]>
Commit | Line | Data |
---|---|---|
2b188cc1 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Shared application/kernel submission and completion ring pairs, for | |
4 | * supporting fast/efficient IO. | |
5 | * | |
6 | * A note on the read/write ordering memory barriers that are matched between | |
1e84b97b SB |
7 | * the application and kernel side. |
8 | * | |
9 | * After the application reads the CQ ring tail, it must use an | |
10 | * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses | |
11 | * before writing the tail (using smp_load_acquire to read the tail will | |
12 | * do). It also needs a smp_mb() before updating CQ head (ordering the | |
13 | * entry load(s) with the head store), pairing with an implicit barrier | |
14 | * through a control-dependency in io_get_cqring (smp_store_release to | |
15 | * store head will do). Failure to do so could lead to reading invalid | |
16 | * CQ entries. | |
17 | * | |
18 | * Likewise, the application must use an appropriate smp_wmb() before | |
19 | * writing the SQ tail (ordering SQ entry stores with the tail store), | |
20 | * which pairs with smp_load_acquire in io_get_sqring (smp_store_release | |
21 | * to store the tail will do). And it needs a barrier ordering the SQ | |
22 | * head load before writing new SQ entries (smp_load_acquire to read | |
23 | * head will do). | |
24 | * | |
25 | * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application | |
26 | * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after* | |
27 | * updating the SQ tail; a full memory barrier smp_mb() is needed | |
28 | * between. | |
2b188cc1 JA |
29 | * |
30 | * Also see the examples in the liburing library: | |
31 | * | |
32 | * git://git.kernel.dk/liburing | |
33 | * | |
34 | * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens | |
35 | * from data shared between the kernel and application. This is done both | |
36 | * for ordering purposes, but also to ensure that once a value is loaded from | |
37 | * data that the application could potentially modify, it remains stable. | |
38 | * | |
39 | * Copyright (C) 2018-2019 Jens Axboe | |
c992fe29 | 40 | * Copyright (c) 2018-2019 Christoph Hellwig |
2b188cc1 JA |
41 | */ |
42 | #include <linux/kernel.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/errno.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/compat.h> | |
52de1fe1 | 47 | #include <net/compat.h> |
2b188cc1 JA |
48 | #include <linux/refcount.h> |
49 | #include <linux/uio.h> | |
6b47ee6e | 50 | #include <linux/bits.h> |
2b188cc1 JA |
51 | |
52 | #include <linux/sched/signal.h> | |
53 | #include <linux/fs.h> | |
54 | #include <linux/file.h> | |
55 | #include <linux/fdtable.h> | |
56 | #include <linux/mm.h> | |
57 | #include <linux/mman.h> | |
2b188cc1 JA |
58 | #include <linux/percpu.h> |
59 | #include <linux/slab.h> | |
2b188cc1 | 60 | #include <linux/blkdev.h> |
edafccee | 61 | #include <linux/bvec.h> |
2b188cc1 JA |
62 | #include <linux/net.h> |
63 | #include <net/sock.h> | |
64 | #include <net/af_unix.h> | |
6b06314c | 65 | #include <net/scm.h> |
2b188cc1 JA |
66 | #include <linux/anon_inodes.h> |
67 | #include <linux/sched/mm.h> | |
68 | #include <linux/uaccess.h> | |
69 | #include <linux/nospec.h> | |
edafccee JA |
70 | #include <linux/sizes.h> |
71 | #include <linux/hugetlb.h> | |
aa4c3967 | 72 | #include <linux/highmem.h> |
15b71abe JA |
73 | #include <linux/namei.h> |
74 | #include <linux/fsnotify.h> | |
4840e418 | 75 | #include <linux/fadvise.h> |
3e4827b0 | 76 | #include <linux/eventpoll.h> |
7d67af2c | 77 | #include <linux/splice.h> |
b41e9852 | 78 | #include <linux/task_work.h> |
bcf5a063 | 79 | #include <linux/pagemap.h> |
0f212204 | 80 | #include <linux/io_uring.h> |
e4b4a13f | 81 | #include <linux/freezer.h> |
2b188cc1 | 82 | |
c826bd7a DD |
83 | #define CREATE_TRACE_POINTS |
84 | #include <trace/events/io_uring.h> | |
85 | ||
2b188cc1 JA |
86 | #include <uapi/linux/io_uring.h> |
87 | ||
88 | #include "internal.h" | |
561fb04a | 89 | #include "io-wq.h" |
2b188cc1 | 90 | |
5277deaa | 91 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 92 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
93 | |
94 | /* | |
95 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
96 | */ | |
97 | #define IORING_FILE_TABLE_SHIFT 9 | |
98 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
99 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
100 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
101 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
102 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 103 | |
b16fed66 PB |
104 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
105 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
106 | IOSQE_BUFFER_SELECT) | |
107 | ||
2b188cc1 JA |
108 | struct io_uring { |
109 | u32 head ____cacheline_aligned_in_smp; | |
110 | u32 tail ____cacheline_aligned_in_smp; | |
111 | }; | |
112 | ||
1e84b97b | 113 | /* |
75b28aff HV |
114 | * This data is shared with the application through the mmap at offsets |
115 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
116 | * |
117 | * The offsets to the member fields are published through struct | |
118 | * io_sqring_offsets when calling io_uring_setup. | |
119 | */ | |
75b28aff | 120 | struct io_rings { |
1e84b97b SB |
121 | /* |
122 | * Head and tail offsets into the ring; the offsets need to be | |
123 | * masked to get valid indices. | |
124 | * | |
75b28aff HV |
125 | * The kernel controls head of the sq ring and the tail of the cq ring, |
126 | * and the application controls tail of the sq ring and the head of the | |
127 | * cq ring. | |
1e84b97b | 128 | */ |
75b28aff | 129 | struct io_uring sq, cq; |
1e84b97b | 130 | /* |
75b28aff | 131 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
132 | * ring_entries - 1) |
133 | */ | |
75b28aff HV |
134 | u32 sq_ring_mask, cq_ring_mask; |
135 | /* Ring sizes (constant, power of 2) */ | |
136 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
137 | /* |
138 | * Number of invalid entries dropped by the kernel due to | |
139 | * invalid index stored in array | |
140 | * | |
141 | * Written by the kernel, shouldn't be modified by the | |
142 | * application (i.e. get number of "new events" by comparing to | |
143 | * cached value). | |
144 | * | |
145 | * After a new SQ head value was read by the application this | |
146 | * counter includes all submissions that were dropped reaching | |
147 | * the new SQ head (and possibly more). | |
148 | */ | |
75b28aff | 149 | u32 sq_dropped; |
1e84b97b | 150 | /* |
0d9b5b3a | 151 | * Runtime SQ flags |
1e84b97b SB |
152 | * |
153 | * Written by the kernel, shouldn't be modified by the | |
154 | * application. | |
155 | * | |
156 | * The application needs a full memory barrier before checking | |
157 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
158 | */ | |
75b28aff | 159 | u32 sq_flags; |
0d9b5b3a SG |
160 | /* |
161 | * Runtime CQ flags | |
162 | * | |
163 | * Written by the application, shouldn't be modified by the | |
164 | * kernel. | |
165 | */ | |
166 | u32 cq_flags; | |
1e84b97b SB |
167 | /* |
168 | * Number of completion events lost because the queue was full; | |
169 | * this should be avoided by the application by making sure | |
0b4295b5 | 170 | * there are not more requests pending than there is space in |
1e84b97b SB |
171 | * the completion queue. |
172 | * | |
173 | * Written by the kernel, shouldn't be modified by the | |
174 | * application (i.e. get number of "new events" by comparing to | |
175 | * cached value). | |
176 | * | |
177 | * As completion events come in out of order this counter is not | |
178 | * ordered with any other data. | |
179 | */ | |
75b28aff | 180 | u32 cq_overflow; |
1e84b97b SB |
181 | /* |
182 | * Ring buffer of completion events. | |
183 | * | |
184 | * The kernel writes completion events fresh every time they are | |
185 | * produced, so the application is allowed to modify pending | |
186 | * entries. | |
187 | */ | |
75b28aff | 188 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
189 | }; |
190 | ||
45d189c6 PB |
191 | enum io_uring_cmd_flags { |
192 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 193 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
194 | }; |
195 | ||
edafccee JA |
196 | struct io_mapped_ubuf { |
197 | u64 ubuf; | |
198 | size_t len; | |
199 | struct bio_vec *bvec; | |
200 | unsigned int nr_bvecs; | |
de293938 | 201 | unsigned long acct_pages; |
edafccee JA |
202 | }; |
203 | ||
50238531 BM |
204 | struct io_ring_ctx; |
205 | ||
269bbe5f BM |
206 | struct io_rsrc_put { |
207 | struct list_head list; | |
50238531 BM |
208 | union { |
209 | void *rsrc; | |
210 | struct file *file; | |
211 | }; | |
269bbe5f BM |
212 | }; |
213 | ||
214 | struct fixed_rsrc_table { | |
65e19f54 | 215 | struct file **files; |
31b51510 JA |
216 | }; |
217 | ||
269bbe5f | 218 | struct fixed_rsrc_ref_node { |
05589553 XW |
219 | struct percpu_ref refs; |
220 | struct list_head node; | |
269bbe5f BM |
221 | struct list_head rsrc_list; |
222 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
223 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
224 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 225 | struct llist_node llist; |
e297822b | 226 | bool done; |
05589553 XW |
227 | }; |
228 | ||
269bbe5f BM |
229 | struct fixed_rsrc_data { |
230 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
231 | struct io_ring_ctx *ctx; |
232 | ||
269bbe5f | 233 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 234 | struct percpu_ref refs; |
05f3fb3c | 235 | struct completion done; |
8bad28d8 | 236 | bool quiesce; |
05f3fb3c JA |
237 | }; |
238 | ||
5a2e745d JA |
239 | struct io_buffer { |
240 | struct list_head list; | |
241 | __u64 addr; | |
242 | __s32 len; | |
243 | __u16 bid; | |
244 | }; | |
245 | ||
21b55dbc SG |
246 | struct io_restriction { |
247 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
248 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
249 | u8 sqe_flags_allowed; | |
250 | u8 sqe_flags_required; | |
7e84e1c7 | 251 | bool registered; |
21b55dbc SG |
252 | }; |
253 | ||
37d1e2e3 JA |
254 | enum { |
255 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
256 | IO_SQ_THREAD_SHOULD_PARK, | |
257 | }; | |
258 | ||
534ca6d6 JA |
259 | struct io_sq_data { |
260 | refcount_t refs; | |
69fb2131 JA |
261 | struct mutex lock; |
262 | ||
263 | /* ctx's that are using this sqd */ | |
264 | struct list_head ctx_list; | |
265 | struct list_head ctx_new_list; | |
266 | struct mutex ctx_lock; | |
267 | ||
534ca6d6 JA |
268 | struct task_struct *thread; |
269 | struct wait_queue_head wait; | |
08369246 XW |
270 | |
271 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
272 | int sq_cpu; |
273 | pid_t task_pid; | |
274 | ||
275 | unsigned long state; | |
276 | struct completion startup; | |
86e0d676 | 277 | struct completion parked; |
37d1e2e3 | 278 | struct completion exited; |
534ca6d6 JA |
279 | }; |
280 | ||
258b29a9 | 281 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 282 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 283 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 284 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
285 | |
286 | struct io_comp_state { | |
6dd0be1e | 287 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 288 | unsigned int nr; |
c7dae4ba JA |
289 | unsigned int locked_free_nr; |
290 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 291 | struct list_head free_list; |
c7dae4ba JA |
292 | /* IRQ completion list, under ->completion_lock */ |
293 | struct list_head locked_free_list; | |
258b29a9 PB |
294 | }; |
295 | ||
a1ab7b35 PB |
296 | struct io_submit_link { |
297 | struct io_kiocb *head; | |
298 | struct io_kiocb *last; | |
299 | }; | |
300 | ||
258b29a9 PB |
301 | struct io_submit_state { |
302 | struct blk_plug plug; | |
a1ab7b35 | 303 | struct io_submit_link link; |
258b29a9 PB |
304 | |
305 | /* | |
306 | * io_kiocb alloc cache | |
307 | */ | |
bf019da7 | 308 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
309 | unsigned int free_reqs; |
310 | ||
311 | bool plug_started; | |
312 | ||
313 | /* | |
314 | * Batch completion logic | |
315 | */ | |
316 | struct io_comp_state comp; | |
317 | ||
318 | /* | |
319 | * File reference cache | |
320 | */ | |
321 | struct file *file; | |
322 | unsigned int fd; | |
323 | unsigned int file_refs; | |
324 | unsigned int ios_left; | |
325 | }; | |
326 | ||
2b188cc1 JA |
327 | struct io_ring_ctx { |
328 | struct { | |
329 | struct percpu_ref refs; | |
330 | } ____cacheline_aligned_in_smp; | |
331 | ||
332 | struct { | |
333 | unsigned int flags; | |
e1d85334 | 334 | unsigned int compat: 1; |
e1d85334 RD |
335 | unsigned int cq_overflow_flushed: 1; |
336 | unsigned int drain_next: 1; | |
337 | unsigned int eventfd_async: 1; | |
21b55dbc | 338 | unsigned int restricted: 1; |
5f3f26f9 | 339 | unsigned int sqo_exec: 1; |
2b188cc1 | 340 | |
75b28aff HV |
341 | /* |
342 | * Ring buffer of indices into array of io_uring_sqe, which is | |
343 | * mmapped by the application using the IORING_OFF_SQES offset. | |
344 | * | |
345 | * This indirection could e.g. be used to assign fixed | |
346 | * io_uring_sqe entries to operations and only submit them to | |
347 | * the queue when needed. | |
348 | * | |
349 | * The kernel modifies neither the indices array nor the entries | |
350 | * array. | |
351 | */ | |
352 | u32 *sq_array; | |
2b188cc1 JA |
353 | unsigned cached_sq_head; |
354 | unsigned sq_entries; | |
355 | unsigned sq_mask; | |
6c271ce2 | 356 | unsigned sq_thread_idle; |
498ccd9e | 357 | unsigned cached_sq_dropped; |
2c3bac6d | 358 | unsigned cached_cq_overflow; |
ad3eb2c8 | 359 | unsigned long sq_check_overflow; |
de0617e4 | 360 | |
e941894e JA |
361 | /* hashed buffered write serialization */ |
362 | struct io_wq_hash *hash_map; | |
363 | ||
de0617e4 | 364 | struct list_head defer_list; |
5262f567 | 365 | struct list_head timeout_list; |
1d7bb1d5 | 366 | struct list_head cq_overflow_list; |
fcb323cc | 367 | |
ad3eb2c8 | 368 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
369 | } ____cacheline_aligned_in_smp; |
370 | ||
3c1a2ead JA |
371 | struct { |
372 | struct mutex uring_lock; | |
373 | wait_queue_head_t wait; | |
374 | } ____cacheline_aligned_in_smp; | |
375 | ||
376 | struct io_submit_state submit_state; | |
377 | ||
206aefde JA |
378 | struct io_rings *rings; |
379 | ||
2aede0e4 JA |
380 | /* Only used for accounting purposes */ |
381 | struct mm_struct *mm_account; | |
382 | ||
534ca6d6 JA |
383 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
384 | ||
90554200 | 385 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 386 | struct list_head sqd_list; |
75b28aff | 387 | |
6b06314c JA |
388 | /* |
389 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
390 | * readers must ensure that ->refs is alive as long as the file* is | |
391 | * used. Only updated through io_uring_register(2). | |
392 | */ | |
269bbe5f | 393 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
394 | unsigned nr_user_files; |
395 | ||
edafccee JA |
396 | /* if used, fixed mapped user buffers */ |
397 | unsigned nr_user_bufs; | |
398 | struct io_mapped_ubuf *user_bufs; | |
399 | ||
2b188cc1 JA |
400 | struct user_struct *user; |
401 | ||
0f158b4c JA |
402 | struct completion ref_comp; |
403 | struct completion sq_thread_comp; | |
206aefde JA |
404 | |
405 | #if defined(CONFIG_UNIX) | |
406 | struct socket *ring_sock; | |
407 | #endif | |
408 | ||
5a2e745d JA |
409 | struct idr io_buffer_idr; |
410 | ||
071698e1 JA |
411 | struct idr personality_idr; |
412 | ||
206aefde JA |
413 | struct { |
414 | unsigned cached_cq_tail; | |
415 | unsigned cq_entries; | |
416 | unsigned cq_mask; | |
417 | atomic_t cq_timeouts; | |
f010505b | 418 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 419 | unsigned long cq_check_overflow; |
206aefde JA |
420 | struct wait_queue_head cq_wait; |
421 | struct fasync_struct *cq_fasync; | |
422 | struct eventfd_ctx *cq_ev_fd; | |
423 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 424 | |
2b188cc1 JA |
425 | struct { |
426 | spinlock_t completion_lock; | |
e94f141b | 427 | |
def596e9 | 428 | /* |
540e32a0 | 429 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
430 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
431 | * For SQPOLL, only the single threaded io_sq_thread() will | |
432 | * manipulate the list, hence no extra locking is needed there. | |
433 | */ | |
540e32a0 | 434 | struct list_head iopoll_list; |
78076bb6 JA |
435 | struct hlist_head *cancel_hash; |
436 | unsigned cancel_hash_bits; | |
e94f141b | 437 | bool poll_multi_file; |
31b51510 | 438 | |
fcb323cc JA |
439 | spinlock_t inflight_lock; |
440 | struct list_head inflight_list; | |
2b188cc1 | 441 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 442 | |
269bbe5f BM |
443 | struct delayed_work rsrc_put_work; |
444 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
445 | struct list_head rsrc_ref_list; |
446 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 447 | |
21b55dbc | 448 | struct io_restriction restrictions; |
3c1a2ead | 449 | |
7c25c0d1 JA |
450 | /* exit task_work */ |
451 | struct callback_head *exit_task_work; | |
452 | ||
e941894e JA |
453 | struct wait_queue_head hash_wait; |
454 | ||
3c1a2ead JA |
455 | /* Keep this last, we don't need it for the fast path */ |
456 | struct work_struct exit_work; | |
13bf43f5 | 457 | struct list_head tctx_list; |
2b188cc1 JA |
458 | }; |
459 | ||
09bb8394 JA |
460 | /* |
461 | * First field must be the file pointer in all the | |
462 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
463 | */ | |
221c5eb2 JA |
464 | struct io_poll_iocb { |
465 | struct file *file; | |
018043be | 466 | struct wait_queue_head *head; |
221c5eb2 | 467 | __poll_t events; |
8c838788 | 468 | bool done; |
221c5eb2 | 469 | bool canceled; |
392edb45 | 470 | struct wait_queue_entry wait; |
221c5eb2 JA |
471 | }; |
472 | ||
018043be PB |
473 | struct io_poll_remove { |
474 | struct file *file; | |
475 | u64 addr; | |
476 | }; | |
477 | ||
b5dba59e JA |
478 | struct io_close { |
479 | struct file *file; | |
b5dba59e JA |
480 | int fd; |
481 | }; | |
482 | ||
ad8a48ac JA |
483 | struct io_timeout_data { |
484 | struct io_kiocb *req; | |
485 | struct hrtimer timer; | |
486 | struct timespec64 ts; | |
487 | enum hrtimer_mode mode; | |
488 | }; | |
489 | ||
8ed8d3c3 JA |
490 | struct io_accept { |
491 | struct file *file; | |
492 | struct sockaddr __user *addr; | |
493 | int __user *addr_len; | |
494 | int flags; | |
09952e3e | 495 | unsigned long nofile; |
8ed8d3c3 JA |
496 | }; |
497 | ||
498 | struct io_sync { | |
499 | struct file *file; | |
500 | loff_t len; | |
501 | loff_t off; | |
502 | int flags; | |
d63d1b5e | 503 | int mode; |
8ed8d3c3 JA |
504 | }; |
505 | ||
fbf23849 JA |
506 | struct io_cancel { |
507 | struct file *file; | |
508 | u64 addr; | |
509 | }; | |
510 | ||
b29472ee JA |
511 | struct io_timeout { |
512 | struct file *file; | |
bfe68a22 PB |
513 | u32 off; |
514 | u32 target_seq; | |
135fcde8 | 515 | struct list_head list; |
90cd7e42 PB |
516 | /* head of the link, used by linked timeouts only */ |
517 | struct io_kiocb *head; | |
b29472ee JA |
518 | }; |
519 | ||
0bdf7a2d PB |
520 | struct io_timeout_rem { |
521 | struct file *file; | |
522 | u64 addr; | |
9c8e11b3 PB |
523 | |
524 | /* timeout update */ | |
525 | struct timespec64 ts; | |
526 | u32 flags; | |
0bdf7a2d PB |
527 | }; |
528 | ||
9adbd45d JA |
529 | struct io_rw { |
530 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
531 | struct kiocb kiocb; | |
532 | u64 addr; | |
533 | u64 len; | |
534 | }; | |
535 | ||
3fbb51c1 JA |
536 | struct io_connect { |
537 | struct file *file; | |
538 | struct sockaddr __user *addr; | |
539 | int addr_len; | |
540 | }; | |
541 | ||
e47293fd JA |
542 | struct io_sr_msg { |
543 | struct file *file; | |
fddaface | 544 | union { |
270a5940 | 545 | struct user_msghdr __user *umsg; |
fddaface JA |
546 | void __user *buf; |
547 | }; | |
e47293fd | 548 | int msg_flags; |
bcda7baa | 549 | int bgid; |
fddaface | 550 | size_t len; |
bcda7baa | 551 | struct io_buffer *kbuf; |
e47293fd JA |
552 | }; |
553 | ||
15b71abe JA |
554 | struct io_open { |
555 | struct file *file; | |
556 | int dfd; | |
15b71abe | 557 | struct filename *filename; |
c12cedf2 | 558 | struct open_how how; |
4022e7af | 559 | unsigned long nofile; |
15b71abe JA |
560 | }; |
561 | ||
269bbe5f | 562 | struct io_rsrc_update { |
05f3fb3c JA |
563 | struct file *file; |
564 | u64 arg; | |
565 | u32 nr_args; | |
566 | u32 offset; | |
567 | }; | |
568 | ||
4840e418 JA |
569 | struct io_fadvise { |
570 | struct file *file; | |
571 | u64 offset; | |
572 | u32 len; | |
573 | u32 advice; | |
574 | }; | |
575 | ||
c1ca757b JA |
576 | struct io_madvise { |
577 | struct file *file; | |
578 | u64 addr; | |
579 | u32 len; | |
580 | u32 advice; | |
581 | }; | |
582 | ||
3e4827b0 JA |
583 | struct io_epoll { |
584 | struct file *file; | |
585 | int epfd; | |
586 | int op; | |
587 | int fd; | |
588 | struct epoll_event event; | |
e47293fd JA |
589 | }; |
590 | ||
7d67af2c PB |
591 | struct io_splice { |
592 | struct file *file_out; | |
593 | struct file *file_in; | |
594 | loff_t off_out; | |
595 | loff_t off_in; | |
596 | u64 len; | |
597 | unsigned int flags; | |
598 | }; | |
599 | ||
ddf0322d JA |
600 | struct io_provide_buf { |
601 | struct file *file; | |
602 | __u64 addr; | |
603 | __s32 len; | |
604 | __u32 bgid; | |
605 | __u16 nbufs; | |
606 | __u16 bid; | |
607 | }; | |
608 | ||
1d9e1288 BM |
609 | struct io_statx { |
610 | struct file *file; | |
611 | int dfd; | |
612 | unsigned int mask; | |
613 | unsigned int flags; | |
e62753e4 | 614 | const char __user *filename; |
1d9e1288 BM |
615 | struct statx __user *buffer; |
616 | }; | |
617 | ||
36f4fa68 JA |
618 | struct io_shutdown { |
619 | struct file *file; | |
620 | int how; | |
621 | }; | |
622 | ||
80a261fd JA |
623 | struct io_rename { |
624 | struct file *file; | |
625 | int old_dfd; | |
626 | int new_dfd; | |
627 | struct filename *oldpath; | |
628 | struct filename *newpath; | |
629 | int flags; | |
630 | }; | |
631 | ||
14a1143b JA |
632 | struct io_unlink { |
633 | struct file *file; | |
634 | int dfd; | |
635 | int flags; | |
636 | struct filename *filename; | |
637 | }; | |
638 | ||
3ca405eb PB |
639 | struct io_completion { |
640 | struct file *file; | |
641 | struct list_head list; | |
0f7e466b | 642 | int cflags; |
3ca405eb PB |
643 | }; |
644 | ||
f499a021 JA |
645 | struct io_async_connect { |
646 | struct sockaddr_storage address; | |
647 | }; | |
648 | ||
03b1230c JA |
649 | struct io_async_msghdr { |
650 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
651 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
652 | struct iovec *free_iov; | |
03b1230c JA |
653 | struct sockaddr __user *uaddr; |
654 | struct msghdr msg; | |
b537916c | 655 | struct sockaddr_storage addr; |
03b1230c JA |
656 | }; |
657 | ||
f67676d1 JA |
658 | struct io_async_rw { |
659 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
660 | const struct iovec *free_iovec; |
661 | struct iov_iter iter; | |
227c0c96 | 662 | size_t bytes_done; |
bcf5a063 | 663 | struct wait_page_queue wpq; |
f67676d1 JA |
664 | }; |
665 | ||
6b47ee6e PB |
666 | enum { |
667 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
668 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
669 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
670 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
671 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 672 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 673 | |
6b47ee6e PB |
674 | REQ_F_FAIL_LINK_BIT, |
675 | REQ_F_INFLIGHT_BIT, | |
676 | REQ_F_CUR_POS_BIT, | |
677 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 678 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 679 | REQ_F_ISREG_BIT, |
99bc4c38 | 680 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 681 | REQ_F_POLLED_BIT, |
bcda7baa | 682 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 683 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 684 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 685 | REQ_F_COMPLETE_INLINE_BIT, |
84557871 JA |
686 | |
687 | /* not a real bit, just to check we're not overflowing the space */ | |
688 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
689 | }; |
690 | ||
691 | enum { | |
692 | /* ctx owns file */ | |
693 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
694 | /* drain existing IO first */ | |
695 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
696 | /* linked sqes */ | |
697 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
698 | /* doesn't sever on completion < 0 */ | |
699 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
700 | /* IOSQE_ASYNC */ | |
701 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
702 | /* IOSQE_BUFFER_SELECT */ |
703 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 704 | |
6b47ee6e PB |
705 | /* fail rest of links */ |
706 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 707 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
708 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
709 | /* read/write uses file position */ | |
710 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
711 | /* must not punt to workers */ | |
712 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 713 | /* has or had linked timeout */ |
6b47ee6e | 714 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
715 | /* regular file */ |
716 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
717 | /* needs cleanup */ |
718 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
719 | /* already went through poll handler */ |
720 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
721 | /* buffer already selected */ |
722 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
723 | /* doesn't need file table for this request */ |
724 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
725 | /* linked timeout is active, i.e. prepared by link's head */ |
726 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
727 | /* completion is deferred through io_comp_state */ |
728 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
d7718a9d JA |
729 | }; |
730 | ||
731 | struct async_poll { | |
732 | struct io_poll_iocb poll; | |
807abcb0 | 733 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
734 | }; |
735 | ||
7cbf1722 JA |
736 | struct io_task_work { |
737 | struct io_wq_work_node node; | |
738 | task_work_func_t func; | |
739 | }; | |
740 | ||
09bb8394 JA |
741 | /* |
742 | * NOTE! Each of the iocb union members has the file pointer | |
743 | * as the first entry in their struct definition. So you can | |
744 | * access the file pointer through any of the sub-structs, | |
745 | * or directly as just 'ki_filp' in this struct. | |
746 | */ | |
2b188cc1 | 747 | struct io_kiocb { |
221c5eb2 | 748 | union { |
09bb8394 | 749 | struct file *file; |
9adbd45d | 750 | struct io_rw rw; |
221c5eb2 | 751 | struct io_poll_iocb poll; |
018043be | 752 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
753 | struct io_accept accept; |
754 | struct io_sync sync; | |
fbf23849 | 755 | struct io_cancel cancel; |
b29472ee | 756 | struct io_timeout timeout; |
0bdf7a2d | 757 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 758 | struct io_connect connect; |
e47293fd | 759 | struct io_sr_msg sr_msg; |
15b71abe | 760 | struct io_open open; |
b5dba59e | 761 | struct io_close close; |
269bbe5f | 762 | struct io_rsrc_update rsrc_update; |
4840e418 | 763 | struct io_fadvise fadvise; |
c1ca757b | 764 | struct io_madvise madvise; |
3e4827b0 | 765 | struct io_epoll epoll; |
7d67af2c | 766 | struct io_splice splice; |
ddf0322d | 767 | struct io_provide_buf pbuf; |
1d9e1288 | 768 | struct io_statx statx; |
36f4fa68 | 769 | struct io_shutdown shutdown; |
80a261fd | 770 | struct io_rename rename; |
14a1143b | 771 | struct io_unlink unlink; |
3ca405eb PB |
772 | /* use only after cleaning per-op data, see io_clean_op() */ |
773 | struct io_completion compl; | |
221c5eb2 | 774 | }; |
2b188cc1 | 775 | |
e8c2bc1f JA |
776 | /* opcode allocated if it needs to store data for async defer */ |
777 | void *async_data; | |
d625c6ee | 778 | u8 opcode; |
65a6543d XW |
779 | /* polled IO has completed */ |
780 | u8 iopoll_completed; | |
2b188cc1 | 781 | |
4f4eeba8 | 782 | u16 buf_index; |
9cf7c104 | 783 | u32 result; |
4f4eeba8 | 784 | |
010e8e6b PB |
785 | struct io_ring_ctx *ctx; |
786 | unsigned int flags; | |
787 | refcount_t refs; | |
788 | struct task_struct *task; | |
789 | u64 user_data; | |
d7718a9d | 790 | |
f2f87370 | 791 | struct io_kiocb *link; |
269bbe5f | 792 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 793 | |
d21ffe7e PB |
794 | /* |
795 | * 1. used with ctx->iopoll_list with reads/writes | |
796 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
797 | */ | |
010e8e6b | 798 | struct list_head inflight_entry; |
7cbf1722 JA |
799 | union { |
800 | struct io_task_work io_task_work; | |
801 | struct callback_head task_work; | |
802 | }; | |
010e8e6b PB |
803 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
804 | struct hlist_node hash_node; | |
805 | struct async_poll *apoll; | |
806 | struct io_wq_work work; | |
2b188cc1 | 807 | }; |
05589553 | 808 | |
13bf43f5 PB |
809 | struct io_tctx_node { |
810 | struct list_head ctx_node; | |
811 | struct task_struct *task; | |
13bf43f5 PB |
812 | struct io_ring_ctx *ctx; |
813 | }; | |
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; | |
d3656344 JA |
841 | }; |
842 | ||
0918682b | 843 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
844 | [IORING_OP_NOP] = {}, |
845 | [IORING_OP_READV] = { | |
d3656344 JA |
846 | .needs_file = 1, |
847 | .unbound_nonreg_file = 1, | |
8a72758c | 848 | .pollin = 1, |
4d954c25 | 849 | .buffer_select = 1, |
e8c2bc1f | 850 | .needs_async_data = 1, |
27926b68 | 851 | .plug = 1, |
e8c2bc1f | 852 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 853 | }, |
0463b6c5 | 854 | [IORING_OP_WRITEV] = { |
d3656344 JA |
855 | .needs_file = 1, |
856 | .hash_reg_file = 1, | |
857 | .unbound_nonreg_file = 1, | |
8a72758c | 858 | .pollout = 1, |
e8c2bc1f | 859 | .needs_async_data = 1, |
27926b68 | 860 | .plug = 1, |
e8c2bc1f | 861 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 862 | }, |
0463b6c5 | 863 | [IORING_OP_FSYNC] = { |
d3656344 JA |
864 | .needs_file = 1, |
865 | }, | |
0463b6c5 | 866 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
867 | .needs_file = 1, |
868 | .unbound_nonreg_file = 1, | |
8a72758c | 869 | .pollin = 1, |
27926b68 | 870 | .plug = 1, |
e8c2bc1f | 871 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 872 | }, |
0463b6c5 | 873 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
874 | .needs_file = 1, |
875 | .hash_reg_file = 1, | |
876 | .unbound_nonreg_file = 1, | |
8a72758c | 877 | .pollout = 1, |
27926b68 | 878 | .plug = 1, |
e8c2bc1f | 879 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 880 | }, |
0463b6c5 | 881 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
882 | .needs_file = 1, |
883 | .unbound_nonreg_file = 1, | |
884 | }, | |
0463b6c5 PB |
885 | [IORING_OP_POLL_REMOVE] = {}, |
886 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
887 | .needs_file = 1, |
888 | }, | |
0463b6c5 | 889 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
890 | .needs_file = 1, |
891 | .unbound_nonreg_file = 1, | |
8a72758c | 892 | .pollout = 1, |
e8c2bc1f JA |
893 | .needs_async_data = 1, |
894 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 895 | }, |
0463b6c5 | 896 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
897 | .needs_file = 1, |
898 | .unbound_nonreg_file = 1, | |
8a72758c | 899 | .pollin = 1, |
52de1fe1 | 900 | .buffer_select = 1, |
e8c2bc1f JA |
901 | .needs_async_data = 1, |
902 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 903 | }, |
0463b6c5 | 904 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
905 | .needs_async_data = 1, |
906 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 907 | }, |
9c8e11b3 PB |
908 | [IORING_OP_TIMEOUT_REMOVE] = { |
909 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 910 | }, |
0463b6c5 | 911 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
912 | .needs_file = 1, |
913 | .unbound_nonreg_file = 1, | |
8a72758c | 914 | .pollin = 1, |
d3656344 | 915 | }, |
0463b6c5 PB |
916 | [IORING_OP_ASYNC_CANCEL] = {}, |
917 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
918 | .needs_async_data = 1, |
919 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 920 | }, |
0463b6c5 | 921 | [IORING_OP_CONNECT] = { |
d3656344 JA |
922 | .needs_file = 1, |
923 | .unbound_nonreg_file = 1, | |
8a72758c | 924 | .pollout = 1, |
e8c2bc1f JA |
925 | .needs_async_data = 1, |
926 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 927 | }, |
0463b6c5 | 928 | [IORING_OP_FALLOCATE] = { |
d3656344 | 929 | .needs_file = 1, |
d3656344 | 930 | }, |
44526bed JA |
931 | [IORING_OP_OPENAT] = {}, |
932 | [IORING_OP_CLOSE] = {}, | |
933 | [IORING_OP_FILES_UPDATE] = {}, | |
934 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 935 | [IORING_OP_READ] = { |
3a6820f2 JA |
936 | .needs_file = 1, |
937 | .unbound_nonreg_file = 1, | |
8a72758c | 938 | .pollin = 1, |
bcda7baa | 939 | .buffer_select = 1, |
27926b68 | 940 | .plug = 1, |
e8c2bc1f | 941 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
944 | .needs_file = 1, |
945 | .unbound_nonreg_file = 1, | |
8a72758c | 946 | .pollout = 1, |
27926b68 | 947 | .plug = 1, |
e8c2bc1f | 948 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 949 | }, |
0463b6c5 | 950 | [IORING_OP_FADVISE] = { |
4840e418 | 951 | .needs_file = 1, |
c1ca757b | 952 | }, |
44526bed | 953 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 954 | [IORING_OP_SEND] = { |
fddaface JA |
955 | .needs_file = 1, |
956 | .unbound_nonreg_file = 1, | |
8a72758c | 957 | .pollout = 1, |
fddaface | 958 | }, |
0463b6c5 | 959 | [IORING_OP_RECV] = { |
fddaface JA |
960 | .needs_file = 1, |
961 | .unbound_nonreg_file = 1, | |
8a72758c | 962 | .pollin = 1, |
bcda7baa | 963 | .buffer_select = 1, |
fddaface | 964 | }, |
0463b6c5 | 965 | [IORING_OP_OPENAT2] = { |
cebdb986 | 966 | }, |
3e4827b0 JA |
967 | [IORING_OP_EPOLL_CTL] = { |
968 | .unbound_nonreg_file = 1, | |
3e4827b0 | 969 | }, |
7d67af2c PB |
970 | [IORING_OP_SPLICE] = { |
971 | .needs_file = 1, | |
972 | .hash_reg_file = 1, | |
973 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
974 | }, |
975 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 976 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
977 | [IORING_OP_TEE] = { |
978 | .needs_file = 1, | |
979 | .hash_reg_file = 1, | |
980 | .unbound_nonreg_file = 1, | |
981 | }, | |
36f4fa68 JA |
982 | [IORING_OP_SHUTDOWN] = { |
983 | .needs_file = 1, | |
984 | }, | |
44526bed JA |
985 | [IORING_OP_RENAMEAT] = {}, |
986 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
987 | }; |
988 | ||
d56d938b | 989 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
990 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
991 | struct task_struct *task, | |
992 | struct files_struct *files); | |
37d1e2e3 | 993 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 994 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 995 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 996 | struct io_ring_ctx *ctx); |
f2303b1f | 997 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 998 | |
23faba36 | 999 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1000 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1001 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1002 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1003 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1004 | static void io_dismantle_req(struct io_kiocb *req); |
1005 | static void io_put_task(struct task_struct *task, int nr); | |
1006 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1007 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1008 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1009 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1010 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1011 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1012 | unsigned nr_args); |
3ca405eb | 1013 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1014 | static struct file *io_file_get(struct io_submit_state *state, |
1015 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1016 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1017 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1018 | |
847595de PB |
1019 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1020 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1021 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1022 | const struct iovec *fast_iov, | |
227c0c96 | 1023 | struct iov_iter *iter, bool force); |
907d1df3 | 1024 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1025 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1026 | struct io_ring_ctx *ctx); | |
de0617e4 | 1027 | |
2b188cc1 JA |
1028 | static struct kmem_cache *req_cachep; |
1029 | ||
0918682b | 1030 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1031 | |
1032 | struct sock *io_uring_get_socket(struct file *file) | |
1033 | { | |
1034 | #if defined(CONFIG_UNIX) | |
1035 | if (file->f_op == &io_uring_fops) { | |
1036 | struct io_ring_ctx *ctx = file->private_data; | |
1037 | ||
1038 | return ctx->ring_sock->sk; | |
1039 | } | |
1040 | #endif | |
1041 | return NULL; | |
1042 | } | |
1043 | EXPORT_SYMBOL(io_uring_get_socket); | |
1044 | ||
f2f87370 PB |
1045 | #define io_for_each_link(pos, head) \ |
1046 | for (pos = (head); pos; pos = pos->link) | |
1047 | ||
3ca405eb PB |
1048 | static inline void io_clean_op(struct io_kiocb *req) |
1049 | { | |
9d5c8190 | 1050 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1051 | __io_clean_op(req); |
1052 | } | |
1053 | ||
36f72fe2 PB |
1054 | static inline void io_set_resource_node(struct io_kiocb *req) |
1055 | { | |
1056 | struct io_ring_ctx *ctx = req->ctx; | |
1057 | ||
269bbe5f BM |
1058 | if (!req->fixed_rsrc_refs) { |
1059 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1060 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1061 | } |
1062 | } | |
1063 | ||
08d23634 PB |
1064 | static bool io_match_task(struct io_kiocb *head, |
1065 | struct task_struct *task, | |
1066 | struct files_struct *files) | |
1067 | { | |
1068 | struct io_kiocb *req; | |
1069 | ||
84965ff8 JA |
1070 | if (task && head->task != task) { |
1071 | /* in terms of cancelation, always match if req task is dead */ | |
1072 | if (head->task->flags & PF_EXITING) | |
1073 | return true; | |
08d23634 | 1074 | return false; |
84965ff8 | 1075 | } |
08d23634 PB |
1076 | if (!files) |
1077 | return true; | |
1078 | ||
1079 | io_for_each_link(req, head) { | |
b05a1bcd | 1080 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1081 | return true; |
4379bf8b | 1082 | if (req->task->files == files) |
08d23634 PB |
1083 | return true; |
1084 | } | |
1085 | return false; | |
1086 | } | |
1087 | ||
c40f6379 JA |
1088 | static inline void req_set_fail_links(struct io_kiocb *req) |
1089 | { | |
1090 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1091 | req->flags |= REQ_F_FAIL_LINK; | |
1092 | } | |
4a38aed2 | 1093 | |
2b188cc1 JA |
1094 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1095 | { | |
1096 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1097 | ||
0f158b4c | 1098 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1099 | } |
1100 | ||
8eb7e2d0 PB |
1101 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1102 | { | |
1103 | return !req->timeout.off; | |
1104 | } | |
1105 | ||
2b188cc1 JA |
1106 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1107 | { | |
1108 | struct io_ring_ctx *ctx; | |
78076bb6 | 1109 | int hash_bits; |
2b188cc1 JA |
1110 | |
1111 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1112 | if (!ctx) | |
1113 | return NULL; | |
1114 | ||
78076bb6 JA |
1115 | /* |
1116 | * Use 5 bits less than the max cq entries, that should give us around | |
1117 | * 32 entries per hash list if totally full and uniformly spread. | |
1118 | */ | |
1119 | hash_bits = ilog2(p->cq_entries); | |
1120 | hash_bits -= 5; | |
1121 | if (hash_bits <= 0) | |
1122 | hash_bits = 1; | |
1123 | ctx->cancel_hash_bits = hash_bits; | |
1124 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1125 | GFP_KERNEL); | |
1126 | if (!ctx->cancel_hash) | |
1127 | goto err; | |
1128 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1129 | ||
21482896 | 1130 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1131 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1132 | goto err; | |
2b188cc1 JA |
1133 | |
1134 | ctx->flags = p->flags; | |
90554200 | 1135 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1136 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1137 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1138 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c JA |
1139 | init_completion(&ctx->ref_comp); |
1140 | init_completion(&ctx->sq_thread_comp); | |
5a2e745d | 1141 | idr_init(&ctx->io_buffer_idr); |
071698e1 | 1142 | idr_init(&ctx->personality_idr); |
2b188cc1 JA |
1143 | mutex_init(&ctx->uring_lock); |
1144 | init_waitqueue_head(&ctx->wait); | |
1145 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1146 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1147 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1148 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1149 | spin_lock_init(&ctx->inflight_lock); |
1150 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1151 | spin_lock_init(&ctx->rsrc_ref_lock); |
1152 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1153 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1154 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1155 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1156 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1157 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1158 | return ctx; |
206aefde | 1159 | err: |
78076bb6 | 1160 | kfree(ctx->cancel_hash); |
206aefde JA |
1161 | kfree(ctx); |
1162 | return NULL; | |
2b188cc1 JA |
1163 | } |
1164 | ||
9cf7c104 | 1165 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1166 | { |
2bc9930e JA |
1167 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1168 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1169 | |
9cf7c104 | 1170 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1171 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1172 | } |
de0617e4 | 1173 | |
9d858b21 | 1174 | return false; |
de0617e4 JA |
1175 | } |
1176 | ||
ce3d5aae PB |
1177 | static void io_req_track_inflight(struct io_kiocb *req) |
1178 | { | |
1179 | struct io_ring_ctx *ctx = req->ctx; | |
1180 | ||
1181 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1182 | req->flags |= REQ_F_INFLIGHT; |
1183 | ||
1184 | spin_lock_irq(&ctx->inflight_lock); | |
1185 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1186 | spin_unlock_irq(&ctx->inflight_lock); | |
1187 | } | |
1188 | } | |
1189 | ||
1e6fa521 JA |
1190 | static void io_prep_async_work(struct io_kiocb *req) |
1191 | { | |
1192 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1193 | struct io_ring_ctx *ctx = req->ctx; |
1194 | ||
003e8dcc JA |
1195 | if (!req->work.creds) |
1196 | req->work.creds = get_current_cred(); | |
1197 | ||
feaadc4f PB |
1198 | if (req->flags & REQ_F_FORCE_ASYNC) |
1199 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1200 | ||
1e6fa521 JA |
1201 | if (req->flags & REQ_F_ISREG) { |
1202 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1203 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
1204 | } else { | |
1205 | if (def->unbound_nonreg_file) | |
1206 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1207 | } | |
561fb04a | 1208 | } |
cccf0ee8 | 1209 | |
cbdcb435 | 1210 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1211 | { |
cbdcb435 | 1212 | struct io_kiocb *cur; |
54a91f3b | 1213 | |
f2f87370 PB |
1214 | io_for_each_link(cur, req) |
1215 | io_prep_async_work(cur); | |
561fb04a JA |
1216 | } |
1217 | ||
ebf93667 | 1218 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1219 | { |
a197f664 | 1220 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1221 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1222 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1223 | |
3bfe6106 JA |
1224 | BUG_ON(!tctx); |
1225 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1226 | |
8766dd51 PB |
1227 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1228 | &req->work, req->flags); | |
cbdcb435 PB |
1229 | /* init ->work of the whole link before punting */ |
1230 | io_prep_async_link(req); | |
ebf93667 | 1231 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1232 | if (link) |
1233 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1234 | } |
1235 | ||
5262f567 JA |
1236 | static void io_kill_timeout(struct io_kiocb *req) |
1237 | { | |
e8c2bc1f | 1238 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1239 | int ret; |
1240 | ||
e8c2bc1f | 1241 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1242 | if (ret != -1) { |
01cec8c1 PB |
1243 | atomic_set(&req->ctx->cq_timeouts, |
1244 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1245 | list_del_init(&req->timeout.list); |
78e19bbe | 1246 | io_cqring_fill_event(req, 0); |
216578e5 | 1247 | io_put_req_deferred(req, 1); |
5262f567 JA |
1248 | } |
1249 | } | |
1250 | ||
76e1b642 JA |
1251 | /* |
1252 | * Returns true if we found and killed one or more timeouts | |
1253 | */ | |
6b81928d PB |
1254 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, |
1255 | struct files_struct *files) | |
5262f567 JA |
1256 | { |
1257 | struct io_kiocb *req, *tmp; | |
76e1b642 | 1258 | int canceled = 0; |
5262f567 JA |
1259 | |
1260 | spin_lock_irq(&ctx->completion_lock); | |
f3606e3a | 1261 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { |
6b81928d | 1262 | if (io_match_task(req, tsk, files)) { |
f3606e3a | 1263 | io_kill_timeout(req); |
76e1b642 JA |
1264 | canceled++; |
1265 | } | |
f3606e3a | 1266 | } |
5262f567 | 1267 | spin_unlock_irq(&ctx->completion_lock); |
76e1b642 | 1268 | return canceled != 0; |
5262f567 JA |
1269 | } |
1270 | ||
04518945 | 1271 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1272 | { |
04518945 | 1273 | do { |
27dc8338 PB |
1274 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1275 | struct io_defer_entry, list); | |
de0617e4 | 1276 | |
9cf7c104 | 1277 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1278 | break; |
27dc8338 | 1279 | list_del_init(&de->list); |
907d1df3 | 1280 | io_req_task_queue(de->req); |
27dc8338 | 1281 | kfree(de); |
04518945 PB |
1282 | } while (!list_empty(&ctx->defer_list)); |
1283 | } | |
1284 | ||
360428f8 | 1285 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1286 | { |
f010505b MDG |
1287 | u32 seq; |
1288 | ||
1289 | if (list_empty(&ctx->timeout_list)) | |
1290 | return; | |
1291 | ||
1292 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1293 | ||
1294 | do { | |
1295 | u32 events_needed, events_got; | |
360428f8 | 1296 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1297 | struct io_kiocb, timeout.list); |
de0617e4 | 1298 | |
8eb7e2d0 | 1299 | if (io_is_timeout_noseq(req)) |
360428f8 | 1300 | break; |
f010505b MDG |
1301 | |
1302 | /* | |
1303 | * Since seq can easily wrap around over time, subtract | |
1304 | * the last seq at which timeouts were flushed before comparing. | |
1305 | * Assuming not more than 2^31-1 events have happened since, | |
1306 | * these subtractions won't have wrapped, so we can check if | |
1307 | * target is in [last_seq, current_seq] by comparing the two. | |
1308 | */ | |
1309 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1310 | events_got = seq - ctx->cq_last_tm_flush; | |
1311 | if (events_got < events_needed) | |
360428f8 | 1312 | break; |
bfe68a22 | 1313 | |
135fcde8 | 1314 | list_del_init(&req->timeout.list); |
5262f567 | 1315 | io_kill_timeout(req); |
f010505b MDG |
1316 | } while (!list_empty(&ctx->timeout_list)); |
1317 | ||
1318 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1319 | } |
5262f567 | 1320 | |
360428f8 PB |
1321 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1322 | { | |
1323 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1324 | |
1325 | /* order cqe stores with ring update */ | |
1326 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1327 | |
04518945 PB |
1328 | if (unlikely(!list_empty(&ctx->defer_list))) |
1329 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1330 | } |
1331 | ||
90554200 JA |
1332 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1333 | { | |
1334 | struct io_rings *r = ctx->rings; | |
1335 | ||
1336 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1337 | } | |
1338 | ||
888aae2e PB |
1339 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1340 | { | |
1341 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1342 | } | |
1343 | ||
2b188cc1 JA |
1344 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1345 | { | |
75b28aff | 1346 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1347 | unsigned tail; |
1348 | ||
115e12e5 SB |
1349 | /* |
1350 | * writes to the cq entry need to come after reading head; the | |
1351 | * control dependency is enough as we're using WRITE_ONCE to | |
1352 | * fill the cq entry | |
1353 | */ | |
888aae2e | 1354 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1355 | return NULL; |
1356 | ||
888aae2e | 1357 | tail = ctx->cached_cq_tail++; |
75b28aff | 1358 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1359 | } |
1360 | ||
f2842ab5 JA |
1361 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1362 | { | |
f0b493e6 JA |
1363 | if (!ctx->cq_ev_fd) |
1364 | return false; | |
7e55a19c SG |
1365 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1366 | return false; | |
f2842ab5 JA |
1367 | if (!ctx->eventfd_async) |
1368 | return true; | |
b41e9852 | 1369 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1370 | } |
1371 | ||
b41e9852 | 1372 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1373 | { |
b1445e59 PB |
1374 | /* see waitqueue_active() comment */ |
1375 | smp_mb(); | |
1376 | ||
1d7bb1d5 JA |
1377 | if (waitqueue_active(&ctx->wait)) |
1378 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1379 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1380 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1381 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1382 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1383 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1384 | wake_up_interruptible(&ctx->cq_wait); |
1385 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1386 | } | |
1d7bb1d5 JA |
1387 | } |
1388 | ||
80c18e4a PB |
1389 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1390 | { | |
b1445e59 PB |
1391 | /* see waitqueue_active() comment */ |
1392 | smp_mb(); | |
1393 | ||
80c18e4a PB |
1394 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1395 | if (waitqueue_active(&ctx->wait)) | |
1396 | wake_up(&ctx->wait); | |
1397 | } | |
1398 | if (io_should_trigger_evfd(ctx)) | |
1399 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1400 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1401 | wake_up_interruptible(&ctx->cq_wait); |
1402 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1403 | } | |
80c18e4a PB |
1404 | } |
1405 | ||
c4a2ed72 | 1406 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1407 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1408 | struct task_struct *tsk, | |
1409 | struct files_struct *files) | |
1d7bb1d5 JA |
1410 | { |
1411 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1412 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1413 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1414 | unsigned long flags; |
b18032bb | 1415 | bool all_flushed, posted; |
1d7bb1d5 JA |
1416 | LIST_HEAD(list); |
1417 | ||
e23de15f PB |
1418 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1419 | return false; | |
1d7bb1d5 | 1420 | |
b18032bb | 1421 | posted = false; |
1d7bb1d5 | 1422 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1423 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1424 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1425 | continue; |
1426 | ||
1d7bb1d5 JA |
1427 | cqe = io_get_cqring(ctx); |
1428 | if (!cqe && !force) | |
1429 | break; | |
1430 | ||
40d8ddd4 | 1431 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1432 | if (cqe) { |
1433 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1434 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1435 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1436 | } else { |
2c3bac6d | 1437 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1438 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1439 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1440 | } |
b18032bb | 1441 | posted = true; |
1d7bb1d5 JA |
1442 | } |
1443 | ||
09e88404 PB |
1444 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1445 | if (all_flushed) { | |
1446 | clear_bit(0, &ctx->sq_check_overflow); | |
1447 | clear_bit(0, &ctx->cq_check_overflow); | |
1448 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1449 | } | |
46930143 | 1450 | |
b18032bb JA |
1451 | if (posted) |
1452 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1453 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1454 | if (posted) |
1455 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1456 | |
1457 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1458 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1459 | list_del(&req->compl.list); | |
ec9c02ad | 1460 | io_put_req(req); |
1d7bb1d5 | 1461 | } |
c4a2ed72 | 1462 | |
09e88404 | 1463 | return all_flushed; |
1d7bb1d5 JA |
1464 | } |
1465 | ||
ca0a2651 | 1466 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1467 | struct task_struct *tsk, |
1468 | struct files_struct *files) | |
1469 | { | |
ca0a2651 JA |
1470 | bool ret = true; |
1471 | ||
6c503150 PB |
1472 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1473 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1474 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1475 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1476 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1477 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1478 | mutex_unlock(&ctx->uring_lock); | |
1479 | } | |
ca0a2651 JA |
1480 | |
1481 | return ret; | |
6c503150 PB |
1482 | } |
1483 | ||
bcda7baa | 1484 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1485 | { |
78e19bbe | 1486 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1487 | struct io_uring_cqe *cqe; |
1488 | ||
78e19bbe | 1489 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1490 | |
2b188cc1 JA |
1491 | /* |
1492 | * If we can't get a cq entry, userspace overflowed the | |
1493 | * submission (by quite a lot). Increment the overflow count in | |
1494 | * the ring. | |
1495 | */ | |
1496 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1497 | if (likely(cqe)) { |
78e19bbe | 1498 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1499 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1500 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1501 | } else if (ctx->cq_overflow_flushed || |
1502 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1503 | /* |
1504 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1505 | * then we cannot store the request for later flushing, we need | |
1506 | * to drop it on the floor. | |
1507 | */ | |
2c3bac6d PB |
1508 | ctx->cached_cq_overflow++; |
1509 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1510 | } else { |
ad3eb2c8 JA |
1511 | if (list_empty(&ctx->cq_overflow_list)) { |
1512 | set_bit(0, &ctx->sq_check_overflow); | |
1513 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1514 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1515 | } |
40d8ddd4 | 1516 | io_clean_op(req); |
1d7bb1d5 | 1517 | req->result = res; |
0f7e466b | 1518 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1519 | refcount_inc(&req->refs); |
1520 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1521 | } |
1522 | } | |
1523 | ||
bcda7baa JA |
1524 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1525 | { | |
1526 | __io_cqring_fill_event(req, res, 0); | |
1527 | } | |
1528 | ||
c7dae4ba JA |
1529 | static inline void io_req_complete_post(struct io_kiocb *req, long res, |
1530 | unsigned int cflags) | |
2b188cc1 | 1531 | { |
78e19bbe | 1532 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1533 | unsigned long flags; |
1534 | ||
1535 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1536 | __io_cqring_fill_event(req, res, cflags); |
2b188cc1 | 1537 | io_commit_cqring(ctx); |
c7dae4ba JA |
1538 | /* |
1539 | * If we're the last reference to this request, add to our locked | |
1540 | * free_list cache. | |
1541 | */ | |
1542 | if (refcount_dec_and_test(&req->refs)) { | |
1543 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1544 | ||
1545 | io_dismantle_req(req); | |
1546 | io_put_task(req->task, 1); | |
1547 | list_add(&req->compl.list, &cs->locked_free_list); | |
1548 | cs->locked_free_nr++; | |
1549 | } else | |
1550 | req = NULL; | |
2b188cc1 JA |
1551 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
1552 | ||
8c838788 | 1553 | io_cqring_ev_posted(ctx); |
c7dae4ba JA |
1554 | if (req) { |
1555 | io_queue_next(req); | |
1556 | percpu_ref_put(&ctx->refs); | |
229a7b63 | 1557 | } |
229a7b63 JA |
1558 | } |
1559 | ||
a38d68db | 1560 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1561 | unsigned int cflags) |
229a7b63 | 1562 | { |
a38d68db PB |
1563 | io_clean_op(req); |
1564 | req->result = res; | |
1565 | req->compl.cflags = cflags; | |
e342c807 | 1566 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1567 | } |
1568 | ||
889fca73 PB |
1569 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1570 | long res, unsigned cflags) | |
bcda7baa | 1571 | { |
889fca73 PB |
1572 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1573 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1574 | else |
c7dae4ba | 1575 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1576 | } |
1577 | ||
a38d68db | 1578 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1579 | { |
889fca73 | 1580 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1581 | } |
1582 | ||
c7dae4ba | 1583 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1584 | { |
c7dae4ba JA |
1585 | struct io_submit_state *state = &ctx->submit_state; |
1586 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1587 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1588 | |
c7dae4ba JA |
1589 | /* |
1590 | * If we have more than a batch's worth of requests in our IRQ side | |
1591 | * locked cache, grab the lock and move them over to our submission | |
1592 | * side cache. | |
1593 | */ | |
1594 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1595 | spin_lock_irq(&ctx->completion_lock); | |
1596 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1597 | cs->locked_free_nr = 0; | |
1598 | spin_unlock_irq(&ctx->completion_lock); | |
1599 | } | |
0ddf92e8 | 1600 | |
c7dae4ba JA |
1601 | while (!list_empty(&cs->free_list)) { |
1602 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1603 | compl.list); |
1604 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1605 | state->reqs[state->free_reqs++] = req; |
1606 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1607 | break; | |
1b4c351f JA |
1608 | } |
1609 | ||
e5d1bc0a | 1610 | return req != NULL; |
0ddf92e8 JA |
1611 | } |
1612 | ||
e5d1bc0a | 1613 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1614 | { |
e5d1bc0a PB |
1615 | struct io_submit_state *state = &ctx->submit_state; |
1616 | ||
1617 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1618 | ||
f6b6c7d6 | 1619 | if (!state->free_reqs) { |
291b2821 | 1620 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1621 | int ret; |
1622 | ||
c7dae4ba | 1623 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1624 | goto got_req; |
1625 | ||
bf019da7 PB |
1626 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1627 | state->reqs); | |
fd6fab2c JA |
1628 | |
1629 | /* | |
1630 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1631 | * retry single alloc to be on the safe side. | |
1632 | */ | |
1633 | if (unlikely(ret <= 0)) { | |
1634 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1635 | if (!state->reqs[0]) | |
3893f39f | 1636 | return NULL; |
fd6fab2c JA |
1637 | ret = 1; |
1638 | } | |
291b2821 | 1639 | state->free_reqs = ret; |
2b188cc1 | 1640 | } |
e5d1bc0a | 1641 | got_req: |
291b2821 PB |
1642 | state->free_reqs--; |
1643 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1644 | } |
1645 | ||
8da11c19 PB |
1646 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1647 | bool fixed) | |
1648 | { | |
36f72fe2 | 1649 | if (!fixed) |
8da11c19 PB |
1650 | fput(file); |
1651 | } | |
1652 | ||
4edf20f9 | 1653 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1654 | { |
3ca405eb | 1655 | io_clean_op(req); |
929a3af9 | 1656 | |
e8c2bc1f JA |
1657 | if (req->async_data) |
1658 | kfree(req->async_data); | |
8da11c19 PB |
1659 | if (req->file) |
1660 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1661 | if (req->fixed_rsrc_refs) |
1662 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1663 | if (req->work.creds) { |
1664 | put_cred(req->work.creds); | |
1665 | req->work.creds = NULL; | |
1666 | } | |
f85c310a PB |
1667 | |
1668 | if (req->flags & REQ_F_INFLIGHT) { | |
1669 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1670 | unsigned long flags; |
1671 | ||
1672 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1673 | list_del(&req->inflight_entry); | |
1674 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1675 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1676 | } |
e65ef56d JA |
1677 | } |
1678 | ||
b23fcf47 | 1679 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1680 | static inline void io_put_task(struct task_struct *task, int nr) |
1681 | { | |
1682 | struct io_uring_task *tctx = task->io_uring; | |
1683 | ||
1684 | percpu_counter_sub(&tctx->inflight, nr); | |
1685 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1686 | wake_up(&tctx->wait); | |
1687 | put_task_struct_many(task, nr); | |
1688 | } | |
1689 | ||
216578e5 | 1690 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1691 | { |
51a4cc11 | 1692 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1693 | |
216578e5 | 1694 | io_dismantle_req(req); |
7c660731 | 1695 | io_put_task(req->task, 1); |
c6ca97b3 | 1696 | |
3893f39f | 1697 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1698 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1699 | } |
1700 | ||
f2f87370 PB |
1701 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1702 | { | |
1703 | struct io_kiocb *nxt = req->link; | |
1704 | ||
1705 | req->link = nxt->link; | |
1706 | nxt->link = NULL; | |
1707 | } | |
1708 | ||
c9abd7ad | 1709 | static void io_kill_linked_timeout(struct io_kiocb *req) |
2665abfd | 1710 | { |
a197f664 | 1711 | struct io_ring_ctx *ctx = req->ctx; |
7c86ffee | 1712 | struct io_kiocb *link; |
c9abd7ad PB |
1713 | bool cancelled = false; |
1714 | unsigned long flags; | |
7c86ffee | 1715 | |
c9abd7ad | 1716 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1717 | link = req->link; |
1718 | ||
900fad45 PB |
1719 | /* |
1720 | * Can happen if a linked timeout fired and link had been like | |
1721 | * req -> link t-out -> link t-out [-> ...] | |
1722 | */ | |
c9abd7ad PB |
1723 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1724 | struct io_timeout_data *io = link->async_data; | |
1725 | int ret; | |
7c86ffee | 1726 | |
f2f87370 | 1727 | io_remove_next_linked(req); |
90cd7e42 | 1728 | link->timeout.head = NULL; |
c9abd7ad PB |
1729 | ret = hrtimer_try_to_cancel(&io->timer); |
1730 | if (ret != -1) { | |
1731 | io_cqring_fill_event(link, -ECANCELED); | |
1732 | io_commit_cqring(ctx); | |
1733 | cancelled = true; | |
1734 | } | |
1735 | } | |
7c86ffee | 1736 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
216578e5 | 1737 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
ab0b6451 | 1738 | |
c9abd7ad | 1739 | if (cancelled) { |
7c86ffee | 1740 | io_cqring_ev_posted(ctx); |
c9abd7ad PB |
1741 | io_put_req(link); |
1742 | } | |
7c86ffee PB |
1743 | } |
1744 | ||
9e645e11 | 1745 | |
d148ca4b | 1746 | static void io_fail_links(struct io_kiocb *req) |
9e645e11 | 1747 | { |
f2f87370 | 1748 | struct io_kiocb *link, *nxt; |
2665abfd | 1749 | struct io_ring_ctx *ctx = req->ctx; |
d148ca4b | 1750 | unsigned long flags; |
9e645e11 | 1751 | |
d148ca4b | 1752 | spin_lock_irqsave(&ctx->completion_lock, flags); |
f2f87370 PB |
1753 | link = req->link; |
1754 | req->link = NULL; | |
9e645e11 | 1755 | |
f2f87370 PB |
1756 | while (link) { |
1757 | nxt = link->link; | |
1758 | link->link = NULL; | |
2665abfd | 1759 | |
f2f87370 | 1760 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1761 | io_cqring_fill_event(link, -ECANCELED); |
216578e5 | 1762 | |
1575f21a | 1763 | io_put_req_deferred(link, 2); |
f2f87370 | 1764 | link = nxt; |
9e645e11 | 1765 | } |
2665abfd | 1766 | io_commit_cqring(ctx); |
216578e5 | 1767 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
9e645e11 | 1768 | |
2665abfd | 1769 | io_cqring_ev_posted(ctx); |
9e645e11 JA |
1770 | } |
1771 | ||
3fa5e0f3 | 1772 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1773 | { |
7c86ffee PB |
1774 | if (req->flags & REQ_F_LINK_TIMEOUT) |
1775 | io_kill_linked_timeout(req); | |
944e58bf | 1776 | |
9e645e11 JA |
1777 | /* |
1778 | * If LINK is set, we have dependent requests in this chain. If we | |
1779 | * didn't fail this request, queue the first one up, moving any other | |
1780 | * dependencies to the next request. In case of failure, fail the rest | |
1781 | * of the chain. | |
1782 | */ | |
f2f87370 PB |
1783 | if (likely(!(req->flags & REQ_F_FAIL_LINK))) { |
1784 | struct io_kiocb *nxt = req->link; | |
1785 | ||
1786 | req->link = NULL; | |
1787 | return nxt; | |
1788 | } | |
9b5f7bd9 PB |
1789 | io_fail_links(req); |
1790 | return NULL; | |
4d7dd462 | 1791 | } |
9e645e11 | 1792 | |
f2f87370 | 1793 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1794 | { |
cdbff982 | 1795 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1796 | return NULL; |
1797 | return __io_req_find_next(req); | |
1798 | } | |
1799 | ||
2c32395d PB |
1800 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1801 | { | |
1802 | if (!ctx) | |
1803 | return; | |
1804 | if (ctx->submit_state.comp.nr) { | |
1805 | mutex_lock(&ctx->uring_lock); | |
1806 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1807 | mutex_unlock(&ctx->uring_lock); | |
1808 | } | |
1809 | percpu_ref_put(&ctx->refs); | |
1810 | } | |
1811 | ||
7cbf1722 | 1812 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1813 | { |
65453d1e | 1814 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1815 | struct io_wq_work_list list; |
1816 | struct io_wq_work_node *node; | |
c2c4c83c | 1817 | |
7cbf1722 JA |
1818 | if (wq_list_empty(&tctx->task_list)) |
1819 | return false; | |
6200b0ae | 1820 | |
0b81e80c | 1821 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1822 | list = tctx->task_list; |
1823 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1824 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1825 | |
7cbf1722 JA |
1826 | node = list.first; |
1827 | while (node) { | |
1828 | struct io_wq_work_node *next = node->next; | |
1829 | struct io_kiocb *req; | |
0ba9c9ed | 1830 | |
7cbf1722 | 1831 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1832 | if (req->ctx != ctx) { |
1833 | ctx_flush_and_put(ctx); | |
1834 | ctx = req->ctx; | |
1835 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1836 | } |
65453d1e | 1837 | |
2c32395d PB |
1838 | req->task_work.func(&req->task_work); |
1839 | node = next; | |
7cbf1722 JA |
1840 | } |
1841 | ||
2c32395d | 1842 | ctx_flush_and_put(ctx); |
7cbf1722 | 1843 | return list.first != NULL; |
c2c4c83c JA |
1844 | } |
1845 | ||
7cbf1722 | 1846 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1847 | { |
7cbf1722 | 1848 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1849 | |
1d5f360d JA |
1850 | clear_bit(0, &tctx->task_state); |
1851 | ||
7cbf1722 JA |
1852 | while (__tctx_task_work(tctx)) |
1853 | cond_resched(); | |
7cbf1722 JA |
1854 | } |
1855 | ||
1856 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1857 | enum task_work_notify_mode notify) | |
1858 | { | |
1859 | struct io_uring_task *tctx = tsk->io_uring; | |
1860 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1861 | unsigned long flags; |
7cbf1722 JA |
1862 | int ret; |
1863 | ||
1864 | WARN_ON_ONCE(!tctx); | |
1865 | ||
0b81e80c | 1866 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1867 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1868 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1869 | |
1870 | /* task_work already pending, we're done */ | |
1871 | if (test_bit(0, &tctx->task_state) || | |
1872 | test_and_set_bit(0, &tctx->task_state)) | |
1873 | return 0; | |
1874 | ||
1875 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1876 | return 0; | |
1877 | ||
1878 | /* | |
1879 | * Slow path - we failed, find and delete work. if the work is not | |
1880 | * in the list, it got run and we're fine. | |
1881 | */ | |
1882 | ret = 0; | |
0b81e80c | 1883 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1884 | wq_list_for_each(node, prev, &tctx->task_list) { |
1885 | if (&req->io_task_work.node == node) { | |
1886 | wq_list_del(&tctx->task_list, node, prev); | |
1887 | ret = 1; | |
1888 | break; | |
1889 | } | |
1890 | } | |
0b81e80c | 1891 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1892 | clear_bit(0, &tctx->task_state); |
1893 | return ret; | |
1894 | } | |
1895 | ||
355fb9e2 | 1896 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1897 | { |
1898 | struct task_struct *tsk = req->task; | |
1899 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1900 | enum task_work_notify_mode notify; |
1901 | int ret; | |
c2c4c83c | 1902 | |
6200b0ae JA |
1903 | if (tsk->flags & PF_EXITING) |
1904 | return -ESRCH; | |
1905 | ||
c2c4c83c | 1906 | /* |
0ba9c9ed JA |
1907 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1908 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1909 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1910 | * will do the job. | |
c2c4c83c | 1911 | */ |
91989c70 | 1912 | notify = TWA_NONE; |
355fb9e2 | 1913 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1914 | notify = TWA_SIGNAL; |
1915 | ||
7cbf1722 | 1916 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1917 | if (!ret) |
1918 | wake_up_process(tsk); | |
0ba9c9ed | 1919 | |
c2c4c83c JA |
1920 | return ret; |
1921 | } | |
1922 | ||
eab30c4d | 1923 | static void io_req_task_work_add_fallback(struct io_kiocb *req, |
7cbf1722 | 1924 | task_work_func_t cb) |
eab30c4d | 1925 | { |
7c25c0d1 JA |
1926 | struct io_ring_ctx *ctx = req->ctx; |
1927 | struct callback_head *head; | |
eab30c4d PB |
1928 | |
1929 | init_task_work(&req->task_work, cb); | |
7c25c0d1 JA |
1930 | do { |
1931 | head = READ_ONCE(ctx->exit_task_work); | |
1932 | req->task_work.next = head; | |
1933 | } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head); | |
eab30c4d PB |
1934 | } |
1935 | ||
c40f6379 JA |
1936 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1937 | { | |
1938 | struct io_ring_ctx *ctx = req->ctx; | |
1939 | ||
1940 | spin_lock_irq(&ctx->completion_lock); | |
1941 | io_cqring_fill_event(req, error); | |
1942 | io_commit_cqring(ctx); | |
1943 | spin_unlock_irq(&ctx->completion_lock); | |
1944 | ||
1945 | io_cqring_ev_posted(ctx); | |
1946 | req_set_fail_links(req); | |
1947 | io_double_put_req(req); | |
1948 | } | |
1949 | ||
1950 | static void io_req_task_cancel(struct callback_head *cb) | |
1951 | { | |
1952 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1953 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1954 | |
792bb6eb | 1955 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1956 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1957 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1958 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1959 | } |
1960 | ||
1961 | static void __io_req_task_submit(struct io_kiocb *req) | |
1962 | { | |
1963 | struct io_ring_ctx *ctx = req->ctx; | |
1964 | ||
04fc6c80 | 1965 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1966 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 1967 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 1968 | __io_queue_sqe(req); |
81b6d05c | 1969 | else |
c40f6379 | 1970 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 1971 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
1972 | } |
1973 | ||
1974 | static void io_req_task_submit(struct callback_head *cb) | |
1975 | { | |
1976 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
1977 | ||
1978 | __io_req_task_submit(req); | |
1979 | } | |
1980 | ||
1981 | static void io_req_task_queue(struct io_kiocb *req) | |
1982 | { | |
c40f6379 JA |
1983 | int ret; |
1984 | ||
7cbf1722 | 1985 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 1986 | ret = io_req_task_work_add(req); |
c40f6379 | 1987 | if (unlikely(ret)) { |
a3df7698 | 1988 | req->result = -ECANCELED; |
04fc6c80 | 1989 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 1990 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 1991 | } |
c40f6379 JA |
1992 | } |
1993 | ||
a3df7698 PB |
1994 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
1995 | { | |
1996 | percpu_ref_get(&req->ctx->refs); | |
1997 | req->result = ret; | |
1998 | req->task_work.func = io_req_task_cancel; | |
1999 | ||
2000 | if (unlikely(io_req_task_work_add(req))) | |
2001 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2002 | } | |
2003 | ||
f2f87370 | 2004 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2005 | { |
9b5f7bd9 | 2006 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2007 | |
2008 | if (nxt) | |
906a8c3f | 2009 | io_req_task_queue(nxt); |
c69f8dbe JL |
2010 | } |
2011 | ||
c3524383 | 2012 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2013 | { |
c3524383 PB |
2014 | io_queue_next(req); |
2015 | __io_free_req(req); | |
2016 | } | |
8766dd51 | 2017 | |
2d6500d4 | 2018 | struct req_batch { |
5af1d13e PB |
2019 | struct task_struct *task; |
2020 | int task_refs; | |
1b4c351f | 2021 | int ctx_refs; |
2d6500d4 PB |
2022 | }; |
2023 | ||
5af1d13e PB |
2024 | static inline void io_init_req_batch(struct req_batch *rb) |
2025 | { | |
5af1d13e | 2026 | rb->task_refs = 0; |
9ae72463 | 2027 | rb->ctx_refs = 0; |
5af1d13e PB |
2028 | rb->task = NULL; |
2029 | } | |
2030 | ||
2d6500d4 PB |
2031 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2032 | struct req_batch *rb) | |
2033 | { | |
6e833d53 | 2034 | if (rb->task) |
7c660731 | 2035 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2036 | if (rb->ctx_refs) |
2037 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2038 | } |
2039 | ||
6ff119a6 PB |
2040 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2041 | struct io_submit_state *state) | |
2d6500d4 | 2042 | { |
f2f87370 | 2043 | io_queue_next(req); |
2d6500d4 | 2044 | |
e3bc8e9d | 2045 | if (req->task != rb->task) { |
7c660731 PB |
2046 | if (rb->task) |
2047 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2048 | rb->task = req->task; |
2049 | rb->task_refs = 0; | |
5af1d13e | 2050 | } |
e3bc8e9d | 2051 | rb->task_refs++; |
9ae72463 | 2052 | rb->ctx_refs++; |
5af1d13e | 2053 | |
4edf20f9 | 2054 | io_dismantle_req(req); |
bd759045 | 2055 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2056 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2057 | else |
2058 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2059 | } |
2060 | ||
905c172f PB |
2061 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2062 | struct io_ring_ctx *ctx) | |
2063 | { | |
2064 | int i, nr = cs->nr; | |
2065 | struct io_kiocb *req; | |
2066 | struct req_batch rb; | |
2067 | ||
2068 | io_init_req_batch(&rb); | |
2069 | spin_lock_irq(&ctx->completion_lock); | |
2070 | for (i = 0; i < nr; i++) { | |
2071 | req = cs->reqs[i]; | |
2072 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2073 | } | |
2074 | io_commit_cqring(ctx); | |
2075 | spin_unlock_irq(&ctx->completion_lock); | |
2076 | ||
2077 | io_cqring_ev_posted(ctx); | |
2078 | for (i = 0; i < nr; i++) { | |
2079 | req = cs->reqs[i]; | |
2080 | ||
2081 | /* submission and completion refs */ | |
2082 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2083 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2084 | } |
2085 | ||
2086 | io_req_free_batch_finish(ctx, &rb); | |
2087 | cs->nr = 0; | |
7a743e22 PB |
2088 | } |
2089 | ||
ba816ad6 JA |
2090 | /* |
2091 | * Drop reference to request, return next in chain (if there is one) if this | |
2092 | * was the last reference to this request. | |
2093 | */ | |
9b5f7bd9 | 2094 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2095 | { |
9b5f7bd9 PB |
2096 | struct io_kiocb *nxt = NULL; |
2097 | ||
2a44f467 | 2098 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2099 | nxt = io_req_find_next(req); |
4d7dd462 | 2100 | __io_free_req(req); |
2a44f467 | 2101 | } |
9b5f7bd9 | 2102 | return nxt; |
2b188cc1 JA |
2103 | } |
2104 | ||
e65ef56d JA |
2105 | static void io_put_req(struct io_kiocb *req) |
2106 | { | |
2107 | if (refcount_dec_and_test(&req->refs)) | |
2108 | io_free_req(req); | |
2b188cc1 JA |
2109 | } |
2110 | ||
216578e5 PB |
2111 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2112 | { | |
2113 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2114 | ||
2115 | io_free_req(req); | |
2116 | } | |
2117 | ||
2118 | static void io_free_req_deferred(struct io_kiocb *req) | |
2119 | { | |
2120 | int ret; | |
2121 | ||
7cbf1722 | 2122 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2123 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2124 | if (unlikely(ret)) |
2125 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2126 | } |
2127 | ||
2128 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2129 | { | |
2130 | if (refcount_sub_and_test(refs, &req->refs)) | |
2131 | io_free_req_deferred(req); | |
2132 | } | |
2133 | ||
978db57e JA |
2134 | static void io_double_put_req(struct io_kiocb *req) |
2135 | { | |
2136 | /* drop both submit and complete references */ | |
2137 | if (refcount_sub_and_test(2, &req->refs)) | |
2138 | io_free_req(req); | |
2139 | } | |
2140 | ||
6c503150 | 2141 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2142 | { |
2143 | /* See comment at the top of this file */ | |
2144 | smp_rmb(); | |
e23de15f | 2145 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2146 | } |
2147 | ||
fb5ccc98 PB |
2148 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2149 | { | |
2150 | struct io_rings *rings = ctx->rings; | |
2151 | ||
2152 | /* make sure SQ entry isn't read before tail */ | |
2153 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2154 | } | |
2155 | ||
8ff069bf | 2156 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2157 | { |
8ff069bf | 2158 | unsigned int cflags; |
e94f141b | 2159 | |
bcda7baa JA |
2160 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2161 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2162 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2163 | kfree(kbuf); |
2164 | return cflags; | |
e94f141b JA |
2165 | } |
2166 | ||
8ff069bf | 2167 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2168 | { |
4d954c25 | 2169 | struct io_buffer *kbuf; |
bcda7baa | 2170 | |
4d954c25 | 2171 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2172 | return io_put_kbuf(req, kbuf); |
2173 | } | |
2174 | ||
4c6e277c JA |
2175 | static inline bool io_run_task_work(void) |
2176 | { | |
6200b0ae JA |
2177 | /* |
2178 | * Not safe to run on exiting task, and the task_work handling will | |
2179 | * not add work to such a task. | |
2180 | */ | |
2181 | if (unlikely(current->flags & PF_EXITING)) | |
2182 | return false; | |
4c6e277c JA |
2183 | if (current->task_works) { |
2184 | __set_current_state(TASK_RUNNING); | |
2185 | task_work_run(); | |
2186 | return true; | |
2187 | } | |
2188 | ||
2189 | return false; | |
bcda7baa JA |
2190 | } |
2191 | ||
def596e9 JA |
2192 | /* |
2193 | * Find and free completed poll iocbs | |
2194 | */ | |
2195 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2196 | struct list_head *done) | |
2197 | { | |
8237e045 | 2198 | struct req_batch rb; |
def596e9 | 2199 | struct io_kiocb *req; |
bbde017a XW |
2200 | |
2201 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2202 | smp_rmb(); | |
def596e9 | 2203 | |
5af1d13e | 2204 | io_init_req_batch(&rb); |
def596e9 | 2205 | while (!list_empty(done)) { |
bcda7baa JA |
2206 | int cflags = 0; |
2207 | ||
d21ffe7e | 2208 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2209 | list_del(&req->inflight_entry); |
2210 | ||
bbde017a XW |
2211 | if (READ_ONCE(req->result) == -EAGAIN) { |
2212 | req->iopoll_completed = 0; | |
23faba36 | 2213 | if (io_rw_reissue(req)) |
f161340d | 2214 | continue; |
bbde017a | 2215 | } |
def596e9 | 2216 | |
bcda7baa | 2217 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2218 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2219 | |
2220 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2221 | (*nr_events)++; |
2222 | ||
c3524383 | 2223 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2224 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2225 | } |
def596e9 | 2226 | |
09bb8394 | 2227 | io_commit_cqring(ctx); |
80c18e4a | 2228 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2229 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2230 | } |
2231 | ||
def596e9 JA |
2232 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2233 | long min) | |
2234 | { | |
2235 | struct io_kiocb *req, *tmp; | |
2236 | LIST_HEAD(done); | |
2237 | bool spin; | |
2238 | int ret; | |
2239 | ||
2240 | /* | |
2241 | * Only spin for completions if we don't have multiple devices hanging | |
2242 | * off our complete list, and we're under the requested amount. | |
2243 | */ | |
2244 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2245 | ||
2246 | ret = 0; | |
d21ffe7e | 2247 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2248 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2249 | |
2250 | /* | |
581f9810 BM |
2251 | * Move completed and retryable entries to our local lists. |
2252 | * If we find a request that requires polling, break out | |
2253 | * and complete those lists first, if we have entries there. | |
def596e9 | 2254 | */ |
65a6543d | 2255 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2256 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2257 | continue; |
2258 | } | |
2259 | if (!list_empty(&done)) | |
2260 | break; | |
2261 | ||
2262 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2263 | if (ret < 0) | |
2264 | break; | |
2265 | ||
3aadc23e PB |
2266 | /* iopoll may have completed current req */ |
2267 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2268 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2269 | |
def596e9 JA |
2270 | if (ret && spin) |
2271 | spin = false; | |
2272 | ret = 0; | |
2273 | } | |
2274 | ||
2275 | if (!list_empty(&done)) | |
2276 | io_iopoll_complete(ctx, nr_events, &done); | |
2277 | ||
2278 | return ret; | |
2279 | } | |
2280 | ||
2281 | /* | |
d195a66e | 2282 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2283 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2284 | * as a non-spinning completion check. | |
2285 | */ | |
2286 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2287 | long min) | |
2288 | { | |
540e32a0 | 2289 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2290 | int ret; |
2291 | ||
2292 | ret = io_do_iopoll(ctx, nr_events, min); | |
2293 | if (ret < 0) | |
2294 | return ret; | |
eba0a4dd | 2295 | if (*nr_events >= min) |
def596e9 JA |
2296 | return 0; |
2297 | } | |
2298 | ||
2299 | return 1; | |
2300 | } | |
2301 | ||
2302 | /* | |
2303 | * We can't just wait for polled events to come to us, we have to actively | |
2304 | * find and complete them. | |
2305 | */ | |
b2edc0a7 | 2306 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2307 | { |
2308 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2309 | return; | |
2310 | ||
2311 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2312 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2313 | unsigned int nr_events = 0; |
2314 | ||
b2edc0a7 | 2315 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2316 | |
b2edc0a7 PB |
2317 | /* let it sleep and repeat later if can't complete a request */ |
2318 | if (nr_events == 0) | |
2319 | break; | |
08f5439f JA |
2320 | /* |
2321 | * Ensure we allow local-to-the-cpu processing to take place, | |
2322 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2323 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2324 | */ |
3fcee5a6 PB |
2325 | if (need_resched()) { |
2326 | mutex_unlock(&ctx->uring_lock); | |
2327 | cond_resched(); | |
2328 | mutex_lock(&ctx->uring_lock); | |
2329 | } | |
def596e9 JA |
2330 | } |
2331 | mutex_unlock(&ctx->uring_lock); | |
2332 | } | |
2333 | ||
7668b92a | 2334 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2335 | { |
7668b92a | 2336 | unsigned int nr_events = 0; |
2b2ed975 | 2337 | int iters = 0, ret = 0; |
500f9fba | 2338 | |
c7849be9 XW |
2339 | /* |
2340 | * We disallow the app entering submit/complete with polling, but we | |
2341 | * still need to lock the ring to prevent racing with polled issue | |
2342 | * that got punted to a workqueue. | |
2343 | */ | |
2344 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2345 | do { |
a3a0e43f JA |
2346 | /* |
2347 | * Don't enter poll loop if we already have events pending. | |
2348 | * If we do, we can potentially be spinning for commands that | |
2349 | * already triggered a CQE (eg in error). | |
2350 | */ | |
6c503150 PB |
2351 | if (test_bit(0, &ctx->cq_check_overflow)) |
2352 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2353 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2354 | break; |
2355 | ||
500f9fba JA |
2356 | /* |
2357 | * If a submit got punted to a workqueue, we can have the | |
2358 | * application entering polling for a command before it gets | |
2359 | * issued. That app will hold the uring_lock for the duration | |
2360 | * of the poll right here, so we need to take a breather every | |
2361 | * now and then to ensure that the issue has a chance to add | |
2362 | * the poll to the issued list. Otherwise we can spin here | |
2363 | * forever, while the workqueue is stuck trying to acquire the | |
2364 | * very same mutex. | |
2365 | */ | |
2366 | if (!(++iters & 7)) { | |
2367 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2368 | io_run_task_work(); |
500f9fba JA |
2369 | mutex_lock(&ctx->uring_lock); |
2370 | } | |
2371 | ||
7668b92a | 2372 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2373 | if (ret <= 0) |
2374 | break; | |
2375 | ret = 0; | |
7668b92a | 2376 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2377 | |
500f9fba | 2378 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2379 | return ret; |
2380 | } | |
2381 | ||
491381ce | 2382 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2383 | { |
491381ce JA |
2384 | /* |
2385 | * Tell lockdep we inherited freeze protection from submission | |
2386 | * thread. | |
2387 | */ | |
2388 | if (req->flags & REQ_F_ISREG) { | |
2389 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2390 | |
491381ce | 2391 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2392 | } |
491381ce | 2393 | file_end_write(req->file); |
2b188cc1 JA |
2394 | } |
2395 | ||
b63534c4 | 2396 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2397 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2398 | { |
2399 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2400 | int rw, ret; |
b63534c4 | 2401 | struct iov_iter iter; |
b63534c4 | 2402 | |
dc2a6e9a PB |
2403 | /* already prepared */ |
2404 | if (req->async_data) | |
2405 | return true; | |
b63534c4 JA |
2406 | |
2407 | switch (req->opcode) { | |
2408 | case IORING_OP_READV: | |
2409 | case IORING_OP_READ_FIXED: | |
2410 | case IORING_OP_READ: | |
2411 | rw = READ; | |
2412 | break; | |
2413 | case IORING_OP_WRITEV: | |
2414 | case IORING_OP_WRITE_FIXED: | |
2415 | case IORING_OP_WRITE: | |
2416 | rw = WRITE; | |
2417 | break; | |
2418 | default: | |
2419 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2420 | req->opcode); | |
dc2a6e9a | 2421 | return false; |
b63534c4 JA |
2422 | } |
2423 | ||
dc2a6e9a PB |
2424 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2425 | if (ret < 0) | |
2426 | return false; | |
6bf985dc | 2427 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2428 | } |
b63534c4 | 2429 | |
3e6a0d3c | 2430 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2431 | { |
355afaeb | 2432 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2433 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2434 | |
355afaeb JA |
2435 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2436 | return false; | |
3e6a0d3c JA |
2437 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2438 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2439 | return false; |
7c977a58 JA |
2440 | /* |
2441 | * If ref is dying, we might be running poll reap from the exit work. | |
2442 | * Don't attempt to reissue from that path, just let it fail with | |
2443 | * -EAGAIN. | |
2444 | */ | |
3e6a0d3c JA |
2445 | if (percpu_ref_is_dying(&ctx->refs)) |
2446 | return false; | |
2447 | return true; | |
2448 | } | |
2449 | #endif | |
2450 | ||
2451 | static bool io_rw_reissue(struct io_kiocb *req) | |
2452 | { | |
2453 | #ifdef CONFIG_BLOCK | |
2454 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2455 | return false; |
b63534c4 | 2456 | |
55e6ac1e PB |
2457 | lockdep_assert_held(&req->ctx->uring_lock); |
2458 | ||
37d1e2e3 | 2459 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2460 | refcount_inc(&req->refs); |
2461 | io_queue_async_work(req); | |
b63534c4 | 2462 | return true; |
fdee946d | 2463 | } |
dc2a6e9a | 2464 | req_set_fail_links(req); |
b63534c4 JA |
2465 | #endif |
2466 | return false; | |
2467 | } | |
2468 | ||
a1d7c393 | 2469 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2470 | unsigned int issue_flags) |
a1d7c393 | 2471 | { |
2f8e45f1 PB |
2472 | int cflags = 0; |
2473 | ||
23faba36 PB |
2474 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req)) |
2475 | return; | |
2f8e45f1 PB |
2476 | if (res != req->result) |
2477 | req_set_fail_links(req); | |
23faba36 | 2478 | |
2f8e45f1 PB |
2479 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2480 | kiocb_end_write(req); | |
2481 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2482 | cflags = io_put_rw_kbuf(req); | |
2483 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2484 | } |
2485 | ||
2486 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2487 | { | |
9adbd45d | 2488 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2489 | |
889fca73 | 2490 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2491 | } |
2492 | ||
def596e9 JA |
2493 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2494 | { | |
9adbd45d | 2495 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2496 | |
3e6a0d3c JA |
2497 | #ifdef CONFIG_BLOCK |
2498 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2499 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2500 | struct io_async_rw *rw = req->async_data; | |
2501 | ||
2502 | if (rw) | |
2503 | iov_iter_revert(&rw->iter, | |
2504 | req->result - iov_iter_count(&rw->iter)); | |
2505 | else if (!io_resubmit_prep(req)) | |
2506 | res = -EIO; | |
2507 | } | |
2508 | #endif | |
2509 | ||
491381ce JA |
2510 | if (kiocb->ki_flags & IOCB_WRITE) |
2511 | kiocb_end_write(req); | |
def596e9 | 2512 | |
2d7d6792 | 2513 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2514 | req_set_fail_links(req); |
bbde017a XW |
2515 | |
2516 | WRITE_ONCE(req->result, res); | |
2517 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2518 | smp_wmb(); |
2519 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2524 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2525 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2526 | * accessing the kiocb cookie. | |
2527 | */ | |
2e9dbe90 | 2528 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2529 | { |
2530 | struct io_ring_ctx *ctx = req->ctx; | |
2531 | ||
2532 | /* | |
2533 | * Track whether we have multiple files in our lists. This will impact | |
2534 | * how we do polling eventually, not spinning if we're on potentially | |
2535 | * different devices. | |
2536 | */ | |
540e32a0 | 2537 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2538 | ctx->poll_multi_file = false; |
2539 | } else if (!ctx->poll_multi_file) { | |
2540 | struct io_kiocb *list_req; | |
2541 | ||
540e32a0 | 2542 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2543 | inflight_entry); |
9adbd45d | 2544 | if (list_req->file != req->file) |
def596e9 JA |
2545 | ctx->poll_multi_file = true; |
2546 | } | |
2547 | ||
2548 | /* | |
2549 | * For fast devices, IO may have already completed. If it has, add | |
2550 | * it to the front so we find it first. | |
2551 | */ | |
65a6543d | 2552 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2553 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2554 | else |
d21ffe7e | 2555 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2556 | |
2e9dbe90 XW |
2557 | /* |
2558 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2559 | * task context or in io worker task context. If current task context is | |
2560 | * sq thread, we don't need to check whether should wake up sq thread. | |
2561 | */ | |
2562 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2563 | wq_has_sleeper(&ctx->sq_data->wait)) |
2564 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2565 | } |
2566 | ||
9f13c35b PB |
2567 | static inline void io_state_file_put(struct io_submit_state *state) |
2568 | { | |
02b23a9a PB |
2569 | if (state->file_refs) { |
2570 | fput_many(state->file, state->file_refs); | |
2571 | state->file_refs = 0; | |
2572 | } | |
9a56a232 JA |
2573 | } |
2574 | ||
2575 | /* | |
2576 | * Get as many references to a file as we have IOs left in this submission, | |
2577 | * assuming most submissions are for one file, or at least that each file | |
2578 | * has more than one submission. | |
2579 | */ | |
8da11c19 | 2580 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2581 | { |
2582 | if (!state) | |
2583 | return fget(fd); | |
2584 | ||
6e1271e6 | 2585 | if (state->file_refs) { |
9a56a232 | 2586 | if (state->fd == fd) { |
6e1271e6 | 2587 | state->file_refs--; |
9a56a232 JA |
2588 | return state->file; |
2589 | } | |
02b23a9a | 2590 | io_state_file_put(state); |
9a56a232 JA |
2591 | } |
2592 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2593 | if (unlikely(!state->file)) |
9a56a232 JA |
2594 | return NULL; |
2595 | ||
2596 | state->fd = fd; | |
6e1271e6 | 2597 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2598 | return state->file; |
2599 | } | |
2600 | ||
4503b767 JA |
2601 | static bool io_bdev_nowait(struct block_device *bdev) |
2602 | { | |
9ba0d0c8 | 2603 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2604 | } |
2605 | ||
2b188cc1 JA |
2606 | /* |
2607 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2608 | * any file. For now, just ensure that anything potentially problematic is done | |
2609 | * inline. | |
2610 | */ | |
af197f50 | 2611 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2612 | { |
2613 | umode_t mode = file_inode(file)->i_mode; | |
2614 | ||
4503b767 | 2615 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2616 | if (IS_ENABLED(CONFIG_BLOCK) && |
2617 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2618 | return true; |
2619 | return false; | |
2620 | } | |
2621 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2622 | return true; |
4503b767 | 2623 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2624 | if (IS_ENABLED(CONFIG_BLOCK) && |
2625 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2626 | file->f_op != &io_uring_fops) |
2627 | return true; | |
2628 | return false; | |
2629 | } | |
2b188cc1 | 2630 | |
c5b85625 JA |
2631 | /* any ->read/write should understand O_NONBLOCK */ |
2632 | if (file->f_flags & O_NONBLOCK) | |
2633 | return true; | |
2634 | ||
af197f50 JA |
2635 | if (!(file->f_mode & FMODE_NOWAIT)) |
2636 | return false; | |
2637 | ||
2638 | if (rw == READ) | |
2639 | return file->f_op->read_iter != NULL; | |
2640 | ||
2641 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2642 | } |
2643 | ||
a88fc400 | 2644 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2645 | { |
def596e9 | 2646 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2647 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2648 | struct file *file = req->file; |
09bb8394 JA |
2649 | unsigned ioprio; |
2650 | int ret; | |
2b188cc1 | 2651 | |
75c668cd | 2652 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2653 | req->flags |= REQ_F_ISREG; |
2654 | ||
2b188cc1 | 2655 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2656 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2657 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2658 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2659 | } |
2b188cc1 | 2660 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2661 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2662 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2663 | if (unlikely(ret)) | |
2664 | return ret; | |
2b188cc1 | 2665 | |
75c668cd PB |
2666 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2667 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2668 | req->flags |= REQ_F_NOWAIT; | |
2669 | ||
2b188cc1 JA |
2670 | ioprio = READ_ONCE(sqe->ioprio); |
2671 | if (ioprio) { | |
2672 | ret = ioprio_check_cap(ioprio); | |
2673 | if (ret) | |
09bb8394 | 2674 | return ret; |
2b188cc1 JA |
2675 | |
2676 | kiocb->ki_ioprio = ioprio; | |
2677 | } else | |
2678 | kiocb->ki_ioprio = get_current_ioprio(); | |
2679 | ||
def596e9 | 2680 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2681 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2682 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2683 | return -EOPNOTSUPP; |
2b188cc1 | 2684 | |
def596e9 JA |
2685 | kiocb->ki_flags |= IOCB_HIPRI; |
2686 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2687 | req->iopoll_completed = 0; |
def596e9 | 2688 | } else { |
09bb8394 JA |
2689 | if (kiocb->ki_flags & IOCB_HIPRI) |
2690 | return -EINVAL; | |
def596e9 JA |
2691 | kiocb->ki_complete = io_complete_rw; |
2692 | } | |
9adbd45d | 2693 | |
3529d8c2 JA |
2694 | req->rw.addr = READ_ONCE(sqe->addr); |
2695 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2696 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2697 | return 0; |
2b188cc1 JA |
2698 | } |
2699 | ||
2700 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2701 | { | |
2702 | switch (ret) { | |
2703 | case -EIOCBQUEUED: | |
2704 | break; | |
2705 | case -ERESTARTSYS: | |
2706 | case -ERESTARTNOINTR: | |
2707 | case -ERESTARTNOHAND: | |
2708 | case -ERESTART_RESTARTBLOCK: | |
2709 | /* | |
2710 | * We can't just restart the syscall, since previously | |
2711 | * submitted sqes may already be in progress. Just fail this | |
2712 | * IO with EINTR. | |
2713 | */ | |
2714 | ret = -EINTR; | |
df561f66 | 2715 | fallthrough; |
2b188cc1 JA |
2716 | default: |
2717 | kiocb->ki_complete(kiocb, ret, 0); | |
2718 | } | |
2719 | } | |
2720 | ||
a1d7c393 | 2721 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2722 | unsigned int issue_flags) |
ba816ad6 | 2723 | { |
ba04291e | 2724 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2725 | struct io_async_rw *io = req->async_data; |
ba04291e | 2726 | |
227c0c96 | 2727 | /* add previously done IO, if any */ |
e8c2bc1f | 2728 | if (io && io->bytes_done > 0) { |
227c0c96 | 2729 | if (ret < 0) |
e8c2bc1f | 2730 | ret = io->bytes_done; |
227c0c96 | 2731 | else |
e8c2bc1f | 2732 | ret += io->bytes_done; |
227c0c96 JA |
2733 | } |
2734 | ||
ba04291e JA |
2735 | if (req->flags & REQ_F_CUR_POS) |
2736 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2737 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2738 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2739 | else |
2740 | io_rw_done(kiocb, ret); | |
2741 | } | |
2742 | ||
847595de | 2743 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2744 | { |
9adbd45d JA |
2745 | struct io_ring_ctx *ctx = req->ctx; |
2746 | size_t len = req->rw.len; | |
edafccee | 2747 | struct io_mapped_ubuf *imu; |
4be1c615 | 2748 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2749 | size_t offset; |
2750 | u64 buf_addr; | |
2751 | ||
edafccee JA |
2752 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2753 | return -EFAULT; | |
edafccee JA |
2754 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2755 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2756 | buf_addr = req->rw.addr; |
edafccee JA |
2757 | |
2758 | /* overflow */ | |
2759 | if (buf_addr + len < buf_addr) | |
2760 | return -EFAULT; | |
2761 | /* not inside the mapped region */ | |
2762 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2763 | return -EFAULT; | |
2764 | ||
2765 | /* | |
2766 | * May not be a start of buffer, set size appropriately | |
2767 | * and advance us to the beginning. | |
2768 | */ | |
2769 | offset = buf_addr - imu->ubuf; | |
2770 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2771 | |
2772 | if (offset) { | |
2773 | /* | |
2774 | * Don't use iov_iter_advance() here, as it's really slow for | |
2775 | * using the latter parts of a big fixed buffer - it iterates | |
2776 | * over each segment manually. We can cheat a bit here, because | |
2777 | * we know that: | |
2778 | * | |
2779 | * 1) it's a BVEC iter, we set it up | |
2780 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2781 | * first and last bvec | |
2782 | * | |
2783 | * So just find our index, and adjust the iterator afterwards. | |
2784 | * If the offset is within the first bvec (or the whole first | |
2785 | * bvec, just use iov_iter_advance(). This makes it easier | |
2786 | * since we can just skip the first segment, which may not | |
2787 | * be PAGE_SIZE aligned. | |
2788 | */ | |
2789 | const struct bio_vec *bvec = imu->bvec; | |
2790 | ||
2791 | if (offset <= bvec->bv_len) { | |
2792 | iov_iter_advance(iter, offset); | |
2793 | } else { | |
2794 | unsigned long seg_skip; | |
2795 | ||
2796 | /* skip first vec */ | |
2797 | offset -= bvec->bv_len; | |
2798 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2799 | ||
2800 | iter->bvec = bvec + seg_skip; | |
2801 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2802 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2803 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2804 | } |
2805 | } | |
2806 | ||
847595de | 2807 | return 0; |
edafccee JA |
2808 | } |
2809 | ||
bcda7baa JA |
2810 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2811 | { | |
2812 | if (needs_lock) | |
2813 | mutex_unlock(&ctx->uring_lock); | |
2814 | } | |
2815 | ||
2816 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2817 | { | |
2818 | /* | |
2819 | * "Normal" inline submissions always hold the uring_lock, since we | |
2820 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2821 | * The only exception is when we've detached the request and issue it | |
2822 | * from an async worker thread, grab the lock for that case. | |
2823 | */ | |
2824 | if (needs_lock) | |
2825 | mutex_lock(&ctx->uring_lock); | |
2826 | } | |
2827 | ||
2828 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2829 | int bgid, struct io_buffer *kbuf, | |
2830 | bool needs_lock) | |
2831 | { | |
2832 | struct io_buffer *head; | |
2833 | ||
2834 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2835 | return kbuf; | |
2836 | ||
2837 | io_ring_submit_lock(req->ctx, needs_lock); | |
2838 | ||
2839 | lockdep_assert_held(&req->ctx->uring_lock); | |
2840 | ||
2841 | head = idr_find(&req->ctx->io_buffer_idr, bgid); | |
2842 | if (head) { | |
2843 | if (!list_empty(&head->list)) { | |
2844 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2845 | list); | |
2846 | list_del(&kbuf->list); | |
2847 | } else { | |
2848 | kbuf = head; | |
2849 | idr_remove(&req->ctx->io_buffer_idr, bgid); | |
2850 | } | |
2851 | if (*len > kbuf->len) | |
2852 | *len = kbuf->len; | |
2853 | } else { | |
2854 | kbuf = ERR_PTR(-ENOBUFS); | |
2855 | } | |
2856 | ||
2857 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2858 | ||
2859 | return kbuf; | |
2860 | } | |
2861 | ||
4d954c25 JA |
2862 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2863 | bool needs_lock) | |
2864 | { | |
2865 | struct io_buffer *kbuf; | |
4f4eeba8 | 2866 | u16 bgid; |
4d954c25 JA |
2867 | |
2868 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2869 | bgid = req->buf_index; |
4d954c25 JA |
2870 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2871 | if (IS_ERR(kbuf)) | |
2872 | return kbuf; | |
2873 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2874 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2875 | return u64_to_user_ptr(kbuf->addr); | |
2876 | } | |
2877 | ||
2878 | #ifdef CONFIG_COMPAT | |
2879 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2880 | bool needs_lock) | |
2881 | { | |
2882 | struct compat_iovec __user *uiov; | |
2883 | compat_ssize_t clen; | |
2884 | void __user *buf; | |
2885 | ssize_t len; | |
2886 | ||
2887 | uiov = u64_to_user_ptr(req->rw.addr); | |
2888 | if (!access_ok(uiov, sizeof(*uiov))) | |
2889 | return -EFAULT; | |
2890 | if (__get_user(clen, &uiov->iov_len)) | |
2891 | return -EFAULT; | |
2892 | if (clen < 0) | |
2893 | return -EINVAL; | |
2894 | ||
2895 | len = clen; | |
2896 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2897 | if (IS_ERR(buf)) | |
2898 | return PTR_ERR(buf); | |
2899 | iov[0].iov_base = buf; | |
2900 | iov[0].iov_len = (compat_size_t) len; | |
2901 | return 0; | |
2902 | } | |
2903 | #endif | |
2904 | ||
2905 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2906 | bool needs_lock) | |
2907 | { | |
2908 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2909 | void __user *buf; | |
2910 | ssize_t len; | |
2911 | ||
2912 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2913 | return -EFAULT; | |
2914 | ||
2915 | len = iov[0].iov_len; | |
2916 | if (len < 0) | |
2917 | return -EINVAL; | |
2918 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2919 | if (IS_ERR(buf)) | |
2920 | return PTR_ERR(buf); | |
2921 | iov[0].iov_base = buf; | |
2922 | iov[0].iov_len = len; | |
2923 | return 0; | |
2924 | } | |
2925 | ||
2926 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2927 | bool needs_lock) | |
2928 | { | |
dddb3e26 JA |
2929 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2930 | struct io_buffer *kbuf; | |
2931 | ||
2932 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2933 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2934 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2935 | return 0; |
dddb3e26 | 2936 | } |
dd201662 | 2937 | if (req->rw.len != 1) |
4d954c25 JA |
2938 | return -EINVAL; |
2939 | ||
2940 | #ifdef CONFIG_COMPAT | |
2941 | if (req->ctx->compat) | |
2942 | return io_compat_import(req, iov, needs_lock); | |
2943 | #endif | |
2944 | ||
2945 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2946 | } | |
2947 | ||
847595de PB |
2948 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2949 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2950 | { |
9adbd45d JA |
2951 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2952 | size_t sqe_len = req->rw.len; | |
847595de | 2953 | u8 opcode = req->opcode; |
4d954c25 | 2954 | ssize_t ret; |
edafccee | 2955 | |
7d009165 | 2956 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2957 | *iovec = NULL; |
9adbd45d | 2958 | return io_import_fixed(req, rw, iter); |
edafccee | 2959 | } |
2b188cc1 | 2960 | |
bcda7baa | 2961 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2962 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2963 | return -EINVAL; |
2964 | ||
3a6820f2 | 2965 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2966 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2967 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2968 | if (IS_ERR(buf)) |
4d954c25 | 2969 | return PTR_ERR(buf); |
3f9d6441 | 2970 | req->rw.len = sqe_len; |
bcda7baa JA |
2971 | } |
2972 | ||
3a6820f2 JA |
2973 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2974 | *iovec = NULL; | |
10fc72e4 | 2975 | return ret; |
3a6820f2 JA |
2976 | } |
2977 | ||
4d954c25 JA |
2978 | if (req->flags & REQ_F_BUFFER_SELECT) { |
2979 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
2980 | if (!ret) |
2981 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
2982 | *iovec = NULL; |
2983 | return ret; | |
2984 | } | |
2985 | ||
89cd35c5 CH |
2986 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
2987 | req->ctx->compat); | |
2b188cc1 JA |
2988 | } |
2989 | ||
0fef9483 JA |
2990 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
2991 | { | |
5b09e37e | 2992 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
2993 | } |
2994 | ||
31b51510 | 2995 | /* |
32960613 JA |
2996 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
2997 | * by looping over ->read() or ->write() manually. | |
31b51510 | 2998 | */ |
4017eb91 | 2999 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3000 | { |
4017eb91 JA |
3001 | struct kiocb *kiocb = &req->rw.kiocb; |
3002 | struct file *file = req->file; | |
32960613 JA |
3003 | ssize_t ret = 0; |
3004 | ||
3005 | /* | |
3006 | * Don't support polled IO through this interface, and we can't | |
3007 | * support non-blocking either. For the latter, this just causes | |
3008 | * the kiocb to be handled from an async context. | |
3009 | */ | |
3010 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3011 | return -EOPNOTSUPP; | |
3012 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3013 | return -EAGAIN; | |
3014 | ||
3015 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3016 | struct iovec iovec; |
32960613 JA |
3017 | ssize_t nr; |
3018 | ||
311ae9e1 PB |
3019 | if (!iov_iter_is_bvec(iter)) { |
3020 | iovec = iov_iter_iovec(iter); | |
3021 | } else { | |
4017eb91 JA |
3022 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3023 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3024 | } |
3025 | ||
32960613 JA |
3026 | if (rw == READ) { |
3027 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3028 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3029 | } else { |
3030 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3031 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3032 | } |
3033 | ||
3034 | if (nr < 0) { | |
3035 | if (!ret) | |
3036 | ret = nr; | |
3037 | break; | |
3038 | } | |
3039 | ret += nr; | |
3040 | if (nr != iovec.iov_len) | |
3041 | break; | |
4017eb91 JA |
3042 | req->rw.len -= nr; |
3043 | req->rw.addr += nr; | |
32960613 JA |
3044 | iov_iter_advance(iter, nr); |
3045 | } | |
3046 | ||
3047 | return ret; | |
3048 | } | |
3049 | ||
ff6165b2 JA |
3050 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3051 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3052 | { |
e8c2bc1f | 3053 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3054 | |
ff6165b2 | 3055 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3056 | rw->free_iovec = iovec; |
227c0c96 | 3057 | rw->bytes_done = 0; |
ff6165b2 | 3058 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3059 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3060 | return; |
b64e3444 | 3061 | if (!iovec) { |
ff6165b2 JA |
3062 | unsigned iov_off = 0; |
3063 | ||
3064 | rw->iter.iov = rw->fast_iov; | |
3065 | if (iter->iov != fast_iov) { | |
3066 | iov_off = iter->iov - fast_iov; | |
3067 | rw->iter.iov += iov_off; | |
3068 | } | |
3069 | if (rw->fast_iov != fast_iov) | |
3070 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3071 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3072 | } else { |
3073 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3074 | } |
3075 | } | |
3076 | ||
e8c2bc1f | 3077 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3078 | { |
e8c2bc1f JA |
3079 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3080 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3081 | return req->async_data == NULL; | |
3d9932a8 XW |
3082 | } |
3083 | ||
e8c2bc1f | 3084 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3085 | { |
e8c2bc1f | 3086 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3087 | return 0; |
3d9932a8 | 3088 | |
e8c2bc1f | 3089 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3090 | } |
3091 | ||
ff6165b2 JA |
3092 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3093 | const struct iovec *fast_iov, | |
227c0c96 | 3094 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3095 | { |
e8c2bc1f | 3096 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3097 | return 0; |
e8c2bc1f | 3098 | if (!req->async_data) { |
6bf985dc PB |
3099 | if (__io_alloc_async_data(req)) { |
3100 | kfree(iovec); | |
5d204bcf | 3101 | return -ENOMEM; |
6bf985dc | 3102 | } |
b7bb4f7d | 3103 | |
ff6165b2 | 3104 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3105 | } |
b7bb4f7d | 3106 | return 0; |
f67676d1 JA |
3107 | } |
3108 | ||
73debe68 | 3109 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3110 | { |
e8c2bc1f | 3111 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3112 | struct iovec *iov = iorw->fast_iov; |
847595de | 3113 | int ret; |
c3e330a4 | 3114 | |
2846c481 | 3115 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3116 | if (unlikely(ret < 0)) |
3117 | return ret; | |
3118 | ||
ab0b196c PB |
3119 | iorw->bytes_done = 0; |
3120 | iorw->free_iovec = iov; | |
3121 | if (iov) | |
3122 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3123 | return 0; |
3124 | } | |
3125 | ||
73debe68 | 3126 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3127 | { |
3529d8c2 JA |
3128 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3129 | return -EBADF; | |
93642ef8 | 3130 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3131 | } |
3132 | ||
c1dd91d1 JA |
3133 | /* |
3134 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3135 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3136 | * This gets called when the page is unlocked, and we generally expect that to | |
3137 | * happen when the page IO is completed and the page is now uptodate. This will | |
3138 | * queue a task_work based retry of the operation, attempting to copy the data | |
3139 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3140 | * do a thread based blocking retry of the operation. That's the unexpected | |
3141 | * slow path. | |
3142 | */ | |
bcf5a063 JA |
3143 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3144 | int sync, void *arg) | |
3145 | { | |
3146 | struct wait_page_queue *wpq; | |
3147 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3148 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3149 | |
3150 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3151 | ||
cdc8fcb4 LT |
3152 | if (!wake_page_match(wpq, key)) |
3153 | return 0; | |
3154 | ||
c8d317aa | 3155 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3156 | list_del_init(&wait->entry); |
3157 | ||
bcf5a063 JA |
3158 | /* submit ref gets dropped, acquire a new one */ |
3159 | refcount_inc(&req->refs); | |
921b9054 | 3160 | io_req_task_queue(req); |
bcf5a063 JA |
3161 | return 1; |
3162 | } | |
3163 | ||
c1dd91d1 JA |
3164 | /* |
3165 | * This controls whether a given IO request should be armed for async page | |
3166 | * based retry. If we return false here, the request is handed to the async | |
3167 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3168 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3169 | * will either succeed because the page is now uptodate and unlocked, or it | |
3170 | * will register a callback when the page is unlocked at IO completion. Through | |
3171 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3172 | * That retry will attempt the buffered read again. The retry will generally | |
3173 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3174 | * async worker threads for a blocking retry. | |
3175 | */ | |
227c0c96 | 3176 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3177 | { |
e8c2bc1f JA |
3178 | struct io_async_rw *rw = req->async_data; |
3179 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3180 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3181 | |
bcf5a063 JA |
3182 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3183 | if (req->flags & REQ_F_NOWAIT) | |
3184 | return false; | |
f67676d1 | 3185 | |
227c0c96 | 3186 | /* Only for buffered IO */ |
3b2a4439 | 3187 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3188 | return false; |
3b2a4439 | 3189 | |
bcf5a063 JA |
3190 | /* |
3191 | * just use poll if we can, and don't attempt if the fs doesn't | |
3192 | * support callback based unlocks | |
3193 | */ | |
3194 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3195 | return false; | |
f67676d1 | 3196 | |
3b2a4439 JA |
3197 | wait->wait.func = io_async_buf_func; |
3198 | wait->wait.private = req; | |
3199 | wait->wait.flags = 0; | |
3200 | INIT_LIST_HEAD(&wait->wait.entry); | |
3201 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3202 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3203 | kiocb->ki_waitq = wait; |
3b2a4439 | 3204 | return true; |
bcf5a063 JA |
3205 | } |
3206 | ||
3207 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3208 | { | |
3209 | if (req->file->f_op->read_iter) | |
3210 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3211 | else if (req->file->f_op->read) |
4017eb91 | 3212 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3213 | else |
3214 | return -EINVAL; | |
f67676d1 JA |
3215 | } |
3216 | ||
889fca73 | 3217 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3218 | { |
3219 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3220 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3221 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3222 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3223 | ssize_t io_size, ret, ret2; |
45d189c6 | 3224 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3225 | |
2846c481 | 3226 | if (rw) { |
e8c2bc1f | 3227 | iter = &rw->iter; |
2846c481 PB |
3228 | iovec = NULL; |
3229 | } else { | |
3230 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3231 | if (ret < 0) | |
3232 | return ret; | |
3233 | } | |
632546c4 | 3234 | io_size = iov_iter_count(iter); |
fa15bafb | 3235 | req->result = io_size; |
2b188cc1 | 3236 | |
fd6c2e4c JA |
3237 | /* Ensure we clear previously set non-block flag */ |
3238 | if (!force_nonblock) | |
29de5f6a | 3239 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3240 | else |
3241 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3242 | ||
24c74678 | 3243 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3244 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3245 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3246 | return ret ?: -EAGAIN; |
6713e7a6 | 3247 | } |
9e645e11 | 3248 | |
632546c4 | 3249 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3250 | if (unlikely(ret)) { |
3251 | kfree(iovec); | |
3252 | return ret; | |
3253 | } | |
2b188cc1 | 3254 | |
227c0c96 | 3255 | ret = io_iter_do_read(req, iter); |
32960613 | 3256 | |
57cd657b | 3257 | if (ret == -EIOCBQUEUED) { |
3e6a0d3c JA |
3258 | if (req->async_data) |
3259 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fe1cdd55 | 3260 | goto out_free; |
227c0c96 | 3261 | } else if (ret == -EAGAIN) { |
eefdf30f JA |
3262 | /* IOPOLL retry should happen for io-wq threads */ |
3263 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3264 | goto done; |
75c668cd PB |
3265 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3266 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3267 | goto done; |
84216315 | 3268 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3269 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3270 | ret = 0; |
7335e3bf | 3271 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3272 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3273 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3274 | goto done; |
227c0c96 JA |
3275 | } |
3276 | ||
227c0c96 | 3277 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3278 | if (ret2) |
3279 | return ret2; | |
3280 | ||
fe1cdd55 | 3281 | iovec = NULL; |
e8c2bc1f | 3282 | rw = req->async_data; |
227c0c96 | 3283 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3284 | iter = &rw->iter; |
227c0c96 | 3285 | |
b23df91b PB |
3286 | do { |
3287 | io_size -= ret; | |
3288 | rw->bytes_done += ret; | |
3289 | /* if we can retry, do so with the callbacks armed */ | |
3290 | if (!io_rw_should_retry(req)) { | |
3291 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3292 | return -EAGAIN; | |
3293 | } | |
3294 | ||
3295 | /* | |
3296 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3297 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3298 | * desired page gets unlocked. We can also get a partial read | |
3299 | * here, and if we do, then just retry at the new offset. | |
3300 | */ | |
3301 | ret = io_iter_do_read(req, iter); | |
3302 | if (ret == -EIOCBQUEUED) | |
3303 | return 0; | |
227c0c96 | 3304 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3305 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3306 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3307 | done: |
889fca73 | 3308 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3309 | out_free: |
3310 | /* it's faster to check here then delegate to kfree */ | |
3311 | if (iovec) | |
3312 | kfree(iovec); | |
5ea5dd45 | 3313 | return 0; |
2b188cc1 JA |
3314 | } |
3315 | ||
73debe68 | 3316 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3317 | { |
3529d8c2 JA |
3318 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3319 | return -EBADF; | |
93642ef8 | 3320 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3321 | } |
3322 | ||
889fca73 | 3323 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3324 | { |
3325 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3326 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3327 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3328 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3329 | ssize_t ret, ret2, io_size; |
45d189c6 | 3330 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3331 | |
2846c481 | 3332 | if (rw) { |
e8c2bc1f | 3333 | iter = &rw->iter; |
2846c481 PB |
3334 | iovec = NULL; |
3335 | } else { | |
3336 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3337 | if (ret < 0) | |
3338 | return ret; | |
3339 | } | |
632546c4 | 3340 | io_size = iov_iter_count(iter); |
fa15bafb | 3341 | req->result = io_size; |
2b188cc1 | 3342 | |
fd6c2e4c JA |
3343 | /* Ensure we clear previously set non-block flag */ |
3344 | if (!force_nonblock) | |
a88fc400 PB |
3345 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3346 | else | |
3347 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3348 | |
24c74678 | 3349 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3350 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3351 | goto copy_iov; |
31b51510 | 3352 | |
10d59345 JA |
3353 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3354 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3355 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3356 | goto copy_iov; |
31b51510 | 3357 | |
632546c4 | 3358 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3359 | if (unlikely(ret)) |
3360 | goto out_free; | |
4ed734b0 | 3361 | |
fa15bafb PB |
3362 | /* |
3363 | * Open-code file_start_write here to grab freeze protection, | |
3364 | * which will be released by another thread in | |
3365 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3366 | * released so that it doesn't complain about the held lock when | |
3367 | * we return to userspace. | |
3368 | */ | |
3369 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3370 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3371 | __sb_writers_release(file_inode(req->file)->i_sb, |
3372 | SB_FREEZE_WRITE); | |
3373 | } | |
3374 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3375 | |
fa15bafb | 3376 | if (req->file->f_op->write_iter) |
ff6165b2 | 3377 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3378 | else if (req->file->f_op->write) |
4017eb91 | 3379 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3380 | else |
3381 | ret2 = -EINVAL; | |
4ed734b0 | 3382 | |
fa15bafb PB |
3383 | /* |
3384 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3385 | * retry them without IOCB_NOWAIT. | |
3386 | */ | |
3387 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3388 | ret2 = -EAGAIN; | |
75c668cd PB |
3389 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3390 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3391 | goto done; |
3e6a0d3c JA |
3392 | if (ret2 == -EIOCBQUEUED && req->async_data) |
3393 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); | |
fa15bafb | 3394 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3395 | /* IOPOLL retry should happen for io-wq threads */ |
3396 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3397 | goto copy_iov; | |
355afaeb | 3398 | done: |
889fca73 | 3399 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3400 | } else { |
f67676d1 | 3401 | copy_iov: |
84216315 | 3402 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3403 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3404 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3405 | return ret ?: -EAGAIN; |
2b188cc1 | 3406 | } |
31b51510 | 3407 | out_free: |
f261c168 | 3408 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3409 | if (iovec) |
6f2cc166 | 3410 | kfree(iovec); |
2b188cc1 JA |
3411 | return ret; |
3412 | } | |
3413 | ||
80a261fd JA |
3414 | static int io_renameat_prep(struct io_kiocb *req, |
3415 | const struct io_uring_sqe *sqe) | |
3416 | { | |
3417 | struct io_rename *ren = &req->rename; | |
3418 | const char __user *oldf, *newf; | |
3419 | ||
3420 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3421 | return -EBADF; | |
3422 | ||
3423 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3424 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3425 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3426 | ren->new_dfd = READ_ONCE(sqe->len); | |
3427 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3428 | ||
3429 | ren->oldpath = getname(oldf); | |
3430 | if (IS_ERR(ren->oldpath)) | |
3431 | return PTR_ERR(ren->oldpath); | |
3432 | ||
3433 | ren->newpath = getname(newf); | |
3434 | if (IS_ERR(ren->newpath)) { | |
3435 | putname(ren->oldpath); | |
3436 | return PTR_ERR(ren->newpath); | |
3437 | } | |
3438 | ||
3439 | req->flags |= REQ_F_NEED_CLEANUP; | |
3440 | return 0; | |
3441 | } | |
3442 | ||
45d189c6 | 3443 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3444 | { |
3445 | struct io_rename *ren = &req->rename; | |
3446 | int ret; | |
3447 | ||
45d189c6 | 3448 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3449 | return -EAGAIN; |
3450 | ||
3451 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3452 | ren->newpath, ren->flags); | |
3453 | ||
3454 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3455 | if (ret < 0) | |
3456 | req_set_fail_links(req); | |
3457 | io_req_complete(req, ret); | |
3458 | return 0; | |
3459 | } | |
3460 | ||
14a1143b JA |
3461 | static int io_unlinkat_prep(struct io_kiocb *req, |
3462 | const struct io_uring_sqe *sqe) | |
3463 | { | |
3464 | struct io_unlink *un = &req->unlink; | |
3465 | const char __user *fname; | |
3466 | ||
3467 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3468 | return -EBADF; | |
3469 | ||
3470 | un->dfd = READ_ONCE(sqe->fd); | |
3471 | ||
3472 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3473 | if (un->flags & ~AT_REMOVEDIR) | |
3474 | return -EINVAL; | |
3475 | ||
3476 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3477 | un->filename = getname(fname); | |
3478 | if (IS_ERR(un->filename)) | |
3479 | return PTR_ERR(un->filename); | |
3480 | ||
3481 | req->flags |= REQ_F_NEED_CLEANUP; | |
3482 | return 0; | |
3483 | } | |
3484 | ||
45d189c6 | 3485 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3486 | { |
3487 | struct io_unlink *un = &req->unlink; | |
3488 | int ret; | |
3489 | ||
45d189c6 | 3490 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3491 | return -EAGAIN; |
3492 | ||
3493 | if (un->flags & AT_REMOVEDIR) | |
3494 | ret = do_rmdir(un->dfd, un->filename); | |
3495 | else | |
3496 | ret = do_unlinkat(un->dfd, un->filename); | |
3497 | ||
3498 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3499 | if (ret < 0) | |
3500 | req_set_fail_links(req); | |
3501 | io_req_complete(req, ret); | |
3502 | return 0; | |
3503 | } | |
3504 | ||
36f4fa68 JA |
3505 | static int io_shutdown_prep(struct io_kiocb *req, |
3506 | const struct io_uring_sqe *sqe) | |
3507 | { | |
3508 | #if defined(CONFIG_NET) | |
3509 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3510 | return -EINVAL; | |
3511 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3512 | sqe->buf_index) | |
3513 | return -EINVAL; | |
3514 | ||
3515 | req->shutdown.how = READ_ONCE(sqe->len); | |
3516 | return 0; | |
3517 | #else | |
3518 | return -EOPNOTSUPP; | |
3519 | #endif | |
3520 | } | |
3521 | ||
45d189c6 | 3522 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3523 | { |
3524 | #if defined(CONFIG_NET) | |
3525 | struct socket *sock; | |
3526 | int ret; | |
3527 | ||
45d189c6 | 3528 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3529 | return -EAGAIN; |
3530 | ||
48aba79b | 3531 | sock = sock_from_file(req->file); |
36f4fa68 | 3532 | if (unlikely(!sock)) |
48aba79b | 3533 | return -ENOTSOCK; |
36f4fa68 JA |
3534 | |
3535 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3536 | if (ret < 0) |
3537 | req_set_fail_links(req); | |
36f4fa68 JA |
3538 | io_req_complete(req, ret); |
3539 | return 0; | |
3540 | #else | |
3541 | return -EOPNOTSUPP; | |
3542 | #endif | |
3543 | } | |
3544 | ||
f2a8d5c7 PB |
3545 | static int __io_splice_prep(struct io_kiocb *req, |
3546 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3547 | { |
3548 | struct io_splice* sp = &req->splice; | |
3549 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3550 | |
3232dd02 PB |
3551 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3552 | return -EINVAL; | |
7d67af2c PB |
3553 | |
3554 | sp->file_in = NULL; | |
7d67af2c PB |
3555 | sp->len = READ_ONCE(sqe->len); |
3556 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3557 | ||
3558 | if (unlikely(sp->flags & ~valid_flags)) | |
3559 | return -EINVAL; | |
3560 | ||
8371adf5 PB |
3561 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3562 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3563 | if (!sp->file_in) | |
3564 | return -EBADF; | |
7d67af2c PB |
3565 | req->flags |= REQ_F_NEED_CLEANUP; |
3566 | ||
7cdaf587 XW |
3567 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3568 | /* | |
3569 | * Splice operation will be punted aync, and here need to | |
3570 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3571 | */ | |
7d67af2c | 3572 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3573 | } |
7d67af2c PB |
3574 | |
3575 | return 0; | |
3576 | } | |
3577 | ||
f2a8d5c7 PB |
3578 | static int io_tee_prep(struct io_kiocb *req, |
3579 | const struct io_uring_sqe *sqe) | |
3580 | { | |
3581 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3582 | return -EINVAL; | |
3583 | return __io_splice_prep(req, sqe); | |
3584 | } | |
3585 | ||
45d189c6 | 3586 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3587 | { |
3588 | struct io_splice *sp = &req->splice; | |
3589 | struct file *in = sp->file_in; | |
3590 | struct file *out = sp->file_out; | |
3591 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3592 | long ret = 0; | |
3593 | ||
45d189c6 | 3594 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3595 | return -EAGAIN; |
3596 | if (sp->len) | |
3597 | ret = do_tee(in, out, sp->len, flags); | |
3598 | ||
3599 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3600 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3601 | ||
f2a8d5c7 PB |
3602 | if (ret != sp->len) |
3603 | req_set_fail_links(req); | |
e1e16097 | 3604 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3605 | return 0; |
3606 | } | |
3607 | ||
3608 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3609 | { | |
3610 | struct io_splice* sp = &req->splice; | |
3611 | ||
3612 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3613 | sp->off_out = READ_ONCE(sqe->off); | |
3614 | return __io_splice_prep(req, sqe); | |
3615 | } | |
3616 | ||
45d189c6 | 3617 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3618 | { |
3619 | struct io_splice *sp = &req->splice; | |
3620 | struct file *in = sp->file_in; | |
3621 | struct file *out = sp->file_out; | |
3622 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3623 | loff_t *poff_in, *poff_out; | |
c9687426 | 3624 | long ret = 0; |
7d67af2c | 3625 | |
45d189c6 | 3626 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3627 | return -EAGAIN; |
7d67af2c PB |
3628 | |
3629 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3630 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3631 | |
948a7749 | 3632 | if (sp->len) |
c9687426 | 3633 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3634 | |
3635 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3636 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3637 | ||
7d67af2c PB |
3638 | if (ret != sp->len) |
3639 | req_set_fail_links(req); | |
e1e16097 | 3640 | io_req_complete(req, ret); |
7d67af2c PB |
3641 | return 0; |
3642 | } | |
3643 | ||
2b188cc1 JA |
3644 | /* |
3645 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3646 | */ | |
889fca73 | 3647 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3648 | { |
3649 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3650 | |
def596e9 JA |
3651 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3652 | return -EINVAL; | |
3653 | ||
889fca73 | 3654 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3655 | return 0; |
3656 | } | |
3657 | ||
1155c76a | 3658 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3659 | { |
6b06314c | 3660 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3661 | |
09bb8394 JA |
3662 | if (!req->file) |
3663 | return -EBADF; | |
c992fe29 | 3664 | |
6b06314c | 3665 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3666 | return -EINVAL; |
edafccee | 3667 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3668 | return -EINVAL; |
3669 | ||
8ed8d3c3 JA |
3670 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3671 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3672 | return -EINVAL; | |
3673 | ||
3674 | req->sync.off = READ_ONCE(sqe->off); | |
3675 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3676 | return 0; |
3677 | } | |
3678 | ||
45d189c6 | 3679 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3680 | { |
8ed8d3c3 | 3681 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3682 | int ret; |
3683 | ||
ac45abc0 | 3684 | /* fsync always requires a blocking context */ |
45d189c6 | 3685 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3686 | return -EAGAIN; |
3687 | ||
9adbd45d | 3688 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3689 | end > 0 ? end : LLONG_MAX, |
3690 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3691 | if (ret < 0) | |
3692 | req_set_fail_links(req); | |
e1e16097 | 3693 | io_req_complete(req, ret); |
c992fe29 CH |
3694 | return 0; |
3695 | } | |
3696 | ||
d63d1b5e JA |
3697 | static int io_fallocate_prep(struct io_kiocb *req, |
3698 | const struct io_uring_sqe *sqe) | |
3699 | { | |
3700 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3701 | return -EINVAL; | |
3232dd02 PB |
3702 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3703 | return -EINVAL; | |
d63d1b5e JA |
3704 | |
3705 | req->sync.off = READ_ONCE(sqe->off); | |
3706 | req->sync.len = READ_ONCE(sqe->addr); | |
3707 | req->sync.mode = READ_ONCE(sqe->len); | |
3708 | return 0; | |
3709 | } | |
3710 | ||
45d189c6 | 3711 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3712 | { |
ac45abc0 PB |
3713 | int ret; |
3714 | ||
d63d1b5e | 3715 | /* fallocate always requiring blocking context */ |
45d189c6 | 3716 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3717 | return -EAGAIN; |
ac45abc0 PB |
3718 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3719 | req->sync.len); | |
ac45abc0 PB |
3720 | if (ret < 0) |
3721 | req_set_fail_links(req); | |
e1e16097 | 3722 | io_req_complete(req, ret); |
5d17b4a4 JA |
3723 | return 0; |
3724 | } | |
3725 | ||
ec65fea5 | 3726 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3727 | { |
f8748881 | 3728 | const char __user *fname; |
15b71abe | 3729 | int ret; |
b7bb4f7d | 3730 | |
ec65fea5 | 3731 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3732 | return -EINVAL; |
ec65fea5 | 3733 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3734 | return -EBADF; |
03b1230c | 3735 | |
ec65fea5 PB |
3736 | /* open.how should be already initialised */ |
3737 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3738 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3739 | |
25e72d10 PB |
3740 | req->open.dfd = READ_ONCE(sqe->fd); |
3741 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3742 | req->open.filename = getname(fname); |
15b71abe JA |
3743 | if (IS_ERR(req->open.filename)) { |
3744 | ret = PTR_ERR(req->open.filename); | |
3745 | req->open.filename = NULL; | |
3746 | return ret; | |
3747 | } | |
4022e7af | 3748 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3749 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3750 | return 0; |
03b1230c JA |
3751 | } |
3752 | ||
ec65fea5 PB |
3753 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3754 | { | |
3755 | u64 flags, mode; | |
3756 | ||
14587a46 | 3757 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3758 | return -EINVAL; |
ec65fea5 PB |
3759 | mode = READ_ONCE(sqe->len); |
3760 | flags = READ_ONCE(sqe->open_flags); | |
3761 | req->open.how = build_open_how(flags, mode); | |
3762 | return __io_openat_prep(req, sqe); | |
3763 | } | |
3764 | ||
cebdb986 | 3765 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3766 | { |
cebdb986 | 3767 | struct open_how __user *how; |
cebdb986 | 3768 | size_t len; |
0fa03c62 JA |
3769 | int ret; |
3770 | ||
14587a46 | 3771 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3772 | return -EINVAL; |
cebdb986 JA |
3773 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3774 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3775 | if (len < OPEN_HOW_SIZE_VER0) |
3776 | return -EINVAL; | |
3529d8c2 | 3777 | |
cebdb986 JA |
3778 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3779 | len); | |
3780 | if (ret) | |
3781 | return ret; | |
3529d8c2 | 3782 | |
ec65fea5 | 3783 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3784 | } |
3785 | ||
45d189c6 | 3786 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3787 | { |
3788 | struct open_flags op; | |
15b71abe | 3789 | struct file *file; |
3a81fd02 JA |
3790 | bool nonblock_set; |
3791 | bool resolve_nonblock; | |
15b71abe JA |
3792 | int ret; |
3793 | ||
cebdb986 | 3794 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3795 | if (ret) |
3796 | goto err; | |
3a81fd02 JA |
3797 | nonblock_set = op.open_flag & O_NONBLOCK; |
3798 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3799 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3800 | /* |
3801 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3802 | * it'll always -EAGAIN | |
3803 | */ | |
3804 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3805 | return -EAGAIN; | |
3806 | op.lookup_flags |= LOOKUP_CACHED; | |
3807 | op.open_flag |= O_NONBLOCK; | |
3808 | } | |
15b71abe | 3809 | |
4022e7af | 3810 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3811 | if (ret < 0) |
3812 | goto err; | |
3813 | ||
3814 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3815 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3816 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3817 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3818 | /* |
3a81fd02 JA |
3819 | * We could hang on to this 'fd', but seems like marginal |
3820 | * gain for something that is now known to be a slower path. | |
3821 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3822 | */ |
3a81fd02 JA |
3823 | put_unused_fd(ret); |
3824 | return -EAGAIN; | |
3825 | } | |
3826 | ||
15b71abe JA |
3827 | if (IS_ERR(file)) { |
3828 | put_unused_fd(ret); | |
3829 | ret = PTR_ERR(file); | |
3830 | } else { | |
45d189c6 | 3831 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3832 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3833 | fsnotify_open(file); |
3834 | fd_install(ret, file); | |
3835 | } | |
3836 | err: | |
3837 | putname(req->open.filename); | |
8fef80bf | 3838 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3839 | if (ret < 0) |
3840 | req_set_fail_links(req); | |
e1e16097 | 3841 | io_req_complete(req, ret); |
15b71abe JA |
3842 | return 0; |
3843 | } | |
3844 | ||
45d189c6 | 3845 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3846 | { |
e45cff58 | 3847 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3848 | } |
3849 | ||
067524e9 JA |
3850 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3851 | const struct io_uring_sqe *sqe) | |
3852 | { | |
3853 | struct io_provide_buf *p = &req->pbuf; | |
3854 | u64 tmp; | |
3855 | ||
3856 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3857 | return -EINVAL; | |
3858 | ||
3859 | tmp = READ_ONCE(sqe->fd); | |
3860 | if (!tmp || tmp > USHRT_MAX) | |
3861 | return -EINVAL; | |
3862 | ||
3863 | memset(p, 0, sizeof(*p)); | |
3864 | p->nbufs = tmp; | |
3865 | p->bgid = READ_ONCE(sqe->buf_group); | |
3866 | return 0; | |
3867 | } | |
3868 | ||
3869 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3870 | int bgid, unsigned nbufs) | |
3871 | { | |
3872 | unsigned i = 0; | |
3873 | ||
3874 | /* shouldn't happen */ | |
3875 | if (!nbufs) | |
3876 | return 0; | |
3877 | ||
3878 | /* the head kbuf is the list itself */ | |
3879 | while (!list_empty(&buf->list)) { | |
3880 | struct io_buffer *nxt; | |
3881 | ||
3882 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3883 | list_del(&nxt->list); | |
3884 | kfree(nxt); | |
3885 | if (++i == nbufs) | |
3886 | return i; | |
3887 | } | |
3888 | i++; | |
3889 | kfree(buf); | |
3890 | idr_remove(&ctx->io_buffer_idr, bgid); | |
3891 | ||
3892 | return i; | |
3893 | } | |
3894 | ||
889fca73 | 3895 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3896 | { |
3897 | struct io_provide_buf *p = &req->pbuf; | |
3898 | struct io_ring_ctx *ctx = req->ctx; | |
3899 | struct io_buffer *head; | |
3900 | int ret = 0; | |
45d189c6 | 3901 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3902 | |
3903 | io_ring_submit_lock(ctx, !force_nonblock); | |
3904 | ||
3905 | lockdep_assert_held(&ctx->uring_lock); | |
3906 | ||
3907 | ret = -ENOENT; | |
3908 | head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3909 | if (head) | |
3910 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3911 | if (ret < 0) |
3912 | req_set_fail_links(req); | |
067524e9 | 3913 | |
31bff9a5 PB |
3914 | /* need to hold the lock to complete IOPOLL requests */ |
3915 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3916 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3917 | io_ring_submit_unlock(ctx, !force_nonblock); |
3918 | } else { | |
3919 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3920 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3921 | } |
067524e9 JA |
3922 | return 0; |
3923 | } | |
3924 | ||
ddf0322d JA |
3925 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3926 | const struct io_uring_sqe *sqe) | |
3927 | { | |
3928 | struct io_provide_buf *p = &req->pbuf; | |
3929 | u64 tmp; | |
3930 | ||
3931 | if (sqe->ioprio || sqe->rw_flags) | |
3932 | return -EINVAL; | |
3933 | ||
3934 | tmp = READ_ONCE(sqe->fd); | |
3935 | if (!tmp || tmp > USHRT_MAX) | |
3936 | return -E2BIG; | |
3937 | p->nbufs = tmp; | |
3938 | p->addr = READ_ONCE(sqe->addr); | |
3939 | p->len = READ_ONCE(sqe->len); | |
3940 | ||
efe68c1c | 3941 | if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs))) |
ddf0322d JA |
3942 | return -EFAULT; |
3943 | ||
3944 | p->bgid = READ_ONCE(sqe->buf_group); | |
3945 | tmp = READ_ONCE(sqe->off); | |
3946 | if (tmp > USHRT_MAX) | |
3947 | return -E2BIG; | |
3948 | p->bid = tmp; | |
3949 | return 0; | |
3950 | } | |
3951 | ||
3952 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3953 | { | |
3954 | struct io_buffer *buf; | |
3955 | u64 addr = pbuf->addr; | |
3956 | int i, bid = pbuf->bid; | |
3957 | ||
3958 | for (i = 0; i < pbuf->nbufs; i++) { | |
3959 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3960 | if (!buf) | |
3961 | break; | |
3962 | ||
3963 | buf->addr = addr; | |
3964 | buf->len = pbuf->len; | |
3965 | buf->bid = bid; | |
3966 | addr += pbuf->len; | |
3967 | bid++; | |
3968 | if (!*head) { | |
3969 | INIT_LIST_HEAD(&buf->list); | |
3970 | *head = buf; | |
3971 | } else { | |
3972 | list_add_tail(&buf->list, &(*head)->list); | |
3973 | } | |
3974 | } | |
3975 | ||
3976 | return i ? i : -ENOMEM; | |
3977 | } | |
3978 | ||
889fca73 | 3979 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3980 | { |
3981 | struct io_provide_buf *p = &req->pbuf; | |
3982 | struct io_ring_ctx *ctx = req->ctx; | |
3983 | struct io_buffer *head, *list; | |
3984 | int ret = 0; | |
45d189c6 | 3985 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
3986 | |
3987 | io_ring_submit_lock(ctx, !force_nonblock); | |
3988 | ||
3989 | lockdep_assert_held(&ctx->uring_lock); | |
3990 | ||
3991 | list = head = idr_find(&ctx->io_buffer_idr, p->bgid); | |
3992 | ||
3993 | ret = io_add_buffers(p, &head); | |
3994 | if (ret < 0) | |
3995 | goto out; | |
3996 | ||
3997 | if (!list) { | |
3998 | ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1, | |
3999 | GFP_KERNEL); | |
4000 | if (ret < 0) { | |
067524e9 | 4001 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d JA |
4002 | goto out; |
4003 | } | |
4004 | } | |
4005 | out: | |
ddf0322d JA |
4006 | if (ret < 0) |
4007 | req_set_fail_links(req); | |
31bff9a5 PB |
4008 | |
4009 | /* need to hold the lock to complete IOPOLL requests */ | |
4010 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4011 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4012 | io_ring_submit_unlock(ctx, !force_nonblock); |
4013 | } else { | |
4014 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4015 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4016 | } |
ddf0322d | 4017 | return 0; |
cebdb986 JA |
4018 | } |
4019 | ||
3e4827b0 JA |
4020 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4021 | const struct io_uring_sqe *sqe) | |
4022 | { | |
4023 | #if defined(CONFIG_EPOLL) | |
4024 | if (sqe->ioprio || sqe->buf_index) | |
4025 | return -EINVAL; | |
6ca56f84 | 4026 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4027 | return -EINVAL; |
3e4827b0 JA |
4028 | |
4029 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4030 | req->epoll.op = READ_ONCE(sqe->len); | |
4031 | req->epoll.fd = READ_ONCE(sqe->off); | |
4032 | ||
4033 | if (ep_op_has_event(req->epoll.op)) { | |
4034 | struct epoll_event __user *ev; | |
4035 | ||
4036 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4037 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4038 | return -EFAULT; | |
4039 | } | |
4040 | ||
4041 | return 0; | |
4042 | #else | |
4043 | return -EOPNOTSUPP; | |
4044 | #endif | |
4045 | } | |
4046 | ||
889fca73 | 4047 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4048 | { |
4049 | #if defined(CONFIG_EPOLL) | |
4050 | struct io_epoll *ie = &req->epoll; | |
4051 | int ret; | |
45d189c6 | 4052 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4053 | |
4054 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4055 | if (force_nonblock && ret == -EAGAIN) | |
4056 | return -EAGAIN; | |
4057 | ||
4058 | if (ret < 0) | |
4059 | req_set_fail_links(req); | |
889fca73 | 4060 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4061 | return 0; |
4062 | #else | |
4063 | return -EOPNOTSUPP; | |
4064 | #endif | |
4065 | } | |
4066 | ||
c1ca757b JA |
4067 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4068 | { | |
4069 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4070 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4071 | return -EINVAL; | |
3232dd02 PB |
4072 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4073 | return -EINVAL; | |
c1ca757b JA |
4074 | |
4075 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4076 | req->madvise.len = READ_ONCE(sqe->len); | |
4077 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4078 | return 0; | |
4079 | #else | |
4080 | return -EOPNOTSUPP; | |
4081 | #endif | |
4082 | } | |
4083 | ||
45d189c6 | 4084 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4085 | { |
4086 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4087 | struct io_madvise *ma = &req->madvise; | |
4088 | int ret; | |
4089 | ||
45d189c6 | 4090 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4091 | return -EAGAIN; |
4092 | ||
0726b01e | 4093 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4094 | if (ret < 0) |
4095 | req_set_fail_links(req); | |
e1e16097 | 4096 | io_req_complete(req, ret); |
c1ca757b JA |
4097 | return 0; |
4098 | #else | |
4099 | return -EOPNOTSUPP; | |
4100 | #endif | |
4101 | } | |
4102 | ||
4840e418 JA |
4103 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4104 | { | |
4105 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4106 | return -EINVAL; | |
3232dd02 PB |
4107 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4108 | return -EINVAL; | |
4840e418 JA |
4109 | |
4110 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4111 | req->fadvise.len = READ_ONCE(sqe->len); | |
4112 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4113 | return 0; | |
4114 | } | |
4115 | ||
45d189c6 | 4116 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4117 | { |
4118 | struct io_fadvise *fa = &req->fadvise; | |
4119 | int ret; | |
4120 | ||
45d189c6 | 4121 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4122 | switch (fa->advice) { |
4123 | case POSIX_FADV_NORMAL: | |
4124 | case POSIX_FADV_RANDOM: | |
4125 | case POSIX_FADV_SEQUENTIAL: | |
4126 | break; | |
4127 | default: | |
4128 | return -EAGAIN; | |
4129 | } | |
4130 | } | |
4840e418 JA |
4131 | |
4132 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4133 | if (ret < 0) | |
4134 | req_set_fail_links(req); | |
e1e16097 | 4135 | io_req_complete(req, ret); |
4840e418 JA |
4136 | return 0; |
4137 | } | |
4138 | ||
eddc7ef5 JA |
4139 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4140 | { | |
6ca56f84 | 4141 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4142 | return -EINVAL; |
eddc7ef5 JA |
4143 | if (sqe->ioprio || sqe->buf_index) |
4144 | return -EINVAL; | |
9c280f90 | 4145 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4146 | return -EBADF; |
eddc7ef5 | 4147 | |
1d9e1288 BM |
4148 | req->statx.dfd = READ_ONCE(sqe->fd); |
4149 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4150 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4151 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4152 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4153 | |
4154 | return 0; | |
4155 | } | |
4156 | ||
45d189c6 | 4157 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4158 | { |
1d9e1288 | 4159 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4160 | int ret; |
4161 | ||
45d189c6 | 4162 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4163 | /* only need file table for an actual valid fd */ |
4164 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4165 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4166 | return -EAGAIN; |
5b0bbee4 | 4167 | } |
eddc7ef5 | 4168 | |
e62753e4 BM |
4169 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4170 | ctx->buffer); | |
eddc7ef5 | 4171 | |
eddc7ef5 JA |
4172 | if (ret < 0) |
4173 | req_set_fail_links(req); | |
e1e16097 | 4174 | io_req_complete(req, ret); |
eddc7ef5 JA |
4175 | return 0; |
4176 | } | |
4177 | ||
b5dba59e JA |
4178 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4179 | { | |
14587a46 | 4180 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4181 | return -EINVAL; |
b5dba59e JA |
4182 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4183 | sqe->rw_flags || sqe->buf_index) | |
4184 | return -EINVAL; | |
9c280f90 | 4185 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4186 | return -EBADF; |
b5dba59e JA |
4187 | |
4188 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4189 | return 0; |
b5dba59e JA |
4190 | } |
4191 | ||
889fca73 | 4192 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4193 | { |
9eac1904 | 4194 | struct files_struct *files = current->files; |
3af73b28 | 4195 | struct io_close *close = &req->close; |
9eac1904 JA |
4196 | struct fdtable *fdt; |
4197 | struct file *file; | |
b5dba59e JA |
4198 | int ret; |
4199 | ||
9eac1904 JA |
4200 | file = NULL; |
4201 | ret = -EBADF; | |
4202 | spin_lock(&files->file_lock); | |
4203 | fdt = files_fdtable(files); | |
4204 | if (close->fd >= fdt->max_fds) { | |
4205 | spin_unlock(&files->file_lock); | |
4206 | goto err; | |
4207 | } | |
4208 | file = fdt->fd[close->fd]; | |
4209 | if (!file) { | |
4210 | spin_unlock(&files->file_lock); | |
4211 | goto err; | |
4212 | } | |
4213 | ||
4214 | if (file->f_op == &io_uring_fops) { | |
4215 | spin_unlock(&files->file_lock); | |
4216 | file = NULL; | |
4217 | goto err; | |
3af73b28 | 4218 | } |
b5dba59e JA |
4219 | |
4220 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4221 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4222 | spin_unlock(&files->file_lock); |
0bf0eefd | 4223 | return -EAGAIN; |
a2100672 | 4224 | } |
b5dba59e | 4225 | |
9eac1904 JA |
4226 | ret = __close_fd_get_file(close->fd, &file); |
4227 | spin_unlock(&files->file_lock); | |
4228 | if (ret < 0) { | |
4229 | if (ret == -ENOENT) | |
4230 | ret = -EBADF; | |
4231 | goto err; | |
4232 | } | |
4233 | ||
3af73b28 | 4234 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4235 | ret = filp_close(file, current->files); |
4236 | err: | |
3af73b28 PB |
4237 | if (ret < 0) |
4238 | req_set_fail_links(req); | |
9eac1904 JA |
4239 | if (file) |
4240 | fput(file); | |
889fca73 | 4241 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4242 | return 0; |
b5dba59e JA |
4243 | } |
4244 | ||
1155c76a | 4245 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4246 | { |
4247 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4248 | |
5d17b4a4 JA |
4249 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4250 | return -EINVAL; | |
4251 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4252 | return -EINVAL; | |
4253 | ||
8ed8d3c3 JA |
4254 | req->sync.off = READ_ONCE(sqe->off); |
4255 | req->sync.len = READ_ONCE(sqe->len); | |
4256 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4257 | return 0; |
4258 | } | |
4259 | ||
45d189c6 | 4260 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4261 | { |
8ed8d3c3 JA |
4262 | int ret; |
4263 | ||
ac45abc0 | 4264 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4265 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4266 | return -EAGAIN; |
4267 | ||
9adbd45d | 4268 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4269 | req->sync.flags); |
4270 | if (ret < 0) | |
4271 | req_set_fail_links(req); | |
e1e16097 | 4272 | io_req_complete(req, ret); |
5d17b4a4 JA |
4273 | return 0; |
4274 | } | |
4275 | ||
469956e8 | 4276 | #if defined(CONFIG_NET) |
02d27d89 PB |
4277 | static int io_setup_async_msg(struct io_kiocb *req, |
4278 | struct io_async_msghdr *kmsg) | |
4279 | { | |
e8c2bc1f JA |
4280 | struct io_async_msghdr *async_msg = req->async_data; |
4281 | ||
4282 | if (async_msg) | |
02d27d89 | 4283 | return -EAGAIN; |
e8c2bc1f | 4284 | if (io_alloc_async_data(req)) { |
257e84a5 | 4285 | kfree(kmsg->free_iov); |
02d27d89 PB |
4286 | return -ENOMEM; |
4287 | } | |
e8c2bc1f | 4288 | async_msg = req->async_data; |
02d27d89 | 4289 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4290 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4291 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4292 | /* if were using fast_iov, set it to the new one */ |
4293 | if (!async_msg->free_iov) | |
4294 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4295 | ||
02d27d89 PB |
4296 | return -EAGAIN; |
4297 | } | |
4298 | ||
2ae523ed PB |
4299 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4300 | struct io_async_msghdr *iomsg) | |
4301 | { | |
2ae523ed | 4302 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4303 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4304 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4305 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4306 | } |
4307 | ||
93642ef8 PB |
4308 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4309 | { | |
4310 | int ret; | |
4311 | ||
4312 | if (!io_op_defs[req->opcode].needs_async_data) | |
4313 | return 0; | |
4314 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4315 | if (!ret) | |
4316 | req->flags |= REQ_F_NEED_CLEANUP; | |
4317 | return ret; | |
4318 | } | |
4319 | ||
3529d8c2 | 4320 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4321 | { |
e47293fd | 4322 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4323 | |
d2b6f48b PB |
4324 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4325 | return -EINVAL; | |
4326 | ||
e47293fd | 4327 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4328 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4329 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4330 | |
d8768362 JA |
4331 | #ifdef CONFIG_COMPAT |
4332 | if (req->ctx->compat) | |
4333 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4334 | #endif | |
93642ef8 | 4335 | return 0; |
03b1230c JA |
4336 | } |
4337 | ||
889fca73 | 4338 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4339 | { |
6b754c8b | 4340 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4341 | struct socket *sock; |
7a7cacba | 4342 | unsigned flags; |
0fa03c62 JA |
4343 | int ret; |
4344 | ||
dba4a925 | 4345 | sock = sock_from_file(req->file); |
7a7cacba | 4346 | if (unlikely(!sock)) |
dba4a925 | 4347 | return -ENOTSOCK; |
3529d8c2 | 4348 | |
257e84a5 PB |
4349 | kmsg = req->async_data; |
4350 | if (!kmsg) { | |
7a7cacba PB |
4351 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4352 | if (ret) | |
4353 | return ret; | |
4354 | kmsg = &iomsg; | |
0fa03c62 | 4355 | } |
0fa03c62 | 4356 | |
7a7cacba PB |
4357 | flags = req->sr_msg.msg_flags; |
4358 | if (flags & MSG_DONTWAIT) | |
4359 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4360 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4361 | flags |= MSG_DONTWAIT; |
e47293fd | 4362 | |
7a7cacba | 4363 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4364 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4365 | return io_setup_async_msg(req, kmsg); |
4366 | if (ret == -ERESTARTSYS) | |
4367 | ret = -EINTR; | |
0fa03c62 | 4368 | |
257e84a5 PB |
4369 | /* fast path, check for non-NULL to avoid function call */ |
4370 | if (kmsg->free_iov) | |
4371 | kfree(kmsg->free_iov); | |
99bc4c38 | 4372 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4373 | if (ret < 0) |
4374 | req_set_fail_links(req); | |
889fca73 | 4375 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4376 | return 0; |
03b1230c | 4377 | } |
aa1fa28f | 4378 | |
889fca73 | 4379 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4380 | { |
7a7cacba PB |
4381 | struct io_sr_msg *sr = &req->sr_msg; |
4382 | struct msghdr msg; | |
4383 | struct iovec iov; | |
fddaface | 4384 | struct socket *sock; |
7a7cacba | 4385 | unsigned flags; |
fddaface JA |
4386 | int ret; |
4387 | ||
dba4a925 | 4388 | sock = sock_from_file(req->file); |
7a7cacba | 4389 | if (unlikely(!sock)) |
dba4a925 | 4390 | return -ENOTSOCK; |
fddaface | 4391 | |
7a7cacba PB |
4392 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4393 | if (unlikely(ret)) | |
14db8411 | 4394 | return ret; |
fddaface | 4395 | |
7a7cacba PB |
4396 | msg.msg_name = NULL; |
4397 | msg.msg_control = NULL; | |
4398 | msg.msg_controllen = 0; | |
4399 | msg.msg_namelen = 0; | |
fddaface | 4400 | |
7a7cacba PB |
4401 | flags = req->sr_msg.msg_flags; |
4402 | if (flags & MSG_DONTWAIT) | |
4403 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4404 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4405 | flags |= MSG_DONTWAIT; |
fddaface | 4406 | |
7a7cacba PB |
4407 | msg.msg_flags = flags; |
4408 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4409 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4410 | return -EAGAIN; |
4411 | if (ret == -ERESTARTSYS) | |
4412 | ret = -EINTR; | |
fddaface | 4413 | |
fddaface JA |
4414 | if (ret < 0) |
4415 | req_set_fail_links(req); | |
889fca73 | 4416 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4417 | return 0; |
fddaface JA |
4418 | } |
4419 | ||
1400e697 PB |
4420 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4421 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4422 | { |
4423 | struct io_sr_msg *sr = &req->sr_msg; | |
4424 | struct iovec __user *uiov; | |
4425 | size_t iov_len; | |
4426 | int ret; | |
4427 | ||
1400e697 PB |
4428 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4429 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4430 | if (ret) |
4431 | return ret; | |
4432 | ||
4433 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4434 | if (iov_len > 1) | |
4435 | return -EINVAL; | |
5476dfed | 4436 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4437 | return -EFAULT; |
5476dfed | 4438 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4439 | iomsg->free_iov = NULL; |
52de1fe1 | 4440 | } else { |
257e84a5 | 4441 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4442 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4443 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4444 | false); |
52de1fe1 JA |
4445 | if (ret > 0) |
4446 | ret = 0; | |
4447 | } | |
4448 | ||
4449 | return ret; | |
4450 | } | |
4451 | ||
4452 | #ifdef CONFIG_COMPAT | |
4453 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4454 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4455 | { |
4456 | struct compat_msghdr __user *msg_compat; | |
4457 | struct io_sr_msg *sr = &req->sr_msg; | |
4458 | struct compat_iovec __user *uiov; | |
4459 | compat_uptr_t ptr; | |
4460 | compat_size_t len; | |
4461 | int ret; | |
4462 | ||
270a5940 | 4463 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4464 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4465 | &ptr, &len); |
4466 | if (ret) | |
4467 | return ret; | |
4468 | ||
4469 | uiov = compat_ptr(ptr); | |
4470 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4471 | compat_ssize_t clen; | |
4472 | ||
4473 | if (len > 1) | |
4474 | return -EINVAL; | |
4475 | if (!access_ok(uiov, sizeof(*uiov))) | |
4476 | return -EFAULT; | |
4477 | if (__get_user(clen, &uiov->iov_len)) | |
4478 | return -EFAULT; | |
4479 | if (clen < 0) | |
4480 | return -EINVAL; | |
2d280bc8 | 4481 | sr->len = clen; |
257e84a5 | 4482 | iomsg->free_iov = NULL; |
52de1fe1 | 4483 | } else { |
257e84a5 | 4484 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4485 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4486 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4487 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4488 | if (ret < 0) |
4489 | return ret; | |
4490 | } | |
4491 | ||
4492 | return 0; | |
4493 | } | |
4494 | #endif | |
4495 | ||
1400e697 PB |
4496 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4497 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4498 | { |
1400e697 | 4499 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4500 | |
4501 | #ifdef CONFIG_COMPAT | |
4502 | if (req->ctx->compat) | |
1400e697 | 4503 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4504 | #endif |
52de1fe1 | 4505 | |
1400e697 | 4506 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4507 | } |
4508 | ||
bcda7baa | 4509 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4510 | bool needs_lock) |
bcda7baa JA |
4511 | { |
4512 | struct io_sr_msg *sr = &req->sr_msg; | |
4513 | struct io_buffer *kbuf; | |
4514 | ||
bcda7baa JA |
4515 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4516 | if (IS_ERR(kbuf)) | |
4517 | return kbuf; | |
4518 | ||
4519 | sr->kbuf = kbuf; | |
4520 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4521 | return kbuf; |
fddaface JA |
4522 | } |
4523 | ||
7fbb1b54 PB |
4524 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4525 | { | |
4526 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4527 | } | |
4528 | ||
93642ef8 | 4529 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4530 | { |
99bc4c38 | 4531 | int ret; |
3529d8c2 | 4532 | |
93642ef8 PB |
4533 | if (!io_op_defs[req->opcode].needs_async_data) |
4534 | return 0; | |
4535 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4536 | if (!ret) | |
4537 | req->flags |= REQ_F_NEED_CLEANUP; | |
4538 | return ret; | |
4539 | } | |
4540 | ||
4541 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4542 | { | |
4543 | struct io_sr_msg *sr = &req->sr_msg; | |
4544 | ||
d2b6f48b PB |
4545 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4546 | return -EINVAL; | |
4547 | ||
3529d8c2 | 4548 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4549 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4550 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4551 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4552 | |
d8768362 JA |
4553 | #ifdef CONFIG_COMPAT |
4554 | if (req->ctx->compat) | |
4555 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4556 | #endif | |
93642ef8 | 4557 | return 0; |
aa1fa28f JA |
4558 | } |
4559 | ||
889fca73 | 4560 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4561 | { |
6b754c8b | 4562 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4563 | struct socket *sock; |
7fbb1b54 | 4564 | struct io_buffer *kbuf; |
7a7cacba | 4565 | unsigned flags; |
52de1fe1 | 4566 | int ret, cflags = 0; |
45d189c6 | 4567 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4568 | |
dba4a925 | 4569 | sock = sock_from_file(req->file); |
7a7cacba | 4570 | if (unlikely(!sock)) |
dba4a925 | 4571 | return -ENOTSOCK; |
3529d8c2 | 4572 | |
257e84a5 PB |
4573 | kmsg = req->async_data; |
4574 | if (!kmsg) { | |
7a7cacba PB |
4575 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4576 | if (ret) | |
681fda8d | 4577 | return ret; |
7a7cacba PB |
4578 | kmsg = &iomsg; |
4579 | } | |
03b1230c | 4580 | |
bc02ef33 | 4581 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4582 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4583 | if (IS_ERR(kbuf)) |
52de1fe1 | 4584 | return PTR_ERR(kbuf); |
7a7cacba | 4585 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4586 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4587 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4588 | 1, req->sr_msg.len); |
4589 | } | |
52de1fe1 | 4590 | |
7a7cacba PB |
4591 | flags = req->sr_msg.msg_flags; |
4592 | if (flags & MSG_DONTWAIT) | |
4593 | req->flags |= REQ_F_NOWAIT; | |
4594 | else if (force_nonblock) | |
4595 | flags |= MSG_DONTWAIT; | |
e47293fd | 4596 | |
7a7cacba PB |
4597 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4598 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4599 | if (force_nonblock && ret == -EAGAIN) |
4600 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4601 | if (ret == -ERESTARTSYS) |
4602 | ret = -EINTR; | |
03b1230c | 4603 | |
7fbb1b54 PB |
4604 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4605 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4606 | /* fast path, check for non-NULL to avoid function call */ |
4607 | if (kmsg->free_iov) | |
4608 | kfree(kmsg->free_iov); | |
99bc4c38 | 4609 | req->flags &= ~REQ_F_NEED_CLEANUP; |
4e88d6e7 JA |
4610 | if (ret < 0) |
4611 | req_set_fail_links(req); | |
889fca73 | 4612 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4613 | return 0; |
0fa03c62 | 4614 | } |
5d17b4a4 | 4615 | |
889fca73 | 4616 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4617 | { |
6b754c8b | 4618 | struct io_buffer *kbuf; |
7a7cacba PB |
4619 | struct io_sr_msg *sr = &req->sr_msg; |
4620 | struct msghdr msg; | |
4621 | void __user *buf = sr->buf; | |
fddaface | 4622 | struct socket *sock; |
7a7cacba PB |
4623 | struct iovec iov; |
4624 | unsigned flags; | |
bcda7baa | 4625 | int ret, cflags = 0; |
45d189c6 | 4626 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4627 | |
dba4a925 | 4628 | sock = sock_from_file(req->file); |
7a7cacba | 4629 | if (unlikely(!sock)) |
dba4a925 | 4630 | return -ENOTSOCK; |
fddaface | 4631 | |
bc02ef33 | 4632 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4633 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4634 | if (IS_ERR(kbuf)) |
4635 | return PTR_ERR(kbuf); | |
7a7cacba | 4636 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4637 | } |
bcda7baa | 4638 | |
7a7cacba | 4639 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4640 | if (unlikely(ret)) |
4641 | goto out_free; | |
fddaface | 4642 | |
7a7cacba PB |
4643 | msg.msg_name = NULL; |
4644 | msg.msg_control = NULL; | |
4645 | msg.msg_controllen = 0; | |
4646 | msg.msg_namelen = 0; | |
4647 | msg.msg_iocb = NULL; | |
4648 | msg.msg_flags = 0; | |
fddaface | 4649 | |
7a7cacba PB |
4650 | flags = req->sr_msg.msg_flags; |
4651 | if (flags & MSG_DONTWAIT) | |
4652 | req->flags |= REQ_F_NOWAIT; | |
4653 | else if (force_nonblock) | |
4654 | flags |= MSG_DONTWAIT; | |
4655 | ||
4656 | ret = sock_recvmsg(sock, &msg, flags); | |
4657 | if (force_nonblock && ret == -EAGAIN) | |
4658 | return -EAGAIN; | |
4659 | if (ret == -ERESTARTSYS) | |
4660 | ret = -EINTR; | |
14c32eee | 4661 | out_free: |
7fbb1b54 PB |
4662 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4663 | cflags = io_put_recv_kbuf(req); | |
fddaface JA |
4664 | if (ret < 0) |
4665 | req_set_fail_links(req); | |
889fca73 | 4666 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4667 | return 0; |
fddaface JA |
4668 | } |
4669 | ||
3529d8c2 | 4670 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4671 | { |
8ed8d3c3 JA |
4672 | struct io_accept *accept = &req->accept; |
4673 | ||
14587a46 | 4674 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4675 | return -EINVAL; |
8042d6ce | 4676 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4677 | return -EINVAL; |
4678 | ||
d55e5f5b JA |
4679 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4680 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4681 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4682 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4683 | return 0; |
8ed8d3c3 | 4684 | } |
17f2fe35 | 4685 | |
889fca73 | 4686 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4687 | { |
4688 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4689 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4690 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4691 | int ret; |
4692 | ||
e697deed JX |
4693 | if (req->file->f_flags & O_NONBLOCK) |
4694 | req->flags |= REQ_F_NOWAIT; | |
4695 | ||
8ed8d3c3 | 4696 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4697 | accept->addr_len, accept->flags, |
4698 | accept->nofile); | |
8ed8d3c3 | 4699 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4700 | return -EAGAIN; |
ac45abc0 PB |
4701 | if (ret < 0) { |
4702 | if (ret == -ERESTARTSYS) | |
4703 | ret = -EINTR; | |
4e88d6e7 | 4704 | req_set_fail_links(req); |
ac45abc0 | 4705 | } |
889fca73 | 4706 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4707 | return 0; |
8ed8d3c3 JA |
4708 | } |
4709 | ||
93642ef8 PB |
4710 | static int io_connect_prep_async(struct io_kiocb *req) |
4711 | { | |
4712 | struct io_async_connect *io = req->async_data; | |
4713 | struct io_connect *conn = &req->connect; | |
4714 | ||
4715 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4716 | } | |
4717 | ||
3529d8c2 | 4718 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4719 | { |
3529d8c2 | 4720 | struct io_connect *conn = &req->connect; |
f499a021 | 4721 | |
14587a46 | 4722 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4723 | return -EINVAL; |
4724 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4725 | return -EINVAL; | |
4726 | ||
3529d8c2 JA |
4727 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4728 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4729 | return 0; |
f499a021 JA |
4730 | } |
4731 | ||
889fca73 | 4732 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4733 | { |
e8c2bc1f | 4734 | struct io_async_connect __io, *io; |
f8e85cf2 | 4735 | unsigned file_flags; |
3fbb51c1 | 4736 | int ret; |
45d189c6 | 4737 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4738 | |
e8c2bc1f JA |
4739 | if (req->async_data) { |
4740 | io = req->async_data; | |
f499a021 | 4741 | } else { |
3529d8c2 JA |
4742 | ret = move_addr_to_kernel(req->connect.addr, |
4743 | req->connect.addr_len, | |
e8c2bc1f | 4744 | &__io.address); |
f499a021 JA |
4745 | if (ret) |
4746 | goto out; | |
4747 | io = &__io; | |
4748 | } | |
4749 | ||
3fbb51c1 JA |
4750 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4751 | ||
e8c2bc1f | 4752 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4753 | req->connect.addr_len, file_flags); |
87f80d62 | 4754 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4755 | if (req->async_data) |
b7bb4f7d | 4756 | return -EAGAIN; |
e8c2bc1f | 4757 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4758 | ret = -ENOMEM; |
4759 | goto out; | |
4760 | } | |
e8c2bc1f JA |
4761 | io = req->async_data; |
4762 | memcpy(req->async_data, &__io, sizeof(__io)); | |
f8e85cf2 | 4763 | return -EAGAIN; |
f499a021 | 4764 | } |
f8e85cf2 JA |
4765 | if (ret == -ERESTARTSYS) |
4766 | ret = -EINTR; | |
f499a021 | 4767 | out: |
4e88d6e7 JA |
4768 | if (ret < 0) |
4769 | req_set_fail_links(req); | |
889fca73 | 4770 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4771 | return 0; |
469956e8 Y |
4772 | } |
4773 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4774 | #define IO_NETOP_FN(op) \ |
4775 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4776 | { \ | |
4777 | return -EOPNOTSUPP; \ | |
4778 | } | |
4779 | ||
4780 | #define IO_NETOP_PREP(op) \ | |
4781 | IO_NETOP_FN(op) \ | |
4782 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4783 | { \ | |
4784 | return -EOPNOTSUPP; \ | |
4785 | } \ | |
4786 | ||
4787 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4788 | IO_NETOP_PREP(op) \ | |
4789 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4790 | { \ | |
4791 | return -EOPNOTSUPP; \ | |
4792 | } | |
4793 | ||
4794 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4795 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4796 | IO_NETOP_PREP_ASYNC(connect); | |
4797 | IO_NETOP_PREP(accept); | |
4798 | IO_NETOP_FN(send); | |
4799 | IO_NETOP_FN(recv); | |
469956e8 | 4800 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4801 | |
d7718a9d JA |
4802 | struct io_poll_table { |
4803 | struct poll_table_struct pt; | |
4804 | struct io_kiocb *req; | |
4805 | int error; | |
4806 | }; | |
ce593a6c | 4807 | |
d7718a9d JA |
4808 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4809 | __poll_t mask, task_work_func_t func) | |
4810 | { | |
aa96bf8a | 4811 | int ret; |
d7718a9d JA |
4812 | |
4813 | /* for instances that support it check for an event match first: */ | |
4814 | if (mask && !(mask & poll->events)) | |
4815 | return 0; | |
4816 | ||
4817 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4818 | ||
4819 | list_del_init(&poll->wait.entry); | |
4820 | ||
d7718a9d | 4821 | req->result = mask; |
7cbf1722 | 4822 | req->task_work.func = func; |
6d816e08 JA |
4823 | percpu_ref_get(&req->ctx->refs); |
4824 | ||
d7718a9d | 4825 | /* |
e3aabf95 JA |
4826 | * If this fails, then the task is exiting. When a task exits, the |
4827 | * work gets canceled, so just cancel this request as well instead | |
4828 | * of executing it. We can't safely execute it anyway, as we may not | |
4829 | * have the needed state needed for it anyway. | |
d7718a9d | 4830 | */ |
355fb9e2 | 4831 | ret = io_req_task_work_add(req); |
aa96bf8a | 4832 | if (unlikely(ret)) { |
e3aabf95 | 4833 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4834 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4835 | } |
d7718a9d JA |
4836 | return 1; |
4837 | } | |
4838 | ||
74ce6ce4 JA |
4839 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4840 | __acquires(&req->ctx->completion_lock) | |
4841 | { | |
4842 | struct io_ring_ctx *ctx = req->ctx; | |
4843 | ||
4844 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4845 | struct poll_table_struct pt = { ._key = poll->events }; | |
4846 | ||
4847 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4848 | } | |
4849 | ||
4850 | spin_lock_irq(&ctx->completion_lock); | |
4851 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4852 | add_wait_queue(poll->head, &poll->wait); | |
4853 | return true; | |
4854 | } | |
4855 | ||
4856 | return false; | |
4857 | } | |
4858 | ||
d4e7cd36 | 4859 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4860 | { |
e8c2bc1f | 4861 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4862 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4863 | return req->async_data; |
d4e7cd36 JA |
4864 | return req->apoll->double_poll; |
4865 | } | |
4866 | ||
4867 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4868 | { | |
4869 | if (req->opcode == IORING_OP_POLL_ADD) | |
4870 | return &req->poll; | |
4871 | return &req->apoll->poll; | |
4872 | } | |
4873 | ||
4874 | static void io_poll_remove_double(struct io_kiocb *req) | |
4875 | { | |
4876 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4877 | |
4878 | lockdep_assert_held(&req->ctx->completion_lock); | |
4879 | ||
4880 | if (poll && poll->head) { | |
4881 | struct wait_queue_head *head = poll->head; | |
4882 | ||
4883 | spin_lock(&head->lock); | |
4884 | list_del_init(&poll->wait.entry); | |
4885 | if (poll->wait.private) | |
4886 | refcount_dec(&req->refs); | |
4887 | poll->head = NULL; | |
4888 | spin_unlock(&head->lock); | |
4889 | } | |
4890 | } | |
4891 | ||
4892 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4893 | { | |
4894 | struct io_ring_ctx *ctx = req->ctx; | |
4895 | ||
d4e7cd36 | 4896 | io_poll_remove_double(req); |
18bceab1 JA |
4897 | req->poll.done = true; |
4898 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4899 | io_commit_cqring(ctx); | |
4900 | } | |
4901 | ||
dd221f46 | 4902 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4903 | { |
dd221f46 | 4904 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4905 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4906 | struct io_kiocb *nxt; |
18bceab1 JA |
4907 | |
4908 | if (io_poll_rewait(req, &req->poll)) { | |
4909 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4910 | } else { |
4911 | hash_del(&req->hash_node); | |
4912 | io_poll_complete(req, req->result, 0); | |
4913 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4914 | |
dd221f46 PB |
4915 | nxt = io_put_req_find_next(req); |
4916 | io_cqring_ev_posted(ctx); | |
4917 | if (nxt) | |
4918 | __io_req_task_submit(nxt); | |
4919 | } | |
18bceab1 | 4920 | |
6d816e08 | 4921 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4922 | } |
4923 | ||
4924 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4925 | int sync, void *key) | |
4926 | { | |
4927 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4928 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4929 | __poll_t mask = key_to_poll(key); |
4930 | ||
4931 | /* for instances that support it check for an event match first: */ | |
4932 | if (mask && !(mask & poll->events)) | |
4933 | return 0; | |
4934 | ||
8706e04e JA |
4935 | list_del_init(&wait->entry); |
4936 | ||
807abcb0 | 4937 | if (poll && poll->head) { |
18bceab1 JA |
4938 | bool done; |
4939 | ||
807abcb0 JA |
4940 | spin_lock(&poll->head->lock); |
4941 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4942 | if (!done) |
807abcb0 | 4943 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4944 | /* make sure double remove sees this as being gone */ |
4945 | wait->private = NULL; | |
807abcb0 | 4946 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4947 | if (!done) { |
4948 | /* use wait func handler, so it matches the rq type */ | |
4949 | poll->wait.func(&poll->wait, mode, sync, key); | |
4950 | } | |
18bceab1 JA |
4951 | } |
4952 | refcount_dec(&req->refs); | |
4953 | return 1; | |
4954 | } | |
4955 | ||
4956 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4957 | wait_queue_func_t wake_func) | |
4958 | { | |
4959 | poll->head = NULL; | |
4960 | poll->done = false; | |
4961 | poll->canceled = false; | |
4962 | poll->events = events; | |
4963 | INIT_LIST_HEAD(&poll->wait.entry); | |
4964 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4965 | } | |
4966 | ||
4967 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4968 | struct wait_queue_head *head, |
4969 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4970 | { |
4971 | struct io_kiocb *req = pt->req; | |
4972 | ||
4973 | /* | |
4974 | * If poll->head is already set, it's because the file being polled | |
4975 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4976 | * for write). Setup a separate io_poll_iocb if this happens. | |
4977 | */ | |
4978 | if (unlikely(poll->head)) { | |
58852d4d PB |
4979 | struct io_poll_iocb *poll_one = poll; |
4980 | ||
18bceab1 | 4981 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4982 | if (*poll_ptr) { |
18bceab1 JA |
4983 | pt->error = -EINVAL; |
4984 | return; | |
4985 | } | |
1c3b3e65 JA |
4986 | /* double add on the same waitqueue head, ignore */ |
4987 | if (poll->head == head) | |
4988 | return; | |
18bceab1 JA |
4989 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
4990 | if (!poll) { | |
4991 | pt->error = -ENOMEM; | |
4992 | return; | |
4993 | } | |
58852d4d | 4994 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
4995 | refcount_inc(&req->refs); |
4996 | poll->wait.private = req; | |
807abcb0 | 4997 | *poll_ptr = poll; |
18bceab1 JA |
4998 | } |
4999 | ||
5000 | pt->error = 0; | |
5001 | poll->head = head; | |
a31eb4a2 JX |
5002 | |
5003 | if (poll->events & EPOLLEXCLUSIVE) | |
5004 | add_wait_queue_exclusive(head, &poll->wait); | |
5005 | else | |
5006 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5007 | } |
5008 | ||
5009 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5010 | struct poll_table_struct *p) | |
5011 | { | |
5012 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5013 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5014 | |
807abcb0 | 5015 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5016 | } |
5017 | ||
d7718a9d JA |
5018 | static void io_async_task_func(struct callback_head *cb) |
5019 | { | |
5020 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5021 | struct async_poll *apoll = req->apoll; | |
5022 | struct io_ring_ctx *ctx = req->ctx; | |
5023 | ||
5024 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5025 | ||
74ce6ce4 | 5026 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5027 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5028 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5029 | return; |
d7718a9d JA |
5030 | } |
5031 | ||
31067255 | 5032 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5033 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5034 | hash_del(&req->hash_node); |
2bae047e | 5035 | |
d4e7cd36 | 5036 | io_poll_remove_double(req); |
74ce6ce4 JA |
5037 | spin_unlock_irq(&ctx->completion_lock); |
5038 | ||
0be0b0e3 PB |
5039 | if (!READ_ONCE(apoll->poll.canceled)) |
5040 | __io_req_task_submit(req); | |
5041 | else | |
5042 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5043 | |
6d816e08 | 5044 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5045 | kfree(apoll->double_poll); |
31067255 | 5046 | kfree(apoll); |
d7718a9d JA |
5047 | } |
5048 | ||
5049 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5050 | void *key) | |
5051 | { | |
5052 | struct io_kiocb *req = wait->private; | |
5053 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5054 | ||
5055 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5056 | key_to_poll(key)); | |
5057 | ||
5058 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5059 | } | |
5060 | ||
5061 | static void io_poll_req_insert(struct io_kiocb *req) | |
5062 | { | |
5063 | struct io_ring_ctx *ctx = req->ctx; | |
5064 | struct hlist_head *list; | |
5065 | ||
5066 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5067 | hlist_add_head(&req->hash_node, list); | |
5068 | } | |
5069 | ||
5070 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5071 | struct io_poll_iocb *poll, | |
5072 | struct io_poll_table *ipt, __poll_t mask, | |
5073 | wait_queue_func_t wake_func) | |
5074 | __acquires(&ctx->completion_lock) | |
5075 | { | |
5076 | struct io_ring_ctx *ctx = req->ctx; | |
5077 | bool cancel = false; | |
5078 | ||
4d52f338 | 5079 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5080 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5081 | poll->file = req->file; |
18bceab1 | 5082 | poll->wait.private = req; |
d7718a9d JA |
5083 | |
5084 | ipt->pt._key = mask; | |
5085 | ipt->req = req; | |
5086 | ipt->error = -EINVAL; | |
5087 | ||
d7718a9d JA |
5088 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5089 | ||
5090 | spin_lock_irq(&ctx->completion_lock); | |
5091 | if (likely(poll->head)) { | |
5092 | spin_lock(&poll->head->lock); | |
5093 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5094 | if (ipt->error) | |
5095 | cancel = true; | |
5096 | ipt->error = 0; | |
5097 | mask = 0; | |
5098 | } | |
5099 | if (mask || ipt->error) | |
5100 | list_del_init(&poll->wait.entry); | |
5101 | else if (cancel) | |
5102 | WRITE_ONCE(poll->canceled, true); | |
5103 | else if (!poll->done) /* actually waiting for an event */ | |
5104 | io_poll_req_insert(req); | |
5105 | spin_unlock(&poll->head->lock); | |
5106 | } | |
5107 | ||
5108 | return mask; | |
5109 | } | |
5110 | ||
5111 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5112 | { | |
5113 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5114 | struct io_ring_ctx *ctx = req->ctx; | |
5115 | struct async_poll *apoll; | |
5116 | struct io_poll_table ipt; | |
5117 | __poll_t mask, ret; | |
9dab14b8 | 5118 | int rw; |
d7718a9d JA |
5119 | |
5120 | if (!req->file || !file_can_poll(req->file)) | |
5121 | return false; | |
24c74678 | 5122 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5123 | return false; |
9dab14b8 JA |
5124 | if (def->pollin) |
5125 | rw = READ; | |
5126 | else if (def->pollout) | |
5127 | rw = WRITE; | |
5128 | else | |
5129 | return false; | |
5130 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5131 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5132 | return false; |
5133 | ||
5134 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5135 | if (unlikely(!apoll)) | |
5136 | return false; | |
807abcb0 | 5137 | apoll->double_poll = NULL; |
d7718a9d JA |
5138 | |
5139 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5140 | req->apoll = apoll; |
d7718a9d | 5141 | |
8755d97a | 5142 | mask = 0; |
d7718a9d | 5143 | if (def->pollin) |
8755d97a | 5144 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5145 | if (def->pollout) |
5146 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5147 | |
5148 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5149 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5150 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5151 | mask &= ~POLLIN; | |
5152 | ||
d7718a9d JA |
5153 | mask |= POLLERR | POLLPRI; |
5154 | ||
5155 | ipt.pt._qproc = io_async_queue_proc; | |
5156 | ||
5157 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5158 | io_async_wake); | |
a36da65c | 5159 | if (ret || ipt.error) { |
d4e7cd36 | 5160 | io_poll_remove_double(req); |
d7718a9d | 5161 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5162 | kfree(apoll->double_poll); |
d7718a9d JA |
5163 | kfree(apoll); |
5164 | return false; | |
5165 | } | |
5166 | spin_unlock_irq(&ctx->completion_lock); | |
5167 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5168 | apoll->poll.events); | |
5169 | return true; | |
5170 | } | |
5171 | ||
5172 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5173 | struct io_poll_iocb *poll) | |
221c5eb2 | 5174 | { |
b41e9852 | 5175 | bool do_complete = false; |
221c5eb2 JA |
5176 | |
5177 | spin_lock(&poll->head->lock); | |
5178 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5179 | if (!list_empty(&poll->wait.entry)) { |
5180 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5181 | do_complete = true; |
221c5eb2 JA |
5182 | } |
5183 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5184 | hash_del(&req->hash_node); |
d7718a9d JA |
5185 | return do_complete; |
5186 | } | |
5187 | ||
5188 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5189 | { | |
5190 | bool do_complete; | |
5191 | ||
d4e7cd36 JA |
5192 | io_poll_remove_double(req); |
5193 | ||
d7718a9d JA |
5194 | if (req->opcode == IORING_OP_POLL_ADD) { |
5195 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5196 | } else { | |
3bfa5bcb JA |
5197 | struct async_poll *apoll = req->apoll; |
5198 | ||
d7718a9d | 5199 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5200 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5201 | if (do_complete) { | |
d7718a9d | 5202 | io_put_req(req); |
807abcb0 | 5203 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5204 | kfree(apoll); |
5205 | } | |
b1f573bd XW |
5206 | } |
5207 | ||
b41e9852 JA |
5208 | if (do_complete) { |
5209 | io_cqring_fill_event(req, -ECANCELED); | |
5210 | io_commit_cqring(req->ctx); | |
f254ac04 | 5211 | req_set_fail_links(req); |
216578e5 | 5212 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5213 | } |
5214 | ||
5215 | return do_complete; | |
221c5eb2 JA |
5216 | } |
5217 | ||
76e1b642 JA |
5218 | /* |
5219 | * Returns true if we found and killed one or more poll requests | |
5220 | */ | |
6b81928d PB |
5221 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5222 | struct files_struct *files) | |
221c5eb2 | 5223 | { |
78076bb6 | 5224 | struct hlist_node *tmp; |
221c5eb2 | 5225 | struct io_kiocb *req; |
8e2e1faf | 5226 | int posted = 0, i; |
221c5eb2 JA |
5227 | |
5228 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5229 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5230 | struct hlist_head *list; | |
5231 | ||
5232 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5233 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5234 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5235 | posted += io_poll_remove_one(req); |
5236 | } | |
221c5eb2 JA |
5237 | } |
5238 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5239 | |
8e2e1faf JA |
5240 | if (posted) |
5241 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5242 | |
5243 | return posted != 0; | |
221c5eb2 JA |
5244 | } |
5245 | ||
47f46768 JA |
5246 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5247 | { | |
78076bb6 | 5248 | struct hlist_head *list; |
47f46768 JA |
5249 | struct io_kiocb *req; |
5250 | ||
78076bb6 JA |
5251 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5252 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5253 | if (sqe_addr != req->user_data) |
5254 | continue; | |
5255 | if (io_poll_remove_one(req)) | |
eac406c6 | 5256 | return 0; |
b41e9852 | 5257 | return -EALREADY; |
47f46768 JA |
5258 | } |
5259 | ||
5260 | return -ENOENT; | |
5261 | } | |
5262 | ||
3529d8c2 JA |
5263 | static int io_poll_remove_prep(struct io_kiocb *req, |
5264 | const struct io_uring_sqe *sqe) | |
0969e783 | 5265 | { |
0969e783 JA |
5266 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5267 | return -EINVAL; | |
5268 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5269 | sqe->poll_events) | |
5270 | return -EINVAL; | |
5271 | ||
018043be | 5272 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5273 | return 0; |
5274 | } | |
5275 | ||
221c5eb2 JA |
5276 | /* |
5277 | * Find a running poll command that matches one specified in sqe->addr, | |
5278 | * and remove it if found. | |
5279 | */ | |
61e98203 | 5280 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5281 | { |
5282 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5283 | int ret; |
221c5eb2 | 5284 | |
221c5eb2 | 5285 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5286 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5287 | spin_unlock_irq(&ctx->completion_lock); |
5288 | ||
4e88d6e7 JA |
5289 | if (ret < 0) |
5290 | req_set_fail_links(req); | |
e1e16097 | 5291 | io_req_complete(req, ret); |
221c5eb2 JA |
5292 | return 0; |
5293 | } | |
5294 | ||
221c5eb2 JA |
5295 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5296 | void *key) | |
5297 | { | |
c2f2eb7d JA |
5298 | struct io_kiocb *req = wait->private; |
5299 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5300 | |
d7718a9d | 5301 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5302 | } |
5303 | ||
221c5eb2 JA |
5304 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5305 | struct poll_table_struct *p) | |
5306 | { | |
5307 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5308 | ||
e8c2bc1f | 5309 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5310 | } |
5311 | ||
3529d8c2 | 5312 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5313 | { |
5314 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5315 | u32 events; |
221c5eb2 JA |
5316 | |
5317 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5318 | return -EINVAL; | |
5319 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5320 | return -EINVAL; | |
5321 | ||
5769a351 JX |
5322 | events = READ_ONCE(sqe->poll32_events); |
5323 | #ifdef __BIG_ENDIAN | |
5324 | events = swahw32(events); | |
5325 | #endif | |
a31eb4a2 JX |
5326 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5327 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5328 | return 0; |
5329 | } | |
5330 | ||
61e98203 | 5331 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5332 | { |
5333 | struct io_poll_iocb *poll = &req->poll; | |
5334 | struct io_ring_ctx *ctx = req->ctx; | |
5335 | struct io_poll_table ipt; | |
0969e783 | 5336 | __poll_t mask; |
0969e783 | 5337 | |
d7718a9d | 5338 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5339 | |
d7718a9d JA |
5340 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5341 | io_poll_wake); | |
221c5eb2 | 5342 | |
8c838788 | 5343 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5344 | ipt.error = 0; |
b0dd8a41 | 5345 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5346 | } |
221c5eb2 JA |
5347 | spin_unlock_irq(&ctx->completion_lock); |
5348 | ||
8c838788 JA |
5349 | if (mask) { |
5350 | io_cqring_ev_posted(ctx); | |
014db007 | 5351 | io_put_req(req); |
221c5eb2 | 5352 | } |
8c838788 | 5353 | return ipt.error; |
221c5eb2 JA |
5354 | } |
5355 | ||
5262f567 JA |
5356 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5357 | { | |
ad8a48ac JA |
5358 | struct io_timeout_data *data = container_of(timer, |
5359 | struct io_timeout_data, timer); | |
5360 | struct io_kiocb *req = data->req; | |
5361 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5362 | unsigned long flags; |
5363 | ||
5262f567 | 5364 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5365 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5366 | atomic_set(&req->ctx->cq_timeouts, |
5367 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5368 | ||
78e19bbe | 5369 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5370 | io_commit_cqring(ctx); |
5371 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5372 | ||
5373 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5374 | req_set_fail_links(req); |
5262f567 JA |
5375 | io_put_req(req); |
5376 | return HRTIMER_NORESTART; | |
5377 | } | |
5378 | ||
fbd15848 PB |
5379 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5380 | __u64 user_data) | |
f254ac04 | 5381 | { |
fbd15848 | 5382 | struct io_timeout_data *io; |
47f46768 JA |
5383 | struct io_kiocb *req; |
5384 | int ret = -ENOENT; | |
f254ac04 | 5385 | |
135fcde8 | 5386 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5387 | if (user_data == req->user_data) { |
47f46768 JA |
5388 | ret = 0; |
5389 | break; | |
5390 | } | |
5391 | } | |
5392 | ||
5393 | if (ret == -ENOENT) | |
fbd15848 PB |
5394 | return ERR_PTR(ret); |
5395 | ||
5396 | io = req->async_data; | |
e8c2bc1f | 5397 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5398 | if (ret == -1) |
fbd15848 | 5399 | return ERR_PTR(-EALREADY); |
a71976f3 | 5400 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5401 | return req; |
5402 | } | |
47f46768 | 5403 | |
fbd15848 PB |
5404 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5405 | { | |
5406 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5407 | ||
5408 | if (IS_ERR(req)) | |
5409 | return PTR_ERR(req); | |
f254ac04 JA |
5410 | |
5411 | req_set_fail_links(req); | |
f254ac04 | 5412 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5413 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5414 | return 0; |
5415 | } | |
5416 | ||
9c8e11b3 PB |
5417 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5418 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5419 | { |
9c8e11b3 PB |
5420 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5421 | struct io_timeout_data *data; | |
47f46768 | 5422 | |
9c8e11b3 PB |
5423 | if (IS_ERR(req)) |
5424 | return PTR_ERR(req); | |
47f46768 | 5425 | |
9c8e11b3 PB |
5426 | req->timeout.off = 0; /* noseq */ |
5427 | data = req->async_data; | |
5428 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5429 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5430 | data->timer.function = io_timeout_fn; | |
5431 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5432 | return 0; | |
47f46768 JA |
5433 | } |
5434 | ||
3529d8c2 JA |
5435 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5436 | const struct io_uring_sqe *sqe) | |
b29472ee | 5437 | { |
9c8e11b3 PB |
5438 | struct io_timeout_rem *tr = &req->timeout_rem; |
5439 | ||
b29472ee JA |
5440 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5441 | return -EINVAL; | |
61710e43 DA |
5442 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5443 | return -EINVAL; | |
9c8e11b3 | 5444 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5445 | return -EINVAL; |
5446 | ||
9c8e11b3 PB |
5447 | tr->addr = READ_ONCE(sqe->addr); |
5448 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5449 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5450 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5451 | return -EINVAL; | |
5452 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5453 | return -EFAULT; | |
5454 | } else if (tr->flags) { | |
5455 | /* timeout removal doesn't support flags */ | |
b29472ee | 5456 | return -EINVAL; |
9c8e11b3 | 5457 | } |
b29472ee | 5458 | |
b29472ee JA |
5459 | return 0; |
5460 | } | |
5461 | ||
8662daec PB |
5462 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5463 | { | |
5464 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5465 | : HRTIMER_MODE_REL; | |
5466 | } | |
5467 | ||
11365043 JA |
5468 | /* |
5469 | * Remove or update an existing timeout command | |
5470 | */ | |
61e98203 | 5471 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5472 | { |
9c8e11b3 | 5473 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5474 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5475 | int ret; |
11365043 | 5476 | |
11365043 | 5477 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5478 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5479 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5480 | else |
5481 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5482 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5483 | |
47f46768 | 5484 | io_cqring_fill_event(req, ret); |
11365043 JA |
5485 | io_commit_cqring(ctx); |
5486 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5487 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5488 | if (ret < 0) |
5489 | req_set_fail_links(req); | |
ec9c02ad | 5490 | io_put_req(req); |
11365043 | 5491 | return 0; |
5262f567 JA |
5492 | } |
5493 | ||
3529d8c2 | 5494 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5495 | bool is_timeout_link) |
5262f567 | 5496 | { |
ad8a48ac | 5497 | struct io_timeout_data *data; |
a41525ab | 5498 | unsigned flags; |
56080b02 | 5499 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5500 | |
ad8a48ac | 5501 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5502 | return -EINVAL; |
ad8a48ac | 5503 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5504 | return -EINVAL; |
56080b02 | 5505 | if (off && is_timeout_link) |
2d28390a | 5506 | return -EINVAL; |
a41525ab JA |
5507 | flags = READ_ONCE(sqe->timeout_flags); |
5508 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5509 | return -EINVAL; |
bdf20073 | 5510 | |
bfe68a22 | 5511 | req->timeout.off = off; |
26a61679 | 5512 | |
e8c2bc1f | 5513 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5514 | return -ENOMEM; |
5515 | ||
e8c2bc1f | 5516 | data = req->async_data; |
ad8a48ac | 5517 | data->req = req; |
ad8a48ac JA |
5518 | |
5519 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5520 | return -EFAULT; |
5521 | ||
8662daec | 5522 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5523 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
dd59a3d5 | 5524 | io_req_track_inflight(req); |
ad8a48ac JA |
5525 | return 0; |
5526 | } | |
5527 | ||
61e98203 | 5528 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5529 | { |
ad8a48ac | 5530 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5531 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5532 | struct list_head *entry; |
bfe68a22 | 5533 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5534 | |
733f5c95 | 5535 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5536 | |
5262f567 JA |
5537 | /* |
5538 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5539 | * timeout event to be satisfied. If it isn't set, then this is |
5540 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5541 | */ |
8eb7e2d0 | 5542 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5543 | entry = ctx->timeout_list.prev; |
5544 | goto add; | |
5545 | } | |
5262f567 | 5546 | |
bfe68a22 PB |
5547 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5548 | req->timeout.target_seq = tail + off; | |
5262f567 | 5549 | |
f010505b MDG |
5550 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5551 | * This is safe because ->completion_lock is held, and submissions | |
5552 | * and completions are never mixed in the same ->completion_lock section. | |
5553 | */ | |
5554 | ctx->cq_last_tm_flush = tail; | |
5555 | ||
5262f567 JA |
5556 | /* |
5557 | * Insertion sort, ensuring the first entry in the list is always | |
5558 | * the one we need first. | |
5559 | */ | |
5262f567 | 5560 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5561 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5562 | timeout.list); | |
5262f567 | 5563 | |
8eb7e2d0 | 5564 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5565 | continue; |
bfe68a22 PB |
5566 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5567 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5568 | break; |
5569 | } | |
93bd25bb | 5570 | add: |
135fcde8 | 5571 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5572 | data->timer.function = io_timeout_fn; |
5573 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5574 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5575 | return 0; |
5576 | } | |
5262f567 | 5577 | |
62755e35 JA |
5578 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5579 | { | |
5580 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
5581 | ||
5582 | return req->user_data == (unsigned long) data; | |
5583 | } | |
5584 | ||
5aa75ed5 | 5585 | static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr) |
62755e35 | 5586 | { |
62755e35 | 5587 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5588 | int ret = 0; |
5589 | ||
5aa75ed5 JA |
5590 | if (!tctx->io_wq) |
5591 | return -ENOENT; | |
5592 | ||
5593 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false); | |
62755e35 JA |
5594 | switch (cancel_ret) { |
5595 | case IO_WQ_CANCEL_OK: | |
5596 | ret = 0; | |
5597 | break; | |
5598 | case IO_WQ_CANCEL_RUNNING: | |
5599 | ret = -EALREADY; | |
5600 | break; | |
5601 | case IO_WQ_CANCEL_NOTFOUND: | |
5602 | ret = -ENOENT; | |
5603 | break; | |
5604 | } | |
5605 | ||
e977d6d3 JA |
5606 | return ret; |
5607 | } | |
5608 | ||
47f46768 JA |
5609 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5610 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5611 | int success_ret) |
47f46768 JA |
5612 | { |
5613 | unsigned long flags; | |
5614 | int ret; | |
5615 | ||
5aa75ed5 JA |
5616 | ret = io_async_cancel_one(req->task->io_uring, |
5617 | (void *) (unsigned long) sqe_addr); | |
47f46768 JA |
5618 | if (ret != -ENOENT) { |
5619 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5620 | goto done; | |
5621 | } | |
5622 | ||
5623 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5624 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5625 | if (ret != -ENOENT) | |
5626 | goto done; | |
5627 | ret = io_poll_cancel(ctx, sqe_addr); | |
5628 | done: | |
b0dd8a41 JA |
5629 | if (!ret) |
5630 | ret = success_ret; | |
47f46768 JA |
5631 | io_cqring_fill_event(req, ret); |
5632 | io_commit_cqring(ctx); | |
5633 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5634 | io_cqring_ev_posted(ctx); | |
5635 | ||
4e88d6e7 JA |
5636 | if (ret < 0) |
5637 | req_set_fail_links(req); | |
014db007 | 5638 | io_put_req(req); |
47f46768 JA |
5639 | } |
5640 | ||
3529d8c2 JA |
5641 | static int io_async_cancel_prep(struct io_kiocb *req, |
5642 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5643 | { |
fbf23849 | 5644 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5645 | return -EINVAL; |
61710e43 DA |
5646 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5647 | return -EINVAL; | |
5648 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5649 | return -EINVAL; |
5650 | ||
fbf23849 JA |
5651 | req->cancel.addr = READ_ONCE(sqe->addr); |
5652 | return 0; | |
5653 | } | |
5654 | ||
61e98203 | 5655 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5656 | { |
5657 | struct io_ring_ctx *ctx = req->ctx; | |
fbf23849 | 5658 | |
014db007 | 5659 | io_async_find_and_cancel(ctx, req, req->cancel.addr, 0); |
5262f567 JA |
5660 | return 0; |
5661 | } | |
5662 | ||
269bbe5f | 5663 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5664 | const struct io_uring_sqe *sqe) |
5665 | { | |
6ca56f84 JA |
5666 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5667 | return -EINVAL; | |
61710e43 DA |
5668 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5669 | return -EINVAL; | |
5670 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5671 | return -EINVAL; |
5672 | ||
269bbe5f BM |
5673 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5674 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5675 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5676 | return -EINVAL; |
269bbe5f | 5677 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5678 | return 0; |
5679 | } | |
5680 | ||
889fca73 | 5681 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5682 | { |
5683 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5684 | struct io_uring_rsrc_update up; |
05f3fb3c | 5685 | int ret; |
fbf23849 | 5686 | |
45d189c6 | 5687 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5688 | return -EAGAIN; |
05f3fb3c | 5689 | |
269bbe5f BM |
5690 | up.offset = req->rsrc_update.offset; |
5691 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5692 | |
5693 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5694 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5695 | mutex_unlock(&ctx->uring_lock); |
5696 | ||
5697 | if (ret < 0) | |
5698 | req_set_fail_links(req); | |
889fca73 | 5699 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5700 | return 0; |
5701 | } | |
5702 | ||
bfe76559 | 5703 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5704 | { |
d625c6ee | 5705 | switch (req->opcode) { |
e781573e | 5706 | case IORING_OP_NOP: |
bfe76559 | 5707 | return 0; |
f67676d1 JA |
5708 | case IORING_OP_READV: |
5709 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5710 | case IORING_OP_READ: |
bfe76559 | 5711 | return io_read_prep(req, sqe); |
f67676d1 JA |
5712 | case IORING_OP_WRITEV: |
5713 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5714 | case IORING_OP_WRITE: |
bfe76559 | 5715 | return io_write_prep(req, sqe); |
0969e783 | 5716 | case IORING_OP_POLL_ADD: |
bfe76559 | 5717 | return io_poll_add_prep(req, sqe); |
0969e783 | 5718 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5719 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5720 | case IORING_OP_FSYNC: |
1155c76a | 5721 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5722 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5723 | return io_sfr_prep(req, sqe); |
03b1230c | 5724 | case IORING_OP_SENDMSG: |
fddaface | 5725 | case IORING_OP_SEND: |
bfe76559 | 5726 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5727 | case IORING_OP_RECVMSG: |
fddaface | 5728 | case IORING_OP_RECV: |
bfe76559 | 5729 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5730 | case IORING_OP_CONNECT: |
bfe76559 | 5731 | return io_connect_prep(req, sqe); |
2d28390a | 5732 | case IORING_OP_TIMEOUT: |
bfe76559 | 5733 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5734 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5735 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5736 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5737 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5738 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5739 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5740 | case IORING_OP_ACCEPT: |
bfe76559 | 5741 | return io_accept_prep(req, sqe); |
d63d1b5e | 5742 | case IORING_OP_FALLOCATE: |
bfe76559 | 5743 | return io_fallocate_prep(req, sqe); |
15b71abe | 5744 | case IORING_OP_OPENAT: |
bfe76559 | 5745 | return io_openat_prep(req, sqe); |
b5dba59e | 5746 | case IORING_OP_CLOSE: |
bfe76559 | 5747 | return io_close_prep(req, sqe); |
05f3fb3c | 5748 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5749 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5750 | case IORING_OP_STATX: |
bfe76559 | 5751 | return io_statx_prep(req, sqe); |
4840e418 | 5752 | case IORING_OP_FADVISE: |
bfe76559 | 5753 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5754 | case IORING_OP_MADVISE: |
bfe76559 | 5755 | return io_madvise_prep(req, sqe); |
cebdb986 | 5756 | case IORING_OP_OPENAT2: |
bfe76559 | 5757 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5758 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5759 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5760 | case IORING_OP_SPLICE: |
bfe76559 | 5761 | return io_splice_prep(req, sqe); |
ddf0322d | 5762 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5763 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5764 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5765 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5766 | case IORING_OP_TEE: |
bfe76559 | 5767 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5768 | case IORING_OP_SHUTDOWN: |
5769 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5770 | case IORING_OP_RENAMEAT: |
5771 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5772 | case IORING_OP_UNLINKAT: |
5773 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5774 | } |
5775 | ||
bfe76559 PB |
5776 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5777 | req->opcode); | |
5778 | return-EINVAL; | |
5779 | } | |
5780 | ||
93642ef8 | 5781 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5782 | { |
93642ef8 PB |
5783 | switch (req->opcode) { |
5784 | case IORING_OP_READV: | |
5785 | case IORING_OP_READ_FIXED: | |
5786 | case IORING_OP_READ: | |
5787 | return io_rw_prep_async(req, READ); | |
5788 | case IORING_OP_WRITEV: | |
5789 | case IORING_OP_WRITE_FIXED: | |
5790 | case IORING_OP_WRITE: | |
5791 | return io_rw_prep_async(req, WRITE); | |
5792 | case IORING_OP_SENDMSG: | |
5793 | case IORING_OP_SEND: | |
5794 | return io_sendmsg_prep_async(req); | |
5795 | case IORING_OP_RECVMSG: | |
5796 | case IORING_OP_RECV: | |
5797 | return io_recvmsg_prep_async(req); | |
5798 | case IORING_OP_CONNECT: | |
5799 | return io_connect_prep_async(req); | |
5800 | } | |
5801 | return 0; | |
5802 | } | |
5803 | ||
be7053b7 | 5804 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5805 | { |
be7053b7 | 5806 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5807 | return 0; |
be7053b7 | 5808 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5809 | if (req->async_data) |
be7053b7 PB |
5810 | return 0; |
5811 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5812 | return -EAGAIN; |
be7053b7 | 5813 | return io_req_prep_async(req); |
f67676d1 JA |
5814 | } |
5815 | ||
9cf7c104 PB |
5816 | static u32 io_get_sequence(struct io_kiocb *req) |
5817 | { | |
5818 | struct io_kiocb *pos; | |
5819 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5820 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5821 | |
f2f87370 PB |
5822 | io_for_each_link(pos, req) |
5823 | nr_reqs++; | |
9cf7c104 PB |
5824 | |
5825 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5826 | return total_submitted - nr_reqs; | |
5827 | } | |
5828 | ||
be7053b7 | 5829 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5830 | { |
a197f664 | 5831 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5832 | struct io_defer_entry *de; |
f67676d1 | 5833 | int ret; |
9cf7c104 | 5834 | u32 seq; |
de0617e4 | 5835 | |
9d858b21 | 5836 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5837 | if (likely(list_empty_careful(&ctx->defer_list) && |
5838 | !(req->flags & REQ_F_IO_DRAIN))) | |
5839 | return 0; | |
5840 | ||
5841 | seq = io_get_sequence(req); | |
5842 | /* Still a chance to pass the sequence check */ | |
5843 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5844 | return 0; |
5845 | ||
be7053b7 PB |
5846 | ret = io_req_defer_prep(req); |
5847 | if (ret) | |
5848 | return ret; | |
cbdcb435 | 5849 | io_prep_async_link(req); |
27dc8338 PB |
5850 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5851 | if (!de) | |
5852 | return -ENOMEM; | |
2d28390a | 5853 | |
de0617e4 | 5854 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5855 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5856 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5857 | kfree(de); |
ae34817b PB |
5858 | io_queue_async_work(req); |
5859 | return -EIOCBQUEUED; | |
de0617e4 JA |
5860 | } |
5861 | ||
915967f6 | 5862 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5863 | de->req = req; |
9cf7c104 | 5864 | de->seq = seq; |
27dc8338 | 5865 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5866 | spin_unlock_irq(&ctx->completion_lock); |
5867 | return -EIOCBQUEUED; | |
5868 | } | |
5869 | ||
3ca405eb | 5870 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5871 | { |
0e1b6fe3 PB |
5872 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5873 | switch (req->opcode) { | |
5874 | case IORING_OP_READV: | |
5875 | case IORING_OP_READ_FIXED: | |
5876 | case IORING_OP_READ: | |
bcda7baa | 5877 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5878 | break; |
5879 | case IORING_OP_RECVMSG: | |
5880 | case IORING_OP_RECV: | |
bcda7baa | 5881 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5882 | break; |
5883 | } | |
5884 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5885 | } |
5886 | ||
0e1b6fe3 PB |
5887 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5888 | switch (req->opcode) { | |
5889 | case IORING_OP_READV: | |
5890 | case IORING_OP_READ_FIXED: | |
5891 | case IORING_OP_READ: | |
5892 | case IORING_OP_WRITEV: | |
5893 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5894 | case IORING_OP_WRITE: { |
5895 | struct io_async_rw *io = req->async_data; | |
5896 | if (io->free_iovec) | |
5897 | kfree(io->free_iovec); | |
0e1b6fe3 | 5898 | break; |
e8c2bc1f | 5899 | } |
0e1b6fe3 | 5900 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5901 | case IORING_OP_SENDMSG: { |
5902 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
5903 | |
5904 | kfree(io->free_iov); | |
0e1b6fe3 | 5905 | break; |
e8c2bc1f | 5906 | } |
0e1b6fe3 PB |
5907 | case IORING_OP_SPLICE: |
5908 | case IORING_OP_TEE: | |
5909 | io_put_file(req, req->splice.file_in, | |
5910 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
5911 | break; | |
f3cd4850 JA |
5912 | case IORING_OP_OPENAT: |
5913 | case IORING_OP_OPENAT2: | |
5914 | if (req->open.filename) | |
5915 | putname(req->open.filename); | |
5916 | break; | |
80a261fd JA |
5917 | case IORING_OP_RENAMEAT: |
5918 | putname(req->rename.oldpath); | |
5919 | putname(req->rename.newpath); | |
5920 | break; | |
14a1143b JA |
5921 | case IORING_OP_UNLINKAT: |
5922 | putname(req->unlink.filename); | |
5923 | break; | |
0e1b6fe3 PB |
5924 | } |
5925 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 5926 | } |
99bc4c38 PB |
5927 | } |
5928 | ||
889fca73 | 5929 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 5930 | { |
a197f664 | 5931 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 5932 | const struct cred *creds = NULL; |
d625c6ee | 5933 | int ret; |
2b188cc1 | 5934 | |
003e8dcc JA |
5935 | if (req->work.creds && req->work.creds != current_cred()) |
5936 | creds = override_creds(req->work.creds); | |
5730b27e | 5937 | |
d625c6ee | 5938 | switch (req->opcode) { |
2b188cc1 | 5939 | case IORING_OP_NOP: |
889fca73 | 5940 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
5941 | break; |
5942 | case IORING_OP_READV: | |
edafccee | 5943 | case IORING_OP_READ_FIXED: |
3a6820f2 | 5944 | case IORING_OP_READ: |
889fca73 | 5945 | ret = io_read(req, issue_flags); |
edafccee | 5946 | break; |
3529d8c2 | 5947 | case IORING_OP_WRITEV: |
edafccee | 5948 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 5949 | case IORING_OP_WRITE: |
889fca73 | 5950 | ret = io_write(req, issue_flags); |
2b188cc1 | 5951 | break; |
c992fe29 | 5952 | case IORING_OP_FSYNC: |
45d189c6 | 5953 | ret = io_fsync(req, issue_flags); |
c992fe29 | 5954 | break; |
221c5eb2 | 5955 | case IORING_OP_POLL_ADD: |
61e98203 | 5956 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
5957 | break; |
5958 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 5959 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 5960 | break; |
5d17b4a4 | 5961 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 5962 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 5963 | break; |
0fa03c62 | 5964 | case IORING_OP_SENDMSG: |
889fca73 | 5965 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 5966 | break; |
fddaface | 5967 | case IORING_OP_SEND: |
889fca73 | 5968 | ret = io_send(req, issue_flags); |
0fa03c62 | 5969 | break; |
aa1fa28f | 5970 | case IORING_OP_RECVMSG: |
889fca73 | 5971 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 5972 | break; |
fddaface | 5973 | case IORING_OP_RECV: |
889fca73 | 5974 | ret = io_recv(req, issue_flags); |
aa1fa28f | 5975 | break; |
5262f567 | 5976 | case IORING_OP_TIMEOUT: |
61e98203 | 5977 | ret = io_timeout(req, issue_flags); |
5262f567 | 5978 | break; |
11365043 | 5979 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 5980 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 5981 | break; |
17f2fe35 | 5982 | case IORING_OP_ACCEPT: |
889fca73 | 5983 | ret = io_accept(req, issue_flags); |
17f2fe35 | 5984 | break; |
f8e85cf2 | 5985 | case IORING_OP_CONNECT: |
889fca73 | 5986 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 5987 | break; |
62755e35 | 5988 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 5989 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 5990 | break; |
d63d1b5e | 5991 | case IORING_OP_FALLOCATE: |
45d189c6 | 5992 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 5993 | break; |
15b71abe | 5994 | case IORING_OP_OPENAT: |
45d189c6 | 5995 | ret = io_openat(req, issue_flags); |
15b71abe | 5996 | break; |
b5dba59e | 5997 | case IORING_OP_CLOSE: |
889fca73 | 5998 | ret = io_close(req, issue_flags); |
b5dba59e | 5999 | break; |
05f3fb3c | 6000 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6001 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6002 | break; |
eddc7ef5 | 6003 | case IORING_OP_STATX: |
45d189c6 | 6004 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6005 | break; |
4840e418 | 6006 | case IORING_OP_FADVISE: |
45d189c6 | 6007 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6008 | break; |
c1ca757b | 6009 | case IORING_OP_MADVISE: |
45d189c6 | 6010 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6011 | break; |
cebdb986 | 6012 | case IORING_OP_OPENAT2: |
45d189c6 | 6013 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6014 | break; |
3e4827b0 | 6015 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6016 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6017 | break; |
7d67af2c | 6018 | case IORING_OP_SPLICE: |
45d189c6 | 6019 | ret = io_splice(req, issue_flags); |
7d67af2c | 6020 | break; |
ddf0322d | 6021 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6022 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6023 | break; |
067524e9 | 6024 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6025 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6026 | break; |
f2a8d5c7 | 6027 | case IORING_OP_TEE: |
45d189c6 | 6028 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6029 | break; |
36f4fa68 | 6030 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6031 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6032 | break; |
80a261fd | 6033 | case IORING_OP_RENAMEAT: |
45d189c6 | 6034 | ret = io_renameat(req, issue_flags); |
80a261fd | 6035 | break; |
14a1143b | 6036 | case IORING_OP_UNLINKAT: |
45d189c6 | 6037 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6038 | break; |
2b188cc1 JA |
6039 | default: |
6040 | ret = -EINVAL; | |
6041 | break; | |
6042 | } | |
6043 | ||
5730b27e JA |
6044 | if (creds) |
6045 | revert_creds(creds); | |
6046 | ||
def596e9 JA |
6047 | if (ret) |
6048 | return ret; | |
6049 | ||
b532576e JA |
6050 | /* If the op doesn't have a file, we're not polling for it */ |
6051 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6052 | const bool in_async = io_wq_current_is_worker(); |
6053 | ||
11ba820b JA |
6054 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6055 | if (in_async) | |
6056 | mutex_lock(&ctx->uring_lock); | |
6057 | ||
2e9dbe90 | 6058 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6059 | |
6060 | if (in_async) | |
6061 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6062 | } |
6063 | ||
6064 | return 0; | |
2b188cc1 JA |
6065 | } |
6066 | ||
5280f7e5 | 6067 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6068 | { |
6069 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6070 | struct io_kiocb *timeout; |
561fb04a | 6071 | int ret = 0; |
2b188cc1 | 6072 | |
6df1db6b PB |
6073 | timeout = io_prep_linked_timeout(req); |
6074 | if (timeout) | |
6075 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6076 | |
4014d943 | 6077 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6078 | ret = -ECANCELED; |
31b51510 | 6079 | |
561fb04a | 6080 | if (!ret) { |
561fb04a | 6081 | do { |
889fca73 | 6082 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6083 | /* |
6084 | * We can get EAGAIN for polled IO even though we're | |
6085 | * forcing a sync submission from here, since we can't | |
6086 | * wait for request slots on the block side. | |
6087 | */ | |
6088 | if (ret != -EAGAIN) | |
6089 | break; | |
6090 | cond_resched(); | |
6091 | } while (1); | |
6092 | } | |
31b51510 | 6093 | |
a3df7698 | 6094 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6095 | if (ret) { |
a3df7698 PB |
6096 | /* io-wq is going to take one down */ |
6097 | refcount_inc(&req->refs); | |
6098 | io_req_task_queue_fail(req, ret); | |
edafccee | 6099 | } |
2b188cc1 JA |
6100 | } |
6101 | ||
65e19f54 JA |
6102 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6103 | int index) | |
6104 | { | |
269bbe5f | 6105 | struct fixed_rsrc_table *table; |
65e19f54 | 6106 | |
05f3fb3c | 6107 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6108 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6109 | } |
6110 | ||
8371adf5 PB |
6111 | static struct file *io_file_get(struct io_submit_state *state, |
6112 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6113 | { |
a197f664 | 6114 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6115 | struct file *file; |
09bb8394 | 6116 | |
8da11c19 | 6117 | if (fixed) { |
479f517b | 6118 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6119 | return NULL; |
b7620121 | 6120 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6121 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6122 | io_set_resource_node(req); |
09bb8394 | 6123 | } else { |
c826bd7a | 6124 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6125 | file = __io_file_get(state, fd); |
09bb8394 JA |
6126 | } |
6127 | ||
ce3d5aae PB |
6128 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6129 | io_req_track_inflight(req); | |
8371adf5 | 6130 | return file; |
09bb8394 JA |
6131 | } |
6132 | ||
2665abfd | 6133 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6134 | { |
ad8a48ac JA |
6135 | struct io_timeout_data *data = container_of(timer, |
6136 | struct io_timeout_data, timer); | |
90cd7e42 | 6137 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6138 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6139 | unsigned long flags; |
2665abfd JA |
6140 | |
6141 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6142 | prev = req->timeout.head; |
6143 | req->timeout.head = NULL; | |
2665abfd JA |
6144 | |
6145 | /* | |
6146 | * We don't expect the list to be empty, that will only happen if we | |
6147 | * race with the completion of the linked work. | |
6148 | */ | |
90cd7e42 | 6149 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6150 | io_remove_next_linked(prev); |
90cd7e42 PB |
6151 | else |
6152 | prev = NULL; | |
2665abfd JA |
6153 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6154 | ||
6155 | if (prev) { | |
4e88d6e7 | 6156 | req_set_fail_links(prev); |
014db007 | 6157 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6158 | io_put_req_deferred(prev, 1); |
47f46768 | 6159 | } else { |
9ae1f8dd PB |
6160 | io_req_complete_post(req, -ETIME, 0); |
6161 | io_put_req_deferred(req, 1); | |
2665abfd | 6162 | } |
2665abfd JA |
6163 | return HRTIMER_NORESTART; |
6164 | } | |
6165 | ||
7271ef3a | 6166 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6167 | { |
76a46e06 | 6168 | /* |
f2f87370 PB |
6169 | * If the back reference is NULL, then our linked request finished |
6170 | * before we got a chance to setup the timer | |
76a46e06 | 6171 | */ |
90cd7e42 | 6172 | if (req->timeout.head) { |
e8c2bc1f | 6173 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6174 | |
ad8a48ac JA |
6175 | data->timer.function = io_link_timeout_fn; |
6176 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6177 | data->mode); | |
2665abfd | 6178 | } |
7271ef3a JA |
6179 | } |
6180 | ||
6181 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6182 | { | |
6183 | struct io_ring_ctx *ctx = req->ctx; | |
6184 | ||
6185 | spin_lock_irq(&ctx->completion_lock); | |
6186 | __io_queue_linked_timeout(req); | |
76a46e06 | 6187 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6188 | |
2665abfd | 6189 | /* drop submission reference */ |
76a46e06 JA |
6190 | io_put_req(req); |
6191 | } | |
2665abfd | 6192 | |
ad8a48ac | 6193 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6194 | { |
f2f87370 | 6195 | struct io_kiocb *nxt = req->link; |
2665abfd | 6196 | |
f2f87370 PB |
6197 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6198 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6199 | return NULL; |
2665abfd | 6200 | |
90cd7e42 | 6201 | nxt->timeout.head = req; |
900fad45 | 6202 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6203 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6204 | return nxt; |
2665abfd JA |
6205 | } |
6206 | ||
c5eef2b9 | 6207 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6208 | { |
d3d7298d | 6209 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6210 | int ret; |
2b188cc1 | 6211 | |
c5eef2b9 | 6212 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6213 | |
491381ce JA |
6214 | /* |
6215 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6216 | * doesn't support non-blocking read/write attempts | |
6217 | */ | |
24c74678 | 6218 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6219 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6220 | /* |
6221 | * Queued up for async execution, worker will release | |
6222 | * submit reference when the iocb is actually submitted. | |
6223 | */ | |
6224 | io_queue_async_work(req); | |
2b188cc1 | 6225 | } |
0d63c148 PB |
6226 | } else if (likely(!ret)) { |
6227 | /* drop submission reference */ | |
e342c807 | 6228 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6229 | struct io_ring_ctx *ctx = req->ctx; |
6230 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6231 | |
6dd0be1e | 6232 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6233 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6234 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6235 | } else { |
d3d7298d | 6236 | io_put_req(req); |
0d63c148 PB |
6237 | } |
6238 | } else { | |
4e88d6e7 | 6239 | req_set_fail_links(req); |
e65ef56d | 6240 | io_put_req(req); |
e1e16097 | 6241 | io_req_complete(req, ret); |
9e645e11 | 6242 | } |
d3d7298d PB |
6243 | if (linked_timeout) |
6244 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6245 | } |
6246 | ||
be7053b7 | 6247 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6248 | { |
6249 | int ret; | |
6250 | ||
be7053b7 | 6251 | ret = io_req_defer(req); |
4fe2c963 JL |
6252 | if (ret) { |
6253 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6254 | fail_req: |
4e88d6e7 | 6255 | req_set_fail_links(req); |
e1e16097 JA |
6256 | io_put_req(req); |
6257 | io_req_complete(req, ret); | |
4fe2c963 | 6258 | } |
2550878f | 6259 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6260 | ret = io_req_defer_prep(req); |
6261 | if (unlikely(ret)) | |
6262 | goto fail_req; | |
ce35a47a JA |
6263 | io_queue_async_work(req); |
6264 | } else { | |
c5eef2b9 | 6265 | __io_queue_sqe(req); |
ce35a47a | 6266 | } |
4fe2c963 JL |
6267 | } |
6268 | ||
b16fed66 PB |
6269 | /* |
6270 | * Check SQE restrictions (opcode and flags). | |
6271 | * | |
6272 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6273 | */ | |
6274 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6275 | struct io_kiocb *req, | |
6276 | unsigned int sqe_flags) | |
4fe2c963 | 6277 | { |
b16fed66 PB |
6278 | if (!ctx->restricted) |
6279 | return true; | |
6280 | ||
6281 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6282 | return false; | |
6283 | ||
6284 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6285 | ctx->restrictions.sqe_flags_required) | |
6286 | return false; | |
6287 | ||
6288 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6289 | ctx->restrictions.sqe_flags_required)) | |
6290 | return false; | |
6291 | ||
6292 | return true; | |
4fe2c963 JL |
6293 | } |
6294 | ||
b16fed66 PB |
6295 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6296 | const struct io_uring_sqe *sqe) | |
6297 | { | |
6298 | struct io_submit_state *state; | |
6299 | unsigned int sqe_flags; | |
003e8dcc | 6300 | int personality, ret = 0; |
b16fed66 PB |
6301 | |
6302 | req->opcode = READ_ONCE(sqe->opcode); | |
6303 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6304 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6305 | req->user_data = READ_ONCE(sqe->user_data); | |
6306 | req->async_data = NULL; | |
6307 | req->file = NULL; | |
6308 | req->ctx = ctx; | |
6309 | req->link = NULL; | |
6310 | req->fixed_rsrc_refs = NULL; | |
6311 | /* one is dropped after submission, the other at completion */ | |
6312 | refcount_set(&req->refs, 2); | |
6313 | req->task = current; | |
6314 | req->result = 0; | |
6315 | ||
6316 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6317 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6318 | req->flags = 0; | |
b16fed66 | 6319 | return -EINVAL; |
ebf4a5db | 6320 | } |
b16fed66 PB |
6321 | |
6322 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6323 | return -EINVAL; | |
6324 | ||
b16fed66 PB |
6325 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6326 | return -EACCES; | |
6327 | ||
6328 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6329 | !io_op_defs[req->opcode].buffer_select) | |
6330 | return -EOPNOTSUPP; | |
863e0560 | 6331 | |
5730b27e | 6332 | req->work.list.next = NULL; |
003e8dcc JA |
6333 | personality = READ_ONCE(sqe->personality); |
6334 | if (personality) { | |
6335 | req->work.creds = idr_find(&ctx->personality_idr, personality); | |
6336 | if (!req->work.creds) | |
6337 | return -EINVAL; | |
6338 | get_cred(req->work.creds); | |
6339 | } else { | |
6340 | req->work.creds = NULL; | |
6341 | } | |
5730b27e | 6342 | req->work.flags = 0; |
b16fed66 PB |
6343 | state = &ctx->submit_state; |
6344 | ||
6345 | /* | |
6346 | * Plug now if we have more than 1 IO left after this, and the target | |
6347 | * is potentially a read/write to block based storage. | |
6348 | */ | |
6349 | if (!state->plug_started && state->ios_left > 1 && | |
6350 | io_op_defs[req->opcode].plug) { | |
6351 | blk_start_plug(&state->plug); | |
6352 | state->plug_started = true; | |
6353 | } | |
6354 | ||
6355 | if (io_op_defs[req->opcode].needs_file) { | |
6356 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6357 | ||
6358 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6359 | if (unlikely(!req->file)) | |
6360 | ret = -EBADF; | |
6361 | } | |
6362 | ||
6363 | state->ios_left--; | |
6364 | return ret; | |
6365 | } | |
6366 | ||
a6b8cadc | 6367 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6368 | const struct io_uring_sqe *sqe) |
9e645e11 | 6369 | { |
a1ab7b35 | 6370 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6371 | int ret; |
9e645e11 | 6372 | |
a6b8cadc PB |
6373 | ret = io_init_req(ctx, req, sqe); |
6374 | if (unlikely(ret)) { | |
6375 | fail_req: | |
6376 | io_put_req(req); | |
6377 | io_req_complete(req, ret); | |
de59bc10 PB |
6378 | if (link->head) { |
6379 | /* fail even hard links since we don't submit */ | |
cf109604 | 6380 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6381 | io_put_req(link->head); |
6382 | io_req_complete(link->head, -ECANCELED); | |
6383 | link->head = NULL; | |
6384 | } | |
a6b8cadc PB |
6385 | return ret; |
6386 | } | |
be7053b7 PB |
6387 | ret = io_req_prep(req, sqe); |
6388 | if (unlikely(ret)) | |
6389 | goto fail_req; | |
a6b8cadc | 6390 | |
be7053b7 | 6391 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6392 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6393 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6394 | ||
9e645e11 JA |
6395 | /* |
6396 | * If we already have a head request, queue this one for async | |
6397 | * submittal once the head completes. If we don't have a head but | |
6398 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6399 | * submitted sync once the chain is complete. If none of those | |
6400 | * conditions are true (normal request), then just queue it. | |
6401 | */ | |
863e0560 PB |
6402 | if (link->head) { |
6403 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6404 | |
8cdf2193 PB |
6405 | /* |
6406 | * Taking sequential execution of a link, draining both sides | |
6407 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6408 | * requests in the link. So, it drains the head and the | |
6409 | * next after the link request. The last one is done via | |
6410 | * drain_next flag to persist the effect across calls. | |
6411 | */ | |
ef4ff581 | 6412 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6413 | head->flags |= REQ_F_IO_DRAIN; |
6414 | ctx->drain_next = 1; | |
6415 | } | |
be7053b7 | 6416 | ret = io_req_defer_prep(req); |
cf109604 | 6417 | if (unlikely(ret)) |
a6b8cadc | 6418 | goto fail_req; |
9d76377f | 6419 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6420 | link->last->link = req; |
863e0560 | 6421 | link->last = req; |
32fe525b PB |
6422 | |
6423 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6424 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6425 | io_queue_sqe(head); |
863e0560 | 6426 | link->head = NULL; |
32fe525b | 6427 | } |
9e645e11 | 6428 | } else { |
711be031 PB |
6429 | if (unlikely(ctx->drain_next)) { |
6430 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6431 | ctx->drain_next = 0; |
711be031 | 6432 | } |
ef4ff581 | 6433 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6434 | link->head = req; |
6435 | link->last = req; | |
711be031 | 6436 | } else { |
be7053b7 | 6437 | io_queue_sqe(req); |
711be031 | 6438 | } |
9e645e11 | 6439 | } |
2e6e1fde | 6440 | |
1d4240cc | 6441 | return 0; |
9e645e11 JA |
6442 | } |
6443 | ||
9a56a232 JA |
6444 | /* |
6445 | * Batched submission is done, ensure local IO is flushed out. | |
6446 | */ | |
ba88ff11 PB |
6447 | static void io_submit_state_end(struct io_submit_state *state, |
6448 | struct io_ring_ctx *ctx) | |
9a56a232 | 6449 | { |
a1ab7b35 | 6450 | if (state->link.head) |
de59bc10 | 6451 | io_queue_sqe(state->link.head); |
6dd0be1e | 6452 | if (state->comp.nr) |
ba88ff11 | 6453 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6454 | if (state->plug_started) |
6455 | blk_finish_plug(&state->plug); | |
9f13c35b | 6456 | io_state_file_put(state); |
9a56a232 JA |
6457 | } |
6458 | ||
6459 | /* | |
6460 | * Start submission side cache. | |
6461 | */ | |
6462 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6463 | unsigned int max_ios) |
9a56a232 | 6464 | { |
27926b68 | 6465 | state->plug_started = false; |
9a56a232 | 6466 | state->ios_left = max_ios; |
a1ab7b35 PB |
6467 | /* set only head, no need to init link_last in advance */ |
6468 | state->link.head = NULL; | |
9a56a232 JA |
6469 | } |
6470 | ||
2b188cc1 JA |
6471 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6472 | { | |
75b28aff | 6473 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6474 | |
caf582c6 PB |
6475 | /* |
6476 | * Ensure any loads from the SQEs are done at this point, | |
6477 | * since once we write the new head, the application could | |
6478 | * write new data to them. | |
6479 | */ | |
6480 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6481 | } |
6482 | ||
2b188cc1 | 6483 | /* |
3529d8c2 | 6484 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6485 | * that is mapped by userspace. This means that care needs to be taken to |
6486 | * ensure that reads are stable, as we cannot rely on userspace always | |
6487 | * being a good citizen. If members of the sqe are validated and then later | |
6488 | * used, it's important that those reads are done through READ_ONCE() to | |
6489 | * prevent a re-load down the line. | |
6490 | */ | |
709b302f | 6491 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6492 | { |
75b28aff | 6493 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6494 | unsigned head; |
6495 | ||
6496 | /* | |
6497 | * The cached sq head (or cq tail) serves two purposes: | |
6498 | * | |
6499 | * 1) allows us to batch the cost of updating the user visible | |
6500 | * head updates. | |
6501 | * 2) allows the kernel side to track the head on its own, even | |
6502 | * though the application is the one updating it. | |
6503 | */ | |
4fccfcbb | 6504 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6505 | if (likely(head < ctx->sq_entries)) |
6506 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6507 | |
6508 | /* drop invalid entries */ | |
498ccd9e | 6509 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6510 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6511 | return NULL; |
6512 | } | |
6513 | ||
0f212204 | 6514 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6515 | { |
46c4e16a | 6516 | int submitted = 0; |
6c271ce2 | 6517 | |
c4a2ed72 | 6518 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6519 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6520 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6521 | return -EBUSY; |
6522 | } | |
6c271ce2 | 6523 | |
ee7d46d9 PB |
6524 | /* make sure SQ entry isn't read before tail */ |
6525 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6526 | |
2b85edfc PB |
6527 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6528 | return -EAGAIN; | |
6c271ce2 | 6529 | |
d8a6df10 | 6530 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6531 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6532 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6533 | |
46c4e16a | 6534 | while (submitted < nr) { |
3529d8c2 | 6535 | const struct io_uring_sqe *sqe; |
196be95c | 6536 | struct io_kiocb *req; |
fb5ccc98 | 6537 | |
258b29a9 | 6538 | req = io_alloc_req(ctx); |
196be95c PB |
6539 | if (unlikely(!req)) { |
6540 | if (!submitted) | |
6541 | submitted = -EAGAIN; | |
fb5ccc98 | 6542 | break; |
196be95c | 6543 | } |
4fccfcbb PB |
6544 | sqe = io_get_sqe(ctx); |
6545 | if (unlikely(!sqe)) { | |
6546 | kmem_cache_free(req_cachep, req); | |
6547 | break; | |
6548 | } | |
d3656344 JA |
6549 | /* will complete beyond this point, count as submitted */ |
6550 | submitted++; | |
a1ab7b35 | 6551 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6552 | break; |
6c271ce2 JA |
6553 | } |
6554 | ||
9466f437 PB |
6555 | if (unlikely(submitted != nr)) { |
6556 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6557 | struct io_uring_task *tctx = current->io_uring; |
6558 | int unused = nr - ref_used; | |
9466f437 | 6559 | |
d8a6df10 JA |
6560 | percpu_ref_put_many(&ctx->refs, unused); |
6561 | percpu_counter_sub(&tctx->inflight, unused); | |
6562 | put_task_struct_many(current, unused); | |
9466f437 | 6563 | } |
6c271ce2 | 6564 | |
a1ab7b35 | 6565 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6566 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6567 | io_commit_sqring(ctx); | |
6568 | ||
6c271ce2 JA |
6569 | return submitted; |
6570 | } | |
6571 | ||
23b3628e XW |
6572 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6573 | { | |
6574 | /* Tell userspace we may need a wakeup call */ | |
6575 | spin_lock_irq(&ctx->completion_lock); | |
6576 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6577 | spin_unlock_irq(&ctx->completion_lock); | |
6578 | } | |
6579 | ||
6580 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6581 | { | |
6582 | spin_lock_irq(&ctx->completion_lock); | |
6583 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6584 | spin_unlock_irq(&ctx->completion_lock); | |
6585 | } | |
6586 | ||
08369246 | 6587 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6588 | { |
c8d1ba58 | 6589 | unsigned int to_submit; |
bdcd3eab | 6590 | int ret = 0; |
6c271ce2 | 6591 | |
c8d1ba58 | 6592 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6593 | /* if we're handling multiple rings, cap submit size for fairness */ |
6594 | if (cap_entries && to_submit > 8) | |
6595 | to_submit = 8; | |
6596 | ||
906a3c6f | 6597 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6598 | unsigned nr_events = 0; |
a4c0b3de | 6599 | |
c8d1ba58 | 6600 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6601 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6602 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6603 | |
70aacfe6 | 6604 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs))) |
08369246 | 6605 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6606 | mutex_unlock(&ctx->uring_lock); |
6607 | } | |
6c271ce2 | 6608 | |
90554200 JA |
6609 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6610 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6611 | |
08369246 XW |
6612 | return ret; |
6613 | } | |
6c271ce2 | 6614 | |
08369246 XW |
6615 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6616 | { | |
6617 | struct io_ring_ctx *ctx; | |
6618 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6619 | |
08369246 XW |
6620 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6621 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6622 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6623 | } |
c1edbf5f | 6624 | |
08369246 | 6625 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6626 | } |
6c271ce2 | 6627 | |
69fb2131 JA |
6628 | static void io_sqd_init_new(struct io_sq_data *sqd) |
6629 | { | |
6630 | struct io_ring_ctx *ctx; | |
6631 | ||
6632 | while (!list_empty(&sqd->ctx_new_list)) { | |
6633 | ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); | |
69fb2131 JA |
6634 | list_move_tail(&ctx->sqd_list, &sqd->ctx_list); |
6635 | complete(&ctx->sq_thread_comp); | |
6636 | } | |
08369246 XW |
6637 | |
6638 | io_sqd_update_thread_idle(sqd); | |
69fb2131 JA |
6639 | } |
6640 | ||
37d1e2e3 JA |
6641 | static bool io_sq_thread_should_stop(struct io_sq_data *sqd) |
6642 | { | |
6643 | return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
6644 | } | |
6645 | ||
6646 | static bool io_sq_thread_should_park(struct io_sq_data *sqd) | |
6647 | { | |
6648 | return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
6649 | } | |
6650 | ||
6651 | static void io_sq_thread_parkme(struct io_sq_data *sqd) | |
6652 | { | |
6653 | for (;;) { | |
6654 | /* | |
6655 | * TASK_PARKED is a special state; we must serialize against | |
6656 | * possible pending wakeups to avoid store-store collisions on | |
6657 | * task->state. | |
6658 | * | |
6659 | * Such a collision might possibly result in the task state | |
6660 | * changin from TASK_PARKED and us failing the | |
6661 | * wait_task_inactive() in kthread_park(). | |
6662 | */ | |
6663 | set_special_state(TASK_PARKED); | |
6664 | if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) | |
6665 | break; | |
6666 | ||
6667 | /* | |
6668 | * Thread is going to call schedule(), do not preempt it, | |
6669 | * or the caller of kthread_park() may spend more time in | |
6670 | * wait_task_inactive(). | |
6671 | */ | |
6672 | preempt_disable(); | |
86e0d676 | 6673 | complete(&sqd->parked); |
37d1e2e3 JA |
6674 | schedule_preempt_disabled(); |
6675 | preempt_enable(); | |
6676 | } | |
6677 | __set_current_state(TASK_RUNNING); | |
6678 | } | |
6679 | ||
c8d1ba58 JA |
6680 | static int io_sq_thread(void *data) |
6681 | { | |
69fb2131 JA |
6682 | struct io_sq_data *sqd = data; |
6683 | struct io_ring_ctx *ctx; | |
a0d9205f | 6684 | unsigned long timeout = 0; |
37d1e2e3 | 6685 | char buf[TASK_COMM_LEN]; |
08369246 | 6686 | DEFINE_WAIT(wait); |
6c271ce2 | 6687 | |
37d1e2e3 JA |
6688 | sprintf(buf, "iou-sqp-%d", sqd->task_pid); |
6689 | set_task_comm(current, buf); | |
37d1e2e3 JA |
6690 | current->pf_io_worker = NULL; |
6691 | ||
6692 | if (sqd->sq_cpu != -1) | |
6693 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6694 | else | |
6695 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6696 | current->flags |= PF_NO_SETAFFINITY; | |
6697 | ||
37d1e2e3 JA |
6698 | wait_for_completion(&sqd->startup); |
6699 | ||
6700 | while (!io_sq_thread_should_stop(sqd)) { | |
08369246 XW |
6701 | int ret; |
6702 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f JA |
6703 | |
6704 | /* | |
69fb2131 | 6705 | * Any changes to the sqd lists are synchronized through the |
37d1e2e3 | 6706 | * thread parking. This synchronizes the thread vs users, |
69fb2131 | 6707 | * the users are synchronized on the sqd->ctx_lock. |
c1edbf5f | 6708 | */ |
37d1e2e3 JA |
6709 | if (io_sq_thread_should_park(sqd)) { |
6710 | io_sq_thread_parkme(sqd); | |
6711 | continue; | |
65b2b213 | 6712 | } |
08369246 | 6713 | if (unlikely(!list_empty(&sqd->ctx_new_list))) { |
69fb2131 | 6714 | io_sqd_init_new(sqd); |
08369246 XW |
6715 | timeout = jiffies + sqd->sq_thread_idle; |
6716 | } | |
37d1e2e3 JA |
6717 | if (fatal_signal_pending(current)) |
6718 | break; | |
08369246 | 6719 | sqt_spin = false; |
e95eee2d | 6720 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6721 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
08369246 XW |
6722 | ret = __io_sq_thread(ctx, cap_entries); |
6723 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) | |
6724 | sqt_spin = true; | |
69fb2131 | 6725 | } |
6c271ce2 | 6726 | |
08369246 | 6727 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6728 | io_run_task_work(); |
6729 | cond_resched(); | |
08369246 XW |
6730 | if (sqt_spin) |
6731 | timeout = jiffies + sqd->sq_thread_idle; | |
6732 | continue; | |
6733 | } | |
6734 | ||
08369246 XW |
6735 | needs_sched = true; |
6736 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6737 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6738 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6739 | !list_empty_careful(&ctx->iopoll_list)) { | |
6740 | needs_sched = false; | |
6741 | break; | |
6742 | } | |
6743 | if (io_sqring_entries(ctx)) { | |
6744 | needs_sched = false; | |
6745 | break; | |
6746 | } | |
6747 | } | |
6748 | ||
37d1e2e3 | 6749 | if (needs_sched && !io_sq_thread_should_park(sqd)) { |
69fb2131 JA |
6750 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6751 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6752 | |
69fb2131 | 6753 | schedule(); |
e4b4a13f | 6754 | try_to_freeze(); |
69fb2131 JA |
6755 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6756 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6757 | } |
08369246 XW |
6758 | |
6759 | finish_wait(&sqd->wait, &wait); | |
6760 | timeout = jiffies + sqd->sq_thread_idle; | |
6c271ce2 JA |
6761 | } |
6762 | ||
37d1e2e3 JA |
6763 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6764 | io_uring_cancel_sqpoll(ctx); | |
06058632 | 6765 | |
37d1e2e3 | 6766 | io_run_task_work(); |
28cea78a | 6767 | |
37d1e2e3 | 6768 | /* |
86e0d676 JA |
6769 | * Ensure that we park properly if racing with someone trying to park |
6770 | * while we're exiting. If we fail to grab the lock, check park and | |
6771 | * park if necessary. The ordering with the park bit and the lock | |
6772 | * ensures that we catch this reliably. | |
37d1e2e3 | 6773 | */ |
86e0d676 JA |
6774 | if (!mutex_trylock(&sqd->lock)) { |
6775 | if (io_sq_thread_should_park(sqd)) | |
6776 | io_sq_thread_parkme(sqd); | |
6777 | mutex_lock(&sqd->lock); | |
6778 | } | |
6779 | ||
37d1e2e3 | 6780 | sqd->thread = NULL; |
5f3f26f9 JA |
6781 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6782 | ctx->sqo_exec = 1; | |
6783 | io_ring_set_wakeup_flag(ctx); | |
6784 | } | |
06058632 | 6785 | |
37d1e2e3 | 6786 | complete(&sqd->exited); |
e54945ae | 6787 | mutex_unlock(&sqd->lock); |
37d1e2e3 | 6788 | do_exit(0); |
6c271ce2 JA |
6789 | } |
6790 | ||
bda52162 JA |
6791 | struct io_wait_queue { |
6792 | struct wait_queue_entry wq; | |
6793 | struct io_ring_ctx *ctx; | |
6794 | unsigned to_wait; | |
6795 | unsigned nr_timeouts; | |
6796 | }; | |
6797 | ||
6c503150 | 6798 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6799 | { |
6800 | struct io_ring_ctx *ctx = iowq->ctx; | |
6801 | ||
6802 | /* | |
d195a66e | 6803 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6804 | * started waiting. For timeouts, we always want to return to userspace, |
6805 | * regardless of event count. | |
6806 | */ | |
6c503150 | 6807 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6808 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6809 | } | |
6810 | ||
6811 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6812 | int wake_flags, void *key) | |
6813 | { | |
6814 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6815 | wq); | |
6816 | ||
6c503150 PB |
6817 | /* |
6818 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6819 | * the task, and the next invocation will do it. | |
6820 | */ | |
6821 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6822 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6823 | return -1; | |
bda52162 JA |
6824 | } |
6825 | ||
af9c1a44 JA |
6826 | static int io_run_task_work_sig(void) |
6827 | { | |
6828 | if (io_run_task_work()) | |
6829 | return 1; | |
6830 | if (!signal_pending(current)) | |
6831 | return 0; | |
792ee0f6 JA |
6832 | if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL)) |
6833 | return -ERESTARTSYS; | |
af9c1a44 JA |
6834 | return -EINTR; |
6835 | } | |
6836 | ||
eeb60b9a PB |
6837 | /* when returns >0, the caller should retry */ |
6838 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6839 | struct io_wait_queue *iowq, | |
6840 | signed long *timeout) | |
6841 | { | |
6842 | int ret; | |
6843 | ||
6844 | /* make sure we run task_work before checking for signals */ | |
6845 | ret = io_run_task_work_sig(); | |
6846 | if (ret || io_should_wake(iowq)) | |
6847 | return ret; | |
6848 | /* let the caller flush overflows, retry */ | |
6849 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6850 | return 1; | |
6851 | ||
6852 | *timeout = schedule_timeout(*timeout); | |
6853 | return !*timeout ? -ETIME : 1; | |
6854 | } | |
6855 | ||
2b188cc1 JA |
6856 | /* |
6857 | * Wait until events become available, if we don't already have some. The | |
6858 | * application must reap them itself, as they reside on the shared cq ring. | |
6859 | */ | |
6860 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6861 | const sigset_t __user *sig, size_t sigsz, |
6862 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6863 | { |
bda52162 JA |
6864 | struct io_wait_queue iowq = { |
6865 | .wq = { | |
6866 | .private = current, | |
6867 | .func = io_wake_function, | |
6868 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6869 | }, | |
6870 | .ctx = ctx, | |
6871 | .to_wait = min_events, | |
6872 | }; | |
75b28aff | 6873 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6874 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6875 | int ret; | |
2b188cc1 | 6876 | |
b41e9852 | 6877 | do { |
6c503150 PB |
6878 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6879 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6880 | return 0; |
4c6e277c | 6881 | if (!io_run_task_work()) |
b41e9852 | 6882 | break; |
b41e9852 | 6883 | } while (1); |
2b188cc1 JA |
6884 | |
6885 | if (sig) { | |
9e75ad5d AB |
6886 | #ifdef CONFIG_COMPAT |
6887 | if (in_compat_syscall()) | |
6888 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6889 | sigsz); |
9e75ad5d AB |
6890 | else |
6891 | #endif | |
b772434b | 6892 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6893 | |
2b188cc1 JA |
6894 | if (ret) |
6895 | return ret; | |
6896 | } | |
6897 | ||
c73ebb68 | 6898 | if (uts) { |
c1d5a224 PB |
6899 | struct timespec64 ts; |
6900 | ||
c73ebb68 HX |
6901 | if (get_timespec64(&ts, uts)) |
6902 | return -EFAULT; | |
6903 | timeout = timespec64_to_jiffies(&ts); | |
6904 | } | |
6905 | ||
bda52162 | 6906 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6907 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6908 | do { |
ca0a2651 JA |
6909 | /* if we can't even flush overflow, don't wait for more */ |
6910 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6911 | ret = -EBUSY; | |
6912 | break; | |
6913 | } | |
bda52162 JA |
6914 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6915 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6916 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6917 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6918 | cond_resched(); |
eeb60b9a | 6919 | } while (ret > 0); |
bda52162 | 6920 | |
b7db41c9 | 6921 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6922 | |
75b28aff | 6923 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6924 | } |
6925 | ||
6b06314c JA |
6926 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6927 | { | |
6928 | #if defined(CONFIG_UNIX) | |
6929 | if (ctx->ring_sock) { | |
6930 | struct sock *sock = ctx->ring_sock->sk; | |
6931 | struct sk_buff *skb; | |
6932 | ||
6933 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6934 | kfree_skb(skb); | |
6935 | } | |
6936 | #else | |
6937 | int i; | |
6938 | ||
65e19f54 JA |
6939 | for (i = 0; i < ctx->nr_user_files; i++) { |
6940 | struct file *file; | |
6941 | ||
6942 | file = io_file_from_index(ctx, i); | |
6943 | if (file) | |
6944 | fput(file); | |
6945 | } | |
6b06314c JA |
6946 | #endif |
6947 | } | |
6948 | ||
00835dce | 6949 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 6950 | { |
269bbe5f | 6951 | struct fixed_rsrc_data *data; |
05f3fb3c | 6952 | |
269bbe5f | 6953 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
6954 | complete(&data->done); |
6955 | } | |
6956 | ||
2a63b2d9 | 6957 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 6958 | { |
2a63b2d9 | 6959 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
6960 | } |
6961 | ||
2a63b2d9 | 6962 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 6963 | { |
2a63b2d9 BM |
6964 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
6965 | } | |
65e19f54 | 6966 | |
d67d2263 BM |
6967 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
6968 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 6969 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 6970 | { |
2a63b2d9 | 6971 | io_rsrc_ref_lock(ctx); |
269bbe5f | 6972 | rsrc_data->node = ref_node; |
d67d2263 | 6973 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 6974 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 6975 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
6976 | } |
6977 | ||
8bad28d8 | 6978 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 6979 | { |
8bad28d8 | 6980 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 6981 | |
2a63b2d9 | 6982 | io_rsrc_ref_lock(ctx); |
1e5d770b | 6983 | ref_node = data->node; |
e6cb007c | 6984 | data->node = NULL; |
2a63b2d9 | 6985 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
6986 | if (ref_node) |
6987 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
6988 | } |
6989 | ||
6990 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
6991 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
6992 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
6993 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 6994 | { |
f2303b1f | 6995 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 6996 | int ret; |
05589553 | 6997 | |
8bad28d8 HX |
6998 | if (data->quiesce) |
6999 | return -ENXIO; | |
05589553 | 7000 | |
8bad28d8 | 7001 | data->quiesce = true; |
1ffc5422 | 7002 | do { |
f2303b1f PB |
7003 | ret = -ENOMEM; |
7004 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7005 | if (!backup_node) | |
7006 | break; | |
7007 | backup_node->rsrc_data = data; | |
7008 | backup_node->rsrc_put = rsrc_put; | |
7009 | ||
8bad28d8 HX |
7010 | io_sqe_rsrc_kill_node(ctx, data); |
7011 | percpu_ref_kill(&data->refs); | |
7012 | flush_delayed_work(&ctx->rsrc_put_work); | |
7013 | ||
1ffc5422 PB |
7014 | ret = wait_for_completion_interruptible(&data->done); |
7015 | if (!ret) | |
7016 | break; | |
8bad28d8 | 7017 | |
cb5e1b81 | 7018 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7019 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7020 | backup_node = NULL; | |
cb5e1b81 | 7021 | reinit_completion(&data->done); |
8bad28d8 | 7022 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7023 | ret = io_run_task_work_sig(); |
8bad28d8 | 7024 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7025 | } while (ret >= 0); |
8bad28d8 | 7026 | data->quiesce = false; |
05f3fb3c | 7027 | |
8bad28d8 HX |
7028 | if (backup_node) |
7029 | destroy_fixed_rsrc_ref_node(backup_node); | |
7030 | return ret; | |
d7954b2b BM |
7031 | } |
7032 | ||
1ad555c6 BM |
7033 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7034 | { | |
7035 | struct fixed_rsrc_data *data; | |
7036 | ||
7037 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7038 | if (!data) | |
7039 | return NULL; | |
7040 | ||
00835dce | 7041 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7042 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7043 | kfree(data); | |
7044 | return NULL; | |
7045 | } | |
7046 | data->ctx = ctx; | |
7047 | init_completion(&data->done); | |
7048 | return data; | |
7049 | } | |
7050 | ||
7051 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7052 | { | |
7053 | percpu_ref_exit(&data->refs); | |
7054 | kfree(data->table); | |
7055 | kfree(data); | |
7056 | } | |
7057 | ||
d7954b2b BM |
7058 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7059 | { | |
7060 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7061 | unsigned nr_tables, i; |
7062 | int ret; | |
7063 | ||
8bad28d8 HX |
7064 | /* |
7065 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7066 | * Since we possibly drop uring lock later in this function to | |
7067 | * run task work. | |
7068 | */ | |
7069 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7070 | return -ENXIO; |
f2303b1f | 7071 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7072 | if (ret) |
7073 | return ret; | |
7074 | ||
6b06314c | 7075 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7076 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7077 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7078 | kfree(data->table[i].files); |
1ad555c6 | 7079 | free_fixed_rsrc_data(data); |
05f3fb3c | 7080 | ctx->file_data = NULL; |
6b06314c JA |
7081 | ctx->nr_user_files = 0; |
7082 | return 0; | |
7083 | } | |
7084 | ||
37d1e2e3 JA |
7085 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
7086 | __releases(&sqd->lock) | |
7087 | { | |
37d1e2e3 JA |
7088 | if (sqd->thread == current) |
7089 | return; | |
7090 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
86e0d676 JA |
7091 | if (sqd->thread) |
7092 | wake_up_state(sqd->thread, TASK_PARKED); | |
37d1e2e3 JA |
7093 | mutex_unlock(&sqd->lock); |
7094 | } | |
7095 | ||
86e0d676 | 7096 | static void io_sq_thread_park(struct io_sq_data *sqd) |
37d1e2e3 JA |
7097 | __acquires(&sqd->lock) |
7098 | { | |
7099 | if (sqd->thread == current) | |
86e0d676 JA |
7100 | return; |
7101 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
37d1e2e3 | 7102 | mutex_lock(&sqd->lock); |
86e0d676 JA |
7103 | if (sqd->thread) { |
7104 | wake_up_process(sqd->thread); | |
7105 | wait_for_completion(&sqd->parked); | |
37d1e2e3 | 7106 | } |
37d1e2e3 JA |
7107 | } |
7108 | ||
7109 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7110 | { | |
e54945ae | 7111 | if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) |
37d1e2e3 | 7112 | return; |
e54945ae JA |
7113 | mutex_lock(&sqd->lock); |
7114 | if (sqd->thread) { | |
7115 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7116 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)); | |
7117 | wake_up_process(sqd->thread); | |
7118 | mutex_unlock(&sqd->lock); | |
7119 | wait_for_completion(&sqd->exited); | |
7120 | WARN_ON_ONCE(sqd->thread); | |
7121 | } else { | |
7122 | mutex_unlock(&sqd->lock); | |
7123 | } | |
37d1e2e3 JA |
7124 | } |
7125 | ||
534ca6d6 | 7126 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7127 | { |
534ca6d6 | 7128 | if (refcount_dec_and_test(&sqd->refs)) { |
37d1e2e3 JA |
7129 | io_sq_thread_stop(sqd); |
7130 | kfree(sqd); | |
7131 | } | |
7132 | } | |
7133 | ||
7134 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7135 | { | |
7136 | struct io_sq_data *sqd = ctx->sq_data; | |
7137 | ||
7138 | if (sqd) { | |
eb85890b | 7139 | complete(&sqd->startup); |
534ca6d6 | 7140 | if (sqd->thread) { |
37d1e2e3 JA |
7141 | wait_for_completion(&ctx->sq_thread_comp); |
7142 | io_sq_thread_park(sqd); | |
534ca6d6 JA |
7143 | } |
7144 | ||
37d1e2e3 JA |
7145 | mutex_lock(&sqd->ctx_lock); |
7146 | list_del(&ctx->sqd_list); | |
7147 | io_sqd_update_thread_idle(sqd); | |
7148 | mutex_unlock(&sqd->ctx_lock); | |
7149 | ||
7150 | if (sqd->thread) | |
7151 | io_sq_thread_unpark(sqd); | |
7152 | ||
7153 | io_put_sq_data(sqd); | |
7154 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7155 | } |
7156 | } | |
7157 | ||
aa06165d JA |
7158 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7159 | { | |
7160 | struct io_ring_ctx *ctx_attach; | |
7161 | struct io_sq_data *sqd; | |
7162 | struct fd f; | |
7163 | ||
7164 | f = fdget(p->wq_fd); | |
7165 | if (!f.file) | |
7166 | return ERR_PTR(-ENXIO); | |
7167 | if (f.file->f_op != &io_uring_fops) { | |
7168 | fdput(f); | |
7169 | return ERR_PTR(-EINVAL); | |
7170 | } | |
7171 | ||
7172 | ctx_attach = f.file->private_data; | |
7173 | sqd = ctx_attach->sq_data; | |
7174 | if (!sqd) { | |
7175 | fdput(f); | |
7176 | return ERR_PTR(-EINVAL); | |
7177 | } | |
7178 | ||
7179 | refcount_inc(&sqd->refs); | |
7180 | fdput(f); | |
7181 | return sqd; | |
7182 | } | |
7183 | ||
534ca6d6 JA |
7184 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) |
7185 | { | |
7186 | struct io_sq_data *sqd; | |
7187 | ||
aa06165d JA |
7188 | if (p->flags & IORING_SETUP_ATTACH_WQ) |
7189 | return io_attach_sq_data(p); | |
7190 | ||
534ca6d6 JA |
7191 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7192 | if (!sqd) | |
7193 | return ERR_PTR(-ENOMEM); | |
7194 | ||
7195 | refcount_set(&sqd->refs, 1); | |
69fb2131 JA |
7196 | INIT_LIST_HEAD(&sqd->ctx_list); |
7197 | INIT_LIST_HEAD(&sqd->ctx_new_list); | |
7198 | mutex_init(&sqd->ctx_lock); | |
7199 | mutex_init(&sqd->lock); | |
534ca6d6 | 7200 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7201 | init_completion(&sqd->startup); |
86e0d676 | 7202 | init_completion(&sqd->parked); |
37d1e2e3 | 7203 | init_completion(&sqd->exited); |
534ca6d6 JA |
7204 | return sqd; |
7205 | } | |
7206 | ||
6b06314c | 7207 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7208 | /* |
7209 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7210 | * the io_uring can be safely unregistered on process exit, even if we have | |
7211 | * loops in the file referencing. | |
7212 | */ | |
7213 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7214 | { | |
7215 | struct sock *sk = ctx->ring_sock->sk; | |
7216 | struct scm_fp_list *fpl; | |
7217 | struct sk_buff *skb; | |
08a45173 | 7218 | int i, nr_files; |
6b06314c | 7219 | |
6b06314c JA |
7220 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7221 | if (!fpl) | |
7222 | return -ENOMEM; | |
7223 | ||
7224 | skb = alloc_skb(0, GFP_KERNEL); | |
7225 | if (!skb) { | |
7226 | kfree(fpl); | |
7227 | return -ENOMEM; | |
7228 | } | |
7229 | ||
7230 | skb->sk = sk; | |
6b06314c | 7231 | |
08a45173 | 7232 | nr_files = 0; |
62e398be | 7233 | fpl->user = get_uid(current_user()); |
6b06314c | 7234 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7235 | struct file *file = io_file_from_index(ctx, i + offset); |
7236 | ||
7237 | if (!file) | |
08a45173 | 7238 | continue; |
65e19f54 | 7239 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7240 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7241 | nr_files++; | |
6b06314c JA |
7242 | } |
7243 | ||
08a45173 JA |
7244 | if (nr_files) { |
7245 | fpl->max = SCM_MAX_FD; | |
7246 | fpl->count = nr_files; | |
7247 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7248 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7249 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7250 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7251 | |
08a45173 JA |
7252 | for (i = 0; i < nr_files; i++) |
7253 | fput(fpl->fp[i]); | |
7254 | } else { | |
7255 | kfree_skb(skb); | |
7256 | kfree(fpl); | |
7257 | } | |
6b06314c JA |
7258 | |
7259 | return 0; | |
7260 | } | |
7261 | ||
7262 | /* | |
7263 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7264 | * causes regular reference counting to break down. We rely on the UNIX | |
7265 | * garbage collection to take care of this problem for us. | |
7266 | */ | |
7267 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7268 | { | |
7269 | unsigned left, total; | |
7270 | int ret = 0; | |
7271 | ||
7272 | total = 0; | |
7273 | left = ctx->nr_user_files; | |
7274 | while (left) { | |
7275 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7276 | |
7277 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7278 | if (ret) | |
7279 | break; | |
7280 | left -= this_files; | |
7281 | total += this_files; | |
7282 | } | |
7283 | ||
7284 | if (!ret) | |
7285 | return 0; | |
7286 | ||
7287 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7288 | struct file *file = io_file_from_index(ctx, total); |
7289 | ||
7290 | if (file) | |
7291 | fput(file); | |
6b06314c JA |
7292 | total++; |
7293 | } | |
7294 | ||
7295 | return ret; | |
7296 | } | |
7297 | #else | |
7298 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7299 | { | |
7300 | return 0; | |
7301 | } | |
7302 | #endif | |
7303 | ||
269bbe5f | 7304 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7305 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7306 | { |
7307 | int i; | |
7308 | ||
7309 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7310 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7311 | unsigned this_files; |
7312 | ||
7313 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7314 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7315 | GFP_KERNEL); | |
7316 | if (!table->files) | |
7317 | break; | |
7318 | nr_files -= this_files; | |
7319 | } | |
7320 | ||
7321 | if (i == nr_tables) | |
7322 | return 0; | |
7323 | ||
7324 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7325 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7326 | kfree(table->files); |
7327 | } | |
7328 | return 1; | |
7329 | } | |
7330 | ||
50238531 | 7331 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7332 | { |
50238531 | 7333 | struct file *file = prsrc->file; |
05f3fb3c JA |
7334 | #if defined(CONFIG_UNIX) |
7335 | struct sock *sock = ctx->ring_sock->sk; | |
7336 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7337 | struct sk_buff *skb; | |
7338 | int i; | |
7339 | ||
7340 | __skb_queue_head_init(&list); | |
7341 | ||
7342 | /* | |
7343 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7344 | * remove this entry and rearrange the file array. | |
7345 | */ | |
7346 | skb = skb_dequeue(head); | |
7347 | while (skb) { | |
7348 | struct scm_fp_list *fp; | |
7349 | ||
7350 | fp = UNIXCB(skb).fp; | |
7351 | for (i = 0; i < fp->count; i++) { | |
7352 | int left; | |
7353 | ||
7354 | if (fp->fp[i] != file) | |
7355 | continue; | |
7356 | ||
7357 | unix_notinflight(fp->user, fp->fp[i]); | |
7358 | left = fp->count - 1 - i; | |
7359 | if (left) { | |
7360 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7361 | left * sizeof(struct file *)); | |
7362 | } | |
7363 | fp->count--; | |
7364 | if (!fp->count) { | |
7365 | kfree_skb(skb); | |
7366 | skb = NULL; | |
7367 | } else { | |
7368 | __skb_queue_tail(&list, skb); | |
7369 | } | |
7370 | fput(file); | |
7371 | file = NULL; | |
7372 | break; | |
7373 | } | |
7374 | ||
7375 | if (!file) | |
7376 | break; | |
7377 | ||
7378 | __skb_queue_tail(&list, skb); | |
7379 | ||
7380 | skb = skb_dequeue(head); | |
7381 | } | |
7382 | ||
7383 | if (skb_peek(&list)) { | |
7384 | spin_lock_irq(&head->lock); | |
7385 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7386 | __skb_queue_tail(head, skb); | |
7387 | spin_unlock_irq(&head->lock); | |
7388 | } | |
7389 | #else | |
7390 | fput(file); | |
7391 | #endif | |
7392 | } | |
7393 | ||
269bbe5f | 7394 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7395 | { |
269bbe5f BM |
7396 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7397 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7398 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7399 | |
269bbe5f BM |
7400 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7401 | list_del(&prsrc->list); | |
50238531 | 7402 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7403 | kfree(prsrc); |
65e19f54 | 7404 | } |
05589553 | 7405 | |
05589553 XW |
7406 | percpu_ref_exit(&ref_node->refs); |
7407 | kfree(ref_node); | |
269bbe5f | 7408 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7409 | } |
65e19f54 | 7410 | |
269bbe5f | 7411 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7412 | { |
7413 | struct io_ring_ctx *ctx; | |
7414 | struct llist_node *node; | |
7415 | ||
269bbe5f BM |
7416 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7417 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7418 | |
7419 | while (node) { | |
269bbe5f | 7420 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7421 | struct llist_node *next = node->next; |
7422 | ||
269bbe5f BM |
7423 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7424 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7425 | node = next; |
7426 | } | |
7427 | } | |
7428 | ||
ea64ec02 PB |
7429 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7430 | unsigned i) | |
2faf852d | 7431 | { |
ea64ec02 PB |
7432 | struct fixed_rsrc_table *table; |
7433 | ||
7434 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7435 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7436 | } | |
7437 | ||
00835dce | 7438 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7439 | { |
269bbe5f BM |
7440 | struct fixed_rsrc_ref_node *ref_node; |
7441 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7442 | struct io_ring_ctx *ctx; |
e297822b | 7443 | bool first_add = false; |
4a38aed2 | 7444 | int delay = HZ; |
65e19f54 | 7445 | |
269bbe5f BM |
7446 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7447 | data = ref_node->rsrc_data; | |
e297822b PB |
7448 | ctx = data->ctx; |
7449 | ||
2a63b2d9 | 7450 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7451 | ref_node->done = true; |
7452 | ||
d67d2263 BM |
7453 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7454 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7455 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7456 | /* recycle ref nodes in order */ |
7457 | if (!ref_node->done) | |
7458 | break; | |
7459 | list_del(&ref_node->node); | |
269bbe5f | 7460 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7461 | } |
2a63b2d9 | 7462 | io_rsrc_ref_unlock(ctx); |
05589553 | 7463 | |
e297822b | 7464 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7465 | delay = 0; |
05589553 | 7466 | |
4a38aed2 | 7467 | if (!delay) |
269bbe5f | 7468 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7469 | else if (first_add) |
269bbe5f | 7470 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7471 | } |
65e19f54 | 7472 | |
6802535d | 7473 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7474 | struct io_ring_ctx *ctx) |
05f3fb3c | 7475 | { |
269bbe5f | 7476 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7477 | |
05589553 XW |
7478 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7479 | if (!ref_node) | |
3e2224c5 | 7480 | return NULL; |
05f3fb3c | 7481 | |
00835dce | 7482 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7483 | 0, GFP_KERNEL)) { |
7484 | kfree(ref_node); | |
3e2224c5 | 7485 | return NULL; |
05589553 XW |
7486 | } |
7487 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7488 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7489 | ref_node->done = false; |
05589553 | 7490 | return ref_node; |
05589553 XW |
7491 | } |
7492 | ||
bc9744cd PB |
7493 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7494 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7495 | { |
269bbe5f | 7496 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7497 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7498 | } |
7499 | ||
269bbe5f | 7500 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7501 | { |
7502 | percpu_ref_exit(&ref_node->refs); | |
7503 | kfree(ref_node); | |
65e19f54 JA |
7504 | } |
7505 | ||
ea64ec02 | 7506 | |
6b06314c JA |
7507 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7508 | unsigned nr_args) | |
7509 | { | |
7510 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7511 | unsigned nr_tables, i; |
05f3fb3c | 7512 | struct file *file; |
600cf3f8 | 7513 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7514 | struct fixed_rsrc_ref_node *ref_node; |
7515 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7516 | |
05f3fb3c | 7517 | if (ctx->file_data) |
6b06314c JA |
7518 | return -EBUSY; |
7519 | if (!nr_args) | |
7520 | return -EINVAL; | |
7521 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7522 | return -EMFILE; | |
7523 | ||
1ad555c6 | 7524 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7525 | if (!file_data) |
05f3fb3c | 7526 | return -ENOMEM; |
13770a71 | 7527 | ctx->file_data = file_data; |
05f3fb3c | 7528 | |
65e19f54 | 7529 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7530 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7531 | GFP_KERNEL); |
600cf3f8 PB |
7532 | if (!file_data->table) |
7533 | goto out_free; | |
05f3fb3c | 7534 | |
600cf3f8 | 7535 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7536 | goto out_free; |
65e19f54 | 7537 | |
08a45173 | 7538 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7539 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7540 | ret = -EFAULT; | |
7541 | goto out_fput; | |
7542 | } | |
08a45173 | 7543 | /* allow sparse sets */ |
600cf3f8 | 7544 | if (fd == -1) |
08a45173 | 7545 | continue; |
6b06314c | 7546 | |
05f3fb3c | 7547 | file = fget(fd); |
6b06314c | 7548 | ret = -EBADF; |
05f3fb3c | 7549 | if (!file) |
600cf3f8 | 7550 | goto out_fput; |
05f3fb3c | 7551 | |
6b06314c JA |
7552 | /* |
7553 | * Don't allow io_uring instances to be registered. If UNIX | |
7554 | * isn't enabled, then this causes a reference cycle and this | |
7555 | * instance can never get freed. If UNIX is enabled we'll | |
7556 | * handle it just fine, but there's still no point in allowing | |
7557 | * a ring fd as it doesn't support regular read/write anyway. | |
7558 | */ | |
05f3fb3c JA |
7559 | if (file->f_op == &io_uring_fops) { |
7560 | fput(file); | |
600cf3f8 | 7561 | goto out_fput; |
6b06314c | 7562 | } |
ea64ec02 | 7563 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7564 | } |
7565 | ||
6b06314c | 7566 | ret = io_sqe_files_scm(ctx); |
05589553 | 7567 | if (ret) { |
6b06314c | 7568 | io_sqe_files_unregister(ctx); |
05589553 XW |
7569 | return ret; |
7570 | } | |
6b06314c | 7571 | |
bc9744cd | 7572 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7573 | if (!ref_node) { |
05589553 | 7574 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7575 | return -ENOMEM; |
05589553 | 7576 | } |
bc9744cd | 7577 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7578 | |
d67d2263 | 7579 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7580 | return ret; |
600cf3f8 PB |
7581 | out_fput: |
7582 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7583 | file = io_file_from_index(ctx, i); | |
7584 | if (file) | |
7585 | fput(file); | |
7586 | } | |
7587 | for (i = 0; i < nr_tables; i++) | |
7588 | kfree(file_data->table[i].files); | |
7589 | ctx->nr_user_files = 0; | |
600cf3f8 | 7590 | out_free: |
1ad555c6 | 7591 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7592 | ctx->file_data = NULL; |
6b06314c JA |
7593 | return ret; |
7594 | } | |
7595 | ||
c3a31e60 JA |
7596 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7597 | int index) | |
7598 | { | |
7599 | #if defined(CONFIG_UNIX) | |
7600 | struct sock *sock = ctx->ring_sock->sk; | |
7601 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7602 | struct sk_buff *skb; | |
7603 | ||
7604 | /* | |
7605 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7606 | * file set. If there's no room, fall back to allocating a new skb | |
7607 | * and filling it in. | |
7608 | */ | |
7609 | spin_lock_irq(&head->lock); | |
7610 | skb = skb_peek(head); | |
7611 | if (skb) { | |
7612 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7613 | ||
7614 | if (fpl->count < SCM_MAX_FD) { | |
7615 | __skb_unlink(skb, head); | |
7616 | spin_unlock_irq(&head->lock); | |
7617 | fpl->fp[fpl->count] = get_file(file); | |
7618 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7619 | fpl->count++; | |
7620 | spin_lock_irq(&head->lock); | |
7621 | __skb_queue_head(head, skb); | |
7622 | } else { | |
7623 | skb = NULL; | |
7624 | } | |
7625 | } | |
7626 | spin_unlock_irq(&head->lock); | |
7627 | ||
7628 | if (skb) { | |
7629 | fput(file); | |
7630 | return 0; | |
7631 | } | |
7632 | ||
7633 | return __io_sqe_files_scm(ctx, 1, index); | |
7634 | #else | |
7635 | return 0; | |
7636 | #endif | |
7637 | } | |
7638 | ||
50238531 | 7639 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7640 | { |
269bbe5f BM |
7641 | struct io_rsrc_put *prsrc; |
7642 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7643 | |
269bbe5f BM |
7644 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7645 | if (!prsrc) | |
a5318d3c | 7646 | return -ENOMEM; |
05f3fb3c | 7647 | |
50238531 | 7648 | prsrc->rsrc = rsrc; |
269bbe5f | 7649 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7650 | |
a5318d3c | 7651 | return 0; |
05f3fb3c JA |
7652 | } |
7653 | ||
269bbe5f BM |
7654 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7655 | struct file *file) | |
7656 | { | |
50238531 | 7657 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7658 | } |
7659 | ||
05f3fb3c | 7660 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7661 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7662 | unsigned nr_args) |
7663 | { | |
269bbe5f BM |
7664 | struct fixed_rsrc_data *data = ctx->file_data; |
7665 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7666 | struct file *file, **file_slot; |
c3a31e60 JA |
7667 | __s32 __user *fds; |
7668 | int fd, i, err; | |
7669 | __u32 done; | |
05589553 | 7670 | bool needs_switch = false; |
c3a31e60 | 7671 | |
05f3fb3c | 7672 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7673 | return -EOVERFLOW; |
7674 | if (done > ctx->nr_user_files) | |
7675 | return -EINVAL; | |
7676 | ||
bc9744cd | 7677 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7678 | if (!ref_node) |
7679 | return -ENOMEM; | |
bc9744cd | 7680 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7681 | |
269bbe5f | 7682 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7683 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7684 | err = 0; |
7685 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7686 | err = -EFAULT; | |
7687 | break; | |
7688 | } | |
4e0377a1 | 7689 | if (fd == IORING_REGISTER_FILES_SKIP) |
7690 | continue; | |
7691 | ||
67973b93 | 7692 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7693 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7694 | ||
7695 | if (*file_slot) { | |
7696 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7697 | if (err) |
7698 | break; | |
ea64ec02 | 7699 | *file_slot = NULL; |
05589553 | 7700 | needs_switch = true; |
c3a31e60 JA |
7701 | } |
7702 | if (fd != -1) { | |
c3a31e60 JA |
7703 | file = fget(fd); |
7704 | if (!file) { | |
7705 | err = -EBADF; | |
7706 | break; | |
7707 | } | |
7708 | /* | |
7709 | * Don't allow io_uring instances to be registered. If | |
7710 | * UNIX isn't enabled, then this causes a reference | |
7711 | * cycle and this instance can never get freed. If UNIX | |
7712 | * is enabled we'll handle it just fine, but there's | |
7713 | * still no point in allowing a ring fd as it doesn't | |
7714 | * support regular read/write anyway. | |
7715 | */ | |
7716 | if (file->f_op == &io_uring_fops) { | |
7717 | fput(file); | |
7718 | err = -EBADF; | |
7719 | break; | |
7720 | } | |
e68a3ff8 | 7721 | *file_slot = file; |
c3a31e60 | 7722 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7723 | if (err) { |
e68a3ff8 | 7724 | *file_slot = NULL; |
f3bd9dae | 7725 | fput(file); |
c3a31e60 | 7726 | break; |
f3bd9dae | 7727 | } |
c3a31e60 | 7728 | } |
05f3fb3c JA |
7729 | } |
7730 | ||
05589553 | 7731 | if (needs_switch) { |
b2e96852 | 7732 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7733 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7734 | } else |
269bbe5f | 7735 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7736 | |
7737 | return done ? done : err; | |
7738 | } | |
05589553 | 7739 | |
05f3fb3c JA |
7740 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7741 | unsigned nr_args) | |
7742 | { | |
269bbe5f | 7743 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7744 | |
7745 | if (!ctx->file_data) | |
7746 | return -ENXIO; | |
7747 | if (!nr_args) | |
7748 | return -EINVAL; | |
7749 | if (copy_from_user(&up, arg, sizeof(up))) | |
7750 | return -EFAULT; | |
7751 | if (up.resv) | |
7752 | return -EINVAL; | |
7753 | ||
7754 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7755 | } | |
c3a31e60 | 7756 | |
5280f7e5 | 7757 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7758 | { |
7759 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7760 | ||
5280f7e5 PB |
7761 | req = io_put_req_find_next(req); |
7762 | return req ? &req->work : NULL; | |
7d723065 JA |
7763 | } |
7764 | ||
5aa75ed5 | 7765 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7766 | { |
e941894e | 7767 | struct io_wq_hash *hash; |
24369c2e | 7768 | struct io_wq_data data; |
24369c2e | 7769 | unsigned int concurrency; |
24369c2e | 7770 | |
e941894e JA |
7771 | hash = ctx->hash_map; |
7772 | if (!hash) { | |
7773 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7774 | if (!hash) | |
7775 | return ERR_PTR(-ENOMEM); | |
7776 | refcount_set(&hash->refs, 1); | |
7777 | init_waitqueue_head(&hash->wait); | |
7778 | ctx->hash_map = hash; | |
24369c2e PB |
7779 | } |
7780 | ||
e941894e | 7781 | data.hash = hash; |
e9fd9396 | 7782 | data.free_work = io_free_work; |
f5fa38c5 | 7783 | data.do_work = io_wq_submit_work; |
24369c2e | 7784 | |
d25e3a3d JA |
7785 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7786 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7787 | |
5aa75ed5 | 7788 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7789 | } |
7790 | ||
5aa75ed5 JA |
7791 | static int io_uring_alloc_task_context(struct task_struct *task, |
7792 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7793 | { |
7794 | struct io_uring_task *tctx; | |
d8a6df10 | 7795 | int ret; |
0f212204 JA |
7796 | |
7797 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7798 | if (unlikely(!tctx)) | |
7799 | return -ENOMEM; | |
7800 | ||
d8a6df10 JA |
7801 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7802 | if (unlikely(ret)) { | |
7803 | kfree(tctx); | |
7804 | return ret; | |
7805 | } | |
7806 | ||
5aa75ed5 JA |
7807 | tctx->io_wq = io_init_wq_offload(ctx); |
7808 | if (IS_ERR(tctx->io_wq)) { | |
7809 | ret = PTR_ERR(tctx->io_wq); | |
7810 | percpu_counter_destroy(&tctx->inflight); | |
7811 | kfree(tctx); | |
7812 | return ret; | |
7813 | } | |
7814 | ||
0f212204 JA |
7815 | xa_init(&tctx->xa); |
7816 | init_waitqueue_head(&tctx->wait); | |
7817 | tctx->last = NULL; | |
fdaf083c JA |
7818 | atomic_set(&tctx->in_idle, 0); |
7819 | tctx->sqpoll = false; | |
0f212204 | 7820 | task->io_uring = tctx; |
7cbf1722 JA |
7821 | spin_lock_init(&tctx->task_lock); |
7822 | INIT_WQ_LIST(&tctx->task_list); | |
7823 | tctx->task_state = 0; | |
7824 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7825 | return 0; |
7826 | } | |
7827 | ||
7828 | void __io_uring_free(struct task_struct *tsk) | |
7829 | { | |
7830 | struct io_uring_task *tctx = tsk->io_uring; | |
7831 | ||
7832 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7833 | WARN_ON_ONCE(tctx->io_wq); |
7834 | ||
d8a6df10 | 7835 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7836 | kfree(tctx); |
7837 | tsk->io_uring = NULL; | |
7838 | } | |
7839 | ||
5f3f26f9 JA |
7840 | static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx) |
7841 | { | |
46fe18b1 | 7842 | struct task_struct *tsk; |
5f3f26f9 JA |
7843 | int ret; |
7844 | ||
7845 | clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
86e0d676 | 7846 | reinit_completion(&sqd->parked); |
70aacfe6 | 7847 | ctx->sqo_exec = 0; |
5f3f26f9 | 7848 | sqd->task_pid = current->pid; |
46fe18b1 JA |
7849 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7850 | if (IS_ERR(tsk)) | |
7851 | return PTR_ERR(tsk); | |
7852 | ret = io_uring_alloc_task_context(tsk, ctx); | |
7853 | if (ret) | |
7854 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7855 | sqd->thread = tsk; | |
7856 | wake_up_new_task(tsk); | |
7857 | return ret; | |
5f3f26f9 JA |
7858 | } |
7859 | ||
7e84e1c7 SG |
7860 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7861 | struct io_uring_params *p) | |
2b188cc1 JA |
7862 | { |
7863 | int ret; | |
7864 | ||
d25e3a3d JA |
7865 | /* Retain compatibility with failing for an invalid attach attempt */ |
7866 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7867 | IORING_SETUP_ATTACH_WQ) { | |
7868 | struct fd f; | |
7869 | ||
7870 | f = fdget(p->wq_fd); | |
7871 | if (!f.file) | |
7872 | return -ENXIO; | |
7873 | if (f.file->f_op != &io_uring_fops) { | |
7874 | fdput(f); | |
7875 | return -EINVAL; | |
7876 | } | |
7877 | fdput(f); | |
7878 | } | |
6c271ce2 | 7879 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7880 | struct task_struct *tsk; |
534ca6d6 JA |
7881 | struct io_sq_data *sqd; |
7882 | ||
3ec482d1 | 7883 | ret = -EPERM; |
ce59fc69 | 7884 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7885 | goto err; |
7886 | ||
534ca6d6 JA |
7887 | sqd = io_get_sq_data(p); |
7888 | if (IS_ERR(sqd)) { | |
7889 | ret = PTR_ERR(sqd); | |
7890 | goto err; | |
7891 | } | |
69fb2131 | 7892 | |
534ca6d6 | 7893 | ctx->sq_data = sqd; |
69fb2131 JA |
7894 | io_sq_thread_park(sqd); |
7895 | mutex_lock(&sqd->ctx_lock); | |
7896 | list_add(&ctx->sqd_list, &sqd->ctx_new_list); | |
7897 | mutex_unlock(&sqd->ctx_lock); | |
7898 | io_sq_thread_unpark(sqd); | |
534ca6d6 | 7899 | |
917257da JA |
7900 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7901 | if (!ctx->sq_thread_idle) | |
7902 | ctx->sq_thread_idle = HZ; | |
7903 | ||
aa06165d | 7904 | if (sqd->thread) |
5aa75ed5 | 7905 | return 0; |
aa06165d | 7906 | |
6c271ce2 | 7907 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7908 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7909 | |
917257da | 7910 | ret = -EINVAL; |
44a9bd18 JA |
7911 | if (cpu >= nr_cpu_ids) |
7912 | goto err; | |
7889f44d | 7913 | if (!cpu_online(cpu)) |
917257da JA |
7914 | goto err; |
7915 | ||
37d1e2e3 | 7916 | sqd->sq_cpu = cpu; |
6c271ce2 | 7917 | } else { |
37d1e2e3 | 7918 | sqd->sq_cpu = -1; |
6c271ce2 | 7919 | } |
37d1e2e3 JA |
7920 | |
7921 | sqd->task_pid = current->pid; | |
46fe18b1 JA |
7922 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7923 | if (IS_ERR(tsk)) { | |
7924 | ret = PTR_ERR(tsk); | |
6c271ce2 JA |
7925 | goto err; |
7926 | } | |
46fe18b1 JA |
7927 | ret = io_uring_alloc_task_context(tsk, ctx); |
7928 | if (ret) | |
7929 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); | |
7930 | sqd->thread = tsk; | |
7931 | wake_up_new_task(tsk); | |
0f212204 JA |
7932 | if (ret) |
7933 | goto err; | |
6c271ce2 JA |
7934 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7935 | /* Can't have SQ_AFF without SQPOLL */ | |
7936 | ret = -EINVAL; | |
7937 | goto err; | |
7938 | } | |
7939 | ||
2b188cc1 JA |
7940 | return 0; |
7941 | err: | |
37d1e2e3 | 7942 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
7943 | return ret; |
7944 | } | |
7945 | ||
7e84e1c7 SG |
7946 | static void io_sq_offload_start(struct io_ring_ctx *ctx) |
7947 | { | |
534ca6d6 JA |
7948 | struct io_sq_data *sqd = ctx->sq_data; |
7949 | ||
3ebba796 | 7950 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
eb85890b | 7951 | if (ctx->flags & IORING_SETUP_SQPOLL) |
37d1e2e3 | 7952 | complete(&sqd->startup); |
7e84e1c7 SG |
7953 | } |
7954 | ||
a087e2b5 BM |
7955 | static inline void __io_unaccount_mem(struct user_struct *user, |
7956 | unsigned long nr_pages) | |
2b188cc1 JA |
7957 | { |
7958 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7959 | } | |
7960 | ||
a087e2b5 BM |
7961 | static inline int __io_account_mem(struct user_struct *user, |
7962 | unsigned long nr_pages) | |
2b188cc1 JA |
7963 | { |
7964 | unsigned long page_limit, cur_pages, new_pages; | |
7965 | ||
7966 | /* Don't allow more pages than we can safely lock */ | |
7967 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
7968 | ||
7969 | do { | |
7970 | cur_pages = atomic_long_read(&user->locked_vm); | |
7971 | new_pages = cur_pages + nr_pages; | |
7972 | if (new_pages > page_limit) | |
7973 | return -ENOMEM; | |
7974 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
7975 | new_pages) != cur_pages); | |
7976 | ||
7977 | return 0; | |
7978 | } | |
7979 | ||
26bfa89e | 7980 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7981 | { |
62e398be | 7982 | if (ctx->user) |
a087e2b5 | 7983 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 7984 | |
26bfa89e JA |
7985 | if (ctx->mm_account) |
7986 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
7987 | } |
7988 | ||
26bfa89e | 7989 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 7990 | { |
30975825 BM |
7991 | int ret; |
7992 | ||
62e398be | 7993 | if (ctx->user) { |
30975825 BM |
7994 | ret = __io_account_mem(ctx->user, nr_pages); |
7995 | if (ret) | |
7996 | return ret; | |
7997 | } | |
7998 | ||
26bfa89e JA |
7999 | if (ctx->mm_account) |
8000 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8001 | |
8002 | return 0; | |
8003 | } | |
8004 | ||
2b188cc1 JA |
8005 | static void io_mem_free(void *ptr) |
8006 | { | |
52e04ef4 MR |
8007 | struct page *page; |
8008 | ||
8009 | if (!ptr) | |
8010 | return; | |
2b188cc1 | 8011 | |
52e04ef4 | 8012 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8013 | if (put_page_testzero(page)) |
8014 | free_compound_page(page); | |
8015 | } | |
8016 | ||
8017 | static void *io_mem_alloc(size_t size) | |
8018 | { | |
8019 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8020 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8021 | |
8022 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8023 | } | |
8024 | ||
75b28aff HV |
8025 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8026 | size_t *sq_offset) | |
8027 | { | |
8028 | struct io_rings *rings; | |
8029 | size_t off, sq_array_size; | |
8030 | ||
8031 | off = struct_size(rings, cqes, cq_entries); | |
8032 | if (off == SIZE_MAX) | |
8033 | return SIZE_MAX; | |
8034 | ||
8035 | #ifdef CONFIG_SMP | |
8036 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8037 | if (off == 0) | |
8038 | return SIZE_MAX; | |
8039 | #endif | |
8040 | ||
b36200f5 DV |
8041 | if (sq_offset) |
8042 | *sq_offset = off; | |
8043 | ||
75b28aff HV |
8044 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8045 | if (sq_array_size == SIZE_MAX) | |
8046 | return SIZE_MAX; | |
8047 | ||
8048 | if (check_add_overflow(off, sq_array_size, &off)) | |
8049 | return SIZE_MAX; | |
8050 | ||
75b28aff HV |
8051 | return off; |
8052 | } | |
8053 | ||
0a96bbe4 | 8054 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8055 | { |
8056 | int i, j; | |
8057 | ||
8058 | if (!ctx->user_bufs) | |
8059 | return -ENXIO; | |
8060 | ||
8061 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8062 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8063 | ||
8064 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8065 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8066 | |
de293938 | 8067 | if (imu->acct_pages) |
26bfa89e | 8068 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8069 | kvfree(imu->bvec); |
edafccee JA |
8070 | imu->nr_bvecs = 0; |
8071 | } | |
8072 | ||
8073 | kfree(ctx->user_bufs); | |
8074 | ctx->user_bufs = NULL; | |
8075 | ctx->nr_user_bufs = 0; | |
8076 | return 0; | |
8077 | } | |
8078 | ||
8079 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8080 | void __user *arg, unsigned index) | |
8081 | { | |
8082 | struct iovec __user *src; | |
8083 | ||
8084 | #ifdef CONFIG_COMPAT | |
8085 | if (ctx->compat) { | |
8086 | struct compat_iovec __user *ciovs; | |
8087 | struct compat_iovec ciov; | |
8088 | ||
8089 | ciovs = (struct compat_iovec __user *) arg; | |
8090 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8091 | return -EFAULT; | |
8092 | ||
d55e5f5b | 8093 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8094 | dst->iov_len = ciov.iov_len; |
8095 | return 0; | |
8096 | } | |
8097 | #endif | |
8098 | src = (struct iovec __user *) arg; | |
8099 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8100 | return -EFAULT; | |
8101 | return 0; | |
8102 | } | |
8103 | ||
de293938 JA |
8104 | /* |
8105 | * Not super efficient, but this is just a registration time. And we do cache | |
8106 | * the last compound head, so generally we'll only do a full search if we don't | |
8107 | * match that one. | |
8108 | * | |
8109 | * We check if the given compound head page has already been accounted, to | |
8110 | * avoid double accounting it. This allows us to account the full size of the | |
8111 | * page, not just the constituent pages of a huge page. | |
8112 | */ | |
8113 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8114 | int nr_pages, struct page *hpage) | |
8115 | { | |
8116 | int i, j; | |
8117 | ||
8118 | /* check current page array */ | |
8119 | for (i = 0; i < nr_pages; i++) { | |
8120 | if (!PageCompound(pages[i])) | |
8121 | continue; | |
8122 | if (compound_head(pages[i]) == hpage) | |
8123 | return true; | |
8124 | } | |
8125 | ||
8126 | /* check previously registered pages */ | |
8127 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8128 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8129 | ||
8130 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8131 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8132 | continue; | |
8133 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8134 | return true; | |
8135 | } | |
8136 | } | |
8137 | ||
8138 | return false; | |
8139 | } | |
8140 | ||
8141 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8142 | int nr_pages, struct io_mapped_ubuf *imu, | |
8143 | struct page **last_hpage) | |
8144 | { | |
8145 | int i, ret; | |
8146 | ||
8147 | for (i = 0; i < nr_pages; i++) { | |
8148 | if (!PageCompound(pages[i])) { | |
8149 | imu->acct_pages++; | |
8150 | } else { | |
8151 | struct page *hpage; | |
8152 | ||
8153 | hpage = compound_head(pages[i]); | |
8154 | if (hpage == *last_hpage) | |
8155 | continue; | |
8156 | *last_hpage = hpage; | |
8157 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8158 | continue; | |
8159 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8160 | } | |
8161 | } | |
8162 | ||
8163 | if (!imu->acct_pages) | |
8164 | return 0; | |
8165 | ||
26bfa89e | 8166 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8167 | if (ret) |
8168 | imu->acct_pages = 0; | |
8169 | return ret; | |
8170 | } | |
8171 | ||
0a96bbe4 BM |
8172 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8173 | struct io_mapped_ubuf *imu, | |
8174 | struct page **last_hpage) | |
edafccee JA |
8175 | { |
8176 | struct vm_area_struct **vmas = NULL; | |
8177 | struct page **pages = NULL; | |
0a96bbe4 BM |
8178 | unsigned long off, start, end, ubuf; |
8179 | size_t size; | |
8180 | int ret, pret, nr_pages, i; | |
8181 | ||
8182 | ubuf = (unsigned long) iov->iov_base; | |
8183 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8184 | start = ubuf >> PAGE_SHIFT; | |
8185 | nr_pages = end - start; | |
8186 | ||
8187 | ret = -ENOMEM; | |
8188 | ||
8189 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8190 | if (!pages) | |
8191 | goto done; | |
8192 | ||
8193 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8194 | GFP_KERNEL); | |
8195 | if (!vmas) | |
8196 | goto done; | |
edafccee | 8197 | |
0a96bbe4 BM |
8198 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8199 | GFP_KERNEL); | |
8200 | if (!imu->bvec) | |
8201 | goto done; | |
8202 | ||
8203 | ret = 0; | |
8204 | mmap_read_lock(current->mm); | |
8205 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8206 | pages, vmas); | |
8207 | if (pret == nr_pages) { | |
8208 | /* don't support file backed memory */ | |
8209 | for (i = 0; i < nr_pages; i++) { | |
8210 | struct vm_area_struct *vma = vmas[i]; | |
8211 | ||
8212 | if (vma->vm_file && | |
8213 | !is_file_hugepages(vma->vm_file)) { | |
8214 | ret = -EOPNOTSUPP; | |
8215 | break; | |
8216 | } | |
8217 | } | |
8218 | } else { | |
8219 | ret = pret < 0 ? pret : -EFAULT; | |
8220 | } | |
8221 | mmap_read_unlock(current->mm); | |
8222 | if (ret) { | |
8223 | /* | |
8224 | * if we did partial map, or found file backed vmas, | |
8225 | * release any pages we did get | |
8226 | */ | |
8227 | if (pret > 0) | |
8228 | unpin_user_pages(pages, pret); | |
8229 | kvfree(imu->bvec); | |
8230 | goto done; | |
8231 | } | |
8232 | ||
8233 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8234 | if (ret) { | |
8235 | unpin_user_pages(pages, pret); | |
8236 | kvfree(imu->bvec); | |
8237 | goto done; | |
8238 | } | |
8239 | ||
8240 | off = ubuf & ~PAGE_MASK; | |
8241 | size = iov->iov_len; | |
8242 | for (i = 0; i < nr_pages; i++) { | |
8243 | size_t vec_len; | |
8244 | ||
8245 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8246 | imu->bvec[i].bv_page = pages[i]; | |
8247 | imu->bvec[i].bv_len = vec_len; | |
8248 | imu->bvec[i].bv_offset = off; | |
8249 | off = 0; | |
8250 | size -= vec_len; | |
8251 | } | |
8252 | /* store original address for later verification */ | |
8253 | imu->ubuf = ubuf; | |
8254 | imu->len = iov->iov_len; | |
8255 | imu->nr_bvecs = nr_pages; | |
8256 | ret = 0; | |
8257 | done: | |
8258 | kvfree(pages); | |
8259 | kvfree(vmas); | |
8260 | return ret; | |
8261 | } | |
8262 | ||
2b358604 | 8263 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8264 | { |
edafccee JA |
8265 | if (ctx->user_bufs) |
8266 | return -EBUSY; | |
8267 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8268 | return -EINVAL; | |
8269 | ||
8270 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8271 | GFP_KERNEL); | |
8272 | if (!ctx->user_bufs) | |
8273 | return -ENOMEM; | |
8274 | ||
2b358604 BM |
8275 | return 0; |
8276 | } | |
edafccee | 8277 | |
2b358604 BM |
8278 | static int io_buffer_validate(struct iovec *iov) |
8279 | { | |
8280 | /* | |
8281 | * Don't impose further limits on the size and buffer | |
8282 | * constraints here, we'll -EINVAL later when IO is | |
8283 | * submitted if they are wrong. | |
8284 | */ | |
8285 | if (!iov->iov_base || !iov->iov_len) | |
8286 | return -EFAULT; | |
edafccee | 8287 | |
2b358604 BM |
8288 | /* arbitrary limit, but we need something */ |
8289 | if (iov->iov_len > SZ_1G) | |
8290 | return -EFAULT; | |
edafccee | 8291 | |
2b358604 BM |
8292 | return 0; |
8293 | } | |
edafccee | 8294 | |
2b358604 BM |
8295 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8296 | unsigned int nr_args) | |
8297 | { | |
8298 | int i, ret; | |
8299 | struct iovec iov; | |
8300 | struct page *last_hpage = NULL; | |
edafccee | 8301 | |
2b358604 BM |
8302 | ret = io_buffers_map_alloc(ctx, nr_args); |
8303 | if (ret) | |
8304 | return ret; | |
edafccee | 8305 | |
edafccee JA |
8306 | for (i = 0; i < nr_args; i++) { |
8307 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8308 | |
edafccee JA |
8309 | ret = io_copy_iov(ctx, &iov, arg, i); |
8310 | if (ret) | |
0a96bbe4 | 8311 | break; |
de293938 | 8312 | |
2b358604 BM |
8313 | ret = io_buffer_validate(&iov); |
8314 | if (ret) | |
0a96bbe4 | 8315 | break; |
edafccee | 8316 | |
0a96bbe4 BM |
8317 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8318 | if (ret) | |
8319 | break; | |
edafccee JA |
8320 | |
8321 | ctx->nr_user_bufs++; | |
8322 | } | |
0a96bbe4 BM |
8323 | |
8324 | if (ret) | |
8325 | io_sqe_buffers_unregister(ctx); | |
8326 | ||
edafccee JA |
8327 | return ret; |
8328 | } | |
8329 | ||
9b402849 JA |
8330 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8331 | { | |
8332 | __s32 __user *fds = arg; | |
8333 | int fd; | |
8334 | ||
8335 | if (ctx->cq_ev_fd) | |
8336 | return -EBUSY; | |
8337 | ||
8338 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8339 | return -EFAULT; | |
8340 | ||
8341 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8342 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8343 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8344 | ctx->cq_ev_fd = NULL; | |
8345 | return ret; | |
8346 | } | |
8347 | ||
8348 | return 0; | |
8349 | } | |
8350 | ||
8351 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8352 | { | |
8353 | if (ctx->cq_ev_fd) { | |
8354 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8355 | ctx->cq_ev_fd = NULL; | |
8356 | return 0; | |
8357 | } | |
8358 | ||
8359 | return -ENXIO; | |
8360 | } | |
8361 | ||
5a2e745d JA |
8362 | static int __io_destroy_buffers(int id, void *p, void *data) |
8363 | { | |
8364 | struct io_ring_ctx *ctx = data; | |
8365 | struct io_buffer *buf = p; | |
8366 | ||
067524e9 | 8367 | __io_remove_buffers(ctx, buf, id, -1U); |
5a2e745d JA |
8368 | return 0; |
8369 | } | |
8370 | ||
8371 | static void io_destroy_buffers(struct io_ring_ctx *ctx) | |
8372 | { | |
8373 | idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx); | |
8374 | idr_destroy(&ctx->io_buffer_idr); | |
8375 | } | |
8376 | ||
68e68ee6 | 8377 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8378 | { |
68e68ee6 | 8379 | struct io_kiocb *req, *nxt; |
1b4c351f | 8380 | |
68e68ee6 JA |
8381 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8382 | if (tsk && req->task != tsk) | |
8383 | continue; | |
1b4c351f JA |
8384 | list_del(&req->compl.list); |
8385 | kmem_cache_free(req_cachep, req); | |
8386 | } | |
8387 | } | |
8388 | ||
4010fec4 | 8389 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8390 | { |
bf019da7 | 8391 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8392 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8393 | |
9a4fdbd8 JA |
8394 | mutex_lock(&ctx->uring_lock); |
8395 | ||
8e5c66c4 | 8396 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8397 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8398 | submit_state->reqs); | |
8e5c66c4 PB |
8399 | submit_state->free_reqs = 0; |
8400 | } | |
9a4fdbd8 JA |
8401 | |
8402 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8403 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8404 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8405 | spin_unlock_irq(&ctx->completion_lock); |
8406 | ||
e5547d2c PB |
8407 | io_req_cache_free(&cs->free_list, NULL); |
8408 | ||
9a4fdbd8 JA |
8409 | mutex_unlock(&ctx->uring_lock); |
8410 | } | |
8411 | ||
2b188cc1 JA |
8412 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8413 | { | |
04fc6c80 PB |
8414 | /* |
8415 | * Some may use context even when all refs and requests have been put, | |
8416 | * and they are free to do so while still holding uring_lock, see | |
8417 | * __io_req_task_submit(). Wait for them to finish. | |
8418 | */ | |
8419 | mutex_lock(&ctx->uring_lock); | |
8420 | mutex_unlock(&ctx->uring_lock); | |
8421 | ||
37d1e2e3 | 8422 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8423 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8424 | |
37d1e2e3 | 8425 | if (ctx->mm_account) { |
2aede0e4 JA |
8426 | mmdrop(ctx->mm_account); |
8427 | ctx->mm_account = NULL; | |
30975825 | 8428 | } |
def596e9 | 8429 | |
8bad28d8 | 8430 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8431 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8432 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8433 | io_eventfd_unregister(ctx); |
5a2e745d | 8434 | io_destroy_buffers(ctx); |
41726c9a | 8435 | idr_destroy(&ctx->personality_idr); |
def596e9 | 8436 | |
2b188cc1 | 8437 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8438 | if (ctx->ring_sock) { |
8439 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8440 | sock_release(ctx->ring_sock); |
355e8d26 | 8441 | } |
2b188cc1 JA |
8442 | #endif |
8443 | ||
75b28aff | 8444 | io_mem_free(ctx->rings); |
2b188cc1 | 8445 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8446 | |
8447 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8448 | free_uid(ctx->user); |
4010fec4 | 8449 | io_req_caches_free(ctx); |
e941894e JA |
8450 | if (ctx->hash_map) |
8451 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8452 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8453 | kfree(ctx); |
8454 | } | |
8455 | ||
8456 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8457 | { | |
8458 | struct io_ring_ctx *ctx = file->private_data; | |
8459 | __poll_t mask = 0; | |
8460 | ||
8461 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8462 | /* |
8463 | * synchronizes with barrier from wq_has_sleeper call in | |
8464 | * io_commit_cqring | |
8465 | */ | |
2b188cc1 | 8466 | smp_rmb(); |
90554200 | 8467 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8468 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8469 | |
8470 | /* | |
8471 | * Don't flush cqring overflow list here, just do a simple check. | |
8472 | * Otherwise there could possible be ABBA deadlock: | |
8473 | * CPU0 CPU1 | |
8474 | * ---- ---- | |
8475 | * lock(&ctx->uring_lock); | |
8476 | * lock(&ep->mtx); | |
8477 | * lock(&ctx->uring_lock); | |
8478 | * lock(&ep->mtx); | |
8479 | * | |
8480 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8481 | * pushs them to do the flush. | |
8482 | */ | |
8483 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8484 | mask |= EPOLLIN | EPOLLRDNORM; |
8485 | ||
8486 | return mask; | |
8487 | } | |
8488 | ||
8489 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8490 | { | |
8491 | struct io_ring_ctx *ctx = file->private_data; | |
8492 | ||
8493 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8494 | } | |
8495 | ||
0bead8cd | 8496 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8497 | { |
4379bf8b | 8498 | const struct cred *creds; |
071698e1 | 8499 | |
4379bf8b JA |
8500 | creds = idr_remove(&ctx->personality_idr, id); |
8501 | if (creds) { | |
8502 | put_cred(creds); | |
0bead8cd | 8503 | return 0; |
1e6fa521 | 8504 | } |
0bead8cd YD |
8505 | |
8506 | return -EINVAL; | |
8507 | } | |
8508 | ||
8509 | static int io_remove_personalities(int id, void *p, void *data) | |
8510 | { | |
8511 | struct io_ring_ctx *ctx = data; | |
8512 | ||
8513 | io_unregister_personality(ctx, id); | |
071698e1 JA |
8514 | return 0; |
8515 | } | |
8516 | ||
ba50a036 | 8517 | static bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8518 | { |
28c4721b | 8519 | struct callback_head *work, *next; |
ba50a036 | 8520 | bool executed = false; |
7c25c0d1 JA |
8521 | |
8522 | do { | |
28c4721b | 8523 | work = xchg(&ctx->exit_task_work, NULL); |
7c25c0d1 JA |
8524 | if (!work) |
8525 | break; | |
8526 | ||
8527 | do { | |
8528 | next = work->next; | |
8529 | work->func(work); | |
8530 | work = next; | |
8531 | cond_resched(); | |
8532 | } while (work); | |
ba50a036 | 8533 | executed = true; |
7c25c0d1 | 8534 | } while (1); |
ba50a036 PB |
8535 | |
8536 | return executed; | |
7c25c0d1 JA |
8537 | } |
8538 | ||
d56d938b PB |
8539 | struct io_tctx_exit { |
8540 | struct callback_head task_work; | |
8541 | struct completion completion; | |
baf186c4 | 8542 | struct io_ring_ctx *ctx; |
d56d938b PB |
8543 | }; |
8544 | ||
8545 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8546 | { | |
8547 | struct io_uring_task *tctx = current->io_uring; | |
8548 | struct io_tctx_exit *work; | |
8549 | ||
8550 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8551 | /* | |
8552 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8553 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8554 | */ | |
8555 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8556 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8557 | complete(&work->completion); |
8558 | } | |
8559 | ||
85faa7b8 JA |
8560 | static void io_ring_exit_work(struct work_struct *work) |
8561 | { | |
d56d938b | 8562 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8563 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8564 | struct io_tctx_exit exit; |
8565 | struct io_tctx_node *node; | |
8566 | int ret; | |
85faa7b8 | 8567 | |
56952e91 JA |
8568 | /* |
8569 | * If we're doing polled IO and end up having requests being | |
8570 | * submitted async (out-of-line), then completions can come in while | |
8571 | * we're waiting for refs to drop. We need to reap these manually, | |
8572 | * as nobody else will be looking for them. | |
8573 | */ | |
b2edc0a7 | 8574 | do { |
9936c7c2 | 8575 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8576 | |
8577 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8578 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8579 | |
8580 | mutex_lock(&ctx->uring_lock); | |
8581 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8582 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8583 | ||
d56d938b PB |
8584 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8585 | ctx_node); | |
baf186c4 | 8586 | exit.ctx = ctx; |
d56d938b PB |
8587 | init_completion(&exit.completion); |
8588 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8589 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8590 | if (WARN_ON_ONCE(ret)) | |
8591 | continue; | |
8592 | wake_up_process(node->task); | |
8593 | ||
8594 | mutex_unlock(&ctx->uring_lock); | |
8595 | wait_for_completion(&exit.completion); | |
8596 | cond_resched(); | |
8597 | mutex_lock(&ctx->uring_lock); | |
8598 | } | |
8599 | mutex_unlock(&ctx->uring_lock); | |
8600 | ||
85faa7b8 JA |
8601 | io_ring_ctx_free(ctx); |
8602 | } | |
8603 | ||
2b188cc1 JA |
8604 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8605 | { | |
8606 | mutex_lock(&ctx->uring_lock); | |
8607 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8608 | /* if force is set, the ring is going away. always drop after that */ |
8609 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8610 | if (ctx->rings) |
6c503150 | 8611 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
5c766a90 | 8612 | idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx); |
2b188cc1 JA |
8613 | mutex_unlock(&ctx->uring_lock); |
8614 | ||
6b81928d PB |
8615 | io_kill_timeouts(ctx, NULL, NULL); |
8616 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8617 | |
15dff286 | 8618 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8619 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8620 | |
85faa7b8 | 8621 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8622 | /* |
8623 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8624 | * if we're exiting a ton of rings at the same time. It just adds | |
8625 | * noise and overhead, there's no discernable change in runtime | |
8626 | * over using system_wq. | |
8627 | */ | |
8628 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8629 | } |
8630 | ||
8631 | static int io_uring_release(struct inode *inode, struct file *file) | |
8632 | { | |
8633 | struct io_ring_ctx *ctx = file->private_data; | |
8634 | ||
8635 | file->private_data = NULL; | |
8636 | io_ring_ctx_wait_and_kill(ctx); | |
8637 | return 0; | |
8638 | } | |
8639 | ||
f6edbabb PB |
8640 | struct io_task_cancel { |
8641 | struct task_struct *task; | |
8642 | struct files_struct *files; | |
8643 | }; | |
f254ac04 | 8644 | |
f6edbabb | 8645 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8646 | { |
9a472ef7 | 8647 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8648 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8649 | bool ret; |
8650 | ||
f6edbabb | 8651 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8652 | unsigned long flags; |
8653 | struct io_ring_ctx *ctx = req->ctx; | |
8654 | ||
8655 | /* protect against races with linked timeouts */ | |
8656 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8657 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8658 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8659 | } else { | |
f6edbabb | 8660 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8661 | } |
8662 | return ret; | |
b711d4ea JA |
8663 | } |
8664 | ||
b7ddce3c | 8665 | static void io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8666 | struct task_struct *task, |
b7ddce3c PB |
8667 | struct files_struct *files) |
8668 | { | |
8669 | struct io_defer_entry *de = NULL; | |
8670 | LIST_HEAD(list); | |
8671 | ||
8672 | spin_lock_irq(&ctx->completion_lock); | |
8673 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8674 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8675 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8676 | break; | |
8677 | } | |
8678 | } | |
8679 | spin_unlock_irq(&ctx->completion_lock); | |
8680 | ||
8681 | while (!list_empty(&list)) { | |
8682 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8683 | list_del_init(&de->list); | |
8684 | req_set_fail_links(de->req); | |
8685 | io_put_req(de->req); | |
8686 | io_req_complete(de->req, -ECANCELED); | |
8687 | kfree(de); | |
8688 | } | |
8689 | } | |
8690 | ||
1b00764f PB |
8691 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8692 | { | |
8693 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8694 | ||
8695 | return req->ctx == data; | |
8696 | } | |
8697 | ||
8698 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8699 | { | |
8700 | struct io_tctx_node *node; | |
8701 | enum io_wq_cancel cret; | |
8702 | bool ret = false; | |
8703 | ||
8704 | mutex_lock(&ctx->uring_lock); | |
8705 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8706 | struct io_uring_task *tctx = node->task->io_uring; | |
8707 | ||
8708 | /* | |
8709 | * io_wq will stay alive while we hold uring_lock, because it's | |
8710 | * killed after ctx nodes, which requires to take the lock. | |
8711 | */ | |
8712 | if (!tctx || !tctx->io_wq) | |
8713 | continue; | |
8714 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8715 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8716 | } | |
8717 | mutex_unlock(&ctx->uring_lock); | |
8718 | ||
8719 | return ret; | |
8720 | } | |
8721 | ||
9936c7c2 PB |
8722 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8723 | struct task_struct *task, | |
8724 | struct files_struct *files) | |
8725 | { | |
8726 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8727 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8728 | |
8729 | while (1) { | |
8730 | enum io_wq_cancel cret; | |
8731 | bool ret = false; | |
8732 | ||
1b00764f PB |
8733 | if (!task) { |
8734 | ret |= io_uring_try_cancel_iowq(ctx); | |
8735 | } else if (tctx && tctx->io_wq) { | |
8736 | /* | |
8737 | * Cancels requests of all rings, not only @ctx, but | |
8738 | * it's fine as the task is in exit/exec. | |
8739 | */ | |
5aa75ed5 | 8740 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8741 | &cancel, true); |
8742 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8743 | } | |
8744 | ||
8745 | /* SQPOLL thread does its own polling */ | |
8746 | if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) { | |
8747 | while (!list_empty_careful(&ctx->iopoll_list)) { | |
8748 | io_iopoll_try_reap_events(ctx); | |
8749 | ret = true; | |
8750 | } | |
8751 | } | |
8752 | ||
8753 | ret |= io_poll_remove_all(ctx, task, files); | |
8754 | ret |= io_kill_timeouts(ctx, task, files); | |
8755 | ret |= io_run_task_work(); | |
ba50a036 | 8756 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8757 | io_cqring_overflow_flush(ctx, true, task, files); |
8758 | if (!ret) | |
8759 | break; | |
8760 | cond_resched(); | |
8761 | } | |
8762 | } | |
8763 | ||
ca70f00b PB |
8764 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8765 | struct task_struct *task, | |
8766 | struct files_struct *files) | |
8767 | { | |
8768 | struct io_kiocb *req; | |
8769 | int cnt = 0; | |
8770 | ||
8771 | spin_lock_irq(&ctx->inflight_lock); | |
8772 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8773 | cnt += io_match_task(req, task, files); | |
8774 | spin_unlock_irq(&ctx->inflight_lock); | |
8775 | return cnt; | |
8776 | } | |
8777 | ||
b52fda00 | 8778 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8779 | struct task_struct *task, |
fcb323cc JA |
8780 | struct files_struct *files) |
8781 | { | |
fcb323cc | 8782 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8783 | DEFINE_WAIT(wait); |
ca70f00b | 8784 | int inflight; |
fcb323cc | 8785 | |
ca70f00b PB |
8786 | inflight = io_uring_count_inflight(ctx, task, files); |
8787 | if (!inflight) | |
fcb323cc | 8788 | break; |
f6edbabb | 8789 | |
9936c7c2 | 8790 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b | 8791 | |
34343786 PB |
8792 | if (ctx->sq_data) |
8793 | io_sq_thread_unpark(ctx->sq_data); | |
ca70f00b PB |
8794 | prepare_to_wait(&task->io_uring->wait, &wait, |
8795 | TASK_UNINTERRUPTIBLE); | |
8796 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8797 | schedule(); | |
c98de08c | 8798 | finish_wait(&task->io_uring->wait, &wait); |
34343786 PB |
8799 | if (ctx->sq_data) |
8800 | io_sq_thread_park(ctx->sq_data); | |
0f212204 | 8801 | } |
0f212204 JA |
8802 | } |
8803 | ||
8804 | /* | |
8805 | * We need to iteratively cancel requests, in case a request has dependent | |
8806 | * hard links. These persist even for failure of cancelations, hence keep | |
8807 | * looping until none are found. | |
8808 | */ | |
8809 | static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx, | |
8810 | struct files_struct *files) | |
8811 | { | |
8812 | struct task_struct *task = current; | |
8813 | ||
fdaf083c | 8814 | if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) { |
70aacfe6 PB |
8815 | /* never started, nothing to cancel */ |
8816 | if (ctx->flags & IORING_SETUP_R_DISABLED) { | |
8817 | io_sq_offload_start(ctx); | |
8818 | return; | |
8819 | } | |
86e0d676 JA |
8820 | io_sq_thread_park(ctx->sq_data); |
8821 | task = ctx->sq_data->thread; | |
8822 | if (task) | |
37d1e2e3 | 8823 | atomic_inc(&task->io_uring->in_idle); |
fdaf083c | 8824 | } |
0f212204 | 8825 | |
df9923f9 | 8826 | io_cancel_defer_files(ctx, task, files); |
0f212204 | 8827 | |
3a7efd1a | 8828 | io_uring_cancel_files(ctx, task, files); |
b52fda00 | 8829 | if (!files) |
9936c7c2 | 8830 | io_uring_try_cancel_requests(ctx, task, NULL); |
fdaf083c | 8831 | |
86e0d676 | 8832 | if (task) |
fdaf083c | 8833 | atomic_dec(&task->io_uring->in_idle); |
86e0d676 | 8834 | if (ctx->sq_data) |
fdaf083c | 8835 | io_sq_thread_unpark(ctx->sq_data); |
0f212204 JA |
8836 | } |
8837 | ||
8838 | /* | |
8839 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8840 | */ | |
baf186c4 | 8841 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8842 | { |
236434c3 | 8843 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8844 | struct io_tctx_node *node; |
a528b04e | 8845 | int ret; |
236434c3 MWO |
8846 | |
8847 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8848 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8849 | if (unlikely(ret)) |
8850 | return ret; | |
236434c3 | 8851 | tctx = current->io_uring; |
0f212204 | 8852 | } |
baf186c4 PB |
8853 | if (tctx->last != ctx) { |
8854 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8855 | |
236434c3 | 8856 | if (!old) { |
13bf43f5 PB |
8857 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8858 | if (!node) | |
8859 | return -ENOMEM; | |
8860 | node->ctx = ctx; | |
13bf43f5 PB |
8861 | node->task = current; |
8862 | ||
baf186c4 | 8863 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8864 | node, GFP_KERNEL)); |
a528b04e | 8865 | if (ret) { |
13bf43f5 | 8866 | kfree(node); |
a528b04e PB |
8867 | return ret; |
8868 | } | |
13bf43f5 PB |
8869 | |
8870 | mutex_lock(&ctx->uring_lock); | |
8871 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8872 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8873 | } |
baf186c4 | 8874 | tctx->last = ctx; |
0f212204 JA |
8875 | } |
8876 | ||
fdaf083c JA |
8877 | /* |
8878 | * This is race safe in that the task itself is doing this, hence it | |
8879 | * cannot be going through the exit/cancel paths at the same time. | |
8880 | * This cannot be modified while exit/cancel is running. | |
8881 | */ | |
8882 | if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL)) | |
8883 | tctx->sqpoll = true; | |
8884 | ||
0f212204 JA |
8885 | return 0; |
8886 | } | |
8887 | ||
8888 | /* | |
8889 | * Remove this io_uring_file -> task mapping. | |
8890 | */ | |
2941267b | 8891 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8892 | { |
8893 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8894 | struct io_tctx_node *node; |
2941267b | 8895 | |
eebd2e37 PB |
8896 | if (!tctx) |
8897 | return; | |
13bf43f5 PB |
8898 | node = xa_erase(&tctx->xa, index); |
8899 | if (!node) | |
2941267b | 8900 | return; |
0f212204 | 8901 | |
13bf43f5 PB |
8902 | WARN_ON_ONCE(current != node->task); |
8903 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8904 | ||
8905 | mutex_lock(&node->ctx->uring_lock); | |
8906 | list_del(&node->ctx_node); | |
8907 | mutex_unlock(&node->ctx->uring_lock); | |
8908 | ||
baf186c4 | 8909 | if (tctx->last == node->ctx) |
0f212204 | 8910 | tctx->last = NULL; |
13bf43f5 | 8911 | kfree(node); |
0f212204 JA |
8912 | } |
8913 | ||
8452d4a6 | 8914 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8915 | { |
13bf43f5 | 8916 | struct io_tctx_node *node; |
de7f1d9e PB |
8917 | unsigned long index; |
8918 | ||
13bf43f5 | 8919 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8920 | io_uring_del_task_file(index); |
8452d4a6 PB |
8921 | if (tctx->io_wq) { |
8922 | io_wq_put_and_exit(tctx->io_wq); | |
8923 | tctx->io_wq = NULL; | |
8924 | } | |
de7f1d9e PB |
8925 | } |
8926 | ||
0f212204 JA |
8927 | void __io_uring_files_cancel(struct files_struct *files) |
8928 | { | |
8929 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8930 | struct io_tctx_node *node; |
ce765372 | 8931 | unsigned long index; |
0f212204 JA |
8932 | |
8933 | /* make sure overflow events are dropped */ | |
fdaf083c | 8934 | atomic_inc(&tctx->in_idle); |
13bf43f5 PB |
8935 | xa_for_each(&tctx->xa, index, node) |
8936 | io_uring_cancel_task_requests(node->ctx, files); | |
fdaf083c | 8937 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8938 | |
8452d4a6 PB |
8939 | if (files) |
8940 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8941 | } |
8942 | ||
8943 | static s64 tctx_inflight(struct io_uring_task *tctx) | |
8944 | { | |
0e9ddb39 PB |
8945 | return percpu_counter_sum(&tctx->inflight); |
8946 | } | |
fdaf083c | 8947 | |
0e9ddb39 PB |
8948 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8949 | { | |
37d1e2e3 | 8950 | struct io_sq_data *sqd = ctx->sq_data; |
0e9ddb39 PB |
8951 | struct io_uring_task *tctx; |
8952 | s64 inflight; | |
8953 | DEFINE_WAIT(wait); | |
fdaf083c | 8954 | |
37d1e2e3 | 8955 | if (!sqd) |
0e9ddb39 | 8956 | return; |
86e0d676 JA |
8957 | io_sq_thread_park(sqd); |
8958 | if (!sqd->thread || !sqd->thread->io_uring) { | |
e54945ae JA |
8959 | io_sq_thread_unpark(sqd); |
8960 | return; | |
8961 | } | |
86e0d676 | 8962 | tctx = ctx->sq_data->thread->io_uring; |
0e9ddb39 PB |
8963 | atomic_inc(&tctx->in_idle); |
8964 | do { | |
8965 | /* read completions before cancelations */ | |
8966 | inflight = tctx_inflight(tctx); | |
8967 | if (!inflight) | |
8968 | break; | |
8969 | io_uring_cancel_task_requests(ctx, NULL); | |
fdaf083c | 8970 | |
0e9ddb39 PB |
8971 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8972 | /* | |
8973 | * If we've seen completions, retry without waiting. This | |
8974 | * avoids a race where a completion comes in before we did | |
8975 | * prepare_to_wait(). | |
8976 | */ | |
8977 | if (inflight == tctx_inflight(tctx)) | |
8978 | schedule(); | |
8979 | finish_wait(&tctx->wait, &wait); | |
8980 | } while (1); | |
8981 | atomic_dec(&tctx->in_idle); | |
37d1e2e3 | 8982 | io_sq_thread_unpark(sqd); |
0f212204 JA |
8983 | } |
8984 | ||
0f212204 JA |
8985 | /* |
8986 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8987 | * requests. | |
8988 | */ | |
8989 | void __io_uring_task_cancel(void) | |
8990 | { | |
8991 | struct io_uring_task *tctx = current->io_uring; | |
8992 | DEFINE_WAIT(wait); | |
d8a6df10 | 8993 | s64 inflight; |
0f212204 JA |
8994 | |
8995 | /* make sure overflow events are dropped */ | |
fdaf083c | 8996 | atomic_inc(&tctx->in_idle); |
0f212204 | 8997 | |
0e9ddb39 | 8998 | if (tctx->sqpoll) { |
13bf43f5 | 8999 | struct io_tctx_node *node; |
0e9ddb39 PB |
9000 | unsigned long index; |
9001 | ||
13bf43f5 PB |
9002 | xa_for_each(&tctx->xa, index, node) |
9003 | io_uring_cancel_sqpoll(node->ctx); | |
0e9ddb39 | 9004 | } |
0b5cd6c3 | 9005 | |
d8a6df10 | 9006 | do { |
0f212204 | 9007 | /* read completions before cancelations */ |
fdaf083c | 9008 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9009 | if (!inflight) |
9010 | break; | |
0f212204 JA |
9011 | __io_uring_files_cancel(NULL); |
9012 | ||
9013 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9014 | ||
9015 | /* | |
a1bb3cd5 PB |
9016 | * If we've seen completions, retry without waiting. This |
9017 | * avoids a race where a completion comes in before we did | |
9018 | * prepare_to_wait(). | |
0f212204 | 9019 | */ |
a1bb3cd5 PB |
9020 | if (inflight == tctx_inflight(tctx)) |
9021 | schedule(); | |
f57555ed | 9022 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9023 | } while (1); |
0f212204 | 9024 | |
fdaf083c | 9025 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9026 | |
8452d4a6 PB |
9027 | io_uring_clean_tctx(tctx); |
9028 | /* all current's requests should be gone, we can kill tctx */ | |
9029 | __io_uring_free(current); | |
44e728b8 PB |
9030 | } |
9031 | ||
6c5c240e RP |
9032 | static void *io_uring_validate_mmap_request(struct file *file, |
9033 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9034 | { |
2b188cc1 | 9035 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9036 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9037 | struct page *page; |
9038 | void *ptr; | |
9039 | ||
9040 | switch (offset) { | |
9041 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9042 | case IORING_OFF_CQ_RING: |
9043 | ptr = ctx->rings; | |
2b188cc1 JA |
9044 | break; |
9045 | case IORING_OFF_SQES: | |
9046 | ptr = ctx->sq_sqes; | |
9047 | break; | |
2b188cc1 | 9048 | default: |
6c5c240e | 9049 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9050 | } |
9051 | ||
9052 | page = virt_to_head_page(ptr); | |
a50b854e | 9053 | if (sz > page_size(page)) |
6c5c240e RP |
9054 | return ERR_PTR(-EINVAL); |
9055 | ||
9056 | return ptr; | |
9057 | } | |
9058 | ||
9059 | #ifdef CONFIG_MMU | |
9060 | ||
9061 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9062 | { | |
9063 | size_t sz = vma->vm_end - vma->vm_start; | |
9064 | unsigned long pfn; | |
9065 | void *ptr; | |
9066 | ||
9067 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9068 | if (IS_ERR(ptr)) | |
9069 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9070 | |
9071 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9072 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9073 | } | |
9074 | ||
6c5c240e RP |
9075 | #else /* !CONFIG_MMU */ |
9076 | ||
9077 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9078 | { | |
9079 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9080 | } | |
9081 | ||
9082 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9083 | { | |
9084 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9085 | } | |
9086 | ||
9087 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9088 | unsigned long addr, unsigned long len, | |
9089 | unsigned long pgoff, unsigned long flags) | |
9090 | { | |
9091 | void *ptr; | |
9092 | ||
9093 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9094 | if (IS_ERR(ptr)) | |
9095 | return PTR_ERR(ptr); | |
9096 | ||
9097 | return (unsigned long) ptr; | |
9098 | } | |
9099 | ||
9100 | #endif /* !CONFIG_MMU */ | |
9101 | ||
d9d05217 | 9102 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 | 9103 | { |
d9d05217 | 9104 | int ret = 0; |
90554200 JA |
9105 | DEFINE_WAIT(wait); |
9106 | ||
9107 | do { | |
9108 | if (!io_sqring_full(ctx)) | |
9109 | break; | |
90554200 JA |
9110 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9111 | ||
9112 | if (!io_sqring_full(ctx)) | |
9113 | break; | |
90554200 JA |
9114 | schedule(); |
9115 | } while (!signal_pending(current)); | |
9116 | ||
9117 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
d9d05217 | 9118 | return ret; |
90554200 JA |
9119 | } |
9120 | ||
c73ebb68 HX |
9121 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9122 | struct __kernel_timespec __user **ts, | |
9123 | const sigset_t __user **sig) | |
9124 | { | |
9125 | struct io_uring_getevents_arg arg; | |
9126 | ||
9127 | /* | |
9128 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9129 | * is just a pointer to the sigset_t. | |
9130 | */ | |
9131 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9132 | *sig = (const sigset_t __user *) argp; | |
9133 | *ts = NULL; | |
9134 | return 0; | |
9135 | } | |
9136 | ||
9137 | /* | |
9138 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9139 | * timespec and sigset_t pointers if good. | |
9140 | */ | |
9141 | if (*argsz != sizeof(arg)) | |
9142 | return -EINVAL; | |
9143 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9144 | return -EFAULT; | |
9145 | *sig = u64_to_user_ptr(arg.sigmask); | |
9146 | *argsz = arg.sigmask_sz; | |
9147 | *ts = u64_to_user_ptr(arg.ts); | |
9148 | return 0; | |
9149 | } | |
9150 | ||
2b188cc1 | 9151 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9152 | u32, min_complete, u32, flags, const void __user *, argp, |
9153 | size_t, argsz) | |
2b188cc1 JA |
9154 | { |
9155 | struct io_ring_ctx *ctx; | |
9156 | long ret = -EBADF; | |
9157 | int submitted = 0; | |
9158 | struct fd f; | |
9159 | ||
4c6e277c | 9160 | io_run_task_work(); |
b41e9852 | 9161 | |
90554200 | 9162 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9163 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9164 | return -EINVAL; |
9165 | ||
9166 | f = fdget(fd); | |
9167 | if (!f.file) | |
9168 | return -EBADF; | |
9169 | ||
9170 | ret = -EOPNOTSUPP; | |
9171 | if (f.file->f_op != &io_uring_fops) | |
9172 | goto out_fput; | |
9173 | ||
9174 | ret = -ENXIO; | |
9175 | ctx = f.file->private_data; | |
9176 | if (!percpu_ref_tryget(&ctx->refs)) | |
9177 | goto out_fput; | |
9178 | ||
7e84e1c7 SG |
9179 | ret = -EBADFD; |
9180 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9181 | goto out; | |
9182 | ||
6c271ce2 JA |
9183 | /* |
9184 | * For SQ polling, the thread will do all submissions and completions. | |
9185 | * Just return the requested submit count, and wake the thread if | |
9186 | * we were asked to. | |
9187 | */ | |
b2a9eada | 9188 | ret = 0; |
6c271ce2 | 9189 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9190 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9191 | |
5f3f26f9 JA |
9192 | if (unlikely(ctx->sqo_exec)) { |
9193 | ret = io_sq_thread_fork(ctx->sq_data, ctx); | |
9194 | if (ret) | |
9195 | goto out; | |
9196 | ctx->sqo_exec = 0; | |
9197 | } | |
d9d05217 | 9198 | ret = -EOWNERDEAD; |
6c271ce2 | 9199 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9200 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9201 | if (flags & IORING_ENTER_SQ_WAIT) { |
9202 | ret = io_sqpoll_wait_sq(ctx); | |
9203 | if (ret) | |
9204 | goto out; | |
9205 | } | |
6c271ce2 | 9206 | submitted = to_submit; |
b2a9eada | 9207 | } else if (to_submit) { |
baf186c4 | 9208 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9209 | if (unlikely(ret)) |
9210 | goto out; | |
2b188cc1 | 9211 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9212 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9213 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9214 | |
9215 | if (submitted != to_submit) | |
9216 | goto out; | |
2b188cc1 JA |
9217 | } |
9218 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9219 | const sigset_t __user *sig; |
9220 | struct __kernel_timespec __user *ts; | |
9221 | ||
9222 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9223 | if (unlikely(ret)) | |
9224 | goto out; | |
9225 | ||
2b188cc1 JA |
9226 | min_complete = min(min_complete, ctx->cq_entries); |
9227 | ||
32b2244a XW |
9228 | /* |
9229 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9230 | * space applications don't need to do io completion events | |
9231 | * polling again, they can rely on io_sq_thread to do polling | |
9232 | * work, which can reduce cpu usage and uring_lock contention. | |
9233 | */ | |
9234 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9235 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9236 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9237 | } else { |
c73ebb68 | 9238 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9239 | } |
2b188cc1 JA |
9240 | } |
9241 | ||
7c504e65 | 9242 | out: |
6805b32e | 9243 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9244 | out_fput: |
9245 | fdput(f); | |
9246 | return submitted ? submitted : ret; | |
9247 | } | |
9248 | ||
bebdb65e | 9249 | #ifdef CONFIG_PROC_FS |
87ce955b JA |
9250 | static int io_uring_show_cred(int id, void *p, void *data) |
9251 | { | |
4379bf8b | 9252 | const struct cred *cred = p; |
87ce955b JA |
9253 | struct seq_file *m = data; |
9254 | struct user_namespace *uns = seq_user_ns(m); | |
9255 | struct group_info *gi; | |
9256 | kernel_cap_t cap; | |
9257 | unsigned __capi; | |
9258 | int g; | |
9259 | ||
9260 | seq_printf(m, "%5d\n", id); | |
9261 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9262 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9263 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9264 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9265 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9266 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9267 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9268 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9269 | seq_puts(m, "\n\tGroups:\t"); | |
9270 | gi = cred->group_info; | |
9271 | for (g = 0; g < gi->ngroups; g++) { | |
9272 | seq_put_decimal_ull(m, g ? " " : "", | |
9273 | from_kgid_munged(uns, gi->gid[g])); | |
9274 | } | |
9275 | seq_puts(m, "\n\tCapEff:\t"); | |
9276 | cap = cred->cap_effective; | |
9277 | CAP_FOR_EACH_U32(__capi) | |
9278 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9279 | seq_putc(m, '\n'); | |
9280 | return 0; | |
9281 | } | |
9282 | ||
9283 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9284 | { | |
dbbe9c64 | 9285 | struct io_sq_data *sq = NULL; |
fad8e0de | 9286 | bool has_lock; |
87ce955b JA |
9287 | int i; |
9288 | ||
fad8e0de JA |
9289 | /* |
9290 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9291 | * since fdinfo case grabs it in the opposite direction of normal use | |
9292 | * cases. If we fail to get the lock, we just don't iterate any | |
9293 | * structures that could be going away outside the io_uring mutex. | |
9294 | */ | |
9295 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9296 | ||
5f3f26f9 | 9297 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9298 | sq = ctx->sq_data; |
5f3f26f9 JA |
9299 | if (!sq->thread) |
9300 | sq = NULL; | |
9301 | } | |
dbbe9c64 JQ |
9302 | |
9303 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9304 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9305 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9306 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9307 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9308 | |
87ce955b JA |
9309 | if (f) |
9310 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9311 | else | |
9312 | seq_printf(m, "%5u: <none>\n", i); | |
9313 | } | |
9314 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9315 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9316 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9317 | ||
9318 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9319 | (unsigned int) buf->len); | |
9320 | } | |
fad8e0de | 9321 | if (has_lock && !idr_is_empty(&ctx->personality_idr)) { |
87ce955b JA |
9322 | seq_printf(m, "Personalities:\n"); |
9323 | idr_for_each(&ctx->personality_idr, io_uring_show_cred, m); | |
9324 | } | |
d7718a9d JA |
9325 | seq_printf(m, "PollList:\n"); |
9326 | spin_lock_irq(&ctx->completion_lock); | |
9327 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9328 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9329 | struct io_kiocb *req; | |
9330 | ||
9331 | hlist_for_each_entry(req, list, hash_node) | |
9332 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9333 | req->task->task_works != NULL); | |
9334 | } | |
9335 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9336 | if (has_lock) |
9337 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9338 | } |
9339 | ||
9340 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9341 | { | |
9342 | struct io_ring_ctx *ctx = f->private_data; | |
9343 | ||
9344 | if (percpu_ref_tryget(&ctx->refs)) { | |
9345 | __io_uring_show_fdinfo(ctx, m); | |
9346 | percpu_ref_put(&ctx->refs); | |
9347 | } | |
9348 | } | |
bebdb65e | 9349 | #endif |
87ce955b | 9350 | |
2b188cc1 JA |
9351 | static const struct file_operations io_uring_fops = { |
9352 | .release = io_uring_release, | |
9353 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9354 | #ifndef CONFIG_MMU |
9355 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9356 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9357 | #endif | |
2b188cc1 JA |
9358 | .poll = io_uring_poll, |
9359 | .fasync = io_uring_fasync, | |
bebdb65e | 9360 | #ifdef CONFIG_PROC_FS |
87ce955b | 9361 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9362 | #endif |
2b188cc1 JA |
9363 | }; |
9364 | ||
9365 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9366 | struct io_uring_params *p) | |
9367 | { | |
75b28aff HV |
9368 | struct io_rings *rings; |
9369 | size_t size, sq_array_offset; | |
2b188cc1 | 9370 | |
bd740481 JA |
9371 | /* make sure these are sane, as we already accounted them */ |
9372 | ctx->sq_entries = p->sq_entries; | |
9373 | ctx->cq_entries = p->cq_entries; | |
9374 | ||
75b28aff HV |
9375 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9376 | if (size == SIZE_MAX) | |
9377 | return -EOVERFLOW; | |
9378 | ||
9379 | rings = io_mem_alloc(size); | |
9380 | if (!rings) | |
2b188cc1 JA |
9381 | return -ENOMEM; |
9382 | ||
75b28aff HV |
9383 | ctx->rings = rings; |
9384 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9385 | rings->sq_ring_mask = p->sq_entries - 1; | |
9386 | rings->cq_ring_mask = p->cq_entries - 1; | |
9387 | rings->sq_ring_entries = p->sq_entries; | |
9388 | rings->cq_ring_entries = p->cq_entries; | |
9389 | ctx->sq_mask = rings->sq_ring_mask; | |
9390 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9391 | |
9392 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9393 | if (size == SIZE_MAX) { |
9394 | io_mem_free(ctx->rings); | |
9395 | ctx->rings = NULL; | |
2b188cc1 | 9396 | return -EOVERFLOW; |
eb065d30 | 9397 | } |
2b188cc1 JA |
9398 | |
9399 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9400 | if (!ctx->sq_sqes) { |
9401 | io_mem_free(ctx->rings); | |
9402 | ctx->rings = NULL; | |
2b188cc1 | 9403 | return -ENOMEM; |
eb065d30 | 9404 | } |
2b188cc1 | 9405 | |
2b188cc1 JA |
9406 | return 0; |
9407 | } | |
9408 | ||
9faadcc8 PB |
9409 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9410 | { | |
9411 | int ret, fd; | |
9412 | ||
9413 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9414 | if (fd < 0) | |
9415 | return fd; | |
9416 | ||
baf186c4 | 9417 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9418 | if (ret) { |
9419 | put_unused_fd(fd); | |
9420 | return ret; | |
9421 | } | |
9422 | fd_install(fd, file); | |
9423 | return fd; | |
9424 | } | |
9425 | ||
2b188cc1 JA |
9426 | /* |
9427 | * Allocate an anonymous fd, this is what constitutes the application | |
9428 | * visible backing of an io_uring instance. The application mmaps this | |
9429 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9430 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9431 | */ | |
9faadcc8 | 9432 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9433 | { |
9434 | struct file *file; | |
9faadcc8 | 9435 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9436 | int ret; |
9437 | ||
2b188cc1 JA |
9438 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9439 | &ctx->ring_sock); | |
9440 | if (ret) | |
9faadcc8 | 9441 | return ERR_PTR(ret); |
2b188cc1 JA |
9442 | #endif |
9443 | ||
2b188cc1 JA |
9444 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9445 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9446 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9447 | if (IS_ERR(file)) { |
9448 | sock_release(ctx->ring_sock); | |
9449 | ctx->ring_sock = NULL; | |
9450 | } else { | |
9451 | ctx->ring_sock->file = file; | |
0f212204 | 9452 | } |
2b188cc1 | 9453 | #endif |
9faadcc8 | 9454 | return file; |
2b188cc1 JA |
9455 | } |
9456 | ||
7f13657d XW |
9457 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9458 | struct io_uring_params __user *params) | |
2b188cc1 | 9459 | { |
2b188cc1 | 9460 | struct io_ring_ctx *ctx; |
9faadcc8 | 9461 | struct file *file; |
2b188cc1 JA |
9462 | int ret; |
9463 | ||
8110c1a6 | 9464 | if (!entries) |
2b188cc1 | 9465 | return -EINVAL; |
8110c1a6 JA |
9466 | if (entries > IORING_MAX_ENTRIES) { |
9467 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9468 | return -EINVAL; | |
9469 | entries = IORING_MAX_ENTRIES; | |
9470 | } | |
2b188cc1 JA |
9471 | |
9472 | /* | |
9473 | * Use twice as many entries for the CQ ring. It's possible for the | |
9474 | * application to drive a higher depth than the size of the SQ ring, | |
9475 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9476 | * some flexibility in overcommitting a bit. If the application has |
9477 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9478 | * of CQ ring entries manually. | |
2b188cc1 JA |
9479 | */ |
9480 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9481 | if (p->flags & IORING_SETUP_CQSIZE) { |
9482 | /* | |
9483 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9484 | * to a power-of-two, if it isn't already. We do NOT impose | |
9485 | * any cq vs sq ring sizing. | |
9486 | */ | |
eb2667b3 | 9487 | if (!p->cq_entries) |
33a107f0 | 9488 | return -EINVAL; |
8110c1a6 JA |
9489 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9490 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9491 | return -EINVAL; | |
9492 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9493 | } | |
eb2667b3 JQ |
9494 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9495 | if (p->cq_entries < p->sq_entries) | |
9496 | return -EINVAL; | |
33a107f0 JA |
9497 | } else { |
9498 | p->cq_entries = 2 * p->sq_entries; | |
9499 | } | |
2b188cc1 | 9500 | |
2b188cc1 | 9501 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9502 | if (!ctx) |
2b188cc1 | 9503 | return -ENOMEM; |
2b188cc1 | 9504 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9505 | if (!capable(CAP_IPC_LOCK)) |
9506 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9507 | |
9508 | /* | |
9509 | * This is just grabbed for accounting purposes. When a process exits, | |
9510 | * the mm is exited and dropped before the files, hence we need to hang | |
9511 | * on to this mm purely for the purposes of being able to unaccount | |
9512 | * memory (locked/pinned vm). It's not used for anything else. | |
9513 | */ | |
6b7898eb | 9514 | mmgrab(current->mm); |
2aede0e4 | 9515 | ctx->mm_account = current->mm; |
6b7898eb | 9516 | |
2b188cc1 JA |
9517 | ret = io_allocate_scq_urings(ctx, p); |
9518 | if (ret) | |
9519 | goto err; | |
9520 | ||
7e84e1c7 | 9521 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9522 | if (ret) |
9523 | goto err; | |
9524 | ||
7e84e1c7 SG |
9525 | if (!(p->flags & IORING_SETUP_R_DISABLED)) |
9526 | io_sq_offload_start(ctx); | |
9527 | ||
2b188cc1 | 9528 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9529 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9530 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9531 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9532 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9533 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9534 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9535 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9536 | |
9537 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9538 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9539 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9540 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9541 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9542 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9543 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9544 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9545 | |
7f13657d XW |
9546 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9547 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9548 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9549 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9550 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9551 | |
9552 | if (copy_to_user(params, p, sizeof(*p))) { | |
9553 | ret = -EFAULT; | |
9554 | goto err; | |
9555 | } | |
d1719f70 | 9556 | |
9faadcc8 PB |
9557 | file = io_uring_get_file(ctx); |
9558 | if (IS_ERR(file)) { | |
9559 | ret = PTR_ERR(file); | |
9560 | goto err; | |
9561 | } | |
9562 | ||
044c1ab3 JA |
9563 | /* |
9564 | * Install ring fd as the very last thing, so we don't risk someone | |
9565 | * having closed it before we finish setup | |
9566 | */ | |
9faadcc8 PB |
9567 | ret = io_uring_install_fd(ctx, file); |
9568 | if (ret < 0) { | |
9569 | /* fput will clean it up */ | |
9570 | fput(file); | |
9571 | return ret; | |
9572 | } | |
044c1ab3 | 9573 | |
c826bd7a | 9574 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9575 | return ret; |
9576 | err: | |
9577 | io_ring_ctx_wait_and_kill(ctx); | |
9578 | return ret; | |
9579 | } | |
9580 | ||
9581 | /* | |
9582 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9583 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9584 | * params structure passed in. | |
9585 | */ | |
9586 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9587 | { | |
9588 | struct io_uring_params p; | |
2b188cc1 JA |
9589 | int i; |
9590 | ||
9591 | if (copy_from_user(&p, params, sizeof(p))) | |
9592 | return -EFAULT; | |
9593 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9594 | if (p.resv[i]) | |
9595 | return -EINVAL; | |
9596 | } | |
9597 | ||
6c271ce2 | 9598 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9599 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9600 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9601 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9602 | return -EINVAL; |
9603 | ||
7f13657d | 9604 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9605 | } |
9606 | ||
9607 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9608 | struct io_uring_params __user *, params) | |
9609 | { | |
9610 | return io_uring_setup(entries, params); | |
9611 | } | |
9612 | ||
66f4af93 JA |
9613 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9614 | { | |
9615 | struct io_uring_probe *p; | |
9616 | size_t size; | |
9617 | int i, ret; | |
9618 | ||
9619 | size = struct_size(p, ops, nr_args); | |
9620 | if (size == SIZE_MAX) | |
9621 | return -EOVERFLOW; | |
9622 | p = kzalloc(size, GFP_KERNEL); | |
9623 | if (!p) | |
9624 | return -ENOMEM; | |
9625 | ||
9626 | ret = -EFAULT; | |
9627 | if (copy_from_user(p, arg, size)) | |
9628 | goto out; | |
9629 | ret = -EINVAL; | |
9630 | if (memchr_inv(p, 0, size)) | |
9631 | goto out; | |
9632 | ||
9633 | p->last_op = IORING_OP_LAST - 1; | |
9634 | if (nr_args > IORING_OP_LAST) | |
9635 | nr_args = IORING_OP_LAST; | |
9636 | ||
9637 | for (i = 0; i < nr_args; i++) { | |
9638 | p->ops[i].op = i; | |
9639 | if (!io_op_defs[i].not_supported) | |
9640 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9641 | } | |
9642 | p->ops_len = i; | |
9643 | ||
9644 | ret = 0; | |
9645 | if (copy_to_user(arg, p, size)) | |
9646 | ret = -EFAULT; | |
9647 | out: | |
9648 | kfree(p); | |
9649 | return ret; | |
9650 | } | |
9651 | ||
071698e1 JA |
9652 | static int io_register_personality(struct io_ring_ctx *ctx) |
9653 | { | |
4379bf8b | 9654 | const struct cred *creds; |
1e6fa521 | 9655 | int ret; |
071698e1 | 9656 | |
4379bf8b | 9657 | creds = get_current_cred(); |
1e6fa521 | 9658 | |
4379bf8b JA |
9659 | ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1, |
9660 | USHRT_MAX, GFP_KERNEL); | |
9661 | if (ret < 0) | |
9662 | put_cred(creds); | |
1e6fa521 | 9663 | return ret; |
071698e1 JA |
9664 | } |
9665 | ||
21b55dbc SG |
9666 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9667 | unsigned int nr_args) | |
9668 | { | |
9669 | struct io_uring_restriction *res; | |
9670 | size_t size; | |
9671 | int i, ret; | |
9672 | ||
7e84e1c7 SG |
9673 | /* Restrictions allowed only if rings started disabled */ |
9674 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9675 | return -EBADFD; | |
9676 | ||
21b55dbc | 9677 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9678 | if (ctx->restrictions.registered) |
21b55dbc SG |
9679 | return -EBUSY; |
9680 | ||
9681 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9682 | return -EINVAL; | |
9683 | ||
9684 | size = array_size(nr_args, sizeof(*res)); | |
9685 | if (size == SIZE_MAX) | |
9686 | return -EOVERFLOW; | |
9687 | ||
9688 | res = memdup_user(arg, size); | |
9689 | if (IS_ERR(res)) | |
9690 | return PTR_ERR(res); | |
9691 | ||
9692 | ret = 0; | |
9693 | ||
9694 | for (i = 0; i < nr_args; i++) { | |
9695 | switch (res[i].opcode) { | |
9696 | case IORING_RESTRICTION_REGISTER_OP: | |
9697 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9698 | ret = -EINVAL; | |
9699 | goto out; | |
9700 | } | |
9701 | ||
9702 | __set_bit(res[i].register_op, | |
9703 | ctx->restrictions.register_op); | |
9704 | break; | |
9705 | case IORING_RESTRICTION_SQE_OP: | |
9706 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9707 | ret = -EINVAL; | |
9708 | goto out; | |
9709 | } | |
9710 | ||
9711 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9712 | break; | |
9713 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9714 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9715 | break; | |
9716 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9717 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9718 | break; | |
9719 | default: | |
9720 | ret = -EINVAL; | |
9721 | goto out; | |
9722 | } | |
9723 | } | |
9724 | ||
9725 | out: | |
9726 | /* Reset all restrictions if an error happened */ | |
9727 | if (ret != 0) | |
9728 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9729 | else | |
7e84e1c7 | 9730 | ctx->restrictions.registered = true; |
21b55dbc SG |
9731 | |
9732 | kfree(res); | |
9733 | return ret; | |
9734 | } | |
9735 | ||
7e84e1c7 SG |
9736 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9737 | { | |
9738 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9739 | return -EBADFD; | |
9740 | ||
9741 | if (ctx->restrictions.registered) | |
9742 | ctx->restricted = 1; | |
9743 | ||
7e84e1c7 | 9744 | io_sq_offload_start(ctx); |
7e84e1c7 SG |
9745 | return 0; |
9746 | } | |
9747 | ||
071698e1 JA |
9748 | static bool io_register_op_must_quiesce(int op) |
9749 | { | |
9750 | switch (op) { | |
9751 | case IORING_UNREGISTER_FILES: | |
9752 | case IORING_REGISTER_FILES_UPDATE: | |
9753 | case IORING_REGISTER_PROBE: | |
9754 | case IORING_REGISTER_PERSONALITY: | |
9755 | case IORING_UNREGISTER_PERSONALITY: | |
9756 | return false; | |
9757 | default: | |
9758 | return true; | |
9759 | } | |
9760 | } | |
9761 | ||
edafccee JA |
9762 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9763 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9764 | __releases(ctx->uring_lock) |
9765 | __acquires(ctx->uring_lock) | |
edafccee JA |
9766 | { |
9767 | int ret; | |
9768 | ||
35fa71a0 JA |
9769 | /* |
9770 | * We're inside the ring mutex, if the ref is already dying, then | |
9771 | * someone else killed the ctx or is already going through | |
9772 | * io_uring_register(). | |
9773 | */ | |
9774 | if (percpu_ref_is_dying(&ctx->refs)) | |
9775 | return -ENXIO; | |
9776 | ||
071698e1 | 9777 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9778 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9779 | |
05f3fb3c JA |
9780 | /* |
9781 | * Drop uring mutex before waiting for references to exit. If | |
9782 | * another thread is currently inside io_uring_enter() it might | |
9783 | * need to grab the uring_lock to make progress. If we hold it | |
9784 | * here across the drain wait, then we can deadlock. It's safe | |
9785 | * to drop the mutex here, since no new references will come in | |
9786 | * after we've killed the percpu ref. | |
9787 | */ | |
9788 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9789 | do { |
9790 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9791 | if (!ret) | |
9792 | break; | |
ed6930c9 JA |
9793 | ret = io_run_task_work_sig(); |
9794 | if (ret < 0) | |
9795 | break; | |
af9c1a44 JA |
9796 | } while (1); |
9797 | ||
05f3fb3c | 9798 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9799 | |
c150368b JA |
9800 | if (ret) { |
9801 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9802 | goto out_quiesce; |
9803 | } | |
9804 | } | |
9805 | ||
9806 | if (ctx->restricted) { | |
9807 | if (opcode >= IORING_REGISTER_LAST) { | |
9808 | ret = -EINVAL; | |
9809 | goto out; | |
9810 | } | |
9811 | ||
9812 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9813 | ret = -EACCES; | |
c150368b JA |
9814 | goto out; |
9815 | } | |
05f3fb3c | 9816 | } |
edafccee JA |
9817 | |
9818 | switch (opcode) { | |
9819 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9820 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9821 | break; |
9822 | case IORING_UNREGISTER_BUFFERS: | |
9823 | ret = -EINVAL; | |
9824 | if (arg || nr_args) | |
9825 | break; | |
0a96bbe4 | 9826 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9827 | break; |
6b06314c JA |
9828 | case IORING_REGISTER_FILES: |
9829 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9830 | break; | |
9831 | case IORING_UNREGISTER_FILES: | |
9832 | ret = -EINVAL; | |
9833 | if (arg || nr_args) | |
9834 | break; | |
9835 | ret = io_sqe_files_unregister(ctx); | |
9836 | break; | |
c3a31e60 JA |
9837 | case IORING_REGISTER_FILES_UPDATE: |
9838 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9839 | break; | |
9b402849 | 9840 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9841 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9842 | ret = -EINVAL; |
9843 | if (nr_args != 1) | |
9844 | break; | |
9845 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9846 | if (ret) |
9847 | break; | |
9848 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9849 | ctx->eventfd_async = 1; | |
9850 | else | |
9851 | ctx->eventfd_async = 0; | |
9b402849 JA |
9852 | break; |
9853 | case IORING_UNREGISTER_EVENTFD: | |
9854 | ret = -EINVAL; | |
9855 | if (arg || nr_args) | |
9856 | break; | |
9857 | ret = io_eventfd_unregister(ctx); | |
9858 | break; | |
66f4af93 JA |
9859 | case IORING_REGISTER_PROBE: |
9860 | ret = -EINVAL; | |
9861 | if (!arg || nr_args > 256) | |
9862 | break; | |
9863 | ret = io_probe(ctx, arg, nr_args); | |
9864 | break; | |
071698e1 JA |
9865 | case IORING_REGISTER_PERSONALITY: |
9866 | ret = -EINVAL; | |
9867 | if (arg || nr_args) | |
9868 | break; | |
9869 | ret = io_register_personality(ctx); | |
9870 | break; | |
9871 | case IORING_UNREGISTER_PERSONALITY: | |
9872 | ret = -EINVAL; | |
9873 | if (arg) | |
9874 | break; | |
9875 | ret = io_unregister_personality(ctx, nr_args); | |
9876 | break; | |
7e84e1c7 SG |
9877 | case IORING_REGISTER_ENABLE_RINGS: |
9878 | ret = -EINVAL; | |
9879 | if (arg || nr_args) | |
9880 | break; | |
9881 | ret = io_register_enable_rings(ctx); | |
9882 | break; | |
21b55dbc SG |
9883 | case IORING_REGISTER_RESTRICTIONS: |
9884 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9885 | break; | |
edafccee JA |
9886 | default: |
9887 | ret = -EINVAL; | |
9888 | break; | |
9889 | } | |
9890 | ||
21b55dbc | 9891 | out: |
071698e1 | 9892 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9893 | /* bring the ctx back to life */ |
05f3fb3c | 9894 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9895 | out_quiesce: |
0f158b4c | 9896 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9897 | } |
edafccee JA |
9898 | return ret; |
9899 | } | |
9900 | ||
9901 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9902 | void __user *, arg, unsigned int, nr_args) | |
9903 | { | |
9904 | struct io_ring_ctx *ctx; | |
9905 | long ret = -EBADF; | |
9906 | struct fd f; | |
9907 | ||
9908 | f = fdget(fd); | |
9909 | if (!f.file) | |
9910 | return -EBADF; | |
9911 | ||
9912 | ret = -EOPNOTSUPP; | |
9913 | if (f.file->f_op != &io_uring_fops) | |
9914 | goto out_fput; | |
9915 | ||
9916 | ctx = f.file->private_data; | |
9917 | ||
b6c23dd5 PB |
9918 | io_run_task_work(); |
9919 | ||
edafccee JA |
9920 | mutex_lock(&ctx->uring_lock); |
9921 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9922 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9923 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9924 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9925 | out_fput: |
9926 | fdput(f); | |
9927 | return ret; | |
9928 | } | |
9929 | ||
2b188cc1 JA |
9930 | static int __init io_uring_init(void) |
9931 | { | |
d7f62e82 SM |
9932 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9933 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9934 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9935 | } while (0) | |
9936 | ||
9937 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9938 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9939 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9940 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9941 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9942 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9943 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9944 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9945 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9946 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9947 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9948 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9949 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9950 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9951 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9952 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9953 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9954 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9955 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9956 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9957 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9958 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9959 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9960 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9961 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9962 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9963 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9964 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9965 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9966 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9967 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9968 | |
d3656344 | 9969 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9970 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9971 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9972 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9973 | return 0; |
9974 | }; | |
9975 | __initcall(io_uring_init); |