call */ |
4950 | if (kmsg->free_iov) | |
4951 | kfree(kmsg->free_iov); | |
99bc4c38 | 4952 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4953 | if (ret < 0) |
4954 | req_set_fail_links(req); | |
889fca73 | 4955 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4956 | return 0; |
0fa03c62 | 4957 | } |
5d17b4a4 | 4958 | |
889fca73 | 4959 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4960 | { |
6b754c8b | 4961 | struct io_buffer *kbuf; |
7a7cacba PB |
4962 | struct io_sr_msg *sr = &req->sr_msg; |
4963 | struct msghdr msg; | |
4964 | void __user *buf = sr->buf; | |
fddaface | 4965 | struct socket *sock; |
7a7cacba PB |
4966 | struct iovec iov; |
4967 | unsigned flags; | |
bcda7baa | 4968 | int ret, cflags = 0; |
45d189c6 | 4969 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4970 | |
dba4a925 | 4971 | sock = sock_from_file(req->file); |
7a7cacba | 4972 | if (unlikely(!sock)) |
dba4a925 | 4973 | return -ENOTSOCK; |
fddaface | 4974 | |
bc02ef33 | 4975 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4976 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4977 | if (IS_ERR(kbuf)) |
4978 | return PTR_ERR(kbuf); | |
7a7cacba | 4979 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4980 | } |
bcda7baa | 4981 | |
7a7cacba | 4982 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4983 | if (unlikely(ret)) |
4984 | goto out_free; | |
fddaface | 4985 | |
7a7cacba PB |
4986 | msg.msg_name = NULL; |
4987 | msg.msg_control = NULL; | |
4988 | msg.msg_controllen = 0; | |
4989 | msg.msg_namelen = 0; | |
4990 | msg.msg_iocb = NULL; | |
4991 | msg.msg_flags = 0; | |
fddaface | 4992 | |
7a7cacba PB |
4993 | flags = req->sr_msg.msg_flags; |
4994 | if (flags & MSG_DONTWAIT) | |
4995 | req->flags |= REQ_F_NOWAIT; | |
4996 | else if (force_nonblock) | |
4997 | flags |= MSG_DONTWAIT; | |
4998 | ||
4999 | ret = sock_recvmsg(sock, &msg, flags); | |
5000 | if (force_nonblock && ret == -EAGAIN) | |
5001 | return -EAGAIN; | |
5002 | if (ret == -ERESTARTSYS) | |
5003 | ret = -EINTR; | |
14c32eee | 5004 | out_free: |
7fbb1b54 PB |
5005 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5006 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
5007 | if (ret < 0) |
5008 | req_set_fail_links(req); | |
889fca73 | 5009 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5010 | return 0; |
fddaface JA |
5011 | } |
5012 | ||
3529d8c2 | 5013 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5014 | { |
8ed8d3c3 JA |
5015 | struct io_accept *accept = &req->accept; |
5016 | ||
14587a46 | 5017 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5018 | return -EINVAL; |
8042d6ce | 5019 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5020 | return -EINVAL; |
5021 | ||
d55e5f5b JA |
5022 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5023 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5024 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5025 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 5026 | return 0; |
8ed8d3c3 | 5027 | } |
17f2fe35 | 5028 | |
889fca73 | 5029 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5030 | { |
5031 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5032 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5033 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
5034 | int ret; |
5035 | ||
e697deed JX |
5036 | if (req->file->f_flags & O_NONBLOCK) |
5037 | req->flags |= REQ_F_NOWAIT; | |
5038 | ||
8ed8d3c3 | 5039 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
5040 | accept->addr_len, accept->flags, |
5041 | accept->nofile); | |
8ed8d3c3 | 5042 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 5043 | return -EAGAIN; |
ac45abc0 PB |
5044 | if (ret < 0) { |
5045 | if (ret == -ERESTARTSYS) | |
5046 | ret = -EINTR; | |
4e88d6e7 | 5047 | req_set_fail_links(req); |
ac45abc0 | 5048 | } |
889fca73 | 5049 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5050 | return 0; |
8ed8d3c3 JA |
5051 | } |
5052 | ||
93642ef8 PB |
5053 | static int io_connect_prep_async(struct io_kiocb *req) |
5054 | { | |
5055 | struct io_async_connect *io = req->async_data; | |
5056 | struct io_connect *conn = &req->connect; | |
5057 | ||
5058 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5059 | } | |
5060 | ||
3529d8c2 | 5061 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5062 | { |
3529d8c2 | 5063 | struct io_connect *conn = &req->connect; |
f499a021 | 5064 | |
14587a46 | 5065 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
5066 | return -EINVAL; |
5067 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
5068 | return -EINVAL; | |
5069 | ||
3529d8c2 JA |
5070 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5071 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5072 | return 0; |
f499a021 JA |
5073 | } |
5074 | ||
889fca73 | 5075 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5076 | { |
e8c2bc1f | 5077 | struct io_async_connect __io, *io; |
f8e85cf2 | 5078 | unsigned file_flags; |
3fbb51c1 | 5079 | int ret; |
45d189c6 | 5080 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5081 | |
e8c2bc1f JA |
5082 | if (req->async_data) { |
5083 | io = req->async_data; | |
f499a021 | 5084 | } else { |
3529d8c2 JA |
5085 | ret = move_addr_to_kernel(req->connect.addr, |
5086 | req->connect.addr_len, | |
e8c2bc1f | 5087 | &__io.address); |
f499a021 JA |
5088 | if (ret) |
5089 | goto out; | |
5090 | io = &__io; | |
5091 | } | |
5092 | ||
3fbb51c1 JA |
5093 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5094 | ||
e8c2bc1f | 5095 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5096 | req->connect.addr_len, file_flags); |
87f80d62 | 5097 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5098 | if (req->async_data) |
b7bb4f7d | 5099 | return -EAGAIN; |
e8c2bc1f | 5100 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5101 | ret = -ENOMEM; |
5102 | goto out; | |
5103 | } | |
e8c2bc1f JA |
5104 | io = req->async_data; |
5105 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 5106 | return -EAGAIN; |
f499a021 | 5107 | } |
f8e85cf2 JA |
5108 | if (ret == -ERESTARTSYS) |
5109 | ret = -EINTR; | |
f499a021 | 5110 | out: |
4e88d6e7 JA |
5111 | if (ret < 0) |
5112 | req_set_fail_links(req); | |
889fca73 | 5113 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5114 | return 0; |
469956e8 Y |
5115 | } |
5116 | #else /* !CONFIG_NET */ | |
5117 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5118 | { | |
f8e85cf2 | 5119 | return -EOPNOTSUPP; |
f8e85cf2 JA |
5120 | } |
5121 | ||
889fca73 | 5122 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5123 | { |
5124 | return -EOPNOTSUPP; | |
5125 | } | |
5126 | ||
889fca73 | 5127 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5128 | { |
5129 | return -EOPNOTSUPP; | |
5130 | } | |
5131 | ||
5132 | static int io_recvmsg_prep(struct io_kiocb *req, | |
5133 | const struct io_uring_sqe *sqe) | |
5134 | { | |
5135 | return -EOPNOTSUPP; | |
5136 | } | |
5137 | ||
889fca73 | 5138 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5139 | { |
5140 | return -EOPNOTSUPP; | |
5141 | } | |
5142 | ||
889fca73 | 5143 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5144 | { |
5145 | return -EOPNOTSUPP; | |
5146 | } | |
5147 | ||
5148 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5149 | { | |
5150 | return -EOPNOTSUPP; | |
5151 | } | |
5152 | ||
889fca73 | 5153 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5154 | { |
5155 | return -EOPNOTSUPP; | |
5156 | } | |
ce593a6c | 5157 | |
469956e8 Y |
5158 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5159 | { | |
5160 | return -EOPNOTSUPP; | |
5161 | } | |
5162 | ||
889fca73 | 5163 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 | 5164 | { |
f8e85cf2 | 5165 | return -EOPNOTSUPP; |
ce593a6c | 5166 | } |
469956e8 | 5167 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5168 | |
d7718a9d JA |
5169 | struct io_poll_table { |
5170 | struct poll_table_struct pt; | |
5171 | struct io_kiocb *req; | |
5172 | int error; | |
5173 | }; | |
ce593a6c | 5174 | |
d7718a9d JA |
5175 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5176 | __poll_t mask, task_work_func_t func) | |
5177 | { | |
aa96bf8a | 5178 | int ret; |
d7718a9d JA |
5179 | |
5180 | /* for instances that support it check for an event match first: */ | |
5181 | if (mask && !(mask & poll->events)) | |
5182 | return 0; | |
5183 | ||
5184 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5185 | ||
5186 | list_del_init(&poll->wait.entry); | |
5187 | ||
d7718a9d | 5188 | req->result = mask; |
7cbf1722 | 5189 | req->task_work.func = func; |
6d816e08 JA |
5190 | percpu_ref_get(&req->ctx->refs); |
5191 | ||
d7718a9d | 5192 | /* |
e3aabf95 JA |
5193 | * If this fails, then the task is exiting. When a task exits, the |
5194 | * work gets canceled, so just cancel this request as well instead | |
5195 | * of executing it. We can't safely execute it anyway, as we may not | |
5196 | * have the needed state needed for it anyway. | |
d7718a9d | 5197 | */ |
355fb9e2 | 5198 | ret = io_req_task_work_add(req); |
aa96bf8a | 5199 | if (unlikely(ret)) { |
e3aabf95 | 5200 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 5201 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 5202 | } |
d7718a9d JA |
5203 | return 1; |
5204 | } | |
5205 | ||
74ce6ce4 JA |
5206 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5207 | __acquires(&req->ctx->completion_lock) | |
5208 | { | |
5209 | struct io_ring_ctx *ctx = req->ctx; | |
5210 | ||
5211 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5212 | struct poll_table_struct pt = { ._key = poll->events }; | |
5213 | ||
5214 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5215 | } | |
5216 | ||
5217 | spin_lock_irq(&ctx->completion_lock); | |
5218 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5219 | add_wait_queue(poll->head, &poll->wait); | |
5220 | return true; | |
5221 | } | |
5222 | ||
5223 | return false; | |
5224 | } | |
5225 | ||
d4e7cd36 | 5226 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5227 | { |
e8c2bc1f | 5228 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5229 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5230 | return req->async_data; |
d4e7cd36 JA |
5231 | return req->apoll->double_poll; |
5232 | } | |
5233 | ||
5234 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5235 | { | |
5236 | if (req->opcode == IORING_OP_POLL_ADD) | |
5237 | return &req->poll; | |
5238 | return &req->apoll->poll; | |
5239 | } | |
5240 | ||
5241 | static void io_poll_remove_double(struct io_kiocb *req) | |
5242 | { | |
5243 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5244 | |
5245 | lockdep_assert_held(&req->ctx->completion_lock); | |
5246 | ||
5247 | if (poll && poll->head) { | |
5248 | struct wait_queue_head *head = poll->head; | |
5249 | ||
5250 | spin_lock(&head->lock); | |
5251 | list_del_init(&poll->wait.entry); | |
5252 | if (poll->wait.private) | |
5253 | refcount_dec(&req->refs); | |
5254 | poll->head = NULL; | |
5255 | spin_unlock(&head->lock); | |
5256 | } | |
5257 | } | |
5258 | ||
5259 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5260 | { | |
5261 | struct io_ring_ctx *ctx = req->ctx; | |
5262 | ||
d4e7cd36 | 5263 | io_poll_remove_double(req); |
18bceab1 JA |
5264 | req->poll.done = true; |
5265 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5266 | io_commit_cqring(ctx); | |
5267 | } | |
5268 | ||
dd221f46 | 5269 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5270 | { |
dd221f46 | 5271 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5272 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5273 | struct io_kiocb *nxt; |
18bceab1 JA |
5274 | |
5275 | if (io_poll_rewait(req, &req->poll)) { | |
5276 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5277 | } else { |
5278 | hash_del(&req->hash_node); | |
5279 | io_poll_complete(req, req->result, 0); | |
5280 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5281 | |
dd221f46 PB |
5282 | nxt = io_put_req_find_next(req); |
5283 | io_cqring_ev_posted(ctx); | |
5284 | if (nxt) | |
5285 | __io_req_task_submit(nxt); | |
5286 | } | |
18bceab1 | 5287 | |
6d816e08 | 5288 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5289 | } |
5290 | ||
5291 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5292 | int sync, void *key) | |
5293 | { | |
5294 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5295 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5296 | __poll_t mask = key_to_poll(key); |
5297 | ||
5298 | /* for instances that support it check for an event match first: */ | |
5299 | if (mask && !(mask & poll->events)) | |
5300 | return 0; | |
5301 | ||
8706e04e JA |
5302 | list_del_init(&wait->entry); |
5303 | ||
807abcb0 | 5304 | if (poll && poll->head) { |
18bceab1 JA |
5305 | bool done; |
5306 | ||
807abcb0 JA |
5307 | spin_lock(&poll->head->lock); |
5308 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5309 | if (!done) |
807abcb0 | 5310 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5311 | /* make sure double remove sees this as being gone */ |
5312 | wait->private = NULL; | |
807abcb0 | 5313 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5314 | if (!done) { |
5315 | /* use wait func handler, so it matches the rq type */ | |
5316 | poll->wait.func(&poll->wait, mode, sync, key); | |
5317 | } | |
18bceab1 JA |
5318 | } |
5319 | refcount_dec(&req->refs); | |
5320 | return 1; | |
5321 | } | |
5322 | ||
5323 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5324 | wait_queue_func_t wake_func) | |
5325 | { | |
5326 | poll->head = NULL; | |
5327 | poll->done = false; | |
5328 | poll->canceled = false; | |
5329 | poll->events = events; | |
5330 | INIT_LIST_HEAD(&poll->wait.entry); | |
5331 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5332 | } | |
5333 | ||
5334 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5335 | struct wait_queue_head *head, |
5336 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5337 | { |
5338 | struct io_kiocb *req = pt->req; | |
5339 | ||
5340 | /* | |
5341 | * If poll->head is already set, it's because the file being polled | |
5342 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5343 | * for write). Setup a separate io_poll_iocb if this happens. | |
5344 | */ | |
5345 | if (unlikely(poll->head)) { | |
58852d4d PB |
5346 | struct io_poll_iocb *poll_one = poll; |
5347 | ||
18bceab1 | 5348 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5349 | if (*poll_ptr) { |
18bceab1 JA |
5350 | pt->error = -EINVAL; |
5351 | return; | |
5352 | } | |
5353 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5354 | if (!poll) { | |
5355 | pt->error = -ENOMEM; | |
5356 | return; | |
5357 | } | |
58852d4d | 5358 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5359 | refcount_inc(&req->refs); |
5360 | poll->wait.private = req; | |
807abcb0 | 5361 | *poll_ptr = poll; |
18bceab1 JA |
5362 | } |
5363 | ||
5364 | pt->error = 0; | |
5365 | poll->head = head; | |
a31eb4a2 JX |
5366 | |
5367 | if (poll->events & EPOLLEXCLUSIVE) | |
5368 | add_wait_queue_exclusive(head, &poll->wait); | |
5369 | else | |
5370 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5371 | } |
5372 | ||
5373 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5374 | struct poll_table_struct *p) | |
5375 | { | |
5376 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5377 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5378 | |
807abcb0 | 5379 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5380 | } |
5381 | ||
d7718a9d JA |
5382 | static void io_async_task_func(struct callback_head *cb) |
5383 | { | |
5384 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5385 | struct async_poll *apoll = req->apoll; | |
5386 | struct io_ring_ctx *ctx = req->ctx; | |
5387 | ||
5388 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5389 | ||
74ce6ce4 | 5390 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5391 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5392 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5393 | return; |
d7718a9d JA |
5394 | } |
5395 | ||
31067255 | 5396 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5397 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5398 | hash_del(&req->hash_node); |
2bae047e | 5399 | |
d4e7cd36 | 5400 | io_poll_remove_double(req); |
74ce6ce4 JA |
5401 | spin_unlock_irq(&ctx->completion_lock); |
5402 | ||
0be0b0e3 PB |
5403 | if (!READ_ONCE(apoll->poll.canceled)) |
5404 | __io_req_task_submit(req); | |
5405 | else | |
5406 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5407 | |
6d816e08 | 5408 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5409 | kfree(apoll->double_poll); |
31067255 | 5410 | kfree(apoll); |
d7718a9d JA |
5411 | } |
5412 | ||
5413 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5414 | void *key) | |
5415 | { | |
5416 | struct io_kiocb *req = wait->private; | |
5417 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5418 | ||
5419 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5420 | key_to_poll(key)); | |
5421 | ||
5422 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5423 | } | |
5424 | ||
5425 | static void io_poll_req_insert(struct io_kiocb *req) | |
5426 | { | |
5427 | struct io_ring_ctx *ctx = req->ctx; | |
5428 | struct hlist_head *list; | |
5429 | ||
5430 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5431 | hlist_add_head(&req->hash_node, list); | |
5432 | } | |
5433 | ||
5434 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5435 | struct io_poll_iocb *poll, | |
5436 | struct io_poll_table *ipt, __poll_t mask, | |
5437 | wait_queue_func_t wake_func) | |
5438 | __acquires(&ctx->completion_lock) | |
5439 | { | |
5440 | struct io_ring_ctx *ctx = req->ctx; | |
5441 | bool cancel = false; | |
5442 | ||
4d52f338 | 5443 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5444 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5445 | poll->file = req->file; |
18bceab1 | 5446 | poll->wait.private = req; |
d7718a9d JA |
5447 | |
5448 | ipt->pt._key = mask; | |
5449 | ipt->req = req; | |
5450 | ipt->error = -EINVAL; | |
5451 | ||
d7718a9d JA |
5452 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5453 | ||
5454 | spin_lock_irq(&ctx->completion_lock); | |
5455 | if (likely(poll->head)) { | |
5456 | spin_lock(&poll->head->lock); | |
5457 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5458 | if (ipt->error) | |
5459 | cancel = true; | |
5460 | ipt->error = 0; | |
5461 | mask = 0; | |
5462 | } | |
5463 | if (mask || ipt->error) | |
5464 | list_del_init(&poll->wait.entry); | |
5465 | else if (cancel) | |
5466 | WRITE_ONCE(poll->canceled, true); | |
5467 | else if (!poll->done) /* actually waiting for an event */ | |
5468 | io_poll_req_insert(req); | |
5469 | spin_unlock(&poll->head->lock); | |
5470 | } | |
5471 | ||
5472 | return mask; | |
5473 | } | |
5474 | ||
5475 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5476 | { | |
5477 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5478 | struct io_ring_ctx *ctx = req->ctx; | |
5479 | struct async_poll *apoll; | |
5480 | struct io_poll_table ipt; | |
5481 | __poll_t mask, ret; | |
9dab14b8 | 5482 | int rw; |
d7718a9d JA |
5483 | |
5484 | if (!req->file || !file_can_poll(req->file)) | |
5485 | return false; | |
24c74678 | 5486 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5487 | return false; |
9dab14b8 JA |
5488 | if (def->pollin) |
5489 | rw = READ; | |
5490 | else if (def->pollout) | |
5491 | rw = WRITE; | |
5492 | else | |
5493 | return false; | |
5494 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5495 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5496 | return false; |
5497 | ||
5498 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5499 | if (unlikely(!apoll)) | |
5500 | return false; | |
807abcb0 | 5501 | apoll->double_poll = NULL; |
d7718a9d JA |
5502 | |
5503 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5504 | req->apoll = apoll; |
d7718a9d | 5505 | |
8755d97a | 5506 | mask = 0; |
d7718a9d | 5507 | if (def->pollin) |
8755d97a | 5508 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5509 | if (def->pollout) |
5510 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5511 | |
5512 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5513 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5514 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5515 | mask &= ~POLLIN; | |
5516 | ||
d7718a9d JA |
5517 | mask |= POLLERR | POLLPRI; |
5518 | ||
5519 | ipt.pt._qproc = io_async_queue_proc; | |
5520 | ||
5521 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5522 | io_async_wake); | |
a36da65c | 5523 | if (ret || ipt.error) { |
d4e7cd36 | 5524 | io_poll_remove_double(req); |
d7718a9d | 5525 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5526 | kfree(apoll->double_poll); |
d7718a9d JA |
5527 | kfree(apoll); |
5528 | return false; | |
5529 | } | |
5530 | spin_unlock_irq(&ctx->completion_lock); | |
5531 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5532 | apoll->poll.events); | |
5533 | return true; | |
5534 | } | |
5535 | ||
5536 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5537 | struct io_poll_iocb *poll) | |
221c5eb2 | 5538 | { |
b41e9852 | 5539 | bool do_complete = false; |
221c5eb2 JA |
5540 | |
5541 | spin_lock(&poll->head->lock); | |
5542 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5543 | if (!list_empty(&poll->wait.entry)) { |
5544 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5545 | do_complete = true; |
221c5eb2 JA |
5546 | } |
5547 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5548 | hash_del(&req->hash_node); |
d7718a9d JA |
5549 | return do_complete; |
5550 | } | |
5551 | ||
5552 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5553 | { | |
5554 | bool do_complete; | |
5555 | ||
d4e7cd36 JA |
5556 | io_poll_remove_double(req); |
5557 | ||
d7718a9d JA |
5558 | if (req->opcode == IORING_OP_POLL_ADD) { |
5559 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5560 | } else { | |
3bfa5bcb JA |
5561 | struct async_poll *apoll = req->apoll; |
5562 | ||
d7718a9d | 5563 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5564 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5565 | if (do_complete) { | |
d7718a9d | 5566 | io_put_req(req); |
807abcb0 | 5567 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5568 | kfree(apoll); |
5569 | } | |
b1f573bd XW |
5570 | } |
5571 | ||
b41e9852 JA |
5572 | if (do_complete) { |
5573 | io_cqring_fill_event(req, -ECANCELED); | |
5574 | io_commit_cqring(req->ctx); | |
f254ac04 | 5575 | req_set_fail_links(req); |
216578e5 | 5576 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5577 | } |
5578 | ||
5579 | return do_complete; | |
221c5eb2 JA |
5580 | } |
5581 | ||
76e1b642 JA |
5582 | /* |
5583 | * Returns true if we found and killed one or more poll requests | |
5584 | */ | |
6b81928d PB |
5585 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5586 | struct files_struct *files) | |
221c5eb2 | 5587 | { |
78076bb6 | 5588 | struct hlist_node *tmp; |
221c5eb2 | 5589 | struct io_kiocb *req; |
8e2e1faf | 5590 | int posted = 0, i; |
221c5eb2 JA |
5591 | |
5592 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5593 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5594 | struct hlist_head *list; | |
5595 | ||
5596 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5597 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5598 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5599 | posted += io_poll_remove_one(req); |
5600 | } | |
221c5eb2 JA |
5601 | } |
5602 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5603 | |
8e2e1faf JA |
5604 | if (posted) |
5605 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5606 | |
5607 | return posted != 0; | |
221c5eb2 JA |
5608 | } |
5609 | ||
47f46768 JA |
5610 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5611 | { | |
78076bb6 | 5612 | struct hlist_head *list; |
47f46768 JA |
5613 | struct io_kiocb *req; |
5614 | ||
78076bb6 JA |
5615 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5616 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5617 | if (sqe_addr != req->user_data) |
5618 | continue; | |
5619 | if (io_poll_remove_one(req)) | |
eac406c6 | 5620 | return 0; |
b41e9852 | 5621 | return -EALREADY; |
47f46768 JA |
5622 | } |
5623 | ||
5624 | return -ENOENT; | |
5625 | } | |
5626 | ||
3529d8c2 JA |
5627 | static int io_poll_remove_prep(struct io_kiocb *req, |
5628 | const struct io_uring_sqe *sqe) | |
0969e783 | 5629 | { |
0969e783 JA |
5630 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5631 | return -EINVAL; | |
5632 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5633 | sqe->poll_events) | |
5634 | return -EINVAL; | |
5635 | ||
018043be | 5636 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5637 | return 0; |
5638 | } | |
5639 | ||
221c5eb2 JA |
5640 | /* |
5641 | * Find a running poll command that matches one specified in sqe->addr, | |
5642 | * and remove it if found. | |
5643 | */ | |
61e98203 | 5644 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5645 | { |
5646 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5647 | int ret; |
221c5eb2 | 5648 | |
221c5eb2 | 5649 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5650 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5651 | spin_unlock_irq(&ctx->completion_lock); |
5652 | ||
4e88d6e7 JA |
5653 | if (ret < 0) |
5654 | req_set_fail_links(req); | |
e1e16097 | 5655 | io_req_complete(req, ret); |
221c5eb2 JA |
5656 | return 0; |
5657 | } | |
5658 | ||
221c5eb2 JA |
5659 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5660 | void *key) | |
5661 | { | |
c2f2eb7d JA |
5662 | struct io_kiocb *req = wait->private; |
5663 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5664 | |
d7718a9d | 5665 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5666 | } |
5667 | ||
221c5eb2 JA |
5668 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5669 | struct poll_table_struct *p) | |
5670 | { | |
5671 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5672 | ||
e8c2bc1f | 5673 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5674 | } |
5675 | ||
3529d8c2 | 5676 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5677 | { |
5678 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5679 | u32 events; |
221c5eb2 JA |
5680 | |
5681 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5682 | return -EINVAL; | |
5683 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5684 | return -EINVAL; | |
5685 | ||
5769a351 JX |
5686 | events = READ_ONCE(sqe->poll32_events); |
5687 | #ifdef __BIG_ENDIAN | |
5688 | events = swahw32(events); | |
5689 | #endif | |
a31eb4a2 JX |
5690 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5691 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5692 | return 0; |
5693 | } | |
5694 | ||
61e98203 | 5695 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5696 | { |
5697 | struct io_poll_iocb *poll = &req->poll; | |
5698 | struct io_ring_ctx *ctx = req->ctx; | |
5699 | struct io_poll_table ipt; | |
0969e783 | 5700 | __poll_t mask; |
0969e783 | 5701 | |
d7718a9d | 5702 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5703 | |
d7718a9d JA |
5704 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5705 | io_poll_wake); | |
221c5eb2 | 5706 | |
8c838788 | 5707 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5708 | ipt.error = 0; |
b0dd8a41 | 5709 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5710 | } |
221c5eb2 JA |
5711 | spin_unlock_irq(&ctx->completion_lock); |
5712 | ||
8c838788 JA |
5713 | if (mask) { |
5714 | io_cqring_ev_posted(ctx); | |
014db007 | 5715 | io_put_req(req); |
221c5eb2 | 5716 | } |
8c838788 | 5717 | return ipt.error; |
221c5eb2 JA |
5718 | } |
5719 | ||
5262f567 JA |
5720 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5721 | { | |
ad8a48ac JA |
5722 | struct io_timeout_data *data = container_of(timer, |
5723 | struct io_timeout_data, timer); | |
5724 | struct io_kiocb *req = data->req; | |
5725 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5726 | unsigned long flags; |
5727 | ||
5262f567 | 5728 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5729 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5730 | atomic_set(&req->ctx->cq_timeouts, |
5731 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5732 | ||
78e19bbe | 5733 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5734 | io_commit_cqring(ctx); |
5735 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5736 | ||
5737 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5738 | req_set_fail_links(req); |
5262f567 JA |
5739 | io_put_req(req); |
5740 | return HRTIMER_NORESTART; | |
5741 | } | |
5742 | ||
fbd15848 PB |
5743 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5744 | __u64 user_data) | |
f254ac04 | 5745 | { |
fbd15848 | 5746 | struct io_timeout_data *io; |
47f46768 JA |
5747 | struct io_kiocb *req; |
5748 | int ret = -ENOENT; | |
f254ac04 | 5749 | |
135fcde8 | 5750 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5751 | if (user_data == req->user_data) { |
47f46768 JA |
5752 | ret = 0; |
5753 | break; | |
5754 | } | |
5755 | } | |
5756 | ||
5757 | if (ret == -ENOENT) | |
fbd15848 PB |
5758 | return ERR_PTR(ret); |
5759 | ||
5760 | io = req->async_data; | |
e8c2bc1f | 5761 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5762 | if (ret == -1) |
fbd15848 | 5763 | return ERR_PTR(-EALREADY); |
a71976f3 | 5764 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5765 | return req; |
5766 | } | |
47f46768 | 5767 | |
fbd15848 PB |
5768 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5769 | { | |
5770 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5771 | ||
5772 | if (IS_ERR(req)) | |
5773 | return PTR_ERR(req); | |
f254ac04 JA |
5774 | |
5775 | req_set_fail_links(req); | |
f254ac04 | 5776 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5777 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5778 | return 0; |
5779 | } | |
5780 | ||
9c8e11b3 PB |
5781 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5782 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5783 | { |
9c8e11b3 PB |
5784 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5785 | struct io_timeout_data *data; | |
47f46768 | 5786 | |
9c8e11b3 PB |
5787 | if (IS_ERR(req)) |
5788 | return PTR_ERR(req); | |
47f46768 | 5789 | |
9c8e11b3 PB |
5790 | req->timeout.off = 0; /* noseq */ |
5791 | data = req->async_data; | |
5792 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5793 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5794 | data->timer.function = io_timeout_fn; | |
5795 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5796 | return 0; | |
47f46768 JA |
5797 | } |
5798 | ||
3529d8c2 JA |
5799 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5800 | const struct io_uring_sqe *sqe) | |
b29472ee | 5801 | { |
9c8e11b3 PB |
5802 | struct io_timeout_rem *tr = &req->timeout_rem; |
5803 | ||
b29472ee JA |
5804 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5805 | return -EINVAL; | |
61710e43 DA |
5806 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5807 | return -EINVAL; | |
9c8e11b3 | 5808 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5809 | return -EINVAL; |
5810 | ||
9c8e11b3 PB |
5811 | tr->addr = READ_ONCE(sqe->addr); |
5812 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5813 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5814 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5815 | return -EINVAL; | |
5816 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5817 | return -EFAULT; | |
5818 | } else if (tr->flags) { | |
5819 | /* timeout removal doesn't support flags */ | |
b29472ee | 5820 | return -EINVAL; |
9c8e11b3 | 5821 | } |
b29472ee | 5822 | |
b29472ee JA |
5823 | return 0; |
5824 | } | |
5825 | ||
8662daec PB |
5826 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5827 | { | |
5828 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5829 | : HRTIMER_MODE_REL; | |
5830 | } | |
5831 | ||
11365043 JA |
5832 | /* |
5833 | * Remove or update an existing timeout command | |
5834 | */ | |
61e98203 | 5835 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5836 | { |
9c8e11b3 | 5837 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5838 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5839 | int ret; |
11365043 | 5840 | |
11365043 | 5841 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5842 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5843 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5844 | else |
5845 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5846 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5847 | |
47f46768 | 5848 | io_cqring_fill_event(req, ret); |
11365043 JA |
5849 | io_commit_cqring(ctx); |
5850 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5851 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5852 | if (ret < 0) |
5853 | req_set_fail_links(req); | |
ec9c02ad | 5854 | io_put_req(req); |
11365043 | 5855 | return 0; |
5262f567 JA |
5856 | } |
5857 | ||
3529d8c2 | 5858 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5859 | bool is_timeout_link) |
5262f567 | 5860 | { |
ad8a48ac | 5861 | struct io_timeout_data *data; |
a41525ab | 5862 | unsigned flags; |
56080b02 | 5863 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5864 | |
ad8a48ac | 5865 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5866 | return -EINVAL; |
ad8a48ac | 5867 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5868 | return -EINVAL; |
56080b02 | 5869 | if (off && is_timeout_link) |
2d28390a | 5870 | return -EINVAL; |
a41525ab JA |
5871 | flags = READ_ONCE(sqe->timeout_flags); |
5872 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5873 | return -EINVAL; |
bdf20073 | 5874 | |
bfe68a22 | 5875 | req->timeout.off = off; |
26a61679 | 5876 | |
e8c2bc1f | 5877 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5878 | return -ENOMEM; |
5879 | ||
e8c2bc1f | 5880 | data = req->async_data; |
ad8a48ac | 5881 | data->req = req; |
ad8a48ac JA |
5882 | |
5883 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5884 | return -EFAULT; |
5885 | ||
8662daec | 5886 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5887 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5888 | return 0; | |
5889 | } | |
5890 | ||
61e98203 | 5891 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5892 | { |
ad8a48ac | 5893 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5894 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5895 | struct list_head *entry; |
bfe68a22 | 5896 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5897 | |
733f5c95 | 5898 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5899 | |
5262f567 JA |
5900 | /* |
5901 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5902 | * timeout event to be satisfied. If it isn't set, then this is |
5903 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5904 | */ |
8eb7e2d0 | 5905 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5906 | entry = ctx->timeout_list.prev; |
5907 | goto add; | |
5908 | } | |
5262f567 | 5909 | |
bfe68a22 PB |
5910 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5911 | req->timeout.target_seq = tail + off; | |
5262f567 | 5912 | |
f010505b MDG |
5913 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5914 | * This is safe because ->completion_lock is held, and submissions | |
5915 | * and completions are never mixed in the same ->completion_lock section. | |
5916 | */ | |
5917 | ctx->cq_last_tm_flush = tail; | |
5918 | ||
5262f567 JA |
5919 | /* |
5920 | * Insertion sort, ensuring the first entry in the list is always | |
5921 | * the one we need first. | |
5922 | */ | |
5262f567 | 5923 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5924 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5925 | timeout.list); | |
5262f567 | 5926 | |
8eb7e2d0 | 5927 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5928 | continue; |
bfe68a22 PB |
5929 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5930 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5931 | break; |
5932 | } | |
93bd25bb | 5933 | add: |
135fcde8 | 5934 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5935 | data->timer.function = io_timeout_fn; |
5936 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5937 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5938 | return 0; |
5939 | } | |
5262f567 | 5940 | |
62755e35 JA |
5941 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5942 | { | |
5943 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5944 | ||
5945 | return req->user_data == (unsigned long) data; | |
5946 | } | |
5947 | ||
e977d6d3 | 5948 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5949 | { |
62755e35 | 5950 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5951 | int ret = 0; |
5952 | ||
4f26bda1 | 5953 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5954 | switch (cancel_ret) { |
5955 | case IO_WQ_CANCEL_OK: | |
5956 | ret = 0; | |
5957 | break; | |
5958 | case IO_WQ_CANCEL_RUNNING: | |
5959 | ret = -EALREADY; | |
5960 | break; | |
5961 | case IO_WQ_CANCEL_NOTFOUND: | |
5962 | ret = -ENOENT; | |
5963 | break; | |
5964 | } | |
5965 | ||
e977d6d3 JA |
5966 | return ret; |
5967 | } | |
5968 | ||
47f46768 JA |
5969 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5970 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5971 | int success_ret) |
47f46768 JA |
5972 | { |
5973 | unsigned long flags; | |
5974 | int ret; | |
5975 | ||
5976 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5977 | if (ret != -ENOENT) { | |
5978 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5979 | goto done; | |
5980 | } | |
5981 | ||
5982 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5983 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5984 | if (ret != -ENOENT) | |
5985 | goto done; | |
5986 | ret = io_poll_cancel(ctx, sqe_addr); | |
5987 | done: | |
b0dd8a41 JA |
5988 | if (!ret) |
5989 | ret = success_ret; | |
47f46768 JA |
5990 | io_cqring_fill_event(req, ret); |
5991 | io_commit_cqring(ctx); | |
5992 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5993 | io_cqring_ev_posted(ctx); | |
5994 | ||
4e88d6e7 JA |
5995 | if (ret < 0) |
5996 | req_set_fail_links(req); | |
014db007 | 5997 | io_put_req(req); |
47f46768 JA |
5998 | } |
5999 | ||
3529d8c2 JA |
6000 | static int io_async_cancel_prep(struct io_kiocb *req, |
6001 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6002 | { |
fbf23849 | 6003 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6004 | return -EINVAL; |
61710e43 DA |
6005 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6006 | return -EINVAL; | |
6007 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
6008 | return -EINVAL; |
6009 | ||
fbf23849 JA |
6010 | req->cancel.addr = READ_ONCE(sqe->addr); |
6011 | return 0; | |
6012 | } | |
6013 | ||
61e98203 | 6014 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6015 | { |
6016 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 6017 | |
014db007 | 6018 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
6019 | return 0; |
6020 | } | |
6021 | ||
269bbe5f | 6022 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6023 | const struct io_uring_sqe *sqe) |
6024 | { | |
6ca56f84 JA |
6025 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
6026 | return -EINVAL; | |
61710e43 DA |
6027 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6028 | return -EINVAL; | |
6029 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
6030 | return -EINVAL; |
6031 | ||
269bbe5f BM |
6032 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6033 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6034 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6035 | return -EINVAL; |
269bbe5f | 6036 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6037 | return 0; |
6038 | } | |
6039 | ||
889fca73 | 6040 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6041 | { |
6042 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 6043 | struct io_uring_rsrc_update up; |
05f3fb3c | 6044 | int ret; |
fbf23849 | 6045 | |
45d189c6 | 6046 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6047 | return -EAGAIN; |
05f3fb3c | 6048 | |
269bbe5f BM |
6049 | up.offset = req->rsrc_update.offset; |
6050 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
6051 | |
6052 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 6053 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6054 | mutex_unlock(&ctx->uring_lock); |
6055 | ||
6056 | if (ret < 0) | |
6057 | req_set_fail_links(req); | |
889fca73 | 6058 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6059 | return 0; |
6060 | } | |
6061 | ||
bfe76559 | 6062 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6063 | { |
d625c6ee | 6064 | switch (req->opcode) { |
e781573e | 6065 | case IORING_OP_NOP: |
bfe76559 | 6066 | return 0; |
f67676d1 JA |
6067 | case IORING_OP_READV: |
6068 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6069 | case IORING_OP_READ: |
bfe76559 | 6070 | return io_read_prep(req, sqe); |
f67676d1 JA |
6071 | case IORING_OP_WRITEV: |
6072 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6073 | case IORING_OP_WRITE: |
bfe76559 | 6074 | return io_write_prep(req, sqe); |
0969e783 | 6075 | case IORING_OP_POLL_ADD: |
bfe76559 | 6076 | return io_poll_add_prep(req, sqe); |
0969e783 | 6077 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 6078 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 6079 | case IORING_OP_FSYNC: |
1155c76a | 6080 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6081 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6082 | return io_sfr_prep(req, sqe); |
03b1230c | 6083 | case IORING_OP_SENDMSG: |
fddaface | 6084 | case IORING_OP_SEND: |
bfe76559 | 6085 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6086 | case IORING_OP_RECVMSG: |
fddaface | 6087 | case IORING_OP_RECV: |
bfe76559 | 6088 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6089 | case IORING_OP_CONNECT: |
bfe76559 | 6090 | return io_connect_prep(req, sqe); |
2d28390a | 6091 | case IORING_OP_TIMEOUT: |
bfe76559 | 6092 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6093 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6094 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6095 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6096 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6097 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6098 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6099 | case IORING_OP_ACCEPT: |
bfe76559 | 6100 | return io_accept_prep(req, sqe); |
d63d1b5e | 6101 | case IORING_OP_FALLOCATE: |
bfe76559 | 6102 | return io_fallocate_prep(req, sqe); |
15b71abe | 6103 | case IORING_OP_OPENAT: |
bfe76559 | 6104 | return io_openat_prep(req, sqe); |
b5dba59e | 6105 | case IORING_OP_CLOSE: |
bfe76559 | 6106 | return io_close_prep(req, sqe); |
05f3fb3c | 6107 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6108 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6109 | case IORING_OP_STATX: |
bfe76559 | 6110 | return io_statx_prep(req, sqe); |
4840e418 | 6111 | case IORING_OP_FADVISE: |
bfe76559 | 6112 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6113 | case IORING_OP_MADVISE: |
bfe76559 | 6114 | return io_madvise_prep(req, sqe); |
cebdb986 | 6115 | case IORING_OP_OPENAT2: |
bfe76559 | 6116 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6117 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6118 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6119 | case IORING_OP_SPLICE: |
bfe76559 | 6120 | return io_splice_prep(req, sqe); |
ddf0322d | 6121 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6122 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6123 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6124 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6125 | case IORING_OP_TEE: |
bfe76559 | 6126 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6127 | case IORING_OP_SHUTDOWN: |
6128 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6129 | case IORING_OP_RENAMEAT: |
6130 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6131 | case IORING_OP_UNLINKAT: |
6132 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6133 | } |
6134 | ||
bfe76559 PB |
6135 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6136 | req->opcode); | |
6137 | return-EINVAL; | |
6138 | } | |
6139 | ||
93642ef8 PB |
6140 | static int io_req_prep_async(struct io_kiocb *req) |
6141 | { | |
6142 | switch (req->opcode) { | |
6143 | case IORING_OP_READV: | |
6144 | case IORING_OP_READ_FIXED: | |
6145 | case IORING_OP_READ: | |
6146 | return io_rw_prep_async(req, READ); | |
6147 | case IORING_OP_WRITEV: | |
6148 | case IORING_OP_WRITE_FIXED: | |
6149 | case IORING_OP_WRITE: | |
6150 | return io_rw_prep_async(req, WRITE); | |
6151 | case IORING_OP_SENDMSG: | |
6152 | case IORING_OP_SEND: | |
6153 | return io_sendmsg_prep_async(req); | |
6154 | case IORING_OP_RECVMSG: | |
6155 | case IORING_OP_RECV: | |
6156 | return io_recvmsg_prep_async(req); | |
6157 | case IORING_OP_CONNECT: | |
6158 | return io_connect_prep_async(req); | |
6159 | } | |
6160 | return 0; | |
6161 | } | |
6162 | ||
bfe76559 PB |
6163 | static int io_req_defer_prep(struct io_kiocb *req, |
6164 | const struct io_uring_sqe *sqe) | |
6165 | { | |
93642ef8 PB |
6166 | int ret; |
6167 | ||
bfe76559 PB |
6168 | if (!sqe) |
6169 | return 0; | |
6170 | if (io_alloc_async_data(req)) | |
6171 | return -EAGAIN; | |
93642ef8 PB |
6172 | ret = io_req_prep(req, sqe); |
6173 | if (ret) | |
6174 | return ret; | |
6175 | if (req->async_data) | |
6176 | return io_req_prep_async(req); | |
6177 | return 0; | |
6178 | ||
f67676d1 JA |
6179 | } |
6180 | ||
9cf7c104 PB |
6181 | static u32 io_get_sequence(struct io_kiocb *req) |
6182 | { | |
6183 | struct io_kiocb *pos; | |
6184 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 6185 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 6186 | |
f2f87370 PB |
6187 | io_for_each_link(pos, req) |
6188 | nr_reqs++; | |
9cf7c104 PB |
6189 | |
6190 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6191 | return total_submitted - nr_reqs; | |
6192 | } | |
6193 | ||
3529d8c2 | 6194 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 6195 | { |
a197f664 | 6196 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6197 | struct io_defer_entry *de; |
f67676d1 | 6198 | int ret; |
9cf7c104 | 6199 | u32 seq; |
de0617e4 | 6200 | |
9d858b21 | 6201 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6202 | if (likely(list_empty_careful(&ctx->defer_list) && |
6203 | !(req->flags & REQ_F_IO_DRAIN))) | |
6204 | return 0; | |
6205 | ||
6206 | seq = io_get_sequence(req); | |
6207 | /* Still a chance to pass the sequence check */ | |
6208 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6209 | return 0; |
6210 | ||
e8c2bc1f | 6211 | if (!req->async_data) { |
650b5481 | 6212 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6213 | if (ret) |
650b5481 PB |
6214 | return ret; |
6215 | } | |
cbdcb435 | 6216 | io_prep_async_link(req); |
27dc8338 PB |
6217 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6218 | if (!de) | |
6219 | return -ENOMEM; | |
2d28390a | 6220 | |
de0617e4 | 6221 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6222 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6223 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6224 | kfree(de); |
ae34817b PB |
6225 | io_queue_async_work(req); |
6226 | return -EIOCBQUEUED; | |
de0617e4 JA |
6227 | } |
6228 | ||
915967f6 | 6229 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6230 | de->req = req; |
9cf7c104 | 6231 | de->seq = seq; |
27dc8338 | 6232 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6233 | spin_unlock_irq(&ctx->completion_lock); |
6234 | return -EIOCBQUEUED; | |
6235 | } | |
6236 | ||
3ca405eb | 6237 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6238 | { |
0e1b6fe3 PB |
6239 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6240 | switch (req->opcode) { | |
6241 | case IORING_OP_READV: | |
6242 | case IORING_OP_READ_FIXED: | |
6243 | case IORING_OP_READ: | |
bcda7baa | 6244 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6245 | break; |
6246 | case IORING_OP_RECVMSG: | |
6247 | case IORING_OP_RECV: | |
bcda7baa | 6248 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6249 | break; |
6250 | } | |
6251 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6252 | } |
6253 | ||
0e1b6fe3 PB |
6254 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6255 | switch (req->opcode) { | |
6256 | case IORING_OP_READV: | |
6257 | case IORING_OP_READ_FIXED: | |
6258 | case IORING_OP_READ: | |
6259 | case IORING_OP_WRITEV: | |
6260 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6261 | case IORING_OP_WRITE: { |
6262 | struct io_async_rw *io = req->async_data; | |
6263 | if (io->free_iovec) | |
6264 | kfree(io->free_iovec); | |
0e1b6fe3 | 6265 | break; |
e8c2bc1f | 6266 | } |
0e1b6fe3 | 6267 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6268 | case IORING_OP_SENDMSG: { |
6269 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6270 | |
6271 | kfree(io->free_iov); | |
0e1b6fe3 | 6272 | break; |
e8c2bc1f | 6273 | } |
0e1b6fe3 PB |
6274 | case IORING_OP_SPLICE: |
6275 | case IORING_OP_TEE: | |
6276 | io_put_file(req, req->splice.file_in, | |
6277 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6278 | break; | |
f3cd4850 JA |
6279 | case IORING_OP_OPENAT: |
6280 | case IORING_OP_OPENAT2: | |
6281 | if (req->open.filename) | |
6282 | putname(req->open.filename); | |
6283 | break; | |
80a261fd JA |
6284 | case IORING_OP_RENAMEAT: |
6285 | putname(req->rename.oldpath); | |
6286 | putname(req->rename.newpath); | |
6287 | break; | |
14a1143b JA |
6288 | case IORING_OP_UNLINKAT: |
6289 | putname(req->unlink.filename); | |
6290 | break; | |
0e1b6fe3 PB |
6291 | } |
6292 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6293 | } |
99bc4c38 PB |
6294 | } |
6295 | ||
889fca73 | 6296 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6297 | { |
a197f664 | 6298 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 6299 | int ret; |
2b188cc1 | 6300 | |
d625c6ee | 6301 | switch (req->opcode) { |
2b188cc1 | 6302 | case IORING_OP_NOP: |
889fca73 | 6303 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6304 | break; |
6305 | case IORING_OP_READV: | |
edafccee | 6306 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6307 | case IORING_OP_READ: |
889fca73 | 6308 | ret = io_read(req, issue_flags); |
edafccee | 6309 | break; |
3529d8c2 | 6310 | case IORING_OP_WRITEV: |
edafccee | 6311 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6312 | case IORING_OP_WRITE: |
889fca73 | 6313 | ret = io_write(req, issue_flags); |
2b188cc1 | 6314 | break; |
c992fe29 | 6315 | case IORING_OP_FSYNC: |
45d189c6 | 6316 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6317 | break; |
221c5eb2 | 6318 | case IORING_OP_POLL_ADD: |
61e98203 | 6319 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6320 | break; |
6321 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6322 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6323 | break; |
5d17b4a4 | 6324 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6325 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6326 | break; |
0fa03c62 | 6327 | case IORING_OP_SENDMSG: |
889fca73 | 6328 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6329 | break; |
fddaface | 6330 | case IORING_OP_SEND: |
889fca73 | 6331 | ret = io_send(req, issue_flags); |
0fa03c62 | 6332 | break; |
aa1fa28f | 6333 | case IORING_OP_RECVMSG: |
889fca73 | 6334 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6335 | break; |
fddaface | 6336 | case IORING_OP_RECV: |
889fca73 | 6337 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6338 | break; |
5262f567 | 6339 | case IORING_OP_TIMEOUT: |
61e98203 | 6340 | ret = io_timeout(req, issue_flags); |
5262f567 | 6341 | break; |
11365043 | 6342 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6343 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6344 | break; |
17f2fe35 | 6345 | case IORING_OP_ACCEPT: |
889fca73 | 6346 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6347 | break; |
f8e85cf2 | 6348 | case IORING_OP_CONNECT: |
889fca73 | 6349 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6350 | break; |
62755e35 | 6351 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6352 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6353 | break; |
d63d1b5e | 6354 | case IORING_OP_FALLOCATE: |
45d189c6 | 6355 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6356 | break; |
15b71abe | 6357 | case IORING_OP_OPENAT: |
45d189c6 | 6358 | ret = io_openat(req, issue_flags); |
15b71abe | 6359 | break; |
b5dba59e | 6360 | case IORING_OP_CLOSE: |
889fca73 | 6361 | ret = io_close(req, issue_flags); |
b5dba59e | 6362 | break; |
05f3fb3c | 6363 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6364 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6365 | break; |
eddc7ef5 | 6366 | case IORING_OP_STATX: |
45d189c6 | 6367 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6368 | break; |
4840e418 | 6369 | case IORING_OP_FADVISE: |
45d189c6 | 6370 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6371 | break; |
c1ca757b | 6372 | case IORING_OP_MADVISE: |
45d189c6 | 6373 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6374 | break; |
cebdb986 | 6375 | case IORING_OP_OPENAT2: |
45d189c6 | 6376 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6377 | break; |
3e4827b0 | 6378 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6379 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6380 | break; |
7d67af2c | 6381 | case IORING_OP_SPLICE: |
45d189c6 | 6382 | ret = io_splice(req, issue_flags); |
7d67af2c | 6383 | break; |
ddf0322d | 6384 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6385 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6386 | break; |
067524e9 | 6387 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6388 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6389 | break; |
f2a8d5c7 | 6390 | case IORING_OP_TEE: |
45d189c6 | 6391 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6392 | break; |
36f4fa68 | 6393 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6394 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6395 | break; |
80a261fd | 6396 | case IORING_OP_RENAMEAT: |
45d189c6 | 6397 | ret = io_renameat(req, issue_flags); |
80a261fd | 6398 | break; |
14a1143b | 6399 | case IORING_OP_UNLINKAT: |
45d189c6 | 6400 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6401 | break; |
2b188cc1 JA |
6402 | default: |
6403 | ret = -EINVAL; | |
6404 | break; | |
6405 | } | |
6406 | ||
def596e9 JA |
6407 | if (ret) |
6408 | return ret; | |
6409 | ||
b532576e JA |
6410 | /* If the op doesn't have a file, we're not polling for it */ |
6411 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6412 | const bool in_async = io_wq_current_is_worker(); |
6413 | ||
11ba820b JA |
6414 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6415 | if (in_async) | |
6416 | mutex_lock(&ctx->uring_lock); | |
6417 | ||
2e9dbe90 | 6418 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6419 | |
6420 | if (in_async) | |
6421 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6422 | } |
6423 | ||
6424 | return 0; | |
2b188cc1 JA |
6425 | } |
6426 | ||
5280f7e5 | 6427 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6428 | { |
6429 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6430 | struct io_kiocb *timeout; |
561fb04a | 6431 | int ret = 0; |
2b188cc1 | 6432 | |
6df1db6b PB |
6433 | timeout = io_prep_linked_timeout(req); |
6434 | if (timeout) | |
6435 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6436 | |
4014d943 | 6437 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6438 | ret = -ECANCELED; |
31b51510 | 6439 | |
561fb04a | 6440 | if (!ret) { |
561fb04a | 6441 | do { |
889fca73 | 6442 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6443 | /* |
6444 | * We can get EAGAIN for polled IO even though we're | |
6445 | * forcing a sync submission from here, since we can't | |
6446 | * wait for request slots on the block side. | |
6447 | */ | |
6448 | if (ret != -EAGAIN) | |
6449 | break; | |
6450 | cond_resched(); | |
6451 | } while (1); | |
6452 | } | |
31b51510 | 6453 | |
561fb04a | 6454 | if (ret) { |
c07e6719 XW |
6455 | struct io_ring_ctx *lock_ctx = NULL; |
6456 | ||
6457 | if (req->ctx->flags & IORING_SETUP_IOPOLL) | |
6458 | lock_ctx = req->ctx; | |
6459 | ||
dad1b124 | 6460 | /* |
c07e6719 XW |
6461 | * io_iopoll_complete() does not hold completion_lock to |
6462 | * complete polled io, so here for polled io, we can not call | |
6463 | * io_req_complete() directly, otherwise there maybe concurrent | |
6464 | * access to cqring, defer_list, etc, which is not safe. Given | |
6465 | * that io_iopoll_complete() is always called under uring_lock, | |
6466 | * so here for polled io, we also get uring_lock to complete | |
6467 | * it. | |
dad1b124 | 6468 | */ |
c07e6719 XW |
6469 | if (lock_ctx) |
6470 | mutex_lock(&lock_ctx->uring_lock); | |
dad1b124 | 6471 | |
c07e6719 XW |
6472 | req_set_fail_links(req); |
6473 | io_req_complete(req, ret); | |
6474 | ||
6475 | if (lock_ctx) | |
6476 | mutex_unlock(&lock_ctx->uring_lock); | |
edafccee | 6477 | } |
2b188cc1 JA |
6478 | } |
6479 | ||
65e19f54 JA |
6480 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6481 | int index) | |
6482 | { | |
269bbe5f | 6483 | struct fixed_rsrc_table *table; |
65e19f54 | 6484 | |
05f3fb3c | 6485 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6486 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6487 | } |
6488 | ||
8371adf5 PB |
6489 | static struct file *io_file_get(struct io_submit_state *state, |
6490 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6491 | { |
a197f664 | 6492 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6493 | struct file *file; |
09bb8394 | 6494 | |
8da11c19 | 6495 | if (fixed) { |
479f517b | 6496 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6497 | return NULL; |
b7620121 | 6498 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6499 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6500 | io_set_resource_node(req); |
09bb8394 | 6501 | } else { |
c826bd7a | 6502 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6503 | file = __io_file_get(state, fd); |
09bb8394 JA |
6504 | } |
6505 | ||
ce3d5aae PB |
6506 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6507 | io_req_track_inflight(req); | |
8371adf5 | 6508 | return file; |
09bb8394 JA |
6509 | } |
6510 | ||
2665abfd | 6511 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6512 | { |
ad8a48ac JA |
6513 | struct io_timeout_data *data = container_of(timer, |
6514 | struct io_timeout_data, timer); | |
90cd7e42 | 6515 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6516 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6517 | unsigned long flags; |
2665abfd JA |
6518 | |
6519 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6520 | prev = req->timeout.head; |
6521 | req->timeout.head = NULL; | |
2665abfd JA |
6522 | |
6523 | /* | |
6524 | * We don't expect the list to be empty, that will only happen if we | |
6525 | * race with the completion of the linked work. | |
6526 | */ | |
90cd7e42 | 6527 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6528 | io_remove_next_linked(prev); |
90cd7e42 PB |
6529 | else |
6530 | prev = NULL; | |
2665abfd JA |
6531 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6532 | ||
6533 | if (prev) { | |
4e88d6e7 | 6534 | req_set_fail_links(prev); |
014db007 | 6535 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6536 | io_put_req_deferred(prev, 1); |
47f46768 | 6537 | } else { |
9ae1f8dd PB |
6538 | io_req_complete_post(req, -ETIME, 0); |
6539 | io_put_req_deferred(req, 1); | |
2665abfd | 6540 | } |
2665abfd JA |
6541 | return HRTIMER_NORESTART; |
6542 | } | |
6543 | ||
7271ef3a | 6544 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6545 | { |
76a46e06 | 6546 | /* |
f2f87370 PB |
6547 | * If the back reference is NULL, then our linked request finished |
6548 | * before we got a chance to setup the timer | |
76a46e06 | 6549 | */ |
90cd7e42 | 6550 | if (req->timeout.head) { |
e8c2bc1f | 6551 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6552 | |
ad8a48ac JA |
6553 | data->timer.function = io_link_timeout_fn; |
6554 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6555 | data->mode); | |
2665abfd | 6556 | } |
7271ef3a JA |
6557 | } |
6558 | ||
6559 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6560 | { | |
6561 | struct io_ring_ctx *ctx = req->ctx; | |
6562 | ||
6563 | spin_lock_irq(&ctx->completion_lock); | |
6564 | __io_queue_linked_timeout(req); | |
76a46e06 | 6565 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6566 | |
2665abfd | 6567 | /* drop submission reference */ |
76a46e06 JA |
6568 | io_put_req(req); |
6569 | } | |
2665abfd | 6570 | |
ad8a48ac | 6571 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6572 | { |
f2f87370 | 6573 | struct io_kiocb *nxt = req->link; |
2665abfd | 6574 | |
f2f87370 PB |
6575 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6576 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6577 | return NULL; |
2665abfd | 6578 | |
90cd7e42 | 6579 | nxt->timeout.head = req; |
900fad45 | 6580 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6581 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6582 | return nxt; |
2665abfd JA |
6583 | } |
6584 | ||
c5eef2b9 | 6585 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6586 | { |
d3d7298d | 6587 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
193155c8 | 6588 | const struct cred *old_creds = NULL; |
c5eef2b9 | 6589 | int ret; |
2b188cc1 | 6590 | |
2e5aa6cb PB |
6591 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6592 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
d3d7298d PB |
6593 | req->work.identity->creds != current_cred()) |
6594 | old_creds = override_creds(req->work.identity->creds); | |
193155c8 | 6595 | |
c5eef2b9 | 6596 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
491381ce | 6597 | |
d3d7298d PB |
6598 | if (old_creds) |
6599 | revert_creds(old_creds); | |
6600 | ||
491381ce JA |
6601 | /* |
6602 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6603 | * doesn't support non-blocking read/write attempts | |
6604 | */ | |
24c74678 | 6605 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6606 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6607 | /* |
6608 | * Queued up for async execution, worker will release | |
6609 | * submit reference when the iocb is actually submitted. | |
6610 | */ | |
6611 | io_queue_async_work(req); | |
2b188cc1 | 6612 | } |
0d63c148 PB |
6613 | } else if (likely(!ret)) { |
6614 | /* drop submission reference */ | |
e342c807 | 6615 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6616 | struct io_ring_ctx *ctx = req->ctx; |
6617 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
6618 | ||
6dd0be1e | 6619 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6620 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6621 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6622 | } else { |
d3d7298d | 6623 | io_put_req(req); |
0d63c148 PB |
6624 | } |
6625 | } else { | |
4e88d6e7 | 6626 | req_set_fail_links(req); |
e65ef56d | 6627 | io_put_req(req); |
e1e16097 | 6628 | io_req_complete(req, ret); |
9e645e11 | 6629 | } |
d3d7298d PB |
6630 | if (linked_timeout) |
6631 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6632 | } |
6633 | ||
c5eef2b9 | 6634 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
6635 | { |
6636 | int ret; | |
6637 | ||
3529d8c2 | 6638 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6639 | if (ret) { |
6640 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6641 | fail_req: |
4e88d6e7 | 6642 | req_set_fail_links(req); |
e1e16097 JA |
6643 | io_put_req(req); |
6644 | io_req_complete(req, ret); | |
4fe2c963 | 6645 | } |
2550878f | 6646 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6647 | if (!req->async_data) { |
bd2ab18a | 6648 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6649 | if (unlikely(ret)) |
bd2ab18a PB |
6650 | goto fail_req; |
6651 | } | |
ce35a47a JA |
6652 | io_queue_async_work(req); |
6653 | } else { | |
c1379e24 PB |
6654 | if (sqe) { |
6655 | ret = io_req_prep(req, sqe); | |
6656 | if (unlikely(ret)) | |
6657 | goto fail_req; | |
6658 | } | |
c5eef2b9 | 6659 | __io_queue_sqe(req); |
ce35a47a | 6660 | } |
4fe2c963 JL |
6661 | } |
6662 | ||
c5eef2b9 | 6663 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 6664 | { |
94ae5e77 | 6665 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6666 | io_put_req(req); |
6667 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6668 | } else |
c5eef2b9 | 6669 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
6670 | } |
6671 | ||
b16fed66 PB |
6672 | /* |
6673 | * Check SQE restrictions (opcode and flags). | |
6674 | * | |
6675 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6676 | */ | |
6677 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6678 | struct io_kiocb *req, | |
6679 | unsigned int sqe_flags) | |
6680 | { | |
6681 | if (!ctx->restricted) | |
6682 | return true; | |
6683 | ||
6684 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6685 | return false; | |
6686 | ||
6687 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6688 | ctx->restrictions.sqe_flags_required) | |
6689 | return false; | |
6690 | ||
6691 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6692 | ctx->restrictions.sqe_flags_required)) | |
6693 | return false; | |
6694 | ||
6695 | return true; | |
6696 | } | |
6697 | ||
6698 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6699 | const struct io_uring_sqe *sqe) | |
6700 | { | |
6701 | struct io_submit_state *state; | |
6702 | unsigned int sqe_flags; | |
6703 | int id, ret = 0; | |
6704 | ||
6705 | req->opcode = READ_ONCE(sqe->opcode); | |
6706 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6707 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6708 | req->user_data = READ_ONCE(sqe->user_data); | |
6709 | req->async_data = NULL; | |
6710 | req->file = NULL; | |
6711 | req->ctx = ctx; | |
6712 | req->link = NULL; | |
6713 | req->fixed_rsrc_refs = NULL; | |
6714 | /* one is dropped after submission, the other at completion */ | |
6715 | refcount_set(&req->refs, 2); | |
6716 | req->task = current; | |
6717 | req->result = 0; | |
6718 | ||
6719 | /* enforce forwards compatibility on users */ | |
6720 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6721 | return -EINVAL; | |
6722 | ||
6723 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6724 | return -EINVAL; | |
6725 | ||
6726 | if (unlikely(io_sq_thread_acquire_mm_files(ctx, req))) | |
6727 | return -EFAULT; | |
6728 | ||
6729 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) | |
6730 | return -EACCES; | |
6731 | ||
6732 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6733 | !io_op_defs[req->opcode].buffer_select) | |
6734 | return -EOPNOTSUPP; | |
6735 | ||
6736 | id = READ_ONCE(sqe->personality); | |
6737 | if (id) { | |
6738 | struct io_identity *iod; | |
6739 | ||
6740 | iod = idr_find(&ctx->personality_idr, id); | |
6741 | if (unlikely(!iod)) | |
6742 | return -EINVAL; | |
6743 | refcount_inc(&iod->count); | |
6744 | ||
6745 | __io_req_init_async(req); | |
6746 | get_cred(iod->creds); | |
6747 | req->work.identity = iod; | |
6748 | req->work.flags |= IO_WQ_WORK_CREDS; | |
6749 | } | |
6750 | ||
6751 | state = &ctx->submit_state; | |
6752 | ||
6753 | /* | |
6754 | * Plug now if we have more than 1 IO left after this, and the target | |
6755 | * is potentially a read/write to block based storage. | |
6756 | */ | |
6757 | if (!state->plug_started && state->ios_left > 1 && | |
6758 | io_op_defs[req->opcode].plug) { | |
6759 | blk_start_plug(&state->plug); | |
6760 | state->plug_started = true; | |
6761 | } | |
6762 | ||
6763 | if (io_op_defs[req->opcode].needs_file) { | |
6764 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6765 | ||
6766 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6767 | if (unlikely(!req->file)) | |
6768 | ret = -EBADF; | |
6769 | } | |
6770 | ||
6771 | state->ios_left--; | |
6772 | return ret; | |
6773 | } | |
6774 | ||
a6b8cadc | 6775 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6776 | const struct io_uring_sqe *sqe) |
9e645e11 | 6777 | { |
a1ab7b35 | 6778 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6779 | int ret; |
9e645e11 | 6780 | |
a6b8cadc PB |
6781 | ret = io_init_req(ctx, req, sqe); |
6782 | if (unlikely(ret)) { | |
6783 | fail_req: | |
6784 | io_put_req(req); | |
6785 | io_req_complete(req, ret); | |
cf109604 PB |
6786 | /* fail even hard links since we don't submit */ |
6787 | if (link->head) | |
6788 | link->head->flags |= REQ_F_FAIL_LINK; | |
a6b8cadc PB |
6789 | return ret; |
6790 | } | |
6791 | ||
6792 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, | |
6793 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6794 | ||
9e645e11 JA |
6795 | /* |
6796 | * If we already have a head request, queue this one for async | |
6797 | * submittal once the head completes. If we don't have a head but | |
6798 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6799 | * submitted sync once the chain is complete. If none of those | |
6800 | * conditions are true (normal request), then just queue it. | |
6801 | */ | |
863e0560 PB |
6802 | if (link->head) { |
6803 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6804 | |
8cdf2193 PB |
6805 | /* |
6806 | * Taking sequential execution of a link, draining both sides | |
6807 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6808 | * requests in the link. So, it drains the head and the | |
6809 | * next after the link request. The last one is done via | |
6810 | * drain_next flag to persist the effect across calls. | |
6811 | */ | |
ef4ff581 | 6812 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6813 | head->flags |= REQ_F_IO_DRAIN; |
6814 | ctx->drain_next = 1; | |
6815 | } | |
3529d8c2 | 6816 | ret = io_req_defer_prep(req, sqe); |
cf109604 | 6817 | if (unlikely(ret)) |
a6b8cadc | 6818 | goto fail_req; |
9d76377f | 6819 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6820 | link->last->link = req; |
863e0560 | 6821 | link->last = req; |
32fe525b PB |
6822 | |
6823 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6824 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
c5eef2b9 | 6825 | io_queue_link_head(head); |
863e0560 | 6826 | link->head = NULL; |
32fe525b | 6827 | } |
9e645e11 | 6828 | } else { |
711be031 PB |
6829 | if (unlikely(ctx->drain_next)) { |
6830 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6831 | ctx->drain_next = 0; |
711be031 | 6832 | } |
ef4ff581 | 6833 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
711be031 | 6834 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6835 | if (unlikely(ret)) |
711be031 | 6836 | req->flags |= REQ_F_FAIL_LINK; |
863e0560 PB |
6837 | link->head = req; |
6838 | link->last = req; | |
711be031 | 6839 | } else { |
c5eef2b9 | 6840 | io_queue_sqe(req, sqe); |
711be031 | 6841 | } |
9e645e11 | 6842 | } |
2e6e1fde | 6843 | |
1d4240cc | 6844 | return 0; |
9e645e11 JA |
6845 | } |
6846 | ||
9a56a232 JA |
6847 | /* |
6848 | * Batched submission is done, ensure local IO is flushed out. | |
6849 | */ | |
ba88ff11 PB |
6850 | static void io_submit_state_end(struct io_submit_state *state, |
6851 | struct io_ring_ctx *ctx) | |
9a56a232 | 6852 | { |
a1ab7b35 PB |
6853 | if (state->link.head) |
6854 | io_queue_link_head(state->link.head); | |
6dd0be1e | 6855 | if (state->comp.nr) |
ba88ff11 | 6856 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6857 | if (state->plug_started) |
6858 | blk_finish_plug(&state->plug); | |
9f13c35b | 6859 | io_state_file_put(state); |
9a56a232 JA |
6860 | } |
6861 | ||
6862 | /* | |
6863 | * Start submission side cache. | |
6864 | */ | |
6865 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6866 | unsigned int max_ios) |
9a56a232 | 6867 | { |
27926b68 | 6868 | state->plug_started = false; |
9a56a232 | 6869 | state->ios_left = max_ios; |
a1ab7b35 PB |
6870 | /* set only head, no need to init link_last in advance */ |
6871 | state->link.head = NULL; | |
9a56a232 JA |
6872 | } |
6873 | ||
2b188cc1 JA |
6874 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6875 | { | |
75b28aff | 6876 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6877 | |
caf582c6 PB |
6878 | /* |
6879 | * Ensure any loads from the SQEs are done at this point, | |
6880 | * since once we write the new head, the application could | |
6881 | * write new data to them. | |
6882 | */ | |
6883 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6884 | } |
6885 | ||
2b188cc1 | 6886 | /* |
3529d8c2 | 6887 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6888 | * that is mapped by userspace. This means that care needs to be taken to |
6889 | * ensure that reads are stable, as we cannot rely on userspace always | |
6890 | * being a good citizen. If members of the sqe are validated and then later | |
6891 | * used, it's important that those reads are done through READ_ONCE() to | |
6892 | * prevent a re-load down the line. | |
6893 | */ | |
709b302f | 6894 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6895 | { |
75b28aff | 6896 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6897 | unsigned head; |
6898 | ||
6899 | /* | |
6900 | * The cached sq head (or cq tail) serves two purposes: | |
6901 | * | |
6902 | * 1) allows us to batch the cost of updating the user visible | |
6903 | * head updates. | |
6904 | * 2) allows the kernel side to track the head on its own, even | |
6905 | * though the application is the one updating it. | |
6906 | */ | |
4fccfcbb | 6907 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6908 | if (likely(head < ctx->sq_entries)) |
6909 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6910 | |
6911 | /* drop invalid entries */ | |
498ccd9e | 6912 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6913 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6914 | return NULL; |
6915 | } | |
6916 | ||
0f212204 | 6917 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6918 | { |
46c4e16a | 6919 | int submitted = 0; |
6c271ce2 | 6920 | |
c4a2ed72 | 6921 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6922 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6923 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6924 | return -EBUSY; |
6925 | } | |
6c271ce2 | 6926 | |
ee7d46d9 PB |
6927 | /* make sure SQ entry isn't read before tail */ |
6928 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6929 | |
2b85edfc PB |
6930 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6931 | return -EAGAIN; | |
6c271ce2 | 6932 | |
d8a6df10 | 6933 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6934 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6935 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6936 | |
46c4e16a | 6937 | while (submitted < nr) { |
3529d8c2 | 6938 | const struct io_uring_sqe *sqe; |
196be95c | 6939 | struct io_kiocb *req; |
fb5ccc98 | 6940 | |
258b29a9 | 6941 | req = io_alloc_req(ctx); |
196be95c PB |
6942 | if (unlikely(!req)) { |
6943 | if (!submitted) | |
6944 | submitted = -EAGAIN; | |
fb5ccc98 | 6945 | break; |
196be95c | 6946 | } |
4fccfcbb PB |
6947 | sqe = io_get_sqe(ctx); |
6948 | if (unlikely(!sqe)) { | |
6949 | kmem_cache_free(req_cachep, req); | |
6950 | break; | |
6951 | } | |
d3656344 JA |
6952 | /* will complete beyond this point, count as submitted */ |
6953 | submitted++; | |
a1ab7b35 | 6954 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6955 | break; |
6c271ce2 JA |
6956 | } |
6957 | ||
9466f437 PB |
6958 | if (unlikely(submitted != nr)) { |
6959 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6960 | struct io_uring_task *tctx = current->io_uring; |
6961 | int unused = nr - ref_used; | |
9466f437 | 6962 | |
d8a6df10 JA |
6963 | percpu_ref_put_many(&ctx->refs, unused); |
6964 | percpu_counter_sub(&tctx->inflight, unused); | |
6965 | put_task_struct_many(current, unused); | |
9466f437 | 6966 | } |
6c271ce2 | 6967 | |
a1ab7b35 | 6968 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6969 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6970 | io_commit_sqring(ctx); | |
6971 | ||
6c271ce2 JA |
6972 | return submitted; |
6973 | } | |
6974 | ||
23b3628e XW |
6975 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6976 | { | |
6977 | /* Tell userspace we may need a wakeup call */ | |
6978 | spin_lock_irq(&ctx->completion_lock); | |
6979 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6980 | spin_unlock_irq(&ctx->completion_lock); | |
6981 | } | |
6982 | ||
6983 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6984 | { | |
6985 | spin_lock_irq(&ctx->completion_lock); | |
6986 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6987 | spin_unlock_irq(&ctx->completion_lock); | |
6988 | } | |
6989 | ||
08369246 | 6990 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6991 | { |
c8d1ba58 | 6992 | unsigned int to_submit; |
bdcd3eab | 6993 | int ret = 0; |
6c271ce2 | 6994 | |
c8d1ba58 | 6995 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6996 | /* if we're handling multiple rings, cap submit size for fairness */ |
6997 | if (cap_entries && to_submit > 8) | |
6998 | to_submit = 8; | |
6999 | ||
906a3c6f | 7000 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 7001 | unsigned nr_events = 0; |
a4c0b3de | 7002 | |
c8d1ba58 | 7003 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 7004 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 7005 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 7006 | |
d9d05217 PB |
7007 | if (to_submit && !ctx->sqo_dead && |
7008 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 7009 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
7010 | mutex_unlock(&ctx->uring_lock); |
7011 | } | |
6c271ce2 | 7012 | |
90554200 JA |
7013 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
7014 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 7015 | |
08369246 XW |
7016 | return ret; |
7017 | } | |
6c271ce2 | 7018 | |
08369246 XW |
7019 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7020 | { | |
7021 | struct io_ring_ctx *ctx; | |
7022 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7023 | |
08369246 XW |
7024 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7025 | if (sq_thread_idle < ctx->sq_thread_idle) | |
7026 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 7027 | } |
c1edbf5f | 7028 | |
08369246 | 7029 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7030 | } |
6c271ce2 | 7031 | |
69fb2131 JA |
7032 | static void io_sqd_init_new(struct io_sq_data *sqd) |
7033 | { | |
7034 | struct io_ring_ctx *ctx; | |
7035 | ||
7036 | while (!list_empty(&sqd->ctx_new_list)) { | |
7037 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
7038 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
7039 | complete(&ctx->sq_thread_comp); | |
7040 | } | |
08369246 XW |
7041 | |
7042 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
7043 | } |
7044 | ||
c8d1ba58 JA |
7045 | static int io_sq_thread(void *data) |
7046 | { | |
91d8f519 | 7047 | struct cgroup_subsys_state *cur_css = NULL; |
28cea78a JA |
7048 | struct files_struct *old_files = current->files; |
7049 | struct nsproxy *old_nsproxy = current->nsproxy; | |
69fb2131 JA |
7050 | const struct cred *old_cred = NULL; |
7051 | struct io_sq_data *sqd = data; | |
7052 | struct io_ring_ctx *ctx; | |
a0d9205f | 7053 | unsigned long timeout = 0; |
08369246 | 7054 | DEFINE_WAIT(wait); |
6c271ce2 | 7055 | |
28cea78a JA |
7056 | task_lock(current); |
7057 | current->files = NULL; | |
7058 | current->nsproxy = NULL; | |
7059 | task_unlock(current); | |
6c271ce2 | 7060 | |
69fb2131 | 7061 | while (!kthread_should_stop()) { |
08369246 XW |
7062 | int ret; |
7063 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
7064 | |
7065 | /* | |
69fb2131 JA |
7066 | * Any changes to the sqd lists are synchronized through the |
7067 | * kthread parking. This synchronizes the thread vs users, | |
7068 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 7069 | */ |
65b2b213 | 7070 | if (kthread_should_park()) { |
69fb2131 | 7071 | kthread_parkme(); |
65b2b213 XW |
7072 | /* |
7073 | * When sq thread is unparked, in case the previous park operation | |
7074 | * comes from io_put_sq_data(), which means that sq thread is going | |
7075 | * to be stopped, so here needs to have a check. | |
7076 | */ | |
7077 | if (kthread_should_stop()) | |
7078 | break; | |
7079 | } | |
7143b5ac | 7080 | |
08369246 | 7081 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 7082 | io_sqd_init_new(sqd); |
08369246 XW |
7083 | timeout = jiffies + sqd->sq_thread_idle; |
7084 | } | |
6c271ce2 | 7085 | |
08369246 | 7086 | sqt_spin = false; |
e95eee2d | 7087 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 JA |
7088 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7089 | if (current->cred != ctx->creds) { | |
7090 | if (old_cred) | |
7091 | revert_creds(old_cred); | |
7092 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 7093 | } |
91d8f519 | 7094 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
7095 | #ifdef CONFIG_AUDIT |
7096 | current->loginuid = ctx->loginuid; | |
7097 | current->sessionid = ctx->sessionid; | |
7098 | #endif | |
bdcd3eab | 7099 | |
08369246 XW |
7100 | ret = __io_sq_thread(ctx, cap_entries); |
7101 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
7102 | sqt_spin = true; | |
6c271ce2 | 7103 | |
28cea78a | 7104 | io_sq_thread_drop_mm_files(); |
69fb2131 | 7105 | } |
6c271ce2 | 7106 | |
08369246 | 7107 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7108 | io_run_task_work(); |
d434ab6d | 7109 | io_sq_thread_drop_mm_files(); |
c8d1ba58 | 7110 | cond_resched(); |
08369246 XW |
7111 | if (sqt_spin) |
7112 | timeout = jiffies + sqd->sq_thread_idle; | |
7113 | continue; | |
7114 | } | |
7115 | ||
08369246 XW |
7116 | needs_sched = true; |
7117 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
7118 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
7119 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
7120 | !list_empty_careful(&ctx->iopoll_list)) { | |
7121 | needs_sched = false; | |
7122 | break; | |
7123 | } | |
7124 | if (io_sqring_entries(ctx)) { | |
7125 | needs_sched = false; | |
7126 | break; | |
7127 | } | |
7128 | } | |
7129 | ||
8b28fdf2 | 7130 | if (needs_sched && !kthread_should_park()) { |
69fb2131 JA |
7131 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7132 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 7133 | |
69fb2131 | 7134 | schedule(); |
69fb2131 JA |
7135 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7136 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7137 | } |
08369246 XW |
7138 | |
7139 | finish_wait(&sqd->wait, &wait); | |
7140 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
7141 | } |
7142 | ||
4c6e277c | 7143 | io_run_task_work(); |
d434ab6d | 7144 | io_sq_thread_drop_mm_files(); |
b41e9852 | 7145 | |
91d8f519 DZ |
7146 | if (cur_css) |
7147 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
7148 | if (old_cred) |
7149 | revert_creds(old_cred); | |
06058632 | 7150 | |
28cea78a JA |
7151 | task_lock(current); |
7152 | current->files = old_files; | |
7153 | current->nsproxy = old_nsproxy; | |
7154 | task_unlock(current); | |
7155 | ||
2bbcd6d3 | 7156 | kthread_parkme(); |
06058632 | 7157 | |
6c271ce2 JA |
7158 | return 0; |
7159 | } | |
7160 | ||
bda52162 JA |
7161 | struct io_wait_queue { |
7162 | struct wait_queue_entry wq; | |
7163 | struct io_ring_ctx *ctx; | |
7164 | unsigned to_wait; | |
7165 | unsigned nr_timeouts; | |
7166 | }; | |
7167 | ||
6c503150 | 7168 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7169 | { |
7170 | struct io_ring_ctx *ctx = iowq->ctx; | |
7171 | ||
7172 | /* | |
d195a66e | 7173 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7174 | * started waiting. For timeouts, we always want to return to userspace, |
7175 | * regardless of event count. | |
7176 | */ | |
6c503150 | 7177 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
7178 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
7179 | } | |
7180 | ||
7181 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7182 | int wake_flags, void *key) | |
7183 | { | |
7184 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7185 | wq); | |
7186 | ||
6c503150 PB |
7187 | /* |
7188 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7189 | * the task, and the next invocation will do it. | |
7190 | */ | |
7191 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
7192 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
7193 | return -1; | |
bda52162 JA |
7194 | } |
7195 | ||
af9c1a44 JA |
7196 | static int io_run_task_work_sig(void) |
7197 | { | |
7198 | if (io_run_task_work()) | |
7199 | return 1; | |
7200 | if (!signal_pending(current)) | |
7201 | return 0; | |
792ee0f6 JA |
7202 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
7203 | return -ERESTARTSYS; | |
af9c1a44 JA |
7204 | return -EINTR; |
7205 | } | |
7206 | ||
eeb60b9a PB |
7207 | /* when returns >0, the caller should retry */ |
7208 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7209 | struct io_wait_queue *iowq, | |
7210 | signed long *timeout) | |
7211 | { | |
7212 | int ret; | |
7213 | ||
7214 | /* make sure we run task_work before checking for signals */ | |
7215 | ret = io_run_task_work_sig(); | |
7216 | if (ret || io_should_wake(iowq)) | |
7217 | return ret; | |
7218 | /* let the caller flush overflows, retry */ | |
7219 | if (test_bit(0, &ctx->cq_check_overflow)) | |
7220 | return 1; | |
7221 | ||
7222 | *timeout = schedule_timeout(*timeout); | |
7223 | return !*timeout ? -ETIME : 1; | |
7224 | } | |
7225 | ||
2b188cc1 JA |
7226 | /* |
7227 | * Wait until events become available, if we don't already have some. The | |
7228 | * application must reap them itself, as they reside on the shared cq ring. | |
7229 | */ | |
7230 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7231 | const sigset_t __user *sig, size_t sigsz, |
7232 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7233 | { |
bda52162 JA |
7234 | struct io_wait_queue iowq = { |
7235 | .wq = { | |
7236 | .private = current, | |
7237 | .func = io_wake_function, | |
7238 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7239 | }, | |
7240 | .ctx = ctx, | |
7241 | .to_wait = min_events, | |
7242 | }; | |
75b28aff | 7243 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7244 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7245 | int ret; | |
2b188cc1 | 7246 | |
b41e9852 | 7247 | do { |
6c503150 PB |
7248 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
7249 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 7250 | return 0; |
4c6e277c | 7251 | if (!io_run_task_work()) |
b41e9852 | 7252 | break; |
b41e9852 | 7253 | } while (1); |
2b188cc1 JA |
7254 | |
7255 | if (sig) { | |
9e75ad5d AB |
7256 | #ifdef CONFIG_COMPAT |
7257 | if (in_compat_syscall()) | |
7258 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7259 | sigsz); |
9e75ad5d AB |
7260 | else |
7261 | #endif | |
b772434b | 7262 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7263 | |
2b188cc1 JA |
7264 | if (ret) |
7265 | return ret; | |
7266 | } | |
7267 | ||
c73ebb68 | 7268 | if (uts) { |
c1d5a224 PB |
7269 | struct timespec64 ts; |
7270 | ||
c73ebb68 HX |
7271 | if (get_timespec64(&ts, uts)) |
7272 | return -EFAULT; | |
7273 | timeout = timespec64_to_jiffies(&ts); | |
7274 | } | |
7275 | ||
bda52162 | 7276 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7277 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7278 | do { |
6c503150 | 7279 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
7280 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7281 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7282 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7283 | finish_wait(&ctx->wait, &iowq.wq); | |
7284 | } while (ret > 0); | |
bda52162 | 7285 | |
b7db41c9 | 7286 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7287 | |
75b28aff | 7288 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7289 | } |
7290 | ||
6b06314c JA |
7291 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7292 | { | |
7293 | #if defined(CONFIG_UNIX) | |
7294 | if (ctx->ring_sock) { | |
7295 | struct sock *sock = ctx->ring_sock->sk; | |
7296 | struct sk_buff *skb; | |
7297 | ||
7298 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7299 | kfree_skb(skb); | |
7300 | } | |
7301 | #else | |
7302 | int i; | |
7303 | ||
65e19f54 JA |
7304 | for (i = 0; i < ctx->nr_user_files; i++) { |
7305 | struct file *file; | |
7306 | ||
7307 | file = io_file_from_index(ctx, i); | |
7308 | if (file) | |
7309 | fput(file); | |
7310 | } | |
6b06314c JA |
7311 | #endif |
7312 | } | |
7313 | ||
00835dce | 7314 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7315 | { |
269bbe5f | 7316 | struct fixed_rsrc_data *data; |
05f3fb3c | 7317 | |
269bbe5f | 7318 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7319 | complete(&data->done); |
7320 | } | |
7321 | ||
2a63b2d9 BM |
7322 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
7323 | { | |
7324 | spin_lock_bh(&ctx->rsrc_ref_lock); | |
7325 | } | |
7326 | ||
7327 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) | |
7328 | { | |
7329 | spin_unlock_bh(&ctx->rsrc_ref_lock); | |
7330 | } | |
7331 | ||
d67d2263 BM |
7332 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7333 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7334 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7335 | { |
2a63b2d9 | 7336 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7337 | rsrc_data->node = ref_node; |
d67d2263 | 7338 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7339 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7340 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7341 | } |
7342 | ||
d7954b2b BM |
7343 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, |
7344 | struct io_ring_ctx *ctx, | |
7345 | struct fixed_rsrc_ref_node *backup_node) | |
6b06314c | 7346 | { |
d7954b2b | 7347 | struct fixed_rsrc_ref_node *ref_node; |
1ffc5422 | 7348 | int ret; |
65e19f54 | 7349 | |
2a63b2d9 | 7350 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7351 | ref_node = data->node; |
2a63b2d9 | 7352 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7353 | if (ref_node) |
7354 | percpu_ref_kill(&ref_node->refs); | |
7355 | ||
7356 | percpu_ref_kill(&data->refs); | |
7357 | ||
7358 | /* wait for all refs nodes to complete */ | |
269bbe5f | 7359 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7360 | do { |
7361 | ret = wait_for_completion_interruptible(&data->done); | |
7362 | if (!ret) | |
7363 | break; | |
7364 | ret = io_run_task_work_sig(); | |
7365 | if (ret < 0) { | |
7366 | percpu_ref_resurrect(&data->refs); | |
7367 | reinit_completion(&data->done); | |
d67d2263 | 7368 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
1ffc5422 PB |
7369 | return ret; |
7370 | } | |
7371 | } while (1); | |
05f3fb3c | 7372 | |
d7954b2b BM |
7373 | destroy_fixed_rsrc_ref_node(backup_node); |
7374 | return 0; | |
7375 | } | |
7376 | ||
1ad555c6 BM |
7377 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7378 | { | |
7379 | struct fixed_rsrc_data *data; | |
7380 | ||
7381 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7382 | if (!data) | |
7383 | return NULL; | |
7384 | ||
00835dce | 7385 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7386 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7387 | kfree(data); | |
7388 | return NULL; | |
7389 | } | |
7390 | data->ctx = ctx; | |
7391 | init_completion(&data->done); | |
7392 | return data; | |
7393 | } | |
7394 | ||
7395 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7396 | { | |
7397 | percpu_ref_exit(&data->refs); | |
7398 | kfree(data->table); | |
7399 | kfree(data); | |
7400 | } | |
7401 | ||
d7954b2b BM |
7402 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7403 | { | |
7404 | struct fixed_rsrc_data *data = ctx->file_data; | |
7405 | struct fixed_rsrc_ref_node *backup_node; | |
7406 | unsigned nr_tables, i; | |
7407 | int ret; | |
7408 | ||
7409 | if (!data) | |
7410 | return -ENXIO; | |
7411 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7412 | if (!backup_node) | |
7413 | return -ENOMEM; | |
7414 | init_fixed_file_ref_node(ctx, backup_node); | |
7415 | ||
7416 | ret = io_rsrc_ref_quiesce(data, ctx, backup_node); | |
7417 | if (ret) | |
7418 | return ret; | |
7419 | ||
6b06314c | 7420 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7421 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7422 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7423 | kfree(data->table[i].files); |
1ad555c6 | 7424 | free_fixed_rsrc_data(data); |
05f3fb3c | 7425 | ctx->file_data = NULL; |
6b06314c JA |
7426 | ctx->nr_user_files = 0; |
7427 | return 0; | |
7428 | } | |
7429 | ||
534ca6d6 | 7430 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7431 | { |
534ca6d6 | 7432 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7433 | /* |
7434 | * The park is a bit of a work-around, without it we get | |
7435 | * warning spews on shutdown with SQPOLL set and affinity | |
7436 | * set to a single CPU. | |
7437 | */ | |
534ca6d6 JA |
7438 | if (sqd->thread) { |
7439 | kthread_park(sqd->thread); | |
7440 | kthread_stop(sqd->thread); | |
7441 | } | |
7442 | ||
7443 | kfree(sqd); | |
7444 | } | |
7445 | } | |
7446 | ||
aa06165d JA |
7447 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7448 | { | |
7449 | struct io_ring_ctx *ctx_attach; | |
7450 | struct io_sq_data *sqd; | |
7451 | struct fd f; | |
7452 | ||
7453 | f = fdget(p->wq_fd); | |
7454 | if (!f.file) | |
7455 | return ERR_PTR(-ENXIO); | |
7456 | if (f.file->f_op != &io_uring_fops) { | |
7457 | fdput(f); | |
7458 | return ERR_PTR(-EINVAL); | |
7459 | } | |
7460 | ||
7461 | ctx_attach = f.file->private_data; | |
7462 | sqd = ctx_attach->sq_data; | |
7463 | if (!sqd) { | |
7464 | fdput(f); | |
7465 | return ERR_PTR(-EINVAL); | |
7466 | } | |
7467 | ||
7468 | refcount_inc(&sqd->refs); | |
7469 | fdput(f); | |
7470 | return sqd; | |
7471 | } | |
7472 | ||
534ca6d6 JA |
7473 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7474 | { | |
7475 | struct io_sq_data *sqd; | |
7476 | ||
aa06165d JA |
7477 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7478 | return io_attach_sq_data(p); | |
7479 | ||
534ca6d6 JA |
7480 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7481 | if (!sqd) | |
7482 | return ERR_PTR(-ENOMEM); | |
7483 | ||
7484 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7485 | INIT_LIST_HEAD(&sqd->ctx_list); |
7486 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7487 | mutex_init(&sqd->ctx_lock); | |
7488 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7489 | init_waitqueue_head(&sqd->wait); |
7490 | return sqd; | |
7491 | } | |
7492 | ||
69fb2131 JA |
7493 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7494 | __releases(&sqd->lock) | |
7495 | { | |
7496 | if (!sqd->thread) | |
7497 | return; | |
7498 | kthread_unpark(sqd->thread); | |
7499 | mutex_unlock(&sqd->lock); | |
7500 | } | |
7501 | ||
7502 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7503 | __acquires(&sqd->lock) | |
7504 | { | |
7505 | if (!sqd->thread) | |
7506 | return; | |
7507 | mutex_lock(&sqd->lock); | |
7508 | kthread_park(sqd->thread); | |
7509 | } | |
7510 | ||
534ca6d6 JA |
7511 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7512 | { | |
7513 | struct io_sq_data *sqd = ctx->sq_data; | |
7514 | ||
7515 | if (sqd) { | |
7516 | if (sqd->thread) { | |
7517 | /* | |
7518 | * We may arrive here from the error branch in | |
7519 | * io_sq_offload_create() where the kthread is created | |
7520 | * without being waked up, thus wake it up now to make | |
7521 | * sure the wait will complete. | |
7522 | */ | |
7523 | wake_up_process(sqd->thread); | |
7524 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7525 | |
7526 | io_sq_thread_park(sqd); | |
7527 | } | |
7528 | ||
7529 | mutex_lock(&sqd->ctx_lock); | |
7530 | list_del(&ctx->sqd_list); | |
08369246 | 7531 | io_sqd_update_thread_idle(sqd); |
69fb2131 JA |
7532 | mutex_unlock(&sqd->ctx_lock); |
7533 | ||
08369246 | 7534 | if (sqd->thread) |
69fb2131 | 7535 | io_sq_thread_unpark(sqd); |
534ca6d6 JA |
7536 | |
7537 | io_put_sq_data(sqd); | |
7538 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7539 | } |
7540 | } | |
7541 | ||
6b06314c JA |
7542 | static void io_finish_async(struct io_ring_ctx *ctx) |
7543 | { | |
6c271ce2 JA |
7544 | io_sq_thread_stop(ctx); |
7545 | ||
561fb04a JA |
7546 | if (ctx->io_wq) { |
7547 | io_wq_destroy(ctx->io_wq); | |
7548 | ctx->io_wq = NULL; | |
6b06314c JA |
7549 | } |
7550 | } | |
7551 | ||
7552 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7553 | /* |
7554 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7555 | * the io_uring can be safely unregistered on process exit, even if we have | |
7556 | * loops in the file referencing. | |
7557 | */ | |
7558 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7559 | { | |
7560 | struct sock *sk = ctx->ring_sock->sk; | |
7561 | struct scm_fp_list *fpl; | |
7562 | struct sk_buff *skb; | |
08a45173 | 7563 | int i, nr_files; |
6b06314c | 7564 | |
6b06314c JA |
7565 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7566 | if (!fpl) | |
7567 | return -ENOMEM; | |
7568 | ||
7569 | skb = alloc_skb(0, GFP_KERNEL); | |
7570 | if (!skb) { | |
7571 | kfree(fpl); | |
7572 | return -ENOMEM; | |
7573 | } | |
7574 | ||
7575 | skb->sk = sk; | |
6b06314c | 7576 | |
08a45173 | 7577 | nr_files = 0; |
6b06314c JA |
7578 | fpl->user = get_uid(ctx->user); |
7579 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7580 | struct file *file = io_file_from_index(ctx, i + offset); |
7581 | ||
7582 | if (!file) | |
08a45173 | 7583 | continue; |
65e19f54 | 7584 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7585 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7586 | nr_files++; | |
6b06314c JA |
7587 | } |
7588 | ||
08a45173 JA |
7589 | if (nr_files) { |
7590 | fpl->max = SCM_MAX_FD; | |
7591 | fpl->count = nr_files; | |
7592 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7593 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7594 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7595 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7596 | |
08a45173 JA |
7597 | for (i = 0; i < nr_files; i++) |
7598 | fput(fpl->fp[i]); | |
7599 | } else { | |
7600 | kfree_skb(skb); | |
7601 | kfree(fpl); | |
7602 | } | |
6b06314c JA |
7603 | |
7604 | return 0; | |
7605 | } | |
7606 | ||
7607 | /* | |
7608 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7609 | * causes regular reference counting to break down. We rely on the UNIX | |
7610 | * garbage collection to take care of this problem for us. | |
7611 | */ | |
7612 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7613 | { | |
7614 | unsigned left, total; | |
7615 | int ret = 0; | |
7616 | ||
7617 | total = 0; | |
7618 | left = ctx->nr_user_files; | |
7619 | while (left) { | |
7620 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7621 | |
7622 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7623 | if (ret) | |
7624 | break; | |
7625 | left -= this_files; | |
7626 | total += this_files; | |
7627 | } | |
7628 | ||
7629 | if (!ret) | |
7630 | return 0; | |
7631 | ||
7632 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7633 | struct file *file = io_file_from_index(ctx, total); |
7634 | ||
7635 | if (file) | |
7636 | fput(file); | |
6b06314c JA |
7637 | total++; |
7638 | } | |
7639 | ||
7640 | return ret; | |
7641 | } | |
7642 | #else | |
7643 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7644 | { | |
7645 | return 0; | |
7646 | } | |
7647 | #endif | |
7648 | ||
269bbe5f | 7649 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7650 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7651 | { |
7652 | int i; | |
7653 | ||
7654 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7655 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7656 | unsigned this_files; |
7657 | ||
7658 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7659 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7660 | GFP_KERNEL); | |
7661 | if (!table->files) | |
7662 | break; | |
7663 | nr_files -= this_files; | |
7664 | } | |
7665 | ||
7666 | if (i == nr_tables) | |
7667 | return 0; | |
7668 | ||
7669 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7670 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7671 | kfree(table->files); |
7672 | } | |
7673 | return 1; | |
7674 | } | |
7675 | ||
50238531 | 7676 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7677 | { |
50238531 | 7678 | struct file *file = prsrc->file; |
05f3fb3c JA |
7679 | #if defined(CONFIG_UNIX) |
7680 | struct sock *sock = ctx->ring_sock->sk; | |
7681 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7682 | struct sk_buff *skb; | |
7683 | int i; | |
7684 | ||
7685 | __skb_queue_head_init(&list); | |
7686 | ||
7687 | /* | |
7688 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7689 | * remove this entry and rearrange the file array. | |
7690 | */ | |
7691 | skb = skb_dequeue(head); | |
7692 | while (skb) { | |
7693 | struct scm_fp_list *fp; | |
7694 | ||
7695 | fp = UNIXCB(skb).fp; | |
7696 | for (i = 0; i < fp->count; i++) { | |
7697 | int left; | |
7698 | ||
7699 | if (fp->fp[i] != file) | |
7700 | continue; | |
7701 | ||
7702 | unix_notinflight(fp->user, fp->fp[i]); | |
7703 | left = fp->count - 1 - i; | |
7704 | if (left) { | |
7705 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7706 | left * sizeof(struct file *)); | |
7707 | } | |
7708 | fp->count--; | |
7709 | if (!fp->count) { | |
7710 | kfree_skb(skb); | |
7711 | skb = NULL; | |
7712 | } else { | |
7713 | __skb_queue_tail(&list, skb); | |
7714 | } | |
7715 | fput(file); | |
7716 | file = NULL; | |
7717 | break; | |
7718 | } | |
7719 | ||
7720 | if (!file) | |
7721 | break; | |
7722 | ||
7723 | __skb_queue_tail(&list, skb); | |
7724 | ||
7725 | skb = skb_dequeue(head); | |
7726 | } | |
7727 | ||
7728 | if (skb_peek(&list)) { | |
7729 | spin_lock_irq(&head->lock); | |
7730 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7731 | __skb_queue_tail(head, skb); | |
7732 | spin_unlock_irq(&head->lock); | |
7733 | } | |
7734 | #else | |
7735 | fput(file); | |
7736 | #endif | |
7737 | } | |
7738 | ||
269bbe5f | 7739 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7740 | { |
269bbe5f BM |
7741 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7742 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7743 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7744 | |
269bbe5f BM |
7745 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7746 | list_del(&prsrc->list); | |
50238531 | 7747 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7748 | kfree(prsrc); |
65e19f54 | 7749 | } |
05589553 | 7750 | |
05589553 XW |
7751 | percpu_ref_exit(&ref_node->refs); |
7752 | kfree(ref_node); | |
269bbe5f | 7753 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7754 | } |
65e19f54 | 7755 | |
269bbe5f | 7756 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7757 | { |
7758 | struct io_ring_ctx *ctx; | |
7759 | struct llist_node *node; | |
7760 | ||
269bbe5f BM |
7761 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7762 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7763 | |
7764 | while (node) { | |
269bbe5f | 7765 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7766 | struct llist_node *next = node->next; |
7767 | ||
269bbe5f BM |
7768 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7769 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7770 | node = next; |
7771 | } | |
7772 | } | |
7773 | ||
ea64ec02 PB |
7774 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7775 | unsigned i) | |
7776 | { | |
7777 | struct fixed_rsrc_table *table; | |
7778 | ||
7779 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7780 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7781 | } | |
7782 | ||
00835dce | 7783 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7784 | { |
269bbe5f BM |
7785 | struct fixed_rsrc_ref_node *ref_node; |
7786 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7787 | struct io_ring_ctx *ctx; |
e297822b | 7788 | bool first_add = false; |
4a38aed2 | 7789 | int delay = HZ; |
65e19f54 | 7790 | |
269bbe5f BM |
7791 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7792 | data = ref_node->rsrc_data; | |
e297822b PB |
7793 | ctx = data->ctx; |
7794 | ||
2a63b2d9 | 7795 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7796 | ref_node->done = true; |
7797 | ||
d67d2263 BM |
7798 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7799 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7800 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7801 | /* recycle ref nodes in order */ |
7802 | if (!ref_node->done) | |
7803 | break; | |
7804 | list_del(&ref_node->node); | |
269bbe5f | 7805 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7806 | } |
2a63b2d9 | 7807 | io_rsrc_ref_unlock(ctx); |
05589553 | 7808 | |
e297822b | 7809 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7810 | delay = 0; |
05589553 | 7811 | |
4a38aed2 | 7812 | if (!delay) |
269bbe5f | 7813 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7814 | else if (first_add) |
269bbe5f | 7815 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7816 | } |
65e19f54 | 7817 | |
6802535d | 7818 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7819 | struct io_ring_ctx *ctx) |
05f3fb3c | 7820 | { |
269bbe5f | 7821 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7822 | |
05589553 XW |
7823 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7824 | if (!ref_node) | |
3e2224c5 | 7825 | return NULL; |
05f3fb3c | 7826 | |
00835dce | 7827 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7828 | 0, GFP_KERNEL)) { |
7829 | kfree(ref_node); | |
3e2224c5 | 7830 | return NULL; |
05589553 XW |
7831 | } |
7832 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7833 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
6802535d BM |
7834 | ref_node->done = false; |
7835 | return ref_node; | |
7836 | } | |
7837 | ||
bc9744cd PB |
7838 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7839 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7840 | { |
269bbe5f | 7841 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7842 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7843 | } |
7844 | ||
269bbe5f | 7845 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7846 | { |
7847 | percpu_ref_exit(&ref_node->refs); | |
7848 | kfree(ref_node); | |
65e19f54 JA |
7849 | } |
7850 | ||
ea64ec02 | 7851 | |
6b06314c JA |
7852 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7853 | unsigned nr_args) | |
7854 | { | |
7855 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7856 | unsigned nr_tables, i; |
05f3fb3c | 7857 | struct file *file; |
600cf3f8 | 7858 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7859 | struct fixed_rsrc_ref_node *ref_node; |
7860 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7861 | |
05f3fb3c | 7862 | if (ctx->file_data) |
6b06314c JA |
7863 | return -EBUSY; |
7864 | if (!nr_args) | |
7865 | return -EINVAL; | |
7866 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7867 | return -EMFILE; | |
7868 | ||
1ad555c6 | 7869 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7870 | if (!file_data) |
05f3fb3c | 7871 | return -ENOMEM; |
13770a71 | 7872 | ctx->file_data = file_data; |
05f3fb3c | 7873 | |
65e19f54 | 7874 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7875 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7876 | GFP_KERNEL); |
600cf3f8 PB |
7877 | if (!file_data->table) |
7878 | goto out_free; | |
05f3fb3c | 7879 | |
600cf3f8 | 7880 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7881 | goto out_free; |
65e19f54 | 7882 | |
08a45173 | 7883 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7884 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7885 | ret = -EFAULT; | |
7886 | goto out_fput; | |
7887 | } | |
08a45173 | 7888 | /* allow sparse sets */ |
600cf3f8 | 7889 | if (fd == -1) |
08a45173 | 7890 | continue; |
6b06314c | 7891 | |
05f3fb3c | 7892 | file = fget(fd); |
6b06314c | 7893 | ret = -EBADF; |
05f3fb3c | 7894 | if (!file) |
600cf3f8 | 7895 | goto out_fput; |
05f3fb3c | 7896 | |
6b06314c JA |
7897 | /* |
7898 | * Don't allow io_uring instances to be registered. If UNIX | |
7899 | * isn't enabled, then this causes a reference cycle and this | |
7900 | * instance can never get freed. If UNIX is enabled we'll | |
7901 | * handle it just fine, but there's still no point in allowing | |
7902 | * a ring fd as it doesn't support regular read/write anyway. | |
7903 | */ | |
05f3fb3c JA |
7904 | if (file->f_op == &io_uring_fops) { |
7905 | fput(file); | |
600cf3f8 | 7906 | goto out_fput; |
6b06314c | 7907 | } |
ea64ec02 | 7908 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7909 | } |
7910 | ||
6b06314c | 7911 | ret = io_sqe_files_scm(ctx); |
05589553 | 7912 | if (ret) { |
6b06314c | 7913 | io_sqe_files_unregister(ctx); |
05589553 XW |
7914 | return ret; |
7915 | } | |
6b06314c | 7916 | |
bc9744cd | 7917 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7918 | if (!ref_node) { |
05589553 | 7919 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7920 | return -ENOMEM; |
05589553 | 7921 | } |
bc9744cd | 7922 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7923 | |
d67d2263 | 7924 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7925 | return ret; |
600cf3f8 PB |
7926 | out_fput: |
7927 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7928 | file = io_file_from_index(ctx, i); | |
7929 | if (file) | |
7930 | fput(file); | |
7931 | } | |
7932 | for (i = 0; i < nr_tables; i++) | |
7933 | kfree(file_data->table[i].files); | |
7934 | ctx->nr_user_files = 0; | |
600cf3f8 | 7935 | out_free: |
1ad555c6 | 7936 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7937 | ctx->file_data = NULL; |
6b06314c JA |
7938 | return ret; |
7939 | } | |
7940 | ||
c3a31e60 JA |
7941 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7942 | int index) | |
7943 | { | |
7944 | #if defined(CONFIG_UNIX) | |
7945 | struct sock *sock = ctx->ring_sock->sk; | |
7946 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7947 | struct sk_buff *skb; | |
7948 | ||
7949 | /* | |
7950 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7951 | * file set. If there's no room, fall back to allocating a new skb | |
7952 | * and filling it in. | |
7953 | */ | |
7954 | spin_lock_irq(&head->lock); | |
7955 | skb = skb_peek(head); | |
7956 | if (skb) { | |
7957 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7958 | ||
7959 | if (fpl->count < SCM_MAX_FD) { | |
7960 | __skb_unlink(skb, head); | |
7961 | spin_unlock_irq(&head->lock); | |
7962 | fpl->fp[fpl->count] = get_file(file); | |
7963 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7964 | fpl->count++; | |
7965 | spin_lock_irq(&head->lock); | |
7966 | __skb_queue_head(head, skb); | |
7967 | } else { | |
7968 | skb = NULL; | |
7969 | } | |
7970 | } | |
7971 | spin_unlock_irq(&head->lock); | |
7972 | ||
7973 | if (skb) { | |
7974 | fput(file); | |
7975 | return 0; | |
7976 | } | |
7977 | ||
7978 | return __io_sqe_files_scm(ctx, 1, index); | |
7979 | #else | |
7980 | return 0; | |
7981 | #endif | |
7982 | } | |
7983 | ||
50238531 | 7984 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7985 | { |
269bbe5f BM |
7986 | struct io_rsrc_put *prsrc; |
7987 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7988 | |
269bbe5f BM |
7989 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7990 | if (!prsrc) | |
a5318d3c | 7991 | return -ENOMEM; |
05f3fb3c | 7992 | |
50238531 | 7993 | prsrc->rsrc = rsrc; |
269bbe5f | 7994 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7995 | |
a5318d3c | 7996 | return 0; |
05f3fb3c JA |
7997 | } |
7998 | ||
269bbe5f BM |
7999 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
8000 | struct file *file) | |
8001 | { | |
50238531 | 8002 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
8003 | } |
8004 | ||
05f3fb3c | 8005 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 8006 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
8007 | unsigned nr_args) |
8008 | { | |
269bbe5f BM |
8009 | struct fixed_rsrc_data *data = ctx->file_data; |
8010 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 8011 | struct file *file, **file_slot; |
c3a31e60 JA |
8012 | __s32 __user *fds; |
8013 | int fd, i, err; | |
8014 | __u32 done; | |
05589553 | 8015 | bool needs_switch = false; |
c3a31e60 | 8016 | |
05f3fb3c | 8017 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
8018 | return -EOVERFLOW; |
8019 | if (done > ctx->nr_user_files) | |
8020 | return -EINVAL; | |
8021 | ||
bc9744cd | 8022 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
8023 | if (!ref_node) |
8024 | return -ENOMEM; | |
bc9744cd | 8025 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 8026 | |
269bbe5f | 8027 | fds = u64_to_user_ptr(up->data); |
67973b93 | 8028 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
8029 | err = 0; |
8030 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
8031 | err = -EFAULT; | |
8032 | break; | |
8033 | } | |
4e0377a1 | 8034 | if (fd == IORING_REGISTER_FILES_SKIP) |
8035 | continue; | |
8036 | ||
67973b93 | 8037 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
8038 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
8039 | ||
8040 | if (*file_slot) { | |
8041 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
8042 | if (err) |
8043 | break; | |
ea64ec02 | 8044 | *file_slot = NULL; |
05589553 | 8045 | needs_switch = true; |
c3a31e60 JA |
8046 | } |
8047 | if (fd != -1) { | |
c3a31e60 JA |
8048 | file = fget(fd); |
8049 | if (!file) { | |
8050 | err = -EBADF; | |
8051 | break; | |
8052 | } | |
8053 | /* | |
8054 | * Don't allow io_uring instances to be registered. If | |
8055 | * UNIX isn't enabled, then this causes a reference | |
8056 | * cycle and this instance can never get freed. If UNIX | |
8057 | * is enabled we'll handle it just fine, but there's | |
8058 | * still no point in allowing a ring fd as it doesn't | |
8059 | * support regular read/write anyway. | |
8060 | */ | |
8061 | if (file->f_op == &io_uring_fops) { | |
8062 | fput(file); | |
8063 | err = -EBADF; | |
8064 | break; | |
8065 | } | |
e68a3ff8 | 8066 | *file_slot = file; |
c3a31e60 | 8067 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8068 | if (err) { |
e68a3ff8 | 8069 | *file_slot = NULL; |
f3bd9dae | 8070 | fput(file); |
c3a31e60 | 8071 | break; |
f3bd9dae | 8072 | } |
c3a31e60 | 8073 | } |
05f3fb3c JA |
8074 | } |
8075 | ||
05589553 | 8076 | if (needs_switch) { |
b2e96852 | 8077 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 8078 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 8079 | } else |
269bbe5f | 8080 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
8081 | |
8082 | return done ? done : err; | |
8083 | } | |
05589553 | 8084 | |
05f3fb3c JA |
8085 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
8086 | unsigned nr_args) | |
8087 | { | |
269bbe5f | 8088 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
8089 | |
8090 | if (!ctx->file_data) | |
8091 | return -ENXIO; | |
8092 | if (!nr_args) | |
8093 | return -EINVAL; | |
8094 | if (copy_from_user(&up, arg, sizeof(up))) | |
8095 | return -EFAULT; | |
8096 | if (up.resv) | |
8097 | return -EINVAL; | |
8098 | ||
8099 | return __io_sqe_files_update(ctx, &up, nr_args); | |
8100 | } | |
c3a31e60 | 8101 | |
5280f7e5 | 8102 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
8103 | { |
8104 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8105 | ||
5280f7e5 PB |
8106 | req = io_put_req_find_next(req); |
8107 | return req ? &req->work : NULL; | |
7d723065 JA |
8108 | } |
8109 | ||
24369c2e PB |
8110 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
8111 | struct io_uring_params *p) | |
8112 | { | |
8113 | struct io_wq_data data; | |
8114 | struct fd f; | |
8115 | struct io_ring_ctx *ctx_attach; | |
8116 | unsigned int concurrency; | |
8117 | int ret = 0; | |
8118 | ||
8119 | data.user = ctx->user; | |
e9fd9396 | 8120 | data.free_work = io_free_work; |
f5fa38c5 | 8121 | data.do_work = io_wq_submit_work; |
24369c2e PB |
8122 | |
8123 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
8124 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
8125 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
8126 | ||
8127 | ctx->io_wq = io_wq_create(concurrency, &data); | |
8128 | if (IS_ERR(ctx->io_wq)) { | |
8129 | ret = PTR_ERR(ctx->io_wq); | |
8130 | ctx->io_wq = NULL; | |
8131 | } | |
8132 | return ret; | |
8133 | } | |
8134 | ||
8135 | f = fdget(p->wq_fd); | |
8136 | if (!f.file) | |
8137 | return -EBADF; | |
8138 | ||
8139 | if (f.file->f_op != &io_uring_fops) { | |
8140 | ret = -EINVAL; | |
8141 | goto out_fput; | |
8142 | } | |
8143 | ||
8144 | ctx_attach = f.file->private_data; | |
8145 | /* @io_wq is protected by holding the fd */ | |
8146 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
8147 | ret = -EINVAL; | |
8148 | goto out_fput; | |
8149 | } | |
8150 | ||
8151 | ctx->io_wq = ctx_attach->io_wq; | |
8152 | out_fput: | |
8153 | fdput(f); | |
8154 | return ret; | |
8155 | } | |
8156 | ||
0f212204 JA |
8157 | static int io_uring_alloc_task_context(struct task_struct *task) |
8158 | { | |
8159 | struct io_uring_task *tctx; | |
d8a6df10 | 8160 | int ret; |
0f212204 JA |
8161 | |
8162 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
8163 | if (unlikely(!tctx)) | |
8164 | return -ENOMEM; | |
8165 | ||
d8a6df10 JA |
8166 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8167 | if (unlikely(ret)) { | |
8168 | kfree(tctx); | |
8169 | return ret; | |
8170 | } | |
8171 | ||
0f212204 JA |
8172 | xa_init(&tctx->xa); |
8173 | init_waitqueue_head(&tctx->wait); | |
8174 | tctx->last = NULL; | |
fdaf083c JA |
8175 | atomic_set(&tctx->in_idle, 0); |
8176 | tctx->sqpoll = false; | |
500a373d JA |
8177 | io_init_identity(&tctx->__identity); |
8178 | tctx->identity = &tctx->__identity; | |
0f212204 | 8179 | task->io_uring = tctx; |
7cbf1722 JA |
8180 | spin_lock_init(&tctx->task_lock); |
8181 | INIT_WQ_LIST(&tctx->task_list); | |
8182 | tctx->task_state = 0; | |
8183 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
8184 | return 0; |
8185 | } | |
8186 | ||
8187 | void __io_uring_free(struct task_struct *tsk) | |
8188 | { | |
8189 | struct io_uring_task *tctx = tsk->io_uring; | |
8190 | ||
8191 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
8192 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
8193 | if (tctx->identity != &tctx->__identity) | |
8194 | kfree(tctx->identity); | |
d8a6df10 | 8195 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8196 | kfree(tctx); |
8197 | tsk->io_uring = NULL; | |
8198 | } | |
8199 | ||
7e84e1c7 SG |
8200 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8201 | struct io_uring_params *p) | |
2b188cc1 JA |
8202 | { |
8203 | int ret; | |
8204 | ||
6c271ce2 | 8205 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
8206 | struct io_sq_data *sqd; |
8207 | ||
3ec482d1 | 8208 | ret = -EPERM; |
ce59fc69 | 8209 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
8210 | goto err; |
8211 | ||
534ca6d6 JA |
8212 | sqd = io_get_sq_data(p); |
8213 | if (IS_ERR(sqd)) { | |
8214 | ret = PTR_ERR(sqd); | |
8215 | goto err; | |
8216 | } | |
69fb2131 | 8217 | |
534ca6d6 | 8218 | ctx->sq_data = sqd; |
69fb2131 JA |
8219 | io_sq_thread_park(sqd); |
8220 | mutex_lock(&sqd->ctx_lock); | |
8221 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
8222 | mutex_unlock(&sqd->ctx_lock); | |
8223 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 8224 | |
917257da JA |
8225 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8226 | if (!ctx->sq_thread_idle) | |
8227 | ctx->sq_thread_idle = HZ; | |
8228 | ||
aa06165d JA |
8229 | if (sqd->thread) |
8230 | goto done; | |
8231 | ||
6c271ce2 | 8232 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8233 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8234 | |
917257da | 8235 | ret = -EINVAL; |
44a9bd18 JA |
8236 | if (cpu >= nr_cpu_ids) |
8237 | goto err; | |
7889f44d | 8238 | if (!cpu_online(cpu)) |
917257da JA |
8239 | goto err; |
8240 | ||
69fb2131 | 8241 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 8242 | cpu, "io_uring-sq"); |
6c271ce2 | 8243 | } else { |
69fb2131 | 8244 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
8245 | "io_uring-sq"); |
8246 | } | |
534ca6d6 JA |
8247 | if (IS_ERR(sqd->thread)) { |
8248 | ret = PTR_ERR(sqd->thread); | |
8249 | sqd->thread = NULL; | |
6c271ce2 JA |
8250 | goto err; |
8251 | } | |
534ca6d6 | 8252 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
8253 | if (ret) |
8254 | goto err; | |
6c271ce2 JA |
8255 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8256 | /* Can't have SQ_AFF without SQPOLL */ | |
8257 | ret = -EINVAL; | |
8258 | goto err; | |
8259 | } | |
8260 | ||
aa06165d | 8261 | done: |
24369c2e PB |
8262 | ret = io_init_wq_offload(ctx, p); |
8263 | if (ret) | |
2b188cc1 | 8264 | goto err; |
2b188cc1 JA |
8265 | |
8266 | return 0; | |
8267 | err: | |
54a91f3b | 8268 | io_finish_async(ctx); |
2b188cc1 JA |
8269 | return ret; |
8270 | } | |
8271 | ||
7e84e1c7 SG |
8272 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
8273 | { | |
534ca6d6 JA |
8274 | struct io_sq_data *sqd = ctx->sq_data; |
8275 | ||
8276 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
8277 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
8278 | } |
8279 | ||
a087e2b5 BM |
8280 | static inline void __io_unaccount_mem(struct user_struct *user, |
8281 | unsigned long nr_pages) | |
2b188cc1 JA |
8282 | { |
8283 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8284 | } | |
8285 | ||
a087e2b5 BM |
8286 | static inline int __io_account_mem(struct user_struct *user, |
8287 | unsigned long nr_pages) | |
2b188cc1 JA |
8288 | { |
8289 | unsigned long page_limit, cur_pages, new_pages; | |
8290 | ||
8291 | /* Don't allow more pages than we can safely lock */ | |
8292 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8293 | ||
8294 | do { | |
8295 | cur_pages = atomic_long_read(&user->locked_vm); | |
8296 | new_pages = cur_pages + nr_pages; | |
8297 | if (new_pages > page_limit) | |
8298 | return -ENOMEM; | |
8299 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8300 | new_pages) != cur_pages); | |
8301 | ||
8302 | return 0; | |
8303 | } | |
8304 | ||
26bfa89e | 8305 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8306 | { |
aad5d8da | 8307 | if (ctx->limit_mem) |
a087e2b5 | 8308 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8309 | |
26bfa89e JA |
8310 | if (ctx->mm_account) |
8311 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8312 | } |
8313 | ||
26bfa89e | 8314 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8315 | { |
30975825 BM |
8316 | int ret; |
8317 | ||
8318 | if (ctx->limit_mem) { | |
8319 | ret = __io_account_mem(ctx->user, nr_pages); | |
8320 | if (ret) | |
8321 | return ret; | |
8322 | } | |
8323 | ||
26bfa89e JA |
8324 | if (ctx->mm_account) |
8325 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8326 | |
8327 | return 0; | |
8328 | } | |
8329 | ||
2b188cc1 JA |
8330 | static void io_mem_free(void *ptr) |
8331 | { | |
52e04ef4 MR |
8332 | struct page *page; |
8333 | ||
8334 | if (!ptr) | |
8335 | return; | |
2b188cc1 | 8336 | |
52e04ef4 | 8337 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8338 | if (put_page_testzero(page)) |
8339 | free_compound_page(page); | |
8340 | } | |
8341 | ||
8342 | static void *io_mem_alloc(size_t size) | |
8343 | { | |
8344 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8345 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8346 | |
8347 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8348 | } | |
8349 | ||
75b28aff HV |
8350 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8351 | size_t *sq_offset) | |
8352 | { | |
8353 | struct io_rings *rings; | |
8354 | size_t off, sq_array_size; | |
8355 | ||
8356 | off = struct_size(rings, cqes, cq_entries); | |
8357 | if (off == SIZE_MAX) | |
8358 | return SIZE_MAX; | |
8359 | ||
8360 | #ifdef CONFIG_SMP | |
8361 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8362 | if (off == 0) | |
8363 | return SIZE_MAX; | |
8364 | #endif | |
8365 | ||
b36200f5 DV |
8366 | if (sq_offset) |
8367 | *sq_offset = off; | |
8368 | ||
75b28aff HV |
8369 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8370 | if (sq_array_size == SIZE_MAX) | |
8371 | return SIZE_MAX; | |
8372 | ||
8373 | if (check_add_overflow(off, sq_array_size, &off)) | |
8374 | return SIZE_MAX; | |
8375 | ||
75b28aff HV |
8376 | return off; |
8377 | } | |
8378 | ||
0a96bbe4 | 8379 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8380 | { |
8381 | int i, j; | |
8382 | ||
8383 | if (!ctx->user_bufs) | |
8384 | return -ENXIO; | |
8385 | ||
8386 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8387 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8388 | ||
8389 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8390 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8391 | |
de293938 | 8392 | if (imu->acct_pages) |
26bfa89e | 8393 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8394 | kvfree(imu->bvec); |
edafccee JA |
8395 | imu->nr_bvecs = 0; |
8396 | } | |
8397 | ||
8398 | kfree(ctx->user_bufs); | |
8399 | ctx->user_bufs = NULL; | |
8400 | ctx->nr_user_bufs = 0; | |
8401 | return 0; | |
8402 | } | |
8403 | ||
8404 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8405 | void __user *arg, unsigned index) | |
8406 | { | |
8407 | struct iovec __user *src; | |
8408 | ||
8409 | #ifdef CONFIG_COMPAT | |
8410 | if (ctx->compat) { | |
8411 | struct compat_iovec __user *ciovs; | |
8412 | struct compat_iovec ciov; | |
8413 | ||
8414 | ciovs = (struct compat_iovec __user *) arg; | |
8415 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8416 | return -EFAULT; | |
8417 | ||
d55e5f5b | 8418 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8419 | dst->iov_len = ciov.iov_len; |
8420 | return 0; | |
8421 | } | |
8422 | #endif | |
8423 | src = (struct iovec __user *) arg; | |
8424 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8425 | return -EFAULT; | |
8426 | return 0; | |
8427 | } | |
8428 | ||
de293938 JA |
8429 | /* |
8430 | * Not super efficient, but this is just a registration time. And we do cache | |
8431 | * the last compound head, so generally we'll only do a full search if we don't | |
8432 | * match that one. | |
8433 | * | |
8434 | * We check if the given compound head page has already been accounted, to | |
8435 | * avoid double accounting it. This allows us to account the full size of the | |
8436 | * page, not just the constituent pages of a huge page. | |
8437 | */ | |
8438 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8439 | int nr_pages, struct page *hpage) | |
8440 | { | |
8441 | int i, j; | |
8442 | ||
8443 | /* check current page array */ | |
8444 | for (i = 0; i < nr_pages; i++) { | |
8445 | if (!PageCompound(pages[i])) | |
8446 | continue; | |
8447 | if (compound_head(pages[i]) == hpage) | |
8448 | return true; | |
8449 | } | |
8450 | ||
8451 | /* check previously registered pages */ | |
8452 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8453 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8454 | ||
8455 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8456 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8457 | continue; | |
8458 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8459 | return true; | |
8460 | } | |
8461 | } | |
8462 | ||
8463 | return false; | |
8464 | } | |
8465 | ||
8466 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8467 | int nr_pages, struct io_mapped_ubuf *imu, | |
8468 | struct page **last_hpage) | |
8469 | { | |
8470 | int i, ret; | |
8471 | ||
8472 | for (i = 0; i < nr_pages; i++) { | |
8473 | if (!PageCompound(pages[i])) { | |
8474 | imu->acct_pages++; | |
8475 | } else { | |
8476 | struct page *hpage; | |
8477 | ||
8478 | hpage = compound_head(pages[i]); | |
8479 | if (hpage == *last_hpage) | |
8480 | continue; | |
8481 | *last_hpage = hpage; | |
8482 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8483 | continue; | |
8484 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8485 | } | |
8486 | } | |
8487 | ||
8488 | if (!imu->acct_pages) | |
8489 | return 0; | |
8490 | ||
26bfa89e | 8491 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8492 | if (ret) |
8493 | imu->acct_pages = 0; | |
8494 | return ret; | |
8495 | } | |
8496 | ||
0a96bbe4 BM |
8497 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8498 | struct io_mapped_ubuf *imu, | |
8499 | struct page **last_hpage) | |
edafccee JA |
8500 | { |
8501 | struct vm_area_struct **vmas = NULL; | |
8502 | struct page **pages = NULL; | |
0a96bbe4 BM |
8503 | unsigned long off, start, end, ubuf; |
8504 | size_t size; | |
8505 | int ret, pret, nr_pages, i; | |
8506 | ||
8507 | ubuf = (unsigned long) iov->iov_base; | |
8508 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8509 | start = ubuf >> PAGE_SHIFT; | |
8510 | nr_pages = end - start; | |
8511 | ||
8512 | ret = -ENOMEM; | |
8513 | ||
8514 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8515 | if (!pages) | |
8516 | goto done; | |
8517 | ||
8518 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8519 | GFP_KERNEL); | |
8520 | if (!vmas) | |
8521 | goto done; | |
8522 | ||
8523 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), | |
8524 | GFP_KERNEL); | |
8525 | if (!imu->bvec) | |
8526 | goto done; | |
8527 | ||
8528 | ret = 0; | |
8529 | mmap_read_lock(current->mm); | |
8530 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8531 | pages, vmas); | |
8532 | if (pret == nr_pages) { | |
8533 | /* don't support file backed memory */ | |
8534 | for (i = 0; i < nr_pages; i++) { | |
8535 | struct vm_area_struct *vma = vmas[i]; | |
8536 | ||
8537 | if (vma->vm_file && | |
8538 | !is_file_hugepages(vma->vm_file)) { | |
8539 | ret = -EOPNOTSUPP; | |
8540 | break; | |
8541 | } | |
8542 | } | |
8543 | } else { | |
8544 | ret = pret < 0 ? pret : -EFAULT; | |
8545 | } | |
8546 | mmap_read_unlock(current->mm); | |
8547 | if (ret) { | |
8548 | /* | |
8549 | * if we did partial map, or found file backed vmas, | |
8550 | * release any pages we did get | |
8551 | */ | |
8552 | if (pret > 0) | |
8553 | unpin_user_pages(pages, pret); | |
8554 | kvfree(imu->bvec); | |
8555 | goto done; | |
8556 | } | |
8557 | ||
8558 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8559 | if (ret) { | |
8560 | unpin_user_pages(pages, pret); | |
8561 | kvfree(imu->bvec); | |
8562 | goto done; | |
8563 | } | |
8564 | ||
8565 | off = ubuf & ~PAGE_MASK; | |
8566 | size = iov->iov_len; | |
8567 | for (i = 0; i < nr_pages; i++) { | |
8568 | size_t vec_len; | |
8569 | ||
8570 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8571 | imu->bvec[i].bv_page = pages[i]; | |
8572 | imu->bvec[i].bv_len = vec_len; | |
8573 | imu->bvec[i].bv_offset = off; | |
8574 | off = 0; | |
8575 | size -= vec_len; | |
8576 | } | |
8577 | /* store original address for later verification */ | |
8578 | imu->ubuf = ubuf; | |
8579 | imu->len = iov->iov_len; | |
8580 | imu->nr_bvecs = nr_pages; | |
8581 | ret = 0; | |
8582 | done: | |
8583 | kvfree(pages); | |
8584 | kvfree(vmas); | |
8585 | return ret; | |
8586 | } | |
8587 | ||
2b358604 | 8588 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8589 | { |
edafccee JA |
8590 | if (ctx->user_bufs) |
8591 | return -EBUSY; | |
8592 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8593 | return -EINVAL; | |
8594 | ||
8595 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8596 | GFP_KERNEL); | |
8597 | if (!ctx->user_bufs) | |
8598 | return -ENOMEM; | |
8599 | ||
2b358604 BM |
8600 | return 0; |
8601 | } | |
8602 | ||
8603 | static int io_buffer_validate(struct iovec *iov) | |
8604 | { | |
8605 | /* | |
8606 | * Don't impose further limits on the size and buffer | |
8607 | * constraints here, we'll -EINVAL later when IO is | |
8608 | * submitted if they are wrong. | |
8609 | */ | |
8610 | if (!iov->iov_base || !iov->iov_len) | |
8611 | return -EFAULT; | |
8612 | ||
8613 | /* arbitrary limit, but we need something */ | |
8614 | if (iov->iov_len > SZ_1G) | |
8615 | return -EFAULT; | |
8616 | ||
8617 | return 0; | |
8618 | } | |
8619 | ||
8620 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, | |
8621 | unsigned int nr_args) | |
8622 | { | |
8623 | int i, ret; | |
8624 | struct iovec iov; | |
8625 | struct page *last_hpage = NULL; | |
8626 | ||
8627 | ret = io_buffers_map_alloc(ctx, nr_args); | |
8628 | if (ret) | |
8629 | return ret; | |
8630 | ||
edafccee JA |
8631 | for (i = 0; i < nr_args; i++) { |
8632 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee JA |
8633 | |
8634 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8635 | if (ret) | |
0a96bbe4 | 8636 | break; |
edafccee | 8637 | |
2b358604 BM |
8638 | ret = io_buffer_validate(&iov); |
8639 | if (ret) | |
0a96bbe4 | 8640 | break; |
edafccee | 8641 | |
0a96bbe4 BM |
8642 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8643 | if (ret) | |
8644 | break; | |
edafccee JA |
8645 | |
8646 | ctx->nr_user_bufs++; | |
8647 | } | |
0a96bbe4 BM |
8648 | |
8649 | if (ret) | |
8650 | io_sqe_buffers_unregister(ctx); | |
8651 | ||
edafccee JA |
8652 | return ret; |
8653 | } | |
8654 | ||
9b402849 JA |
8655 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8656 | { | |
8657 | __s32 __user *fds = arg; | |
8658 | int fd; | |
8659 | ||
8660 | if (ctx->cq_ev_fd) | |
8661 | return -EBUSY; | |
8662 | ||
8663 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8664 | return -EFAULT; | |
8665 | ||
8666 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8667 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8668 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8669 | ctx->cq_ev_fd = NULL; | |
8670 | return ret; | |
8671 | } | |
8672 | ||
8673 | return 0; | |
8674 | } | |
8675 | ||
8676 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8677 | { | |
8678 | if (ctx->cq_ev_fd) { | |
8679 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8680 | ctx->cq_ev_fd = NULL; | |
8681 | return 0; | |
8682 | } | |
8683 | ||
8684 | return -ENXIO; | |
8685 | } | |
8686 | ||
5a2e745d JA |
8687 | static int __io_destroy_buffers(int id, void *p, void *data) |
8688 | { | |
8689 | struct io_ring_ctx *ctx = data; | |
8690 | struct io_buffer *buf = p; | |
8691 | ||
067524e9 | 8692 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8693 | return 0; |
8694 | } | |
8695 | ||
8696 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8697 | { | |
8698 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8699 | idr_destroy(&ctx->io_buffer_idr); | |
8700 | } | |
8701 | ||
68e68ee6 | 8702 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8703 | { |
68e68ee6 | 8704 | struct io_kiocb *req, *nxt; |
1b4c351f | 8705 | |
68e68ee6 JA |
8706 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8707 | if (tsk && req->task != tsk) | |
8708 | continue; | |
1b4c351f JA |
8709 | list_del(&req->compl.list); |
8710 | kmem_cache_free(req_cachep, req); | |
8711 | } | |
8712 | } | |
8713 | ||
9a4fdbd8 | 8714 | static void io_req_caches_free(struct io_ring_ctx *ctx, struct task_struct *tsk) |
2b188cc1 | 8715 | { |
bf019da7 PB |
8716 | struct io_submit_state *submit_state = &ctx->submit_state; |
8717 | ||
9a4fdbd8 JA |
8718 | mutex_lock(&ctx->uring_lock); |
8719 | ||
8720 | if (submit_state->free_reqs) | |
8721 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, | |
8722 | submit_state->reqs); | |
8723 | ||
8724 | io_req_cache_free(&submit_state->comp.free_list, NULL); | |
8725 | ||
8726 | spin_lock_irq(&ctx->completion_lock); | |
8727 | io_req_cache_free(&submit_state->comp.locked_free_list, NULL); | |
8728 | spin_unlock_irq(&ctx->completion_lock); | |
8729 | ||
8730 | mutex_unlock(&ctx->uring_lock); | |
8731 | } | |
8732 | ||
8733 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) | |
8734 | { | |
04fc6c80 PB |
8735 | /* |
8736 | * Some may use context even when all refs and requests have been put, | |
8737 | * and they are free to do so while still holding uring_lock, see | |
8738 | * __io_req_task_submit(). Wait for them to finish. | |
8739 | */ | |
8740 | mutex_lock(&ctx->uring_lock); | |
8741 | mutex_unlock(&ctx->uring_lock); | |
8742 | ||
6b06314c | 8743 | io_finish_async(ctx); |
0a96bbe4 | 8744 | io_sqe_buffers_unregister(ctx); |
2aede0e4 JA |
8745 | |
8746 | if (ctx->sqo_task) { | |
8747 | put_task_struct(ctx->sqo_task); | |
8748 | ctx->sqo_task = NULL; | |
8749 | mmdrop(ctx->mm_account); | |
8750 | ctx->mm_account = NULL; | |
30975825 | 8751 | } |
def596e9 | 8752 | |
91d8f519 DZ |
8753 | #ifdef CONFIG_BLK_CGROUP |
8754 | if (ctx->sqo_blkcg_css) | |
8755 | css_put(ctx->sqo_blkcg_css); | |
8756 | #endif | |
8757 | ||
6b06314c | 8758 | io_sqe_files_unregister(ctx); |
9b402849 | 8759 | io_eventfd_unregister(ctx); |
5a2e745d | 8760 | io_destroy_buffers(ctx); |
41726c9a | 8761 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8762 | |
2b188cc1 | 8763 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8764 | if (ctx->ring_sock) { |
8765 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8766 | sock_release(ctx->ring_sock); |
355e8d26 | 8767 | } |
2b188cc1 JA |
8768 | #endif |
8769 | ||
75b28aff | 8770 | io_mem_free(ctx->rings); |
2b188cc1 | 8771 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8772 | |
8773 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8774 | free_uid(ctx->user); |
181e448d | 8775 | put_cred(ctx->creds); |
9a4fdbd8 | 8776 | io_req_caches_free(ctx, NULL); |
78076bb6 | 8777 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8778 | kfree(ctx); |
8779 | } | |
8780 | ||
8781 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8782 | { | |
8783 | struct io_ring_ctx *ctx = file->private_data; | |
8784 | __poll_t mask = 0; | |
8785 | ||
8786 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8787 | /* |
8788 | * synchronizes with barrier from wq_has_sleeper call in | |
8789 | * io_commit_cqring | |
8790 | */ | |
2b188cc1 | 8791 | smp_rmb(); |
90554200 | 8792 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8793 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8794 | |
8795 | /* | |
8796 | * Don't flush cqring overflow list here, just do a simple check. | |
8797 | * Otherwise there could possible be ABBA deadlock: | |
8798 | * CPU0 CPU1 | |
8799 | * ---- ---- | |
8800 | * lock(&ctx->uring_lock); | |
8801 | * lock(&ep->mtx); | |
8802 | * lock(&ctx->uring_lock); | |
8803 | * lock(&ep->mtx); | |
8804 | * | |
8805 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8806 | * pushs them to do the flush. | |
8807 | */ | |
8808 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8809 | mask |= EPOLLIN | EPOLLRDNORM; |
8810 | ||
8811 | return mask; | |
8812 | } | |
8813 | ||
8814 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8815 | { | |
8816 | struct io_ring_ctx *ctx = file->private_data; | |
8817 | ||
8818 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8819 | } | |
8820 | ||
0bead8cd | 8821 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8822 | { |
1e6fa521 | 8823 | struct io_identity *iod; |
071698e1 | 8824 | |
1e6fa521 JA |
8825 | iod = idr_remove(&ctx->personality_idr, id); |
8826 | if (iod) { | |
8827 | put_cred(iod->creds); | |
8828 | if (refcount_dec_and_test(&iod->count)) | |
8829 | kfree(iod); | |
0bead8cd | 8830 | return 0; |
1e6fa521 | 8831 | } |
0bead8cd YD |
8832 | |
8833 | return -EINVAL; | |
8834 | } | |
8835 | ||
8836 | static int io_remove_personalities(int id, void *p, void *data) | |
8837 | { | |
8838 | struct io_ring_ctx *ctx = data; | |
8839 | ||
8840 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8841 | return 0; |
8842 | } | |
8843 | ||
85faa7b8 JA |
8844 | static void io_ring_exit_work(struct work_struct *work) |
8845 | { | |
b2edc0a7 PB |
8846 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8847 | exit_work); | |
85faa7b8 | 8848 | |
56952e91 JA |
8849 | /* |
8850 | * If we're doing polled IO and end up having requests being | |
8851 | * submitted async (out-of-line), then completions can come in while | |
8852 | * we're waiting for refs to drop. We need to reap these manually, | |
8853 | * as nobody else will be looking for them. | |
8854 | */ | |
b2edc0a7 | 8855 | do { |
9936c7c2 | 8856 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8857 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8858 | io_ring_ctx_free(ctx); |
8859 | } | |
8860 | ||
00c18640 JA |
8861 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8862 | { | |
8863 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8864 | ||
8865 | return req->ctx == data; | |
8866 | } | |
8867 | ||
2b188cc1 JA |
8868 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8869 | { | |
8870 | mutex_lock(&ctx->uring_lock); | |
8871 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8872 | |
8873 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8874 | ctx->sqo_dead = 1; | |
8875 | ||
cda286f0 PB |
8876 | /* if force is set, the ring is going away. always drop after that */ |
8877 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8878 | if (ctx->rings) |
6c503150 | 8879 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8880 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8881 | mutex_unlock(&ctx->uring_lock); |
8882 | ||
6b81928d PB |
8883 | io_kill_timeouts(ctx, NULL, NULL); |
8884 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a JA |
8885 | |
8886 | if (ctx->io_wq) | |
00c18640 | 8887 | io_wq_cancel_cb(ctx->io_wq, io_cancel_ctx_cb, ctx, true); |
561fb04a | 8888 | |
15dff286 | 8889 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8890 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8891 | |
85faa7b8 | 8892 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8893 | /* |
8894 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8895 | * if we're exiting a ton of rings at the same time. It just adds | |
8896 | * noise and overhead, there's no discernable change in runtime | |
8897 | * over using system_wq. | |
8898 | */ | |
8899 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8900 | } |
8901 | ||
8902 | static int io_uring_release(struct inode *inode, struct file *file) | |
8903 | { | |
8904 | struct io_ring_ctx *ctx = file->private_data; | |
8905 | ||
8906 | file->private_data = NULL; | |
8907 | io_ring_ctx_wait_and_kill(ctx); | |
8908 | return 0; | |
8909 | } | |
8910 | ||
f6edbabb PB |
8911 | struct io_task_cancel { |
8912 | struct task_struct *task; | |
8913 | struct files_struct *files; | |
8914 | }; | |
f254ac04 | 8915 | |
f6edbabb | 8916 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8917 | { |
9a472ef7 | 8918 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8919 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8920 | bool ret; |
8921 | ||
f6edbabb | 8922 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8923 | unsigned long flags; |
8924 | struct io_ring_ctx *ctx = req->ctx; | |
8925 | ||
8926 | /* protect against races with linked timeouts */ | |
8927 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8928 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8929 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8930 | } else { | |
f6edbabb | 8931 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8932 | } |
8933 | return ret; | |
b711d4ea JA |
8934 | } |
8935 | ||
b7ddce3c | 8936 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8937 | struct task_struct *task, |
b7ddce3c PB |
8938 | struct files_struct *files) |
8939 | { | |
8940 | struct io_defer_entry *de = NULL; | |
8941 | LIST_HEAD(list); | |
8942 | ||
8943 | spin_lock_irq(&ctx->completion_lock); | |
8944 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8945 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8946 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8947 | break; | |
8948 | } | |
8949 | } | |
8950 | spin_unlock_irq(&ctx->completion_lock); | |
8951 | ||
8952 | while (!list_empty(&list)) { | |
8953 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8954 | list_del_init(&de->list); | |
8955 | req_set_fail_links(de->req); | |
8956 | io_put_req(de->req); | |
8957 | io_req_complete(de->req, -ECANCELED); | |
8958 | kfree(de); | |
8959 | } | |
8960 | } | |
8961 | ||
9936c7c2 PB |
8962 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8963 | struct task_struct *task, | |
8964 | struct files_struct *files) | |
8965 | { | |
8966 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
8967 | ||
8968 | while (1) { | |
8969 | enum io_wq_cancel cret; | |
8970 | bool ret = false; | |
8971 | ||
8972 | if (ctx->io_wq) { | |
8973 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, | |
8974 | &cancel, true); | |
8975 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8976 | } | |
8977 | ||
8978 | /* SQPOLL thread does its own polling */ | |
8979 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8980 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8981 | io_iopoll_try_reap_events(ctx); | |
8982 | ret = true; | |
8983 | } | |
8984 | } | |
8985 | ||
8986 | ret |= io_poll_remove_all(ctx, task, files); | |
8987 | ret |= io_kill_timeouts(ctx, task, files); | |
8988 | ret |= io_run_task_work(); | |
8989 | io_cqring_overflow_flush(ctx, true, task, files); | |
8990 | if (!ret) | |
8991 | break; | |
8992 | cond_resched(); | |
8993 | } | |
8994 | } | |
8995 | ||
ca70f00b PB |
8996 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8997 | struct task_struct *task, | |
8998 | struct files_struct *files) | |
8999 | { | |
9000 | struct io_kiocb *req; | |
9001 | int cnt = 0; | |
9002 | ||
9003 | spin_lock_irq(&ctx->inflight_lock); | |
9004 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
9005 | cnt += io_match_task(req, task, files); | |
9006 | spin_unlock_irq(&ctx->inflight_lock); | |
9007 | return cnt; | |
9008 | } | |
9009 | ||
b52fda00 | 9010 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 9011 | struct task_struct *task, |
fcb323cc JA |
9012 | struct files_struct *files) |
9013 | { | |
fcb323cc | 9014 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 9015 | DEFINE_WAIT(wait); |
ca70f00b | 9016 | int inflight; |
fcb323cc | 9017 | |
ca70f00b PB |
9018 | inflight = io_uring_count_inflight(ctx, task, files); |
9019 | if (!inflight) | |
fcb323cc | 9020 | break; |
f6edbabb | 9021 | |
9936c7c2 | 9022 | io_uring_try_cancel_requests(ctx, task, files); |
34343786 PB |
9023 | |
9024 | if (ctx->sq_data) | |
9025 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
9026 | prepare_to_wait(&task->io_uring->wait, &wait, |
9027 | TASK_UNINTERRUPTIBLE); | |
9028 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
9029 | schedule(); | |
c98de08c | 9030 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
9031 | if (ctx->sq_data) |
9032 | io_sq_thread_park(ctx->sq_data); | |
fcb323cc | 9033 | } |
44e728b8 PB |
9034 | } |
9035 | ||
d9d05217 PB |
9036 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
9037 | { | |
d9d05217 PB |
9038 | mutex_lock(&ctx->uring_lock); |
9039 | ctx->sqo_dead = 1; | |
9040 | mutex_unlock(&ctx->uring_lock); | |
9041 | ||
9042 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
9043 | if (ctx->rings) |
9044 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
9045 | } |
9046 | ||
0f212204 JA |
9047 | /* |
9048 | * We need to iteratively cancel requests, in case a request has dependent | |
9049 | * hard links. These persist even for failure of cancelations, hence keep | |
9050 | * looping until none are found. | |
9051 | */ | |
9052 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
9053 | struct files_struct *files) | |
9054 | { | |
9055 | struct task_struct *task = current; | |
9056 | ||
fdaf083c | 9057 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 | 9058 | io_disable_sqo_submit(ctx); |
534ca6d6 | 9059 | task = ctx->sq_data->thread; |
fdaf083c JA |
9060 | atomic_inc(&task->io_uring->in_idle); |
9061 | io_sq_thread_park(ctx->sq_data); | |
9062 | } | |
0f212204 | 9063 | |
df9923f9 | 9064 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 9065 | |
3a7efd1a | 9066 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 9067 | if (!files) |
9936c7c2 | 9068 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c JA |
9069 | |
9070 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
9071 | atomic_dec(&task->io_uring->in_idle); | |
9072 | /* | |
9073 | * If the files that are going away are the ones in the thread | |
9074 | * identity, clear them out. | |
9075 | */ | |
9076 | if (task->io_uring->identity->files == files) | |
9077 | task->io_uring->identity->files = NULL; | |
9078 | io_sq_thread_unpark(ctx->sq_data); | |
9079 | } | |
0f212204 JA |
9080 | } |
9081 | ||
9082 | /* | |
9083 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9084 | */ | |
fdaf083c | 9085 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 9086 | { |
236434c3 | 9087 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 9088 | int ret; |
236434c3 MWO |
9089 | |
9090 | if (unlikely(!tctx)) { | |
0f212204 JA |
9091 | ret = io_uring_alloc_task_context(current); |
9092 | if (unlikely(ret)) | |
9093 | return ret; | |
236434c3 | 9094 | tctx = current->io_uring; |
0f212204 | 9095 | } |
236434c3 MWO |
9096 | if (tctx->last != file) { |
9097 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 9098 | |
236434c3 | 9099 | if (!old) { |
0f212204 | 9100 | get_file(file); |
a528b04e PB |
9101 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
9102 | file, GFP_KERNEL)); | |
9103 | if (ret) { | |
9104 | fput(file); | |
9105 | return ret; | |
9106 | } | |
ecfc8492 PB |
9107 | |
9108 | /* one and only SQPOLL file note, held by sqo_task */ | |
9109 | WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && | |
9110 | current != ctx->sqo_task); | |
0f212204 | 9111 | } |
236434c3 | 9112 | tctx->last = file; |
0f212204 JA |
9113 | } |
9114 | ||
fdaf083c JA |
9115 | /* |
9116 | * This is race safe in that the task itself is doing this, hence it | |
9117 | * cannot be going through the exit/cancel paths at the same time. | |
9118 | * This cannot be modified while exit/cancel is running. | |
9119 | */ | |
9120 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
9121 | tctx->sqpoll = true; | |
9122 | ||
0f212204 JA |
9123 | return 0; |
9124 | } | |
9125 | ||
9126 | /* | |
9127 | * Remove this io_uring_file -> task mapping. | |
9128 | */ | |
9129 | static void io_uring_del_task_file(struct file *file) | |
9130 | { | |
9131 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
9132 | |
9133 | if (tctx->last == file) | |
9134 | tctx->last = NULL; | |
5e2ed8c4 | 9135 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
9136 | if (file) |
9137 | fput(file); | |
9138 | } | |
9139 | ||
de7f1d9e PB |
9140 | static void io_uring_remove_task_files(struct io_uring_task *tctx) |
9141 | { | |
9142 | struct file *file; | |
9143 | unsigned long index; | |
9144 | ||
9145 | xa_for_each(&tctx->xa, index, file) | |
9146 | io_uring_del_task_file(file); | |
9147 | } | |
9148 | ||
0f212204 JA |
9149 | void __io_uring_files_cancel(struct files_struct *files) |
9150 | { | |
9151 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
9152 | struct file *file; |
9153 | unsigned long index; | |
0f212204 JA |
9154 | |
9155 | /* make sure overflow events are dropped */ | |
fdaf083c | 9156 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
9157 | xa_for_each(&tctx->xa, index, file) |
9158 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 9159 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9160 | |
9161 | if (files) | |
9162 | io_uring_remove_task_files(tctx); | |
fdaf083c JA |
9163 | } |
9164 | ||
9165 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
9166 | { | |
0e9ddb39 PB |
9167 | return percpu_counter_sum(&tctx->inflight); |
9168 | } | |
fdaf083c | 9169 | |
0e9ddb39 PB |
9170 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
9171 | { | |
9172 | struct io_uring_task *tctx; | |
9173 | s64 inflight; | |
9174 | DEFINE_WAIT(wait); | |
fdaf083c | 9175 | |
0e9ddb39 PB |
9176 | if (!ctx->sq_data) |
9177 | return; | |
9178 | tctx = ctx->sq_data->thread->io_uring; | |
9179 | io_disable_sqo_submit(ctx); | |
fdaf083c | 9180 | |
0e9ddb39 PB |
9181 | atomic_inc(&tctx->in_idle); |
9182 | do { | |
9183 | /* read completions before cancelations */ | |
9184 | inflight = tctx_inflight(tctx); | |
9185 | if (!inflight) | |
9186 | break; | |
9187 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 9188 | |
0e9ddb39 PB |
9189 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9190 | /* | |
9191 | * If we've seen completions, retry without waiting. This | |
9192 | * avoids a race where a completion comes in before we did | |
9193 | * prepare_to_wait(). | |
9194 | */ | |
9195 | if (inflight == tctx_inflight(tctx)) | |
9196 | schedule(); | |
9197 | finish_wait(&tctx->wait, &wait); | |
9198 | } while (1); | |
9199 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9200 | } |
9201 | ||
0f212204 JA |
9202 | /* |
9203 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9204 | * requests. | |
9205 | */ | |
9206 | void __io_uring_task_cancel(void) | |
9207 | { | |
9208 | struct io_uring_task *tctx = current->io_uring; | |
9209 | DEFINE_WAIT(wait); | |
d8a6df10 | 9210 | s64 inflight; |
0f212204 JA |
9211 | |
9212 | /* make sure overflow events are dropped */ | |
fdaf083c | 9213 | atomic_inc(&tctx->in_idle); |
0f212204 | 9214 | |
0b5cd6c3 | 9215 | /* trigger io_disable_sqo_submit() */ |
0e9ddb39 PB |
9216 | if (tctx->sqpoll) { |
9217 | struct file *file; | |
9218 | unsigned long index; | |
9219 | ||
9220 | xa_for_each(&tctx->xa, index, file) | |
9221 | io_uring_cancel_sqpoll(file->private_data); | |
9222 | } | |
0b5cd6c3 | 9223 | |
d8a6df10 | 9224 | do { |
0f212204 | 9225 | /* read completions before cancelations */ |
fdaf083c | 9226 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9227 | if (!inflight) |
9228 | break; | |
0f212204 JA |
9229 | __io_uring_files_cancel(NULL); |
9230 | ||
9231 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9232 | ||
9233 | /* | |
a1bb3cd5 PB |
9234 | * If we've seen completions, retry without waiting. This |
9235 | * avoids a race where a completion comes in before we did | |
9236 | * prepare_to_wait(). | |
0f212204 | 9237 | */ |
a1bb3cd5 PB |
9238 | if (inflight == tctx_inflight(tctx)) |
9239 | schedule(); | |
f57555ed | 9240 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9241 | } while (1); |
0f212204 | 9242 | |
fdaf083c | 9243 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9244 | |
9245 | io_uring_remove_task_files(tctx); | |
44e728b8 PB |
9246 | } |
9247 | ||
fcb323cc JA |
9248 | static int io_uring_flush(struct file *file, void *data) |
9249 | { | |
6b5733eb | 9250 | struct io_uring_task *tctx = current->io_uring; |
d9d05217 | 9251 | struct io_ring_ctx *ctx = file->private_data; |
6b5733eb | 9252 | |
41be53e9 | 9253 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) { |
84965ff8 | 9254 | io_uring_cancel_task_requests(ctx, NULL); |
41be53e9 JA |
9255 | io_req_caches_free(ctx, current); |
9256 | } | |
84965ff8 | 9257 | |
6b5733eb | 9258 | if (!tctx) |
4f793dc4 PB |
9259 | return 0; |
9260 | ||
6b5733eb PB |
9261 | /* we should have cancelled and erased it before PF_EXITING */ |
9262 | WARN_ON_ONCE((current->flags & PF_EXITING) && | |
9263 | xa_load(&tctx->xa, (unsigned long)file)); | |
9264 | ||
4f793dc4 PB |
9265 | /* |
9266 | * fput() is pending, will be 2 if the only other ref is our potential | |
9267 | * task file note. If the task is exiting, drop regardless of count. | |
9268 | */ | |
6b5733eb PB |
9269 | if (atomic_long_read(&file->f_count) != 2) |
9270 | return 0; | |
4f793dc4 | 9271 | |
d9d05217 PB |
9272 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
9273 | /* there is only one file note, which is owned by sqo_task */ | |
4325cb49 PB |
9274 | WARN_ON_ONCE(ctx->sqo_task != current && |
9275 | xa_load(&tctx->xa, (unsigned long)file)); | |
9276 | /* sqo_dead check is for when this happens after cancellation */ | |
9277 | WARN_ON_ONCE(ctx->sqo_task == current && !ctx->sqo_dead && | |
d9d05217 PB |
9278 | !xa_load(&tctx->xa, (unsigned long)file)); |
9279 | ||
9280 | io_disable_sqo_submit(ctx); | |
9281 | } | |
9282 | ||
9283 | if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current) | |
9284 | io_uring_del_task_file(file); | |
fcb323cc JA |
9285 | return 0; |
9286 | } | |
9287 | ||
6c5c240e RP |
9288 | static void *io_uring_validate_mmap_request(struct file *file, |
9289 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9290 | { |
2b188cc1 | 9291 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9292 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9293 | struct page *page; |
9294 | void *ptr; | |
9295 | ||
9296 | switch (offset) { | |
9297 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9298 | case IORING_OFF_CQ_RING: |
9299 | ptr = ctx->rings; | |
2b188cc1 JA |
9300 | break; |
9301 | case IORING_OFF_SQES: | |
9302 | ptr = ctx->sq_sqes; | |
9303 | break; | |
2b188cc1 | 9304 | default: |
6c5c240e | 9305 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9306 | } |
9307 | ||
9308 | page = virt_to_head_page(ptr); | |
a50b854e | 9309 | if (sz > page_size(page)) |
6c5c240e RP |
9310 | return ERR_PTR(-EINVAL); |
9311 | ||
9312 | return ptr; | |
9313 | } | |
9314 | ||
9315 | #ifdef CONFIG_MMU | |
9316 | ||
9317 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9318 | { | |
9319 | size_t sz = vma->vm_end - vma->vm_start; | |
9320 | unsigned long pfn; | |
9321 | void *ptr; | |
9322 | ||
9323 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9324 | if (IS_ERR(ptr)) | |
9325 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9326 | |
9327 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9328 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9329 | } | |
9330 | ||
6c5c240e RP |
9331 | #else /* !CONFIG_MMU */ |
9332 | ||
9333 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9334 | { | |
9335 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9336 | } | |
9337 | ||
9338 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9339 | { | |
9340 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9341 | } | |
9342 | ||
9343 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9344 | unsigned long addr, unsigned long len, | |
9345 | unsigned long pgoff, unsigned long flags) | |
9346 | { | |
9347 | void *ptr; | |
9348 | ||
9349 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9350 | if (IS_ERR(ptr)) | |
9351 | return PTR_ERR(ptr); | |
9352 | ||
9353 | return (unsigned long) ptr; | |
9354 | } | |
9355 | ||
9356 | #endif /* !CONFIG_MMU */ | |
9357 | ||
d9d05217 | 9358 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9359 | { |
d9d05217 | 9360 | int ret = 0; |
90554200 JA |
9361 | DEFINE_WAIT(wait); |
9362 | ||
9363 | do { | |
9364 | if (!io_sqring_full(ctx)) | |
9365 | break; | |
9366 | ||
9367 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
9368 | ||
d9d05217 PB |
9369 | if (unlikely(ctx->sqo_dead)) { |
9370 | ret = -EOWNERDEAD; | |
9371 | goto out; | |
9372 | } | |
9373 | ||
90554200 JA |
9374 | if (!io_sqring_full(ctx)) |
9375 | break; | |
9376 | ||
9377 | schedule(); | |
9378 | } while (!signal_pending(current)); | |
9379 | ||
9380 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9381 | out: |
9382 | return ret; | |
90554200 JA |
9383 | } |
9384 | ||
c73ebb68 HX |
9385 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9386 | struct __kernel_timespec __user **ts, | |
9387 | const sigset_t __user **sig) | |
9388 | { | |
9389 | struct io_uring_getevents_arg arg; | |
9390 | ||
9391 | /* | |
9392 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9393 | * is just a pointer to the sigset_t. | |
9394 | */ | |
9395 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9396 | *sig = (const sigset_t __user *) argp; | |
9397 | *ts = NULL; | |
9398 | return 0; | |
9399 | } | |
9400 | ||
9401 | /* | |
9402 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9403 | * timespec and sigset_t pointers if good. | |
9404 | */ | |
9405 | if (*argsz != sizeof(arg)) | |
9406 | return -EINVAL; | |
9407 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9408 | return -EFAULT; | |
9409 | *sig = u64_to_user_ptr(arg.sigmask); | |
9410 | *argsz = arg.sigmask_sz; | |
9411 | *ts = u64_to_user_ptr(arg.ts); | |
9412 | return 0; | |
9413 | } | |
9414 | ||
2b188cc1 | 9415 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9416 | u32, min_complete, u32, flags, const void __user *, argp, |
9417 | size_t, argsz) | |
2b188cc1 JA |
9418 | { |
9419 | struct io_ring_ctx *ctx; | |
9420 | long ret = -EBADF; | |
9421 | int submitted = 0; | |
9422 | struct fd f; | |
9423 | ||
4c6e277c | 9424 | io_run_task_work(); |
b41e9852 | 9425 | |
90554200 | 9426 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9427 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9428 | return -EINVAL; |
9429 | ||
9430 | f = fdget(fd); | |
9431 | if (!f.file) | |
9432 | return -EBADF; | |
9433 | ||
9434 | ret = -EOPNOTSUPP; | |
9435 | if (f.file->f_op != &io_uring_fops) | |
9436 | goto out_fput; | |
9437 | ||
9438 | ret = -ENXIO; | |
9439 | ctx = f.file->private_data; | |
9440 | if (!percpu_ref_tryget(&ctx->refs)) | |
9441 | goto out_fput; | |
9442 | ||
7e84e1c7 SG |
9443 | ret = -EBADFD; |
9444 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9445 | goto out; | |
9446 | ||
6c271ce2 JA |
9447 | /* |
9448 | * For SQ polling, the thread will do all submissions and completions. | |
9449 | * Just return the requested submit count, and wake the thread if | |
9450 | * we were asked to. | |
9451 | */ | |
b2a9eada | 9452 | ret = 0; |
6c271ce2 | 9453 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9454 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9455 | |
d9d05217 PB |
9456 | ret = -EOWNERDEAD; |
9457 | if (unlikely(ctx->sqo_dead)) | |
9458 | goto out; | |
6c271ce2 | 9459 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9460 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9461 | if (flags & IORING_ENTER_SQ_WAIT) { |
9462 | ret = io_sqpoll_wait_sq(ctx); | |
9463 | if (ret) | |
9464 | goto out; | |
9465 | } | |
6c271ce2 | 9466 | submitted = to_submit; |
b2a9eada | 9467 | } else if (to_submit) { |
fdaf083c | 9468 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9469 | if (unlikely(ret)) |
9470 | goto out; | |
2b188cc1 | 9471 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9472 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9473 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9474 | |
9475 | if (submitted != to_submit) | |
9476 | goto out; | |
2b188cc1 JA |
9477 | } |
9478 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9479 | const sigset_t __user *sig; |
9480 | struct __kernel_timespec __user *ts; | |
9481 | ||
9482 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9483 | if (unlikely(ret)) | |
9484 | goto out; | |
9485 | ||
2b188cc1 JA |
9486 | min_complete = min(min_complete, ctx->cq_entries); |
9487 | ||
32b2244a XW |
9488 | /* |
9489 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9490 | * space applications don't need to do io completion events | |
9491 | * polling again, they can rely on io_sq_thread to do polling | |
9492 | * work, which can reduce cpu usage and uring_lock contention. | |
9493 | */ | |
9494 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9495 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9496 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9497 | } else { |
c73ebb68 | 9498 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9499 | } |
2b188cc1 JA |
9500 | } |
9501 | ||
7c504e65 | 9502 | out: |
6805b32e | 9503 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9504 | out_fput: |
9505 | fdput(f); | |
9506 | return submitted ? submitted : ret; | |
9507 | } | |
9508 | ||
bebdb65e | 9509 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9510 | static int io_uring_show_cred(int id, void *p, void *data) |
9511 | { | |
6b47ab81 JA |
9512 | struct io_identity *iod = p; |
9513 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9514 | struct seq_file *m = data; |
9515 | struct user_namespace *uns = seq_user_ns(m); | |
9516 | struct group_info *gi; | |
9517 | kernel_cap_t cap; | |
9518 | unsigned __capi; | |
9519 | int g; | |
9520 | ||
9521 | seq_printf(m, "%5d\n", id); | |
9522 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9523 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9524 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9525 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9526 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9527 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9528 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9529 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9530 | seq_puts(m, "\n\tGroups:\t"); | |
9531 | gi = cred->group_info; | |
9532 | for (g = 0; g < gi->ngroups; g++) { | |
9533 | seq_put_decimal_ull(m, g ? " " : "", | |
9534 | from_kgid_munged(uns, gi->gid[g])); | |
9535 | } | |
9536 | seq_puts(m, "\n\tCapEff:\t"); | |
9537 | cap = cred->cap_effective; | |
9538 | CAP_FOR_EACH_U32(__capi) | |
9539 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9540 | seq_putc(m, '\n'); | |
9541 | return 0; | |
9542 | } | |
9543 | ||
9544 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9545 | { | |
dbbe9c64 | 9546 | struct io_sq_data *sq = NULL; |
fad8e0de | 9547 | bool has_lock; |
87ce955b JA |
9548 | int i; |
9549 | ||
fad8e0de JA |
9550 | /* |
9551 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9552 | * since fdinfo case grabs it in the opposite direction of normal use | |
9553 | * cases. If we fail to get the lock, we just don't iterate any | |
9554 | * structures that could be going away outside the io_uring mutex. | |
9555 | */ | |
9556 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9557 | ||
dbbe9c64 JQ |
9558 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9559 | sq = ctx->sq_data; | |
9560 | ||
9561 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9562 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9563 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9564 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9565 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9566 | |
87ce955b JA |
9567 | if (f) |
9568 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9569 | else | |
9570 | seq_printf(m, "%5u: <none>\n", i); | |
9571 | } | |
9572 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9573 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9574 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9575 | ||
9576 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9577 | (unsigned int) buf->len); | |
9578 | } | |
fad8e0de | 9579 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9580 | seq_printf(m, "Personalities:\n"); |
9581 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9582 | } | |
d7718a9d JA |
9583 | seq_printf(m, "PollList:\n"); |
9584 | spin_lock_irq(&ctx->completion_lock); | |
9585 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9586 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9587 | struct io_kiocb *req; | |
9588 | ||
9589 | hlist_for_each_entry(req, list, hash_node) | |
9590 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9591 | req->task->task_works != NULL); | |
9592 | } | |
9593 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9594 | if (has_lock) |
9595 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9596 | } |
9597 | ||
9598 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9599 | { | |
9600 | struct io_ring_ctx *ctx = f->private_data; | |
9601 | ||
9602 | if (percpu_ref_tryget(&ctx->refs)) { | |
9603 | __io_uring_show_fdinfo(ctx, m); | |
9604 | percpu_ref_put(&ctx->refs); | |
9605 | } | |
9606 | } | |
bebdb65e | 9607 | #endif |
87ce955b | 9608 | |
2b188cc1 JA |
9609 | static const struct file_operations io_uring_fops = { |
9610 | .release = io_uring_release, | |
fcb323cc | 9611 | .flush = io_uring_flush, |
2b188cc1 | 9612 | .mmap = io_uring_mmap, |
6c5c240e RP |
9613 | #ifndef CONFIG_MMU |
9614 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9615 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9616 | #endif | |
2b188cc1 JA |
9617 | .poll = io_uring_poll, |
9618 | .fasync = io_uring_fasync, | |
bebdb65e | 9619 | #ifdef CONFIG_PROC_FS |
87ce955b | 9620 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9621 | #endif |
2b188cc1 JA |
9622 | }; |
9623 | ||
9624 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9625 | struct io_uring_params *p) | |
9626 | { | |
75b28aff HV |
9627 | struct io_rings *rings; |
9628 | size_t size, sq_array_offset; | |
2b188cc1 | 9629 | |
bd740481 JA |
9630 | /* make sure these are sane, as we already accounted them */ |
9631 | ctx->sq_entries = p->sq_entries; | |
9632 | ctx->cq_entries = p->cq_entries; | |
9633 | ||
75b28aff HV |
9634 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9635 | if (size == SIZE_MAX) | |
9636 | return -EOVERFLOW; | |
9637 | ||
9638 | rings = io_mem_alloc(size); | |
9639 | if (!rings) | |
2b188cc1 JA |
9640 | return -ENOMEM; |
9641 | ||
75b28aff HV |
9642 | ctx->rings = rings; |
9643 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9644 | rings->sq_ring_mask = p->sq_entries - 1; | |
9645 | rings->cq_ring_mask = p->cq_entries - 1; | |
9646 | rings->sq_ring_entries = p->sq_entries; | |
9647 | rings->cq_ring_entries = p->cq_entries; | |
9648 | ctx->sq_mask = rings->sq_ring_mask; | |
9649 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9650 | |
9651 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9652 | if (size == SIZE_MAX) { |
9653 | io_mem_free(ctx->rings); | |
9654 | ctx->rings = NULL; | |
2b188cc1 | 9655 | return -EOVERFLOW; |
eb065d30 | 9656 | } |
2b188cc1 JA |
9657 | |
9658 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9659 | if (!ctx->sq_sqes) { |
9660 | io_mem_free(ctx->rings); | |
9661 | ctx->rings = NULL; | |
2b188cc1 | 9662 | return -ENOMEM; |
eb065d30 | 9663 | } |
2b188cc1 | 9664 | |
2b188cc1 JA |
9665 | return 0; |
9666 | } | |
9667 | ||
9faadcc8 PB |
9668 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9669 | { | |
9670 | int ret, fd; | |
9671 | ||
9672 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9673 | if (fd < 0) | |
9674 | return fd; | |
9675 | ||
9676 | ret = io_uring_add_task_file(ctx, file); | |
9677 | if (ret) { | |
9678 | put_unused_fd(fd); | |
9679 | return ret; | |
9680 | } | |
9681 | fd_install(fd, file); | |
9682 | return fd; | |
9683 | } | |
9684 | ||
2b188cc1 JA |
9685 | /* |
9686 | * Allocate an anonymous fd, this is what constitutes the application | |
9687 | * visible backing of an io_uring instance. The application mmaps this | |
9688 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9689 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9690 | */ | |
9faadcc8 | 9691 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9692 | { |
9693 | struct file *file; | |
9faadcc8 | 9694 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9695 | int ret; |
9696 | ||
2b188cc1 JA |
9697 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9698 | &ctx->ring_sock); | |
9699 | if (ret) | |
9faadcc8 | 9700 | return ERR_PTR(ret); |
2b188cc1 JA |
9701 | #endif |
9702 | ||
2b188cc1 JA |
9703 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9704 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9705 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9706 | if (IS_ERR(file)) { |
9707 | sock_release(ctx->ring_sock); | |
9708 | ctx->ring_sock = NULL; | |
9709 | } else { | |
9710 | ctx->ring_sock->file = file; | |
0f212204 | 9711 | } |
2b188cc1 | 9712 | #endif |
9faadcc8 | 9713 | return file; |
2b188cc1 JA |
9714 | } |
9715 | ||
7f13657d XW |
9716 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9717 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9718 | { |
9719 | struct user_struct *user = NULL; | |
9720 | struct io_ring_ctx *ctx; | |
9faadcc8 | 9721 | struct file *file; |
2b188cc1 JA |
9722 | int ret; |
9723 | ||
8110c1a6 | 9724 | if (!entries) |
2b188cc1 | 9725 | return -EINVAL; |
8110c1a6 JA |
9726 | if (entries > IORING_MAX_ENTRIES) { |
9727 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9728 | return -EINVAL; | |
9729 | entries = IORING_MAX_ENTRIES; | |
9730 | } | |
2b188cc1 JA |
9731 | |
9732 | /* | |
9733 | * Use twice as many entries for the CQ ring. It's possible for the | |
9734 | * application to drive a higher depth than the size of the SQ ring, | |
9735 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9736 | * some flexibility in overcommitting a bit. If the application has |
9737 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9738 | * of CQ ring entries manually. | |
2b188cc1 JA |
9739 | */ |
9740 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9741 | if (p->flags & IORING_SETUP_CQSIZE) { |
9742 | /* | |
9743 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9744 | * to a power-of-two, if it isn't already. We do NOT impose | |
9745 | * any cq vs sq ring sizing. | |
9746 | */ | |
eb2667b3 | 9747 | if (!p->cq_entries) |
33a107f0 | 9748 | return -EINVAL; |
8110c1a6 JA |
9749 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9750 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9751 | return -EINVAL; | |
9752 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9753 | } | |
eb2667b3 JQ |
9754 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9755 | if (p->cq_entries < p->sq_entries) | |
9756 | return -EINVAL; | |
33a107f0 JA |
9757 | } else { |
9758 | p->cq_entries = 2 * p->sq_entries; | |
9759 | } | |
2b188cc1 JA |
9760 | |
9761 | user = get_uid(current_user()); | |
2b188cc1 JA |
9762 | |
9763 | ctx = io_ring_ctx_alloc(p); | |
9764 | if (!ctx) { | |
2b188cc1 JA |
9765 | free_uid(user); |
9766 | return -ENOMEM; | |
9767 | } | |
9768 | ctx->compat = in_compat_syscall(); | |
26bfa89e | 9769 | ctx->limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9770 | ctx->user = user; |
0b8c0ec7 | 9771 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9772 | #ifdef CONFIG_AUDIT |
9773 | ctx->loginuid = current->loginuid; | |
9774 | ctx->sessionid = current->sessionid; | |
9775 | #endif | |
2aede0e4 JA |
9776 | ctx->sqo_task = get_task_struct(current); |
9777 | ||
9778 | /* | |
9779 | * This is just grabbed for accounting purposes. When a process exits, | |
9780 | * the mm is exited and dropped before the files, hence we need to hang | |
9781 | * on to this mm purely for the purposes of being able to unaccount | |
9782 | * memory (locked/pinned vm). It's not used for anything else. | |
9783 | */ | |
6b7898eb | 9784 | mmgrab(current->mm); |
2aede0e4 | 9785 | ctx->mm_account = current->mm; |
6b7898eb | 9786 | |
91d8f519 DZ |
9787 | #ifdef CONFIG_BLK_CGROUP |
9788 | /* | |
9789 | * The sq thread will belong to the original cgroup it was inited in. | |
9790 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9791 | * issued bios will be associated with the closest cgroup later in the | |
9792 | * block layer. | |
9793 | */ | |
9794 | rcu_read_lock(); | |
9795 | ctx->sqo_blkcg_css = blkcg_css(); | |
9796 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9797 | rcu_read_unlock(); | |
9798 | if (!ret) { | |
9799 | /* don't init against a dying cgroup, have the user try again */ | |
9800 | ctx->sqo_blkcg_css = NULL; | |
9801 | ret = -ENODEV; | |
9802 | goto err; | |
9803 | } | |
9804 | #endif | |
2b188cc1 JA |
9805 | ret = io_allocate_scq_urings(ctx, p); |
9806 | if (ret) | |
9807 | goto err; | |
9808 | ||
7e84e1c7 | 9809 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9810 | if (ret) |
9811 | goto err; | |
9812 | ||
7e84e1c7 SG |
9813 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9814 | io_sq_offload_start(ctx); | |
9815 | ||
2b188cc1 | 9816 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9817 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9818 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9819 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9820 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9821 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9822 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9823 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9824 | |
9825 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9826 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9827 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9828 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9829 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9830 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9831 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9832 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9833 | |
7f13657d XW |
9834 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9835 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9836 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 HX |
9837 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9838 | IORING_FEAT_EXT_ARG; | |
7f13657d XW |
9839 | |
9840 | if (copy_to_user(params, p, sizeof(*p))) { | |
9841 | ret = -EFAULT; | |
9842 | goto err; | |
9843 | } | |
d1719f70 | 9844 | |
9faadcc8 PB |
9845 | file = io_uring_get_file(ctx); |
9846 | if (IS_ERR(file)) { | |
9847 | ret = PTR_ERR(file); | |
9848 | goto err; | |
9849 | } | |
9850 | ||
044c1ab3 JA |
9851 | /* |
9852 | * Install ring fd as the very last thing, so we don't risk someone | |
9853 | * having closed it before we finish setup | |
9854 | */ | |
9faadcc8 PB |
9855 | ret = io_uring_install_fd(ctx, file); |
9856 | if (ret < 0) { | |
06585c49 | 9857 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9858 | /* fput will clean it up */ |
9859 | fput(file); | |
9860 | return ret; | |
9861 | } | |
044c1ab3 | 9862 | |
c826bd7a | 9863 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9864 | return ret; |
9865 | err: | |
d9d05217 | 9866 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9867 | io_ring_ctx_wait_and_kill(ctx); |
9868 | return ret; | |
9869 | } | |
9870 | ||
9871 | /* | |
9872 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9873 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9874 | * params structure passed in. | |
9875 | */ | |
9876 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9877 | { | |
9878 | struct io_uring_params p; | |
2b188cc1 JA |
9879 | int i; |
9880 | ||
9881 | if (copy_from_user(&p, params, sizeof(p))) | |
9882 | return -EFAULT; | |
9883 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9884 | if (p.resv[i]) | |
9885 | return -EINVAL; | |
9886 | } | |
9887 | ||
6c271ce2 | 9888 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9889 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9890 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9891 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9892 | return -EINVAL; |
9893 | ||
7f13657d | 9894 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9895 | } |
9896 | ||
9897 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9898 | struct io_uring_params __user *, params) | |
9899 | { | |
9900 | return io_uring_setup(entries, params); | |
9901 | } | |
9902 | ||
66f4af93 JA |
9903 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9904 | { | |
9905 | struct io_uring_probe *p; | |
9906 | size_t size; | |
9907 | int i, ret; | |
9908 | ||
9909 | size = struct_size(p, ops, nr_args); | |
9910 | if (size == SIZE_MAX) | |
9911 | return -EOVERFLOW; | |
9912 | p = kzalloc(size, GFP_KERNEL); | |
9913 | if (!p) | |
9914 | return -ENOMEM; | |
9915 | ||
9916 | ret = -EFAULT; | |
9917 | if (copy_from_user(p, arg, size)) | |
9918 | goto out; | |
9919 | ret = -EINVAL; | |
9920 | if (memchr_inv(p, 0, size)) | |
9921 | goto out; | |
9922 | ||
9923 | p->last_op = IORING_OP_LAST - 1; | |
9924 | if (nr_args > IORING_OP_LAST) | |
9925 | nr_args = IORING_OP_LAST; | |
9926 | ||
9927 | for (i = 0; i < nr_args; i++) { | |
9928 | p->ops[i].op = i; | |
9929 | if (!io_op_defs[i].not_supported) | |
9930 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9931 | } | |
9932 | p->ops_len = i; | |
9933 | ||
9934 | ret = 0; | |
9935 | if (copy_to_user(arg, p, size)) | |
9936 | ret = -EFAULT; | |
9937 | out: | |
9938 | kfree(p); | |
9939 | return ret; | |
9940 | } | |
9941 | ||
071698e1 JA |
9942 | static int io_register_personality(struct io_ring_ctx *ctx) |
9943 | { | |
1e6fa521 JA |
9944 | struct io_identity *id; |
9945 | int ret; | |
071698e1 | 9946 | |
1e6fa521 JA |
9947 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9948 | if (unlikely(!id)) | |
9949 | return -ENOMEM; | |
9950 | ||
9951 | io_init_identity(id); | |
9952 | id->creds = get_current_cred(); | |
9953 | ||
9954 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9955 | if (ret < 0) { | |
9956 | put_cred(id->creds); | |
9957 | kfree(id); | |
9958 | } | |
9959 | return ret; | |
071698e1 JA |
9960 | } |
9961 | ||
21b55dbc SG |
9962 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9963 | unsigned int nr_args) | |
9964 | { | |
9965 | struct io_uring_restriction *res; | |
9966 | size_t size; | |
9967 | int i, ret; | |
9968 | ||
7e84e1c7 SG |
9969 | /* Restrictions allowed only if rings started disabled */ |
9970 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9971 | return -EBADFD; | |
9972 | ||
21b55dbc | 9973 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9974 | if (ctx->restrictions.registered) |
21b55dbc SG |
9975 | return -EBUSY; |
9976 | ||
9977 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9978 | return -EINVAL; | |
9979 | ||
9980 | size = array_size(nr_args, sizeof(*res)); | |
9981 | if (size == SIZE_MAX) | |
9982 | return -EOVERFLOW; | |
9983 | ||
9984 | res = memdup_user(arg, size); | |
9985 | if (IS_ERR(res)) | |
9986 | return PTR_ERR(res); | |
9987 | ||
9988 | ret = 0; | |
9989 | ||
9990 | for (i = 0; i < nr_args; i++) { | |
9991 | switch (res[i].opcode) { | |
9992 | case IORING_RESTRICTION_REGISTER_OP: | |
9993 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9994 | ret = -EINVAL; | |
9995 | goto out; | |
9996 | } | |
9997 | ||
9998 | __set_bit(res[i].register_op, | |
9999 | ctx->restrictions.register_op); | |
10000 | break; | |
10001 | case IORING_RESTRICTION_SQE_OP: | |
10002 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
10003 | ret = -EINVAL; | |
10004 | goto out; | |
10005 | } | |
10006 | ||
10007 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
10008 | break; | |
10009 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
10010 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
10011 | break; | |
10012 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
10013 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
10014 | break; | |
10015 | default: | |
10016 | ret = -EINVAL; | |
10017 | goto out; | |
10018 | } | |
10019 | } | |
10020 | ||
10021 | out: | |
10022 | /* Reset all restrictions if an error happened */ | |
10023 | if (ret != 0) | |
10024 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10025 | else | |
7e84e1c7 | 10026 | ctx->restrictions.registered = true; |
21b55dbc SG |
10027 | |
10028 | kfree(res); | |
10029 | return ret; | |
10030 | } | |
10031 | ||
7e84e1c7 SG |
10032 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10033 | { | |
10034 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10035 | return -EBADFD; | |
10036 | ||
10037 | if (ctx->restrictions.registered) | |
10038 | ctx->restricted = 1; | |
10039 | ||
10040 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
10041 | ||
10042 | io_sq_offload_start(ctx); | |
10043 | ||
10044 | return 0; | |
10045 | } | |
10046 | ||
071698e1 JA |
10047 | static bool io_register_op_must_quiesce(int op) |
10048 | { | |
10049 | switch (op) { | |
10050 | case IORING_UNREGISTER_FILES: | |
10051 | case IORING_REGISTER_FILES_UPDATE: | |
10052 | case IORING_REGISTER_PROBE: | |
10053 | case IORING_REGISTER_PERSONALITY: | |
10054 | case IORING_UNREGISTER_PERSONALITY: | |
10055 | return false; | |
10056 | default: | |
10057 | return true; | |
10058 | } | |
10059 | } | |
10060 | ||
edafccee JA |
10061 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10062 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10063 | __releases(ctx->uring_lock) |
10064 | __acquires(ctx->uring_lock) | |
edafccee JA |
10065 | { |
10066 | int ret; | |
10067 | ||
35fa71a0 JA |
10068 | /* |
10069 | * We're inside the ring mutex, if the ref is already dying, then | |
10070 | * someone else killed the ctx or is already going through | |
10071 | * io_uring_register(). | |
10072 | */ | |
10073 | if (percpu_ref_is_dying(&ctx->refs)) | |
10074 | return -ENXIO; | |
10075 | ||
071698e1 | 10076 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10077 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10078 | |
05f3fb3c JA |
10079 | /* |
10080 | * Drop uring mutex before waiting for references to exit. If | |
10081 | * another thread is currently inside io_uring_enter() it might | |
10082 | * need to grab the uring_lock to make progress. If we hold it | |
10083 | * here across the drain wait, then we can deadlock. It's safe | |
10084 | * to drop the mutex here, since no new references will come in | |
10085 | * after we've killed the percpu ref. | |
10086 | */ | |
10087 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10088 | do { |
10089 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10090 | if (!ret) | |
10091 | break; | |
ed6930c9 JA |
10092 | ret = io_run_task_work_sig(); |
10093 | if (ret < 0) | |
10094 | break; | |
af9c1a44 JA |
10095 | } while (1); |
10096 | ||
05f3fb3c | 10097 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10098 | |
c150368b JA |
10099 | if (ret) { |
10100 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
10101 | goto out_quiesce; |
10102 | } | |
10103 | } | |
10104 | ||
10105 | if (ctx->restricted) { | |
10106 | if (opcode >= IORING_REGISTER_LAST) { | |
10107 | ret = -EINVAL; | |
10108 | goto out; | |
10109 | } | |
10110 | ||
10111 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
10112 | ret = -EACCES; | |
c150368b JA |
10113 | goto out; |
10114 | } | |
05f3fb3c | 10115 | } |
edafccee JA |
10116 | |
10117 | switch (opcode) { | |
10118 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 10119 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
10120 | break; |
10121 | case IORING_UNREGISTER_BUFFERS: | |
10122 | ret = -EINVAL; | |
10123 | if (arg || nr_args) | |
10124 | break; | |
0a96bbe4 | 10125 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10126 | break; |
6b06314c JA |
10127 | case IORING_REGISTER_FILES: |
10128 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
10129 | break; | |
10130 | case IORING_UNREGISTER_FILES: | |
10131 | ret = -EINVAL; | |
10132 | if (arg || nr_args) | |
10133 | break; | |
10134 | ret = io_sqe_files_unregister(ctx); | |
10135 | break; | |
c3a31e60 JA |
10136 | case IORING_REGISTER_FILES_UPDATE: |
10137 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
10138 | break; | |
9b402849 | 10139 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10140 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10141 | ret = -EINVAL; |
10142 | if (nr_args != 1) | |
10143 | break; | |
10144 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10145 | if (ret) |
10146 | break; | |
10147 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10148 | ctx->eventfd_async = 1; | |
10149 | else | |
10150 | ctx->eventfd_async = 0; | |
9b402849 JA |
10151 | break; |
10152 | case IORING_UNREGISTER_EVENTFD: | |
10153 | ret = -EINVAL; | |
10154 | if (arg || nr_args) | |
10155 | break; | |
10156 | ret = io_eventfd_unregister(ctx); | |
10157 | break; | |
66f4af93 JA |
10158 | case IORING_REGISTER_PROBE: |
10159 | ret = -EINVAL; | |
10160 | if (!arg || nr_args > 256) | |
10161 | break; | |
10162 | ret = io_probe(ctx, arg, nr_args); | |
10163 | break; | |
071698e1 JA |
10164 | case IORING_REGISTER_PERSONALITY: |
10165 | ret = -EINVAL; | |
10166 | if (arg || nr_args) | |
10167 | break; | |
10168 | ret = io_register_personality(ctx); | |
10169 | break; | |
10170 | case IORING_UNREGISTER_PERSONALITY: | |
10171 | ret = -EINVAL; | |
10172 | if (arg) | |
10173 | break; | |
10174 | ret = io_unregister_personality(ctx, nr_args); | |
10175 | break; | |
7e84e1c7 SG |
10176 | case IORING_REGISTER_ENABLE_RINGS: |
10177 | ret = -EINVAL; | |
10178 | if (arg || nr_args) | |
10179 | break; | |
10180 | ret = io_register_enable_rings(ctx); | |
10181 | break; | |
21b55dbc SG |
10182 | case IORING_REGISTER_RESTRICTIONS: |
10183 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10184 | break; | |
edafccee JA |
10185 | default: |
10186 | ret = -EINVAL; | |
10187 | break; | |
10188 | } | |
10189 | ||
21b55dbc | 10190 | out: |
071698e1 | 10191 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10192 | /* bring the ctx back to life */ |
05f3fb3c | 10193 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 10194 | out_quiesce: |
0f158b4c | 10195 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10196 | } |
edafccee JA |
10197 | return ret; |
10198 | } | |
10199 | ||
10200 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10201 | void __user *, arg, unsigned int, nr_args) | |
10202 | { | |
10203 | struct io_ring_ctx *ctx; | |
10204 | long ret = -EBADF; | |
10205 | struct fd f; | |
10206 | ||
10207 | f = fdget(fd); | |
10208 | if (!f.file) | |
10209 | return -EBADF; | |
10210 | ||
10211 | ret = -EOPNOTSUPP; | |
10212 | if (f.file->f_op != &io_uring_fops) | |
10213 | goto out_fput; | |
10214 | ||
10215 | ctx = f.file->private_data; | |
10216 | ||
10217 | mutex_lock(&ctx->uring_lock); | |
10218 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10219 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10220 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10221 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10222 | out_fput: |
10223 | fdput(f); | |
10224 | return ret; | |
10225 | } | |
10226 | ||
2b188cc1 JA |
10227 | static int __init io_uring_init(void) |
10228 | { | |
d7f62e82 SM |
10229 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10230 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10231 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10232 | } while (0) | |
10233 | ||
10234 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10235 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10236 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10237 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10238 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10239 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10240 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10241 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10242 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10243 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10244 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10245 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10246 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10247 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10248 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10249 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10250 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10251 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10252 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10253 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10254 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10255 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10256 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10257 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10258 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10259 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10260 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10261 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10262 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10263 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10264 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10265 | |
d3656344 | 10266 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10267 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10268 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10269 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10270 | return 0; |
10271 | }; | |
10272 | __initcall(io_uring_init); |