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 | ||
286 | struct io_submit_state { | |
287 | struct blk_plug plug; | |
288 | ||
289 | /* | |
290 | * io_kiocb alloc cache | |
291 | */ | |
bf019da7 | 292 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
293 | unsigned int free_reqs; |
294 | ||
295 | bool plug_started; | |
296 | ||
297 | /* | |
298 | * Batch completion logic | |
299 | */ | |
300 | struct io_comp_state comp; | |
301 | ||
302 | /* | |
303 | * File reference cache | |
304 | */ | |
305 | struct file *file; | |
306 | unsigned int fd; | |
307 | unsigned int file_refs; | |
308 | unsigned int ios_left; | |
309 | }; | |
310 | ||
2b188cc1 JA |
311 | struct io_ring_ctx { |
312 | struct { | |
313 | struct percpu_ref refs; | |
314 | } ____cacheline_aligned_in_smp; | |
315 | ||
316 | struct { | |
317 | unsigned int flags; | |
e1d85334 | 318 | unsigned int compat: 1; |
aad5d8da | 319 | unsigned int limit_mem: 1; |
e1d85334 RD |
320 | unsigned int cq_overflow_flushed: 1; |
321 | unsigned int drain_next: 1; | |
322 | unsigned int eventfd_async: 1; | |
21b55dbc | 323 | unsigned int restricted: 1; |
d9d05217 | 324 | unsigned int sqo_dead: 1; |
2b188cc1 | 325 | |
75b28aff HV |
326 | /* |
327 | * Ring buffer of indices into array of io_uring_sqe, which is | |
328 | * mmapped by the application using the IORING_OFF_SQES offset. | |
329 | * | |
330 | * This indirection could e.g. be used to assign fixed | |
331 | * io_uring_sqe entries to operations and only submit them to | |
332 | * the queue when needed. | |
333 | * | |
334 | * The kernel modifies neither the indices array nor the entries | |
335 | * array. | |
336 | */ | |
337 | u32 *sq_array; | |
2b188cc1 JA |
338 | unsigned cached_sq_head; |
339 | unsigned sq_entries; | |
340 | unsigned sq_mask; | |
6c271ce2 | 341 | unsigned sq_thread_idle; |
498ccd9e | 342 | unsigned cached_sq_dropped; |
2c3bac6d | 343 | unsigned cached_cq_overflow; |
ad3eb2c8 | 344 | unsigned long sq_check_overflow; |
de0617e4 JA |
345 | |
346 | struct list_head defer_list; | |
5262f567 | 347 | struct list_head timeout_list; |
1d7bb1d5 | 348 | struct list_head cq_overflow_list; |
fcb323cc | 349 | |
ad3eb2c8 | 350 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
351 | } ____cacheline_aligned_in_smp; |
352 | ||
3c1a2ead JA |
353 | struct { |
354 | struct mutex uring_lock; | |
355 | wait_queue_head_t wait; | |
356 | } ____cacheline_aligned_in_smp; | |
357 | ||
358 | struct io_submit_state submit_state; | |
359 | ||
206aefde JA |
360 | struct io_rings *rings; |
361 | ||
2b188cc1 | 362 | /* IO offload */ |
561fb04a | 363 | struct io_wq *io_wq; |
2aede0e4 JA |
364 | |
365 | /* | |
366 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
367 | * have access to the ->files | |
368 | */ | |
369 | struct task_struct *sqo_task; | |
370 | ||
371 | /* Only used for accounting purposes */ | |
372 | struct mm_struct *mm_account; | |
373 | ||
91d8f519 DZ |
374 | #ifdef CONFIG_BLK_CGROUP |
375 | struct cgroup_subsys_state *sqo_blkcg_css; | |
376 | #endif | |
377 | ||
534ca6d6 JA |
378 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
379 | ||
90554200 | 380 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 381 | struct list_head sqd_list; |
75b28aff | 382 | |
6b06314c JA |
383 | /* |
384 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
385 | * readers must ensure that ->refs is alive as long as the file* is | |
386 | * used. Only updated through io_uring_register(2). | |
387 | */ | |
269bbe5f | 388 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
389 | unsigned nr_user_files; |
390 | ||
edafccee JA |
391 | /* if used, fixed mapped user buffers */ |
392 | unsigned nr_user_bufs; | |
393 | struct io_mapped_ubuf *user_bufs; | |
394 | ||
2b188cc1 JA |
395 | struct user_struct *user; |
396 | ||
0b8c0ec7 | 397 | const struct cred *creds; |
181e448d | 398 | |
4ea33a97 JA |
399 | #ifdef CONFIG_AUDIT |
400 | kuid_t loginuid; | |
401 | unsigned int sessionid; | |
402 | #endif | |
403 | ||
0f158b4c JA |
404 | struct completion ref_comp; |
405 | struct completion sq_thread_comp; | |
206aefde JA |
406 | |
407 | #if defined(CONFIG_UNIX) | |
408 | struct socket *ring_sock; | |
409 | #endif | |
410 | ||
5a2e745d JA |
411 | struct idr io_buffer_idr; |
412 | ||
071698e1 JA |
413 | struct idr personality_idr; |
414 | ||
206aefde JA |
415 | struct { |
416 | unsigned cached_cq_tail; | |
417 | unsigned cq_entries; | |
418 | unsigned cq_mask; | |
419 | atomic_t cq_timeouts; | |
f010505b | 420 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 421 | unsigned long cq_check_overflow; |
206aefde JA |
422 | struct wait_queue_head cq_wait; |
423 | struct fasync_struct *cq_fasync; | |
424 | struct eventfd_ctx *cq_ev_fd; | |
425 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 426 | |
2b188cc1 JA |
427 | struct { |
428 | spinlock_t completion_lock; | |
e94f141b | 429 | |
def596e9 | 430 | /* |
540e32a0 | 431 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
432 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
433 | * For SQPOLL, only the single threaded io_sq_thread() will | |
434 | * manipulate the list, hence no extra locking is needed there. | |
435 | */ | |
540e32a0 | 436 | struct list_head iopoll_list; |
78076bb6 JA |
437 | struct hlist_head *cancel_hash; |
438 | unsigned cancel_hash_bits; | |
e94f141b | 439 | bool poll_multi_file; |
31b51510 | 440 | |
fcb323cc JA |
441 | spinlock_t inflight_lock; |
442 | struct list_head inflight_list; | |
2b188cc1 | 443 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 444 | |
269bbe5f BM |
445 | struct delayed_work rsrc_put_work; |
446 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
447 | struct list_head rsrc_ref_list; |
448 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 449 | |
21b55dbc | 450 | struct io_restriction restrictions; |
3c1a2ead JA |
451 | |
452 | /* Keep this last, we don't need it for the fast path */ | |
453 | struct work_struct exit_work; | |
2b188cc1 JA |
454 | }; |
455 | ||
09bb8394 JA |
456 | /* |
457 | * First field must be the file pointer in all the | |
458 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
459 | */ | |
221c5eb2 JA |
460 | struct io_poll_iocb { |
461 | struct file *file; | |
018043be | 462 | struct wait_queue_head *head; |
221c5eb2 | 463 | __poll_t events; |
8c838788 | 464 | bool done; |
221c5eb2 | 465 | bool canceled; |
392edb45 | 466 | struct wait_queue_entry wait; |
221c5eb2 JA |
467 | }; |
468 | ||
018043be PB |
469 | struct io_poll_remove { |
470 | struct file *file; | |
471 | u64 addr; | |
472 | }; | |
473 | ||
b5dba59e JA |
474 | struct io_close { |
475 | struct file *file; | |
b5dba59e JA |
476 | int fd; |
477 | }; | |
478 | ||
ad8a48ac JA |
479 | struct io_timeout_data { |
480 | struct io_kiocb *req; | |
481 | struct hrtimer timer; | |
482 | struct timespec64 ts; | |
483 | enum hrtimer_mode mode; | |
484 | }; | |
485 | ||
8ed8d3c3 JA |
486 | struct io_accept { |
487 | struct file *file; | |
488 | struct sockaddr __user *addr; | |
489 | int __user *addr_len; | |
490 | int flags; | |
09952e3e | 491 | unsigned long nofile; |
8ed8d3c3 JA |
492 | }; |
493 | ||
494 | struct io_sync { | |
495 | struct file *file; | |
496 | loff_t len; | |
497 | loff_t off; | |
498 | int flags; | |
d63d1b5e | 499 | int mode; |
8ed8d3c3 JA |
500 | }; |
501 | ||
fbf23849 JA |
502 | struct io_cancel { |
503 | struct file *file; | |
504 | u64 addr; | |
505 | }; | |
506 | ||
b29472ee JA |
507 | struct io_timeout { |
508 | struct file *file; | |
bfe68a22 PB |
509 | u32 off; |
510 | u32 target_seq; | |
135fcde8 | 511 | struct list_head list; |
90cd7e42 PB |
512 | /* head of the link, used by linked timeouts only */ |
513 | struct io_kiocb *head; | |
b29472ee JA |
514 | }; |
515 | ||
0bdf7a2d PB |
516 | struct io_timeout_rem { |
517 | struct file *file; | |
518 | u64 addr; | |
9c8e11b3 PB |
519 | |
520 | /* timeout update */ | |
521 | struct timespec64 ts; | |
522 | u32 flags; | |
0bdf7a2d PB |
523 | }; |
524 | ||
9adbd45d JA |
525 | struct io_rw { |
526 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
527 | struct kiocb kiocb; | |
528 | u64 addr; | |
529 | u64 len; | |
530 | }; | |
531 | ||
3fbb51c1 JA |
532 | struct io_connect { |
533 | struct file *file; | |
534 | struct sockaddr __user *addr; | |
535 | int addr_len; | |
536 | }; | |
537 | ||
e47293fd JA |
538 | struct io_sr_msg { |
539 | struct file *file; | |
fddaface | 540 | union { |
270a5940 | 541 | struct user_msghdr __user *umsg; |
fddaface JA |
542 | void __user *buf; |
543 | }; | |
e47293fd | 544 | int msg_flags; |
bcda7baa | 545 | int bgid; |
fddaface | 546 | size_t len; |
bcda7baa | 547 | struct io_buffer *kbuf; |
e47293fd JA |
548 | }; |
549 | ||
15b71abe JA |
550 | struct io_open { |
551 | struct file *file; | |
552 | int dfd; | |
15b71abe | 553 | struct filename *filename; |
c12cedf2 | 554 | struct open_how how; |
4022e7af | 555 | unsigned long nofile; |
15b71abe JA |
556 | }; |
557 | ||
269bbe5f | 558 | struct io_rsrc_update { |
05f3fb3c JA |
559 | struct file *file; |
560 | u64 arg; | |
561 | u32 nr_args; | |
562 | u32 offset; | |
563 | }; | |
564 | ||
4840e418 JA |
565 | struct io_fadvise { |
566 | struct file *file; | |
567 | u64 offset; | |
568 | u32 len; | |
569 | u32 advice; | |
570 | }; | |
571 | ||
c1ca757b JA |
572 | struct io_madvise { |
573 | struct file *file; | |
574 | u64 addr; | |
575 | u32 len; | |
576 | u32 advice; | |
577 | }; | |
578 | ||
3e4827b0 JA |
579 | struct io_epoll { |
580 | struct file *file; | |
581 | int epfd; | |
582 | int op; | |
583 | int fd; | |
584 | struct epoll_event event; | |
e47293fd JA |
585 | }; |
586 | ||
7d67af2c PB |
587 | struct io_splice { |
588 | struct file *file_out; | |
589 | struct file *file_in; | |
590 | loff_t off_out; | |
591 | loff_t off_in; | |
592 | u64 len; | |
593 | unsigned int flags; | |
594 | }; | |
595 | ||
ddf0322d JA |
596 | struct io_provide_buf { |
597 | struct file *file; | |
598 | __u64 addr; | |
599 | __s32 len; | |
600 | __u32 bgid; | |
601 | __u16 nbufs; | |
602 | __u16 bid; | |
603 | }; | |
604 | ||
1d9e1288 BM |
605 | struct io_statx { |
606 | struct file *file; | |
607 | int dfd; | |
608 | unsigned int mask; | |
609 | unsigned int flags; | |
e62753e4 | 610 | const char __user *filename; |
1d9e1288 BM |
611 | struct statx __user *buffer; |
612 | }; | |
613 | ||
36f4fa68 JA |
614 | struct io_shutdown { |
615 | struct file *file; | |
616 | int how; | |
617 | }; | |
618 | ||
80a261fd JA |
619 | struct io_rename { |
620 | struct file *file; | |
621 | int old_dfd; | |
622 | int new_dfd; | |
623 | struct filename *oldpath; | |
624 | struct filename *newpath; | |
625 | int flags; | |
626 | }; | |
627 | ||
14a1143b JA |
628 | struct io_unlink { |
629 | struct file *file; | |
630 | int dfd; | |
631 | int flags; | |
632 | struct filename *filename; | |
633 | }; | |
634 | ||
3ca405eb PB |
635 | struct io_completion { |
636 | struct file *file; | |
637 | struct list_head list; | |
0f7e466b | 638 | int cflags; |
3ca405eb PB |
639 | }; |
640 | ||
f499a021 JA |
641 | struct io_async_connect { |
642 | struct sockaddr_storage address; | |
643 | }; | |
644 | ||
03b1230c JA |
645 | struct io_async_msghdr { |
646 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
647 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
648 | struct iovec *free_iov; | |
03b1230c JA |
649 | struct sockaddr __user *uaddr; |
650 | struct msghdr msg; | |
b537916c | 651 | struct sockaddr_storage addr; |
03b1230c JA |
652 | }; |
653 | ||
f67676d1 JA |
654 | struct io_async_rw { |
655 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
656 | const struct iovec *free_iovec; |
657 | struct iov_iter iter; | |
227c0c96 | 658 | size_t bytes_done; |
bcf5a063 | 659 | struct wait_page_queue wpq; |
f67676d1 JA |
660 | }; |
661 | ||
6b47ee6e PB |
662 | enum { |
663 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
664 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
665 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
666 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
667 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 668 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 669 | |
6b47ee6e PB |
670 | REQ_F_FAIL_LINK_BIT, |
671 | REQ_F_INFLIGHT_BIT, | |
672 | REQ_F_CUR_POS_BIT, | |
673 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 674 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 675 | REQ_F_ISREG_BIT, |
99bc4c38 | 676 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 677 | REQ_F_POLLED_BIT, |
bcda7baa | 678 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 679 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 680 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 681 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 682 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
683 | |
684 | /* not a real bit, just to check we're not overflowing the space */ | |
685 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
686 | }; |
687 | ||
688 | enum { | |
689 | /* ctx owns file */ | |
690 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
691 | /* drain existing IO first */ | |
692 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
693 | /* linked sqes */ | |
694 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
695 | /* doesn't sever on completion < 0 */ | |
696 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
697 | /* IOSQE_ASYNC */ | |
698 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
699 | /* IOSQE_BUFFER_SELECT */ |
700 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 701 | |
6b47ee6e PB |
702 | /* fail rest of links */ |
703 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
704 | /* on inflight list */ | |
705 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
706 | /* read/write uses file position */ | |
707 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
708 | /* must not punt to workers */ | |
709 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 710 | /* has or had linked timeout */ |
6b47ee6e | 711 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
712 | /* regular file */ |
713 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
714 | /* needs cleanup */ |
715 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
716 | /* already went through poll handler */ |
717 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
718 | /* buffer already selected */ |
719 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
720 | /* doesn't need file table for this request */ |
721 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
722 | /* io_wq_work is initialized */ |
723 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
724 | /* linked timeout is active, i.e. prepared by link's head */ |
725 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
726 | /* completion is deferred through io_comp_state */ |
727 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
728 | }; |
729 | ||
730 | struct async_poll { | |
731 | struct io_poll_iocb poll; | |
807abcb0 | 732 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
733 | }; |
734 | ||
7cbf1722 JA |
735 | struct io_task_work { |
736 | struct io_wq_work_node node; | |
737 | task_work_func_t func; | |
738 | }; | |
739 | ||
09bb8394 JA |
740 | /* |
741 | * NOTE! Each of the iocb union members has the file pointer | |
742 | * as the first entry in their struct definition. So you can | |
743 | * access the file pointer through any of the sub-structs, | |
744 | * or directly as just 'ki_filp' in this struct. | |
745 | */ | |
2b188cc1 | 746 | struct io_kiocb { |
221c5eb2 | 747 | union { |
09bb8394 | 748 | struct file *file; |
9adbd45d | 749 | struct io_rw rw; |
221c5eb2 | 750 | struct io_poll_iocb poll; |
018043be | 751 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
752 | struct io_accept accept; |
753 | struct io_sync sync; | |
fbf23849 | 754 | struct io_cancel cancel; |
b29472ee | 755 | struct io_timeout timeout; |
0bdf7a2d | 756 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 757 | struct io_connect connect; |
e47293fd | 758 | struct io_sr_msg sr_msg; |
15b71abe | 759 | struct io_open open; |
b5dba59e | 760 | struct io_close close; |
269bbe5f | 761 | struct io_rsrc_update rsrc_update; |
4840e418 | 762 | struct io_fadvise fadvise; |
c1ca757b | 763 | struct io_madvise madvise; |
3e4827b0 | 764 | struct io_epoll epoll; |
7d67af2c | 765 | struct io_splice splice; |
ddf0322d | 766 | struct io_provide_buf pbuf; |
1d9e1288 | 767 | struct io_statx statx; |
36f4fa68 | 768 | struct io_shutdown shutdown; |
80a261fd | 769 | struct io_rename rename; |
14a1143b | 770 | struct io_unlink unlink; |
3ca405eb PB |
771 | /* use only after cleaning per-op data, see io_clean_op() */ |
772 | struct io_completion compl; | |
221c5eb2 | 773 | }; |
2b188cc1 | 774 | |
e8c2bc1f JA |
775 | /* opcode allocated if it needs to store data for async defer */ |
776 | void *async_data; | |
d625c6ee | 777 | u8 opcode; |
65a6543d XW |
778 | /* polled IO has completed */ |
779 | u8 iopoll_completed; | |
2b188cc1 | 780 | |
4f4eeba8 | 781 | u16 buf_index; |
9cf7c104 | 782 | u32 result; |
4f4eeba8 | 783 | |
010e8e6b PB |
784 | struct io_ring_ctx *ctx; |
785 | unsigned int flags; | |
786 | refcount_t refs; | |
787 | struct task_struct *task; | |
788 | u64 user_data; | |
d7718a9d | 789 | |
f2f87370 | 790 | struct io_kiocb *link; |
269bbe5f | 791 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 792 | |
d21ffe7e PB |
793 | /* |
794 | * 1. used with ctx->iopoll_list with reads/writes | |
795 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
796 | */ | |
010e8e6b | 797 | struct list_head inflight_entry; |
7cbf1722 JA |
798 | union { |
799 | struct io_task_work io_task_work; | |
800 | struct callback_head task_work; | |
801 | }; | |
010e8e6b PB |
802 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
803 | struct hlist_node hash_node; | |
804 | struct async_poll *apoll; | |
805 | struct io_wq_work work; | |
2b188cc1 | 806 | }; |
05589553 | 807 | |
27dc8338 PB |
808 | struct io_defer_entry { |
809 | struct list_head list; | |
810 | struct io_kiocb *req; | |
9cf7c104 | 811 | u32 seq; |
2b188cc1 JA |
812 | }; |
813 | ||
d3656344 | 814 | struct io_op_def { |
d3656344 JA |
815 | /* needs req->file assigned */ |
816 | unsigned needs_file : 1; | |
d3656344 JA |
817 | /* hash wq insertion if file is a regular file */ |
818 | unsigned hash_reg_file : 1; | |
819 | /* unbound wq insertion if file is a non-regular file */ | |
820 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
821 | /* opcode is not supported by this kernel */ |
822 | unsigned not_supported : 1; | |
8a72758c JA |
823 | /* set if opcode supports polled "wait" */ |
824 | unsigned pollin : 1; | |
825 | unsigned pollout : 1; | |
bcda7baa JA |
826 | /* op supports buffer selection */ |
827 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
828 | /* must always have async data allocated */ |
829 | unsigned needs_async_data : 1; | |
27926b68 JA |
830 | /* should block plug */ |
831 | unsigned plug : 1; | |
e8c2bc1f JA |
832 | /* size of async data needed, if any */ |
833 | unsigned short async_size; | |
0f203765 | 834 | unsigned work_flags; |
d3656344 JA |
835 | }; |
836 | ||
0918682b | 837 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
838 | [IORING_OP_NOP] = {}, |
839 | [IORING_OP_READV] = { | |
d3656344 JA |
840 | .needs_file = 1, |
841 | .unbound_nonreg_file = 1, | |
8a72758c | 842 | .pollin = 1, |
4d954c25 | 843 | .buffer_select = 1, |
e8c2bc1f | 844 | .needs_async_data = 1, |
27926b68 | 845 | .plug = 1, |
e8c2bc1f | 846 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 847 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 848 | }, |
0463b6c5 | 849 | [IORING_OP_WRITEV] = { |
d3656344 JA |
850 | .needs_file = 1, |
851 | .hash_reg_file = 1, | |
852 | .unbound_nonreg_file = 1, | |
8a72758c | 853 | .pollout = 1, |
e8c2bc1f | 854 | .needs_async_data = 1, |
27926b68 | 855 | .plug = 1, |
e8c2bc1f | 856 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
857 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
858 | IO_WQ_WORK_FSIZE, | |
d3656344 | 859 | }, |
0463b6c5 | 860 | [IORING_OP_FSYNC] = { |
d3656344 | 861 | .needs_file = 1, |
0f203765 | 862 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 863 | }, |
0463b6c5 | 864 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
865 | .needs_file = 1, |
866 | .unbound_nonreg_file = 1, | |
8a72758c | 867 | .pollin = 1, |
27926b68 | 868 | .plug = 1, |
e8c2bc1f | 869 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 870 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 871 | }, |
0463b6c5 | 872 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
873 | .needs_file = 1, |
874 | .hash_reg_file = 1, | |
875 | .unbound_nonreg_file = 1, | |
8a72758c | 876 | .pollout = 1, |
27926b68 | 877 | .plug = 1, |
e8c2bc1f | 878 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
879 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
880 | IO_WQ_WORK_MM, | |
d3656344 | 881 | }, |
0463b6c5 | 882 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
883 | .needs_file = 1, |
884 | .unbound_nonreg_file = 1, | |
885 | }, | |
0463b6c5 PB |
886 | [IORING_OP_POLL_REMOVE] = {}, |
887 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 888 | .needs_file = 1, |
0f203765 | 889 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 890 | }, |
0463b6c5 | 891 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
892 | .needs_file = 1, |
893 | .unbound_nonreg_file = 1, | |
8a72758c | 894 | .pollout = 1, |
e8c2bc1f JA |
895 | .needs_async_data = 1, |
896 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 897 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 898 | }, |
0463b6c5 | 899 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
900 | .needs_file = 1, |
901 | .unbound_nonreg_file = 1, | |
8a72758c | 902 | .pollin = 1, |
52de1fe1 | 903 | .buffer_select = 1, |
e8c2bc1f JA |
904 | .needs_async_data = 1, |
905 | .async_size = sizeof(struct io_async_msghdr), | |
10cad2c4 | 906 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 907 | }, |
0463b6c5 | 908 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
909 | .needs_async_data = 1, |
910 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 911 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 912 | }, |
9c8e11b3 PB |
913 | [IORING_OP_TIMEOUT_REMOVE] = { |
914 | /* used by timeout updates' prep() */ | |
915 | .work_flags = IO_WQ_WORK_MM, | |
916 | }, | |
0463b6c5 | 917 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
918 | .needs_file = 1, |
919 | .unbound_nonreg_file = 1, | |
8a72758c | 920 | .pollin = 1, |
0f203765 | 921 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 922 | }, |
0463b6c5 PB |
923 | [IORING_OP_ASYNC_CANCEL] = {}, |
924 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
925 | .needs_async_data = 1, |
926 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 927 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 928 | }, |
0463b6c5 | 929 | [IORING_OP_CONNECT] = { |
d3656344 JA |
930 | .needs_file = 1, |
931 | .unbound_nonreg_file = 1, | |
8a72758c | 932 | .pollout = 1, |
e8c2bc1f JA |
933 | .needs_async_data = 1, |
934 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 935 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 936 | }, |
0463b6c5 | 937 | [IORING_OP_FALLOCATE] = { |
d3656344 | 938 | .needs_file = 1, |
69228338 | 939 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 940 | }, |
0463b6c5 | 941 | [IORING_OP_OPENAT] = { |
0f203765 | 942 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
14587a46 | 943 | IO_WQ_WORK_FS | IO_WQ_WORK_MM, |
d3656344 | 944 | }, |
0463b6c5 | 945 | [IORING_OP_CLOSE] = { |
0f203765 | 946 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 947 | }, |
0463b6c5 | 948 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 949 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 950 | }, |
0463b6c5 | 951 | [IORING_OP_STATX] = { |
0f203765 JA |
952 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
953 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 954 | }, |
0463b6c5 | 955 | [IORING_OP_READ] = { |
3a6820f2 JA |
956 | .needs_file = 1, |
957 | .unbound_nonreg_file = 1, | |
8a72758c | 958 | .pollin = 1, |
bcda7baa | 959 | .buffer_select = 1, |
27926b68 | 960 | .plug = 1, |
e8c2bc1f | 961 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 962 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 963 | }, |
0463b6c5 | 964 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
965 | .needs_file = 1, |
966 | .unbound_nonreg_file = 1, | |
8a72758c | 967 | .pollout = 1, |
27926b68 | 968 | .plug = 1, |
e8c2bc1f | 969 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
970 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
971 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 972 | }, |
0463b6c5 | 973 | [IORING_OP_FADVISE] = { |
4840e418 | 974 | .needs_file = 1, |
0f203765 | 975 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 976 | }, |
0463b6c5 | 977 | [IORING_OP_MADVISE] = { |
0f203765 | 978 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 979 | }, |
0463b6c5 | 980 | [IORING_OP_SEND] = { |
fddaface JA |
981 | .needs_file = 1, |
982 | .unbound_nonreg_file = 1, | |
8a72758c | 983 | .pollout = 1, |
0f203765 | 984 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 985 | }, |
0463b6c5 | 986 | [IORING_OP_RECV] = { |
fddaface JA |
987 | .needs_file = 1, |
988 | .unbound_nonreg_file = 1, | |
8a72758c | 989 | .pollin = 1, |
bcda7baa | 990 | .buffer_select = 1, |
0f203765 | 991 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 992 | }, |
0463b6c5 | 993 | [IORING_OP_OPENAT2] = { |
0f203765 | 994 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
14587a46 | 995 | IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
cebdb986 | 996 | }, |
3e4827b0 JA |
997 | [IORING_OP_EPOLL_CTL] = { |
998 | .unbound_nonreg_file = 1, | |
0f203765 | 999 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 1000 | }, |
7d67af2c PB |
1001 | [IORING_OP_SPLICE] = { |
1002 | .needs_file = 1, | |
1003 | .hash_reg_file = 1, | |
1004 | .unbound_nonreg_file = 1, | |
0f203765 | 1005 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
1006 | }, |
1007 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1008 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1009 | [IORING_OP_TEE] = { |
1010 | .needs_file = 1, | |
1011 | .hash_reg_file = 1, | |
1012 | .unbound_nonreg_file = 1, | |
1013 | }, | |
36f4fa68 JA |
1014 | [IORING_OP_SHUTDOWN] = { |
1015 | .needs_file = 1, | |
1016 | }, | |
80a261fd JA |
1017 | [IORING_OP_RENAMEAT] = { |
1018 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1019 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1020 | }, | |
14a1143b JA |
1021 | [IORING_OP_UNLINKAT] = { |
1022 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1023 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1024 | }, | |
d3656344 JA |
1025 | }; |
1026 | ||
9936c7c2 PB |
1027 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1028 | struct task_struct *task, | |
1029 | struct files_struct *files); | |
269bbe5f | 1030 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1031 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1032 | struct io_ring_ctx *ctx); |
bc9744cd PB |
1033 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
1034 | struct fixed_rsrc_ref_node *ref_node); | |
1ffc5422 | 1035 | |
23faba36 | 1036 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1037 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1038 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1039 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1040 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1041 | static void io_dismantle_req(struct io_kiocb *req); |
1042 | static void io_put_task(struct task_struct *task, int nr); | |
1043 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1044 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1045 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1046 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1047 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1048 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1049 | unsigned nr_args); |
3ca405eb | 1050 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1051 | static struct file *io_file_get(struct io_submit_state *state, |
1052 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1053 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1054 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1055 | |
847595de PB |
1056 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1057 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1058 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1059 | const struct iovec *fast_iov, | |
227c0c96 | 1060 | struct iov_iter *iter, bool force); |
907d1df3 | 1061 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1062 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1063 | struct io_ring_ctx *ctx); | |
de0617e4 | 1064 | |
2b188cc1 JA |
1065 | static struct kmem_cache *req_cachep; |
1066 | ||
0918682b | 1067 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1068 | |
1069 | struct sock *io_uring_get_socket(struct file *file) | |
1070 | { | |
1071 | #if defined(CONFIG_UNIX) | |
1072 | if (file->f_op == &io_uring_fops) { | |
1073 | struct io_ring_ctx *ctx = file->private_data; | |
1074 | ||
1075 | return ctx->ring_sock->sk; | |
1076 | } | |
1077 | #endif | |
1078 | return NULL; | |
1079 | } | |
1080 | EXPORT_SYMBOL(io_uring_get_socket); | |
1081 | ||
f2f87370 PB |
1082 | #define io_for_each_link(pos, head) \ |
1083 | for (pos = (head); pos; pos = pos->link) | |
1084 | ||
3ca405eb PB |
1085 | static inline void io_clean_op(struct io_kiocb *req) |
1086 | { | |
9d5c8190 | 1087 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1088 | __io_clean_op(req); |
1089 | } | |
1090 | ||
36f72fe2 PB |
1091 | static inline void io_set_resource_node(struct io_kiocb *req) |
1092 | { | |
1093 | struct io_ring_ctx *ctx = req->ctx; | |
1094 | ||
269bbe5f BM |
1095 | if (!req->fixed_rsrc_refs) { |
1096 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1097 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1098 | } |
1099 | } | |
1100 | ||
08d23634 PB |
1101 | static bool io_match_task(struct io_kiocb *head, |
1102 | struct task_struct *task, | |
1103 | struct files_struct *files) | |
1104 | { | |
1105 | struct io_kiocb *req; | |
1106 | ||
84965ff8 JA |
1107 | if (task && head->task != task) { |
1108 | /* in terms of cancelation, always match if req task is dead */ | |
1109 | if (head->task->flags & PF_EXITING) | |
1110 | return true; | |
08d23634 | 1111 | return false; |
84965ff8 | 1112 | } |
08d23634 PB |
1113 | if (!files) |
1114 | return true; | |
1115 | ||
1116 | io_for_each_link(req, head) { | |
02a13674 JA |
1117 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1118 | continue; | |
1119 | if (req->file && req->file->f_op == &io_uring_fops) | |
1120 | return true; | |
1121 | if ((req->work.flags & IO_WQ_WORK_FILES) && | |
08d23634 PB |
1122 | req->work.identity->files == files) |
1123 | return true; | |
1124 | } | |
1125 | return false; | |
1126 | } | |
1127 | ||
28cea78a | 1128 | static void io_sq_thread_drop_mm_files(void) |
c40f6379 | 1129 | { |
28cea78a | 1130 | struct files_struct *files = current->files; |
c40f6379 JA |
1131 | struct mm_struct *mm = current->mm; |
1132 | ||
1133 | if (mm) { | |
1134 | kthread_unuse_mm(mm); | |
1135 | mmput(mm); | |
4b70cf9d | 1136 | current->mm = NULL; |
c40f6379 | 1137 | } |
28cea78a JA |
1138 | if (files) { |
1139 | struct nsproxy *nsproxy = current->nsproxy; | |
1140 | ||
1141 | task_lock(current); | |
1142 | current->files = NULL; | |
1143 | current->nsproxy = NULL; | |
1144 | task_unlock(current); | |
1145 | put_files_struct(files); | |
1146 | put_nsproxy(nsproxy); | |
1147 | } | |
1148 | } | |
1149 | ||
1a38ffc9 | 1150 | static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx) |
28cea78a JA |
1151 | { |
1152 | if (!current->files) { | |
1153 | struct files_struct *files; | |
1154 | struct nsproxy *nsproxy; | |
1155 | ||
1156 | task_lock(ctx->sqo_task); | |
1157 | files = ctx->sqo_task->files; | |
1158 | if (!files) { | |
1159 | task_unlock(ctx->sqo_task); | |
1a38ffc9 | 1160 | return -EOWNERDEAD; |
28cea78a JA |
1161 | } |
1162 | atomic_inc(&files->count); | |
1163 | get_nsproxy(ctx->sqo_task->nsproxy); | |
1164 | nsproxy = ctx->sqo_task->nsproxy; | |
1165 | task_unlock(ctx->sqo_task); | |
1166 | ||
1167 | task_lock(current); | |
1168 | current->files = files; | |
1169 | current->nsproxy = nsproxy; | |
1170 | task_unlock(current); | |
1171 | } | |
1a38ffc9 | 1172 | return 0; |
c40f6379 JA |
1173 | } |
1174 | ||
1175 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1176 | { | |
4b70cf9d JA |
1177 | struct mm_struct *mm; |
1178 | ||
1179 | if (current->mm) | |
1180 | return 0; | |
1181 | ||
4b70cf9d JA |
1182 | task_lock(ctx->sqo_task); |
1183 | mm = ctx->sqo_task->mm; | |
1184 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1185 | mm = NULL; | |
1186 | task_unlock(ctx->sqo_task); | |
1187 | ||
1188 | if (mm) { | |
1189 | kthread_use_mm(mm); | |
1190 | return 0; | |
c40f6379 JA |
1191 | } |
1192 | ||
4b70cf9d | 1193 | return -EFAULT; |
c40f6379 JA |
1194 | } |
1195 | ||
4e326358 PB |
1196 | static int __io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1197 | struct io_kiocb *req) | |
c40f6379 | 1198 | { |
28cea78a | 1199 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1a38ffc9 | 1200 | int ret; |
28cea78a JA |
1201 | |
1202 | if (def->work_flags & IO_WQ_WORK_MM) { | |
1a38ffc9 | 1203 | ret = __io_sq_thread_acquire_mm(ctx); |
28cea78a JA |
1204 | if (unlikely(ret)) |
1205 | return ret; | |
1206 | } | |
1207 | ||
1a38ffc9 PB |
1208 | if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) { |
1209 | ret = __io_sq_thread_acquire_files(ctx); | |
1210 | if (unlikely(ret)) | |
1211 | return ret; | |
1212 | } | |
28cea78a JA |
1213 | |
1214 | return 0; | |
c40f6379 JA |
1215 | } |
1216 | ||
4e326358 PB |
1217 | static inline int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1218 | struct io_kiocb *req) | |
1219 | { | |
4e326358 PB |
1220 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
1221 | return 0; | |
1222 | return __io_sq_thread_acquire_mm_files(ctx, req); | |
1223 | } | |
1224 | ||
91d8f519 DZ |
1225 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1226 | struct cgroup_subsys_state **cur_css) | |
1227 | ||
1228 | { | |
1229 | #ifdef CONFIG_BLK_CGROUP | |
1230 | /* puts the old one when swapping */ | |
1231 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1232 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1233 | *cur_css = ctx->sqo_blkcg_css; | |
1234 | } | |
1235 | #endif | |
1236 | } | |
1237 | ||
1238 | static void io_sq_thread_unassociate_blkcg(void) | |
1239 | { | |
1240 | #ifdef CONFIG_BLK_CGROUP | |
1241 | kthread_associate_blkcg(NULL); | |
1242 | #endif | |
1243 | } | |
1244 | ||
c40f6379 JA |
1245 | static inline void req_set_fail_links(struct io_kiocb *req) |
1246 | { | |
1247 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1248 | req->flags |= REQ_F_FAIL_LINK; | |
1249 | } | |
4a38aed2 | 1250 | |
1e6fa521 JA |
1251 | /* |
1252 | * None of these are dereferenced, they are simply used to check if any of | |
1253 | * them have changed. If we're under current and check they are still the | |
1254 | * same, we're fine to grab references to them for actual out-of-line use. | |
1255 | */ | |
1256 | static void io_init_identity(struct io_identity *id) | |
1257 | { | |
1258 | id->files = current->files; | |
1259 | id->mm = current->mm; | |
1260 | #ifdef CONFIG_BLK_CGROUP | |
1261 | rcu_read_lock(); | |
1262 | id->blkcg_css = blkcg_css(); | |
1263 | rcu_read_unlock(); | |
1264 | #endif | |
1265 | id->creds = current_cred(); | |
1266 | id->nsproxy = current->nsproxy; | |
1267 | id->fs = current->fs; | |
1268 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1269 | #ifdef CONFIG_AUDIT |
1270 | id->loginuid = current->loginuid; | |
1271 | id->sessionid = current->sessionid; | |
1272 | #endif | |
1e6fa521 JA |
1273 | refcount_set(&id->count, 1); |
1274 | } | |
1275 | ||
ec99ca6c PB |
1276 | static inline void __io_req_init_async(struct io_kiocb *req) |
1277 | { | |
1278 | memset(&req->work, 0, sizeof(req->work)); | |
1279 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1280 | } | |
1281 | ||
7cdaf587 XW |
1282 | /* |
1283 | * Note: must call io_req_init_async() for the first time you | |
1284 | * touch any members of io_wq_work. | |
1285 | */ | |
1286 | static inline void io_req_init_async(struct io_kiocb *req) | |
1287 | { | |
500a373d JA |
1288 | struct io_uring_task *tctx = current->io_uring; |
1289 | ||
7cdaf587 XW |
1290 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1291 | return; | |
1292 | ||
ec99ca6c | 1293 | __io_req_init_async(req); |
500a373d JA |
1294 | |
1295 | /* Grab a ref if this isn't our static identity */ | |
1296 | req->work.identity = tctx->identity; | |
1297 | if (tctx->identity != &tctx->__identity) | |
1298 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1299 | } |
1300 | ||
2b188cc1 JA |
1301 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1302 | { | |
1303 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1304 | ||
0f158b4c | 1305 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1306 | } |
1307 | ||
8eb7e2d0 PB |
1308 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1309 | { | |
1310 | return !req->timeout.off; | |
1311 | } | |
1312 | ||
2b188cc1 JA |
1313 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1314 | { | |
1315 | struct io_ring_ctx *ctx; | |
78076bb6 | 1316 | int hash_bits; |
2b188cc1 JA |
1317 | |
1318 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1319 | if (!ctx) | |
1320 | return NULL; | |
1321 | ||
78076bb6 JA |
1322 | /* |
1323 | * Use 5 bits less than the max cq entries, that should give us around | |
1324 | * 32 entries per hash list if totally full and uniformly spread. | |
1325 | */ | |
1326 | hash_bits = ilog2(p->cq_entries); | |
1327 | hash_bits -= 5; | |
1328 | if (hash_bits <= 0) | |
1329 | hash_bits = 1; | |
1330 | ctx->cancel_hash_bits = hash_bits; | |
1331 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1332 | GFP_KERNEL); | |
1333 | if (!ctx->cancel_hash) | |
1334 | goto err; | |
1335 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1336 | ||
21482896 | 1337 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1338 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1339 | goto err; | |
2b188cc1 JA |
1340 | |
1341 | ctx->flags = p->flags; | |
90554200 | 1342 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1343 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1344 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1345 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1346 | init_completion(&ctx->ref_comp); |
1347 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1348 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1349 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1350 | mutex_init(&ctx->uring_lock); |
1351 | init_waitqueue_head(&ctx->wait); | |
1352 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1353 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1354 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1355 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1356 | spin_lock_init(&ctx->inflight_lock); |
1357 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1358 | spin_lock_init(&ctx->rsrc_ref_lock); |
1359 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1360 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1361 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1362 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1363 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1364 | return ctx; |
206aefde | 1365 | err: |
78076bb6 | 1366 | kfree(ctx->cancel_hash); |
206aefde JA |
1367 | kfree(ctx); |
1368 | return NULL; | |
2b188cc1 JA |
1369 | } |
1370 | ||
9cf7c104 | 1371 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1372 | { |
2bc9930e JA |
1373 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1374 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1375 | |
9cf7c104 | 1376 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1377 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1378 | } |
de0617e4 | 1379 | |
9d858b21 | 1380 | return false; |
de0617e4 JA |
1381 | } |
1382 | ||
5c3462cf | 1383 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1384 | { |
500a373d | 1385 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1386 | return; |
1387 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1388 | kfree(req->work.identity); | |
1389 | } | |
1390 | ||
4edf20f9 | 1391 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1392 | { |
7cdaf587 | 1393 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1394 | return; |
51a4cc11 | 1395 | |
e86d0047 | 1396 | if (req->work.flags & IO_WQ_WORK_MM) |
98447d65 | 1397 | mmdrop(req->work.identity->mm); |
91d8f519 | 1398 | #ifdef CONFIG_BLK_CGROUP |
e86d0047 | 1399 | if (req->work.flags & IO_WQ_WORK_BLKCG) |
98447d65 | 1400 | css_put(req->work.identity->blkcg_css); |
91d8f519 | 1401 | #endif |
e86d0047 | 1402 | if (req->work.flags & IO_WQ_WORK_CREDS) |
98447d65 | 1403 | put_cred(req->work.identity->creds); |
dfead8a8 | 1404 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1405 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1406 | |
98447d65 | 1407 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1408 | if (--fs->users) |
1409 | fs = NULL; | |
98447d65 | 1410 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1411 | if (fs) |
1412 | free_fs_struct(fs); | |
1413 | } | |
34e08fed PB |
1414 | if (req->work.flags & IO_WQ_WORK_FILES) { |
1415 | put_files_struct(req->work.identity->files); | |
1416 | put_nsproxy(req->work.identity->nsproxy); | |
34e08fed PB |
1417 | } |
1418 | if (req->flags & REQ_F_INFLIGHT) { | |
1419 | struct io_ring_ctx *ctx = req->ctx; | |
1420 | struct io_uring_task *tctx = req->task->io_uring; | |
1421 | unsigned long flags; | |
1422 | ||
1423 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1424 | list_del(&req->inflight_entry); | |
1425 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1426 | req->flags &= ~REQ_F_INFLIGHT; | |
1427 | if (atomic_read(&tctx->in_idle)) | |
1428 | wake_up(&tctx->wait); | |
1429 | } | |
51a4cc11 | 1430 | |
e86d0047 PB |
1431 | req->flags &= ~REQ_F_WORK_INITIALIZED; |
1432 | req->work.flags &= ~(IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | IO_WQ_WORK_FS | | |
1433 | IO_WQ_WORK_CREDS | IO_WQ_WORK_FILES); | |
5c3462cf | 1434 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1435 | } |
1436 | ||
1e6fa521 JA |
1437 | /* |
1438 | * Create a private copy of io_identity, since some fields don't match | |
1439 | * the current context. | |
1440 | */ | |
1441 | static bool io_identity_cow(struct io_kiocb *req) | |
1442 | { | |
5c3462cf | 1443 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1444 | const struct cred *creds = NULL; |
1445 | struct io_identity *id; | |
1446 | ||
1447 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1448 | creds = req->work.identity->creds; | |
1449 | ||
1450 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1451 | if (unlikely(!id)) { | |
1452 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1453 | return false; | |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * We can safely just re-init the creds we copied Either the field | |
1458 | * matches the current one, or we haven't grabbed it yet. The only | |
1459 | * exception is ->creds, through registered personalities, so handle | |
1460 | * that one separately. | |
1461 | */ | |
1462 | io_init_identity(id); | |
1463 | if (creds) | |
e8c954df | 1464 | id->creds = creds; |
1e6fa521 JA |
1465 | |
1466 | /* add one for this request */ | |
1467 | refcount_inc(&id->count); | |
1468 | ||
cb8a8ae3 JA |
1469 | /* drop tctx and req identity references, if needed */ |
1470 | if (tctx->identity != &tctx->__identity && | |
1471 | refcount_dec_and_test(&tctx->identity->count)) | |
1472 | kfree(tctx->identity); | |
1473 | if (req->work.identity != &tctx->__identity && | |
1474 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1475 | kfree(req->work.identity); |
1476 | ||
1477 | req->work.identity = id; | |
500a373d | 1478 | tctx->identity = id; |
1e6fa521 JA |
1479 | return true; |
1480 | } | |
1481 | ||
ce3d5aae PB |
1482 | static void io_req_track_inflight(struct io_kiocb *req) |
1483 | { | |
1484 | struct io_ring_ctx *ctx = req->ctx; | |
1485 | ||
1486 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
1487 | io_req_init_async(req); | |
1488 | req->flags |= REQ_F_INFLIGHT; | |
1489 | ||
1490 | spin_lock_irq(&ctx->inflight_lock); | |
1491 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1492 | spin_unlock_irq(&ctx->inflight_lock); | |
1493 | } | |
1494 | } | |
1495 | ||
1e6fa521 | 1496 | static bool io_grab_identity(struct io_kiocb *req) |
18d9be1a | 1497 | { |
d3656344 | 1498 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1499 | struct io_identity *id = req->work.identity; |
54a91f3b | 1500 | |
69228338 JA |
1501 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1502 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1503 | return false; | |
1504 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1505 | } | |
91d8f519 | 1506 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1507 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1508 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1509 | rcu_read_lock(); |
1e6fa521 JA |
1510 | if (id->blkcg_css != blkcg_css()) { |
1511 | rcu_read_unlock(); | |
1512 | return false; | |
1513 | } | |
91d8f519 DZ |
1514 | /* |
1515 | * This should be rare, either the cgroup is dying or the task | |
1516 | * is moving cgroups. Just punt to root for the handful of ios. | |
1517 | */ | |
1e6fa521 | 1518 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1519 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1520 | rcu_read_unlock(); |
1521 | } | |
1522 | #endif | |
dfead8a8 | 1523 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1524 | if (id->creds != current_cred()) |
1525 | return false; | |
1526 | get_cred(id->creds); | |
dfead8a8 JA |
1527 | req->work.flags |= IO_WQ_WORK_CREDS; |
1528 | } | |
4ea33a97 JA |
1529 | #ifdef CONFIG_AUDIT |
1530 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1531 | current->sessionid != id->sessionid) | |
1532 | return false; | |
1533 | #endif | |
dfead8a8 JA |
1534 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1535 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1536 | if (current->fs != id->fs) |
1537 | return false; | |
1538 | spin_lock(&id->fs->lock); | |
1539 | if (!id->fs->in_exec) { | |
1540 | id->fs->users++; | |
dfead8a8 | 1541 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1542 | } else { |
1543 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1544 | } | |
1545 | spin_unlock(¤t->fs->lock); | |
1546 | } | |
af604703 PB |
1547 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1548 | (def->work_flags & IO_WQ_WORK_FILES) && | |
1549 | !(req->flags & REQ_F_NO_FILE_TABLE)) { | |
1550 | if (id->files != current->files || | |
1551 | id->nsproxy != current->nsproxy) | |
1552 | return false; | |
1553 | atomic_inc(&id->files->count); | |
1554 | get_nsproxy(id->nsproxy); | |
af604703 | 1555 | req->work.flags |= IO_WQ_WORK_FILES; |
ce3d5aae | 1556 | io_req_track_inflight(req); |
af604703 | 1557 | } |
77788775 JA |
1558 | if (!(req->work.flags & IO_WQ_WORK_MM) && |
1559 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1560 | if (id->mm != current->mm) | |
1561 | return false; | |
1562 | mmgrab(id->mm); | |
1563 | req->work.flags |= IO_WQ_WORK_MM; | |
1564 | } | |
1e6fa521 JA |
1565 | |
1566 | return true; | |
1567 | } | |
1568 | ||
1569 | static void io_prep_async_work(struct io_kiocb *req) | |
1570 | { | |
1571 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1572 | struct io_ring_ctx *ctx = req->ctx; |
1573 | ||
1574 | io_req_init_async(req); | |
1575 | ||
feaadc4f PB |
1576 | if (req->flags & REQ_F_FORCE_ASYNC) |
1577 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1578 | ||
1e6fa521 JA |
1579 | if (req->flags & REQ_F_ISREG) { |
1580 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1581 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1582 | } else { | |
1583 | if (def->unbound_nonreg_file) | |
1584 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1585 | } | |
1586 | ||
1e6fa521 JA |
1587 | /* if we fail grabbing identity, we must COW, regrab, and retry */ |
1588 | if (io_grab_identity(req)) | |
1589 | return; | |
1590 | ||
1591 | if (!io_identity_cow(req)) | |
1592 | return; | |
1593 | ||
1594 | /* can't fail at this point */ | |
1595 | if (!io_grab_identity(req)) | |
1596 | WARN_ON(1); | |
561fb04a | 1597 | } |
cccf0ee8 | 1598 | |
cbdcb435 | 1599 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1600 | { |
cbdcb435 | 1601 | struct io_kiocb *cur; |
54a91f3b | 1602 | |
f2f87370 PB |
1603 | io_for_each_link(cur, req) |
1604 | io_prep_async_work(cur); | |
561fb04a JA |
1605 | } |
1606 | ||
7271ef3a | 1607 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1608 | { |
a197f664 | 1609 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1610 | struct io_kiocb *link = io_prep_linked_timeout(req); |
561fb04a | 1611 | |
8766dd51 PB |
1612 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1613 | &req->work, req->flags); | |
1614 | io_wq_enqueue(ctx->io_wq, &req->work); | |
7271ef3a | 1615 | return link; |
18d9be1a JA |
1616 | } |
1617 | ||
cbdcb435 PB |
1618 | static void io_queue_async_work(struct io_kiocb *req) |
1619 | { | |
7271ef3a JA |
1620 | struct io_kiocb *link; |
1621 | ||
cbdcb435 PB |
1622 | /* init ->work of the whole link before punting */ |
1623 | io_prep_async_link(req); | |
7271ef3a JA |
1624 | link = __io_queue_async_work(req); |
1625 | ||
1626 | if (link) | |
1627 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1628 | } |
1629 | ||
5262f567 JA |
1630 | static void io_kill_timeout(struct io_kiocb *req) |
1631 | { | |
e8c2bc1f | 1632 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1633 | int ret; |
1634 | ||
e8c2bc1f | 1635 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1636 | if (ret != -1) { |
01cec8c1 PB |
1637 | atomic_set(&req->ctx->cq_timeouts, |
1638 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1639 | list_del_init(&req->timeout.list); |
78e19bbe | 1640 | io_cqring_fill_event(req, 0); |
216578e5 | 1641 | io_put_req_deferred(req, 1); |
5262f567 JA |
1642 | } |
1643 | } | |
1644 | ||
76e1b642 JA |
1645 | /* |
1646 | * Returns true if we found and killed one or more timeouts | |
1647 | */ | |
6b81928d PB |
1648 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1649 | struct files_struct *files) | |
5262f567 JA |
1650 | { |
1651 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1652 | int canceled = 0; |
5262f567 JA |
1653 | |
1654 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1655 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1656 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1657 | io_kill_timeout(req); |
76e1b642 JA |
1658 | canceled++; |
1659 | } | |
f3606e3a | 1660 | } |
5262f567 | 1661 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1662 | return canceled != 0; |
5262f567 JA |
1663 | } |
1664 | ||
04518945 | 1665 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1666 | { |
04518945 | 1667 | do { |
27dc8338 PB |
1668 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1669 | struct io_defer_entry, list); | |
de0617e4 | 1670 | |
9cf7c104 | 1671 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1672 | break; |
27dc8338 | 1673 | list_del_init(&de->list); |
907d1df3 | 1674 | io_req_task_queue(de->req); |
27dc8338 | 1675 | kfree(de); |
04518945 PB |
1676 | } while (!list_empty(&ctx->defer_list)); |
1677 | } | |
1678 | ||
360428f8 | 1679 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1680 | { |
f010505b MDG |
1681 | u32 seq; |
1682 | ||
1683 | if (list_empty(&ctx->timeout_list)) | |
1684 | return; | |
1685 | ||
1686 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1687 | ||
1688 | do { | |
1689 | u32 events_needed, events_got; | |
360428f8 | 1690 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1691 | struct io_kiocb, timeout.list); |
de0617e4 | 1692 | |
8eb7e2d0 | 1693 | if (io_is_timeout_noseq(req)) |
360428f8 | 1694 | break; |
f010505b MDG |
1695 | |
1696 | /* | |
1697 | * Since seq can easily wrap around over time, subtract | |
1698 | * the last seq at which timeouts were flushed before comparing. | |
1699 | * Assuming not more than 2^31-1 events have happened since, | |
1700 | * these subtractions won't have wrapped, so we can check if | |
1701 | * target is in [last_seq, current_seq] by comparing the two. | |
1702 | */ | |
1703 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1704 | events_got = seq - ctx->cq_last_tm_flush; | |
1705 | if (events_got < events_needed) | |
360428f8 | 1706 | break; |
bfe68a22 | 1707 | |
135fcde8 | 1708 | list_del_init(&req->timeout.list); |
5262f567 | 1709 | io_kill_timeout(req); |
f010505b MDG |
1710 | } while (!list_empty(&ctx->timeout_list)); |
1711 | ||
1712 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1713 | } |
5262f567 | 1714 | |
360428f8 PB |
1715 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1716 | { | |
1717 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1718 | |
1719 | /* order cqe stores with ring update */ | |
1720 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1721 | |
04518945 PB |
1722 | if (unlikely(!list_empty(&ctx->defer_list))) |
1723 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1724 | } |
1725 | ||
90554200 JA |
1726 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1727 | { | |
1728 | struct io_rings *r = ctx->rings; | |
1729 | ||
1730 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1731 | } | |
1732 | ||
888aae2e PB |
1733 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1734 | { | |
1735 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1736 | } | |
1737 | ||
2b188cc1 JA |
1738 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1739 | { | |
75b28aff | 1740 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1741 | unsigned tail; |
1742 | ||
115e12e5 SB |
1743 | /* |
1744 | * writes to the cq entry need to come after reading head; the | |
1745 | * control dependency is enough as we're using WRITE_ONCE to | |
1746 | * fill the cq entry | |
1747 | */ | |
888aae2e | 1748 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1749 | return NULL; |
1750 | ||
888aae2e | 1751 | tail = ctx->cached_cq_tail++; |
75b28aff | 1752 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1753 | } |
1754 | ||
f2842ab5 JA |
1755 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1756 | { | |
f0b493e6 JA |
1757 | if (!ctx->cq_ev_fd) |
1758 | return false; | |
7e55a19c SG |
1759 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1760 | return false; | |
f2842ab5 JA |
1761 | if (!ctx->eventfd_async) |
1762 | return true; | |
b41e9852 | 1763 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1764 | } |
1765 | ||
b41e9852 | 1766 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1767 | { |
b1445e59 PB |
1768 | /* see waitqueue_active() comment */ |
1769 | smp_mb(); | |
1770 | ||
1d7bb1d5 JA |
1771 | if (waitqueue_active(&ctx->wait)) |
1772 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1773 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1774 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1775 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1776 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1777 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1778 | wake_up_interruptible(&ctx->cq_wait); |
1779 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1780 | } | |
1d7bb1d5 JA |
1781 | } |
1782 | ||
80c18e4a PB |
1783 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1784 | { | |
b1445e59 PB |
1785 | /* see waitqueue_active() comment */ |
1786 | smp_mb(); | |
1787 | ||
80c18e4a PB |
1788 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1789 | if (waitqueue_active(&ctx->wait)) | |
1790 | wake_up(&ctx->wait); | |
1791 | } | |
1792 | if (io_should_trigger_evfd(ctx)) | |
1793 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1794 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1795 | wake_up_interruptible(&ctx->cq_wait); |
1796 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1797 | } | |
80c18e4a PB |
1798 | } |
1799 | ||
c4a2ed72 | 1800 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1801 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1802 | struct task_struct *tsk, | |
1803 | struct files_struct *files) | |
1d7bb1d5 JA |
1804 | { |
1805 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1806 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1807 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1808 | unsigned long flags; |
b18032bb | 1809 | bool all_flushed, posted; |
1d7bb1d5 JA |
1810 | LIST_HEAD(list); |
1811 | ||
e23de15f PB |
1812 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1813 | return false; | |
1d7bb1d5 | 1814 | |
b18032bb | 1815 | posted = false; |
1d7bb1d5 | 1816 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1817 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1818 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1819 | continue; |
1820 | ||
1d7bb1d5 JA |
1821 | cqe = io_get_cqring(ctx); |
1822 | if (!cqe && !force) | |
1823 | break; | |
1824 | ||
40d8ddd4 | 1825 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1826 | if (cqe) { |
1827 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1828 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1829 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1830 | } else { |
2c3bac6d | 1831 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1832 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1833 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1834 | } |
b18032bb | 1835 | posted = true; |
1d7bb1d5 JA |
1836 | } |
1837 | ||
09e88404 PB |
1838 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1839 | if (all_flushed) { | |
1840 | clear_bit(0, &ctx->sq_check_overflow); | |
1841 | clear_bit(0, &ctx->cq_check_overflow); | |
1842 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1843 | } | |
46930143 | 1844 | |
b18032bb JA |
1845 | if (posted) |
1846 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1847 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1848 | if (posted) |
1849 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1850 | |
1851 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1852 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1853 | list_del(&req->compl.list); | |
ec9c02ad | 1854 | io_put_req(req); |
1d7bb1d5 | 1855 | } |
c4a2ed72 | 1856 | |
09e88404 | 1857 | return all_flushed; |
1d7bb1d5 JA |
1858 | } |
1859 | ||
6c503150 PB |
1860 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1861 | struct task_struct *tsk, | |
1862 | struct files_struct *files) | |
1863 | { | |
1864 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1865 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1866 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1867 | mutex_lock(&ctx->uring_lock); | |
1868 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1869 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1870 | mutex_unlock(&ctx->uring_lock); | |
1871 | } | |
1872 | } | |
1873 | ||
bcda7baa | 1874 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1875 | { |
78e19bbe | 1876 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1877 | struct io_uring_cqe *cqe; |
1878 | ||
78e19bbe | 1879 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1880 | |
2b188cc1 JA |
1881 | /* |
1882 | * If we can't get a cq entry, userspace overflowed the | |
1883 | * submission (by quite a lot). Increment the overflow count in | |
1884 | * the ring. | |
1885 | */ | |
1886 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1887 | if (likely(cqe)) { |
78e19bbe | 1888 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1889 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1890 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1891 | } else if (ctx->cq_overflow_flushed || |
1892 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1893 | /* |
1894 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1895 | * then we cannot store the request for later flushing, we need | |
1896 | * to drop it on the floor. | |
1897 | */ | |
2c3bac6d PB |
1898 | ctx->cached_cq_overflow++; |
1899 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1900 | } else { |
ad3eb2c8 JA |
1901 | if (list_empty(&ctx->cq_overflow_list)) { |
1902 | set_bit(0, &ctx->sq_check_overflow); | |
1903 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1904 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1905 | } |
40d8ddd4 | 1906 | io_clean_op(req); |
1d7bb1d5 | 1907 | req->result = res; |
0f7e466b | 1908 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1909 | refcount_inc(&req->refs); |
1910 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1911 | } |
1912 | } | |
1913 | ||
bcda7baa JA |
1914 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1915 | { | |
1916 | __io_cqring_fill_event(req, res, 0); | |
1917 | } | |
1918 | ||
c7dae4ba JA |
1919 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1920 | unsigned int cflags) | |
2b188cc1 | 1921 | { |
78e19bbe | 1922 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1923 | unsigned long flags; |
1924 | ||
1925 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1926 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1927 | io_commit_cqring(ctx); |
c7dae4ba JA |
1928 | /* |
1929 | * If we're the last reference to this request, add to our locked | |
1930 | * free_list cache. | |
1931 | */ | |
1932 | if (refcount_dec_and_test(&req->refs)) { | |
1933 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1934 | ||
1935 | io_dismantle_req(req); | |
1936 | io_put_task(req->task, 1); | |
1937 | list_add(&req->compl.list, &cs->locked_free_list); | |
1938 | cs->locked_free_nr++; | |
1939 | } else | |
1940 | req = NULL; | |
2b188cc1 JA |
1941 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1942 | ||
8c838788 | 1943 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1944 | if (req) { |
1945 | io_queue_next(req); | |
1946 | percpu_ref_put(&ctx->refs); | |
1947 | } | |
2b188cc1 JA |
1948 | } |
1949 | ||
a38d68db | 1950 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1951 | unsigned int cflags) |
229a7b63 | 1952 | { |
a38d68db PB |
1953 | io_clean_op(req); |
1954 | req->result = res; | |
1955 | req->compl.cflags = cflags; | |
e342c807 | 1956 | req->flags |= REQ_F_COMPLETE_INLINE; |
a38d68db PB |
1957 | } |
1958 | ||
889fca73 PB |
1959 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1960 | long res, unsigned cflags) | |
a38d68db | 1961 | { |
889fca73 PB |
1962 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1963 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1964 | else |
c7dae4ba | 1965 | io_req_complete_post(req, res, cflags); |
e1e16097 JA |
1966 | } |
1967 | ||
a38d68db | 1968 | static inline void io_req_complete(struct io_kiocb *req, long res) |
bcda7baa | 1969 | { |
889fca73 | 1970 | __io_req_complete(req, 0, res, 0); |
bcda7baa JA |
1971 | } |
1972 | ||
c7dae4ba | 1973 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
2b188cc1 | 1974 | { |
c7dae4ba JA |
1975 | struct io_submit_state *state = &ctx->submit_state; |
1976 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1977 | struct io_kiocb *req = NULL; |
1b4c351f | 1978 | |
c7dae4ba JA |
1979 | /* |
1980 | * If we have more than a batch's worth of requests in our IRQ side | |
1981 | * locked cache, grab the lock and move them over to our submission | |
1982 | * side cache. | |
1983 | */ | |
1984 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1985 | spin_lock_irq(&ctx->completion_lock); | |
1986 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1987 | cs->locked_free_nr = 0; | |
1988 | spin_unlock_irq(&ctx->completion_lock); | |
1989 | } | |
1990 | ||
1991 | while (!list_empty(&cs->free_list)) { | |
1992 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1993 | compl.list); |
1994 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1995 | state->reqs[state->free_reqs++] = req; |
1996 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1997 | break; | |
1b4c351f JA |
1998 | } |
1999 | ||
e5d1bc0a PB |
2000 | return req != NULL; |
2001 | } | |
2002 | ||
2003 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
2004 | { | |
2005 | struct io_submit_state *state = &ctx->submit_state; | |
2006 | ||
2007 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
2008 | ||
f6b6c7d6 | 2009 | if (!state->free_reqs) { |
291b2821 | 2010 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
2011 | int ret; |
2012 | ||
c7dae4ba | 2013 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
2014 | goto got_req; |
2015 | ||
bf019da7 PB |
2016 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
2017 | state->reqs); | |
fd6fab2c JA |
2018 | |
2019 | /* | |
2020 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2021 | * retry single alloc to be on the safe side. | |
2022 | */ | |
2023 | if (unlikely(ret <= 0)) { | |
2024 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
2025 | if (!state->reqs[0]) | |
3893f39f | 2026 | return NULL; |
fd6fab2c JA |
2027 | ret = 1; |
2028 | } | |
291b2821 | 2029 | state->free_reqs = ret; |
2b188cc1 | 2030 | } |
e5d1bc0a | 2031 | got_req: |
291b2821 PB |
2032 | state->free_reqs--; |
2033 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
2034 | } |
2035 | ||
8da11c19 PB |
2036 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
2037 | bool fixed) | |
2038 | { | |
36f72fe2 | 2039 | if (!fixed) |
8da11c19 PB |
2040 | fput(file); |
2041 | } | |
2042 | ||
4edf20f9 | 2043 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2044 | { |
3ca405eb | 2045 | io_clean_op(req); |
929a3af9 | 2046 | |
e8c2bc1f JA |
2047 | if (req->async_data) |
2048 | kfree(req->async_data); | |
8da11c19 PB |
2049 | if (req->file) |
2050 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
2051 | if (req->fixed_rsrc_refs) |
2052 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 2053 | io_req_clean_work(req); |
e65ef56d JA |
2054 | } |
2055 | ||
7c660731 PB |
2056 | static inline void io_put_task(struct task_struct *task, int nr) |
2057 | { | |
2058 | struct io_uring_task *tctx = task->io_uring; | |
2059 | ||
2060 | percpu_counter_sub(&tctx->inflight, nr); | |
2061 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2062 | wake_up(&tctx->wait); | |
2063 | put_task_struct_many(task, nr); | |
2064 | } | |
2065 | ||
216578e5 | 2066 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2067 | { |
51a4cc11 | 2068 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2069 | |
216578e5 | 2070 | io_dismantle_req(req); |
7c660731 | 2071 | io_put_task(req->task, 1); |
e3bc8e9d | 2072 | |
3893f39f | 2073 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 2074 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
2075 | } |
2076 | ||
f2f87370 PB |
2077 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2078 | { | |
2079 | struct io_kiocb *nxt = req->link; | |
2080 | ||
2081 | req->link = nxt->link; | |
2082 | nxt->link = NULL; | |
2083 | } | |
2084 | ||
c9abd7ad | 2085 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 2086 | { |
a197f664 | 2087 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 2088 | struct io_kiocb *link; |
c9abd7ad PB |
2089 | bool cancelled = false; |
2090 | unsigned long flags; | |
7c86ffee | 2091 | |
c9abd7ad | 2092 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2093 | link = req->link; |
2094 | ||
900fad45 PB |
2095 | /* |
2096 | * Can happen if a linked timeout fired and link had been like | |
2097 | * req -> link t-out -> link t-out [-> ...] | |
2098 | */ | |
c9abd7ad PB |
2099 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
2100 | struct io_timeout_data *io = link->async_data; | |
2101 | int ret; | |
7c86ffee | 2102 | |
f2f87370 | 2103 | io_remove_next_linked(req); |
90cd7e42 | 2104 | link->timeout.head = NULL; |
c9abd7ad PB |
2105 | ret = hrtimer_try_to_cancel(&io->timer); |
2106 | if (ret != -1) { | |
2107 | io_cqring_fill_event(link, -ECANCELED); | |
2108 | io_commit_cqring(ctx); | |
2109 | cancelled = true; | |
2110 | } | |
2111 | } | |
7c86ffee | 2112 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 2113 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 2114 | |
c9abd7ad | 2115 | if (cancelled) { |
7c86ffee | 2116 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
2117 | io_put_req(link); |
2118 | } | |
7c86ffee PB |
2119 | } |
2120 | ||
9e645e11 | 2121 | |
d148ca4b | 2122 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 2123 | { |
f2f87370 | 2124 | struct io_kiocb *link, *nxt; |
2665abfd | 2125 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 2126 | unsigned long flags; |
9e645e11 | 2127 | |
d148ca4b | 2128 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2129 | link = req->link; |
2130 | req->link = NULL; | |
9e645e11 | 2131 | |
f2f87370 PB |
2132 | while (link) { |
2133 | nxt = link->link; | |
2134 | link->link = NULL; | |
2665abfd | 2135 | |
f2f87370 | 2136 | trace_io_uring_fail_link(req, link); |
7c86ffee | 2137 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
2138 | |
2139 | /* | |
2140 | * It's ok to free under spinlock as they're not linked anymore, | |
2141 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
2142 | * work.fs->lock. | |
2143 | */ | |
2144 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
2145 | io_put_req_deferred(link, 2); | |
2146 | else | |
2147 | io_double_put_req(link); | |
f2f87370 | 2148 | link = nxt; |
9e645e11 | 2149 | } |
2665abfd | 2150 | io_commit_cqring(ctx); |
216578e5 | 2151 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 2152 | |
2665abfd | 2153 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
2154 | } |
2155 | ||
3fa5e0f3 | 2156 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2157 | { |
7c86ffee PB |
2158 | if (req->flags & REQ_F_LINK_TIMEOUT) |
2159 | io_kill_linked_timeout(req); | |
944e58bf | 2160 | |
9e645e11 JA |
2161 | /* |
2162 | * If LINK is set, we have dependent requests in this chain. If we | |
2163 | * didn't fail this request, queue the first one up, moving any other | |
2164 | * dependencies to the next request. In case of failure, fail the rest | |
2165 | * of the chain. | |
2166 | */ | |
f2f87370 PB |
2167 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
2168 | struct io_kiocb *nxt = req->link; | |
2169 | ||
2170 | req->link = NULL; | |
2171 | return nxt; | |
2172 | } | |
9b5f7bd9 PB |
2173 | io_fail_links(req); |
2174 | return NULL; | |
4d7dd462 | 2175 | } |
9e645e11 | 2176 | |
f2f87370 | 2177 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2178 | { |
cdbff982 | 2179 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2180 | return NULL; |
2181 | return __io_req_find_next(req); | |
2182 | } | |
2183 | ||
7cbf1722 JA |
2184 | static bool __tctx_task_work(struct io_uring_task *tctx) |
2185 | { | |
65453d1e | 2186 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
2187 | struct io_wq_work_list list; |
2188 | struct io_wq_work_node *node; | |
2189 | ||
2190 | if (wq_list_empty(&tctx->task_list)) | |
2191 | return false; | |
2192 | ||
0b81e80c | 2193 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
2194 | list = tctx->task_list; |
2195 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2196 | spin_unlock_irq(&tctx->task_lock); |
7cbf1722 JA |
2197 | |
2198 | node = list.first; | |
2199 | while (node) { | |
2200 | struct io_wq_work_node *next = node->next; | |
65453d1e | 2201 | struct io_ring_ctx *this_ctx; |
7cbf1722 JA |
2202 | struct io_kiocb *req; |
2203 | ||
2204 | req = container_of(node, struct io_kiocb, io_task_work.node); | |
65453d1e | 2205 | this_ctx = req->ctx; |
7cbf1722 JA |
2206 | req->task_work.func(&req->task_work); |
2207 | node = next; | |
65453d1e JA |
2208 | |
2209 | if (!ctx) { | |
2210 | ctx = this_ctx; | |
2211 | } else if (ctx != this_ctx) { | |
2212 | mutex_lock(&ctx->uring_lock); | |
2213 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2214 | mutex_unlock(&ctx->uring_lock); | |
2215 | ctx = this_ctx; | |
2216 | } | |
2217 | } | |
2218 | ||
2219 | if (ctx && ctx->submit_state.comp.nr) { | |
2220 | mutex_lock(&ctx->uring_lock); | |
2221 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2222 | mutex_unlock(&ctx->uring_lock); | |
7cbf1722 JA |
2223 | } |
2224 | ||
2225 | return list.first != NULL; | |
2226 | } | |
2227 | ||
2228 | static void tctx_task_work(struct callback_head *cb) | |
2229 | { | |
2230 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); | |
2231 | ||
2232 | while (__tctx_task_work(tctx)) | |
2233 | cond_resched(); | |
2234 | ||
2235 | clear_bit(0, &tctx->task_state); | |
2236 | } | |
2237 | ||
2238 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
2239 | enum task_work_notify_mode notify) | |
2240 | { | |
2241 | struct io_uring_task *tctx = tsk->io_uring; | |
2242 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 2243 | unsigned long flags; |
7cbf1722 JA |
2244 | int ret; |
2245 | ||
2246 | WARN_ON_ONCE(!tctx); | |
2247 | ||
0b81e80c | 2248 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2249 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 2250 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2251 | |
2252 | /* task_work already pending, we're done */ | |
2253 | if (test_bit(0, &tctx->task_state) || | |
2254 | test_and_set_bit(0, &tctx->task_state)) | |
2255 | return 0; | |
2256 | ||
2257 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
2258 | return 0; | |
2259 | ||
2260 | /* | |
2261 | * Slow path - we failed, find and delete work. if the work is not | |
2262 | * in the list, it got run and we're fine. | |
2263 | */ | |
2264 | ret = 0; | |
0b81e80c | 2265 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
2266 | wq_list_for_each(node, prev, &tctx->task_list) { |
2267 | if (&req->io_task_work.node == node) { | |
2268 | wq_list_del(&tctx->task_list, node, prev); | |
2269 | ret = 1; | |
2270 | break; | |
2271 | } | |
2272 | } | |
0b81e80c | 2273 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2274 | clear_bit(0, &tctx->task_state); |
2275 | return ret; | |
2276 | } | |
2277 | ||
355fb9e2 | 2278 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
2279 | { |
2280 | struct task_struct *tsk = req->task; | |
2281 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2282 | enum task_work_notify_mode notify; |
2283 | int ret; | |
c2c4c83c | 2284 | |
6200b0ae JA |
2285 | if (tsk->flags & PF_EXITING) |
2286 | return -ESRCH; | |
2287 | ||
c2c4c83c | 2288 | /* |
0ba9c9ed JA |
2289 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2290 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2291 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2292 | * will do the job. | |
c2c4c83c | 2293 | */ |
91989c70 | 2294 | notify = TWA_NONE; |
355fb9e2 | 2295 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
2296 | notify = TWA_SIGNAL; |
2297 | ||
7cbf1722 | 2298 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
2299 | if (!ret) |
2300 | wake_up_process(tsk); | |
0ba9c9ed | 2301 | |
c2c4c83c JA |
2302 | return ret; |
2303 | } | |
2304 | ||
eab30c4d | 2305 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 2306 | task_work_func_t cb) |
eab30c4d PB |
2307 | { |
2308 | struct task_struct *tsk = io_wq_get_task(req->ctx->io_wq); | |
2309 | ||
2310 | init_task_work(&req->task_work, cb); | |
2311 | task_work_add(tsk, &req->task_work, TWA_NONE); | |
2312 | wake_up_process(tsk); | |
2313 | } | |
2314 | ||
c40f6379 JA |
2315 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2316 | { | |
2317 | struct io_ring_ctx *ctx = req->ctx; | |
2318 | ||
2319 | spin_lock_irq(&ctx->completion_lock); | |
2320 | io_cqring_fill_event(req, error); | |
2321 | io_commit_cqring(ctx); | |
2322 | spin_unlock_irq(&ctx->completion_lock); | |
2323 | ||
2324 | io_cqring_ev_posted(ctx); | |
2325 | req_set_fail_links(req); | |
2326 | io_double_put_req(req); | |
2327 | } | |
2328 | ||
2329 | static void io_req_task_cancel(struct callback_head *cb) | |
2330 | { | |
2331 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2332 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 JA |
2333 | |
2334 | __io_req_task_cancel(req, -ECANCELED); | |
87ceb6a6 | 2335 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2336 | } |
2337 | ||
2338 | static void __io_req_task_submit(struct io_kiocb *req) | |
2339 | { | |
2340 | struct io_ring_ctx *ctx = req->ctx; | |
2341 | ||
04fc6c80 | 2342 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2343 | mutex_lock(&ctx->uring_lock); |
dc0eced5 PB |
2344 | if (!ctx->sqo_dead && !(current->flags & PF_EXITING) && |
2345 | !io_sq_thread_acquire_mm_files(ctx, req)) | |
c5eef2b9 | 2346 | __io_queue_sqe(req); |
81b6d05c | 2347 | else |
c40f6379 | 2348 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2349 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2350 | } |
2351 | ||
2352 | static void io_req_task_submit(struct callback_head *cb) | |
2353 | { | |
2354 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2355 | ||
2356 | __io_req_task_submit(req); | |
2357 | } | |
2358 | ||
2359 | static void io_req_task_queue(struct io_kiocb *req) | |
2360 | { | |
c40f6379 JA |
2361 | int ret; |
2362 | ||
7cbf1722 | 2363 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2364 | ret = io_req_task_work_add(req); |
04fc6c80 PB |
2365 | if (unlikely(ret)) { |
2366 | percpu_ref_get(&req->ctx->refs); | |
eab30c4d | 2367 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
04fc6c80 | 2368 | } |
c40f6379 JA |
2369 | } |
2370 | ||
f2f87370 | 2371 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2372 | { |
9b5f7bd9 | 2373 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2374 | |
2375 | if (nxt) | |
906a8c3f | 2376 | io_req_task_queue(nxt); |
c69f8dbe JL |
2377 | } |
2378 | ||
c3524383 | 2379 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2380 | { |
c3524383 PB |
2381 | io_queue_next(req); |
2382 | __io_free_req(req); | |
2383 | } | |
8766dd51 | 2384 | |
2d6500d4 | 2385 | struct req_batch { |
5af1d13e PB |
2386 | struct task_struct *task; |
2387 | int task_refs; | |
1b4c351f | 2388 | int ctx_refs; |
2d6500d4 PB |
2389 | }; |
2390 | ||
5af1d13e PB |
2391 | static inline void io_init_req_batch(struct req_batch *rb) |
2392 | { | |
5af1d13e | 2393 | rb->task_refs = 0; |
9ae72463 | 2394 | rb->ctx_refs = 0; |
5af1d13e PB |
2395 | rb->task = NULL; |
2396 | } | |
2397 | ||
2d6500d4 PB |
2398 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2399 | struct req_batch *rb) | |
2400 | { | |
6e833d53 | 2401 | if (rb->task) |
7c660731 | 2402 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2403 | if (rb->ctx_refs) |
2404 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2405 | } |
2406 | ||
6ff119a6 PB |
2407 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2408 | struct io_submit_state *state) | |
2d6500d4 | 2409 | { |
f2f87370 | 2410 | io_queue_next(req); |
2d6500d4 | 2411 | |
e3bc8e9d | 2412 | if (req->task != rb->task) { |
7c660731 PB |
2413 | if (rb->task) |
2414 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2415 | rb->task = req->task; |
2416 | rb->task_refs = 0; | |
5af1d13e | 2417 | } |
e3bc8e9d | 2418 | rb->task_refs++; |
9ae72463 | 2419 | rb->ctx_refs++; |
5af1d13e | 2420 | |
4edf20f9 | 2421 | io_dismantle_req(req); |
bd759045 | 2422 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2423 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2424 | else |
2425 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2426 | } |
2427 | ||
905c172f PB |
2428 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2429 | struct io_ring_ctx *ctx) | |
2430 | { | |
2431 | int i, nr = cs->nr; | |
2432 | struct io_kiocb *req; | |
2433 | struct req_batch rb; | |
2434 | ||
2435 | io_init_req_batch(&rb); | |
2436 | spin_lock_irq(&ctx->completion_lock); | |
2437 | for (i = 0; i < nr; i++) { | |
2438 | req = cs->reqs[i]; | |
2439 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2440 | } | |
2441 | io_commit_cqring(ctx); | |
2442 | spin_unlock_irq(&ctx->completion_lock); | |
2443 | ||
2444 | io_cqring_ev_posted(ctx); | |
2445 | for (i = 0; i < nr; i++) { | |
2446 | req = cs->reqs[i]; | |
2447 | ||
2448 | /* submission and completion refs */ | |
2449 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2450 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2451 | } |
2452 | ||
2453 | io_req_free_batch_finish(ctx, &rb); | |
2454 | cs->nr = 0; | |
2455 | } | |
2456 | ||
ba816ad6 JA |
2457 | /* |
2458 | * Drop reference to request, return next in chain (if there is one) if this | |
2459 | * was the last reference to this request. | |
2460 | */ | |
9b5f7bd9 | 2461 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2462 | { |
9b5f7bd9 PB |
2463 | struct io_kiocb *nxt = NULL; |
2464 | ||
2a44f467 | 2465 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2466 | nxt = io_req_find_next(req); |
4d7dd462 | 2467 | __io_free_req(req); |
2a44f467 | 2468 | } |
9b5f7bd9 | 2469 | return nxt; |
2b188cc1 JA |
2470 | } |
2471 | ||
e65ef56d JA |
2472 | static void io_put_req(struct io_kiocb *req) |
2473 | { | |
2474 | if (refcount_dec_and_test(&req->refs)) | |
2475 | io_free_req(req); | |
2b188cc1 JA |
2476 | } |
2477 | ||
216578e5 PB |
2478 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2479 | { | |
2480 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2481 | ||
2482 | io_free_req(req); | |
2483 | } | |
2484 | ||
2485 | static void io_free_req_deferred(struct io_kiocb *req) | |
2486 | { | |
2487 | int ret; | |
2488 | ||
7cbf1722 | 2489 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2490 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2491 | if (unlikely(ret)) |
2492 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2493 | } |
2494 | ||
2495 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2496 | { | |
2497 | if (refcount_sub_and_test(refs, &req->refs)) | |
2498 | io_free_req_deferred(req); | |
2499 | } | |
2500 | ||
978db57e JA |
2501 | static void io_double_put_req(struct io_kiocb *req) |
2502 | { | |
2503 | /* drop both submit and complete references */ | |
2504 | if (refcount_sub_and_test(2, &req->refs)) | |
2505 | io_free_req(req); | |
2506 | } | |
2507 | ||
6c503150 | 2508 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2509 | { |
2510 | /* See comment at the top of this file */ | |
2511 | smp_rmb(); | |
e23de15f | 2512 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2513 | } |
2514 | ||
fb5ccc98 PB |
2515 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2516 | { | |
2517 | struct io_rings *rings = ctx->rings; | |
2518 | ||
2519 | /* make sure SQ entry isn't read before tail */ | |
2520 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2521 | } | |
2522 | ||
8ff069bf | 2523 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2524 | { |
8ff069bf | 2525 | unsigned int cflags; |
e94f141b | 2526 | |
bcda7baa JA |
2527 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2528 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2529 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2530 | kfree(kbuf); |
2531 | return cflags; | |
e94f141b JA |
2532 | } |
2533 | ||
8ff069bf | 2534 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2535 | { |
4d954c25 | 2536 | struct io_buffer *kbuf; |
bcda7baa | 2537 | |
4d954c25 | 2538 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2539 | return io_put_kbuf(req, kbuf); |
2540 | } | |
2541 | ||
4c6e277c JA |
2542 | static inline bool io_run_task_work(void) |
2543 | { | |
6200b0ae JA |
2544 | /* |
2545 | * Not safe to run on exiting task, and the task_work handling will | |
2546 | * not add work to such a task. | |
2547 | */ | |
2548 | if (unlikely(current->flags & PF_EXITING)) | |
2549 | return false; | |
4c6e277c JA |
2550 | if (current->task_works) { |
2551 | __set_current_state(TASK_RUNNING); | |
2552 | task_work_run(); | |
2553 | return true; | |
2554 | } | |
2555 | ||
2556 | return false; | |
bcda7baa JA |
2557 | } |
2558 | ||
def596e9 JA |
2559 | /* |
2560 | * Find and free completed poll iocbs | |
2561 | */ | |
2562 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2563 | struct list_head *done) | |
2564 | { | |
8237e045 | 2565 | struct req_batch rb; |
def596e9 | 2566 | struct io_kiocb *req; |
bbde017a XW |
2567 | |
2568 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2569 | smp_rmb(); | |
def596e9 | 2570 | |
5af1d13e | 2571 | io_init_req_batch(&rb); |
def596e9 | 2572 | while (!list_empty(done)) { |
bcda7baa JA |
2573 | int cflags = 0; |
2574 | ||
d21ffe7e | 2575 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2576 | list_del(&req->inflight_entry); |
2577 | ||
bbde017a XW |
2578 | if (READ_ONCE(req->result) == -EAGAIN) { |
2579 | req->iopoll_completed = 0; | |
23faba36 | 2580 | if (io_rw_reissue(req)) |
f161340d | 2581 | continue; |
bbde017a | 2582 | } |
def596e9 | 2583 | |
bcda7baa | 2584 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2585 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2586 | |
2587 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2588 | (*nr_events)++; |
2589 | ||
c3524383 | 2590 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2591 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2592 | } |
def596e9 | 2593 | |
09bb8394 | 2594 | io_commit_cqring(ctx); |
80c18e4a | 2595 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2596 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2597 | } |
2598 | ||
def596e9 JA |
2599 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2600 | long min) | |
2601 | { | |
2602 | struct io_kiocb *req, *tmp; | |
2603 | LIST_HEAD(done); | |
2604 | bool spin; | |
2605 | int ret; | |
2606 | ||
2607 | /* | |
2608 | * Only spin for completions if we don't have multiple devices hanging | |
2609 | * off our complete list, and we're under the requested amount. | |
2610 | */ | |
2611 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2612 | ||
2613 | ret = 0; | |
d21ffe7e | 2614 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2615 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2616 | |
2617 | /* | |
581f9810 BM |
2618 | * Move completed and retryable entries to our local lists. |
2619 | * If we find a request that requires polling, break out | |
2620 | * and complete those lists first, if we have entries there. | |
def596e9 | 2621 | */ |
65a6543d | 2622 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2623 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2624 | continue; |
2625 | } | |
2626 | if (!list_empty(&done)) | |
2627 | break; | |
2628 | ||
2629 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2630 | if (ret < 0) | |
2631 | break; | |
2632 | ||
3aadc23e PB |
2633 | /* iopoll may have completed current req */ |
2634 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2635 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2636 | |
def596e9 JA |
2637 | if (ret && spin) |
2638 | spin = false; | |
2639 | ret = 0; | |
2640 | } | |
2641 | ||
2642 | if (!list_empty(&done)) | |
2643 | io_iopoll_complete(ctx, nr_events, &done); | |
2644 | ||
2645 | return ret; | |
2646 | } | |
2647 | ||
2648 | /* | |
d195a66e | 2649 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2650 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2651 | * as a non-spinning completion check. | |
2652 | */ | |
2653 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2654 | long min) | |
2655 | { | |
540e32a0 | 2656 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2657 | int ret; |
2658 | ||
2659 | ret = io_do_iopoll(ctx, nr_events, min); | |
2660 | if (ret < 0) | |
2661 | return ret; | |
eba0a4dd | 2662 | if (*nr_events >= min) |
def596e9 JA |
2663 | return 0; |
2664 | } | |
2665 | ||
2666 | return 1; | |
2667 | } | |
2668 | ||
2669 | /* | |
2670 | * We can't just wait for polled events to come to us, we have to actively | |
2671 | * find and complete them. | |
2672 | */ | |
b2edc0a7 | 2673 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2674 | { |
2675 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2676 | return; | |
2677 | ||
2678 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2679 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2680 | unsigned int nr_events = 0; |
2681 | ||
b2edc0a7 | 2682 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2683 | |
b2edc0a7 PB |
2684 | /* let it sleep and repeat later if can't complete a request */ |
2685 | if (nr_events == 0) | |
2686 | break; | |
08f5439f JA |
2687 | /* |
2688 | * Ensure we allow local-to-the-cpu processing to take place, | |
2689 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2690 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2691 | */ |
3fcee5a6 PB |
2692 | if (need_resched()) { |
2693 | mutex_unlock(&ctx->uring_lock); | |
2694 | cond_resched(); | |
2695 | mutex_lock(&ctx->uring_lock); | |
2696 | } | |
def596e9 JA |
2697 | } |
2698 | mutex_unlock(&ctx->uring_lock); | |
2699 | } | |
2700 | ||
7668b92a | 2701 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2702 | { |
7668b92a | 2703 | unsigned int nr_events = 0; |
2b2ed975 | 2704 | int iters = 0, ret = 0; |
500f9fba | 2705 | |
c7849be9 XW |
2706 | /* |
2707 | * We disallow the app entering submit/complete with polling, but we | |
2708 | * still need to lock the ring to prevent racing with polled issue | |
2709 | * that got punted to a workqueue. | |
2710 | */ | |
2711 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2712 | do { |
a3a0e43f JA |
2713 | /* |
2714 | * Don't enter poll loop if we already have events pending. | |
2715 | * If we do, we can potentially be spinning for commands that | |
2716 | * already triggered a CQE (eg in error). | |
2717 | */ | |
6c503150 PB |
2718 | if (test_bit(0, &ctx->cq_check_overflow)) |
2719 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2720 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2721 | break; |
2722 | ||
500f9fba JA |
2723 | /* |
2724 | * If a submit got punted to a workqueue, we can have the | |
2725 | * application entering polling for a command before it gets | |
2726 | * issued. That app will hold the uring_lock for the duration | |
2727 | * of the poll right here, so we need to take a breather every | |
2728 | * now and then to ensure that the issue has a chance to add | |
2729 | * the poll to the issued list. Otherwise we can spin here | |
2730 | * forever, while the workqueue is stuck trying to acquire the | |
2731 | * very same mutex. | |
2732 | */ | |
2733 | if (!(++iters & 7)) { | |
2734 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2735 | io_run_task_work(); |
500f9fba JA |
2736 | mutex_lock(&ctx->uring_lock); |
2737 | } | |
2738 | ||
7668b92a | 2739 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2740 | if (ret <= 0) |
2741 | break; | |
2742 | ret = 0; | |
7668b92a | 2743 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2744 | |
500f9fba | 2745 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2746 | return ret; |
2747 | } | |
2748 | ||
491381ce | 2749 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2750 | { |
491381ce JA |
2751 | /* |
2752 | * Tell lockdep we inherited freeze protection from submission | |
2753 | * thread. | |
2754 | */ | |
2755 | if (req->flags & REQ_F_ISREG) { | |
2756 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2757 | |
491381ce | 2758 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2759 | } |
491381ce | 2760 | file_end_write(req->file); |
2b188cc1 JA |
2761 | } |
2762 | ||
b63534c4 | 2763 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2764 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2765 | { |
2766 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2767 | int rw, ret; |
b63534c4 | 2768 | struct iov_iter iter; |
b63534c4 | 2769 | |
dc2a6e9a PB |
2770 | /* already prepared */ |
2771 | if (req->async_data) | |
2772 | return true; | |
b63534c4 JA |
2773 | |
2774 | switch (req->opcode) { | |
2775 | case IORING_OP_READV: | |
2776 | case IORING_OP_READ_FIXED: | |
2777 | case IORING_OP_READ: | |
2778 | rw = READ; | |
2779 | break; | |
2780 | case IORING_OP_WRITEV: | |
2781 | case IORING_OP_WRITE_FIXED: | |
2782 | case IORING_OP_WRITE: | |
2783 | rw = WRITE; | |
2784 | break; | |
2785 | default: | |
2786 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2787 | req->opcode); | |
dc2a6e9a | 2788 | return false; |
b63534c4 JA |
2789 | } |
2790 | ||
dc2a6e9a PB |
2791 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2792 | if (ret < 0) | |
2793 | return false; | |
6bf985dc | 2794 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2795 | } |
b63534c4 JA |
2796 | #endif |
2797 | ||
23faba36 | 2798 | static bool io_rw_reissue(struct io_kiocb *req) |
b63534c4 JA |
2799 | { |
2800 | #ifdef CONFIG_BLOCK | |
23faba36 | 2801 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2802 | int ret; |
2803 | ||
75c668cd PB |
2804 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2805 | return false; | |
2806 | if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker()) | |
b63534c4 JA |
2807 | return false; |
2808 | ||
55e6ac1e PB |
2809 | lockdep_assert_held(&req->ctx->uring_lock); |
2810 | ||
28cea78a | 2811 | ret = io_sq_thread_acquire_mm_files(req->ctx, req); |
6d816e08 | 2812 | |
dc2a6e9a | 2813 | if (!ret && io_resubmit_prep(req)) { |
fdee946d JA |
2814 | refcount_inc(&req->refs); |
2815 | io_queue_async_work(req); | |
b63534c4 | 2816 | return true; |
fdee946d | 2817 | } |
dc2a6e9a | 2818 | req_set_fail_links(req); |
b63534c4 JA |
2819 | #endif |
2820 | return false; | |
2821 | } | |
2822 | ||
a1d7c393 | 2823 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2824 | unsigned int issue_flags) |
a1d7c393 | 2825 | { |
2f8e45f1 PB |
2826 | int cflags = 0; |
2827 | ||
23faba36 PB |
2828 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2829 | return; | |
2f8e45f1 PB |
2830 | if (res != req->result) |
2831 | req_set_fail_links(req); | |
23faba36 | 2832 | |
2f8e45f1 PB |
2833 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2834 | kiocb_end_write(req); | |
2835 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2836 | cflags = io_put_rw_kbuf(req); | |
2837 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2838 | } |
2839 | ||
2840 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2841 | { | |
9adbd45d | 2842 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2843 | |
889fca73 | 2844 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2845 | } |
2846 | ||
def596e9 JA |
2847 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2848 | { | |
9adbd45d | 2849 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2850 | |
491381ce JA |
2851 | if (kiocb->ki_flags & IOCB_WRITE) |
2852 | kiocb_end_write(req); | |
def596e9 | 2853 | |
2d7d6792 | 2854 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2855 | req_set_fail_links(req); |
bbde017a XW |
2856 | |
2857 | WRITE_ONCE(req->result, res); | |
2858 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2859 | smp_wmb(); |
2860 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2861 | } |
2862 | ||
2863 | /* | |
2864 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2865 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2866 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2867 | * accessing the kiocb cookie. | |
2868 | */ | |
2e9dbe90 | 2869 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2870 | { |
2871 | struct io_ring_ctx *ctx = req->ctx; | |
2872 | ||
2873 | /* | |
2874 | * Track whether we have multiple files in our lists. This will impact | |
2875 | * how we do polling eventually, not spinning if we're on potentially | |
2876 | * different devices. | |
2877 | */ | |
540e32a0 | 2878 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2879 | ctx->poll_multi_file = false; |
2880 | } else if (!ctx->poll_multi_file) { | |
2881 | struct io_kiocb *list_req; | |
2882 | ||
540e32a0 | 2883 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2884 | inflight_entry); |
9adbd45d | 2885 | if (list_req->file != req->file) |
def596e9 JA |
2886 | ctx->poll_multi_file = true; |
2887 | } | |
2888 | ||
2889 | /* | |
2890 | * For fast devices, IO may have already completed. If it has, add | |
2891 | * it to the front so we find it first. | |
2892 | */ | |
65a6543d | 2893 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2894 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2895 | else |
d21ffe7e | 2896 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2897 | |
2e9dbe90 XW |
2898 | /* |
2899 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2900 | * task context or in io worker task context. If current task context is | |
2901 | * sq thread, we don't need to check whether should wake up sq thread. | |
2902 | */ | |
2903 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2904 | wq_has_sleeper(&ctx->sq_data->wait)) |
2905 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2906 | } |
2907 | ||
9f13c35b PB |
2908 | static inline void io_state_file_put(struct io_submit_state *state) |
2909 | { | |
02b23a9a PB |
2910 | if (state->file_refs) { |
2911 | fput_many(state->file, state->file_refs); | |
2912 | state->file_refs = 0; | |
2913 | } | |
9a56a232 JA |
2914 | } |
2915 | ||
2916 | /* | |
2917 | * Get as many references to a file as we have IOs left in this submission, | |
2918 | * assuming most submissions are for one file, or at least that each file | |
2919 | * has more than one submission. | |
2920 | */ | |
8da11c19 | 2921 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2922 | { |
2923 | if (!state) | |
2924 | return fget(fd); | |
2925 | ||
6e1271e6 | 2926 | if (state->file_refs) { |
9a56a232 | 2927 | if (state->fd == fd) { |
6e1271e6 | 2928 | state->file_refs--; |
9a56a232 JA |
2929 | return state->file; |
2930 | } | |
02b23a9a | 2931 | io_state_file_put(state); |
9a56a232 JA |
2932 | } |
2933 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2934 | if (unlikely(!state->file)) |
9a56a232 JA |
2935 | return NULL; |
2936 | ||
2937 | state->fd = fd; | |
6e1271e6 | 2938 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2939 | return state->file; |
2940 | } | |
2941 | ||
4503b767 JA |
2942 | static bool io_bdev_nowait(struct block_device *bdev) |
2943 | { | |
9ba0d0c8 | 2944 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2945 | } |
2946 | ||
2b188cc1 JA |
2947 | /* |
2948 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2949 | * any file. For now, just ensure that anything potentially problematic is done | |
2950 | * inline. | |
2951 | */ | |
af197f50 | 2952 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2953 | { |
2954 | umode_t mode = file_inode(file)->i_mode; | |
2955 | ||
4503b767 | 2956 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2957 | if (IS_ENABLED(CONFIG_BLOCK) && |
2958 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2959 | return true; |
2960 | return false; | |
2961 | } | |
2962 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2963 | return true; |
4503b767 | 2964 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2965 | if (IS_ENABLED(CONFIG_BLOCK) && |
2966 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2967 | file->f_op != &io_uring_fops) |
2968 | return true; | |
2969 | return false; | |
2970 | } | |
2b188cc1 | 2971 | |
c5b85625 JA |
2972 | /* any ->read/write should understand O_NONBLOCK */ |
2973 | if (file->f_flags & O_NONBLOCK) | |
2974 | return true; | |
2975 | ||
af197f50 JA |
2976 | if (!(file->f_mode & FMODE_NOWAIT)) |
2977 | return false; | |
2978 | ||
2979 | if (rw == READ) | |
2980 | return file->f_op->read_iter != NULL; | |
2981 | ||
2982 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2983 | } |
2984 | ||
a88fc400 | 2985 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2986 | { |
def596e9 | 2987 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2988 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2989 | struct file *file = req->file; |
09bb8394 JA |
2990 | unsigned ioprio; |
2991 | int ret; | |
2b188cc1 | 2992 | |
75c668cd | 2993 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2994 | req->flags |= REQ_F_ISREG; |
2995 | ||
2b188cc1 | 2996 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2997 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2998 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2999 | kiocb->ki_pos = file->f_pos; |
ba04291e | 3000 | } |
2b188cc1 | 3001 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
3002 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
3003 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
3004 | if (unlikely(ret)) | |
3005 | return ret; | |
2b188cc1 | 3006 | |
75c668cd PB |
3007 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
3008 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
3009 | req->flags |= REQ_F_NOWAIT; | |
3010 | ||
2b188cc1 JA |
3011 | ioprio = READ_ONCE(sqe->ioprio); |
3012 | if (ioprio) { | |
3013 | ret = ioprio_check_cap(ioprio); | |
3014 | if (ret) | |
09bb8394 | 3015 | return ret; |
2b188cc1 JA |
3016 | |
3017 | kiocb->ki_ioprio = ioprio; | |
3018 | } else | |
3019 | kiocb->ki_ioprio = get_current_ioprio(); | |
3020 | ||
def596e9 | 3021 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
3022 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
3023 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 3024 | return -EOPNOTSUPP; |
2b188cc1 | 3025 | |
def596e9 JA |
3026 | kiocb->ki_flags |= IOCB_HIPRI; |
3027 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 3028 | req->iopoll_completed = 0; |
def596e9 | 3029 | } else { |
09bb8394 JA |
3030 | if (kiocb->ki_flags & IOCB_HIPRI) |
3031 | return -EINVAL; | |
def596e9 JA |
3032 | kiocb->ki_complete = io_complete_rw; |
3033 | } | |
9adbd45d | 3034 | |
3529d8c2 JA |
3035 | req->rw.addr = READ_ONCE(sqe->addr); |
3036 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 3037 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 3038 | return 0; |
2b188cc1 JA |
3039 | } |
3040 | ||
3041 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3042 | { | |
3043 | switch (ret) { | |
3044 | case -EIOCBQUEUED: | |
3045 | break; | |
3046 | case -ERESTARTSYS: | |
3047 | case -ERESTARTNOINTR: | |
3048 | case -ERESTARTNOHAND: | |
3049 | case -ERESTART_RESTARTBLOCK: | |
3050 | /* | |
3051 | * We can't just restart the syscall, since previously | |
3052 | * submitted sqes may already be in progress. Just fail this | |
3053 | * IO with EINTR. | |
3054 | */ | |
3055 | ret = -EINTR; | |
df561f66 | 3056 | fallthrough; |
2b188cc1 JA |
3057 | default: |
3058 | kiocb->ki_complete(kiocb, ret, 0); | |
3059 | } | |
3060 | } | |
3061 | ||
a1d7c393 | 3062 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 3063 | unsigned int issue_flags) |
ba816ad6 | 3064 | { |
ba04291e | 3065 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 3066 | struct io_async_rw *io = req->async_data; |
ba04291e | 3067 | |
227c0c96 | 3068 | /* add previously done IO, if any */ |
e8c2bc1f | 3069 | if (io && io->bytes_done > 0) { |
227c0c96 | 3070 | if (ret < 0) |
e8c2bc1f | 3071 | ret = io->bytes_done; |
227c0c96 | 3072 | else |
e8c2bc1f | 3073 | ret += io->bytes_done; |
227c0c96 JA |
3074 | } |
3075 | ||
ba04291e JA |
3076 | if (req->flags & REQ_F_CUR_POS) |
3077 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 3078 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 3079 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
3080 | else |
3081 | io_rw_done(kiocb, ret); | |
3082 | } | |
3083 | ||
847595de | 3084 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 3085 | { |
9adbd45d JA |
3086 | struct io_ring_ctx *ctx = req->ctx; |
3087 | size_t len = req->rw.len; | |
edafccee | 3088 | struct io_mapped_ubuf *imu; |
4be1c615 | 3089 | u16 index, buf_index = req->buf_index; |
edafccee JA |
3090 | size_t offset; |
3091 | u64 buf_addr; | |
3092 | ||
edafccee JA |
3093 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3094 | return -EFAULT; | |
edafccee JA |
3095 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3096 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 3097 | buf_addr = req->rw.addr; |
edafccee JA |
3098 | |
3099 | /* overflow */ | |
3100 | if (buf_addr + len < buf_addr) | |
3101 | return -EFAULT; | |
3102 | /* not inside the mapped region */ | |
3103 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
3104 | return -EFAULT; | |
3105 | ||
3106 | /* | |
3107 | * May not be a start of buffer, set size appropriately | |
3108 | * and advance us to the beginning. | |
3109 | */ | |
3110 | offset = buf_addr - imu->ubuf; | |
3111 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3112 | |
3113 | if (offset) { | |
3114 | /* | |
3115 | * Don't use iov_iter_advance() here, as it's really slow for | |
3116 | * using the latter parts of a big fixed buffer - it iterates | |
3117 | * over each segment manually. We can cheat a bit here, because | |
3118 | * we know that: | |
3119 | * | |
3120 | * 1) it's a BVEC iter, we set it up | |
3121 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3122 | * first and last bvec | |
3123 | * | |
3124 | * So just find our index, and adjust the iterator afterwards. | |
3125 | * If the offset is within the first bvec (or the whole first | |
3126 | * bvec, just use iov_iter_advance(). This makes it easier | |
3127 | * since we can just skip the first segment, which may not | |
3128 | * be PAGE_SIZE aligned. | |
3129 | */ | |
3130 | const struct bio_vec *bvec = imu->bvec; | |
3131 | ||
3132 | if (offset <= bvec->bv_len) { | |
3133 | iov_iter_advance(iter, offset); | |
3134 | } else { | |
3135 | unsigned long seg_skip; | |
3136 | ||
3137 | /* skip first vec */ | |
3138 | offset -= bvec->bv_len; | |
3139 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3140 | ||
3141 | iter->bvec = bvec + seg_skip; | |
3142 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3143 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3144 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3145 | } |
3146 | } | |
3147 | ||
847595de | 3148 | return 0; |
edafccee JA |
3149 | } |
3150 | ||
bcda7baa JA |
3151 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3152 | { | |
3153 | if (needs_lock) | |
3154 | mutex_unlock(&ctx->uring_lock); | |
3155 | } | |
3156 | ||
3157 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3158 | { | |
3159 | /* | |
3160 | * "Normal" inline submissions always hold the uring_lock, since we | |
3161 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3162 | * The only exception is when we've detached the request and issue it | |
3163 | * from an async worker thread, grab the lock for that case. | |
3164 | */ | |
3165 | if (needs_lock) | |
3166 | mutex_lock(&ctx->uring_lock); | |
3167 | } | |
3168 | ||
3169 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3170 | int bgid, struct io_buffer *kbuf, | |
3171 | bool needs_lock) | |
3172 | { | |
3173 | struct io_buffer *head; | |
3174 | ||
3175 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3176 | return kbuf; | |
3177 | ||
3178 | io_ring_submit_lock(req->ctx, needs_lock); | |
3179 | ||
3180 | lockdep_assert_held(&req->ctx->uring_lock); | |
3181 | ||
3182 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
3183 | if (head) { | |
3184 | if (!list_empty(&head->list)) { | |
3185 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3186 | list); | |
3187 | list_del(&kbuf->list); | |
3188 | } else { | |
3189 | kbuf = head; | |
3190 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
3191 | } | |
3192 | if (*len > kbuf->len) | |
3193 | *len = kbuf->len; | |
3194 | } else { | |
3195 | kbuf = ERR_PTR(-ENOBUFS); | |
3196 | } | |
3197 | ||
3198 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3199 | ||
3200 | return kbuf; | |
3201 | } | |
3202 | ||
4d954c25 JA |
3203 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3204 | bool needs_lock) | |
3205 | { | |
3206 | struct io_buffer *kbuf; | |
4f4eeba8 | 3207 | u16 bgid; |
4d954c25 JA |
3208 | |
3209 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3210 | bgid = req->buf_index; |
4d954c25 JA |
3211 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3212 | if (IS_ERR(kbuf)) | |
3213 | return kbuf; | |
3214 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3215 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3216 | return u64_to_user_ptr(kbuf->addr); | |
3217 | } | |
3218 | ||
3219 | #ifdef CONFIG_COMPAT | |
3220 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3221 | bool needs_lock) | |
3222 | { | |
3223 | struct compat_iovec __user *uiov; | |
3224 | compat_ssize_t clen; | |
3225 | void __user *buf; | |
3226 | ssize_t len; | |
3227 | ||
3228 | uiov = u64_to_user_ptr(req->rw.addr); | |
3229 | if (!access_ok(uiov, sizeof(*uiov))) | |
3230 | return -EFAULT; | |
3231 | if (__get_user(clen, &uiov->iov_len)) | |
3232 | return -EFAULT; | |
3233 | if (clen < 0) | |
3234 | return -EINVAL; | |
3235 | ||
3236 | len = clen; | |
3237 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3238 | if (IS_ERR(buf)) | |
3239 | return PTR_ERR(buf); | |
3240 | iov[0].iov_base = buf; | |
3241 | iov[0].iov_len = (compat_size_t) len; | |
3242 | return 0; | |
3243 | } | |
3244 | #endif | |
3245 | ||
3246 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3247 | bool needs_lock) | |
3248 | { | |
3249 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3250 | void __user *buf; | |
3251 | ssize_t len; | |
3252 | ||
3253 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3254 | return -EFAULT; | |
3255 | ||
3256 | len = iov[0].iov_len; | |
3257 | if (len < 0) | |
3258 | return -EINVAL; | |
3259 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3260 | if (IS_ERR(buf)) | |
3261 | return PTR_ERR(buf); | |
3262 | iov[0].iov_base = buf; | |
3263 | iov[0].iov_len = len; | |
3264 | return 0; | |
3265 | } | |
3266 | ||
3267 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3268 | bool needs_lock) | |
3269 | { | |
dddb3e26 JA |
3270 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3271 | struct io_buffer *kbuf; | |
3272 | ||
3273 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3274 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3275 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3276 | return 0; |
dddb3e26 | 3277 | } |
dd201662 | 3278 | if (req->rw.len != 1) |
4d954c25 JA |
3279 | return -EINVAL; |
3280 | ||
3281 | #ifdef CONFIG_COMPAT | |
3282 | if (req->ctx->compat) | |
3283 | return io_compat_import(req, iov, needs_lock); | |
3284 | #endif | |
3285 | ||
3286 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3287 | } | |
3288 | ||
847595de PB |
3289 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3290 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3291 | { |
9adbd45d JA |
3292 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3293 | size_t sqe_len = req->rw.len; | |
847595de | 3294 | u8 opcode = req->opcode; |
4d954c25 | 3295 | ssize_t ret; |
edafccee | 3296 | |
7d009165 | 3297 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3298 | *iovec = NULL; |
9adbd45d | 3299 | return io_import_fixed(req, rw, iter); |
edafccee | 3300 | } |
2b188cc1 | 3301 | |
bcda7baa | 3302 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3303 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3304 | return -EINVAL; |
3305 | ||
3a6820f2 | 3306 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3307 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3308 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3309 | if (IS_ERR(buf)) |
4d954c25 | 3310 | return PTR_ERR(buf); |
3f9d6441 | 3311 | req->rw.len = sqe_len; |
bcda7baa JA |
3312 | } |
3313 | ||
3a6820f2 JA |
3314 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3315 | *iovec = NULL; | |
10fc72e4 | 3316 | return ret; |
3a6820f2 JA |
3317 | } |
3318 | ||
4d954c25 JA |
3319 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3320 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3321 | if (!ret) |
3322 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3323 | *iovec = NULL; |
3324 | return ret; | |
3325 | } | |
3326 | ||
89cd35c5 CH |
3327 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3328 | req->ctx->compat); | |
2b188cc1 JA |
3329 | } |
3330 | ||
0fef9483 JA |
3331 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3332 | { | |
5b09e37e | 3333 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3334 | } |
3335 | ||
31b51510 | 3336 | /* |
32960613 JA |
3337 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3338 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3339 | */ |
4017eb91 | 3340 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3341 | { |
4017eb91 JA |
3342 | struct kiocb *kiocb = &req->rw.kiocb; |
3343 | struct file *file = req->file; | |
32960613 JA |
3344 | ssize_t ret = 0; |
3345 | ||
3346 | /* | |
3347 | * Don't support polled IO through this interface, and we can't | |
3348 | * support non-blocking either. For the latter, this just causes | |
3349 | * the kiocb to be handled from an async context. | |
3350 | */ | |
3351 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3352 | return -EOPNOTSUPP; | |
3353 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3354 | return -EAGAIN; | |
3355 | ||
3356 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3357 | struct iovec iovec; |
32960613 JA |
3358 | ssize_t nr; |
3359 | ||
311ae9e1 PB |
3360 | if (!iov_iter_is_bvec(iter)) { |
3361 | iovec = iov_iter_iovec(iter); | |
3362 | } else { | |
4017eb91 JA |
3363 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3364 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3365 | } |
3366 | ||
32960613 JA |
3367 | if (rw == READ) { |
3368 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3369 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3370 | } else { |
3371 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3372 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3373 | } |
3374 | ||
3375 | if (nr < 0) { | |
3376 | if (!ret) | |
3377 | ret = nr; | |
3378 | break; | |
3379 | } | |
3380 | ret += nr; | |
3381 | if (nr != iovec.iov_len) | |
3382 | break; | |
4017eb91 JA |
3383 | req->rw.len -= nr; |
3384 | req->rw.addr += nr; | |
32960613 JA |
3385 | iov_iter_advance(iter, nr); |
3386 | } | |
3387 | ||
3388 | return ret; | |
3389 | } | |
3390 | ||
ff6165b2 JA |
3391 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3392 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3393 | { |
e8c2bc1f | 3394 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3395 | |
ff6165b2 | 3396 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3397 | rw->free_iovec = iovec; |
227c0c96 | 3398 | rw->bytes_done = 0; |
ff6165b2 | 3399 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3400 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3401 | return; |
b64e3444 | 3402 | if (!iovec) { |
ff6165b2 JA |
3403 | unsigned iov_off = 0; |
3404 | ||
3405 | rw->iter.iov = rw->fast_iov; | |
3406 | if (iter->iov != fast_iov) { | |
3407 | iov_off = iter->iov - fast_iov; | |
3408 | rw->iter.iov += iov_off; | |
3409 | } | |
3410 | if (rw->fast_iov != fast_iov) | |
3411 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3412 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3413 | } else { |
3414 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3415 | } |
3416 | } | |
3417 | ||
e8c2bc1f | 3418 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3419 | { |
e8c2bc1f JA |
3420 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3421 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3422 | return req->async_data == NULL; | |
3d9932a8 XW |
3423 | } |
3424 | ||
e8c2bc1f | 3425 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3426 | { |
e8c2bc1f | 3427 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3428 | return 0; |
3d9932a8 | 3429 | |
e8c2bc1f | 3430 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3431 | } |
3432 | ||
ff6165b2 JA |
3433 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3434 | const struct iovec *fast_iov, | |
227c0c96 | 3435 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3436 | { |
e8c2bc1f | 3437 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3438 | return 0; |
e8c2bc1f | 3439 | if (!req->async_data) { |
6bf985dc PB |
3440 | if (__io_alloc_async_data(req)) { |
3441 | kfree(iovec); | |
5d204bcf | 3442 | return -ENOMEM; |
6bf985dc | 3443 | } |
b7bb4f7d | 3444 | |
ff6165b2 | 3445 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3446 | } |
b7bb4f7d | 3447 | return 0; |
f67676d1 JA |
3448 | } |
3449 | ||
73debe68 | 3450 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3451 | { |
e8c2bc1f | 3452 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3453 | struct iovec *iov = iorw->fast_iov; |
847595de | 3454 | int ret; |
c3e330a4 | 3455 | |
2846c481 | 3456 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3457 | if (unlikely(ret < 0)) |
3458 | return ret; | |
3459 | ||
ab0b196c PB |
3460 | iorw->bytes_done = 0; |
3461 | iorw->free_iovec = iov; | |
3462 | if (iov) | |
3463 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3464 | return 0; |
3465 | } | |
3466 | ||
73debe68 | 3467 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3468 | { |
3469 | ssize_t ret; | |
3470 | ||
a88fc400 | 3471 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3472 | if (ret) |
3473 | return ret; | |
f67676d1 | 3474 | |
3529d8c2 JA |
3475 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3476 | return -EBADF; | |
f67676d1 | 3477 | |
5f798bea | 3478 | /* either don't need iovec imported or already have it */ |
2d199895 | 3479 | if (!req->async_data) |
3529d8c2 | 3480 | return 0; |
73debe68 | 3481 | return io_rw_prep_async(req, READ); |
f67676d1 JA |
3482 | } |
3483 | ||
c1dd91d1 JA |
3484 | /* |
3485 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3486 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3487 | * This gets called when the page is unlocked, and we generally expect that to | |
3488 | * happen when the page IO is completed and the page is now uptodate. This will | |
3489 | * queue a task_work based retry of the operation, attempting to copy the data | |
3490 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3491 | * do a thread based blocking retry of the operation. That's the unexpected | |
3492 | * slow path. | |
3493 | */ | |
bcf5a063 JA |
3494 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3495 | int sync, void *arg) | |
3496 | { | |
3497 | struct wait_page_queue *wpq; | |
3498 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3499 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3500 | |
3501 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3502 | ||
cdc8fcb4 LT |
3503 | if (!wake_page_match(wpq, key)) |
3504 | return 0; | |
3505 | ||
c8d317aa | 3506 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3507 | list_del_init(&wait->entry); |
3508 | ||
bcf5a063 JA |
3509 | /* submit ref gets dropped, acquire a new one */ |
3510 | refcount_inc(&req->refs); | |
921b9054 | 3511 | io_req_task_queue(req); |
bcf5a063 JA |
3512 | return 1; |
3513 | } | |
3514 | ||
c1dd91d1 JA |
3515 | /* |
3516 | * This controls whether a given IO request should be armed for async page | |
3517 | * based retry. If we return false here, the request is handed to the async | |
3518 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3519 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3520 | * will either succeed because the page is now uptodate and unlocked, or it | |
3521 | * will register a callback when the page is unlocked at IO completion. Through | |
3522 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3523 | * That retry will attempt the buffered read again. The retry will generally | |
3524 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3525 | * async worker threads for a blocking retry. | |
3526 | */ | |
227c0c96 | 3527 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3528 | { |
e8c2bc1f JA |
3529 | struct io_async_rw *rw = req->async_data; |
3530 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3531 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3532 | |
bcf5a063 JA |
3533 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3534 | if (req->flags & REQ_F_NOWAIT) | |
3535 | return false; | |
f67676d1 | 3536 | |
227c0c96 | 3537 | /* Only for buffered IO */ |
3b2a4439 | 3538 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3539 | return false; |
3b2a4439 | 3540 | |
bcf5a063 JA |
3541 | /* |
3542 | * just use poll if we can, and don't attempt if the fs doesn't | |
3543 | * support callback based unlocks | |
3544 | */ | |
3545 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3546 | return false; | |
f67676d1 | 3547 | |
3b2a4439 JA |
3548 | wait->wait.func = io_async_buf_func; |
3549 | wait->wait.private = req; | |
3550 | wait->wait.flags = 0; | |
3551 | INIT_LIST_HEAD(&wait->wait.entry); | |
3552 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3553 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3554 | kiocb->ki_waitq = wait; |
3b2a4439 | 3555 | return true; |
bcf5a063 JA |
3556 | } |
3557 | ||
3558 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3559 | { | |
3560 | if (req->file->f_op->read_iter) | |
3561 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3562 | else if (req->file->f_op->read) |
4017eb91 | 3563 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3564 | else |
3565 | return -EINVAL; | |
f67676d1 JA |
3566 | } |
3567 | ||
889fca73 | 3568 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3569 | { |
3570 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3571 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3572 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3573 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3574 | ssize_t io_size, ret, ret2; |
45d189c6 | 3575 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3576 | |
2846c481 | 3577 | if (rw) { |
e8c2bc1f | 3578 | iter = &rw->iter; |
2846c481 PB |
3579 | iovec = NULL; |
3580 | } else { | |
3581 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3582 | if (ret < 0) | |
3583 | return ret; | |
3584 | } | |
632546c4 | 3585 | io_size = iov_iter_count(iter); |
fa15bafb | 3586 | req->result = io_size; |
2b188cc1 | 3587 | |
fd6c2e4c JA |
3588 | /* Ensure we clear previously set non-block flag */ |
3589 | if (!force_nonblock) | |
29de5f6a | 3590 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3591 | else |
3592 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3593 | ||
24c74678 | 3594 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3595 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3596 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3597 | return ret ?: -EAGAIN; |
6713e7a6 | 3598 | } |
9e645e11 | 3599 | |
632546c4 | 3600 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3601 | if (unlikely(ret)) { |
3602 | kfree(iovec); | |
3603 | return ret; | |
3604 | } | |
2b188cc1 | 3605 | |
227c0c96 | 3606 | ret = io_iter_do_read(req, iter); |
32960613 | 3607 | |
57cd657b | 3608 | if (ret == -EIOCBQUEUED) { |
fe1cdd55 | 3609 | goto out_free; |
227c0c96 | 3610 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3611 | /* IOPOLL retry should happen for io-wq threads */ |
3612 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3613 | goto done; |
75c668cd PB |
3614 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3615 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3616 | goto done; |
84216315 | 3617 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3618 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3619 | ret = 0; |
7335e3bf | 3620 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3621 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3622 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3623 | goto done; |
1a2cc0ce | 3624 | } |
227c0c96 | 3625 | |
227c0c96 | 3626 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3627 | if (ret2) |
3628 | return ret2; | |
3629 | ||
fe1cdd55 | 3630 | iovec = NULL; |
e8c2bc1f | 3631 | rw = req->async_data; |
227c0c96 | 3632 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3633 | iter = &rw->iter; |
227c0c96 | 3634 | |
b23df91b PB |
3635 | do { |
3636 | io_size -= ret; | |
3637 | rw->bytes_done += ret; | |
3638 | /* if we can retry, do so with the callbacks armed */ | |
3639 | if (!io_rw_should_retry(req)) { | |
3640 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3641 | return -EAGAIN; | |
3642 | } | |
3643 | ||
3644 | /* | |
3645 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3646 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3647 | * desired page gets unlocked. We can also get a partial read | |
3648 | * here, and if we do, then just retry at the new offset. | |
3649 | */ | |
3650 | ret = io_iter_do_read(req, iter); | |
3651 | if (ret == -EIOCBQUEUED) | |
3652 | return 0; | |
3653 | /* we got some bytes, but not all. retry. */ | |
3654 | } while (ret > 0 && ret < io_size); | |
227c0c96 | 3655 | done: |
889fca73 | 3656 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3657 | out_free: |
3658 | /* it's faster to check here then delegate to kfree */ | |
3659 | if (iovec) | |
3660 | kfree(iovec); | |
5ea5dd45 | 3661 | return 0; |
2b188cc1 JA |
3662 | } |
3663 | ||
73debe68 | 3664 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 JA |
3665 | { |
3666 | ssize_t ret; | |
3667 | ||
a88fc400 | 3668 | ret = io_prep_rw(req, sqe); |
3529d8c2 JA |
3669 | if (ret) |
3670 | return ret; | |
f67676d1 | 3671 | |
3529d8c2 JA |
3672 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3673 | return -EBADF; | |
f67676d1 | 3674 | |
5f798bea | 3675 | /* either don't need iovec imported or already have it */ |
2d199895 | 3676 | if (!req->async_data) |
3529d8c2 | 3677 | return 0; |
73debe68 | 3678 | return io_rw_prep_async(req, WRITE); |
f67676d1 JA |
3679 | } |
3680 | ||
889fca73 | 3681 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3682 | { |
3683 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3684 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3685 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3686 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3687 | ssize_t ret, ret2, io_size; |
45d189c6 | 3688 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3689 | |
2846c481 | 3690 | if (rw) { |
e8c2bc1f | 3691 | iter = &rw->iter; |
2846c481 PB |
3692 | iovec = NULL; |
3693 | } else { | |
3694 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3695 | if (ret < 0) | |
3696 | return ret; | |
3697 | } | |
632546c4 | 3698 | io_size = iov_iter_count(iter); |
fa15bafb | 3699 | req->result = io_size; |
2b188cc1 | 3700 | |
fd6c2e4c JA |
3701 | /* Ensure we clear previously set non-block flag */ |
3702 | if (!force_nonblock) | |
a88fc400 PB |
3703 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3704 | else | |
3705 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3706 | |
24c74678 | 3707 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3708 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3709 | goto copy_iov; |
31b51510 | 3710 | |
10d59345 JA |
3711 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3712 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3713 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3714 | goto copy_iov; |
31b51510 | 3715 | |
632546c4 | 3716 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3717 | if (unlikely(ret)) |
3718 | goto out_free; | |
4ed734b0 | 3719 | |
fa15bafb PB |
3720 | /* |
3721 | * Open-code file_start_write here to grab freeze protection, | |
3722 | * which will be released by another thread in | |
3723 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3724 | * released so that it doesn't complain about the held lock when | |
3725 | * we return to userspace. | |
3726 | */ | |
3727 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3728 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3729 | __sb_writers_release(file_inode(req->file)->i_sb, |
3730 | SB_FREEZE_WRITE); | |
3731 | } | |
3732 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3733 | |
fa15bafb | 3734 | if (req->file->f_op->write_iter) |
ff6165b2 | 3735 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3736 | else if (req->file->f_op->write) |
4017eb91 | 3737 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3738 | else |
3739 | ret2 = -EINVAL; | |
4ed734b0 | 3740 | |
fa15bafb PB |
3741 | /* |
3742 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3743 | * retry them without IOCB_NOWAIT. | |
3744 | */ | |
3745 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3746 | ret2 = -EAGAIN; | |
75c668cd PB |
3747 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3748 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3749 | goto done; |
fa15bafb | 3750 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3751 | /* IOPOLL retry should happen for io-wq threads */ |
3752 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3753 | goto copy_iov; | |
355afaeb | 3754 | done: |
889fca73 | 3755 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3756 | } else { |
f67676d1 | 3757 | copy_iov: |
84216315 | 3758 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3759 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3760 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3761 | return ret ?: -EAGAIN; |
2b188cc1 | 3762 | } |
31b51510 | 3763 | out_free: |
f261c168 | 3764 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3765 | if (iovec) |
6f2cc166 | 3766 | kfree(iovec); |
2b188cc1 JA |
3767 | return ret; |
3768 | } | |
3769 | ||
80a261fd JA |
3770 | static int io_renameat_prep(struct io_kiocb *req, |
3771 | const struct io_uring_sqe *sqe) | |
3772 | { | |
3773 | struct io_rename *ren = &req->rename; | |
3774 | const char __user *oldf, *newf; | |
3775 | ||
3776 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3777 | return -EBADF; | |
3778 | ||
3779 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3780 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3781 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3782 | ren->new_dfd = READ_ONCE(sqe->len); | |
3783 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3784 | ||
3785 | ren->oldpath = getname(oldf); | |
3786 | if (IS_ERR(ren->oldpath)) | |
3787 | return PTR_ERR(ren->oldpath); | |
3788 | ||
3789 | ren->newpath = getname(newf); | |
3790 | if (IS_ERR(ren->newpath)) { | |
3791 | putname(ren->oldpath); | |
3792 | return PTR_ERR(ren->newpath); | |
3793 | } | |
3794 | ||
3795 | req->flags |= REQ_F_NEED_CLEANUP; | |
3796 | return 0; | |
3797 | } | |
3798 | ||
45d189c6 | 3799 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3800 | { |
3801 | struct io_rename *ren = &req->rename; | |
3802 | int ret; | |
3803 | ||
45d189c6 | 3804 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3805 | return -EAGAIN; |
3806 | ||
3807 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3808 | ren->newpath, ren->flags); | |
3809 | ||
3810 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3811 | if (ret < 0) | |
3812 | req_set_fail_links(req); | |
3813 | io_req_complete(req, ret); | |
3814 | return 0; | |
3815 | } | |
3816 | ||
14a1143b JA |
3817 | static int io_unlinkat_prep(struct io_kiocb *req, |
3818 | const struct io_uring_sqe *sqe) | |
3819 | { | |
3820 | struct io_unlink *un = &req->unlink; | |
3821 | const char __user *fname; | |
3822 | ||
3823 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3824 | return -EBADF; | |
3825 | ||
3826 | un->dfd = READ_ONCE(sqe->fd); | |
3827 | ||
3828 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3829 | if (un->flags & ~AT_REMOVEDIR) | |
3830 | return -EINVAL; | |
3831 | ||
3832 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3833 | un->filename = getname(fname); | |
3834 | if (IS_ERR(un->filename)) | |
3835 | return PTR_ERR(un->filename); | |
3836 | ||
3837 | req->flags |= REQ_F_NEED_CLEANUP; | |
3838 | return 0; | |
3839 | } | |
3840 | ||
45d189c6 | 3841 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3842 | { |
3843 | struct io_unlink *un = &req->unlink; | |
3844 | int ret; | |
3845 | ||
45d189c6 | 3846 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3847 | return -EAGAIN; |
3848 | ||
3849 | if (un->flags & AT_REMOVEDIR) | |
3850 | ret = do_rmdir(un->dfd, un->filename); | |
3851 | else | |
3852 | ret = do_unlinkat(un->dfd, un->filename); | |
3853 | ||
3854 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3855 | if (ret < 0) | |
3856 | req_set_fail_links(req); | |
3857 | io_req_complete(req, ret); | |
3858 | return 0; | |
3859 | } | |
3860 | ||
36f4fa68 JA |
3861 | static int io_shutdown_prep(struct io_kiocb *req, |
3862 | const struct io_uring_sqe *sqe) | |
3863 | { | |
3864 | #if defined(CONFIG_NET) | |
3865 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3866 | return -EINVAL; | |
3867 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3868 | sqe->buf_index) | |
3869 | return -EINVAL; | |
3870 | ||
3871 | req->shutdown.how = READ_ONCE(sqe->len); | |
3872 | return 0; | |
3873 | #else | |
3874 | return -EOPNOTSUPP; | |
3875 | #endif | |
3876 | } | |
3877 | ||
45d189c6 | 3878 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3879 | { |
3880 | #if defined(CONFIG_NET) | |
3881 | struct socket *sock; | |
3882 | int ret; | |
3883 | ||
45d189c6 | 3884 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3885 | return -EAGAIN; |
3886 | ||
48aba79b | 3887 | sock = sock_from_file(req->file); |
36f4fa68 | 3888 | if (unlikely(!sock)) |
48aba79b | 3889 | return -ENOTSOCK; |
36f4fa68 JA |
3890 | |
3891 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3892 | if (ret < 0) |
3893 | req_set_fail_links(req); | |
36f4fa68 JA |
3894 | io_req_complete(req, ret); |
3895 | return 0; | |
3896 | #else | |
3897 | return -EOPNOTSUPP; | |
3898 | #endif | |
3899 | } | |
3900 | ||
f2a8d5c7 PB |
3901 | static int __io_splice_prep(struct io_kiocb *req, |
3902 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3903 | { |
3904 | struct io_splice* sp = &req->splice; | |
3905 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3906 | |
3232dd02 PB |
3907 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3908 | return -EINVAL; | |
7d67af2c PB |
3909 | |
3910 | sp->file_in = NULL; | |
7d67af2c PB |
3911 | sp->len = READ_ONCE(sqe->len); |
3912 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3913 | ||
3914 | if (unlikely(sp->flags & ~valid_flags)) | |
3915 | return -EINVAL; | |
3916 | ||
8371adf5 PB |
3917 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3918 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3919 | if (!sp->file_in) | |
3920 | return -EBADF; | |
7d67af2c PB |
3921 | req->flags |= REQ_F_NEED_CLEANUP; |
3922 | ||
7cdaf587 XW |
3923 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3924 | /* | |
3925 | * Splice operation will be punted aync, and here need to | |
3926 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3927 | */ | |
3928 | io_req_init_async(req); | |
7d67af2c | 3929 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3930 | } |
7d67af2c PB |
3931 | |
3932 | return 0; | |
3933 | } | |
3934 | ||
f2a8d5c7 PB |
3935 | static int io_tee_prep(struct io_kiocb *req, |
3936 | const struct io_uring_sqe *sqe) | |
3937 | { | |
3938 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3939 | return -EINVAL; | |
3940 | return __io_splice_prep(req, sqe); | |
3941 | } | |
3942 | ||
45d189c6 | 3943 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3944 | { |
3945 | struct io_splice *sp = &req->splice; | |
3946 | struct file *in = sp->file_in; | |
3947 | struct file *out = sp->file_out; | |
3948 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3949 | long ret = 0; | |
3950 | ||
45d189c6 | 3951 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3952 | return -EAGAIN; |
3953 | if (sp->len) | |
3954 | ret = do_tee(in, out, sp->len, flags); | |
3955 | ||
3956 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3957 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3958 | ||
f2a8d5c7 PB |
3959 | if (ret != sp->len) |
3960 | req_set_fail_links(req); | |
e1e16097 | 3961 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3962 | return 0; |
3963 | } | |
3964 | ||
3965 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3966 | { | |
3967 | struct io_splice* sp = &req->splice; | |
3968 | ||
3969 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3970 | sp->off_out = READ_ONCE(sqe->off); | |
3971 | return __io_splice_prep(req, sqe); | |
3972 | } | |
3973 | ||
45d189c6 | 3974 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3975 | { |
3976 | struct io_splice *sp = &req->splice; | |
3977 | struct file *in = sp->file_in; | |
3978 | struct file *out = sp->file_out; | |
3979 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3980 | loff_t *poff_in, *poff_out; | |
c9687426 | 3981 | long ret = 0; |
7d67af2c | 3982 | |
45d189c6 | 3983 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3984 | return -EAGAIN; |
7d67af2c PB |
3985 | |
3986 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3987 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3988 | |
948a7749 | 3989 | if (sp->len) |
c9687426 | 3990 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3991 | |
3992 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3993 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3994 | ||
7d67af2c PB |
3995 | if (ret != sp->len) |
3996 | req_set_fail_links(req); | |
e1e16097 | 3997 | io_req_complete(req, ret); |
7d67af2c PB |
3998 | return 0; |
3999 | } | |
4000 | ||
2b188cc1 JA |
4001 | /* |
4002 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4003 | */ | |
889fca73 | 4004 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4005 | { |
4006 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4007 | |
def596e9 JA |
4008 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4009 | return -EINVAL; | |
4010 | ||
889fca73 | 4011 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4012 | return 0; |
4013 | } | |
4014 | ||
1155c76a | 4015 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4016 | { |
6b06314c | 4017 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4018 | |
09bb8394 JA |
4019 | if (!req->file) |
4020 | return -EBADF; | |
c992fe29 | 4021 | |
6b06314c | 4022 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4023 | return -EINVAL; |
edafccee | 4024 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
4025 | return -EINVAL; |
4026 | ||
8ed8d3c3 JA |
4027 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4028 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4029 | return -EINVAL; | |
4030 | ||
4031 | req->sync.off = READ_ONCE(sqe->off); | |
4032 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4033 | return 0; |
4034 | } | |
4035 | ||
45d189c6 | 4036 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4037 | { |
8ed8d3c3 | 4038 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4039 | int ret; |
4040 | ||
ac45abc0 | 4041 | /* fsync always requires a blocking context */ |
45d189c6 | 4042 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4043 | return -EAGAIN; |
4044 | ||
9adbd45d | 4045 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4046 | end > 0 ? end : LLONG_MAX, |
4047 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4048 | if (ret < 0) | |
4049 | req_set_fail_links(req); | |
e1e16097 | 4050 | io_req_complete(req, ret); |
c992fe29 CH |
4051 | return 0; |
4052 | } | |
4053 | ||
d63d1b5e JA |
4054 | static int io_fallocate_prep(struct io_kiocb *req, |
4055 | const struct io_uring_sqe *sqe) | |
4056 | { | |
4057 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
4058 | return -EINVAL; | |
3232dd02 PB |
4059 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4060 | return -EINVAL; | |
d63d1b5e JA |
4061 | |
4062 | req->sync.off = READ_ONCE(sqe->off); | |
4063 | req->sync.len = READ_ONCE(sqe->addr); | |
4064 | req->sync.mode = READ_ONCE(sqe->len); | |
4065 | return 0; | |
4066 | } | |
4067 | ||
45d189c6 | 4068 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4069 | { |
ac45abc0 PB |
4070 | int ret; |
4071 | ||
d63d1b5e | 4072 | /* fallocate always requiring blocking context */ |
45d189c6 | 4073 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4074 | return -EAGAIN; |
ac45abc0 PB |
4075 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4076 | req->sync.len); | |
ac45abc0 PB |
4077 | if (ret < 0) |
4078 | req_set_fail_links(req); | |
e1e16097 | 4079 | io_req_complete(req, ret); |
5d17b4a4 JA |
4080 | return 0; |
4081 | } | |
4082 | ||
ec65fea5 | 4083 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4084 | { |
f8748881 | 4085 | const char __user *fname; |
15b71abe | 4086 | int ret; |
b7bb4f7d | 4087 | |
ec65fea5 | 4088 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4089 | return -EINVAL; |
ec65fea5 | 4090 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4091 | return -EBADF; |
03b1230c | 4092 | |
ec65fea5 PB |
4093 | /* open.how should be already initialised */ |
4094 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4095 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4096 | |
25e72d10 PB |
4097 | req->open.dfd = READ_ONCE(sqe->fd); |
4098 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4099 | req->open.filename = getname(fname); |
15b71abe JA |
4100 | if (IS_ERR(req->open.filename)) { |
4101 | ret = PTR_ERR(req->open.filename); | |
4102 | req->open.filename = NULL; | |
4103 | return ret; | |
4104 | } | |
4022e7af | 4105 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4106 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4107 | return 0; |
03b1230c JA |
4108 | } |
4109 | ||
ec65fea5 PB |
4110 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4111 | { | |
4112 | u64 flags, mode; | |
4113 | ||
14587a46 | 4114 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4115 | return -EINVAL; |
ec65fea5 PB |
4116 | mode = READ_ONCE(sqe->len); |
4117 | flags = READ_ONCE(sqe->open_flags); | |
4118 | req->open.how = build_open_how(flags, mode); | |
4119 | return __io_openat_prep(req, sqe); | |
4120 | } | |
4121 | ||
cebdb986 | 4122 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4123 | { |
cebdb986 | 4124 | struct open_how __user *how; |
cebdb986 | 4125 | size_t len; |
0fa03c62 JA |
4126 | int ret; |
4127 | ||
14587a46 | 4128 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4129 | return -EINVAL; |
cebdb986 JA |
4130 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4131 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4132 | if (len < OPEN_HOW_SIZE_VER0) |
4133 | return -EINVAL; | |
3529d8c2 | 4134 | |
cebdb986 JA |
4135 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4136 | len); | |
4137 | if (ret) | |
4138 | return ret; | |
3529d8c2 | 4139 | |
ec65fea5 | 4140 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4141 | } |
4142 | ||
45d189c6 | 4143 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4144 | { |
4145 | struct open_flags op; | |
15b71abe | 4146 | struct file *file; |
3a81fd02 JA |
4147 | bool nonblock_set; |
4148 | bool resolve_nonblock; | |
15b71abe JA |
4149 | int ret; |
4150 | ||
cebdb986 | 4151 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4152 | if (ret) |
4153 | goto err; | |
3a81fd02 JA |
4154 | nonblock_set = op.open_flag & O_NONBLOCK; |
4155 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4156 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4157 | /* |
4158 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4159 | * it'll always -EAGAIN | |
4160 | */ | |
4161 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4162 | return -EAGAIN; | |
4163 | op.lookup_flags |= LOOKUP_CACHED; | |
4164 | op.open_flag |= O_NONBLOCK; | |
4165 | } | |
15b71abe | 4166 | |
4022e7af | 4167 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
4168 | if (ret < 0) |
4169 | goto err; | |
4170 | ||
4171 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 4172 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
4173 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
4174 | file == ERR_PTR(-EAGAIN)) { | |
3a81fd02 JA |
4175 | /* |
4176 | * We could hang on to this 'fd', but seems like marginal | |
4177 | * gain for something that is now known to be a slower path. | |
4178 | * So just put it, and we'll get a new one when we retry. | |
4179 | */ | |
4180 | put_unused_fd(ret); | |
4181 | return -EAGAIN; | |
4182 | } | |
4183 | ||
15b71abe JA |
4184 | if (IS_ERR(file)) { |
4185 | put_unused_fd(ret); | |
4186 | ret = PTR_ERR(file); | |
4187 | } else { | |
45d189c6 | 4188 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 4189 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
4190 | fsnotify_open(file); |
4191 | fd_install(ret, file); | |
4192 | } | |
4193 | err: | |
4194 | putname(req->open.filename); | |
8fef80bf | 4195 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
4196 | if (ret < 0) |
4197 | req_set_fail_links(req); | |
e1e16097 | 4198 | io_req_complete(req, ret); |
15b71abe JA |
4199 | return 0; |
4200 | } | |
4201 | ||
45d189c6 | 4202 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4203 | { |
45d189c6 | 4204 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
4205 | } |
4206 | ||
067524e9 JA |
4207 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4208 | const struct io_uring_sqe *sqe) | |
4209 | { | |
4210 | struct io_provide_buf *p = &req->pbuf; | |
4211 | u64 tmp; | |
4212 | ||
4213 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
4214 | return -EINVAL; | |
4215 | ||
4216 | tmp = READ_ONCE(sqe->fd); | |
4217 | if (!tmp || tmp > USHRT_MAX) | |
4218 | return -EINVAL; | |
4219 | ||
4220 | memset(p, 0, sizeof(*p)); | |
4221 | p->nbufs = tmp; | |
4222 | p->bgid = READ_ONCE(sqe->buf_group); | |
4223 | return 0; | |
4224 | } | |
4225 | ||
4226 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4227 | int bgid, unsigned nbufs) | |
4228 | { | |
4229 | unsigned i = 0; | |
4230 | ||
4231 | /* shouldn't happen */ | |
4232 | if (!nbufs) | |
4233 | return 0; | |
4234 | ||
4235 | /* the head kbuf is the list itself */ | |
4236 | while (!list_empty(&buf->list)) { | |
4237 | struct io_buffer *nxt; | |
4238 | ||
4239 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4240 | list_del(&nxt->list); | |
4241 | kfree(nxt); | |
4242 | if (++i == nbufs) | |
4243 | return i; | |
4244 | } | |
4245 | i++; | |
4246 | kfree(buf); | |
4247 | idr_remove(&ctx->io_buffer_idr, bgid); | |
4248 | ||
4249 | return i; | |
4250 | } | |
4251 | ||
889fca73 | 4252 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4253 | { |
4254 | struct io_provide_buf *p = &req->pbuf; | |
4255 | struct io_ring_ctx *ctx = req->ctx; | |
4256 | struct io_buffer *head; | |
4257 | int ret = 0; | |
45d189c6 | 4258 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4259 | |
4260 | io_ring_submit_lock(ctx, !force_nonblock); | |
4261 | ||
4262 | lockdep_assert_held(&ctx->uring_lock); | |
4263 | ||
4264 | ret = -ENOENT; | |
4265 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4266 | if (head) | |
4267 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
4268 | if (ret < 0) |
4269 | req_set_fail_links(req); | |
067524e9 | 4270 | |
31bff9a5 PB |
4271 | /* need to hold the lock to complete IOPOLL requests */ |
4272 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4273 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4274 | io_ring_submit_unlock(ctx, !force_nonblock); |
4275 | } else { | |
4276 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4277 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4278 | } |
067524e9 JA |
4279 | return 0; |
4280 | } | |
4281 | ||
ddf0322d JA |
4282 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4283 | const struct io_uring_sqe *sqe) | |
4284 | { | |
4285 | struct io_provide_buf *p = &req->pbuf; | |
4286 | u64 tmp; | |
4287 | ||
4288 | if (sqe->ioprio || sqe->rw_flags) | |
4289 | return -EINVAL; | |
4290 | ||
4291 | tmp = READ_ONCE(sqe->fd); | |
4292 | if (!tmp || tmp > USHRT_MAX) | |
4293 | return -E2BIG; | |
4294 | p->nbufs = tmp; | |
4295 | p->addr = READ_ONCE(sqe->addr); | |
4296 | p->len = READ_ONCE(sqe->len); | |
4297 | ||
efe68c1c | 4298 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4299 | return -EFAULT; |
4300 | ||
4301 | p->bgid = READ_ONCE(sqe->buf_group); | |
4302 | tmp = READ_ONCE(sqe->off); | |
4303 | if (tmp > USHRT_MAX) | |
4304 | return -E2BIG; | |
4305 | p->bid = tmp; | |
4306 | return 0; | |
4307 | } | |
4308 | ||
4309 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4310 | { | |
4311 | struct io_buffer *buf; | |
4312 | u64 addr = pbuf->addr; | |
4313 | int i, bid = pbuf->bid; | |
4314 | ||
4315 | for (i = 0; i < pbuf->nbufs; i++) { | |
4316 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4317 | if (!buf) | |
4318 | break; | |
4319 | ||
4320 | buf->addr = addr; | |
4321 | buf->len = pbuf->len; | |
4322 | buf->bid = bid; | |
4323 | addr += pbuf->len; | |
4324 | bid++; | |
4325 | if (!*head) { | |
4326 | INIT_LIST_HEAD(&buf->list); | |
4327 | *head = buf; | |
4328 | } else { | |
4329 | list_add_tail(&buf->list, &(*head)->list); | |
4330 | } | |
4331 | } | |
4332 | ||
4333 | return i ? i : -ENOMEM; | |
4334 | } | |
4335 | ||
889fca73 | 4336 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4337 | { |
4338 | struct io_provide_buf *p = &req->pbuf; | |
4339 | struct io_ring_ctx *ctx = req->ctx; | |
4340 | struct io_buffer *head, *list; | |
4341 | int ret = 0; | |
45d189c6 | 4342 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4343 | |
4344 | io_ring_submit_lock(ctx, !force_nonblock); | |
4345 | ||
4346 | lockdep_assert_held(&ctx->uring_lock); | |
4347 | ||
4348 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4349 | ||
4350 | ret = io_add_buffers(p, &head); | |
4351 | if (ret < 0) | |
4352 | goto out; | |
4353 | ||
4354 | if (!list) { | |
4355 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4356 | GFP_KERNEL); | |
4357 | if (ret < 0) { | |
067524e9 | 4358 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4359 | goto out; |
4360 | } | |
4361 | } | |
4362 | out: | |
ddf0322d JA |
4363 | if (ret < 0) |
4364 | req_set_fail_links(req); | |
31bff9a5 PB |
4365 | |
4366 | /* need to hold the lock to complete IOPOLL requests */ | |
4367 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4368 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4369 | io_ring_submit_unlock(ctx, !force_nonblock); |
4370 | } else { | |
4371 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4372 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4373 | } |
ddf0322d | 4374 | return 0; |
cebdb986 JA |
4375 | } |
4376 | ||
3e4827b0 JA |
4377 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4378 | const struct io_uring_sqe *sqe) | |
4379 | { | |
4380 | #if defined(CONFIG_EPOLL) | |
4381 | if (sqe->ioprio || sqe->buf_index) | |
4382 | return -EINVAL; | |
6ca56f84 | 4383 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4384 | return -EINVAL; |
3e4827b0 JA |
4385 | |
4386 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4387 | req->epoll.op = READ_ONCE(sqe->len); | |
4388 | req->epoll.fd = READ_ONCE(sqe->off); | |
4389 | ||
4390 | if (ep_op_has_event(req->epoll.op)) { | |
4391 | struct epoll_event __user *ev; | |
4392 | ||
4393 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4394 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4395 | return -EFAULT; | |
4396 | } | |
4397 | ||
4398 | return 0; | |
4399 | #else | |
4400 | return -EOPNOTSUPP; | |
4401 | #endif | |
4402 | } | |
4403 | ||
889fca73 | 4404 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4405 | { |
4406 | #if defined(CONFIG_EPOLL) | |
4407 | struct io_epoll *ie = &req->epoll; | |
4408 | int ret; | |
45d189c6 | 4409 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4410 | |
4411 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4412 | if (force_nonblock && ret == -EAGAIN) | |
4413 | return -EAGAIN; | |
4414 | ||
4415 | if (ret < 0) | |
4416 | req_set_fail_links(req); | |
889fca73 | 4417 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4418 | return 0; |
4419 | #else | |
4420 | return -EOPNOTSUPP; | |
4421 | #endif | |
4422 | } | |
4423 | ||
c1ca757b JA |
4424 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4425 | { | |
4426 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4427 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4428 | return -EINVAL; | |
3232dd02 PB |
4429 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4430 | return -EINVAL; | |
c1ca757b JA |
4431 | |
4432 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4433 | req->madvise.len = READ_ONCE(sqe->len); | |
4434 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4435 | return 0; | |
4436 | #else | |
4437 | return -EOPNOTSUPP; | |
4438 | #endif | |
4439 | } | |
4440 | ||
45d189c6 | 4441 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4442 | { |
4443 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4444 | struct io_madvise *ma = &req->madvise; | |
4445 | int ret; | |
4446 | ||
45d189c6 | 4447 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4448 | return -EAGAIN; |
4449 | ||
0726b01e | 4450 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4451 | if (ret < 0) |
4452 | req_set_fail_links(req); | |
e1e16097 | 4453 | io_req_complete(req, ret); |
c1ca757b JA |
4454 | return 0; |
4455 | #else | |
4456 | return -EOPNOTSUPP; | |
4457 | #endif | |
4458 | } | |
4459 | ||
4840e418 JA |
4460 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4461 | { | |
4462 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4463 | return -EINVAL; | |
3232dd02 PB |
4464 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4465 | return -EINVAL; | |
4840e418 JA |
4466 | |
4467 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4468 | req->fadvise.len = READ_ONCE(sqe->len); | |
4469 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4470 | return 0; | |
4471 | } | |
4472 | ||
45d189c6 | 4473 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4474 | { |
4475 | struct io_fadvise *fa = &req->fadvise; | |
4476 | int ret; | |
4477 | ||
45d189c6 | 4478 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4479 | switch (fa->advice) { |
4480 | case POSIX_FADV_NORMAL: | |
4481 | case POSIX_FADV_RANDOM: | |
4482 | case POSIX_FADV_SEQUENTIAL: | |
4483 | break; | |
4484 | default: | |
4485 | return -EAGAIN; | |
4486 | } | |
4487 | } | |
4840e418 JA |
4488 | |
4489 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4490 | if (ret < 0) | |
4491 | req_set_fail_links(req); | |
e1e16097 | 4492 | io_req_complete(req, ret); |
4840e418 JA |
4493 | return 0; |
4494 | } | |
4495 | ||
eddc7ef5 JA |
4496 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4497 | { | |
6ca56f84 | 4498 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4499 | return -EINVAL; |
eddc7ef5 JA |
4500 | if (sqe->ioprio || sqe->buf_index) |
4501 | return -EINVAL; | |
9c280f90 | 4502 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4503 | return -EBADF; |
eddc7ef5 | 4504 | |
1d9e1288 BM |
4505 | req->statx.dfd = READ_ONCE(sqe->fd); |
4506 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4507 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4508 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4509 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4510 | |
4511 | return 0; | |
4512 | } | |
4513 | ||
45d189c6 | 4514 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4515 | { |
1d9e1288 | 4516 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4517 | int ret; |
4518 | ||
45d189c6 | 4519 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4520 | /* only need file table for an actual valid fd */ |
4521 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4522 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4523 | return -EAGAIN; |
5b0bbee4 | 4524 | } |
eddc7ef5 | 4525 | |
e62753e4 BM |
4526 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4527 | ctx->buffer); | |
eddc7ef5 | 4528 | |
eddc7ef5 JA |
4529 | if (ret < 0) |
4530 | req_set_fail_links(req); | |
e1e16097 | 4531 | io_req_complete(req, ret); |
eddc7ef5 JA |
4532 | return 0; |
4533 | } | |
4534 | ||
b5dba59e JA |
4535 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4536 | { | |
14587a46 | 4537 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4538 | return -EINVAL; |
b5dba59e JA |
4539 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4540 | sqe->rw_flags || sqe->buf_index) | |
4541 | return -EINVAL; | |
9c280f90 | 4542 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4543 | return -EBADF; |
b5dba59e JA |
4544 | |
4545 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4546 | return 0; |
b5dba59e JA |
4547 | } |
4548 | ||
889fca73 | 4549 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4550 | { |
9eac1904 | 4551 | struct files_struct *files = current->files; |
3af73b28 | 4552 | struct io_close *close = &req->close; |
9eac1904 JA |
4553 | struct fdtable *fdt; |
4554 | struct file *file; | |
b5dba59e JA |
4555 | int ret; |
4556 | ||
9eac1904 JA |
4557 | file = NULL; |
4558 | ret = -EBADF; | |
4559 | spin_lock(&files->file_lock); | |
4560 | fdt = files_fdtable(files); | |
4561 | if (close->fd >= fdt->max_fds) { | |
4562 | spin_unlock(&files->file_lock); | |
4563 | goto err; | |
4564 | } | |
4565 | file = fdt->fd[close->fd]; | |
4566 | if (!file) { | |
4567 | spin_unlock(&files->file_lock); | |
4568 | goto err; | |
4569 | } | |
4570 | ||
4571 | if (file->f_op == &io_uring_fops) { | |
4572 | spin_unlock(&files->file_lock); | |
4573 | file = NULL; | |
4574 | goto err; | |
3af73b28 | 4575 | } |
b5dba59e JA |
4576 | |
4577 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4578 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4579 | spin_unlock(&files->file_lock); |
0bf0eefd | 4580 | return -EAGAIN; |
a2100672 | 4581 | } |
b5dba59e | 4582 | |
9eac1904 JA |
4583 | ret = __close_fd_get_file(close->fd, &file); |
4584 | spin_unlock(&files->file_lock); | |
4585 | if (ret < 0) { | |
4586 | if (ret == -ENOENT) | |
4587 | ret = -EBADF; | |
4588 | goto err; | |
4589 | } | |
4590 | ||
3af73b28 | 4591 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4592 | ret = filp_close(file, current->files); |
4593 | err: | |
3af73b28 PB |
4594 | if (ret < 0) |
4595 | req_set_fail_links(req); | |
9eac1904 JA |
4596 | if (file) |
4597 | fput(file); | |
889fca73 | 4598 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4599 | return 0; |
b5dba59e JA |
4600 | } |
4601 | ||
1155c76a | 4602 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4603 | { |
4604 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4605 | |
5d17b4a4 JA |
4606 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4607 | return -EINVAL; | |
4608 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4609 | return -EINVAL; | |
4610 | ||
8ed8d3c3 JA |
4611 | req->sync.off = READ_ONCE(sqe->off); |
4612 | req->sync.len = READ_ONCE(sqe->len); | |
4613 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4614 | return 0; |
4615 | } | |
4616 | ||
45d189c6 | 4617 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4618 | { |
8ed8d3c3 JA |
4619 | int ret; |
4620 | ||
ac45abc0 | 4621 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4622 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4623 | return -EAGAIN; |
4624 | ||
9adbd45d | 4625 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4626 | req->sync.flags); |
4627 | if (ret < 0) | |
4628 | req_set_fail_links(req); | |
e1e16097 | 4629 | io_req_complete(req, ret); |
5d17b4a4 JA |
4630 | return 0; |
4631 | } | |
4632 | ||
469956e8 | 4633 | #if defined(CONFIG_NET) |
02d27d89 PB |
4634 | static int io_setup_async_msg(struct io_kiocb *req, |
4635 | struct io_async_msghdr *kmsg) | |
4636 | { | |
e8c2bc1f JA |
4637 | struct io_async_msghdr *async_msg = req->async_data; |
4638 | ||
4639 | if (async_msg) | |
02d27d89 | 4640 | return -EAGAIN; |
e8c2bc1f | 4641 | if (io_alloc_async_data(req)) { |
257e84a5 | 4642 | kfree(kmsg->free_iov); |
02d27d89 PB |
4643 | return -ENOMEM; |
4644 | } | |
e8c2bc1f | 4645 | async_msg = req->async_data; |
02d27d89 | 4646 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4647 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4648 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4649 | /* if were using fast_iov, set it to the new one */ |
4650 | if (!async_msg->free_iov) | |
4651 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4652 | ||
02d27d89 PB |
4653 | return -EAGAIN; |
4654 | } | |
4655 | ||
2ae523ed PB |
4656 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4657 | struct io_async_msghdr *iomsg) | |
4658 | { | |
2ae523ed | 4659 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4660 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4661 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4662 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4663 | } |
4664 | ||
3529d8c2 | 4665 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4666 | { |
e8c2bc1f | 4667 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4668 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4669 | int ret; |
03b1230c | 4670 | |
d2b6f48b PB |
4671 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4672 | return -EINVAL; | |
4673 | ||
e47293fd | 4674 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4675 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4676 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4677 | |
d8768362 JA |
4678 | #ifdef CONFIG_COMPAT |
4679 | if (req->ctx->compat) | |
4680 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4681 | #endif | |
4682 | ||
e8c2bc1f | 4683 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
3529d8c2 | 4684 | return 0; |
e8c2bc1f | 4685 | ret = io_sendmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4686 | if (!ret) |
4687 | req->flags |= REQ_F_NEED_CLEANUP; | |
4688 | return ret; | |
03b1230c JA |
4689 | } |
4690 | ||
889fca73 | 4691 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4692 | { |
6b754c8b | 4693 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4694 | struct socket *sock; |
7a7cacba | 4695 | unsigned flags; |
0fa03c62 JA |
4696 | int ret; |
4697 | ||
dba4a925 | 4698 | sock = sock_from_file(req->file); |
7a7cacba | 4699 | if (unlikely(!sock)) |
dba4a925 | 4700 | return -ENOTSOCK; |
3529d8c2 | 4701 | |
257e84a5 PB |
4702 | kmsg = req->async_data; |
4703 | if (!kmsg) { | |
7a7cacba PB |
4704 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4705 | if (ret) | |
4706 | return ret; | |
4707 | kmsg = &iomsg; | |
0fa03c62 | 4708 | } |
0fa03c62 | 4709 | |
7a7cacba PB |
4710 | flags = req->sr_msg.msg_flags; |
4711 | if (flags & MSG_DONTWAIT) | |
4712 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4713 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4714 | flags |= MSG_DONTWAIT; |
e47293fd | 4715 | |
7a7cacba | 4716 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4717 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4718 | return io_setup_async_msg(req, kmsg); |
4719 | if (ret == -ERESTARTSYS) | |
4720 | ret = -EINTR; | |
0fa03c62 | 4721 | |
257e84a5 PB |
4722 | /* fast path, check for non-NULL to avoid function call */ |
4723 | if (kmsg->free_iov) | |
4724 | kfree(kmsg->free_iov); | |
99bc4c38 | 4725 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4726 | if (ret < 0) |
4727 | req_set_fail_links(req); | |
889fca73 | 4728 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4729 | return 0; |
03b1230c | 4730 | } |
aa1fa28f | 4731 | |
889fca73 | 4732 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4733 | { |
7a7cacba PB |
4734 | struct io_sr_msg *sr = &req->sr_msg; |
4735 | struct msghdr msg; | |
4736 | struct iovec iov; | |
fddaface | 4737 | struct socket *sock; |
7a7cacba | 4738 | unsigned flags; |
fddaface JA |
4739 | int ret; |
4740 | ||
dba4a925 | 4741 | sock = sock_from_file(req->file); |
7a7cacba | 4742 | if (unlikely(!sock)) |
dba4a925 | 4743 | return -ENOTSOCK; |
fddaface | 4744 | |
7a7cacba PB |
4745 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4746 | if (unlikely(ret)) | |
14db8411 | 4747 | return ret; |
fddaface | 4748 | |
7a7cacba PB |
4749 | msg.msg_name = NULL; |
4750 | msg.msg_control = NULL; | |
4751 | msg.msg_controllen = 0; | |
4752 | msg.msg_namelen = 0; | |
fddaface | 4753 | |
7a7cacba PB |
4754 | flags = req->sr_msg.msg_flags; |
4755 | if (flags & MSG_DONTWAIT) | |
4756 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4757 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4758 | flags |= MSG_DONTWAIT; |
fddaface | 4759 | |
7a7cacba PB |
4760 | msg.msg_flags = flags; |
4761 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4762 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4763 | return -EAGAIN; |
4764 | if (ret == -ERESTARTSYS) | |
4765 | ret = -EINTR; | |
fddaface | 4766 | |
fddaface JA |
4767 | if (ret < 0) |
4768 | req_set_fail_links(req); | |
889fca73 | 4769 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4770 | return 0; |
fddaface JA |
4771 | } |
4772 | ||
1400e697 PB |
4773 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4774 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4775 | { |
4776 | struct io_sr_msg *sr = &req->sr_msg; | |
4777 | struct iovec __user *uiov; | |
4778 | size_t iov_len; | |
4779 | int ret; | |
4780 | ||
1400e697 PB |
4781 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4782 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4783 | if (ret) |
4784 | return ret; | |
4785 | ||
4786 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4787 | if (iov_len > 1) | |
4788 | return -EINVAL; | |
5476dfed | 4789 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4790 | return -EFAULT; |
5476dfed | 4791 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4792 | iomsg->free_iov = NULL; |
52de1fe1 | 4793 | } else { |
257e84a5 | 4794 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4795 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4796 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4797 | false); |
52de1fe1 JA |
4798 | if (ret > 0) |
4799 | ret = 0; | |
4800 | } | |
4801 | ||
4802 | return ret; | |
4803 | } | |
4804 | ||
4805 | #ifdef CONFIG_COMPAT | |
4806 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4807 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4808 | { |
4809 | struct compat_msghdr __user *msg_compat; | |
4810 | struct io_sr_msg *sr = &req->sr_msg; | |
4811 | struct compat_iovec __user *uiov; | |
4812 | compat_uptr_t ptr; | |
4813 | compat_size_t len; | |
4814 | int ret; | |
4815 | ||
270a5940 | 4816 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4817 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4818 | &ptr, &len); |
4819 | if (ret) | |
4820 | return ret; | |
4821 | ||
4822 | uiov = compat_ptr(ptr); | |
4823 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4824 | compat_ssize_t clen; | |
4825 | ||
4826 | if (len > 1) | |
4827 | return -EINVAL; | |
4828 | if (!access_ok(uiov, sizeof(*uiov))) | |
4829 | return -EFAULT; | |
4830 | if (__get_user(clen, &uiov->iov_len)) | |
4831 | return -EFAULT; | |
4832 | if (clen < 0) | |
4833 | return -EINVAL; | |
2d280bc8 | 4834 | sr->len = clen; |
257e84a5 | 4835 | iomsg->free_iov = NULL; |
52de1fe1 | 4836 | } else { |
257e84a5 | 4837 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4838 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4839 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4840 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4841 | if (ret < 0) |
4842 | return ret; | |
4843 | } | |
4844 | ||
4845 | return 0; | |
4846 | } | |
4847 | #endif | |
4848 | ||
1400e697 PB |
4849 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4850 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4851 | { |
1400e697 | 4852 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4853 | |
4854 | #ifdef CONFIG_COMPAT | |
4855 | if (req->ctx->compat) | |
1400e697 | 4856 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4857 | #endif |
52de1fe1 | 4858 | |
1400e697 | 4859 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4860 | } |
4861 | ||
bcda7baa | 4862 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4863 | bool needs_lock) |
bcda7baa JA |
4864 | { |
4865 | struct io_sr_msg *sr = &req->sr_msg; | |
4866 | struct io_buffer *kbuf; | |
4867 | ||
bcda7baa JA |
4868 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4869 | if (IS_ERR(kbuf)) | |
4870 | return kbuf; | |
4871 | ||
4872 | sr->kbuf = kbuf; | |
4873 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4874 | return kbuf; |
fddaface JA |
4875 | } |
4876 | ||
7fbb1b54 PB |
4877 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4878 | { | |
4879 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4880 | } | |
4881 | ||
3529d8c2 JA |
4882 | static int io_recvmsg_prep(struct io_kiocb *req, |
4883 | const struct io_uring_sqe *sqe) | |
aa1fa28f | 4884 | { |
e8c2bc1f | 4885 | struct io_async_msghdr *async_msg = req->async_data; |
e47293fd | 4886 | struct io_sr_msg *sr = &req->sr_msg; |
99bc4c38 | 4887 | int ret; |
3529d8c2 | 4888 | |
d2b6f48b PB |
4889 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4890 | return -EINVAL; | |
4891 | ||
3529d8c2 | 4892 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4893 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4894 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4895 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4896 | |
d8768362 JA |
4897 | #ifdef CONFIG_COMPAT |
4898 | if (req->ctx->compat) | |
4899 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4900 | #endif | |
4901 | ||
e8c2bc1f | 4902 | if (!async_msg || !io_op_defs[req->opcode].needs_async_data) |
5f798bea | 4903 | return 0; |
e8c2bc1f | 4904 | ret = io_recvmsg_copy_hdr(req, async_msg); |
99bc4c38 PB |
4905 | if (!ret) |
4906 | req->flags |= REQ_F_NEED_CLEANUP; | |
4907 | return ret; | |
aa1fa28f JA |
4908 | } |
4909 | ||
889fca73 | 4910 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4911 | { |
6b754c8b | 4912 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4913 | struct socket *sock; |
7fbb1b54 | 4914 | struct io_buffer *kbuf; |
7a7cacba | 4915 | unsigned flags; |
52de1fe1 | 4916 | int ret, cflags = 0; |
45d189c6 | 4917 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4918 | |
dba4a925 | 4919 | sock = sock_from_file(req->file); |
7a7cacba | 4920 | if (unlikely(!sock)) |
dba4a925 | 4921 | return -ENOTSOCK; |
3529d8c2 | 4922 | |
257e84a5 PB |
4923 | kmsg = req->async_data; |
4924 | if (!kmsg) { | |
7a7cacba PB |
4925 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4926 | if (ret) | |
681fda8d | 4927 | return ret; |
7a7cacba PB |
4928 | kmsg = &iomsg; |
4929 | } | |
03b1230c | 4930 | |
bc02ef33 | 4931 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4932 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4933 | if (IS_ERR(kbuf)) |
52de1fe1 | 4934 | return PTR_ERR(kbuf); |
7a7cacba | 4935 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4936 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4937 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4938 | 1, req->sr_msg.len); |
4939 | } | |
52de1fe1 | 4940 | |
7a7cacba PB |
4941 | flags = req->sr_msg.msg_flags; |
4942 | if (flags & MSG_DONTWAIT) | |
4943 | req->flags |= REQ_F_NOWAIT; | |
4944 | else if (force_nonblock) | |
4945 | flags |= MSG_DONTWAIT; | |
e47293fd | 4946 | |
7a7cacba PB |
4947 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4948 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4949 | if (force_nonblock && ret == -EAGAIN) |
4950 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4951 | if (ret == -ERESTARTSYS) |
4952 | ret = -EINTR; | |
03b1230c | 4953 | |
7fbb1b54 PB |
4954 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4955 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4956 | /* fast path, check for non-NULL to avoid function call */ |
4957 | if (kmsg->free_iov) | |
4958 | kfree(kmsg->free_iov); | |
99bc4c38 | 4959 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4960 | if (ret < 0) |
4961 | req_set_fail_links(req); | |
889fca73 | 4962 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4963 | return 0; |
0fa03c62 | 4964 | } |
5d17b4a4 | 4965 | |
889fca73 | 4966 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4967 | { |
6b754c8b | 4968 | struct io_buffer *kbuf; |
7a7cacba PB |
4969 | struct io_sr_msg *sr = &req->sr_msg; |
4970 | struct msghdr msg; | |
4971 | void __user *buf = sr->buf; | |
fddaface | 4972 | struct socket *sock; |
7a7cacba PB |
4973 | struct iovec iov; |
4974 | unsigned flags; | |
bcda7baa | 4975 | int ret, cflags = 0; |
45d189c6 | 4976 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4977 | |
dba4a925 | 4978 | sock = sock_from_file(req->file); |
7a7cacba | 4979 | if (unlikely(!sock)) |
dba4a925 | 4980 | return -ENOTSOCK; |
fddaface | 4981 | |
bc02ef33 | 4982 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4983 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4984 | if (IS_ERR(kbuf)) |
4985 | return PTR_ERR(kbuf); | |
7a7cacba | 4986 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4987 | } |
bcda7baa | 4988 | |
7a7cacba | 4989 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4990 | if (unlikely(ret)) |
4991 | goto out_free; | |
fddaface | 4992 | |
7a7cacba PB |
4993 | msg.msg_name = NULL; |
4994 | msg.msg_control = NULL; | |
4995 | msg.msg_controllen = 0; | |
4996 | msg.msg_namelen = 0; | |
4997 | msg.msg_iocb = NULL; | |
4998 | msg.msg_flags = 0; | |
fddaface | 4999 | |
7a7cacba PB |
5000 | flags = req->sr_msg.msg_flags; |
5001 | if (flags & MSG_DONTWAIT) | |
5002 | req->flags |= REQ_F_NOWAIT; | |
5003 | else if (force_nonblock) | |
5004 | flags |= MSG_DONTWAIT; | |
5005 | ||
5006 | ret = sock_recvmsg(sock, &msg, flags); | |
5007 | if (force_nonblock && ret == -EAGAIN) | |
5008 | return -EAGAIN; | |
5009 | if (ret == -ERESTARTSYS) | |
5010 | ret = -EINTR; | |
14c32eee | 5011 | out_free: |
7fbb1b54 PB |
5012 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5013 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
5014 | if (ret < 0) |
5015 | req_set_fail_links(req); | |
889fca73 | 5016 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5017 | return 0; |
fddaface JA |
5018 | } |
5019 | ||
3529d8c2 | 5020 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5021 | { |
8ed8d3c3 JA |
5022 | struct io_accept *accept = &req->accept; |
5023 | ||
14587a46 | 5024 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5025 | return -EINVAL; |
8042d6ce | 5026 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5027 | return -EINVAL; |
5028 | ||
d55e5f5b JA |
5029 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5030 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5031 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5032 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 5033 | return 0; |
8ed8d3c3 | 5034 | } |
17f2fe35 | 5035 | |
889fca73 | 5036 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5037 | { |
5038 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5039 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5040 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
5041 | int ret; |
5042 | ||
e697deed JX |
5043 | if (req->file->f_flags & O_NONBLOCK) |
5044 | req->flags |= REQ_F_NOWAIT; | |
5045 | ||
8ed8d3c3 | 5046 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
5047 | accept->addr_len, accept->flags, |
5048 | accept->nofile); | |
8ed8d3c3 | 5049 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 5050 | return -EAGAIN; |
ac45abc0 PB |
5051 | if (ret < 0) { |
5052 | if (ret == -ERESTARTSYS) | |
5053 | ret = -EINTR; | |
4e88d6e7 | 5054 | req_set_fail_links(req); |
ac45abc0 | 5055 | } |
889fca73 | 5056 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5057 | return 0; |
8ed8d3c3 JA |
5058 | } |
5059 | ||
3529d8c2 | 5060 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5061 | { |
3529d8c2 | 5062 | struct io_connect *conn = &req->connect; |
e8c2bc1f | 5063 | struct io_async_connect *io = req->async_data; |
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); | |
5072 | ||
5073 | if (!io) | |
5074 | return 0; | |
5075 | ||
5076 | return move_addr_to_kernel(conn->addr, conn->addr_len, | |
e8c2bc1f | 5077 | &io->address); |
f499a021 JA |
5078 | } |
5079 | ||
889fca73 | 5080 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5081 | { |
e8c2bc1f | 5082 | struct io_async_connect __io, *io; |
f8e85cf2 | 5083 | unsigned file_flags; |
3fbb51c1 | 5084 | int ret; |
45d189c6 | 5085 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5086 | |
e8c2bc1f JA |
5087 | if (req->async_data) { |
5088 | io = req->async_data; | |
f499a021 | 5089 | } else { |
3529d8c2 JA |
5090 | ret = move_addr_to_kernel(req->connect.addr, |
5091 | req->connect.addr_len, | |
e8c2bc1f | 5092 | &__io.address); |
f499a021 JA |
5093 | if (ret) |
5094 | goto out; | |
5095 | io = &__io; | |
5096 | } | |
5097 | ||
3fbb51c1 JA |
5098 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5099 | ||
e8c2bc1f | 5100 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5101 | req->connect.addr_len, file_flags); |
87f80d62 | 5102 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5103 | if (req->async_data) |
b7bb4f7d | 5104 | return -EAGAIN; |
e8c2bc1f | 5105 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5106 | ret = -ENOMEM; |
5107 | goto out; | |
5108 | } | |
e8c2bc1f JA |
5109 | io = req->async_data; |
5110 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 5111 | return -EAGAIN; |
f499a021 | 5112 | } |
f8e85cf2 JA |
5113 | if (ret == -ERESTARTSYS) |
5114 | ret = -EINTR; | |
f499a021 | 5115 | out: |
4e88d6e7 JA |
5116 | if (ret < 0) |
5117 | req_set_fail_links(req); | |
889fca73 | 5118 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5119 | return 0; |
469956e8 Y |
5120 | } |
5121 | #else /* !CONFIG_NET */ | |
5122 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5123 | { | |
f8e85cf2 | 5124 | return -EOPNOTSUPP; |
f8e85cf2 JA |
5125 | } |
5126 | ||
889fca73 | 5127 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5128 | { |
5129 | return -EOPNOTSUPP; | |
5130 | } | |
5131 | ||
889fca73 | 5132 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5133 | { |
5134 | return -EOPNOTSUPP; | |
5135 | } | |
5136 | ||
5137 | static int io_recvmsg_prep(struct io_kiocb *req, | |
5138 | const struct io_uring_sqe *sqe) | |
5139 | { | |
5140 | return -EOPNOTSUPP; | |
5141 | } | |
5142 | ||
889fca73 | 5143 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5144 | { |
5145 | return -EOPNOTSUPP; | |
5146 | } | |
5147 | ||
889fca73 | 5148 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5149 | { |
5150 | return -EOPNOTSUPP; | |
5151 | } | |
5152 | ||
5153 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
5154 | { | |
5155 | return -EOPNOTSUPP; | |
5156 | } | |
5157 | ||
889fca73 | 5158 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 Y |
5159 | { |
5160 | return -EOPNOTSUPP; | |
5161 | } | |
ce593a6c | 5162 | |
469956e8 Y |
5163 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5164 | { | |
5165 | return -EOPNOTSUPP; | |
5166 | } | |
5167 | ||
889fca73 | 5168 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
469956e8 | 5169 | { |
f8e85cf2 | 5170 | return -EOPNOTSUPP; |
ce593a6c | 5171 | } |
469956e8 | 5172 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5173 | |
d7718a9d JA |
5174 | struct io_poll_table { |
5175 | struct poll_table_struct pt; | |
5176 | struct io_kiocb *req; | |
5177 | int error; | |
5178 | }; | |
ce593a6c | 5179 | |
d7718a9d JA |
5180 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5181 | __poll_t mask, task_work_func_t func) | |
5182 | { | |
aa96bf8a | 5183 | int ret; |
d7718a9d JA |
5184 | |
5185 | /* for instances that support it check for an event match first: */ | |
5186 | if (mask && !(mask & poll->events)) | |
5187 | return 0; | |
5188 | ||
5189 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5190 | ||
5191 | list_del_init(&poll->wait.entry); | |
5192 | ||
d7718a9d | 5193 | req->result = mask; |
7cbf1722 | 5194 | req->task_work.func = func; |
6d816e08 JA |
5195 | percpu_ref_get(&req->ctx->refs); |
5196 | ||
d7718a9d | 5197 | /* |
e3aabf95 JA |
5198 | * If this fails, then the task is exiting. When a task exits, the |
5199 | * work gets canceled, so just cancel this request as well instead | |
5200 | * of executing it. We can't safely execute it anyway, as we may not | |
5201 | * have the needed state needed for it anyway. | |
d7718a9d | 5202 | */ |
355fb9e2 | 5203 | ret = io_req_task_work_add(req); |
aa96bf8a | 5204 | if (unlikely(ret)) { |
e3aabf95 | 5205 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 5206 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 5207 | } |
d7718a9d JA |
5208 | return 1; |
5209 | } | |
5210 | ||
74ce6ce4 JA |
5211 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5212 | __acquires(&req->ctx->completion_lock) | |
5213 | { | |
5214 | struct io_ring_ctx *ctx = req->ctx; | |
5215 | ||
5216 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5217 | struct poll_table_struct pt = { ._key = poll->events }; | |
5218 | ||
5219 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5220 | } | |
5221 | ||
5222 | spin_lock_irq(&ctx->completion_lock); | |
5223 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5224 | add_wait_queue(poll->head, &poll->wait); | |
5225 | return true; | |
5226 | } | |
5227 | ||
5228 | return false; | |
5229 | } | |
5230 | ||
d4e7cd36 | 5231 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5232 | { |
e8c2bc1f | 5233 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5234 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5235 | return req->async_data; |
d4e7cd36 JA |
5236 | return req->apoll->double_poll; |
5237 | } | |
5238 | ||
5239 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5240 | { | |
5241 | if (req->opcode == IORING_OP_POLL_ADD) | |
5242 | return &req->poll; | |
5243 | return &req->apoll->poll; | |
5244 | } | |
5245 | ||
5246 | static void io_poll_remove_double(struct io_kiocb *req) | |
5247 | { | |
5248 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5249 | |
5250 | lockdep_assert_held(&req->ctx->completion_lock); | |
5251 | ||
5252 | if (poll && poll->head) { | |
5253 | struct wait_queue_head *head = poll->head; | |
5254 | ||
5255 | spin_lock(&head->lock); | |
5256 | list_del_init(&poll->wait.entry); | |
5257 | if (poll->wait.private) | |
5258 | refcount_dec(&req->refs); | |
5259 | poll->head = NULL; | |
5260 | spin_unlock(&head->lock); | |
5261 | } | |
5262 | } | |
5263 | ||
5264 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5265 | { | |
5266 | struct io_ring_ctx *ctx = req->ctx; | |
5267 | ||
d4e7cd36 | 5268 | io_poll_remove_double(req); |
18bceab1 JA |
5269 | req->poll.done = true; |
5270 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5271 | io_commit_cqring(ctx); | |
5272 | } | |
5273 | ||
dd221f46 | 5274 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5275 | { |
dd221f46 | 5276 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5277 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5278 | struct io_kiocb *nxt; |
18bceab1 JA |
5279 | |
5280 | if (io_poll_rewait(req, &req->poll)) { | |
5281 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5282 | } else { |
5283 | hash_del(&req->hash_node); | |
5284 | io_poll_complete(req, req->result, 0); | |
5285 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5286 | |
dd221f46 PB |
5287 | nxt = io_put_req_find_next(req); |
5288 | io_cqring_ev_posted(ctx); | |
5289 | if (nxt) | |
5290 | __io_req_task_submit(nxt); | |
5291 | } | |
18bceab1 | 5292 | |
6d816e08 | 5293 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5294 | } |
5295 | ||
5296 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5297 | int sync, void *key) | |
5298 | { | |
5299 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5300 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5301 | __poll_t mask = key_to_poll(key); |
5302 | ||
5303 | /* for instances that support it check for an event match first: */ | |
5304 | if (mask && !(mask & poll->events)) | |
5305 | return 0; | |
5306 | ||
8706e04e JA |
5307 | list_del_init(&wait->entry); |
5308 | ||
807abcb0 | 5309 | if (poll && poll->head) { |
18bceab1 JA |
5310 | bool done; |
5311 | ||
807abcb0 JA |
5312 | spin_lock(&poll->head->lock); |
5313 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5314 | if (!done) |
807abcb0 | 5315 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5316 | /* make sure double remove sees this as being gone */ |
5317 | wait->private = NULL; | |
807abcb0 | 5318 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5319 | if (!done) { |
5320 | /* use wait func handler, so it matches the rq type */ | |
5321 | poll->wait.func(&poll->wait, mode, sync, key); | |
5322 | } | |
18bceab1 JA |
5323 | } |
5324 | refcount_dec(&req->refs); | |
5325 | return 1; | |
5326 | } | |
5327 | ||
5328 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5329 | wait_queue_func_t wake_func) | |
5330 | { | |
5331 | poll->head = NULL; | |
5332 | poll->done = false; | |
5333 | poll->canceled = false; | |
5334 | poll->events = events; | |
5335 | INIT_LIST_HEAD(&poll->wait.entry); | |
5336 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5337 | } | |
5338 | ||
5339 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5340 | struct wait_queue_head *head, |
5341 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5342 | { |
5343 | struct io_kiocb *req = pt->req; | |
5344 | ||
5345 | /* | |
5346 | * If poll->head is already set, it's because the file being polled | |
5347 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5348 | * for write). Setup a separate io_poll_iocb if this happens. | |
5349 | */ | |
5350 | if (unlikely(poll->head)) { | |
58852d4d PB |
5351 | struct io_poll_iocb *poll_one = poll; |
5352 | ||
18bceab1 | 5353 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5354 | if (*poll_ptr) { |
18bceab1 JA |
5355 | pt->error = -EINVAL; |
5356 | return; | |
5357 | } | |
5358 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5359 | if (!poll) { | |
5360 | pt->error = -ENOMEM; | |
5361 | return; | |
5362 | } | |
58852d4d | 5363 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5364 | refcount_inc(&req->refs); |
5365 | poll->wait.private = req; | |
807abcb0 | 5366 | *poll_ptr = poll; |
18bceab1 JA |
5367 | } |
5368 | ||
5369 | pt->error = 0; | |
5370 | poll->head = head; | |
a31eb4a2 JX |
5371 | |
5372 | if (poll->events & EPOLLEXCLUSIVE) | |
5373 | add_wait_queue_exclusive(head, &poll->wait); | |
5374 | else | |
5375 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5376 | } |
5377 | ||
5378 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5379 | struct poll_table_struct *p) | |
5380 | { | |
5381 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5382 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5383 | |
807abcb0 | 5384 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5385 | } |
5386 | ||
d7718a9d JA |
5387 | static void io_async_task_func(struct callback_head *cb) |
5388 | { | |
5389 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5390 | struct async_poll *apoll = req->apoll; | |
5391 | struct io_ring_ctx *ctx = req->ctx; | |
5392 | ||
5393 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5394 | ||
74ce6ce4 | 5395 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5396 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5397 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5398 | return; |
d7718a9d JA |
5399 | } |
5400 | ||
31067255 | 5401 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5402 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5403 | hash_del(&req->hash_node); |
2bae047e | 5404 | |
d4e7cd36 | 5405 | io_poll_remove_double(req); |
74ce6ce4 JA |
5406 | spin_unlock_irq(&ctx->completion_lock); |
5407 | ||
0be0b0e3 PB |
5408 | if (!READ_ONCE(apoll->poll.canceled)) |
5409 | __io_req_task_submit(req); | |
5410 | else | |
5411 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5412 | |
6d816e08 | 5413 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5414 | kfree(apoll->double_poll); |
31067255 | 5415 | kfree(apoll); |
d7718a9d JA |
5416 | } |
5417 | ||
5418 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5419 | void *key) | |
5420 | { | |
5421 | struct io_kiocb *req = wait->private; | |
5422 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5423 | ||
5424 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5425 | key_to_poll(key)); | |
5426 | ||
5427 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5428 | } | |
5429 | ||
5430 | static void io_poll_req_insert(struct io_kiocb *req) | |
5431 | { | |
5432 | struct io_ring_ctx *ctx = req->ctx; | |
5433 | struct hlist_head *list; | |
5434 | ||
5435 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5436 | hlist_add_head(&req->hash_node, list); | |
5437 | } | |
5438 | ||
5439 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5440 | struct io_poll_iocb *poll, | |
5441 | struct io_poll_table *ipt, __poll_t mask, | |
5442 | wait_queue_func_t wake_func) | |
5443 | __acquires(&ctx->completion_lock) | |
5444 | { | |
5445 | struct io_ring_ctx *ctx = req->ctx; | |
5446 | bool cancel = false; | |
5447 | ||
4d52f338 | 5448 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5449 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5450 | poll->file = req->file; |
18bceab1 | 5451 | poll->wait.private = req; |
d7718a9d JA |
5452 | |
5453 | ipt->pt._key = mask; | |
5454 | ipt->req = req; | |
5455 | ipt->error = -EINVAL; | |
5456 | ||
d7718a9d JA |
5457 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5458 | ||
5459 | spin_lock_irq(&ctx->completion_lock); | |
5460 | if (likely(poll->head)) { | |
5461 | spin_lock(&poll->head->lock); | |
5462 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5463 | if (ipt->error) | |
5464 | cancel = true; | |
5465 | ipt->error = 0; | |
5466 | mask = 0; | |
5467 | } | |
5468 | if (mask || ipt->error) | |
5469 | list_del_init(&poll->wait.entry); | |
5470 | else if (cancel) | |
5471 | WRITE_ONCE(poll->canceled, true); | |
5472 | else if (!poll->done) /* actually waiting for an event */ | |
5473 | io_poll_req_insert(req); | |
5474 | spin_unlock(&poll->head->lock); | |
5475 | } | |
5476 | ||
5477 | return mask; | |
5478 | } | |
5479 | ||
5480 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5481 | { | |
5482 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5483 | struct io_ring_ctx *ctx = req->ctx; | |
5484 | struct async_poll *apoll; | |
5485 | struct io_poll_table ipt; | |
5486 | __poll_t mask, ret; | |
9dab14b8 | 5487 | int rw; |
d7718a9d JA |
5488 | |
5489 | if (!req->file || !file_can_poll(req->file)) | |
5490 | return false; | |
24c74678 | 5491 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5492 | return false; |
9dab14b8 JA |
5493 | if (def->pollin) |
5494 | rw = READ; | |
5495 | else if (def->pollout) | |
5496 | rw = WRITE; | |
5497 | else | |
5498 | return false; | |
5499 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5500 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5501 | return false; |
5502 | ||
5503 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5504 | if (unlikely(!apoll)) | |
5505 | return false; | |
807abcb0 | 5506 | apoll->double_poll = NULL; |
d7718a9d JA |
5507 | |
5508 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5509 | req->apoll = apoll; |
d7718a9d | 5510 | |
8755d97a | 5511 | mask = 0; |
d7718a9d | 5512 | if (def->pollin) |
8755d97a | 5513 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5514 | if (def->pollout) |
5515 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5516 | |
5517 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5518 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5519 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5520 | mask &= ~POLLIN; | |
5521 | ||
d7718a9d JA |
5522 | mask |= POLLERR | POLLPRI; |
5523 | ||
5524 | ipt.pt._qproc = io_async_queue_proc; | |
5525 | ||
5526 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5527 | io_async_wake); | |
a36da65c | 5528 | if (ret || ipt.error) { |
d4e7cd36 | 5529 | io_poll_remove_double(req); |
d7718a9d | 5530 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5531 | kfree(apoll->double_poll); |
d7718a9d JA |
5532 | kfree(apoll); |
5533 | return false; | |
5534 | } | |
5535 | spin_unlock_irq(&ctx->completion_lock); | |
5536 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5537 | apoll->poll.events); | |
5538 | return true; | |
5539 | } | |
5540 | ||
5541 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5542 | struct io_poll_iocb *poll) | |
221c5eb2 | 5543 | { |
b41e9852 | 5544 | bool do_complete = false; |
221c5eb2 JA |
5545 | |
5546 | spin_lock(&poll->head->lock); | |
5547 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5548 | if (!list_empty(&poll->wait.entry)) { |
5549 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5550 | do_complete = true; |
221c5eb2 JA |
5551 | } |
5552 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5553 | hash_del(&req->hash_node); |
d7718a9d JA |
5554 | return do_complete; |
5555 | } | |
5556 | ||
5557 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5558 | { | |
5559 | bool do_complete; | |
5560 | ||
d4e7cd36 JA |
5561 | io_poll_remove_double(req); |
5562 | ||
d7718a9d JA |
5563 | if (req->opcode == IORING_OP_POLL_ADD) { |
5564 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5565 | } else { | |
3bfa5bcb JA |
5566 | struct async_poll *apoll = req->apoll; |
5567 | ||
d7718a9d | 5568 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5569 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5570 | if (do_complete) { | |
d7718a9d | 5571 | io_put_req(req); |
807abcb0 | 5572 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5573 | kfree(apoll); |
5574 | } | |
b1f573bd XW |
5575 | } |
5576 | ||
b41e9852 JA |
5577 | if (do_complete) { |
5578 | io_cqring_fill_event(req, -ECANCELED); | |
5579 | io_commit_cqring(req->ctx); | |
f254ac04 | 5580 | req_set_fail_links(req); |
216578e5 | 5581 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5582 | } |
5583 | ||
5584 | return do_complete; | |
221c5eb2 JA |
5585 | } |
5586 | ||
76e1b642 JA |
5587 | /* |
5588 | * Returns true if we found and killed one or more poll requests | |
5589 | */ | |
6b81928d PB |
5590 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5591 | struct files_struct *files) | |
221c5eb2 | 5592 | { |
78076bb6 | 5593 | struct hlist_node *tmp; |
221c5eb2 | 5594 | struct io_kiocb *req; |
8e2e1faf | 5595 | int posted = 0, i; |
221c5eb2 JA |
5596 | |
5597 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5598 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5599 | struct hlist_head *list; | |
5600 | ||
5601 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5602 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5603 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5604 | posted += io_poll_remove_one(req); |
5605 | } | |
221c5eb2 JA |
5606 | } |
5607 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5608 | |
8e2e1faf JA |
5609 | if (posted) |
5610 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5611 | |
5612 | return posted != 0; | |
221c5eb2 JA |
5613 | } |
5614 | ||
47f46768 JA |
5615 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5616 | { | |
78076bb6 | 5617 | struct hlist_head *list; |
47f46768 JA |
5618 | struct io_kiocb *req; |
5619 | ||
78076bb6 JA |
5620 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5621 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5622 | if (sqe_addr != req->user_data) |
5623 | continue; | |
5624 | if (io_poll_remove_one(req)) | |
eac406c6 | 5625 | return 0; |
b41e9852 | 5626 | return -EALREADY; |
47f46768 JA |
5627 | } |
5628 | ||
5629 | return -ENOENT; | |
5630 | } | |
5631 | ||
3529d8c2 JA |
5632 | static int io_poll_remove_prep(struct io_kiocb *req, |
5633 | const struct io_uring_sqe *sqe) | |
0969e783 | 5634 | { |
0969e783 JA |
5635 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5636 | return -EINVAL; | |
5637 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5638 | sqe->poll_events) | |
5639 | return -EINVAL; | |
5640 | ||
018043be | 5641 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5642 | return 0; |
5643 | } | |
5644 | ||
221c5eb2 JA |
5645 | /* |
5646 | * Find a running poll command that matches one specified in sqe->addr, | |
5647 | * and remove it if found. | |
5648 | */ | |
61e98203 | 5649 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5650 | { |
5651 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5652 | int ret; |
221c5eb2 | 5653 | |
221c5eb2 | 5654 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5655 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5656 | spin_unlock_irq(&ctx->completion_lock); |
5657 | ||
4e88d6e7 JA |
5658 | if (ret < 0) |
5659 | req_set_fail_links(req); | |
e1e16097 | 5660 | io_req_complete(req, ret); |
221c5eb2 JA |
5661 | return 0; |
5662 | } | |
5663 | ||
221c5eb2 JA |
5664 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5665 | void *key) | |
5666 | { | |
c2f2eb7d JA |
5667 | struct io_kiocb *req = wait->private; |
5668 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5669 | |
d7718a9d | 5670 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5671 | } |
5672 | ||
221c5eb2 JA |
5673 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5674 | struct poll_table_struct *p) | |
5675 | { | |
5676 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5677 | ||
e8c2bc1f | 5678 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5679 | } |
5680 | ||
3529d8c2 | 5681 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5682 | { |
5683 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5684 | u32 events; |
221c5eb2 JA |
5685 | |
5686 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5687 | return -EINVAL; | |
5688 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5689 | return -EINVAL; | |
5690 | ||
5769a351 JX |
5691 | events = READ_ONCE(sqe->poll32_events); |
5692 | #ifdef __BIG_ENDIAN | |
5693 | events = swahw32(events); | |
5694 | #endif | |
a31eb4a2 JX |
5695 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5696 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5697 | return 0; |
5698 | } | |
5699 | ||
61e98203 | 5700 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5701 | { |
5702 | struct io_poll_iocb *poll = &req->poll; | |
5703 | struct io_ring_ctx *ctx = req->ctx; | |
5704 | struct io_poll_table ipt; | |
0969e783 | 5705 | __poll_t mask; |
0969e783 | 5706 | |
d7718a9d | 5707 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5708 | |
d7718a9d JA |
5709 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5710 | io_poll_wake); | |
221c5eb2 | 5711 | |
8c838788 | 5712 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5713 | ipt.error = 0; |
b0dd8a41 | 5714 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5715 | } |
221c5eb2 JA |
5716 | spin_unlock_irq(&ctx->completion_lock); |
5717 | ||
8c838788 JA |
5718 | if (mask) { |
5719 | io_cqring_ev_posted(ctx); | |
014db007 | 5720 | io_put_req(req); |
221c5eb2 | 5721 | } |
8c838788 | 5722 | return ipt.error; |
221c5eb2 JA |
5723 | } |
5724 | ||
5262f567 JA |
5725 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5726 | { | |
ad8a48ac JA |
5727 | struct io_timeout_data *data = container_of(timer, |
5728 | struct io_timeout_data, timer); | |
5729 | struct io_kiocb *req = data->req; | |
5730 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5731 | unsigned long flags; |
5732 | ||
5262f567 | 5733 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5734 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5735 | atomic_set(&req->ctx->cq_timeouts, |
5736 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5737 | ||
78e19bbe | 5738 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5739 | io_commit_cqring(ctx); |
5740 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5741 | ||
5742 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5743 | req_set_fail_links(req); |
5262f567 JA |
5744 | io_put_req(req); |
5745 | return HRTIMER_NORESTART; | |
5746 | } | |
5747 | ||
fbd15848 PB |
5748 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5749 | __u64 user_data) | |
f254ac04 | 5750 | { |
fbd15848 | 5751 | struct io_timeout_data *io; |
47f46768 JA |
5752 | struct io_kiocb *req; |
5753 | int ret = -ENOENT; | |
f254ac04 | 5754 | |
135fcde8 | 5755 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5756 | if (user_data == req->user_data) { |
47f46768 JA |
5757 | ret = 0; |
5758 | break; | |
5759 | } | |
5760 | } | |
5761 | ||
5762 | if (ret == -ENOENT) | |
fbd15848 PB |
5763 | return ERR_PTR(ret); |
5764 | ||
5765 | io = req->async_data; | |
e8c2bc1f | 5766 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5767 | if (ret == -1) |
fbd15848 | 5768 | return ERR_PTR(-EALREADY); |
a71976f3 | 5769 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5770 | return req; |
5771 | } | |
47f46768 | 5772 | |
fbd15848 PB |
5773 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5774 | { | |
5775 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5776 | ||
5777 | if (IS_ERR(req)) | |
5778 | return PTR_ERR(req); | |
f254ac04 JA |
5779 | |
5780 | req_set_fail_links(req); | |
f254ac04 | 5781 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5782 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5783 | return 0; |
5784 | } | |
5785 | ||
9c8e11b3 PB |
5786 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5787 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5788 | { |
9c8e11b3 PB |
5789 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5790 | struct io_timeout_data *data; | |
47f46768 | 5791 | |
9c8e11b3 PB |
5792 | if (IS_ERR(req)) |
5793 | return PTR_ERR(req); | |
47f46768 | 5794 | |
9c8e11b3 PB |
5795 | req->timeout.off = 0; /* noseq */ |
5796 | data = req->async_data; | |
5797 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5798 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5799 | data->timer.function = io_timeout_fn; | |
5800 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5801 | return 0; | |
47f46768 JA |
5802 | } |
5803 | ||
3529d8c2 JA |
5804 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5805 | const struct io_uring_sqe *sqe) | |
b29472ee | 5806 | { |
9c8e11b3 PB |
5807 | struct io_timeout_rem *tr = &req->timeout_rem; |
5808 | ||
b29472ee JA |
5809 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5810 | return -EINVAL; | |
61710e43 DA |
5811 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5812 | return -EINVAL; | |
9c8e11b3 | 5813 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5814 | return -EINVAL; |
5815 | ||
9c8e11b3 PB |
5816 | tr->addr = READ_ONCE(sqe->addr); |
5817 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5818 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5819 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5820 | return -EINVAL; | |
5821 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5822 | return -EFAULT; | |
5823 | } else if (tr->flags) { | |
5824 | /* timeout removal doesn't support flags */ | |
b29472ee | 5825 | return -EINVAL; |
9c8e11b3 | 5826 | } |
b29472ee | 5827 | |
b29472ee JA |
5828 | return 0; |
5829 | } | |
5830 | ||
8662daec PB |
5831 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5832 | { | |
5833 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5834 | : HRTIMER_MODE_REL; | |
5835 | } | |
5836 | ||
11365043 JA |
5837 | /* |
5838 | * Remove or update an existing timeout command | |
5839 | */ | |
61e98203 | 5840 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5841 | { |
9c8e11b3 | 5842 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5843 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5844 | int ret; |
11365043 | 5845 | |
11365043 | 5846 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5847 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5848 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5849 | else |
5850 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5851 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5852 | |
47f46768 | 5853 | io_cqring_fill_event(req, ret); |
11365043 JA |
5854 | io_commit_cqring(ctx); |
5855 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5856 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5857 | if (ret < 0) |
5858 | req_set_fail_links(req); | |
ec9c02ad | 5859 | io_put_req(req); |
11365043 | 5860 | return 0; |
5262f567 JA |
5861 | } |
5862 | ||
3529d8c2 | 5863 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5864 | bool is_timeout_link) |
5262f567 | 5865 | { |
ad8a48ac | 5866 | struct io_timeout_data *data; |
a41525ab | 5867 | unsigned flags; |
56080b02 | 5868 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5869 | |
ad8a48ac | 5870 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5871 | return -EINVAL; |
ad8a48ac | 5872 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5873 | return -EINVAL; |
56080b02 | 5874 | if (off && is_timeout_link) |
2d28390a | 5875 | return -EINVAL; |
a41525ab JA |
5876 | flags = READ_ONCE(sqe->timeout_flags); |
5877 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5878 | return -EINVAL; |
bdf20073 | 5879 | |
bfe68a22 | 5880 | req->timeout.off = off; |
26a61679 | 5881 | |
e8c2bc1f | 5882 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5883 | return -ENOMEM; |
5884 | ||
e8c2bc1f | 5885 | data = req->async_data; |
ad8a48ac | 5886 | data->req = req; |
ad8a48ac JA |
5887 | |
5888 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5889 | return -EFAULT; |
5890 | ||
8662daec | 5891 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5892 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5893 | return 0; | |
5894 | } | |
5895 | ||
61e98203 | 5896 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5897 | { |
ad8a48ac | 5898 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5899 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5900 | struct list_head *entry; |
bfe68a22 | 5901 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5902 | |
733f5c95 | 5903 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5904 | |
5262f567 JA |
5905 | /* |
5906 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5907 | * timeout event to be satisfied. If it isn't set, then this is |
5908 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5909 | */ |
8eb7e2d0 | 5910 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5911 | entry = ctx->timeout_list.prev; |
5912 | goto add; | |
5913 | } | |
5262f567 | 5914 | |
bfe68a22 PB |
5915 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5916 | req->timeout.target_seq = tail + off; | |
5262f567 | 5917 | |
f010505b MDG |
5918 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5919 | * This is safe because ->completion_lock is held, and submissions | |
5920 | * and completions are never mixed in the same ->completion_lock section. | |
5921 | */ | |
5922 | ctx->cq_last_tm_flush = tail; | |
5923 | ||
5262f567 JA |
5924 | /* |
5925 | * Insertion sort, ensuring the first entry in the list is always | |
5926 | * the one we need first. | |
5927 | */ | |
5262f567 | 5928 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5929 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5930 | timeout.list); | |
5262f567 | 5931 | |
8eb7e2d0 | 5932 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5933 | continue; |
bfe68a22 PB |
5934 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5935 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5936 | break; |
5937 | } | |
93bd25bb | 5938 | add: |
135fcde8 | 5939 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5940 | data->timer.function = io_timeout_fn; |
5941 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5942 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5943 | return 0; |
5944 | } | |
5262f567 | 5945 | |
62755e35 JA |
5946 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5947 | { | |
5948 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5949 | ||
5950 | return req->user_data == (unsigned long) data; | |
5951 | } | |
5952 | ||
e977d6d3 | 5953 | static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr) |
62755e35 | 5954 | { |
62755e35 | 5955 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5956 | int ret = 0; |
5957 | ||
4f26bda1 | 5958 | cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false); |
62755e35 JA |
5959 | switch (cancel_ret) { |
5960 | case IO_WQ_CANCEL_OK: | |
5961 | ret = 0; | |
5962 | break; | |
5963 | case IO_WQ_CANCEL_RUNNING: | |
5964 | ret = -EALREADY; | |
5965 | break; | |
5966 | case IO_WQ_CANCEL_NOTFOUND: | |
5967 | ret = -ENOENT; | |
5968 | break; | |
5969 | } | |
5970 | ||
e977d6d3 JA |
5971 | return ret; |
5972 | } | |
5973 | ||
47f46768 JA |
5974 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5975 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5976 | int success_ret) |
47f46768 JA |
5977 | { |
5978 | unsigned long flags; | |
5979 | int ret; | |
5980 | ||
5981 | ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr); | |
5982 | if (ret != -ENOENT) { | |
5983 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5984 | goto done; | |
5985 | } | |
5986 | ||
5987 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5988 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5989 | if (ret != -ENOENT) | |
5990 | goto done; | |
5991 | ret = io_poll_cancel(ctx, sqe_addr); | |
5992 | done: | |
b0dd8a41 JA |
5993 | if (!ret) |
5994 | ret = success_ret; | |
47f46768 JA |
5995 | io_cqring_fill_event(req, ret); |
5996 | io_commit_cqring(ctx); | |
5997 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5998 | io_cqring_ev_posted(ctx); | |
5999 | ||
4e88d6e7 JA |
6000 | if (ret < 0) |
6001 | req_set_fail_links(req); | |
014db007 | 6002 | io_put_req(req); |
47f46768 JA |
6003 | } |
6004 | ||
3529d8c2 JA |
6005 | static int io_async_cancel_prep(struct io_kiocb *req, |
6006 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6007 | { |
fbf23849 | 6008 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6009 | return -EINVAL; |
61710e43 DA |
6010 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6011 | return -EINVAL; | |
6012 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
6013 | return -EINVAL; |
6014 | ||
fbf23849 JA |
6015 | req->cancel.addr = READ_ONCE(sqe->addr); |
6016 | return 0; | |
6017 | } | |
6018 | ||
61e98203 | 6019 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6020 | { |
6021 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 6022 | |
014db007 | 6023 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
6024 | return 0; |
6025 | } | |
6026 | ||
269bbe5f | 6027 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6028 | const struct io_uring_sqe *sqe) |
6029 | { | |
6ca56f84 JA |
6030 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
6031 | return -EINVAL; | |
61710e43 DA |
6032 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6033 | return -EINVAL; | |
6034 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
6035 | return -EINVAL; |
6036 | ||
269bbe5f BM |
6037 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6038 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6039 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6040 | return -EINVAL; |
269bbe5f | 6041 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6042 | return 0; |
6043 | } | |
6044 | ||
889fca73 | 6045 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6046 | { |
6047 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 6048 | struct io_uring_rsrc_update up; |
05f3fb3c | 6049 | int ret; |
fbf23849 | 6050 | |
45d189c6 | 6051 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6052 | return -EAGAIN; |
05f3fb3c | 6053 | |
269bbe5f BM |
6054 | up.offset = req->rsrc_update.offset; |
6055 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
6056 | |
6057 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 6058 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6059 | mutex_unlock(&ctx->uring_lock); |
6060 | ||
6061 | if (ret < 0) | |
6062 | req_set_fail_links(req); | |
889fca73 | 6063 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6064 | return 0; |
6065 | } | |
6066 | ||
bfe76559 | 6067 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6068 | { |
d625c6ee | 6069 | switch (req->opcode) { |
e781573e | 6070 | case IORING_OP_NOP: |
bfe76559 | 6071 | return 0; |
f67676d1 JA |
6072 | case IORING_OP_READV: |
6073 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6074 | case IORING_OP_READ: |
bfe76559 | 6075 | return io_read_prep(req, sqe); |
f67676d1 JA |
6076 | case IORING_OP_WRITEV: |
6077 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6078 | case IORING_OP_WRITE: |
bfe76559 | 6079 | return io_write_prep(req, sqe); |
0969e783 | 6080 | case IORING_OP_POLL_ADD: |
bfe76559 | 6081 | return io_poll_add_prep(req, sqe); |
0969e783 | 6082 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 6083 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 6084 | case IORING_OP_FSYNC: |
1155c76a | 6085 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6086 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6087 | return io_sfr_prep(req, sqe); |
03b1230c | 6088 | case IORING_OP_SENDMSG: |
fddaface | 6089 | case IORING_OP_SEND: |
bfe76559 | 6090 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6091 | case IORING_OP_RECVMSG: |
fddaface | 6092 | case IORING_OP_RECV: |
bfe76559 | 6093 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6094 | case IORING_OP_CONNECT: |
bfe76559 | 6095 | return io_connect_prep(req, sqe); |
2d28390a | 6096 | case IORING_OP_TIMEOUT: |
bfe76559 | 6097 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6098 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6099 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6100 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6101 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6102 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6103 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6104 | case IORING_OP_ACCEPT: |
bfe76559 | 6105 | return io_accept_prep(req, sqe); |
d63d1b5e | 6106 | case IORING_OP_FALLOCATE: |
bfe76559 | 6107 | return io_fallocate_prep(req, sqe); |
15b71abe | 6108 | case IORING_OP_OPENAT: |
bfe76559 | 6109 | return io_openat_prep(req, sqe); |
b5dba59e | 6110 | case IORING_OP_CLOSE: |
bfe76559 | 6111 | return io_close_prep(req, sqe); |
05f3fb3c | 6112 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6113 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6114 | case IORING_OP_STATX: |
bfe76559 | 6115 | return io_statx_prep(req, sqe); |
4840e418 | 6116 | case IORING_OP_FADVISE: |
bfe76559 | 6117 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6118 | case IORING_OP_MADVISE: |
bfe76559 | 6119 | return io_madvise_prep(req, sqe); |
cebdb986 | 6120 | case IORING_OP_OPENAT2: |
bfe76559 | 6121 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6122 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6123 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6124 | case IORING_OP_SPLICE: |
bfe76559 | 6125 | return io_splice_prep(req, sqe); |
ddf0322d | 6126 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6127 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6128 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6129 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6130 | case IORING_OP_TEE: |
bfe76559 | 6131 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6132 | case IORING_OP_SHUTDOWN: |
6133 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6134 | case IORING_OP_RENAMEAT: |
6135 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6136 | case IORING_OP_UNLINKAT: |
6137 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6138 | } |
6139 | ||
bfe76559 PB |
6140 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6141 | req->opcode); | |
6142 | return-EINVAL; | |
6143 | } | |
6144 | ||
6145 | static int io_req_defer_prep(struct io_kiocb *req, | |
6146 | const struct io_uring_sqe *sqe) | |
6147 | { | |
bfe76559 PB |
6148 | if (!sqe) |
6149 | return 0; | |
6150 | if (io_alloc_async_data(req)) | |
6151 | return -EAGAIN; | |
bfe76559 | 6152 | return io_req_prep(req, sqe); |
f67676d1 JA |
6153 | } |
6154 | ||
9cf7c104 PB |
6155 | static u32 io_get_sequence(struct io_kiocb *req) |
6156 | { | |
6157 | struct io_kiocb *pos; | |
6158 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 6159 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 6160 | |
f2f87370 PB |
6161 | io_for_each_link(pos, req) |
6162 | nr_reqs++; | |
9cf7c104 PB |
6163 | |
6164 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6165 | return total_submitted - nr_reqs; | |
6166 | } | |
6167 | ||
3529d8c2 | 6168 | static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
de0617e4 | 6169 | { |
a197f664 | 6170 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6171 | struct io_defer_entry *de; |
f67676d1 | 6172 | int ret; |
9cf7c104 | 6173 | u32 seq; |
de0617e4 | 6174 | |
9d858b21 | 6175 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6176 | if (likely(list_empty_careful(&ctx->defer_list) && |
6177 | !(req->flags & REQ_F_IO_DRAIN))) | |
6178 | return 0; | |
6179 | ||
6180 | seq = io_get_sequence(req); | |
6181 | /* Still a chance to pass the sequence check */ | |
6182 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6183 | return 0; |
6184 | ||
e8c2bc1f | 6185 | if (!req->async_data) { |
650b5481 | 6186 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6187 | if (ret) |
650b5481 PB |
6188 | return ret; |
6189 | } | |
cbdcb435 | 6190 | io_prep_async_link(req); |
27dc8338 PB |
6191 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6192 | if (!de) | |
6193 | return -ENOMEM; | |
2d28390a | 6194 | |
de0617e4 | 6195 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6196 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6197 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6198 | kfree(de); |
ae34817b PB |
6199 | io_queue_async_work(req); |
6200 | return -EIOCBQUEUED; | |
de0617e4 JA |
6201 | } |
6202 | ||
915967f6 | 6203 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6204 | de->req = req; |
9cf7c104 | 6205 | de->seq = seq; |
27dc8338 | 6206 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6207 | spin_unlock_irq(&ctx->completion_lock); |
6208 | return -EIOCBQUEUED; | |
6209 | } | |
6210 | ||
3ca405eb | 6211 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6212 | { |
0e1b6fe3 PB |
6213 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6214 | switch (req->opcode) { | |
6215 | case IORING_OP_READV: | |
6216 | case IORING_OP_READ_FIXED: | |
6217 | case IORING_OP_READ: | |
bcda7baa | 6218 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6219 | break; |
6220 | case IORING_OP_RECVMSG: | |
6221 | case IORING_OP_RECV: | |
bcda7baa | 6222 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6223 | break; |
6224 | } | |
6225 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6226 | } |
6227 | ||
0e1b6fe3 PB |
6228 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6229 | switch (req->opcode) { | |
6230 | case IORING_OP_READV: | |
6231 | case IORING_OP_READ_FIXED: | |
6232 | case IORING_OP_READ: | |
6233 | case IORING_OP_WRITEV: | |
6234 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6235 | case IORING_OP_WRITE: { |
6236 | struct io_async_rw *io = req->async_data; | |
6237 | if (io->free_iovec) | |
6238 | kfree(io->free_iovec); | |
0e1b6fe3 | 6239 | break; |
e8c2bc1f | 6240 | } |
0e1b6fe3 | 6241 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6242 | case IORING_OP_SENDMSG: { |
6243 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6244 | |
6245 | kfree(io->free_iov); | |
0e1b6fe3 | 6246 | break; |
e8c2bc1f | 6247 | } |
0e1b6fe3 PB |
6248 | case IORING_OP_SPLICE: |
6249 | case IORING_OP_TEE: | |
6250 | io_put_file(req, req->splice.file_in, | |
6251 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6252 | break; | |
f3cd4850 JA |
6253 | case IORING_OP_OPENAT: |
6254 | case IORING_OP_OPENAT2: | |
6255 | if (req->open.filename) | |
6256 | putname(req->open.filename); | |
6257 | break; | |
80a261fd JA |
6258 | case IORING_OP_RENAMEAT: |
6259 | putname(req->rename.oldpath); | |
6260 | putname(req->rename.newpath); | |
6261 | break; | |
14a1143b JA |
6262 | case IORING_OP_UNLINKAT: |
6263 | putname(req->unlink.filename); | |
6264 | break; | |
0e1b6fe3 PB |
6265 | } |
6266 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6267 | } |
99bc4c38 PB |
6268 | } |
6269 | ||
889fca73 | 6270 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6271 | { |
a197f664 | 6272 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 6273 | int ret; |
2b188cc1 | 6274 | |
d625c6ee | 6275 | switch (req->opcode) { |
2b188cc1 | 6276 | case IORING_OP_NOP: |
889fca73 | 6277 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6278 | break; |
6279 | case IORING_OP_READV: | |
edafccee | 6280 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6281 | case IORING_OP_READ: |
889fca73 | 6282 | ret = io_read(req, issue_flags); |
edafccee | 6283 | break; |
3529d8c2 | 6284 | case IORING_OP_WRITEV: |
edafccee | 6285 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6286 | case IORING_OP_WRITE: |
889fca73 | 6287 | ret = io_write(req, issue_flags); |
2b188cc1 | 6288 | break; |
c992fe29 | 6289 | case IORING_OP_FSYNC: |
45d189c6 | 6290 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6291 | break; |
221c5eb2 | 6292 | case IORING_OP_POLL_ADD: |
61e98203 | 6293 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6294 | break; |
6295 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6296 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6297 | break; |
5d17b4a4 | 6298 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6299 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6300 | break; |
0fa03c62 | 6301 | case IORING_OP_SENDMSG: |
889fca73 | 6302 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6303 | break; |
fddaface | 6304 | case IORING_OP_SEND: |
889fca73 | 6305 | ret = io_send(req, issue_flags); |
0fa03c62 | 6306 | break; |
aa1fa28f | 6307 | case IORING_OP_RECVMSG: |
889fca73 | 6308 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6309 | break; |
fddaface | 6310 | case IORING_OP_RECV: |
889fca73 | 6311 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6312 | break; |
5262f567 | 6313 | case IORING_OP_TIMEOUT: |
61e98203 | 6314 | ret = io_timeout(req, issue_flags); |
5262f567 | 6315 | break; |
11365043 | 6316 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6317 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6318 | break; |
17f2fe35 | 6319 | case IORING_OP_ACCEPT: |
889fca73 | 6320 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6321 | break; |
f8e85cf2 | 6322 | case IORING_OP_CONNECT: |
889fca73 | 6323 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6324 | break; |
62755e35 | 6325 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6326 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6327 | break; |
d63d1b5e | 6328 | case IORING_OP_FALLOCATE: |
45d189c6 | 6329 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6330 | break; |
15b71abe | 6331 | case IORING_OP_OPENAT: |
45d189c6 | 6332 | ret = io_openat(req, issue_flags); |
15b71abe | 6333 | break; |
b5dba59e | 6334 | case IORING_OP_CLOSE: |
889fca73 | 6335 | ret = io_close(req, issue_flags); |
b5dba59e | 6336 | break; |
05f3fb3c | 6337 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6338 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6339 | break; |
eddc7ef5 | 6340 | case IORING_OP_STATX: |
45d189c6 | 6341 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6342 | break; |
4840e418 | 6343 | case IORING_OP_FADVISE: |
45d189c6 | 6344 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6345 | break; |
c1ca757b | 6346 | case IORING_OP_MADVISE: |
45d189c6 | 6347 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6348 | break; |
cebdb986 | 6349 | case IORING_OP_OPENAT2: |
45d189c6 | 6350 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6351 | break; |
3e4827b0 | 6352 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6353 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6354 | break; |
7d67af2c | 6355 | case IORING_OP_SPLICE: |
45d189c6 | 6356 | ret = io_splice(req, issue_flags); |
7d67af2c | 6357 | break; |
ddf0322d | 6358 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6359 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6360 | break; |
067524e9 | 6361 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6362 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6363 | break; |
f2a8d5c7 | 6364 | case IORING_OP_TEE: |
45d189c6 | 6365 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6366 | break; |
36f4fa68 | 6367 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6368 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6369 | break; |
80a261fd | 6370 | case IORING_OP_RENAMEAT: |
45d189c6 | 6371 | ret = io_renameat(req, issue_flags); |
80a261fd | 6372 | break; |
14a1143b | 6373 | case IORING_OP_UNLINKAT: |
45d189c6 | 6374 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6375 | break; |
2b188cc1 JA |
6376 | default: |
6377 | ret = -EINVAL; | |
6378 | break; | |
6379 | } | |
6380 | ||
def596e9 JA |
6381 | if (ret) |
6382 | return ret; | |
6383 | ||
b532576e JA |
6384 | /* If the op doesn't have a file, we're not polling for it */ |
6385 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6386 | const bool in_async = io_wq_current_is_worker(); |
6387 | ||
11ba820b JA |
6388 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6389 | if (in_async) | |
6390 | mutex_lock(&ctx->uring_lock); | |
6391 | ||
2e9dbe90 | 6392 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6393 | |
6394 | if (in_async) | |
6395 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6396 | } |
6397 | ||
6398 | return 0; | |
2b188cc1 JA |
6399 | } |
6400 | ||
5280f7e5 | 6401 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6402 | { |
6403 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6404 | struct io_kiocb *timeout; |
561fb04a | 6405 | int ret = 0; |
2b188cc1 | 6406 | |
6df1db6b PB |
6407 | timeout = io_prep_linked_timeout(req); |
6408 | if (timeout) | |
6409 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6410 | |
4014d943 | 6411 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6412 | ret = -ECANCELED; |
31b51510 | 6413 | |
561fb04a | 6414 | if (!ret) { |
561fb04a | 6415 | do { |
889fca73 | 6416 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6417 | /* |
6418 | * We can get EAGAIN for polled IO even though we're | |
6419 | * forcing a sync submission from here, since we can't | |
6420 | * wait for request slots on the block side. | |
6421 | */ | |
6422 | if (ret != -EAGAIN) | |
6423 | break; | |
6424 | cond_resched(); | |
6425 | } while (1); | |
6426 | } | |
31b51510 | 6427 | |
561fb04a | 6428 | if (ret) { |
c07e6719 XW |
6429 | struct io_ring_ctx *lock_ctx = NULL; |
6430 | ||
6431 | if (req->ctx->flags & IORING_SETUP_IOPOLL) | |
6432 | lock_ctx = req->ctx; | |
6433 | ||
dad1b124 | 6434 | /* |
c07e6719 XW |
6435 | * io_iopoll_complete() does not hold completion_lock to |
6436 | * complete polled io, so here for polled io, we can not call | |
6437 | * io_req_complete() directly, otherwise there maybe concurrent | |
6438 | * access to cqring, defer_list, etc, which is not safe. Given | |
6439 | * that io_iopoll_complete() is always called under uring_lock, | |
6440 | * so here for polled io, we also get uring_lock to complete | |
6441 | * it. | |
dad1b124 | 6442 | */ |
c07e6719 XW |
6443 | if (lock_ctx) |
6444 | mutex_lock(&lock_ctx->uring_lock); | |
dad1b124 | 6445 | |
c07e6719 XW |
6446 | req_set_fail_links(req); |
6447 | io_req_complete(req, ret); | |
6448 | ||
6449 | if (lock_ctx) | |
6450 | mutex_unlock(&lock_ctx->uring_lock); | |
edafccee | 6451 | } |
2b188cc1 JA |
6452 | } |
6453 | ||
65e19f54 JA |
6454 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6455 | int index) | |
6456 | { | |
269bbe5f | 6457 | struct fixed_rsrc_table *table; |
65e19f54 | 6458 | |
05f3fb3c | 6459 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6460 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6461 | } |
6462 | ||
8371adf5 PB |
6463 | static struct file *io_file_get(struct io_submit_state *state, |
6464 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6465 | { |
a197f664 | 6466 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6467 | struct file *file; |
09bb8394 | 6468 | |
8da11c19 | 6469 | if (fixed) { |
479f517b | 6470 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6471 | return NULL; |
b7620121 | 6472 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6473 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6474 | io_set_resource_node(req); |
09bb8394 | 6475 | } else { |
c826bd7a | 6476 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6477 | file = __io_file_get(state, fd); |
09bb8394 JA |
6478 | } |
6479 | ||
ce3d5aae PB |
6480 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6481 | io_req_track_inflight(req); | |
8371adf5 | 6482 | return file; |
09bb8394 JA |
6483 | } |
6484 | ||
2665abfd | 6485 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6486 | { |
ad8a48ac JA |
6487 | struct io_timeout_data *data = container_of(timer, |
6488 | struct io_timeout_data, timer); | |
90cd7e42 | 6489 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6490 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6491 | unsigned long flags; |
2665abfd JA |
6492 | |
6493 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6494 | prev = req->timeout.head; |
6495 | req->timeout.head = NULL; | |
2665abfd JA |
6496 | |
6497 | /* | |
6498 | * We don't expect the list to be empty, that will only happen if we | |
6499 | * race with the completion of the linked work. | |
6500 | */ | |
90cd7e42 | 6501 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6502 | io_remove_next_linked(prev); |
90cd7e42 PB |
6503 | else |
6504 | prev = NULL; | |
2665abfd JA |
6505 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6506 | ||
6507 | if (prev) { | |
4e88d6e7 | 6508 | req_set_fail_links(prev); |
014db007 | 6509 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6510 | io_put_req_deferred(prev, 1); |
47f46768 | 6511 | } else { |
9ae1f8dd PB |
6512 | io_req_complete_post(req, -ETIME, 0); |
6513 | io_put_req_deferred(req, 1); | |
2665abfd | 6514 | } |
2665abfd JA |
6515 | return HRTIMER_NORESTART; |
6516 | } | |
6517 | ||
7271ef3a | 6518 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6519 | { |
76a46e06 | 6520 | /* |
f2f87370 PB |
6521 | * If the back reference is NULL, then our linked request finished |
6522 | * before we got a chance to setup the timer | |
76a46e06 | 6523 | */ |
90cd7e42 | 6524 | if (req->timeout.head) { |
e8c2bc1f | 6525 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6526 | |
ad8a48ac JA |
6527 | data->timer.function = io_link_timeout_fn; |
6528 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6529 | data->mode); | |
2665abfd | 6530 | } |
7271ef3a JA |
6531 | } |
6532 | ||
6533 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6534 | { | |
6535 | struct io_ring_ctx *ctx = req->ctx; | |
6536 | ||
6537 | spin_lock_irq(&ctx->completion_lock); | |
6538 | __io_queue_linked_timeout(req); | |
76a46e06 | 6539 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6540 | |
2665abfd | 6541 | /* drop submission reference */ |
76a46e06 JA |
6542 | io_put_req(req); |
6543 | } | |
2665abfd | 6544 | |
ad8a48ac | 6545 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6546 | { |
f2f87370 | 6547 | struct io_kiocb *nxt = req->link; |
2665abfd | 6548 | |
f2f87370 PB |
6549 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6550 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6551 | return NULL; |
2665abfd | 6552 | |
90cd7e42 | 6553 | nxt->timeout.head = req; |
900fad45 | 6554 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6555 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6556 | return nxt; |
2665abfd JA |
6557 | } |
6558 | ||
c5eef2b9 | 6559 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6560 | { |
d3d7298d | 6561 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
193155c8 | 6562 | const struct cred *old_creds = NULL; |
c5eef2b9 | 6563 | int ret; |
2b188cc1 | 6564 | |
2e5aa6cb PB |
6565 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6566 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
d3d7298d PB |
6567 | req->work.identity->creds != current_cred()) |
6568 | old_creds = override_creds(req->work.identity->creds); | |
193155c8 | 6569 | |
c5eef2b9 | 6570 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
491381ce | 6571 | |
d3d7298d PB |
6572 | if (old_creds) |
6573 | revert_creds(old_creds); | |
6574 | ||
491381ce JA |
6575 | /* |
6576 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6577 | * doesn't support non-blocking read/write attempts | |
6578 | */ | |
24c74678 | 6579 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6580 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6581 | /* |
6582 | * Queued up for async execution, worker will release | |
6583 | * submit reference when the iocb is actually submitted. | |
6584 | */ | |
6585 | io_queue_async_work(req); | |
2b188cc1 | 6586 | } |
0d63c148 PB |
6587 | } else if (likely(!ret)) { |
6588 | /* drop submission reference */ | |
e342c807 | 6589 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6590 | struct io_ring_ctx *ctx = req->ctx; |
6591 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
6592 | ||
6dd0be1e | 6593 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6594 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6595 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6596 | } else { |
d3d7298d | 6597 | io_put_req(req); |
0d63c148 PB |
6598 | } |
6599 | } else { | |
4e88d6e7 | 6600 | req_set_fail_links(req); |
e65ef56d | 6601 | io_put_req(req); |
e1e16097 | 6602 | io_req_complete(req, ret); |
9e645e11 | 6603 | } |
d3d7298d PB |
6604 | if (linked_timeout) |
6605 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6606 | } |
6607 | ||
c5eef2b9 | 6608 | static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4fe2c963 JL |
6609 | { |
6610 | int ret; | |
6611 | ||
3529d8c2 | 6612 | ret = io_req_defer(req, sqe); |
4fe2c963 JL |
6613 | if (ret) { |
6614 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6615 | fail_req: |
4e88d6e7 | 6616 | req_set_fail_links(req); |
e1e16097 JA |
6617 | io_put_req(req); |
6618 | io_req_complete(req, ret); | |
4fe2c963 | 6619 | } |
2550878f | 6620 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
e8c2bc1f | 6621 | if (!req->async_data) { |
bd2ab18a | 6622 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6623 | if (unlikely(ret)) |
bd2ab18a PB |
6624 | goto fail_req; |
6625 | } | |
ce35a47a JA |
6626 | io_queue_async_work(req); |
6627 | } else { | |
c1379e24 PB |
6628 | if (sqe) { |
6629 | ret = io_req_prep(req, sqe); | |
6630 | if (unlikely(ret)) | |
6631 | goto fail_req; | |
6632 | } | |
c5eef2b9 | 6633 | __io_queue_sqe(req); |
ce35a47a | 6634 | } |
4fe2c963 JL |
6635 | } |
6636 | ||
c5eef2b9 | 6637 | static inline void io_queue_link_head(struct io_kiocb *req) |
4fe2c963 | 6638 | { |
94ae5e77 | 6639 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { |
e1e16097 JA |
6640 | io_put_req(req); |
6641 | io_req_complete(req, -ECANCELED); | |
1b4a51b6 | 6642 | } else |
c5eef2b9 | 6643 | io_queue_sqe(req, NULL); |
4fe2c963 JL |
6644 | } |
6645 | ||
b16fed66 PB |
6646 | /* |
6647 | * Check SQE restrictions (opcode and flags). | |
6648 | * | |
6649 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6650 | */ | |
6651 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6652 | struct io_kiocb *req, | |
6653 | unsigned int sqe_flags) | |
6654 | { | |
6655 | if (!ctx->restricted) | |
6656 | return true; | |
6657 | ||
6658 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6659 | return false; | |
6660 | ||
6661 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6662 | ctx->restrictions.sqe_flags_required) | |
6663 | return false; | |
6664 | ||
6665 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6666 | ctx->restrictions.sqe_flags_required)) | |
6667 | return false; | |
6668 | ||
6669 | return true; | |
6670 | } | |
6671 | ||
6672 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, | |
6673 | const struct io_uring_sqe *sqe) | |
6674 | { | |
6675 | struct io_submit_state *state; | |
6676 | unsigned int sqe_flags; | |
6677 | int id, ret = 0; | |
6678 | ||
6679 | req->opcode = READ_ONCE(sqe->opcode); | |
6680 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6681 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6682 | req->user_data = READ_ONCE(sqe->user_data); | |
6683 | req->async_data = NULL; | |
6684 | req->file = NULL; | |
6685 | req->ctx = ctx; | |
6686 | req->link = NULL; | |
6687 | req->fixed_rsrc_refs = NULL; | |
6688 | /* one is dropped after submission, the other at completion */ | |
6689 | refcount_set(&req->refs, 2); | |
6690 | req->task = current; | |
6691 | req->result = 0; | |
6692 | ||
6693 | /* enforce forwards compatibility on users */ | |
6694 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) | |
6695 | return -EINVAL; | |
6696 | ||
6697 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6698 | return -EINVAL; | |
6699 | ||
6700 | if (unlikely(io_sq_thread_acquire_mm_files(ctx, req))) | |
6701 | return -EFAULT; | |
6702 | ||
6703 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) | |
6704 | return -EACCES; | |
6705 | ||
6706 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6707 | !io_op_defs[req->opcode].buffer_select) | |
6708 | return -EOPNOTSUPP; | |
6709 | ||
6710 | id = READ_ONCE(sqe->personality); | |
6711 | if (id) { | |
6712 | struct io_identity *iod; | |
6713 | ||
6714 | iod = idr_find(&ctx->personality_idr, id); | |
6715 | if (unlikely(!iod)) | |
6716 | return -EINVAL; | |
6717 | refcount_inc(&iod->count); | |
6718 | ||
6719 | __io_req_init_async(req); | |
6720 | get_cred(iod->creds); | |
6721 | req->work.identity = iod; | |
6722 | req->work.flags |= IO_WQ_WORK_CREDS; | |
6723 | } | |
6724 | ||
6725 | state = &ctx->submit_state; | |
6726 | ||
6727 | /* | |
6728 | * Plug now if we have more than 1 IO left after this, and the target | |
6729 | * is potentially a read/write to block based storage. | |
6730 | */ | |
6731 | if (!state->plug_started && state->ios_left > 1 && | |
6732 | io_op_defs[req->opcode].plug) { | |
6733 | blk_start_plug(&state->plug); | |
6734 | state->plug_started = true; | |
6735 | } | |
6736 | ||
6737 | if (io_op_defs[req->opcode].needs_file) { | |
6738 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6739 | ||
6740 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6741 | if (unlikely(!req->file)) | |
6742 | ret = -EBADF; | |
6743 | } | |
6744 | ||
6745 | state->ios_left--; | |
6746 | return ret; | |
6747 | } | |
6748 | ||
863e0560 PB |
6749 | struct io_submit_link { |
6750 | struct io_kiocb *head; | |
6751 | struct io_kiocb *last; | |
6752 | }; | |
6753 | ||
1d4240cc | 6754 | static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
c5eef2b9 | 6755 | struct io_submit_link *link) |
9e645e11 | 6756 | { |
a197f664 | 6757 | struct io_ring_ctx *ctx = req->ctx; |
ef4ff581 | 6758 | int ret; |
9e645e11 | 6759 | |
9e645e11 JA |
6760 | /* |
6761 | * If we already have a head request, queue this one for async | |
6762 | * submittal once the head completes. If we don't have a head but | |
6763 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6764 | * submitted sync once the chain is complete. If none of those | |
6765 | * conditions are true (normal request), then just queue it. | |
6766 | */ | |
863e0560 PB |
6767 | if (link->head) { |
6768 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6769 | |
8cdf2193 PB |
6770 | /* |
6771 | * Taking sequential execution of a link, draining both sides | |
6772 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6773 | * requests in the link. So, it drains the head and the | |
6774 | * next after the link request. The last one is done via | |
6775 | * drain_next flag to persist the effect across calls. | |
6776 | */ | |
ef4ff581 | 6777 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6778 | head->flags |= REQ_F_IO_DRAIN; |
6779 | ctx->drain_next = 1; | |
6780 | } | |
3529d8c2 | 6781 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6782 | if (unlikely(ret)) { |
4e88d6e7 | 6783 | /* fail even hard links since we don't submit */ |
9d76377f | 6784 | head->flags |= REQ_F_FAIL_LINK; |
1d4240cc | 6785 | return ret; |
2d28390a | 6786 | } |
9d76377f | 6787 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6788 | link->last->link = req; |
863e0560 | 6789 | link->last = req; |
32fe525b PB |
6790 | |
6791 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6792 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
c5eef2b9 | 6793 | io_queue_link_head(head); |
863e0560 | 6794 | link->head = NULL; |
32fe525b | 6795 | } |
9e645e11 | 6796 | } else { |
711be031 PB |
6797 | if (unlikely(ctx->drain_next)) { |
6798 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6799 | ctx->drain_next = 0; |
711be031 | 6800 | } |
ef4ff581 | 6801 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
711be031 | 6802 | ret = io_req_defer_prep(req, sqe); |
327d6d96 | 6803 | if (unlikely(ret)) |
711be031 | 6804 | req->flags |= REQ_F_FAIL_LINK; |
863e0560 PB |
6805 | link->head = req; |
6806 | link->last = req; | |
711be031 | 6807 | } else { |
c5eef2b9 | 6808 | io_queue_sqe(req, sqe); |
711be031 | 6809 | } |
9e645e11 | 6810 | } |
2e6e1fde | 6811 | |
1d4240cc | 6812 | return 0; |
9e645e11 JA |
6813 | } |
6814 | ||
9a56a232 JA |
6815 | /* |
6816 | * Batched submission is done, ensure local IO is flushed out. | |
6817 | */ | |
ba88ff11 PB |
6818 | static void io_submit_state_end(struct io_submit_state *state, |
6819 | struct io_ring_ctx *ctx) | |
9a56a232 | 6820 | { |
6dd0be1e | 6821 | if (state->comp.nr) |
ba88ff11 | 6822 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6823 | if (state->plug_started) |
6824 | blk_finish_plug(&state->plug); | |
9f13c35b | 6825 | io_state_file_put(state); |
9a56a232 JA |
6826 | } |
6827 | ||
6828 | /* | |
6829 | * Start submission side cache. | |
6830 | */ | |
6831 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6832 | unsigned int max_ios) |
9a56a232 | 6833 | { |
27926b68 | 6834 | state->plug_started = false; |
9a56a232 JA |
6835 | state->ios_left = max_ios; |
6836 | } | |
6837 | ||
2b188cc1 JA |
6838 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6839 | { | |
75b28aff | 6840 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6841 | |
caf582c6 PB |
6842 | /* |
6843 | * Ensure any loads from the SQEs are done at this point, | |
6844 | * since once we write the new head, the application could | |
6845 | * write new data to them. | |
6846 | */ | |
6847 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6848 | } |
6849 | ||
2b188cc1 | 6850 | /* |
3529d8c2 | 6851 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6852 | * that is mapped by userspace. This means that care needs to be taken to |
6853 | * ensure that reads are stable, as we cannot rely on userspace always | |
6854 | * being a good citizen. If members of the sqe are validated and then later | |
6855 | * used, it's important that those reads are done through READ_ONCE() to | |
6856 | * prevent a re-load down the line. | |
6857 | */ | |
709b302f | 6858 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6859 | { |
75b28aff | 6860 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6861 | unsigned head; |
6862 | ||
6863 | /* | |
6864 | * The cached sq head (or cq tail) serves two purposes: | |
6865 | * | |
6866 | * 1) allows us to batch the cost of updating the user visible | |
6867 | * head updates. | |
6868 | * 2) allows the kernel side to track the head on its own, even | |
6869 | * though the application is the one updating it. | |
6870 | */ | |
4fccfcbb | 6871 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6872 | if (likely(head < ctx->sq_entries)) |
6873 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6874 | |
6875 | /* drop invalid entries */ | |
498ccd9e | 6876 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6877 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6878 | return NULL; |
6879 | } | |
6880 | ||
0f212204 | 6881 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6882 | { |
863e0560 | 6883 | struct io_submit_link link; |
46c4e16a | 6884 | int submitted = 0; |
6c271ce2 | 6885 | |
c4a2ed72 | 6886 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6887 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6888 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6889 | return -EBUSY; |
6890 | } | |
6c271ce2 | 6891 | |
ee7d46d9 PB |
6892 | /* make sure SQ entry isn't read before tail */ |
6893 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6894 | |
2b85edfc PB |
6895 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6896 | return -EAGAIN; | |
6c271ce2 | 6897 | |
d8a6df10 | 6898 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6899 | refcount_add(nr, ¤t->usage); |
6c271ce2 | 6900 | |
ba88ff11 | 6901 | io_submit_state_start(&ctx->submit_state, nr); |
863e0560 | 6902 | link.head = NULL; |
b14cca0c | 6903 | |
46c4e16a | 6904 | while (submitted < nr) { |
3529d8c2 | 6905 | const struct io_uring_sqe *sqe; |
196be95c | 6906 | struct io_kiocb *req; |
1cb1edb2 | 6907 | int err; |
fb5ccc98 | 6908 | |
258b29a9 | 6909 | req = io_alloc_req(ctx); |
196be95c PB |
6910 | if (unlikely(!req)) { |
6911 | if (!submitted) | |
6912 | submitted = -EAGAIN; | |
fb5ccc98 | 6913 | break; |
196be95c | 6914 | } |
4fccfcbb PB |
6915 | sqe = io_get_sqe(ctx); |
6916 | if (unlikely(!sqe)) { | |
6917 | kmem_cache_free(req_cachep, req); | |
6918 | break; | |
6919 | } | |
d3656344 JA |
6920 | /* will complete beyond this point, count as submitted */ |
6921 | submitted++; | |
6922 | ||
258b29a9 | 6923 | err = io_init_req(ctx, req, sqe); |
ef4ff581 | 6924 | if (unlikely(err)) { |
1cb1edb2 | 6925 | fail_req: |
e1e16097 JA |
6926 | io_put_req(req); |
6927 | io_req_complete(req, err); | |
196be95c PB |
6928 | break; |
6929 | } | |
fb5ccc98 | 6930 | |
354420f7 | 6931 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
2d7e9358 | 6932 | true, ctx->flags & IORING_SETUP_SQPOLL); |
c5eef2b9 | 6933 | err = io_submit_sqe(req, sqe, &link); |
1d4240cc PB |
6934 | if (err) |
6935 | goto fail_req; | |
6c271ce2 JA |
6936 | } |
6937 | ||
9466f437 PB |
6938 | if (unlikely(submitted != nr)) { |
6939 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6940 | struct io_uring_task *tctx = current->io_uring; |
6941 | int unused = nr - ref_used; | |
9466f437 | 6942 | |
d8a6df10 JA |
6943 | percpu_ref_put_many(&ctx->refs, unused); |
6944 | percpu_counter_sub(&tctx->inflight, unused); | |
6945 | put_task_struct_many(current, unused); | |
9466f437 | 6946 | } |
863e0560 | 6947 | if (link.head) |
c5eef2b9 | 6948 | io_queue_link_head(link.head); |
ba88ff11 | 6949 | io_submit_state_end(&ctx->submit_state, ctx); |
6c271ce2 | 6950 | |
ae9428ca PB |
6951 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6952 | io_commit_sqring(ctx); | |
6953 | ||
6c271ce2 JA |
6954 | return submitted; |
6955 | } | |
6956 | ||
23b3628e XW |
6957 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6958 | { | |
6959 | /* Tell userspace we may need a wakeup call */ | |
6960 | spin_lock_irq(&ctx->completion_lock); | |
6961 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6962 | spin_unlock_irq(&ctx->completion_lock); | |
6963 | } | |
6964 | ||
6965 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6966 | { | |
6967 | spin_lock_irq(&ctx->completion_lock); | |
6968 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6969 | spin_unlock_irq(&ctx->completion_lock); | |
6970 | } | |
6971 | ||
08369246 | 6972 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6973 | { |
c8d1ba58 | 6974 | unsigned int to_submit; |
bdcd3eab | 6975 | int ret = 0; |
6c271ce2 | 6976 | |
c8d1ba58 | 6977 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6978 | /* if we're handling multiple rings, cap submit size for fairness */ |
6979 | if (cap_entries && to_submit > 8) | |
6980 | to_submit = 8; | |
6981 | ||
906a3c6f | 6982 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6983 | unsigned nr_events = 0; |
a4c0b3de | 6984 | |
c8d1ba58 | 6985 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6986 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6987 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6988 | |
d9d05217 PB |
6989 | if (to_submit && !ctx->sqo_dead && |
6990 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 6991 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6992 | mutex_unlock(&ctx->uring_lock); |
6993 | } | |
6c271ce2 | 6994 | |
90554200 JA |
6995 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6996 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6997 | |
08369246 XW |
6998 | return ret; |
6999 | } | |
6c271ce2 | 7000 | |
08369246 XW |
7001 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7002 | { | |
7003 | struct io_ring_ctx *ctx; | |
7004 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7005 | |
08369246 XW |
7006 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7007 | if (sq_thread_idle < ctx->sq_thread_idle) | |
7008 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 7009 | } |
c1edbf5f | 7010 | |
08369246 | 7011 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7012 | } |
6c271ce2 | 7013 | |
69fb2131 JA |
7014 | static void io_sqd_init_new(struct io_sq_data *sqd) |
7015 | { | |
7016 | struct io_ring_ctx *ctx; | |
7017 | ||
7018 | while (!list_empty(&sqd->ctx_new_list)) { | |
7019 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
7020 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
7021 | complete(&ctx->sq_thread_comp); | |
7022 | } | |
08369246 XW |
7023 | |
7024 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
7025 | } |
7026 | ||
c8d1ba58 JA |
7027 | static int io_sq_thread(void *data) |
7028 | { | |
91d8f519 | 7029 | struct cgroup_subsys_state *cur_css = NULL; |
28cea78a JA |
7030 | struct files_struct *old_files = current->files; |
7031 | struct nsproxy *old_nsproxy = current->nsproxy; | |
69fb2131 JA |
7032 | const struct cred *old_cred = NULL; |
7033 | struct io_sq_data *sqd = data; | |
7034 | struct io_ring_ctx *ctx; | |
a0d9205f | 7035 | unsigned long timeout = 0; |
08369246 | 7036 | DEFINE_WAIT(wait); |
6c271ce2 | 7037 | |
28cea78a JA |
7038 | task_lock(current); |
7039 | current->files = NULL; | |
7040 | current->nsproxy = NULL; | |
7041 | task_unlock(current); | |
6c271ce2 | 7042 | |
69fb2131 | 7043 | while (!kthread_should_stop()) { |
08369246 XW |
7044 | int ret; |
7045 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
7046 | |
7047 | /* | |
69fb2131 JA |
7048 | * Any changes to the sqd lists are synchronized through the |
7049 | * kthread parking. This synchronizes the thread vs users, | |
7050 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 7051 | */ |
65b2b213 | 7052 | if (kthread_should_park()) { |
69fb2131 | 7053 | kthread_parkme(); |
65b2b213 XW |
7054 | /* |
7055 | * When sq thread is unparked, in case the previous park operation | |
7056 | * comes from io_put_sq_data(), which means that sq thread is going | |
7057 | * to be stopped, so here needs to have a check. | |
7058 | */ | |
7059 | if (kthread_should_stop()) | |
7060 | break; | |
7061 | } | |
7143b5ac | 7062 | |
08369246 | 7063 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 7064 | io_sqd_init_new(sqd); |
08369246 XW |
7065 | timeout = jiffies + sqd->sq_thread_idle; |
7066 | } | |
6c271ce2 | 7067 | |
08369246 | 7068 | sqt_spin = false; |
e95eee2d | 7069 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 JA |
7070 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7071 | if (current->cred != ctx->creds) { | |
7072 | if (old_cred) | |
7073 | revert_creds(old_cred); | |
7074 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 7075 | } |
91d8f519 | 7076 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
7077 | #ifdef CONFIG_AUDIT |
7078 | current->loginuid = ctx->loginuid; | |
7079 | current->sessionid = ctx->sessionid; | |
7080 | #endif | |
bdcd3eab | 7081 | |
08369246 XW |
7082 | ret = __io_sq_thread(ctx, cap_entries); |
7083 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
7084 | sqt_spin = true; | |
6c271ce2 | 7085 | |
28cea78a | 7086 | io_sq_thread_drop_mm_files(); |
69fb2131 | 7087 | } |
6c271ce2 | 7088 | |
08369246 | 7089 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7090 | io_run_task_work(); |
d434ab6d | 7091 | io_sq_thread_drop_mm_files(); |
c8d1ba58 | 7092 | cond_resched(); |
08369246 XW |
7093 | if (sqt_spin) |
7094 | timeout = jiffies + sqd->sq_thread_idle; | |
7095 | continue; | |
7096 | } | |
7097 | ||
08369246 XW |
7098 | needs_sched = true; |
7099 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
7100 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
7101 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
7102 | !list_empty_careful(&ctx->iopoll_list)) { | |
7103 | needs_sched = false; | |
7104 | break; | |
7105 | } | |
7106 | if (io_sqring_entries(ctx)) { | |
7107 | needs_sched = false; | |
7108 | break; | |
7109 | } | |
7110 | } | |
7111 | ||
8b28fdf2 | 7112 | if (needs_sched && !kthread_should_park()) { |
69fb2131 JA |
7113 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7114 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 7115 | |
69fb2131 | 7116 | schedule(); |
69fb2131 JA |
7117 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7118 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7119 | } |
08369246 XW |
7120 | |
7121 | finish_wait(&sqd->wait, &wait); | |
7122 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
7123 | } |
7124 | ||
4c6e277c | 7125 | io_run_task_work(); |
d434ab6d | 7126 | io_sq_thread_drop_mm_files(); |
b41e9852 | 7127 | |
91d8f519 DZ |
7128 | if (cur_css) |
7129 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
7130 | if (old_cred) |
7131 | revert_creds(old_cred); | |
06058632 | 7132 | |
28cea78a JA |
7133 | task_lock(current); |
7134 | current->files = old_files; | |
7135 | current->nsproxy = old_nsproxy; | |
7136 | task_unlock(current); | |
7137 | ||
2bbcd6d3 | 7138 | kthread_parkme(); |
06058632 | 7139 | |
6c271ce2 JA |
7140 | return 0; |
7141 | } | |
7142 | ||
bda52162 JA |
7143 | struct io_wait_queue { |
7144 | struct wait_queue_entry wq; | |
7145 | struct io_ring_ctx *ctx; | |
7146 | unsigned to_wait; | |
7147 | unsigned nr_timeouts; | |
7148 | }; | |
7149 | ||
6c503150 | 7150 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7151 | { |
7152 | struct io_ring_ctx *ctx = iowq->ctx; | |
7153 | ||
7154 | /* | |
d195a66e | 7155 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7156 | * started waiting. For timeouts, we always want to return to userspace, |
7157 | * regardless of event count. | |
7158 | */ | |
6c503150 | 7159 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
7160 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
7161 | } | |
7162 | ||
7163 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7164 | int wake_flags, void *key) | |
7165 | { | |
7166 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7167 | wq); | |
7168 | ||
6c503150 PB |
7169 | /* |
7170 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7171 | * the task, and the next invocation will do it. | |
7172 | */ | |
7173 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
7174 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
7175 | return -1; | |
bda52162 JA |
7176 | } |
7177 | ||
af9c1a44 JA |
7178 | static int io_run_task_work_sig(void) |
7179 | { | |
7180 | if (io_run_task_work()) | |
7181 | return 1; | |
7182 | if (!signal_pending(current)) | |
7183 | return 0; | |
792ee0f6 JA |
7184 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
7185 | return -ERESTARTSYS; | |
af9c1a44 JA |
7186 | return -EINTR; |
7187 | } | |
7188 | ||
eeb60b9a PB |
7189 | /* when returns >0, the caller should retry */ |
7190 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7191 | struct io_wait_queue *iowq, | |
7192 | signed long *timeout) | |
7193 | { | |
7194 | int ret; | |
7195 | ||
7196 | /* make sure we run task_work before checking for signals */ | |
7197 | ret = io_run_task_work_sig(); | |
7198 | if (ret || io_should_wake(iowq)) | |
7199 | return ret; | |
7200 | /* let the caller flush overflows, retry */ | |
7201 | if (test_bit(0, &ctx->cq_check_overflow)) | |
7202 | return 1; | |
7203 | ||
7204 | *timeout = schedule_timeout(*timeout); | |
7205 | return !*timeout ? -ETIME : 1; | |
7206 | } | |
7207 | ||
2b188cc1 JA |
7208 | /* |
7209 | * Wait until events become available, if we don't already have some. The | |
7210 | * application must reap them itself, as they reside on the shared cq ring. | |
7211 | */ | |
7212 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7213 | const sigset_t __user *sig, size_t sigsz, |
7214 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7215 | { |
bda52162 JA |
7216 | struct io_wait_queue iowq = { |
7217 | .wq = { | |
7218 | .private = current, | |
7219 | .func = io_wake_function, | |
7220 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7221 | }, | |
7222 | .ctx = ctx, | |
7223 | .to_wait = min_events, | |
7224 | }; | |
75b28aff | 7225 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7226 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7227 | int ret; | |
2b188cc1 | 7228 | |
b41e9852 | 7229 | do { |
6c503150 PB |
7230 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
7231 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 7232 | return 0; |
4c6e277c | 7233 | if (!io_run_task_work()) |
b41e9852 | 7234 | break; |
b41e9852 | 7235 | } while (1); |
2b188cc1 JA |
7236 | |
7237 | if (sig) { | |
9e75ad5d AB |
7238 | #ifdef CONFIG_COMPAT |
7239 | if (in_compat_syscall()) | |
7240 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7241 | sigsz); |
9e75ad5d AB |
7242 | else |
7243 | #endif | |
b772434b | 7244 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7245 | |
2b188cc1 JA |
7246 | if (ret) |
7247 | return ret; | |
7248 | } | |
7249 | ||
c73ebb68 | 7250 | if (uts) { |
c1d5a224 PB |
7251 | struct timespec64 ts; |
7252 | ||
c73ebb68 HX |
7253 | if (get_timespec64(&ts, uts)) |
7254 | return -EFAULT; | |
7255 | timeout = timespec64_to_jiffies(&ts); | |
7256 | } | |
7257 | ||
bda52162 | 7258 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7259 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7260 | do { |
6c503150 | 7261 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
7262 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7263 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7264 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7265 | finish_wait(&ctx->wait, &iowq.wq); | |
7266 | } while (ret > 0); | |
bda52162 | 7267 | |
b7db41c9 | 7268 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7269 | |
75b28aff | 7270 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7271 | } |
7272 | ||
6b06314c JA |
7273 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7274 | { | |
7275 | #if defined(CONFIG_UNIX) | |
7276 | if (ctx->ring_sock) { | |
7277 | struct sock *sock = ctx->ring_sock->sk; | |
7278 | struct sk_buff *skb; | |
7279 | ||
7280 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7281 | kfree_skb(skb); | |
7282 | } | |
7283 | #else | |
7284 | int i; | |
7285 | ||
65e19f54 JA |
7286 | for (i = 0; i < ctx->nr_user_files; i++) { |
7287 | struct file *file; | |
7288 | ||
7289 | file = io_file_from_index(ctx, i); | |
7290 | if (file) | |
7291 | fput(file); | |
7292 | } | |
6b06314c JA |
7293 | #endif |
7294 | } | |
7295 | ||
00835dce | 7296 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7297 | { |
269bbe5f | 7298 | struct fixed_rsrc_data *data; |
05f3fb3c | 7299 | |
269bbe5f | 7300 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7301 | complete(&data->done); |
7302 | } | |
7303 | ||
2a63b2d9 BM |
7304 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
7305 | { | |
7306 | spin_lock_bh(&ctx->rsrc_ref_lock); | |
7307 | } | |
7308 | ||
7309 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) | |
7310 | { | |
7311 | spin_unlock_bh(&ctx->rsrc_ref_lock); | |
7312 | } | |
7313 | ||
d67d2263 BM |
7314 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7315 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7316 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7317 | { |
2a63b2d9 | 7318 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7319 | rsrc_data->node = ref_node; |
d67d2263 | 7320 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7321 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7322 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7323 | } |
7324 | ||
d7954b2b BM |
7325 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, |
7326 | struct io_ring_ctx *ctx, | |
7327 | struct fixed_rsrc_ref_node *backup_node) | |
6b06314c | 7328 | { |
d7954b2b | 7329 | struct fixed_rsrc_ref_node *ref_node; |
1ffc5422 | 7330 | int ret; |
65e19f54 | 7331 | |
2a63b2d9 | 7332 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7333 | ref_node = data->node; |
2a63b2d9 | 7334 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7335 | if (ref_node) |
7336 | percpu_ref_kill(&ref_node->refs); | |
7337 | ||
7338 | percpu_ref_kill(&data->refs); | |
7339 | ||
7340 | /* wait for all refs nodes to complete */ | |
269bbe5f | 7341 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7342 | do { |
7343 | ret = wait_for_completion_interruptible(&data->done); | |
7344 | if (!ret) | |
7345 | break; | |
7346 | ret = io_run_task_work_sig(); | |
7347 | if (ret < 0) { | |
7348 | percpu_ref_resurrect(&data->refs); | |
7349 | reinit_completion(&data->done); | |
d67d2263 | 7350 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
1ffc5422 PB |
7351 | return ret; |
7352 | } | |
7353 | } while (1); | |
05f3fb3c | 7354 | |
d7954b2b BM |
7355 | destroy_fixed_rsrc_ref_node(backup_node); |
7356 | return 0; | |
7357 | } | |
7358 | ||
1ad555c6 BM |
7359 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7360 | { | |
7361 | struct fixed_rsrc_data *data; | |
7362 | ||
7363 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7364 | if (!data) | |
7365 | return NULL; | |
7366 | ||
00835dce | 7367 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7368 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7369 | kfree(data); | |
7370 | return NULL; | |
7371 | } | |
7372 | data->ctx = ctx; | |
7373 | init_completion(&data->done); | |
7374 | return data; | |
7375 | } | |
7376 | ||
7377 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7378 | { | |
7379 | percpu_ref_exit(&data->refs); | |
7380 | kfree(data->table); | |
7381 | kfree(data); | |
7382 | } | |
7383 | ||
d7954b2b BM |
7384 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7385 | { | |
7386 | struct fixed_rsrc_data *data = ctx->file_data; | |
7387 | struct fixed_rsrc_ref_node *backup_node; | |
7388 | unsigned nr_tables, i; | |
7389 | int ret; | |
7390 | ||
7391 | if (!data) | |
7392 | return -ENXIO; | |
7393 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7394 | if (!backup_node) | |
7395 | return -ENOMEM; | |
7396 | init_fixed_file_ref_node(ctx, backup_node); | |
7397 | ||
7398 | ret = io_rsrc_ref_quiesce(data, ctx, backup_node); | |
7399 | if (ret) | |
7400 | return ret; | |
7401 | ||
6b06314c | 7402 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7403 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7404 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7405 | kfree(data->table[i].files); |
1ad555c6 | 7406 | free_fixed_rsrc_data(data); |
05f3fb3c | 7407 | ctx->file_data = NULL; |
6b06314c JA |
7408 | ctx->nr_user_files = 0; |
7409 | return 0; | |
7410 | } | |
7411 | ||
534ca6d6 | 7412 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7413 | { |
534ca6d6 | 7414 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7415 | /* |
7416 | * The park is a bit of a work-around, without it we get | |
7417 | * warning spews on shutdown with SQPOLL set and affinity | |
7418 | * set to a single CPU. | |
7419 | */ | |
534ca6d6 JA |
7420 | if (sqd->thread) { |
7421 | kthread_park(sqd->thread); | |
7422 | kthread_stop(sqd->thread); | |
7423 | } | |
7424 | ||
7425 | kfree(sqd); | |
7426 | } | |
7427 | } | |
7428 | ||
aa06165d JA |
7429 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7430 | { | |
7431 | struct io_ring_ctx *ctx_attach; | |
7432 | struct io_sq_data *sqd; | |
7433 | struct fd f; | |
7434 | ||
7435 | f = fdget(p->wq_fd); | |
7436 | if (!f.file) | |
7437 | return ERR_PTR(-ENXIO); | |
7438 | if (f.file->f_op != &io_uring_fops) { | |
7439 | fdput(f); | |
7440 | return ERR_PTR(-EINVAL); | |
7441 | } | |
7442 | ||
7443 | ctx_attach = f.file->private_data; | |
7444 | sqd = ctx_attach->sq_data; | |
7445 | if (!sqd) { | |
7446 | fdput(f); | |
7447 | return ERR_PTR(-EINVAL); | |
7448 | } | |
7449 | ||
7450 | refcount_inc(&sqd->refs); | |
7451 | fdput(f); | |
7452 | return sqd; | |
7453 | } | |
7454 | ||
534ca6d6 JA |
7455 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7456 | { | |
7457 | struct io_sq_data *sqd; | |
7458 | ||
aa06165d JA |
7459 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7460 | return io_attach_sq_data(p); | |
7461 | ||
534ca6d6 JA |
7462 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7463 | if (!sqd) | |
7464 | return ERR_PTR(-ENOMEM); | |
7465 | ||
7466 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7467 | INIT_LIST_HEAD(&sqd->ctx_list); |
7468 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7469 | mutex_init(&sqd->ctx_lock); | |
7470 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7471 | init_waitqueue_head(&sqd->wait); |
7472 | return sqd; | |
7473 | } | |
7474 | ||
69fb2131 JA |
7475 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7476 | __releases(&sqd->lock) | |
7477 | { | |
7478 | if (!sqd->thread) | |
7479 | return; | |
7480 | kthread_unpark(sqd->thread); | |
7481 | mutex_unlock(&sqd->lock); | |
7482 | } | |
7483 | ||
7484 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7485 | __acquires(&sqd->lock) | |
7486 | { | |
7487 | if (!sqd->thread) | |
7488 | return; | |
7489 | mutex_lock(&sqd->lock); | |
7490 | kthread_park(sqd->thread); | |
7491 | } | |
7492 | ||
534ca6d6 JA |
7493 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7494 | { | |
7495 | struct io_sq_data *sqd = ctx->sq_data; | |
7496 | ||
7497 | if (sqd) { | |
7498 | if (sqd->thread) { | |
7499 | /* | |
7500 | * We may arrive here from the error branch in | |
7501 | * io_sq_offload_create() where the kthread is created | |
7502 | * without being waked up, thus wake it up now to make | |
7503 | * sure the wait will complete. | |
7504 | */ | |
7505 | wake_up_process(sqd->thread); | |
7506 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7507 | |
7508 | io_sq_thread_park(sqd); | |
7509 | } | |
7510 | ||
7511 | mutex_lock(&sqd->ctx_lock); | |
7512 | list_del(&ctx->sqd_list); | |
08369246 | 7513 | io_sqd_update_thread_idle(sqd); |
69fb2131 JA |
7514 | mutex_unlock(&sqd->ctx_lock); |
7515 | ||
08369246 | 7516 | if (sqd->thread) |
69fb2131 | 7517 | io_sq_thread_unpark(sqd); |
534ca6d6 JA |
7518 | |
7519 | io_put_sq_data(sqd); | |
7520 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7521 | } |
7522 | } | |
7523 | ||
6b06314c JA |
7524 | static void io_finish_async(struct io_ring_ctx *ctx) |
7525 | { | |
6c271ce2 JA |
7526 | io_sq_thread_stop(ctx); |
7527 | ||
561fb04a JA |
7528 | if (ctx->io_wq) { |
7529 | io_wq_destroy(ctx->io_wq); | |
7530 | ctx->io_wq = NULL; | |
6b06314c JA |
7531 | } |
7532 | } | |
7533 | ||
7534 | #if defined(CONFIG_UNIX) | |
6b06314c JA |
7535 | /* |
7536 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7537 | * the io_uring can be safely unregistered on process exit, even if we have | |
7538 | * loops in the file referencing. | |
7539 | */ | |
7540 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7541 | { | |
7542 | struct sock *sk = ctx->ring_sock->sk; | |
7543 | struct scm_fp_list *fpl; | |
7544 | struct sk_buff *skb; | |
08a45173 | 7545 | int i, nr_files; |
6b06314c | 7546 | |
6b06314c JA |
7547 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7548 | if (!fpl) | |
7549 | return -ENOMEM; | |
7550 | ||
7551 | skb = alloc_skb(0, GFP_KERNEL); | |
7552 | if (!skb) { | |
7553 | kfree(fpl); | |
7554 | return -ENOMEM; | |
7555 | } | |
7556 | ||
7557 | skb->sk = sk; | |
6b06314c | 7558 | |
08a45173 | 7559 | nr_files = 0; |
6b06314c JA |
7560 | fpl->user = get_uid(ctx->user); |
7561 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7562 | struct file *file = io_file_from_index(ctx, i + offset); |
7563 | ||
7564 | if (!file) | |
08a45173 | 7565 | continue; |
65e19f54 | 7566 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7567 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7568 | nr_files++; | |
6b06314c JA |
7569 | } |
7570 | ||
08a45173 JA |
7571 | if (nr_files) { |
7572 | fpl->max = SCM_MAX_FD; | |
7573 | fpl->count = nr_files; | |
7574 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7575 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7576 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7577 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7578 | |
08a45173 JA |
7579 | for (i = 0; i < nr_files; i++) |
7580 | fput(fpl->fp[i]); | |
7581 | } else { | |
7582 | kfree_skb(skb); | |
7583 | kfree(fpl); | |
7584 | } | |
6b06314c JA |
7585 | |
7586 | return 0; | |
7587 | } | |
7588 | ||
7589 | /* | |
7590 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7591 | * causes regular reference counting to break down. We rely on the UNIX | |
7592 | * garbage collection to take care of this problem for us. | |
7593 | */ | |
7594 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7595 | { | |
7596 | unsigned left, total; | |
7597 | int ret = 0; | |
7598 | ||
7599 | total = 0; | |
7600 | left = ctx->nr_user_files; | |
7601 | while (left) { | |
7602 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7603 | |
7604 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7605 | if (ret) | |
7606 | break; | |
7607 | left -= this_files; | |
7608 | total += this_files; | |
7609 | } | |
7610 | ||
7611 | if (!ret) | |
7612 | return 0; | |
7613 | ||
7614 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7615 | struct file *file = io_file_from_index(ctx, total); |
7616 | ||
7617 | if (file) | |
7618 | fput(file); | |
6b06314c JA |
7619 | total++; |
7620 | } | |
7621 | ||
7622 | return ret; | |
7623 | } | |
7624 | #else | |
7625 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7626 | { | |
7627 | return 0; | |
7628 | } | |
7629 | #endif | |
7630 | ||
269bbe5f | 7631 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7632 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7633 | { |
7634 | int i; | |
7635 | ||
7636 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7637 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7638 | unsigned this_files; |
7639 | ||
7640 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7641 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7642 | GFP_KERNEL); | |
7643 | if (!table->files) | |
7644 | break; | |
7645 | nr_files -= this_files; | |
7646 | } | |
7647 | ||
7648 | if (i == nr_tables) | |
7649 | return 0; | |
7650 | ||
7651 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7652 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7653 | kfree(table->files); |
7654 | } | |
7655 | return 1; | |
7656 | } | |
7657 | ||
50238531 | 7658 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7659 | { |
50238531 | 7660 | struct file *file = prsrc->file; |
05f3fb3c JA |
7661 | #if defined(CONFIG_UNIX) |
7662 | struct sock *sock = ctx->ring_sock->sk; | |
7663 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7664 | struct sk_buff *skb; | |
7665 | int i; | |
7666 | ||
7667 | __skb_queue_head_init(&list); | |
7668 | ||
7669 | /* | |
7670 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7671 | * remove this entry and rearrange the file array. | |
7672 | */ | |
7673 | skb = skb_dequeue(head); | |
7674 | while (skb) { | |
7675 | struct scm_fp_list *fp; | |
7676 | ||
7677 | fp = UNIXCB(skb).fp; | |
7678 | for (i = 0; i < fp->count; i++) { | |
7679 | int left; | |
7680 | ||
7681 | if (fp->fp[i] != file) | |
7682 | continue; | |
7683 | ||
7684 | unix_notinflight(fp->user, fp->fp[i]); | |
7685 | left = fp->count - 1 - i; | |
7686 | if (left) { | |
7687 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7688 | left * sizeof(struct file *)); | |
7689 | } | |
7690 | fp->count--; | |
7691 | if (!fp->count) { | |
7692 | kfree_skb(skb); | |
7693 | skb = NULL; | |
7694 | } else { | |
7695 | __skb_queue_tail(&list, skb); | |
7696 | } | |
7697 | fput(file); | |
7698 | file = NULL; | |
7699 | break; | |
7700 | } | |
7701 | ||
7702 | if (!file) | |
7703 | break; | |
7704 | ||
7705 | __skb_queue_tail(&list, skb); | |
7706 | ||
7707 | skb = skb_dequeue(head); | |
7708 | } | |
7709 | ||
7710 | if (skb_peek(&list)) { | |
7711 | spin_lock_irq(&head->lock); | |
7712 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7713 | __skb_queue_tail(head, skb); | |
7714 | spin_unlock_irq(&head->lock); | |
7715 | } | |
7716 | #else | |
7717 | fput(file); | |
7718 | #endif | |
7719 | } | |
7720 | ||
269bbe5f | 7721 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7722 | { |
269bbe5f BM |
7723 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7724 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7725 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7726 | |
269bbe5f BM |
7727 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7728 | list_del(&prsrc->list); | |
50238531 | 7729 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7730 | kfree(prsrc); |
65e19f54 | 7731 | } |
05589553 | 7732 | |
05589553 XW |
7733 | percpu_ref_exit(&ref_node->refs); |
7734 | kfree(ref_node); | |
269bbe5f | 7735 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7736 | } |
65e19f54 | 7737 | |
269bbe5f | 7738 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7739 | { |
7740 | struct io_ring_ctx *ctx; | |
7741 | struct llist_node *node; | |
7742 | ||
269bbe5f BM |
7743 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7744 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7745 | |
7746 | while (node) { | |
269bbe5f | 7747 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7748 | struct llist_node *next = node->next; |
7749 | ||
269bbe5f BM |
7750 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7751 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7752 | node = next; |
7753 | } | |
7754 | } | |
7755 | ||
ea64ec02 PB |
7756 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7757 | unsigned i) | |
7758 | { | |
7759 | struct fixed_rsrc_table *table; | |
7760 | ||
7761 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7762 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7763 | } | |
7764 | ||
00835dce | 7765 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7766 | { |
269bbe5f BM |
7767 | struct fixed_rsrc_ref_node *ref_node; |
7768 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7769 | struct io_ring_ctx *ctx; |
e297822b | 7770 | bool first_add = false; |
4a38aed2 | 7771 | int delay = HZ; |
65e19f54 | 7772 | |
269bbe5f BM |
7773 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7774 | data = ref_node->rsrc_data; | |
e297822b PB |
7775 | ctx = data->ctx; |
7776 | ||
2a63b2d9 | 7777 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7778 | ref_node->done = true; |
7779 | ||
d67d2263 BM |
7780 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7781 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7782 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7783 | /* recycle ref nodes in order */ |
7784 | if (!ref_node->done) | |
7785 | break; | |
7786 | list_del(&ref_node->node); | |
269bbe5f | 7787 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7788 | } |
2a63b2d9 | 7789 | io_rsrc_ref_unlock(ctx); |
05589553 | 7790 | |
e297822b | 7791 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7792 | delay = 0; |
05589553 | 7793 | |
4a38aed2 | 7794 | if (!delay) |
269bbe5f | 7795 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7796 | else if (first_add) |
269bbe5f | 7797 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7798 | } |
65e19f54 | 7799 | |
6802535d | 7800 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7801 | struct io_ring_ctx *ctx) |
05f3fb3c | 7802 | { |
269bbe5f | 7803 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7804 | |
05589553 XW |
7805 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7806 | if (!ref_node) | |
3e2224c5 | 7807 | return NULL; |
05f3fb3c | 7808 | |
00835dce | 7809 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7810 | 0, GFP_KERNEL)) { |
7811 | kfree(ref_node); | |
3e2224c5 | 7812 | return NULL; |
05589553 XW |
7813 | } |
7814 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7815 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
6802535d BM |
7816 | ref_node->done = false; |
7817 | return ref_node; | |
7818 | } | |
7819 | ||
bc9744cd PB |
7820 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7821 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7822 | { |
269bbe5f | 7823 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7824 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7825 | } |
7826 | ||
269bbe5f | 7827 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7828 | { |
7829 | percpu_ref_exit(&ref_node->refs); | |
7830 | kfree(ref_node); | |
65e19f54 JA |
7831 | } |
7832 | ||
ea64ec02 | 7833 | |
6b06314c JA |
7834 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7835 | unsigned nr_args) | |
7836 | { | |
7837 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7838 | unsigned nr_tables, i; |
05f3fb3c | 7839 | struct file *file; |
600cf3f8 | 7840 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7841 | struct fixed_rsrc_ref_node *ref_node; |
7842 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7843 | |
05f3fb3c | 7844 | if (ctx->file_data) |
6b06314c JA |
7845 | return -EBUSY; |
7846 | if (!nr_args) | |
7847 | return -EINVAL; | |
7848 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7849 | return -EMFILE; | |
7850 | ||
1ad555c6 | 7851 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7852 | if (!file_data) |
05f3fb3c | 7853 | return -ENOMEM; |
13770a71 | 7854 | ctx->file_data = file_data; |
05f3fb3c | 7855 | |
65e19f54 | 7856 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7857 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7858 | GFP_KERNEL); |
600cf3f8 PB |
7859 | if (!file_data->table) |
7860 | goto out_free; | |
05f3fb3c | 7861 | |
600cf3f8 | 7862 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7863 | goto out_free; |
65e19f54 | 7864 | |
08a45173 | 7865 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7866 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7867 | ret = -EFAULT; | |
7868 | goto out_fput; | |
7869 | } | |
08a45173 | 7870 | /* allow sparse sets */ |
600cf3f8 | 7871 | if (fd == -1) |
08a45173 | 7872 | continue; |
6b06314c | 7873 | |
05f3fb3c | 7874 | file = fget(fd); |
6b06314c | 7875 | ret = -EBADF; |
05f3fb3c | 7876 | if (!file) |
600cf3f8 | 7877 | goto out_fput; |
05f3fb3c | 7878 | |
6b06314c JA |
7879 | /* |
7880 | * Don't allow io_uring instances to be registered. If UNIX | |
7881 | * isn't enabled, then this causes a reference cycle and this | |
7882 | * instance can never get freed. If UNIX is enabled we'll | |
7883 | * handle it just fine, but there's still no point in allowing | |
7884 | * a ring fd as it doesn't support regular read/write anyway. | |
7885 | */ | |
05f3fb3c JA |
7886 | if (file->f_op == &io_uring_fops) { |
7887 | fput(file); | |
600cf3f8 | 7888 | goto out_fput; |
6b06314c | 7889 | } |
ea64ec02 | 7890 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7891 | } |
7892 | ||
6b06314c | 7893 | ret = io_sqe_files_scm(ctx); |
05589553 | 7894 | if (ret) { |
6b06314c | 7895 | io_sqe_files_unregister(ctx); |
05589553 XW |
7896 | return ret; |
7897 | } | |
6b06314c | 7898 | |
bc9744cd | 7899 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7900 | if (!ref_node) { |
05589553 | 7901 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7902 | return -ENOMEM; |
05589553 | 7903 | } |
bc9744cd | 7904 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7905 | |
d67d2263 | 7906 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7907 | return ret; |
600cf3f8 PB |
7908 | out_fput: |
7909 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7910 | file = io_file_from_index(ctx, i); | |
7911 | if (file) | |
7912 | fput(file); | |
7913 | } | |
7914 | for (i = 0; i < nr_tables; i++) | |
7915 | kfree(file_data->table[i].files); | |
7916 | ctx->nr_user_files = 0; | |
600cf3f8 | 7917 | out_free: |
1ad555c6 | 7918 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7919 | ctx->file_data = NULL; |
6b06314c JA |
7920 | return ret; |
7921 | } | |
7922 | ||
c3a31e60 JA |
7923 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7924 | int index) | |
7925 | { | |
7926 | #if defined(CONFIG_UNIX) | |
7927 | struct sock *sock = ctx->ring_sock->sk; | |
7928 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7929 | struct sk_buff *skb; | |
7930 | ||
7931 | /* | |
7932 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7933 | * file set. If there's no room, fall back to allocating a new skb | |
7934 | * and filling it in. | |
7935 | */ | |
7936 | spin_lock_irq(&head->lock); | |
7937 | skb = skb_peek(head); | |
7938 | if (skb) { | |
7939 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7940 | ||
7941 | if (fpl->count < SCM_MAX_FD) { | |
7942 | __skb_unlink(skb, head); | |
7943 | spin_unlock_irq(&head->lock); | |
7944 | fpl->fp[fpl->count] = get_file(file); | |
7945 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7946 | fpl->count++; | |
7947 | spin_lock_irq(&head->lock); | |
7948 | __skb_queue_head(head, skb); | |
7949 | } else { | |
7950 | skb = NULL; | |
7951 | } | |
7952 | } | |
7953 | spin_unlock_irq(&head->lock); | |
7954 | ||
7955 | if (skb) { | |
7956 | fput(file); | |
7957 | return 0; | |
7958 | } | |
7959 | ||
7960 | return __io_sqe_files_scm(ctx, 1, index); | |
7961 | #else | |
7962 | return 0; | |
7963 | #endif | |
7964 | } | |
7965 | ||
50238531 | 7966 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7967 | { |
269bbe5f BM |
7968 | struct io_rsrc_put *prsrc; |
7969 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7970 | |
269bbe5f BM |
7971 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7972 | if (!prsrc) | |
a5318d3c | 7973 | return -ENOMEM; |
05f3fb3c | 7974 | |
50238531 | 7975 | prsrc->rsrc = rsrc; |
269bbe5f | 7976 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7977 | |
a5318d3c | 7978 | return 0; |
05f3fb3c JA |
7979 | } |
7980 | ||
269bbe5f BM |
7981 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7982 | struct file *file) | |
7983 | { | |
50238531 | 7984 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7985 | } |
7986 | ||
05f3fb3c | 7987 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7988 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7989 | unsigned nr_args) |
7990 | { | |
269bbe5f BM |
7991 | struct fixed_rsrc_data *data = ctx->file_data; |
7992 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7993 | struct file *file, **file_slot; |
c3a31e60 JA |
7994 | __s32 __user *fds; |
7995 | int fd, i, err; | |
7996 | __u32 done; | |
05589553 | 7997 | bool needs_switch = false; |
c3a31e60 | 7998 | |
05f3fb3c | 7999 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
8000 | return -EOVERFLOW; |
8001 | if (done > ctx->nr_user_files) | |
8002 | return -EINVAL; | |
8003 | ||
bc9744cd | 8004 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
8005 | if (!ref_node) |
8006 | return -ENOMEM; | |
bc9744cd | 8007 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 8008 | |
269bbe5f | 8009 | fds = u64_to_user_ptr(up->data); |
67973b93 | 8010 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
8011 | err = 0; |
8012 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
8013 | err = -EFAULT; | |
8014 | break; | |
8015 | } | |
4e0377a1 | 8016 | if (fd == IORING_REGISTER_FILES_SKIP) |
8017 | continue; | |
8018 | ||
67973b93 | 8019 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
8020 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
8021 | ||
8022 | if (*file_slot) { | |
8023 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
8024 | if (err) |
8025 | break; | |
ea64ec02 | 8026 | *file_slot = NULL; |
05589553 | 8027 | needs_switch = true; |
c3a31e60 JA |
8028 | } |
8029 | if (fd != -1) { | |
c3a31e60 JA |
8030 | file = fget(fd); |
8031 | if (!file) { | |
8032 | err = -EBADF; | |
8033 | break; | |
8034 | } | |
8035 | /* | |
8036 | * Don't allow io_uring instances to be registered. If | |
8037 | * UNIX isn't enabled, then this causes a reference | |
8038 | * cycle and this instance can never get freed. If UNIX | |
8039 | * is enabled we'll handle it just fine, but there's | |
8040 | * still no point in allowing a ring fd as it doesn't | |
8041 | * support regular read/write anyway. | |
8042 | */ | |
8043 | if (file->f_op == &io_uring_fops) { | |
8044 | fput(file); | |
8045 | err = -EBADF; | |
8046 | break; | |
8047 | } | |
e68a3ff8 | 8048 | *file_slot = file; |
c3a31e60 | 8049 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8050 | if (err) { |
e68a3ff8 | 8051 | *file_slot = NULL; |
f3bd9dae | 8052 | fput(file); |
c3a31e60 | 8053 | break; |
f3bd9dae | 8054 | } |
c3a31e60 | 8055 | } |
05f3fb3c JA |
8056 | } |
8057 | ||
05589553 | 8058 | if (needs_switch) { |
b2e96852 | 8059 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 8060 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 8061 | } else |
269bbe5f | 8062 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
8063 | |
8064 | return done ? done : err; | |
8065 | } | |
05589553 | 8066 | |
05f3fb3c JA |
8067 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
8068 | unsigned nr_args) | |
8069 | { | |
269bbe5f | 8070 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
8071 | |
8072 | if (!ctx->file_data) | |
8073 | return -ENXIO; | |
8074 | if (!nr_args) | |
8075 | return -EINVAL; | |
8076 | if (copy_from_user(&up, arg, sizeof(up))) | |
8077 | return -EFAULT; | |
8078 | if (up.resv) | |
8079 | return -EINVAL; | |
8080 | ||
8081 | return __io_sqe_files_update(ctx, &up, nr_args); | |
8082 | } | |
c3a31e60 | 8083 | |
5280f7e5 | 8084 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
8085 | { |
8086 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8087 | ||
5280f7e5 PB |
8088 | req = io_put_req_find_next(req); |
8089 | return req ? &req->work : NULL; | |
7d723065 JA |
8090 | } |
8091 | ||
24369c2e PB |
8092 | static int io_init_wq_offload(struct io_ring_ctx *ctx, |
8093 | struct io_uring_params *p) | |
8094 | { | |
8095 | struct io_wq_data data; | |
8096 | struct fd f; | |
8097 | struct io_ring_ctx *ctx_attach; | |
8098 | unsigned int concurrency; | |
8099 | int ret = 0; | |
8100 | ||
8101 | data.user = ctx->user; | |
e9fd9396 | 8102 | data.free_work = io_free_work; |
f5fa38c5 | 8103 | data.do_work = io_wq_submit_work; |
24369c2e PB |
8104 | |
8105 | if (!(p->flags & IORING_SETUP_ATTACH_WQ)) { | |
8106 | /* Do QD, or 4 * CPUS, whatever is smallest */ | |
8107 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
8108 | ||
8109 | ctx->io_wq = io_wq_create(concurrency, &data); | |
8110 | if (IS_ERR(ctx->io_wq)) { | |
8111 | ret = PTR_ERR(ctx->io_wq); | |
8112 | ctx->io_wq = NULL; | |
8113 | } | |
8114 | return ret; | |
8115 | } | |
8116 | ||
8117 | f = fdget(p->wq_fd); | |
8118 | if (!f.file) | |
8119 | return -EBADF; | |
8120 | ||
8121 | if (f.file->f_op != &io_uring_fops) { | |
8122 | ret = -EINVAL; | |
8123 | goto out_fput; | |
8124 | } | |
8125 | ||
8126 | ctx_attach = f.file->private_data; | |
8127 | /* @io_wq is protected by holding the fd */ | |
8128 | if (!io_wq_get(ctx_attach->io_wq, &data)) { | |
8129 | ret = -EINVAL; | |
8130 | goto out_fput; | |
8131 | } | |
8132 | ||
8133 | ctx->io_wq = ctx_attach->io_wq; | |
8134 | out_fput: | |
8135 | fdput(f); | |
8136 | return ret; | |
8137 | } | |
8138 | ||
0f212204 JA |
8139 | static int io_uring_alloc_task_context(struct task_struct *task) |
8140 | { | |
8141 | struct io_uring_task *tctx; | |
d8a6df10 | 8142 | int ret; |
0f212204 JA |
8143 | |
8144 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
8145 | if (unlikely(!tctx)) | |
8146 | return -ENOMEM; | |
8147 | ||
d8a6df10 JA |
8148 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8149 | if (unlikely(ret)) { | |
8150 | kfree(tctx); | |
8151 | return ret; | |
8152 | } | |
8153 | ||
0f212204 JA |
8154 | xa_init(&tctx->xa); |
8155 | init_waitqueue_head(&tctx->wait); | |
8156 | tctx->last = NULL; | |
fdaf083c JA |
8157 | atomic_set(&tctx->in_idle, 0); |
8158 | tctx->sqpoll = false; | |
500a373d JA |
8159 | io_init_identity(&tctx->__identity); |
8160 | tctx->identity = &tctx->__identity; | |
0f212204 | 8161 | task->io_uring = tctx; |
7cbf1722 JA |
8162 | spin_lock_init(&tctx->task_lock); |
8163 | INIT_WQ_LIST(&tctx->task_list); | |
8164 | tctx->task_state = 0; | |
8165 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
8166 | return 0; |
8167 | } | |
8168 | ||
8169 | void __io_uring_free(struct task_struct *tsk) | |
8170 | { | |
8171 | struct io_uring_task *tctx = tsk->io_uring; | |
8172 | ||
8173 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
8174 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
8175 | if (tctx->identity != &tctx->__identity) | |
8176 | kfree(tctx->identity); | |
d8a6df10 | 8177 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8178 | kfree(tctx); |
8179 | tsk->io_uring = NULL; | |
8180 | } | |
8181 | ||
7e84e1c7 SG |
8182 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8183 | struct io_uring_params *p) | |
2b188cc1 JA |
8184 | { |
8185 | int ret; | |
8186 | ||
6c271ce2 | 8187 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
8188 | struct io_sq_data *sqd; |
8189 | ||
3ec482d1 | 8190 | ret = -EPERM; |
ce59fc69 | 8191 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
8192 | goto err; |
8193 | ||
534ca6d6 JA |
8194 | sqd = io_get_sq_data(p); |
8195 | if (IS_ERR(sqd)) { | |
8196 | ret = PTR_ERR(sqd); | |
8197 | goto err; | |
8198 | } | |
69fb2131 | 8199 | |
534ca6d6 | 8200 | ctx->sq_data = sqd; |
69fb2131 JA |
8201 | io_sq_thread_park(sqd); |
8202 | mutex_lock(&sqd->ctx_lock); | |
8203 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
8204 | mutex_unlock(&sqd->ctx_lock); | |
8205 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 8206 | |
917257da JA |
8207 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8208 | if (!ctx->sq_thread_idle) | |
8209 | ctx->sq_thread_idle = HZ; | |
8210 | ||
aa06165d JA |
8211 | if (sqd->thread) |
8212 | goto done; | |
8213 | ||
6c271ce2 | 8214 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8215 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8216 | |
917257da | 8217 | ret = -EINVAL; |
44a9bd18 JA |
8218 | if (cpu >= nr_cpu_ids) |
8219 | goto err; | |
7889f44d | 8220 | if (!cpu_online(cpu)) |
917257da JA |
8221 | goto err; |
8222 | ||
69fb2131 | 8223 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 8224 | cpu, "io_uring-sq"); |
6c271ce2 | 8225 | } else { |
69fb2131 | 8226 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
8227 | "io_uring-sq"); |
8228 | } | |
534ca6d6 JA |
8229 | if (IS_ERR(sqd->thread)) { |
8230 | ret = PTR_ERR(sqd->thread); | |
8231 | sqd->thread = NULL; | |
6c271ce2 JA |
8232 | goto err; |
8233 | } | |
534ca6d6 | 8234 | ret = io_uring_alloc_task_context(sqd->thread); |
0f212204 JA |
8235 | if (ret) |
8236 | goto err; | |
6c271ce2 JA |
8237 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8238 | /* Can't have SQ_AFF without SQPOLL */ | |
8239 | ret = -EINVAL; | |
8240 | goto err; | |
8241 | } | |
8242 | ||
aa06165d | 8243 | done: |
24369c2e PB |
8244 | ret = io_init_wq_offload(ctx, p); |
8245 | if (ret) | |
2b188cc1 | 8246 | goto err; |
2b188cc1 JA |
8247 | |
8248 | return 0; | |
8249 | err: | |
54a91f3b | 8250 | io_finish_async(ctx); |
2b188cc1 JA |
8251 | return ret; |
8252 | } | |
8253 | ||
7e84e1c7 SG |
8254 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
8255 | { | |
534ca6d6 JA |
8256 | struct io_sq_data *sqd = ctx->sq_data; |
8257 | ||
8258 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
8259 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
8260 | } |
8261 | ||
a087e2b5 BM |
8262 | static inline void __io_unaccount_mem(struct user_struct *user, |
8263 | unsigned long nr_pages) | |
2b188cc1 JA |
8264 | { |
8265 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8266 | } | |
8267 | ||
a087e2b5 BM |
8268 | static inline int __io_account_mem(struct user_struct *user, |
8269 | unsigned long nr_pages) | |
2b188cc1 JA |
8270 | { |
8271 | unsigned long page_limit, cur_pages, new_pages; | |
8272 | ||
8273 | /* Don't allow more pages than we can safely lock */ | |
8274 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8275 | ||
8276 | do { | |
8277 | cur_pages = atomic_long_read(&user->locked_vm); | |
8278 | new_pages = cur_pages + nr_pages; | |
8279 | if (new_pages > page_limit) | |
8280 | return -ENOMEM; | |
8281 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8282 | new_pages) != cur_pages); | |
8283 | ||
8284 | return 0; | |
8285 | } | |
8286 | ||
26bfa89e | 8287 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8288 | { |
aad5d8da | 8289 | if (ctx->limit_mem) |
a087e2b5 | 8290 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8291 | |
26bfa89e JA |
8292 | if (ctx->mm_account) |
8293 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8294 | } |
8295 | ||
26bfa89e | 8296 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8297 | { |
30975825 BM |
8298 | int ret; |
8299 | ||
8300 | if (ctx->limit_mem) { | |
8301 | ret = __io_account_mem(ctx->user, nr_pages); | |
8302 | if (ret) | |
8303 | return ret; | |
8304 | } | |
8305 | ||
26bfa89e JA |
8306 | if (ctx->mm_account) |
8307 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8308 | |
8309 | return 0; | |
8310 | } | |
8311 | ||
2b188cc1 JA |
8312 | static void io_mem_free(void *ptr) |
8313 | { | |
52e04ef4 MR |
8314 | struct page *page; |
8315 | ||
8316 | if (!ptr) | |
8317 | return; | |
2b188cc1 | 8318 | |
52e04ef4 | 8319 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8320 | if (put_page_testzero(page)) |
8321 | free_compound_page(page); | |
8322 | } | |
8323 | ||
8324 | static void *io_mem_alloc(size_t size) | |
8325 | { | |
8326 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8327 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8328 | |
8329 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8330 | } | |
8331 | ||
75b28aff HV |
8332 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8333 | size_t *sq_offset) | |
8334 | { | |
8335 | struct io_rings *rings; | |
8336 | size_t off, sq_array_size; | |
8337 | ||
8338 | off = struct_size(rings, cqes, cq_entries); | |
8339 | if (off == SIZE_MAX) | |
8340 | return SIZE_MAX; | |
8341 | ||
8342 | #ifdef CONFIG_SMP | |
8343 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8344 | if (off == 0) | |
8345 | return SIZE_MAX; | |
8346 | #endif | |
8347 | ||
b36200f5 DV |
8348 | if (sq_offset) |
8349 | *sq_offset = off; | |
8350 | ||
75b28aff HV |
8351 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8352 | if (sq_array_size == SIZE_MAX) | |
8353 | return SIZE_MAX; | |
8354 | ||
8355 | if (check_add_overflow(off, sq_array_size, &off)) | |
8356 | return SIZE_MAX; | |
8357 | ||
75b28aff HV |
8358 | return off; |
8359 | } | |
8360 | ||
0a96bbe4 | 8361 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8362 | { |
8363 | int i, j; | |
8364 | ||
8365 | if (!ctx->user_bufs) | |
8366 | return -ENXIO; | |
8367 | ||
8368 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8369 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8370 | ||
8371 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8372 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8373 | |
de293938 | 8374 | if (imu->acct_pages) |
26bfa89e | 8375 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8376 | kvfree(imu->bvec); |
edafccee JA |
8377 | imu->nr_bvecs = 0; |
8378 | } | |
8379 | ||
8380 | kfree(ctx->user_bufs); | |
8381 | ctx->user_bufs = NULL; | |
8382 | ctx->nr_user_bufs = 0; | |
8383 | return 0; | |
8384 | } | |
8385 | ||
8386 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8387 | void __user *arg, unsigned index) | |
8388 | { | |
8389 | struct iovec __user *src; | |
8390 | ||
8391 | #ifdef CONFIG_COMPAT | |
8392 | if (ctx->compat) { | |
8393 | struct compat_iovec __user *ciovs; | |
8394 | struct compat_iovec ciov; | |
8395 | ||
8396 | ciovs = (struct compat_iovec __user *) arg; | |
8397 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8398 | return -EFAULT; | |
8399 | ||
d55e5f5b | 8400 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8401 | dst->iov_len = ciov.iov_len; |
8402 | return 0; | |
8403 | } | |
8404 | #endif | |
8405 | src = (struct iovec __user *) arg; | |
8406 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8407 | return -EFAULT; | |
8408 | return 0; | |
8409 | } | |
8410 | ||
de293938 JA |
8411 | /* |
8412 | * Not super efficient, but this is just a registration time. And we do cache | |
8413 | * the last compound head, so generally we'll only do a full search if we don't | |
8414 | * match that one. | |
8415 | * | |
8416 | * We check if the given compound head page has already been accounted, to | |
8417 | * avoid double accounting it. This allows us to account the full size of the | |
8418 | * page, not just the constituent pages of a huge page. | |
8419 | */ | |
8420 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8421 | int nr_pages, struct page *hpage) | |
8422 | { | |
8423 | int i, j; | |
8424 | ||
8425 | /* check current page array */ | |
8426 | for (i = 0; i < nr_pages; i++) { | |
8427 | if (!PageCompound(pages[i])) | |
8428 | continue; | |
8429 | if (compound_head(pages[i]) == hpage) | |
8430 | return true; | |
8431 | } | |
8432 | ||
8433 | /* check previously registered pages */ | |
8434 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8435 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8436 | ||
8437 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8438 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8439 | continue; | |
8440 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8441 | return true; | |
8442 | } | |
8443 | } | |
8444 | ||
8445 | return false; | |
8446 | } | |
8447 | ||
8448 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8449 | int nr_pages, struct io_mapped_ubuf *imu, | |
8450 | struct page **last_hpage) | |
8451 | { | |
8452 | int i, ret; | |
8453 | ||
8454 | for (i = 0; i < nr_pages; i++) { | |
8455 | if (!PageCompound(pages[i])) { | |
8456 | imu->acct_pages++; | |
8457 | } else { | |
8458 | struct page *hpage; | |
8459 | ||
8460 | hpage = compound_head(pages[i]); | |
8461 | if (hpage == *last_hpage) | |
8462 | continue; | |
8463 | *last_hpage = hpage; | |
8464 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8465 | continue; | |
8466 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8467 | } | |
8468 | } | |
8469 | ||
8470 | if (!imu->acct_pages) | |
8471 | return 0; | |
8472 | ||
26bfa89e | 8473 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8474 | if (ret) |
8475 | imu->acct_pages = 0; | |
8476 | return ret; | |
8477 | } | |
8478 | ||
0a96bbe4 BM |
8479 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8480 | struct io_mapped_ubuf *imu, | |
8481 | struct page **last_hpage) | |
edafccee JA |
8482 | { |
8483 | struct vm_area_struct **vmas = NULL; | |
8484 | struct page **pages = NULL; | |
0a96bbe4 BM |
8485 | unsigned long off, start, end, ubuf; |
8486 | size_t size; | |
8487 | int ret, pret, nr_pages, i; | |
8488 | ||
8489 | ubuf = (unsigned long) iov->iov_base; | |
8490 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8491 | start = ubuf >> PAGE_SHIFT; | |
8492 | nr_pages = end - start; | |
8493 | ||
8494 | ret = -ENOMEM; | |
8495 | ||
8496 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8497 | if (!pages) | |
8498 | goto done; | |
8499 | ||
8500 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8501 | GFP_KERNEL); | |
8502 | if (!vmas) | |
8503 | goto done; | |
8504 | ||
8505 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), | |
8506 | GFP_KERNEL); | |
8507 | if (!imu->bvec) | |
8508 | goto done; | |
8509 | ||
8510 | ret = 0; | |
8511 | mmap_read_lock(current->mm); | |
8512 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8513 | pages, vmas); | |
8514 | if (pret == nr_pages) { | |
8515 | /* don't support file backed memory */ | |
8516 | for (i = 0; i < nr_pages; i++) { | |
8517 | struct vm_area_struct *vma = vmas[i]; | |
8518 | ||
8519 | if (vma->vm_file && | |
8520 | !is_file_hugepages(vma->vm_file)) { | |
8521 | ret = -EOPNOTSUPP; | |
8522 | break; | |
8523 | } | |
8524 | } | |
8525 | } else { | |
8526 | ret = pret < 0 ? pret : -EFAULT; | |
8527 | } | |
8528 | mmap_read_unlock(current->mm); | |
8529 | if (ret) { | |
8530 | /* | |
8531 | * if we did partial map, or found file backed vmas, | |
8532 | * release any pages we did get | |
8533 | */ | |
8534 | if (pret > 0) | |
8535 | unpin_user_pages(pages, pret); | |
8536 | kvfree(imu->bvec); | |
8537 | goto done; | |
8538 | } | |
8539 | ||
8540 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8541 | if (ret) { | |
8542 | unpin_user_pages(pages, pret); | |
8543 | kvfree(imu->bvec); | |
8544 | goto done; | |
8545 | } | |
8546 | ||
8547 | off = ubuf & ~PAGE_MASK; | |
8548 | size = iov->iov_len; | |
8549 | for (i = 0; i < nr_pages; i++) { | |
8550 | size_t vec_len; | |
8551 | ||
8552 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8553 | imu->bvec[i].bv_page = pages[i]; | |
8554 | imu->bvec[i].bv_len = vec_len; | |
8555 | imu->bvec[i].bv_offset = off; | |
8556 | off = 0; | |
8557 | size -= vec_len; | |
8558 | } | |
8559 | /* store original address for later verification */ | |
8560 | imu->ubuf = ubuf; | |
8561 | imu->len = iov->iov_len; | |
8562 | imu->nr_bvecs = nr_pages; | |
8563 | ret = 0; | |
8564 | done: | |
8565 | kvfree(pages); | |
8566 | kvfree(vmas); | |
8567 | return ret; | |
8568 | } | |
8569 | ||
2b358604 | 8570 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8571 | { |
edafccee JA |
8572 | if (ctx->user_bufs) |
8573 | return -EBUSY; | |
8574 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8575 | return -EINVAL; | |
8576 | ||
8577 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8578 | GFP_KERNEL); | |
8579 | if (!ctx->user_bufs) | |
8580 | return -ENOMEM; | |
8581 | ||
2b358604 BM |
8582 | return 0; |
8583 | } | |
8584 | ||
8585 | static int io_buffer_validate(struct iovec *iov) | |
8586 | { | |
8587 | /* | |
8588 | * Don't impose further limits on the size and buffer | |
8589 | * constraints here, we'll -EINVAL later when IO is | |
8590 | * submitted if they are wrong. | |
8591 | */ | |
8592 | if (!iov->iov_base || !iov->iov_len) | |
8593 | return -EFAULT; | |
8594 | ||
8595 | /* arbitrary limit, but we need something */ | |
8596 | if (iov->iov_len > SZ_1G) | |
8597 | return -EFAULT; | |
8598 | ||
8599 | return 0; | |
8600 | } | |
8601 | ||
8602 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, | |
8603 | unsigned int nr_args) | |
8604 | { | |
8605 | int i, ret; | |
8606 | struct iovec iov; | |
8607 | struct page *last_hpage = NULL; | |
8608 | ||
8609 | ret = io_buffers_map_alloc(ctx, nr_args); | |
8610 | if (ret) | |
8611 | return ret; | |
8612 | ||
edafccee JA |
8613 | for (i = 0; i < nr_args; i++) { |
8614 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee JA |
8615 | |
8616 | ret = io_copy_iov(ctx, &iov, arg, i); | |
8617 | if (ret) | |
0a96bbe4 | 8618 | break; |
edafccee | 8619 | |
2b358604 BM |
8620 | ret = io_buffer_validate(&iov); |
8621 | if (ret) | |
0a96bbe4 | 8622 | break; |
edafccee | 8623 | |
0a96bbe4 BM |
8624 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8625 | if (ret) | |
8626 | break; | |
edafccee JA |
8627 | |
8628 | ctx->nr_user_bufs++; | |
8629 | } | |
0a96bbe4 BM |
8630 | |
8631 | if (ret) | |
8632 | io_sqe_buffers_unregister(ctx); | |
8633 | ||
edafccee JA |
8634 | return ret; |
8635 | } | |
8636 | ||
9b402849 JA |
8637 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8638 | { | |
8639 | __s32 __user *fds = arg; | |
8640 | int fd; | |
8641 | ||
8642 | if (ctx->cq_ev_fd) | |
8643 | return -EBUSY; | |
8644 | ||
8645 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8646 | return -EFAULT; | |
8647 | ||
8648 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8649 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8650 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8651 | ctx->cq_ev_fd = NULL; | |
8652 | return ret; | |
8653 | } | |
8654 | ||
8655 | return 0; | |
8656 | } | |
8657 | ||
8658 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8659 | { | |
8660 | if (ctx->cq_ev_fd) { | |
8661 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8662 | ctx->cq_ev_fd = NULL; | |
8663 | return 0; | |
8664 | } | |
8665 | ||
8666 | return -ENXIO; | |
8667 | } | |
8668 | ||
5a2e745d JA |
8669 | static int __io_destroy_buffers(int id, void *p, void *data) |
8670 | { | |
8671 | struct io_ring_ctx *ctx = data; | |
8672 | struct io_buffer *buf = p; | |
8673 | ||
067524e9 | 8674 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8675 | return 0; |
8676 | } | |
8677 | ||
8678 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8679 | { | |
8680 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8681 | idr_destroy(&ctx->io_buffer_idr); | |
8682 | } | |
8683 | ||
68e68ee6 | 8684 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8685 | { |
68e68ee6 | 8686 | struct io_kiocb *req, *nxt; |
1b4c351f | 8687 | |
68e68ee6 JA |
8688 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8689 | if (tsk && req->task != tsk) | |
8690 | continue; | |
1b4c351f JA |
8691 | list_del(&req->compl.list); |
8692 | kmem_cache_free(req_cachep, req); | |
8693 | } | |
8694 | } | |
8695 | ||
9a4fdbd8 | 8696 | static void io_req_caches_free(struct io_ring_ctx *ctx, struct task_struct *tsk) |
2b188cc1 | 8697 | { |
bf019da7 PB |
8698 | struct io_submit_state *submit_state = &ctx->submit_state; |
8699 | ||
9a4fdbd8 JA |
8700 | mutex_lock(&ctx->uring_lock); |
8701 | ||
8702 | if (submit_state->free_reqs) | |
8703 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, | |
8704 | submit_state->reqs); | |
8705 | ||
8706 | io_req_cache_free(&submit_state->comp.free_list, NULL); | |
8707 | ||
8708 | spin_lock_irq(&ctx->completion_lock); | |
8709 | io_req_cache_free(&submit_state->comp.locked_free_list, NULL); | |
8710 | spin_unlock_irq(&ctx->completion_lock); | |
8711 | ||
8712 | mutex_unlock(&ctx->uring_lock); | |
8713 | } | |
8714 | ||
8715 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) | |
8716 | { | |
04fc6c80 PB |
8717 | /* |
8718 | * Some may use context even when all refs and requests have been put, | |
8719 | * and they are free to do so while still holding uring_lock, see | |
8720 | * __io_req_task_submit(). Wait for them to finish. | |
8721 | */ | |
8722 | mutex_lock(&ctx->uring_lock); | |
8723 | mutex_unlock(&ctx->uring_lock); | |
8724 | ||
6b06314c | 8725 | io_finish_async(ctx); |
0a96bbe4 | 8726 | io_sqe_buffers_unregister(ctx); |
2aede0e4 JA |
8727 | |
8728 | if (ctx->sqo_task) { | |
8729 | put_task_struct(ctx->sqo_task); | |
8730 | ctx->sqo_task = NULL; | |
8731 | mmdrop(ctx->mm_account); | |
8732 | ctx->mm_account = NULL; | |
30975825 | 8733 | } |
def596e9 | 8734 | |
91d8f519 DZ |
8735 | #ifdef CONFIG_BLK_CGROUP |
8736 | if (ctx->sqo_blkcg_css) | |
8737 | css_put(ctx->sqo_blkcg_css); | |
8738 | #endif | |
8739 | ||
6b06314c | 8740 | io_sqe_files_unregister(ctx); |
9b402849 | 8741 | io_eventfd_unregister(ctx); |
5a2e745d | 8742 | io_destroy_buffers(ctx); |
41726c9a | 8743 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8744 | |
2b188cc1 | 8745 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8746 | if (ctx->ring_sock) { |
8747 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8748 | sock_release(ctx->ring_sock); |
355e8d26 | 8749 | } |
2b188cc1 JA |
8750 | #endif |
8751 | ||
75b28aff | 8752 | io_mem_free(ctx->rings); |
2b188cc1 | 8753 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8754 | |
8755 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8756 | free_uid(ctx->user); |
181e448d | 8757 | put_cred(ctx->creds); |
9a4fdbd8 | 8758 | io_req_caches_free(ctx, NULL); |
78076bb6 | 8759 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8760 | kfree(ctx); |
8761 | } | |
8762 | ||
8763 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8764 | { | |
8765 | struct io_ring_ctx *ctx = file->private_data; | |
8766 | __poll_t mask = 0; | |
8767 | ||
8768 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8769 | /* |
8770 | * synchronizes with barrier from wq_has_sleeper call in | |
8771 | * io_commit_cqring | |
8772 | */ | |
2b188cc1 | 8773 | smp_rmb(); |
90554200 | 8774 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8775 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8776 | |
8777 | /* | |
8778 | * Don't flush cqring overflow list here, just do a simple check. | |
8779 | * Otherwise there could possible be ABBA deadlock: | |
8780 | * CPU0 CPU1 | |
8781 | * ---- ---- | |
8782 | * lock(&ctx->uring_lock); | |
8783 | * lock(&ep->mtx); | |
8784 | * lock(&ctx->uring_lock); | |
8785 | * lock(&ep->mtx); | |
8786 | * | |
8787 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8788 | * pushs them to do the flush. | |
8789 | */ | |
8790 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8791 | mask |= EPOLLIN | EPOLLRDNORM; |
8792 | ||
8793 | return mask; | |
8794 | } | |
8795 | ||
8796 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8797 | { | |
8798 | struct io_ring_ctx *ctx = file->private_data; | |
8799 | ||
8800 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8801 | } | |
8802 | ||
0bead8cd | 8803 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8804 | { |
1e6fa521 | 8805 | struct io_identity *iod; |
071698e1 | 8806 | |
1e6fa521 JA |
8807 | iod = idr_remove(&ctx->personality_idr, id); |
8808 | if (iod) { | |
8809 | put_cred(iod->creds); | |
8810 | if (refcount_dec_and_test(&iod->count)) | |
8811 | kfree(iod); | |
0bead8cd | 8812 | return 0; |
1e6fa521 | 8813 | } |
0bead8cd YD |
8814 | |
8815 | return -EINVAL; | |
8816 | } | |
8817 | ||
8818 | static int io_remove_personalities(int id, void *p, void *data) | |
8819 | { | |
8820 | struct io_ring_ctx *ctx = data; | |
8821 | ||
8822 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8823 | return 0; |
8824 | } | |
8825 | ||
85faa7b8 JA |
8826 | static void io_ring_exit_work(struct work_struct *work) |
8827 | { | |
b2edc0a7 PB |
8828 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8829 | exit_work); | |
85faa7b8 | 8830 | |
56952e91 JA |
8831 | /* |
8832 | * If we're doing polled IO and end up having requests being | |
8833 | * submitted async (out-of-line), then completions can come in while | |
8834 | * we're waiting for refs to drop. We need to reap these manually, | |
8835 | * as nobody else will be looking for them. | |
8836 | */ | |
b2edc0a7 | 8837 | do { |
9936c7c2 | 8838 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b2edc0a7 | 8839 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8840 | io_ring_ctx_free(ctx); |
8841 | } | |
8842 | ||
00c18640 JA |
8843 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8844 | { | |
8845 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8846 | ||
8847 | return req->ctx == data; | |
8848 | } | |
8849 | ||
2b188cc1 JA |
8850 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8851 | { | |
8852 | mutex_lock(&ctx->uring_lock); | |
8853 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8854 | |
8855 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8856 | ctx->sqo_dead = 1; | |
8857 | ||
cda286f0 PB |
8858 | /* if force is set, the ring is going away. always drop after that */ |
8859 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8860 | if (ctx->rings) |
6c503150 | 8861 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8862 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8863 | mutex_unlock(&ctx->uring_lock); |
8864 | ||
6b81928d PB |
8865 | io_kill_timeouts(ctx, NULL, NULL); |
8866 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a JA |
8867 | |
8868 | if (ctx->io_wq) | |
00c18640 | 8869 | io_wq_cancel_cb(ctx->io_wq, io_cancel_ctx_cb, ctx, true); |
561fb04a | 8870 | |
15dff286 | 8871 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8872 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8873 | |
85faa7b8 | 8874 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8875 | /* |
8876 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8877 | * if we're exiting a ton of rings at the same time. It just adds | |
8878 | * noise and overhead, there's no discernable change in runtime | |
8879 | * over using system_wq. | |
8880 | */ | |
8881 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8882 | } |
8883 | ||
8884 | static int io_uring_release(struct inode *inode, struct file *file) | |
8885 | { | |
8886 | struct io_ring_ctx *ctx = file->private_data; | |
8887 | ||
8888 | file->private_data = NULL; | |
8889 | io_ring_ctx_wait_and_kill(ctx); | |
8890 | return 0; | |
8891 | } | |
8892 | ||
f6edbabb PB |
8893 | struct io_task_cancel { |
8894 | struct task_struct *task; | |
8895 | struct files_struct *files; | |
8896 | }; | |
f254ac04 | 8897 | |
f6edbabb | 8898 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8899 | { |
9a472ef7 | 8900 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8901 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8902 | bool ret; |
8903 | ||
f6edbabb | 8904 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8905 | unsigned long flags; |
8906 | struct io_ring_ctx *ctx = req->ctx; | |
8907 | ||
8908 | /* protect against races with linked timeouts */ | |
8909 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8910 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8911 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8912 | } else { | |
f6edbabb | 8913 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8914 | } |
8915 | return ret; | |
b711d4ea JA |
8916 | } |
8917 | ||
b7ddce3c | 8918 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8919 | struct task_struct *task, |
b7ddce3c PB |
8920 | struct files_struct *files) |
8921 | { | |
8922 | struct io_defer_entry *de = NULL; | |
8923 | LIST_HEAD(list); | |
8924 | ||
8925 | spin_lock_irq(&ctx->completion_lock); | |
8926 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8927 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8928 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8929 | break; | |
8930 | } | |
8931 | } | |
8932 | spin_unlock_irq(&ctx->completion_lock); | |
8933 | ||
8934 | while (!list_empty(&list)) { | |
8935 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8936 | list_del_init(&de->list); | |
8937 | req_set_fail_links(de->req); | |
8938 | io_put_req(de->req); | |
8939 | io_req_complete(de->req, -ECANCELED); | |
8940 | kfree(de); | |
8941 | } | |
8942 | } | |
8943 | ||
9936c7c2 PB |
8944 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8945 | struct task_struct *task, | |
8946 | struct files_struct *files) | |
8947 | { | |
8948 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
8949 | ||
8950 | while (1) { | |
8951 | enum io_wq_cancel cret; | |
8952 | bool ret = false; | |
8953 | ||
8954 | if (ctx->io_wq) { | |
8955 | cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, | |
8956 | &cancel, true); | |
8957 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8958 | } | |
8959 | ||
8960 | /* SQPOLL thread does its own polling */ | |
8961 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8962 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8963 | io_iopoll_try_reap_events(ctx); | |
8964 | ret = true; | |
8965 | } | |
8966 | } | |
8967 | ||
8968 | ret |= io_poll_remove_all(ctx, task, files); | |
8969 | ret |= io_kill_timeouts(ctx, task, files); | |
8970 | ret |= io_run_task_work(); | |
8971 | io_cqring_overflow_flush(ctx, true, task, files); | |
8972 | if (!ret) | |
8973 | break; | |
8974 | cond_resched(); | |
8975 | } | |
8976 | } | |
8977 | ||
ca70f00b PB |
8978 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8979 | struct task_struct *task, | |
8980 | struct files_struct *files) | |
8981 | { | |
8982 | struct io_kiocb *req; | |
8983 | int cnt = 0; | |
8984 | ||
8985 | spin_lock_irq(&ctx->inflight_lock); | |
8986 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8987 | cnt += io_match_task(req, task, files); | |
8988 | spin_unlock_irq(&ctx->inflight_lock); | |
8989 | return cnt; | |
8990 | } | |
8991 | ||
b52fda00 | 8992 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8993 | struct task_struct *task, |
fcb323cc JA |
8994 | struct files_struct *files) |
8995 | { | |
fcb323cc | 8996 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8997 | DEFINE_WAIT(wait); |
ca70f00b | 8998 | int inflight; |
fcb323cc | 8999 | |
ca70f00b PB |
9000 | inflight = io_uring_count_inflight(ctx, task, files); |
9001 | if (!inflight) | |
fcb323cc | 9002 | break; |
f6edbabb | 9003 | |
9936c7c2 | 9004 | io_uring_try_cancel_requests(ctx, task, files); |
34343786 PB |
9005 | |
9006 | if (ctx->sq_data) | |
9007 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
9008 | prepare_to_wait(&task->io_uring->wait, &wait, |
9009 | TASK_UNINTERRUPTIBLE); | |
9010 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
9011 | schedule(); | |
c98de08c | 9012 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
9013 | if (ctx->sq_data) |
9014 | io_sq_thread_park(ctx->sq_data); | |
fcb323cc | 9015 | } |
44e728b8 PB |
9016 | } |
9017 | ||
d9d05217 PB |
9018 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
9019 | { | |
d9d05217 PB |
9020 | mutex_lock(&ctx->uring_lock); |
9021 | ctx->sqo_dead = 1; | |
9022 | mutex_unlock(&ctx->uring_lock); | |
9023 | ||
9024 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
9025 | if (ctx->rings) |
9026 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
9027 | } |
9028 | ||
0f212204 JA |
9029 | /* |
9030 | * We need to iteratively cancel requests, in case a request has dependent | |
9031 | * hard links. These persist even for failure of cancelations, hence keep | |
9032 | * looping until none are found. | |
9033 | */ | |
9034 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
9035 | struct files_struct *files) | |
9036 | { | |
9037 | struct task_struct *task = current; | |
9038 | ||
fdaf083c | 9039 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 | 9040 | io_disable_sqo_submit(ctx); |
534ca6d6 | 9041 | task = ctx->sq_data->thread; |
fdaf083c JA |
9042 | atomic_inc(&task->io_uring->in_idle); |
9043 | io_sq_thread_park(ctx->sq_data); | |
9044 | } | |
0f212204 | 9045 | |
df9923f9 | 9046 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 9047 | |
3a7efd1a | 9048 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 9049 | if (!files) |
9936c7c2 | 9050 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c JA |
9051 | |
9052 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
9053 | atomic_dec(&task->io_uring->in_idle); | |
9054 | /* | |
9055 | * If the files that are going away are the ones in the thread | |
9056 | * identity, clear them out. | |
9057 | */ | |
9058 | if (task->io_uring->identity->files == files) | |
9059 | task->io_uring->identity->files = NULL; | |
9060 | io_sq_thread_unpark(ctx->sq_data); | |
9061 | } | |
0f212204 JA |
9062 | } |
9063 | ||
9064 | /* | |
9065 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9066 | */ | |
fdaf083c | 9067 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 9068 | { |
236434c3 | 9069 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 9070 | int ret; |
236434c3 MWO |
9071 | |
9072 | if (unlikely(!tctx)) { | |
0f212204 JA |
9073 | ret = io_uring_alloc_task_context(current); |
9074 | if (unlikely(ret)) | |
9075 | return ret; | |
236434c3 | 9076 | tctx = current->io_uring; |
0f212204 | 9077 | } |
236434c3 MWO |
9078 | if (tctx->last != file) { |
9079 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 9080 | |
236434c3 | 9081 | if (!old) { |
0f212204 | 9082 | get_file(file); |
a528b04e PB |
9083 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
9084 | file, GFP_KERNEL)); | |
9085 | if (ret) { | |
9086 | fput(file); | |
9087 | return ret; | |
9088 | } | |
ecfc8492 PB |
9089 | |
9090 | /* one and only SQPOLL file note, held by sqo_task */ | |
9091 | WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && | |
9092 | current != ctx->sqo_task); | |
0f212204 | 9093 | } |
236434c3 | 9094 | tctx->last = file; |
0f212204 JA |
9095 | } |
9096 | ||
fdaf083c JA |
9097 | /* |
9098 | * This is race safe in that the task itself is doing this, hence it | |
9099 | * cannot be going through the exit/cancel paths at the same time. | |
9100 | * This cannot be modified while exit/cancel is running. | |
9101 | */ | |
9102 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
9103 | tctx->sqpoll = true; | |
9104 | ||
0f212204 JA |
9105 | return 0; |
9106 | } | |
9107 | ||
9108 | /* | |
9109 | * Remove this io_uring_file -> task mapping. | |
9110 | */ | |
9111 | static void io_uring_del_task_file(struct file *file) | |
9112 | { | |
9113 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
9114 | |
9115 | if (tctx->last == file) | |
9116 | tctx->last = NULL; | |
5e2ed8c4 | 9117 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
9118 | if (file) |
9119 | fput(file); | |
9120 | } | |
9121 | ||
de7f1d9e PB |
9122 | static void io_uring_remove_task_files(struct io_uring_task *tctx) |
9123 | { | |
9124 | struct file *file; | |
9125 | unsigned long index; | |
9126 | ||
9127 | xa_for_each(&tctx->xa, index, file) | |
9128 | io_uring_del_task_file(file); | |
9129 | } | |
9130 | ||
0f212204 JA |
9131 | void __io_uring_files_cancel(struct files_struct *files) |
9132 | { | |
9133 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
9134 | struct file *file; |
9135 | unsigned long index; | |
0f212204 JA |
9136 | |
9137 | /* make sure overflow events are dropped */ | |
fdaf083c | 9138 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
9139 | xa_for_each(&tctx->xa, index, file) |
9140 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 9141 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9142 | |
9143 | if (files) | |
9144 | io_uring_remove_task_files(tctx); | |
fdaf083c JA |
9145 | } |
9146 | ||
9147 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
9148 | { | |
0e9ddb39 PB |
9149 | return percpu_counter_sum(&tctx->inflight); |
9150 | } | |
fdaf083c | 9151 | |
0e9ddb39 PB |
9152 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
9153 | { | |
9154 | struct io_uring_task *tctx; | |
9155 | s64 inflight; | |
9156 | DEFINE_WAIT(wait); | |
fdaf083c | 9157 | |
0e9ddb39 PB |
9158 | if (!ctx->sq_data) |
9159 | return; | |
9160 | tctx = ctx->sq_data->thread->io_uring; | |
9161 | io_disable_sqo_submit(ctx); | |
fdaf083c | 9162 | |
0e9ddb39 PB |
9163 | atomic_inc(&tctx->in_idle); |
9164 | do { | |
9165 | /* read completions before cancelations */ | |
9166 | inflight = tctx_inflight(tctx); | |
9167 | if (!inflight) | |
9168 | break; | |
9169 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 9170 | |
0e9ddb39 PB |
9171 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9172 | /* | |
9173 | * If we've seen completions, retry without waiting. This | |
9174 | * avoids a race where a completion comes in before we did | |
9175 | * prepare_to_wait(). | |
9176 | */ | |
9177 | if (inflight == tctx_inflight(tctx)) | |
9178 | schedule(); | |
9179 | finish_wait(&tctx->wait, &wait); | |
9180 | } while (1); | |
9181 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9182 | } |
9183 | ||
0f212204 JA |
9184 | /* |
9185 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9186 | * requests. | |
9187 | */ | |
9188 | void __io_uring_task_cancel(void) | |
9189 | { | |
9190 | struct io_uring_task *tctx = current->io_uring; | |
9191 | DEFINE_WAIT(wait); | |
d8a6df10 | 9192 | s64 inflight; |
0f212204 JA |
9193 | |
9194 | /* make sure overflow events are dropped */ | |
fdaf083c | 9195 | atomic_inc(&tctx->in_idle); |
0f212204 | 9196 | |
0b5cd6c3 | 9197 | /* trigger io_disable_sqo_submit() */ |
0e9ddb39 PB |
9198 | if (tctx->sqpoll) { |
9199 | struct file *file; | |
9200 | unsigned long index; | |
9201 | ||
9202 | xa_for_each(&tctx->xa, index, file) | |
9203 | io_uring_cancel_sqpoll(file->private_data); | |
9204 | } | |
0b5cd6c3 | 9205 | |
d8a6df10 | 9206 | do { |
0f212204 | 9207 | /* read completions before cancelations */ |
fdaf083c | 9208 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9209 | if (!inflight) |
9210 | break; | |
0f212204 JA |
9211 | __io_uring_files_cancel(NULL); |
9212 | ||
9213 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9214 | ||
9215 | /* | |
a1bb3cd5 PB |
9216 | * If we've seen completions, retry without waiting. This |
9217 | * avoids a race where a completion comes in before we did | |
9218 | * prepare_to_wait(). | |
0f212204 | 9219 | */ |
a1bb3cd5 PB |
9220 | if (inflight == tctx_inflight(tctx)) |
9221 | schedule(); | |
f57555ed | 9222 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9223 | } while (1); |
0f212204 | 9224 | |
fdaf083c | 9225 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9226 | |
9227 | io_uring_remove_task_files(tctx); | |
44e728b8 PB |
9228 | } |
9229 | ||
fcb323cc JA |
9230 | static int io_uring_flush(struct file *file, void *data) |
9231 | { | |
6b5733eb | 9232 | struct io_uring_task *tctx = current->io_uring; |
d9d05217 | 9233 | struct io_ring_ctx *ctx = file->private_data; |
6b5733eb | 9234 | |
41be53e9 | 9235 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) { |
84965ff8 | 9236 | io_uring_cancel_task_requests(ctx, NULL); |
41be53e9 JA |
9237 | io_req_caches_free(ctx, current); |
9238 | } | |
84965ff8 | 9239 | |
6b5733eb | 9240 | if (!tctx) |
4f793dc4 PB |
9241 | return 0; |
9242 | ||
6b5733eb PB |
9243 | /* we should have cancelled and erased it before PF_EXITING */ |
9244 | WARN_ON_ONCE((current->flags & PF_EXITING) && | |
9245 | xa_load(&tctx->xa, (unsigned long)file)); | |
9246 | ||
4f793dc4 PB |
9247 | /* |
9248 | * fput() is pending, will be 2 if the only other ref is our potential | |
9249 | * task file note. If the task is exiting, drop regardless of count. | |
9250 | */ | |
6b5733eb PB |
9251 | if (atomic_long_read(&file->f_count) != 2) |
9252 | return 0; | |
4f793dc4 | 9253 | |
d9d05217 PB |
9254 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
9255 | /* there is only one file note, which is owned by sqo_task */ | |
4325cb49 PB |
9256 | WARN_ON_ONCE(ctx->sqo_task != current && |
9257 | xa_load(&tctx->xa, (unsigned long)file)); | |
9258 | /* sqo_dead check is for when this happens after cancellation */ | |
9259 | WARN_ON_ONCE(ctx->sqo_task == current && !ctx->sqo_dead && | |
d9d05217 PB |
9260 | !xa_load(&tctx->xa, (unsigned long)file)); |
9261 | ||
9262 | io_disable_sqo_submit(ctx); | |
9263 | } | |
9264 | ||
9265 | if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current) | |
9266 | io_uring_del_task_file(file); | |
fcb323cc JA |
9267 | return 0; |
9268 | } | |
9269 | ||
6c5c240e RP |
9270 | static void *io_uring_validate_mmap_request(struct file *file, |
9271 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9272 | { |
2b188cc1 | 9273 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9274 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9275 | struct page *page; |
9276 | void *ptr; | |
9277 | ||
9278 | switch (offset) { | |
9279 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9280 | case IORING_OFF_CQ_RING: |
9281 | ptr = ctx->rings; | |
2b188cc1 JA |
9282 | break; |
9283 | case IORING_OFF_SQES: | |
9284 | ptr = ctx->sq_sqes; | |
9285 | break; | |
2b188cc1 | 9286 | default: |
6c5c240e | 9287 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9288 | } |
9289 | ||
9290 | page = virt_to_head_page(ptr); | |
a50b854e | 9291 | if (sz > page_size(page)) |
6c5c240e RP |
9292 | return ERR_PTR(-EINVAL); |
9293 | ||
9294 | return ptr; | |
9295 | } | |
9296 | ||
9297 | #ifdef CONFIG_MMU | |
9298 | ||
9299 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9300 | { | |
9301 | size_t sz = vma->vm_end - vma->vm_start; | |
9302 | unsigned long pfn; | |
9303 | void *ptr; | |
9304 | ||
9305 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9306 | if (IS_ERR(ptr)) | |
9307 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9308 | |
9309 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9310 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9311 | } | |
9312 | ||
6c5c240e RP |
9313 | #else /* !CONFIG_MMU */ |
9314 | ||
9315 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9316 | { | |
9317 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9318 | } | |
9319 | ||
9320 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9321 | { | |
9322 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9323 | } | |
9324 | ||
9325 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9326 | unsigned long addr, unsigned long len, | |
9327 | unsigned long pgoff, unsigned long flags) | |
9328 | { | |
9329 | void *ptr; | |
9330 | ||
9331 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9332 | if (IS_ERR(ptr)) | |
9333 | return PTR_ERR(ptr); | |
9334 | ||
9335 | return (unsigned long) ptr; | |
9336 | } | |
9337 | ||
9338 | #endif /* !CONFIG_MMU */ | |
9339 | ||
d9d05217 | 9340 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9341 | { |
d9d05217 | 9342 | int ret = 0; |
90554200 JA |
9343 | DEFINE_WAIT(wait); |
9344 | ||
9345 | do { | |
9346 | if (!io_sqring_full(ctx)) | |
9347 | break; | |
9348 | ||
9349 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
9350 | ||
d9d05217 PB |
9351 | if (unlikely(ctx->sqo_dead)) { |
9352 | ret = -EOWNERDEAD; | |
9353 | goto out; | |
9354 | } | |
9355 | ||
90554200 JA |
9356 | if (!io_sqring_full(ctx)) |
9357 | break; | |
9358 | ||
9359 | schedule(); | |
9360 | } while (!signal_pending(current)); | |
9361 | ||
9362 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9363 | out: |
9364 | return ret; | |
90554200 JA |
9365 | } |
9366 | ||
c73ebb68 HX |
9367 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9368 | struct __kernel_timespec __user **ts, | |
9369 | const sigset_t __user **sig) | |
9370 | { | |
9371 | struct io_uring_getevents_arg arg; | |
9372 | ||
9373 | /* | |
9374 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9375 | * is just a pointer to the sigset_t. | |
9376 | */ | |
9377 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9378 | *sig = (const sigset_t __user *) argp; | |
9379 | *ts = NULL; | |
9380 | return 0; | |
9381 | } | |
9382 | ||
9383 | /* | |
9384 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9385 | * timespec and sigset_t pointers if good. | |
9386 | */ | |
9387 | if (*argsz != sizeof(arg)) | |
9388 | return -EINVAL; | |
9389 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9390 | return -EFAULT; | |
9391 | *sig = u64_to_user_ptr(arg.sigmask); | |
9392 | *argsz = arg.sigmask_sz; | |
9393 | *ts = u64_to_user_ptr(arg.ts); | |
9394 | return 0; | |
9395 | } | |
9396 | ||
2b188cc1 | 9397 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9398 | u32, min_complete, u32, flags, const void __user *, argp, |
9399 | size_t, argsz) | |
2b188cc1 JA |
9400 | { |
9401 | struct io_ring_ctx *ctx; | |
9402 | long ret = -EBADF; | |
9403 | int submitted = 0; | |
9404 | struct fd f; | |
9405 | ||
4c6e277c | 9406 | io_run_task_work(); |
b41e9852 | 9407 | |
90554200 | 9408 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9409 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9410 | return -EINVAL; |
9411 | ||
9412 | f = fdget(fd); | |
9413 | if (!f.file) | |
9414 | return -EBADF; | |
9415 | ||
9416 | ret = -EOPNOTSUPP; | |
9417 | if (f.file->f_op != &io_uring_fops) | |
9418 | goto out_fput; | |
9419 | ||
9420 | ret = -ENXIO; | |
9421 | ctx = f.file->private_data; | |
9422 | if (!percpu_ref_tryget(&ctx->refs)) | |
9423 | goto out_fput; | |
9424 | ||
7e84e1c7 SG |
9425 | ret = -EBADFD; |
9426 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9427 | goto out; | |
9428 | ||
6c271ce2 JA |
9429 | /* |
9430 | * For SQ polling, the thread will do all submissions and completions. | |
9431 | * Just return the requested submit count, and wake the thread if | |
9432 | * we were asked to. | |
9433 | */ | |
b2a9eada | 9434 | ret = 0; |
6c271ce2 | 9435 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9436 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9437 | |
d9d05217 PB |
9438 | ret = -EOWNERDEAD; |
9439 | if (unlikely(ctx->sqo_dead)) | |
9440 | goto out; | |
6c271ce2 | 9441 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9442 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9443 | if (flags & IORING_ENTER_SQ_WAIT) { |
9444 | ret = io_sqpoll_wait_sq(ctx); | |
9445 | if (ret) | |
9446 | goto out; | |
9447 | } | |
6c271ce2 | 9448 | submitted = to_submit; |
b2a9eada | 9449 | } else if (to_submit) { |
fdaf083c | 9450 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9451 | if (unlikely(ret)) |
9452 | goto out; | |
2b188cc1 | 9453 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9454 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9455 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9456 | |
9457 | if (submitted != to_submit) | |
9458 | goto out; | |
2b188cc1 JA |
9459 | } |
9460 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9461 | const sigset_t __user *sig; |
9462 | struct __kernel_timespec __user *ts; | |
9463 | ||
9464 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9465 | if (unlikely(ret)) | |
9466 | goto out; | |
9467 | ||
2b188cc1 JA |
9468 | min_complete = min(min_complete, ctx->cq_entries); |
9469 | ||
32b2244a XW |
9470 | /* |
9471 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9472 | * space applications don't need to do io completion events | |
9473 | * polling again, they can rely on io_sq_thread to do polling | |
9474 | * work, which can reduce cpu usage and uring_lock contention. | |
9475 | */ | |
9476 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9477 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9478 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9479 | } else { |
c73ebb68 | 9480 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9481 | } |
2b188cc1 JA |
9482 | } |
9483 | ||
7c504e65 | 9484 | out: |
6805b32e | 9485 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9486 | out_fput: |
9487 | fdput(f); | |
9488 | return submitted ? submitted : ret; | |
9489 | } | |
9490 | ||
bebdb65e | 9491 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9492 | static int io_uring_show_cred(int id, void *p, void *data) |
9493 | { | |
6b47ab81 JA |
9494 | struct io_identity *iod = p; |
9495 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9496 | struct seq_file *m = data; |
9497 | struct user_namespace *uns = seq_user_ns(m); | |
9498 | struct group_info *gi; | |
9499 | kernel_cap_t cap; | |
9500 | unsigned __capi; | |
9501 | int g; | |
9502 | ||
9503 | seq_printf(m, "%5d\n", id); | |
9504 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9505 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9506 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9507 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9508 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9509 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9510 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9511 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9512 | seq_puts(m, "\n\tGroups:\t"); | |
9513 | gi = cred->group_info; | |
9514 | for (g = 0; g < gi->ngroups; g++) { | |
9515 | seq_put_decimal_ull(m, g ? " " : "", | |
9516 | from_kgid_munged(uns, gi->gid[g])); | |
9517 | } | |
9518 | seq_puts(m, "\n\tCapEff:\t"); | |
9519 | cap = cred->cap_effective; | |
9520 | CAP_FOR_EACH_U32(__capi) | |
9521 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9522 | seq_putc(m, '\n'); | |
9523 | return 0; | |
9524 | } | |
9525 | ||
9526 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9527 | { | |
dbbe9c64 | 9528 | struct io_sq_data *sq = NULL; |
fad8e0de | 9529 | bool has_lock; |
87ce955b JA |
9530 | int i; |
9531 | ||
fad8e0de JA |
9532 | /* |
9533 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9534 | * since fdinfo case grabs it in the opposite direction of normal use | |
9535 | * cases. If we fail to get the lock, we just don't iterate any | |
9536 | * structures that could be going away outside the io_uring mutex. | |
9537 | */ | |
9538 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9539 | ||
dbbe9c64 JQ |
9540 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9541 | sq = ctx->sq_data; | |
9542 | ||
9543 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9544 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9545 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9546 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9547 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9548 | |
87ce955b JA |
9549 | if (f) |
9550 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9551 | else | |
9552 | seq_printf(m, "%5u: <none>\n", i); | |
9553 | } | |
9554 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9555 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9556 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9557 | ||
9558 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9559 | (unsigned int) buf->len); | |
9560 | } | |
fad8e0de | 9561 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9562 | seq_printf(m, "Personalities:\n"); |
9563 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9564 | } | |
d7718a9d JA |
9565 | seq_printf(m, "PollList:\n"); |
9566 | spin_lock_irq(&ctx->completion_lock); | |
9567 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9568 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9569 | struct io_kiocb *req; | |
9570 | ||
9571 | hlist_for_each_entry(req, list, hash_node) | |
9572 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9573 | req->task->task_works != NULL); | |
9574 | } | |
9575 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9576 | if (has_lock) |
9577 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9578 | } |
9579 | ||
9580 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9581 | { | |
9582 | struct io_ring_ctx *ctx = f->private_data; | |
9583 | ||
9584 | if (percpu_ref_tryget(&ctx->refs)) { | |
9585 | __io_uring_show_fdinfo(ctx, m); | |
9586 | percpu_ref_put(&ctx->refs); | |
9587 | } | |
9588 | } | |
bebdb65e | 9589 | #endif |
87ce955b | 9590 | |
2b188cc1 JA |
9591 | static const struct file_operations io_uring_fops = { |
9592 | .release = io_uring_release, | |
fcb323cc | 9593 | .flush = io_uring_flush, |
2b188cc1 | 9594 | .mmap = io_uring_mmap, |
6c5c240e RP |
9595 | #ifndef CONFIG_MMU |
9596 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9597 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9598 | #endif | |
2b188cc1 JA |
9599 | .poll = io_uring_poll, |
9600 | .fasync = io_uring_fasync, | |
bebdb65e | 9601 | #ifdef CONFIG_PROC_FS |
87ce955b | 9602 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9603 | #endif |
2b188cc1 JA |
9604 | }; |
9605 | ||
9606 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9607 | struct io_uring_params *p) | |
9608 | { | |
75b28aff HV |
9609 | struct io_rings *rings; |
9610 | size_t size, sq_array_offset; | |
2b188cc1 | 9611 | |
bd740481 JA |
9612 | /* make sure these are sane, as we already accounted them */ |
9613 | ctx->sq_entries = p->sq_entries; | |
9614 | ctx->cq_entries = p->cq_entries; | |
9615 | ||
75b28aff HV |
9616 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9617 | if (size == SIZE_MAX) | |
9618 | return -EOVERFLOW; | |
9619 | ||
9620 | rings = io_mem_alloc(size); | |
9621 | if (!rings) | |
2b188cc1 JA |
9622 | return -ENOMEM; |
9623 | ||
75b28aff HV |
9624 | ctx->rings = rings; |
9625 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9626 | rings->sq_ring_mask = p->sq_entries - 1; | |
9627 | rings->cq_ring_mask = p->cq_entries - 1; | |
9628 | rings->sq_ring_entries = p->sq_entries; | |
9629 | rings->cq_ring_entries = p->cq_entries; | |
9630 | ctx->sq_mask = rings->sq_ring_mask; | |
9631 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9632 | |
9633 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9634 | if (size == SIZE_MAX) { |
9635 | io_mem_free(ctx->rings); | |
9636 | ctx->rings = NULL; | |
2b188cc1 | 9637 | return -EOVERFLOW; |
eb065d30 | 9638 | } |
2b188cc1 JA |
9639 | |
9640 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9641 | if (!ctx->sq_sqes) { |
9642 | io_mem_free(ctx->rings); | |
9643 | ctx->rings = NULL; | |
2b188cc1 | 9644 | return -ENOMEM; |
eb065d30 | 9645 | } |
2b188cc1 | 9646 | |
2b188cc1 JA |
9647 | return 0; |
9648 | } | |
9649 | ||
9faadcc8 PB |
9650 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9651 | { | |
9652 | int ret, fd; | |
9653 | ||
9654 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9655 | if (fd < 0) | |
9656 | return fd; | |
9657 | ||
9658 | ret = io_uring_add_task_file(ctx, file); | |
9659 | if (ret) { | |
9660 | put_unused_fd(fd); | |
9661 | return ret; | |
9662 | } | |
9663 | fd_install(fd, file); | |
9664 | return fd; | |
9665 | } | |
9666 | ||
2b188cc1 JA |
9667 | /* |
9668 | * Allocate an anonymous fd, this is what constitutes the application | |
9669 | * visible backing of an io_uring instance. The application mmaps this | |
9670 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9671 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9672 | */ | |
9faadcc8 | 9673 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9674 | { |
9675 | struct file *file; | |
9faadcc8 | 9676 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9677 | int ret; |
9678 | ||
2b188cc1 JA |
9679 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9680 | &ctx->ring_sock); | |
9681 | if (ret) | |
9faadcc8 | 9682 | return ERR_PTR(ret); |
2b188cc1 JA |
9683 | #endif |
9684 | ||
2b188cc1 JA |
9685 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9686 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9687 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9688 | if (IS_ERR(file)) { |
9689 | sock_release(ctx->ring_sock); | |
9690 | ctx->ring_sock = NULL; | |
9691 | } else { | |
9692 | ctx->ring_sock->file = file; | |
0f212204 | 9693 | } |
2b188cc1 | 9694 | #endif |
9faadcc8 | 9695 | return file; |
2b188cc1 JA |
9696 | } |
9697 | ||
7f13657d XW |
9698 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9699 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9700 | { |
9701 | struct user_struct *user = NULL; | |
9702 | struct io_ring_ctx *ctx; | |
9faadcc8 | 9703 | struct file *file; |
2b188cc1 JA |
9704 | int ret; |
9705 | ||
8110c1a6 | 9706 | if (!entries) |
2b188cc1 | 9707 | return -EINVAL; |
8110c1a6 JA |
9708 | if (entries > IORING_MAX_ENTRIES) { |
9709 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9710 | return -EINVAL; | |
9711 | entries = IORING_MAX_ENTRIES; | |
9712 | } | |
2b188cc1 JA |
9713 | |
9714 | /* | |
9715 | * Use twice as many entries for the CQ ring. It's possible for the | |
9716 | * application to drive a higher depth than the size of the SQ ring, | |
9717 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9718 | * some flexibility in overcommitting a bit. If the application has |
9719 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9720 | * of CQ ring entries manually. | |
2b188cc1 JA |
9721 | */ |
9722 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9723 | if (p->flags & IORING_SETUP_CQSIZE) { |
9724 | /* | |
9725 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9726 | * to a power-of-two, if it isn't already. We do NOT impose | |
9727 | * any cq vs sq ring sizing. | |
9728 | */ | |
eb2667b3 | 9729 | if (!p->cq_entries) |
33a107f0 | 9730 | return -EINVAL; |
8110c1a6 JA |
9731 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9732 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9733 | return -EINVAL; | |
9734 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9735 | } | |
eb2667b3 JQ |
9736 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9737 | if (p->cq_entries < p->sq_entries) | |
9738 | return -EINVAL; | |
33a107f0 JA |
9739 | } else { |
9740 | p->cq_entries = 2 * p->sq_entries; | |
9741 | } | |
2b188cc1 JA |
9742 | |
9743 | user = get_uid(current_user()); | |
2b188cc1 JA |
9744 | |
9745 | ctx = io_ring_ctx_alloc(p); | |
9746 | if (!ctx) { | |
2b188cc1 JA |
9747 | free_uid(user); |
9748 | return -ENOMEM; | |
9749 | } | |
9750 | ctx->compat = in_compat_syscall(); | |
26bfa89e | 9751 | ctx->limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9752 | ctx->user = user; |
0b8c0ec7 | 9753 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9754 | #ifdef CONFIG_AUDIT |
9755 | ctx->loginuid = current->loginuid; | |
9756 | ctx->sessionid = current->sessionid; | |
9757 | #endif | |
2aede0e4 JA |
9758 | ctx->sqo_task = get_task_struct(current); |
9759 | ||
9760 | /* | |
9761 | * This is just grabbed for accounting purposes. When a process exits, | |
9762 | * the mm is exited and dropped before the files, hence we need to hang | |
9763 | * on to this mm purely for the purposes of being able to unaccount | |
9764 | * memory (locked/pinned vm). It's not used for anything else. | |
9765 | */ | |
6b7898eb | 9766 | mmgrab(current->mm); |
2aede0e4 | 9767 | ctx->mm_account = current->mm; |
6b7898eb | 9768 | |
91d8f519 DZ |
9769 | #ifdef CONFIG_BLK_CGROUP |
9770 | /* | |
9771 | * The sq thread will belong to the original cgroup it was inited in. | |
9772 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9773 | * issued bios will be associated with the closest cgroup later in the | |
9774 | * block layer. | |
9775 | */ | |
9776 | rcu_read_lock(); | |
9777 | ctx->sqo_blkcg_css = blkcg_css(); | |
9778 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9779 | rcu_read_unlock(); | |
9780 | if (!ret) { | |
9781 | /* don't init against a dying cgroup, have the user try again */ | |
9782 | ctx->sqo_blkcg_css = NULL; | |
9783 | ret = -ENODEV; | |
9784 | goto err; | |
9785 | } | |
9786 | #endif | |
2b188cc1 JA |
9787 | ret = io_allocate_scq_urings(ctx, p); |
9788 | if (ret) | |
9789 | goto err; | |
9790 | ||
7e84e1c7 | 9791 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9792 | if (ret) |
9793 | goto err; | |
9794 | ||
7e84e1c7 SG |
9795 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9796 | io_sq_offload_start(ctx); | |
9797 | ||
2b188cc1 | 9798 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9799 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9800 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9801 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9802 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9803 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9804 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9805 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9806 | |
9807 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9808 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9809 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9810 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9811 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9812 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9813 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9814 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9815 | |
7f13657d XW |
9816 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9817 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9818 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 HX |
9819 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9820 | IORING_FEAT_EXT_ARG; | |
7f13657d XW |
9821 | |
9822 | if (copy_to_user(params, p, sizeof(*p))) { | |
9823 | ret = -EFAULT; | |
9824 | goto err; | |
9825 | } | |
d1719f70 | 9826 | |
9faadcc8 PB |
9827 | file = io_uring_get_file(ctx); |
9828 | if (IS_ERR(file)) { | |
9829 | ret = PTR_ERR(file); | |
9830 | goto err; | |
9831 | } | |
9832 | ||
044c1ab3 JA |
9833 | /* |
9834 | * Install ring fd as the very last thing, so we don't risk someone | |
9835 | * having closed it before we finish setup | |
9836 | */ | |
9faadcc8 PB |
9837 | ret = io_uring_install_fd(ctx, file); |
9838 | if (ret < 0) { | |
06585c49 | 9839 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9840 | /* fput will clean it up */ |
9841 | fput(file); | |
9842 | return ret; | |
9843 | } | |
044c1ab3 | 9844 | |
c826bd7a | 9845 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9846 | return ret; |
9847 | err: | |
d9d05217 | 9848 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9849 | io_ring_ctx_wait_and_kill(ctx); |
9850 | return ret; | |
9851 | } | |
9852 | ||
9853 | /* | |
9854 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9855 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9856 | * params structure passed in. | |
9857 | */ | |
9858 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9859 | { | |
9860 | struct io_uring_params p; | |
2b188cc1 JA |
9861 | int i; |
9862 | ||
9863 | if (copy_from_user(&p, params, sizeof(p))) | |
9864 | return -EFAULT; | |
9865 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9866 | if (p.resv[i]) | |
9867 | return -EINVAL; | |
9868 | } | |
9869 | ||
6c271ce2 | 9870 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9871 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9872 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9873 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9874 | return -EINVAL; |
9875 | ||
7f13657d | 9876 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9877 | } |
9878 | ||
9879 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9880 | struct io_uring_params __user *, params) | |
9881 | { | |
9882 | return io_uring_setup(entries, params); | |
9883 | } | |
9884 | ||
66f4af93 JA |
9885 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9886 | { | |
9887 | struct io_uring_probe *p; | |
9888 | size_t size; | |
9889 | int i, ret; | |
9890 | ||
9891 | size = struct_size(p, ops, nr_args); | |
9892 | if (size == SIZE_MAX) | |
9893 | return -EOVERFLOW; | |
9894 | p = kzalloc(size, GFP_KERNEL); | |
9895 | if (!p) | |
9896 | return -ENOMEM; | |
9897 | ||
9898 | ret = -EFAULT; | |
9899 | if (copy_from_user(p, arg, size)) | |
9900 | goto out; | |
9901 | ret = -EINVAL; | |
9902 | if (memchr_inv(p, 0, size)) | |
9903 | goto out; | |
9904 | ||
9905 | p->last_op = IORING_OP_LAST - 1; | |
9906 | if (nr_args > IORING_OP_LAST) | |
9907 | nr_args = IORING_OP_LAST; | |
9908 | ||
9909 | for (i = 0; i < nr_args; i++) { | |
9910 | p->ops[i].op = i; | |
9911 | if (!io_op_defs[i].not_supported) | |
9912 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9913 | } | |
9914 | p->ops_len = i; | |
9915 | ||
9916 | ret = 0; | |
9917 | if (copy_to_user(arg, p, size)) | |
9918 | ret = -EFAULT; | |
9919 | out: | |
9920 | kfree(p); | |
9921 | return ret; | |
9922 | } | |
9923 | ||
071698e1 JA |
9924 | static int io_register_personality(struct io_ring_ctx *ctx) |
9925 | { | |
1e6fa521 JA |
9926 | struct io_identity *id; |
9927 | int ret; | |
071698e1 | 9928 | |
1e6fa521 JA |
9929 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9930 | if (unlikely(!id)) | |
9931 | return -ENOMEM; | |
9932 | ||
9933 | io_init_identity(id); | |
9934 | id->creds = get_current_cred(); | |
9935 | ||
9936 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9937 | if (ret < 0) { | |
9938 | put_cred(id->creds); | |
9939 | kfree(id); | |
9940 | } | |
9941 | return ret; | |
071698e1 JA |
9942 | } |
9943 | ||
21b55dbc SG |
9944 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9945 | unsigned int nr_args) | |
9946 | { | |
9947 | struct io_uring_restriction *res; | |
9948 | size_t size; | |
9949 | int i, ret; | |
9950 | ||
7e84e1c7 SG |
9951 | /* Restrictions allowed only if rings started disabled */ |
9952 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9953 | return -EBADFD; | |
9954 | ||
21b55dbc | 9955 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9956 | if (ctx->restrictions.registered) |
21b55dbc SG |
9957 | return -EBUSY; |
9958 | ||
9959 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9960 | return -EINVAL; | |
9961 | ||
9962 | size = array_size(nr_args, sizeof(*res)); | |
9963 | if (size == SIZE_MAX) | |
9964 | return -EOVERFLOW; | |
9965 | ||
9966 | res = memdup_user(arg, size); | |
9967 | if (IS_ERR(res)) | |
9968 | return PTR_ERR(res); | |
9969 | ||
9970 | ret = 0; | |
9971 | ||
9972 | for (i = 0; i < nr_args; i++) { | |
9973 | switch (res[i].opcode) { | |
9974 | case IORING_RESTRICTION_REGISTER_OP: | |
9975 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9976 | ret = -EINVAL; | |
9977 | goto out; | |
9978 | } | |
9979 | ||
9980 | __set_bit(res[i].register_op, | |
9981 | ctx->restrictions.register_op); | |
9982 | break; | |
9983 | case IORING_RESTRICTION_SQE_OP: | |
9984 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9985 | ret = -EINVAL; | |
9986 | goto out; | |
9987 | } | |
9988 | ||
9989 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9990 | break; | |
9991 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9992 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9993 | break; | |
9994 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9995 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9996 | break; | |
9997 | default: | |
9998 | ret = -EINVAL; | |
9999 | goto out; | |
10000 | } | |
10001 | } | |
10002 | ||
10003 | out: | |
10004 | /* Reset all restrictions if an error happened */ | |
10005 | if (ret != 0) | |
10006 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10007 | else | |
7e84e1c7 | 10008 | ctx->restrictions.registered = true; |
21b55dbc SG |
10009 | |
10010 | kfree(res); | |
10011 | return ret; | |
10012 | } | |
10013 | ||
7e84e1c7 SG |
10014 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10015 | { | |
10016 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10017 | return -EBADFD; | |
10018 | ||
10019 | if (ctx->restrictions.registered) | |
10020 | ctx->restricted = 1; | |
10021 | ||
10022 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
10023 | ||
10024 | io_sq_offload_start(ctx); | |
10025 | ||
10026 | return 0; | |
10027 | } | |
10028 | ||
071698e1 JA |
10029 | static bool io_register_op_must_quiesce(int op) |
10030 | { | |
10031 | switch (op) { | |
10032 | case IORING_UNREGISTER_FILES: | |
10033 | case IORING_REGISTER_FILES_UPDATE: | |
10034 | case IORING_REGISTER_PROBE: | |
10035 | case IORING_REGISTER_PERSONALITY: | |
10036 | case IORING_UNREGISTER_PERSONALITY: | |
10037 | return false; | |
10038 | default: | |
10039 | return true; | |
10040 | } | |
10041 | } | |
10042 | ||
edafccee JA |
10043 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10044 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10045 | __releases(ctx->uring_lock) |
10046 | __acquires(ctx->uring_lock) | |
edafccee JA |
10047 | { |
10048 | int ret; | |
10049 | ||
35fa71a0 JA |
10050 | /* |
10051 | * We're inside the ring mutex, if the ref is already dying, then | |
10052 | * someone else killed the ctx or is already going through | |
10053 | * io_uring_register(). | |
10054 | */ | |
10055 | if (percpu_ref_is_dying(&ctx->refs)) | |
10056 | return -ENXIO; | |
10057 | ||
071698e1 | 10058 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10059 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10060 | |
05f3fb3c JA |
10061 | /* |
10062 | * Drop uring mutex before waiting for references to exit. If | |
10063 | * another thread is currently inside io_uring_enter() it might | |
10064 | * need to grab the uring_lock to make progress. If we hold it | |
10065 | * here across the drain wait, then we can deadlock. It's safe | |
10066 | * to drop the mutex here, since no new references will come in | |
10067 | * after we've killed the percpu ref. | |
10068 | */ | |
10069 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10070 | do { |
10071 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10072 | if (!ret) | |
10073 | break; | |
ed6930c9 JA |
10074 | ret = io_run_task_work_sig(); |
10075 | if (ret < 0) | |
10076 | break; | |
af9c1a44 JA |
10077 | } while (1); |
10078 | ||
05f3fb3c | 10079 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10080 | |
c150368b JA |
10081 | if (ret) { |
10082 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
10083 | goto out_quiesce; |
10084 | } | |
10085 | } | |
10086 | ||
10087 | if (ctx->restricted) { | |
10088 | if (opcode >= IORING_REGISTER_LAST) { | |
10089 | ret = -EINVAL; | |
10090 | goto out; | |
10091 | } | |
10092 | ||
10093 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
10094 | ret = -EACCES; | |
c150368b JA |
10095 | goto out; |
10096 | } | |
05f3fb3c | 10097 | } |
edafccee JA |
10098 | |
10099 | switch (opcode) { | |
10100 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 10101 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
10102 | break; |
10103 | case IORING_UNREGISTER_BUFFERS: | |
10104 | ret = -EINVAL; | |
10105 | if (arg || nr_args) | |
10106 | break; | |
0a96bbe4 | 10107 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10108 | break; |
6b06314c JA |
10109 | case IORING_REGISTER_FILES: |
10110 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
10111 | break; | |
10112 | case IORING_UNREGISTER_FILES: | |
10113 | ret = -EINVAL; | |
10114 | if (arg || nr_args) | |
10115 | break; | |
10116 | ret = io_sqe_files_unregister(ctx); | |
10117 | break; | |
c3a31e60 JA |
10118 | case IORING_REGISTER_FILES_UPDATE: |
10119 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
10120 | break; | |
9b402849 | 10121 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10122 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10123 | ret = -EINVAL; |
10124 | if (nr_args != 1) | |
10125 | break; | |
10126 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10127 | if (ret) |
10128 | break; | |
10129 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10130 | ctx->eventfd_async = 1; | |
10131 | else | |
10132 | ctx->eventfd_async = 0; | |
9b402849 JA |
10133 | break; |
10134 | case IORING_UNREGISTER_EVENTFD: | |
10135 | ret = -EINVAL; | |
10136 | if (arg || nr_args) | |
10137 | break; | |
10138 | ret = io_eventfd_unregister(ctx); | |
10139 | break; | |
66f4af93 JA |
10140 | case IORING_REGISTER_PROBE: |
10141 | ret = -EINVAL; | |
10142 | if (!arg || nr_args > 256) | |
10143 | break; | |
10144 | ret = io_probe(ctx, arg, nr_args); | |
10145 | break; | |
071698e1 JA |
10146 | case IORING_REGISTER_PERSONALITY: |
10147 | ret = -EINVAL; | |
10148 | if (arg || nr_args) | |
10149 | break; | |
10150 | ret = io_register_personality(ctx); | |
10151 | break; | |
10152 | case IORING_UNREGISTER_PERSONALITY: | |
10153 | ret = -EINVAL; | |
10154 | if (arg) | |
10155 | break; | |
10156 | ret = io_unregister_personality(ctx, nr_args); | |
10157 | break; | |
7e84e1c7 SG |
10158 | case IORING_REGISTER_ENABLE_RINGS: |
10159 | ret = -EINVAL; | |
10160 | if (arg || nr_args) | |
10161 | break; | |
10162 | ret = io_register_enable_rings(ctx); | |
10163 | break; | |
21b55dbc SG |
10164 | case IORING_REGISTER_RESTRICTIONS: |
10165 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10166 | break; | |
edafccee JA |
10167 | default: |
10168 | ret = -EINVAL; | |
10169 | break; | |
10170 | } | |
10171 | ||
21b55dbc | 10172 | out: |
071698e1 | 10173 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10174 | /* bring the ctx back to life */ |
05f3fb3c | 10175 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 10176 | out_quiesce: |
0f158b4c | 10177 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10178 | } |
edafccee JA |
10179 | return ret; |
10180 | } | |
10181 | ||
10182 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10183 | void __user *, arg, unsigned int, nr_args) | |
10184 | { | |
10185 | struct io_ring_ctx *ctx; | |
10186 | long ret = -EBADF; | |
10187 | struct fd f; | |
10188 | ||
10189 | f = fdget(fd); | |
10190 | if (!f.file) | |
10191 | return -EBADF; | |
10192 | ||
10193 | ret = -EOPNOTSUPP; | |
10194 | if (f.file->f_op != &io_uring_fops) | |
10195 | goto out_fput; | |
10196 | ||
10197 | ctx = f.file->private_data; | |
10198 | ||
10199 | mutex_lock(&ctx->uring_lock); | |
10200 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10201 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10202 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10203 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10204 | out_fput: |
10205 | fdput(f); | |
10206 | return ret; | |
10207 | } | |
10208 | ||
2b188cc1 JA |
10209 | static int __init io_uring_init(void) |
10210 | { | |
d7f62e82 SM |
10211 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10212 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10213 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10214 | } while (0) | |
10215 | ||
10216 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10217 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10218 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10219 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10220 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10221 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10222 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10223 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10224 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10225 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10226 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10227 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10228 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10229 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10230 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10231 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10232 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10233 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10234 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10235 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10236 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10237 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10238 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10239 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10240 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10241 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10242 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10243 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10244 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10245 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10246 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10247 | |
d3656344 | 10248 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10249 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10250 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10251 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10252 | return 0; |
10253 | }; | |
10254 | __initcall(io_uring_init); |