]>
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 | 238 | struct completion done; |
8bad28d8 | 239 | bool quiesce; |
05f3fb3c JA |
240 | }; |
241 | ||
5a2e745d JA |
242 | struct io_buffer { |
243 | struct list_head list; | |
244 | __u64 addr; | |
245 | __s32 len; | |
246 | __u16 bid; | |
247 | }; | |
248 | ||
21b55dbc SG |
249 | struct io_restriction { |
250 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
251 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
252 | u8 sqe_flags_allowed; | |
253 | u8 sqe_flags_required; | |
7e84e1c7 | 254 | bool registered; |
21b55dbc SG |
255 | }; |
256 | ||
534ca6d6 JA |
257 | struct io_sq_data { |
258 | refcount_t refs; | |
69fb2131 JA |
259 | struct mutex lock; |
260 | ||
261 | /* ctx's that are using this sqd */ | |
262 | struct list_head ctx_list; | |
263 | struct list_head ctx_new_list; | |
264 | struct mutex ctx_lock; | |
265 | ||
534ca6d6 JA |
266 | struct task_struct *thread; |
267 | struct wait_queue_head wait; | |
08369246 XW |
268 | |
269 | unsigned sq_thread_idle; | |
534ca6d6 JA |
270 | }; |
271 | ||
258b29a9 | 272 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 273 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 274 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 275 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
276 | |
277 | struct io_comp_state { | |
6dd0be1e | 278 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 279 | unsigned int nr; |
c7dae4ba JA |
280 | unsigned int locked_free_nr; |
281 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 282 | struct list_head free_list; |
c7dae4ba JA |
283 | /* IRQ completion list, under ->completion_lock */ |
284 | struct list_head locked_free_list; | |
258b29a9 PB |
285 | }; |
286 | ||
a1ab7b35 PB |
287 | struct io_submit_link { |
288 | struct io_kiocb *head; | |
289 | struct io_kiocb *last; | |
290 | }; | |
291 | ||
258b29a9 PB |
292 | struct io_submit_state { |
293 | struct blk_plug plug; | |
a1ab7b35 | 294 | struct io_submit_link link; |
258b29a9 PB |
295 | |
296 | /* | |
297 | * io_kiocb alloc cache | |
298 | */ | |
bf019da7 | 299 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
300 | unsigned int free_reqs; |
301 | ||
302 | bool plug_started; | |
303 | ||
304 | /* | |
305 | * Batch completion logic | |
306 | */ | |
307 | struct io_comp_state comp; | |
308 | ||
309 | /* | |
310 | * File reference cache | |
311 | */ | |
312 | struct file *file; | |
313 | unsigned int fd; | |
314 | unsigned int file_refs; | |
315 | unsigned int ios_left; | |
316 | }; | |
317 | ||
2b188cc1 JA |
318 | struct io_ring_ctx { |
319 | struct { | |
320 | struct percpu_ref refs; | |
321 | } ____cacheline_aligned_in_smp; | |
322 | ||
323 | struct { | |
324 | unsigned int flags; | |
e1d85334 | 325 | unsigned int compat: 1; |
aad5d8da | 326 | unsigned int limit_mem: 1; |
e1d85334 RD |
327 | unsigned int cq_overflow_flushed: 1; |
328 | unsigned int drain_next: 1; | |
329 | unsigned int eventfd_async: 1; | |
21b55dbc | 330 | unsigned int restricted: 1; |
d9d05217 | 331 | unsigned int sqo_dead: 1; |
2b188cc1 | 332 | |
75b28aff HV |
333 | /* |
334 | * Ring buffer of indices into array of io_uring_sqe, which is | |
335 | * mmapped by the application using the IORING_OFF_SQES offset. | |
336 | * | |
337 | * This indirection could e.g. be used to assign fixed | |
338 | * io_uring_sqe entries to operations and only submit them to | |
339 | * the queue when needed. | |
340 | * | |
341 | * The kernel modifies neither the indices array nor the entries | |
342 | * array. | |
343 | */ | |
344 | u32 *sq_array; | |
2b188cc1 JA |
345 | unsigned cached_sq_head; |
346 | unsigned sq_entries; | |
347 | unsigned sq_mask; | |
6c271ce2 | 348 | unsigned sq_thread_idle; |
498ccd9e | 349 | unsigned cached_sq_dropped; |
2c3bac6d | 350 | unsigned cached_cq_overflow; |
ad3eb2c8 | 351 | unsigned long sq_check_overflow; |
de0617e4 JA |
352 | |
353 | struct list_head defer_list; | |
5262f567 | 354 | struct list_head timeout_list; |
1d7bb1d5 | 355 | struct list_head cq_overflow_list; |
fcb323cc | 356 | |
ad3eb2c8 | 357 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
358 | } ____cacheline_aligned_in_smp; |
359 | ||
3c1a2ead JA |
360 | struct { |
361 | struct mutex uring_lock; | |
362 | wait_queue_head_t wait; | |
363 | } ____cacheline_aligned_in_smp; | |
364 | ||
365 | struct io_submit_state submit_state; | |
366 | ||
206aefde JA |
367 | struct io_rings *rings; |
368 | ||
2aede0e4 JA |
369 | /* |
370 | * For SQPOLL usage - we hold a reference to the parent task, so we | |
371 | * have access to the ->files | |
372 | */ | |
373 | struct task_struct *sqo_task; | |
374 | ||
375 | /* Only used for accounting purposes */ | |
376 | struct mm_struct *mm_account; | |
377 | ||
91d8f519 DZ |
378 | #ifdef CONFIG_BLK_CGROUP |
379 | struct cgroup_subsys_state *sqo_blkcg_css; | |
380 | #endif | |
381 | ||
534ca6d6 JA |
382 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
383 | ||
90554200 | 384 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 385 | struct list_head sqd_list; |
75b28aff | 386 | |
6b06314c JA |
387 | /* |
388 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
389 | * readers must ensure that ->refs is alive as long as the file* is | |
390 | * used. Only updated through io_uring_register(2). | |
391 | */ | |
269bbe5f | 392 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
393 | unsigned nr_user_files; |
394 | ||
edafccee JA |
395 | /* if used, fixed mapped user buffers */ |
396 | unsigned nr_user_bufs; | |
397 | struct io_mapped_ubuf *user_bufs; | |
398 | ||
2b188cc1 JA |
399 | struct user_struct *user; |
400 | ||
0b8c0ec7 | 401 | const struct cred *creds; |
181e448d | 402 | |
4ea33a97 JA |
403 | #ifdef CONFIG_AUDIT |
404 | kuid_t loginuid; | |
405 | unsigned int sessionid; | |
406 | #endif | |
407 | ||
0f158b4c JA |
408 | struct completion ref_comp; |
409 | struct completion sq_thread_comp; | |
206aefde JA |
410 | |
411 | #if defined(CONFIG_UNIX) | |
412 | struct socket *ring_sock; | |
413 | #endif | |
414 | ||
5a2e745d JA |
415 | struct idr io_buffer_idr; |
416 | ||
071698e1 JA |
417 | struct idr personality_idr; |
418 | ||
206aefde JA |
419 | struct { |
420 | unsigned cached_cq_tail; | |
421 | unsigned cq_entries; | |
422 | unsigned cq_mask; | |
423 | atomic_t cq_timeouts; | |
f010505b | 424 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 425 | unsigned long cq_check_overflow; |
206aefde JA |
426 | struct wait_queue_head cq_wait; |
427 | struct fasync_struct *cq_fasync; | |
428 | struct eventfd_ctx *cq_ev_fd; | |
429 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 430 | |
2b188cc1 JA |
431 | struct { |
432 | spinlock_t completion_lock; | |
e94f141b | 433 | |
def596e9 | 434 | /* |
540e32a0 | 435 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
436 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
437 | * For SQPOLL, only the single threaded io_sq_thread() will | |
438 | * manipulate the list, hence no extra locking is needed there. | |
439 | */ | |
540e32a0 | 440 | struct list_head iopoll_list; |
78076bb6 JA |
441 | struct hlist_head *cancel_hash; |
442 | unsigned cancel_hash_bits; | |
e94f141b | 443 | bool poll_multi_file; |
31b51510 | 444 | |
fcb323cc JA |
445 | spinlock_t inflight_lock; |
446 | struct list_head inflight_list; | |
2b188cc1 | 447 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 448 | |
269bbe5f BM |
449 | struct delayed_work rsrc_put_work; |
450 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
451 | struct list_head rsrc_ref_list; |
452 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 453 | |
21b55dbc | 454 | struct io_restriction restrictions; |
3c1a2ead | 455 | |
7c25c0d1 JA |
456 | /* exit task_work */ |
457 | struct callback_head *exit_task_work; | |
458 | ||
3c1a2ead JA |
459 | /* Keep this last, we don't need it for the fast path */ |
460 | struct work_struct exit_work; | |
2b188cc1 JA |
461 | }; |
462 | ||
09bb8394 JA |
463 | /* |
464 | * First field must be the file pointer in all the | |
465 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
466 | */ | |
221c5eb2 JA |
467 | struct io_poll_iocb { |
468 | struct file *file; | |
018043be | 469 | struct wait_queue_head *head; |
221c5eb2 | 470 | __poll_t events; |
8c838788 | 471 | bool done; |
221c5eb2 | 472 | bool canceled; |
392edb45 | 473 | struct wait_queue_entry wait; |
221c5eb2 JA |
474 | }; |
475 | ||
018043be PB |
476 | struct io_poll_remove { |
477 | struct file *file; | |
478 | u64 addr; | |
479 | }; | |
480 | ||
b5dba59e JA |
481 | struct io_close { |
482 | struct file *file; | |
b5dba59e JA |
483 | int fd; |
484 | }; | |
485 | ||
ad8a48ac JA |
486 | struct io_timeout_data { |
487 | struct io_kiocb *req; | |
488 | struct hrtimer timer; | |
489 | struct timespec64 ts; | |
490 | enum hrtimer_mode mode; | |
491 | }; | |
492 | ||
8ed8d3c3 JA |
493 | struct io_accept { |
494 | struct file *file; | |
495 | struct sockaddr __user *addr; | |
496 | int __user *addr_len; | |
497 | int flags; | |
09952e3e | 498 | unsigned long nofile; |
8ed8d3c3 JA |
499 | }; |
500 | ||
501 | struct io_sync { | |
502 | struct file *file; | |
503 | loff_t len; | |
504 | loff_t off; | |
505 | int flags; | |
d63d1b5e | 506 | int mode; |
8ed8d3c3 JA |
507 | }; |
508 | ||
fbf23849 JA |
509 | struct io_cancel { |
510 | struct file *file; | |
511 | u64 addr; | |
512 | }; | |
513 | ||
b29472ee JA |
514 | struct io_timeout { |
515 | struct file *file; | |
bfe68a22 PB |
516 | u32 off; |
517 | u32 target_seq; | |
135fcde8 | 518 | struct list_head list; |
90cd7e42 PB |
519 | /* head of the link, used by linked timeouts only */ |
520 | struct io_kiocb *head; | |
b29472ee JA |
521 | }; |
522 | ||
0bdf7a2d PB |
523 | struct io_timeout_rem { |
524 | struct file *file; | |
525 | u64 addr; | |
9c8e11b3 PB |
526 | |
527 | /* timeout update */ | |
528 | struct timespec64 ts; | |
529 | u32 flags; | |
0bdf7a2d PB |
530 | }; |
531 | ||
9adbd45d JA |
532 | struct io_rw { |
533 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
534 | struct kiocb kiocb; | |
535 | u64 addr; | |
536 | u64 len; | |
537 | }; | |
538 | ||
3fbb51c1 JA |
539 | struct io_connect { |
540 | struct file *file; | |
541 | struct sockaddr __user *addr; | |
542 | int addr_len; | |
543 | }; | |
544 | ||
e47293fd JA |
545 | struct io_sr_msg { |
546 | struct file *file; | |
fddaface | 547 | union { |
270a5940 | 548 | struct user_msghdr __user *umsg; |
fddaface JA |
549 | void __user *buf; |
550 | }; | |
e47293fd | 551 | int msg_flags; |
bcda7baa | 552 | int bgid; |
fddaface | 553 | size_t len; |
bcda7baa | 554 | struct io_buffer *kbuf; |
e47293fd JA |
555 | }; |
556 | ||
15b71abe JA |
557 | struct io_open { |
558 | struct file *file; | |
559 | int dfd; | |
15b71abe | 560 | struct filename *filename; |
c12cedf2 | 561 | struct open_how how; |
4022e7af | 562 | unsigned long nofile; |
15b71abe JA |
563 | }; |
564 | ||
269bbe5f | 565 | struct io_rsrc_update { |
05f3fb3c JA |
566 | struct file *file; |
567 | u64 arg; | |
568 | u32 nr_args; | |
569 | u32 offset; | |
570 | }; | |
571 | ||
4840e418 JA |
572 | struct io_fadvise { |
573 | struct file *file; | |
574 | u64 offset; | |
575 | u32 len; | |
576 | u32 advice; | |
577 | }; | |
578 | ||
c1ca757b JA |
579 | struct io_madvise { |
580 | struct file *file; | |
581 | u64 addr; | |
582 | u32 len; | |
583 | u32 advice; | |
584 | }; | |
585 | ||
3e4827b0 JA |
586 | struct io_epoll { |
587 | struct file *file; | |
588 | int epfd; | |
589 | int op; | |
590 | int fd; | |
591 | struct epoll_event event; | |
e47293fd JA |
592 | }; |
593 | ||
7d67af2c PB |
594 | struct io_splice { |
595 | struct file *file_out; | |
596 | struct file *file_in; | |
597 | loff_t off_out; | |
598 | loff_t off_in; | |
599 | u64 len; | |
600 | unsigned int flags; | |
601 | }; | |
602 | ||
ddf0322d JA |
603 | struct io_provide_buf { |
604 | struct file *file; | |
605 | __u64 addr; | |
606 | __s32 len; | |
607 | __u32 bgid; | |
608 | __u16 nbufs; | |
609 | __u16 bid; | |
610 | }; | |
611 | ||
1d9e1288 BM |
612 | struct io_statx { |
613 | struct file *file; | |
614 | int dfd; | |
615 | unsigned int mask; | |
616 | unsigned int flags; | |
e62753e4 | 617 | const char __user *filename; |
1d9e1288 BM |
618 | struct statx __user *buffer; |
619 | }; | |
620 | ||
36f4fa68 JA |
621 | struct io_shutdown { |
622 | struct file *file; | |
623 | int how; | |
624 | }; | |
625 | ||
80a261fd JA |
626 | struct io_rename { |
627 | struct file *file; | |
628 | int old_dfd; | |
629 | int new_dfd; | |
630 | struct filename *oldpath; | |
631 | struct filename *newpath; | |
632 | int flags; | |
633 | }; | |
634 | ||
14a1143b JA |
635 | struct io_unlink { |
636 | struct file *file; | |
637 | int dfd; | |
638 | int flags; | |
639 | struct filename *filename; | |
640 | }; | |
641 | ||
3ca405eb PB |
642 | struct io_completion { |
643 | struct file *file; | |
644 | struct list_head list; | |
0f7e466b | 645 | int cflags; |
3ca405eb PB |
646 | }; |
647 | ||
f499a021 JA |
648 | struct io_async_connect { |
649 | struct sockaddr_storage address; | |
650 | }; | |
651 | ||
03b1230c JA |
652 | struct io_async_msghdr { |
653 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
654 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
655 | struct iovec *free_iov; | |
03b1230c JA |
656 | struct sockaddr __user *uaddr; |
657 | struct msghdr msg; | |
b537916c | 658 | struct sockaddr_storage addr; |
03b1230c JA |
659 | }; |
660 | ||
f67676d1 JA |
661 | struct io_async_rw { |
662 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
663 | const struct iovec *free_iovec; |
664 | struct iov_iter iter; | |
227c0c96 | 665 | size_t bytes_done; |
bcf5a063 | 666 | struct wait_page_queue wpq; |
f67676d1 JA |
667 | }; |
668 | ||
6b47ee6e PB |
669 | enum { |
670 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
671 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
672 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
673 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
674 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 675 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 676 | |
6b47ee6e PB |
677 | REQ_F_FAIL_LINK_BIT, |
678 | REQ_F_INFLIGHT_BIT, | |
679 | REQ_F_CUR_POS_BIT, | |
680 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 681 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 682 | REQ_F_ISREG_BIT, |
99bc4c38 | 683 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 684 | REQ_F_POLLED_BIT, |
bcda7baa | 685 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 686 | REQ_F_NO_FILE_TABLE_BIT, |
7cdaf587 | 687 | REQ_F_WORK_INITIALIZED_BIT, |
900fad45 | 688 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 689 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
690 | |
691 | /* not a real bit, just to check we're not overflowing the space */ | |
692 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
693 | }; |
694 | ||
695 | enum { | |
696 | /* ctx owns file */ | |
697 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
698 | /* drain existing IO first */ | |
699 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
700 | /* linked sqes */ | |
701 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
702 | /* doesn't sever on completion < 0 */ | |
703 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
704 | /* IOSQE_ASYNC */ | |
705 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
706 | /* IOSQE_BUFFER_SELECT */ |
707 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 708 | |
6b47ee6e PB |
709 | /* fail rest of links */ |
710 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
711 | /* on inflight list */ | |
712 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), | |
713 | /* read/write uses file position */ | |
714 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
715 | /* must not punt to workers */ | |
716 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 717 | /* has or had linked timeout */ |
6b47ee6e | 718 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
719 | /* regular file */ |
720 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
721 | /* needs cleanup */ |
722 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
723 | /* already went through poll handler */ |
724 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
725 | /* buffer already selected */ |
726 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
727 | /* doesn't need file table for this request */ |
728 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
7cdaf587 XW |
729 | /* io_wq_work is initialized */ |
730 | REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT), | |
900fad45 PB |
731 | /* linked timeout is active, i.e. prepared by link's head */ |
732 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
733 | /* completion is deferred through io_comp_state */ |
734 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
735 | }; |
736 | ||
737 | struct async_poll { | |
738 | struct io_poll_iocb poll; | |
807abcb0 | 739 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
740 | }; |
741 | ||
7cbf1722 JA |
742 | struct io_task_work { |
743 | struct io_wq_work_node node; | |
744 | task_work_func_t func; | |
745 | }; | |
746 | ||
09bb8394 JA |
747 | /* |
748 | * NOTE! Each of the iocb union members has the file pointer | |
749 | * as the first entry in their struct definition. So you can | |
750 | * access the file pointer through any of the sub-structs, | |
751 | * or directly as just 'ki_filp' in this struct. | |
752 | */ | |
2b188cc1 | 753 | struct io_kiocb { |
221c5eb2 | 754 | union { |
09bb8394 | 755 | struct file *file; |
9adbd45d | 756 | struct io_rw rw; |
221c5eb2 | 757 | struct io_poll_iocb poll; |
018043be | 758 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
759 | struct io_accept accept; |
760 | struct io_sync sync; | |
fbf23849 | 761 | struct io_cancel cancel; |
b29472ee | 762 | struct io_timeout timeout; |
0bdf7a2d | 763 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 764 | struct io_connect connect; |
e47293fd | 765 | struct io_sr_msg sr_msg; |
15b71abe | 766 | struct io_open open; |
b5dba59e | 767 | struct io_close close; |
269bbe5f | 768 | struct io_rsrc_update rsrc_update; |
4840e418 | 769 | struct io_fadvise fadvise; |
c1ca757b | 770 | struct io_madvise madvise; |
3e4827b0 | 771 | struct io_epoll epoll; |
7d67af2c | 772 | struct io_splice splice; |
ddf0322d | 773 | struct io_provide_buf pbuf; |
1d9e1288 | 774 | struct io_statx statx; |
36f4fa68 | 775 | struct io_shutdown shutdown; |
80a261fd | 776 | struct io_rename rename; |
14a1143b | 777 | struct io_unlink unlink; |
3ca405eb PB |
778 | /* use only after cleaning per-op data, see io_clean_op() */ |
779 | struct io_completion compl; | |
221c5eb2 | 780 | }; |
2b188cc1 | 781 | |
e8c2bc1f JA |
782 | /* opcode allocated if it needs to store data for async defer */ |
783 | void *async_data; | |
d625c6ee | 784 | u8 opcode; |
65a6543d XW |
785 | /* polled IO has completed */ |
786 | u8 iopoll_completed; | |
2b188cc1 | 787 | |
4f4eeba8 | 788 | u16 buf_index; |
9cf7c104 | 789 | u32 result; |
4f4eeba8 | 790 | |
010e8e6b PB |
791 | struct io_ring_ctx *ctx; |
792 | unsigned int flags; | |
793 | refcount_t refs; | |
794 | struct task_struct *task; | |
795 | u64 user_data; | |
d7718a9d | 796 | |
f2f87370 | 797 | struct io_kiocb *link; |
269bbe5f | 798 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 799 | |
d21ffe7e PB |
800 | /* |
801 | * 1. used with ctx->iopoll_list with reads/writes | |
802 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
803 | */ | |
010e8e6b | 804 | struct list_head inflight_entry; |
7cbf1722 JA |
805 | union { |
806 | struct io_task_work io_task_work; | |
807 | struct callback_head task_work; | |
808 | }; | |
010e8e6b PB |
809 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
810 | struct hlist_node hash_node; | |
811 | struct async_poll *apoll; | |
812 | struct io_wq_work work; | |
2b188cc1 | 813 | }; |
05589553 | 814 | |
27dc8338 PB |
815 | struct io_defer_entry { |
816 | struct list_head list; | |
817 | struct io_kiocb *req; | |
9cf7c104 | 818 | u32 seq; |
2b188cc1 JA |
819 | }; |
820 | ||
d3656344 | 821 | struct io_op_def { |
d3656344 JA |
822 | /* needs req->file assigned */ |
823 | unsigned needs_file : 1; | |
d3656344 JA |
824 | /* hash wq insertion if file is a regular file */ |
825 | unsigned hash_reg_file : 1; | |
826 | /* unbound wq insertion if file is a non-regular file */ | |
827 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
828 | /* opcode is not supported by this kernel */ |
829 | unsigned not_supported : 1; | |
8a72758c JA |
830 | /* set if opcode supports polled "wait" */ |
831 | unsigned pollin : 1; | |
832 | unsigned pollout : 1; | |
bcda7baa JA |
833 | /* op supports buffer selection */ |
834 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
835 | /* must always have async data allocated */ |
836 | unsigned needs_async_data : 1; | |
27926b68 JA |
837 | /* should block plug */ |
838 | unsigned plug : 1; | |
e8c2bc1f JA |
839 | /* size of async data needed, if any */ |
840 | unsigned short async_size; | |
0f203765 | 841 | unsigned work_flags; |
d3656344 JA |
842 | }; |
843 | ||
0918682b | 844 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
845 | [IORING_OP_NOP] = {}, |
846 | [IORING_OP_READV] = { | |
d3656344 JA |
847 | .needs_file = 1, |
848 | .unbound_nonreg_file = 1, | |
8a72758c | 849 | .pollin = 1, |
4d954c25 | 850 | .buffer_select = 1, |
e8c2bc1f | 851 | .needs_async_data = 1, |
27926b68 | 852 | .plug = 1, |
e8c2bc1f | 853 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 854 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
d3656344 | 855 | }, |
0463b6c5 | 856 | [IORING_OP_WRITEV] = { |
d3656344 JA |
857 | .needs_file = 1, |
858 | .hash_reg_file = 1, | |
859 | .unbound_nonreg_file = 1, | |
8a72758c | 860 | .pollout = 1, |
e8c2bc1f | 861 | .needs_async_data = 1, |
27926b68 | 862 | .plug = 1, |
e8c2bc1f | 863 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
864 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
865 | IO_WQ_WORK_FSIZE, | |
d3656344 | 866 | }, |
0463b6c5 | 867 | [IORING_OP_FSYNC] = { |
d3656344 | 868 | .needs_file = 1, |
0f203765 | 869 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 870 | }, |
0463b6c5 | 871 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
872 | .needs_file = 1, |
873 | .unbound_nonreg_file = 1, | |
8a72758c | 874 | .pollin = 1, |
27926b68 | 875 | .plug = 1, |
e8c2bc1f | 876 | .async_size = sizeof(struct io_async_rw), |
4017eb91 | 877 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
d3656344 | 878 | }, |
0463b6c5 | 879 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
880 | .needs_file = 1, |
881 | .hash_reg_file = 1, | |
882 | .unbound_nonreg_file = 1, | |
8a72758c | 883 | .pollout = 1, |
27926b68 | 884 | .plug = 1, |
e8c2bc1f | 885 | .async_size = sizeof(struct io_async_rw), |
4017eb91 JA |
886 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE | |
887 | IO_WQ_WORK_MM, | |
d3656344 | 888 | }, |
0463b6c5 | 889 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
890 | .needs_file = 1, |
891 | .unbound_nonreg_file = 1, | |
892 | }, | |
0463b6c5 PB |
893 | [IORING_OP_POLL_REMOVE] = {}, |
894 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 | 895 | .needs_file = 1, |
0f203765 | 896 | .work_flags = IO_WQ_WORK_BLKCG, |
d3656344 | 897 | }, |
0463b6c5 | 898 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
899 | .needs_file = 1, |
900 | .unbound_nonreg_file = 1, | |
8a72758c | 901 | .pollout = 1, |
e8c2bc1f JA |
902 | .needs_async_data = 1, |
903 | .async_size = sizeof(struct io_async_msghdr), | |
92c75f75 JA |
904 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
905 | IO_WQ_WORK_FS, | |
d3656344 | 906 | }, |
0463b6c5 | 907 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
908 | .needs_file = 1, |
909 | .unbound_nonreg_file = 1, | |
8a72758c | 910 | .pollin = 1, |
52de1fe1 | 911 | .buffer_select = 1, |
e8c2bc1f JA |
912 | .needs_async_data = 1, |
913 | .async_size = sizeof(struct io_async_msghdr), | |
92c75f75 JA |
914 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
915 | IO_WQ_WORK_FS, | |
d3656344 | 916 | }, |
0463b6c5 | 917 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
918 | .needs_async_data = 1, |
919 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 920 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 921 | }, |
9c8e11b3 PB |
922 | [IORING_OP_TIMEOUT_REMOVE] = { |
923 | /* used by timeout updates' prep() */ | |
924 | .work_flags = IO_WQ_WORK_MM, | |
925 | }, | |
0463b6c5 | 926 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
927 | .needs_file = 1, |
928 | .unbound_nonreg_file = 1, | |
8a72758c | 929 | .pollin = 1, |
0f203765 | 930 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES, |
d3656344 | 931 | }, |
0463b6c5 PB |
932 | [IORING_OP_ASYNC_CANCEL] = {}, |
933 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
934 | .needs_async_data = 1, |
935 | .async_size = sizeof(struct io_timeout_data), | |
0f203765 | 936 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_CONNECT] = { |
d3656344 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
8a72758c | 941 | .pollout = 1, |
e8c2bc1f JA |
942 | .needs_async_data = 1, |
943 | .async_size = sizeof(struct io_async_connect), | |
0f203765 | 944 | .work_flags = IO_WQ_WORK_MM, |
d3656344 | 945 | }, |
0463b6c5 | 946 | [IORING_OP_FALLOCATE] = { |
d3656344 | 947 | .needs_file = 1, |
69228338 | 948 | .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE, |
d3656344 | 949 | }, |
0463b6c5 | 950 | [IORING_OP_OPENAT] = { |
0f203765 | 951 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG | |
14587a46 | 952 | IO_WQ_WORK_FS | IO_WQ_WORK_MM, |
d3656344 | 953 | }, |
0463b6c5 | 954 | [IORING_OP_CLOSE] = { |
0f203765 | 955 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG, |
d3656344 | 956 | }, |
0463b6c5 | 957 | [IORING_OP_FILES_UPDATE] = { |
0f203765 | 958 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM, |
d3656344 | 959 | }, |
0463b6c5 | 960 | [IORING_OP_STATX] = { |
0f203765 JA |
961 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM | |
962 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
d3656344 | 963 | }, |
0463b6c5 | 964 | [IORING_OP_READ] = { |
3a6820f2 JA |
965 | .needs_file = 1, |
966 | .unbound_nonreg_file = 1, | |
8a72758c | 967 | .pollin = 1, |
bcda7baa | 968 | .buffer_select = 1, |
27926b68 | 969 | .plug = 1, |
e8c2bc1f | 970 | .async_size = sizeof(struct io_async_rw), |
0f203765 | 971 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
3a6820f2 | 972 | }, |
0463b6c5 | 973 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
974 | .needs_file = 1, |
975 | .unbound_nonreg_file = 1, | |
8a72758c | 976 | .pollout = 1, |
27926b68 | 977 | .plug = 1, |
e8c2bc1f | 978 | .async_size = sizeof(struct io_async_rw), |
69228338 JA |
979 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | |
980 | IO_WQ_WORK_FSIZE, | |
3a6820f2 | 981 | }, |
0463b6c5 | 982 | [IORING_OP_FADVISE] = { |
4840e418 | 983 | .needs_file = 1, |
0f203765 | 984 | .work_flags = IO_WQ_WORK_BLKCG, |
4840e418 | 985 | }, |
0463b6c5 | 986 | [IORING_OP_MADVISE] = { |
0f203765 | 987 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
c1ca757b | 988 | }, |
0463b6c5 | 989 | [IORING_OP_SEND] = { |
fddaface JA |
990 | .needs_file = 1, |
991 | .unbound_nonreg_file = 1, | |
8a72758c | 992 | .pollout = 1, |
0f203765 | 993 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 994 | }, |
0463b6c5 | 995 | [IORING_OP_RECV] = { |
fddaface JA |
996 | .needs_file = 1, |
997 | .unbound_nonreg_file = 1, | |
8a72758c | 998 | .pollin = 1, |
bcda7baa | 999 | .buffer_select = 1, |
0f203765 | 1000 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG, |
fddaface | 1001 | }, |
0463b6c5 | 1002 | [IORING_OP_OPENAT2] = { |
0f203765 | 1003 | .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS | |
14587a46 | 1004 | IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM, |
cebdb986 | 1005 | }, |
3e4827b0 JA |
1006 | [IORING_OP_EPOLL_CTL] = { |
1007 | .unbound_nonreg_file = 1, | |
0f203765 | 1008 | .work_flags = IO_WQ_WORK_FILES, |
3e4827b0 | 1009 | }, |
7d67af2c PB |
1010 | [IORING_OP_SPLICE] = { |
1011 | .needs_file = 1, | |
1012 | .hash_reg_file = 1, | |
1013 | .unbound_nonreg_file = 1, | |
0f203765 | 1014 | .work_flags = IO_WQ_WORK_BLKCG, |
ddf0322d JA |
1015 | }, |
1016 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1017 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1018 | [IORING_OP_TEE] = { |
1019 | .needs_file = 1, | |
1020 | .hash_reg_file = 1, | |
1021 | .unbound_nonreg_file = 1, | |
1022 | }, | |
36f4fa68 JA |
1023 | [IORING_OP_SHUTDOWN] = { |
1024 | .needs_file = 1, | |
1025 | }, | |
80a261fd JA |
1026 | [IORING_OP_RENAMEAT] = { |
1027 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1028 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1029 | }, | |
14a1143b JA |
1030 | [IORING_OP_UNLINKAT] = { |
1031 | .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES | | |
1032 | IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG, | |
1033 | }, | |
d3656344 JA |
1034 | }; |
1035 | ||
9936c7c2 PB |
1036 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1037 | struct task_struct *task, | |
1038 | struct files_struct *files); | |
269bbe5f | 1039 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1040 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1041 | struct io_ring_ctx *ctx); |
f2303b1f | 1042 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 1043 | |
23faba36 | 1044 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1045 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1046 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1047 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1048 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1049 | static void io_dismantle_req(struct io_kiocb *req); |
1050 | static void io_put_task(struct task_struct *task, int nr); | |
1051 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1052 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1053 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1054 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1055 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1056 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1057 | unsigned nr_args); |
3ca405eb | 1058 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1059 | static struct file *io_file_get(struct io_submit_state *state, |
1060 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1061 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1062 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1063 | |
847595de PB |
1064 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1065 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1066 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1067 | const struct iovec *fast_iov, | |
227c0c96 | 1068 | struct iov_iter *iter, bool force); |
907d1df3 | 1069 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1070 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1071 | struct io_ring_ctx *ctx); | |
de0617e4 | 1072 | |
2b188cc1 JA |
1073 | static struct kmem_cache *req_cachep; |
1074 | ||
0918682b | 1075 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1076 | |
1077 | struct sock *io_uring_get_socket(struct file *file) | |
1078 | { | |
1079 | #if defined(CONFIG_UNIX) | |
1080 | if (file->f_op == &io_uring_fops) { | |
1081 | struct io_ring_ctx *ctx = file->private_data; | |
1082 | ||
1083 | return ctx->ring_sock->sk; | |
1084 | } | |
1085 | #endif | |
1086 | return NULL; | |
1087 | } | |
1088 | EXPORT_SYMBOL(io_uring_get_socket); | |
1089 | ||
f2f87370 PB |
1090 | #define io_for_each_link(pos, head) \ |
1091 | for (pos = (head); pos; pos = pos->link) | |
1092 | ||
3ca405eb PB |
1093 | static inline void io_clean_op(struct io_kiocb *req) |
1094 | { | |
9d5c8190 | 1095 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1096 | __io_clean_op(req); |
1097 | } | |
1098 | ||
36f72fe2 PB |
1099 | static inline void io_set_resource_node(struct io_kiocb *req) |
1100 | { | |
1101 | struct io_ring_ctx *ctx = req->ctx; | |
1102 | ||
269bbe5f BM |
1103 | if (!req->fixed_rsrc_refs) { |
1104 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1105 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1106 | } |
1107 | } | |
1108 | ||
88f171ab PB |
1109 | static bool io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1110 | { | |
1111 | if (!percpu_ref_tryget(ref)) { | |
1112 | /* already at zero, wait for ->release() */ | |
1113 | if (!try_wait_for_completion(compl)) | |
1114 | synchronize_rcu(); | |
1115 | return false; | |
1116 | } | |
1117 | ||
1118 | percpu_ref_resurrect(ref); | |
1119 | reinit_completion(compl); | |
1120 | percpu_ref_put(ref); | |
1121 | return true; | |
1122 | } | |
1123 | ||
08d23634 PB |
1124 | static bool io_match_task(struct io_kiocb *head, |
1125 | struct task_struct *task, | |
1126 | struct files_struct *files) | |
1127 | { | |
1128 | struct io_kiocb *req; | |
1129 | ||
84965ff8 JA |
1130 | if (task && head->task != task) { |
1131 | /* in terms of cancelation, always match if req task is dead */ | |
1132 | if (head->task->flags & PF_EXITING) | |
1133 | return true; | |
08d23634 | 1134 | return false; |
84965ff8 | 1135 | } |
08d23634 PB |
1136 | if (!files) |
1137 | return true; | |
1138 | ||
1139 | io_for_each_link(req, head) { | |
02a13674 JA |
1140 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
1141 | continue; | |
1142 | if (req->file && req->file->f_op == &io_uring_fops) | |
1143 | return true; | |
1144 | if ((req->work.flags & IO_WQ_WORK_FILES) && | |
08d23634 PB |
1145 | req->work.identity->files == files) |
1146 | return true; | |
1147 | } | |
1148 | return false; | |
1149 | } | |
1150 | ||
28cea78a | 1151 | static void io_sq_thread_drop_mm_files(void) |
c40f6379 | 1152 | { |
28cea78a | 1153 | struct files_struct *files = current->files; |
c40f6379 JA |
1154 | struct mm_struct *mm = current->mm; |
1155 | ||
1156 | if (mm) { | |
1157 | kthread_unuse_mm(mm); | |
1158 | mmput(mm); | |
4b70cf9d | 1159 | current->mm = NULL; |
c40f6379 | 1160 | } |
28cea78a JA |
1161 | if (files) { |
1162 | struct nsproxy *nsproxy = current->nsproxy; | |
1163 | ||
1164 | task_lock(current); | |
1165 | current->files = NULL; | |
1166 | current->nsproxy = NULL; | |
1167 | task_unlock(current); | |
1168 | put_files_struct(files); | |
1169 | put_nsproxy(nsproxy); | |
1170 | } | |
1171 | } | |
1172 | ||
1a38ffc9 | 1173 | static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx) |
28cea78a JA |
1174 | { |
1175 | if (!current->files) { | |
1176 | struct files_struct *files; | |
1177 | struct nsproxy *nsproxy; | |
1178 | ||
1179 | task_lock(ctx->sqo_task); | |
1180 | files = ctx->sqo_task->files; | |
1181 | if (!files) { | |
1182 | task_unlock(ctx->sqo_task); | |
1a38ffc9 | 1183 | return -EOWNERDEAD; |
28cea78a JA |
1184 | } |
1185 | atomic_inc(&files->count); | |
1186 | get_nsproxy(ctx->sqo_task->nsproxy); | |
1187 | nsproxy = ctx->sqo_task->nsproxy; | |
1188 | task_unlock(ctx->sqo_task); | |
1189 | ||
1190 | task_lock(current); | |
1191 | current->files = files; | |
1192 | current->nsproxy = nsproxy; | |
1193 | task_unlock(current); | |
1194 | } | |
1a38ffc9 | 1195 | return 0; |
c40f6379 JA |
1196 | } |
1197 | ||
1198 | static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx) | |
1199 | { | |
4b70cf9d JA |
1200 | struct mm_struct *mm; |
1201 | ||
1202 | if (current->mm) | |
1203 | return 0; | |
1204 | ||
4b70cf9d JA |
1205 | task_lock(ctx->sqo_task); |
1206 | mm = ctx->sqo_task->mm; | |
1207 | if (unlikely(!mm || !mmget_not_zero(mm))) | |
1208 | mm = NULL; | |
1209 | task_unlock(ctx->sqo_task); | |
1210 | ||
1211 | if (mm) { | |
1212 | kthread_use_mm(mm); | |
1213 | return 0; | |
c40f6379 JA |
1214 | } |
1215 | ||
4b70cf9d | 1216 | return -EFAULT; |
c40f6379 JA |
1217 | } |
1218 | ||
4e326358 PB |
1219 | static int __io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1220 | struct io_kiocb *req) | |
c40f6379 | 1221 | { |
28cea78a | 1222 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
1a38ffc9 | 1223 | int ret; |
28cea78a JA |
1224 | |
1225 | if (def->work_flags & IO_WQ_WORK_MM) { | |
1a38ffc9 | 1226 | ret = __io_sq_thread_acquire_mm(ctx); |
28cea78a JA |
1227 | if (unlikely(ret)) |
1228 | return ret; | |
1229 | } | |
1230 | ||
1a38ffc9 PB |
1231 | if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) { |
1232 | ret = __io_sq_thread_acquire_files(ctx); | |
1233 | if (unlikely(ret)) | |
1234 | return ret; | |
1235 | } | |
28cea78a JA |
1236 | |
1237 | return 0; | |
c40f6379 JA |
1238 | } |
1239 | ||
4e326358 PB |
1240 | static inline int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx, |
1241 | struct io_kiocb *req) | |
1242 | { | |
4e326358 PB |
1243 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
1244 | return 0; | |
1245 | return __io_sq_thread_acquire_mm_files(ctx, req); | |
1246 | } | |
1247 | ||
91d8f519 DZ |
1248 | static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx, |
1249 | struct cgroup_subsys_state **cur_css) | |
1250 | ||
1251 | { | |
1252 | #ifdef CONFIG_BLK_CGROUP | |
1253 | /* puts the old one when swapping */ | |
1254 | if (*cur_css != ctx->sqo_blkcg_css) { | |
1255 | kthread_associate_blkcg(ctx->sqo_blkcg_css); | |
1256 | *cur_css = ctx->sqo_blkcg_css; | |
1257 | } | |
1258 | #endif | |
1259 | } | |
1260 | ||
1261 | static void io_sq_thread_unassociate_blkcg(void) | |
1262 | { | |
1263 | #ifdef CONFIG_BLK_CGROUP | |
1264 | kthread_associate_blkcg(NULL); | |
1265 | #endif | |
1266 | } | |
1267 | ||
c40f6379 JA |
1268 | static inline void req_set_fail_links(struct io_kiocb *req) |
1269 | { | |
1270 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1271 | req->flags |= REQ_F_FAIL_LINK; | |
1272 | } | |
4a38aed2 | 1273 | |
1e6fa521 JA |
1274 | /* |
1275 | * None of these are dereferenced, they are simply used to check if any of | |
1276 | * them have changed. If we're under current and check they are still the | |
1277 | * same, we're fine to grab references to them for actual out-of-line use. | |
1278 | */ | |
1279 | static void io_init_identity(struct io_identity *id) | |
1280 | { | |
1281 | id->files = current->files; | |
1282 | id->mm = current->mm; | |
1283 | #ifdef CONFIG_BLK_CGROUP | |
1284 | rcu_read_lock(); | |
1285 | id->blkcg_css = blkcg_css(); | |
1286 | rcu_read_unlock(); | |
1287 | #endif | |
1288 | id->creds = current_cred(); | |
1289 | id->nsproxy = current->nsproxy; | |
1290 | id->fs = current->fs; | |
1291 | id->fsize = rlimit(RLIMIT_FSIZE); | |
4ea33a97 JA |
1292 | #ifdef CONFIG_AUDIT |
1293 | id->loginuid = current->loginuid; | |
1294 | id->sessionid = current->sessionid; | |
1295 | #endif | |
1e6fa521 JA |
1296 | refcount_set(&id->count, 1); |
1297 | } | |
1298 | ||
ec99ca6c PB |
1299 | static inline void __io_req_init_async(struct io_kiocb *req) |
1300 | { | |
1301 | memset(&req->work, 0, sizeof(req->work)); | |
1302 | req->flags |= REQ_F_WORK_INITIALIZED; | |
1303 | } | |
1304 | ||
7cdaf587 XW |
1305 | /* |
1306 | * Note: must call io_req_init_async() for the first time you | |
1307 | * touch any members of io_wq_work. | |
1308 | */ | |
1309 | static inline void io_req_init_async(struct io_kiocb *req) | |
1310 | { | |
500a373d JA |
1311 | struct io_uring_task *tctx = current->io_uring; |
1312 | ||
7cdaf587 XW |
1313 | if (req->flags & REQ_F_WORK_INITIALIZED) |
1314 | return; | |
1315 | ||
ec99ca6c | 1316 | __io_req_init_async(req); |
500a373d JA |
1317 | |
1318 | /* Grab a ref if this isn't our static identity */ | |
1319 | req->work.identity = tctx->identity; | |
1320 | if (tctx->identity != &tctx->__identity) | |
1321 | refcount_inc(&req->work.identity->count); | |
7cdaf587 XW |
1322 | } |
1323 | ||
2b188cc1 JA |
1324 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1325 | { | |
1326 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1327 | ||
0f158b4c | 1328 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1329 | } |
1330 | ||
8eb7e2d0 PB |
1331 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1332 | { | |
1333 | return !req->timeout.off; | |
1334 | } | |
1335 | ||
2b188cc1 JA |
1336 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1337 | { | |
1338 | struct io_ring_ctx *ctx; | |
78076bb6 | 1339 | int hash_bits; |
2b188cc1 JA |
1340 | |
1341 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1342 | if (!ctx) | |
1343 | return NULL; | |
1344 | ||
78076bb6 JA |
1345 | /* |
1346 | * Use 5 bits less than the max cq entries, that should give us around | |
1347 | * 32 entries per hash list if totally full and uniformly spread. | |
1348 | */ | |
1349 | hash_bits = ilog2(p->cq_entries); | |
1350 | hash_bits -= 5; | |
1351 | if (hash_bits <= 0) | |
1352 | hash_bits = 1; | |
1353 | ctx->cancel_hash_bits = hash_bits; | |
1354 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1355 | GFP_KERNEL); | |
1356 | if (!ctx->cancel_hash) | |
1357 | goto err; | |
1358 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1359 | ||
21482896 | 1360 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1361 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1362 | goto err; | |
2b188cc1 JA |
1363 | |
1364 | ctx->flags = p->flags; | |
90554200 | 1365 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1366 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1367 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1368 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1369 | init_completion(&ctx->ref_comp); |
1370 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1371 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1372 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1373 | mutex_init(&ctx->uring_lock); |
1374 | init_waitqueue_head(&ctx->wait); | |
1375 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1376 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1377 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1378 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1379 | spin_lock_init(&ctx->inflight_lock); |
1380 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1381 | spin_lock_init(&ctx->rsrc_ref_lock); |
1382 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1383 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1384 | init_llist_head(&ctx->rsrc_put_llist); | |
1b4c351f | 1385 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1386 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1387 | return ctx; |
206aefde | 1388 | err: |
78076bb6 | 1389 | kfree(ctx->cancel_hash); |
206aefde JA |
1390 | kfree(ctx); |
1391 | return NULL; | |
2b188cc1 JA |
1392 | } |
1393 | ||
9cf7c104 | 1394 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1395 | { |
2bc9930e JA |
1396 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1397 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1398 | |
9cf7c104 | 1399 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1400 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1401 | } |
de0617e4 | 1402 | |
9d858b21 | 1403 | return false; |
de0617e4 JA |
1404 | } |
1405 | ||
5c3462cf | 1406 | static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req) |
1e6fa521 | 1407 | { |
500a373d | 1408 | if (req->work.identity == &tctx->__identity) |
1e6fa521 JA |
1409 | return; |
1410 | if (refcount_dec_and_test(&req->work.identity->count)) | |
1411 | kfree(req->work.identity); | |
1412 | } | |
1413 | ||
4edf20f9 | 1414 | static void io_req_clean_work(struct io_kiocb *req) |
18d9be1a | 1415 | { |
7cdaf587 | 1416 | if (!(req->flags & REQ_F_WORK_INITIALIZED)) |
4edf20f9 | 1417 | return; |
51a4cc11 | 1418 | |
e86d0047 | 1419 | if (req->work.flags & IO_WQ_WORK_MM) |
98447d65 | 1420 | mmdrop(req->work.identity->mm); |
91d8f519 | 1421 | #ifdef CONFIG_BLK_CGROUP |
e86d0047 | 1422 | if (req->work.flags & IO_WQ_WORK_BLKCG) |
98447d65 | 1423 | css_put(req->work.identity->blkcg_css); |
91d8f519 | 1424 | #endif |
e86d0047 | 1425 | if (req->work.flags & IO_WQ_WORK_CREDS) |
98447d65 | 1426 | put_cred(req->work.identity->creds); |
dfead8a8 | 1427 | if (req->work.flags & IO_WQ_WORK_FS) { |
98447d65 | 1428 | struct fs_struct *fs = req->work.identity->fs; |
51a4cc11 | 1429 | |
98447d65 | 1430 | spin_lock(&req->work.identity->fs->lock); |
ff002b30 JA |
1431 | if (--fs->users) |
1432 | fs = NULL; | |
98447d65 | 1433 | spin_unlock(&req->work.identity->fs->lock); |
ff002b30 JA |
1434 | if (fs) |
1435 | free_fs_struct(fs); | |
1436 | } | |
34e08fed PB |
1437 | if (req->work.flags & IO_WQ_WORK_FILES) { |
1438 | put_files_struct(req->work.identity->files); | |
1439 | put_nsproxy(req->work.identity->nsproxy); | |
34e08fed PB |
1440 | } |
1441 | if (req->flags & REQ_F_INFLIGHT) { | |
1442 | struct io_ring_ctx *ctx = req->ctx; | |
1443 | struct io_uring_task *tctx = req->task->io_uring; | |
1444 | unsigned long flags; | |
1445 | ||
1446 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1447 | list_del(&req->inflight_entry); | |
1448 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1449 | req->flags &= ~REQ_F_INFLIGHT; | |
1450 | if (atomic_read(&tctx->in_idle)) | |
1451 | wake_up(&tctx->wait); | |
1452 | } | |
51a4cc11 | 1453 | |
e86d0047 PB |
1454 | req->flags &= ~REQ_F_WORK_INITIALIZED; |
1455 | req->work.flags &= ~(IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | IO_WQ_WORK_FS | | |
1456 | IO_WQ_WORK_CREDS | IO_WQ_WORK_FILES); | |
5c3462cf | 1457 | io_put_identity(req->task->io_uring, req); |
561fb04a JA |
1458 | } |
1459 | ||
1e6fa521 JA |
1460 | /* |
1461 | * Create a private copy of io_identity, since some fields don't match | |
1462 | * the current context. | |
1463 | */ | |
1464 | static bool io_identity_cow(struct io_kiocb *req) | |
1465 | { | |
5c3462cf | 1466 | struct io_uring_task *tctx = current->io_uring; |
1e6fa521 JA |
1467 | const struct cred *creds = NULL; |
1468 | struct io_identity *id; | |
1469 | ||
1470 | if (req->work.flags & IO_WQ_WORK_CREDS) | |
1471 | creds = req->work.identity->creds; | |
1472 | ||
1473 | id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL); | |
1474 | if (unlikely(!id)) { | |
1475 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1476 | return false; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * We can safely just re-init the creds we copied Either the field | |
1481 | * matches the current one, or we haven't grabbed it yet. The only | |
1482 | * exception is ->creds, through registered personalities, so handle | |
1483 | * that one separately. | |
1484 | */ | |
1485 | io_init_identity(id); | |
1486 | if (creds) | |
e8c954df | 1487 | id->creds = creds; |
1e6fa521 JA |
1488 | |
1489 | /* add one for this request */ | |
1490 | refcount_inc(&id->count); | |
1491 | ||
cb8a8ae3 JA |
1492 | /* drop tctx and req identity references, if needed */ |
1493 | if (tctx->identity != &tctx->__identity && | |
1494 | refcount_dec_and_test(&tctx->identity->count)) | |
1495 | kfree(tctx->identity); | |
1496 | if (req->work.identity != &tctx->__identity && | |
1497 | refcount_dec_and_test(&req->work.identity->count)) | |
1e6fa521 JA |
1498 | kfree(req->work.identity); |
1499 | ||
1500 | req->work.identity = id; | |
500a373d | 1501 | tctx->identity = id; |
1e6fa521 JA |
1502 | return true; |
1503 | } | |
1504 | ||
ce3d5aae PB |
1505 | static void io_req_track_inflight(struct io_kiocb *req) |
1506 | { | |
1507 | struct io_ring_ctx *ctx = req->ctx; | |
1508 | ||
1509 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
1510 | io_req_init_async(req); | |
1511 | req->flags |= REQ_F_INFLIGHT; | |
1512 | ||
1513 | spin_lock_irq(&ctx->inflight_lock); | |
1514 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1515 | spin_unlock_irq(&ctx->inflight_lock); | |
1516 | } | |
1517 | } | |
1518 | ||
1e6fa521 | 1519 | static bool io_grab_identity(struct io_kiocb *req) |
18d9be1a | 1520 | { |
d3656344 | 1521 | const struct io_op_def *def = &io_op_defs[req->opcode]; |
5c3462cf | 1522 | struct io_identity *id = req->work.identity; |
54a91f3b | 1523 | |
69228338 JA |
1524 | if (def->work_flags & IO_WQ_WORK_FSIZE) { |
1525 | if (id->fsize != rlimit(RLIMIT_FSIZE)) | |
1526 | return false; | |
1527 | req->work.flags |= IO_WQ_WORK_FSIZE; | |
1528 | } | |
91d8f519 | 1529 | #ifdef CONFIG_BLK_CGROUP |
dfead8a8 JA |
1530 | if (!(req->work.flags & IO_WQ_WORK_BLKCG) && |
1531 | (def->work_flags & IO_WQ_WORK_BLKCG)) { | |
91d8f519 | 1532 | rcu_read_lock(); |
1e6fa521 JA |
1533 | if (id->blkcg_css != blkcg_css()) { |
1534 | rcu_read_unlock(); | |
1535 | return false; | |
1536 | } | |
91d8f519 DZ |
1537 | /* |
1538 | * This should be rare, either the cgroup is dying or the task | |
1539 | * is moving cgroups. Just punt to root for the handful of ios. | |
1540 | */ | |
1e6fa521 | 1541 | if (css_tryget_online(id->blkcg_css)) |
dfead8a8 | 1542 | req->work.flags |= IO_WQ_WORK_BLKCG; |
91d8f519 DZ |
1543 | rcu_read_unlock(); |
1544 | } | |
1545 | #endif | |
dfead8a8 | 1546 | if (!(req->work.flags & IO_WQ_WORK_CREDS)) { |
1e6fa521 JA |
1547 | if (id->creds != current_cred()) |
1548 | return false; | |
1549 | get_cred(id->creds); | |
dfead8a8 JA |
1550 | req->work.flags |= IO_WQ_WORK_CREDS; |
1551 | } | |
4ea33a97 JA |
1552 | #ifdef CONFIG_AUDIT |
1553 | if (!uid_eq(current->loginuid, id->loginuid) || | |
1554 | current->sessionid != id->sessionid) | |
1555 | return false; | |
1556 | #endif | |
dfead8a8 JA |
1557 | if (!(req->work.flags & IO_WQ_WORK_FS) && |
1558 | (def->work_flags & IO_WQ_WORK_FS)) { | |
1e6fa521 JA |
1559 | if (current->fs != id->fs) |
1560 | return false; | |
1561 | spin_lock(&id->fs->lock); | |
1562 | if (!id->fs->in_exec) { | |
1563 | id->fs->users++; | |
dfead8a8 | 1564 | req->work.flags |= IO_WQ_WORK_FS; |
dca9cf8b PB |
1565 | } else { |
1566 | req->work.flags |= IO_WQ_WORK_CANCEL; | |
1567 | } | |
1568 | spin_unlock(¤t->fs->lock); | |
1569 | } | |
af604703 PB |
1570 | if (!(req->work.flags & IO_WQ_WORK_FILES) && |
1571 | (def->work_flags & IO_WQ_WORK_FILES) && | |
1572 | !(req->flags & REQ_F_NO_FILE_TABLE)) { | |
1573 | if (id->files != current->files || | |
1574 | id->nsproxy != current->nsproxy) | |
1575 | return false; | |
1576 | atomic_inc(&id->files->count); | |
1577 | get_nsproxy(id->nsproxy); | |
af604703 | 1578 | req->work.flags |= IO_WQ_WORK_FILES; |
ce3d5aae | 1579 | io_req_track_inflight(req); |
af604703 | 1580 | } |
77788775 JA |
1581 | if (!(req->work.flags & IO_WQ_WORK_MM) && |
1582 | (def->work_flags & IO_WQ_WORK_MM)) { | |
1583 | if (id->mm != current->mm) | |
1584 | return false; | |
1585 | mmgrab(id->mm); | |
1586 | req->work.flags |= IO_WQ_WORK_MM; | |
1587 | } | |
1e6fa521 JA |
1588 | |
1589 | return true; | |
1590 | } | |
1591 | ||
1592 | static void io_prep_async_work(struct io_kiocb *req) | |
1593 | { | |
1594 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1595 | struct io_ring_ctx *ctx = req->ctx; |
1596 | ||
1597 | io_req_init_async(req); | |
1598 | ||
feaadc4f PB |
1599 | if (req->flags & REQ_F_FORCE_ASYNC) |
1600 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1601 | ||
1e6fa521 JA |
1602 | if (req->flags & REQ_F_ISREG) { |
1603 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1604 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1605 | } else { | |
1606 | if (def->unbound_nonreg_file) | |
1607 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1608 | } | |
1609 | ||
1e6fa521 JA |
1610 | /* if we fail grabbing identity, we must COW, regrab, and retry */ |
1611 | if (io_grab_identity(req)) | |
1612 | return; | |
1613 | ||
1614 | if (!io_identity_cow(req)) | |
1615 | return; | |
1616 | ||
1617 | /* can't fail at this point */ | |
1618 | if (!io_grab_identity(req)) | |
1619 | WARN_ON(1); | |
561fb04a | 1620 | } |
cccf0ee8 | 1621 | |
cbdcb435 | 1622 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1623 | { |
cbdcb435 | 1624 | struct io_kiocb *cur; |
54a91f3b | 1625 | |
f2f87370 PB |
1626 | io_for_each_link(cur, req) |
1627 | io_prep_async_work(cur); | |
561fb04a JA |
1628 | } |
1629 | ||
7271ef3a | 1630 | static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1631 | { |
a197f664 | 1632 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1633 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1634 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1635 | |
3bfe6106 JA |
1636 | BUG_ON(!tctx); |
1637 | BUG_ON(!tctx->io_wq); | |
1638 | ||
8766dd51 PB |
1639 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1640 | &req->work, req->flags); | |
5aa75ed5 | 1641 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a | 1642 | return link; |
18d9be1a JA |
1643 | } |
1644 | ||
cbdcb435 PB |
1645 | static void io_queue_async_work(struct io_kiocb *req) |
1646 | { | |
7271ef3a JA |
1647 | struct io_kiocb *link; |
1648 | ||
cbdcb435 PB |
1649 | /* init ->work of the whole link before punting */ |
1650 | io_prep_async_link(req); | |
7271ef3a JA |
1651 | link = __io_queue_async_work(req); |
1652 | ||
1653 | if (link) | |
1654 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1655 | } |
1656 | ||
5262f567 JA |
1657 | static void io_kill_timeout(struct io_kiocb *req) |
1658 | { | |
e8c2bc1f | 1659 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1660 | int ret; |
1661 | ||
e8c2bc1f | 1662 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1663 | if (ret != -1) { |
01cec8c1 PB |
1664 | atomic_set(&req->ctx->cq_timeouts, |
1665 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1666 | list_del_init(&req->timeout.list); |
78e19bbe | 1667 | io_cqring_fill_event(req, 0); |
216578e5 | 1668 | io_put_req_deferred(req, 1); |
5262f567 JA |
1669 | } |
1670 | } | |
1671 | ||
76e1b642 JA |
1672 | /* |
1673 | * Returns true if we found and killed one or more timeouts | |
1674 | */ | |
6b81928d PB |
1675 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1676 | struct files_struct *files) | |
5262f567 JA |
1677 | { |
1678 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1679 | int canceled = 0; |
5262f567 JA |
1680 | |
1681 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1682 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1683 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1684 | io_kill_timeout(req); |
76e1b642 JA |
1685 | canceled++; |
1686 | } | |
f3606e3a | 1687 | } |
5262f567 | 1688 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1689 | return canceled != 0; |
5262f567 JA |
1690 | } |
1691 | ||
04518945 | 1692 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1693 | { |
04518945 | 1694 | do { |
27dc8338 PB |
1695 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1696 | struct io_defer_entry, list); | |
de0617e4 | 1697 | |
9cf7c104 | 1698 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1699 | break; |
27dc8338 | 1700 | list_del_init(&de->list); |
907d1df3 | 1701 | io_req_task_queue(de->req); |
27dc8338 | 1702 | kfree(de); |
04518945 PB |
1703 | } while (!list_empty(&ctx->defer_list)); |
1704 | } | |
1705 | ||
360428f8 | 1706 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1707 | { |
f010505b MDG |
1708 | u32 seq; |
1709 | ||
1710 | if (list_empty(&ctx->timeout_list)) | |
1711 | return; | |
1712 | ||
1713 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1714 | ||
1715 | do { | |
1716 | u32 events_needed, events_got; | |
360428f8 | 1717 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1718 | struct io_kiocb, timeout.list); |
de0617e4 | 1719 | |
8eb7e2d0 | 1720 | if (io_is_timeout_noseq(req)) |
360428f8 | 1721 | break; |
f010505b MDG |
1722 | |
1723 | /* | |
1724 | * Since seq can easily wrap around over time, subtract | |
1725 | * the last seq at which timeouts were flushed before comparing. | |
1726 | * Assuming not more than 2^31-1 events have happened since, | |
1727 | * these subtractions won't have wrapped, so we can check if | |
1728 | * target is in [last_seq, current_seq] by comparing the two. | |
1729 | */ | |
1730 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1731 | events_got = seq - ctx->cq_last_tm_flush; | |
1732 | if (events_got < events_needed) | |
360428f8 | 1733 | break; |
bfe68a22 | 1734 | |
135fcde8 | 1735 | list_del_init(&req->timeout.list); |
5262f567 | 1736 | io_kill_timeout(req); |
f010505b MDG |
1737 | } while (!list_empty(&ctx->timeout_list)); |
1738 | ||
1739 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1740 | } |
5262f567 | 1741 | |
360428f8 PB |
1742 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1743 | { | |
1744 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1745 | |
1746 | /* order cqe stores with ring update */ | |
1747 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1748 | |
04518945 PB |
1749 | if (unlikely(!list_empty(&ctx->defer_list))) |
1750 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1751 | } |
1752 | ||
90554200 JA |
1753 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1754 | { | |
1755 | struct io_rings *r = ctx->rings; | |
1756 | ||
1757 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1758 | } | |
1759 | ||
888aae2e PB |
1760 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1761 | { | |
1762 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1763 | } | |
1764 | ||
2b188cc1 JA |
1765 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1766 | { | |
75b28aff | 1767 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1768 | unsigned tail; |
1769 | ||
115e12e5 SB |
1770 | /* |
1771 | * writes to the cq entry need to come after reading head; the | |
1772 | * control dependency is enough as we're using WRITE_ONCE to | |
1773 | * fill the cq entry | |
1774 | */ | |
888aae2e | 1775 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1776 | return NULL; |
1777 | ||
888aae2e | 1778 | tail = ctx->cached_cq_tail++; |
75b28aff | 1779 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1780 | } |
1781 | ||
f2842ab5 JA |
1782 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1783 | { | |
f0b493e6 JA |
1784 | if (!ctx->cq_ev_fd) |
1785 | return false; | |
7e55a19c SG |
1786 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1787 | return false; | |
f2842ab5 JA |
1788 | if (!ctx->eventfd_async) |
1789 | return true; | |
b41e9852 | 1790 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1791 | } |
1792 | ||
b41e9852 | 1793 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1794 | { |
b1445e59 PB |
1795 | /* see waitqueue_active() comment */ |
1796 | smp_mb(); | |
1797 | ||
1d7bb1d5 JA |
1798 | if (waitqueue_active(&ctx->wait)) |
1799 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1800 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1801 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1802 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1803 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1804 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1805 | wake_up_interruptible(&ctx->cq_wait); |
1806 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1807 | } | |
1d7bb1d5 JA |
1808 | } |
1809 | ||
80c18e4a PB |
1810 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1811 | { | |
b1445e59 PB |
1812 | /* see waitqueue_active() comment */ |
1813 | smp_mb(); | |
1814 | ||
80c18e4a PB |
1815 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1816 | if (waitqueue_active(&ctx->wait)) | |
1817 | wake_up(&ctx->wait); | |
1818 | } | |
1819 | if (io_should_trigger_evfd(ctx)) | |
1820 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1821 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1822 | wake_up_interruptible(&ctx->cq_wait); |
1823 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1824 | } | |
80c18e4a PB |
1825 | } |
1826 | ||
c4a2ed72 | 1827 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1828 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1829 | struct task_struct *tsk, | |
1830 | struct files_struct *files) | |
1d7bb1d5 JA |
1831 | { |
1832 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1833 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1834 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1835 | unsigned long flags; |
b18032bb | 1836 | bool all_flushed, posted; |
1d7bb1d5 JA |
1837 | LIST_HEAD(list); |
1838 | ||
e23de15f PB |
1839 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1840 | return false; | |
1d7bb1d5 | 1841 | |
b18032bb | 1842 | posted = false; |
1d7bb1d5 | 1843 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1844 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1845 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1846 | continue; |
1847 | ||
1d7bb1d5 JA |
1848 | cqe = io_get_cqring(ctx); |
1849 | if (!cqe && !force) | |
1850 | break; | |
1851 | ||
40d8ddd4 | 1852 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1853 | if (cqe) { |
1854 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1855 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1856 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1857 | } else { |
2c3bac6d | 1858 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1859 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1860 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1861 | } |
b18032bb | 1862 | posted = true; |
1d7bb1d5 JA |
1863 | } |
1864 | ||
09e88404 PB |
1865 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1866 | if (all_flushed) { | |
1867 | clear_bit(0, &ctx->sq_check_overflow); | |
1868 | clear_bit(0, &ctx->cq_check_overflow); | |
1869 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1870 | } | |
46930143 | 1871 | |
b18032bb JA |
1872 | if (posted) |
1873 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1874 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1875 | if (posted) |
1876 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1877 | |
1878 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1879 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1880 | list_del(&req->compl.list); | |
ec9c02ad | 1881 | io_put_req(req); |
1d7bb1d5 | 1882 | } |
c4a2ed72 | 1883 | |
09e88404 | 1884 | return all_flushed; |
1d7bb1d5 JA |
1885 | } |
1886 | ||
6c503150 PB |
1887 | static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1888 | struct task_struct *tsk, | |
1889 | struct files_struct *files) | |
1890 | { | |
1891 | if (test_bit(0, &ctx->cq_check_overflow)) { | |
1892 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1893 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1894 | mutex_lock(&ctx->uring_lock); | |
1895 | __io_cqring_overflow_flush(ctx, force, tsk, files); | |
1896 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1897 | mutex_unlock(&ctx->uring_lock); | |
1898 | } | |
1899 | } | |
1900 | ||
bcda7baa | 1901 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1902 | { |
78e19bbe | 1903 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1904 | struct io_uring_cqe *cqe; |
1905 | ||
78e19bbe | 1906 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1907 | |
2b188cc1 JA |
1908 | /* |
1909 | * If we can't get a cq entry, userspace overflowed the | |
1910 | * submission (by quite a lot). Increment the overflow count in | |
1911 | * the ring. | |
1912 | */ | |
1913 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1914 | if (likely(cqe)) { |
78e19bbe | 1915 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1916 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1917 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1918 | } else if (ctx->cq_overflow_flushed || |
1919 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1920 | /* |
1921 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1922 | * then we cannot store the request for later flushing, we need | |
1923 | * to drop it on the floor. | |
1924 | */ | |
2c3bac6d PB |
1925 | ctx->cached_cq_overflow++; |
1926 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1927 | } else { |
ad3eb2c8 JA |
1928 | if (list_empty(&ctx->cq_overflow_list)) { |
1929 | set_bit(0, &ctx->sq_check_overflow); | |
1930 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1931 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1932 | } |
40d8ddd4 | 1933 | io_clean_op(req); |
1d7bb1d5 | 1934 | req->result = res; |
0f7e466b | 1935 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1936 | refcount_inc(&req->refs); |
1937 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1938 | } |
1939 | } | |
1940 | ||
bcda7baa JA |
1941 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1942 | { | |
1943 | __io_cqring_fill_event(req, res, 0); | |
1944 | } | |
1945 | ||
c7dae4ba JA |
1946 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1947 | unsigned int cflags) | |
2b188cc1 | 1948 | { |
78e19bbe | 1949 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1950 | unsigned long flags; |
1951 | ||
1952 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1953 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1954 | io_commit_cqring(ctx); |
c7dae4ba JA |
1955 | /* |
1956 | * If we're the last reference to this request, add to our locked | |
1957 | * free_list cache. | |
1958 | */ | |
1959 | if (refcount_dec_and_test(&req->refs)) { | |
1960 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1961 | ||
1962 | io_dismantle_req(req); | |
1963 | io_put_task(req->task, 1); | |
1964 | list_add(&req->compl.list, &cs->locked_free_list); | |
1965 | cs->locked_free_nr++; | |
1966 | } else | |
1967 | req = NULL; | |
2b188cc1 JA |
1968 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1969 | ||
8c838788 | 1970 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1971 | if (req) { |
1972 | io_queue_next(req); | |
1973 | percpu_ref_put(&ctx->refs); | |
229a7b63 | 1974 | } |
229a7b63 JA |
1975 | } |
1976 | ||
a38d68db | 1977 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1978 | unsigned int cflags) |
229a7b63 | 1979 | { |
a38d68db PB |
1980 | io_clean_op(req); |
1981 | req->result = res; | |
1982 | req->compl.cflags = cflags; | |
e342c807 | 1983 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1984 | } |
1985 | ||
889fca73 PB |
1986 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1987 | long res, unsigned cflags) | |
bcda7baa | 1988 | { |
889fca73 PB |
1989 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1990 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1991 | else |
c7dae4ba | 1992 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1993 | } |
1994 | ||
a38d68db | 1995 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1996 | { |
889fca73 | 1997 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1998 | } |
1999 | ||
c7dae4ba | 2000 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 2001 | { |
c7dae4ba JA |
2002 | struct io_submit_state *state = &ctx->submit_state; |
2003 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 2004 | struct io_kiocb *req = NULL; |
0ddf92e8 | 2005 | |
c7dae4ba JA |
2006 | /* |
2007 | * If we have more than a batch's worth of requests in our IRQ side | |
2008 | * locked cache, grab the lock and move them over to our submission | |
2009 | * side cache. | |
2010 | */ | |
2011 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
2012 | spin_lock_irq(&ctx->completion_lock); | |
2013 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
2014 | cs->locked_free_nr = 0; | |
2015 | spin_unlock_irq(&ctx->completion_lock); | |
2016 | } | |
0ddf92e8 | 2017 | |
c7dae4ba JA |
2018 | while (!list_empty(&cs->free_list)) { |
2019 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
2020 | compl.list); |
2021 | list_del(&req->compl.list); | |
e5d1bc0a PB |
2022 | state->reqs[state->free_reqs++] = req; |
2023 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
2024 | break; | |
1b4c351f JA |
2025 | } |
2026 | ||
e5d1bc0a | 2027 | return req != NULL; |
0ddf92e8 JA |
2028 | } |
2029 | ||
e5d1bc0a | 2030 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 2031 | { |
e5d1bc0a PB |
2032 | struct io_submit_state *state = &ctx->submit_state; |
2033 | ||
2034 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
2035 | ||
f6b6c7d6 | 2036 | if (!state->free_reqs) { |
291b2821 | 2037 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
2038 | int ret; |
2039 | ||
c7dae4ba | 2040 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
2041 | goto got_req; |
2042 | ||
bf019da7 PB |
2043 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
2044 | state->reqs); | |
fd6fab2c JA |
2045 | |
2046 | /* | |
2047 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
2048 | * retry single alloc to be on the safe side. | |
2049 | */ | |
2050 | if (unlikely(ret <= 0)) { | |
2051 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
2052 | if (!state->reqs[0]) | |
3893f39f | 2053 | return NULL; |
fd6fab2c JA |
2054 | ret = 1; |
2055 | } | |
291b2821 | 2056 | state->free_reqs = ret; |
2b188cc1 | 2057 | } |
e5d1bc0a | 2058 | got_req: |
291b2821 PB |
2059 | state->free_reqs--; |
2060 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
2061 | } |
2062 | ||
8da11c19 PB |
2063 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
2064 | bool fixed) | |
2065 | { | |
36f72fe2 | 2066 | if (!fixed) |
8da11c19 PB |
2067 | fput(file); |
2068 | } | |
2069 | ||
4edf20f9 | 2070 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 2071 | { |
3ca405eb | 2072 | io_clean_op(req); |
929a3af9 | 2073 | |
e8c2bc1f JA |
2074 | if (req->async_data) |
2075 | kfree(req->async_data); | |
8da11c19 PB |
2076 | if (req->file) |
2077 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
2078 | if (req->fixed_rsrc_refs) |
2079 | percpu_ref_put(req->fixed_rsrc_refs); | |
4edf20f9 | 2080 | io_req_clean_work(req); |
e65ef56d JA |
2081 | } |
2082 | ||
7c660731 PB |
2083 | static inline void io_put_task(struct task_struct *task, int nr) |
2084 | { | |
2085 | struct io_uring_task *tctx = task->io_uring; | |
2086 | ||
2087 | percpu_counter_sub(&tctx->inflight, nr); | |
2088 | if (unlikely(atomic_read(&tctx->in_idle))) | |
2089 | wake_up(&tctx->wait); | |
2090 | put_task_struct_many(task, nr); | |
2091 | } | |
2092 | ||
216578e5 | 2093 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 2094 | { |
51a4cc11 | 2095 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 2096 | |
216578e5 | 2097 | io_dismantle_req(req); |
7c660731 | 2098 | io_put_task(req->task, 1); |
c6ca97b3 | 2099 | |
3893f39f | 2100 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 2101 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
2102 | } |
2103 | ||
f2f87370 PB |
2104 | static inline void io_remove_next_linked(struct io_kiocb *req) |
2105 | { | |
2106 | struct io_kiocb *nxt = req->link; | |
2107 | ||
2108 | req->link = nxt->link; | |
2109 | nxt->link = NULL; | |
2110 | } | |
2111 | ||
c9abd7ad | 2112 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 2113 | { |
a197f664 | 2114 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 2115 | struct io_kiocb *link; |
c9abd7ad PB |
2116 | bool cancelled = false; |
2117 | unsigned long flags; | |
7c86ffee | 2118 | |
c9abd7ad | 2119 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2120 | link = req->link; |
2121 | ||
900fad45 PB |
2122 | /* |
2123 | * Can happen if a linked timeout fired and link had been like | |
2124 | * req -> link t-out -> link t-out [-> ...] | |
2125 | */ | |
c9abd7ad PB |
2126 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
2127 | struct io_timeout_data *io = link->async_data; | |
2128 | int ret; | |
7c86ffee | 2129 | |
f2f87370 | 2130 | io_remove_next_linked(req); |
90cd7e42 | 2131 | link->timeout.head = NULL; |
c9abd7ad PB |
2132 | ret = hrtimer_try_to_cancel(&io->timer); |
2133 | if (ret != -1) { | |
2134 | io_cqring_fill_event(link, -ECANCELED); | |
2135 | io_commit_cqring(ctx); | |
2136 | cancelled = true; | |
2137 | } | |
2138 | } | |
7c86ffee | 2139 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 2140 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 2141 | |
c9abd7ad | 2142 | if (cancelled) { |
7c86ffee | 2143 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
2144 | io_put_req(link); |
2145 | } | |
7c86ffee PB |
2146 | } |
2147 | ||
9e645e11 | 2148 | |
d148ca4b | 2149 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 2150 | { |
f2f87370 | 2151 | struct io_kiocb *link, *nxt; |
2665abfd | 2152 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 2153 | unsigned long flags; |
9e645e11 | 2154 | |
d148ca4b | 2155 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
2156 | link = req->link; |
2157 | req->link = NULL; | |
9e645e11 | 2158 | |
f2f87370 PB |
2159 | while (link) { |
2160 | nxt = link->link; | |
2161 | link->link = NULL; | |
2665abfd | 2162 | |
f2f87370 | 2163 | trace_io_uring_fail_link(req, link); |
7c86ffee | 2164 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 PB |
2165 | |
2166 | /* | |
2167 | * It's ok to free under spinlock as they're not linked anymore, | |
2168 | * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on | |
2169 | * work.fs->lock. | |
2170 | */ | |
2171 | if (link->flags & REQ_F_WORK_INITIALIZED) | |
2172 | io_put_req_deferred(link, 2); | |
2173 | else | |
2174 | io_double_put_req(link); | |
f2f87370 | 2175 | link = nxt; |
9e645e11 | 2176 | } |
2665abfd | 2177 | io_commit_cqring(ctx); |
216578e5 | 2178 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 2179 | |
2665abfd | 2180 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
2181 | } |
2182 | ||
3fa5e0f3 | 2183 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 2184 | { |
7c86ffee PB |
2185 | if (req->flags & REQ_F_LINK_TIMEOUT) |
2186 | io_kill_linked_timeout(req); | |
944e58bf | 2187 | |
9e645e11 JA |
2188 | /* |
2189 | * If LINK is set, we have dependent requests in this chain. If we | |
2190 | * didn't fail this request, queue the first one up, moving any other | |
2191 | * dependencies to the next request. In case of failure, fail the rest | |
2192 | * of the chain. | |
2193 | */ | |
f2f87370 PB |
2194 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
2195 | struct io_kiocb *nxt = req->link; | |
2196 | ||
2197 | req->link = NULL; | |
2198 | return nxt; | |
2199 | } | |
9b5f7bd9 PB |
2200 | io_fail_links(req); |
2201 | return NULL; | |
4d7dd462 | 2202 | } |
9e645e11 | 2203 | |
f2f87370 | 2204 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 2205 | { |
cdbff982 | 2206 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
2207 | return NULL; |
2208 | return __io_req_find_next(req); | |
2209 | } | |
2210 | ||
7cbf1722 | 2211 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 2212 | { |
65453d1e | 2213 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
2214 | struct io_wq_work_list list; |
2215 | struct io_wq_work_node *node; | |
c2c4c83c | 2216 | |
7cbf1722 JA |
2217 | if (wq_list_empty(&tctx->task_list)) |
2218 | return false; | |
6200b0ae | 2219 | |
0b81e80c | 2220 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
2221 | list = tctx->task_list; |
2222 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 2223 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 2224 | |
7cbf1722 JA |
2225 | node = list.first; |
2226 | while (node) { | |
2227 | struct io_wq_work_node *next = node->next; | |
65453d1e | 2228 | struct io_ring_ctx *this_ctx; |
7cbf1722 | 2229 | struct io_kiocb *req; |
0ba9c9ed | 2230 | |
7cbf1722 | 2231 | req = container_of(node, struct io_kiocb, io_task_work.node); |
65453d1e | 2232 | this_ctx = req->ctx; |
7cbf1722 JA |
2233 | req->task_work.func(&req->task_work); |
2234 | node = next; | |
65453d1e JA |
2235 | |
2236 | if (!ctx) { | |
2237 | ctx = this_ctx; | |
2238 | } else if (ctx != this_ctx) { | |
2239 | mutex_lock(&ctx->uring_lock); | |
2240 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2241 | mutex_unlock(&ctx->uring_lock); | |
2242 | ctx = this_ctx; | |
2243 | } | |
2244 | } | |
2245 | ||
2246 | if (ctx && ctx->submit_state.comp.nr) { | |
2247 | mutex_lock(&ctx->uring_lock); | |
2248 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
2249 | mutex_unlock(&ctx->uring_lock); | |
7cbf1722 JA |
2250 | } |
2251 | ||
2252 | return list.first != NULL; | |
c2c4c83c JA |
2253 | } |
2254 | ||
7cbf1722 | 2255 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 2256 | { |
7cbf1722 | 2257 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 2258 | |
7cbf1722 JA |
2259 | while (__tctx_task_work(tctx)) |
2260 | cond_resched(); | |
2261 | ||
2262 | clear_bit(0, &tctx->task_state); | |
2263 | } | |
2264 | ||
2265 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
2266 | enum task_work_notify_mode notify) | |
2267 | { | |
2268 | struct io_uring_task *tctx = tsk->io_uring; | |
2269 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 2270 | unsigned long flags; |
7cbf1722 JA |
2271 | int ret; |
2272 | ||
2273 | WARN_ON_ONCE(!tctx); | |
2274 | ||
0b81e80c | 2275 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 2276 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 2277 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2278 | |
2279 | /* task_work already pending, we're done */ | |
2280 | if (test_bit(0, &tctx->task_state) || | |
2281 | test_and_set_bit(0, &tctx->task_state)) | |
2282 | return 0; | |
2283 | ||
2284 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
2285 | return 0; | |
2286 | ||
2287 | /* | |
2288 | * Slow path - we failed, find and delete work. if the work is not | |
2289 | * in the list, it got run and we're fine. | |
2290 | */ | |
2291 | ret = 0; | |
0b81e80c | 2292 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
2293 | wq_list_for_each(node, prev, &tctx->task_list) { |
2294 | if (&req->io_task_work.node == node) { | |
2295 | wq_list_del(&tctx->task_list, node, prev); | |
2296 | ret = 1; | |
2297 | break; | |
2298 | } | |
2299 | } | |
0b81e80c | 2300 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
2301 | clear_bit(0, &tctx->task_state); |
2302 | return ret; | |
2303 | } | |
2304 | ||
355fb9e2 | 2305 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
2306 | { |
2307 | struct task_struct *tsk = req->task; | |
2308 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
2309 | enum task_work_notify_mode notify; |
2310 | int ret; | |
c2c4c83c | 2311 | |
6200b0ae JA |
2312 | if (tsk->flags & PF_EXITING) |
2313 | return -ESRCH; | |
2314 | ||
c2c4c83c | 2315 | /* |
0ba9c9ed JA |
2316 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
2317 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
2318 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
2319 | * will do the job. | |
c2c4c83c | 2320 | */ |
91989c70 | 2321 | notify = TWA_NONE; |
355fb9e2 | 2322 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
2323 | notify = TWA_SIGNAL; |
2324 | ||
7cbf1722 | 2325 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
2326 | if (!ret) |
2327 | wake_up_process(tsk); | |
0ba9c9ed | 2328 | |
c2c4c83c JA |
2329 | return ret; |
2330 | } | |
2331 | ||
eab30c4d | 2332 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 2333 | task_work_func_t cb) |
eab30c4d | 2334 | { |
7c25c0d1 JA |
2335 | struct io_ring_ctx *ctx = req->ctx; |
2336 | struct callback_head *head; | |
eab30c4d PB |
2337 | |
2338 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
2339 | do { |
2340 | head = READ_ONCE(ctx->exit_task_work); | |
2341 | req->task_work.next = head; | |
2342 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
2343 | } |
2344 | ||
c40f6379 JA |
2345 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2346 | { | |
2347 | struct io_ring_ctx *ctx = req->ctx; | |
2348 | ||
2349 | spin_lock_irq(&ctx->completion_lock); | |
2350 | io_cqring_fill_event(req, error); | |
2351 | io_commit_cqring(ctx); | |
2352 | spin_unlock_irq(&ctx->completion_lock); | |
2353 | ||
2354 | io_cqring_ev_posted(ctx); | |
2355 | req_set_fail_links(req); | |
2356 | io_double_put_req(req); | |
2357 | } | |
2358 | ||
2359 | static void io_req_task_cancel(struct callback_head *cb) | |
2360 | { | |
2361 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2362 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2363 | |
792bb6eb | 2364 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 2365 | __io_req_task_cancel(req, req->result); |
792bb6eb | 2366 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 2367 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
2368 | } |
2369 | ||
2370 | static void __io_req_task_submit(struct io_kiocb *req) | |
2371 | { | |
2372 | struct io_ring_ctx *ctx = req->ctx; | |
2373 | ||
04fc6c80 | 2374 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2375 | mutex_lock(&ctx->uring_lock); |
dc0eced5 PB |
2376 | if (!ctx->sqo_dead && !(current->flags & PF_EXITING) && |
2377 | !io_sq_thread_acquire_mm_files(ctx, req)) | |
c5eef2b9 | 2378 | __io_queue_sqe(req); |
81b6d05c | 2379 | else |
c40f6379 | 2380 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2381 | mutex_unlock(&ctx->uring_lock); |
aec18a57 PB |
2382 | |
2383 | if (ctx->flags & IORING_SETUP_SQPOLL) | |
2384 | io_sq_thread_drop_mm_files(); | |
c40f6379 JA |
2385 | } |
2386 | ||
2387 | static void io_req_task_submit(struct callback_head *cb) | |
2388 | { | |
2389 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2390 | ||
2391 | __io_req_task_submit(req); | |
2392 | } | |
2393 | ||
2394 | static void io_req_task_queue(struct io_kiocb *req) | |
2395 | { | |
c40f6379 JA |
2396 | int ret; |
2397 | ||
7cbf1722 | 2398 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2399 | ret = io_req_task_work_add(req); |
c40f6379 | 2400 | if (unlikely(ret)) { |
a3df7698 | 2401 | req->result = -ECANCELED; |
04fc6c80 | 2402 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 2403 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 2404 | } |
c40f6379 JA |
2405 | } |
2406 | ||
a3df7698 PB |
2407 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2408 | { | |
2409 | percpu_ref_get(&req->ctx->refs); | |
2410 | req->result = ret; | |
2411 | req->task_work.func = io_req_task_cancel; | |
2412 | ||
2413 | if (unlikely(io_req_task_work_add(req))) | |
2414 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2415 | } | |
2416 | ||
f2f87370 | 2417 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2418 | { |
9b5f7bd9 | 2419 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2420 | |
2421 | if (nxt) | |
906a8c3f | 2422 | io_req_task_queue(nxt); |
c69f8dbe JL |
2423 | } |
2424 | ||
c3524383 | 2425 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2426 | { |
c3524383 PB |
2427 | io_queue_next(req); |
2428 | __io_free_req(req); | |
2429 | } | |
8766dd51 | 2430 | |
2d6500d4 | 2431 | struct req_batch { |
5af1d13e PB |
2432 | struct task_struct *task; |
2433 | int task_refs; | |
1b4c351f | 2434 | int ctx_refs; |
2d6500d4 PB |
2435 | }; |
2436 | ||
5af1d13e PB |
2437 | static inline void io_init_req_batch(struct req_batch *rb) |
2438 | { | |
5af1d13e | 2439 | rb->task_refs = 0; |
9ae72463 | 2440 | rb->ctx_refs = 0; |
5af1d13e PB |
2441 | rb->task = NULL; |
2442 | } | |
2443 | ||
2d6500d4 PB |
2444 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2445 | struct req_batch *rb) | |
2446 | { | |
6e833d53 | 2447 | if (rb->task) |
7c660731 | 2448 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2449 | if (rb->ctx_refs) |
2450 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2451 | } |
2452 | ||
6ff119a6 PB |
2453 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2454 | struct io_submit_state *state) | |
2d6500d4 | 2455 | { |
f2f87370 | 2456 | io_queue_next(req); |
2d6500d4 | 2457 | |
e3bc8e9d | 2458 | if (req->task != rb->task) { |
7c660731 PB |
2459 | if (rb->task) |
2460 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2461 | rb->task = req->task; |
2462 | rb->task_refs = 0; | |
5af1d13e | 2463 | } |
e3bc8e9d | 2464 | rb->task_refs++; |
9ae72463 | 2465 | rb->ctx_refs++; |
5af1d13e | 2466 | |
4edf20f9 | 2467 | io_dismantle_req(req); |
bd759045 | 2468 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2469 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2470 | else |
2471 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2472 | } |
2473 | ||
905c172f PB |
2474 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2475 | struct io_ring_ctx *ctx) | |
2476 | { | |
2477 | int i, nr = cs->nr; | |
2478 | struct io_kiocb *req; | |
2479 | struct req_batch rb; | |
2480 | ||
2481 | io_init_req_batch(&rb); | |
2482 | spin_lock_irq(&ctx->completion_lock); | |
2483 | for (i = 0; i < nr; i++) { | |
2484 | req = cs->reqs[i]; | |
2485 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2486 | } | |
2487 | io_commit_cqring(ctx); | |
2488 | spin_unlock_irq(&ctx->completion_lock); | |
2489 | ||
2490 | io_cqring_ev_posted(ctx); | |
2491 | for (i = 0; i < nr; i++) { | |
2492 | req = cs->reqs[i]; | |
2493 | ||
2494 | /* submission and completion refs */ | |
2495 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2496 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2497 | } |
2498 | ||
2499 | io_req_free_batch_finish(ctx, &rb); | |
2500 | cs->nr = 0; | |
7a743e22 PB |
2501 | } |
2502 | ||
ba816ad6 JA |
2503 | /* |
2504 | * Drop reference to request, return next in chain (if there is one) if this | |
2505 | * was the last reference to this request. | |
2506 | */ | |
9b5f7bd9 | 2507 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2508 | { |
9b5f7bd9 PB |
2509 | struct io_kiocb *nxt = NULL; |
2510 | ||
2a44f467 | 2511 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2512 | nxt = io_req_find_next(req); |
4d7dd462 | 2513 | __io_free_req(req); |
2a44f467 | 2514 | } |
9b5f7bd9 | 2515 | return nxt; |
2b188cc1 JA |
2516 | } |
2517 | ||
e65ef56d JA |
2518 | static void io_put_req(struct io_kiocb *req) |
2519 | { | |
2520 | if (refcount_dec_and_test(&req->refs)) | |
2521 | io_free_req(req); | |
2b188cc1 JA |
2522 | } |
2523 | ||
216578e5 PB |
2524 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2525 | { | |
2526 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2527 | ||
2528 | io_free_req(req); | |
2529 | } | |
2530 | ||
2531 | static void io_free_req_deferred(struct io_kiocb *req) | |
2532 | { | |
2533 | int ret; | |
2534 | ||
7cbf1722 | 2535 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2536 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2537 | if (unlikely(ret)) |
2538 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2539 | } |
2540 | ||
2541 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2542 | { | |
2543 | if (refcount_sub_and_test(refs, &req->refs)) | |
2544 | io_free_req_deferred(req); | |
2545 | } | |
2546 | ||
978db57e JA |
2547 | static void io_double_put_req(struct io_kiocb *req) |
2548 | { | |
2549 | /* drop both submit and complete references */ | |
2550 | if (refcount_sub_and_test(2, &req->refs)) | |
2551 | io_free_req(req); | |
2552 | } | |
2553 | ||
6c503150 | 2554 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2555 | { |
2556 | /* See comment at the top of this file */ | |
2557 | smp_rmb(); | |
e23de15f | 2558 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2559 | } |
2560 | ||
fb5ccc98 PB |
2561 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2562 | { | |
2563 | struct io_rings *rings = ctx->rings; | |
2564 | ||
2565 | /* make sure SQ entry isn't read before tail */ | |
2566 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2567 | } | |
2568 | ||
8ff069bf | 2569 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2570 | { |
8ff069bf | 2571 | unsigned int cflags; |
e94f141b | 2572 | |
bcda7baa JA |
2573 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2574 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2575 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2576 | kfree(kbuf); |
2577 | return cflags; | |
e94f141b JA |
2578 | } |
2579 | ||
8ff069bf | 2580 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2581 | { |
4d954c25 | 2582 | struct io_buffer *kbuf; |
bcda7baa | 2583 | |
4d954c25 | 2584 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2585 | return io_put_kbuf(req, kbuf); |
2586 | } | |
2587 | ||
4c6e277c JA |
2588 | static inline bool io_run_task_work(void) |
2589 | { | |
6200b0ae JA |
2590 | /* |
2591 | * Not safe to run on exiting task, and the task_work handling will | |
2592 | * not add work to such a task. | |
2593 | */ | |
2594 | if (unlikely(current->flags & PF_EXITING)) | |
2595 | return false; | |
4c6e277c JA |
2596 | if (current->task_works) { |
2597 | __set_current_state(TASK_RUNNING); | |
2598 | task_work_run(); | |
2599 | return true; | |
2600 | } | |
2601 | ||
2602 | return false; | |
bcda7baa JA |
2603 | } |
2604 | ||
def596e9 JA |
2605 | /* |
2606 | * Find and free completed poll iocbs | |
2607 | */ | |
2608 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2609 | struct list_head *done) | |
2610 | { | |
8237e045 | 2611 | struct req_batch rb; |
def596e9 | 2612 | struct io_kiocb *req; |
bbde017a XW |
2613 | |
2614 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2615 | smp_rmb(); | |
def596e9 | 2616 | |
5af1d13e | 2617 | io_init_req_batch(&rb); |
def596e9 | 2618 | while (!list_empty(done)) { |
bcda7baa JA |
2619 | int cflags = 0; |
2620 | ||
d21ffe7e | 2621 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2622 | list_del(&req->inflight_entry); |
2623 | ||
bbde017a XW |
2624 | if (READ_ONCE(req->result) == -EAGAIN) { |
2625 | req->iopoll_completed = 0; | |
23faba36 | 2626 | if (io_rw_reissue(req)) |
f161340d | 2627 | continue; |
bbde017a | 2628 | } |
def596e9 | 2629 | |
bcda7baa | 2630 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2631 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2632 | |
2633 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2634 | (*nr_events)++; |
2635 | ||
c3524383 | 2636 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2637 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2638 | } |
def596e9 | 2639 | |
09bb8394 | 2640 | io_commit_cqring(ctx); |
80c18e4a | 2641 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2642 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2643 | } |
2644 | ||
def596e9 JA |
2645 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2646 | long min) | |
2647 | { | |
2648 | struct io_kiocb *req, *tmp; | |
2649 | LIST_HEAD(done); | |
2650 | bool spin; | |
2651 | int ret; | |
2652 | ||
2653 | /* | |
2654 | * Only spin for completions if we don't have multiple devices hanging | |
2655 | * off our complete list, and we're under the requested amount. | |
2656 | */ | |
2657 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2658 | ||
2659 | ret = 0; | |
d21ffe7e | 2660 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2661 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2662 | |
2663 | /* | |
581f9810 BM |
2664 | * Move completed and retryable entries to our local lists. |
2665 | * If we find a request that requires polling, break out | |
2666 | * and complete those lists first, if we have entries there. | |
def596e9 | 2667 | */ |
65a6543d | 2668 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2669 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2670 | continue; |
2671 | } | |
2672 | if (!list_empty(&done)) | |
2673 | break; | |
2674 | ||
2675 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2676 | if (ret < 0) | |
2677 | break; | |
2678 | ||
3aadc23e PB |
2679 | /* iopoll may have completed current req */ |
2680 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2681 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2682 | |
def596e9 JA |
2683 | if (ret && spin) |
2684 | spin = false; | |
2685 | ret = 0; | |
2686 | } | |
2687 | ||
2688 | if (!list_empty(&done)) | |
2689 | io_iopoll_complete(ctx, nr_events, &done); | |
2690 | ||
2691 | return ret; | |
2692 | } | |
2693 | ||
2694 | /* | |
d195a66e | 2695 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2696 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2697 | * as a non-spinning completion check. | |
2698 | */ | |
2699 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2700 | long min) | |
2701 | { | |
540e32a0 | 2702 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2703 | int ret; |
2704 | ||
2705 | ret = io_do_iopoll(ctx, nr_events, min); | |
2706 | if (ret < 0) | |
2707 | return ret; | |
eba0a4dd | 2708 | if (*nr_events >= min) |
def596e9 JA |
2709 | return 0; |
2710 | } | |
2711 | ||
2712 | return 1; | |
2713 | } | |
2714 | ||
2715 | /* | |
2716 | * We can't just wait for polled events to come to us, we have to actively | |
2717 | * find and complete them. | |
2718 | */ | |
b2edc0a7 | 2719 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2720 | { |
2721 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2722 | return; | |
2723 | ||
2724 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2725 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2726 | unsigned int nr_events = 0; |
2727 | ||
b2edc0a7 | 2728 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2729 | |
b2edc0a7 PB |
2730 | /* let it sleep and repeat later if can't complete a request */ |
2731 | if (nr_events == 0) | |
2732 | break; | |
08f5439f JA |
2733 | /* |
2734 | * Ensure we allow local-to-the-cpu processing to take place, | |
2735 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2736 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2737 | */ |
3fcee5a6 PB |
2738 | if (need_resched()) { |
2739 | mutex_unlock(&ctx->uring_lock); | |
2740 | cond_resched(); | |
2741 | mutex_lock(&ctx->uring_lock); | |
2742 | } | |
def596e9 JA |
2743 | } |
2744 | mutex_unlock(&ctx->uring_lock); | |
2745 | } | |
2746 | ||
7668b92a | 2747 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2748 | { |
7668b92a | 2749 | unsigned int nr_events = 0; |
2b2ed975 | 2750 | int iters = 0, ret = 0; |
500f9fba | 2751 | |
c7849be9 XW |
2752 | /* |
2753 | * We disallow the app entering submit/complete with polling, but we | |
2754 | * still need to lock the ring to prevent racing with polled issue | |
2755 | * that got punted to a workqueue. | |
2756 | */ | |
2757 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2758 | do { |
a3a0e43f JA |
2759 | /* |
2760 | * Don't enter poll loop if we already have events pending. | |
2761 | * If we do, we can potentially be spinning for commands that | |
2762 | * already triggered a CQE (eg in error). | |
2763 | */ | |
6c503150 PB |
2764 | if (test_bit(0, &ctx->cq_check_overflow)) |
2765 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2766 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2767 | break; |
2768 | ||
500f9fba JA |
2769 | /* |
2770 | * If a submit got punted to a workqueue, we can have the | |
2771 | * application entering polling for a command before it gets | |
2772 | * issued. That app will hold the uring_lock for the duration | |
2773 | * of the poll right here, so we need to take a breather every | |
2774 | * now and then to ensure that the issue has a chance to add | |
2775 | * the poll to the issued list. Otherwise we can spin here | |
2776 | * forever, while the workqueue is stuck trying to acquire the | |
2777 | * very same mutex. | |
2778 | */ | |
2779 | if (!(++iters & 7)) { | |
2780 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2781 | io_run_task_work(); |
500f9fba JA |
2782 | mutex_lock(&ctx->uring_lock); |
2783 | } | |
2784 | ||
7668b92a | 2785 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2786 | if (ret <= 0) |
2787 | break; | |
2788 | ret = 0; | |
7668b92a | 2789 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2790 | |
500f9fba | 2791 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2792 | return ret; |
2793 | } | |
2794 | ||
491381ce | 2795 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2796 | { |
491381ce JA |
2797 | /* |
2798 | * Tell lockdep we inherited freeze protection from submission | |
2799 | * thread. | |
2800 | */ | |
2801 | if (req->flags & REQ_F_ISREG) { | |
2802 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2803 | |
491381ce | 2804 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2805 | } |
491381ce | 2806 | file_end_write(req->file); |
2b188cc1 JA |
2807 | } |
2808 | ||
b63534c4 | 2809 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2810 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2811 | { |
2812 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2813 | int rw, ret; |
b63534c4 | 2814 | struct iov_iter iter; |
b63534c4 | 2815 | |
dc2a6e9a PB |
2816 | /* already prepared */ |
2817 | if (req->async_data) | |
2818 | return true; | |
b63534c4 JA |
2819 | |
2820 | switch (req->opcode) { | |
2821 | case IORING_OP_READV: | |
2822 | case IORING_OP_READ_FIXED: | |
2823 | case IORING_OP_READ: | |
2824 | rw = READ; | |
2825 | break; | |
2826 | case IORING_OP_WRITEV: | |
2827 | case IORING_OP_WRITE_FIXED: | |
2828 | case IORING_OP_WRITE: | |
2829 | rw = WRITE; | |
2830 | break; | |
2831 | default: | |
2832 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2833 | req->opcode); | |
dc2a6e9a | 2834 | return false; |
b63534c4 JA |
2835 | } |
2836 | ||
dc2a6e9a PB |
2837 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2838 | if (ret < 0) | |
2839 | return false; | |
6bf985dc | 2840 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2841 | } |
b63534c4 JA |
2842 | #endif |
2843 | ||
23faba36 | 2844 | static bool io_rw_reissue(struct io_kiocb *req) |
b63534c4 JA |
2845 | { |
2846 | #ifdef CONFIG_BLOCK | |
355afaeb | 2847 | umode_t mode = file_inode(req->file)->i_mode; |
b63534c4 JA |
2848 | int ret; |
2849 | ||
355afaeb JA |
2850 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2851 | return false; | |
75c668cd | 2852 | if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker()) |
b63534c4 JA |
2853 | return false; |
2854 | ||
55e6ac1e PB |
2855 | lockdep_assert_held(&req->ctx->uring_lock); |
2856 | ||
28cea78a | 2857 | ret = io_sq_thread_acquire_mm_files(req->ctx, req); |
6d816e08 | 2858 | |
dc2a6e9a | 2859 | if (!ret && io_resubmit_prep(req)) { |
fdee946d JA |
2860 | refcount_inc(&req->refs); |
2861 | io_queue_async_work(req); | |
b63534c4 | 2862 | return true; |
fdee946d | 2863 | } |
dc2a6e9a | 2864 | req_set_fail_links(req); |
b63534c4 JA |
2865 | #endif |
2866 | return false; | |
2867 | } | |
2868 | ||
a1d7c393 | 2869 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2870 | unsigned int issue_flags) |
a1d7c393 | 2871 | { |
2f8e45f1 PB |
2872 | int cflags = 0; |
2873 | ||
23faba36 PB |
2874 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2875 | return; | |
2f8e45f1 PB |
2876 | if (res != req->result) |
2877 | req_set_fail_links(req); | |
23faba36 | 2878 | |
2f8e45f1 PB |
2879 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2880 | kiocb_end_write(req); | |
2881 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2882 | cflags = io_put_rw_kbuf(req); | |
2883 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2884 | } |
2885 | ||
2886 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2887 | { | |
9adbd45d | 2888 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2889 | |
889fca73 | 2890 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2891 | } |
2892 | ||
def596e9 JA |
2893 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2894 | { | |
9adbd45d | 2895 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2896 | |
491381ce JA |
2897 | if (kiocb->ki_flags & IOCB_WRITE) |
2898 | kiocb_end_write(req); | |
def596e9 | 2899 | |
2d7d6792 | 2900 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2901 | req_set_fail_links(req); |
bbde017a XW |
2902 | |
2903 | WRITE_ONCE(req->result, res); | |
2904 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2905 | smp_wmb(); |
2906 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2907 | } |
2908 | ||
2909 | /* | |
2910 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2911 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2912 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2913 | * accessing the kiocb cookie. | |
2914 | */ | |
2e9dbe90 | 2915 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2916 | { |
2917 | struct io_ring_ctx *ctx = req->ctx; | |
2918 | ||
2919 | /* | |
2920 | * Track whether we have multiple files in our lists. This will impact | |
2921 | * how we do polling eventually, not spinning if we're on potentially | |
2922 | * different devices. | |
2923 | */ | |
540e32a0 | 2924 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2925 | ctx->poll_multi_file = false; |
2926 | } else if (!ctx->poll_multi_file) { | |
2927 | struct io_kiocb *list_req; | |
2928 | ||
540e32a0 | 2929 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2930 | inflight_entry); |
9adbd45d | 2931 | if (list_req->file != req->file) |
def596e9 JA |
2932 | ctx->poll_multi_file = true; |
2933 | } | |
2934 | ||
2935 | /* | |
2936 | * For fast devices, IO may have already completed. If it has, add | |
2937 | * it to the front so we find it first. | |
2938 | */ | |
65a6543d | 2939 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2940 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2941 | else |
d21ffe7e | 2942 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2943 | |
2e9dbe90 XW |
2944 | /* |
2945 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2946 | * task context or in io worker task context. If current task context is | |
2947 | * sq thread, we don't need to check whether should wake up sq thread. | |
2948 | */ | |
2949 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2950 | wq_has_sleeper(&ctx->sq_data->wait)) |
2951 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2952 | } |
2953 | ||
9f13c35b PB |
2954 | static inline void io_state_file_put(struct io_submit_state *state) |
2955 | { | |
02b23a9a PB |
2956 | if (state->file_refs) { |
2957 | fput_many(state->file, state->file_refs); | |
2958 | state->file_refs = 0; | |
2959 | } | |
9a56a232 JA |
2960 | } |
2961 | ||
2962 | /* | |
2963 | * Get as many references to a file as we have IOs left in this submission, | |
2964 | * assuming most submissions are for one file, or at least that each file | |
2965 | * has more than one submission. | |
2966 | */ | |
8da11c19 | 2967 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2968 | { |
2969 | if (!state) | |
2970 | return fget(fd); | |
2971 | ||
6e1271e6 | 2972 | if (state->file_refs) { |
9a56a232 | 2973 | if (state->fd == fd) { |
6e1271e6 | 2974 | state->file_refs--; |
9a56a232 JA |
2975 | return state->file; |
2976 | } | |
02b23a9a | 2977 | io_state_file_put(state); |
9a56a232 JA |
2978 | } |
2979 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2980 | if (unlikely(!state->file)) |
9a56a232 JA |
2981 | return NULL; |
2982 | ||
2983 | state->fd = fd; | |
6e1271e6 | 2984 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2985 | return state->file; |
2986 | } | |
2987 | ||
4503b767 JA |
2988 | static bool io_bdev_nowait(struct block_device *bdev) |
2989 | { | |
9ba0d0c8 | 2990 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2991 | } |
2992 | ||
2b188cc1 JA |
2993 | /* |
2994 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2995 | * any file. For now, just ensure that anything potentially problematic is done | |
2996 | * inline. | |
2997 | */ | |
af197f50 | 2998 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2999 | { |
3000 | umode_t mode = file_inode(file)->i_mode; | |
3001 | ||
4503b767 | 3002 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
3003 | if (IS_ENABLED(CONFIG_BLOCK) && |
3004 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
3005 | return true; |
3006 | return false; | |
3007 | } | |
3008 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 3009 | return true; |
4503b767 | 3010 | if (S_ISREG(mode)) { |
4e7b5671 CH |
3011 | if (IS_ENABLED(CONFIG_BLOCK) && |
3012 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
3013 | file->f_op != &io_uring_fops) |
3014 | return true; | |
3015 | return false; | |
3016 | } | |
2b188cc1 | 3017 | |
c5b85625 JA |
3018 | /* any ->read/write should understand O_NONBLOCK */ |
3019 | if (file->f_flags & O_NONBLOCK) | |
3020 | return true; | |
3021 | ||
af197f50 JA |
3022 | if (!(file->f_mode & FMODE_NOWAIT)) |
3023 | return false; | |
3024 | ||
3025 | if (rw == READ) | |
3026 | return file->f_op->read_iter != NULL; | |
3027 | ||
3028 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
3029 | } |
3030 | ||
a88fc400 | 3031 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 3032 | { |
def596e9 | 3033 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 3034 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 3035 | struct file *file = req->file; |
09bb8394 JA |
3036 | unsigned ioprio; |
3037 | int ret; | |
2b188cc1 | 3038 | |
75c668cd | 3039 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
3040 | req->flags |= REQ_F_ISREG; |
3041 | ||
2b188cc1 | 3042 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 3043 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 3044 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 3045 | kiocb->ki_pos = file->f_pos; |
ba04291e | 3046 | } |
2b188cc1 | 3047 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
3048 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
3049 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
3050 | if (unlikely(ret)) | |
3051 | return ret; | |
2b188cc1 | 3052 | |
75c668cd PB |
3053 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
3054 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
3055 | req->flags |= REQ_F_NOWAIT; | |
3056 | ||
2b188cc1 JA |
3057 | ioprio = READ_ONCE(sqe->ioprio); |
3058 | if (ioprio) { | |
3059 | ret = ioprio_check_cap(ioprio); | |
3060 | if (ret) | |
09bb8394 | 3061 | return ret; |
2b188cc1 JA |
3062 | |
3063 | kiocb->ki_ioprio = ioprio; | |
3064 | } else | |
3065 | kiocb->ki_ioprio = get_current_ioprio(); | |
3066 | ||
def596e9 | 3067 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
3068 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
3069 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 3070 | return -EOPNOTSUPP; |
2b188cc1 | 3071 | |
def596e9 JA |
3072 | kiocb->ki_flags |= IOCB_HIPRI; |
3073 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 3074 | req->iopoll_completed = 0; |
def596e9 | 3075 | } else { |
09bb8394 JA |
3076 | if (kiocb->ki_flags & IOCB_HIPRI) |
3077 | return -EINVAL; | |
def596e9 JA |
3078 | kiocb->ki_complete = io_complete_rw; |
3079 | } | |
9adbd45d | 3080 | |
3529d8c2 JA |
3081 | req->rw.addr = READ_ONCE(sqe->addr); |
3082 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 3083 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 3084 | return 0; |
2b188cc1 JA |
3085 | } |
3086 | ||
3087 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
3088 | { | |
3089 | switch (ret) { | |
3090 | case -EIOCBQUEUED: | |
3091 | break; | |
3092 | case -ERESTARTSYS: | |
3093 | case -ERESTARTNOINTR: | |
3094 | case -ERESTARTNOHAND: | |
3095 | case -ERESTART_RESTARTBLOCK: | |
3096 | /* | |
3097 | * We can't just restart the syscall, since previously | |
3098 | * submitted sqes may already be in progress. Just fail this | |
3099 | * IO with EINTR. | |
3100 | */ | |
3101 | ret = -EINTR; | |
df561f66 | 3102 | fallthrough; |
2b188cc1 JA |
3103 | default: |
3104 | kiocb->ki_complete(kiocb, ret, 0); | |
3105 | } | |
3106 | } | |
3107 | ||
a1d7c393 | 3108 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 3109 | unsigned int issue_flags) |
ba816ad6 | 3110 | { |
ba04291e | 3111 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 3112 | struct io_async_rw *io = req->async_data; |
ba04291e | 3113 | |
227c0c96 | 3114 | /* add previously done IO, if any */ |
e8c2bc1f | 3115 | if (io && io->bytes_done > 0) { |
227c0c96 | 3116 | if (ret < 0) |
e8c2bc1f | 3117 | ret = io->bytes_done; |
227c0c96 | 3118 | else |
e8c2bc1f | 3119 | ret += io->bytes_done; |
227c0c96 JA |
3120 | } |
3121 | ||
ba04291e JA |
3122 | if (req->flags & REQ_F_CUR_POS) |
3123 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 3124 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 3125 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
3126 | else |
3127 | io_rw_done(kiocb, ret); | |
3128 | } | |
3129 | ||
847595de | 3130 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 3131 | { |
9adbd45d JA |
3132 | struct io_ring_ctx *ctx = req->ctx; |
3133 | size_t len = req->rw.len; | |
edafccee | 3134 | struct io_mapped_ubuf *imu; |
4be1c615 | 3135 | u16 index, buf_index = req->buf_index; |
edafccee JA |
3136 | size_t offset; |
3137 | u64 buf_addr; | |
3138 | ||
edafccee JA |
3139 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
3140 | return -EFAULT; | |
edafccee JA |
3141 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
3142 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 3143 | buf_addr = req->rw.addr; |
edafccee JA |
3144 | |
3145 | /* overflow */ | |
3146 | if (buf_addr + len < buf_addr) | |
3147 | return -EFAULT; | |
3148 | /* not inside the mapped region */ | |
3149 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
3150 | return -EFAULT; | |
3151 | ||
3152 | /* | |
3153 | * May not be a start of buffer, set size appropriately | |
3154 | * and advance us to the beginning. | |
3155 | */ | |
3156 | offset = buf_addr - imu->ubuf; | |
3157 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
3158 | |
3159 | if (offset) { | |
3160 | /* | |
3161 | * Don't use iov_iter_advance() here, as it's really slow for | |
3162 | * using the latter parts of a big fixed buffer - it iterates | |
3163 | * over each segment manually. We can cheat a bit here, because | |
3164 | * we know that: | |
3165 | * | |
3166 | * 1) it's a BVEC iter, we set it up | |
3167 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
3168 | * first and last bvec | |
3169 | * | |
3170 | * So just find our index, and adjust the iterator afterwards. | |
3171 | * If the offset is within the first bvec (or the whole first | |
3172 | * bvec, just use iov_iter_advance(). This makes it easier | |
3173 | * since we can just skip the first segment, which may not | |
3174 | * be PAGE_SIZE aligned. | |
3175 | */ | |
3176 | const struct bio_vec *bvec = imu->bvec; | |
3177 | ||
3178 | if (offset <= bvec->bv_len) { | |
3179 | iov_iter_advance(iter, offset); | |
3180 | } else { | |
3181 | unsigned long seg_skip; | |
3182 | ||
3183 | /* skip first vec */ | |
3184 | offset -= bvec->bv_len; | |
3185 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
3186 | ||
3187 | iter->bvec = bvec + seg_skip; | |
3188 | iter->nr_segs -= seg_skip; | |
99c79f66 | 3189 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 3190 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
3191 | } |
3192 | } | |
3193 | ||
847595de | 3194 | return 0; |
edafccee JA |
3195 | } |
3196 | ||
bcda7baa JA |
3197 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
3198 | { | |
3199 | if (needs_lock) | |
3200 | mutex_unlock(&ctx->uring_lock); | |
3201 | } | |
3202 | ||
3203 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
3204 | { | |
3205 | /* | |
3206 | * "Normal" inline submissions always hold the uring_lock, since we | |
3207 | * grab it from the system call. Same is true for the SQPOLL offload. | |
3208 | * The only exception is when we've detached the request and issue it | |
3209 | * from an async worker thread, grab the lock for that case. | |
3210 | */ | |
3211 | if (needs_lock) | |
3212 | mutex_lock(&ctx->uring_lock); | |
3213 | } | |
3214 | ||
3215 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
3216 | int bgid, struct io_buffer *kbuf, | |
3217 | bool needs_lock) | |
3218 | { | |
3219 | struct io_buffer *head; | |
3220 | ||
3221 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
3222 | return kbuf; | |
3223 | ||
3224 | io_ring_submit_lock(req->ctx, needs_lock); | |
3225 | ||
3226 | lockdep_assert_held(&req->ctx->uring_lock); | |
3227 | ||
3228 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
3229 | if (head) { | |
3230 | if (!list_empty(&head->list)) { | |
3231 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
3232 | list); | |
3233 | list_del(&kbuf->list); | |
3234 | } else { | |
3235 | kbuf = head; | |
3236 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
3237 | } | |
3238 | if (*len > kbuf->len) | |
3239 | *len = kbuf->len; | |
3240 | } else { | |
3241 | kbuf = ERR_PTR(-ENOBUFS); | |
3242 | } | |
3243 | ||
3244 | io_ring_submit_unlock(req->ctx, needs_lock); | |
3245 | ||
3246 | return kbuf; | |
3247 | } | |
3248 | ||
4d954c25 JA |
3249 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
3250 | bool needs_lock) | |
3251 | { | |
3252 | struct io_buffer *kbuf; | |
4f4eeba8 | 3253 | u16 bgid; |
4d954c25 JA |
3254 | |
3255 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 3256 | bgid = req->buf_index; |
4d954c25 JA |
3257 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
3258 | if (IS_ERR(kbuf)) | |
3259 | return kbuf; | |
3260 | req->rw.addr = (u64) (unsigned long) kbuf; | |
3261 | req->flags |= REQ_F_BUFFER_SELECTED; | |
3262 | return u64_to_user_ptr(kbuf->addr); | |
3263 | } | |
3264 | ||
3265 | #ifdef CONFIG_COMPAT | |
3266 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
3267 | bool needs_lock) | |
3268 | { | |
3269 | struct compat_iovec __user *uiov; | |
3270 | compat_ssize_t clen; | |
3271 | void __user *buf; | |
3272 | ssize_t len; | |
3273 | ||
3274 | uiov = u64_to_user_ptr(req->rw.addr); | |
3275 | if (!access_ok(uiov, sizeof(*uiov))) | |
3276 | return -EFAULT; | |
3277 | if (__get_user(clen, &uiov->iov_len)) | |
3278 | return -EFAULT; | |
3279 | if (clen < 0) | |
3280 | return -EINVAL; | |
3281 | ||
3282 | len = clen; | |
3283 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3284 | if (IS_ERR(buf)) | |
3285 | return PTR_ERR(buf); | |
3286 | iov[0].iov_base = buf; | |
3287 | iov[0].iov_len = (compat_size_t) len; | |
3288 | return 0; | |
3289 | } | |
3290 | #endif | |
3291 | ||
3292 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3293 | bool needs_lock) | |
3294 | { | |
3295 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
3296 | void __user *buf; | |
3297 | ssize_t len; | |
3298 | ||
3299 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
3300 | return -EFAULT; | |
3301 | ||
3302 | len = iov[0].iov_len; | |
3303 | if (len < 0) | |
3304 | return -EINVAL; | |
3305 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
3306 | if (IS_ERR(buf)) | |
3307 | return PTR_ERR(buf); | |
3308 | iov[0].iov_base = buf; | |
3309 | iov[0].iov_len = len; | |
3310 | return 0; | |
3311 | } | |
3312 | ||
3313 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
3314 | bool needs_lock) | |
3315 | { | |
dddb3e26 JA |
3316 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3317 | struct io_buffer *kbuf; | |
3318 | ||
3319 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3320 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3321 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3322 | return 0; |
dddb3e26 | 3323 | } |
dd201662 | 3324 | if (req->rw.len != 1) |
4d954c25 JA |
3325 | return -EINVAL; |
3326 | ||
3327 | #ifdef CONFIG_COMPAT | |
3328 | if (req->ctx->compat) | |
3329 | return io_compat_import(req, iov, needs_lock); | |
3330 | #endif | |
3331 | ||
3332 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3333 | } | |
3334 | ||
847595de PB |
3335 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3336 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3337 | { |
9adbd45d JA |
3338 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3339 | size_t sqe_len = req->rw.len; | |
847595de | 3340 | u8 opcode = req->opcode; |
4d954c25 | 3341 | ssize_t ret; |
edafccee | 3342 | |
7d009165 | 3343 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3344 | *iovec = NULL; |
9adbd45d | 3345 | return io_import_fixed(req, rw, iter); |
edafccee | 3346 | } |
2b188cc1 | 3347 | |
bcda7baa | 3348 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3349 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3350 | return -EINVAL; |
3351 | ||
3a6820f2 | 3352 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3353 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3354 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3355 | if (IS_ERR(buf)) |
4d954c25 | 3356 | return PTR_ERR(buf); |
3f9d6441 | 3357 | req->rw.len = sqe_len; |
bcda7baa JA |
3358 | } |
3359 | ||
3a6820f2 JA |
3360 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3361 | *iovec = NULL; | |
10fc72e4 | 3362 | return ret; |
3a6820f2 JA |
3363 | } |
3364 | ||
4d954c25 JA |
3365 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3366 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3367 | if (!ret) |
3368 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3369 | *iovec = NULL; |
3370 | return ret; | |
3371 | } | |
3372 | ||
89cd35c5 CH |
3373 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3374 | req->ctx->compat); | |
2b188cc1 JA |
3375 | } |
3376 | ||
0fef9483 JA |
3377 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3378 | { | |
5b09e37e | 3379 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3380 | } |
3381 | ||
31b51510 | 3382 | /* |
32960613 JA |
3383 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3384 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3385 | */ |
4017eb91 | 3386 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3387 | { |
4017eb91 JA |
3388 | struct kiocb *kiocb = &req->rw.kiocb; |
3389 | struct file *file = req->file; | |
32960613 JA |
3390 | ssize_t ret = 0; |
3391 | ||
3392 | /* | |
3393 | * Don't support polled IO through this interface, and we can't | |
3394 | * support non-blocking either. For the latter, this just causes | |
3395 | * the kiocb to be handled from an async context. | |
3396 | */ | |
3397 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3398 | return -EOPNOTSUPP; | |
3399 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3400 | return -EAGAIN; | |
3401 | ||
3402 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3403 | struct iovec iovec; |
32960613 JA |
3404 | ssize_t nr; |
3405 | ||
311ae9e1 PB |
3406 | if (!iov_iter_is_bvec(iter)) { |
3407 | iovec = iov_iter_iovec(iter); | |
3408 | } else { | |
4017eb91 JA |
3409 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3410 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3411 | } |
3412 | ||
32960613 JA |
3413 | if (rw == READ) { |
3414 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3415 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3416 | } else { |
3417 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3418 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3419 | } |
3420 | ||
3421 | if (nr < 0) { | |
3422 | if (!ret) | |
3423 | ret = nr; | |
3424 | break; | |
3425 | } | |
3426 | ret += nr; | |
3427 | if (nr != iovec.iov_len) | |
3428 | break; | |
4017eb91 JA |
3429 | req->rw.len -= nr; |
3430 | req->rw.addr += nr; | |
32960613 JA |
3431 | iov_iter_advance(iter, nr); |
3432 | } | |
3433 | ||
3434 | return ret; | |
3435 | } | |
3436 | ||
ff6165b2 JA |
3437 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3438 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3439 | { |
e8c2bc1f | 3440 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3441 | |
ff6165b2 | 3442 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3443 | rw->free_iovec = iovec; |
227c0c96 | 3444 | rw->bytes_done = 0; |
ff6165b2 | 3445 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3446 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3447 | return; |
b64e3444 | 3448 | if (!iovec) { |
ff6165b2 JA |
3449 | unsigned iov_off = 0; |
3450 | ||
3451 | rw->iter.iov = rw->fast_iov; | |
3452 | if (iter->iov != fast_iov) { | |
3453 | iov_off = iter->iov - fast_iov; | |
3454 | rw->iter.iov += iov_off; | |
3455 | } | |
3456 | if (rw->fast_iov != fast_iov) | |
3457 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3458 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3459 | } else { |
3460 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3461 | } |
3462 | } | |
3463 | ||
e8c2bc1f | 3464 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3465 | { |
e8c2bc1f JA |
3466 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3467 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3468 | return req->async_data == NULL; | |
3d9932a8 XW |
3469 | } |
3470 | ||
e8c2bc1f | 3471 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3472 | { |
e8c2bc1f | 3473 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3474 | return 0; |
3d9932a8 | 3475 | |
e8c2bc1f | 3476 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3477 | } |
3478 | ||
ff6165b2 JA |
3479 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3480 | const struct iovec *fast_iov, | |
227c0c96 | 3481 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3482 | { |
e8c2bc1f | 3483 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3484 | return 0; |
e8c2bc1f | 3485 | if (!req->async_data) { |
6bf985dc PB |
3486 | if (__io_alloc_async_data(req)) { |
3487 | kfree(iovec); | |
5d204bcf | 3488 | return -ENOMEM; |
6bf985dc | 3489 | } |
b7bb4f7d | 3490 | |
ff6165b2 | 3491 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3492 | } |
b7bb4f7d | 3493 | return 0; |
f67676d1 JA |
3494 | } |
3495 | ||
73debe68 | 3496 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3497 | { |
e8c2bc1f | 3498 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3499 | struct iovec *iov = iorw->fast_iov; |
847595de | 3500 | int ret; |
c3e330a4 | 3501 | |
2846c481 | 3502 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3503 | if (unlikely(ret < 0)) |
3504 | return ret; | |
3505 | ||
ab0b196c PB |
3506 | iorw->bytes_done = 0; |
3507 | iorw->free_iovec = iov; | |
3508 | if (iov) | |
3509 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3510 | return 0; |
3511 | } | |
3512 | ||
73debe68 | 3513 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3514 | { |
3529d8c2 JA |
3515 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3516 | return -EBADF; | |
93642ef8 | 3517 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3518 | } |
3519 | ||
c1dd91d1 JA |
3520 | /* |
3521 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3522 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3523 | * This gets called when the page is unlocked, and we generally expect that to | |
3524 | * happen when the page IO is completed and the page is now uptodate. This will | |
3525 | * queue a task_work based retry of the operation, attempting to copy the data | |
3526 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3527 | * do a thread based blocking retry of the operation. That's the unexpected | |
3528 | * slow path. | |
3529 | */ | |
bcf5a063 JA |
3530 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3531 | int sync, void *arg) | |
3532 | { | |
3533 | struct wait_page_queue *wpq; | |
3534 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3535 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3536 | |
3537 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3538 | ||
cdc8fcb4 LT |
3539 | if (!wake_page_match(wpq, key)) |
3540 | return 0; | |
3541 | ||
c8d317aa | 3542 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3543 | list_del_init(&wait->entry); |
3544 | ||
bcf5a063 JA |
3545 | /* submit ref gets dropped, acquire a new one */ |
3546 | refcount_inc(&req->refs); | |
921b9054 | 3547 | io_req_task_queue(req); |
bcf5a063 JA |
3548 | return 1; |
3549 | } | |
3550 | ||
c1dd91d1 JA |
3551 | /* |
3552 | * This controls whether a given IO request should be armed for async page | |
3553 | * based retry. If we return false here, the request is handed to the async | |
3554 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3555 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3556 | * will either succeed because the page is now uptodate and unlocked, or it | |
3557 | * will register a callback when the page is unlocked at IO completion. Through | |
3558 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3559 | * That retry will attempt the buffered read again. The retry will generally | |
3560 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3561 | * async worker threads for a blocking retry. | |
3562 | */ | |
227c0c96 | 3563 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3564 | { |
e8c2bc1f JA |
3565 | struct io_async_rw *rw = req->async_data; |
3566 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3567 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3568 | |
bcf5a063 JA |
3569 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3570 | if (req->flags & REQ_F_NOWAIT) | |
3571 | return false; | |
f67676d1 | 3572 | |
227c0c96 | 3573 | /* Only for buffered IO */ |
3b2a4439 | 3574 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3575 | return false; |
3b2a4439 | 3576 | |
bcf5a063 JA |
3577 | /* |
3578 | * just use poll if we can, and don't attempt if the fs doesn't | |
3579 | * support callback based unlocks | |
3580 | */ | |
3581 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3582 | return false; | |
f67676d1 | 3583 | |
3b2a4439 JA |
3584 | wait->wait.func = io_async_buf_func; |
3585 | wait->wait.private = req; | |
3586 | wait->wait.flags = 0; | |
3587 | INIT_LIST_HEAD(&wait->wait.entry); | |
3588 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3589 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3590 | kiocb->ki_waitq = wait; |
3b2a4439 | 3591 | return true; |
bcf5a063 JA |
3592 | } |
3593 | ||
3594 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3595 | { | |
3596 | if (req->file->f_op->read_iter) | |
3597 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3598 | else if (req->file->f_op->read) |
4017eb91 | 3599 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3600 | else |
3601 | return -EINVAL; | |
f67676d1 JA |
3602 | } |
3603 | ||
889fca73 | 3604 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3605 | { |
3606 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3607 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3608 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3609 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3610 | ssize_t io_size, ret, ret2; |
45d189c6 | 3611 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3612 | |
2846c481 | 3613 | if (rw) { |
e8c2bc1f | 3614 | iter = &rw->iter; |
2846c481 PB |
3615 | iovec = NULL; |
3616 | } else { | |
3617 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3618 | if (ret < 0) | |
3619 | return ret; | |
3620 | } | |
632546c4 | 3621 | io_size = iov_iter_count(iter); |
fa15bafb | 3622 | req->result = io_size; |
2b188cc1 | 3623 | |
fd6c2e4c JA |
3624 | /* Ensure we clear previously set non-block flag */ |
3625 | if (!force_nonblock) | |
29de5f6a | 3626 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3627 | else |
3628 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3629 | ||
24c74678 | 3630 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3631 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3632 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3633 | return ret ?: -EAGAIN; |
6713e7a6 | 3634 | } |
9e645e11 | 3635 | |
632546c4 | 3636 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3637 | if (unlikely(ret)) { |
3638 | kfree(iovec); | |
3639 | return ret; | |
3640 | } | |
2b188cc1 | 3641 | |
227c0c96 | 3642 | ret = io_iter_do_read(req, iter); |
32960613 | 3643 | |
57cd657b | 3644 | if (ret == -EIOCBQUEUED) { |
fe1cdd55 | 3645 | goto out_free; |
227c0c96 | 3646 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3647 | /* IOPOLL retry should happen for io-wq threads */ |
3648 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3649 | goto done; |
75c668cd PB |
3650 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3651 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3652 | goto done; |
84216315 | 3653 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3654 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3655 | ret = 0; |
7335e3bf | 3656 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3657 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3658 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3659 | goto done; |
227c0c96 JA |
3660 | } |
3661 | ||
227c0c96 | 3662 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3663 | if (ret2) |
3664 | return ret2; | |
3665 | ||
fe1cdd55 | 3666 | iovec = NULL; |
e8c2bc1f | 3667 | rw = req->async_data; |
227c0c96 | 3668 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3669 | iter = &rw->iter; |
227c0c96 | 3670 | |
b23df91b PB |
3671 | do { |
3672 | io_size -= ret; | |
3673 | rw->bytes_done += ret; | |
3674 | /* if we can retry, do so with the callbacks armed */ | |
3675 | if (!io_rw_should_retry(req)) { | |
3676 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3677 | return -EAGAIN; | |
3678 | } | |
3679 | ||
3680 | /* | |
3681 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3682 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3683 | * desired page gets unlocked. We can also get a partial read | |
3684 | * here, and if we do, then just retry at the new offset. | |
3685 | */ | |
3686 | ret = io_iter_do_read(req, iter); | |
3687 | if (ret == -EIOCBQUEUED) | |
3688 | return 0; | |
227c0c96 | 3689 | /* we got some bytes, but not all. retry. */ |
b23df91b | 3690 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3691 | done: |
889fca73 | 3692 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3693 | out_free: |
3694 | /* it's faster to check here then delegate to kfree */ | |
3695 | if (iovec) | |
3696 | kfree(iovec); | |
5ea5dd45 | 3697 | return 0; |
2b188cc1 JA |
3698 | } |
3699 | ||
73debe68 | 3700 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3701 | { |
3529d8c2 JA |
3702 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3703 | return -EBADF; | |
93642ef8 | 3704 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3705 | } |
3706 | ||
889fca73 | 3707 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3708 | { |
3709 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3710 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3711 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3712 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3713 | ssize_t ret, ret2, io_size; |
45d189c6 | 3714 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3715 | |
2846c481 | 3716 | if (rw) { |
e8c2bc1f | 3717 | iter = &rw->iter; |
2846c481 PB |
3718 | iovec = NULL; |
3719 | } else { | |
3720 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3721 | if (ret < 0) | |
3722 | return ret; | |
3723 | } | |
632546c4 | 3724 | io_size = iov_iter_count(iter); |
fa15bafb | 3725 | req->result = io_size; |
2b188cc1 | 3726 | |
fd6c2e4c JA |
3727 | /* Ensure we clear previously set non-block flag */ |
3728 | if (!force_nonblock) | |
a88fc400 PB |
3729 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3730 | else | |
3731 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3732 | |
24c74678 | 3733 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3734 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3735 | goto copy_iov; |
31b51510 | 3736 | |
10d59345 JA |
3737 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3738 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3739 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3740 | goto copy_iov; |
31b51510 | 3741 | |
632546c4 | 3742 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3743 | if (unlikely(ret)) |
3744 | goto out_free; | |
4ed734b0 | 3745 | |
fa15bafb PB |
3746 | /* |
3747 | * Open-code file_start_write here to grab freeze protection, | |
3748 | * which will be released by another thread in | |
3749 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3750 | * released so that it doesn't complain about the held lock when | |
3751 | * we return to userspace. | |
3752 | */ | |
3753 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3754 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3755 | __sb_writers_release(file_inode(req->file)->i_sb, |
3756 | SB_FREEZE_WRITE); | |
3757 | } | |
3758 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3759 | |
fa15bafb | 3760 | if (req->file->f_op->write_iter) |
ff6165b2 | 3761 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3762 | else if (req->file->f_op->write) |
4017eb91 | 3763 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3764 | else |
3765 | ret2 = -EINVAL; | |
4ed734b0 | 3766 | |
fa15bafb PB |
3767 | /* |
3768 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3769 | * retry them without IOCB_NOWAIT. | |
3770 | */ | |
3771 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3772 | ret2 = -EAGAIN; | |
75c668cd PB |
3773 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3774 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3775 | goto done; |
fa15bafb | 3776 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3777 | /* IOPOLL retry should happen for io-wq threads */ |
3778 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3779 | goto copy_iov; | |
355afaeb | 3780 | done: |
889fca73 | 3781 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3782 | } else { |
f67676d1 | 3783 | copy_iov: |
84216315 | 3784 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3785 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3786 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3787 | return ret ?: -EAGAIN; |
2b188cc1 | 3788 | } |
31b51510 | 3789 | out_free: |
f261c168 | 3790 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3791 | if (iovec) |
6f2cc166 | 3792 | kfree(iovec); |
2b188cc1 JA |
3793 | return ret; |
3794 | } | |
3795 | ||
80a261fd JA |
3796 | static int io_renameat_prep(struct io_kiocb *req, |
3797 | const struct io_uring_sqe *sqe) | |
3798 | { | |
3799 | struct io_rename *ren = &req->rename; | |
3800 | const char __user *oldf, *newf; | |
3801 | ||
3802 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3803 | return -EBADF; | |
3804 | ||
3805 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3806 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3807 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3808 | ren->new_dfd = READ_ONCE(sqe->len); | |
3809 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3810 | ||
3811 | ren->oldpath = getname(oldf); | |
3812 | if (IS_ERR(ren->oldpath)) | |
3813 | return PTR_ERR(ren->oldpath); | |
3814 | ||
3815 | ren->newpath = getname(newf); | |
3816 | if (IS_ERR(ren->newpath)) { | |
3817 | putname(ren->oldpath); | |
3818 | return PTR_ERR(ren->newpath); | |
3819 | } | |
3820 | ||
3821 | req->flags |= REQ_F_NEED_CLEANUP; | |
3822 | return 0; | |
3823 | } | |
3824 | ||
45d189c6 | 3825 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3826 | { |
3827 | struct io_rename *ren = &req->rename; | |
3828 | int ret; | |
3829 | ||
45d189c6 | 3830 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3831 | return -EAGAIN; |
3832 | ||
3833 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3834 | ren->newpath, ren->flags); | |
3835 | ||
3836 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3837 | if (ret < 0) | |
3838 | req_set_fail_links(req); | |
3839 | io_req_complete(req, ret); | |
3840 | return 0; | |
3841 | } | |
3842 | ||
14a1143b JA |
3843 | static int io_unlinkat_prep(struct io_kiocb *req, |
3844 | const struct io_uring_sqe *sqe) | |
3845 | { | |
3846 | struct io_unlink *un = &req->unlink; | |
3847 | const char __user *fname; | |
3848 | ||
3849 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3850 | return -EBADF; | |
3851 | ||
3852 | un->dfd = READ_ONCE(sqe->fd); | |
3853 | ||
3854 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3855 | if (un->flags & ~AT_REMOVEDIR) | |
3856 | return -EINVAL; | |
3857 | ||
3858 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3859 | un->filename = getname(fname); | |
3860 | if (IS_ERR(un->filename)) | |
3861 | return PTR_ERR(un->filename); | |
3862 | ||
3863 | req->flags |= REQ_F_NEED_CLEANUP; | |
3864 | return 0; | |
3865 | } | |
3866 | ||
45d189c6 | 3867 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3868 | { |
3869 | struct io_unlink *un = &req->unlink; | |
3870 | int ret; | |
3871 | ||
45d189c6 | 3872 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3873 | return -EAGAIN; |
3874 | ||
3875 | if (un->flags & AT_REMOVEDIR) | |
3876 | ret = do_rmdir(un->dfd, un->filename); | |
3877 | else | |
3878 | ret = do_unlinkat(un->dfd, un->filename); | |
3879 | ||
3880 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3881 | if (ret < 0) | |
3882 | req_set_fail_links(req); | |
3883 | io_req_complete(req, ret); | |
3884 | return 0; | |
3885 | } | |
3886 | ||
36f4fa68 JA |
3887 | static int io_shutdown_prep(struct io_kiocb *req, |
3888 | const struct io_uring_sqe *sqe) | |
3889 | { | |
3890 | #if defined(CONFIG_NET) | |
3891 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3892 | return -EINVAL; | |
3893 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3894 | sqe->buf_index) | |
3895 | return -EINVAL; | |
3896 | ||
3897 | req->shutdown.how = READ_ONCE(sqe->len); | |
3898 | return 0; | |
3899 | #else | |
3900 | return -EOPNOTSUPP; | |
3901 | #endif | |
3902 | } | |
3903 | ||
45d189c6 | 3904 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3905 | { |
3906 | #if defined(CONFIG_NET) | |
3907 | struct socket *sock; | |
3908 | int ret; | |
3909 | ||
45d189c6 | 3910 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3911 | return -EAGAIN; |
3912 | ||
48aba79b | 3913 | sock = sock_from_file(req->file); |
36f4fa68 | 3914 | if (unlikely(!sock)) |
48aba79b | 3915 | return -ENOTSOCK; |
36f4fa68 JA |
3916 | |
3917 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3918 | if (ret < 0) |
3919 | req_set_fail_links(req); | |
36f4fa68 JA |
3920 | io_req_complete(req, ret); |
3921 | return 0; | |
3922 | #else | |
3923 | return -EOPNOTSUPP; | |
3924 | #endif | |
3925 | } | |
3926 | ||
f2a8d5c7 PB |
3927 | static int __io_splice_prep(struct io_kiocb *req, |
3928 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3929 | { |
3930 | struct io_splice* sp = &req->splice; | |
3931 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3932 | |
3232dd02 PB |
3933 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3934 | return -EINVAL; | |
7d67af2c PB |
3935 | |
3936 | sp->file_in = NULL; | |
7d67af2c PB |
3937 | sp->len = READ_ONCE(sqe->len); |
3938 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3939 | ||
3940 | if (unlikely(sp->flags & ~valid_flags)) | |
3941 | return -EINVAL; | |
3942 | ||
8371adf5 PB |
3943 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3944 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3945 | if (!sp->file_in) | |
3946 | return -EBADF; | |
7d67af2c PB |
3947 | req->flags |= REQ_F_NEED_CLEANUP; |
3948 | ||
7cdaf587 XW |
3949 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3950 | /* | |
3951 | * Splice operation will be punted aync, and here need to | |
3952 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3953 | */ | |
3954 | io_req_init_async(req); | |
7d67af2c | 3955 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3956 | } |
7d67af2c PB |
3957 | |
3958 | return 0; | |
3959 | } | |
3960 | ||
f2a8d5c7 PB |
3961 | static int io_tee_prep(struct io_kiocb *req, |
3962 | const struct io_uring_sqe *sqe) | |
3963 | { | |
3964 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3965 | return -EINVAL; | |
3966 | return __io_splice_prep(req, sqe); | |
3967 | } | |
3968 | ||
45d189c6 | 3969 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3970 | { |
3971 | struct io_splice *sp = &req->splice; | |
3972 | struct file *in = sp->file_in; | |
3973 | struct file *out = sp->file_out; | |
3974 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3975 | long ret = 0; | |
3976 | ||
45d189c6 | 3977 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3978 | return -EAGAIN; |
3979 | if (sp->len) | |
3980 | ret = do_tee(in, out, sp->len, flags); | |
3981 | ||
3982 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3983 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3984 | ||
f2a8d5c7 PB |
3985 | if (ret != sp->len) |
3986 | req_set_fail_links(req); | |
e1e16097 | 3987 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3988 | return 0; |
3989 | } | |
3990 | ||
3991 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3992 | { | |
3993 | struct io_splice* sp = &req->splice; | |
3994 | ||
3995 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3996 | sp->off_out = READ_ONCE(sqe->off); | |
3997 | return __io_splice_prep(req, sqe); | |
3998 | } | |
3999 | ||
45d189c6 | 4000 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
4001 | { |
4002 | struct io_splice *sp = &req->splice; | |
4003 | struct file *in = sp->file_in; | |
4004 | struct file *out = sp->file_out; | |
4005 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
4006 | loff_t *poff_in, *poff_out; | |
c9687426 | 4007 | long ret = 0; |
7d67af2c | 4008 | |
45d189c6 | 4009 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 4010 | return -EAGAIN; |
7d67af2c PB |
4011 | |
4012 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
4013 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 4014 | |
948a7749 | 4015 | if (sp->len) |
c9687426 | 4016 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
4017 | |
4018 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
4019 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
4020 | ||
7d67af2c PB |
4021 | if (ret != sp->len) |
4022 | req_set_fail_links(req); | |
e1e16097 | 4023 | io_req_complete(req, ret); |
7d67af2c PB |
4024 | return 0; |
4025 | } | |
4026 | ||
2b188cc1 JA |
4027 | /* |
4028 | * IORING_OP_NOP just posts a completion event, nothing else. | |
4029 | */ | |
889fca73 | 4030 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
4031 | { |
4032 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 4033 | |
def596e9 JA |
4034 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4035 | return -EINVAL; | |
4036 | ||
889fca73 | 4037 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
4038 | return 0; |
4039 | } | |
4040 | ||
1155c76a | 4041 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 4042 | { |
6b06314c | 4043 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 4044 | |
09bb8394 JA |
4045 | if (!req->file) |
4046 | return -EBADF; | |
c992fe29 | 4047 | |
6b06314c | 4048 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 4049 | return -EINVAL; |
edafccee | 4050 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
4051 | return -EINVAL; |
4052 | ||
8ed8d3c3 JA |
4053 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
4054 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
4055 | return -EINVAL; | |
4056 | ||
4057 | req->sync.off = READ_ONCE(sqe->off); | |
4058 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
4059 | return 0; |
4060 | } | |
4061 | ||
45d189c6 | 4062 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4063 | { |
8ed8d3c3 | 4064 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
4065 | int ret; |
4066 | ||
ac45abc0 | 4067 | /* fsync always requires a blocking context */ |
45d189c6 | 4068 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4069 | return -EAGAIN; |
4070 | ||
9adbd45d | 4071 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
4072 | end > 0 ? end : LLONG_MAX, |
4073 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
4074 | if (ret < 0) | |
4075 | req_set_fail_links(req); | |
e1e16097 | 4076 | io_req_complete(req, ret); |
c992fe29 CH |
4077 | return 0; |
4078 | } | |
4079 | ||
d63d1b5e JA |
4080 | static int io_fallocate_prep(struct io_kiocb *req, |
4081 | const struct io_uring_sqe *sqe) | |
4082 | { | |
4083 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
4084 | return -EINVAL; | |
3232dd02 PB |
4085 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4086 | return -EINVAL; | |
d63d1b5e JA |
4087 | |
4088 | req->sync.off = READ_ONCE(sqe->off); | |
4089 | req->sync.len = READ_ONCE(sqe->addr); | |
4090 | req->sync.mode = READ_ONCE(sqe->len); | |
4091 | return 0; | |
4092 | } | |
4093 | ||
45d189c6 | 4094 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 4095 | { |
ac45abc0 PB |
4096 | int ret; |
4097 | ||
d63d1b5e | 4098 | /* fallocate always requiring blocking context */ |
45d189c6 | 4099 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 4100 | return -EAGAIN; |
ac45abc0 PB |
4101 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
4102 | req->sync.len); | |
ac45abc0 PB |
4103 | if (ret < 0) |
4104 | req_set_fail_links(req); | |
e1e16097 | 4105 | io_req_complete(req, ret); |
5d17b4a4 JA |
4106 | return 0; |
4107 | } | |
4108 | ||
ec65fea5 | 4109 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 4110 | { |
f8748881 | 4111 | const char __user *fname; |
15b71abe | 4112 | int ret; |
b7bb4f7d | 4113 | |
ec65fea5 | 4114 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 4115 | return -EINVAL; |
ec65fea5 | 4116 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 4117 | return -EBADF; |
03b1230c | 4118 | |
ec65fea5 PB |
4119 | /* open.how should be already initialised */ |
4120 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 4121 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 4122 | |
25e72d10 PB |
4123 | req->open.dfd = READ_ONCE(sqe->fd); |
4124 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 4125 | req->open.filename = getname(fname); |
15b71abe JA |
4126 | if (IS_ERR(req->open.filename)) { |
4127 | ret = PTR_ERR(req->open.filename); | |
4128 | req->open.filename = NULL; | |
4129 | return ret; | |
4130 | } | |
4022e7af | 4131 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 4132 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 4133 | return 0; |
03b1230c JA |
4134 | } |
4135 | ||
ec65fea5 PB |
4136 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4137 | { | |
4138 | u64 flags, mode; | |
4139 | ||
14587a46 | 4140 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4141 | return -EINVAL; |
ec65fea5 PB |
4142 | mode = READ_ONCE(sqe->len); |
4143 | flags = READ_ONCE(sqe->open_flags); | |
4144 | req->open.how = build_open_how(flags, mode); | |
4145 | return __io_openat_prep(req, sqe); | |
4146 | } | |
4147 | ||
cebdb986 | 4148 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 4149 | { |
cebdb986 | 4150 | struct open_how __user *how; |
cebdb986 | 4151 | size_t len; |
0fa03c62 JA |
4152 | int ret; |
4153 | ||
14587a46 | 4154 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 4155 | return -EINVAL; |
cebdb986 JA |
4156 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4157 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
4158 | if (len < OPEN_HOW_SIZE_VER0) |
4159 | return -EINVAL; | |
3529d8c2 | 4160 | |
cebdb986 JA |
4161 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
4162 | len); | |
4163 | if (ret) | |
4164 | return ret; | |
3529d8c2 | 4165 | |
ec65fea5 | 4166 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
4167 | } |
4168 | ||
45d189c6 | 4169 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
4170 | { |
4171 | struct open_flags op; | |
15b71abe | 4172 | struct file *file; |
3a81fd02 JA |
4173 | bool nonblock_set; |
4174 | bool resolve_nonblock; | |
15b71abe JA |
4175 | int ret; |
4176 | ||
cebdb986 | 4177 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
4178 | if (ret) |
4179 | goto err; | |
3a81fd02 JA |
4180 | nonblock_set = op.open_flag & O_NONBLOCK; |
4181 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 4182 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
4183 | /* |
4184 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
4185 | * it'll always -EAGAIN | |
4186 | */ | |
4187 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
4188 | return -EAGAIN; | |
4189 | op.lookup_flags |= LOOKUP_CACHED; | |
4190 | op.open_flag |= O_NONBLOCK; | |
4191 | } | |
15b71abe | 4192 | |
4022e7af | 4193 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
4194 | if (ret < 0) |
4195 | goto err; | |
4196 | ||
4197 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 4198 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
4199 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
4200 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 4201 | /* |
3a81fd02 JA |
4202 | * We could hang on to this 'fd', but seems like marginal |
4203 | * gain for something that is now known to be a slower path. | |
4204 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 4205 | */ |
3a81fd02 JA |
4206 | put_unused_fd(ret); |
4207 | return -EAGAIN; | |
4208 | } | |
4209 | ||
15b71abe JA |
4210 | if (IS_ERR(file)) { |
4211 | put_unused_fd(ret); | |
4212 | ret = PTR_ERR(file); | |
4213 | } else { | |
45d189c6 | 4214 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 4215 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
4216 | fsnotify_open(file); |
4217 | fd_install(ret, file); | |
4218 | } | |
4219 | err: | |
4220 | putname(req->open.filename); | |
8fef80bf | 4221 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
4222 | if (ret < 0) |
4223 | req_set_fail_links(req); | |
e1e16097 | 4224 | io_req_complete(req, ret); |
15b71abe JA |
4225 | return 0; |
4226 | } | |
4227 | ||
45d189c6 | 4228 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 4229 | { |
45d189c6 | 4230 | return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK); |
cebdb986 JA |
4231 | } |
4232 | ||
067524e9 JA |
4233 | static int io_remove_buffers_prep(struct io_kiocb *req, |
4234 | const struct io_uring_sqe *sqe) | |
4235 | { | |
4236 | struct io_provide_buf *p = &req->pbuf; | |
4237 | u64 tmp; | |
4238 | ||
4239 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
4240 | return -EINVAL; | |
4241 | ||
4242 | tmp = READ_ONCE(sqe->fd); | |
4243 | if (!tmp || tmp > USHRT_MAX) | |
4244 | return -EINVAL; | |
4245 | ||
4246 | memset(p, 0, sizeof(*p)); | |
4247 | p->nbufs = tmp; | |
4248 | p->bgid = READ_ONCE(sqe->buf_group); | |
4249 | return 0; | |
4250 | } | |
4251 | ||
4252 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
4253 | int bgid, unsigned nbufs) | |
4254 | { | |
4255 | unsigned i = 0; | |
4256 | ||
4257 | /* shouldn't happen */ | |
4258 | if (!nbufs) | |
4259 | return 0; | |
4260 | ||
4261 | /* the head kbuf is the list itself */ | |
4262 | while (!list_empty(&buf->list)) { | |
4263 | struct io_buffer *nxt; | |
4264 | ||
4265 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
4266 | list_del(&nxt->list); | |
4267 | kfree(nxt); | |
4268 | if (++i == nbufs) | |
4269 | return i; | |
4270 | } | |
4271 | i++; | |
4272 | kfree(buf); | |
4273 | idr_remove(&ctx->io_buffer_idr, bgid); | |
4274 | ||
4275 | return i; | |
4276 | } | |
4277 | ||
889fca73 | 4278 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
4279 | { |
4280 | struct io_provide_buf *p = &req->pbuf; | |
4281 | struct io_ring_ctx *ctx = req->ctx; | |
4282 | struct io_buffer *head; | |
4283 | int ret = 0; | |
45d189c6 | 4284 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
4285 | |
4286 | io_ring_submit_lock(ctx, !force_nonblock); | |
4287 | ||
4288 | lockdep_assert_held(&ctx->uring_lock); | |
4289 | ||
4290 | ret = -ENOENT; | |
4291 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4292 | if (head) | |
4293 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
4294 | if (ret < 0) |
4295 | req_set_fail_links(req); | |
067524e9 | 4296 | |
31bff9a5 PB |
4297 | /* need to hold the lock to complete IOPOLL requests */ |
4298 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4299 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4300 | io_ring_submit_unlock(ctx, !force_nonblock); |
4301 | } else { | |
4302 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4303 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4304 | } |
067524e9 JA |
4305 | return 0; |
4306 | } | |
4307 | ||
ddf0322d JA |
4308 | static int io_provide_buffers_prep(struct io_kiocb *req, |
4309 | const struct io_uring_sqe *sqe) | |
4310 | { | |
4311 | struct io_provide_buf *p = &req->pbuf; | |
4312 | u64 tmp; | |
4313 | ||
4314 | if (sqe->ioprio || sqe->rw_flags) | |
4315 | return -EINVAL; | |
4316 | ||
4317 | tmp = READ_ONCE(sqe->fd); | |
4318 | if (!tmp || tmp > USHRT_MAX) | |
4319 | return -E2BIG; | |
4320 | p->nbufs = tmp; | |
4321 | p->addr = READ_ONCE(sqe->addr); | |
4322 | p->len = READ_ONCE(sqe->len); | |
4323 | ||
efe68c1c | 4324 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
4325 | return -EFAULT; |
4326 | ||
4327 | p->bgid = READ_ONCE(sqe->buf_group); | |
4328 | tmp = READ_ONCE(sqe->off); | |
4329 | if (tmp > USHRT_MAX) | |
4330 | return -E2BIG; | |
4331 | p->bid = tmp; | |
4332 | return 0; | |
4333 | } | |
4334 | ||
4335 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4336 | { | |
4337 | struct io_buffer *buf; | |
4338 | u64 addr = pbuf->addr; | |
4339 | int i, bid = pbuf->bid; | |
4340 | ||
4341 | for (i = 0; i < pbuf->nbufs; i++) { | |
4342 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4343 | if (!buf) | |
4344 | break; | |
4345 | ||
4346 | buf->addr = addr; | |
4347 | buf->len = pbuf->len; | |
4348 | buf->bid = bid; | |
4349 | addr += pbuf->len; | |
4350 | bid++; | |
4351 | if (!*head) { | |
4352 | INIT_LIST_HEAD(&buf->list); | |
4353 | *head = buf; | |
4354 | } else { | |
4355 | list_add_tail(&buf->list, &(*head)->list); | |
4356 | } | |
4357 | } | |
4358 | ||
4359 | return i ? i : -ENOMEM; | |
4360 | } | |
4361 | ||
889fca73 | 4362 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4363 | { |
4364 | struct io_provide_buf *p = &req->pbuf; | |
4365 | struct io_ring_ctx *ctx = req->ctx; | |
4366 | struct io_buffer *head, *list; | |
4367 | int ret = 0; | |
45d189c6 | 4368 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4369 | |
4370 | io_ring_submit_lock(ctx, !force_nonblock); | |
4371 | ||
4372 | lockdep_assert_held(&ctx->uring_lock); | |
4373 | ||
4374 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
4375 | ||
4376 | ret = io_add_buffers(p, &head); | |
4377 | if (ret < 0) | |
4378 | goto out; | |
4379 | ||
4380 | if (!list) { | |
4381 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
4382 | GFP_KERNEL); | |
4383 | if (ret < 0) { | |
067524e9 | 4384 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4385 | goto out; |
4386 | } | |
4387 | } | |
4388 | out: | |
ddf0322d JA |
4389 | if (ret < 0) |
4390 | req_set_fail_links(req); | |
31bff9a5 PB |
4391 | |
4392 | /* need to hold the lock to complete IOPOLL requests */ | |
4393 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4394 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4395 | io_ring_submit_unlock(ctx, !force_nonblock); |
4396 | } else { | |
4397 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4398 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4399 | } |
ddf0322d | 4400 | return 0; |
cebdb986 JA |
4401 | } |
4402 | ||
3e4827b0 JA |
4403 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4404 | const struct io_uring_sqe *sqe) | |
4405 | { | |
4406 | #if defined(CONFIG_EPOLL) | |
4407 | if (sqe->ioprio || sqe->buf_index) | |
4408 | return -EINVAL; | |
6ca56f84 | 4409 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4410 | return -EINVAL; |
3e4827b0 JA |
4411 | |
4412 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4413 | req->epoll.op = READ_ONCE(sqe->len); | |
4414 | req->epoll.fd = READ_ONCE(sqe->off); | |
4415 | ||
4416 | if (ep_op_has_event(req->epoll.op)) { | |
4417 | struct epoll_event __user *ev; | |
4418 | ||
4419 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4420 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4421 | return -EFAULT; | |
4422 | } | |
4423 | ||
4424 | return 0; | |
4425 | #else | |
4426 | return -EOPNOTSUPP; | |
4427 | #endif | |
4428 | } | |
4429 | ||
889fca73 | 4430 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4431 | { |
4432 | #if defined(CONFIG_EPOLL) | |
4433 | struct io_epoll *ie = &req->epoll; | |
4434 | int ret; | |
45d189c6 | 4435 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4436 | |
4437 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4438 | if (force_nonblock && ret == -EAGAIN) | |
4439 | return -EAGAIN; | |
4440 | ||
4441 | if (ret < 0) | |
4442 | req_set_fail_links(req); | |
889fca73 | 4443 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4444 | return 0; |
4445 | #else | |
4446 | return -EOPNOTSUPP; | |
4447 | #endif | |
4448 | } | |
4449 | ||
c1ca757b JA |
4450 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4451 | { | |
4452 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4453 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4454 | return -EINVAL; | |
3232dd02 PB |
4455 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4456 | return -EINVAL; | |
c1ca757b JA |
4457 | |
4458 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4459 | req->madvise.len = READ_ONCE(sqe->len); | |
4460 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4461 | return 0; | |
4462 | #else | |
4463 | return -EOPNOTSUPP; | |
4464 | #endif | |
4465 | } | |
4466 | ||
45d189c6 | 4467 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4468 | { |
4469 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4470 | struct io_madvise *ma = &req->madvise; | |
4471 | int ret; | |
4472 | ||
45d189c6 | 4473 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4474 | return -EAGAIN; |
4475 | ||
0726b01e | 4476 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4477 | if (ret < 0) |
4478 | req_set_fail_links(req); | |
e1e16097 | 4479 | io_req_complete(req, ret); |
c1ca757b JA |
4480 | return 0; |
4481 | #else | |
4482 | return -EOPNOTSUPP; | |
4483 | #endif | |
4484 | } | |
4485 | ||
4840e418 JA |
4486 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4487 | { | |
4488 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4489 | return -EINVAL; | |
3232dd02 PB |
4490 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4491 | return -EINVAL; | |
4840e418 JA |
4492 | |
4493 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4494 | req->fadvise.len = READ_ONCE(sqe->len); | |
4495 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4496 | return 0; | |
4497 | } | |
4498 | ||
45d189c6 | 4499 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4500 | { |
4501 | struct io_fadvise *fa = &req->fadvise; | |
4502 | int ret; | |
4503 | ||
45d189c6 | 4504 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4505 | switch (fa->advice) { |
4506 | case POSIX_FADV_NORMAL: | |
4507 | case POSIX_FADV_RANDOM: | |
4508 | case POSIX_FADV_SEQUENTIAL: | |
4509 | break; | |
4510 | default: | |
4511 | return -EAGAIN; | |
4512 | } | |
4513 | } | |
4840e418 JA |
4514 | |
4515 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4516 | if (ret < 0) | |
4517 | req_set_fail_links(req); | |
e1e16097 | 4518 | io_req_complete(req, ret); |
4840e418 JA |
4519 | return 0; |
4520 | } | |
4521 | ||
eddc7ef5 JA |
4522 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4523 | { | |
6ca56f84 | 4524 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4525 | return -EINVAL; |
eddc7ef5 JA |
4526 | if (sqe->ioprio || sqe->buf_index) |
4527 | return -EINVAL; | |
9c280f90 | 4528 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4529 | return -EBADF; |
eddc7ef5 | 4530 | |
1d9e1288 BM |
4531 | req->statx.dfd = READ_ONCE(sqe->fd); |
4532 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4533 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4534 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4535 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4536 | |
4537 | return 0; | |
4538 | } | |
4539 | ||
45d189c6 | 4540 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4541 | { |
1d9e1288 | 4542 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4543 | int ret; |
4544 | ||
45d189c6 | 4545 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4546 | /* only need file table for an actual valid fd */ |
4547 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4548 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4549 | return -EAGAIN; |
5b0bbee4 | 4550 | } |
eddc7ef5 | 4551 | |
e62753e4 BM |
4552 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4553 | ctx->buffer); | |
eddc7ef5 | 4554 | |
eddc7ef5 JA |
4555 | if (ret < 0) |
4556 | req_set_fail_links(req); | |
e1e16097 | 4557 | io_req_complete(req, ret); |
eddc7ef5 JA |
4558 | return 0; |
4559 | } | |
4560 | ||
b5dba59e JA |
4561 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4562 | { | |
14587a46 | 4563 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4564 | return -EINVAL; |
b5dba59e JA |
4565 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4566 | sqe->rw_flags || sqe->buf_index) | |
4567 | return -EINVAL; | |
9c280f90 | 4568 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4569 | return -EBADF; |
b5dba59e JA |
4570 | |
4571 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4572 | return 0; |
b5dba59e JA |
4573 | } |
4574 | ||
889fca73 | 4575 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4576 | { |
9eac1904 | 4577 | struct files_struct *files = current->files; |
3af73b28 | 4578 | struct io_close *close = &req->close; |
9eac1904 JA |
4579 | struct fdtable *fdt; |
4580 | struct file *file; | |
b5dba59e JA |
4581 | int ret; |
4582 | ||
9eac1904 JA |
4583 | file = NULL; |
4584 | ret = -EBADF; | |
4585 | spin_lock(&files->file_lock); | |
4586 | fdt = files_fdtable(files); | |
4587 | if (close->fd >= fdt->max_fds) { | |
4588 | spin_unlock(&files->file_lock); | |
4589 | goto err; | |
4590 | } | |
4591 | file = fdt->fd[close->fd]; | |
4592 | if (!file) { | |
4593 | spin_unlock(&files->file_lock); | |
4594 | goto err; | |
4595 | } | |
4596 | ||
4597 | if (file->f_op == &io_uring_fops) { | |
4598 | spin_unlock(&files->file_lock); | |
4599 | file = NULL; | |
4600 | goto err; | |
3af73b28 | 4601 | } |
b5dba59e JA |
4602 | |
4603 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4604 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4605 | spin_unlock(&files->file_lock); |
0bf0eefd | 4606 | return -EAGAIN; |
a2100672 | 4607 | } |
b5dba59e | 4608 | |
9eac1904 JA |
4609 | ret = __close_fd_get_file(close->fd, &file); |
4610 | spin_unlock(&files->file_lock); | |
4611 | if (ret < 0) { | |
4612 | if (ret == -ENOENT) | |
4613 | ret = -EBADF; | |
4614 | goto err; | |
4615 | } | |
4616 | ||
3af73b28 | 4617 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4618 | ret = filp_close(file, current->files); |
4619 | err: | |
3af73b28 PB |
4620 | if (ret < 0) |
4621 | req_set_fail_links(req); | |
9eac1904 JA |
4622 | if (file) |
4623 | fput(file); | |
889fca73 | 4624 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4625 | return 0; |
b5dba59e JA |
4626 | } |
4627 | ||
1155c76a | 4628 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4629 | { |
4630 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4631 | |
5d17b4a4 JA |
4632 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4633 | return -EINVAL; | |
4634 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4635 | return -EINVAL; | |
4636 | ||
8ed8d3c3 JA |
4637 | req->sync.off = READ_ONCE(sqe->off); |
4638 | req->sync.len = READ_ONCE(sqe->len); | |
4639 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4640 | return 0; |
4641 | } | |
4642 | ||
45d189c6 | 4643 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4644 | { |
8ed8d3c3 JA |
4645 | int ret; |
4646 | ||
ac45abc0 | 4647 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4648 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4649 | return -EAGAIN; |
4650 | ||
9adbd45d | 4651 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4652 | req->sync.flags); |
4653 | if (ret < 0) | |
4654 | req_set_fail_links(req); | |
e1e16097 | 4655 | io_req_complete(req, ret); |
5d17b4a4 JA |
4656 | return 0; |
4657 | } | |
4658 | ||
469956e8 | 4659 | #if defined(CONFIG_NET) |
02d27d89 PB |
4660 | static int io_setup_async_msg(struct io_kiocb *req, |
4661 | struct io_async_msghdr *kmsg) | |
4662 | { | |
e8c2bc1f JA |
4663 | struct io_async_msghdr *async_msg = req->async_data; |
4664 | ||
4665 | if (async_msg) | |
02d27d89 | 4666 | return -EAGAIN; |
e8c2bc1f | 4667 | if (io_alloc_async_data(req)) { |
257e84a5 | 4668 | kfree(kmsg->free_iov); |
02d27d89 PB |
4669 | return -ENOMEM; |
4670 | } | |
e8c2bc1f | 4671 | async_msg = req->async_data; |
02d27d89 | 4672 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4673 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4674 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4675 | /* if were using fast_iov, set it to the new one */ |
4676 | if (!async_msg->free_iov) | |
4677 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4678 | ||
02d27d89 PB |
4679 | return -EAGAIN; |
4680 | } | |
4681 | ||
2ae523ed PB |
4682 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4683 | struct io_async_msghdr *iomsg) | |
4684 | { | |
2ae523ed | 4685 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4686 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4687 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4688 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4689 | } |
4690 | ||
93642ef8 PB |
4691 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4692 | { | |
4693 | int ret; | |
4694 | ||
4695 | if (!io_op_defs[req->opcode].needs_async_data) | |
4696 | return 0; | |
4697 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4698 | if (!ret) | |
4699 | req->flags |= REQ_F_NEED_CLEANUP; | |
4700 | return ret; | |
4701 | } | |
4702 | ||
3529d8c2 | 4703 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4704 | { |
e47293fd | 4705 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4706 | |
d2b6f48b PB |
4707 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4708 | return -EINVAL; | |
4709 | ||
e47293fd | 4710 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4711 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4712 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4713 | |
d8768362 JA |
4714 | #ifdef CONFIG_COMPAT |
4715 | if (req->ctx->compat) | |
4716 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4717 | #endif | |
93642ef8 | 4718 | return 0; |
03b1230c JA |
4719 | } |
4720 | ||
889fca73 | 4721 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4722 | { |
6b754c8b | 4723 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4724 | struct socket *sock; |
7a7cacba | 4725 | unsigned flags; |
0fa03c62 JA |
4726 | int ret; |
4727 | ||
dba4a925 | 4728 | sock = sock_from_file(req->file); |
7a7cacba | 4729 | if (unlikely(!sock)) |
dba4a925 | 4730 | return -ENOTSOCK; |
3529d8c2 | 4731 | |
257e84a5 PB |
4732 | kmsg = req->async_data; |
4733 | if (!kmsg) { | |
7a7cacba PB |
4734 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4735 | if (ret) | |
4736 | return ret; | |
4737 | kmsg = &iomsg; | |
0fa03c62 | 4738 | } |
0fa03c62 | 4739 | |
7a7cacba PB |
4740 | flags = req->sr_msg.msg_flags; |
4741 | if (flags & MSG_DONTWAIT) | |
4742 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4743 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4744 | flags |= MSG_DONTWAIT; |
e47293fd | 4745 | |
7a7cacba | 4746 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4747 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4748 | return io_setup_async_msg(req, kmsg); |
4749 | if (ret == -ERESTARTSYS) | |
4750 | ret = -EINTR; | |
0fa03c62 | 4751 | |
257e84a5 PB |
4752 | /* fast path, check for non-NULL to avoid function call */ |
4753 | if (kmsg->free_iov) | |
4754 | kfree(kmsg->free_iov); | |
99bc4c38 | 4755 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4756 | if (ret < 0) |
4757 | req_set_fail_links(req); | |
889fca73 | 4758 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4759 | return 0; |
03b1230c | 4760 | } |
aa1fa28f | 4761 | |
889fca73 | 4762 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4763 | { |
7a7cacba PB |
4764 | struct io_sr_msg *sr = &req->sr_msg; |
4765 | struct msghdr msg; | |
4766 | struct iovec iov; | |
fddaface | 4767 | struct socket *sock; |
7a7cacba | 4768 | unsigned flags; |
fddaface JA |
4769 | int ret; |
4770 | ||
dba4a925 | 4771 | sock = sock_from_file(req->file); |
7a7cacba | 4772 | if (unlikely(!sock)) |
dba4a925 | 4773 | return -ENOTSOCK; |
fddaface | 4774 | |
7a7cacba PB |
4775 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4776 | if (unlikely(ret)) | |
14db8411 | 4777 | return ret; |
fddaface | 4778 | |
7a7cacba PB |
4779 | msg.msg_name = NULL; |
4780 | msg.msg_control = NULL; | |
4781 | msg.msg_controllen = 0; | |
4782 | msg.msg_namelen = 0; | |
fddaface | 4783 | |
7a7cacba PB |
4784 | flags = req->sr_msg.msg_flags; |
4785 | if (flags & MSG_DONTWAIT) | |
4786 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4787 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4788 | flags |= MSG_DONTWAIT; |
fddaface | 4789 | |
7a7cacba PB |
4790 | msg.msg_flags = flags; |
4791 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4792 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4793 | return -EAGAIN; |
4794 | if (ret == -ERESTARTSYS) | |
4795 | ret = -EINTR; | |
fddaface | 4796 | |
fddaface JA |
4797 | if (ret < 0) |
4798 | req_set_fail_links(req); | |
889fca73 | 4799 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4800 | return 0; |
fddaface JA |
4801 | } |
4802 | ||
1400e697 PB |
4803 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4804 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4805 | { |
4806 | struct io_sr_msg *sr = &req->sr_msg; | |
4807 | struct iovec __user *uiov; | |
4808 | size_t iov_len; | |
4809 | int ret; | |
4810 | ||
1400e697 PB |
4811 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4812 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4813 | if (ret) |
4814 | return ret; | |
4815 | ||
4816 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4817 | if (iov_len > 1) | |
4818 | return -EINVAL; | |
5476dfed | 4819 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4820 | return -EFAULT; |
5476dfed | 4821 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4822 | iomsg->free_iov = NULL; |
52de1fe1 | 4823 | } else { |
257e84a5 | 4824 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4825 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4826 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4827 | false); |
52de1fe1 JA |
4828 | if (ret > 0) |
4829 | ret = 0; | |
4830 | } | |
4831 | ||
4832 | return ret; | |
4833 | } | |
4834 | ||
4835 | #ifdef CONFIG_COMPAT | |
4836 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4837 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4838 | { |
4839 | struct compat_msghdr __user *msg_compat; | |
4840 | struct io_sr_msg *sr = &req->sr_msg; | |
4841 | struct compat_iovec __user *uiov; | |
4842 | compat_uptr_t ptr; | |
4843 | compat_size_t len; | |
4844 | int ret; | |
4845 | ||
270a5940 | 4846 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4847 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4848 | &ptr, &len); |
4849 | if (ret) | |
4850 | return ret; | |
4851 | ||
4852 | uiov = compat_ptr(ptr); | |
4853 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4854 | compat_ssize_t clen; | |
4855 | ||
4856 | if (len > 1) | |
4857 | return -EINVAL; | |
4858 | if (!access_ok(uiov, sizeof(*uiov))) | |
4859 | return -EFAULT; | |
4860 | if (__get_user(clen, &uiov->iov_len)) | |
4861 | return -EFAULT; | |
4862 | if (clen < 0) | |
4863 | return -EINVAL; | |
2d280bc8 | 4864 | sr->len = clen; |
257e84a5 | 4865 | iomsg->free_iov = NULL; |
52de1fe1 | 4866 | } else { |
257e84a5 | 4867 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4868 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4869 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4870 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4871 | if (ret < 0) |
4872 | return ret; | |
4873 | } | |
4874 | ||
4875 | return 0; | |
4876 | } | |
4877 | #endif | |
4878 | ||
1400e697 PB |
4879 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4880 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4881 | { |
1400e697 | 4882 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4883 | |
4884 | #ifdef CONFIG_COMPAT | |
4885 | if (req->ctx->compat) | |
1400e697 | 4886 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4887 | #endif |
52de1fe1 | 4888 | |
1400e697 | 4889 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4890 | } |
4891 | ||
bcda7baa | 4892 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4893 | bool needs_lock) |
bcda7baa JA |
4894 | { |
4895 | struct io_sr_msg *sr = &req->sr_msg; | |
4896 | struct io_buffer *kbuf; | |
4897 | ||
bcda7baa JA |
4898 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4899 | if (IS_ERR(kbuf)) | |
4900 | return kbuf; | |
4901 | ||
4902 | sr->kbuf = kbuf; | |
4903 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4904 | return kbuf; |
fddaface JA |
4905 | } |
4906 | ||
7fbb1b54 PB |
4907 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4908 | { | |
4909 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4910 | } | |
4911 | ||
93642ef8 | 4912 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4913 | { |
99bc4c38 | 4914 | int ret; |
3529d8c2 | 4915 | |
93642ef8 PB |
4916 | if (!io_op_defs[req->opcode].needs_async_data) |
4917 | return 0; | |
4918 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4919 | if (!ret) | |
4920 | req->flags |= REQ_F_NEED_CLEANUP; | |
4921 | return ret; | |
4922 | } | |
4923 | ||
4924 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4925 | { | |
4926 | struct io_sr_msg *sr = &req->sr_msg; | |
4927 | ||
d2b6f48b PB |
4928 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4929 | return -EINVAL; | |
4930 | ||
3529d8c2 | 4931 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4932 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4933 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4934 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4935 | |
d8768362 JA |
4936 | #ifdef CONFIG_COMPAT |
4937 | if (req->ctx->compat) | |
4938 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4939 | #endif | |
93642ef8 | 4940 | return 0; |
aa1fa28f JA |
4941 | } |
4942 | ||
889fca73 | 4943 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4944 | { |
6b754c8b | 4945 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4946 | struct socket *sock; |
7fbb1b54 | 4947 | struct io_buffer *kbuf; |
7a7cacba | 4948 | unsigned flags; |
52de1fe1 | 4949 | int ret, cflags = 0; |
45d189c6 | 4950 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4951 | |
dba4a925 | 4952 | sock = sock_from_file(req->file); |
7a7cacba | 4953 | if (unlikely(!sock)) |
dba4a925 | 4954 | return -ENOTSOCK; |
3529d8c2 | 4955 | |
257e84a5 PB |
4956 | kmsg = req->async_data; |
4957 | if (!kmsg) { | |
7a7cacba PB |
4958 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4959 | if (ret) | |
681fda8d | 4960 | return ret; |
7a7cacba PB |
4961 | kmsg = &iomsg; |
4962 | } | |
03b1230c | 4963 | |
bc02ef33 | 4964 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4965 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4966 | if (IS_ERR(kbuf)) |
52de1fe1 | 4967 | return PTR_ERR(kbuf); |
7a7cacba | 4968 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4969 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4970 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4971 | 1, req->sr_msg.len); |
4972 | } | |
52de1fe1 | 4973 | |
7a7cacba PB |
4974 | flags = req->sr_msg.msg_flags; |
4975 | if (flags & MSG_DONTWAIT) | |
4976 | req->flags |= REQ_F_NOWAIT; | |
4977 | else if (force_nonblock) | |
4978 | flags |= MSG_DONTWAIT; | |
e47293fd | 4979 | |
7a7cacba PB |
4980 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4981 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4982 | if (force_nonblock && ret == -EAGAIN) |
4983 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4984 | if (ret == -ERESTARTSYS) |
4985 | ret = -EINTR; | |
03b1230c | 4986 | |
7fbb1b54 PB |
4987 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4988 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4989 | /* fast path, check for non-NULL to avoid function call */ |
4990 | if (kmsg->free_iov) | |
4991 | kfree(kmsg->free_iov); | |
99bc4c38 | 4992 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4993 | if (ret < 0) |
4994 | req_set_fail_links(req); | |
889fca73 | 4995 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4996 | return 0; |
0fa03c62 | 4997 | } |
5d17b4a4 | 4998 | |
889fca73 | 4999 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 5000 | { |
6b754c8b | 5001 | struct io_buffer *kbuf; |
7a7cacba PB |
5002 | struct io_sr_msg *sr = &req->sr_msg; |
5003 | struct msghdr msg; | |
5004 | void __user *buf = sr->buf; | |
fddaface | 5005 | struct socket *sock; |
7a7cacba PB |
5006 | struct iovec iov; |
5007 | unsigned flags; | |
bcda7baa | 5008 | int ret, cflags = 0; |
45d189c6 | 5009 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 5010 | |
dba4a925 | 5011 | sock = sock_from_file(req->file); |
7a7cacba | 5012 | if (unlikely(!sock)) |
dba4a925 | 5013 | return -ENOTSOCK; |
fddaface | 5014 | |
bc02ef33 | 5015 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 5016 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
5017 | if (IS_ERR(kbuf)) |
5018 | return PTR_ERR(kbuf); | |
7a7cacba | 5019 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 5020 | } |
bcda7baa | 5021 | |
7a7cacba | 5022 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
5023 | if (unlikely(ret)) |
5024 | goto out_free; | |
fddaface | 5025 | |
7a7cacba PB |
5026 | msg.msg_name = NULL; |
5027 | msg.msg_control = NULL; | |
5028 | msg.msg_controllen = 0; | |
5029 | msg.msg_namelen = 0; | |
5030 | msg.msg_iocb = NULL; | |
5031 | msg.msg_flags = 0; | |
fddaface | 5032 | |
7a7cacba PB |
5033 | flags = req->sr_msg.msg_flags; |
5034 | if (flags & MSG_DONTWAIT) | |
5035 | req->flags |= REQ_F_NOWAIT; | |
5036 | else if (force_nonblock) | |
5037 | flags |= MSG_DONTWAIT; | |
5038 | ||
5039 | ret = sock_recvmsg(sock, &msg, flags); | |
5040 | if (force_nonblock && ret == -EAGAIN) | |
5041 | return -EAGAIN; | |
5042 | if (ret == -ERESTARTSYS) | |
5043 | ret = -EINTR; | |
14c32eee | 5044 | out_free: |
7fbb1b54 PB |
5045 | if (req->flags & REQ_F_BUFFER_SELECTED) |
5046 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
5047 | if (ret < 0) |
5048 | req_set_fail_links(req); | |
889fca73 | 5049 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 5050 | return 0; |
fddaface JA |
5051 | } |
5052 | ||
3529d8c2 | 5053 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 5054 | { |
8ed8d3c3 JA |
5055 | struct io_accept *accept = &req->accept; |
5056 | ||
14587a46 | 5057 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 5058 | return -EINVAL; |
8042d6ce | 5059 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
5060 | return -EINVAL; |
5061 | ||
d55e5f5b JA |
5062 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5063 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 5064 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 5065 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 5066 | return 0; |
8ed8d3c3 | 5067 | } |
17f2fe35 | 5068 | |
889fca73 | 5069 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
5070 | { |
5071 | struct io_accept *accept = &req->accept; | |
45d189c6 | 5072 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 5073 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
5074 | int ret; |
5075 | ||
e697deed JX |
5076 | if (req->file->f_flags & O_NONBLOCK) |
5077 | req->flags |= REQ_F_NOWAIT; | |
5078 | ||
8ed8d3c3 | 5079 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
5080 | accept->addr_len, accept->flags, |
5081 | accept->nofile); | |
8ed8d3c3 | 5082 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 5083 | return -EAGAIN; |
ac45abc0 PB |
5084 | if (ret < 0) { |
5085 | if (ret == -ERESTARTSYS) | |
5086 | ret = -EINTR; | |
4e88d6e7 | 5087 | req_set_fail_links(req); |
ac45abc0 | 5088 | } |
889fca73 | 5089 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 5090 | return 0; |
8ed8d3c3 JA |
5091 | } |
5092 | ||
93642ef8 PB |
5093 | static int io_connect_prep_async(struct io_kiocb *req) |
5094 | { | |
5095 | struct io_async_connect *io = req->async_data; | |
5096 | struct io_connect *conn = &req->connect; | |
5097 | ||
5098 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
5099 | } | |
5100 | ||
3529d8c2 | 5101 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 5102 | { |
3529d8c2 | 5103 | struct io_connect *conn = &req->connect; |
f499a021 | 5104 | |
14587a46 | 5105 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
5106 | return -EINVAL; |
5107 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
5108 | return -EINVAL; | |
5109 | ||
3529d8c2 JA |
5110 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
5111 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 5112 | return 0; |
f499a021 JA |
5113 | } |
5114 | ||
889fca73 | 5115 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 5116 | { |
e8c2bc1f | 5117 | struct io_async_connect __io, *io; |
f8e85cf2 | 5118 | unsigned file_flags; |
3fbb51c1 | 5119 | int ret; |
45d189c6 | 5120 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 5121 | |
e8c2bc1f JA |
5122 | if (req->async_data) { |
5123 | io = req->async_data; | |
f499a021 | 5124 | } else { |
3529d8c2 JA |
5125 | ret = move_addr_to_kernel(req->connect.addr, |
5126 | req->connect.addr_len, | |
e8c2bc1f | 5127 | &__io.address); |
f499a021 JA |
5128 | if (ret) |
5129 | goto out; | |
5130 | io = &__io; | |
5131 | } | |
5132 | ||
3fbb51c1 JA |
5133 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
5134 | ||
e8c2bc1f | 5135 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 5136 | req->connect.addr_len, file_flags); |
87f80d62 | 5137 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 5138 | if (req->async_data) |
b7bb4f7d | 5139 | return -EAGAIN; |
e8c2bc1f | 5140 | if (io_alloc_async_data(req)) { |
f499a021 JA |
5141 | ret = -ENOMEM; |
5142 | goto out; | |
5143 | } | |
e8c2bc1f JA |
5144 | io = req->async_data; |
5145 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 5146 | return -EAGAIN; |
f499a021 | 5147 | } |
f8e85cf2 JA |
5148 | if (ret == -ERESTARTSYS) |
5149 | ret = -EINTR; | |
f499a021 | 5150 | out: |
4e88d6e7 JA |
5151 | if (ret < 0) |
5152 | req_set_fail_links(req); | |
889fca73 | 5153 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 5154 | return 0; |
469956e8 Y |
5155 | } |
5156 | #else /* !CONFIG_NET */ | |
99a10081 JA |
5157 | #define IO_NETOP_FN(op) \ |
5158 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
5159 | { \ | |
5160 | return -EOPNOTSUPP; \ | |
5161 | } | |
5162 | ||
5163 | #define IO_NETOP_PREP(op) \ | |
5164 | IO_NETOP_FN(op) \ | |
5165 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
5166 | { \ | |
5167 | return -EOPNOTSUPP; \ | |
5168 | } \ | |
5169 | ||
5170 | #define IO_NETOP_PREP_ASYNC(op) \ | |
5171 | IO_NETOP_PREP(op) \ | |
5172 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
5173 | { \ | |
5174 | return -EOPNOTSUPP; \ | |
5175 | } | |
5176 | ||
5177 | IO_NETOP_PREP_ASYNC(sendmsg); | |
5178 | IO_NETOP_PREP_ASYNC(recvmsg); | |
5179 | IO_NETOP_PREP_ASYNC(connect); | |
5180 | IO_NETOP_PREP(accept); | |
5181 | IO_NETOP_FN(send); | |
5182 | IO_NETOP_FN(recv); | |
469956e8 | 5183 | #endif /* CONFIG_NET */ |
f8e85cf2 | 5184 | |
d7718a9d JA |
5185 | struct io_poll_table { |
5186 | struct poll_table_struct pt; | |
5187 | struct io_kiocb *req; | |
5188 | int error; | |
5189 | }; | |
ce593a6c | 5190 | |
d7718a9d JA |
5191 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
5192 | __poll_t mask, task_work_func_t func) | |
5193 | { | |
aa96bf8a | 5194 | int ret; |
d7718a9d JA |
5195 | |
5196 | /* for instances that support it check for an event match first: */ | |
5197 | if (mask && !(mask & poll->events)) | |
5198 | return 0; | |
5199 | ||
5200 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
5201 | ||
5202 | list_del_init(&poll->wait.entry); | |
5203 | ||
d7718a9d | 5204 | req->result = mask; |
7cbf1722 | 5205 | req->task_work.func = func; |
6d816e08 JA |
5206 | percpu_ref_get(&req->ctx->refs); |
5207 | ||
d7718a9d | 5208 | /* |
e3aabf95 JA |
5209 | * If this fails, then the task is exiting. When a task exits, the |
5210 | * work gets canceled, so just cancel this request as well instead | |
5211 | * of executing it. We can't safely execute it anyway, as we may not | |
5212 | * have the needed state needed for it anyway. | |
d7718a9d | 5213 | */ |
355fb9e2 | 5214 | ret = io_req_task_work_add(req); |
aa96bf8a | 5215 | if (unlikely(ret)) { |
e3aabf95 | 5216 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 5217 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 5218 | } |
d7718a9d JA |
5219 | return 1; |
5220 | } | |
5221 | ||
74ce6ce4 JA |
5222 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
5223 | __acquires(&req->ctx->completion_lock) | |
5224 | { | |
5225 | struct io_ring_ctx *ctx = req->ctx; | |
5226 | ||
5227 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5228 | struct poll_table_struct pt = { ._key = poll->events }; | |
5229 | ||
5230 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
5231 | } | |
5232 | ||
5233 | spin_lock_irq(&ctx->completion_lock); | |
5234 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
5235 | add_wait_queue(poll->head, &poll->wait); | |
5236 | return true; | |
5237 | } | |
5238 | ||
5239 | return false; | |
5240 | } | |
5241 | ||
d4e7cd36 | 5242 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 5243 | { |
e8c2bc1f | 5244 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 5245 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 5246 | return req->async_data; |
d4e7cd36 JA |
5247 | return req->apoll->double_poll; |
5248 | } | |
5249 | ||
5250 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
5251 | { | |
5252 | if (req->opcode == IORING_OP_POLL_ADD) | |
5253 | return &req->poll; | |
5254 | return &req->apoll->poll; | |
5255 | } | |
5256 | ||
5257 | static void io_poll_remove_double(struct io_kiocb *req) | |
5258 | { | |
5259 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
5260 | |
5261 | lockdep_assert_held(&req->ctx->completion_lock); | |
5262 | ||
5263 | if (poll && poll->head) { | |
5264 | struct wait_queue_head *head = poll->head; | |
5265 | ||
5266 | spin_lock(&head->lock); | |
5267 | list_del_init(&poll->wait.entry); | |
5268 | if (poll->wait.private) | |
5269 | refcount_dec(&req->refs); | |
5270 | poll->head = NULL; | |
5271 | spin_unlock(&head->lock); | |
5272 | } | |
5273 | } | |
5274 | ||
5275 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
5276 | { | |
5277 | struct io_ring_ctx *ctx = req->ctx; | |
5278 | ||
d4e7cd36 | 5279 | io_poll_remove_double(req); |
18bceab1 JA |
5280 | req->poll.done = true; |
5281 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
5282 | io_commit_cqring(ctx); | |
5283 | } | |
5284 | ||
dd221f46 | 5285 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 5286 | { |
dd221f46 | 5287 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 5288 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 5289 | struct io_kiocb *nxt; |
18bceab1 JA |
5290 | |
5291 | if (io_poll_rewait(req, &req->poll)) { | |
5292 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
5293 | } else { |
5294 | hash_del(&req->hash_node); | |
5295 | io_poll_complete(req, req->result, 0); | |
5296 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 5297 | |
dd221f46 PB |
5298 | nxt = io_put_req_find_next(req); |
5299 | io_cqring_ev_posted(ctx); | |
5300 | if (nxt) | |
5301 | __io_req_task_submit(nxt); | |
5302 | } | |
18bceab1 | 5303 | |
6d816e08 | 5304 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
5305 | } |
5306 | ||
5307 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
5308 | int sync, void *key) | |
5309 | { | |
5310 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 5311 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
5312 | __poll_t mask = key_to_poll(key); |
5313 | ||
5314 | /* for instances that support it check for an event match first: */ | |
5315 | if (mask && !(mask & poll->events)) | |
5316 | return 0; | |
5317 | ||
8706e04e JA |
5318 | list_del_init(&wait->entry); |
5319 | ||
807abcb0 | 5320 | if (poll && poll->head) { |
18bceab1 JA |
5321 | bool done; |
5322 | ||
807abcb0 JA |
5323 | spin_lock(&poll->head->lock); |
5324 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5325 | if (!done) |
807abcb0 | 5326 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5327 | /* make sure double remove sees this as being gone */ |
5328 | wait->private = NULL; | |
807abcb0 | 5329 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5330 | if (!done) { |
5331 | /* use wait func handler, so it matches the rq type */ | |
5332 | poll->wait.func(&poll->wait, mode, sync, key); | |
5333 | } | |
18bceab1 JA |
5334 | } |
5335 | refcount_dec(&req->refs); | |
5336 | return 1; | |
5337 | } | |
5338 | ||
5339 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5340 | wait_queue_func_t wake_func) | |
5341 | { | |
5342 | poll->head = NULL; | |
5343 | poll->done = false; | |
5344 | poll->canceled = false; | |
5345 | poll->events = events; | |
5346 | INIT_LIST_HEAD(&poll->wait.entry); | |
5347 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5348 | } | |
5349 | ||
5350 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5351 | struct wait_queue_head *head, |
5352 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5353 | { |
5354 | struct io_kiocb *req = pt->req; | |
5355 | ||
5356 | /* | |
5357 | * If poll->head is already set, it's because the file being polled | |
5358 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5359 | * for write). Setup a separate io_poll_iocb if this happens. | |
5360 | */ | |
5361 | if (unlikely(poll->head)) { | |
58852d4d PB |
5362 | struct io_poll_iocb *poll_one = poll; |
5363 | ||
18bceab1 | 5364 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5365 | if (*poll_ptr) { |
18bceab1 JA |
5366 | pt->error = -EINVAL; |
5367 | return; | |
5368 | } | |
5369 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); | |
5370 | if (!poll) { | |
5371 | pt->error = -ENOMEM; | |
5372 | return; | |
5373 | } | |
58852d4d | 5374 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5375 | refcount_inc(&req->refs); |
5376 | poll->wait.private = req; | |
807abcb0 | 5377 | *poll_ptr = poll; |
18bceab1 JA |
5378 | } |
5379 | ||
5380 | pt->error = 0; | |
5381 | poll->head = head; | |
a31eb4a2 JX |
5382 | |
5383 | if (poll->events & EPOLLEXCLUSIVE) | |
5384 | add_wait_queue_exclusive(head, &poll->wait); | |
5385 | else | |
5386 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5387 | } |
5388 | ||
5389 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5390 | struct poll_table_struct *p) | |
5391 | { | |
5392 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5393 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5394 | |
807abcb0 | 5395 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5396 | } |
5397 | ||
d7718a9d JA |
5398 | static void io_async_task_func(struct callback_head *cb) |
5399 | { | |
5400 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5401 | struct async_poll *apoll = req->apoll; | |
5402 | struct io_ring_ctx *ctx = req->ctx; | |
5403 | ||
5404 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5405 | ||
74ce6ce4 | 5406 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5407 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5408 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5409 | return; |
d7718a9d JA |
5410 | } |
5411 | ||
31067255 | 5412 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5413 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5414 | hash_del(&req->hash_node); |
2bae047e | 5415 | |
d4e7cd36 | 5416 | io_poll_remove_double(req); |
74ce6ce4 JA |
5417 | spin_unlock_irq(&ctx->completion_lock); |
5418 | ||
0be0b0e3 PB |
5419 | if (!READ_ONCE(apoll->poll.canceled)) |
5420 | __io_req_task_submit(req); | |
5421 | else | |
5422 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5423 | |
6d816e08 | 5424 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5425 | kfree(apoll->double_poll); |
31067255 | 5426 | kfree(apoll); |
d7718a9d JA |
5427 | } |
5428 | ||
5429 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5430 | void *key) | |
5431 | { | |
5432 | struct io_kiocb *req = wait->private; | |
5433 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5434 | ||
5435 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5436 | key_to_poll(key)); | |
5437 | ||
5438 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5439 | } | |
5440 | ||
5441 | static void io_poll_req_insert(struct io_kiocb *req) | |
5442 | { | |
5443 | struct io_ring_ctx *ctx = req->ctx; | |
5444 | struct hlist_head *list; | |
5445 | ||
5446 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5447 | hlist_add_head(&req->hash_node, list); | |
5448 | } | |
5449 | ||
5450 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5451 | struct io_poll_iocb *poll, | |
5452 | struct io_poll_table *ipt, __poll_t mask, | |
5453 | wait_queue_func_t wake_func) | |
5454 | __acquires(&ctx->completion_lock) | |
5455 | { | |
5456 | struct io_ring_ctx *ctx = req->ctx; | |
5457 | bool cancel = false; | |
5458 | ||
4d52f338 | 5459 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5460 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5461 | poll->file = req->file; |
18bceab1 | 5462 | poll->wait.private = req; |
d7718a9d JA |
5463 | |
5464 | ipt->pt._key = mask; | |
5465 | ipt->req = req; | |
5466 | ipt->error = -EINVAL; | |
5467 | ||
d7718a9d JA |
5468 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5469 | ||
5470 | spin_lock_irq(&ctx->completion_lock); | |
5471 | if (likely(poll->head)) { | |
5472 | spin_lock(&poll->head->lock); | |
5473 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5474 | if (ipt->error) | |
5475 | cancel = true; | |
5476 | ipt->error = 0; | |
5477 | mask = 0; | |
5478 | } | |
5479 | if (mask || ipt->error) | |
5480 | list_del_init(&poll->wait.entry); | |
5481 | else if (cancel) | |
5482 | WRITE_ONCE(poll->canceled, true); | |
5483 | else if (!poll->done) /* actually waiting for an event */ | |
5484 | io_poll_req_insert(req); | |
5485 | spin_unlock(&poll->head->lock); | |
5486 | } | |
5487 | ||
5488 | return mask; | |
5489 | } | |
5490 | ||
5491 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5492 | { | |
5493 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5494 | struct io_ring_ctx *ctx = req->ctx; | |
5495 | struct async_poll *apoll; | |
5496 | struct io_poll_table ipt; | |
5497 | __poll_t mask, ret; | |
9dab14b8 | 5498 | int rw; |
d7718a9d JA |
5499 | |
5500 | if (!req->file || !file_can_poll(req->file)) | |
5501 | return false; | |
24c74678 | 5502 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5503 | return false; |
9dab14b8 JA |
5504 | if (def->pollin) |
5505 | rw = READ; | |
5506 | else if (def->pollout) | |
5507 | rw = WRITE; | |
5508 | else | |
5509 | return false; | |
5510 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5511 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5512 | return false; |
5513 | ||
5514 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5515 | if (unlikely(!apoll)) | |
5516 | return false; | |
807abcb0 | 5517 | apoll->double_poll = NULL; |
d7718a9d JA |
5518 | |
5519 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5520 | req->apoll = apoll; |
d7718a9d | 5521 | |
8755d97a | 5522 | mask = 0; |
d7718a9d | 5523 | if (def->pollin) |
8755d97a | 5524 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5525 | if (def->pollout) |
5526 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5527 | |
5528 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5529 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5530 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5531 | mask &= ~POLLIN; | |
5532 | ||
d7718a9d JA |
5533 | mask |= POLLERR | POLLPRI; |
5534 | ||
5535 | ipt.pt._qproc = io_async_queue_proc; | |
5536 | ||
5537 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5538 | io_async_wake); | |
a36da65c | 5539 | if (ret || ipt.error) { |
d4e7cd36 | 5540 | io_poll_remove_double(req); |
d7718a9d | 5541 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5542 | kfree(apoll->double_poll); |
d7718a9d JA |
5543 | kfree(apoll); |
5544 | return false; | |
5545 | } | |
5546 | spin_unlock_irq(&ctx->completion_lock); | |
5547 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5548 | apoll->poll.events); | |
5549 | return true; | |
5550 | } | |
5551 | ||
5552 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5553 | struct io_poll_iocb *poll) | |
221c5eb2 | 5554 | { |
b41e9852 | 5555 | bool do_complete = false; |
221c5eb2 JA |
5556 | |
5557 | spin_lock(&poll->head->lock); | |
5558 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5559 | if (!list_empty(&poll->wait.entry)) { |
5560 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5561 | do_complete = true; |
221c5eb2 JA |
5562 | } |
5563 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5564 | hash_del(&req->hash_node); |
d7718a9d JA |
5565 | return do_complete; |
5566 | } | |
5567 | ||
5568 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5569 | { | |
5570 | bool do_complete; | |
5571 | ||
d4e7cd36 JA |
5572 | io_poll_remove_double(req); |
5573 | ||
d7718a9d JA |
5574 | if (req->opcode == IORING_OP_POLL_ADD) { |
5575 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5576 | } else { | |
3bfa5bcb JA |
5577 | struct async_poll *apoll = req->apoll; |
5578 | ||
d7718a9d | 5579 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5580 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5581 | if (do_complete) { | |
d7718a9d | 5582 | io_put_req(req); |
807abcb0 | 5583 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5584 | kfree(apoll); |
5585 | } | |
b1f573bd XW |
5586 | } |
5587 | ||
b41e9852 JA |
5588 | if (do_complete) { |
5589 | io_cqring_fill_event(req, -ECANCELED); | |
5590 | io_commit_cqring(req->ctx); | |
f254ac04 | 5591 | req_set_fail_links(req); |
216578e5 | 5592 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5593 | } |
5594 | ||
5595 | return do_complete; | |
221c5eb2 JA |
5596 | } |
5597 | ||
76e1b642 JA |
5598 | /* |
5599 | * Returns true if we found and killed one or more poll requests | |
5600 | */ | |
6b81928d PB |
5601 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5602 | struct files_struct *files) | |
221c5eb2 | 5603 | { |
78076bb6 | 5604 | struct hlist_node *tmp; |
221c5eb2 | 5605 | struct io_kiocb *req; |
8e2e1faf | 5606 | int posted = 0, i; |
221c5eb2 JA |
5607 | |
5608 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5609 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5610 | struct hlist_head *list; | |
5611 | ||
5612 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5613 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5614 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5615 | posted += io_poll_remove_one(req); |
5616 | } | |
221c5eb2 JA |
5617 | } |
5618 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5619 | |
8e2e1faf JA |
5620 | if (posted) |
5621 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5622 | |
5623 | return posted != 0; | |
221c5eb2 JA |
5624 | } |
5625 | ||
47f46768 JA |
5626 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5627 | { | |
78076bb6 | 5628 | struct hlist_head *list; |
47f46768 JA |
5629 | struct io_kiocb *req; |
5630 | ||
78076bb6 JA |
5631 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5632 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5633 | if (sqe_addr != req->user_data) |
5634 | continue; | |
5635 | if (io_poll_remove_one(req)) | |
eac406c6 | 5636 | return 0; |
b41e9852 | 5637 | return -EALREADY; |
47f46768 JA |
5638 | } |
5639 | ||
5640 | return -ENOENT; | |
5641 | } | |
5642 | ||
3529d8c2 JA |
5643 | static int io_poll_remove_prep(struct io_kiocb *req, |
5644 | const struct io_uring_sqe *sqe) | |
0969e783 | 5645 | { |
0969e783 JA |
5646 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5647 | return -EINVAL; | |
5648 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5649 | sqe->poll_events) | |
5650 | return -EINVAL; | |
5651 | ||
018043be | 5652 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5653 | return 0; |
5654 | } | |
5655 | ||
221c5eb2 JA |
5656 | /* |
5657 | * Find a running poll command that matches one specified in sqe->addr, | |
5658 | * and remove it if found. | |
5659 | */ | |
61e98203 | 5660 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5661 | { |
5662 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5663 | int ret; |
221c5eb2 | 5664 | |
221c5eb2 | 5665 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5666 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5667 | spin_unlock_irq(&ctx->completion_lock); |
5668 | ||
4e88d6e7 JA |
5669 | if (ret < 0) |
5670 | req_set_fail_links(req); | |
e1e16097 | 5671 | io_req_complete(req, ret); |
221c5eb2 JA |
5672 | return 0; |
5673 | } | |
5674 | ||
221c5eb2 JA |
5675 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5676 | void *key) | |
5677 | { | |
c2f2eb7d JA |
5678 | struct io_kiocb *req = wait->private; |
5679 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5680 | |
d7718a9d | 5681 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5682 | } |
5683 | ||
221c5eb2 JA |
5684 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5685 | struct poll_table_struct *p) | |
5686 | { | |
5687 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5688 | ||
e8c2bc1f | 5689 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5690 | } |
5691 | ||
3529d8c2 | 5692 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5693 | { |
5694 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5695 | u32 events; |
221c5eb2 JA |
5696 | |
5697 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5698 | return -EINVAL; | |
5699 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5700 | return -EINVAL; | |
5701 | ||
5769a351 JX |
5702 | events = READ_ONCE(sqe->poll32_events); |
5703 | #ifdef __BIG_ENDIAN | |
5704 | events = swahw32(events); | |
5705 | #endif | |
a31eb4a2 JX |
5706 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5707 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5708 | return 0; |
5709 | } | |
5710 | ||
61e98203 | 5711 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5712 | { |
5713 | struct io_poll_iocb *poll = &req->poll; | |
5714 | struct io_ring_ctx *ctx = req->ctx; | |
5715 | struct io_poll_table ipt; | |
0969e783 | 5716 | __poll_t mask; |
0969e783 | 5717 | |
d7718a9d | 5718 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5719 | |
d7718a9d JA |
5720 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5721 | io_poll_wake); | |
221c5eb2 | 5722 | |
8c838788 | 5723 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5724 | ipt.error = 0; |
b0dd8a41 | 5725 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5726 | } |
221c5eb2 JA |
5727 | spin_unlock_irq(&ctx->completion_lock); |
5728 | ||
8c838788 JA |
5729 | if (mask) { |
5730 | io_cqring_ev_posted(ctx); | |
014db007 | 5731 | io_put_req(req); |
221c5eb2 | 5732 | } |
8c838788 | 5733 | return ipt.error; |
221c5eb2 JA |
5734 | } |
5735 | ||
5262f567 JA |
5736 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5737 | { | |
ad8a48ac JA |
5738 | struct io_timeout_data *data = container_of(timer, |
5739 | struct io_timeout_data, timer); | |
5740 | struct io_kiocb *req = data->req; | |
5741 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5742 | unsigned long flags; |
5743 | ||
5262f567 | 5744 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5745 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5746 | atomic_set(&req->ctx->cq_timeouts, |
5747 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5748 | ||
78e19bbe | 5749 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5750 | io_commit_cqring(ctx); |
5751 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5752 | ||
5753 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5754 | req_set_fail_links(req); |
5262f567 JA |
5755 | io_put_req(req); |
5756 | return HRTIMER_NORESTART; | |
5757 | } | |
5758 | ||
fbd15848 PB |
5759 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5760 | __u64 user_data) | |
f254ac04 | 5761 | { |
fbd15848 | 5762 | struct io_timeout_data *io; |
47f46768 JA |
5763 | struct io_kiocb *req; |
5764 | int ret = -ENOENT; | |
f254ac04 | 5765 | |
135fcde8 | 5766 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5767 | if (user_data == req->user_data) { |
47f46768 JA |
5768 | ret = 0; |
5769 | break; | |
5770 | } | |
5771 | } | |
5772 | ||
5773 | if (ret == -ENOENT) | |
fbd15848 PB |
5774 | return ERR_PTR(ret); |
5775 | ||
5776 | io = req->async_data; | |
e8c2bc1f | 5777 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5778 | if (ret == -1) |
fbd15848 | 5779 | return ERR_PTR(-EALREADY); |
a71976f3 | 5780 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5781 | return req; |
5782 | } | |
47f46768 | 5783 | |
fbd15848 PB |
5784 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5785 | { | |
5786 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5787 | ||
5788 | if (IS_ERR(req)) | |
5789 | return PTR_ERR(req); | |
f254ac04 JA |
5790 | |
5791 | req_set_fail_links(req); | |
f254ac04 | 5792 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5793 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5794 | return 0; |
5795 | } | |
5796 | ||
9c8e11b3 PB |
5797 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5798 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5799 | { |
9c8e11b3 PB |
5800 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5801 | struct io_timeout_data *data; | |
47f46768 | 5802 | |
9c8e11b3 PB |
5803 | if (IS_ERR(req)) |
5804 | return PTR_ERR(req); | |
47f46768 | 5805 | |
9c8e11b3 PB |
5806 | req->timeout.off = 0; /* noseq */ |
5807 | data = req->async_data; | |
5808 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5809 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5810 | data->timer.function = io_timeout_fn; | |
5811 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5812 | return 0; | |
47f46768 JA |
5813 | } |
5814 | ||
3529d8c2 JA |
5815 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5816 | const struct io_uring_sqe *sqe) | |
b29472ee | 5817 | { |
9c8e11b3 PB |
5818 | struct io_timeout_rem *tr = &req->timeout_rem; |
5819 | ||
b29472ee JA |
5820 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5821 | return -EINVAL; | |
61710e43 DA |
5822 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5823 | return -EINVAL; | |
9c8e11b3 | 5824 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5825 | return -EINVAL; |
5826 | ||
9c8e11b3 PB |
5827 | tr->addr = READ_ONCE(sqe->addr); |
5828 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5829 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5830 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5831 | return -EINVAL; | |
5832 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5833 | return -EFAULT; | |
5834 | } else if (tr->flags) { | |
5835 | /* timeout removal doesn't support flags */ | |
b29472ee | 5836 | return -EINVAL; |
9c8e11b3 | 5837 | } |
b29472ee | 5838 | |
b29472ee JA |
5839 | return 0; |
5840 | } | |
5841 | ||
8662daec PB |
5842 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5843 | { | |
5844 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5845 | : HRTIMER_MODE_REL; | |
5846 | } | |
5847 | ||
11365043 JA |
5848 | /* |
5849 | * Remove or update an existing timeout command | |
5850 | */ | |
61e98203 | 5851 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5852 | { |
9c8e11b3 | 5853 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5854 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5855 | int ret; |
11365043 | 5856 | |
11365043 | 5857 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5858 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5859 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5860 | else |
5861 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5862 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5863 | |
47f46768 | 5864 | io_cqring_fill_event(req, ret); |
11365043 JA |
5865 | io_commit_cqring(ctx); |
5866 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5867 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5868 | if (ret < 0) |
5869 | req_set_fail_links(req); | |
ec9c02ad | 5870 | io_put_req(req); |
11365043 | 5871 | return 0; |
5262f567 JA |
5872 | } |
5873 | ||
3529d8c2 | 5874 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5875 | bool is_timeout_link) |
5262f567 | 5876 | { |
ad8a48ac | 5877 | struct io_timeout_data *data; |
a41525ab | 5878 | unsigned flags; |
56080b02 | 5879 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5880 | |
ad8a48ac | 5881 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5882 | return -EINVAL; |
ad8a48ac | 5883 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5884 | return -EINVAL; |
56080b02 | 5885 | if (off && is_timeout_link) |
2d28390a | 5886 | return -EINVAL; |
a41525ab JA |
5887 | flags = READ_ONCE(sqe->timeout_flags); |
5888 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5889 | return -EINVAL; |
bdf20073 | 5890 | |
bfe68a22 | 5891 | req->timeout.off = off; |
26a61679 | 5892 | |
e8c2bc1f | 5893 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5894 | return -ENOMEM; |
5895 | ||
e8c2bc1f | 5896 | data = req->async_data; |
ad8a48ac | 5897 | data->req = req; |
ad8a48ac JA |
5898 | |
5899 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5900 | return -EFAULT; |
5901 | ||
8662daec | 5902 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac JA |
5903 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
5904 | return 0; | |
5905 | } | |
5906 | ||
61e98203 | 5907 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5908 | { |
ad8a48ac | 5909 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5910 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5911 | struct list_head *entry; |
bfe68a22 | 5912 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5913 | |
733f5c95 | 5914 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5915 | |
5262f567 JA |
5916 | /* |
5917 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5918 | * timeout event to be satisfied. If it isn't set, then this is |
5919 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5920 | */ |
8eb7e2d0 | 5921 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5922 | entry = ctx->timeout_list.prev; |
5923 | goto add; | |
5924 | } | |
5262f567 | 5925 | |
bfe68a22 PB |
5926 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5927 | req->timeout.target_seq = tail + off; | |
5262f567 | 5928 | |
f010505b MDG |
5929 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5930 | * This is safe because ->completion_lock is held, and submissions | |
5931 | * and completions are never mixed in the same ->completion_lock section. | |
5932 | */ | |
5933 | ctx->cq_last_tm_flush = tail; | |
5934 | ||
5262f567 JA |
5935 | /* |
5936 | * Insertion sort, ensuring the first entry in the list is always | |
5937 | * the one we need first. | |
5938 | */ | |
5262f567 | 5939 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5940 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5941 | timeout.list); | |
5262f567 | 5942 | |
8eb7e2d0 | 5943 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5944 | continue; |
bfe68a22 PB |
5945 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5946 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5947 | break; |
5948 | } | |
93bd25bb | 5949 | add: |
135fcde8 | 5950 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5951 | data->timer.function = io_timeout_fn; |
5952 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5953 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5954 | return 0; |
5955 | } | |
5262f567 | 5956 | |
62755e35 JA |
5957 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5958 | { | |
5959 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5960 | ||
5961 | return req->user_data == (unsigned long) data; | |
5962 | } | |
5963 | ||
5aa75ed5 | 5964 | static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr) |
62755e35 | 5965 | { |
62755e35 | 5966 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5967 | int ret = 0; |
5968 | ||
5aa75ed5 JA |
5969 | if (!tctx->io_wq) |
5970 | return -ENOENT; | |
5971 | ||
5972 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false); | |
62755e35 JA |
5973 | switch (cancel_ret) { |
5974 | case IO_WQ_CANCEL_OK: | |
5975 | ret = 0; | |
5976 | break; | |
5977 | case IO_WQ_CANCEL_RUNNING: | |
5978 | ret = -EALREADY; | |
5979 | break; | |
5980 | case IO_WQ_CANCEL_NOTFOUND: | |
5981 | ret = -ENOENT; | |
5982 | break; | |
5983 | } | |
5984 | ||
e977d6d3 JA |
5985 | return ret; |
5986 | } | |
5987 | ||
47f46768 JA |
5988 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5989 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5990 | int success_ret) |
47f46768 JA |
5991 | { |
5992 | unsigned long flags; | |
5993 | int ret; | |
5994 | ||
5aa75ed5 JA |
5995 | ret = io_async_cancel_one(req->task->io_uring, |
5996 | (void *) (unsigned long) sqe_addr); | |
47f46768 JA |
5997 | if (ret != -ENOENT) { |
5998 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5999 | goto done; | |
6000 | } | |
6001 | ||
6002 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
6003 | ret = io_timeout_cancel(ctx, sqe_addr); | |
6004 | if (ret != -ENOENT) | |
6005 | goto done; | |
6006 | ret = io_poll_cancel(ctx, sqe_addr); | |
6007 | done: | |
b0dd8a41 JA |
6008 | if (!ret) |
6009 | ret = success_ret; | |
47f46768 JA |
6010 | io_cqring_fill_event(req, ret); |
6011 | io_commit_cqring(ctx); | |
6012 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
6013 | io_cqring_ev_posted(ctx); | |
6014 | ||
4e88d6e7 JA |
6015 | if (ret < 0) |
6016 | req_set_fail_links(req); | |
014db007 | 6017 | io_put_req(req); |
47f46768 JA |
6018 | } |
6019 | ||
3529d8c2 JA |
6020 | static int io_async_cancel_prep(struct io_kiocb *req, |
6021 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 6022 | { |
fbf23849 | 6023 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 6024 | return -EINVAL; |
61710e43 DA |
6025 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6026 | return -EINVAL; | |
6027 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
6028 | return -EINVAL; |
6029 | ||
fbf23849 JA |
6030 | req->cancel.addr = READ_ONCE(sqe->addr); |
6031 | return 0; | |
6032 | } | |
6033 | ||
61e98203 | 6034 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6035 | { |
6036 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 6037 | |
014db007 | 6038 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
6039 | return 0; |
6040 | } | |
6041 | ||
269bbe5f | 6042 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
6043 | const struct io_uring_sqe *sqe) |
6044 | { | |
6ca56f84 JA |
6045 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
6046 | return -EINVAL; | |
61710e43 DA |
6047 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
6048 | return -EINVAL; | |
6049 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
6050 | return -EINVAL; |
6051 | ||
269bbe5f BM |
6052 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
6053 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
6054 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 6055 | return -EINVAL; |
269bbe5f | 6056 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
6057 | return 0; |
6058 | } | |
6059 | ||
889fca73 | 6060 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
6061 | { |
6062 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 6063 | struct io_uring_rsrc_update up; |
05f3fb3c | 6064 | int ret; |
fbf23849 | 6065 | |
45d189c6 | 6066 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 6067 | return -EAGAIN; |
05f3fb3c | 6068 | |
269bbe5f BM |
6069 | up.offset = req->rsrc_update.offset; |
6070 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
6071 | |
6072 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 6073 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
6074 | mutex_unlock(&ctx->uring_lock); |
6075 | ||
6076 | if (ret < 0) | |
6077 | req_set_fail_links(req); | |
889fca73 | 6078 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
6079 | return 0; |
6080 | } | |
6081 | ||
bfe76559 | 6082 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 6083 | { |
d625c6ee | 6084 | switch (req->opcode) { |
e781573e | 6085 | case IORING_OP_NOP: |
bfe76559 | 6086 | return 0; |
f67676d1 JA |
6087 | case IORING_OP_READV: |
6088 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 6089 | case IORING_OP_READ: |
bfe76559 | 6090 | return io_read_prep(req, sqe); |
f67676d1 JA |
6091 | case IORING_OP_WRITEV: |
6092 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 6093 | case IORING_OP_WRITE: |
bfe76559 | 6094 | return io_write_prep(req, sqe); |
0969e783 | 6095 | case IORING_OP_POLL_ADD: |
bfe76559 | 6096 | return io_poll_add_prep(req, sqe); |
0969e783 | 6097 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 6098 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 6099 | case IORING_OP_FSYNC: |
1155c76a | 6100 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 6101 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 6102 | return io_sfr_prep(req, sqe); |
03b1230c | 6103 | case IORING_OP_SENDMSG: |
fddaface | 6104 | case IORING_OP_SEND: |
bfe76559 | 6105 | return io_sendmsg_prep(req, sqe); |
03b1230c | 6106 | case IORING_OP_RECVMSG: |
fddaface | 6107 | case IORING_OP_RECV: |
bfe76559 | 6108 | return io_recvmsg_prep(req, sqe); |
f499a021 | 6109 | case IORING_OP_CONNECT: |
bfe76559 | 6110 | return io_connect_prep(req, sqe); |
2d28390a | 6111 | case IORING_OP_TIMEOUT: |
bfe76559 | 6112 | return io_timeout_prep(req, sqe, false); |
b29472ee | 6113 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 6114 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 6115 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 6116 | return io_async_cancel_prep(req, sqe); |
2d28390a | 6117 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 6118 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 6119 | case IORING_OP_ACCEPT: |
bfe76559 | 6120 | return io_accept_prep(req, sqe); |
d63d1b5e | 6121 | case IORING_OP_FALLOCATE: |
bfe76559 | 6122 | return io_fallocate_prep(req, sqe); |
15b71abe | 6123 | case IORING_OP_OPENAT: |
bfe76559 | 6124 | return io_openat_prep(req, sqe); |
b5dba59e | 6125 | case IORING_OP_CLOSE: |
bfe76559 | 6126 | return io_close_prep(req, sqe); |
05f3fb3c | 6127 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 6128 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 6129 | case IORING_OP_STATX: |
bfe76559 | 6130 | return io_statx_prep(req, sqe); |
4840e418 | 6131 | case IORING_OP_FADVISE: |
bfe76559 | 6132 | return io_fadvise_prep(req, sqe); |
c1ca757b | 6133 | case IORING_OP_MADVISE: |
bfe76559 | 6134 | return io_madvise_prep(req, sqe); |
cebdb986 | 6135 | case IORING_OP_OPENAT2: |
bfe76559 | 6136 | return io_openat2_prep(req, sqe); |
3e4827b0 | 6137 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 6138 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 6139 | case IORING_OP_SPLICE: |
bfe76559 | 6140 | return io_splice_prep(req, sqe); |
ddf0322d | 6141 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 6142 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 6143 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 6144 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 6145 | case IORING_OP_TEE: |
bfe76559 | 6146 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
6147 | case IORING_OP_SHUTDOWN: |
6148 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
6149 | case IORING_OP_RENAMEAT: |
6150 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
6151 | case IORING_OP_UNLINKAT: |
6152 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
6153 | } |
6154 | ||
bfe76559 PB |
6155 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
6156 | req->opcode); | |
6157 | return-EINVAL; | |
6158 | } | |
6159 | ||
93642ef8 | 6160 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 6161 | { |
93642ef8 PB |
6162 | switch (req->opcode) { |
6163 | case IORING_OP_READV: | |
6164 | case IORING_OP_READ_FIXED: | |
6165 | case IORING_OP_READ: | |
6166 | return io_rw_prep_async(req, READ); | |
6167 | case IORING_OP_WRITEV: | |
6168 | case IORING_OP_WRITE_FIXED: | |
6169 | case IORING_OP_WRITE: | |
6170 | return io_rw_prep_async(req, WRITE); | |
6171 | case IORING_OP_SENDMSG: | |
6172 | case IORING_OP_SEND: | |
6173 | return io_sendmsg_prep_async(req); | |
6174 | case IORING_OP_RECVMSG: | |
6175 | case IORING_OP_RECV: | |
6176 | return io_recvmsg_prep_async(req); | |
6177 | case IORING_OP_CONNECT: | |
6178 | return io_connect_prep_async(req); | |
6179 | } | |
6180 | return 0; | |
6181 | } | |
6182 | ||
be7053b7 | 6183 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 6184 | { |
be7053b7 | 6185 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 6186 | return 0; |
be7053b7 | 6187 | /* some opcodes init it during the inital prep */ |
93642ef8 | 6188 | if (req->async_data) |
be7053b7 PB |
6189 | return 0; |
6190 | if (__io_alloc_async_data(req)) | |
bfe76559 | 6191 | return -EAGAIN; |
be7053b7 | 6192 | return io_req_prep_async(req); |
f67676d1 JA |
6193 | } |
6194 | ||
9cf7c104 PB |
6195 | static u32 io_get_sequence(struct io_kiocb *req) |
6196 | { | |
6197 | struct io_kiocb *pos; | |
6198 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 6199 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 6200 | |
f2f87370 PB |
6201 | io_for_each_link(pos, req) |
6202 | nr_reqs++; | |
9cf7c104 PB |
6203 | |
6204 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6205 | return total_submitted - nr_reqs; | |
6206 | } | |
6207 | ||
be7053b7 | 6208 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 6209 | { |
a197f664 | 6210 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6211 | struct io_defer_entry *de; |
f67676d1 | 6212 | int ret; |
9cf7c104 | 6213 | u32 seq; |
de0617e4 | 6214 | |
9d858b21 | 6215 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6216 | if (likely(list_empty_careful(&ctx->defer_list) && |
6217 | !(req->flags & REQ_F_IO_DRAIN))) | |
6218 | return 0; | |
6219 | ||
6220 | seq = io_get_sequence(req); | |
6221 | /* Still a chance to pass the sequence check */ | |
6222 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6223 | return 0; |
6224 | ||
be7053b7 PB |
6225 | ret = io_req_defer_prep(req); |
6226 | if (ret) | |
6227 | return ret; | |
cbdcb435 | 6228 | io_prep_async_link(req); |
27dc8338 PB |
6229 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6230 | if (!de) | |
6231 | return -ENOMEM; | |
2d28390a | 6232 | |
de0617e4 | 6233 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6234 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6235 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6236 | kfree(de); |
ae34817b PB |
6237 | io_queue_async_work(req); |
6238 | return -EIOCBQUEUED; | |
de0617e4 JA |
6239 | } |
6240 | ||
915967f6 | 6241 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6242 | de->req = req; |
9cf7c104 | 6243 | de->seq = seq; |
27dc8338 | 6244 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6245 | spin_unlock_irq(&ctx->completion_lock); |
6246 | return -EIOCBQUEUED; | |
6247 | } | |
6248 | ||
3ca405eb | 6249 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6250 | { |
0e1b6fe3 PB |
6251 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6252 | switch (req->opcode) { | |
6253 | case IORING_OP_READV: | |
6254 | case IORING_OP_READ_FIXED: | |
6255 | case IORING_OP_READ: | |
bcda7baa | 6256 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6257 | break; |
6258 | case IORING_OP_RECVMSG: | |
6259 | case IORING_OP_RECV: | |
bcda7baa | 6260 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6261 | break; |
6262 | } | |
6263 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6264 | } |
6265 | ||
0e1b6fe3 PB |
6266 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6267 | switch (req->opcode) { | |
6268 | case IORING_OP_READV: | |
6269 | case IORING_OP_READ_FIXED: | |
6270 | case IORING_OP_READ: | |
6271 | case IORING_OP_WRITEV: | |
6272 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6273 | case IORING_OP_WRITE: { |
6274 | struct io_async_rw *io = req->async_data; | |
6275 | if (io->free_iovec) | |
6276 | kfree(io->free_iovec); | |
0e1b6fe3 | 6277 | break; |
e8c2bc1f | 6278 | } |
0e1b6fe3 | 6279 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6280 | case IORING_OP_SENDMSG: { |
6281 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6282 | |
6283 | kfree(io->free_iov); | |
0e1b6fe3 | 6284 | break; |
e8c2bc1f | 6285 | } |
0e1b6fe3 PB |
6286 | case IORING_OP_SPLICE: |
6287 | case IORING_OP_TEE: | |
6288 | io_put_file(req, req->splice.file_in, | |
6289 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6290 | break; | |
f3cd4850 JA |
6291 | case IORING_OP_OPENAT: |
6292 | case IORING_OP_OPENAT2: | |
6293 | if (req->open.filename) | |
6294 | putname(req->open.filename); | |
6295 | break; | |
80a261fd JA |
6296 | case IORING_OP_RENAMEAT: |
6297 | putname(req->rename.oldpath); | |
6298 | putname(req->rename.newpath); | |
6299 | break; | |
14a1143b JA |
6300 | case IORING_OP_UNLINKAT: |
6301 | putname(req->unlink.filename); | |
6302 | break; | |
0e1b6fe3 PB |
6303 | } |
6304 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6305 | } |
99bc4c38 PB |
6306 | } |
6307 | ||
889fca73 | 6308 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6309 | { |
a197f664 | 6310 | struct io_ring_ctx *ctx = req->ctx; |
d625c6ee | 6311 | int ret; |
2b188cc1 | 6312 | |
d625c6ee | 6313 | switch (req->opcode) { |
2b188cc1 | 6314 | case IORING_OP_NOP: |
889fca73 | 6315 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6316 | break; |
6317 | case IORING_OP_READV: | |
edafccee | 6318 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6319 | case IORING_OP_READ: |
889fca73 | 6320 | ret = io_read(req, issue_flags); |
edafccee | 6321 | break; |
3529d8c2 | 6322 | case IORING_OP_WRITEV: |
edafccee | 6323 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6324 | case IORING_OP_WRITE: |
889fca73 | 6325 | ret = io_write(req, issue_flags); |
2b188cc1 | 6326 | break; |
c992fe29 | 6327 | case IORING_OP_FSYNC: |
45d189c6 | 6328 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6329 | break; |
221c5eb2 | 6330 | case IORING_OP_POLL_ADD: |
61e98203 | 6331 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6332 | break; |
6333 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6334 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6335 | break; |
5d17b4a4 | 6336 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6337 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6338 | break; |
0fa03c62 | 6339 | case IORING_OP_SENDMSG: |
889fca73 | 6340 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6341 | break; |
fddaface | 6342 | case IORING_OP_SEND: |
889fca73 | 6343 | ret = io_send(req, issue_flags); |
0fa03c62 | 6344 | break; |
aa1fa28f | 6345 | case IORING_OP_RECVMSG: |
889fca73 | 6346 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6347 | break; |
fddaface | 6348 | case IORING_OP_RECV: |
889fca73 | 6349 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6350 | break; |
5262f567 | 6351 | case IORING_OP_TIMEOUT: |
61e98203 | 6352 | ret = io_timeout(req, issue_flags); |
5262f567 | 6353 | break; |
11365043 | 6354 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6355 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6356 | break; |
17f2fe35 | 6357 | case IORING_OP_ACCEPT: |
889fca73 | 6358 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6359 | break; |
f8e85cf2 | 6360 | case IORING_OP_CONNECT: |
889fca73 | 6361 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6362 | break; |
62755e35 | 6363 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6364 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6365 | break; |
d63d1b5e | 6366 | case IORING_OP_FALLOCATE: |
45d189c6 | 6367 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6368 | break; |
15b71abe | 6369 | case IORING_OP_OPENAT: |
45d189c6 | 6370 | ret = io_openat(req, issue_flags); |
15b71abe | 6371 | break; |
b5dba59e | 6372 | case IORING_OP_CLOSE: |
889fca73 | 6373 | ret = io_close(req, issue_flags); |
b5dba59e | 6374 | break; |
05f3fb3c | 6375 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6376 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6377 | break; |
eddc7ef5 | 6378 | case IORING_OP_STATX: |
45d189c6 | 6379 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6380 | break; |
4840e418 | 6381 | case IORING_OP_FADVISE: |
45d189c6 | 6382 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6383 | break; |
c1ca757b | 6384 | case IORING_OP_MADVISE: |
45d189c6 | 6385 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6386 | break; |
cebdb986 | 6387 | case IORING_OP_OPENAT2: |
45d189c6 | 6388 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6389 | break; |
3e4827b0 | 6390 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6391 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6392 | break; |
7d67af2c | 6393 | case IORING_OP_SPLICE: |
45d189c6 | 6394 | ret = io_splice(req, issue_flags); |
7d67af2c | 6395 | break; |
ddf0322d | 6396 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6397 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6398 | break; |
067524e9 | 6399 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6400 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6401 | break; |
f2a8d5c7 | 6402 | case IORING_OP_TEE: |
45d189c6 | 6403 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6404 | break; |
36f4fa68 | 6405 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6406 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6407 | break; |
80a261fd | 6408 | case IORING_OP_RENAMEAT: |
45d189c6 | 6409 | ret = io_renameat(req, issue_flags); |
80a261fd | 6410 | break; |
14a1143b | 6411 | case IORING_OP_UNLINKAT: |
45d189c6 | 6412 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6413 | break; |
2b188cc1 JA |
6414 | default: |
6415 | ret = -EINVAL; | |
6416 | break; | |
6417 | } | |
6418 | ||
def596e9 JA |
6419 | if (ret) |
6420 | return ret; | |
6421 | ||
b532576e JA |
6422 | /* If the op doesn't have a file, we're not polling for it */ |
6423 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6424 | const bool in_async = io_wq_current_is_worker(); |
6425 | ||
11ba820b JA |
6426 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6427 | if (in_async) | |
6428 | mutex_lock(&ctx->uring_lock); | |
6429 | ||
2e9dbe90 | 6430 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6431 | |
6432 | if (in_async) | |
6433 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6434 | } |
6435 | ||
6436 | return 0; | |
2b188cc1 JA |
6437 | } |
6438 | ||
5280f7e5 | 6439 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6440 | { |
6441 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6442 | struct io_kiocb *timeout; |
561fb04a | 6443 | int ret = 0; |
2b188cc1 | 6444 | |
6df1db6b PB |
6445 | timeout = io_prep_linked_timeout(req); |
6446 | if (timeout) | |
6447 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6448 | |
4014d943 | 6449 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6450 | ret = -ECANCELED; |
31b51510 | 6451 | |
561fb04a | 6452 | if (!ret) { |
561fb04a | 6453 | do { |
889fca73 | 6454 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6455 | /* |
6456 | * We can get EAGAIN for polled IO even though we're | |
6457 | * forcing a sync submission from here, since we can't | |
6458 | * wait for request slots on the block side. | |
6459 | */ | |
6460 | if (ret != -EAGAIN) | |
6461 | break; | |
6462 | cond_resched(); | |
6463 | } while (1); | |
6464 | } | |
31b51510 | 6465 | |
a3df7698 | 6466 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6467 | if (ret) { |
a3df7698 PB |
6468 | /* io-wq is going to take one down */ |
6469 | refcount_inc(&req->refs); | |
6470 | io_req_task_queue_fail(req, ret); | |
edafccee | 6471 | } |
2b188cc1 JA |
6472 | } |
6473 | ||
65e19f54 JA |
6474 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6475 | int index) | |
6476 | { | |
269bbe5f | 6477 | struct fixed_rsrc_table *table; |
65e19f54 | 6478 | |
05f3fb3c | 6479 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6480 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6481 | } |
6482 | ||
8371adf5 PB |
6483 | static struct file *io_file_get(struct io_submit_state *state, |
6484 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6485 | { |
a197f664 | 6486 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6487 | struct file *file; |
09bb8394 | 6488 | |
8da11c19 | 6489 | if (fixed) { |
479f517b | 6490 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6491 | return NULL; |
b7620121 | 6492 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6493 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6494 | io_set_resource_node(req); |
09bb8394 | 6495 | } else { |
c826bd7a | 6496 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6497 | file = __io_file_get(state, fd); |
09bb8394 JA |
6498 | } |
6499 | ||
ce3d5aae PB |
6500 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6501 | io_req_track_inflight(req); | |
8371adf5 | 6502 | return file; |
09bb8394 JA |
6503 | } |
6504 | ||
2665abfd | 6505 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6506 | { |
ad8a48ac JA |
6507 | struct io_timeout_data *data = container_of(timer, |
6508 | struct io_timeout_data, timer); | |
90cd7e42 | 6509 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6510 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6511 | unsigned long flags; |
2665abfd JA |
6512 | |
6513 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6514 | prev = req->timeout.head; |
6515 | req->timeout.head = NULL; | |
2665abfd JA |
6516 | |
6517 | /* | |
6518 | * We don't expect the list to be empty, that will only happen if we | |
6519 | * race with the completion of the linked work. | |
6520 | */ | |
90cd7e42 | 6521 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6522 | io_remove_next_linked(prev); |
90cd7e42 PB |
6523 | else |
6524 | prev = NULL; | |
2665abfd JA |
6525 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6526 | ||
6527 | if (prev) { | |
4e88d6e7 | 6528 | req_set_fail_links(prev); |
014db007 | 6529 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6530 | io_put_req_deferred(prev, 1); |
47f46768 | 6531 | } else { |
9ae1f8dd PB |
6532 | io_req_complete_post(req, -ETIME, 0); |
6533 | io_put_req_deferred(req, 1); | |
2665abfd | 6534 | } |
2665abfd JA |
6535 | return HRTIMER_NORESTART; |
6536 | } | |
6537 | ||
7271ef3a | 6538 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6539 | { |
76a46e06 | 6540 | /* |
f2f87370 PB |
6541 | * If the back reference is NULL, then our linked request finished |
6542 | * before we got a chance to setup the timer | |
76a46e06 | 6543 | */ |
90cd7e42 | 6544 | if (req->timeout.head) { |
e8c2bc1f | 6545 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6546 | |
ad8a48ac JA |
6547 | data->timer.function = io_link_timeout_fn; |
6548 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6549 | data->mode); | |
2665abfd | 6550 | } |
7271ef3a JA |
6551 | } |
6552 | ||
6553 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6554 | { | |
6555 | struct io_ring_ctx *ctx = req->ctx; | |
6556 | ||
6557 | spin_lock_irq(&ctx->completion_lock); | |
6558 | __io_queue_linked_timeout(req); | |
76a46e06 | 6559 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6560 | |
2665abfd | 6561 | /* drop submission reference */ |
76a46e06 JA |
6562 | io_put_req(req); |
6563 | } | |
2665abfd | 6564 | |
ad8a48ac | 6565 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6566 | { |
f2f87370 | 6567 | struct io_kiocb *nxt = req->link; |
2665abfd | 6568 | |
f2f87370 PB |
6569 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6570 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6571 | return NULL; |
2665abfd | 6572 | |
90cd7e42 | 6573 | nxt->timeout.head = req; |
900fad45 | 6574 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6575 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6576 | return nxt; |
2665abfd JA |
6577 | } |
6578 | ||
c5eef2b9 | 6579 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6580 | { |
d3d7298d | 6581 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
193155c8 | 6582 | const struct cred *old_creds = NULL; |
e0c5c576 | 6583 | int ret; |
2b188cc1 | 6584 | |
2e5aa6cb PB |
6585 | if ((req->flags & REQ_F_WORK_INITIALIZED) && |
6586 | (req->work.flags & IO_WQ_WORK_CREDS) && | |
d3d7298d PB |
6587 | req->work.identity->creds != current_cred()) |
6588 | old_creds = override_creds(req->work.identity->creds); | |
193155c8 | 6589 | |
c5eef2b9 | 6590 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6591 | |
d3d7298d PB |
6592 | if (old_creds) |
6593 | revert_creds(old_creds); | |
491381ce JA |
6594 | |
6595 | /* | |
6596 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6597 | * doesn't support non-blocking read/write attempts | |
6598 | */ | |
24c74678 | 6599 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6600 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6601 | /* |
6602 | * Queued up for async execution, worker will release | |
6603 | * submit reference when the iocb is actually submitted. | |
6604 | */ | |
6605 | io_queue_async_work(req); | |
2b188cc1 | 6606 | } |
0d63c148 PB |
6607 | } else if (likely(!ret)) { |
6608 | /* drop submission reference */ | |
e342c807 | 6609 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6610 | struct io_ring_ctx *ctx = req->ctx; |
6611 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6612 | |
6dd0be1e | 6613 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6614 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6615 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6616 | } else { |
d3d7298d | 6617 | io_put_req(req); |
0d63c148 PB |
6618 | } |
6619 | } else { | |
4e88d6e7 | 6620 | req_set_fail_links(req); |
e65ef56d | 6621 | io_put_req(req); |
e1e16097 | 6622 | io_req_complete(req, ret); |
9e645e11 | 6623 | } |
d3d7298d PB |
6624 | if (linked_timeout) |
6625 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6626 | } |
6627 | ||
be7053b7 | 6628 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6629 | { |
6630 | int ret; | |
6631 | ||
be7053b7 | 6632 | ret = io_req_defer(req); |
4fe2c963 JL |
6633 | if (ret) { |
6634 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6635 | fail_req: |
4e88d6e7 | 6636 | req_set_fail_links(req); |
e1e16097 JA |
6637 | io_put_req(req); |
6638 | io_req_complete(req, ret); | |
4fe2c963 | 6639 | } |
2550878f | 6640 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6641 | ret = io_req_defer_prep(req); |
6642 | if (unlikely(ret)) | |
6643 | goto fail_req; | |
ce35a47a JA |
6644 | io_queue_async_work(req); |
6645 | } else { | |
c5eef2b9 | 6646 | __io_queue_sqe(req); |
ce35a47a | 6647 | } |
4fe2c963 JL |
6648 | } |
6649 | ||
b16fed66 PB |
6650 | /* |
6651 | * Check SQE restrictions (opcode and flags). | |
6652 | * | |
6653 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6654 | */ | |
6655 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6656 | struct io_kiocb *req, | |
6657 | unsigned int sqe_flags) | |
4fe2c963 | 6658 | { |
b16fed66 PB |
6659 | if (!ctx->restricted) |
6660 | return true; | |
6661 | ||
6662 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6663 | return false; | |
6664 | ||
6665 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6666 | ctx->restrictions.sqe_flags_required) | |
6667 | return false; | |
6668 | ||
6669 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6670 | ctx->restrictions.sqe_flags_required)) | |
6671 | return false; | |
6672 | ||
6673 | return true; | |
4fe2c963 JL |
6674 | } |
6675 | ||
b16fed66 PB |
6676 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6677 | const struct io_uring_sqe *sqe) | |
6678 | { | |
6679 | struct io_submit_state *state; | |
6680 | unsigned int sqe_flags; | |
6681 | int id, ret = 0; | |
6682 | ||
6683 | req->opcode = READ_ONCE(sqe->opcode); | |
6684 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6685 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6686 | req->user_data = READ_ONCE(sqe->user_data); | |
6687 | req->async_data = NULL; | |
6688 | req->file = NULL; | |
6689 | req->ctx = ctx; | |
6690 | req->link = NULL; | |
6691 | req->fixed_rsrc_refs = NULL; | |
6692 | /* one is dropped after submission, the other at completion */ | |
6693 | refcount_set(&req->refs, 2); | |
6694 | req->task = current; | |
6695 | req->result = 0; | |
863e0560 | 6696 | |
b16fed66 | 6697 | /* enforce forwards compatibility on users */ |
ebf4a5db PB |
6698 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6699 | req->flags = 0; | |
b16fed66 | 6700 | return -EINVAL; |
ebf4a5db | 6701 | } |
b16fed66 PB |
6702 | |
6703 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6704 | return -EINVAL; | |
6705 | ||
6706 | if (unlikely(io_sq_thread_acquire_mm_files(ctx, req))) | |
6707 | return -EFAULT; | |
6708 | ||
6709 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) | |
6710 | return -EACCES; | |
6711 | ||
6712 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6713 | !io_op_defs[req->opcode].buffer_select) | |
6714 | return -EOPNOTSUPP; | |
6715 | ||
6716 | id = READ_ONCE(sqe->personality); | |
6717 | if (id) { | |
6718 | struct io_identity *iod; | |
6719 | ||
6720 | iod = idr_find(&ctx->personality_idr, id); | |
6721 | if (unlikely(!iod)) | |
6722 | return -EINVAL; | |
6723 | refcount_inc(&iod->count); | |
6724 | ||
6725 | __io_req_init_async(req); | |
6726 | get_cred(iod->creds); | |
6727 | req->work.identity = iod; | |
6728 | req->work.flags |= IO_WQ_WORK_CREDS; | |
6729 | } | |
6730 | ||
6731 | state = &ctx->submit_state; | |
6732 | ||
6733 | /* | |
6734 | * Plug now if we have more than 1 IO left after this, and the target | |
6735 | * is potentially a read/write to block based storage. | |
6736 | */ | |
6737 | if (!state->plug_started && state->ios_left > 1 && | |
6738 | io_op_defs[req->opcode].plug) { | |
6739 | blk_start_plug(&state->plug); | |
6740 | state->plug_started = true; | |
6741 | } | |
6742 | ||
6743 | if (io_op_defs[req->opcode].needs_file) { | |
6744 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6745 | ||
6746 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6747 | if (unlikely(!req->file)) | |
6748 | ret = -EBADF; | |
6749 | } | |
6750 | ||
6751 | state->ios_left--; | |
6752 | return ret; | |
6753 | } | |
6754 | ||
a6b8cadc | 6755 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6756 | const struct io_uring_sqe *sqe) |
9e645e11 | 6757 | { |
a1ab7b35 | 6758 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6759 | int ret; |
9e645e11 | 6760 | |
a6b8cadc PB |
6761 | ret = io_init_req(ctx, req, sqe); |
6762 | if (unlikely(ret)) { | |
6763 | fail_req: | |
6764 | io_put_req(req); | |
6765 | io_req_complete(req, ret); | |
de59bc10 PB |
6766 | if (link->head) { |
6767 | /* fail even hard links since we don't submit */ | |
cf109604 | 6768 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6769 | io_put_req(link->head); |
6770 | io_req_complete(link->head, -ECANCELED); | |
6771 | link->head = NULL; | |
6772 | } | |
a6b8cadc PB |
6773 | return ret; |
6774 | } | |
be7053b7 PB |
6775 | ret = io_req_prep(req, sqe); |
6776 | if (unlikely(ret)) | |
6777 | goto fail_req; | |
a6b8cadc | 6778 | |
be7053b7 | 6779 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6780 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6781 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6782 | ||
9e645e11 JA |
6783 | /* |
6784 | * If we already have a head request, queue this one for async | |
6785 | * submittal once the head completes. If we don't have a head but | |
6786 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6787 | * submitted sync once the chain is complete. If none of those | |
6788 | * conditions are true (normal request), then just queue it. | |
6789 | */ | |
863e0560 PB |
6790 | if (link->head) { |
6791 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6792 | |
8cdf2193 PB |
6793 | /* |
6794 | * Taking sequential execution of a link, draining both sides | |
6795 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6796 | * requests in the link. So, it drains the head and the | |
6797 | * next after the link request. The last one is done via | |
6798 | * drain_next flag to persist the effect across calls. | |
6799 | */ | |
ef4ff581 | 6800 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6801 | head->flags |= REQ_F_IO_DRAIN; |
6802 | ctx->drain_next = 1; | |
6803 | } | |
be7053b7 | 6804 | ret = io_req_defer_prep(req); |
cf109604 | 6805 | if (unlikely(ret)) |
a6b8cadc | 6806 | goto fail_req; |
9d76377f | 6807 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6808 | link->last->link = req; |
863e0560 | 6809 | link->last = req; |
32fe525b PB |
6810 | |
6811 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6812 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6813 | io_queue_sqe(head); |
863e0560 | 6814 | link->head = NULL; |
32fe525b | 6815 | } |
9e645e11 | 6816 | } else { |
711be031 PB |
6817 | if (unlikely(ctx->drain_next)) { |
6818 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6819 | ctx->drain_next = 0; |
711be031 | 6820 | } |
ef4ff581 | 6821 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6822 | link->head = req; |
6823 | link->last = req; | |
711be031 | 6824 | } else { |
be7053b7 | 6825 | io_queue_sqe(req); |
711be031 | 6826 | } |
9e645e11 | 6827 | } |
2e6e1fde | 6828 | |
1d4240cc | 6829 | return 0; |
9e645e11 JA |
6830 | } |
6831 | ||
9a56a232 JA |
6832 | /* |
6833 | * Batched submission is done, ensure local IO is flushed out. | |
6834 | */ | |
ba88ff11 PB |
6835 | static void io_submit_state_end(struct io_submit_state *state, |
6836 | struct io_ring_ctx *ctx) | |
9a56a232 | 6837 | { |
a1ab7b35 | 6838 | if (state->link.head) |
de59bc10 | 6839 | io_queue_sqe(state->link.head); |
6dd0be1e | 6840 | if (state->comp.nr) |
ba88ff11 | 6841 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6842 | if (state->plug_started) |
6843 | blk_finish_plug(&state->plug); | |
9f13c35b | 6844 | io_state_file_put(state); |
9a56a232 JA |
6845 | } |
6846 | ||
6847 | /* | |
6848 | * Start submission side cache. | |
6849 | */ | |
6850 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6851 | unsigned int max_ios) |
9a56a232 | 6852 | { |
27926b68 | 6853 | state->plug_started = false; |
9a56a232 | 6854 | state->ios_left = max_ios; |
a1ab7b35 PB |
6855 | /* set only head, no need to init link_last in advance */ |
6856 | state->link.head = NULL; | |
9a56a232 JA |
6857 | } |
6858 | ||
2b188cc1 JA |
6859 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6860 | { | |
75b28aff | 6861 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6862 | |
caf582c6 PB |
6863 | /* |
6864 | * Ensure any loads from the SQEs are done at this point, | |
6865 | * since once we write the new head, the application could | |
6866 | * write new data to them. | |
6867 | */ | |
6868 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6869 | } |
6870 | ||
2b188cc1 | 6871 | /* |
3529d8c2 | 6872 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6873 | * that is mapped by userspace. This means that care needs to be taken to |
6874 | * ensure that reads are stable, as we cannot rely on userspace always | |
6875 | * being a good citizen. If members of the sqe are validated and then later | |
6876 | * used, it's important that those reads are done through READ_ONCE() to | |
6877 | * prevent a re-load down the line. | |
6878 | */ | |
709b302f | 6879 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6880 | { |
75b28aff | 6881 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6882 | unsigned head; |
6883 | ||
6884 | /* | |
6885 | * The cached sq head (or cq tail) serves two purposes: | |
6886 | * | |
6887 | * 1) allows us to batch the cost of updating the user visible | |
6888 | * head updates. | |
6889 | * 2) allows the kernel side to track the head on its own, even | |
6890 | * though the application is the one updating it. | |
6891 | */ | |
4fccfcbb | 6892 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6893 | if (likely(head < ctx->sq_entries)) |
6894 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6895 | |
6896 | /* drop invalid entries */ | |
498ccd9e | 6897 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6898 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6899 | return NULL; |
6900 | } | |
6901 | ||
0f212204 | 6902 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6903 | { |
46c4e16a | 6904 | int submitted = 0; |
6c271ce2 | 6905 | |
c4a2ed72 | 6906 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6907 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6908 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6909 | return -EBUSY; |
6910 | } | |
6c271ce2 | 6911 | |
ee7d46d9 PB |
6912 | /* make sure SQ entry isn't read before tail */ |
6913 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6914 | |
2b85edfc PB |
6915 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6916 | return -EAGAIN; | |
6c271ce2 | 6917 | |
d8a6df10 | 6918 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6919 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6920 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6921 | |
46c4e16a | 6922 | while (submitted < nr) { |
3529d8c2 | 6923 | const struct io_uring_sqe *sqe; |
196be95c | 6924 | struct io_kiocb *req; |
fb5ccc98 | 6925 | |
258b29a9 | 6926 | req = io_alloc_req(ctx); |
196be95c PB |
6927 | if (unlikely(!req)) { |
6928 | if (!submitted) | |
6929 | submitted = -EAGAIN; | |
fb5ccc98 | 6930 | break; |
196be95c | 6931 | } |
4fccfcbb PB |
6932 | sqe = io_get_sqe(ctx); |
6933 | if (unlikely(!sqe)) { | |
6934 | kmem_cache_free(req_cachep, req); | |
6935 | break; | |
6936 | } | |
d3656344 JA |
6937 | /* will complete beyond this point, count as submitted */ |
6938 | submitted++; | |
a1ab7b35 | 6939 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6940 | break; |
6c271ce2 JA |
6941 | } |
6942 | ||
9466f437 PB |
6943 | if (unlikely(submitted != nr)) { |
6944 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6945 | struct io_uring_task *tctx = current->io_uring; |
6946 | int unused = nr - ref_used; | |
9466f437 | 6947 | |
d8a6df10 JA |
6948 | percpu_ref_put_many(&ctx->refs, unused); |
6949 | percpu_counter_sub(&tctx->inflight, unused); | |
6950 | put_task_struct_many(current, unused); | |
9466f437 | 6951 | } |
6c271ce2 | 6952 | |
a1ab7b35 | 6953 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6954 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6955 | io_commit_sqring(ctx); | |
6956 | ||
6c271ce2 JA |
6957 | return submitted; |
6958 | } | |
6959 | ||
23b3628e XW |
6960 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6961 | { | |
6962 | /* Tell userspace we may need a wakeup call */ | |
6963 | spin_lock_irq(&ctx->completion_lock); | |
6964 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6965 | spin_unlock_irq(&ctx->completion_lock); | |
6966 | } | |
6967 | ||
6968 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6969 | { | |
6970 | spin_lock_irq(&ctx->completion_lock); | |
6971 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6972 | spin_unlock_irq(&ctx->completion_lock); | |
6973 | } | |
6974 | ||
08369246 | 6975 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6976 | { |
c8d1ba58 | 6977 | unsigned int to_submit; |
bdcd3eab | 6978 | int ret = 0; |
6c271ce2 | 6979 | |
c8d1ba58 | 6980 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6981 | /* if we're handling multiple rings, cap submit size for fairness */ |
6982 | if (cap_entries && to_submit > 8) | |
6983 | to_submit = 8; | |
6984 | ||
906a3c6f | 6985 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6986 | unsigned nr_events = 0; |
a4c0b3de | 6987 | |
c8d1ba58 | 6988 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6989 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6990 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6991 | |
d9d05217 PB |
6992 | if (to_submit && !ctx->sqo_dead && |
6993 | likely(!percpu_ref_is_dying(&ctx->refs))) | |
08369246 | 6994 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6995 | mutex_unlock(&ctx->uring_lock); |
6996 | } | |
6c271ce2 | 6997 | |
90554200 JA |
6998 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6999 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 7000 | |
08369246 XW |
7001 | return ret; |
7002 | } | |
6c271ce2 | 7003 | |
08369246 XW |
7004 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
7005 | { | |
7006 | struct io_ring_ctx *ctx; | |
7007 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 7008 | |
08369246 XW |
7009 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7010 | if (sq_thread_idle < ctx->sq_thread_idle) | |
7011 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 7012 | } |
c1edbf5f | 7013 | |
08369246 | 7014 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 7015 | } |
6c271ce2 | 7016 | |
69fb2131 JA |
7017 | static void io_sqd_init_new(struct io_sq_data *sqd) |
7018 | { | |
7019 | struct io_ring_ctx *ctx; | |
7020 | ||
7021 | while (!list_empty(&sqd->ctx_new_list)) { | |
7022 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
7023 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
7024 | complete(&ctx->sq_thread_comp); | |
7025 | } | |
08369246 XW |
7026 | |
7027 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
7028 | } |
7029 | ||
c8d1ba58 JA |
7030 | static int io_sq_thread(void *data) |
7031 | { | |
91d8f519 | 7032 | struct cgroup_subsys_state *cur_css = NULL; |
28cea78a JA |
7033 | struct files_struct *old_files = current->files; |
7034 | struct nsproxy *old_nsproxy = current->nsproxy; | |
69fb2131 JA |
7035 | const struct cred *old_cred = NULL; |
7036 | struct io_sq_data *sqd = data; | |
7037 | struct io_ring_ctx *ctx; | |
a0d9205f | 7038 | unsigned long timeout = 0; |
08369246 | 7039 | DEFINE_WAIT(wait); |
6c271ce2 | 7040 | |
28cea78a JA |
7041 | task_lock(current); |
7042 | current->files = NULL; | |
7043 | current->nsproxy = NULL; | |
7044 | task_unlock(current); | |
6c271ce2 | 7045 | |
69fb2131 | 7046 | while (!kthread_should_stop()) { |
08369246 XW |
7047 | int ret; |
7048 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
7049 | |
7050 | /* | |
69fb2131 JA |
7051 | * Any changes to the sqd lists are synchronized through the |
7052 | * kthread parking. This synchronizes the thread vs users, | |
7053 | * the users are synchronized on the sqd->ctx_lock. | |
c1edbf5f | 7054 | */ |
65b2b213 | 7055 | if (kthread_should_park()) { |
69fb2131 | 7056 | kthread_parkme(); |
65b2b213 XW |
7057 | /* |
7058 | * When sq thread is unparked, in case the previous park operation | |
7059 | * comes from io_put_sq_data(), which means that sq thread is going | |
7060 | * to be stopped, so here needs to have a check. | |
7061 | */ | |
7062 | if (kthread_should_stop()) | |
7063 | break; | |
7064 | } | |
7143b5ac | 7065 | |
08369246 | 7066 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 7067 | io_sqd_init_new(sqd); |
08369246 XW |
7068 | timeout = jiffies + sqd->sq_thread_idle; |
7069 | } | |
6c271ce2 | 7070 | |
08369246 | 7071 | sqt_spin = false; |
e95eee2d | 7072 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 JA |
7073 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7074 | if (current->cred != ctx->creds) { | |
7075 | if (old_cred) | |
7076 | revert_creds(old_cred); | |
7077 | old_cred = override_creds(ctx->creds); | |
bdcd3eab | 7078 | } |
91d8f519 | 7079 | io_sq_thread_associate_blkcg(ctx, &cur_css); |
4ea33a97 JA |
7080 | #ifdef CONFIG_AUDIT |
7081 | current->loginuid = ctx->loginuid; | |
7082 | current->sessionid = ctx->sessionid; | |
7083 | #endif | |
bdcd3eab | 7084 | |
08369246 XW |
7085 | ret = __io_sq_thread(ctx, cap_entries); |
7086 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
7087 | sqt_spin = true; | |
6c271ce2 | 7088 | |
28cea78a | 7089 | io_sq_thread_drop_mm_files(); |
69fb2131 | 7090 | } |
6c271ce2 | 7091 | |
08369246 | 7092 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 | 7093 | io_run_task_work(); |
d434ab6d | 7094 | io_sq_thread_drop_mm_files(); |
c8d1ba58 | 7095 | cond_resched(); |
08369246 XW |
7096 | if (sqt_spin) |
7097 | timeout = jiffies + sqd->sq_thread_idle; | |
7098 | continue; | |
7099 | } | |
7100 | ||
08369246 XW |
7101 | needs_sched = true; |
7102 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
7103 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
7104 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
7105 | !list_empty_careful(&ctx->iopoll_list)) { | |
7106 | needs_sched = false; | |
7107 | break; | |
7108 | } | |
7109 | if (io_sqring_entries(ctx)) { | |
7110 | needs_sched = false; | |
7111 | break; | |
7112 | } | |
7113 | } | |
7114 | ||
8b28fdf2 | 7115 | if (needs_sched && !kthread_should_park()) { |
69fb2131 JA |
7116 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7117 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 7118 | |
69fb2131 | 7119 | schedule(); |
69fb2131 JA |
7120 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
7121 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 7122 | } |
08369246 XW |
7123 | |
7124 | finish_wait(&sqd->wait, &wait); | |
7125 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
7126 | } |
7127 | ||
4c6e277c | 7128 | io_run_task_work(); |
d434ab6d | 7129 | io_sq_thread_drop_mm_files(); |
b41e9852 | 7130 | |
91d8f519 DZ |
7131 | if (cur_css) |
7132 | io_sq_thread_unassociate_blkcg(); | |
69fb2131 JA |
7133 | if (old_cred) |
7134 | revert_creds(old_cred); | |
06058632 | 7135 | |
28cea78a JA |
7136 | task_lock(current); |
7137 | current->files = old_files; | |
7138 | current->nsproxy = old_nsproxy; | |
7139 | task_unlock(current); | |
7140 | ||
2bbcd6d3 | 7141 | kthread_parkme(); |
06058632 | 7142 | |
6c271ce2 JA |
7143 | return 0; |
7144 | } | |
7145 | ||
bda52162 JA |
7146 | struct io_wait_queue { |
7147 | struct wait_queue_entry wq; | |
7148 | struct io_ring_ctx *ctx; | |
7149 | unsigned to_wait; | |
7150 | unsigned nr_timeouts; | |
7151 | }; | |
7152 | ||
6c503150 | 7153 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
7154 | { |
7155 | struct io_ring_ctx *ctx = iowq->ctx; | |
7156 | ||
7157 | /* | |
d195a66e | 7158 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
7159 | * started waiting. For timeouts, we always want to return to userspace, |
7160 | * regardless of event count. | |
7161 | */ | |
6c503150 | 7162 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
7163 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
7164 | } | |
7165 | ||
7166 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
7167 | int wake_flags, void *key) | |
7168 | { | |
7169 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
7170 | wq); | |
7171 | ||
6c503150 PB |
7172 | /* |
7173 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
7174 | * the task, and the next invocation will do it. | |
7175 | */ | |
7176 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
7177 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
7178 | return -1; | |
bda52162 JA |
7179 | } |
7180 | ||
af9c1a44 JA |
7181 | static int io_run_task_work_sig(void) |
7182 | { | |
7183 | if (io_run_task_work()) | |
7184 | return 1; | |
7185 | if (!signal_pending(current)) | |
7186 | return 0; | |
792ee0f6 JA |
7187 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
7188 | return -ERESTARTSYS; | |
af9c1a44 JA |
7189 | return -EINTR; |
7190 | } | |
7191 | ||
eeb60b9a PB |
7192 | /* when returns >0, the caller should retry */ |
7193 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
7194 | struct io_wait_queue *iowq, | |
7195 | signed long *timeout) | |
7196 | { | |
7197 | int ret; | |
7198 | ||
7199 | /* make sure we run task_work before checking for signals */ | |
7200 | ret = io_run_task_work_sig(); | |
7201 | if (ret || io_should_wake(iowq)) | |
7202 | return ret; | |
7203 | /* let the caller flush overflows, retry */ | |
7204 | if (test_bit(0, &ctx->cq_check_overflow)) | |
7205 | return 1; | |
7206 | ||
7207 | *timeout = schedule_timeout(*timeout); | |
7208 | return !*timeout ? -ETIME : 1; | |
7209 | } | |
7210 | ||
2b188cc1 JA |
7211 | /* |
7212 | * Wait until events become available, if we don't already have some. The | |
7213 | * application must reap them itself, as they reside on the shared cq ring. | |
7214 | */ | |
7215 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
7216 | const sigset_t __user *sig, size_t sigsz, |
7217 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 7218 | { |
bda52162 JA |
7219 | struct io_wait_queue iowq = { |
7220 | .wq = { | |
7221 | .private = current, | |
7222 | .func = io_wake_function, | |
7223 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7224 | }, | |
7225 | .ctx = ctx, | |
7226 | .to_wait = min_events, | |
7227 | }; | |
75b28aff | 7228 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7229 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7230 | int ret; | |
2b188cc1 | 7231 | |
b41e9852 | 7232 | do { |
6c503150 PB |
7233 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
7234 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 7235 | return 0; |
4c6e277c | 7236 | if (!io_run_task_work()) |
b41e9852 | 7237 | break; |
b41e9852 | 7238 | } while (1); |
2b188cc1 JA |
7239 | |
7240 | if (sig) { | |
9e75ad5d AB |
7241 | #ifdef CONFIG_COMPAT |
7242 | if (in_compat_syscall()) | |
7243 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7244 | sigsz); |
9e75ad5d AB |
7245 | else |
7246 | #endif | |
b772434b | 7247 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7248 | |
2b188cc1 JA |
7249 | if (ret) |
7250 | return ret; | |
7251 | } | |
7252 | ||
c73ebb68 | 7253 | if (uts) { |
c1d5a224 PB |
7254 | struct timespec64 ts; |
7255 | ||
c73ebb68 HX |
7256 | if (get_timespec64(&ts, uts)) |
7257 | return -EFAULT; | |
7258 | timeout = timespec64_to_jiffies(&ts); | |
7259 | } | |
7260 | ||
bda52162 | 7261 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7262 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7263 | do { |
6c503150 | 7264 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
bda52162 JA |
7265 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7266 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7267 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7268 | finish_wait(&ctx->wait, &iowq.wq); | |
7269 | } while (ret > 0); | |
bda52162 | 7270 | |
b7db41c9 | 7271 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7272 | |
75b28aff | 7273 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7274 | } |
7275 | ||
6b06314c JA |
7276 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7277 | { | |
7278 | #if defined(CONFIG_UNIX) | |
7279 | if (ctx->ring_sock) { | |
7280 | struct sock *sock = ctx->ring_sock->sk; | |
7281 | struct sk_buff *skb; | |
7282 | ||
7283 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7284 | kfree_skb(skb); | |
7285 | } | |
7286 | #else | |
7287 | int i; | |
7288 | ||
65e19f54 JA |
7289 | for (i = 0; i < ctx->nr_user_files; i++) { |
7290 | struct file *file; | |
7291 | ||
7292 | file = io_file_from_index(ctx, i); | |
7293 | if (file) | |
7294 | fput(file); | |
7295 | } | |
6b06314c JA |
7296 | #endif |
7297 | } | |
7298 | ||
00835dce | 7299 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7300 | { |
269bbe5f | 7301 | struct fixed_rsrc_data *data; |
05f3fb3c | 7302 | |
269bbe5f | 7303 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7304 | complete(&data->done); |
7305 | } | |
7306 | ||
2a63b2d9 | 7307 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7308 | { |
2a63b2d9 | 7309 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7310 | } |
7311 | ||
2a63b2d9 | 7312 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7313 | { |
2a63b2d9 BM |
7314 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7315 | } | |
65e19f54 | 7316 | |
d67d2263 BM |
7317 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7318 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7319 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7320 | { |
2a63b2d9 | 7321 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7322 | rsrc_data->node = ref_node; |
d67d2263 | 7323 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7324 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7325 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7326 | } |
7327 | ||
8bad28d8 | 7328 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 7329 | { |
8bad28d8 | 7330 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 7331 | |
2a63b2d9 | 7332 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7333 | ref_node = data->node; |
e6cb007c | 7334 | data->node = NULL; |
2a63b2d9 | 7335 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7336 | if (ref_node) |
7337 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 | 7338 | } |
05589553 | 7339 | |
8bad28d8 HX |
7340 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, |
7341 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
7342 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7343 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7344 | { |
f2303b1f | 7345 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 7346 | int ret; |
05589553 | 7347 | |
8bad28d8 HX |
7348 | if (data->quiesce) |
7349 | return -ENXIO; | |
05589553 | 7350 | |
8bad28d8 | 7351 | data->quiesce = true; |
1ffc5422 | 7352 | do { |
f2303b1f PB |
7353 | ret = -ENOMEM; |
7354 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7355 | if (!backup_node) | |
7356 | break; | |
7357 | backup_node->rsrc_data = data; | |
7358 | backup_node->rsrc_put = rsrc_put; | |
7359 | ||
8bad28d8 HX |
7360 | io_sqe_rsrc_kill_node(ctx, data); |
7361 | percpu_ref_kill(&data->refs); | |
7362 | flush_delayed_work(&ctx->rsrc_put_work); | |
7363 | ||
1ffc5422 | 7364 | ret = wait_for_completion_interruptible(&data->done); |
88f171ab | 7365 | if (!ret || !io_refs_resurrect(&data->refs, &data->done)) |
1ffc5422 | 7366 | break; |
8bad28d8 | 7367 | |
8bad28d8 HX |
7368 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7369 | backup_node = NULL; | |
8bad28d8 | 7370 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7371 | ret = io_run_task_work_sig(); |
8bad28d8 | 7372 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7373 | } while (ret >= 0); |
8bad28d8 | 7374 | data->quiesce = false; |
05f3fb3c | 7375 | |
8bad28d8 HX |
7376 | if (backup_node) |
7377 | destroy_fixed_rsrc_ref_node(backup_node); | |
7378 | return ret; | |
d7954b2b BM |
7379 | } |
7380 | ||
1ad555c6 BM |
7381 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7382 | { | |
7383 | struct fixed_rsrc_data *data; | |
7384 | ||
7385 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7386 | if (!data) | |
7387 | return NULL; | |
7388 | ||
00835dce | 7389 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7390 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7391 | kfree(data); | |
7392 | return NULL; | |
7393 | } | |
7394 | data->ctx = ctx; | |
7395 | init_completion(&data->done); | |
7396 | return data; | |
7397 | } | |
7398 | ||
7399 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7400 | { | |
7401 | percpu_ref_exit(&data->refs); | |
7402 | kfree(data->table); | |
7403 | kfree(data); | |
7404 | } | |
7405 | ||
d7954b2b BM |
7406 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7407 | { | |
7408 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7409 | unsigned nr_tables, i; |
7410 | int ret; | |
7411 | ||
8bad28d8 HX |
7412 | /* |
7413 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7414 | * Since we possibly drop uring lock later in this function to | |
7415 | * run task work. | |
7416 | */ | |
7417 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7418 | return -ENXIO; |
f2303b1f | 7419 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7420 | if (ret) |
7421 | return ret; | |
7422 | ||
6b06314c | 7423 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7424 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7425 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7426 | kfree(data->table[i].files); |
1ad555c6 | 7427 | free_fixed_rsrc_data(data); |
05f3fb3c | 7428 | ctx->file_data = NULL; |
6b06314c JA |
7429 | ctx->nr_user_files = 0; |
7430 | return 0; | |
7431 | } | |
7432 | ||
534ca6d6 | 7433 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7434 | { |
534ca6d6 | 7435 | if (refcount_dec_and_test(&sqd->refs)) { |
2bbcd6d3 RP |
7436 | /* |
7437 | * The park is a bit of a work-around, without it we get | |
7438 | * warning spews on shutdown with SQPOLL set and affinity | |
7439 | * set to a single CPU. | |
7440 | */ | |
534ca6d6 JA |
7441 | if (sqd->thread) { |
7442 | kthread_park(sqd->thread); | |
7443 | kthread_stop(sqd->thread); | |
7444 | } | |
7445 | ||
7446 | kfree(sqd); | |
7447 | } | |
7448 | } | |
7449 | ||
aa06165d JA |
7450 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7451 | { | |
7452 | struct io_ring_ctx *ctx_attach; | |
7453 | struct io_sq_data *sqd; | |
7454 | struct fd f; | |
7455 | ||
7456 | f = fdget(p->wq_fd); | |
7457 | if (!f.file) | |
7458 | return ERR_PTR(-ENXIO); | |
7459 | if (f.file->f_op != &io_uring_fops) { | |
7460 | fdput(f); | |
7461 | return ERR_PTR(-EINVAL); | |
7462 | } | |
7463 | ||
7464 | ctx_attach = f.file->private_data; | |
7465 | sqd = ctx_attach->sq_data; | |
7466 | if (!sqd) { | |
7467 | fdput(f); | |
7468 | return ERR_PTR(-EINVAL); | |
7469 | } | |
7470 | ||
7471 | refcount_inc(&sqd->refs); | |
7472 | fdput(f); | |
7473 | return sqd; | |
7474 | } | |
7475 | ||
534ca6d6 JA |
7476 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7477 | { | |
7478 | struct io_sq_data *sqd; | |
7479 | ||
aa06165d JA |
7480 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7481 | return io_attach_sq_data(p); | |
7482 | ||
534ca6d6 JA |
7483 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7484 | if (!sqd) | |
7485 | return ERR_PTR(-ENOMEM); | |
7486 | ||
7487 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7488 | INIT_LIST_HEAD(&sqd->ctx_list); |
7489 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7490 | mutex_init(&sqd->ctx_lock); | |
7491 | mutex_init(&sqd->lock); | |
534ca6d6 JA |
7492 | init_waitqueue_head(&sqd->wait); |
7493 | return sqd; | |
7494 | } | |
7495 | ||
69fb2131 JA |
7496 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7497 | __releases(&sqd->lock) | |
7498 | { | |
7499 | if (!sqd->thread) | |
7500 | return; | |
7501 | kthread_unpark(sqd->thread); | |
7502 | mutex_unlock(&sqd->lock); | |
7503 | } | |
7504 | ||
7505 | static void io_sq_thread_park(struct io_sq_data *sqd) | |
7506 | __acquires(&sqd->lock) | |
7507 | { | |
7508 | if (!sqd->thread) | |
7509 | return; | |
7510 | mutex_lock(&sqd->lock); | |
7511 | kthread_park(sqd->thread); | |
7512 | } | |
7513 | ||
534ca6d6 JA |
7514 | static void io_sq_thread_stop(struct io_ring_ctx *ctx) |
7515 | { | |
7516 | struct io_sq_data *sqd = ctx->sq_data; | |
7517 | ||
7518 | if (sqd) { | |
7519 | if (sqd->thread) { | |
7520 | /* | |
7521 | * We may arrive here from the error branch in | |
7522 | * io_sq_offload_create() where the kthread is created | |
7523 | * without being waked up, thus wake it up now to make | |
7524 | * sure the wait will complete. | |
7525 | */ | |
7526 | wake_up_process(sqd->thread); | |
7527 | wait_for_completion(&ctx->sq_thread_comp); | |
69fb2131 JA |
7528 | |
7529 | io_sq_thread_park(sqd); | |
7530 | } | |
7531 | ||
7532 | mutex_lock(&sqd->ctx_lock); | |
7533 | list_del(&ctx->sqd_list); | |
08369246 | 7534 | io_sqd_update_thread_idle(sqd); |
69fb2131 JA |
7535 | mutex_unlock(&sqd->ctx_lock); |
7536 | ||
08369246 | 7537 | if (sqd->thread) |
69fb2131 | 7538 | io_sq_thread_unpark(sqd); |
534ca6d6 JA |
7539 | |
7540 | io_put_sq_data(sqd); | |
7541 | ctx->sq_data = NULL; | |
6c271ce2 JA |
7542 | } |
7543 | } | |
7544 | ||
6b06314c | 7545 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7546 | /* |
7547 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7548 | * the io_uring can be safely unregistered on process exit, even if we have | |
7549 | * loops in the file referencing. | |
7550 | */ | |
7551 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7552 | { | |
7553 | struct sock *sk = ctx->ring_sock->sk; | |
7554 | struct scm_fp_list *fpl; | |
7555 | struct sk_buff *skb; | |
08a45173 | 7556 | int i, nr_files; |
6b06314c | 7557 | |
6b06314c JA |
7558 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7559 | if (!fpl) | |
7560 | return -ENOMEM; | |
7561 | ||
7562 | skb = alloc_skb(0, GFP_KERNEL); | |
7563 | if (!skb) { | |
7564 | kfree(fpl); | |
7565 | return -ENOMEM; | |
7566 | } | |
7567 | ||
7568 | skb->sk = sk; | |
6b06314c | 7569 | |
08a45173 | 7570 | nr_files = 0; |
6b06314c JA |
7571 | fpl->user = get_uid(ctx->user); |
7572 | for (i = 0; i < nr; i++) { | |
65e19f54 JA |
7573 | struct file *file = io_file_from_index(ctx, i + offset); |
7574 | ||
7575 | if (!file) | |
08a45173 | 7576 | continue; |
65e19f54 | 7577 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7578 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7579 | nr_files++; | |
6b06314c JA |
7580 | } |
7581 | ||
08a45173 JA |
7582 | if (nr_files) { |
7583 | fpl->max = SCM_MAX_FD; | |
7584 | fpl->count = nr_files; | |
7585 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7586 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7587 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7588 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7589 | |
08a45173 JA |
7590 | for (i = 0; i < nr_files; i++) |
7591 | fput(fpl->fp[i]); | |
7592 | } else { | |
7593 | kfree_skb(skb); | |
7594 | kfree(fpl); | |
7595 | } | |
6b06314c JA |
7596 | |
7597 | return 0; | |
7598 | } | |
7599 | ||
7600 | /* | |
7601 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7602 | * causes regular reference counting to break down. We rely on the UNIX | |
7603 | * garbage collection to take care of this problem for us. | |
7604 | */ | |
7605 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7606 | { | |
7607 | unsigned left, total; | |
7608 | int ret = 0; | |
7609 | ||
7610 | total = 0; | |
7611 | left = ctx->nr_user_files; | |
7612 | while (left) { | |
7613 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7614 | |
7615 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7616 | if (ret) | |
7617 | break; | |
7618 | left -= this_files; | |
7619 | total += this_files; | |
7620 | } | |
7621 | ||
7622 | if (!ret) | |
7623 | return 0; | |
7624 | ||
7625 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7626 | struct file *file = io_file_from_index(ctx, total); |
7627 | ||
7628 | if (file) | |
7629 | fput(file); | |
6b06314c JA |
7630 | total++; |
7631 | } | |
7632 | ||
7633 | return ret; | |
7634 | } | |
7635 | #else | |
7636 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7637 | { | |
7638 | return 0; | |
7639 | } | |
7640 | #endif | |
7641 | ||
269bbe5f | 7642 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7643 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7644 | { |
7645 | int i; | |
7646 | ||
7647 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7648 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7649 | unsigned this_files; |
7650 | ||
7651 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7652 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7653 | GFP_KERNEL); | |
7654 | if (!table->files) | |
7655 | break; | |
7656 | nr_files -= this_files; | |
7657 | } | |
7658 | ||
7659 | if (i == nr_tables) | |
7660 | return 0; | |
7661 | ||
7662 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7663 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7664 | kfree(table->files); |
7665 | } | |
7666 | return 1; | |
7667 | } | |
7668 | ||
50238531 | 7669 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7670 | { |
50238531 | 7671 | struct file *file = prsrc->file; |
05f3fb3c JA |
7672 | #if defined(CONFIG_UNIX) |
7673 | struct sock *sock = ctx->ring_sock->sk; | |
7674 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7675 | struct sk_buff *skb; | |
7676 | int i; | |
7677 | ||
7678 | __skb_queue_head_init(&list); | |
7679 | ||
7680 | /* | |
7681 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7682 | * remove this entry and rearrange the file array. | |
7683 | */ | |
7684 | skb = skb_dequeue(head); | |
7685 | while (skb) { | |
7686 | struct scm_fp_list *fp; | |
7687 | ||
7688 | fp = UNIXCB(skb).fp; | |
7689 | for (i = 0; i < fp->count; i++) { | |
7690 | int left; | |
7691 | ||
7692 | if (fp->fp[i] != file) | |
7693 | continue; | |
7694 | ||
7695 | unix_notinflight(fp->user, fp->fp[i]); | |
7696 | left = fp->count - 1 - i; | |
7697 | if (left) { | |
7698 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7699 | left * sizeof(struct file *)); | |
7700 | } | |
7701 | fp->count--; | |
7702 | if (!fp->count) { | |
7703 | kfree_skb(skb); | |
7704 | skb = NULL; | |
7705 | } else { | |
7706 | __skb_queue_tail(&list, skb); | |
7707 | } | |
7708 | fput(file); | |
7709 | file = NULL; | |
7710 | break; | |
7711 | } | |
7712 | ||
7713 | if (!file) | |
7714 | break; | |
7715 | ||
7716 | __skb_queue_tail(&list, skb); | |
7717 | ||
7718 | skb = skb_dequeue(head); | |
7719 | } | |
7720 | ||
7721 | if (skb_peek(&list)) { | |
7722 | spin_lock_irq(&head->lock); | |
7723 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7724 | __skb_queue_tail(head, skb); | |
7725 | spin_unlock_irq(&head->lock); | |
7726 | } | |
7727 | #else | |
7728 | fput(file); | |
7729 | #endif | |
7730 | } | |
7731 | ||
269bbe5f | 7732 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7733 | { |
269bbe5f BM |
7734 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7735 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7736 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7737 | |
269bbe5f BM |
7738 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7739 | list_del(&prsrc->list); | |
50238531 | 7740 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7741 | kfree(prsrc); |
65e19f54 | 7742 | } |
05589553 | 7743 | |
05589553 XW |
7744 | percpu_ref_exit(&ref_node->refs); |
7745 | kfree(ref_node); | |
269bbe5f | 7746 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7747 | } |
65e19f54 | 7748 | |
269bbe5f | 7749 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7750 | { |
7751 | struct io_ring_ctx *ctx; | |
7752 | struct llist_node *node; | |
7753 | ||
269bbe5f BM |
7754 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7755 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7756 | |
7757 | while (node) { | |
269bbe5f | 7758 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7759 | struct llist_node *next = node->next; |
7760 | ||
269bbe5f BM |
7761 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7762 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7763 | node = next; |
7764 | } | |
7765 | } | |
7766 | ||
ea64ec02 PB |
7767 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7768 | unsigned i) | |
2faf852d | 7769 | { |
ea64ec02 PB |
7770 | struct fixed_rsrc_table *table; |
7771 | ||
7772 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7773 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7774 | } | |
7775 | ||
00835dce | 7776 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7777 | { |
269bbe5f BM |
7778 | struct fixed_rsrc_ref_node *ref_node; |
7779 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7780 | struct io_ring_ctx *ctx; |
e297822b | 7781 | bool first_add = false; |
4a38aed2 | 7782 | int delay = HZ; |
65e19f54 | 7783 | |
269bbe5f BM |
7784 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7785 | data = ref_node->rsrc_data; | |
e297822b PB |
7786 | ctx = data->ctx; |
7787 | ||
2a63b2d9 | 7788 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7789 | ref_node->done = true; |
7790 | ||
d67d2263 BM |
7791 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7792 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7793 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7794 | /* recycle ref nodes in order */ |
7795 | if (!ref_node->done) | |
7796 | break; | |
7797 | list_del(&ref_node->node); | |
269bbe5f | 7798 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7799 | } |
2a63b2d9 | 7800 | io_rsrc_ref_unlock(ctx); |
05589553 | 7801 | |
e297822b | 7802 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7803 | delay = 0; |
05589553 | 7804 | |
4a38aed2 | 7805 | if (!delay) |
269bbe5f | 7806 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7807 | else if (first_add) |
269bbe5f | 7808 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7809 | } |
65e19f54 | 7810 | |
6802535d | 7811 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7812 | struct io_ring_ctx *ctx) |
05f3fb3c | 7813 | { |
269bbe5f | 7814 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7815 | |
05589553 XW |
7816 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7817 | if (!ref_node) | |
3e2224c5 | 7818 | return NULL; |
05f3fb3c | 7819 | |
00835dce | 7820 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7821 | 0, GFP_KERNEL)) { |
7822 | kfree(ref_node); | |
3e2224c5 | 7823 | return NULL; |
05589553 XW |
7824 | } |
7825 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7826 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7827 | ref_node->done = false; |
05589553 | 7828 | return ref_node; |
05589553 XW |
7829 | } |
7830 | ||
bc9744cd PB |
7831 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7832 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7833 | { |
269bbe5f | 7834 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7835 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7836 | } |
7837 | ||
269bbe5f | 7838 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7839 | { |
7840 | percpu_ref_exit(&ref_node->refs); | |
7841 | kfree(ref_node); | |
65e19f54 JA |
7842 | } |
7843 | ||
ea64ec02 | 7844 | |
6b06314c JA |
7845 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7846 | unsigned nr_args) | |
7847 | { | |
7848 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7849 | unsigned nr_tables, i; |
05f3fb3c | 7850 | struct file *file; |
600cf3f8 | 7851 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7852 | struct fixed_rsrc_ref_node *ref_node; |
7853 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7854 | |
05f3fb3c | 7855 | if (ctx->file_data) |
6b06314c JA |
7856 | return -EBUSY; |
7857 | if (!nr_args) | |
7858 | return -EINVAL; | |
7859 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7860 | return -EMFILE; | |
7861 | ||
1ad555c6 | 7862 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7863 | if (!file_data) |
05f3fb3c | 7864 | return -ENOMEM; |
13770a71 | 7865 | ctx->file_data = file_data; |
05f3fb3c | 7866 | |
65e19f54 | 7867 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7868 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7869 | GFP_KERNEL); |
600cf3f8 PB |
7870 | if (!file_data->table) |
7871 | goto out_free; | |
05f3fb3c | 7872 | |
600cf3f8 | 7873 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7874 | goto out_free; |
65e19f54 | 7875 | |
08a45173 | 7876 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7877 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7878 | ret = -EFAULT; | |
7879 | goto out_fput; | |
7880 | } | |
08a45173 | 7881 | /* allow sparse sets */ |
600cf3f8 | 7882 | if (fd == -1) |
08a45173 | 7883 | continue; |
6b06314c | 7884 | |
05f3fb3c | 7885 | file = fget(fd); |
6b06314c | 7886 | ret = -EBADF; |
05f3fb3c | 7887 | if (!file) |
600cf3f8 | 7888 | goto out_fput; |
05f3fb3c | 7889 | |
6b06314c JA |
7890 | /* |
7891 | * Don't allow io_uring instances to be registered. If UNIX | |
7892 | * isn't enabled, then this causes a reference cycle and this | |
7893 | * instance can never get freed. If UNIX is enabled we'll | |
7894 | * handle it just fine, but there's still no point in allowing | |
7895 | * a ring fd as it doesn't support regular read/write anyway. | |
7896 | */ | |
05f3fb3c JA |
7897 | if (file->f_op == &io_uring_fops) { |
7898 | fput(file); | |
600cf3f8 | 7899 | goto out_fput; |
6b06314c | 7900 | } |
ea64ec02 | 7901 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7902 | } |
7903 | ||
6b06314c | 7904 | ret = io_sqe_files_scm(ctx); |
05589553 | 7905 | if (ret) { |
6b06314c | 7906 | io_sqe_files_unregister(ctx); |
05589553 XW |
7907 | return ret; |
7908 | } | |
6b06314c | 7909 | |
bc9744cd | 7910 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7911 | if (!ref_node) { |
05589553 | 7912 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7913 | return -ENOMEM; |
05589553 | 7914 | } |
bc9744cd | 7915 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7916 | |
d67d2263 | 7917 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7918 | return ret; |
600cf3f8 PB |
7919 | out_fput: |
7920 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7921 | file = io_file_from_index(ctx, i); | |
7922 | if (file) | |
7923 | fput(file); | |
7924 | } | |
7925 | for (i = 0; i < nr_tables; i++) | |
7926 | kfree(file_data->table[i].files); | |
7927 | ctx->nr_user_files = 0; | |
600cf3f8 | 7928 | out_free: |
1ad555c6 | 7929 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7930 | ctx->file_data = NULL; |
6b06314c JA |
7931 | return ret; |
7932 | } | |
7933 | ||
c3a31e60 JA |
7934 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7935 | int index) | |
7936 | { | |
7937 | #if defined(CONFIG_UNIX) | |
7938 | struct sock *sock = ctx->ring_sock->sk; | |
7939 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7940 | struct sk_buff *skb; | |
7941 | ||
7942 | /* | |
7943 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7944 | * file set. If there's no room, fall back to allocating a new skb | |
7945 | * and filling it in. | |
7946 | */ | |
7947 | spin_lock_irq(&head->lock); | |
7948 | skb = skb_peek(head); | |
7949 | if (skb) { | |
7950 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7951 | ||
7952 | if (fpl->count < SCM_MAX_FD) { | |
7953 | __skb_unlink(skb, head); | |
7954 | spin_unlock_irq(&head->lock); | |
7955 | fpl->fp[fpl->count] = get_file(file); | |
7956 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7957 | fpl->count++; | |
7958 | spin_lock_irq(&head->lock); | |
7959 | __skb_queue_head(head, skb); | |
7960 | } else { | |
7961 | skb = NULL; | |
7962 | } | |
7963 | } | |
7964 | spin_unlock_irq(&head->lock); | |
7965 | ||
7966 | if (skb) { | |
7967 | fput(file); | |
7968 | return 0; | |
7969 | } | |
7970 | ||
7971 | return __io_sqe_files_scm(ctx, 1, index); | |
7972 | #else | |
7973 | return 0; | |
7974 | #endif | |
7975 | } | |
7976 | ||
50238531 | 7977 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7978 | { |
269bbe5f BM |
7979 | struct io_rsrc_put *prsrc; |
7980 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7981 | |
269bbe5f BM |
7982 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7983 | if (!prsrc) | |
a5318d3c | 7984 | return -ENOMEM; |
05f3fb3c | 7985 | |
50238531 | 7986 | prsrc->rsrc = rsrc; |
269bbe5f | 7987 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7988 | |
a5318d3c | 7989 | return 0; |
05f3fb3c JA |
7990 | } |
7991 | ||
269bbe5f BM |
7992 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7993 | struct file *file) | |
7994 | { | |
50238531 | 7995 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7996 | } |
7997 | ||
05f3fb3c | 7998 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7999 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
8000 | unsigned nr_args) |
8001 | { | |
269bbe5f BM |
8002 | struct fixed_rsrc_data *data = ctx->file_data; |
8003 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 8004 | struct file *file, **file_slot; |
c3a31e60 JA |
8005 | __s32 __user *fds; |
8006 | int fd, i, err; | |
8007 | __u32 done; | |
05589553 | 8008 | bool needs_switch = false; |
c3a31e60 | 8009 | |
05f3fb3c | 8010 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
8011 | return -EOVERFLOW; |
8012 | if (done > ctx->nr_user_files) | |
8013 | return -EINVAL; | |
8014 | ||
bc9744cd | 8015 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
8016 | if (!ref_node) |
8017 | return -ENOMEM; | |
bc9744cd | 8018 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 8019 | |
269bbe5f | 8020 | fds = u64_to_user_ptr(up->data); |
67973b93 | 8021 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
8022 | err = 0; |
8023 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
8024 | err = -EFAULT; | |
8025 | break; | |
8026 | } | |
4e0377a1 | 8027 | if (fd == IORING_REGISTER_FILES_SKIP) |
8028 | continue; | |
8029 | ||
67973b93 | 8030 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
8031 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
8032 | ||
8033 | if (*file_slot) { | |
8034 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
8035 | if (err) |
8036 | break; | |
ea64ec02 | 8037 | *file_slot = NULL; |
05589553 | 8038 | needs_switch = true; |
c3a31e60 JA |
8039 | } |
8040 | if (fd != -1) { | |
c3a31e60 JA |
8041 | file = fget(fd); |
8042 | if (!file) { | |
8043 | err = -EBADF; | |
8044 | break; | |
8045 | } | |
8046 | /* | |
8047 | * Don't allow io_uring instances to be registered. If | |
8048 | * UNIX isn't enabled, then this causes a reference | |
8049 | * cycle and this instance can never get freed. If UNIX | |
8050 | * is enabled we'll handle it just fine, but there's | |
8051 | * still no point in allowing a ring fd as it doesn't | |
8052 | * support regular read/write anyway. | |
8053 | */ | |
8054 | if (file->f_op == &io_uring_fops) { | |
8055 | fput(file); | |
8056 | err = -EBADF; | |
8057 | break; | |
8058 | } | |
e68a3ff8 | 8059 | *file_slot = file; |
c3a31e60 | 8060 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 8061 | if (err) { |
e68a3ff8 | 8062 | *file_slot = NULL; |
f3bd9dae | 8063 | fput(file); |
c3a31e60 | 8064 | break; |
f3bd9dae | 8065 | } |
c3a31e60 | 8066 | } |
05f3fb3c JA |
8067 | } |
8068 | ||
05589553 | 8069 | if (needs_switch) { |
b2e96852 | 8070 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 8071 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 8072 | } else |
269bbe5f | 8073 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
8074 | |
8075 | return done ? done : err; | |
8076 | } | |
05589553 | 8077 | |
05f3fb3c JA |
8078 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
8079 | unsigned nr_args) | |
8080 | { | |
269bbe5f | 8081 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
8082 | |
8083 | if (!ctx->file_data) | |
8084 | return -ENXIO; | |
8085 | if (!nr_args) | |
8086 | return -EINVAL; | |
8087 | if (copy_from_user(&up, arg, sizeof(up))) | |
8088 | return -EFAULT; | |
8089 | if (up.resv) | |
8090 | return -EINVAL; | |
8091 | ||
8092 | return __io_sqe_files_update(ctx, &up, nr_args); | |
8093 | } | |
c3a31e60 | 8094 | |
5280f7e5 | 8095 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
8096 | { |
8097 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8098 | ||
5280f7e5 PB |
8099 | req = io_put_req_find_next(req); |
8100 | return req ? &req->work : NULL; | |
7d723065 JA |
8101 | } |
8102 | ||
5aa75ed5 | 8103 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e PB |
8104 | { |
8105 | struct io_wq_data data; | |
24369c2e | 8106 | unsigned int concurrency; |
24369c2e PB |
8107 | |
8108 | data.user = ctx->user; | |
e9fd9396 | 8109 | data.free_work = io_free_work; |
f5fa38c5 | 8110 | data.do_work = io_wq_submit_work; |
24369c2e | 8111 | |
d25e3a3d JA |
8112 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
8113 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 8114 | |
5aa75ed5 | 8115 | return io_wq_create(concurrency, &data); |
24369c2e PB |
8116 | } |
8117 | ||
5aa75ed5 JA |
8118 | static int io_uring_alloc_task_context(struct task_struct *task, |
8119 | struct io_ring_ctx *ctx) | |
0f212204 JA |
8120 | { |
8121 | struct io_uring_task *tctx; | |
d8a6df10 | 8122 | int ret; |
0f212204 JA |
8123 | |
8124 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
8125 | if (unlikely(!tctx)) | |
8126 | return -ENOMEM; | |
8127 | ||
d8a6df10 JA |
8128 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
8129 | if (unlikely(ret)) { | |
8130 | kfree(tctx); | |
8131 | return ret; | |
8132 | } | |
8133 | ||
5aa75ed5 JA |
8134 | tctx->io_wq = io_init_wq_offload(ctx); |
8135 | if (IS_ERR(tctx->io_wq)) { | |
8136 | ret = PTR_ERR(tctx->io_wq); | |
8137 | percpu_counter_destroy(&tctx->inflight); | |
8138 | kfree(tctx); | |
8139 | return ret; | |
8140 | } | |
8141 | ||
0f212204 JA |
8142 | xa_init(&tctx->xa); |
8143 | init_waitqueue_head(&tctx->wait); | |
8144 | tctx->last = NULL; | |
fdaf083c JA |
8145 | atomic_set(&tctx->in_idle, 0); |
8146 | tctx->sqpoll = false; | |
500a373d JA |
8147 | io_init_identity(&tctx->__identity); |
8148 | tctx->identity = &tctx->__identity; | |
0f212204 | 8149 | task->io_uring = tctx; |
7cbf1722 JA |
8150 | spin_lock_init(&tctx->task_lock); |
8151 | INIT_WQ_LIST(&tctx->task_list); | |
8152 | tctx->task_state = 0; | |
8153 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
8154 | return 0; |
8155 | } | |
8156 | ||
8157 | void __io_uring_free(struct task_struct *tsk) | |
8158 | { | |
8159 | struct io_uring_task *tctx = tsk->io_uring; | |
8160 | ||
8161 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
500a373d JA |
8162 | WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1); |
8163 | if (tctx->identity != &tctx->__identity) | |
8164 | kfree(tctx->identity); | |
d8a6df10 | 8165 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
8166 | kfree(tctx); |
8167 | tsk->io_uring = NULL; | |
8168 | } | |
8169 | ||
7e84e1c7 SG |
8170 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
8171 | struct io_uring_params *p) | |
2b188cc1 JA |
8172 | { |
8173 | int ret; | |
8174 | ||
d25e3a3d JA |
8175 | /* Retain compatibility with failing for an invalid attach attempt */ |
8176 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
8177 | IORING_SETUP_ATTACH_WQ) { | |
8178 | struct fd f; | |
8179 | ||
8180 | f = fdget(p->wq_fd); | |
8181 | if (!f.file) | |
8182 | return -ENXIO; | |
8183 | if (f.file->f_op != &io_uring_fops) { | |
8184 | fdput(f); | |
8185 | return -EINVAL; | |
8186 | } | |
8187 | fdput(f); | |
8188 | } | |
6c271ce2 | 8189 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
534ca6d6 JA |
8190 | struct io_sq_data *sqd; |
8191 | ||
3ec482d1 | 8192 | ret = -EPERM; |
ce59fc69 | 8193 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
8194 | goto err; |
8195 | ||
534ca6d6 JA |
8196 | sqd = io_get_sq_data(p); |
8197 | if (IS_ERR(sqd)) { | |
8198 | ret = PTR_ERR(sqd); | |
8199 | goto err; | |
8200 | } | |
69fb2131 | 8201 | |
534ca6d6 | 8202 | ctx->sq_data = sqd; |
69fb2131 JA |
8203 | io_sq_thread_park(sqd); |
8204 | mutex_lock(&sqd->ctx_lock); | |
8205 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
8206 | mutex_unlock(&sqd->ctx_lock); | |
8207 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 8208 | |
917257da JA |
8209 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8210 | if (!ctx->sq_thread_idle) | |
8211 | ctx->sq_thread_idle = HZ; | |
8212 | ||
aa06165d | 8213 | if (sqd->thread) |
5aa75ed5 | 8214 | return 0; |
aa06165d | 8215 | |
6c271ce2 | 8216 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8217 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8218 | |
917257da | 8219 | ret = -EINVAL; |
44a9bd18 JA |
8220 | if (cpu >= nr_cpu_ids) |
8221 | goto err; | |
7889f44d | 8222 | if (!cpu_online(cpu)) |
917257da JA |
8223 | goto err; |
8224 | ||
69fb2131 | 8225 | sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd, |
534ca6d6 | 8226 | cpu, "io_uring-sq"); |
6c271ce2 | 8227 | } else { |
69fb2131 | 8228 | sqd->thread = kthread_create(io_sq_thread, sqd, |
6c271ce2 JA |
8229 | "io_uring-sq"); |
8230 | } | |
534ca6d6 JA |
8231 | if (IS_ERR(sqd->thread)) { |
8232 | ret = PTR_ERR(sqd->thread); | |
8233 | sqd->thread = NULL; | |
6c271ce2 JA |
8234 | goto err; |
8235 | } | |
5aa75ed5 | 8236 | ret = io_uring_alloc_task_context(sqd->thread, ctx); |
0f212204 JA |
8237 | if (ret) |
8238 | goto err; | |
6c271ce2 JA |
8239 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8240 | /* Can't have SQ_AFF without SQPOLL */ | |
8241 | ret = -EINVAL; | |
8242 | goto err; | |
8243 | } | |
8244 | ||
2b188cc1 JA |
8245 | return 0; |
8246 | err: | |
5aa75ed5 | 8247 | io_sq_thread_stop(ctx); |
2b188cc1 JA |
8248 | return ret; |
8249 | } | |
8250 | ||
7e84e1c7 SG |
8251 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
8252 | { | |
534ca6d6 JA |
8253 | struct io_sq_data *sqd = ctx->sq_data; |
8254 | ||
8255 | if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread) | |
8256 | wake_up_process(sqd->thread); | |
7e84e1c7 SG |
8257 | } |
8258 | ||
a087e2b5 BM |
8259 | static inline void __io_unaccount_mem(struct user_struct *user, |
8260 | unsigned long nr_pages) | |
2b188cc1 JA |
8261 | { |
8262 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8263 | } | |
8264 | ||
a087e2b5 BM |
8265 | static inline int __io_account_mem(struct user_struct *user, |
8266 | unsigned long nr_pages) | |
2b188cc1 JA |
8267 | { |
8268 | unsigned long page_limit, cur_pages, new_pages; | |
8269 | ||
8270 | /* Don't allow more pages than we can safely lock */ | |
8271 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8272 | ||
8273 | do { | |
8274 | cur_pages = atomic_long_read(&user->locked_vm); | |
8275 | new_pages = cur_pages + nr_pages; | |
8276 | if (new_pages > page_limit) | |
8277 | return -ENOMEM; | |
8278 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8279 | new_pages) != cur_pages); | |
8280 | ||
8281 | return 0; | |
8282 | } | |
8283 | ||
26bfa89e | 8284 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8285 | { |
aad5d8da | 8286 | if (ctx->limit_mem) |
a087e2b5 | 8287 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8288 | |
26bfa89e JA |
8289 | if (ctx->mm_account) |
8290 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8291 | } |
8292 | ||
26bfa89e | 8293 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8294 | { |
30975825 BM |
8295 | int ret; |
8296 | ||
8297 | if (ctx->limit_mem) { | |
8298 | ret = __io_account_mem(ctx->user, nr_pages); | |
8299 | if (ret) | |
8300 | return ret; | |
8301 | } | |
8302 | ||
26bfa89e JA |
8303 | if (ctx->mm_account) |
8304 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8305 | |
8306 | return 0; | |
8307 | } | |
8308 | ||
2b188cc1 JA |
8309 | static void io_mem_free(void *ptr) |
8310 | { | |
52e04ef4 MR |
8311 | struct page *page; |
8312 | ||
8313 | if (!ptr) | |
8314 | return; | |
2b188cc1 | 8315 | |
52e04ef4 | 8316 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8317 | if (put_page_testzero(page)) |
8318 | free_compound_page(page); | |
8319 | } | |
8320 | ||
8321 | static void *io_mem_alloc(size_t size) | |
8322 | { | |
8323 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8324 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8325 | |
8326 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8327 | } | |
8328 | ||
75b28aff HV |
8329 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8330 | size_t *sq_offset) | |
8331 | { | |
8332 | struct io_rings *rings; | |
8333 | size_t off, sq_array_size; | |
8334 | ||
8335 | off = struct_size(rings, cqes, cq_entries); | |
8336 | if (off == SIZE_MAX) | |
8337 | return SIZE_MAX; | |
8338 | ||
8339 | #ifdef CONFIG_SMP | |
8340 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8341 | if (off == 0) | |
8342 | return SIZE_MAX; | |
8343 | #endif | |
8344 | ||
b36200f5 DV |
8345 | if (sq_offset) |
8346 | *sq_offset = off; | |
8347 | ||
75b28aff HV |
8348 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8349 | if (sq_array_size == SIZE_MAX) | |
8350 | return SIZE_MAX; | |
8351 | ||
8352 | if (check_add_overflow(off, sq_array_size, &off)) | |
8353 | return SIZE_MAX; | |
8354 | ||
75b28aff HV |
8355 | return off; |
8356 | } | |
8357 | ||
0a96bbe4 | 8358 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8359 | { |
8360 | int i, j; | |
8361 | ||
8362 | if (!ctx->user_bufs) | |
8363 | return -ENXIO; | |
8364 | ||
8365 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8366 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8367 | ||
8368 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8369 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8370 | |
de293938 | 8371 | if (imu->acct_pages) |
26bfa89e | 8372 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8373 | kvfree(imu->bvec); |
edafccee JA |
8374 | imu->nr_bvecs = 0; |
8375 | } | |
8376 | ||
8377 | kfree(ctx->user_bufs); | |
8378 | ctx->user_bufs = NULL; | |
8379 | ctx->nr_user_bufs = 0; | |
8380 | return 0; | |
8381 | } | |
8382 | ||
8383 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8384 | void __user *arg, unsigned index) | |
8385 | { | |
8386 | struct iovec __user *src; | |
8387 | ||
8388 | #ifdef CONFIG_COMPAT | |
8389 | if (ctx->compat) { | |
8390 | struct compat_iovec __user *ciovs; | |
8391 | struct compat_iovec ciov; | |
8392 | ||
8393 | ciovs = (struct compat_iovec __user *) arg; | |
8394 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8395 | return -EFAULT; | |
8396 | ||
d55e5f5b | 8397 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8398 | dst->iov_len = ciov.iov_len; |
8399 | return 0; | |
8400 | } | |
8401 | #endif | |
8402 | src = (struct iovec __user *) arg; | |
8403 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8404 | return -EFAULT; | |
8405 | return 0; | |
8406 | } | |
8407 | ||
de293938 JA |
8408 | /* |
8409 | * Not super efficient, but this is just a registration time. And we do cache | |
8410 | * the last compound head, so generally we'll only do a full search if we don't | |
8411 | * match that one. | |
8412 | * | |
8413 | * We check if the given compound head page has already been accounted, to | |
8414 | * avoid double accounting it. This allows us to account the full size of the | |
8415 | * page, not just the constituent pages of a huge page. | |
8416 | */ | |
8417 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8418 | int nr_pages, struct page *hpage) | |
8419 | { | |
8420 | int i, j; | |
8421 | ||
8422 | /* check current page array */ | |
8423 | for (i = 0; i < nr_pages; i++) { | |
8424 | if (!PageCompound(pages[i])) | |
8425 | continue; | |
8426 | if (compound_head(pages[i]) == hpage) | |
8427 | return true; | |
8428 | } | |
8429 | ||
8430 | /* check previously registered pages */ | |
8431 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8432 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8433 | ||
8434 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8435 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8436 | continue; | |
8437 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8438 | return true; | |
8439 | } | |
8440 | } | |
8441 | ||
8442 | return false; | |
8443 | } | |
8444 | ||
8445 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8446 | int nr_pages, struct io_mapped_ubuf *imu, | |
8447 | struct page **last_hpage) | |
8448 | { | |
8449 | int i, ret; | |
8450 | ||
8451 | for (i = 0; i < nr_pages; i++) { | |
8452 | if (!PageCompound(pages[i])) { | |
8453 | imu->acct_pages++; | |
8454 | } else { | |
8455 | struct page *hpage; | |
8456 | ||
8457 | hpage = compound_head(pages[i]); | |
8458 | if (hpage == *last_hpage) | |
8459 | continue; | |
8460 | *last_hpage = hpage; | |
8461 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8462 | continue; | |
8463 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8464 | } | |
8465 | } | |
8466 | ||
8467 | if (!imu->acct_pages) | |
8468 | return 0; | |
8469 | ||
26bfa89e | 8470 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8471 | if (ret) |
8472 | imu->acct_pages = 0; | |
8473 | return ret; | |
8474 | } | |
8475 | ||
0a96bbe4 BM |
8476 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8477 | struct io_mapped_ubuf *imu, | |
8478 | struct page **last_hpage) | |
edafccee JA |
8479 | { |
8480 | struct vm_area_struct **vmas = NULL; | |
8481 | struct page **pages = NULL; | |
0a96bbe4 BM |
8482 | unsigned long off, start, end, ubuf; |
8483 | size_t size; | |
8484 | int ret, pret, nr_pages, i; | |
8485 | ||
8486 | ubuf = (unsigned long) iov->iov_base; | |
8487 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8488 | start = ubuf >> PAGE_SHIFT; | |
8489 | nr_pages = end - start; | |
8490 | ||
8491 | ret = -ENOMEM; | |
8492 | ||
8493 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8494 | if (!pages) | |
8495 | goto done; | |
8496 | ||
8497 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8498 | GFP_KERNEL); | |
8499 | if (!vmas) | |
8500 | goto done; | |
edafccee | 8501 | |
0a96bbe4 BM |
8502 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8503 | GFP_KERNEL); | |
8504 | if (!imu->bvec) | |
8505 | goto done; | |
8506 | ||
8507 | ret = 0; | |
8508 | mmap_read_lock(current->mm); | |
8509 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8510 | pages, vmas); | |
8511 | if (pret == nr_pages) { | |
8512 | /* don't support file backed memory */ | |
8513 | for (i = 0; i < nr_pages; i++) { | |
8514 | struct vm_area_struct *vma = vmas[i]; | |
8515 | ||
8516 | if (vma->vm_file && | |
8517 | !is_file_hugepages(vma->vm_file)) { | |
8518 | ret = -EOPNOTSUPP; | |
8519 | break; | |
8520 | } | |
8521 | } | |
8522 | } else { | |
8523 | ret = pret < 0 ? pret : -EFAULT; | |
8524 | } | |
8525 | mmap_read_unlock(current->mm); | |
8526 | if (ret) { | |
8527 | /* | |
8528 | * if we did partial map, or found file backed vmas, | |
8529 | * release any pages we did get | |
8530 | */ | |
8531 | if (pret > 0) | |
8532 | unpin_user_pages(pages, pret); | |
8533 | kvfree(imu->bvec); | |
8534 | goto done; | |
8535 | } | |
8536 | ||
8537 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8538 | if (ret) { | |
8539 | unpin_user_pages(pages, pret); | |
8540 | kvfree(imu->bvec); | |
8541 | goto done; | |
8542 | } | |
8543 | ||
8544 | off = ubuf & ~PAGE_MASK; | |
8545 | size = iov->iov_len; | |
8546 | for (i = 0; i < nr_pages; i++) { | |
8547 | size_t vec_len; | |
8548 | ||
8549 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8550 | imu->bvec[i].bv_page = pages[i]; | |
8551 | imu->bvec[i].bv_len = vec_len; | |
8552 | imu->bvec[i].bv_offset = off; | |
8553 | off = 0; | |
8554 | size -= vec_len; | |
8555 | } | |
8556 | /* store original address for later verification */ | |
8557 | imu->ubuf = ubuf; | |
8558 | imu->len = iov->iov_len; | |
8559 | imu->nr_bvecs = nr_pages; | |
8560 | ret = 0; | |
8561 | done: | |
8562 | kvfree(pages); | |
8563 | kvfree(vmas); | |
8564 | return ret; | |
8565 | } | |
8566 | ||
2b358604 | 8567 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8568 | { |
edafccee JA |
8569 | if (ctx->user_bufs) |
8570 | return -EBUSY; | |
8571 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8572 | return -EINVAL; | |
8573 | ||
8574 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8575 | GFP_KERNEL); | |
8576 | if (!ctx->user_bufs) | |
8577 | return -ENOMEM; | |
8578 | ||
2b358604 BM |
8579 | return 0; |
8580 | } | |
edafccee | 8581 | |
2b358604 BM |
8582 | static int io_buffer_validate(struct iovec *iov) |
8583 | { | |
8584 | /* | |
8585 | * Don't impose further limits on the size and buffer | |
8586 | * constraints here, we'll -EINVAL later when IO is | |
8587 | * submitted if they are wrong. | |
8588 | */ | |
8589 | if (!iov->iov_base || !iov->iov_len) | |
8590 | return -EFAULT; | |
edafccee | 8591 | |
2b358604 BM |
8592 | /* arbitrary limit, but we need something */ |
8593 | if (iov->iov_len > SZ_1G) | |
8594 | return -EFAULT; | |
edafccee | 8595 | |
2b358604 BM |
8596 | return 0; |
8597 | } | |
edafccee | 8598 | |
2b358604 BM |
8599 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8600 | unsigned int nr_args) | |
8601 | { | |
8602 | int i, ret; | |
8603 | struct iovec iov; | |
8604 | struct page *last_hpage = NULL; | |
edafccee | 8605 | |
2b358604 BM |
8606 | ret = io_buffers_map_alloc(ctx, nr_args); |
8607 | if (ret) | |
8608 | return ret; | |
edafccee | 8609 | |
edafccee JA |
8610 | for (i = 0; i < nr_args; i++) { |
8611 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8612 | |
edafccee JA |
8613 | ret = io_copy_iov(ctx, &iov, arg, i); |
8614 | if (ret) | |
0a96bbe4 | 8615 | break; |
de293938 | 8616 | |
2b358604 BM |
8617 | ret = io_buffer_validate(&iov); |
8618 | if (ret) | |
0a96bbe4 | 8619 | break; |
edafccee | 8620 | |
0a96bbe4 BM |
8621 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8622 | if (ret) | |
8623 | break; | |
edafccee JA |
8624 | |
8625 | ctx->nr_user_bufs++; | |
8626 | } | |
0a96bbe4 BM |
8627 | |
8628 | if (ret) | |
8629 | io_sqe_buffers_unregister(ctx); | |
8630 | ||
edafccee JA |
8631 | return ret; |
8632 | } | |
8633 | ||
9b402849 JA |
8634 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8635 | { | |
8636 | __s32 __user *fds = arg; | |
8637 | int fd; | |
8638 | ||
8639 | if (ctx->cq_ev_fd) | |
8640 | return -EBUSY; | |
8641 | ||
8642 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8643 | return -EFAULT; | |
8644 | ||
8645 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8646 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8647 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8648 | ctx->cq_ev_fd = NULL; | |
8649 | return ret; | |
8650 | } | |
8651 | ||
8652 | return 0; | |
8653 | } | |
8654 | ||
8655 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8656 | { | |
8657 | if (ctx->cq_ev_fd) { | |
8658 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8659 | ctx->cq_ev_fd = NULL; | |
8660 | return 0; | |
8661 | } | |
8662 | ||
8663 | return -ENXIO; | |
8664 | } | |
8665 | ||
5a2e745d JA |
8666 | static int __io_destroy_buffers(int id, void *p, void *data) |
8667 | { | |
8668 | struct io_ring_ctx *ctx = data; | |
8669 | struct io_buffer *buf = p; | |
8670 | ||
067524e9 | 8671 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8672 | return 0; |
8673 | } | |
8674 | ||
8675 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8676 | { | |
8677 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8678 | idr_destroy(&ctx->io_buffer_idr); | |
8679 | } | |
8680 | ||
68e68ee6 | 8681 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8682 | { |
68e68ee6 | 8683 | struct io_kiocb *req, *nxt; |
1b4c351f | 8684 | |
68e68ee6 JA |
8685 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8686 | if (tsk && req->task != tsk) | |
8687 | continue; | |
1b4c351f JA |
8688 | list_del(&req->compl.list); |
8689 | kmem_cache_free(req_cachep, req); | |
8690 | } | |
8691 | } | |
8692 | ||
9a4fdbd8 | 8693 | static void io_req_caches_free(struct io_ring_ctx *ctx, struct task_struct *tsk) |
2b188cc1 | 8694 | { |
bf019da7 PB |
8695 | struct io_submit_state *submit_state = &ctx->submit_state; |
8696 | ||
9a4fdbd8 JA |
8697 | mutex_lock(&ctx->uring_lock); |
8698 | ||
8699 | if (submit_state->free_reqs) | |
8700 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, | |
8701 | submit_state->reqs); | |
8702 | ||
8703 | io_req_cache_free(&submit_state->comp.free_list, NULL); | |
8704 | ||
8705 | spin_lock_irq(&ctx->completion_lock); | |
8706 | io_req_cache_free(&submit_state->comp.locked_free_list, NULL); | |
8707 | spin_unlock_irq(&ctx->completion_lock); | |
8708 | ||
8709 | mutex_unlock(&ctx->uring_lock); | |
8710 | } | |
8711 | ||
2b188cc1 JA |
8712 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8713 | { | |
04fc6c80 PB |
8714 | /* |
8715 | * Some may use context even when all refs and requests have been put, | |
8716 | * and they are free to do so while still holding uring_lock, see | |
8717 | * __io_req_task_submit(). Wait for them to finish. | |
8718 | */ | |
8719 | mutex_lock(&ctx->uring_lock); | |
8720 | mutex_unlock(&ctx->uring_lock); | |
8721 | ||
5aa75ed5 | 8722 | io_sq_thread_stop(ctx); |
0a96bbe4 | 8723 | io_sqe_buffers_unregister(ctx); |
2aede0e4 JA |
8724 | |
8725 | if (ctx->sqo_task) { | |
8726 | put_task_struct(ctx->sqo_task); | |
8727 | ctx->sqo_task = NULL; | |
8728 | mmdrop(ctx->mm_account); | |
8729 | ctx->mm_account = NULL; | |
30975825 | 8730 | } |
def596e9 | 8731 | |
91d8f519 DZ |
8732 | #ifdef CONFIG_BLK_CGROUP |
8733 | if (ctx->sqo_blkcg_css) | |
8734 | css_put(ctx->sqo_blkcg_css); | |
8735 | #endif | |
8736 | ||
8bad28d8 | 8737 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8738 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8739 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8740 | io_eventfd_unregister(ctx); |
5a2e745d | 8741 | io_destroy_buffers(ctx); |
41726c9a | 8742 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8743 | |
2b188cc1 | 8744 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8745 | if (ctx->ring_sock) { |
8746 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8747 | sock_release(ctx->ring_sock); |
355e8d26 | 8748 | } |
2b188cc1 JA |
8749 | #endif |
8750 | ||
75b28aff | 8751 | io_mem_free(ctx->rings); |
2b188cc1 | 8752 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8753 | |
8754 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8755 | free_uid(ctx->user); |
181e448d | 8756 | put_cred(ctx->creds); |
9a4fdbd8 | 8757 | io_req_caches_free(ctx, NULL); |
78076bb6 | 8758 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8759 | kfree(ctx); |
8760 | } | |
8761 | ||
8762 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8763 | { | |
8764 | struct io_ring_ctx *ctx = file->private_data; | |
8765 | __poll_t mask = 0; | |
8766 | ||
8767 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8768 | /* |
8769 | * synchronizes with barrier from wq_has_sleeper call in | |
8770 | * io_commit_cqring | |
8771 | */ | |
2b188cc1 | 8772 | smp_rmb(); |
90554200 | 8773 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8774 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8775 | |
8776 | /* | |
8777 | * Don't flush cqring overflow list here, just do a simple check. | |
8778 | * Otherwise there could possible be ABBA deadlock: | |
8779 | * CPU0 CPU1 | |
8780 | * ---- ---- | |
8781 | * lock(&ctx->uring_lock); | |
8782 | * lock(&ep->mtx); | |
8783 | * lock(&ctx->uring_lock); | |
8784 | * lock(&ep->mtx); | |
8785 | * | |
8786 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8787 | * pushs them to do the flush. | |
8788 | */ | |
8789 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8790 | mask |= EPOLLIN | EPOLLRDNORM; |
8791 | ||
8792 | return mask; | |
8793 | } | |
8794 | ||
8795 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8796 | { | |
8797 | struct io_ring_ctx *ctx = file->private_data; | |
8798 | ||
8799 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8800 | } | |
8801 | ||
0bead8cd | 8802 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8803 | { |
1e6fa521 | 8804 | struct io_identity *iod; |
071698e1 | 8805 | |
1e6fa521 JA |
8806 | iod = idr_remove(&ctx->personality_idr, id); |
8807 | if (iod) { | |
8808 | put_cred(iod->creds); | |
8809 | if (refcount_dec_and_test(&iod->count)) | |
8810 | kfree(iod); | |
0bead8cd | 8811 | return 0; |
1e6fa521 | 8812 | } |
0bead8cd YD |
8813 | |
8814 | return -EINVAL; | |
8815 | } | |
8816 | ||
8817 | static int io_remove_personalities(int id, void *p, void *data) | |
8818 | { | |
8819 | struct io_ring_ctx *ctx = data; | |
8820 | ||
8821 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8822 | return 0; |
8823 | } | |
8824 | ||
7c25c0d1 JA |
8825 | static void io_run_ctx_fallback(struct io_ring_ctx *ctx) |
8826 | { | |
8827 | struct callback_head *work, *head, *next; | |
8828 | ||
8829 | do { | |
8830 | do { | |
8831 | head = NULL; | |
8832 | work = READ_ONCE(ctx->exit_task_work); | |
8833 | } while (cmpxchg(&ctx->exit_task_work, work, head) != work); | |
8834 | ||
8835 | if (!work) | |
8836 | break; | |
8837 | ||
8838 | do { | |
8839 | next = work->next; | |
8840 | work->func(work); | |
8841 | work = next; | |
8842 | cond_resched(); | |
8843 | } while (work); | |
8844 | } while (1); | |
8845 | } | |
8846 | ||
85faa7b8 JA |
8847 | static void io_ring_exit_work(struct work_struct *work) |
8848 | { | |
b2edc0a7 PB |
8849 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, |
8850 | exit_work); | |
85faa7b8 | 8851 | |
56952e91 JA |
8852 | /* |
8853 | * If we're doing polled IO and end up having requests being | |
8854 | * submitted async (out-of-line), then completions can come in while | |
8855 | * we're waiting for refs to drop. We need to reap these manually, | |
8856 | * as nobody else will be looking for them. | |
8857 | */ | |
b2edc0a7 | 8858 | do { |
9936c7c2 | 8859 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
7c25c0d1 | 8860 | io_run_ctx_fallback(ctx); |
b2edc0a7 | 8861 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
85faa7b8 JA |
8862 | io_ring_ctx_free(ctx); |
8863 | } | |
8864 | ||
2b188cc1 JA |
8865 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8866 | { | |
8867 | mutex_lock(&ctx->uring_lock); | |
8868 | percpu_ref_kill(&ctx->refs); | |
d9d05217 PB |
8869 | |
8870 | if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead)) | |
8871 | ctx->sqo_dead = 1; | |
8872 | ||
cda286f0 PB |
8873 | /* if force is set, the ring is going away. always drop after that */ |
8874 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8875 | if (ctx->rings) |
6c503150 | 8876 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8877 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8878 | mutex_unlock(&ctx->uring_lock); |
8879 | ||
6b81928d PB |
8880 | io_kill_timeouts(ctx, NULL, NULL); |
8881 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8882 | |
15dff286 | 8883 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8884 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8885 | |
85faa7b8 | 8886 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8887 | /* |
8888 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8889 | * if we're exiting a ton of rings at the same time. It just adds | |
8890 | * noise and overhead, there's no discernable change in runtime | |
8891 | * over using system_wq. | |
8892 | */ | |
8893 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8894 | } |
8895 | ||
8896 | static int io_uring_release(struct inode *inode, struct file *file) | |
8897 | { | |
8898 | struct io_ring_ctx *ctx = file->private_data; | |
8899 | ||
8900 | file->private_data = NULL; | |
8901 | io_ring_ctx_wait_and_kill(ctx); | |
8902 | return 0; | |
8903 | } | |
8904 | ||
f6edbabb PB |
8905 | struct io_task_cancel { |
8906 | struct task_struct *task; | |
8907 | struct files_struct *files; | |
8908 | }; | |
f254ac04 | 8909 | |
f6edbabb | 8910 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8911 | { |
9a472ef7 | 8912 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8913 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8914 | bool ret; |
8915 | ||
f6edbabb | 8916 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8917 | unsigned long flags; |
8918 | struct io_ring_ctx *ctx = req->ctx; | |
8919 | ||
8920 | /* protect against races with linked timeouts */ | |
8921 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8922 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8923 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8924 | } else { | |
f6edbabb | 8925 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8926 | } |
8927 | return ret; | |
b711d4ea JA |
8928 | } |
8929 | ||
b7ddce3c | 8930 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8931 | struct task_struct *task, |
b7ddce3c PB |
8932 | struct files_struct *files) |
8933 | { | |
8934 | struct io_defer_entry *de = NULL; | |
8935 | LIST_HEAD(list); | |
8936 | ||
8937 | spin_lock_irq(&ctx->completion_lock); | |
8938 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8939 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8940 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8941 | break; | |
8942 | } | |
8943 | } | |
8944 | spin_unlock_irq(&ctx->completion_lock); | |
8945 | ||
8946 | while (!list_empty(&list)) { | |
8947 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8948 | list_del_init(&de->list); | |
8949 | req_set_fail_links(de->req); | |
8950 | io_put_req(de->req); | |
8951 | io_req_complete(de->req, -ECANCELED); | |
8952 | kfree(de); | |
8953 | } | |
8954 | } | |
8955 | ||
9936c7c2 PB |
8956 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8957 | struct task_struct *task, | |
8958 | struct files_struct *files) | |
8959 | { | |
8960 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
5aa75ed5 | 8961 | struct io_uring_task *tctx = current->io_uring; |
9936c7c2 PB |
8962 | |
8963 | while (1) { | |
8964 | enum io_wq_cancel cret; | |
8965 | bool ret = false; | |
8966 | ||
5aa75ed5 JA |
8967 | if (tctx && tctx->io_wq) { |
8968 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, | |
9936c7c2 PB |
8969 | &cancel, true); |
8970 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8971 | } | |
8972 | ||
8973 | /* SQPOLL thread does its own polling */ | |
8974 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8975 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8976 | io_iopoll_try_reap_events(ctx); | |
8977 | ret = true; | |
8978 | } | |
8979 | } | |
8980 | ||
8981 | ret |= io_poll_remove_all(ctx, task, files); | |
8982 | ret |= io_kill_timeouts(ctx, task, files); | |
8983 | ret |= io_run_task_work(); | |
8984 | io_cqring_overflow_flush(ctx, true, task, files); | |
8985 | if (!ret) | |
8986 | break; | |
8987 | cond_resched(); | |
8988 | } | |
8989 | } | |
8990 | ||
ca70f00b PB |
8991 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8992 | struct task_struct *task, | |
8993 | struct files_struct *files) | |
8994 | { | |
8995 | struct io_kiocb *req; | |
8996 | int cnt = 0; | |
8997 | ||
8998 | spin_lock_irq(&ctx->inflight_lock); | |
8999 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
9000 | cnt += io_match_task(req, task, files); | |
9001 | spin_unlock_irq(&ctx->inflight_lock); | |
9002 | return cnt; | |
9003 | } | |
9004 | ||
b52fda00 | 9005 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 9006 | struct task_struct *task, |
fcb323cc JA |
9007 | struct files_struct *files) |
9008 | { | |
fcb323cc | 9009 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 9010 | DEFINE_WAIT(wait); |
ca70f00b | 9011 | int inflight; |
fcb323cc | 9012 | |
ca70f00b PB |
9013 | inflight = io_uring_count_inflight(ctx, task, files); |
9014 | if (!inflight) | |
fcb323cc | 9015 | break; |
f6edbabb | 9016 | |
9936c7c2 | 9017 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b | 9018 | |
34343786 PB |
9019 | if (ctx->sq_data) |
9020 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
9021 | prepare_to_wait(&task->io_uring->wait, &wait, |
9022 | TASK_UNINTERRUPTIBLE); | |
9023 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
9024 | schedule(); | |
c98de08c | 9025 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
9026 | if (ctx->sq_data) |
9027 | io_sq_thread_park(ctx->sq_data); | |
0f212204 | 9028 | } |
0f212204 JA |
9029 | } |
9030 | ||
d9d05217 PB |
9031 | static void io_disable_sqo_submit(struct io_ring_ctx *ctx) |
9032 | { | |
d9d05217 PB |
9033 | mutex_lock(&ctx->uring_lock); |
9034 | ctx->sqo_dead = 1; | |
9035 | mutex_unlock(&ctx->uring_lock); | |
9036 | ||
9037 | /* make sure callers enter the ring to get error */ | |
b4411616 PB |
9038 | if (ctx->rings) |
9039 | io_ring_set_wakeup_flag(ctx); | |
d9d05217 PB |
9040 | } |
9041 | ||
0f212204 JA |
9042 | /* |
9043 | * We need to iteratively cancel requests, in case a request has dependent | |
9044 | * hard links. These persist even for failure of cancelations, hence keep | |
9045 | * looping until none are found. | |
9046 | */ | |
9047 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
9048 | struct files_struct *files) | |
9049 | { | |
9050 | struct task_struct *task = current; | |
9051 | ||
fdaf083c | 9052 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
d9d05217 | 9053 | io_disable_sqo_submit(ctx); |
534ca6d6 | 9054 | task = ctx->sq_data->thread; |
fdaf083c JA |
9055 | atomic_inc(&task->io_uring->in_idle); |
9056 | io_sq_thread_park(ctx->sq_data); | |
9057 | } | |
0f212204 | 9058 | |
df9923f9 | 9059 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 9060 | |
3a7efd1a | 9061 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 9062 | if (!files) |
9936c7c2 | 9063 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c JA |
9064 | |
9065 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { | |
9066 | atomic_dec(&task->io_uring->in_idle); | |
fdaf083c JA |
9067 | io_sq_thread_unpark(ctx->sq_data); |
9068 | } | |
0f212204 JA |
9069 | } |
9070 | ||
9071 | /* | |
9072 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
9073 | */ | |
fdaf083c | 9074 | static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file) |
0f212204 | 9075 | { |
236434c3 | 9076 | struct io_uring_task *tctx = current->io_uring; |
a528b04e | 9077 | int ret; |
236434c3 MWO |
9078 | |
9079 | if (unlikely(!tctx)) { | |
5aa75ed5 | 9080 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
9081 | if (unlikely(ret)) |
9082 | return ret; | |
236434c3 | 9083 | tctx = current->io_uring; |
0f212204 | 9084 | } |
236434c3 MWO |
9085 | if (tctx->last != file) { |
9086 | void *old = xa_load(&tctx->xa, (unsigned long)file); | |
0f212204 | 9087 | |
236434c3 | 9088 | if (!old) { |
0f212204 | 9089 | get_file(file); |
a528b04e PB |
9090 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)file, |
9091 | file, GFP_KERNEL)); | |
9092 | if (ret) { | |
9093 | fput(file); | |
9094 | return ret; | |
9095 | } | |
ecfc8492 PB |
9096 | |
9097 | /* one and only SQPOLL file note, held by sqo_task */ | |
9098 | WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && | |
9099 | current != ctx->sqo_task); | |
0f212204 | 9100 | } |
236434c3 | 9101 | tctx->last = file; |
0f212204 JA |
9102 | } |
9103 | ||
fdaf083c JA |
9104 | /* |
9105 | * This is race safe in that the task itself is doing this, hence it | |
9106 | * cannot be going through the exit/cancel paths at the same time. | |
9107 | * This cannot be modified while exit/cancel is running. | |
9108 | */ | |
9109 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
9110 | tctx->sqpoll = true; | |
9111 | ||
0f212204 JA |
9112 | return 0; |
9113 | } | |
9114 | ||
9115 | /* | |
9116 | * Remove this io_uring_file -> task mapping. | |
9117 | */ | |
9118 | static void io_uring_del_task_file(struct file *file) | |
9119 | { | |
9120 | struct io_uring_task *tctx = current->io_uring; | |
0f212204 JA |
9121 | |
9122 | if (tctx->last == file) | |
9123 | tctx->last = NULL; | |
5e2ed8c4 | 9124 | file = xa_erase(&tctx->xa, (unsigned long)file); |
0f212204 JA |
9125 | if (file) |
9126 | fput(file); | |
9127 | } | |
9128 | ||
de7f1d9e PB |
9129 | static void io_uring_remove_task_files(struct io_uring_task *tctx) |
9130 | { | |
9131 | struct file *file; | |
9132 | unsigned long index; | |
9133 | ||
9134 | xa_for_each(&tctx->xa, index, file) | |
9135 | io_uring_del_task_file(file); | |
9136 | } | |
9137 | ||
0f212204 JA |
9138 | void __io_uring_files_cancel(struct files_struct *files) |
9139 | { | |
9140 | struct io_uring_task *tctx = current->io_uring; | |
ce765372 MWO |
9141 | struct file *file; |
9142 | unsigned long index; | |
0f212204 JA |
9143 | |
9144 | /* make sure overflow events are dropped */ | |
fdaf083c | 9145 | atomic_inc(&tctx->in_idle); |
de7f1d9e PB |
9146 | xa_for_each(&tctx->xa, index, file) |
9147 | io_uring_cancel_task_requests(file->private_data, files); | |
fdaf083c | 9148 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9149 | |
5aa75ed5 | 9150 | if (files) { |
de7f1d9e | 9151 | io_uring_remove_task_files(tctx); |
5aa75ed5 JA |
9152 | } else if (tctx->io_wq && current->flags & PF_EXITING) { |
9153 | io_wq_destroy(tctx->io_wq); | |
9154 | tctx->io_wq = NULL; | |
9155 | } | |
fdaf083c JA |
9156 | } |
9157 | ||
9158 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
9159 | { | |
0e9ddb39 PB |
9160 | return percpu_counter_sum(&tctx->inflight); |
9161 | } | |
fdaf083c | 9162 | |
0e9ddb39 PB |
9163 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
9164 | { | |
9165 | struct io_uring_task *tctx; | |
9166 | s64 inflight; | |
9167 | DEFINE_WAIT(wait); | |
fdaf083c | 9168 | |
0e9ddb39 PB |
9169 | if (!ctx->sq_data) |
9170 | return; | |
9171 | tctx = ctx->sq_data->thread->io_uring; | |
9172 | io_disable_sqo_submit(ctx); | |
fdaf083c | 9173 | |
0e9ddb39 PB |
9174 | atomic_inc(&tctx->in_idle); |
9175 | do { | |
9176 | /* read completions before cancelations */ | |
9177 | inflight = tctx_inflight(tctx); | |
9178 | if (!inflight) | |
9179 | break; | |
9180 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 9181 | |
0e9ddb39 PB |
9182 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9183 | /* | |
9184 | * If we've seen completions, retry without waiting. This | |
9185 | * avoids a race where a completion comes in before we did | |
9186 | * prepare_to_wait(). | |
9187 | */ | |
9188 | if (inflight == tctx_inflight(tctx)) | |
9189 | schedule(); | |
9190 | finish_wait(&tctx->wait, &wait); | |
9191 | } while (1); | |
9192 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9193 | } |
9194 | ||
0f212204 JA |
9195 | /* |
9196 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9197 | * requests. | |
9198 | */ | |
9199 | void __io_uring_task_cancel(void) | |
9200 | { | |
9201 | struct io_uring_task *tctx = current->io_uring; | |
9202 | DEFINE_WAIT(wait); | |
d8a6df10 | 9203 | s64 inflight; |
0f212204 JA |
9204 | |
9205 | /* make sure overflow events are dropped */ | |
fdaf083c | 9206 | atomic_inc(&tctx->in_idle); |
0f212204 | 9207 | |
0b5cd6c3 | 9208 | /* trigger io_disable_sqo_submit() */ |
0e9ddb39 PB |
9209 | if (tctx->sqpoll) { |
9210 | struct file *file; | |
9211 | unsigned long index; | |
9212 | ||
9213 | xa_for_each(&tctx->xa, index, file) | |
9214 | io_uring_cancel_sqpoll(file->private_data); | |
9215 | } | |
0b5cd6c3 | 9216 | |
d8a6df10 | 9217 | do { |
0f212204 | 9218 | /* read completions before cancelations */ |
fdaf083c | 9219 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9220 | if (!inflight) |
9221 | break; | |
0f212204 JA |
9222 | __io_uring_files_cancel(NULL); |
9223 | ||
9224 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9225 | ||
9226 | /* | |
a1bb3cd5 PB |
9227 | * If we've seen completions, retry without waiting. This |
9228 | * avoids a race where a completion comes in before we did | |
9229 | * prepare_to_wait(). | |
0f212204 | 9230 | */ |
a1bb3cd5 PB |
9231 | if (inflight == tctx_inflight(tctx)) |
9232 | schedule(); | |
f57555ed | 9233 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9234 | } while (1); |
0f212204 | 9235 | |
fdaf083c | 9236 | atomic_dec(&tctx->in_idle); |
de7f1d9e PB |
9237 | |
9238 | io_uring_remove_task_files(tctx); | |
44e728b8 PB |
9239 | } |
9240 | ||
fcb323cc JA |
9241 | static int io_uring_flush(struct file *file, void *data) |
9242 | { | |
6b5733eb | 9243 | struct io_uring_task *tctx = current->io_uring; |
d9d05217 | 9244 | struct io_ring_ctx *ctx = file->private_data; |
6b5733eb | 9245 | |
3bfe6106 JA |
9246 | /* Ignore helper thread files exit */ |
9247 | if (current->flags & PF_IO_WORKER) | |
9248 | return 0; | |
9249 | ||
41be53e9 | 9250 | if (fatal_signal_pending(current) || (current->flags & PF_EXITING)) { |
84965ff8 | 9251 | io_uring_cancel_task_requests(ctx, NULL); |
41be53e9 JA |
9252 | io_req_caches_free(ctx, current); |
9253 | } | |
84965ff8 | 9254 | |
7c25c0d1 JA |
9255 | io_run_ctx_fallback(ctx); |
9256 | ||
6b5733eb | 9257 | if (!tctx) |
4f793dc4 PB |
9258 | return 0; |
9259 | ||
6b5733eb PB |
9260 | /* we should have cancelled and erased it before PF_EXITING */ |
9261 | WARN_ON_ONCE((current->flags & PF_EXITING) && | |
9262 | xa_load(&tctx->xa, (unsigned long)file)); | |
9263 | ||
4f793dc4 PB |
9264 | /* |
9265 | * fput() is pending, will be 2 if the only other ref is our potential | |
9266 | * task file note. If the task is exiting, drop regardless of count. | |
9267 | */ | |
6b5733eb PB |
9268 | if (atomic_long_read(&file->f_count) != 2) |
9269 | return 0; | |
4f793dc4 | 9270 | |
d9d05217 PB |
9271 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
9272 | /* there is only one file note, which is owned by sqo_task */ | |
4325cb49 PB |
9273 | WARN_ON_ONCE(ctx->sqo_task != current && |
9274 | xa_load(&tctx->xa, (unsigned long)file)); | |
9275 | /* sqo_dead check is for when this happens after cancellation */ | |
9276 | WARN_ON_ONCE(ctx->sqo_task == current && !ctx->sqo_dead && | |
d9d05217 PB |
9277 | !xa_load(&tctx->xa, (unsigned long)file)); |
9278 | ||
9279 | io_disable_sqo_submit(ctx); | |
9280 | } | |
9281 | ||
9282 | if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current) | |
9283 | io_uring_del_task_file(file); | |
fcb323cc JA |
9284 | return 0; |
9285 | } | |
9286 | ||
6c5c240e RP |
9287 | static void *io_uring_validate_mmap_request(struct file *file, |
9288 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9289 | { |
2b188cc1 | 9290 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9291 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9292 | struct page *page; |
9293 | void *ptr; | |
9294 | ||
9295 | switch (offset) { | |
9296 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9297 | case IORING_OFF_CQ_RING: |
9298 | ptr = ctx->rings; | |
2b188cc1 JA |
9299 | break; |
9300 | case IORING_OFF_SQES: | |
9301 | ptr = ctx->sq_sqes; | |
9302 | break; | |
2b188cc1 | 9303 | default: |
6c5c240e | 9304 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9305 | } |
9306 | ||
9307 | page = virt_to_head_page(ptr); | |
a50b854e | 9308 | if (sz > page_size(page)) |
6c5c240e RP |
9309 | return ERR_PTR(-EINVAL); |
9310 | ||
9311 | return ptr; | |
9312 | } | |
9313 | ||
9314 | #ifdef CONFIG_MMU | |
9315 | ||
9316 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9317 | { | |
9318 | size_t sz = vma->vm_end - vma->vm_start; | |
9319 | unsigned long pfn; | |
9320 | void *ptr; | |
9321 | ||
9322 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9323 | if (IS_ERR(ptr)) | |
9324 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9325 | |
9326 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9327 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9328 | } | |
9329 | ||
6c5c240e RP |
9330 | #else /* !CONFIG_MMU */ |
9331 | ||
9332 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9333 | { | |
9334 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9335 | } | |
9336 | ||
9337 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9338 | { | |
9339 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9340 | } | |
9341 | ||
9342 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9343 | unsigned long addr, unsigned long len, | |
9344 | unsigned long pgoff, unsigned long flags) | |
9345 | { | |
9346 | void *ptr; | |
9347 | ||
9348 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9349 | if (IS_ERR(ptr)) | |
9350 | return PTR_ERR(ptr); | |
9351 | ||
9352 | return (unsigned long) ptr; | |
9353 | } | |
9354 | ||
9355 | #endif /* !CONFIG_MMU */ | |
9356 | ||
d9d05217 | 9357 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9358 | { |
d9d05217 | 9359 | int ret = 0; |
90554200 JA |
9360 | DEFINE_WAIT(wait); |
9361 | ||
9362 | do { | |
9363 | if (!io_sqring_full(ctx)) | |
9364 | break; | |
9365 | ||
9366 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); | |
9367 | ||
d9d05217 PB |
9368 | if (unlikely(ctx->sqo_dead)) { |
9369 | ret = -EOWNERDEAD; | |
9370 | goto out; | |
9371 | } | |
9372 | ||
90554200 JA |
9373 | if (!io_sqring_full(ctx)) |
9374 | break; | |
9375 | ||
9376 | schedule(); | |
9377 | } while (!signal_pending(current)); | |
9378 | ||
9379 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 PB |
9380 | out: |
9381 | return ret; | |
90554200 JA |
9382 | } |
9383 | ||
c73ebb68 HX |
9384 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9385 | struct __kernel_timespec __user **ts, | |
9386 | const sigset_t __user **sig) | |
9387 | { | |
9388 | struct io_uring_getevents_arg arg; | |
9389 | ||
9390 | /* | |
9391 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9392 | * is just a pointer to the sigset_t. | |
9393 | */ | |
9394 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9395 | *sig = (const sigset_t __user *) argp; | |
9396 | *ts = NULL; | |
9397 | return 0; | |
9398 | } | |
9399 | ||
9400 | /* | |
9401 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9402 | * timespec and sigset_t pointers if good. | |
9403 | */ | |
9404 | if (*argsz != sizeof(arg)) | |
9405 | return -EINVAL; | |
9406 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9407 | return -EFAULT; | |
9408 | *sig = u64_to_user_ptr(arg.sigmask); | |
9409 | *argsz = arg.sigmask_sz; | |
9410 | *ts = u64_to_user_ptr(arg.ts); | |
9411 | return 0; | |
9412 | } | |
9413 | ||
2b188cc1 | 9414 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9415 | u32, min_complete, u32, flags, const void __user *, argp, |
9416 | size_t, argsz) | |
2b188cc1 JA |
9417 | { |
9418 | struct io_ring_ctx *ctx; | |
9419 | long ret = -EBADF; | |
9420 | int submitted = 0; | |
9421 | struct fd f; | |
9422 | ||
4c6e277c | 9423 | io_run_task_work(); |
b41e9852 | 9424 | |
90554200 | 9425 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9426 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9427 | return -EINVAL; |
9428 | ||
9429 | f = fdget(fd); | |
9430 | if (!f.file) | |
9431 | return -EBADF; | |
9432 | ||
9433 | ret = -EOPNOTSUPP; | |
9434 | if (f.file->f_op != &io_uring_fops) | |
9435 | goto out_fput; | |
9436 | ||
9437 | ret = -ENXIO; | |
9438 | ctx = f.file->private_data; | |
9439 | if (!percpu_ref_tryget(&ctx->refs)) | |
9440 | goto out_fput; | |
9441 | ||
7e84e1c7 SG |
9442 | ret = -EBADFD; |
9443 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9444 | goto out; | |
9445 | ||
6c271ce2 JA |
9446 | /* |
9447 | * For SQ polling, the thread will do all submissions and completions. | |
9448 | * Just return the requested submit count, and wake the thread if | |
9449 | * we were asked to. | |
9450 | */ | |
b2a9eada | 9451 | ret = 0; |
6c271ce2 | 9452 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9453 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9454 | |
d9d05217 PB |
9455 | ret = -EOWNERDEAD; |
9456 | if (unlikely(ctx->sqo_dead)) | |
9457 | goto out; | |
6c271ce2 | 9458 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9459 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9460 | if (flags & IORING_ENTER_SQ_WAIT) { |
9461 | ret = io_sqpoll_wait_sq(ctx); | |
9462 | if (ret) | |
9463 | goto out; | |
9464 | } | |
6c271ce2 | 9465 | submitted = to_submit; |
b2a9eada | 9466 | } else if (to_submit) { |
fdaf083c | 9467 | ret = io_uring_add_task_file(ctx, f.file); |
0f212204 JA |
9468 | if (unlikely(ret)) |
9469 | goto out; | |
2b188cc1 | 9470 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9471 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9472 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9473 | |
9474 | if (submitted != to_submit) | |
9475 | goto out; | |
2b188cc1 JA |
9476 | } |
9477 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9478 | const sigset_t __user *sig; |
9479 | struct __kernel_timespec __user *ts; | |
9480 | ||
9481 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9482 | if (unlikely(ret)) | |
9483 | goto out; | |
9484 | ||
2b188cc1 JA |
9485 | min_complete = min(min_complete, ctx->cq_entries); |
9486 | ||
32b2244a XW |
9487 | /* |
9488 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9489 | * space applications don't need to do io completion events | |
9490 | * polling again, they can rely on io_sq_thread to do polling | |
9491 | * work, which can reduce cpu usage and uring_lock contention. | |
9492 | */ | |
9493 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9494 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9495 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9496 | } else { |
c73ebb68 | 9497 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9498 | } |
2b188cc1 JA |
9499 | } |
9500 | ||
7c504e65 | 9501 | out: |
6805b32e | 9502 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9503 | out_fput: |
9504 | fdput(f); | |
9505 | return submitted ? submitted : ret; | |
9506 | } | |
9507 | ||
bebdb65e | 9508 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9509 | static int io_uring_show_cred(int id, void *p, void *data) |
9510 | { | |
6b47ab81 JA |
9511 | struct io_identity *iod = p; |
9512 | const struct cred *cred = iod->creds; | |
87ce955b JA |
9513 | struct seq_file *m = data; |
9514 | struct user_namespace *uns = seq_user_ns(m); | |
9515 | struct group_info *gi; | |
9516 | kernel_cap_t cap; | |
9517 | unsigned __capi; | |
9518 | int g; | |
9519 | ||
9520 | seq_printf(m, "%5d\n", id); | |
9521 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9522 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9523 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9524 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9525 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9526 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9527 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9528 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9529 | seq_puts(m, "\n\tGroups:\t"); | |
9530 | gi = cred->group_info; | |
9531 | for (g = 0; g < gi->ngroups; g++) { | |
9532 | seq_put_decimal_ull(m, g ? " " : "", | |
9533 | from_kgid_munged(uns, gi->gid[g])); | |
9534 | } | |
9535 | seq_puts(m, "\n\tCapEff:\t"); | |
9536 | cap = cred->cap_effective; | |
9537 | CAP_FOR_EACH_U32(__capi) | |
9538 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9539 | seq_putc(m, '\n'); | |
9540 | return 0; | |
9541 | } | |
9542 | ||
9543 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9544 | { | |
dbbe9c64 | 9545 | struct io_sq_data *sq = NULL; |
fad8e0de | 9546 | bool has_lock; |
87ce955b JA |
9547 | int i; |
9548 | ||
fad8e0de JA |
9549 | /* |
9550 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9551 | * since fdinfo case grabs it in the opposite direction of normal use | |
9552 | * cases. If we fail to get the lock, we just don't iterate any | |
9553 | * structures that could be going away outside the io_uring mutex. | |
9554 | */ | |
9555 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9556 | ||
dbbe9c64 JQ |
9557 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) |
9558 | sq = ctx->sq_data; | |
9559 | ||
9560 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9561 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9562 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9563 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9564 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9565 | |
87ce955b JA |
9566 | if (f) |
9567 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9568 | else | |
9569 | seq_printf(m, "%5u: <none>\n", i); | |
9570 | } | |
9571 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9572 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9573 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9574 | ||
9575 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9576 | (unsigned int) buf->len); | |
9577 | } | |
fad8e0de | 9578 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9579 | seq_printf(m, "Personalities:\n"); |
9580 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9581 | } | |
d7718a9d JA |
9582 | seq_printf(m, "PollList:\n"); |
9583 | spin_lock_irq(&ctx->completion_lock); | |
9584 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9585 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9586 | struct io_kiocb *req; | |
9587 | ||
9588 | hlist_for_each_entry(req, list, hash_node) | |
9589 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9590 | req->task->task_works != NULL); | |
9591 | } | |
9592 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9593 | if (has_lock) |
9594 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9595 | } |
9596 | ||
9597 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9598 | { | |
9599 | struct io_ring_ctx *ctx = f->private_data; | |
9600 | ||
9601 | if (percpu_ref_tryget(&ctx->refs)) { | |
9602 | __io_uring_show_fdinfo(ctx, m); | |
9603 | percpu_ref_put(&ctx->refs); | |
9604 | } | |
9605 | } | |
bebdb65e | 9606 | #endif |
87ce955b | 9607 | |
2b188cc1 JA |
9608 | static const struct file_operations io_uring_fops = { |
9609 | .release = io_uring_release, | |
fcb323cc | 9610 | .flush = io_uring_flush, |
2b188cc1 | 9611 | .mmap = io_uring_mmap, |
6c5c240e RP |
9612 | #ifndef CONFIG_MMU |
9613 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9614 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9615 | #endif | |
2b188cc1 JA |
9616 | .poll = io_uring_poll, |
9617 | .fasync = io_uring_fasync, | |
bebdb65e | 9618 | #ifdef CONFIG_PROC_FS |
87ce955b | 9619 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9620 | #endif |
2b188cc1 JA |
9621 | }; |
9622 | ||
9623 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9624 | struct io_uring_params *p) | |
9625 | { | |
75b28aff HV |
9626 | struct io_rings *rings; |
9627 | size_t size, sq_array_offset; | |
2b188cc1 | 9628 | |
bd740481 JA |
9629 | /* make sure these are sane, as we already accounted them */ |
9630 | ctx->sq_entries = p->sq_entries; | |
9631 | ctx->cq_entries = p->cq_entries; | |
9632 | ||
75b28aff HV |
9633 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9634 | if (size == SIZE_MAX) | |
9635 | return -EOVERFLOW; | |
9636 | ||
9637 | rings = io_mem_alloc(size); | |
9638 | if (!rings) | |
2b188cc1 JA |
9639 | return -ENOMEM; |
9640 | ||
75b28aff HV |
9641 | ctx->rings = rings; |
9642 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9643 | rings->sq_ring_mask = p->sq_entries - 1; | |
9644 | rings->cq_ring_mask = p->cq_entries - 1; | |
9645 | rings->sq_ring_entries = p->sq_entries; | |
9646 | rings->cq_ring_entries = p->cq_entries; | |
9647 | ctx->sq_mask = rings->sq_ring_mask; | |
9648 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9649 | |
9650 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9651 | if (size == SIZE_MAX) { |
9652 | io_mem_free(ctx->rings); | |
9653 | ctx->rings = NULL; | |
2b188cc1 | 9654 | return -EOVERFLOW; |
eb065d30 | 9655 | } |
2b188cc1 JA |
9656 | |
9657 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9658 | if (!ctx->sq_sqes) { |
9659 | io_mem_free(ctx->rings); | |
9660 | ctx->rings = NULL; | |
2b188cc1 | 9661 | return -ENOMEM; |
eb065d30 | 9662 | } |
2b188cc1 | 9663 | |
2b188cc1 JA |
9664 | return 0; |
9665 | } | |
9666 | ||
9faadcc8 PB |
9667 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9668 | { | |
9669 | int ret, fd; | |
9670 | ||
9671 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9672 | if (fd < 0) | |
9673 | return fd; | |
9674 | ||
9675 | ret = io_uring_add_task_file(ctx, file); | |
9676 | if (ret) { | |
9677 | put_unused_fd(fd); | |
9678 | return ret; | |
9679 | } | |
9680 | fd_install(fd, file); | |
9681 | return fd; | |
9682 | } | |
9683 | ||
2b188cc1 JA |
9684 | /* |
9685 | * Allocate an anonymous fd, this is what constitutes the application | |
9686 | * visible backing of an io_uring instance. The application mmaps this | |
9687 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9688 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9689 | */ | |
9faadcc8 | 9690 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9691 | { |
9692 | struct file *file; | |
9faadcc8 | 9693 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9694 | int ret; |
9695 | ||
2b188cc1 JA |
9696 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9697 | &ctx->ring_sock); | |
9698 | if (ret) | |
9faadcc8 | 9699 | return ERR_PTR(ret); |
2b188cc1 JA |
9700 | #endif |
9701 | ||
2b188cc1 JA |
9702 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9703 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9704 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9705 | if (IS_ERR(file)) { |
9706 | sock_release(ctx->ring_sock); | |
9707 | ctx->ring_sock = NULL; | |
9708 | } else { | |
9709 | ctx->ring_sock->file = file; | |
0f212204 | 9710 | } |
2b188cc1 | 9711 | #endif |
9faadcc8 | 9712 | return file; |
2b188cc1 JA |
9713 | } |
9714 | ||
7f13657d XW |
9715 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9716 | struct io_uring_params __user *params) | |
2b188cc1 JA |
9717 | { |
9718 | struct user_struct *user = NULL; | |
9719 | struct io_ring_ctx *ctx; | |
9faadcc8 | 9720 | struct file *file; |
2b188cc1 JA |
9721 | int ret; |
9722 | ||
8110c1a6 | 9723 | if (!entries) |
2b188cc1 | 9724 | return -EINVAL; |
8110c1a6 JA |
9725 | if (entries > IORING_MAX_ENTRIES) { |
9726 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9727 | return -EINVAL; | |
9728 | entries = IORING_MAX_ENTRIES; | |
9729 | } | |
2b188cc1 JA |
9730 | |
9731 | /* | |
9732 | * Use twice as many entries for the CQ ring. It's possible for the | |
9733 | * application to drive a higher depth than the size of the SQ ring, | |
9734 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9735 | * some flexibility in overcommitting a bit. If the application has |
9736 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9737 | * of CQ ring entries manually. | |
2b188cc1 JA |
9738 | */ |
9739 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9740 | if (p->flags & IORING_SETUP_CQSIZE) { |
9741 | /* | |
9742 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9743 | * to a power-of-two, if it isn't already. We do NOT impose | |
9744 | * any cq vs sq ring sizing. | |
9745 | */ | |
eb2667b3 | 9746 | if (!p->cq_entries) |
33a107f0 | 9747 | return -EINVAL; |
8110c1a6 JA |
9748 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9749 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9750 | return -EINVAL; | |
9751 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9752 | } | |
eb2667b3 JQ |
9753 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9754 | if (p->cq_entries < p->sq_entries) | |
9755 | return -EINVAL; | |
33a107f0 JA |
9756 | } else { |
9757 | p->cq_entries = 2 * p->sq_entries; | |
9758 | } | |
2b188cc1 JA |
9759 | |
9760 | user = get_uid(current_user()); | |
2b188cc1 JA |
9761 | |
9762 | ctx = io_ring_ctx_alloc(p); | |
9763 | if (!ctx) { | |
2b188cc1 JA |
9764 | free_uid(user); |
9765 | return -ENOMEM; | |
9766 | } | |
9767 | ctx->compat = in_compat_syscall(); | |
26bfa89e | 9768 | ctx->limit_mem = !capable(CAP_IPC_LOCK); |
2b188cc1 | 9769 | ctx->user = user; |
0b8c0ec7 | 9770 | ctx->creds = get_current_cred(); |
4ea33a97 JA |
9771 | #ifdef CONFIG_AUDIT |
9772 | ctx->loginuid = current->loginuid; | |
9773 | ctx->sessionid = current->sessionid; | |
9774 | #endif | |
2aede0e4 JA |
9775 | ctx->sqo_task = get_task_struct(current); |
9776 | ||
9777 | /* | |
9778 | * This is just grabbed for accounting purposes. When a process exits, | |
9779 | * the mm is exited and dropped before the files, hence we need to hang | |
9780 | * on to this mm purely for the purposes of being able to unaccount | |
9781 | * memory (locked/pinned vm). It's not used for anything else. | |
9782 | */ | |
6b7898eb | 9783 | mmgrab(current->mm); |
2aede0e4 | 9784 | ctx->mm_account = current->mm; |
6b7898eb | 9785 | |
91d8f519 DZ |
9786 | #ifdef CONFIG_BLK_CGROUP |
9787 | /* | |
9788 | * The sq thread will belong to the original cgroup it was inited in. | |
9789 | * If the cgroup goes offline (e.g. disabling the io controller), then | |
9790 | * issued bios will be associated with the closest cgroup later in the | |
9791 | * block layer. | |
9792 | */ | |
9793 | rcu_read_lock(); | |
9794 | ctx->sqo_blkcg_css = blkcg_css(); | |
9795 | ret = css_tryget_online(ctx->sqo_blkcg_css); | |
9796 | rcu_read_unlock(); | |
9797 | if (!ret) { | |
9798 | /* don't init against a dying cgroup, have the user try again */ | |
9799 | ctx->sqo_blkcg_css = NULL; | |
9800 | ret = -ENODEV; | |
9801 | goto err; | |
9802 | } | |
9803 | #endif | |
2b188cc1 JA |
9804 | ret = io_allocate_scq_urings(ctx, p); |
9805 | if (ret) | |
9806 | goto err; | |
9807 | ||
7e84e1c7 | 9808 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9809 | if (ret) |
9810 | goto err; | |
9811 | ||
7e84e1c7 SG |
9812 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9813 | io_sq_offload_start(ctx); | |
9814 | ||
2b188cc1 | 9815 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9816 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9817 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9818 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9819 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9820 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9821 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9822 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9823 | |
9824 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9825 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9826 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9827 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9828 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9829 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9830 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9831 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9832 | |
7f13657d XW |
9833 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9834 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9835 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 HX |
9836 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
9837 | IORING_FEAT_EXT_ARG; | |
7f13657d XW |
9838 | |
9839 | if (copy_to_user(params, p, sizeof(*p))) { | |
9840 | ret = -EFAULT; | |
9841 | goto err; | |
9842 | } | |
d1719f70 | 9843 | |
9faadcc8 PB |
9844 | file = io_uring_get_file(ctx); |
9845 | if (IS_ERR(file)) { | |
9846 | ret = PTR_ERR(file); | |
9847 | goto err; | |
9848 | } | |
9849 | ||
044c1ab3 JA |
9850 | /* |
9851 | * Install ring fd as the very last thing, so we don't risk someone | |
9852 | * having closed it before we finish setup | |
9853 | */ | |
9faadcc8 PB |
9854 | ret = io_uring_install_fd(ctx, file); |
9855 | if (ret < 0) { | |
06585c49 | 9856 | io_disable_sqo_submit(ctx); |
9faadcc8 PB |
9857 | /* fput will clean it up */ |
9858 | fput(file); | |
9859 | return ret; | |
9860 | } | |
044c1ab3 | 9861 | |
c826bd7a | 9862 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9863 | return ret; |
9864 | err: | |
d9d05217 | 9865 | io_disable_sqo_submit(ctx); |
2b188cc1 JA |
9866 | io_ring_ctx_wait_and_kill(ctx); |
9867 | return ret; | |
9868 | } | |
9869 | ||
9870 | /* | |
9871 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9872 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9873 | * params structure passed in. | |
9874 | */ | |
9875 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9876 | { | |
9877 | struct io_uring_params p; | |
2b188cc1 JA |
9878 | int i; |
9879 | ||
9880 | if (copy_from_user(&p, params, sizeof(p))) | |
9881 | return -EFAULT; | |
9882 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9883 | if (p.resv[i]) | |
9884 | return -EINVAL; | |
9885 | } | |
9886 | ||
6c271ce2 | 9887 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9888 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9889 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9890 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9891 | return -EINVAL; |
9892 | ||
7f13657d | 9893 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9894 | } |
9895 | ||
9896 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9897 | struct io_uring_params __user *, params) | |
9898 | { | |
9899 | return io_uring_setup(entries, params); | |
9900 | } | |
9901 | ||
66f4af93 JA |
9902 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9903 | { | |
9904 | struct io_uring_probe *p; | |
9905 | size_t size; | |
9906 | int i, ret; | |
9907 | ||
9908 | size = struct_size(p, ops, nr_args); | |
9909 | if (size == SIZE_MAX) | |
9910 | return -EOVERFLOW; | |
9911 | p = kzalloc(size, GFP_KERNEL); | |
9912 | if (!p) | |
9913 | return -ENOMEM; | |
9914 | ||
9915 | ret = -EFAULT; | |
9916 | if (copy_from_user(p, arg, size)) | |
9917 | goto out; | |
9918 | ret = -EINVAL; | |
9919 | if (memchr_inv(p, 0, size)) | |
9920 | goto out; | |
9921 | ||
9922 | p->last_op = IORING_OP_LAST - 1; | |
9923 | if (nr_args > IORING_OP_LAST) | |
9924 | nr_args = IORING_OP_LAST; | |
9925 | ||
9926 | for (i = 0; i < nr_args; i++) { | |
9927 | p->ops[i].op = i; | |
9928 | if (!io_op_defs[i].not_supported) | |
9929 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9930 | } | |
9931 | p->ops_len = i; | |
9932 | ||
9933 | ret = 0; | |
9934 | if (copy_to_user(arg, p, size)) | |
9935 | ret = -EFAULT; | |
9936 | out: | |
9937 | kfree(p); | |
9938 | return ret; | |
9939 | } | |
9940 | ||
071698e1 JA |
9941 | static int io_register_personality(struct io_ring_ctx *ctx) |
9942 | { | |
1e6fa521 JA |
9943 | struct io_identity *id; |
9944 | int ret; | |
071698e1 | 9945 | |
1e6fa521 JA |
9946 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
9947 | if (unlikely(!id)) | |
9948 | return -ENOMEM; | |
9949 | ||
9950 | io_init_identity(id); | |
9951 | id->creds = get_current_cred(); | |
9952 | ||
9953 | ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL); | |
9954 | if (ret < 0) { | |
9955 | put_cred(id->creds); | |
9956 | kfree(id); | |
9957 | } | |
9958 | return ret; | |
071698e1 JA |
9959 | } |
9960 | ||
21b55dbc SG |
9961 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9962 | unsigned int nr_args) | |
9963 | { | |
9964 | struct io_uring_restriction *res; | |
9965 | size_t size; | |
9966 | int i, ret; | |
9967 | ||
7e84e1c7 SG |
9968 | /* Restrictions allowed only if rings started disabled */ |
9969 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9970 | return -EBADFD; | |
9971 | ||
21b55dbc | 9972 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9973 | if (ctx->restrictions.registered) |
21b55dbc SG |
9974 | return -EBUSY; |
9975 | ||
9976 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9977 | return -EINVAL; | |
9978 | ||
9979 | size = array_size(nr_args, sizeof(*res)); | |
9980 | if (size == SIZE_MAX) | |
9981 | return -EOVERFLOW; | |
9982 | ||
9983 | res = memdup_user(arg, size); | |
9984 | if (IS_ERR(res)) | |
9985 | return PTR_ERR(res); | |
9986 | ||
9987 | ret = 0; | |
9988 | ||
9989 | for (i = 0; i < nr_args; i++) { | |
9990 | switch (res[i].opcode) { | |
9991 | case IORING_RESTRICTION_REGISTER_OP: | |
9992 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9993 | ret = -EINVAL; | |
9994 | goto out; | |
9995 | } | |
9996 | ||
9997 | __set_bit(res[i].register_op, | |
9998 | ctx->restrictions.register_op); | |
9999 | break; | |
10000 | case IORING_RESTRICTION_SQE_OP: | |
10001 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
10002 | ret = -EINVAL; | |
10003 | goto out; | |
10004 | } | |
10005 | ||
10006 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
10007 | break; | |
10008 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
10009 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
10010 | break; | |
10011 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
10012 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
10013 | break; | |
10014 | default: | |
10015 | ret = -EINVAL; | |
10016 | goto out; | |
10017 | } | |
10018 | } | |
10019 | ||
10020 | out: | |
10021 | /* Reset all restrictions if an error happened */ | |
10022 | if (ret != 0) | |
10023 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
10024 | else | |
7e84e1c7 | 10025 | ctx->restrictions.registered = true; |
21b55dbc SG |
10026 | |
10027 | kfree(res); | |
10028 | return ret; | |
10029 | } | |
10030 | ||
7e84e1c7 SG |
10031 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
10032 | { | |
10033 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
10034 | return -EBADFD; | |
10035 | ||
10036 | if (ctx->restrictions.registered) | |
10037 | ctx->restricted = 1; | |
10038 | ||
10039 | ctx->flags &= ~IORING_SETUP_R_DISABLED; | |
10040 | ||
10041 | io_sq_offload_start(ctx); | |
10042 | ||
10043 | return 0; | |
10044 | } | |
10045 | ||
071698e1 JA |
10046 | static bool io_register_op_must_quiesce(int op) |
10047 | { | |
10048 | switch (op) { | |
10049 | case IORING_UNREGISTER_FILES: | |
10050 | case IORING_REGISTER_FILES_UPDATE: | |
10051 | case IORING_REGISTER_PROBE: | |
10052 | case IORING_REGISTER_PERSONALITY: | |
10053 | case IORING_UNREGISTER_PERSONALITY: | |
10054 | return false; | |
10055 | default: | |
10056 | return true; | |
10057 | } | |
10058 | } | |
10059 | ||
edafccee JA |
10060 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
10061 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
10062 | __releases(ctx->uring_lock) |
10063 | __acquires(ctx->uring_lock) | |
edafccee JA |
10064 | { |
10065 | int ret; | |
10066 | ||
35fa71a0 JA |
10067 | /* |
10068 | * We're inside the ring mutex, if the ref is already dying, then | |
10069 | * someone else killed the ctx or is already going through | |
10070 | * io_uring_register(). | |
10071 | */ | |
10072 | if (percpu_ref_is_dying(&ctx->refs)) | |
10073 | return -ENXIO; | |
10074 | ||
071698e1 | 10075 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10076 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 10077 | |
05f3fb3c JA |
10078 | /* |
10079 | * Drop uring mutex before waiting for references to exit. If | |
10080 | * another thread is currently inside io_uring_enter() it might | |
10081 | * need to grab the uring_lock to make progress. If we hold it | |
10082 | * here across the drain wait, then we can deadlock. It's safe | |
10083 | * to drop the mutex here, since no new references will come in | |
10084 | * after we've killed the percpu ref. | |
10085 | */ | |
10086 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
10087 | do { |
10088 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
10089 | if (!ret) | |
10090 | break; | |
ed6930c9 JA |
10091 | ret = io_run_task_work_sig(); |
10092 | if (ret < 0) | |
10093 | break; | |
af9c1a44 JA |
10094 | } while (1); |
10095 | ||
05f3fb3c | 10096 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 10097 | |
88f171ab PB |
10098 | if (ret && io_refs_resurrect(&ctx->refs, &ctx->ref_comp)) |
10099 | return ret; | |
21b55dbc SG |
10100 | } |
10101 | ||
10102 | if (ctx->restricted) { | |
10103 | if (opcode >= IORING_REGISTER_LAST) { | |
10104 | ret = -EINVAL; | |
10105 | goto out; | |
10106 | } | |
10107 | ||
10108 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
10109 | ret = -EACCES; | |
c150368b JA |
10110 | goto out; |
10111 | } | |
05f3fb3c | 10112 | } |
edafccee JA |
10113 | |
10114 | switch (opcode) { | |
10115 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 10116 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
10117 | break; |
10118 | case IORING_UNREGISTER_BUFFERS: | |
10119 | ret = -EINVAL; | |
10120 | if (arg || nr_args) | |
10121 | break; | |
0a96bbe4 | 10122 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 10123 | break; |
6b06314c JA |
10124 | case IORING_REGISTER_FILES: |
10125 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
10126 | break; | |
10127 | case IORING_UNREGISTER_FILES: | |
10128 | ret = -EINVAL; | |
10129 | if (arg || nr_args) | |
10130 | break; | |
10131 | ret = io_sqe_files_unregister(ctx); | |
10132 | break; | |
c3a31e60 JA |
10133 | case IORING_REGISTER_FILES_UPDATE: |
10134 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
10135 | break; | |
9b402849 | 10136 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 10137 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
10138 | ret = -EINVAL; |
10139 | if (nr_args != 1) | |
10140 | break; | |
10141 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
10142 | if (ret) |
10143 | break; | |
10144 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
10145 | ctx->eventfd_async = 1; | |
10146 | else | |
10147 | ctx->eventfd_async = 0; | |
9b402849 JA |
10148 | break; |
10149 | case IORING_UNREGISTER_EVENTFD: | |
10150 | ret = -EINVAL; | |
10151 | if (arg || nr_args) | |
10152 | break; | |
10153 | ret = io_eventfd_unregister(ctx); | |
10154 | break; | |
66f4af93 JA |
10155 | case IORING_REGISTER_PROBE: |
10156 | ret = -EINVAL; | |
10157 | if (!arg || nr_args > 256) | |
10158 | break; | |
10159 | ret = io_probe(ctx, arg, nr_args); | |
10160 | break; | |
071698e1 JA |
10161 | case IORING_REGISTER_PERSONALITY: |
10162 | ret = -EINVAL; | |
10163 | if (arg || nr_args) | |
10164 | break; | |
10165 | ret = io_register_personality(ctx); | |
10166 | break; | |
10167 | case IORING_UNREGISTER_PERSONALITY: | |
10168 | ret = -EINVAL; | |
10169 | if (arg) | |
10170 | break; | |
10171 | ret = io_unregister_personality(ctx, nr_args); | |
10172 | break; | |
7e84e1c7 SG |
10173 | case IORING_REGISTER_ENABLE_RINGS: |
10174 | ret = -EINVAL; | |
10175 | if (arg || nr_args) | |
10176 | break; | |
10177 | ret = io_register_enable_rings(ctx); | |
10178 | break; | |
21b55dbc SG |
10179 | case IORING_REGISTER_RESTRICTIONS: |
10180 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10181 | break; | |
edafccee JA |
10182 | default: |
10183 | ret = -EINVAL; | |
10184 | break; | |
10185 | } | |
10186 | ||
21b55dbc | 10187 | out: |
071698e1 | 10188 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10189 | /* bring the ctx back to life */ |
05f3fb3c | 10190 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10191 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10192 | } |
edafccee JA |
10193 | return ret; |
10194 | } | |
10195 | ||
10196 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10197 | void __user *, arg, unsigned int, nr_args) | |
10198 | { | |
10199 | struct io_ring_ctx *ctx; | |
10200 | long ret = -EBADF; | |
10201 | struct fd f; | |
10202 | ||
10203 | f = fdget(fd); | |
10204 | if (!f.file) | |
10205 | return -EBADF; | |
10206 | ||
10207 | ret = -EOPNOTSUPP; | |
10208 | if (f.file->f_op != &io_uring_fops) | |
10209 | goto out_fput; | |
10210 | ||
10211 | ctx = f.file->private_data; | |
10212 | ||
b6c23dd5 PB |
10213 | io_run_task_work(); |
10214 | ||
edafccee JA |
10215 | mutex_lock(&ctx->uring_lock); |
10216 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10217 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10218 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10219 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10220 | out_fput: |
10221 | fdput(f); | |
10222 | return ret; | |
10223 | } | |
10224 | ||
2b188cc1 JA |
10225 | static int __init io_uring_init(void) |
10226 | { | |
d7f62e82 SM |
10227 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10228 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10229 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10230 | } while (0) | |
10231 | ||
10232 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10233 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10234 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10235 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10236 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10237 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10238 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10239 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10240 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10241 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10242 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10243 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10244 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10245 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10246 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10247 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10248 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10249 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10250 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10251 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10252 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10253 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10254 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10255 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10256 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10257 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10258 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10259 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10260 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10261 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10262 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10263 | |
d3656344 | 10264 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10265 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10266 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10267 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10268 | return 0; |
10269 | }; | |
10270 | __initcall(io_uring_init); |