]>
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> |
2b188cc1 | 81 | |
c826bd7a DD |
82 | #define CREATE_TRACE_POINTS |
83 | #include <trace/events/io_uring.h> | |
84 | ||
2b188cc1 JA |
85 | #include <uapi/linux/io_uring.h> |
86 | ||
87 | #include "internal.h" | |
561fb04a | 88 | #include "io-wq.h" |
2b188cc1 | 89 | |
5277deaa | 90 | #define IORING_MAX_ENTRIES 32768 |
33a107f0 | 91 | #define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES) |
65e19f54 JA |
92 | |
93 | /* | |
94 | * Shift of 9 is 512 entries, or exactly one page on 64-bit archs | |
95 | */ | |
96 | #define IORING_FILE_TABLE_SHIFT 9 | |
97 | #define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT) | |
98 | #define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1) | |
99 | #define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE) | |
21b55dbc SG |
100 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
101 | IORING_REGISTER_LAST + IORING_OP_LAST) | |
2b188cc1 | 102 | |
b16fed66 PB |
103 | #define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \ |
104 | IOSQE_IO_HARDLINK | IOSQE_ASYNC | \ | |
105 | IOSQE_BUFFER_SELECT) | |
106 | ||
2b188cc1 JA |
107 | struct io_uring { |
108 | u32 head ____cacheline_aligned_in_smp; | |
109 | u32 tail ____cacheline_aligned_in_smp; | |
110 | }; | |
111 | ||
1e84b97b | 112 | /* |
75b28aff HV |
113 | * This data is shared with the application through the mmap at offsets |
114 | * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. | |
1e84b97b SB |
115 | * |
116 | * The offsets to the member fields are published through struct | |
117 | * io_sqring_offsets when calling io_uring_setup. | |
118 | */ | |
75b28aff | 119 | struct io_rings { |
1e84b97b SB |
120 | /* |
121 | * Head and tail offsets into the ring; the offsets need to be | |
122 | * masked to get valid indices. | |
123 | * | |
75b28aff HV |
124 | * The kernel controls head of the sq ring and the tail of the cq ring, |
125 | * and the application controls tail of the sq ring and the head of the | |
126 | * cq ring. | |
1e84b97b | 127 | */ |
75b28aff | 128 | struct io_uring sq, cq; |
1e84b97b | 129 | /* |
75b28aff | 130 | * Bitmasks to apply to head and tail offsets (constant, equals |
1e84b97b SB |
131 | * ring_entries - 1) |
132 | */ | |
75b28aff HV |
133 | u32 sq_ring_mask, cq_ring_mask; |
134 | /* Ring sizes (constant, power of 2) */ | |
135 | u32 sq_ring_entries, cq_ring_entries; | |
1e84b97b SB |
136 | /* |
137 | * Number of invalid entries dropped by the kernel due to | |
138 | * invalid index stored in array | |
139 | * | |
140 | * Written by the kernel, shouldn't be modified by the | |
141 | * application (i.e. get number of "new events" by comparing to | |
142 | * cached value). | |
143 | * | |
144 | * After a new SQ head value was read by the application this | |
145 | * counter includes all submissions that were dropped reaching | |
146 | * the new SQ head (and possibly more). | |
147 | */ | |
75b28aff | 148 | u32 sq_dropped; |
1e84b97b | 149 | /* |
0d9b5b3a | 150 | * Runtime SQ flags |
1e84b97b SB |
151 | * |
152 | * Written by the kernel, shouldn't be modified by the | |
153 | * application. | |
154 | * | |
155 | * The application needs a full memory barrier before checking | |
156 | * for IORING_SQ_NEED_WAKEUP after updating the sq tail. | |
157 | */ | |
75b28aff | 158 | u32 sq_flags; |
0d9b5b3a SG |
159 | /* |
160 | * Runtime CQ flags | |
161 | * | |
162 | * Written by the application, shouldn't be modified by the | |
163 | * kernel. | |
164 | */ | |
165 | u32 cq_flags; | |
1e84b97b SB |
166 | /* |
167 | * Number of completion events lost because the queue was full; | |
168 | * this should be avoided by the application by making sure | |
0b4295b5 | 169 | * there are not more requests pending than there is space in |
1e84b97b SB |
170 | * the completion queue. |
171 | * | |
172 | * Written by the kernel, shouldn't be modified by the | |
173 | * application (i.e. get number of "new events" by comparing to | |
174 | * cached value). | |
175 | * | |
176 | * As completion events come in out of order this counter is not | |
177 | * ordered with any other data. | |
178 | */ | |
75b28aff | 179 | u32 cq_overflow; |
1e84b97b SB |
180 | /* |
181 | * Ring buffer of completion events. | |
182 | * | |
183 | * The kernel writes completion events fresh every time they are | |
184 | * produced, so the application is allowed to modify pending | |
185 | * entries. | |
186 | */ | |
75b28aff | 187 | struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; |
2b188cc1 JA |
188 | }; |
189 | ||
45d189c6 PB |
190 | enum io_uring_cmd_flags { |
191 | IO_URING_F_NONBLOCK = 1, | |
889fca73 | 192 | IO_URING_F_COMPLETE_DEFER = 2, |
45d189c6 PB |
193 | }; |
194 | ||
edafccee JA |
195 | struct io_mapped_ubuf { |
196 | u64 ubuf; | |
197 | size_t len; | |
198 | struct bio_vec *bvec; | |
199 | unsigned int nr_bvecs; | |
de293938 | 200 | unsigned long acct_pages; |
edafccee JA |
201 | }; |
202 | ||
50238531 BM |
203 | struct io_ring_ctx; |
204 | ||
269bbe5f BM |
205 | struct io_rsrc_put { |
206 | struct list_head list; | |
50238531 BM |
207 | union { |
208 | void *rsrc; | |
209 | struct file *file; | |
210 | }; | |
269bbe5f BM |
211 | }; |
212 | ||
213 | struct fixed_rsrc_table { | |
65e19f54 | 214 | struct file **files; |
31b51510 JA |
215 | }; |
216 | ||
269bbe5f | 217 | struct fixed_rsrc_ref_node { |
05589553 XW |
218 | struct percpu_ref refs; |
219 | struct list_head node; | |
269bbe5f BM |
220 | struct list_head rsrc_list; |
221 | struct fixed_rsrc_data *rsrc_data; | |
50238531 BM |
222 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
223 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 224 | struct llist_node llist; |
e297822b | 225 | bool done; |
05589553 XW |
226 | }; |
227 | ||
269bbe5f BM |
228 | struct fixed_rsrc_data { |
229 | struct fixed_rsrc_table *table; | |
05f3fb3c JA |
230 | struct io_ring_ctx *ctx; |
231 | ||
269bbe5f | 232 | struct fixed_rsrc_ref_node *node; |
05f3fb3c | 233 | struct percpu_ref refs; |
05f3fb3c | 234 | struct completion done; |
8bad28d8 | 235 | bool quiesce; |
05f3fb3c JA |
236 | }; |
237 | ||
5a2e745d JA |
238 | struct io_buffer { |
239 | struct list_head list; | |
240 | __u64 addr; | |
241 | __s32 len; | |
242 | __u16 bid; | |
243 | }; | |
244 | ||
21b55dbc SG |
245 | struct io_restriction { |
246 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
247 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
248 | u8 sqe_flags_allowed; | |
249 | u8 sqe_flags_required; | |
7e84e1c7 | 250 | bool registered; |
21b55dbc SG |
251 | }; |
252 | ||
37d1e2e3 JA |
253 | enum { |
254 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
255 | IO_SQ_THREAD_SHOULD_PARK, | |
256 | }; | |
257 | ||
534ca6d6 JA |
258 | struct io_sq_data { |
259 | refcount_t refs; | |
9e138a48 | 260 | atomic_t park_pending; |
09a6f4ef | 261 | struct mutex lock; |
69fb2131 JA |
262 | |
263 | /* ctx's that are using this sqd */ | |
264 | struct list_head ctx_list; | |
69fb2131 | 265 | |
534ca6d6 JA |
266 | struct task_struct *thread; |
267 | struct wait_queue_head wait; | |
08369246 XW |
268 | |
269 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
270 | int sq_cpu; |
271 | pid_t task_pid; | |
5c2469e0 | 272 | pid_t task_tgid; |
37d1e2e3 JA |
273 | |
274 | unsigned long state; | |
37d1e2e3 | 275 | struct completion exited; |
b7f5a0bf | 276 | struct callback_head *park_task_work; |
534ca6d6 JA |
277 | }; |
278 | ||
258b29a9 | 279 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 280 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 281 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 282 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
283 | |
284 | struct io_comp_state { | |
6dd0be1e | 285 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 286 | unsigned int nr; |
c7dae4ba JA |
287 | unsigned int locked_free_nr; |
288 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 289 | struct list_head free_list; |
c7dae4ba JA |
290 | /* IRQ completion list, under ->completion_lock */ |
291 | struct list_head locked_free_list; | |
258b29a9 PB |
292 | }; |
293 | ||
a1ab7b35 PB |
294 | struct io_submit_link { |
295 | struct io_kiocb *head; | |
296 | struct io_kiocb *last; | |
297 | }; | |
298 | ||
258b29a9 PB |
299 | struct io_submit_state { |
300 | struct blk_plug plug; | |
a1ab7b35 | 301 | struct io_submit_link link; |
258b29a9 PB |
302 | |
303 | /* | |
304 | * io_kiocb alloc cache | |
305 | */ | |
bf019da7 | 306 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
307 | unsigned int free_reqs; |
308 | ||
309 | bool plug_started; | |
310 | ||
311 | /* | |
312 | * Batch completion logic | |
313 | */ | |
314 | struct io_comp_state comp; | |
315 | ||
316 | /* | |
317 | * File reference cache | |
318 | */ | |
319 | struct file *file; | |
320 | unsigned int fd; | |
321 | unsigned int file_refs; | |
322 | unsigned int ios_left; | |
323 | }; | |
324 | ||
2b188cc1 JA |
325 | struct io_ring_ctx { |
326 | struct { | |
327 | struct percpu_ref refs; | |
328 | } ____cacheline_aligned_in_smp; | |
329 | ||
330 | struct { | |
331 | unsigned int flags; | |
e1d85334 | 332 | unsigned int compat: 1; |
e1d85334 RD |
333 | unsigned int cq_overflow_flushed: 1; |
334 | unsigned int drain_next: 1; | |
335 | unsigned int eventfd_async: 1; | |
21b55dbc | 336 | unsigned int restricted: 1; |
2b188cc1 | 337 | |
75b28aff HV |
338 | /* |
339 | * Ring buffer of indices into array of io_uring_sqe, which is | |
340 | * mmapped by the application using the IORING_OFF_SQES offset. | |
341 | * | |
342 | * This indirection could e.g. be used to assign fixed | |
343 | * io_uring_sqe entries to operations and only submit them to | |
344 | * the queue when needed. | |
345 | * | |
346 | * The kernel modifies neither the indices array nor the entries | |
347 | * array. | |
348 | */ | |
349 | u32 *sq_array; | |
2b188cc1 JA |
350 | unsigned cached_sq_head; |
351 | unsigned sq_entries; | |
352 | unsigned sq_mask; | |
6c271ce2 | 353 | unsigned sq_thread_idle; |
498ccd9e | 354 | unsigned cached_sq_dropped; |
2c3bac6d | 355 | unsigned cached_cq_overflow; |
ad3eb2c8 | 356 | unsigned long sq_check_overflow; |
de0617e4 | 357 | |
e941894e JA |
358 | /* hashed buffered write serialization */ |
359 | struct io_wq_hash *hash_map; | |
360 | ||
de0617e4 | 361 | struct list_head defer_list; |
5262f567 | 362 | struct list_head timeout_list; |
1d7bb1d5 | 363 | struct list_head cq_overflow_list; |
fcb323cc | 364 | |
ad3eb2c8 | 365 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
366 | } ____cacheline_aligned_in_smp; |
367 | ||
3c1a2ead JA |
368 | struct { |
369 | struct mutex uring_lock; | |
370 | wait_queue_head_t wait; | |
371 | } ____cacheline_aligned_in_smp; | |
372 | ||
373 | struct io_submit_state submit_state; | |
374 | ||
206aefde JA |
375 | struct io_rings *rings; |
376 | ||
2aede0e4 JA |
377 | /* Only used for accounting purposes */ |
378 | struct mm_struct *mm_account; | |
379 | ||
7c30f36a | 380 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
381 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
382 | ||
90554200 | 383 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 384 | struct list_head sqd_list; |
75b28aff | 385 | |
6b06314c JA |
386 | /* |
387 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
388 | * readers must ensure that ->refs is alive as long as the file* is | |
389 | * used. Only updated through io_uring_register(2). | |
390 | */ | |
269bbe5f | 391 | struct fixed_rsrc_data *file_data; |
6b06314c JA |
392 | unsigned nr_user_files; |
393 | ||
edafccee JA |
394 | /* if used, fixed mapped user buffers */ |
395 | unsigned nr_user_bufs; | |
396 | struct io_mapped_ubuf *user_bufs; | |
397 | ||
2b188cc1 JA |
398 | struct user_struct *user; |
399 | ||
0f158b4c | 400 | struct completion ref_comp; |
206aefde JA |
401 | |
402 | #if defined(CONFIG_UNIX) | |
403 | struct socket *ring_sock; | |
404 | #endif | |
405 | ||
9e15c3a0 | 406 | struct xarray io_buffers; |
5a2e745d | 407 | |
61cf9370 MWO |
408 | struct xarray personalities; |
409 | u32 pers_next; | |
071698e1 | 410 | |
206aefde JA |
411 | struct { |
412 | unsigned cached_cq_tail; | |
413 | unsigned cq_entries; | |
414 | unsigned cq_mask; | |
415 | atomic_t cq_timeouts; | |
f010505b | 416 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 417 | unsigned long cq_check_overflow; |
206aefde JA |
418 | struct wait_queue_head cq_wait; |
419 | struct fasync_struct *cq_fasync; | |
420 | struct eventfd_ctx *cq_ev_fd; | |
421 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 422 | |
2b188cc1 JA |
423 | struct { |
424 | spinlock_t completion_lock; | |
e94f141b | 425 | |
def596e9 | 426 | /* |
540e32a0 | 427 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
428 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
429 | * For SQPOLL, only the single threaded io_sq_thread() will | |
430 | * manipulate the list, hence no extra locking is needed there. | |
431 | */ | |
540e32a0 | 432 | struct list_head iopoll_list; |
78076bb6 JA |
433 | struct hlist_head *cancel_hash; |
434 | unsigned cancel_hash_bits; | |
e94f141b | 435 | bool poll_multi_file; |
31b51510 | 436 | |
fcb323cc JA |
437 | spinlock_t inflight_lock; |
438 | struct list_head inflight_list; | |
2b188cc1 | 439 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 440 | |
269bbe5f BM |
441 | struct delayed_work rsrc_put_work; |
442 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
443 | struct list_head rsrc_ref_list; |
444 | spinlock_t rsrc_ref_lock; | |
4a38aed2 | 445 | |
21b55dbc | 446 | struct io_restriction restrictions; |
3c1a2ead | 447 | |
7c25c0d1 JA |
448 | /* exit task_work */ |
449 | struct callback_head *exit_task_work; | |
450 | ||
e941894e JA |
451 | struct wait_queue_head hash_wait; |
452 | ||
3c1a2ead JA |
453 | /* Keep this last, we don't need it for the fast path */ |
454 | struct work_struct exit_work; | |
13bf43f5 | 455 | struct list_head tctx_list; |
2b188cc1 JA |
456 | }; |
457 | ||
53e043b2 SM |
458 | struct io_uring_task { |
459 | /* submission side */ | |
460 | struct xarray xa; | |
461 | struct wait_queue_head wait; | |
ee53fb2b SM |
462 | const struct io_ring_ctx *last; |
463 | struct io_wq *io_wq; | |
53e043b2 SM |
464 | struct percpu_counter inflight; |
465 | atomic_t in_idle; | |
466 | bool sqpoll; | |
467 | ||
468 | spinlock_t task_lock; | |
469 | struct io_wq_work_list task_list; | |
470 | unsigned long task_state; | |
471 | struct callback_head task_work; | |
472 | }; | |
473 | ||
09bb8394 JA |
474 | /* |
475 | * First field must be the file pointer in all the | |
476 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
477 | */ | |
221c5eb2 JA |
478 | struct io_poll_iocb { |
479 | struct file *file; | |
018043be | 480 | struct wait_queue_head *head; |
221c5eb2 | 481 | __poll_t events; |
8c838788 | 482 | bool done; |
221c5eb2 | 483 | bool canceled; |
392edb45 | 484 | struct wait_queue_entry wait; |
221c5eb2 JA |
485 | }; |
486 | ||
018043be PB |
487 | struct io_poll_remove { |
488 | struct file *file; | |
489 | u64 addr; | |
490 | }; | |
491 | ||
b5dba59e JA |
492 | struct io_close { |
493 | struct file *file; | |
b5dba59e JA |
494 | int fd; |
495 | }; | |
496 | ||
ad8a48ac JA |
497 | struct io_timeout_data { |
498 | struct io_kiocb *req; | |
499 | struct hrtimer timer; | |
500 | struct timespec64 ts; | |
501 | enum hrtimer_mode mode; | |
502 | }; | |
503 | ||
8ed8d3c3 JA |
504 | struct io_accept { |
505 | struct file *file; | |
506 | struct sockaddr __user *addr; | |
507 | int __user *addr_len; | |
508 | int flags; | |
09952e3e | 509 | unsigned long nofile; |
8ed8d3c3 JA |
510 | }; |
511 | ||
512 | struct io_sync { | |
513 | struct file *file; | |
514 | loff_t len; | |
515 | loff_t off; | |
516 | int flags; | |
d63d1b5e | 517 | int mode; |
8ed8d3c3 JA |
518 | }; |
519 | ||
fbf23849 JA |
520 | struct io_cancel { |
521 | struct file *file; | |
522 | u64 addr; | |
523 | }; | |
524 | ||
b29472ee JA |
525 | struct io_timeout { |
526 | struct file *file; | |
bfe68a22 PB |
527 | u32 off; |
528 | u32 target_seq; | |
135fcde8 | 529 | struct list_head list; |
90cd7e42 PB |
530 | /* head of the link, used by linked timeouts only */ |
531 | struct io_kiocb *head; | |
b29472ee JA |
532 | }; |
533 | ||
0bdf7a2d PB |
534 | struct io_timeout_rem { |
535 | struct file *file; | |
536 | u64 addr; | |
9c8e11b3 PB |
537 | |
538 | /* timeout update */ | |
539 | struct timespec64 ts; | |
540 | u32 flags; | |
0bdf7a2d PB |
541 | }; |
542 | ||
9adbd45d JA |
543 | struct io_rw { |
544 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
545 | struct kiocb kiocb; | |
546 | u64 addr; | |
547 | u64 len; | |
548 | }; | |
549 | ||
3fbb51c1 JA |
550 | struct io_connect { |
551 | struct file *file; | |
552 | struct sockaddr __user *addr; | |
553 | int addr_len; | |
554 | }; | |
555 | ||
e47293fd JA |
556 | struct io_sr_msg { |
557 | struct file *file; | |
fddaface | 558 | union { |
270a5940 | 559 | struct user_msghdr __user *umsg; |
fddaface JA |
560 | void __user *buf; |
561 | }; | |
e47293fd | 562 | int msg_flags; |
bcda7baa | 563 | int bgid; |
fddaface | 564 | size_t len; |
bcda7baa | 565 | struct io_buffer *kbuf; |
e47293fd JA |
566 | }; |
567 | ||
15b71abe JA |
568 | struct io_open { |
569 | struct file *file; | |
570 | int dfd; | |
15b71abe | 571 | struct filename *filename; |
c12cedf2 | 572 | struct open_how how; |
4022e7af | 573 | unsigned long nofile; |
15b71abe JA |
574 | }; |
575 | ||
269bbe5f | 576 | struct io_rsrc_update { |
05f3fb3c JA |
577 | struct file *file; |
578 | u64 arg; | |
579 | u32 nr_args; | |
580 | u32 offset; | |
581 | }; | |
582 | ||
4840e418 JA |
583 | struct io_fadvise { |
584 | struct file *file; | |
585 | u64 offset; | |
586 | u32 len; | |
587 | u32 advice; | |
588 | }; | |
589 | ||
c1ca757b JA |
590 | struct io_madvise { |
591 | struct file *file; | |
592 | u64 addr; | |
593 | u32 len; | |
594 | u32 advice; | |
595 | }; | |
596 | ||
3e4827b0 JA |
597 | struct io_epoll { |
598 | struct file *file; | |
599 | int epfd; | |
600 | int op; | |
601 | int fd; | |
602 | struct epoll_event event; | |
e47293fd JA |
603 | }; |
604 | ||
7d67af2c PB |
605 | struct io_splice { |
606 | struct file *file_out; | |
607 | struct file *file_in; | |
608 | loff_t off_out; | |
609 | loff_t off_in; | |
610 | u64 len; | |
611 | unsigned int flags; | |
612 | }; | |
613 | ||
ddf0322d JA |
614 | struct io_provide_buf { |
615 | struct file *file; | |
616 | __u64 addr; | |
617 | __s32 len; | |
618 | __u32 bgid; | |
619 | __u16 nbufs; | |
620 | __u16 bid; | |
621 | }; | |
622 | ||
1d9e1288 BM |
623 | struct io_statx { |
624 | struct file *file; | |
625 | int dfd; | |
626 | unsigned int mask; | |
627 | unsigned int flags; | |
e62753e4 | 628 | const char __user *filename; |
1d9e1288 BM |
629 | struct statx __user *buffer; |
630 | }; | |
631 | ||
36f4fa68 JA |
632 | struct io_shutdown { |
633 | struct file *file; | |
634 | int how; | |
635 | }; | |
636 | ||
80a261fd JA |
637 | struct io_rename { |
638 | struct file *file; | |
639 | int old_dfd; | |
640 | int new_dfd; | |
641 | struct filename *oldpath; | |
642 | struct filename *newpath; | |
643 | int flags; | |
644 | }; | |
645 | ||
14a1143b JA |
646 | struct io_unlink { |
647 | struct file *file; | |
648 | int dfd; | |
649 | int flags; | |
650 | struct filename *filename; | |
651 | }; | |
652 | ||
3ca405eb PB |
653 | struct io_completion { |
654 | struct file *file; | |
655 | struct list_head list; | |
0f7e466b | 656 | int cflags; |
3ca405eb PB |
657 | }; |
658 | ||
f499a021 JA |
659 | struct io_async_connect { |
660 | struct sockaddr_storage address; | |
661 | }; | |
662 | ||
03b1230c JA |
663 | struct io_async_msghdr { |
664 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
665 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
666 | struct iovec *free_iov; | |
03b1230c JA |
667 | struct sockaddr __user *uaddr; |
668 | struct msghdr msg; | |
b537916c | 669 | struct sockaddr_storage addr; |
03b1230c JA |
670 | }; |
671 | ||
f67676d1 JA |
672 | struct io_async_rw { |
673 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
674 | const struct iovec *free_iovec; |
675 | struct iov_iter iter; | |
227c0c96 | 676 | size_t bytes_done; |
bcf5a063 | 677 | struct wait_page_queue wpq; |
f67676d1 JA |
678 | }; |
679 | ||
6b47ee6e PB |
680 | enum { |
681 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
682 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
683 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
684 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
685 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 686 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 687 | |
6b47ee6e PB |
688 | REQ_F_FAIL_LINK_BIT, |
689 | REQ_F_INFLIGHT_BIT, | |
690 | REQ_F_CUR_POS_BIT, | |
691 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 692 | REQ_F_LINK_TIMEOUT_BIT, |
6b47ee6e | 693 | REQ_F_ISREG_BIT, |
99bc4c38 | 694 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 695 | REQ_F_POLLED_BIT, |
bcda7baa | 696 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 697 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 698 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 699 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 700 | REQ_F_REISSUE_BIT, |
84557871 JA |
701 | |
702 | /* not a real bit, just to check we're not overflowing the space */ | |
703 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
704 | }; |
705 | ||
706 | enum { | |
707 | /* ctx owns file */ | |
708 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
709 | /* drain existing IO first */ | |
710 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
711 | /* linked sqes */ | |
712 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
713 | /* doesn't sever on completion < 0 */ | |
714 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
715 | /* IOSQE_ASYNC */ | |
716 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
717 | /* IOSQE_BUFFER_SELECT */ |
718 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 719 | |
6b47ee6e PB |
720 | /* fail rest of links */ |
721 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 722 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
723 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
724 | /* read/write uses file position */ | |
725 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
726 | /* must not punt to workers */ | |
727 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 728 | /* has or had linked timeout */ |
6b47ee6e | 729 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
6b47ee6e PB |
730 | /* regular file */ |
731 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
99bc4c38 PB |
732 | /* needs cleanup */ |
733 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
734 | /* already went through poll handler */ |
735 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
736 | /* buffer already selected */ |
737 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
738 | /* doesn't need file table for this request */ |
739 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
740 | /* linked timeout is active, i.e. prepared by link's head */ |
741 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
742 | /* completion is deferred through io_comp_state */ |
743 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
744 | /* caller should reissue async */ |
745 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
d7718a9d JA |
746 | }; |
747 | ||
748 | struct async_poll { | |
749 | struct io_poll_iocb poll; | |
807abcb0 | 750 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
751 | }; |
752 | ||
7cbf1722 JA |
753 | struct io_task_work { |
754 | struct io_wq_work_node node; | |
755 | task_work_func_t func; | |
756 | }; | |
757 | ||
09bb8394 JA |
758 | /* |
759 | * NOTE! Each of the iocb union members has the file pointer | |
760 | * as the first entry in their struct definition. So you can | |
761 | * access the file pointer through any of the sub-structs, | |
762 | * or directly as just 'ki_filp' in this struct. | |
763 | */ | |
2b188cc1 | 764 | struct io_kiocb { |
221c5eb2 | 765 | union { |
09bb8394 | 766 | struct file *file; |
9adbd45d | 767 | struct io_rw rw; |
221c5eb2 | 768 | struct io_poll_iocb poll; |
018043be | 769 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
770 | struct io_accept accept; |
771 | struct io_sync sync; | |
fbf23849 | 772 | struct io_cancel cancel; |
b29472ee | 773 | struct io_timeout timeout; |
0bdf7a2d | 774 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 775 | struct io_connect connect; |
e47293fd | 776 | struct io_sr_msg sr_msg; |
15b71abe | 777 | struct io_open open; |
b5dba59e | 778 | struct io_close close; |
269bbe5f | 779 | struct io_rsrc_update rsrc_update; |
4840e418 | 780 | struct io_fadvise fadvise; |
c1ca757b | 781 | struct io_madvise madvise; |
3e4827b0 | 782 | struct io_epoll epoll; |
7d67af2c | 783 | struct io_splice splice; |
ddf0322d | 784 | struct io_provide_buf pbuf; |
1d9e1288 | 785 | struct io_statx statx; |
36f4fa68 | 786 | struct io_shutdown shutdown; |
80a261fd | 787 | struct io_rename rename; |
14a1143b | 788 | struct io_unlink unlink; |
3ca405eb PB |
789 | /* use only after cleaning per-op data, see io_clean_op() */ |
790 | struct io_completion compl; | |
221c5eb2 | 791 | }; |
2b188cc1 | 792 | |
e8c2bc1f JA |
793 | /* opcode allocated if it needs to store data for async defer */ |
794 | void *async_data; | |
d625c6ee | 795 | u8 opcode; |
65a6543d XW |
796 | /* polled IO has completed */ |
797 | u8 iopoll_completed; | |
2b188cc1 | 798 | |
4f4eeba8 | 799 | u16 buf_index; |
9cf7c104 | 800 | u32 result; |
4f4eeba8 | 801 | |
010e8e6b PB |
802 | struct io_ring_ctx *ctx; |
803 | unsigned int flags; | |
804 | refcount_t refs; | |
805 | struct task_struct *task; | |
806 | u64 user_data; | |
d7718a9d | 807 | |
f2f87370 | 808 | struct io_kiocb *link; |
269bbe5f | 809 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 810 | |
d21ffe7e PB |
811 | /* |
812 | * 1. used with ctx->iopoll_list with reads/writes | |
813 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
814 | */ | |
010e8e6b | 815 | struct list_head inflight_entry; |
7cbf1722 JA |
816 | union { |
817 | struct io_task_work io_task_work; | |
818 | struct callback_head task_work; | |
819 | }; | |
010e8e6b PB |
820 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
821 | struct hlist_node hash_node; | |
822 | struct async_poll *apoll; | |
823 | struct io_wq_work work; | |
2b188cc1 | 824 | }; |
05589553 | 825 | |
13bf43f5 PB |
826 | struct io_tctx_node { |
827 | struct list_head ctx_node; | |
828 | struct task_struct *task; | |
13bf43f5 PB |
829 | struct io_ring_ctx *ctx; |
830 | }; | |
831 | ||
27dc8338 PB |
832 | struct io_defer_entry { |
833 | struct list_head list; | |
834 | struct io_kiocb *req; | |
9cf7c104 | 835 | u32 seq; |
2b188cc1 JA |
836 | }; |
837 | ||
d3656344 | 838 | struct io_op_def { |
d3656344 JA |
839 | /* needs req->file assigned */ |
840 | unsigned needs_file : 1; | |
d3656344 JA |
841 | /* hash wq insertion if file is a regular file */ |
842 | unsigned hash_reg_file : 1; | |
843 | /* unbound wq insertion if file is a non-regular file */ | |
844 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
845 | /* opcode is not supported by this kernel */ |
846 | unsigned not_supported : 1; | |
8a72758c JA |
847 | /* set if opcode supports polled "wait" */ |
848 | unsigned pollin : 1; | |
849 | unsigned pollout : 1; | |
bcda7baa JA |
850 | /* op supports buffer selection */ |
851 | unsigned buffer_select : 1; | |
e8c2bc1f JA |
852 | /* must always have async data allocated */ |
853 | unsigned needs_async_data : 1; | |
27926b68 JA |
854 | /* should block plug */ |
855 | unsigned plug : 1; | |
e8c2bc1f JA |
856 | /* size of async data needed, if any */ |
857 | unsigned short async_size; | |
d3656344 JA |
858 | }; |
859 | ||
0918682b | 860 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
861 | [IORING_OP_NOP] = {}, |
862 | [IORING_OP_READV] = { | |
d3656344 JA |
863 | .needs_file = 1, |
864 | .unbound_nonreg_file = 1, | |
8a72758c | 865 | .pollin = 1, |
4d954c25 | 866 | .buffer_select = 1, |
e8c2bc1f | 867 | .needs_async_data = 1, |
27926b68 | 868 | .plug = 1, |
e8c2bc1f | 869 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 870 | }, |
0463b6c5 | 871 | [IORING_OP_WRITEV] = { |
d3656344 JA |
872 | .needs_file = 1, |
873 | .hash_reg_file = 1, | |
874 | .unbound_nonreg_file = 1, | |
8a72758c | 875 | .pollout = 1, |
e8c2bc1f | 876 | .needs_async_data = 1, |
27926b68 | 877 | .plug = 1, |
e8c2bc1f | 878 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 879 | }, |
0463b6c5 | 880 | [IORING_OP_FSYNC] = { |
d3656344 JA |
881 | .needs_file = 1, |
882 | }, | |
0463b6c5 | 883 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
884 | .needs_file = 1, |
885 | .unbound_nonreg_file = 1, | |
8a72758c | 886 | .pollin = 1, |
27926b68 | 887 | .plug = 1, |
e8c2bc1f | 888 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 889 | }, |
0463b6c5 | 890 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
891 | .needs_file = 1, |
892 | .hash_reg_file = 1, | |
893 | .unbound_nonreg_file = 1, | |
8a72758c | 894 | .pollout = 1, |
27926b68 | 895 | .plug = 1, |
e8c2bc1f | 896 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 897 | }, |
0463b6c5 | 898 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
899 | .needs_file = 1, |
900 | .unbound_nonreg_file = 1, | |
901 | }, | |
0463b6c5 PB |
902 | [IORING_OP_POLL_REMOVE] = {}, |
903 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
904 | .needs_file = 1, |
905 | }, | |
0463b6c5 | 906 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
907 | .needs_file = 1, |
908 | .unbound_nonreg_file = 1, | |
8a72758c | 909 | .pollout = 1, |
e8c2bc1f JA |
910 | .needs_async_data = 1, |
911 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 912 | }, |
0463b6c5 | 913 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
914 | .needs_file = 1, |
915 | .unbound_nonreg_file = 1, | |
8a72758c | 916 | .pollin = 1, |
52de1fe1 | 917 | .buffer_select = 1, |
e8c2bc1f JA |
918 | .needs_async_data = 1, |
919 | .async_size = sizeof(struct io_async_msghdr), | |
d3656344 | 920 | }, |
0463b6c5 | 921 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f JA |
922 | .needs_async_data = 1, |
923 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 924 | }, |
9c8e11b3 PB |
925 | [IORING_OP_TIMEOUT_REMOVE] = { |
926 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 927 | }, |
0463b6c5 | 928 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
929 | .needs_file = 1, |
930 | .unbound_nonreg_file = 1, | |
8a72758c | 931 | .pollin = 1, |
d3656344 | 932 | }, |
0463b6c5 PB |
933 | [IORING_OP_ASYNC_CANCEL] = {}, |
934 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f JA |
935 | .needs_async_data = 1, |
936 | .async_size = sizeof(struct io_timeout_data), | |
d3656344 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_CONNECT] = { |
d3656344 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
8a72758c | 941 | .pollout = 1, |
e8c2bc1f JA |
942 | .needs_async_data = 1, |
943 | .async_size = sizeof(struct io_async_connect), | |
d3656344 | 944 | }, |
0463b6c5 | 945 | [IORING_OP_FALLOCATE] = { |
d3656344 | 946 | .needs_file = 1, |
d3656344 | 947 | }, |
44526bed JA |
948 | [IORING_OP_OPENAT] = {}, |
949 | [IORING_OP_CLOSE] = {}, | |
950 | [IORING_OP_FILES_UPDATE] = {}, | |
951 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 952 | [IORING_OP_READ] = { |
3a6820f2 JA |
953 | .needs_file = 1, |
954 | .unbound_nonreg_file = 1, | |
8a72758c | 955 | .pollin = 1, |
bcda7baa | 956 | .buffer_select = 1, |
27926b68 | 957 | .plug = 1, |
e8c2bc1f | 958 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 959 | }, |
0463b6c5 | 960 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
961 | .needs_file = 1, |
962 | .unbound_nonreg_file = 1, | |
8a72758c | 963 | .pollout = 1, |
27926b68 | 964 | .plug = 1, |
e8c2bc1f | 965 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 966 | }, |
0463b6c5 | 967 | [IORING_OP_FADVISE] = { |
4840e418 | 968 | .needs_file = 1, |
c1ca757b | 969 | }, |
44526bed | 970 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 971 | [IORING_OP_SEND] = { |
fddaface JA |
972 | .needs_file = 1, |
973 | .unbound_nonreg_file = 1, | |
8a72758c | 974 | .pollout = 1, |
fddaface | 975 | }, |
0463b6c5 | 976 | [IORING_OP_RECV] = { |
fddaface JA |
977 | .needs_file = 1, |
978 | .unbound_nonreg_file = 1, | |
8a72758c | 979 | .pollin = 1, |
bcda7baa | 980 | .buffer_select = 1, |
fddaface | 981 | }, |
0463b6c5 | 982 | [IORING_OP_OPENAT2] = { |
cebdb986 | 983 | }, |
3e4827b0 JA |
984 | [IORING_OP_EPOLL_CTL] = { |
985 | .unbound_nonreg_file = 1, | |
3e4827b0 | 986 | }, |
7d67af2c PB |
987 | [IORING_OP_SPLICE] = { |
988 | .needs_file = 1, | |
989 | .hash_reg_file = 1, | |
990 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
991 | }, |
992 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 993 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
994 | [IORING_OP_TEE] = { |
995 | .needs_file = 1, | |
996 | .hash_reg_file = 1, | |
997 | .unbound_nonreg_file = 1, | |
998 | }, | |
36f4fa68 JA |
999 | [IORING_OP_SHUTDOWN] = { |
1000 | .needs_file = 1, | |
1001 | }, | |
44526bed JA |
1002 | [IORING_OP_RENAMEAT] = {}, |
1003 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1004 | }; |
1005 | ||
7a612350 | 1006 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1007 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1008 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1009 | struct task_struct *task, | |
1010 | struct files_struct *files); | |
37d1e2e3 | 1011 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 1012 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1013 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1014 | struct io_ring_ctx *ctx); |
f2303b1f | 1015 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 1016 | |
23faba36 | 1017 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1018 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1019 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1020 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1021 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1022 | static void io_dismantle_req(struct io_kiocb *req); |
1023 | static void io_put_task(struct task_struct *task, int nr); | |
1024 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 | 1025 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
7271ef3a | 1026 | static void __io_queue_linked_timeout(struct io_kiocb *req); |
94ae5e77 | 1027 | static void io_queue_linked_timeout(struct io_kiocb *req); |
05f3fb3c | 1028 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1029 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1030 | unsigned nr_args); |
3ca405eb | 1031 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1032 | static struct file *io_file_get(struct io_submit_state *state, |
1033 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1034 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1035 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1036 | |
847595de PB |
1037 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
1038 | struct iov_iter *iter, bool needs_lock); | |
ff6165b2 JA |
1039 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
1040 | const struct iovec *fast_iov, | |
227c0c96 | 1041 | struct iov_iter *iter, bool force); |
907d1df3 | 1042 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1043 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1044 | struct io_ring_ctx *ctx); | |
de0617e4 | 1045 | |
2b188cc1 JA |
1046 | static struct kmem_cache *req_cachep; |
1047 | ||
0918682b | 1048 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1049 | |
1050 | struct sock *io_uring_get_socket(struct file *file) | |
1051 | { | |
1052 | #if defined(CONFIG_UNIX) | |
1053 | if (file->f_op == &io_uring_fops) { | |
1054 | struct io_ring_ctx *ctx = file->private_data; | |
1055 | ||
1056 | return ctx->ring_sock->sk; | |
1057 | } | |
1058 | #endif | |
1059 | return NULL; | |
1060 | } | |
1061 | EXPORT_SYMBOL(io_uring_get_socket); | |
1062 | ||
f2f87370 PB |
1063 | #define io_for_each_link(pos, head) \ |
1064 | for (pos = (head); pos; pos = pos->link) | |
1065 | ||
3ca405eb PB |
1066 | static inline void io_clean_op(struct io_kiocb *req) |
1067 | { | |
9d5c8190 | 1068 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
3ca405eb PB |
1069 | __io_clean_op(req); |
1070 | } | |
1071 | ||
36f72fe2 PB |
1072 | static inline void io_set_resource_node(struct io_kiocb *req) |
1073 | { | |
1074 | struct io_ring_ctx *ctx = req->ctx; | |
1075 | ||
269bbe5f BM |
1076 | if (!req->fixed_rsrc_refs) { |
1077 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1078 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1079 | } |
1080 | } | |
1081 | ||
08d23634 PB |
1082 | static bool io_match_task(struct io_kiocb *head, |
1083 | struct task_struct *task, | |
1084 | struct files_struct *files) | |
1085 | { | |
1086 | struct io_kiocb *req; | |
1087 | ||
84965ff8 JA |
1088 | if (task && head->task != task) { |
1089 | /* in terms of cancelation, always match if req task is dead */ | |
1090 | if (head->task->flags & PF_EXITING) | |
1091 | return true; | |
08d23634 | 1092 | return false; |
84965ff8 | 1093 | } |
08d23634 PB |
1094 | if (!files) |
1095 | return true; | |
1096 | ||
1097 | io_for_each_link(req, head) { | |
b05a1bcd | 1098 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1099 | return true; |
08d23634 PB |
1100 | } |
1101 | return false; | |
1102 | } | |
1103 | ||
c40f6379 JA |
1104 | static inline void req_set_fail_links(struct io_kiocb *req) |
1105 | { | |
1106 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1107 | req->flags |= REQ_F_FAIL_LINK; | |
1108 | } | |
4a38aed2 | 1109 | |
2b188cc1 JA |
1110 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1111 | { | |
1112 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1113 | ||
0f158b4c | 1114 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1115 | } |
1116 | ||
8eb7e2d0 PB |
1117 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1118 | { | |
1119 | return !req->timeout.off; | |
1120 | } | |
1121 | ||
2b188cc1 JA |
1122 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1123 | { | |
1124 | struct io_ring_ctx *ctx; | |
78076bb6 | 1125 | int hash_bits; |
2b188cc1 JA |
1126 | |
1127 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1128 | if (!ctx) | |
1129 | return NULL; | |
1130 | ||
78076bb6 JA |
1131 | /* |
1132 | * Use 5 bits less than the max cq entries, that should give us around | |
1133 | * 32 entries per hash list if totally full and uniformly spread. | |
1134 | */ | |
1135 | hash_bits = ilog2(p->cq_entries); | |
1136 | hash_bits -= 5; | |
1137 | if (hash_bits <= 0) | |
1138 | hash_bits = 1; | |
1139 | ctx->cancel_hash_bits = hash_bits; | |
1140 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1141 | GFP_KERNEL); | |
1142 | if (!ctx->cancel_hash) | |
1143 | goto err; | |
1144 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1145 | ||
21482896 | 1146 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1147 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1148 | goto err; | |
2b188cc1 JA |
1149 | |
1150 | ctx->flags = p->flags; | |
90554200 | 1151 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1152 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1153 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1154 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1155 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1156 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1157 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1158 | mutex_init(&ctx->uring_lock); |
1159 | init_waitqueue_head(&ctx->wait); | |
1160 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1161 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1162 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1163 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1164 | spin_lock_init(&ctx->inflight_lock); |
1165 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1166 | spin_lock_init(&ctx->rsrc_ref_lock); |
1167 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1168 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1169 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1170 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1171 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1172 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1173 | return ctx; |
206aefde | 1174 | err: |
78076bb6 | 1175 | kfree(ctx->cancel_hash); |
206aefde JA |
1176 | kfree(ctx); |
1177 | return NULL; | |
2b188cc1 JA |
1178 | } |
1179 | ||
9cf7c104 | 1180 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1181 | { |
2bc9930e JA |
1182 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1183 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1184 | |
9cf7c104 | 1185 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1186 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1187 | } |
de0617e4 | 1188 | |
9d858b21 | 1189 | return false; |
de0617e4 JA |
1190 | } |
1191 | ||
ce3d5aae PB |
1192 | static void io_req_track_inflight(struct io_kiocb *req) |
1193 | { | |
1194 | struct io_ring_ctx *ctx = req->ctx; | |
1195 | ||
1196 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1197 | req->flags |= REQ_F_INFLIGHT; |
1198 | ||
1199 | spin_lock_irq(&ctx->inflight_lock); | |
1200 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1201 | spin_unlock_irq(&ctx->inflight_lock); | |
1202 | } | |
1203 | } | |
1204 | ||
1e6fa521 JA |
1205 | static void io_prep_async_work(struct io_kiocb *req) |
1206 | { | |
1207 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1208 | struct io_ring_ctx *ctx = req->ctx; |
1209 | ||
003e8dcc JA |
1210 | if (!req->work.creds) |
1211 | req->work.creds = get_current_cred(); | |
1212 | ||
feaadc4f PB |
1213 | if (req->flags & REQ_F_FORCE_ASYNC) |
1214 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1215 | ||
1e6fa521 JA |
1216 | if (req->flags & REQ_F_ISREG) { |
1217 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1218 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1219 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1220 | if (def->unbound_nonreg_file) |
1221 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1222 | } | |
561fb04a | 1223 | } |
cccf0ee8 | 1224 | |
cbdcb435 | 1225 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1226 | { |
cbdcb435 | 1227 | struct io_kiocb *cur; |
54a91f3b | 1228 | |
f2f87370 PB |
1229 | io_for_each_link(cur, req) |
1230 | io_prep_async_work(cur); | |
561fb04a JA |
1231 | } |
1232 | ||
ebf93667 | 1233 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1234 | { |
a197f664 | 1235 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1236 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1237 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1238 | |
3bfe6106 JA |
1239 | BUG_ON(!tctx); |
1240 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1241 | |
cbdcb435 PB |
1242 | /* init ->work of the whole link before punting */ |
1243 | io_prep_async_link(req); | |
d07f1e8a PB |
1244 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1245 | &req->work, req->flags); | |
ebf93667 | 1246 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1247 | if (link) |
1248 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1249 | } |
1250 | ||
1ee4160c | 1251 | static void io_kill_timeout(struct io_kiocb *req, int status) |
5262f567 | 1252 | { |
e8c2bc1f | 1253 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1254 | int ret; |
1255 | ||
e8c2bc1f | 1256 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1257 | if (ret != -1) { |
01cec8c1 PB |
1258 | atomic_set(&req->ctx->cq_timeouts, |
1259 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1260 | list_del_init(&req->timeout.list); |
1ee4160c | 1261 | io_cqring_fill_event(req, status); |
216578e5 | 1262 | io_put_req_deferred(req, 1); |
5262f567 JA |
1263 | } |
1264 | } | |
1265 | ||
04518945 | 1266 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1267 | { |
04518945 | 1268 | do { |
27dc8338 PB |
1269 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1270 | struct io_defer_entry, list); | |
de0617e4 | 1271 | |
9cf7c104 | 1272 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1273 | break; |
27dc8338 | 1274 | list_del_init(&de->list); |
907d1df3 | 1275 | io_req_task_queue(de->req); |
27dc8338 | 1276 | kfree(de); |
04518945 PB |
1277 | } while (!list_empty(&ctx->defer_list)); |
1278 | } | |
1279 | ||
360428f8 | 1280 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1281 | { |
f010505b MDG |
1282 | u32 seq; |
1283 | ||
1284 | if (list_empty(&ctx->timeout_list)) | |
1285 | return; | |
1286 | ||
1287 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1288 | ||
1289 | do { | |
1290 | u32 events_needed, events_got; | |
360428f8 | 1291 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1292 | struct io_kiocb, timeout.list); |
de0617e4 | 1293 | |
8eb7e2d0 | 1294 | if (io_is_timeout_noseq(req)) |
360428f8 | 1295 | break; |
f010505b MDG |
1296 | |
1297 | /* | |
1298 | * Since seq can easily wrap around over time, subtract | |
1299 | * the last seq at which timeouts were flushed before comparing. | |
1300 | * Assuming not more than 2^31-1 events have happened since, | |
1301 | * these subtractions won't have wrapped, so we can check if | |
1302 | * target is in [last_seq, current_seq] by comparing the two. | |
1303 | */ | |
1304 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1305 | events_got = seq - ctx->cq_last_tm_flush; | |
1306 | if (events_got < events_needed) | |
360428f8 | 1307 | break; |
bfe68a22 | 1308 | |
135fcde8 | 1309 | list_del_init(&req->timeout.list); |
1ee4160c | 1310 | io_kill_timeout(req, 0); |
f010505b MDG |
1311 | } while (!list_empty(&ctx->timeout_list)); |
1312 | ||
1313 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1314 | } |
5262f567 | 1315 | |
360428f8 PB |
1316 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1317 | { | |
1318 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1319 | |
1320 | /* order cqe stores with ring update */ | |
1321 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1322 | |
04518945 PB |
1323 | if (unlikely(!list_empty(&ctx->defer_list))) |
1324 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1325 | } |
1326 | ||
90554200 JA |
1327 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1328 | { | |
1329 | struct io_rings *r = ctx->rings; | |
1330 | ||
1331 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1332 | } | |
1333 | ||
888aae2e PB |
1334 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1335 | { | |
1336 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1337 | } | |
1338 | ||
2b188cc1 JA |
1339 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1340 | { | |
75b28aff | 1341 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1342 | unsigned tail; |
1343 | ||
115e12e5 SB |
1344 | /* |
1345 | * writes to the cq entry need to come after reading head; the | |
1346 | * control dependency is enough as we're using WRITE_ONCE to | |
1347 | * fill the cq entry | |
1348 | */ | |
888aae2e | 1349 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1350 | return NULL; |
1351 | ||
888aae2e | 1352 | tail = ctx->cached_cq_tail++; |
75b28aff | 1353 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1354 | } |
1355 | ||
f2842ab5 JA |
1356 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1357 | { | |
f0b493e6 JA |
1358 | if (!ctx->cq_ev_fd) |
1359 | return false; | |
7e55a19c SG |
1360 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1361 | return false; | |
f2842ab5 JA |
1362 | if (!ctx->eventfd_async) |
1363 | return true; | |
b41e9852 | 1364 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1365 | } |
1366 | ||
b41e9852 | 1367 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1368 | { |
b1445e59 PB |
1369 | /* see waitqueue_active() comment */ |
1370 | smp_mb(); | |
1371 | ||
1d7bb1d5 JA |
1372 | if (waitqueue_active(&ctx->wait)) |
1373 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1374 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1375 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1376 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1377 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1378 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1379 | wake_up_interruptible(&ctx->cq_wait); |
1380 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1381 | } | |
1d7bb1d5 JA |
1382 | } |
1383 | ||
80c18e4a PB |
1384 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1385 | { | |
b1445e59 PB |
1386 | /* see waitqueue_active() comment */ |
1387 | smp_mb(); | |
1388 | ||
80c18e4a PB |
1389 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1390 | if (waitqueue_active(&ctx->wait)) | |
1391 | wake_up(&ctx->wait); | |
1392 | } | |
1393 | if (io_should_trigger_evfd(ctx)) | |
1394 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1395 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1396 | wake_up_interruptible(&ctx->cq_wait); |
1397 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1398 | } | |
80c18e4a PB |
1399 | } |
1400 | ||
c4a2ed72 | 1401 | /* Returns true if there are no backlogged entries after the flush */ |
6c503150 PB |
1402 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
1403 | struct task_struct *tsk, | |
1404 | struct files_struct *files) | |
1d7bb1d5 JA |
1405 | { |
1406 | struct io_rings *rings = ctx->rings; | |
e6c8aa9a | 1407 | struct io_kiocb *req, *tmp; |
1d7bb1d5 | 1408 | struct io_uring_cqe *cqe; |
1d7bb1d5 | 1409 | unsigned long flags; |
b18032bb | 1410 | bool all_flushed, posted; |
1d7bb1d5 JA |
1411 | LIST_HEAD(list); |
1412 | ||
e23de15f PB |
1413 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1414 | return false; | |
1d7bb1d5 | 1415 | |
b18032bb | 1416 | posted = false; |
1d7bb1d5 | 1417 | spin_lock_irqsave(&ctx->completion_lock, flags); |
e6c8aa9a | 1418 | list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) { |
08d23634 | 1419 | if (!io_match_task(req, tsk, files)) |
e6c8aa9a JA |
1420 | continue; |
1421 | ||
1d7bb1d5 JA |
1422 | cqe = io_get_cqring(ctx); |
1423 | if (!cqe && !force) | |
1424 | break; | |
1425 | ||
40d8ddd4 | 1426 | list_move(&req->compl.list, &list); |
1d7bb1d5 JA |
1427 | if (cqe) { |
1428 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1429 | WRITE_ONCE(cqe->res, req->result); | |
0f7e466b | 1430 | WRITE_ONCE(cqe->flags, req->compl.cflags); |
1d7bb1d5 | 1431 | } else { |
2c3bac6d | 1432 | ctx->cached_cq_overflow++; |
1d7bb1d5 | 1433 | WRITE_ONCE(ctx->rings->cq_overflow, |
2c3bac6d | 1434 | ctx->cached_cq_overflow); |
1d7bb1d5 | 1435 | } |
b18032bb | 1436 | posted = true; |
1d7bb1d5 JA |
1437 | } |
1438 | ||
09e88404 PB |
1439 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1440 | if (all_flushed) { | |
1441 | clear_bit(0, &ctx->sq_check_overflow); | |
1442 | clear_bit(0, &ctx->cq_check_overflow); | |
1443 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1444 | } | |
46930143 | 1445 | |
b18032bb JA |
1446 | if (posted) |
1447 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1448 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1449 | if (posted) |
1450 | io_cqring_ev_posted(ctx); | |
1d7bb1d5 JA |
1451 | |
1452 | while (!list_empty(&list)) { | |
40d8ddd4 PB |
1453 | req = list_first_entry(&list, struct io_kiocb, compl.list); |
1454 | list_del(&req->compl.list); | |
ec9c02ad | 1455 | io_put_req(req); |
1d7bb1d5 | 1456 | } |
c4a2ed72 | 1457 | |
09e88404 | 1458 | return all_flushed; |
1d7bb1d5 JA |
1459 | } |
1460 | ||
ca0a2651 | 1461 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force, |
6c503150 PB |
1462 | struct task_struct *tsk, |
1463 | struct files_struct *files) | |
1464 | { | |
ca0a2651 JA |
1465 | bool ret = true; |
1466 | ||
6c503150 PB |
1467 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1468 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1469 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1470 | mutex_lock(&ctx->uring_lock); | |
ca0a2651 | 1471 | ret = __io_cqring_overflow_flush(ctx, force, tsk, files); |
6c503150 PB |
1472 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1473 | mutex_unlock(&ctx->uring_lock); | |
1474 | } | |
ca0a2651 JA |
1475 | |
1476 | return ret; | |
6c503150 PB |
1477 | } |
1478 | ||
bcda7baa | 1479 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) |
2b188cc1 | 1480 | { |
78e19bbe | 1481 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1482 | struct io_uring_cqe *cqe; |
1483 | ||
78e19bbe | 1484 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1485 | |
2b188cc1 JA |
1486 | /* |
1487 | * If we can't get a cq entry, userspace overflowed the | |
1488 | * submission (by quite a lot). Increment the overflow count in | |
1489 | * the ring. | |
1490 | */ | |
1491 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1492 | if (likely(cqe)) { |
78e19bbe | 1493 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1494 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1495 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1496 | } else if (ctx->cq_overflow_flushed || |
1497 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1498 | /* |
1499 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1500 | * then we cannot store the request for later flushing, we need | |
1501 | * to drop it on the floor. | |
1502 | */ | |
2c3bac6d PB |
1503 | ctx->cached_cq_overflow++; |
1504 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1505 | } else { |
ad3eb2c8 JA |
1506 | if (list_empty(&ctx->cq_overflow_list)) { |
1507 | set_bit(0, &ctx->sq_check_overflow); | |
1508 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1509 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1510 | } |
40d8ddd4 | 1511 | io_clean_op(req); |
1d7bb1d5 | 1512 | req->result = res; |
0f7e466b | 1513 | req->compl.cflags = cflags; |
40d8ddd4 PB |
1514 | refcount_inc(&req->refs); |
1515 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); | |
2b188cc1 JA |
1516 | } |
1517 | } | |
1518 | ||
bcda7baa JA |
1519 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1520 | { | |
1521 | __io_cqring_fill_event(req, res, 0); | |
1522 | } | |
1523 | ||
7a612350 PB |
1524 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1525 | unsigned int cflags) | |
2b188cc1 | 1526 | { |
78e19bbe | 1527 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1528 | unsigned long flags; |
1529 | ||
1530 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1531 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1532 | /* |
1533 | * If we're the last reference to this request, add to our locked | |
1534 | * free_list cache. | |
1535 | */ | |
1536 | if (refcount_dec_and_test(&req->refs)) { | |
1537 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
1538 | ||
7a612350 PB |
1539 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1540 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1541 | io_disarm_next(req); | |
1542 | if (req->link) { | |
1543 | io_req_task_queue(req->link); | |
1544 | req->link = NULL; | |
1545 | } | |
1546 | } | |
c7dae4ba JA |
1547 | io_dismantle_req(req); |
1548 | io_put_task(req->task, 1); | |
1549 | list_add(&req->compl.list, &cs->locked_free_list); | |
1550 | cs->locked_free_nr++; | |
180f829f PB |
1551 | } else { |
1552 | if (!percpu_ref_tryget(&ctx->refs)) | |
1553 | req = NULL; | |
1554 | } | |
7a612350 | 1555 | io_commit_cqring(ctx); |
2b188cc1 | 1556 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1557 | |
180f829f PB |
1558 | if (req) { |
1559 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1560 | percpu_ref_put(&ctx->refs); |
180f829f | 1561 | } |
229a7b63 JA |
1562 | } |
1563 | ||
a38d68db | 1564 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1565 | unsigned int cflags) |
229a7b63 | 1566 | { |
a38d68db PB |
1567 | io_clean_op(req); |
1568 | req->result = res; | |
1569 | req->compl.cflags = cflags; | |
e342c807 | 1570 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1571 | } |
1572 | ||
889fca73 PB |
1573 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1574 | long res, unsigned cflags) | |
bcda7baa | 1575 | { |
889fca73 PB |
1576 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1577 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1578 | else |
c7dae4ba | 1579 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1580 | } |
1581 | ||
a38d68db | 1582 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1583 | { |
889fca73 | 1584 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1585 | } |
1586 | ||
c7dae4ba | 1587 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1588 | { |
c7dae4ba JA |
1589 | struct io_submit_state *state = &ctx->submit_state; |
1590 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1591 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1592 | |
c7dae4ba JA |
1593 | /* |
1594 | * If we have more than a batch's worth of requests in our IRQ side | |
1595 | * locked cache, grab the lock and move them over to our submission | |
1596 | * side cache. | |
1597 | */ | |
1598 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1599 | spin_lock_irq(&ctx->completion_lock); | |
1600 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1601 | cs->locked_free_nr = 0; | |
1602 | spin_unlock_irq(&ctx->completion_lock); | |
1603 | } | |
0ddf92e8 | 1604 | |
c7dae4ba JA |
1605 | while (!list_empty(&cs->free_list)) { |
1606 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1607 | compl.list); |
1608 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1609 | state->reqs[state->free_reqs++] = req; |
1610 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1611 | break; | |
1b4c351f JA |
1612 | } |
1613 | ||
e5d1bc0a | 1614 | return req != NULL; |
0ddf92e8 JA |
1615 | } |
1616 | ||
e5d1bc0a | 1617 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1618 | { |
e5d1bc0a PB |
1619 | struct io_submit_state *state = &ctx->submit_state; |
1620 | ||
1621 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1622 | ||
f6b6c7d6 | 1623 | if (!state->free_reqs) { |
291b2821 | 1624 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1625 | int ret; |
1626 | ||
c7dae4ba | 1627 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1628 | goto got_req; |
1629 | ||
bf019da7 PB |
1630 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1631 | state->reqs); | |
fd6fab2c JA |
1632 | |
1633 | /* | |
1634 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1635 | * retry single alloc to be on the safe side. | |
1636 | */ | |
1637 | if (unlikely(ret <= 0)) { | |
1638 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1639 | if (!state->reqs[0]) | |
3893f39f | 1640 | return NULL; |
fd6fab2c JA |
1641 | ret = 1; |
1642 | } | |
291b2821 | 1643 | state->free_reqs = ret; |
2b188cc1 | 1644 | } |
e5d1bc0a | 1645 | got_req: |
291b2821 PB |
1646 | state->free_reqs--; |
1647 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1648 | } |
1649 | ||
8da11c19 PB |
1650 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1651 | bool fixed) | |
1652 | { | |
36f72fe2 | 1653 | if (!fixed) |
8da11c19 PB |
1654 | fput(file); |
1655 | } | |
1656 | ||
4edf20f9 | 1657 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1658 | { |
3ca405eb | 1659 | io_clean_op(req); |
929a3af9 | 1660 | |
e8c2bc1f JA |
1661 | if (req->async_data) |
1662 | kfree(req->async_data); | |
8da11c19 PB |
1663 | if (req->file) |
1664 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1665 | if (req->fixed_rsrc_refs) |
1666 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1667 | if (req->work.creds) { |
1668 | put_cred(req->work.creds); | |
1669 | req->work.creds = NULL; | |
1670 | } | |
f85c310a PB |
1671 | |
1672 | if (req->flags & REQ_F_INFLIGHT) { | |
1673 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1674 | unsigned long flags; |
1675 | ||
1676 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1677 | list_del(&req->inflight_entry); | |
1678 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1679 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1680 | } |
e65ef56d JA |
1681 | } |
1682 | ||
b23fcf47 | 1683 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1684 | static inline void io_put_task(struct task_struct *task, int nr) |
1685 | { | |
1686 | struct io_uring_task *tctx = task->io_uring; | |
1687 | ||
1688 | percpu_counter_sub(&tctx->inflight, nr); | |
1689 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1690 | wake_up(&tctx->wait); | |
1691 | put_task_struct_many(task, nr); | |
1692 | } | |
1693 | ||
216578e5 | 1694 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1695 | { |
51a4cc11 | 1696 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1697 | |
216578e5 | 1698 | io_dismantle_req(req); |
7c660731 | 1699 | io_put_task(req->task, 1); |
c6ca97b3 | 1700 | |
3893f39f | 1701 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1702 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1703 | } |
1704 | ||
f2f87370 PB |
1705 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1706 | { | |
1707 | struct io_kiocb *nxt = req->link; | |
1708 | ||
1709 | req->link = nxt->link; | |
1710 | nxt->link = NULL; | |
1711 | } | |
1712 | ||
33cc89a9 PB |
1713 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1714 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1715 | { |
33cc89a9 | 1716 | struct io_kiocb *link = req->link; |
c9abd7ad | 1717 | bool cancelled = false; |
f2f87370 | 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); | |
33cc89a9 | 1732 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1733 | cancelled = true; |
1734 | } | |
1735 | } | |
7c86ffee | 1736 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1737 | return cancelled; |
7c86ffee PB |
1738 | } |
1739 | ||
d148ca4b | 1740 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1741 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1742 | { |
33cc89a9 | 1743 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1744 | |
f2f87370 | 1745 | req->link = NULL; |
f2f87370 PB |
1746 | while (link) { |
1747 | nxt = link->link; | |
1748 | link->link = NULL; | |
2665abfd | 1749 | |
f2f87370 | 1750 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1751 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1752 | io_put_req_deferred(link, 2); |
f2f87370 | 1753 | link = nxt; |
9e645e11 | 1754 | } |
33cc89a9 | 1755 | } |
9e645e11 | 1756 | |
33cc89a9 PB |
1757 | static bool io_disarm_next(struct io_kiocb *req) |
1758 | __must_hold(&req->ctx->completion_lock) | |
1759 | { | |
1760 | bool posted = false; | |
1761 | ||
1762 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1763 | posted = io_kill_linked_timeout(req); | |
1764 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1765 | posted |= (req->link != NULL); | |
1766 | io_fail_links(req); | |
1767 | } | |
1768 | return posted; | |
9e645e11 JA |
1769 | } |
1770 | ||
3fa5e0f3 | 1771 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1772 | { |
33cc89a9 | 1773 | struct io_kiocb *nxt; |
944e58bf | 1774 | |
9e645e11 JA |
1775 | /* |
1776 | * If LINK is set, we have dependent requests in this chain. If we | |
1777 | * didn't fail this request, queue the first one up, moving any other | |
1778 | * dependencies to the next request. In case of failure, fail the rest | |
1779 | * of the chain. | |
1780 | */ | |
33cc89a9 PB |
1781 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1782 | struct io_ring_ctx *ctx = req->ctx; | |
1783 | unsigned long flags; | |
1784 | bool posted; | |
1785 | ||
1786 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1787 | posted = io_disarm_next(req); | |
1788 | if (posted) | |
1789 | io_commit_cqring(req->ctx); | |
1790 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1791 | if (posted) | |
1792 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1793 | } |
33cc89a9 PB |
1794 | nxt = req->link; |
1795 | req->link = NULL; | |
1796 | return nxt; | |
4d7dd462 | 1797 | } |
9e645e11 | 1798 | |
f2f87370 | 1799 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1800 | { |
cdbff982 | 1801 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1802 | return NULL; |
1803 | return __io_req_find_next(req); | |
1804 | } | |
1805 | ||
2c32395d PB |
1806 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1807 | { | |
1808 | if (!ctx) | |
1809 | return; | |
1810 | if (ctx->submit_state.comp.nr) { | |
1811 | mutex_lock(&ctx->uring_lock); | |
1812 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1813 | mutex_unlock(&ctx->uring_lock); | |
1814 | } | |
1815 | percpu_ref_put(&ctx->refs); | |
1816 | } | |
1817 | ||
7cbf1722 | 1818 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1819 | { |
65453d1e | 1820 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1821 | struct io_wq_work_list list; |
1822 | struct io_wq_work_node *node; | |
c2c4c83c | 1823 | |
7cbf1722 JA |
1824 | if (wq_list_empty(&tctx->task_list)) |
1825 | return false; | |
6200b0ae | 1826 | |
0b81e80c | 1827 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1828 | list = tctx->task_list; |
1829 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1830 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1831 | |
7cbf1722 JA |
1832 | node = list.first; |
1833 | while (node) { | |
1834 | struct io_wq_work_node *next = node->next; | |
1835 | struct io_kiocb *req; | |
0ba9c9ed | 1836 | |
7cbf1722 | 1837 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1838 | if (req->ctx != ctx) { |
1839 | ctx_flush_and_put(ctx); | |
1840 | ctx = req->ctx; | |
1841 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1842 | } |
65453d1e | 1843 | |
2c32395d PB |
1844 | req->task_work.func(&req->task_work); |
1845 | node = next; | |
7cbf1722 JA |
1846 | } |
1847 | ||
2c32395d | 1848 | ctx_flush_and_put(ctx); |
7cbf1722 | 1849 | return list.first != NULL; |
c2c4c83c JA |
1850 | } |
1851 | ||
7cbf1722 | 1852 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1853 | { |
7cbf1722 | 1854 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1855 | |
1d5f360d JA |
1856 | clear_bit(0, &tctx->task_state); |
1857 | ||
7cbf1722 JA |
1858 | while (__tctx_task_work(tctx)) |
1859 | cond_resched(); | |
7cbf1722 JA |
1860 | } |
1861 | ||
1862 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1863 | enum task_work_notify_mode notify) | |
1864 | { | |
1865 | struct io_uring_task *tctx = tsk->io_uring; | |
1866 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1867 | unsigned long flags; |
7cbf1722 JA |
1868 | int ret; |
1869 | ||
1870 | WARN_ON_ONCE(!tctx); | |
1871 | ||
0b81e80c | 1872 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1873 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1874 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1875 | |
1876 | /* task_work already pending, we're done */ | |
1877 | if (test_bit(0, &tctx->task_state) || | |
1878 | test_and_set_bit(0, &tctx->task_state)) | |
1879 | return 0; | |
1880 | ||
1881 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1882 | return 0; | |
1883 | ||
1884 | /* | |
1885 | * Slow path - we failed, find and delete work. if the work is not | |
1886 | * in the list, it got run and we're fine. | |
1887 | */ | |
1888 | ret = 0; | |
0b81e80c | 1889 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1890 | wq_list_for_each(node, prev, &tctx->task_list) { |
1891 | if (&req->io_task_work.node == node) { | |
1892 | wq_list_del(&tctx->task_list, node, prev); | |
1893 | ret = 1; | |
1894 | break; | |
1895 | } | |
1896 | } | |
0b81e80c | 1897 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1898 | clear_bit(0, &tctx->task_state); |
1899 | return ret; | |
1900 | } | |
1901 | ||
355fb9e2 | 1902 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1903 | { |
1904 | struct task_struct *tsk = req->task; | |
1905 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1906 | enum task_work_notify_mode notify; |
1907 | int ret; | |
c2c4c83c | 1908 | |
6200b0ae JA |
1909 | if (tsk->flags & PF_EXITING) |
1910 | return -ESRCH; | |
1911 | ||
c2c4c83c | 1912 | /* |
0ba9c9ed JA |
1913 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1914 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1915 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1916 | * will do the job. | |
c2c4c83c | 1917 | */ |
91989c70 | 1918 | notify = TWA_NONE; |
355fb9e2 | 1919 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1920 | notify = TWA_SIGNAL; |
1921 | ||
7cbf1722 | 1922 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1923 | if (!ret) |
1924 | wake_up_process(tsk); | |
0ba9c9ed | 1925 | |
c2c4c83c JA |
1926 | return ret; |
1927 | } | |
1928 | ||
9b465711 PB |
1929 | static bool io_run_task_work_head(struct callback_head **work_head) |
1930 | { | |
1931 | struct callback_head *work, *next; | |
1932 | bool executed = false; | |
1933 | ||
1934 | do { | |
1935 | work = xchg(work_head, NULL); | |
1936 | if (!work) | |
1937 | break; | |
1938 | ||
1939 | do { | |
1940 | next = work->next; | |
1941 | work->func(work); | |
1942 | work = next; | |
1943 | cond_resched(); | |
1944 | } while (work); | |
1945 | executed = true; | |
1946 | } while (1); | |
1947 | ||
1948 | return executed; | |
1949 | } | |
1950 | ||
1951 | static void io_task_work_add_head(struct callback_head **work_head, | |
1952 | struct callback_head *task_work) | |
eab30c4d | 1953 | { |
7c25c0d1 | 1954 | struct callback_head *head; |
eab30c4d | 1955 | |
7c25c0d1 | 1956 | do { |
9b465711 PB |
1957 | head = READ_ONCE(*work_head); |
1958 | task_work->next = head; | |
1959 | } while (cmpxchg(work_head, head, task_work) != head); | |
1960 | } | |
1961 | ||
1962 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
1963 | task_work_func_t cb) | |
1964 | { | |
1965 | init_task_work(&req->task_work, cb); | |
1966 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
1967 | } |
1968 | ||
c40f6379 JA |
1969 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
1970 | { | |
1971 | struct io_ring_ctx *ctx = req->ctx; | |
1972 | ||
1973 | spin_lock_irq(&ctx->completion_lock); | |
1974 | io_cqring_fill_event(req, error); | |
1975 | io_commit_cqring(ctx); | |
1976 | spin_unlock_irq(&ctx->completion_lock); | |
1977 | ||
1978 | io_cqring_ev_posted(ctx); | |
1979 | req_set_fail_links(req); | |
1980 | io_double_put_req(req); | |
1981 | } | |
1982 | ||
1983 | static void io_req_task_cancel(struct callback_head *cb) | |
1984 | { | |
1985 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 1986 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 1987 | |
792bb6eb | 1988 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 1989 | __io_req_task_cancel(req, req->result); |
792bb6eb | 1990 | mutex_unlock(&ctx->uring_lock); |
87ceb6a6 | 1991 | percpu_ref_put(&ctx->refs); |
c40f6379 JA |
1992 | } |
1993 | ||
1994 | static void __io_req_task_submit(struct io_kiocb *req) | |
1995 | { | |
1996 | struct io_ring_ctx *ctx = req->ctx; | |
1997 | ||
04fc6c80 | 1998 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 1999 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2000 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2001 | __io_queue_sqe(req); |
81b6d05c | 2002 | else |
c40f6379 | 2003 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2004 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2005 | } |
2006 | ||
2007 | static void io_req_task_submit(struct callback_head *cb) | |
2008 | { | |
2009 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2010 | ||
2011 | __io_req_task_submit(req); | |
2012 | } | |
2013 | ||
2014 | static void io_req_task_queue(struct io_kiocb *req) | |
2015 | { | |
c40f6379 JA |
2016 | int ret; |
2017 | ||
7cbf1722 | 2018 | req->task_work.func = io_req_task_submit; |
355fb9e2 | 2019 | ret = io_req_task_work_add(req); |
c40f6379 | 2020 | if (unlikely(ret)) { |
a3df7698 | 2021 | req->result = -ECANCELED; |
04fc6c80 | 2022 | percpu_ref_get(&req->ctx->refs); |
eab30c4d | 2023 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 | 2024 | } |
c40f6379 JA |
2025 | } |
2026 | ||
a3df7698 PB |
2027 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
2028 | { | |
2029 | percpu_ref_get(&req->ctx->refs); | |
2030 | req->result = ret; | |
2031 | req->task_work.func = io_req_task_cancel; | |
2032 | ||
2033 | if (unlikely(io_req_task_work_add(req))) | |
2034 | io_req_task_work_add_fallback(req, io_req_task_cancel); | |
2035 | } | |
2036 | ||
f2f87370 | 2037 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2038 | { |
9b5f7bd9 | 2039 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2040 | |
2041 | if (nxt) | |
906a8c3f | 2042 | io_req_task_queue(nxt); |
c69f8dbe JL |
2043 | } |
2044 | ||
c3524383 | 2045 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2046 | { |
c3524383 PB |
2047 | io_queue_next(req); |
2048 | __io_free_req(req); | |
2049 | } | |
8766dd51 | 2050 | |
2d6500d4 | 2051 | struct req_batch { |
5af1d13e PB |
2052 | struct task_struct *task; |
2053 | int task_refs; | |
1b4c351f | 2054 | int ctx_refs; |
2d6500d4 PB |
2055 | }; |
2056 | ||
5af1d13e PB |
2057 | static inline void io_init_req_batch(struct req_batch *rb) |
2058 | { | |
5af1d13e | 2059 | rb->task_refs = 0; |
9ae72463 | 2060 | rb->ctx_refs = 0; |
5af1d13e PB |
2061 | rb->task = NULL; |
2062 | } | |
2063 | ||
2d6500d4 PB |
2064 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2065 | struct req_batch *rb) | |
2066 | { | |
6e833d53 | 2067 | if (rb->task) |
7c660731 | 2068 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2069 | if (rb->ctx_refs) |
2070 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2071 | } |
2072 | ||
6ff119a6 PB |
2073 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2074 | struct io_submit_state *state) | |
2d6500d4 | 2075 | { |
f2f87370 | 2076 | io_queue_next(req); |
2d6500d4 | 2077 | |
e3bc8e9d | 2078 | if (req->task != rb->task) { |
7c660731 PB |
2079 | if (rb->task) |
2080 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2081 | rb->task = req->task; |
2082 | rb->task_refs = 0; | |
5af1d13e | 2083 | } |
e3bc8e9d | 2084 | rb->task_refs++; |
9ae72463 | 2085 | rb->ctx_refs++; |
5af1d13e | 2086 | |
4edf20f9 | 2087 | io_dismantle_req(req); |
bd759045 | 2088 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2089 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2090 | else |
2091 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2092 | } |
2093 | ||
905c172f PB |
2094 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2095 | struct io_ring_ctx *ctx) | |
2096 | { | |
2097 | int i, nr = cs->nr; | |
2098 | struct io_kiocb *req; | |
2099 | struct req_batch rb; | |
2100 | ||
2101 | io_init_req_batch(&rb); | |
2102 | spin_lock_irq(&ctx->completion_lock); | |
2103 | for (i = 0; i < nr; i++) { | |
2104 | req = cs->reqs[i]; | |
2105 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2106 | } | |
2107 | io_commit_cqring(ctx); | |
2108 | spin_unlock_irq(&ctx->completion_lock); | |
2109 | ||
2110 | io_cqring_ev_posted(ctx); | |
2111 | for (i = 0; i < nr; i++) { | |
2112 | req = cs->reqs[i]; | |
2113 | ||
2114 | /* submission and completion refs */ | |
2115 | if (refcount_sub_and_test(2, &req->refs)) | |
6ff119a6 | 2116 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2117 | } |
2118 | ||
2119 | io_req_free_batch_finish(ctx, &rb); | |
2120 | cs->nr = 0; | |
7a743e22 PB |
2121 | } |
2122 | ||
ba816ad6 JA |
2123 | /* |
2124 | * Drop reference to request, return next in chain (if there is one) if this | |
2125 | * was the last reference to this request. | |
2126 | */ | |
9b5f7bd9 | 2127 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2128 | { |
9b5f7bd9 PB |
2129 | struct io_kiocb *nxt = NULL; |
2130 | ||
2a44f467 | 2131 | if (refcount_dec_and_test(&req->refs)) { |
9b5f7bd9 | 2132 | nxt = io_req_find_next(req); |
4d7dd462 | 2133 | __io_free_req(req); |
2a44f467 | 2134 | } |
9b5f7bd9 | 2135 | return nxt; |
2b188cc1 JA |
2136 | } |
2137 | ||
e65ef56d JA |
2138 | static void io_put_req(struct io_kiocb *req) |
2139 | { | |
2140 | if (refcount_dec_and_test(&req->refs)) | |
2141 | io_free_req(req); | |
2b188cc1 JA |
2142 | } |
2143 | ||
216578e5 PB |
2144 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2145 | { | |
2146 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2147 | ||
2148 | io_free_req(req); | |
2149 | } | |
2150 | ||
2151 | static void io_free_req_deferred(struct io_kiocb *req) | |
2152 | { | |
2153 | int ret; | |
2154 | ||
7cbf1722 | 2155 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2156 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2157 | if (unlikely(ret)) |
2158 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2159 | } |
2160 | ||
2161 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2162 | { | |
2163 | if (refcount_sub_and_test(refs, &req->refs)) | |
2164 | io_free_req_deferred(req); | |
2165 | } | |
2166 | ||
978db57e JA |
2167 | static void io_double_put_req(struct io_kiocb *req) |
2168 | { | |
2169 | /* drop both submit and complete references */ | |
2170 | if (refcount_sub_and_test(2, &req->refs)) | |
2171 | io_free_req(req); | |
2172 | } | |
2173 | ||
6c503150 | 2174 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2175 | { |
2176 | /* See comment at the top of this file */ | |
2177 | smp_rmb(); | |
e23de15f | 2178 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2179 | } |
2180 | ||
fb5ccc98 PB |
2181 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2182 | { | |
2183 | struct io_rings *rings = ctx->rings; | |
2184 | ||
2185 | /* make sure SQ entry isn't read before tail */ | |
2186 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2187 | } | |
2188 | ||
8ff069bf | 2189 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2190 | { |
8ff069bf | 2191 | unsigned int cflags; |
e94f141b | 2192 | |
bcda7baa JA |
2193 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2194 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2195 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2196 | kfree(kbuf); |
2197 | return cflags; | |
e94f141b JA |
2198 | } |
2199 | ||
8ff069bf | 2200 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2201 | { |
4d954c25 | 2202 | struct io_buffer *kbuf; |
bcda7baa | 2203 | |
4d954c25 | 2204 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2205 | return io_put_kbuf(req, kbuf); |
2206 | } | |
2207 | ||
4c6e277c JA |
2208 | static inline bool io_run_task_work(void) |
2209 | { | |
6200b0ae JA |
2210 | /* |
2211 | * Not safe to run on exiting task, and the task_work handling will | |
2212 | * not add work to such a task. | |
2213 | */ | |
2214 | if (unlikely(current->flags & PF_EXITING)) | |
2215 | return false; | |
4c6e277c JA |
2216 | if (current->task_works) { |
2217 | __set_current_state(TASK_RUNNING); | |
2218 | task_work_run(); | |
2219 | return true; | |
2220 | } | |
2221 | ||
2222 | return false; | |
bcda7baa JA |
2223 | } |
2224 | ||
def596e9 JA |
2225 | /* |
2226 | * Find and free completed poll iocbs | |
2227 | */ | |
2228 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2229 | struct list_head *done) | |
2230 | { | |
8237e045 | 2231 | struct req_batch rb; |
def596e9 | 2232 | struct io_kiocb *req; |
bbde017a XW |
2233 | |
2234 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2235 | smp_rmb(); | |
def596e9 | 2236 | |
5af1d13e | 2237 | io_init_req_batch(&rb); |
def596e9 | 2238 | while (!list_empty(done)) { |
bcda7baa JA |
2239 | int cflags = 0; |
2240 | ||
d21ffe7e | 2241 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2242 | list_del(&req->inflight_entry); |
2243 | ||
bbde017a XW |
2244 | if (READ_ONCE(req->result) == -EAGAIN) { |
2245 | req->iopoll_completed = 0; | |
23faba36 | 2246 | if (io_rw_reissue(req)) |
f161340d | 2247 | continue; |
bbde017a | 2248 | } |
def596e9 | 2249 | |
bcda7baa | 2250 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2251 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2252 | |
2253 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2254 | (*nr_events)++; |
2255 | ||
c3524383 | 2256 | if (refcount_dec_and_test(&req->refs)) |
6ff119a6 | 2257 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2258 | } |
def596e9 | 2259 | |
09bb8394 | 2260 | io_commit_cqring(ctx); |
80c18e4a | 2261 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2262 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2263 | } |
2264 | ||
def596e9 JA |
2265 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2266 | long min) | |
2267 | { | |
2268 | struct io_kiocb *req, *tmp; | |
2269 | LIST_HEAD(done); | |
2270 | bool spin; | |
2271 | int ret; | |
2272 | ||
2273 | /* | |
2274 | * Only spin for completions if we don't have multiple devices hanging | |
2275 | * off our complete list, and we're under the requested amount. | |
2276 | */ | |
2277 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2278 | ||
2279 | ret = 0; | |
d21ffe7e | 2280 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2281 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2282 | |
2283 | /* | |
581f9810 BM |
2284 | * Move completed and retryable entries to our local lists. |
2285 | * If we find a request that requires polling, break out | |
2286 | * and complete those lists first, if we have entries there. | |
def596e9 | 2287 | */ |
65a6543d | 2288 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2289 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2290 | continue; |
2291 | } | |
2292 | if (!list_empty(&done)) | |
2293 | break; | |
2294 | ||
2295 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2296 | if (ret < 0) | |
2297 | break; | |
2298 | ||
3aadc23e PB |
2299 | /* iopoll may have completed current req */ |
2300 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2301 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2302 | |
def596e9 JA |
2303 | if (ret && spin) |
2304 | spin = false; | |
2305 | ret = 0; | |
2306 | } | |
2307 | ||
2308 | if (!list_empty(&done)) | |
2309 | io_iopoll_complete(ctx, nr_events, &done); | |
2310 | ||
2311 | return ret; | |
2312 | } | |
2313 | ||
2314 | /* | |
d195a66e | 2315 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2316 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2317 | * as a non-spinning completion check. | |
2318 | */ | |
2319 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2320 | long min) | |
2321 | { | |
540e32a0 | 2322 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2323 | int ret; |
2324 | ||
2325 | ret = io_do_iopoll(ctx, nr_events, min); | |
2326 | if (ret < 0) | |
2327 | return ret; | |
eba0a4dd | 2328 | if (*nr_events >= min) |
def596e9 JA |
2329 | return 0; |
2330 | } | |
2331 | ||
2332 | return 1; | |
2333 | } | |
2334 | ||
2335 | /* | |
2336 | * We can't just wait for polled events to come to us, we have to actively | |
2337 | * find and complete them. | |
2338 | */ | |
b2edc0a7 | 2339 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2340 | { |
2341 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2342 | return; | |
2343 | ||
2344 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2345 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2346 | unsigned int nr_events = 0; |
2347 | ||
b2edc0a7 | 2348 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2349 | |
b2edc0a7 PB |
2350 | /* let it sleep and repeat later if can't complete a request */ |
2351 | if (nr_events == 0) | |
2352 | break; | |
08f5439f JA |
2353 | /* |
2354 | * Ensure we allow local-to-the-cpu processing to take place, | |
2355 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2356 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2357 | */ |
3fcee5a6 PB |
2358 | if (need_resched()) { |
2359 | mutex_unlock(&ctx->uring_lock); | |
2360 | cond_resched(); | |
2361 | mutex_lock(&ctx->uring_lock); | |
2362 | } | |
def596e9 JA |
2363 | } |
2364 | mutex_unlock(&ctx->uring_lock); | |
2365 | } | |
2366 | ||
7668b92a | 2367 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2368 | { |
7668b92a | 2369 | unsigned int nr_events = 0; |
2b2ed975 | 2370 | int iters = 0, ret = 0; |
500f9fba | 2371 | |
c7849be9 XW |
2372 | /* |
2373 | * We disallow the app entering submit/complete with polling, but we | |
2374 | * still need to lock the ring to prevent racing with polled issue | |
2375 | * that got punted to a workqueue. | |
2376 | */ | |
2377 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2378 | do { |
a3a0e43f JA |
2379 | /* |
2380 | * Don't enter poll loop if we already have events pending. | |
2381 | * If we do, we can potentially be spinning for commands that | |
2382 | * already triggered a CQE (eg in error). | |
2383 | */ | |
6c503150 PB |
2384 | if (test_bit(0, &ctx->cq_check_overflow)) |
2385 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2386 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2387 | break; |
2388 | ||
500f9fba JA |
2389 | /* |
2390 | * If a submit got punted to a workqueue, we can have the | |
2391 | * application entering polling for a command before it gets | |
2392 | * issued. That app will hold the uring_lock for the duration | |
2393 | * of the poll right here, so we need to take a breather every | |
2394 | * now and then to ensure that the issue has a chance to add | |
2395 | * the poll to the issued list. Otherwise we can spin here | |
2396 | * forever, while the workqueue is stuck trying to acquire the | |
2397 | * very same mutex. | |
2398 | */ | |
2399 | if (!(++iters & 7)) { | |
2400 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2401 | io_run_task_work(); |
500f9fba JA |
2402 | mutex_lock(&ctx->uring_lock); |
2403 | } | |
2404 | ||
7668b92a | 2405 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2406 | if (ret <= 0) |
2407 | break; | |
2408 | ret = 0; | |
7668b92a | 2409 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2410 | |
500f9fba | 2411 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2412 | return ret; |
2413 | } | |
2414 | ||
491381ce | 2415 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2416 | { |
491381ce JA |
2417 | /* |
2418 | * Tell lockdep we inherited freeze protection from submission | |
2419 | * thread. | |
2420 | */ | |
2421 | if (req->flags & REQ_F_ISREG) { | |
2422 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2423 | |
491381ce | 2424 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2425 | } |
491381ce | 2426 | file_end_write(req->file); |
2b188cc1 JA |
2427 | } |
2428 | ||
b63534c4 | 2429 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2430 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 JA |
2431 | { |
2432 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
4a245479 | 2433 | int rw, ret; |
b63534c4 | 2434 | struct iov_iter iter; |
b63534c4 | 2435 | |
dc2a6e9a PB |
2436 | /* already prepared */ |
2437 | if (req->async_data) | |
2438 | return true; | |
b63534c4 JA |
2439 | |
2440 | switch (req->opcode) { | |
2441 | case IORING_OP_READV: | |
2442 | case IORING_OP_READ_FIXED: | |
2443 | case IORING_OP_READ: | |
2444 | rw = READ; | |
2445 | break; | |
2446 | case IORING_OP_WRITEV: | |
2447 | case IORING_OP_WRITE_FIXED: | |
2448 | case IORING_OP_WRITE: | |
2449 | rw = WRITE; | |
2450 | break; | |
2451 | default: | |
2452 | printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n", | |
2453 | req->opcode); | |
dc2a6e9a | 2454 | return false; |
b63534c4 JA |
2455 | } |
2456 | ||
dc2a6e9a PB |
2457 | ret = io_import_iovec(rw, req, &iovec, &iter, false); |
2458 | if (ret < 0) | |
2459 | return false; | |
6bf985dc | 2460 | return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false); |
b63534c4 | 2461 | } |
b63534c4 | 2462 | |
3e6a0d3c | 2463 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2464 | { |
355afaeb | 2465 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2466 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2467 | |
355afaeb JA |
2468 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2469 | return false; | |
3e6a0d3c JA |
2470 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2471 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2472 | return false; |
7c977a58 JA |
2473 | /* |
2474 | * If ref is dying, we might be running poll reap from the exit work. | |
2475 | * Don't attempt to reissue from that path, just let it fail with | |
2476 | * -EAGAIN. | |
2477 | */ | |
3e6a0d3c JA |
2478 | if (percpu_ref_is_dying(&ctx->refs)) |
2479 | return false; | |
2480 | return true; | |
2481 | } | |
e82ad485 JA |
2482 | #else |
2483 | static bool io_rw_should_reissue(struct io_kiocb *req) | |
2484 | { | |
2485 | return false; | |
2486 | } | |
3e6a0d3c JA |
2487 | #endif |
2488 | ||
2489 | static bool io_rw_reissue(struct io_kiocb *req) | |
2490 | { | |
2491 | #ifdef CONFIG_BLOCK | |
2492 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2493 | return false; |
b63534c4 | 2494 | |
55e6ac1e PB |
2495 | lockdep_assert_held(&req->ctx->uring_lock); |
2496 | ||
37d1e2e3 | 2497 | if (io_resubmit_prep(req)) { |
fdee946d JA |
2498 | refcount_inc(&req->refs); |
2499 | io_queue_async_work(req); | |
b63534c4 | 2500 | return true; |
fdee946d | 2501 | } |
dc2a6e9a | 2502 | req_set_fail_links(req); |
b63534c4 JA |
2503 | #endif |
2504 | return false; | |
2505 | } | |
2506 | ||
a1d7c393 | 2507 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2508 | unsigned int issue_flags) |
a1d7c393 | 2509 | { |
2f8e45f1 PB |
2510 | int cflags = 0; |
2511 | ||
b65c128f PB |
2512 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2513 | kiocb_end_write(req); | |
230d50d4 JA |
2514 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_should_reissue(req)) { |
2515 | req->flags |= REQ_F_REISSUE; | |
23faba36 | 2516 | return; |
230d50d4 | 2517 | } |
2f8e45f1 PB |
2518 | if (res != req->result) |
2519 | req_set_fail_links(req); | |
2f8e45f1 PB |
2520 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2521 | cflags = io_put_rw_kbuf(req); | |
2522 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2523 | } |
2524 | ||
2525 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2526 | { | |
9adbd45d | 2527 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2528 | |
889fca73 | 2529 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2530 | } |
2531 | ||
def596e9 JA |
2532 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2533 | { | |
9adbd45d | 2534 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2535 | |
3e6a0d3c JA |
2536 | #ifdef CONFIG_BLOCK |
2537 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2538 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2539 | struct io_async_rw *rw = req->async_data; | |
2540 | ||
2541 | if (rw) | |
2542 | iov_iter_revert(&rw->iter, | |
2543 | req->result - iov_iter_count(&rw->iter)); | |
2544 | else if (!io_resubmit_prep(req)) | |
2545 | res = -EIO; | |
2546 | } | |
2547 | #endif | |
2548 | ||
491381ce JA |
2549 | if (kiocb->ki_flags & IOCB_WRITE) |
2550 | kiocb_end_write(req); | |
def596e9 | 2551 | |
2d7d6792 | 2552 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2553 | req_set_fail_links(req); |
bbde017a XW |
2554 | |
2555 | WRITE_ONCE(req->result, res); | |
2556 | /* order with io_poll_complete() checking ->result */ | |
cd664b0e PB |
2557 | smp_wmb(); |
2558 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2559 | } |
2560 | ||
2561 | /* | |
2562 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2563 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2564 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2565 | * accessing the kiocb cookie. | |
2566 | */ | |
2e9dbe90 | 2567 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2568 | { |
2569 | struct io_ring_ctx *ctx = req->ctx; | |
2570 | ||
2571 | /* | |
2572 | * Track whether we have multiple files in our lists. This will impact | |
2573 | * how we do polling eventually, not spinning if we're on potentially | |
2574 | * different devices. | |
2575 | */ | |
540e32a0 | 2576 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2577 | ctx->poll_multi_file = false; |
2578 | } else if (!ctx->poll_multi_file) { | |
2579 | struct io_kiocb *list_req; | |
2580 | ||
540e32a0 | 2581 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2582 | inflight_entry); |
9adbd45d | 2583 | if (list_req->file != req->file) |
def596e9 JA |
2584 | ctx->poll_multi_file = true; |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * For fast devices, IO may have already completed. If it has, add | |
2589 | * it to the front so we find it first. | |
2590 | */ | |
65a6543d | 2591 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2592 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2593 | else |
d21ffe7e | 2594 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2595 | |
2e9dbe90 XW |
2596 | /* |
2597 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2598 | * task context or in io worker task context. If current task context is | |
2599 | * sq thread, we don't need to check whether should wake up sq thread. | |
2600 | */ | |
2601 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2602 | wq_has_sleeper(&ctx->sq_data->wait)) |
2603 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2604 | } |
2605 | ||
9f13c35b PB |
2606 | static inline void io_state_file_put(struct io_submit_state *state) |
2607 | { | |
02b23a9a PB |
2608 | if (state->file_refs) { |
2609 | fput_many(state->file, state->file_refs); | |
2610 | state->file_refs = 0; | |
2611 | } | |
9a56a232 JA |
2612 | } |
2613 | ||
2614 | /* | |
2615 | * Get as many references to a file as we have IOs left in this submission, | |
2616 | * assuming most submissions are for one file, or at least that each file | |
2617 | * has more than one submission. | |
2618 | */ | |
8da11c19 | 2619 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2620 | { |
2621 | if (!state) | |
2622 | return fget(fd); | |
2623 | ||
6e1271e6 | 2624 | if (state->file_refs) { |
9a56a232 | 2625 | if (state->fd == fd) { |
6e1271e6 | 2626 | state->file_refs--; |
9a56a232 JA |
2627 | return state->file; |
2628 | } | |
02b23a9a | 2629 | io_state_file_put(state); |
9a56a232 JA |
2630 | } |
2631 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2632 | if (unlikely(!state->file)) |
9a56a232 JA |
2633 | return NULL; |
2634 | ||
2635 | state->fd = fd; | |
6e1271e6 | 2636 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2637 | return state->file; |
2638 | } | |
2639 | ||
4503b767 JA |
2640 | static bool io_bdev_nowait(struct block_device *bdev) |
2641 | { | |
9ba0d0c8 | 2642 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2643 | } |
2644 | ||
2b188cc1 JA |
2645 | /* |
2646 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2647 | * any file. For now, just ensure that anything potentially problematic is done | |
2648 | * inline. | |
2649 | */ | |
af197f50 | 2650 | static bool io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2651 | { |
2652 | umode_t mode = file_inode(file)->i_mode; | |
2653 | ||
4503b767 | 2654 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2655 | if (IS_ENABLED(CONFIG_BLOCK) && |
2656 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2657 | return true; |
2658 | return false; | |
2659 | } | |
2660 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2661 | return true; |
4503b767 | 2662 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2663 | if (IS_ENABLED(CONFIG_BLOCK) && |
2664 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2665 | file->f_op != &io_uring_fops) |
2666 | return true; | |
2667 | return false; | |
2668 | } | |
2b188cc1 | 2669 | |
c5b85625 JA |
2670 | /* any ->read/write should understand O_NONBLOCK */ |
2671 | if (file->f_flags & O_NONBLOCK) | |
2672 | return true; | |
2673 | ||
af197f50 JA |
2674 | if (!(file->f_mode & FMODE_NOWAIT)) |
2675 | return false; | |
2676 | ||
2677 | if (rw == READ) | |
2678 | return file->f_op->read_iter != NULL; | |
2679 | ||
2680 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2681 | } |
2682 | ||
a88fc400 | 2683 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2684 | { |
def596e9 | 2685 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2686 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2687 | struct file *file = req->file; |
09bb8394 JA |
2688 | unsigned ioprio; |
2689 | int ret; | |
2b188cc1 | 2690 | |
75c668cd | 2691 | if (S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2692 | req->flags |= REQ_F_ISREG; |
2693 | ||
2b188cc1 | 2694 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2695 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2696 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2697 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2698 | } |
2b188cc1 | 2699 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2700 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2701 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2702 | if (unlikely(ret)) | |
2703 | return ret; | |
2b188cc1 | 2704 | |
75c668cd PB |
2705 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2706 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2707 | req->flags |= REQ_F_NOWAIT; | |
2708 | ||
2b188cc1 JA |
2709 | ioprio = READ_ONCE(sqe->ioprio); |
2710 | if (ioprio) { | |
2711 | ret = ioprio_check_cap(ioprio); | |
2712 | if (ret) | |
09bb8394 | 2713 | return ret; |
2b188cc1 JA |
2714 | |
2715 | kiocb->ki_ioprio = ioprio; | |
2716 | } else | |
2717 | kiocb->ki_ioprio = get_current_ioprio(); | |
2718 | ||
def596e9 | 2719 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2720 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2721 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2722 | return -EOPNOTSUPP; |
2b188cc1 | 2723 | |
def596e9 JA |
2724 | kiocb->ki_flags |= IOCB_HIPRI; |
2725 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2726 | req->iopoll_completed = 0; |
def596e9 | 2727 | } else { |
09bb8394 JA |
2728 | if (kiocb->ki_flags & IOCB_HIPRI) |
2729 | return -EINVAL; | |
def596e9 JA |
2730 | kiocb->ki_complete = io_complete_rw; |
2731 | } | |
9adbd45d | 2732 | |
3529d8c2 JA |
2733 | req->rw.addr = READ_ONCE(sqe->addr); |
2734 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2735 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2736 | return 0; |
2b188cc1 JA |
2737 | } |
2738 | ||
2739 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2740 | { | |
2741 | switch (ret) { | |
2742 | case -EIOCBQUEUED: | |
2743 | break; | |
2744 | case -ERESTARTSYS: | |
2745 | case -ERESTARTNOINTR: | |
2746 | case -ERESTARTNOHAND: | |
2747 | case -ERESTART_RESTARTBLOCK: | |
2748 | /* | |
2749 | * We can't just restart the syscall, since previously | |
2750 | * submitted sqes may already be in progress. Just fail this | |
2751 | * IO with EINTR. | |
2752 | */ | |
2753 | ret = -EINTR; | |
df561f66 | 2754 | fallthrough; |
2b188cc1 JA |
2755 | default: |
2756 | kiocb->ki_complete(kiocb, ret, 0); | |
2757 | } | |
2758 | } | |
2759 | ||
a1d7c393 | 2760 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2761 | unsigned int issue_flags) |
ba816ad6 | 2762 | { |
ba04291e | 2763 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2764 | struct io_async_rw *io = req->async_data; |
97284637 | 2765 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2766 | |
227c0c96 | 2767 | /* add previously done IO, if any */ |
e8c2bc1f | 2768 | if (io && io->bytes_done > 0) { |
227c0c96 | 2769 | if (ret < 0) |
e8c2bc1f | 2770 | ret = io->bytes_done; |
227c0c96 | 2771 | else |
e8c2bc1f | 2772 | ret += io->bytes_done; |
227c0c96 JA |
2773 | } |
2774 | ||
ba04291e JA |
2775 | if (req->flags & REQ_F_CUR_POS) |
2776 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2777 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2778 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2779 | else |
2780 | io_rw_done(kiocb, ret); | |
97284637 PB |
2781 | |
2782 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2783 | req->flags &= ~REQ_F_REISSUE; | |
2784 | if (!io_rw_reissue(req)) { | |
2785 | int cflags = 0; | |
2786 | ||
2787 | req_set_fail_links(req); | |
2788 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2789 | cflags = io_put_rw_kbuf(req); | |
2790 | __io_req_complete(req, issue_flags, ret, cflags); | |
2791 | } | |
2792 | } | |
ba816ad6 JA |
2793 | } |
2794 | ||
847595de | 2795 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2796 | { |
9adbd45d JA |
2797 | struct io_ring_ctx *ctx = req->ctx; |
2798 | size_t len = req->rw.len; | |
edafccee | 2799 | struct io_mapped_ubuf *imu; |
4be1c615 | 2800 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2801 | size_t offset; |
2802 | u64 buf_addr; | |
2803 | ||
edafccee JA |
2804 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2805 | return -EFAULT; | |
edafccee JA |
2806 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2807 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2808 | buf_addr = req->rw.addr; |
edafccee JA |
2809 | |
2810 | /* overflow */ | |
2811 | if (buf_addr + len < buf_addr) | |
2812 | return -EFAULT; | |
2813 | /* not inside the mapped region */ | |
2814 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2815 | return -EFAULT; | |
2816 | ||
2817 | /* | |
2818 | * May not be a start of buffer, set size appropriately | |
2819 | * and advance us to the beginning. | |
2820 | */ | |
2821 | offset = buf_addr - imu->ubuf; | |
2822 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2823 | |
2824 | if (offset) { | |
2825 | /* | |
2826 | * Don't use iov_iter_advance() here, as it's really slow for | |
2827 | * using the latter parts of a big fixed buffer - it iterates | |
2828 | * over each segment manually. We can cheat a bit here, because | |
2829 | * we know that: | |
2830 | * | |
2831 | * 1) it's a BVEC iter, we set it up | |
2832 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2833 | * first and last bvec | |
2834 | * | |
2835 | * So just find our index, and adjust the iterator afterwards. | |
2836 | * If the offset is within the first bvec (or the whole first | |
2837 | * bvec, just use iov_iter_advance(). This makes it easier | |
2838 | * since we can just skip the first segment, which may not | |
2839 | * be PAGE_SIZE aligned. | |
2840 | */ | |
2841 | const struct bio_vec *bvec = imu->bvec; | |
2842 | ||
2843 | if (offset <= bvec->bv_len) { | |
2844 | iov_iter_advance(iter, offset); | |
2845 | } else { | |
2846 | unsigned long seg_skip; | |
2847 | ||
2848 | /* skip first vec */ | |
2849 | offset -= bvec->bv_len; | |
2850 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2851 | ||
2852 | iter->bvec = bvec + seg_skip; | |
2853 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2854 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2855 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2856 | } |
2857 | } | |
2858 | ||
847595de | 2859 | return 0; |
edafccee JA |
2860 | } |
2861 | ||
bcda7baa JA |
2862 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2863 | { | |
2864 | if (needs_lock) | |
2865 | mutex_unlock(&ctx->uring_lock); | |
2866 | } | |
2867 | ||
2868 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2869 | { | |
2870 | /* | |
2871 | * "Normal" inline submissions always hold the uring_lock, since we | |
2872 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2873 | * The only exception is when we've detached the request and issue it | |
2874 | * from an async worker thread, grab the lock for that case. | |
2875 | */ | |
2876 | if (needs_lock) | |
2877 | mutex_lock(&ctx->uring_lock); | |
2878 | } | |
2879 | ||
2880 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2881 | int bgid, struct io_buffer *kbuf, | |
2882 | bool needs_lock) | |
2883 | { | |
2884 | struct io_buffer *head; | |
2885 | ||
2886 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2887 | return kbuf; | |
2888 | ||
2889 | io_ring_submit_lock(req->ctx, needs_lock); | |
2890 | ||
2891 | lockdep_assert_held(&req->ctx->uring_lock); | |
2892 | ||
9e15c3a0 | 2893 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2894 | if (head) { |
2895 | if (!list_empty(&head->list)) { | |
2896 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2897 | list); | |
2898 | list_del(&kbuf->list); | |
2899 | } else { | |
2900 | kbuf = head; | |
9e15c3a0 | 2901 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2902 | } |
2903 | if (*len > kbuf->len) | |
2904 | *len = kbuf->len; | |
2905 | } else { | |
2906 | kbuf = ERR_PTR(-ENOBUFS); | |
2907 | } | |
2908 | ||
2909 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2910 | ||
2911 | return kbuf; | |
2912 | } | |
2913 | ||
4d954c25 JA |
2914 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2915 | bool needs_lock) | |
2916 | { | |
2917 | struct io_buffer *kbuf; | |
4f4eeba8 | 2918 | u16 bgid; |
4d954c25 JA |
2919 | |
2920 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2921 | bgid = req->buf_index; |
4d954c25 JA |
2922 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2923 | if (IS_ERR(kbuf)) | |
2924 | return kbuf; | |
2925 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2926 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2927 | return u64_to_user_ptr(kbuf->addr); | |
2928 | } | |
2929 | ||
2930 | #ifdef CONFIG_COMPAT | |
2931 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2932 | bool needs_lock) | |
2933 | { | |
2934 | struct compat_iovec __user *uiov; | |
2935 | compat_ssize_t clen; | |
2936 | void __user *buf; | |
2937 | ssize_t len; | |
2938 | ||
2939 | uiov = u64_to_user_ptr(req->rw.addr); | |
2940 | if (!access_ok(uiov, sizeof(*uiov))) | |
2941 | return -EFAULT; | |
2942 | if (__get_user(clen, &uiov->iov_len)) | |
2943 | return -EFAULT; | |
2944 | if (clen < 0) | |
2945 | return -EINVAL; | |
2946 | ||
2947 | len = clen; | |
2948 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2949 | if (IS_ERR(buf)) | |
2950 | return PTR_ERR(buf); | |
2951 | iov[0].iov_base = buf; | |
2952 | iov[0].iov_len = (compat_size_t) len; | |
2953 | return 0; | |
2954 | } | |
2955 | #endif | |
2956 | ||
2957 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2958 | bool needs_lock) | |
2959 | { | |
2960 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2961 | void __user *buf; | |
2962 | ssize_t len; | |
2963 | ||
2964 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2965 | return -EFAULT; | |
2966 | ||
2967 | len = iov[0].iov_len; | |
2968 | if (len < 0) | |
2969 | return -EINVAL; | |
2970 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2971 | if (IS_ERR(buf)) | |
2972 | return PTR_ERR(buf); | |
2973 | iov[0].iov_base = buf; | |
2974 | iov[0].iov_len = len; | |
2975 | return 0; | |
2976 | } | |
2977 | ||
2978 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2979 | bool needs_lock) | |
2980 | { | |
dddb3e26 JA |
2981 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2982 | struct io_buffer *kbuf; | |
2983 | ||
2984 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2985 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2986 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2987 | return 0; |
dddb3e26 | 2988 | } |
dd201662 | 2989 | if (req->rw.len != 1) |
4d954c25 JA |
2990 | return -EINVAL; |
2991 | ||
2992 | #ifdef CONFIG_COMPAT | |
2993 | if (req->ctx->compat) | |
2994 | return io_compat_import(req, iov, needs_lock); | |
2995 | #endif | |
2996 | ||
2997 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2998 | } | |
2999 | ||
847595de PB |
3000 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3001 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3002 | { |
9adbd45d JA |
3003 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3004 | size_t sqe_len = req->rw.len; | |
847595de | 3005 | u8 opcode = req->opcode; |
4d954c25 | 3006 | ssize_t ret; |
edafccee | 3007 | |
7d009165 | 3008 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3009 | *iovec = NULL; |
9adbd45d | 3010 | return io_import_fixed(req, rw, iter); |
edafccee | 3011 | } |
2b188cc1 | 3012 | |
bcda7baa | 3013 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3014 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3015 | return -EINVAL; |
3016 | ||
3a6820f2 | 3017 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3018 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3019 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3020 | if (IS_ERR(buf)) |
4d954c25 | 3021 | return PTR_ERR(buf); |
3f9d6441 | 3022 | req->rw.len = sqe_len; |
bcda7baa JA |
3023 | } |
3024 | ||
3a6820f2 JA |
3025 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3026 | *iovec = NULL; | |
10fc72e4 | 3027 | return ret; |
3a6820f2 JA |
3028 | } |
3029 | ||
4d954c25 JA |
3030 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3031 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3032 | if (!ret) |
3033 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3034 | *iovec = NULL; |
3035 | return ret; | |
3036 | } | |
3037 | ||
89cd35c5 CH |
3038 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3039 | req->ctx->compat); | |
2b188cc1 JA |
3040 | } |
3041 | ||
0fef9483 JA |
3042 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3043 | { | |
5b09e37e | 3044 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3045 | } |
3046 | ||
31b51510 | 3047 | /* |
32960613 JA |
3048 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3049 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3050 | */ |
4017eb91 | 3051 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3052 | { |
4017eb91 JA |
3053 | struct kiocb *kiocb = &req->rw.kiocb; |
3054 | struct file *file = req->file; | |
32960613 JA |
3055 | ssize_t ret = 0; |
3056 | ||
3057 | /* | |
3058 | * Don't support polled IO through this interface, and we can't | |
3059 | * support non-blocking either. For the latter, this just causes | |
3060 | * the kiocb to be handled from an async context. | |
3061 | */ | |
3062 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3063 | return -EOPNOTSUPP; | |
3064 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3065 | return -EAGAIN; | |
3066 | ||
3067 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3068 | struct iovec iovec; |
32960613 JA |
3069 | ssize_t nr; |
3070 | ||
311ae9e1 PB |
3071 | if (!iov_iter_is_bvec(iter)) { |
3072 | iovec = iov_iter_iovec(iter); | |
3073 | } else { | |
4017eb91 JA |
3074 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3075 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3076 | } |
3077 | ||
32960613 JA |
3078 | if (rw == READ) { |
3079 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3080 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3081 | } else { |
3082 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3083 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3084 | } |
3085 | ||
3086 | if (nr < 0) { | |
3087 | if (!ret) | |
3088 | ret = nr; | |
3089 | break; | |
3090 | } | |
3091 | ret += nr; | |
3092 | if (nr != iovec.iov_len) | |
3093 | break; | |
4017eb91 JA |
3094 | req->rw.len -= nr; |
3095 | req->rw.addr += nr; | |
32960613 JA |
3096 | iov_iter_advance(iter, nr); |
3097 | } | |
3098 | ||
3099 | return ret; | |
3100 | } | |
3101 | ||
ff6165b2 JA |
3102 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3103 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3104 | { |
e8c2bc1f | 3105 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3106 | |
ff6165b2 | 3107 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3108 | rw->free_iovec = iovec; |
227c0c96 | 3109 | rw->bytes_done = 0; |
ff6165b2 | 3110 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3111 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3112 | return; |
b64e3444 | 3113 | if (!iovec) { |
ff6165b2 JA |
3114 | unsigned iov_off = 0; |
3115 | ||
3116 | rw->iter.iov = rw->fast_iov; | |
3117 | if (iter->iov != fast_iov) { | |
3118 | iov_off = iter->iov - fast_iov; | |
3119 | rw->iter.iov += iov_off; | |
3120 | } | |
3121 | if (rw->fast_iov != fast_iov) | |
3122 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3123 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3124 | } else { |
3125 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3126 | } |
3127 | } | |
3128 | ||
e8c2bc1f | 3129 | static inline int __io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3130 | { |
e8c2bc1f JA |
3131 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3132 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3133 | return req->async_data == NULL; | |
3d9932a8 XW |
3134 | } |
3135 | ||
e8c2bc1f | 3136 | static int io_alloc_async_data(struct io_kiocb *req) |
f67676d1 | 3137 | { |
e8c2bc1f | 3138 | if (!io_op_defs[req->opcode].needs_async_data) |
d3656344 | 3139 | return 0; |
3d9932a8 | 3140 | |
e8c2bc1f | 3141 | return __io_alloc_async_data(req); |
b7bb4f7d JA |
3142 | } |
3143 | ||
ff6165b2 JA |
3144 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3145 | const struct iovec *fast_iov, | |
227c0c96 | 3146 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3147 | { |
e8c2bc1f | 3148 | if (!force && !io_op_defs[req->opcode].needs_async_data) |
74566df3 | 3149 | return 0; |
e8c2bc1f | 3150 | if (!req->async_data) { |
6bf985dc PB |
3151 | if (__io_alloc_async_data(req)) { |
3152 | kfree(iovec); | |
5d204bcf | 3153 | return -ENOMEM; |
6bf985dc | 3154 | } |
b7bb4f7d | 3155 | |
ff6165b2 | 3156 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3157 | } |
b7bb4f7d | 3158 | return 0; |
f67676d1 JA |
3159 | } |
3160 | ||
73debe68 | 3161 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3162 | { |
e8c2bc1f | 3163 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3164 | struct iovec *iov = iorw->fast_iov; |
847595de | 3165 | int ret; |
c3e330a4 | 3166 | |
2846c481 | 3167 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3168 | if (unlikely(ret < 0)) |
3169 | return ret; | |
3170 | ||
ab0b196c PB |
3171 | iorw->bytes_done = 0; |
3172 | iorw->free_iovec = iov; | |
3173 | if (iov) | |
3174 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3175 | return 0; |
3176 | } | |
3177 | ||
73debe68 | 3178 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3179 | { |
3529d8c2 JA |
3180 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3181 | return -EBADF; | |
93642ef8 | 3182 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3183 | } |
3184 | ||
c1dd91d1 JA |
3185 | /* |
3186 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3187 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3188 | * This gets called when the page is unlocked, and we generally expect that to | |
3189 | * happen when the page IO is completed and the page is now uptodate. This will | |
3190 | * queue a task_work based retry of the operation, attempting to copy the data | |
3191 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3192 | * do a thread based blocking retry of the operation. That's the unexpected | |
3193 | * slow path. | |
3194 | */ | |
bcf5a063 JA |
3195 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3196 | int sync, void *arg) | |
3197 | { | |
3198 | struct wait_page_queue *wpq; | |
3199 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3200 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3201 | |
3202 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3203 | ||
cdc8fcb4 LT |
3204 | if (!wake_page_match(wpq, key)) |
3205 | return 0; | |
3206 | ||
c8d317aa | 3207 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3208 | list_del_init(&wait->entry); |
3209 | ||
bcf5a063 JA |
3210 | /* submit ref gets dropped, acquire a new one */ |
3211 | refcount_inc(&req->refs); | |
921b9054 | 3212 | io_req_task_queue(req); |
bcf5a063 JA |
3213 | return 1; |
3214 | } | |
3215 | ||
c1dd91d1 JA |
3216 | /* |
3217 | * This controls whether a given IO request should be armed for async page | |
3218 | * based retry. If we return false here, the request is handed to the async | |
3219 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3220 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3221 | * will either succeed because the page is now uptodate and unlocked, or it | |
3222 | * will register a callback when the page is unlocked at IO completion. Through | |
3223 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3224 | * That retry will attempt the buffered read again. The retry will generally | |
3225 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3226 | * async worker threads for a blocking retry. | |
3227 | */ | |
227c0c96 | 3228 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3229 | { |
e8c2bc1f JA |
3230 | struct io_async_rw *rw = req->async_data; |
3231 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3232 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3233 | |
bcf5a063 JA |
3234 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3235 | if (req->flags & REQ_F_NOWAIT) | |
3236 | return false; | |
f67676d1 | 3237 | |
227c0c96 | 3238 | /* Only for buffered IO */ |
3b2a4439 | 3239 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3240 | return false; |
3b2a4439 | 3241 | |
bcf5a063 JA |
3242 | /* |
3243 | * just use poll if we can, and don't attempt if the fs doesn't | |
3244 | * support callback based unlocks | |
3245 | */ | |
3246 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3247 | return false; | |
f67676d1 | 3248 | |
3b2a4439 JA |
3249 | wait->wait.func = io_async_buf_func; |
3250 | wait->wait.private = req; | |
3251 | wait->wait.flags = 0; | |
3252 | INIT_LIST_HEAD(&wait->wait.entry); | |
3253 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3254 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3255 | kiocb->ki_waitq = wait; |
3b2a4439 | 3256 | return true; |
bcf5a063 JA |
3257 | } |
3258 | ||
3259 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3260 | { | |
3261 | if (req->file->f_op->read_iter) | |
3262 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3263 | else if (req->file->f_op->read) |
4017eb91 | 3264 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3265 | else |
3266 | return -EINVAL; | |
f67676d1 JA |
3267 | } |
3268 | ||
889fca73 | 3269 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3270 | { |
3271 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3272 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3273 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3274 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3275 | ssize_t io_size, ret, ret2; |
45d189c6 | 3276 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3277 | |
2846c481 | 3278 | if (rw) { |
e8c2bc1f | 3279 | iter = &rw->iter; |
2846c481 PB |
3280 | iovec = NULL; |
3281 | } else { | |
3282 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3283 | if (ret < 0) | |
3284 | return ret; | |
3285 | } | |
632546c4 | 3286 | io_size = iov_iter_count(iter); |
fa15bafb | 3287 | req->result = io_size; |
2b188cc1 | 3288 | |
fd6c2e4c JA |
3289 | /* Ensure we clear previously set non-block flag */ |
3290 | if (!force_nonblock) | |
29de5f6a | 3291 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3292 | else |
3293 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3294 | ||
24c74678 | 3295 | /* If the file doesn't support async, just async punt */ |
6713e7a6 PB |
3296 | if (force_nonblock && !io_file_supports_async(req->file, READ)) { |
3297 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); | |
6bf985dc | 3298 | return ret ?: -EAGAIN; |
6713e7a6 | 3299 | } |
9e645e11 | 3300 | |
632546c4 | 3301 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3302 | if (unlikely(ret)) { |
3303 | kfree(iovec); | |
3304 | return ret; | |
3305 | } | |
2b188cc1 | 3306 | |
227c0c96 | 3307 | ret = io_iter_do_read(req, iter); |
32960613 | 3308 | |
230d50d4 | 3309 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3310 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3311 | /* IOPOLL retry should happen for io-wq threads */ |
3312 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3313 | goto done; |
75c668cd PB |
3314 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3315 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3316 | goto done; |
84216315 | 3317 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3318 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3319 | ret = 0; |
230d50d4 JA |
3320 | } else if (ret == -EIOCBQUEUED) { |
3321 | goto out_free; | |
7335e3bf | 3322 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3323 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3324 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3325 | goto done; |
227c0c96 JA |
3326 | } |
3327 | ||
227c0c96 | 3328 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3329 | if (ret2) |
3330 | return ret2; | |
3331 | ||
fe1cdd55 | 3332 | iovec = NULL; |
e8c2bc1f | 3333 | rw = req->async_data; |
227c0c96 | 3334 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3335 | iter = &rw->iter; |
227c0c96 | 3336 | |
b23df91b PB |
3337 | do { |
3338 | io_size -= ret; | |
3339 | rw->bytes_done += ret; | |
3340 | /* if we can retry, do so with the callbacks armed */ | |
3341 | if (!io_rw_should_retry(req)) { | |
3342 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3343 | return -EAGAIN; | |
3344 | } | |
3345 | ||
3346 | /* | |
3347 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3348 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3349 | * desired page gets unlocked. We can also get a partial read | |
3350 | * here, and if we do, then just retry at the new offset. | |
3351 | */ | |
3352 | ret = io_iter_do_read(req, iter); | |
3353 | if (ret == -EIOCBQUEUED) | |
3354 | return 0; | |
227c0c96 | 3355 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3356 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3357 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3358 | done: |
889fca73 | 3359 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3360 | out_free: |
3361 | /* it's faster to check here then delegate to kfree */ | |
3362 | if (iovec) | |
3363 | kfree(iovec); | |
5ea5dd45 | 3364 | return 0; |
2b188cc1 JA |
3365 | } |
3366 | ||
73debe68 | 3367 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3368 | { |
3529d8c2 JA |
3369 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3370 | return -EBADF; | |
93642ef8 | 3371 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3372 | } |
3373 | ||
889fca73 | 3374 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3375 | { |
3376 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3377 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3378 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3379 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3380 | ssize_t ret, ret2, io_size; |
45d189c6 | 3381 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3382 | |
2846c481 | 3383 | if (rw) { |
e8c2bc1f | 3384 | iter = &rw->iter; |
2846c481 PB |
3385 | iovec = NULL; |
3386 | } else { | |
3387 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3388 | if (ret < 0) | |
3389 | return ret; | |
3390 | } | |
632546c4 | 3391 | io_size = iov_iter_count(iter); |
fa15bafb | 3392 | req->result = io_size; |
2b188cc1 | 3393 | |
fd6c2e4c JA |
3394 | /* Ensure we clear previously set non-block flag */ |
3395 | if (!force_nonblock) | |
a88fc400 PB |
3396 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3397 | else | |
3398 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3399 | |
24c74678 | 3400 | /* If the file doesn't support async, just async punt */ |
af197f50 | 3401 | if (force_nonblock && !io_file_supports_async(req->file, WRITE)) |
f67676d1 | 3402 | goto copy_iov; |
31b51510 | 3403 | |
10d59345 JA |
3404 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3405 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3406 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3407 | goto copy_iov; |
31b51510 | 3408 | |
632546c4 | 3409 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3410 | if (unlikely(ret)) |
3411 | goto out_free; | |
4ed734b0 | 3412 | |
fa15bafb PB |
3413 | /* |
3414 | * Open-code file_start_write here to grab freeze protection, | |
3415 | * which will be released by another thread in | |
3416 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3417 | * released so that it doesn't complain about the held lock when | |
3418 | * we return to userspace. | |
3419 | */ | |
3420 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3421 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3422 | __sb_writers_release(file_inode(req->file)->i_sb, |
3423 | SB_FREEZE_WRITE); | |
3424 | } | |
3425 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3426 | |
fa15bafb | 3427 | if (req->file->f_op->write_iter) |
ff6165b2 | 3428 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3429 | else if (req->file->f_op->write) |
4017eb91 | 3430 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3431 | else |
3432 | ret2 = -EINVAL; | |
4ed734b0 | 3433 | |
6ad7f233 PB |
3434 | if (req->flags & REQ_F_REISSUE) { |
3435 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3436 | ret2 = -EAGAIN; |
6ad7f233 | 3437 | } |
230d50d4 | 3438 | |
fa15bafb PB |
3439 | /* |
3440 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3441 | * retry them without IOCB_NOWAIT. | |
3442 | */ | |
3443 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3444 | ret2 = -EAGAIN; | |
75c668cd PB |
3445 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3446 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3447 | goto done; |
fa15bafb | 3448 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3449 | /* IOPOLL retry should happen for io-wq threads */ |
3450 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3451 | goto copy_iov; | |
355afaeb | 3452 | done: |
889fca73 | 3453 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3454 | } else { |
f67676d1 | 3455 | copy_iov: |
84216315 | 3456 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3457 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3458 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3459 | return ret ?: -EAGAIN; |
2b188cc1 | 3460 | } |
31b51510 | 3461 | out_free: |
f261c168 | 3462 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3463 | if (iovec) |
6f2cc166 | 3464 | kfree(iovec); |
2b188cc1 JA |
3465 | return ret; |
3466 | } | |
3467 | ||
80a261fd JA |
3468 | static int io_renameat_prep(struct io_kiocb *req, |
3469 | const struct io_uring_sqe *sqe) | |
3470 | { | |
3471 | struct io_rename *ren = &req->rename; | |
3472 | const char __user *oldf, *newf; | |
3473 | ||
3474 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3475 | return -EBADF; | |
3476 | ||
3477 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3478 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3479 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3480 | ren->new_dfd = READ_ONCE(sqe->len); | |
3481 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3482 | ||
3483 | ren->oldpath = getname(oldf); | |
3484 | if (IS_ERR(ren->oldpath)) | |
3485 | return PTR_ERR(ren->oldpath); | |
3486 | ||
3487 | ren->newpath = getname(newf); | |
3488 | if (IS_ERR(ren->newpath)) { | |
3489 | putname(ren->oldpath); | |
3490 | return PTR_ERR(ren->newpath); | |
3491 | } | |
3492 | ||
3493 | req->flags |= REQ_F_NEED_CLEANUP; | |
3494 | return 0; | |
3495 | } | |
3496 | ||
45d189c6 | 3497 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3498 | { |
3499 | struct io_rename *ren = &req->rename; | |
3500 | int ret; | |
3501 | ||
45d189c6 | 3502 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3503 | return -EAGAIN; |
3504 | ||
3505 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3506 | ren->newpath, ren->flags); | |
3507 | ||
3508 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3509 | if (ret < 0) | |
3510 | req_set_fail_links(req); | |
3511 | io_req_complete(req, ret); | |
3512 | return 0; | |
3513 | } | |
3514 | ||
14a1143b JA |
3515 | static int io_unlinkat_prep(struct io_kiocb *req, |
3516 | const struct io_uring_sqe *sqe) | |
3517 | { | |
3518 | struct io_unlink *un = &req->unlink; | |
3519 | const char __user *fname; | |
3520 | ||
3521 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3522 | return -EBADF; | |
3523 | ||
3524 | un->dfd = READ_ONCE(sqe->fd); | |
3525 | ||
3526 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3527 | if (un->flags & ~AT_REMOVEDIR) | |
3528 | return -EINVAL; | |
3529 | ||
3530 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3531 | un->filename = getname(fname); | |
3532 | if (IS_ERR(un->filename)) | |
3533 | return PTR_ERR(un->filename); | |
3534 | ||
3535 | req->flags |= REQ_F_NEED_CLEANUP; | |
3536 | return 0; | |
3537 | } | |
3538 | ||
45d189c6 | 3539 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3540 | { |
3541 | struct io_unlink *un = &req->unlink; | |
3542 | int ret; | |
3543 | ||
45d189c6 | 3544 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3545 | return -EAGAIN; |
3546 | ||
3547 | if (un->flags & AT_REMOVEDIR) | |
3548 | ret = do_rmdir(un->dfd, un->filename); | |
3549 | else | |
3550 | ret = do_unlinkat(un->dfd, un->filename); | |
3551 | ||
3552 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3553 | if (ret < 0) | |
3554 | req_set_fail_links(req); | |
3555 | io_req_complete(req, ret); | |
3556 | return 0; | |
3557 | } | |
3558 | ||
36f4fa68 JA |
3559 | static int io_shutdown_prep(struct io_kiocb *req, |
3560 | const struct io_uring_sqe *sqe) | |
3561 | { | |
3562 | #if defined(CONFIG_NET) | |
3563 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3564 | return -EINVAL; | |
3565 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3566 | sqe->buf_index) | |
3567 | return -EINVAL; | |
3568 | ||
3569 | req->shutdown.how = READ_ONCE(sqe->len); | |
3570 | return 0; | |
3571 | #else | |
3572 | return -EOPNOTSUPP; | |
3573 | #endif | |
3574 | } | |
3575 | ||
45d189c6 | 3576 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3577 | { |
3578 | #if defined(CONFIG_NET) | |
3579 | struct socket *sock; | |
3580 | int ret; | |
3581 | ||
45d189c6 | 3582 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3583 | return -EAGAIN; |
3584 | ||
48aba79b | 3585 | sock = sock_from_file(req->file); |
36f4fa68 | 3586 | if (unlikely(!sock)) |
48aba79b | 3587 | return -ENOTSOCK; |
36f4fa68 JA |
3588 | |
3589 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3590 | if (ret < 0) |
3591 | req_set_fail_links(req); | |
36f4fa68 JA |
3592 | io_req_complete(req, ret); |
3593 | return 0; | |
3594 | #else | |
3595 | return -EOPNOTSUPP; | |
3596 | #endif | |
3597 | } | |
3598 | ||
f2a8d5c7 PB |
3599 | static int __io_splice_prep(struct io_kiocb *req, |
3600 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3601 | { |
3602 | struct io_splice* sp = &req->splice; | |
3603 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3604 | |
3232dd02 PB |
3605 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3606 | return -EINVAL; | |
7d67af2c PB |
3607 | |
3608 | sp->file_in = NULL; | |
7d67af2c PB |
3609 | sp->len = READ_ONCE(sqe->len); |
3610 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3611 | ||
3612 | if (unlikely(sp->flags & ~valid_flags)) | |
3613 | return -EINVAL; | |
3614 | ||
8371adf5 PB |
3615 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3616 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3617 | if (!sp->file_in) | |
3618 | return -EBADF; | |
7d67af2c PB |
3619 | req->flags |= REQ_F_NEED_CLEANUP; |
3620 | ||
7cdaf587 XW |
3621 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3622 | /* | |
3623 | * Splice operation will be punted aync, and here need to | |
3624 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3625 | */ | |
7d67af2c | 3626 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3627 | } |
7d67af2c PB |
3628 | |
3629 | return 0; | |
3630 | } | |
3631 | ||
f2a8d5c7 PB |
3632 | static int io_tee_prep(struct io_kiocb *req, |
3633 | const struct io_uring_sqe *sqe) | |
3634 | { | |
3635 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3636 | return -EINVAL; | |
3637 | return __io_splice_prep(req, sqe); | |
3638 | } | |
3639 | ||
45d189c6 | 3640 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3641 | { |
3642 | struct io_splice *sp = &req->splice; | |
3643 | struct file *in = sp->file_in; | |
3644 | struct file *out = sp->file_out; | |
3645 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3646 | long ret = 0; | |
3647 | ||
45d189c6 | 3648 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3649 | return -EAGAIN; |
3650 | if (sp->len) | |
3651 | ret = do_tee(in, out, sp->len, flags); | |
3652 | ||
3653 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3654 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3655 | ||
f2a8d5c7 PB |
3656 | if (ret != sp->len) |
3657 | req_set_fail_links(req); | |
e1e16097 | 3658 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3659 | return 0; |
3660 | } | |
3661 | ||
3662 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3663 | { | |
3664 | struct io_splice* sp = &req->splice; | |
3665 | ||
3666 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3667 | sp->off_out = READ_ONCE(sqe->off); | |
3668 | return __io_splice_prep(req, sqe); | |
3669 | } | |
3670 | ||
45d189c6 | 3671 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3672 | { |
3673 | struct io_splice *sp = &req->splice; | |
3674 | struct file *in = sp->file_in; | |
3675 | struct file *out = sp->file_out; | |
3676 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3677 | loff_t *poff_in, *poff_out; | |
c9687426 | 3678 | long ret = 0; |
7d67af2c | 3679 | |
45d189c6 | 3680 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3681 | return -EAGAIN; |
7d67af2c PB |
3682 | |
3683 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3684 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3685 | |
948a7749 | 3686 | if (sp->len) |
c9687426 | 3687 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3688 | |
3689 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3690 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3691 | ||
7d67af2c PB |
3692 | if (ret != sp->len) |
3693 | req_set_fail_links(req); | |
e1e16097 | 3694 | io_req_complete(req, ret); |
7d67af2c PB |
3695 | return 0; |
3696 | } | |
3697 | ||
2b188cc1 JA |
3698 | /* |
3699 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3700 | */ | |
889fca73 | 3701 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3702 | { |
3703 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3704 | |
def596e9 JA |
3705 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3706 | return -EINVAL; | |
3707 | ||
889fca73 | 3708 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3709 | return 0; |
3710 | } | |
3711 | ||
1155c76a | 3712 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3713 | { |
6b06314c | 3714 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3715 | |
09bb8394 JA |
3716 | if (!req->file) |
3717 | return -EBADF; | |
c992fe29 | 3718 | |
6b06314c | 3719 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3720 | return -EINVAL; |
edafccee | 3721 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3722 | return -EINVAL; |
3723 | ||
8ed8d3c3 JA |
3724 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3725 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3726 | return -EINVAL; | |
3727 | ||
3728 | req->sync.off = READ_ONCE(sqe->off); | |
3729 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3730 | return 0; |
3731 | } | |
3732 | ||
45d189c6 | 3733 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3734 | { |
8ed8d3c3 | 3735 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3736 | int ret; |
3737 | ||
ac45abc0 | 3738 | /* fsync always requires a blocking context */ |
45d189c6 | 3739 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3740 | return -EAGAIN; |
3741 | ||
9adbd45d | 3742 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3743 | end > 0 ? end : LLONG_MAX, |
3744 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3745 | if (ret < 0) | |
3746 | req_set_fail_links(req); | |
e1e16097 | 3747 | io_req_complete(req, ret); |
c992fe29 CH |
3748 | return 0; |
3749 | } | |
3750 | ||
d63d1b5e JA |
3751 | static int io_fallocate_prep(struct io_kiocb *req, |
3752 | const struct io_uring_sqe *sqe) | |
3753 | { | |
3754 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3755 | return -EINVAL; | |
3232dd02 PB |
3756 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3757 | return -EINVAL; | |
d63d1b5e JA |
3758 | |
3759 | req->sync.off = READ_ONCE(sqe->off); | |
3760 | req->sync.len = READ_ONCE(sqe->addr); | |
3761 | req->sync.mode = READ_ONCE(sqe->len); | |
3762 | return 0; | |
3763 | } | |
3764 | ||
45d189c6 | 3765 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3766 | { |
ac45abc0 PB |
3767 | int ret; |
3768 | ||
d63d1b5e | 3769 | /* fallocate always requiring blocking context */ |
45d189c6 | 3770 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3771 | return -EAGAIN; |
ac45abc0 PB |
3772 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3773 | req->sync.len); | |
ac45abc0 PB |
3774 | if (ret < 0) |
3775 | req_set_fail_links(req); | |
e1e16097 | 3776 | io_req_complete(req, ret); |
5d17b4a4 JA |
3777 | return 0; |
3778 | } | |
3779 | ||
ec65fea5 | 3780 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3781 | { |
f8748881 | 3782 | const char __user *fname; |
15b71abe | 3783 | int ret; |
b7bb4f7d | 3784 | |
ec65fea5 | 3785 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3786 | return -EINVAL; |
ec65fea5 | 3787 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3788 | return -EBADF; |
03b1230c | 3789 | |
ec65fea5 PB |
3790 | /* open.how should be already initialised */ |
3791 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3792 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3793 | |
25e72d10 PB |
3794 | req->open.dfd = READ_ONCE(sqe->fd); |
3795 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3796 | req->open.filename = getname(fname); |
15b71abe JA |
3797 | if (IS_ERR(req->open.filename)) { |
3798 | ret = PTR_ERR(req->open.filename); | |
3799 | req->open.filename = NULL; | |
3800 | return ret; | |
3801 | } | |
4022e7af | 3802 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3803 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3804 | return 0; |
03b1230c JA |
3805 | } |
3806 | ||
ec65fea5 PB |
3807 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3808 | { | |
3809 | u64 flags, mode; | |
3810 | ||
14587a46 | 3811 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3812 | return -EINVAL; |
ec65fea5 PB |
3813 | mode = READ_ONCE(sqe->len); |
3814 | flags = READ_ONCE(sqe->open_flags); | |
3815 | req->open.how = build_open_how(flags, mode); | |
3816 | return __io_openat_prep(req, sqe); | |
3817 | } | |
3818 | ||
cebdb986 | 3819 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3820 | { |
cebdb986 | 3821 | struct open_how __user *how; |
cebdb986 | 3822 | size_t len; |
0fa03c62 JA |
3823 | int ret; |
3824 | ||
14587a46 | 3825 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3826 | return -EINVAL; |
cebdb986 JA |
3827 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3828 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3829 | if (len < OPEN_HOW_SIZE_VER0) |
3830 | return -EINVAL; | |
3529d8c2 | 3831 | |
cebdb986 JA |
3832 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3833 | len); | |
3834 | if (ret) | |
3835 | return ret; | |
3529d8c2 | 3836 | |
ec65fea5 | 3837 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3838 | } |
3839 | ||
45d189c6 | 3840 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3841 | { |
3842 | struct open_flags op; | |
15b71abe | 3843 | struct file *file; |
3a81fd02 JA |
3844 | bool nonblock_set; |
3845 | bool resolve_nonblock; | |
15b71abe JA |
3846 | int ret; |
3847 | ||
cebdb986 | 3848 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3849 | if (ret) |
3850 | goto err; | |
3a81fd02 JA |
3851 | nonblock_set = op.open_flag & O_NONBLOCK; |
3852 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3853 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3854 | /* |
3855 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3856 | * it'll always -EAGAIN | |
3857 | */ | |
3858 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3859 | return -EAGAIN; | |
3860 | op.lookup_flags |= LOOKUP_CACHED; | |
3861 | op.open_flag |= O_NONBLOCK; | |
3862 | } | |
15b71abe | 3863 | |
4022e7af | 3864 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3865 | if (ret < 0) |
3866 | goto err; | |
3867 | ||
3868 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3869 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3870 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3871 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3872 | /* |
3a81fd02 JA |
3873 | * We could hang on to this 'fd', but seems like marginal |
3874 | * gain for something that is now known to be a slower path. | |
3875 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3876 | */ |
3a81fd02 JA |
3877 | put_unused_fd(ret); |
3878 | return -EAGAIN; | |
3879 | } | |
3880 | ||
15b71abe JA |
3881 | if (IS_ERR(file)) { |
3882 | put_unused_fd(ret); | |
3883 | ret = PTR_ERR(file); | |
3884 | } else { | |
45d189c6 | 3885 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3886 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3887 | fsnotify_open(file); |
3888 | fd_install(ret, file); | |
3889 | } | |
3890 | err: | |
3891 | putname(req->open.filename); | |
8fef80bf | 3892 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3893 | if (ret < 0) |
3894 | req_set_fail_links(req); | |
e1e16097 | 3895 | io_req_complete(req, ret); |
15b71abe JA |
3896 | return 0; |
3897 | } | |
3898 | ||
45d189c6 | 3899 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3900 | { |
e45cff58 | 3901 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3902 | } |
3903 | ||
067524e9 JA |
3904 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3905 | const struct io_uring_sqe *sqe) | |
3906 | { | |
3907 | struct io_provide_buf *p = &req->pbuf; | |
3908 | u64 tmp; | |
3909 | ||
3910 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3911 | return -EINVAL; | |
3912 | ||
3913 | tmp = READ_ONCE(sqe->fd); | |
3914 | if (!tmp || tmp > USHRT_MAX) | |
3915 | return -EINVAL; | |
3916 | ||
3917 | memset(p, 0, sizeof(*p)); | |
3918 | p->nbufs = tmp; | |
3919 | p->bgid = READ_ONCE(sqe->buf_group); | |
3920 | return 0; | |
3921 | } | |
3922 | ||
3923 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3924 | int bgid, unsigned nbufs) | |
3925 | { | |
3926 | unsigned i = 0; | |
3927 | ||
3928 | /* shouldn't happen */ | |
3929 | if (!nbufs) | |
3930 | return 0; | |
3931 | ||
3932 | /* the head kbuf is the list itself */ | |
3933 | while (!list_empty(&buf->list)) { | |
3934 | struct io_buffer *nxt; | |
3935 | ||
3936 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3937 | list_del(&nxt->list); | |
3938 | kfree(nxt); | |
3939 | if (++i == nbufs) | |
3940 | return i; | |
3941 | } | |
3942 | i++; | |
3943 | kfree(buf); | |
9e15c3a0 | 3944 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3945 | |
3946 | return i; | |
3947 | } | |
3948 | ||
889fca73 | 3949 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3950 | { |
3951 | struct io_provide_buf *p = &req->pbuf; | |
3952 | struct io_ring_ctx *ctx = req->ctx; | |
3953 | struct io_buffer *head; | |
3954 | int ret = 0; | |
45d189c6 | 3955 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3956 | |
3957 | io_ring_submit_lock(ctx, !force_nonblock); | |
3958 | ||
3959 | lockdep_assert_held(&ctx->uring_lock); | |
3960 | ||
3961 | ret = -ENOENT; | |
9e15c3a0 | 3962 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3963 | if (head) |
3964 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3965 | if (ret < 0) |
3966 | req_set_fail_links(req); | |
067524e9 | 3967 | |
31bff9a5 PB |
3968 | /* need to hold the lock to complete IOPOLL requests */ |
3969 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 3970 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
3971 | io_ring_submit_unlock(ctx, !force_nonblock); |
3972 | } else { | |
3973 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 3974 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 3975 | } |
067524e9 JA |
3976 | return 0; |
3977 | } | |
3978 | ||
ddf0322d JA |
3979 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3980 | const struct io_uring_sqe *sqe) | |
3981 | { | |
d81269fe | 3982 | unsigned long size; |
ddf0322d JA |
3983 | struct io_provide_buf *p = &req->pbuf; |
3984 | u64 tmp; | |
3985 | ||
3986 | if (sqe->ioprio || sqe->rw_flags) | |
3987 | return -EINVAL; | |
3988 | ||
3989 | tmp = READ_ONCE(sqe->fd); | |
3990 | if (!tmp || tmp > USHRT_MAX) | |
3991 | return -E2BIG; | |
3992 | p->nbufs = tmp; | |
3993 | p->addr = READ_ONCE(sqe->addr); | |
3994 | p->len = READ_ONCE(sqe->len); | |
3995 | ||
d81269fe PB |
3996 | size = (unsigned long)p->len * p->nbufs; |
3997 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3998 | return -EFAULT; |
3999 | ||
4000 | p->bgid = READ_ONCE(sqe->buf_group); | |
4001 | tmp = READ_ONCE(sqe->off); | |
4002 | if (tmp > USHRT_MAX) | |
4003 | return -E2BIG; | |
4004 | p->bid = tmp; | |
4005 | return 0; | |
4006 | } | |
4007 | ||
4008 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4009 | { | |
4010 | struct io_buffer *buf; | |
4011 | u64 addr = pbuf->addr; | |
4012 | int i, bid = pbuf->bid; | |
4013 | ||
4014 | for (i = 0; i < pbuf->nbufs; i++) { | |
4015 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4016 | if (!buf) | |
4017 | break; | |
4018 | ||
4019 | buf->addr = addr; | |
4020 | buf->len = pbuf->len; | |
4021 | buf->bid = bid; | |
4022 | addr += pbuf->len; | |
4023 | bid++; | |
4024 | if (!*head) { | |
4025 | INIT_LIST_HEAD(&buf->list); | |
4026 | *head = buf; | |
4027 | } else { | |
4028 | list_add_tail(&buf->list, &(*head)->list); | |
4029 | } | |
4030 | } | |
4031 | ||
4032 | return i ? i : -ENOMEM; | |
4033 | } | |
4034 | ||
889fca73 | 4035 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4036 | { |
4037 | struct io_provide_buf *p = &req->pbuf; | |
4038 | struct io_ring_ctx *ctx = req->ctx; | |
4039 | struct io_buffer *head, *list; | |
4040 | int ret = 0; | |
45d189c6 | 4041 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4042 | |
4043 | io_ring_submit_lock(ctx, !force_nonblock); | |
4044 | ||
4045 | lockdep_assert_held(&ctx->uring_lock); | |
4046 | ||
9e15c3a0 | 4047 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4048 | |
4049 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4050 | if (ret >= 0 && !list) { |
4051 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4052 | if (ret < 0) | |
067524e9 | 4053 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4054 | } |
ddf0322d JA |
4055 | if (ret < 0) |
4056 | req_set_fail_links(req); | |
31bff9a5 PB |
4057 | |
4058 | /* need to hold the lock to complete IOPOLL requests */ | |
4059 | if (ctx->flags & IORING_SETUP_IOPOLL) { | |
889fca73 | 4060 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 PB |
4061 | io_ring_submit_unlock(ctx, !force_nonblock); |
4062 | } else { | |
4063 | io_ring_submit_unlock(ctx, !force_nonblock); | |
889fca73 | 4064 | __io_req_complete(req, issue_flags, ret, 0); |
31bff9a5 | 4065 | } |
ddf0322d | 4066 | return 0; |
cebdb986 JA |
4067 | } |
4068 | ||
3e4827b0 JA |
4069 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4070 | const struct io_uring_sqe *sqe) | |
4071 | { | |
4072 | #if defined(CONFIG_EPOLL) | |
4073 | if (sqe->ioprio || sqe->buf_index) | |
4074 | return -EINVAL; | |
6ca56f84 | 4075 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4076 | return -EINVAL; |
3e4827b0 JA |
4077 | |
4078 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4079 | req->epoll.op = READ_ONCE(sqe->len); | |
4080 | req->epoll.fd = READ_ONCE(sqe->off); | |
4081 | ||
4082 | if (ep_op_has_event(req->epoll.op)) { | |
4083 | struct epoll_event __user *ev; | |
4084 | ||
4085 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4086 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4087 | return -EFAULT; | |
4088 | } | |
4089 | ||
4090 | return 0; | |
4091 | #else | |
4092 | return -EOPNOTSUPP; | |
4093 | #endif | |
4094 | } | |
4095 | ||
889fca73 | 4096 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4097 | { |
4098 | #if defined(CONFIG_EPOLL) | |
4099 | struct io_epoll *ie = &req->epoll; | |
4100 | int ret; | |
45d189c6 | 4101 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4102 | |
4103 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4104 | if (force_nonblock && ret == -EAGAIN) | |
4105 | return -EAGAIN; | |
4106 | ||
4107 | if (ret < 0) | |
4108 | req_set_fail_links(req); | |
889fca73 | 4109 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4110 | return 0; |
4111 | #else | |
4112 | return -EOPNOTSUPP; | |
4113 | #endif | |
4114 | } | |
4115 | ||
c1ca757b JA |
4116 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4117 | { | |
4118 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4119 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4120 | return -EINVAL; | |
3232dd02 PB |
4121 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4122 | return -EINVAL; | |
c1ca757b JA |
4123 | |
4124 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4125 | req->madvise.len = READ_ONCE(sqe->len); | |
4126 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4127 | return 0; | |
4128 | #else | |
4129 | return -EOPNOTSUPP; | |
4130 | #endif | |
4131 | } | |
4132 | ||
45d189c6 | 4133 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4134 | { |
4135 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4136 | struct io_madvise *ma = &req->madvise; | |
4137 | int ret; | |
4138 | ||
45d189c6 | 4139 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4140 | return -EAGAIN; |
4141 | ||
0726b01e | 4142 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4143 | if (ret < 0) |
4144 | req_set_fail_links(req); | |
e1e16097 | 4145 | io_req_complete(req, ret); |
c1ca757b JA |
4146 | return 0; |
4147 | #else | |
4148 | return -EOPNOTSUPP; | |
4149 | #endif | |
4150 | } | |
4151 | ||
4840e418 JA |
4152 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4153 | { | |
4154 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4155 | return -EINVAL; | |
3232dd02 PB |
4156 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4157 | return -EINVAL; | |
4840e418 JA |
4158 | |
4159 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4160 | req->fadvise.len = READ_ONCE(sqe->len); | |
4161 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4162 | return 0; | |
4163 | } | |
4164 | ||
45d189c6 | 4165 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4166 | { |
4167 | struct io_fadvise *fa = &req->fadvise; | |
4168 | int ret; | |
4169 | ||
45d189c6 | 4170 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4171 | switch (fa->advice) { |
4172 | case POSIX_FADV_NORMAL: | |
4173 | case POSIX_FADV_RANDOM: | |
4174 | case POSIX_FADV_SEQUENTIAL: | |
4175 | break; | |
4176 | default: | |
4177 | return -EAGAIN; | |
4178 | } | |
4179 | } | |
4840e418 JA |
4180 | |
4181 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4182 | if (ret < 0) | |
4183 | req_set_fail_links(req); | |
e1e16097 | 4184 | io_req_complete(req, ret); |
4840e418 JA |
4185 | return 0; |
4186 | } | |
4187 | ||
eddc7ef5 JA |
4188 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4189 | { | |
6ca56f84 | 4190 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4191 | return -EINVAL; |
eddc7ef5 JA |
4192 | if (sqe->ioprio || sqe->buf_index) |
4193 | return -EINVAL; | |
9c280f90 | 4194 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4195 | return -EBADF; |
eddc7ef5 | 4196 | |
1d9e1288 BM |
4197 | req->statx.dfd = READ_ONCE(sqe->fd); |
4198 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4199 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4200 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4201 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4202 | |
4203 | return 0; | |
4204 | } | |
4205 | ||
45d189c6 | 4206 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4207 | { |
1d9e1288 | 4208 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4209 | int ret; |
4210 | ||
45d189c6 | 4211 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4212 | /* only need file table for an actual valid fd */ |
4213 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4214 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4215 | return -EAGAIN; |
5b0bbee4 | 4216 | } |
eddc7ef5 | 4217 | |
e62753e4 BM |
4218 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4219 | ctx->buffer); | |
eddc7ef5 | 4220 | |
eddc7ef5 JA |
4221 | if (ret < 0) |
4222 | req_set_fail_links(req); | |
e1e16097 | 4223 | io_req_complete(req, ret); |
eddc7ef5 JA |
4224 | return 0; |
4225 | } | |
4226 | ||
b5dba59e JA |
4227 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4228 | { | |
14587a46 | 4229 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4230 | return -EINVAL; |
b5dba59e JA |
4231 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4232 | sqe->rw_flags || sqe->buf_index) | |
4233 | return -EINVAL; | |
9c280f90 | 4234 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4235 | return -EBADF; |
b5dba59e JA |
4236 | |
4237 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4238 | return 0; |
b5dba59e JA |
4239 | } |
4240 | ||
889fca73 | 4241 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4242 | { |
9eac1904 | 4243 | struct files_struct *files = current->files; |
3af73b28 | 4244 | struct io_close *close = &req->close; |
9eac1904 JA |
4245 | struct fdtable *fdt; |
4246 | struct file *file; | |
b5dba59e JA |
4247 | int ret; |
4248 | ||
9eac1904 JA |
4249 | file = NULL; |
4250 | ret = -EBADF; | |
4251 | spin_lock(&files->file_lock); | |
4252 | fdt = files_fdtable(files); | |
4253 | if (close->fd >= fdt->max_fds) { | |
4254 | spin_unlock(&files->file_lock); | |
4255 | goto err; | |
4256 | } | |
4257 | file = fdt->fd[close->fd]; | |
4258 | if (!file) { | |
4259 | spin_unlock(&files->file_lock); | |
4260 | goto err; | |
4261 | } | |
4262 | ||
4263 | if (file->f_op == &io_uring_fops) { | |
4264 | spin_unlock(&files->file_lock); | |
4265 | file = NULL; | |
4266 | goto err; | |
3af73b28 | 4267 | } |
b5dba59e JA |
4268 | |
4269 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4270 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4271 | spin_unlock(&files->file_lock); |
0bf0eefd | 4272 | return -EAGAIN; |
a2100672 | 4273 | } |
b5dba59e | 4274 | |
9eac1904 JA |
4275 | ret = __close_fd_get_file(close->fd, &file); |
4276 | spin_unlock(&files->file_lock); | |
4277 | if (ret < 0) { | |
4278 | if (ret == -ENOENT) | |
4279 | ret = -EBADF; | |
4280 | goto err; | |
4281 | } | |
4282 | ||
3af73b28 | 4283 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4284 | ret = filp_close(file, current->files); |
4285 | err: | |
3af73b28 PB |
4286 | if (ret < 0) |
4287 | req_set_fail_links(req); | |
9eac1904 JA |
4288 | if (file) |
4289 | fput(file); | |
889fca73 | 4290 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4291 | return 0; |
b5dba59e JA |
4292 | } |
4293 | ||
1155c76a | 4294 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4295 | { |
4296 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4297 | |
5d17b4a4 JA |
4298 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4299 | return -EINVAL; | |
4300 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4301 | return -EINVAL; | |
4302 | ||
8ed8d3c3 JA |
4303 | req->sync.off = READ_ONCE(sqe->off); |
4304 | req->sync.len = READ_ONCE(sqe->len); | |
4305 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4306 | return 0; |
4307 | } | |
4308 | ||
45d189c6 | 4309 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4310 | { |
8ed8d3c3 JA |
4311 | int ret; |
4312 | ||
ac45abc0 | 4313 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4314 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4315 | return -EAGAIN; |
4316 | ||
9adbd45d | 4317 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4318 | req->sync.flags); |
4319 | if (ret < 0) | |
4320 | req_set_fail_links(req); | |
e1e16097 | 4321 | io_req_complete(req, ret); |
5d17b4a4 JA |
4322 | return 0; |
4323 | } | |
4324 | ||
469956e8 | 4325 | #if defined(CONFIG_NET) |
02d27d89 PB |
4326 | static int io_setup_async_msg(struct io_kiocb *req, |
4327 | struct io_async_msghdr *kmsg) | |
4328 | { | |
e8c2bc1f JA |
4329 | struct io_async_msghdr *async_msg = req->async_data; |
4330 | ||
4331 | if (async_msg) | |
02d27d89 | 4332 | return -EAGAIN; |
e8c2bc1f | 4333 | if (io_alloc_async_data(req)) { |
257e84a5 | 4334 | kfree(kmsg->free_iov); |
02d27d89 PB |
4335 | return -ENOMEM; |
4336 | } | |
e8c2bc1f | 4337 | async_msg = req->async_data; |
02d27d89 | 4338 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4339 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4340 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4341 | /* if were using fast_iov, set it to the new one */ |
4342 | if (!async_msg->free_iov) | |
4343 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4344 | ||
02d27d89 PB |
4345 | return -EAGAIN; |
4346 | } | |
4347 | ||
2ae523ed PB |
4348 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4349 | struct io_async_msghdr *iomsg) | |
4350 | { | |
2ae523ed | 4351 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4352 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4353 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4354 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4355 | } |
4356 | ||
93642ef8 PB |
4357 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4358 | { | |
4359 | int ret; | |
4360 | ||
4361 | if (!io_op_defs[req->opcode].needs_async_data) | |
4362 | return 0; | |
4363 | ret = io_sendmsg_copy_hdr(req, req->async_data); | |
4364 | if (!ret) | |
4365 | req->flags |= REQ_F_NEED_CLEANUP; | |
4366 | return ret; | |
4367 | } | |
4368 | ||
3529d8c2 | 4369 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4370 | { |
e47293fd | 4371 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4372 | |
d2b6f48b PB |
4373 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4374 | return -EINVAL; | |
4375 | ||
e47293fd | 4376 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4377 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4378 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4379 | |
d8768362 JA |
4380 | #ifdef CONFIG_COMPAT |
4381 | if (req->ctx->compat) | |
4382 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4383 | #endif | |
93642ef8 | 4384 | return 0; |
03b1230c JA |
4385 | } |
4386 | ||
889fca73 | 4387 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4388 | { |
6b754c8b | 4389 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4390 | struct socket *sock; |
7a7cacba | 4391 | unsigned flags; |
0031275d | 4392 | int min_ret = 0; |
0fa03c62 JA |
4393 | int ret; |
4394 | ||
dba4a925 | 4395 | sock = sock_from_file(req->file); |
7a7cacba | 4396 | if (unlikely(!sock)) |
dba4a925 | 4397 | return -ENOTSOCK; |
3529d8c2 | 4398 | |
257e84a5 PB |
4399 | kmsg = req->async_data; |
4400 | if (!kmsg) { | |
7a7cacba PB |
4401 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4402 | if (ret) | |
4403 | return ret; | |
4404 | kmsg = &iomsg; | |
0fa03c62 | 4405 | } |
0fa03c62 | 4406 | |
76cd979f | 4407 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4408 | if (flags & MSG_DONTWAIT) |
4409 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4410 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4411 | flags |= MSG_DONTWAIT; |
e47293fd | 4412 | |
0031275d SM |
4413 | if (flags & MSG_WAITALL) |
4414 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4415 | ||
7a7cacba | 4416 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4417 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4418 | return io_setup_async_msg(req, kmsg); |
4419 | if (ret == -ERESTARTSYS) | |
4420 | ret = -EINTR; | |
0fa03c62 | 4421 | |
257e84a5 PB |
4422 | /* fast path, check for non-NULL to avoid function call */ |
4423 | if (kmsg->free_iov) | |
4424 | kfree(kmsg->free_iov); | |
99bc4c38 | 4425 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4426 | if (ret < min_ret) |
4e88d6e7 | 4427 | req_set_fail_links(req); |
889fca73 | 4428 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4429 | return 0; |
03b1230c | 4430 | } |
aa1fa28f | 4431 | |
889fca73 | 4432 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4433 | { |
7a7cacba PB |
4434 | struct io_sr_msg *sr = &req->sr_msg; |
4435 | struct msghdr msg; | |
4436 | struct iovec iov; | |
fddaface | 4437 | struct socket *sock; |
7a7cacba | 4438 | unsigned flags; |
0031275d | 4439 | int min_ret = 0; |
fddaface JA |
4440 | int ret; |
4441 | ||
dba4a925 | 4442 | sock = sock_from_file(req->file); |
7a7cacba | 4443 | if (unlikely(!sock)) |
dba4a925 | 4444 | return -ENOTSOCK; |
fddaface | 4445 | |
7a7cacba PB |
4446 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4447 | if (unlikely(ret)) | |
14db8411 | 4448 | return ret; |
fddaface | 4449 | |
7a7cacba PB |
4450 | msg.msg_name = NULL; |
4451 | msg.msg_control = NULL; | |
4452 | msg.msg_controllen = 0; | |
4453 | msg.msg_namelen = 0; | |
fddaface | 4454 | |
76cd979f | 4455 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4456 | if (flags & MSG_DONTWAIT) |
4457 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4458 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4459 | flags |= MSG_DONTWAIT; |
fddaface | 4460 | |
0031275d SM |
4461 | if (flags & MSG_WAITALL) |
4462 | min_ret = iov_iter_count(&msg.msg_iter); | |
4463 | ||
7a7cacba PB |
4464 | msg.msg_flags = flags; |
4465 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4466 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4467 | return -EAGAIN; |
4468 | if (ret == -ERESTARTSYS) | |
4469 | ret = -EINTR; | |
fddaface | 4470 | |
0031275d | 4471 | if (ret < min_ret) |
fddaface | 4472 | req_set_fail_links(req); |
889fca73 | 4473 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4474 | return 0; |
fddaface JA |
4475 | } |
4476 | ||
1400e697 PB |
4477 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4478 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4479 | { |
4480 | struct io_sr_msg *sr = &req->sr_msg; | |
4481 | struct iovec __user *uiov; | |
4482 | size_t iov_len; | |
4483 | int ret; | |
4484 | ||
1400e697 PB |
4485 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4486 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4487 | if (ret) |
4488 | return ret; | |
4489 | ||
4490 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4491 | if (iov_len > 1) | |
4492 | return -EINVAL; | |
5476dfed | 4493 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4494 | return -EFAULT; |
5476dfed | 4495 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4496 | iomsg->free_iov = NULL; |
52de1fe1 | 4497 | } else { |
257e84a5 | 4498 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4499 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4500 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4501 | false); |
52de1fe1 JA |
4502 | if (ret > 0) |
4503 | ret = 0; | |
4504 | } | |
4505 | ||
4506 | return ret; | |
4507 | } | |
4508 | ||
4509 | #ifdef CONFIG_COMPAT | |
4510 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4511 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4512 | { |
4513 | struct compat_msghdr __user *msg_compat; | |
4514 | struct io_sr_msg *sr = &req->sr_msg; | |
4515 | struct compat_iovec __user *uiov; | |
4516 | compat_uptr_t ptr; | |
4517 | compat_size_t len; | |
4518 | int ret; | |
4519 | ||
270a5940 | 4520 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4521 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4522 | &ptr, &len); |
4523 | if (ret) | |
4524 | return ret; | |
4525 | ||
4526 | uiov = compat_ptr(ptr); | |
4527 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4528 | compat_ssize_t clen; | |
4529 | ||
4530 | if (len > 1) | |
4531 | return -EINVAL; | |
4532 | if (!access_ok(uiov, sizeof(*uiov))) | |
4533 | return -EFAULT; | |
4534 | if (__get_user(clen, &uiov->iov_len)) | |
4535 | return -EFAULT; | |
4536 | if (clen < 0) | |
4537 | return -EINVAL; | |
2d280bc8 | 4538 | sr->len = clen; |
257e84a5 | 4539 | iomsg->free_iov = NULL; |
52de1fe1 | 4540 | } else { |
257e84a5 | 4541 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4542 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4543 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4544 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4545 | if (ret < 0) |
4546 | return ret; | |
4547 | } | |
4548 | ||
4549 | return 0; | |
4550 | } | |
4551 | #endif | |
4552 | ||
1400e697 PB |
4553 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4554 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4555 | { |
1400e697 | 4556 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4557 | |
4558 | #ifdef CONFIG_COMPAT | |
4559 | if (req->ctx->compat) | |
1400e697 | 4560 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4561 | #endif |
52de1fe1 | 4562 | |
1400e697 | 4563 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4564 | } |
4565 | ||
bcda7baa | 4566 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4567 | bool needs_lock) |
bcda7baa JA |
4568 | { |
4569 | struct io_sr_msg *sr = &req->sr_msg; | |
4570 | struct io_buffer *kbuf; | |
4571 | ||
bcda7baa JA |
4572 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4573 | if (IS_ERR(kbuf)) | |
4574 | return kbuf; | |
4575 | ||
4576 | sr->kbuf = kbuf; | |
4577 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4578 | return kbuf; |
fddaface JA |
4579 | } |
4580 | ||
7fbb1b54 PB |
4581 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4582 | { | |
4583 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4584 | } | |
4585 | ||
93642ef8 | 4586 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4587 | { |
99bc4c38 | 4588 | int ret; |
3529d8c2 | 4589 | |
93642ef8 PB |
4590 | if (!io_op_defs[req->opcode].needs_async_data) |
4591 | return 0; | |
4592 | ret = io_recvmsg_copy_hdr(req, req->async_data); | |
4593 | if (!ret) | |
4594 | req->flags |= REQ_F_NEED_CLEANUP; | |
4595 | return ret; | |
4596 | } | |
4597 | ||
4598 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4599 | { | |
4600 | struct io_sr_msg *sr = &req->sr_msg; | |
4601 | ||
d2b6f48b PB |
4602 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4603 | return -EINVAL; | |
4604 | ||
3529d8c2 | 4605 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4606 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4607 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4608 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4609 | |
d8768362 JA |
4610 | #ifdef CONFIG_COMPAT |
4611 | if (req->ctx->compat) | |
4612 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4613 | #endif | |
93642ef8 | 4614 | return 0; |
aa1fa28f JA |
4615 | } |
4616 | ||
889fca73 | 4617 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4618 | { |
6b754c8b | 4619 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4620 | struct socket *sock; |
7fbb1b54 | 4621 | struct io_buffer *kbuf; |
7a7cacba | 4622 | unsigned flags; |
0031275d | 4623 | int min_ret = 0; |
52de1fe1 | 4624 | int ret, cflags = 0; |
45d189c6 | 4625 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4626 | |
dba4a925 | 4627 | sock = sock_from_file(req->file); |
7a7cacba | 4628 | if (unlikely(!sock)) |
dba4a925 | 4629 | return -ENOTSOCK; |
3529d8c2 | 4630 | |
257e84a5 PB |
4631 | kmsg = req->async_data; |
4632 | if (!kmsg) { | |
7a7cacba PB |
4633 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4634 | if (ret) | |
681fda8d | 4635 | return ret; |
7a7cacba PB |
4636 | kmsg = &iomsg; |
4637 | } | |
03b1230c | 4638 | |
bc02ef33 | 4639 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4640 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4641 | if (IS_ERR(kbuf)) |
52de1fe1 | 4642 | return PTR_ERR(kbuf); |
7a7cacba | 4643 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4644 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4645 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4646 | 1, req->sr_msg.len); |
4647 | } | |
52de1fe1 | 4648 | |
76cd979f | 4649 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4650 | if (flags & MSG_DONTWAIT) |
4651 | req->flags |= REQ_F_NOWAIT; | |
4652 | else if (force_nonblock) | |
4653 | flags |= MSG_DONTWAIT; | |
e47293fd | 4654 | |
0031275d SM |
4655 | if (flags & MSG_WAITALL) |
4656 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4657 | ||
7a7cacba PB |
4658 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4659 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4660 | if (force_nonblock && ret == -EAGAIN) |
4661 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4662 | if (ret == -ERESTARTSYS) |
4663 | ret = -EINTR; | |
03b1230c | 4664 | |
7fbb1b54 PB |
4665 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4666 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4667 | /* fast path, check for non-NULL to avoid function call */ |
4668 | if (kmsg->free_iov) | |
4669 | kfree(kmsg->free_iov); | |
99bc4c38 | 4670 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4671 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4672 | req_set_fail_links(req); |
889fca73 | 4673 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4674 | return 0; |
0fa03c62 | 4675 | } |
5d17b4a4 | 4676 | |
889fca73 | 4677 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4678 | { |
6b754c8b | 4679 | struct io_buffer *kbuf; |
7a7cacba PB |
4680 | struct io_sr_msg *sr = &req->sr_msg; |
4681 | struct msghdr msg; | |
4682 | void __user *buf = sr->buf; | |
fddaface | 4683 | struct socket *sock; |
7a7cacba PB |
4684 | struct iovec iov; |
4685 | unsigned flags; | |
0031275d | 4686 | int min_ret = 0; |
bcda7baa | 4687 | int ret, cflags = 0; |
45d189c6 | 4688 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4689 | |
dba4a925 | 4690 | sock = sock_from_file(req->file); |
7a7cacba | 4691 | if (unlikely(!sock)) |
dba4a925 | 4692 | return -ENOTSOCK; |
fddaface | 4693 | |
bc02ef33 | 4694 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4695 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4696 | if (IS_ERR(kbuf)) |
4697 | return PTR_ERR(kbuf); | |
7a7cacba | 4698 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4699 | } |
bcda7baa | 4700 | |
7a7cacba | 4701 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4702 | if (unlikely(ret)) |
4703 | goto out_free; | |
fddaface | 4704 | |
7a7cacba PB |
4705 | msg.msg_name = NULL; |
4706 | msg.msg_control = NULL; | |
4707 | msg.msg_controllen = 0; | |
4708 | msg.msg_namelen = 0; | |
4709 | msg.msg_iocb = NULL; | |
4710 | msg.msg_flags = 0; | |
fddaface | 4711 | |
76cd979f | 4712 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4713 | if (flags & MSG_DONTWAIT) |
4714 | req->flags |= REQ_F_NOWAIT; | |
4715 | else if (force_nonblock) | |
4716 | flags |= MSG_DONTWAIT; | |
4717 | ||
0031275d SM |
4718 | if (flags & MSG_WAITALL) |
4719 | min_ret = iov_iter_count(&msg.msg_iter); | |
4720 | ||
7a7cacba PB |
4721 | ret = sock_recvmsg(sock, &msg, flags); |
4722 | if (force_nonblock && ret == -EAGAIN) | |
4723 | return -EAGAIN; | |
4724 | if (ret == -ERESTARTSYS) | |
4725 | ret = -EINTR; | |
14c32eee | 4726 | out_free: |
7fbb1b54 PB |
4727 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4728 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4729 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4730 | req_set_fail_links(req); |
889fca73 | 4731 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4732 | return 0; |
fddaface JA |
4733 | } |
4734 | ||
3529d8c2 | 4735 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4736 | { |
8ed8d3c3 JA |
4737 | struct io_accept *accept = &req->accept; |
4738 | ||
14587a46 | 4739 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4740 | return -EINVAL; |
8042d6ce | 4741 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4742 | return -EINVAL; |
4743 | ||
d55e5f5b JA |
4744 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4745 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4746 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4747 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4748 | return 0; |
8ed8d3c3 | 4749 | } |
17f2fe35 | 4750 | |
889fca73 | 4751 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4752 | { |
4753 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4754 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4755 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4756 | int ret; |
4757 | ||
e697deed JX |
4758 | if (req->file->f_flags & O_NONBLOCK) |
4759 | req->flags |= REQ_F_NOWAIT; | |
4760 | ||
8ed8d3c3 | 4761 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4762 | accept->addr_len, accept->flags, |
4763 | accept->nofile); | |
8ed8d3c3 | 4764 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4765 | return -EAGAIN; |
ac45abc0 PB |
4766 | if (ret < 0) { |
4767 | if (ret == -ERESTARTSYS) | |
4768 | ret = -EINTR; | |
4e88d6e7 | 4769 | req_set_fail_links(req); |
ac45abc0 | 4770 | } |
889fca73 | 4771 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4772 | return 0; |
8ed8d3c3 JA |
4773 | } |
4774 | ||
93642ef8 PB |
4775 | static int io_connect_prep_async(struct io_kiocb *req) |
4776 | { | |
4777 | struct io_async_connect *io = req->async_data; | |
4778 | struct io_connect *conn = &req->connect; | |
4779 | ||
4780 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4781 | } | |
4782 | ||
3529d8c2 | 4783 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4784 | { |
3529d8c2 | 4785 | struct io_connect *conn = &req->connect; |
f499a021 | 4786 | |
14587a46 | 4787 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4788 | return -EINVAL; |
4789 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4790 | return -EINVAL; | |
4791 | ||
3529d8c2 JA |
4792 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4793 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4794 | return 0; |
f499a021 JA |
4795 | } |
4796 | ||
889fca73 | 4797 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4798 | { |
e8c2bc1f | 4799 | struct io_async_connect __io, *io; |
f8e85cf2 | 4800 | unsigned file_flags; |
3fbb51c1 | 4801 | int ret; |
45d189c6 | 4802 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4803 | |
e8c2bc1f JA |
4804 | if (req->async_data) { |
4805 | io = req->async_data; | |
f499a021 | 4806 | } else { |
3529d8c2 JA |
4807 | ret = move_addr_to_kernel(req->connect.addr, |
4808 | req->connect.addr_len, | |
e8c2bc1f | 4809 | &__io.address); |
f499a021 JA |
4810 | if (ret) |
4811 | goto out; | |
4812 | io = &__io; | |
4813 | } | |
4814 | ||
3fbb51c1 JA |
4815 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4816 | ||
e8c2bc1f | 4817 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4818 | req->connect.addr_len, file_flags); |
87f80d62 | 4819 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4820 | if (req->async_data) |
b7bb4f7d | 4821 | return -EAGAIN; |
e8c2bc1f | 4822 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4823 | ret = -ENOMEM; |
4824 | goto out; | |
4825 | } | |
e8c2bc1f | 4826 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4827 | return -EAGAIN; |
f499a021 | 4828 | } |
f8e85cf2 JA |
4829 | if (ret == -ERESTARTSYS) |
4830 | ret = -EINTR; | |
f499a021 | 4831 | out: |
4e88d6e7 JA |
4832 | if (ret < 0) |
4833 | req_set_fail_links(req); | |
889fca73 | 4834 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4835 | return 0; |
469956e8 Y |
4836 | } |
4837 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4838 | #define IO_NETOP_FN(op) \ |
4839 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4840 | { \ | |
4841 | return -EOPNOTSUPP; \ | |
4842 | } | |
4843 | ||
4844 | #define IO_NETOP_PREP(op) \ | |
4845 | IO_NETOP_FN(op) \ | |
4846 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4847 | { \ | |
4848 | return -EOPNOTSUPP; \ | |
4849 | } \ | |
4850 | ||
4851 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4852 | IO_NETOP_PREP(op) \ | |
4853 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4854 | { \ | |
4855 | return -EOPNOTSUPP; \ | |
4856 | } | |
4857 | ||
4858 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4859 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4860 | IO_NETOP_PREP_ASYNC(connect); | |
4861 | IO_NETOP_PREP(accept); | |
4862 | IO_NETOP_FN(send); | |
4863 | IO_NETOP_FN(recv); | |
469956e8 | 4864 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4865 | |
d7718a9d JA |
4866 | struct io_poll_table { |
4867 | struct poll_table_struct pt; | |
4868 | struct io_kiocb *req; | |
4869 | int error; | |
4870 | }; | |
ce593a6c | 4871 | |
d7718a9d JA |
4872 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4873 | __poll_t mask, task_work_func_t func) | |
4874 | { | |
aa96bf8a | 4875 | int ret; |
d7718a9d JA |
4876 | |
4877 | /* for instances that support it check for an event match first: */ | |
4878 | if (mask && !(mask & poll->events)) | |
4879 | return 0; | |
4880 | ||
4881 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4882 | ||
4883 | list_del_init(&poll->wait.entry); | |
4884 | ||
d7718a9d | 4885 | req->result = mask; |
7cbf1722 | 4886 | req->task_work.func = func; |
6d816e08 JA |
4887 | percpu_ref_get(&req->ctx->refs); |
4888 | ||
d7718a9d | 4889 | /* |
e3aabf95 JA |
4890 | * If this fails, then the task is exiting. When a task exits, the |
4891 | * work gets canceled, so just cancel this request as well instead | |
4892 | * of executing it. We can't safely execute it anyway, as we may not | |
4893 | * have the needed state needed for it anyway. | |
d7718a9d | 4894 | */ |
355fb9e2 | 4895 | ret = io_req_task_work_add(req); |
aa96bf8a | 4896 | if (unlikely(ret)) { |
e3aabf95 | 4897 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4898 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4899 | } |
d7718a9d JA |
4900 | return 1; |
4901 | } | |
4902 | ||
74ce6ce4 JA |
4903 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4904 | __acquires(&req->ctx->completion_lock) | |
4905 | { | |
4906 | struct io_ring_ctx *ctx = req->ctx; | |
4907 | ||
4908 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4909 | struct poll_table_struct pt = { ._key = poll->events }; | |
4910 | ||
4911 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4912 | } | |
4913 | ||
4914 | spin_lock_irq(&ctx->completion_lock); | |
4915 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4916 | add_wait_queue(poll->head, &poll->wait); | |
4917 | return true; | |
4918 | } | |
4919 | ||
4920 | return false; | |
4921 | } | |
4922 | ||
d4e7cd36 | 4923 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4924 | { |
e8c2bc1f | 4925 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4926 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4927 | return req->async_data; |
d4e7cd36 JA |
4928 | return req->apoll->double_poll; |
4929 | } | |
4930 | ||
4931 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4932 | { | |
4933 | if (req->opcode == IORING_OP_POLL_ADD) | |
4934 | return &req->poll; | |
4935 | return &req->apoll->poll; | |
4936 | } | |
4937 | ||
4938 | static void io_poll_remove_double(struct io_kiocb *req) | |
4939 | { | |
4940 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4941 | |
4942 | lockdep_assert_held(&req->ctx->completion_lock); | |
4943 | ||
4944 | if (poll && poll->head) { | |
4945 | struct wait_queue_head *head = poll->head; | |
4946 | ||
4947 | spin_lock(&head->lock); | |
4948 | list_del_init(&poll->wait.entry); | |
4949 | if (poll->wait.private) | |
4950 | refcount_dec(&req->refs); | |
4951 | poll->head = NULL; | |
4952 | spin_unlock(&head->lock); | |
4953 | } | |
4954 | } | |
4955 | ||
4956 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4957 | { | |
4958 | struct io_ring_ctx *ctx = req->ctx; | |
4959 | ||
d4e7cd36 | 4960 | io_poll_remove_double(req); |
18bceab1 JA |
4961 | req->poll.done = true; |
4962 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4963 | io_commit_cqring(ctx); | |
4964 | } | |
4965 | ||
dd221f46 | 4966 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4967 | { |
dd221f46 | 4968 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4969 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4970 | struct io_kiocb *nxt; |
18bceab1 JA |
4971 | |
4972 | if (io_poll_rewait(req, &req->poll)) { | |
4973 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4974 | } else { |
4975 | hash_del(&req->hash_node); | |
4976 | io_poll_complete(req, req->result, 0); | |
4977 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4978 | |
dd221f46 PB |
4979 | nxt = io_put_req_find_next(req); |
4980 | io_cqring_ev_posted(ctx); | |
4981 | if (nxt) | |
4982 | __io_req_task_submit(nxt); | |
4983 | } | |
18bceab1 | 4984 | |
6d816e08 | 4985 | percpu_ref_put(&ctx->refs); |
18bceab1 JA |
4986 | } |
4987 | ||
4988 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4989 | int sync, void *key) | |
4990 | { | |
4991 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4992 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4993 | __poll_t mask = key_to_poll(key); |
4994 | ||
4995 | /* for instances that support it check for an event match first: */ | |
4996 | if (mask && !(mask & poll->events)) | |
4997 | return 0; | |
4998 | ||
8706e04e JA |
4999 | list_del_init(&wait->entry); |
5000 | ||
807abcb0 | 5001 | if (poll && poll->head) { |
18bceab1 JA |
5002 | bool done; |
5003 | ||
807abcb0 JA |
5004 | spin_lock(&poll->head->lock); |
5005 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5006 | if (!done) |
807abcb0 | 5007 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5008 | /* make sure double remove sees this as being gone */ |
5009 | wait->private = NULL; | |
807abcb0 | 5010 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5011 | if (!done) { |
5012 | /* use wait func handler, so it matches the rq type */ | |
5013 | poll->wait.func(&poll->wait, mode, sync, key); | |
5014 | } | |
18bceab1 JA |
5015 | } |
5016 | refcount_dec(&req->refs); | |
5017 | return 1; | |
5018 | } | |
5019 | ||
5020 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5021 | wait_queue_func_t wake_func) | |
5022 | { | |
5023 | poll->head = NULL; | |
5024 | poll->done = false; | |
5025 | poll->canceled = false; | |
5026 | poll->events = events; | |
5027 | INIT_LIST_HEAD(&poll->wait.entry); | |
5028 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5029 | } | |
5030 | ||
5031 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5032 | struct wait_queue_head *head, |
5033 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5034 | { |
5035 | struct io_kiocb *req = pt->req; | |
5036 | ||
5037 | /* | |
5038 | * If poll->head is already set, it's because the file being polled | |
5039 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5040 | * for write). Setup a separate io_poll_iocb if this happens. | |
5041 | */ | |
5042 | if (unlikely(poll->head)) { | |
58852d4d PB |
5043 | struct io_poll_iocb *poll_one = poll; |
5044 | ||
18bceab1 | 5045 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5046 | if (*poll_ptr) { |
18bceab1 JA |
5047 | pt->error = -EINVAL; |
5048 | return; | |
5049 | } | |
1c3b3e65 JA |
5050 | /* double add on the same waitqueue head, ignore */ |
5051 | if (poll->head == head) | |
5052 | return; | |
18bceab1 JA |
5053 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5054 | if (!poll) { | |
5055 | pt->error = -ENOMEM; | |
5056 | return; | |
5057 | } | |
58852d4d | 5058 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
18bceab1 JA |
5059 | refcount_inc(&req->refs); |
5060 | poll->wait.private = req; | |
807abcb0 | 5061 | *poll_ptr = poll; |
18bceab1 JA |
5062 | } |
5063 | ||
5064 | pt->error = 0; | |
5065 | poll->head = head; | |
a31eb4a2 JX |
5066 | |
5067 | if (poll->events & EPOLLEXCLUSIVE) | |
5068 | add_wait_queue_exclusive(head, &poll->wait); | |
5069 | else | |
5070 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5071 | } |
5072 | ||
5073 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5074 | struct poll_table_struct *p) | |
5075 | { | |
5076 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5077 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5078 | |
807abcb0 | 5079 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5080 | } |
5081 | ||
d7718a9d JA |
5082 | static void io_async_task_func(struct callback_head *cb) |
5083 | { | |
5084 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5085 | struct async_poll *apoll = req->apoll; | |
5086 | struct io_ring_ctx *ctx = req->ctx; | |
5087 | ||
5088 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5089 | ||
74ce6ce4 | 5090 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5091 | spin_unlock_irq(&ctx->completion_lock); |
6d816e08 | 5092 | percpu_ref_put(&ctx->refs); |
74ce6ce4 | 5093 | return; |
d7718a9d JA |
5094 | } |
5095 | ||
31067255 | 5096 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5097 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5098 | hash_del(&req->hash_node); |
2bae047e | 5099 | |
d4e7cd36 | 5100 | io_poll_remove_double(req); |
74ce6ce4 JA |
5101 | spin_unlock_irq(&ctx->completion_lock); |
5102 | ||
0be0b0e3 PB |
5103 | if (!READ_ONCE(apoll->poll.canceled)) |
5104 | __io_req_task_submit(req); | |
5105 | else | |
5106 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5107 | |
6d816e08 | 5108 | percpu_ref_put(&ctx->refs); |
807abcb0 | 5109 | kfree(apoll->double_poll); |
31067255 | 5110 | kfree(apoll); |
d7718a9d JA |
5111 | } |
5112 | ||
5113 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5114 | void *key) | |
5115 | { | |
5116 | struct io_kiocb *req = wait->private; | |
5117 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5118 | ||
5119 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5120 | key_to_poll(key)); | |
5121 | ||
5122 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5123 | } | |
5124 | ||
5125 | static void io_poll_req_insert(struct io_kiocb *req) | |
5126 | { | |
5127 | struct io_ring_ctx *ctx = req->ctx; | |
5128 | struct hlist_head *list; | |
5129 | ||
5130 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5131 | hlist_add_head(&req->hash_node, list); | |
5132 | } | |
5133 | ||
5134 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5135 | struct io_poll_iocb *poll, | |
5136 | struct io_poll_table *ipt, __poll_t mask, | |
5137 | wait_queue_func_t wake_func) | |
5138 | __acquires(&ctx->completion_lock) | |
5139 | { | |
5140 | struct io_ring_ctx *ctx = req->ctx; | |
5141 | bool cancel = false; | |
5142 | ||
4d52f338 | 5143 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5144 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5145 | poll->file = req->file; |
18bceab1 | 5146 | poll->wait.private = req; |
d7718a9d JA |
5147 | |
5148 | ipt->pt._key = mask; | |
5149 | ipt->req = req; | |
5150 | ipt->error = -EINVAL; | |
5151 | ||
d7718a9d JA |
5152 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5153 | ||
5154 | spin_lock_irq(&ctx->completion_lock); | |
5155 | if (likely(poll->head)) { | |
5156 | spin_lock(&poll->head->lock); | |
5157 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5158 | if (ipt->error) | |
5159 | cancel = true; | |
5160 | ipt->error = 0; | |
5161 | mask = 0; | |
5162 | } | |
5163 | if (mask || ipt->error) | |
5164 | list_del_init(&poll->wait.entry); | |
5165 | else if (cancel) | |
5166 | WRITE_ONCE(poll->canceled, true); | |
5167 | else if (!poll->done) /* actually waiting for an event */ | |
5168 | io_poll_req_insert(req); | |
5169 | spin_unlock(&poll->head->lock); | |
5170 | } | |
5171 | ||
5172 | return mask; | |
5173 | } | |
5174 | ||
5175 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5176 | { | |
5177 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5178 | struct io_ring_ctx *ctx = req->ctx; | |
5179 | struct async_poll *apoll; | |
5180 | struct io_poll_table ipt; | |
5181 | __poll_t mask, ret; | |
9dab14b8 | 5182 | int rw; |
d7718a9d JA |
5183 | |
5184 | if (!req->file || !file_can_poll(req->file)) | |
5185 | return false; | |
24c74678 | 5186 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5187 | return false; |
9dab14b8 JA |
5188 | if (def->pollin) |
5189 | rw = READ; | |
5190 | else if (def->pollout) | |
5191 | rw = WRITE; | |
5192 | else | |
5193 | return false; | |
5194 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
5195 | if (!io_file_supports_async(req->file, rw)) | |
d7718a9d JA |
5196 | return false; |
5197 | ||
5198 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5199 | if (unlikely(!apoll)) | |
5200 | return false; | |
807abcb0 | 5201 | apoll->double_poll = NULL; |
d7718a9d JA |
5202 | |
5203 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5204 | req->apoll = apoll; |
d7718a9d | 5205 | |
8755d97a | 5206 | mask = 0; |
d7718a9d | 5207 | if (def->pollin) |
8755d97a | 5208 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5209 | if (def->pollout) |
5210 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5211 | |
5212 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5213 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5214 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5215 | mask &= ~POLLIN; | |
5216 | ||
d7718a9d JA |
5217 | mask |= POLLERR | POLLPRI; |
5218 | ||
5219 | ipt.pt._qproc = io_async_queue_proc; | |
5220 | ||
5221 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5222 | io_async_wake); | |
a36da65c | 5223 | if (ret || ipt.error) { |
d4e7cd36 | 5224 | io_poll_remove_double(req); |
d7718a9d | 5225 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5226 | kfree(apoll->double_poll); |
d7718a9d JA |
5227 | kfree(apoll); |
5228 | return false; | |
5229 | } | |
5230 | spin_unlock_irq(&ctx->completion_lock); | |
5231 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5232 | apoll->poll.events); | |
5233 | return true; | |
5234 | } | |
5235 | ||
5236 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5237 | struct io_poll_iocb *poll) | |
221c5eb2 | 5238 | { |
b41e9852 | 5239 | bool do_complete = false; |
221c5eb2 JA |
5240 | |
5241 | spin_lock(&poll->head->lock); | |
5242 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5243 | if (!list_empty(&poll->wait.entry)) { |
5244 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5245 | do_complete = true; |
221c5eb2 JA |
5246 | } |
5247 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5248 | hash_del(&req->hash_node); |
d7718a9d JA |
5249 | return do_complete; |
5250 | } | |
5251 | ||
5252 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5253 | { | |
5254 | bool do_complete; | |
5255 | ||
d4e7cd36 JA |
5256 | io_poll_remove_double(req); |
5257 | ||
d7718a9d JA |
5258 | if (req->opcode == IORING_OP_POLL_ADD) { |
5259 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5260 | } else { | |
3bfa5bcb JA |
5261 | struct async_poll *apoll = req->apoll; |
5262 | ||
d7718a9d | 5263 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5264 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5265 | if (do_complete) { | |
d7718a9d | 5266 | io_put_req(req); |
807abcb0 | 5267 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5268 | kfree(apoll); |
5269 | } | |
b1f573bd XW |
5270 | } |
5271 | ||
b41e9852 JA |
5272 | if (do_complete) { |
5273 | io_cqring_fill_event(req, -ECANCELED); | |
5274 | io_commit_cqring(req->ctx); | |
f254ac04 | 5275 | req_set_fail_links(req); |
216578e5 | 5276 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5277 | } |
5278 | ||
5279 | return do_complete; | |
221c5eb2 JA |
5280 | } |
5281 | ||
76e1b642 JA |
5282 | /* |
5283 | * Returns true if we found and killed one or more poll requests | |
5284 | */ | |
6b81928d PB |
5285 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5286 | struct files_struct *files) | |
221c5eb2 | 5287 | { |
78076bb6 | 5288 | struct hlist_node *tmp; |
221c5eb2 | 5289 | struct io_kiocb *req; |
8e2e1faf | 5290 | int posted = 0, i; |
221c5eb2 JA |
5291 | |
5292 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5293 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5294 | struct hlist_head *list; | |
5295 | ||
5296 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5297 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5298 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5299 | posted += io_poll_remove_one(req); |
5300 | } | |
221c5eb2 JA |
5301 | } |
5302 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5303 | |
8e2e1faf JA |
5304 | if (posted) |
5305 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5306 | |
5307 | return posted != 0; | |
221c5eb2 JA |
5308 | } |
5309 | ||
47f46768 JA |
5310 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5311 | { | |
78076bb6 | 5312 | struct hlist_head *list; |
47f46768 JA |
5313 | struct io_kiocb *req; |
5314 | ||
78076bb6 JA |
5315 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5316 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5317 | if (sqe_addr != req->user_data) |
5318 | continue; | |
5319 | if (io_poll_remove_one(req)) | |
eac406c6 | 5320 | return 0; |
b41e9852 | 5321 | return -EALREADY; |
47f46768 JA |
5322 | } |
5323 | ||
5324 | return -ENOENT; | |
5325 | } | |
5326 | ||
3529d8c2 JA |
5327 | static int io_poll_remove_prep(struct io_kiocb *req, |
5328 | const struct io_uring_sqe *sqe) | |
0969e783 | 5329 | { |
0969e783 JA |
5330 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5331 | return -EINVAL; | |
5332 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5333 | sqe->poll_events) | |
5334 | return -EINVAL; | |
5335 | ||
018043be | 5336 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5337 | return 0; |
5338 | } | |
5339 | ||
221c5eb2 JA |
5340 | /* |
5341 | * Find a running poll command that matches one specified in sqe->addr, | |
5342 | * and remove it if found. | |
5343 | */ | |
61e98203 | 5344 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5345 | { |
5346 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5347 | int ret; |
221c5eb2 | 5348 | |
221c5eb2 | 5349 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5350 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5351 | spin_unlock_irq(&ctx->completion_lock); |
5352 | ||
4e88d6e7 JA |
5353 | if (ret < 0) |
5354 | req_set_fail_links(req); | |
e1e16097 | 5355 | io_req_complete(req, ret); |
221c5eb2 JA |
5356 | return 0; |
5357 | } | |
5358 | ||
221c5eb2 JA |
5359 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5360 | void *key) | |
5361 | { | |
c2f2eb7d JA |
5362 | struct io_kiocb *req = wait->private; |
5363 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5364 | |
d7718a9d | 5365 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5366 | } |
5367 | ||
221c5eb2 JA |
5368 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5369 | struct poll_table_struct *p) | |
5370 | { | |
5371 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5372 | ||
e8c2bc1f | 5373 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5374 | } |
5375 | ||
3529d8c2 | 5376 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5377 | { |
5378 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5379 | u32 events; |
221c5eb2 JA |
5380 | |
5381 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5382 | return -EINVAL; | |
5383 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5384 | return -EINVAL; | |
5385 | ||
5769a351 JX |
5386 | events = READ_ONCE(sqe->poll32_events); |
5387 | #ifdef __BIG_ENDIAN | |
5388 | events = swahw32(events); | |
5389 | #endif | |
a31eb4a2 JX |
5390 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5391 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5392 | return 0; |
5393 | } | |
5394 | ||
61e98203 | 5395 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5396 | { |
5397 | struct io_poll_iocb *poll = &req->poll; | |
5398 | struct io_ring_ctx *ctx = req->ctx; | |
5399 | struct io_poll_table ipt; | |
0969e783 | 5400 | __poll_t mask; |
0969e783 | 5401 | |
d7718a9d | 5402 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5403 | |
d7718a9d JA |
5404 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5405 | io_poll_wake); | |
221c5eb2 | 5406 | |
8c838788 | 5407 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5408 | ipt.error = 0; |
b0dd8a41 | 5409 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5410 | } |
221c5eb2 JA |
5411 | spin_unlock_irq(&ctx->completion_lock); |
5412 | ||
8c838788 JA |
5413 | if (mask) { |
5414 | io_cqring_ev_posted(ctx); | |
014db007 | 5415 | io_put_req(req); |
221c5eb2 | 5416 | } |
8c838788 | 5417 | return ipt.error; |
221c5eb2 JA |
5418 | } |
5419 | ||
5262f567 JA |
5420 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5421 | { | |
ad8a48ac JA |
5422 | struct io_timeout_data *data = container_of(timer, |
5423 | struct io_timeout_data, timer); | |
5424 | struct io_kiocb *req = data->req; | |
5425 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5426 | unsigned long flags; |
5427 | ||
5262f567 | 5428 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5429 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5430 | atomic_set(&req->ctx->cq_timeouts, |
5431 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5432 | ||
78e19bbe | 5433 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5434 | io_commit_cqring(ctx); |
5435 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5436 | ||
5437 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5438 | req_set_fail_links(req); |
5262f567 JA |
5439 | io_put_req(req); |
5440 | return HRTIMER_NORESTART; | |
5441 | } | |
5442 | ||
fbd15848 PB |
5443 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5444 | __u64 user_data) | |
f254ac04 | 5445 | { |
fbd15848 | 5446 | struct io_timeout_data *io; |
47f46768 JA |
5447 | struct io_kiocb *req; |
5448 | int ret = -ENOENT; | |
f254ac04 | 5449 | |
135fcde8 | 5450 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5451 | if (user_data == req->user_data) { |
47f46768 JA |
5452 | ret = 0; |
5453 | break; | |
5454 | } | |
5455 | } | |
5456 | ||
5457 | if (ret == -ENOENT) | |
fbd15848 PB |
5458 | return ERR_PTR(ret); |
5459 | ||
5460 | io = req->async_data; | |
e8c2bc1f | 5461 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5462 | if (ret == -1) |
fbd15848 | 5463 | return ERR_PTR(-EALREADY); |
a71976f3 | 5464 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5465 | return req; |
5466 | } | |
47f46768 | 5467 | |
fbd15848 PB |
5468 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5469 | { | |
5470 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5471 | ||
5472 | if (IS_ERR(req)) | |
5473 | return PTR_ERR(req); | |
f254ac04 JA |
5474 | |
5475 | req_set_fail_links(req); | |
f254ac04 | 5476 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5477 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5478 | return 0; |
5479 | } | |
5480 | ||
9c8e11b3 PB |
5481 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5482 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5483 | { |
9c8e11b3 PB |
5484 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5485 | struct io_timeout_data *data; | |
47f46768 | 5486 | |
9c8e11b3 PB |
5487 | if (IS_ERR(req)) |
5488 | return PTR_ERR(req); | |
47f46768 | 5489 | |
9c8e11b3 PB |
5490 | req->timeout.off = 0; /* noseq */ |
5491 | data = req->async_data; | |
5492 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5493 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5494 | data->timer.function = io_timeout_fn; | |
5495 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5496 | return 0; | |
47f46768 JA |
5497 | } |
5498 | ||
3529d8c2 JA |
5499 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5500 | const struct io_uring_sqe *sqe) | |
b29472ee | 5501 | { |
9c8e11b3 PB |
5502 | struct io_timeout_rem *tr = &req->timeout_rem; |
5503 | ||
b29472ee JA |
5504 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5505 | return -EINVAL; | |
61710e43 DA |
5506 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5507 | return -EINVAL; | |
9c8e11b3 | 5508 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5509 | return -EINVAL; |
5510 | ||
9c8e11b3 PB |
5511 | tr->addr = READ_ONCE(sqe->addr); |
5512 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5513 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5514 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5515 | return -EINVAL; | |
5516 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5517 | return -EFAULT; | |
5518 | } else if (tr->flags) { | |
5519 | /* timeout removal doesn't support flags */ | |
b29472ee | 5520 | return -EINVAL; |
9c8e11b3 | 5521 | } |
b29472ee | 5522 | |
b29472ee JA |
5523 | return 0; |
5524 | } | |
5525 | ||
8662daec PB |
5526 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5527 | { | |
5528 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5529 | : HRTIMER_MODE_REL; | |
5530 | } | |
5531 | ||
11365043 JA |
5532 | /* |
5533 | * Remove or update an existing timeout command | |
5534 | */ | |
61e98203 | 5535 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5536 | { |
9c8e11b3 | 5537 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5538 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5539 | int ret; |
11365043 | 5540 | |
11365043 | 5541 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5542 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5543 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5544 | else |
5545 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5546 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5547 | |
47f46768 | 5548 | io_cqring_fill_event(req, ret); |
11365043 JA |
5549 | io_commit_cqring(ctx); |
5550 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5551 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5552 | if (ret < 0) |
5553 | req_set_fail_links(req); | |
ec9c02ad | 5554 | io_put_req(req); |
11365043 | 5555 | return 0; |
5262f567 JA |
5556 | } |
5557 | ||
3529d8c2 | 5558 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5559 | bool is_timeout_link) |
5262f567 | 5560 | { |
ad8a48ac | 5561 | struct io_timeout_data *data; |
a41525ab | 5562 | unsigned flags; |
56080b02 | 5563 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5564 | |
ad8a48ac | 5565 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5566 | return -EINVAL; |
ad8a48ac | 5567 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5568 | return -EINVAL; |
56080b02 | 5569 | if (off && is_timeout_link) |
2d28390a | 5570 | return -EINVAL; |
a41525ab JA |
5571 | flags = READ_ONCE(sqe->timeout_flags); |
5572 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5573 | return -EINVAL; |
bdf20073 | 5574 | |
bfe68a22 | 5575 | req->timeout.off = off; |
26a61679 | 5576 | |
e8c2bc1f | 5577 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5578 | return -ENOMEM; |
5579 | ||
e8c2bc1f | 5580 | data = req->async_data; |
ad8a48ac | 5581 | data->req = req; |
ad8a48ac JA |
5582 | |
5583 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5584 | return -EFAULT; |
5585 | ||
8662daec | 5586 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5587 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5588 | if (is_timeout_link) |
5589 | io_req_track_inflight(req); | |
ad8a48ac JA |
5590 | return 0; |
5591 | } | |
5592 | ||
61e98203 | 5593 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5594 | { |
ad8a48ac | 5595 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5596 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5597 | struct list_head *entry; |
bfe68a22 | 5598 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5599 | |
733f5c95 | 5600 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5601 | |
5262f567 JA |
5602 | /* |
5603 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5604 | * timeout event to be satisfied. If it isn't set, then this is |
5605 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5606 | */ |
8eb7e2d0 | 5607 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5608 | entry = ctx->timeout_list.prev; |
5609 | goto add; | |
5610 | } | |
5262f567 | 5611 | |
bfe68a22 PB |
5612 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5613 | req->timeout.target_seq = tail + off; | |
5262f567 | 5614 | |
f010505b MDG |
5615 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5616 | * This is safe because ->completion_lock is held, and submissions | |
5617 | * and completions are never mixed in the same ->completion_lock section. | |
5618 | */ | |
5619 | ctx->cq_last_tm_flush = tail; | |
5620 | ||
5262f567 JA |
5621 | /* |
5622 | * Insertion sort, ensuring the first entry in the list is always | |
5623 | * the one we need first. | |
5624 | */ | |
5262f567 | 5625 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5626 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5627 | timeout.list); | |
5262f567 | 5628 | |
8eb7e2d0 | 5629 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5630 | continue; |
bfe68a22 PB |
5631 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5632 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5633 | break; |
5634 | } | |
93bd25bb | 5635 | add: |
135fcde8 | 5636 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5637 | data->timer.function = io_timeout_fn; |
5638 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5639 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5640 | return 0; |
5641 | } | |
5262f567 | 5642 | |
f458dd84 PB |
5643 | struct io_cancel_data { |
5644 | struct io_ring_ctx *ctx; | |
5645 | u64 user_data; | |
5646 | }; | |
5647 | ||
62755e35 JA |
5648 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5649 | { | |
5650 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5651 | struct io_cancel_data *cd = data; |
62755e35 | 5652 | |
f458dd84 | 5653 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5654 | } |
5655 | ||
f458dd84 PB |
5656 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5657 | struct io_ring_ctx *ctx) | |
62755e35 | 5658 | { |
f458dd84 | 5659 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5660 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5661 | int ret = 0; |
5662 | ||
f458dd84 | 5663 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5664 | return -ENOENT; |
5665 | ||
f458dd84 | 5666 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5667 | switch (cancel_ret) { |
5668 | case IO_WQ_CANCEL_OK: | |
5669 | ret = 0; | |
5670 | break; | |
5671 | case IO_WQ_CANCEL_RUNNING: | |
5672 | ret = -EALREADY; | |
5673 | break; | |
5674 | case IO_WQ_CANCEL_NOTFOUND: | |
5675 | ret = -ENOENT; | |
5676 | break; | |
5677 | } | |
5678 | ||
e977d6d3 JA |
5679 | return ret; |
5680 | } | |
5681 | ||
47f46768 JA |
5682 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5683 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5684 | int success_ret) |
47f46768 JA |
5685 | { |
5686 | unsigned long flags; | |
5687 | int ret; | |
5688 | ||
f458dd84 | 5689 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5690 | if (ret != -ENOENT) { |
5691 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5692 | goto done; | |
5693 | } | |
5694 | ||
5695 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5696 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5697 | if (ret != -ENOENT) | |
5698 | goto done; | |
5699 | ret = io_poll_cancel(ctx, sqe_addr); | |
5700 | done: | |
b0dd8a41 JA |
5701 | if (!ret) |
5702 | ret = success_ret; | |
47f46768 JA |
5703 | io_cqring_fill_event(req, ret); |
5704 | io_commit_cqring(ctx); | |
5705 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5706 | io_cqring_ev_posted(ctx); | |
5707 | ||
4e88d6e7 JA |
5708 | if (ret < 0) |
5709 | req_set_fail_links(req); | |
014db007 | 5710 | io_put_req(req); |
47f46768 JA |
5711 | } |
5712 | ||
3529d8c2 JA |
5713 | static int io_async_cancel_prep(struct io_kiocb *req, |
5714 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5715 | { |
fbf23849 | 5716 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5717 | return -EINVAL; |
61710e43 DA |
5718 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5719 | return -EINVAL; | |
5720 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5721 | return -EINVAL; |
5722 | ||
fbf23849 JA |
5723 | req->cancel.addr = READ_ONCE(sqe->addr); |
5724 | return 0; | |
5725 | } | |
5726 | ||
61e98203 | 5727 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5728 | { |
5729 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5730 | u64 sqe_addr = req->cancel.addr; |
5731 | struct io_tctx_node *node; | |
5732 | int ret; | |
5733 | ||
5734 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5735 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5736 | spin_lock_irq(&ctx->completion_lock); | |
5737 | if (ret != -ENOENT) | |
5738 | goto done; | |
5739 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5740 | if (ret != -ENOENT) | |
5741 | goto done; | |
5742 | ret = io_poll_cancel(ctx, sqe_addr); | |
5743 | if (ret != -ENOENT) | |
5744 | goto done; | |
5745 | spin_unlock_irq(&ctx->completion_lock); | |
5746 | ||
5747 | /* slow path, try all io-wq's */ | |
5748 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5749 | ret = -ENOENT; | |
5750 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5751 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5752 | |
58f99373 PB |
5753 | if (!tctx || !tctx->io_wq) |
5754 | continue; | |
5755 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5756 | if (ret != -ENOENT) | |
5757 | break; | |
5758 | } | |
5759 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5760 | ||
5761 | spin_lock_irq(&ctx->completion_lock); | |
5762 | done: | |
5763 | io_cqring_fill_event(req, ret); | |
5764 | io_commit_cqring(ctx); | |
5765 | spin_unlock_irq(&ctx->completion_lock); | |
5766 | io_cqring_ev_posted(ctx); | |
5767 | ||
5768 | if (ret < 0) | |
5769 | req_set_fail_links(req); | |
5770 | io_put_req(req); | |
5262f567 JA |
5771 | return 0; |
5772 | } | |
5773 | ||
269bbe5f | 5774 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5775 | const struct io_uring_sqe *sqe) |
5776 | { | |
6ca56f84 JA |
5777 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5778 | return -EINVAL; | |
61710e43 DA |
5779 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5780 | return -EINVAL; | |
5781 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5782 | return -EINVAL; |
5783 | ||
269bbe5f BM |
5784 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5785 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5786 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5787 | return -EINVAL; |
269bbe5f | 5788 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5789 | return 0; |
5790 | } | |
5791 | ||
889fca73 | 5792 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5793 | { |
5794 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5795 | struct io_uring_rsrc_update up; |
05f3fb3c | 5796 | int ret; |
fbf23849 | 5797 | |
45d189c6 | 5798 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5799 | return -EAGAIN; |
05f3fb3c | 5800 | |
269bbe5f BM |
5801 | up.offset = req->rsrc_update.offset; |
5802 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5803 | |
5804 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5805 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5806 | mutex_unlock(&ctx->uring_lock); |
5807 | ||
5808 | if (ret < 0) | |
5809 | req_set_fail_links(req); | |
889fca73 | 5810 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5811 | return 0; |
5812 | } | |
5813 | ||
bfe76559 | 5814 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5815 | { |
d625c6ee | 5816 | switch (req->opcode) { |
e781573e | 5817 | case IORING_OP_NOP: |
bfe76559 | 5818 | return 0; |
f67676d1 JA |
5819 | case IORING_OP_READV: |
5820 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5821 | case IORING_OP_READ: |
bfe76559 | 5822 | return io_read_prep(req, sqe); |
f67676d1 JA |
5823 | case IORING_OP_WRITEV: |
5824 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5825 | case IORING_OP_WRITE: |
bfe76559 | 5826 | return io_write_prep(req, sqe); |
0969e783 | 5827 | case IORING_OP_POLL_ADD: |
bfe76559 | 5828 | return io_poll_add_prep(req, sqe); |
0969e783 | 5829 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5830 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5831 | case IORING_OP_FSYNC: |
1155c76a | 5832 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5833 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5834 | return io_sfr_prep(req, sqe); |
03b1230c | 5835 | case IORING_OP_SENDMSG: |
fddaface | 5836 | case IORING_OP_SEND: |
bfe76559 | 5837 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5838 | case IORING_OP_RECVMSG: |
fddaface | 5839 | case IORING_OP_RECV: |
bfe76559 | 5840 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5841 | case IORING_OP_CONNECT: |
bfe76559 | 5842 | return io_connect_prep(req, sqe); |
2d28390a | 5843 | case IORING_OP_TIMEOUT: |
bfe76559 | 5844 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5845 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5846 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5847 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5848 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5849 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5850 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5851 | case IORING_OP_ACCEPT: |
bfe76559 | 5852 | return io_accept_prep(req, sqe); |
d63d1b5e | 5853 | case IORING_OP_FALLOCATE: |
bfe76559 | 5854 | return io_fallocate_prep(req, sqe); |
15b71abe | 5855 | case IORING_OP_OPENAT: |
bfe76559 | 5856 | return io_openat_prep(req, sqe); |
b5dba59e | 5857 | case IORING_OP_CLOSE: |
bfe76559 | 5858 | return io_close_prep(req, sqe); |
05f3fb3c | 5859 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5860 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5861 | case IORING_OP_STATX: |
bfe76559 | 5862 | return io_statx_prep(req, sqe); |
4840e418 | 5863 | case IORING_OP_FADVISE: |
bfe76559 | 5864 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5865 | case IORING_OP_MADVISE: |
bfe76559 | 5866 | return io_madvise_prep(req, sqe); |
cebdb986 | 5867 | case IORING_OP_OPENAT2: |
bfe76559 | 5868 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5869 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5870 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5871 | case IORING_OP_SPLICE: |
bfe76559 | 5872 | return io_splice_prep(req, sqe); |
ddf0322d | 5873 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5874 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5875 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5876 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5877 | case IORING_OP_TEE: |
bfe76559 | 5878 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5879 | case IORING_OP_SHUTDOWN: |
5880 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5881 | case IORING_OP_RENAMEAT: |
5882 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5883 | case IORING_OP_UNLINKAT: |
5884 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5885 | } |
5886 | ||
bfe76559 PB |
5887 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5888 | req->opcode); | |
5889 | return-EINVAL; | |
5890 | } | |
5891 | ||
93642ef8 | 5892 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5893 | { |
93642ef8 PB |
5894 | switch (req->opcode) { |
5895 | case IORING_OP_READV: | |
5896 | case IORING_OP_READ_FIXED: | |
5897 | case IORING_OP_READ: | |
5898 | return io_rw_prep_async(req, READ); | |
5899 | case IORING_OP_WRITEV: | |
5900 | case IORING_OP_WRITE_FIXED: | |
5901 | case IORING_OP_WRITE: | |
5902 | return io_rw_prep_async(req, WRITE); | |
5903 | case IORING_OP_SENDMSG: | |
5904 | case IORING_OP_SEND: | |
5905 | return io_sendmsg_prep_async(req); | |
5906 | case IORING_OP_RECVMSG: | |
5907 | case IORING_OP_RECV: | |
5908 | return io_recvmsg_prep_async(req); | |
5909 | case IORING_OP_CONNECT: | |
5910 | return io_connect_prep_async(req); | |
5911 | } | |
5912 | return 0; | |
5913 | } | |
5914 | ||
be7053b7 | 5915 | static int io_req_defer_prep(struct io_kiocb *req) |
bfe76559 | 5916 | { |
be7053b7 | 5917 | if (!io_op_defs[req->opcode].needs_async_data) |
bfe76559 | 5918 | return 0; |
be7053b7 | 5919 | /* some opcodes init it during the inital prep */ |
93642ef8 | 5920 | if (req->async_data) |
be7053b7 PB |
5921 | return 0; |
5922 | if (__io_alloc_async_data(req)) | |
bfe76559 | 5923 | return -EAGAIN; |
be7053b7 | 5924 | return io_req_prep_async(req); |
f67676d1 JA |
5925 | } |
5926 | ||
9cf7c104 PB |
5927 | static u32 io_get_sequence(struct io_kiocb *req) |
5928 | { | |
5929 | struct io_kiocb *pos; | |
5930 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5931 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5932 | |
f2f87370 PB |
5933 | io_for_each_link(pos, req) |
5934 | nr_reqs++; | |
9cf7c104 PB |
5935 | |
5936 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5937 | return total_submitted - nr_reqs; | |
5938 | } | |
5939 | ||
be7053b7 | 5940 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5941 | { |
a197f664 | 5942 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5943 | struct io_defer_entry *de; |
f67676d1 | 5944 | int ret; |
9cf7c104 | 5945 | u32 seq; |
de0617e4 | 5946 | |
9d858b21 | 5947 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5948 | if (likely(list_empty_careful(&ctx->defer_list) && |
5949 | !(req->flags & REQ_F_IO_DRAIN))) | |
5950 | return 0; | |
5951 | ||
5952 | seq = io_get_sequence(req); | |
5953 | /* Still a chance to pass the sequence check */ | |
5954 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5955 | return 0; |
5956 | ||
be7053b7 PB |
5957 | ret = io_req_defer_prep(req); |
5958 | if (ret) | |
5959 | return ret; | |
cbdcb435 | 5960 | io_prep_async_link(req); |
27dc8338 PB |
5961 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5962 | if (!de) | |
5963 | return -ENOMEM; | |
2d28390a | 5964 | |
de0617e4 | 5965 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5966 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5967 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5968 | kfree(de); |
ae34817b PB |
5969 | io_queue_async_work(req); |
5970 | return -EIOCBQUEUED; | |
de0617e4 JA |
5971 | } |
5972 | ||
915967f6 | 5973 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5974 | de->req = req; |
9cf7c104 | 5975 | de->seq = seq; |
27dc8338 | 5976 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5977 | spin_unlock_irq(&ctx->completion_lock); |
5978 | return -EIOCBQUEUED; | |
5979 | } | |
5980 | ||
3ca405eb | 5981 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5982 | { |
0e1b6fe3 PB |
5983 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5984 | switch (req->opcode) { | |
5985 | case IORING_OP_READV: | |
5986 | case IORING_OP_READ_FIXED: | |
5987 | case IORING_OP_READ: | |
bcda7baa | 5988 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5989 | break; |
5990 | case IORING_OP_RECVMSG: | |
5991 | case IORING_OP_RECV: | |
bcda7baa | 5992 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5993 | break; |
5994 | } | |
5995 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5996 | } |
5997 | ||
0e1b6fe3 PB |
5998 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5999 | switch (req->opcode) { | |
6000 | case IORING_OP_READV: | |
6001 | case IORING_OP_READ_FIXED: | |
6002 | case IORING_OP_READ: | |
6003 | case IORING_OP_WRITEV: | |
6004 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6005 | case IORING_OP_WRITE: { |
6006 | struct io_async_rw *io = req->async_data; | |
6007 | if (io->free_iovec) | |
6008 | kfree(io->free_iovec); | |
0e1b6fe3 | 6009 | break; |
e8c2bc1f | 6010 | } |
0e1b6fe3 | 6011 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6012 | case IORING_OP_SENDMSG: { |
6013 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6014 | |
6015 | kfree(io->free_iov); | |
0e1b6fe3 | 6016 | break; |
e8c2bc1f | 6017 | } |
0e1b6fe3 PB |
6018 | case IORING_OP_SPLICE: |
6019 | case IORING_OP_TEE: | |
6020 | io_put_file(req, req->splice.file_in, | |
6021 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6022 | break; | |
f3cd4850 JA |
6023 | case IORING_OP_OPENAT: |
6024 | case IORING_OP_OPENAT2: | |
6025 | if (req->open.filename) | |
6026 | putname(req->open.filename); | |
6027 | break; | |
80a261fd JA |
6028 | case IORING_OP_RENAMEAT: |
6029 | putname(req->rename.oldpath); | |
6030 | putname(req->rename.newpath); | |
6031 | break; | |
14a1143b JA |
6032 | case IORING_OP_UNLINKAT: |
6033 | putname(req->unlink.filename); | |
6034 | break; | |
0e1b6fe3 PB |
6035 | } |
6036 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6037 | } |
99bc4c38 PB |
6038 | } |
6039 | ||
889fca73 | 6040 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6041 | { |
a197f664 | 6042 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6043 | const struct cred *creds = NULL; |
d625c6ee | 6044 | int ret; |
2b188cc1 | 6045 | |
003e8dcc JA |
6046 | if (req->work.creds && req->work.creds != current_cred()) |
6047 | creds = override_creds(req->work.creds); | |
5730b27e | 6048 | |
d625c6ee | 6049 | switch (req->opcode) { |
2b188cc1 | 6050 | case IORING_OP_NOP: |
889fca73 | 6051 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6052 | break; |
6053 | case IORING_OP_READV: | |
edafccee | 6054 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6055 | case IORING_OP_READ: |
889fca73 | 6056 | ret = io_read(req, issue_flags); |
edafccee | 6057 | break; |
3529d8c2 | 6058 | case IORING_OP_WRITEV: |
edafccee | 6059 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6060 | case IORING_OP_WRITE: |
889fca73 | 6061 | ret = io_write(req, issue_flags); |
2b188cc1 | 6062 | break; |
c992fe29 | 6063 | case IORING_OP_FSYNC: |
45d189c6 | 6064 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6065 | break; |
221c5eb2 | 6066 | case IORING_OP_POLL_ADD: |
61e98203 | 6067 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6068 | break; |
6069 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6070 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6071 | break; |
5d17b4a4 | 6072 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6073 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6074 | break; |
0fa03c62 | 6075 | case IORING_OP_SENDMSG: |
889fca73 | 6076 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6077 | break; |
fddaface | 6078 | case IORING_OP_SEND: |
889fca73 | 6079 | ret = io_send(req, issue_flags); |
0fa03c62 | 6080 | break; |
aa1fa28f | 6081 | case IORING_OP_RECVMSG: |
889fca73 | 6082 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6083 | break; |
fddaface | 6084 | case IORING_OP_RECV: |
889fca73 | 6085 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6086 | break; |
5262f567 | 6087 | case IORING_OP_TIMEOUT: |
61e98203 | 6088 | ret = io_timeout(req, issue_flags); |
5262f567 | 6089 | break; |
11365043 | 6090 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6091 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6092 | break; |
17f2fe35 | 6093 | case IORING_OP_ACCEPT: |
889fca73 | 6094 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6095 | break; |
f8e85cf2 | 6096 | case IORING_OP_CONNECT: |
889fca73 | 6097 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6098 | break; |
62755e35 | 6099 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6100 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6101 | break; |
d63d1b5e | 6102 | case IORING_OP_FALLOCATE: |
45d189c6 | 6103 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6104 | break; |
15b71abe | 6105 | case IORING_OP_OPENAT: |
45d189c6 | 6106 | ret = io_openat(req, issue_flags); |
15b71abe | 6107 | break; |
b5dba59e | 6108 | case IORING_OP_CLOSE: |
889fca73 | 6109 | ret = io_close(req, issue_flags); |
b5dba59e | 6110 | break; |
05f3fb3c | 6111 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6112 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6113 | break; |
eddc7ef5 | 6114 | case IORING_OP_STATX: |
45d189c6 | 6115 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6116 | break; |
4840e418 | 6117 | case IORING_OP_FADVISE: |
45d189c6 | 6118 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6119 | break; |
c1ca757b | 6120 | case IORING_OP_MADVISE: |
45d189c6 | 6121 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6122 | break; |
cebdb986 | 6123 | case IORING_OP_OPENAT2: |
45d189c6 | 6124 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6125 | break; |
3e4827b0 | 6126 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6127 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6128 | break; |
7d67af2c | 6129 | case IORING_OP_SPLICE: |
45d189c6 | 6130 | ret = io_splice(req, issue_flags); |
7d67af2c | 6131 | break; |
ddf0322d | 6132 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6133 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6134 | break; |
067524e9 | 6135 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6136 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6137 | break; |
f2a8d5c7 | 6138 | case IORING_OP_TEE: |
45d189c6 | 6139 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6140 | break; |
36f4fa68 | 6141 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6142 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6143 | break; |
80a261fd | 6144 | case IORING_OP_RENAMEAT: |
45d189c6 | 6145 | ret = io_renameat(req, issue_flags); |
80a261fd | 6146 | break; |
14a1143b | 6147 | case IORING_OP_UNLINKAT: |
45d189c6 | 6148 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6149 | break; |
2b188cc1 JA |
6150 | default: |
6151 | ret = -EINVAL; | |
6152 | break; | |
6153 | } | |
6154 | ||
5730b27e JA |
6155 | if (creds) |
6156 | revert_creds(creds); | |
6157 | ||
def596e9 JA |
6158 | if (ret) |
6159 | return ret; | |
6160 | ||
b532576e JA |
6161 | /* If the op doesn't have a file, we're not polling for it */ |
6162 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6163 | const bool in_async = io_wq_current_is_worker(); |
6164 | ||
11ba820b JA |
6165 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6166 | if (in_async) | |
6167 | mutex_lock(&ctx->uring_lock); | |
6168 | ||
2e9dbe90 | 6169 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6170 | |
6171 | if (in_async) | |
6172 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6173 | } |
6174 | ||
6175 | return 0; | |
2b188cc1 JA |
6176 | } |
6177 | ||
5280f7e5 | 6178 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6179 | { |
6180 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6181 | struct io_kiocb *timeout; |
561fb04a | 6182 | int ret = 0; |
2b188cc1 | 6183 | |
6df1db6b PB |
6184 | timeout = io_prep_linked_timeout(req); |
6185 | if (timeout) | |
6186 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6187 | |
4014d943 | 6188 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6189 | ret = -ECANCELED; |
31b51510 | 6190 | |
561fb04a | 6191 | if (!ret) { |
561fb04a | 6192 | do { |
889fca73 | 6193 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6194 | /* |
6195 | * We can get EAGAIN for polled IO even though we're | |
6196 | * forcing a sync submission from here, since we can't | |
6197 | * wait for request slots on the block side. | |
6198 | */ | |
6199 | if (ret != -EAGAIN) | |
6200 | break; | |
6201 | cond_resched(); | |
6202 | } while (1); | |
6203 | } | |
31b51510 | 6204 | |
a3df7698 | 6205 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6206 | if (ret) { |
a3df7698 PB |
6207 | /* io-wq is going to take one down */ |
6208 | refcount_inc(&req->refs); | |
6209 | io_req_task_queue_fail(req, ret); | |
edafccee | 6210 | } |
2b188cc1 JA |
6211 | } |
6212 | ||
65e19f54 JA |
6213 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, |
6214 | int index) | |
6215 | { | |
269bbe5f | 6216 | struct fixed_rsrc_table *table; |
65e19f54 | 6217 | |
05f3fb3c | 6218 | table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT]; |
84695089 | 6219 | return table->files[index & IORING_FILE_TABLE_MASK]; |
65e19f54 JA |
6220 | } |
6221 | ||
8371adf5 PB |
6222 | static struct file *io_file_get(struct io_submit_state *state, |
6223 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6224 | { |
a197f664 | 6225 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6226 | struct file *file; |
09bb8394 | 6227 | |
8da11c19 | 6228 | if (fixed) { |
479f517b | 6229 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6230 | return NULL; |
b7620121 | 6231 | fd = array_index_nospec(fd, ctx->nr_user_files); |
8da11c19 | 6232 | file = io_file_from_index(ctx, fd); |
36f72fe2 | 6233 | io_set_resource_node(req); |
09bb8394 | 6234 | } else { |
c826bd7a | 6235 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6236 | file = __io_file_get(state, fd); |
09bb8394 JA |
6237 | } |
6238 | ||
ce3d5aae PB |
6239 | if (file && unlikely(file->f_op == &io_uring_fops)) |
6240 | io_req_track_inflight(req); | |
8371adf5 | 6241 | return file; |
09bb8394 JA |
6242 | } |
6243 | ||
2665abfd | 6244 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6245 | { |
ad8a48ac JA |
6246 | struct io_timeout_data *data = container_of(timer, |
6247 | struct io_timeout_data, timer); | |
90cd7e42 | 6248 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6249 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6250 | unsigned long flags; |
2665abfd JA |
6251 | |
6252 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6253 | prev = req->timeout.head; |
6254 | req->timeout.head = NULL; | |
2665abfd JA |
6255 | |
6256 | /* | |
6257 | * We don't expect the list to be empty, that will only happen if we | |
6258 | * race with the completion of the linked work. | |
6259 | */ | |
90cd7e42 | 6260 | if (prev && refcount_inc_not_zero(&prev->refs)) |
f2f87370 | 6261 | io_remove_next_linked(prev); |
90cd7e42 PB |
6262 | else |
6263 | prev = NULL; | |
2665abfd JA |
6264 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6265 | ||
6266 | if (prev) { | |
014db007 | 6267 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6268 | io_put_req_deferred(prev, 1); |
47f46768 | 6269 | } else { |
9ae1f8dd PB |
6270 | io_req_complete_post(req, -ETIME, 0); |
6271 | io_put_req_deferred(req, 1); | |
2665abfd | 6272 | } |
2665abfd JA |
6273 | return HRTIMER_NORESTART; |
6274 | } | |
6275 | ||
7271ef3a | 6276 | static void __io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6277 | { |
76a46e06 | 6278 | /* |
f2f87370 PB |
6279 | * If the back reference is NULL, then our linked request finished |
6280 | * before we got a chance to setup the timer | |
76a46e06 | 6281 | */ |
90cd7e42 | 6282 | if (req->timeout.head) { |
e8c2bc1f | 6283 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6284 | |
ad8a48ac JA |
6285 | data->timer.function = io_link_timeout_fn; |
6286 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6287 | data->mode); | |
2665abfd | 6288 | } |
7271ef3a JA |
6289 | } |
6290 | ||
6291 | static void io_queue_linked_timeout(struct io_kiocb *req) | |
6292 | { | |
6293 | struct io_ring_ctx *ctx = req->ctx; | |
6294 | ||
6295 | spin_lock_irq(&ctx->completion_lock); | |
6296 | __io_queue_linked_timeout(req); | |
76a46e06 | 6297 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6298 | |
2665abfd | 6299 | /* drop submission reference */ |
76a46e06 JA |
6300 | io_put_req(req); |
6301 | } | |
2665abfd | 6302 | |
ad8a48ac | 6303 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6304 | { |
f2f87370 | 6305 | struct io_kiocb *nxt = req->link; |
2665abfd | 6306 | |
f2f87370 PB |
6307 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6308 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6309 | return NULL; |
2665abfd | 6310 | |
90cd7e42 | 6311 | nxt->timeout.head = req; |
900fad45 | 6312 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6313 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6314 | return nxt; |
2665abfd JA |
6315 | } |
6316 | ||
c5eef2b9 | 6317 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6318 | { |
d3d7298d | 6319 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6320 | int ret; |
2b188cc1 | 6321 | |
c5eef2b9 | 6322 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6323 | |
491381ce JA |
6324 | /* |
6325 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6326 | * doesn't support non-blocking read/write attempts | |
6327 | */ | |
24c74678 | 6328 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6329 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6330 | /* |
6331 | * Queued up for async execution, worker will release | |
6332 | * submit reference when the iocb is actually submitted. | |
6333 | */ | |
6334 | io_queue_async_work(req); | |
2b188cc1 | 6335 | } |
0d63c148 PB |
6336 | } else if (likely(!ret)) { |
6337 | /* drop submission reference */ | |
e342c807 | 6338 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6339 | struct io_ring_ctx *ctx = req->ctx; |
6340 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6341 | |
6dd0be1e | 6342 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6343 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6344 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6345 | } else { |
d3d7298d | 6346 | io_put_req(req); |
0d63c148 PB |
6347 | } |
6348 | } else { | |
4e88d6e7 | 6349 | req_set_fail_links(req); |
e65ef56d | 6350 | io_put_req(req); |
e1e16097 | 6351 | io_req_complete(req, ret); |
9e645e11 | 6352 | } |
d3d7298d PB |
6353 | if (linked_timeout) |
6354 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6355 | } |
6356 | ||
be7053b7 | 6357 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6358 | { |
6359 | int ret; | |
6360 | ||
be7053b7 | 6361 | ret = io_req_defer(req); |
4fe2c963 JL |
6362 | if (ret) { |
6363 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6364 | fail_req: |
4e88d6e7 | 6365 | req_set_fail_links(req); |
e1e16097 JA |
6366 | io_put_req(req); |
6367 | io_req_complete(req, ret); | |
4fe2c963 | 6368 | } |
2550878f | 6369 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
be7053b7 PB |
6370 | ret = io_req_defer_prep(req); |
6371 | if (unlikely(ret)) | |
6372 | goto fail_req; | |
ce35a47a JA |
6373 | io_queue_async_work(req); |
6374 | } else { | |
c5eef2b9 | 6375 | __io_queue_sqe(req); |
ce35a47a | 6376 | } |
4fe2c963 JL |
6377 | } |
6378 | ||
b16fed66 PB |
6379 | /* |
6380 | * Check SQE restrictions (opcode and flags). | |
6381 | * | |
6382 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6383 | */ | |
6384 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6385 | struct io_kiocb *req, | |
6386 | unsigned int sqe_flags) | |
4fe2c963 | 6387 | { |
b16fed66 PB |
6388 | if (!ctx->restricted) |
6389 | return true; | |
6390 | ||
6391 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6392 | return false; | |
6393 | ||
6394 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6395 | ctx->restrictions.sqe_flags_required) | |
6396 | return false; | |
6397 | ||
6398 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6399 | ctx->restrictions.sqe_flags_required)) | |
6400 | return false; | |
6401 | ||
6402 | return true; | |
4fe2c963 JL |
6403 | } |
6404 | ||
b16fed66 PB |
6405 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6406 | const struct io_uring_sqe *sqe) | |
6407 | { | |
6408 | struct io_submit_state *state; | |
6409 | unsigned int sqe_flags; | |
003e8dcc | 6410 | int personality, ret = 0; |
b16fed66 PB |
6411 | |
6412 | req->opcode = READ_ONCE(sqe->opcode); | |
6413 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6414 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6415 | req->user_data = READ_ONCE(sqe->user_data); | |
6416 | req->async_data = NULL; | |
6417 | req->file = NULL; | |
6418 | req->ctx = ctx; | |
6419 | req->link = NULL; | |
6420 | req->fixed_rsrc_refs = NULL; | |
6421 | /* one is dropped after submission, the other at completion */ | |
6422 | refcount_set(&req->refs, 2); | |
6423 | req->task = current; | |
6424 | req->result = 0; | |
93e68e03 JA |
6425 | req->work.list.next = NULL; |
6426 | req->work.creds = NULL; | |
6427 | req->work.flags = 0; | |
b16fed66 PB |
6428 | |
6429 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6430 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6431 | req->flags = 0; | |
b16fed66 | 6432 | return -EINVAL; |
ebf4a5db | 6433 | } |
b16fed66 PB |
6434 | |
6435 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6436 | return -EINVAL; | |
6437 | ||
b16fed66 PB |
6438 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6439 | return -EACCES; | |
6440 | ||
6441 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6442 | !io_op_defs[req->opcode].buffer_select) | |
6443 | return -EOPNOTSUPP; | |
863e0560 | 6444 | |
003e8dcc JA |
6445 | personality = READ_ONCE(sqe->personality); |
6446 | if (personality) { | |
61cf9370 | 6447 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6448 | if (!req->work.creds) |
6449 | return -EINVAL; | |
6450 | get_cred(req->work.creds); | |
003e8dcc | 6451 | } |
b16fed66 PB |
6452 | state = &ctx->submit_state; |
6453 | ||
6454 | /* | |
6455 | * Plug now if we have more than 1 IO left after this, and the target | |
6456 | * is potentially a read/write to block based storage. | |
6457 | */ | |
6458 | if (!state->plug_started && state->ios_left > 1 && | |
6459 | io_op_defs[req->opcode].plug) { | |
6460 | blk_start_plug(&state->plug); | |
6461 | state->plug_started = true; | |
6462 | } | |
6463 | ||
6464 | if (io_op_defs[req->opcode].needs_file) { | |
6465 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6466 | ||
6467 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6468 | if (unlikely(!req->file)) | |
6469 | ret = -EBADF; | |
6470 | } | |
6471 | ||
6472 | state->ios_left--; | |
6473 | return ret; | |
6474 | } | |
6475 | ||
a6b8cadc | 6476 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6477 | const struct io_uring_sqe *sqe) |
9e645e11 | 6478 | { |
a1ab7b35 | 6479 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6480 | int ret; |
9e645e11 | 6481 | |
a6b8cadc PB |
6482 | ret = io_init_req(ctx, req, sqe); |
6483 | if (unlikely(ret)) { | |
6484 | fail_req: | |
de59bc10 PB |
6485 | if (link->head) { |
6486 | /* fail even hard links since we don't submit */ | |
cf109604 | 6487 | link->head->flags |= REQ_F_FAIL_LINK; |
de59bc10 PB |
6488 | io_put_req(link->head); |
6489 | io_req_complete(link->head, -ECANCELED); | |
6490 | link->head = NULL; | |
6491 | } | |
90b87490 PB |
6492 | io_put_req(req); |
6493 | io_req_complete(req, ret); | |
a6b8cadc PB |
6494 | return ret; |
6495 | } | |
be7053b7 PB |
6496 | ret = io_req_prep(req, sqe); |
6497 | if (unlikely(ret)) | |
6498 | goto fail_req; | |
a6b8cadc | 6499 | |
be7053b7 | 6500 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6501 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6502 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6503 | ||
9e645e11 JA |
6504 | /* |
6505 | * If we already have a head request, queue this one for async | |
6506 | * submittal once the head completes. If we don't have a head but | |
6507 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6508 | * submitted sync once the chain is complete. If none of those | |
6509 | * conditions are true (normal request), then just queue it. | |
6510 | */ | |
863e0560 PB |
6511 | if (link->head) { |
6512 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6513 | |
8cdf2193 PB |
6514 | /* |
6515 | * Taking sequential execution of a link, draining both sides | |
6516 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6517 | * requests in the link. So, it drains the head and the | |
6518 | * next after the link request. The last one is done via | |
6519 | * drain_next flag to persist the effect across calls. | |
6520 | */ | |
ef4ff581 | 6521 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6522 | head->flags |= REQ_F_IO_DRAIN; |
6523 | ctx->drain_next = 1; | |
6524 | } | |
be7053b7 | 6525 | ret = io_req_defer_prep(req); |
cf109604 | 6526 | if (unlikely(ret)) |
a6b8cadc | 6527 | goto fail_req; |
9d76377f | 6528 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6529 | link->last->link = req; |
863e0560 | 6530 | link->last = req; |
32fe525b PB |
6531 | |
6532 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6533 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6534 | io_queue_sqe(head); |
863e0560 | 6535 | link->head = NULL; |
32fe525b | 6536 | } |
9e645e11 | 6537 | } else { |
711be031 PB |
6538 | if (unlikely(ctx->drain_next)) { |
6539 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6540 | ctx->drain_next = 0; |
711be031 | 6541 | } |
ef4ff581 | 6542 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6543 | link->head = req; |
6544 | link->last = req; | |
711be031 | 6545 | } else { |
be7053b7 | 6546 | io_queue_sqe(req); |
711be031 | 6547 | } |
9e645e11 | 6548 | } |
2e6e1fde | 6549 | |
1d4240cc | 6550 | return 0; |
9e645e11 JA |
6551 | } |
6552 | ||
9a56a232 JA |
6553 | /* |
6554 | * Batched submission is done, ensure local IO is flushed out. | |
6555 | */ | |
ba88ff11 PB |
6556 | static void io_submit_state_end(struct io_submit_state *state, |
6557 | struct io_ring_ctx *ctx) | |
9a56a232 | 6558 | { |
a1ab7b35 | 6559 | if (state->link.head) |
de59bc10 | 6560 | io_queue_sqe(state->link.head); |
6dd0be1e | 6561 | if (state->comp.nr) |
ba88ff11 | 6562 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6563 | if (state->plug_started) |
6564 | blk_finish_plug(&state->plug); | |
9f13c35b | 6565 | io_state_file_put(state); |
9a56a232 JA |
6566 | } |
6567 | ||
6568 | /* | |
6569 | * Start submission side cache. | |
6570 | */ | |
6571 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6572 | unsigned int max_ios) |
9a56a232 | 6573 | { |
27926b68 | 6574 | state->plug_started = false; |
9a56a232 | 6575 | state->ios_left = max_ios; |
a1ab7b35 PB |
6576 | /* set only head, no need to init link_last in advance */ |
6577 | state->link.head = NULL; | |
9a56a232 JA |
6578 | } |
6579 | ||
2b188cc1 JA |
6580 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6581 | { | |
75b28aff | 6582 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6583 | |
caf582c6 PB |
6584 | /* |
6585 | * Ensure any loads from the SQEs are done at this point, | |
6586 | * since once we write the new head, the application could | |
6587 | * write new data to them. | |
6588 | */ | |
6589 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6590 | } |
6591 | ||
2b188cc1 | 6592 | /* |
3529d8c2 | 6593 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6594 | * that is mapped by userspace. This means that care needs to be taken to |
6595 | * ensure that reads are stable, as we cannot rely on userspace always | |
6596 | * being a good citizen. If members of the sqe are validated and then later | |
6597 | * used, it's important that those reads are done through READ_ONCE() to | |
6598 | * prevent a re-load down the line. | |
6599 | */ | |
709b302f | 6600 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6601 | { |
75b28aff | 6602 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6603 | unsigned head; |
6604 | ||
6605 | /* | |
6606 | * The cached sq head (or cq tail) serves two purposes: | |
6607 | * | |
6608 | * 1) allows us to batch the cost of updating the user visible | |
6609 | * head updates. | |
6610 | * 2) allows the kernel side to track the head on its own, even | |
6611 | * though the application is the one updating it. | |
6612 | */ | |
4fccfcbb | 6613 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6614 | if (likely(head < ctx->sq_entries)) |
6615 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6616 | |
6617 | /* drop invalid entries */ | |
498ccd9e | 6618 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6619 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6620 | return NULL; |
6621 | } | |
6622 | ||
0f212204 | 6623 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6624 | { |
46c4e16a | 6625 | int submitted = 0; |
6c271ce2 | 6626 | |
c4a2ed72 | 6627 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6628 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6629 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6630 | return -EBUSY; |
6631 | } | |
6c271ce2 | 6632 | |
ee7d46d9 PB |
6633 | /* make sure SQ entry isn't read before tail */ |
6634 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6635 | |
2b85edfc PB |
6636 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6637 | return -EAGAIN; | |
6c271ce2 | 6638 | |
d8a6df10 | 6639 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6640 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6641 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6642 | |
46c4e16a | 6643 | while (submitted < nr) { |
3529d8c2 | 6644 | const struct io_uring_sqe *sqe; |
196be95c | 6645 | struct io_kiocb *req; |
fb5ccc98 | 6646 | |
258b29a9 | 6647 | req = io_alloc_req(ctx); |
196be95c PB |
6648 | if (unlikely(!req)) { |
6649 | if (!submitted) | |
6650 | submitted = -EAGAIN; | |
fb5ccc98 | 6651 | break; |
196be95c | 6652 | } |
4fccfcbb PB |
6653 | sqe = io_get_sqe(ctx); |
6654 | if (unlikely(!sqe)) { | |
6655 | kmem_cache_free(req_cachep, req); | |
6656 | break; | |
6657 | } | |
d3656344 JA |
6658 | /* will complete beyond this point, count as submitted */ |
6659 | submitted++; | |
a1ab7b35 | 6660 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6661 | break; |
6c271ce2 JA |
6662 | } |
6663 | ||
9466f437 PB |
6664 | if (unlikely(submitted != nr)) { |
6665 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6666 | struct io_uring_task *tctx = current->io_uring; |
6667 | int unused = nr - ref_used; | |
9466f437 | 6668 | |
d8a6df10 JA |
6669 | percpu_ref_put_many(&ctx->refs, unused); |
6670 | percpu_counter_sub(&tctx->inflight, unused); | |
6671 | put_task_struct_many(current, unused); | |
9466f437 | 6672 | } |
6c271ce2 | 6673 | |
a1ab7b35 | 6674 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6675 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6676 | io_commit_sqring(ctx); | |
6677 | ||
6c271ce2 JA |
6678 | return submitted; |
6679 | } | |
6680 | ||
23b3628e XW |
6681 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6682 | { | |
6683 | /* Tell userspace we may need a wakeup call */ | |
6684 | spin_lock_irq(&ctx->completion_lock); | |
6685 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6686 | spin_unlock_irq(&ctx->completion_lock); | |
6687 | } | |
6688 | ||
6689 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6690 | { | |
6691 | spin_lock_irq(&ctx->completion_lock); | |
6692 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6693 | spin_unlock_irq(&ctx->completion_lock); | |
6694 | } | |
6695 | ||
08369246 | 6696 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6697 | { |
c8d1ba58 | 6698 | unsigned int to_submit; |
bdcd3eab | 6699 | int ret = 0; |
6c271ce2 | 6700 | |
c8d1ba58 | 6701 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6702 | /* if we're handling multiple rings, cap submit size for fairness */ |
6703 | if (cap_entries && to_submit > 8) | |
6704 | to_submit = 8; | |
6705 | ||
906a3c6f | 6706 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6707 | unsigned nr_events = 0; |
a4c0b3de | 6708 | |
c8d1ba58 | 6709 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6710 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6711 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6712 | |
0298ef96 PB |
6713 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6714 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6715 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6716 | mutex_unlock(&ctx->uring_lock); |
6717 | } | |
6c271ce2 | 6718 | |
90554200 JA |
6719 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6720 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6721 | |
08369246 XW |
6722 | return ret; |
6723 | } | |
6c271ce2 | 6724 | |
08369246 XW |
6725 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6726 | { | |
6727 | struct io_ring_ctx *ctx; | |
6728 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6729 | |
08369246 XW |
6730 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
6731 | if (sq_thread_idle < ctx->sq_thread_idle) | |
6732 | sq_thread_idle = ctx->sq_thread_idle; | |
c8d1ba58 | 6733 | } |
c1edbf5f | 6734 | |
08369246 | 6735 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6736 | } |
6c271ce2 | 6737 | |
c8d1ba58 JA |
6738 | static int io_sq_thread(void *data) |
6739 | { | |
69fb2131 JA |
6740 | struct io_sq_data *sqd = data; |
6741 | struct io_ring_ctx *ctx; | |
a0d9205f | 6742 | unsigned long timeout = 0; |
37d1e2e3 | 6743 | char buf[TASK_COMM_LEN]; |
08369246 | 6744 | DEFINE_WAIT(wait); |
6c271ce2 | 6745 | |
696ee88a | 6746 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6747 | set_task_comm(current, buf); |
37d1e2e3 JA |
6748 | current->pf_io_worker = NULL; |
6749 | ||
6750 | if (sqd->sq_cpu != -1) | |
6751 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6752 | else | |
6753 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6754 | current->flags |= PF_NO_SETAFFINITY; | |
6755 | ||
09a6f4ef | 6756 | mutex_lock(&sqd->lock); |
c7d95613 PB |
6757 | /* a user may had exited before the thread started */ |
6758 | io_run_task_work_head(&sqd->park_task_work); | |
6759 | ||
05962f95 | 6760 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6761 | int ret; |
6762 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6763 | |
82734c5b JA |
6764 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6765 | signal_pending(current)) { | |
6766 | bool did_sig = false; | |
6767 | ||
09a6f4ef | 6768 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6769 | if (signal_pending(current)) { |
6770 | struct ksignal ksig; | |
6771 | ||
6772 | did_sig = get_signal(&ksig); | |
6773 | } | |
05962f95 | 6774 | cond_resched(); |
09a6f4ef | 6775 | mutex_lock(&sqd->lock); |
521d6a73 | 6776 | io_run_task_work(); |
b7f5a0bf | 6777 | io_run_task_work_head(&sqd->park_task_work); |
c7d95613 PB |
6778 | if (did_sig) |
6779 | break; | |
08369246 | 6780 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6781 | continue; |
08369246 | 6782 | } |
08369246 | 6783 | sqt_spin = false; |
e95eee2d | 6784 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6785 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6786 | const struct cred *creds = NULL; |
6787 | ||
6788 | if (ctx->sq_creds != current_cred()) | |
6789 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6790 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6791 | if (creds) |
6792 | revert_creds(creds); | |
08369246 XW |
6793 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6794 | sqt_spin = true; | |
69fb2131 | 6795 | } |
6c271ce2 | 6796 | |
08369246 | 6797 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6798 | io_run_task_work(); |
6799 | cond_resched(); | |
08369246 XW |
6800 | if (sqt_spin) |
6801 | timeout = jiffies + sqd->sq_thread_idle; | |
6802 | continue; | |
6803 | } | |
6804 | ||
08369246 XW |
6805 | needs_sched = true; |
6806 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6807 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6808 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6809 | !list_empty_careful(&ctx->iopoll_list)) { | |
6810 | needs_sched = false; | |
6811 | break; | |
6812 | } | |
6813 | if (io_sqring_entries(ctx)) { | |
6814 | needs_sched = false; | |
6815 | break; | |
6816 | } | |
6817 | } | |
6818 | ||
05962f95 | 6819 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6820 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6821 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6822 | |
09a6f4ef | 6823 | mutex_unlock(&sqd->lock); |
69fb2131 | 6824 | schedule(); |
09a6f4ef | 6825 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6826 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6827 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6828 | } |
08369246 XW |
6829 | |
6830 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6831 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6832 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6833 | } |
28cea78a | 6834 | |
521d6a73 PB |
6835 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6836 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6837 | sqd->thread = NULL; |
05962f95 | 6838 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6839 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6840 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6841 | |
6842 | io_run_task_work(); | |
b7f5a0bf | 6843 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6844 | complete(&sqd->exited); |
6845 | do_exit(0); | |
6c271ce2 JA |
6846 | } |
6847 | ||
bda52162 JA |
6848 | struct io_wait_queue { |
6849 | struct wait_queue_entry wq; | |
6850 | struct io_ring_ctx *ctx; | |
6851 | unsigned to_wait; | |
6852 | unsigned nr_timeouts; | |
6853 | }; | |
6854 | ||
6c503150 | 6855 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6856 | { |
6857 | struct io_ring_ctx *ctx = iowq->ctx; | |
6858 | ||
6859 | /* | |
d195a66e | 6860 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6861 | * started waiting. For timeouts, we always want to return to userspace, |
6862 | * regardless of event count. | |
6863 | */ | |
6c503150 | 6864 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6865 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6866 | } | |
6867 | ||
6868 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6869 | int wake_flags, void *key) | |
6870 | { | |
6871 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6872 | wq); | |
6873 | ||
6c503150 PB |
6874 | /* |
6875 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6876 | * the task, and the next invocation will do it. | |
6877 | */ | |
6878 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6879 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6880 | return -1; | |
bda52162 JA |
6881 | } |
6882 | ||
af9c1a44 JA |
6883 | static int io_run_task_work_sig(void) |
6884 | { | |
6885 | if (io_run_task_work()) | |
6886 | return 1; | |
6887 | if (!signal_pending(current)) | |
6888 | return 0; | |
0b8cfa97 | 6889 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6890 | return -ERESTARTSYS; |
af9c1a44 JA |
6891 | return -EINTR; |
6892 | } | |
6893 | ||
eeb60b9a PB |
6894 | /* when returns >0, the caller should retry */ |
6895 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6896 | struct io_wait_queue *iowq, | |
6897 | signed long *timeout) | |
6898 | { | |
6899 | int ret; | |
6900 | ||
6901 | /* make sure we run task_work before checking for signals */ | |
6902 | ret = io_run_task_work_sig(); | |
6903 | if (ret || io_should_wake(iowq)) | |
6904 | return ret; | |
6905 | /* let the caller flush overflows, retry */ | |
6906 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6907 | return 1; | |
6908 | ||
6909 | *timeout = schedule_timeout(*timeout); | |
6910 | return !*timeout ? -ETIME : 1; | |
6911 | } | |
6912 | ||
2b188cc1 JA |
6913 | /* |
6914 | * Wait until events become available, if we don't already have some. The | |
6915 | * application must reap them itself, as they reside on the shared cq ring. | |
6916 | */ | |
6917 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6918 | const sigset_t __user *sig, size_t sigsz, |
6919 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6920 | { |
bda52162 JA |
6921 | struct io_wait_queue iowq = { |
6922 | .wq = { | |
6923 | .private = current, | |
6924 | .func = io_wake_function, | |
6925 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6926 | }, | |
6927 | .ctx = ctx, | |
6928 | .to_wait = min_events, | |
6929 | }; | |
75b28aff | 6930 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6931 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6932 | int ret; | |
2b188cc1 | 6933 | |
b41e9852 | 6934 | do { |
6c503150 PB |
6935 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6936 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6937 | return 0; |
4c6e277c | 6938 | if (!io_run_task_work()) |
b41e9852 | 6939 | break; |
b41e9852 | 6940 | } while (1); |
2b188cc1 JA |
6941 | |
6942 | if (sig) { | |
9e75ad5d AB |
6943 | #ifdef CONFIG_COMPAT |
6944 | if (in_compat_syscall()) | |
6945 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6946 | sigsz); |
9e75ad5d AB |
6947 | else |
6948 | #endif | |
b772434b | 6949 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6950 | |
2b188cc1 JA |
6951 | if (ret) |
6952 | return ret; | |
6953 | } | |
6954 | ||
c73ebb68 | 6955 | if (uts) { |
c1d5a224 PB |
6956 | struct timespec64 ts; |
6957 | ||
c73ebb68 HX |
6958 | if (get_timespec64(&ts, uts)) |
6959 | return -EFAULT; | |
6960 | timeout = timespec64_to_jiffies(&ts); | |
6961 | } | |
6962 | ||
bda52162 | 6963 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6964 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6965 | do { |
ca0a2651 JA |
6966 | /* if we can't even flush overflow, don't wait for more */ |
6967 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6968 | ret = -EBUSY; | |
6969 | break; | |
6970 | } | |
bda52162 JA |
6971 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6972 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6973 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6974 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6975 | cond_resched(); |
eeb60b9a | 6976 | } while (ret > 0); |
bda52162 | 6977 | |
b7db41c9 | 6978 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6979 | |
75b28aff | 6980 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6981 | } |
6982 | ||
6b06314c JA |
6983 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6984 | { | |
6985 | #if defined(CONFIG_UNIX) | |
6986 | if (ctx->ring_sock) { | |
6987 | struct sock *sock = ctx->ring_sock->sk; | |
6988 | struct sk_buff *skb; | |
6989 | ||
6990 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6991 | kfree_skb(skb); | |
6992 | } | |
6993 | #else | |
6994 | int i; | |
6995 | ||
65e19f54 JA |
6996 | for (i = 0; i < ctx->nr_user_files; i++) { |
6997 | struct file *file; | |
6998 | ||
6999 | file = io_file_from_index(ctx, i); | |
7000 | if (file) | |
7001 | fput(file); | |
7002 | } | |
6b06314c JA |
7003 | #endif |
7004 | } | |
7005 | ||
00835dce | 7006 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7007 | { |
269bbe5f | 7008 | struct fixed_rsrc_data *data; |
05f3fb3c | 7009 | |
269bbe5f | 7010 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7011 | complete(&data->done); |
7012 | } | |
7013 | ||
2a63b2d9 | 7014 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7015 | { |
2a63b2d9 | 7016 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7017 | } |
7018 | ||
2a63b2d9 | 7019 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7020 | { |
2a63b2d9 BM |
7021 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7022 | } | |
65e19f54 | 7023 | |
d67d2263 BM |
7024 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7025 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7026 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7027 | { |
2a63b2d9 | 7028 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7029 | rsrc_data->node = ref_node; |
d67d2263 | 7030 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7031 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7032 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7033 | } |
7034 | ||
8bad28d8 | 7035 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 7036 | { |
8bad28d8 | 7037 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 7038 | |
2a63b2d9 | 7039 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7040 | ref_node = data->node; |
e6cb007c | 7041 | data->node = NULL; |
2a63b2d9 | 7042 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7043 | if (ref_node) |
7044 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
7045 | } |
7046 | ||
7047 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
7048 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
7049 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7050 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7051 | { |
f2303b1f | 7052 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 7053 | int ret; |
05589553 | 7054 | |
8bad28d8 HX |
7055 | if (data->quiesce) |
7056 | return -ENXIO; | |
05589553 | 7057 | |
8bad28d8 | 7058 | data->quiesce = true; |
1ffc5422 | 7059 | do { |
f2303b1f PB |
7060 | ret = -ENOMEM; |
7061 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7062 | if (!backup_node) | |
7063 | break; | |
7064 | backup_node->rsrc_data = data; | |
7065 | backup_node->rsrc_put = rsrc_put; | |
7066 | ||
8bad28d8 HX |
7067 | io_sqe_rsrc_kill_node(ctx, data); |
7068 | percpu_ref_kill(&data->refs); | |
7069 | flush_delayed_work(&ctx->rsrc_put_work); | |
7070 | ||
1ffc5422 PB |
7071 | ret = wait_for_completion_interruptible(&data->done); |
7072 | if (!ret) | |
7073 | break; | |
8bad28d8 | 7074 | |
cb5e1b81 | 7075 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7076 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7077 | backup_node = NULL; | |
cb5e1b81 | 7078 | reinit_completion(&data->done); |
8bad28d8 | 7079 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7080 | ret = io_run_task_work_sig(); |
8bad28d8 | 7081 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7082 | } while (ret >= 0); |
8bad28d8 | 7083 | data->quiesce = false; |
05f3fb3c | 7084 | |
8bad28d8 HX |
7085 | if (backup_node) |
7086 | destroy_fixed_rsrc_ref_node(backup_node); | |
7087 | return ret; | |
d7954b2b BM |
7088 | } |
7089 | ||
1ad555c6 BM |
7090 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7091 | { | |
7092 | struct fixed_rsrc_data *data; | |
7093 | ||
7094 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7095 | if (!data) | |
7096 | return NULL; | |
7097 | ||
00835dce | 7098 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7099 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7100 | kfree(data); | |
7101 | return NULL; | |
7102 | } | |
7103 | data->ctx = ctx; | |
7104 | init_completion(&data->done); | |
7105 | return data; | |
7106 | } | |
7107 | ||
7108 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7109 | { | |
7110 | percpu_ref_exit(&data->refs); | |
7111 | kfree(data->table); | |
7112 | kfree(data); | |
7113 | } | |
7114 | ||
d7954b2b BM |
7115 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7116 | { | |
7117 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7118 | unsigned nr_tables, i; |
7119 | int ret; | |
7120 | ||
8bad28d8 HX |
7121 | /* |
7122 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7123 | * Since we possibly drop uring lock later in this function to | |
7124 | * run task work. | |
7125 | */ | |
7126 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7127 | return -ENXIO; |
f2303b1f | 7128 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7129 | if (ret) |
7130 | return ret; | |
7131 | ||
6b06314c | 7132 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7133 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7134 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7135 | kfree(data->table[i].files); |
1ad555c6 | 7136 | free_fixed_rsrc_data(data); |
05f3fb3c | 7137 | ctx->file_data = NULL; |
6b06314c JA |
7138 | ctx->nr_user_files = 0; |
7139 | return 0; | |
7140 | } | |
7141 | ||
37d1e2e3 | 7142 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7143 | __releases(&sqd->lock) |
37d1e2e3 | 7144 | { |
521d6a73 PB |
7145 | WARN_ON_ONCE(sqd->thread == current); |
7146 | ||
9e138a48 PB |
7147 | /* |
7148 | * Do the dance but not conditional clear_bit() because it'd race with | |
7149 | * other threads incrementing park_pending and setting the bit. | |
7150 | */ | |
37d1e2e3 | 7151 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7152 | if (atomic_dec_return(&sqd->park_pending)) |
7153 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7154 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7155 | } |
7156 | ||
86e0d676 | 7157 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7158 | __acquires(&sqd->lock) |
37d1e2e3 | 7159 | { |
521d6a73 PB |
7160 | WARN_ON_ONCE(sqd->thread == current); |
7161 | ||
9e138a48 | 7162 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7163 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7164 | mutex_lock(&sqd->lock); |
05962f95 | 7165 | if (sqd->thread) |
86e0d676 | 7166 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7167 | } |
7168 | ||
7169 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7170 | { | |
521d6a73 PB |
7171 | WARN_ON_ONCE(sqd->thread == current); |
7172 | ||
09a6f4ef | 7173 | mutex_lock(&sqd->lock); |
05962f95 | 7174 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7175 | if (sqd->thread) |
7176 | wake_up_process(sqd->thread); | |
09a6f4ef | 7177 | mutex_unlock(&sqd->lock); |
05962f95 | 7178 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7179 | } |
7180 | ||
534ca6d6 | 7181 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7182 | { |
534ca6d6 | 7183 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7184 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7185 | ||
37d1e2e3 JA |
7186 | io_sq_thread_stop(sqd); |
7187 | kfree(sqd); | |
7188 | } | |
7189 | } | |
7190 | ||
7191 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7192 | { | |
7193 | struct io_sq_data *sqd = ctx->sq_data; | |
7194 | ||
7195 | if (sqd) { | |
05962f95 | 7196 | io_sq_thread_park(sqd); |
521d6a73 | 7197 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7198 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7199 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7200 | |
7201 | io_put_sq_data(sqd); | |
7202 | ctx->sq_data = NULL; | |
7c30f36a SM |
7203 | if (ctx->sq_creds) |
7204 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7205 | } |
7206 | } | |
7207 | ||
aa06165d JA |
7208 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7209 | { | |
7210 | struct io_ring_ctx *ctx_attach; | |
7211 | struct io_sq_data *sqd; | |
7212 | struct fd f; | |
7213 | ||
7214 | f = fdget(p->wq_fd); | |
7215 | if (!f.file) | |
7216 | return ERR_PTR(-ENXIO); | |
7217 | if (f.file->f_op != &io_uring_fops) { | |
7218 | fdput(f); | |
7219 | return ERR_PTR(-EINVAL); | |
7220 | } | |
7221 | ||
7222 | ctx_attach = f.file->private_data; | |
7223 | sqd = ctx_attach->sq_data; | |
7224 | if (!sqd) { | |
7225 | fdput(f); | |
7226 | return ERR_PTR(-EINVAL); | |
7227 | } | |
5c2469e0 JA |
7228 | if (sqd->task_tgid != current->tgid) { |
7229 | fdput(f); | |
7230 | return ERR_PTR(-EPERM); | |
7231 | } | |
aa06165d JA |
7232 | |
7233 | refcount_inc(&sqd->refs); | |
7234 | fdput(f); | |
7235 | return sqd; | |
7236 | } | |
7237 | ||
26984fbf PB |
7238 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7239 | bool *attached) | |
534ca6d6 JA |
7240 | { |
7241 | struct io_sq_data *sqd; | |
7242 | ||
26984fbf | 7243 | *attached = false; |
5c2469e0 JA |
7244 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7245 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7246 | if (!IS_ERR(sqd)) { |
7247 | *attached = true; | |
5c2469e0 | 7248 | return sqd; |
26984fbf | 7249 | } |
5c2469e0 JA |
7250 | /* fall through for EPERM case, setup new sqd/task */ |
7251 | if (PTR_ERR(sqd) != -EPERM) | |
7252 | return sqd; | |
7253 | } | |
aa06165d | 7254 | |
534ca6d6 JA |
7255 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7256 | if (!sqd) | |
7257 | return ERR_PTR(-ENOMEM); | |
7258 | ||
9e138a48 | 7259 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7260 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7261 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7262 | mutex_init(&sqd->lock); |
534ca6d6 | 7263 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7264 | init_completion(&sqd->exited); |
534ca6d6 JA |
7265 | return sqd; |
7266 | } | |
7267 | ||
6b06314c | 7268 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7269 | /* |
7270 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7271 | * the io_uring can be safely unregistered on process exit, even if we have | |
7272 | * loops in the file referencing. | |
7273 | */ | |
7274 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7275 | { | |
7276 | struct sock *sk = ctx->ring_sock->sk; | |
7277 | struct scm_fp_list *fpl; | |
7278 | struct sk_buff *skb; | |
08a45173 | 7279 | int i, nr_files; |
6b06314c | 7280 | |
6b06314c JA |
7281 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7282 | if (!fpl) | |
7283 | return -ENOMEM; | |
7284 | ||
7285 | skb = alloc_skb(0, GFP_KERNEL); | |
7286 | if (!skb) { | |
7287 | kfree(fpl); | |
7288 | return -ENOMEM; | |
7289 | } | |
7290 | ||
7291 | skb->sk = sk; | |
6b06314c | 7292 | |
08a45173 | 7293 | nr_files = 0; |
62e398be | 7294 | fpl->user = get_uid(current_user()); |
6b06314c | 7295 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7296 | struct file *file = io_file_from_index(ctx, i + offset); |
7297 | ||
7298 | if (!file) | |
08a45173 | 7299 | continue; |
65e19f54 | 7300 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7301 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7302 | nr_files++; | |
6b06314c JA |
7303 | } |
7304 | ||
08a45173 JA |
7305 | if (nr_files) { |
7306 | fpl->max = SCM_MAX_FD; | |
7307 | fpl->count = nr_files; | |
7308 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7309 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7310 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7311 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7312 | |
08a45173 JA |
7313 | for (i = 0; i < nr_files; i++) |
7314 | fput(fpl->fp[i]); | |
7315 | } else { | |
7316 | kfree_skb(skb); | |
7317 | kfree(fpl); | |
7318 | } | |
6b06314c JA |
7319 | |
7320 | return 0; | |
7321 | } | |
7322 | ||
7323 | /* | |
7324 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7325 | * causes regular reference counting to break down. We rely on the UNIX | |
7326 | * garbage collection to take care of this problem for us. | |
7327 | */ | |
7328 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7329 | { | |
7330 | unsigned left, total; | |
7331 | int ret = 0; | |
7332 | ||
7333 | total = 0; | |
7334 | left = ctx->nr_user_files; | |
7335 | while (left) { | |
7336 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7337 | |
7338 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7339 | if (ret) | |
7340 | break; | |
7341 | left -= this_files; | |
7342 | total += this_files; | |
7343 | } | |
7344 | ||
7345 | if (!ret) | |
7346 | return 0; | |
7347 | ||
7348 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7349 | struct file *file = io_file_from_index(ctx, total); |
7350 | ||
7351 | if (file) | |
7352 | fput(file); | |
6b06314c JA |
7353 | total++; |
7354 | } | |
7355 | ||
7356 | return ret; | |
7357 | } | |
7358 | #else | |
7359 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7360 | { | |
7361 | return 0; | |
7362 | } | |
7363 | #endif | |
7364 | ||
269bbe5f | 7365 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7366 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7367 | { |
7368 | int i; | |
7369 | ||
7370 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7371 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7372 | unsigned this_files; |
7373 | ||
7374 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7375 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7376 | GFP_KERNEL); | |
7377 | if (!table->files) | |
7378 | break; | |
7379 | nr_files -= this_files; | |
7380 | } | |
7381 | ||
7382 | if (i == nr_tables) | |
7383 | return 0; | |
7384 | ||
7385 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7386 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7387 | kfree(table->files); |
7388 | } | |
7389 | return 1; | |
7390 | } | |
7391 | ||
50238531 | 7392 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7393 | { |
50238531 | 7394 | struct file *file = prsrc->file; |
05f3fb3c JA |
7395 | #if defined(CONFIG_UNIX) |
7396 | struct sock *sock = ctx->ring_sock->sk; | |
7397 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7398 | struct sk_buff *skb; | |
7399 | int i; | |
7400 | ||
7401 | __skb_queue_head_init(&list); | |
7402 | ||
7403 | /* | |
7404 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7405 | * remove this entry and rearrange the file array. | |
7406 | */ | |
7407 | skb = skb_dequeue(head); | |
7408 | while (skb) { | |
7409 | struct scm_fp_list *fp; | |
7410 | ||
7411 | fp = UNIXCB(skb).fp; | |
7412 | for (i = 0; i < fp->count; i++) { | |
7413 | int left; | |
7414 | ||
7415 | if (fp->fp[i] != file) | |
7416 | continue; | |
7417 | ||
7418 | unix_notinflight(fp->user, fp->fp[i]); | |
7419 | left = fp->count - 1 - i; | |
7420 | if (left) { | |
7421 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7422 | left * sizeof(struct file *)); | |
7423 | } | |
7424 | fp->count--; | |
7425 | if (!fp->count) { | |
7426 | kfree_skb(skb); | |
7427 | skb = NULL; | |
7428 | } else { | |
7429 | __skb_queue_tail(&list, skb); | |
7430 | } | |
7431 | fput(file); | |
7432 | file = NULL; | |
7433 | break; | |
7434 | } | |
7435 | ||
7436 | if (!file) | |
7437 | break; | |
7438 | ||
7439 | __skb_queue_tail(&list, skb); | |
7440 | ||
7441 | skb = skb_dequeue(head); | |
7442 | } | |
7443 | ||
7444 | if (skb_peek(&list)) { | |
7445 | spin_lock_irq(&head->lock); | |
7446 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7447 | __skb_queue_tail(head, skb); | |
7448 | spin_unlock_irq(&head->lock); | |
7449 | } | |
7450 | #else | |
7451 | fput(file); | |
7452 | #endif | |
7453 | } | |
7454 | ||
269bbe5f | 7455 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7456 | { |
269bbe5f BM |
7457 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7458 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7459 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7460 | |
269bbe5f BM |
7461 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7462 | list_del(&prsrc->list); | |
50238531 | 7463 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7464 | kfree(prsrc); |
65e19f54 | 7465 | } |
05589553 | 7466 | |
05589553 XW |
7467 | percpu_ref_exit(&ref_node->refs); |
7468 | kfree(ref_node); | |
269bbe5f | 7469 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7470 | } |
65e19f54 | 7471 | |
269bbe5f | 7472 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7473 | { |
7474 | struct io_ring_ctx *ctx; | |
7475 | struct llist_node *node; | |
7476 | ||
269bbe5f BM |
7477 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7478 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7479 | |
7480 | while (node) { | |
269bbe5f | 7481 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7482 | struct llist_node *next = node->next; |
7483 | ||
269bbe5f BM |
7484 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7485 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7486 | node = next; |
7487 | } | |
7488 | } | |
7489 | ||
ea64ec02 PB |
7490 | static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
7491 | unsigned i) | |
2faf852d | 7492 | { |
ea64ec02 PB |
7493 | struct fixed_rsrc_table *table; |
7494 | ||
7495 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; | |
7496 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
7497 | } | |
7498 | ||
00835dce | 7499 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7500 | { |
269bbe5f BM |
7501 | struct fixed_rsrc_ref_node *ref_node; |
7502 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7503 | struct io_ring_ctx *ctx; |
e297822b | 7504 | bool first_add = false; |
4a38aed2 | 7505 | int delay = HZ; |
65e19f54 | 7506 | |
269bbe5f BM |
7507 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7508 | data = ref_node->rsrc_data; | |
e297822b PB |
7509 | ctx = data->ctx; |
7510 | ||
2a63b2d9 | 7511 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7512 | ref_node->done = true; |
7513 | ||
d67d2263 BM |
7514 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7515 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7516 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7517 | /* recycle ref nodes in order */ |
7518 | if (!ref_node->done) | |
7519 | break; | |
7520 | list_del(&ref_node->node); | |
269bbe5f | 7521 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7522 | } |
2a63b2d9 | 7523 | io_rsrc_ref_unlock(ctx); |
05589553 | 7524 | |
e297822b | 7525 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7526 | delay = 0; |
05589553 | 7527 | |
4a38aed2 | 7528 | if (!delay) |
269bbe5f | 7529 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7530 | else if (first_add) |
269bbe5f | 7531 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7532 | } |
65e19f54 | 7533 | |
6802535d | 7534 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7535 | struct io_ring_ctx *ctx) |
05f3fb3c | 7536 | { |
269bbe5f | 7537 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7538 | |
05589553 XW |
7539 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7540 | if (!ref_node) | |
3e2224c5 | 7541 | return NULL; |
05f3fb3c | 7542 | |
00835dce | 7543 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7544 | 0, GFP_KERNEL)) { |
7545 | kfree(ref_node); | |
3e2224c5 | 7546 | return NULL; |
05589553 XW |
7547 | } |
7548 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7549 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7550 | ref_node->done = false; |
05589553 | 7551 | return ref_node; |
05589553 XW |
7552 | } |
7553 | ||
bc9744cd PB |
7554 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7555 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7556 | { |
269bbe5f | 7557 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7558 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7559 | } |
7560 | ||
269bbe5f | 7561 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7562 | { |
7563 | percpu_ref_exit(&ref_node->refs); | |
7564 | kfree(ref_node); | |
65e19f54 JA |
7565 | } |
7566 | ||
ea64ec02 | 7567 | |
6b06314c JA |
7568 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7569 | unsigned nr_args) | |
7570 | { | |
7571 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7572 | unsigned nr_tables, i; |
05f3fb3c | 7573 | struct file *file; |
600cf3f8 | 7574 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7575 | struct fixed_rsrc_ref_node *ref_node; |
7576 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7577 | |
05f3fb3c | 7578 | if (ctx->file_data) |
6b06314c JA |
7579 | return -EBUSY; |
7580 | if (!nr_args) | |
7581 | return -EINVAL; | |
7582 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7583 | return -EMFILE; | |
7584 | ||
1ad555c6 | 7585 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7586 | if (!file_data) |
05f3fb3c | 7587 | return -ENOMEM; |
13770a71 | 7588 | ctx->file_data = file_data; |
05f3fb3c | 7589 | |
65e19f54 | 7590 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7591 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7592 | GFP_KERNEL); |
600cf3f8 PB |
7593 | if (!file_data->table) |
7594 | goto out_free; | |
05f3fb3c | 7595 | |
600cf3f8 | 7596 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7597 | goto out_free; |
65e19f54 | 7598 | |
08a45173 | 7599 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7600 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7601 | ret = -EFAULT; | |
7602 | goto out_fput; | |
7603 | } | |
08a45173 | 7604 | /* allow sparse sets */ |
600cf3f8 | 7605 | if (fd == -1) |
08a45173 | 7606 | continue; |
6b06314c | 7607 | |
05f3fb3c | 7608 | file = fget(fd); |
6b06314c | 7609 | ret = -EBADF; |
05f3fb3c | 7610 | if (!file) |
600cf3f8 | 7611 | goto out_fput; |
05f3fb3c | 7612 | |
6b06314c JA |
7613 | /* |
7614 | * Don't allow io_uring instances to be registered. If UNIX | |
7615 | * isn't enabled, then this causes a reference cycle and this | |
7616 | * instance can never get freed. If UNIX is enabled we'll | |
7617 | * handle it just fine, but there's still no point in allowing | |
7618 | * a ring fd as it doesn't support regular read/write anyway. | |
7619 | */ | |
05f3fb3c JA |
7620 | if (file->f_op == &io_uring_fops) { |
7621 | fput(file); | |
600cf3f8 | 7622 | goto out_fput; |
6b06314c | 7623 | } |
ea64ec02 | 7624 | *io_fixed_file_slot(file_data, i) = file; |
6b06314c JA |
7625 | } |
7626 | ||
6b06314c | 7627 | ret = io_sqe_files_scm(ctx); |
05589553 | 7628 | if (ret) { |
6b06314c | 7629 | io_sqe_files_unregister(ctx); |
05589553 XW |
7630 | return ret; |
7631 | } | |
6b06314c | 7632 | |
bc9744cd | 7633 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7634 | if (!ref_node) { |
05589553 | 7635 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7636 | return -ENOMEM; |
05589553 | 7637 | } |
bc9744cd | 7638 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7639 | |
d67d2263 | 7640 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7641 | return ret; |
600cf3f8 PB |
7642 | out_fput: |
7643 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7644 | file = io_file_from_index(ctx, i); | |
7645 | if (file) | |
7646 | fput(file); | |
7647 | } | |
7648 | for (i = 0; i < nr_tables; i++) | |
7649 | kfree(file_data->table[i].files); | |
7650 | ctx->nr_user_files = 0; | |
600cf3f8 | 7651 | out_free: |
1ad555c6 | 7652 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7653 | ctx->file_data = NULL; |
6b06314c JA |
7654 | return ret; |
7655 | } | |
7656 | ||
c3a31e60 JA |
7657 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7658 | int index) | |
7659 | { | |
7660 | #if defined(CONFIG_UNIX) | |
7661 | struct sock *sock = ctx->ring_sock->sk; | |
7662 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7663 | struct sk_buff *skb; | |
7664 | ||
7665 | /* | |
7666 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7667 | * file set. If there's no room, fall back to allocating a new skb | |
7668 | * and filling it in. | |
7669 | */ | |
7670 | spin_lock_irq(&head->lock); | |
7671 | skb = skb_peek(head); | |
7672 | if (skb) { | |
7673 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7674 | ||
7675 | if (fpl->count < SCM_MAX_FD) { | |
7676 | __skb_unlink(skb, head); | |
7677 | spin_unlock_irq(&head->lock); | |
7678 | fpl->fp[fpl->count] = get_file(file); | |
7679 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7680 | fpl->count++; | |
7681 | spin_lock_irq(&head->lock); | |
7682 | __skb_queue_head(head, skb); | |
7683 | } else { | |
7684 | skb = NULL; | |
7685 | } | |
7686 | } | |
7687 | spin_unlock_irq(&head->lock); | |
7688 | ||
7689 | if (skb) { | |
7690 | fput(file); | |
7691 | return 0; | |
7692 | } | |
7693 | ||
7694 | return __io_sqe_files_scm(ctx, 1, index); | |
7695 | #else | |
7696 | return 0; | |
7697 | #endif | |
7698 | } | |
7699 | ||
50238531 | 7700 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7701 | { |
269bbe5f BM |
7702 | struct io_rsrc_put *prsrc; |
7703 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7704 | |
269bbe5f BM |
7705 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7706 | if (!prsrc) | |
a5318d3c | 7707 | return -ENOMEM; |
05f3fb3c | 7708 | |
50238531 | 7709 | prsrc->rsrc = rsrc; |
269bbe5f | 7710 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7711 | |
a5318d3c | 7712 | return 0; |
05f3fb3c JA |
7713 | } |
7714 | ||
269bbe5f BM |
7715 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7716 | struct file *file) | |
7717 | { | |
50238531 | 7718 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7719 | } |
7720 | ||
05f3fb3c | 7721 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7722 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7723 | unsigned nr_args) |
7724 | { | |
269bbe5f BM |
7725 | struct fixed_rsrc_data *data = ctx->file_data; |
7726 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7727 | struct file *file, **file_slot; |
c3a31e60 JA |
7728 | __s32 __user *fds; |
7729 | int fd, i, err; | |
7730 | __u32 done; | |
05589553 | 7731 | bool needs_switch = false; |
c3a31e60 | 7732 | |
05f3fb3c | 7733 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7734 | return -EOVERFLOW; |
7735 | if (done > ctx->nr_user_files) | |
7736 | return -EINVAL; | |
7737 | ||
bc9744cd | 7738 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7739 | if (!ref_node) |
7740 | return -ENOMEM; | |
bc9744cd | 7741 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7742 | |
269bbe5f | 7743 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7744 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7745 | err = 0; |
7746 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7747 | err = -EFAULT; | |
7748 | break; | |
7749 | } | |
4e0377a1 | 7750 | if (fd == IORING_REGISTER_FILES_SKIP) |
7751 | continue; | |
7752 | ||
67973b93 | 7753 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7754 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7755 | ||
7756 | if (*file_slot) { | |
7757 | err = io_queue_file_removal(data, *file_slot); | |
a5318d3c HD |
7758 | if (err) |
7759 | break; | |
ea64ec02 | 7760 | *file_slot = NULL; |
05589553 | 7761 | needs_switch = true; |
c3a31e60 JA |
7762 | } |
7763 | if (fd != -1) { | |
c3a31e60 JA |
7764 | file = fget(fd); |
7765 | if (!file) { | |
7766 | err = -EBADF; | |
7767 | break; | |
7768 | } | |
7769 | /* | |
7770 | * Don't allow io_uring instances to be registered. If | |
7771 | * UNIX isn't enabled, then this causes a reference | |
7772 | * cycle and this instance can never get freed. If UNIX | |
7773 | * is enabled we'll handle it just fine, but there's | |
7774 | * still no point in allowing a ring fd as it doesn't | |
7775 | * support regular read/write anyway. | |
7776 | */ | |
7777 | if (file->f_op == &io_uring_fops) { | |
7778 | fput(file); | |
7779 | err = -EBADF; | |
7780 | break; | |
7781 | } | |
e68a3ff8 | 7782 | *file_slot = file; |
c3a31e60 | 7783 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7784 | if (err) { |
e68a3ff8 | 7785 | *file_slot = NULL; |
f3bd9dae | 7786 | fput(file); |
c3a31e60 | 7787 | break; |
f3bd9dae | 7788 | } |
c3a31e60 | 7789 | } |
05f3fb3c JA |
7790 | } |
7791 | ||
05589553 | 7792 | if (needs_switch) { |
b2e96852 | 7793 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7794 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7795 | } else |
269bbe5f | 7796 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7797 | |
7798 | return done ? done : err; | |
7799 | } | |
05589553 | 7800 | |
05f3fb3c JA |
7801 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7802 | unsigned nr_args) | |
7803 | { | |
269bbe5f | 7804 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7805 | |
7806 | if (!ctx->file_data) | |
7807 | return -ENXIO; | |
7808 | if (!nr_args) | |
7809 | return -EINVAL; | |
7810 | if (copy_from_user(&up, arg, sizeof(up))) | |
7811 | return -EFAULT; | |
7812 | if (up.resv) | |
7813 | return -EINVAL; | |
7814 | ||
7815 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7816 | } | |
c3a31e60 | 7817 | |
5280f7e5 | 7818 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7819 | { |
7820 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7821 | ||
5280f7e5 PB |
7822 | req = io_put_req_find_next(req); |
7823 | return req ? &req->work : NULL; | |
7d723065 JA |
7824 | } |
7825 | ||
5aa75ed5 | 7826 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7827 | { |
e941894e | 7828 | struct io_wq_hash *hash; |
24369c2e | 7829 | struct io_wq_data data; |
24369c2e | 7830 | unsigned int concurrency; |
24369c2e | 7831 | |
e941894e JA |
7832 | hash = ctx->hash_map; |
7833 | if (!hash) { | |
7834 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7835 | if (!hash) | |
7836 | return ERR_PTR(-ENOMEM); | |
7837 | refcount_set(&hash->refs, 1); | |
7838 | init_waitqueue_head(&hash->wait); | |
7839 | ctx->hash_map = hash; | |
24369c2e PB |
7840 | } |
7841 | ||
e941894e | 7842 | data.hash = hash; |
e9fd9396 | 7843 | data.free_work = io_free_work; |
f5fa38c5 | 7844 | data.do_work = io_wq_submit_work; |
24369c2e | 7845 | |
d25e3a3d JA |
7846 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7847 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7848 | |
5aa75ed5 | 7849 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7850 | } |
7851 | ||
5aa75ed5 JA |
7852 | static int io_uring_alloc_task_context(struct task_struct *task, |
7853 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7854 | { |
7855 | struct io_uring_task *tctx; | |
d8a6df10 | 7856 | int ret; |
0f212204 JA |
7857 | |
7858 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7859 | if (unlikely(!tctx)) | |
7860 | return -ENOMEM; | |
7861 | ||
d8a6df10 JA |
7862 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7863 | if (unlikely(ret)) { | |
7864 | kfree(tctx); | |
7865 | return ret; | |
7866 | } | |
7867 | ||
5aa75ed5 JA |
7868 | tctx->io_wq = io_init_wq_offload(ctx); |
7869 | if (IS_ERR(tctx->io_wq)) { | |
7870 | ret = PTR_ERR(tctx->io_wq); | |
7871 | percpu_counter_destroy(&tctx->inflight); | |
7872 | kfree(tctx); | |
7873 | return ret; | |
7874 | } | |
7875 | ||
0f212204 JA |
7876 | xa_init(&tctx->xa); |
7877 | init_waitqueue_head(&tctx->wait); | |
7878 | tctx->last = NULL; | |
fdaf083c | 7879 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7880 | task->io_uring = tctx; |
7cbf1722 JA |
7881 | spin_lock_init(&tctx->task_lock); |
7882 | INIT_WQ_LIST(&tctx->task_list); | |
7883 | tctx->task_state = 0; | |
7884 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7885 | return 0; |
7886 | } | |
7887 | ||
7888 | void __io_uring_free(struct task_struct *tsk) | |
7889 | { | |
7890 | struct io_uring_task *tctx = tsk->io_uring; | |
7891 | ||
7892 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7893 | WARN_ON_ONCE(tctx->io_wq); |
7894 | ||
d8a6df10 | 7895 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7896 | kfree(tctx); |
7897 | tsk->io_uring = NULL; | |
7898 | } | |
7899 | ||
7e84e1c7 SG |
7900 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7901 | struct io_uring_params *p) | |
2b188cc1 JA |
7902 | { |
7903 | int ret; | |
7904 | ||
d25e3a3d JA |
7905 | /* Retain compatibility with failing for an invalid attach attempt */ |
7906 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7907 | IORING_SETUP_ATTACH_WQ) { | |
7908 | struct fd f; | |
7909 | ||
7910 | f = fdget(p->wq_fd); | |
7911 | if (!f.file) | |
7912 | return -ENXIO; | |
7913 | if (f.file->f_op != &io_uring_fops) { | |
7914 | fdput(f); | |
7915 | return -EINVAL; | |
7916 | } | |
7917 | fdput(f); | |
7918 | } | |
6c271ce2 | 7919 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7920 | struct task_struct *tsk; |
534ca6d6 | 7921 | struct io_sq_data *sqd; |
26984fbf | 7922 | bool attached; |
534ca6d6 | 7923 | |
3ec482d1 | 7924 | ret = -EPERM; |
ce59fc69 | 7925 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7926 | goto err; |
7927 | ||
26984fbf | 7928 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7929 | if (IS_ERR(sqd)) { |
7930 | ret = PTR_ERR(sqd); | |
7931 | goto err; | |
7932 | } | |
69fb2131 | 7933 | |
7c30f36a | 7934 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7935 | ctx->sq_data = sqd; |
917257da JA |
7936 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7937 | if (!ctx->sq_thread_idle) | |
7938 | ctx->sq_thread_idle = HZ; | |
7939 | ||
26984fbf | 7940 | ret = 0; |
78d7f6ba | 7941 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7942 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7943 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7944 | /* don't attach to a dying SQPOLL thread, would be racy */ |
de75a3d3 | 7945 | if (attached && !sqd->thread) |
26984fbf | 7946 | ret = -ENXIO; |
78d7f6ba PB |
7947 | io_sq_thread_unpark(sqd); |
7948 | ||
de75a3d3 PB |
7949 | if (ret < 0) |
7950 | goto err; | |
7951 | if (attached) | |
5aa75ed5 | 7952 | return 0; |
aa06165d | 7953 | |
6c271ce2 | 7954 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7955 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7956 | |
917257da | 7957 | ret = -EINVAL; |
44a9bd18 | 7958 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7959 | goto err_sqpoll; |
7889f44d | 7960 | if (!cpu_online(cpu)) |
e8f98f24 | 7961 | goto err_sqpoll; |
917257da | 7962 | |
37d1e2e3 | 7963 | sqd->sq_cpu = cpu; |
6c271ce2 | 7964 | } else { |
37d1e2e3 | 7965 | sqd->sq_cpu = -1; |
6c271ce2 | 7966 | } |
37d1e2e3 JA |
7967 | |
7968 | sqd->task_pid = current->pid; | |
5c2469e0 | 7969 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7970 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7971 | if (IS_ERR(tsk)) { | |
7972 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7973 | goto err_sqpoll; |
6c271ce2 | 7974 | } |
97a73a0f | 7975 | |
46fe18b1 | 7976 | sqd->thread = tsk; |
97a73a0f | 7977 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7978 | wake_up_new_task(tsk); |
0f212204 JA |
7979 | if (ret) |
7980 | goto err; | |
6c271ce2 JA |
7981 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7982 | /* Can't have SQ_AFF without SQPOLL */ | |
7983 | ret = -EINVAL; | |
7984 | goto err; | |
7985 | } | |
7986 | ||
2b188cc1 JA |
7987 | return 0; |
7988 | err: | |
37d1e2e3 | 7989 | io_sq_thread_finish(ctx); |
2b188cc1 | 7990 | return ret; |
e8f98f24 JA |
7991 | err_sqpoll: |
7992 | complete(&ctx->sq_data->exited); | |
7993 | goto err; | |
2b188cc1 JA |
7994 | } |
7995 | ||
a087e2b5 BM |
7996 | static inline void __io_unaccount_mem(struct user_struct *user, |
7997 | unsigned long nr_pages) | |
2b188cc1 JA |
7998 | { |
7999 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8000 | } | |
8001 | ||
a087e2b5 BM |
8002 | static inline int __io_account_mem(struct user_struct *user, |
8003 | unsigned long nr_pages) | |
2b188cc1 JA |
8004 | { |
8005 | unsigned long page_limit, cur_pages, new_pages; | |
8006 | ||
8007 | /* Don't allow more pages than we can safely lock */ | |
8008 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8009 | ||
8010 | do { | |
8011 | cur_pages = atomic_long_read(&user->locked_vm); | |
8012 | new_pages = cur_pages + nr_pages; | |
8013 | if (new_pages > page_limit) | |
8014 | return -ENOMEM; | |
8015 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8016 | new_pages) != cur_pages); | |
8017 | ||
8018 | return 0; | |
8019 | } | |
8020 | ||
26bfa89e | 8021 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8022 | { |
62e398be | 8023 | if (ctx->user) |
a087e2b5 | 8024 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8025 | |
26bfa89e JA |
8026 | if (ctx->mm_account) |
8027 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8028 | } |
8029 | ||
26bfa89e | 8030 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8031 | { |
30975825 BM |
8032 | int ret; |
8033 | ||
62e398be | 8034 | if (ctx->user) { |
30975825 BM |
8035 | ret = __io_account_mem(ctx->user, nr_pages); |
8036 | if (ret) | |
8037 | return ret; | |
8038 | } | |
8039 | ||
26bfa89e JA |
8040 | if (ctx->mm_account) |
8041 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8042 | |
8043 | return 0; | |
8044 | } | |
8045 | ||
2b188cc1 JA |
8046 | static void io_mem_free(void *ptr) |
8047 | { | |
52e04ef4 MR |
8048 | struct page *page; |
8049 | ||
8050 | if (!ptr) | |
8051 | return; | |
2b188cc1 | 8052 | |
52e04ef4 | 8053 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8054 | if (put_page_testzero(page)) |
8055 | free_compound_page(page); | |
8056 | } | |
8057 | ||
8058 | static void *io_mem_alloc(size_t size) | |
8059 | { | |
8060 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8061 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8062 | |
8063 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8064 | } | |
8065 | ||
75b28aff HV |
8066 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8067 | size_t *sq_offset) | |
8068 | { | |
8069 | struct io_rings *rings; | |
8070 | size_t off, sq_array_size; | |
8071 | ||
8072 | off = struct_size(rings, cqes, cq_entries); | |
8073 | if (off == SIZE_MAX) | |
8074 | return SIZE_MAX; | |
8075 | ||
8076 | #ifdef CONFIG_SMP | |
8077 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8078 | if (off == 0) | |
8079 | return SIZE_MAX; | |
8080 | #endif | |
8081 | ||
b36200f5 DV |
8082 | if (sq_offset) |
8083 | *sq_offset = off; | |
8084 | ||
75b28aff HV |
8085 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8086 | if (sq_array_size == SIZE_MAX) | |
8087 | return SIZE_MAX; | |
8088 | ||
8089 | if (check_add_overflow(off, sq_array_size, &off)) | |
8090 | return SIZE_MAX; | |
8091 | ||
75b28aff HV |
8092 | return off; |
8093 | } | |
8094 | ||
0a96bbe4 | 8095 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8096 | { |
8097 | int i, j; | |
8098 | ||
8099 | if (!ctx->user_bufs) | |
8100 | return -ENXIO; | |
8101 | ||
8102 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8103 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8104 | ||
8105 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8106 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8107 | |
de293938 | 8108 | if (imu->acct_pages) |
26bfa89e | 8109 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8110 | kvfree(imu->bvec); |
edafccee JA |
8111 | imu->nr_bvecs = 0; |
8112 | } | |
8113 | ||
8114 | kfree(ctx->user_bufs); | |
8115 | ctx->user_bufs = NULL; | |
8116 | ctx->nr_user_bufs = 0; | |
8117 | return 0; | |
8118 | } | |
8119 | ||
8120 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8121 | void __user *arg, unsigned index) | |
8122 | { | |
8123 | struct iovec __user *src; | |
8124 | ||
8125 | #ifdef CONFIG_COMPAT | |
8126 | if (ctx->compat) { | |
8127 | struct compat_iovec __user *ciovs; | |
8128 | struct compat_iovec ciov; | |
8129 | ||
8130 | ciovs = (struct compat_iovec __user *) arg; | |
8131 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8132 | return -EFAULT; | |
8133 | ||
d55e5f5b | 8134 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8135 | dst->iov_len = ciov.iov_len; |
8136 | return 0; | |
8137 | } | |
8138 | #endif | |
8139 | src = (struct iovec __user *) arg; | |
8140 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8141 | return -EFAULT; | |
8142 | return 0; | |
8143 | } | |
8144 | ||
de293938 JA |
8145 | /* |
8146 | * Not super efficient, but this is just a registration time. And we do cache | |
8147 | * the last compound head, so generally we'll only do a full search if we don't | |
8148 | * match that one. | |
8149 | * | |
8150 | * We check if the given compound head page has already been accounted, to | |
8151 | * avoid double accounting it. This allows us to account the full size of the | |
8152 | * page, not just the constituent pages of a huge page. | |
8153 | */ | |
8154 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8155 | int nr_pages, struct page *hpage) | |
8156 | { | |
8157 | int i, j; | |
8158 | ||
8159 | /* check current page array */ | |
8160 | for (i = 0; i < nr_pages; i++) { | |
8161 | if (!PageCompound(pages[i])) | |
8162 | continue; | |
8163 | if (compound_head(pages[i]) == hpage) | |
8164 | return true; | |
8165 | } | |
8166 | ||
8167 | /* check previously registered pages */ | |
8168 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8169 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8170 | ||
8171 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8172 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8173 | continue; | |
8174 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8175 | return true; | |
8176 | } | |
8177 | } | |
8178 | ||
8179 | return false; | |
8180 | } | |
8181 | ||
8182 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8183 | int nr_pages, struct io_mapped_ubuf *imu, | |
8184 | struct page **last_hpage) | |
8185 | { | |
8186 | int i, ret; | |
8187 | ||
8188 | for (i = 0; i < nr_pages; i++) { | |
8189 | if (!PageCompound(pages[i])) { | |
8190 | imu->acct_pages++; | |
8191 | } else { | |
8192 | struct page *hpage; | |
8193 | ||
8194 | hpage = compound_head(pages[i]); | |
8195 | if (hpage == *last_hpage) | |
8196 | continue; | |
8197 | *last_hpage = hpage; | |
8198 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8199 | continue; | |
8200 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8201 | } | |
8202 | } | |
8203 | ||
8204 | if (!imu->acct_pages) | |
8205 | return 0; | |
8206 | ||
26bfa89e | 8207 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8208 | if (ret) |
8209 | imu->acct_pages = 0; | |
8210 | return ret; | |
8211 | } | |
8212 | ||
0a96bbe4 BM |
8213 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8214 | struct io_mapped_ubuf *imu, | |
8215 | struct page **last_hpage) | |
edafccee JA |
8216 | { |
8217 | struct vm_area_struct **vmas = NULL; | |
8218 | struct page **pages = NULL; | |
0a96bbe4 BM |
8219 | unsigned long off, start, end, ubuf; |
8220 | size_t size; | |
8221 | int ret, pret, nr_pages, i; | |
8222 | ||
8223 | ubuf = (unsigned long) iov->iov_base; | |
8224 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8225 | start = ubuf >> PAGE_SHIFT; | |
8226 | nr_pages = end - start; | |
8227 | ||
8228 | ret = -ENOMEM; | |
8229 | ||
8230 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8231 | if (!pages) | |
8232 | goto done; | |
8233 | ||
8234 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8235 | GFP_KERNEL); | |
8236 | if (!vmas) | |
8237 | goto done; | |
edafccee | 8238 | |
0a96bbe4 BM |
8239 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8240 | GFP_KERNEL); | |
8241 | if (!imu->bvec) | |
8242 | goto done; | |
8243 | ||
8244 | ret = 0; | |
8245 | mmap_read_lock(current->mm); | |
8246 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8247 | pages, vmas); | |
8248 | if (pret == nr_pages) { | |
8249 | /* don't support file backed memory */ | |
8250 | for (i = 0; i < nr_pages; i++) { | |
8251 | struct vm_area_struct *vma = vmas[i]; | |
8252 | ||
8253 | if (vma->vm_file && | |
8254 | !is_file_hugepages(vma->vm_file)) { | |
8255 | ret = -EOPNOTSUPP; | |
8256 | break; | |
8257 | } | |
8258 | } | |
8259 | } else { | |
8260 | ret = pret < 0 ? pret : -EFAULT; | |
8261 | } | |
8262 | mmap_read_unlock(current->mm); | |
8263 | if (ret) { | |
8264 | /* | |
8265 | * if we did partial map, or found file backed vmas, | |
8266 | * release any pages we did get | |
8267 | */ | |
8268 | if (pret > 0) | |
8269 | unpin_user_pages(pages, pret); | |
8270 | kvfree(imu->bvec); | |
8271 | goto done; | |
8272 | } | |
8273 | ||
8274 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8275 | if (ret) { | |
8276 | unpin_user_pages(pages, pret); | |
8277 | kvfree(imu->bvec); | |
8278 | goto done; | |
8279 | } | |
8280 | ||
8281 | off = ubuf & ~PAGE_MASK; | |
8282 | size = iov->iov_len; | |
8283 | for (i = 0; i < nr_pages; i++) { | |
8284 | size_t vec_len; | |
8285 | ||
8286 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8287 | imu->bvec[i].bv_page = pages[i]; | |
8288 | imu->bvec[i].bv_len = vec_len; | |
8289 | imu->bvec[i].bv_offset = off; | |
8290 | off = 0; | |
8291 | size -= vec_len; | |
8292 | } | |
8293 | /* store original address for later verification */ | |
8294 | imu->ubuf = ubuf; | |
8295 | imu->len = iov->iov_len; | |
8296 | imu->nr_bvecs = nr_pages; | |
8297 | ret = 0; | |
8298 | done: | |
8299 | kvfree(pages); | |
8300 | kvfree(vmas); | |
8301 | return ret; | |
8302 | } | |
8303 | ||
2b358604 | 8304 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8305 | { |
edafccee JA |
8306 | if (ctx->user_bufs) |
8307 | return -EBUSY; | |
8308 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8309 | return -EINVAL; | |
8310 | ||
8311 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8312 | GFP_KERNEL); | |
8313 | if (!ctx->user_bufs) | |
8314 | return -ENOMEM; | |
8315 | ||
2b358604 BM |
8316 | return 0; |
8317 | } | |
edafccee | 8318 | |
2b358604 BM |
8319 | static int io_buffer_validate(struct iovec *iov) |
8320 | { | |
8321 | /* | |
8322 | * Don't impose further limits on the size and buffer | |
8323 | * constraints here, we'll -EINVAL later when IO is | |
8324 | * submitted if they are wrong. | |
8325 | */ | |
8326 | if (!iov->iov_base || !iov->iov_len) | |
8327 | return -EFAULT; | |
edafccee | 8328 | |
2b358604 BM |
8329 | /* arbitrary limit, but we need something */ |
8330 | if (iov->iov_len > SZ_1G) | |
8331 | return -EFAULT; | |
edafccee | 8332 | |
2b358604 BM |
8333 | return 0; |
8334 | } | |
edafccee | 8335 | |
2b358604 BM |
8336 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8337 | unsigned int nr_args) | |
8338 | { | |
8339 | int i, ret; | |
8340 | struct iovec iov; | |
8341 | struct page *last_hpage = NULL; | |
edafccee | 8342 | |
2b358604 BM |
8343 | ret = io_buffers_map_alloc(ctx, nr_args); |
8344 | if (ret) | |
8345 | return ret; | |
edafccee | 8346 | |
edafccee JA |
8347 | for (i = 0; i < nr_args; i++) { |
8348 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8349 | |
edafccee JA |
8350 | ret = io_copy_iov(ctx, &iov, arg, i); |
8351 | if (ret) | |
0a96bbe4 | 8352 | break; |
de293938 | 8353 | |
2b358604 BM |
8354 | ret = io_buffer_validate(&iov); |
8355 | if (ret) | |
0a96bbe4 | 8356 | break; |
edafccee | 8357 | |
0a96bbe4 BM |
8358 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8359 | if (ret) | |
8360 | break; | |
edafccee JA |
8361 | |
8362 | ctx->nr_user_bufs++; | |
8363 | } | |
0a96bbe4 BM |
8364 | |
8365 | if (ret) | |
8366 | io_sqe_buffers_unregister(ctx); | |
8367 | ||
edafccee JA |
8368 | return ret; |
8369 | } | |
8370 | ||
9b402849 JA |
8371 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8372 | { | |
8373 | __s32 __user *fds = arg; | |
8374 | int fd; | |
8375 | ||
8376 | if (ctx->cq_ev_fd) | |
8377 | return -EBUSY; | |
8378 | ||
8379 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8380 | return -EFAULT; | |
8381 | ||
8382 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8383 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8384 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8385 | ctx->cq_ev_fd = NULL; | |
8386 | return ret; | |
8387 | } | |
8388 | ||
8389 | return 0; | |
8390 | } | |
8391 | ||
8392 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8393 | { | |
8394 | if (ctx->cq_ev_fd) { | |
8395 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8396 | ctx->cq_ev_fd = NULL; | |
8397 | return 0; | |
8398 | } | |
8399 | ||
8400 | return -ENXIO; | |
8401 | } | |
8402 | ||
5a2e745d JA |
8403 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8404 | { | |
9e15c3a0 JA |
8405 | struct io_buffer *buf; |
8406 | unsigned long index; | |
8407 | ||
8408 | xa_for_each(&ctx->io_buffers, index, buf) | |
8409 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8410 | } |
8411 | ||
68e68ee6 | 8412 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8413 | { |
68e68ee6 | 8414 | struct io_kiocb *req, *nxt; |
1b4c351f | 8415 | |
68e68ee6 JA |
8416 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8417 | if (tsk && req->task != tsk) | |
8418 | continue; | |
1b4c351f JA |
8419 | list_del(&req->compl.list); |
8420 | kmem_cache_free(req_cachep, req); | |
8421 | } | |
8422 | } | |
8423 | ||
4010fec4 | 8424 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8425 | { |
bf019da7 | 8426 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8427 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8428 | |
9a4fdbd8 JA |
8429 | mutex_lock(&ctx->uring_lock); |
8430 | ||
8e5c66c4 | 8431 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8432 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8433 | submit_state->reqs); | |
8e5c66c4 PB |
8434 | submit_state->free_reqs = 0; |
8435 | } | |
9a4fdbd8 JA |
8436 | |
8437 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8438 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8439 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8440 | spin_unlock_irq(&ctx->completion_lock); |
8441 | ||
e5547d2c PB |
8442 | io_req_cache_free(&cs->free_list, NULL); |
8443 | ||
9a4fdbd8 JA |
8444 | mutex_unlock(&ctx->uring_lock); |
8445 | } | |
8446 | ||
2b188cc1 JA |
8447 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8448 | { | |
04fc6c80 PB |
8449 | /* |
8450 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8451 | * and they are free to do so while still holding uring_lock or |
8452 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8453 | */ |
8454 | mutex_lock(&ctx->uring_lock); | |
8455 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8456 | spin_lock_irq(&ctx->completion_lock); |
8457 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8458 | |
37d1e2e3 | 8459 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8460 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8461 | |
37d1e2e3 | 8462 | if (ctx->mm_account) { |
2aede0e4 JA |
8463 | mmdrop(ctx->mm_account); |
8464 | ctx->mm_account = NULL; | |
30975825 | 8465 | } |
def596e9 | 8466 | |
8bad28d8 | 8467 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8468 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8469 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8470 | io_eventfd_unregister(ctx); |
5a2e745d | 8471 | io_destroy_buffers(ctx); |
def596e9 | 8472 | |
2b188cc1 | 8473 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8474 | if (ctx->ring_sock) { |
8475 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8476 | sock_release(ctx->ring_sock); |
355e8d26 | 8477 | } |
2b188cc1 JA |
8478 | #endif |
8479 | ||
75b28aff | 8480 | io_mem_free(ctx->rings); |
2b188cc1 | 8481 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8482 | |
8483 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8484 | free_uid(ctx->user); |
4010fec4 | 8485 | io_req_caches_free(ctx); |
e941894e JA |
8486 | if (ctx->hash_map) |
8487 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8488 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8489 | kfree(ctx); |
8490 | } | |
8491 | ||
8492 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8493 | { | |
8494 | struct io_ring_ctx *ctx = file->private_data; | |
8495 | __poll_t mask = 0; | |
8496 | ||
8497 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8498 | /* |
8499 | * synchronizes with barrier from wq_has_sleeper call in | |
8500 | * io_commit_cqring | |
8501 | */ | |
2b188cc1 | 8502 | smp_rmb(); |
90554200 | 8503 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8504 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8505 | |
8506 | /* | |
8507 | * Don't flush cqring overflow list here, just do a simple check. | |
8508 | * Otherwise there could possible be ABBA deadlock: | |
8509 | * CPU0 CPU1 | |
8510 | * ---- ---- | |
8511 | * lock(&ctx->uring_lock); | |
8512 | * lock(&ep->mtx); | |
8513 | * lock(&ctx->uring_lock); | |
8514 | * lock(&ep->mtx); | |
8515 | * | |
8516 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8517 | * pushs them to do the flush. | |
8518 | */ | |
8519 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8520 | mask |= EPOLLIN | EPOLLRDNORM; |
8521 | ||
8522 | return mask; | |
8523 | } | |
8524 | ||
8525 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8526 | { | |
8527 | struct io_ring_ctx *ctx = file->private_data; | |
8528 | ||
8529 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8530 | } | |
8531 | ||
0bead8cd | 8532 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8533 | { |
4379bf8b | 8534 | const struct cred *creds; |
071698e1 | 8535 | |
61cf9370 | 8536 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8537 | if (creds) { |
8538 | put_cred(creds); | |
0bead8cd | 8539 | return 0; |
1e6fa521 | 8540 | } |
0bead8cd YD |
8541 | |
8542 | return -EINVAL; | |
8543 | } | |
8544 | ||
9b465711 | 8545 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8546 | { |
9b465711 | 8547 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8548 | } |
8549 | ||
d56d938b PB |
8550 | struct io_tctx_exit { |
8551 | struct callback_head task_work; | |
8552 | struct completion completion; | |
baf186c4 | 8553 | struct io_ring_ctx *ctx; |
d56d938b PB |
8554 | }; |
8555 | ||
8556 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8557 | { | |
8558 | struct io_uring_task *tctx = current->io_uring; | |
8559 | struct io_tctx_exit *work; | |
8560 | ||
8561 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8562 | /* | |
8563 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8564 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8565 | */ | |
8566 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8567 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8568 | complete(&work->completion); |
8569 | } | |
8570 | ||
85faa7b8 JA |
8571 | static void io_ring_exit_work(struct work_struct *work) |
8572 | { | |
d56d938b | 8573 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8574 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8575 | struct io_tctx_exit exit; |
8576 | struct io_tctx_node *node; | |
8577 | int ret; | |
85faa7b8 | 8578 | |
a185f1db PB |
8579 | /* prevent SQPOLL from submitting new requests */ |
8580 | if (ctx->sq_data) { | |
8581 | io_sq_thread_park(ctx->sq_data); | |
8582 | list_del_init(&ctx->sqd_list); | |
8583 | io_sqd_update_thread_idle(ctx->sq_data); | |
8584 | io_sq_thread_unpark(ctx->sq_data); | |
8585 | } | |
8586 | ||
56952e91 JA |
8587 | /* |
8588 | * If we're doing polled IO and end up having requests being | |
8589 | * submitted async (out-of-line), then completions can come in while | |
8590 | * we're waiting for refs to drop. We need to reap these manually, | |
8591 | * as nobody else will be looking for them. | |
8592 | */ | |
b2edc0a7 | 8593 | do { |
9936c7c2 | 8594 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8595 | |
8596 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8597 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8598 | |
8599 | mutex_lock(&ctx->uring_lock); | |
8600 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8601 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8602 | ||
d56d938b PB |
8603 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8604 | ctx_node); | |
baf186c4 | 8605 | exit.ctx = ctx; |
d56d938b PB |
8606 | init_completion(&exit.completion); |
8607 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8608 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8609 | if (WARN_ON_ONCE(ret)) | |
8610 | continue; | |
8611 | wake_up_process(node->task); | |
8612 | ||
8613 | mutex_unlock(&ctx->uring_lock); | |
8614 | wait_for_completion(&exit.completion); | |
8615 | cond_resched(); | |
8616 | mutex_lock(&ctx->uring_lock); | |
8617 | } | |
8618 | mutex_unlock(&ctx->uring_lock); | |
8619 | ||
85faa7b8 JA |
8620 | io_ring_ctx_free(ctx); |
8621 | } | |
8622 | ||
80c4cbdb PB |
8623 | /* Returns true if we found and killed one or more timeouts */ |
8624 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8625 | struct files_struct *files) | |
8626 | { | |
8627 | struct io_kiocb *req, *tmp; | |
8628 | int canceled = 0; | |
8629 | ||
8630 | spin_lock_irq(&ctx->completion_lock); | |
8631 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8632 | if (io_match_task(req, tsk, files)) { | |
8633 | io_kill_timeout(req, -ECANCELED); | |
8634 | canceled++; | |
8635 | } | |
8636 | } | |
51520426 PB |
8637 | if (canceled != 0) |
8638 | io_commit_cqring(ctx); | |
80c4cbdb | 8639 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8640 | if (canceled != 0) |
8641 | io_cqring_ev_posted(ctx); | |
8642 | return canceled != 0; | |
8643 | } | |
8644 | ||
2b188cc1 JA |
8645 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8646 | { | |
61cf9370 MWO |
8647 | unsigned long index; |
8648 | struct creds *creds; | |
8649 | ||
2b188cc1 JA |
8650 | mutex_lock(&ctx->uring_lock); |
8651 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8652 | /* if force is set, the ring is going away. always drop after that */ |
8653 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8654 | if (ctx->rings) |
6c503150 | 8655 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8656 | xa_for_each(&ctx->personalities, index, creds) |
8657 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8658 | mutex_unlock(&ctx->uring_lock); |
8659 | ||
6b81928d PB |
8660 | io_kill_timeouts(ctx, NULL, NULL); |
8661 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8662 | |
15dff286 | 8663 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8664 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8665 | |
85faa7b8 | 8666 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8667 | /* |
8668 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8669 | * if we're exiting a ton of rings at the same time. It just adds | |
8670 | * noise and overhead, there's no discernable change in runtime | |
8671 | * over using system_wq. | |
8672 | */ | |
8673 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8674 | } |
8675 | ||
8676 | static int io_uring_release(struct inode *inode, struct file *file) | |
8677 | { | |
8678 | struct io_ring_ctx *ctx = file->private_data; | |
8679 | ||
8680 | file->private_data = NULL; | |
8681 | io_ring_ctx_wait_and_kill(ctx); | |
8682 | return 0; | |
8683 | } | |
8684 | ||
f6edbabb PB |
8685 | struct io_task_cancel { |
8686 | struct task_struct *task; | |
8687 | struct files_struct *files; | |
8688 | }; | |
f254ac04 | 8689 | |
f6edbabb | 8690 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8691 | { |
9a472ef7 | 8692 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8693 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8694 | bool ret; |
8695 | ||
f6edbabb | 8696 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8697 | unsigned long flags; |
8698 | struct io_ring_ctx *ctx = req->ctx; | |
8699 | ||
8700 | /* protect against races with linked timeouts */ | |
8701 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8702 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8703 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8704 | } else { | |
f6edbabb | 8705 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8706 | } |
8707 | return ret; | |
b711d4ea JA |
8708 | } |
8709 | ||
e1915f76 | 8710 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8711 | struct task_struct *task, |
b7ddce3c PB |
8712 | struct files_struct *files) |
8713 | { | |
e1915f76 | 8714 | struct io_defer_entry *de; |
b7ddce3c PB |
8715 | LIST_HEAD(list); |
8716 | ||
8717 | spin_lock_irq(&ctx->completion_lock); | |
8718 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8719 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8720 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8721 | break; | |
8722 | } | |
8723 | } | |
8724 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8725 | if (list_empty(&list)) |
8726 | return false; | |
b7ddce3c PB |
8727 | |
8728 | while (!list_empty(&list)) { | |
8729 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8730 | list_del_init(&de->list); | |
8731 | req_set_fail_links(de->req); | |
8732 | io_put_req(de->req); | |
8733 | io_req_complete(de->req, -ECANCELED); | |
8734 | kfree(de); | |
8735 | } | |
e1915f76 | 8736 | return true; |
b7ddce3c PB |
8737 | } |
8738 | ||
1b00764f PB |
8739 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8740 | { | |
8741 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8742 | ||
8743 | return req->ctx == data; | |
8744 | } | |
8745 | ||
8746 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8747 | { | |
8748 | struct io_tctx_node *node; | |
8749 | enum io_wq_cancel cret; | |
8750 | bool ret = false; | |
8751 | ||
8752 | mutex_lock(&ctx->uring_lock); | |
8753 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8754 | struct io_uring_task *tctx = node->task->io_uring; | |
8755 | ||
8756 | /* | |
8757 | * io_wq will stay alive while we hold uring_lock, because it's | |
8758 | * killed after ctx nodes, which requires to take the lock. | |
8759 | */ | |
8760 | if (!tctx || !tctx->io_wq) | |
8761 | continue; | |
8762 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8763 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8764 | } | |
8765 | mutex_unlock(&ctx->uring_lock); | |
8766 | ||
8767 | return ret; | |
8768 | } | |
8769 | ||
9936c7c2 PB |
8770 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8771 | struct task_struct *task, | |
8772 | struct files_struct *files) | |
8773 | { | |
8774 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8775 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8776 | |
8777 | while (1) { | |
8778 | enum io_wq_cancel cret; | |
8779 | bool ret = false; | |
8780 | ||
1b00764f PB |
8781 | if (!task) { |
8782 | ret |= io_uring_try_cancel_iowq(ctx); | |
8783 | } else if (tctx && tctx->io_wq) { | |
8784 | /* | |
8785 | * Cancels requests of all rings, not only @ctx, but | |
8786 | * it's fine as the task is in exit/exec. | |
8787 | */ | |
5aa75ed5 | 8788 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8789 | &cancel, true); |
8790 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8791 | } | |
8792 | ||
8793 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8794 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8795 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8796 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8797 | io_iopoll_try_reap_events(ctx); | |
8798 | ret = true; | |
8799 | } | |
8800 | } | |
8801 | ||
e1915f76 | 8802 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8803 | ret |= io_poll_remove_all(ctx, task, files); |
8804 | ret |= io_kill_timeouts(ctx, task, files); | |
8805 | ret |= io_run_task_work(); | |
ba50a036 | 8806 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8807 | io_cqring_overflow_flush(ctx, true, task, files); |
8808 | if (!ret) | |
8809 | break; | |
8810 | cond_resched(); | |
8811 | } | |
8812 | } | |
8813 | ||
ca70f00b PB |
8814 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8815 | struct task_struct *task, | |
8816 | struct files_struct *files) | |
8817 | { | |
8818 | struct io_kiocb *req; | |
8819 | int cnt = 0; | |
8820 | ||
8821 | spin_lock_irq(&ctx->inflight_lock); | |
8822 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8823 | cnt += io_match_task(req, task, files); | |
8824 | spin_unlock_irq(&ctx->inflight_lock); | |
8825 | return cnt; | |
8826 | } | |
8827 | ||
b52fda00 | 8828 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8829 | struct task_struct *task, |
fcb323cc JA |
8830 | struct files_struct *files) |
8831 | { | |
fcb323cc | 8832 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8833 | DEFINE_WAIT(wait); |
ca70f00b | 8834 | int inflight; |
fcb323cc | 8835 | |
ca70f00b PB |
8836 | inflight = io_uring_count_inflight(ctx, task, files); |
8837 | if (!inflight) | |
fcb323cc | 8838 | break; |
f6edbabb | 8839 | |
9936c7c2 | 8840 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8841 | |
8842 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8843 | TASK_UNINTERRUPTIBLE); | |
8844 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8845 | schedule(); | |
c98de08c | 8846 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8847 | } |
0f212204 JA |
8848 | } |
8849 | ||
0f212204 JA |
8850 | /* |
8851 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8852 | */ | |
baf186c4 | 8853 | static int io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8854 | { |
236434c3 | 8855 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8856 | struct io_tctx_node *node; |
a528b04e | 8857 | int ret; |
236434c3 MWO |
8858 | |
8859 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8860 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8861 | if (unlikely(ret)) |
8862 | return ret; | |
236434c3 | 8863 | tctx = current->io_uring; |
0f212204 | 8864 | } |
baf186c4 PB |
8865 | if (tctx->last != ctx) { |
8866 | void *old = xa_load(&tctx->xa, (unsigned long)ctx); | |
0f212204 | 8867 | |
236434c3 | 8868 | if (!old) { |
13bf43f5 PB |
8869 | node = kmalloc(sizeof(*node), GFP_KERNEL); |
8870 | if (!node) | |
8871 | return -ENOMEM; | |
8872 | node->ctx = ctx; | |
13bf43f5 PB |
8873 | node->task = current; |
8874 | ||
baf186c4 | 8875 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
13bf43f5 | 8876 | node, GFP_KERNEL)); |
a528b04e | 8877 | if (ret) { |
13bf43f5 | 8878 | kfree(node); |
a528b04e PB |
8879 | return ret; |
8880 | } | |
13bf43f5 PB |
8881 | |
8882 | mutex_lock(&ctx->uring_lock); | |
8883 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8884 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8885 | } |
baf186c4 | 8886 | tctx->last = ctx; |
0f212204 | 8887 | } |
0f212204 JA |
8888 | return 0; |
8889 | } | |
8890 | ||
8891 | /* | |
8892 | * Remove this io_uring_file -> task mapping. | |
8893 | */ | |
2941267b | 8894 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8895 | { |
8896 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8897 | struct io_tctx_node *node; |
2941267b | 8898 | |
eebd2e37 PB |
8899 | if (!tctx) |
8900 | return; | |
13bf43f5 PB |
8901 | node = xa_erase(&tctx->xa, index); |
8902 | if (!node) | |
2941267b | 8903 | return; |
0f212204 | 8904 | |
13bf43f5 PB |
8905 | WARN_ON_ONCE(current != node->task); |
8906 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8907 | ||
8908 | mutex_lock(&node->ctx->uring_lock); | |
8909 | list_del(&node->ctx_node); | |
8910 | mutex_unlock(&node->ctx->uring_lock); | |
8911 | ||
baf186c4 | 8912 | if (tctx->last == node->ctx) |
0f212204 | 8913 | tctx->last = NULL; |
13bf43f5 | 8914 | kfree(node); |
0f212204 JA |
8915 | } |
8916 | ||
8452d4a6 | 8917 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8918 | { |
13bf43f5 | 8919 | struct io_tctx_node *node; |
de7f1d9e PB |
8920 | unsigned long index; |
8921 | ||
13bf43f5 | 8922 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8923 | io_uring_del_task_file(index); |
8452d4a6 PB |
8924 | if (tctx->io_wq) { |
8925 | io_wq_put_and_exit(tctx->io_wq); | |
8926 | tctx->io_wq = NULL; | |
8927 | } | |
de7f1d9e PB |
8928 | } |
8929 | ||
521d6a73 PB |
8930 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8931 | { | |
8932 | return percpu_counter_sum(&tctx->inflight); | |
8933 | } | |
8934 | ||
8935 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8936 | { | |
8937 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8938 | struct io_ring_ctx *ctx = work->ctx; | |
8939 | struct io_sq_data *sqd = ctx->sq_data; | |
8940 | ||
8941 | if (sqd->thread) | |
8942 | io_uring_cancel_sqpoll(ctx); | |
8943 | complete(&work->completion); | |
8944 | } | |
8945 | ||
8946 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8947 | { | |
8948 | struct io_sq_data *sqd = ctx->sq_data; | |
8949 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8950 | struct task_struct *task; | |
8951 | ||
8952 | io_sq_thread_park(sqd); | |
8953 | list_del_init(&ctx->sqd_list); | |
8954 | io_sqd_update_thread_idle(sqd); | |
8955 | task = sqd->thread; | |
8956 | if (task) { | |
8957 | init_completion(&work.completion); | |
8958 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8959 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8960 | wake_up_process(task); |
8961 | } | |
8962 | io_sq_thread_unpark(sqd); | |
8963 | ||
8964 | if (task) | |
8965 | wait_for_completion(&work.completion); | |
8966 | } | |
8967 | ||
0f212204 JA |
8968 | void __io_uring_files_cancel(struct files_struct *files) |
8969 | { | |
8970 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8971 | struct io_tctx_node *node; |
ce765372 | 8972 | unsigned long index; |
0f212204 JA |
8973 | |
8974 | /* make sure overflow events are dropped */ | |
fdaf083c | 8975 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8976 | xa_for_each(&tctx->xa, index, node) { |
8977 | struct io_ring_ctx *ctx = node->ctx; | |
8978 | ||
8979 | if (ctx->sq_data) { | |
8980 | io_sqpoll_cancel_sync(ctx); | |
8981 | continue; | |
8982 | } | |
8983 | io_uring_cancel_files(ctx, current, files); | |
8984 | if (!files) | |
8985 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8986 | } | |
fdaf083c | 8987 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8988 | |
8452d4a6 PB |
8989 | if (files) |
8990 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8991 | } |
8992 | ||
521d6a73 | 8993 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8994 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8995 | { | |
37d1e2e3 | 8996 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8997 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8998 | s64 inflight; |
8999 | DEFINE_WAIT(wait); | |
fdaf083c | 9000 | |
521d6a73 PB |
9001 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
9002 | ||
0e9ddb39 PB |
9003 | atomic_inc(&tctx->in_idle); |
9004 | do { | |
9005 | /* read completions before cancelations */ | |
9006 | inflight = tctx_inflight(tctx); | |
9007 | if (!inflight) | |
9008 | break; | |
521d6a73 | 9009 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 9010 | |
0e9ddb39 PB |
9011 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9012 | /* | |
9013 | * If we've seen completions, retry without waiting. This | |
9014 | * avoids a race where a completion comes in before we did | |
9015 | * prepare_to_wait(). | |
9016 | */ | |
9017 | if (inflight == tctx_inflight(tctx)) | |
9018 | schedule(); | |
9019 | finish_wait(&tctx->wait, &wait); | |
9020 | } while (1); | |
9021 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9022 | } |
9023 | ||
0f212204 JA |
9024 | /* |
9025 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9026 | * requests. | |
9027 | */ | |
9028 | void __io_uring_task_cancel(void) | |
9029 | { | |
9030 | struct io_uring_task *tctx = current->io_uring; | |
9031 | DEFINE_WAIT(wait); | |
d8a6df10 | 9032 | s64 inflight; |
0f212204 JA |
9033 | |
9034 | /* make sure overflow events are dropped */ | |
fdaf083c | 9035 | atomic_inc(&tctx->in_idle); |
5a978dcf PB |
9036 | __io_uring_files_cancel(NULL); |
9037 | ||
d8a6df10 | 9038 | do { |
0f212204 | 9039 | /* read completions before cancelations */ |
fdaf083c | 9040 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9041 | if (!inflight) |
9042 | break; | |
0f212204 JA |
9043 | __io_uring_files_cancel(NULL); |
9044 | ||
9045 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9046 | ||
9047 | /* | |
a1bb3cd5 PB |
9048 | * If we've seen completions, retry without waiting. This |
9049 | * avoids a race where a completion comes in before we did | |
9050 | * prepare_to_wait(). | |
0f212204 | 9051 | */ |
a1bb3cd5 PB |
9052 | if (inflight == tctx_inflight(tctx)) |
9053 | schedule(); | |
f57555ed | 9054 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9055 | } while (1); |
0f212204 | 9056 | |
fdaf083c | 9057 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9058 | |
8452d4a6 PB |
9059 | io_uring_clean_tctx(tctx); |
9060 | /* all current's requests should be gone, we can kill tctx */ | |
9061 | __io_uring_free(current); | |
44e728b8 PB |
9062 | } |
9063 | ||
6c5c240e RP |
9064 | static void *io_uring_validate_mmap_request(struct file *file, |
9065 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9066 | { |
2b188cc1 | 9067 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9068 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9069 | struct page *page; |
9070 | void *ptr; | |
9071 | ||
9072 | switch (offset) { | |
9073 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9074 | case IORING_OFF_CQ_RING: |
9075 | ptr = ctx->rings; | |
2b188cc1 JA |
9076 | break; |
9077 | case IORING_OFF_SQES: | |
9078 | ptr = ctx->sq_sqes; | |
9079 | break; | |
2b188cc1 | 9080 | default: |
6c5c240e | 9081 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9082 | } |
9083 | ||
9084 | page = virt_to_head_page(ptr); | |
a50b854e | 9085 | if (sz > page_size(page)) |
6c5c240e RP |
9086 | return ERR_PTR(-EINVAL); |
9087 | ||
9088 | return ptr; | |
9089 | } | |
9090 | ||
9091 | #ifdef CONFIG_MMU | |
9092 | ||
9093 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9094 | { | |
9095 | size_t sz = vma->vm_end - vma->vm_start; | |
9096 | unsigned long pfn; | |
9097 | void *ptr; | |
9098 | ||
9099 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9100 | if (IS_ERR(ptr)) | |
9101 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9102 | |
9103 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9104 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9105 | } | |
9106 | ||
6c5c240e RP |
9107 | #else /* !CONFIG_MMU */ |
9108 | ||
9109 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9110 | { | |
9111 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9112 | } | |
9113 | ||
9114 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9115 | { | |
9116 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9117 | } | |
9118 | ||
9119 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9120 | unsigned long addr, unsigned long len, | |
9121 | unsigned long pgoff, unsigned long flags) | |
9122 | { | |
9123 | void *ptr; | |
9124 | ||
9125 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9126 | if (IS_ERR(ptr)) | |
9127 | return PTR_ERR(ptr); | |
9128 | ||
9129 | return (unsigned long) ptr; | |
9130 | } | |
9131 | ||
9132 | #endif /* !CONFIG_MMU */ | |
9133 | ||
d9d05217 | 9134 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9135 | { |
9136 | DEFINE_WAIT(wait); | |
9137 | ||
9138 | do { | |
9139 | if (!io_sqring_full(ctx)) | |
9140 | break; | |
90554200 JA |
9141 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9142 | ||
9143 | if (!io_sqring_full(ctx)) | |
9144 | break; | |
90554200 JA |
9145 | schedule(); |
9146 | } while (!signal_pending(current)); | |
9147 | ||
9148 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9149 | return 0; |
90554200 JA |
9150 | } |
9151 | ||
c73ebb68 HX |
9152 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9153 | struct __kernel_timespec __user **ts, | |
9154 | const sigset_t __user **sig) | |
9155 | { | |
9156 | struct io_uring_getevents_arg arg; | |
9157 | ||
9158 | /* | |
9159 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9160 | * is just a pointer to the sigset_t. | |
9161 | */ | |
9162 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9163 | *sig = (const sigset_t __user *) argp; | |
9164 | *ts = NULL; | |
9165 | return 0; | |
9166 | } | |
9167 | ||
9168 | /* | |
9169 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9170 | * timespec and sigset_t pointers if good. | |
9171 | */ | |
9172 | if (*argsz != sizeof(arg)) | |
9173 | return -EINVAL; | |
9174 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9175 | return -EFAULT; | |
9176 | *sig = u64_to_user_ptr(arg.sigmask); | |
9177 | *argsz = arg.sigmask_sz; | |
9178 | *ts = u64_to_user_ptr(arg.ts); | |
9179 | return 0; | |
9180 | } | |
9181 | ||
2b188cc1 | 9182 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9183 | u32, min_complete, u32, flags, const void __user *, argp, |
9184 | size_t, argsz) | |
2b188cc1 JA |
9185 | { |
9186 | struct io_ring_ctx *ctx; | |
9187 | long ret = -EBADF; | |
9188 | int submitted = 0; | |
9189 | struct fd f; | |
9190 | ||
4c6e277c | 9191 | io_run_task_work(); |
b41e9852 | 9192 | |
90554200 | 9193 | if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
c73ebb68 | 9194 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)) |
2b188cc1 JA |
9195 | return -EINVAL; |
9196 | ||
9197 | f = fdget(fd); | |
9198 | if (!f.file) | |
9199 | return -EBADF; | |
9200 | ||
9201 | ret = -EOPNOTSUPP; | |
9202 | if (f.file->f_op != &io_uring_fops) | |
9203 | goto out_fput; | |
9204 | ||
9205 | ret = -ENXIO; | |
9206 | ctx = f.file->private_data; | |
9207 | if (!percpu_ref_tryget(&ctx->refs)) | |
9208 | goto out_fput; | |
9209 | ||
7e84e1c7 SG |
9210 | ret = -EBADFD; |
9211 | if (ctx->flags & IORING_SETUP_R_DISABLED) | |
9212 | goto out; | |
9213 | ||
6c271ce2 JA |
9214 | /* |
9215 | * For SQ polling, the thread will do all submissions and completions. | |
9216 | * Just return the requested submit count, and wake the thread if | |
9217 | * we were asked to. | |
9218 | */ | |
b2a9eada | 9219 | ret = 0; |
6c271ce2 | 9220 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9221 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9222 | |
d9d05217 | 9223 | ret = -EOWNERDEAD; |
04147488 SM |
9224 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9225 | goto out; | |
9226 | } | |
6c271ce2 | 9227 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9228 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9229 | if (flags & IORING_ENTER_SQ_WAIT) { |
9230 | ret = io_sqpoll_wait_sq(ctx); | |
9231 | if (ret) | |
9232 | goto out; | |
9233 | } | |
6c271ce2 | 9234 | submitted = to_submit; |
b2a9eada | 9235 | } else if (to_submit) { |
baf186c4 | 9236 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9237 | if (unlikely(ret)) |
9238 | goto out; | |
2b188cc1 | 9239 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9240 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9241 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9242 | |
9243 | if (submitted != to_submit) | |
9244 | goto out; | |
2b188cc1 JA |
9245 | } |
9246 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9247 | const sigset_t __user *sig; |
9248 | struct __kernel_timespec __user *ts; | |
9249 | ||
9250 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9251 | if (unlikely(ret)) | |
9252 | goto out; | |
9253 | ||
2b188cc1 JA |
9254 | min_complete = min(min_complete, ctx->cq_entries); |
9255 | ||
32b2244a XW |
9256 | /* |
9257 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9258 | * space applications don't need to do io completion events | |
9259 | * polling again, they can rely on io_sq_thread to do polling | |
9260 | * work, which can reduce cpu usage and uring_lock contention. | |
9261 | */ | |
9262 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9263 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9264 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9265 | } else { |
c73ebb68 | 9266 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9267 | } |
2b188cc1 JA |
9268 | } |
9269 | ||
7c504e65 | 9270 | out: |
6805b32e | 9271 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9272 | out_fput: |
9273 | fdput(f); | |
9274 | return submitted ? submitted : ret; | |
9275 | } | |
9276 | ||
bebdb65e | 9277 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9278 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9279 | const struct cred *cred) | |
87ce955b | 9280 | { |
87ce955b JA |
9281 | struct user_namespace *uns = seq_user_ns(m); |
9282 | struct group_info *gi; | |
9283 | kernel_cap_t cap; | |
9284 | unsigned __capi; | |
9285 | int g; | |
9286 | ||
9287 | seq_printf(m, "%5d\n", id); | |
9288 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9289 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9290 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9291 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9292 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9293 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9294 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9295 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9296 | seq_puts(m, "\n\tGroups:\t"); | |
9297 | gi = cred->group_info; | |
9298 | for (g = 0; g < gi->ngroups; g++) { | |
9299 | seq_put_decimal_ull(m, g ? " " : "", | |
9300 | from_kgid_munged(uns, gi->gid[g])); | |
9301 | } | |
9302 | seq_puts(m, "\n\tCapEff:\t"); | |
9303 | cap = cred->cap_effective; | |
9304 | CAP_FOR_EACH_U32(__capi) | |
9305 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9306 | seq_putc(m, '\n'); | |
9307 | return 0; | |
9308 | } | |
9309 | ||
9310 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9311 | { | |
dbbe9c64 | 9312 | struct io_sq_data *sq = NULL; |
fad8e0de | 9313 | bool has_lock; |
87ce955b JA |
9314 | int i; |
9315 | ||
fad8e0de JA |
9316 | /* |
9317 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9318 | * since fdinfo case grabs it in the opposite direction of normal use | |
9319 | * cases. If we fail to get the lock, we just don't iterate any | |
9320 | * structures that could be going away outside the io_uring mutex. | |
9321 | */ | |
9322 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9323 | ||
5f3f26f9 | 9324 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9325 | sq = ctx->sq_data; |
5f3f26f9 JA |
9326 | if (!sq->thread) |
9327 | sq = NULL; | |
9328 | } | |
dbbe9c64 JQ |
9329 | |
9330 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9331 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9332 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9333 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
ea64ec02 | 9334 | struct file *f = *io_fixed_file_slot(ctx->file_data, i); |
87ce955b | 9335 | |
87ce955b JA |
9336 | if (f) |
9337 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9338 | else | |
9339 | seq_printf(m, "%5u: <none>\n", i); | |
9340 | } | |
9341 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9342 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9343 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9344 | ||
9345 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9346 | (unsigned int) buf->len); | |
9347 | } | |
61cf9370 MWO |
9348 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9349 | unsigned long index; | |
9350 | const struct cred *cred; | |
9351 | ||
87ce955b | 9352 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9353 | xa_for_each(&ctx->personalities, index, cred) |
9354 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9355 | } |
d7718a9d JA |
9356 | seq_printf(m, "PollList:\n"); |
9357 | spin_lock_irq(&ctx->completion_lock); | |
9358 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9359 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9360 | struct io_kiocb *req; | |
9361 | ||
9362 | hlist_for_each_entry(req, list, hash_node) | |
9363 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9364 | req->task->task_works != NULL); | |
9365 | } | |
9366 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9367 | if (has_lock) |
9368 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9369 | } |
9370 | ||
9371 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9372 | { | |
9373 | struct io_ring_ctx *ctx = f->private_data; | |
9374 | ||
9375 | if (percpu_ref_tryget(&ctx->refs)) { | |
9376 | __io_uring_show_fdinfo(ctx, m); | |
9377 | percpu_ref_put(&ctx->refs); | |
9378 | } | |
9379 | } | |
bebdb65e | 9380 | #endif |
87ce955b | 9381 | |
2b188cc1 JA |
9382 | static const struct file_operations io_uring_fops = { |
9383 | .release = io_uring_release, | |
9384 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9385 | #ifndef CONFIG_MMU |
9386 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9387 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9388 | #endif | |
2b188cc1 JA |
9389 | .poll = io_uring_poll, |
9390 | .fasync = io_uring_fasync, | |
bebdb65e | 9391 | #ifdef CONFIG_PROC_FS |
87ce955b | 9392 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9393 | #endif |
2b188cc1 JA |
9394 | }; |
9395 | ||
9396 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9397 | struct io_uring_params *p) | |
9398 | { | |
75b28aff HV |
9399 | struct io_rings *rings; |
9400 | size_t size, sq_array_offset; | |
2b188cc1 | 9401 | |
bd740481 JA |
9402 | /* make sure these are sane, as we already accounted them */ |
9403 | ctx->sq_entries = p->sq_entries; | |
9404 | ctx->cq_entries = p->cq_entries; | |
9405 | ||
75b28aff HV |
9406 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9407 | if (size == SIZE_MAX) | |
9408 | return -EOVERFLOW; | |
9409 | ||
9410 | rings = io_mem_alloc(size); | |
9411 | if (!rings) | |
2b188cc1 JA |
9412 | return -ENOMEM; |
9413 | ||
75b28aff HV |
9414 | ctx->rings = rings; |
9415 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9416 | rings->sq_ring_mask = p->sq_entries - 1; | |
9417 | rings->cq_ring_mask = p->cq_entries - 1; | |
9418 | rings->sq_ring_entries = p->sq_entries; | |
9419 | rings->cq_ring_entries = p->cq_entries; | |
9420 | ctx->sq_mask = rings->sq_ring_mask; | |
9421 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9422 | |
9423 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9424 | if (size == SIZE_MAX) { |
9425 | io_mem_free(ctx->rings); | |
9426 | ctx->rings = NULL; | |
2b188cc1 | 9427 | return -EOVERFLOW; |
eb065d30 | 9428 | } |
2b188cc1 JA |
9429 | |
9430 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9431 | if (!ctx->sq_sqes) { |
9432 | io_mem_free(ctx->rings); | |
9433 | ctx->rings = NULL; | |
2b188cc1 | 9434 | return -ENOMEM; |
eb065d30 | 9435 | } |
2b188cc1 | 9436 | |
2b188cc1 JA |
9437 | return 0; |
9438 | } | |
9439 | ||
9faadcc8 PB |
9440 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9441 | { | |
9442 | int ret, fd; | |
9443 | ||
9444 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9445 | if (fd < 0) | |
9446 | return fd; | |
9447 | ||
baf186c4 | 9448 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9449 | if (ret) { |
9450 | put_unused_fd(fd); | |
9451 | return ret; | |
9452 | } | |
9453 | fd_install(fd, file); | |
9454 | return fd; | |
9455 | } | |
9456 | ||
2b188cc1 JA |
9457 | /* |
9458 | * Allocate an anonymous fd, this is what constitutes the application | |
9459 | * visible backing of an io_uring instance. The application mmaps this | |
9460 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9461 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9462 | */ | |
9faadcc8 | 9463 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9464 | { |
9465 | struct file *file; | |
9faadcc8 | 9466 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9467 | int ret; |
9468 | ||
2b188cc1 JA |
9469 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9470 | &ctx->ring_sock); | |
9471 | if (ret) | |
9faadcc8 | 9472 | return ERR_PTR(ret); |
2b188cc1 JA |
9473 | #endif |
9474 | ||
2b188cc1 JA |
9475 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9476 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9477 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9478 | if (IS_ERR(file)) { |
9479 | sock_release(ctx->ring_sock); | |
9480 | ctx->ring_sock = NULL; | |
9481 | } else { | |
9482 | ctx->ring_sock->file = file; | |
0f212204 | 9483 | } |
2b188cc1 | 9484 | #endif |
9faadcc8 | 9485 | return file; |
2b188cc1 JA |
9486 | } |
9487 | ||
7f13657d XW |
9488 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9489 | struct io_uring_params __user *params) | |
2b188cc1 | 9490 | { |
2b188cc1 | 9491 | struct io_ring_ctx *ctx; |
9faadcc8 | 9492 | struct file *file; |
2b188cc1 JA |
9493 | int ret; |
9494 | ||
8110c1a6 | 9495 | if (!entries) |
2b188cc1 | 9496 | return -EINVAL; |
8110c1a6 JA |
9497 | if (entries > IORING_MAX_ENTRIES) { |
9498 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9499 | return -EINVAL; | |
9500 | entries = IORING_MAX_ENTRIES; | |
9501 | } | |
2b188cc1 JA |
9502 | |
9503 | /* | |
9504 | * Use twice as many entries for the CQ ring. It's possible for the | |
9505 | * application to drive a higher depth than the size of the SQ ring, | |
9506 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9507 | * some flexibility in overcommitting a bit. If the application has |
9508 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9509 | * of CQ ring entries manually. | |
2b188cc1 JA |
9510 | */ |
9511 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9512 | if (p->flags & IORING_SETUP_CQSIZE) { |
9513 | /* | |
9514 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9515 | * to a power-of-two, if it isn't already. We do NOT impose | |
9516 | * any cq vs sq ring sizing. | |
9517 | */ | |
eb2667b3 | 9518 | if (!p->cq_entries) |
33a107f0 | 9519 | return -EINVAL; |
8110c1a6 JA |
9520 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9521 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9522 | return -EINVAL; | |
9523 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9524 | } | |
eb2667b3 JQ |
9525 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9526 | if (p->cq_entries < p->sq_entries) | |
9527 | return -EINVAL; | |
33a107f0 JA |
9528 | } else { |
9529 | p->cq_entries = 2 * p->sq_entries; | |
9530 | } | |
2b188cc1 | 9531 | |
2b188cc1 | 9532 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9533 | if (!ctx) |
2b188cc1 | 9534 | return -ENOMEM; |
2b188cc1 | 9535 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9536 | if (!capable(CAP_IPC_LOCK)) |
9537 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9538 | |
9539 | /* | |
9540 | * This is just grabbed for accounting purposes. When a process exits, | |
9541 | * the mm is exited and dropped before the files, hence we need to hang | |
9542 | * on to this mm purely for the purposes of being able to unaccount | |
9543 | * memory (locked/pinned vm). It's not used for anything else. | |
9544 | */ | |
6b7898eb | 9545 | mmgrab(current->mm); |
2aede0e4 | 9546 | ctx->mm_account = current->mm; |
6b7898eb | 9547 | |
2b188cc1 JA |
9548 | ret = io_allocate_scq_urings(ctx, p); |
9549 | if (ret) | |
9550 | goto err; | |
9551 | ||
7e84e1c7 | 9552 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9553 | if (ret) |
9554 | goto err; | |
9555 | ||
2b188cc1 | 9556 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9557 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9558 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9559 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9560 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9561 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9562 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9563 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9564 | |
9565 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9566 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9567 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9568 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9569 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9570 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9571 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9572 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9573 | |
7f13657d XW |
9574 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9575 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9576 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9577 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9578 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9579 | |
9580 | if (copy_to_user(params, p, sizeof(*p))) { | |
9581 | ret = -EFAULT; | |
9582 | goto err; | |
9583 | } | |
d1719f70 | 9584 | |
9faadcc8 PB |
9585 | file = io_uring_get_file(ctx); |
9586 | if (IS_ERR(file)) { | |
9587 | ret = PTR_ERR(file); | |
9588 | goto err; | |
9589 | } | |
9590 | ||
044c1ab3 JA |
9591 | /* |
9592 | * Install ring fd as the very last thing, so we don't risk someone | |
9593 | * having closed it before we finish setup | |
9594 | */ | |
9faadcc8 PB |
9595 | ret = io_uring_install_fd(ctx, file); |
9596 | if (ret < 0) { | |
9597 | /* fput will clean it up */ | |
9598 | fput(file); | |
9599 | return ret; | |
9600 | } | |
044c1ab3 | 9601 | |
c826bd7a | 9602 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9603 | return ret; |
9604 | err: | |
9605 | io_ring_ctx_wait_and_kill(ctx); | |
9606 | return ret; | |
9607 | } | |
9608 | ||
9609 | /* | |
9610 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9611 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9612 | * params structure passed in. | |
9613 | */ | |
9614 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9615 | { | |
9616 | struct io_uring_params p; | |
2b188cc1 JA |
9617 | int i; |
9618 | ||
9619 | if (copy_from_user(&p, params, sizeof(p))) | |
9620 | return -EFAULT; | |
9621 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9622 | if (p.resv[i]) | |
9623 | return -EINVAL; | |
9624 | } | |
9625 | ||
6c271ce2 | 9626 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9627 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9628 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9629 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9630 | return -EINVAL; |
9631 | ||
7f13657d | 9632 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9633 | } |
9634 | ||
9635 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9636 | struct io_uring_params __user *, params) | |
9637 | { | |
9638 | return io_uring_setup(entries, params); | |
9639 | } | |
9640 | ||
66f4af93 JA |
9641 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9642 | { | |
9643 | struct io_uring_probe *p; | |
9644 | size_t size; | |
9645 | int i, ret; | |
9646 | ||
9647 | size = struct_size(p, ops, nr_args); | |
9648 | if (size == SIZE_MAX) | |
9649 | return -EOVERFLOW; | |
9650 | p = kzalloc(size, GFP_KERNEL); | |
9651 | if (!p) | |
9652 | return -ENOMEM; | |
9653 | ||
9654 | ret = -EFAULT; | |
9655 | if (copy_from_user(p, arg, size)) | |
9656 | goto out; | |
9657 | ret = -EINVAL; | |
9658 | if (memchr_inv(p, 0, size)) | |
9659 | goto out; | |
9660 | ||
9661 | p->last_op = IORING_OP_LAST - 1; | |
9662 | if (nr_args > IORING_OP_LAST) | |
9663 | nr_args = IORING_OP_LAST; | |
9664 | ||
9665 | for (i = 0; i < nr_args; i++) { | |
9666 | p->ops[i].op = i; | |
9667 | if (!io_op_defs[i].not_supported) | |
9668 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9669 | } | |
9670 | p->ops_len = i; | |
9671 | ||
9672 | ret = 0; | |
9673 | if (copy_to_user(arg, p, size)) | |
9674 | ret = -EFAULT; | |
9675 | out: | |
9676 | kfree(p); | |
9677 | return ret; | |
9678 | } | |
9679 | ||
071698e1 JA |
9680 | static int io_register_personality(struct io_ring_ctx *ctx) |
9681 | { | |
4379bf8b | 9682 | const struct cred *creds; |
61cf9370 | 9683 | u32 id; |
1e6fa521 | 9684 | int ret; |
071698e1 | 9685 | |
4379bf8b | 9686 | creds = get_current_cred(); |
1e6fa521 | 9687 | |
61cf9370 MWO |
9688 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9689 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9690 | if (!ret) | |
9691 | return id; | |
9692 | put_cred(creds); | |
1e6fa521 | 9693 | return ret; |
071698e1 JA |
9694 | } |
9695 | ||
21b55dbc SG |
9696 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9697 | unsigned int nr_args) | |
9698 | { | |
9699 | struct io_uring_restriction *res; | |
9700 | size_t size; | |
9701 | int i, ret; | |
9702 | ||
7e84e1c7 SG |
9703 | /* Restrictions allowed only if rings started disabled */ |
9704 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9705 | return -EBADFD; | |
9706 | ||
21b55dbc | 9707 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9708 | if (ctx->restrictions.registered) |
21b55dbc SG |
9709 | return -EBUSY; |
9710 | ||
9711 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9712 | return -EINVAL; | |
9713 | ||
9714 | size = array_size(nr_args, sizeof(*res)); | |
9715 | if (size == SIZE_MAX) | |
9716 | return -EOVERFLOW; | |
9717 | ||
9718 | res = memdup_user(arg, size); | |
9719 | if (IS_ERR(res)) | |
9720 | return PTR_ERR(res); | |
9721 | ||
9722 | ret = 0; | |
9723 | ||
9724 | for (i = 0; i < nr_args; i++) { | |
9725 | switch (res[i].opcode) { | |
9726 | case IORING_RESTRICTION_REGISTER_OP: | |
9727 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9728 | ret = -EINVAL; | |
9729 | goto out; | |
9730 | } | |
9731 | ||
9732 | __set_bit(res[i].register_op, | |
9733 | ctx->restrictions.register_op); | |
9734 | break; | |
9735 | case IORING_RESTRICTION_SQE_OP: | |
9736 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9737 | ret = -EINVAL; | |
9738 | goto out; | |
9739 | } | |
9740 | ||
9741 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9742 | break; | |
9743 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9744 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9745 | break; | |
9746 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9747 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9748 | break; | |
9749 | default: | |
9750 | ret = -EINVAL; | |
9751 | goto out; | |
9752 | } | |
9753 | } | |
9754 | ||
9755 | out: | |
9756 | /* Reset all restrictions if an error happened */ | |
9757 | if (ret != 0) | |
9758 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9759 | else | |
7e84e1c7 | 9760 | ctx->restrictions.registered = true; |
21b55dbc SG |
9761 | |
9762 | kfree(res); | |
9763 | return ret; | |
9764 | } | |
9765 | ||
7e84e1c7 SG |
9766 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9767 | { | |
9768 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9769 | return -EBADFD; | |
9770 | ||
9771 | if (ctx->restrictions.registered) | |
9772 | ctx->restricted = 1; | |
9773 | ||
0298ef96 PB |
9774 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9775 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9776 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9777 | return 0; |
9778 | } | |
9779 | ||
071698e1 JA |
9780 | static bool io_register_op_must_quiesce(int op) |
9781 | { | |
9782 | switch (op) { | |
9783 | case IORING_UNREGISTER_FILES: | |
9784 | case IORING_REGISTER_FILES_UPDATE: | |
9785 | case IORING_REGISTER_PROBE: | |
9786 | case IORING_REGISTER_PERSONALITY: | |
9787 | case IORING_UNREGISTER_PERSONALITY: | |
9788 | return false; | |
9789 | default: | |
9790 | return true; | |
9791 | } | |
9792 | } | |
9793 | ||
edafccee JA |
9794 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9795 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9796 | __releases(ctx->uring_lock) |
9797 | __acquires(ctx->uring_lock) | |
edafccee JA |
9798 | { |
9799 | int ret; | |
9800 | ||
35fa71a0 JA |
9801 | /* |
9802 | * We're inside the ring mutex, if the ref is already dying, then | |
9803 | * someone else killed the ctx or is already going through | |
9804 | * io_uring_register(). | |
9805 | */ | |
9806 | if (percpu_ref_is_dying(&ctx->refs)) | |
9807 | return -ENXIO; | |
9808 | ||
071698e1 | 9809 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9810 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9811 | |
05f3fb3c JA |
9812 | /* |
9813 | * Drop uring mutex before waiting for references to exit. If | |
9814 | * another thread is currently inside io_uring_enter() it might | |
9815 | * need to grab the uring_lock to make progress. If we hold it | |
9816 | * here across the drain wait, then we can deadlock. It's safe | |
9817 | * to drop the mutex here, since no new references will come in | |
9818 | * after we've killed the percpu ref. | |
9819 | */ | |
9820 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9821 | do { |
9822 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9823 | if (!ret) | |
9824 | break; | |
ed6930c9 JA |
9825 | ret = io_run_task_work_sig(); |
9826 | if (ret < 0) | |
9827 | break; | |
af9c1a44 JA |
9828 | } while (1); |
9829 | ||
05f3fb3c | 9830 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9831 | |
c150368b JA |
9832 | if (ret) { |
9833 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9834 | goto out_quiesce; |
9835 | } | |
9836 | } | |
9837 | ||
9838 | if (ctx->restricted) { | |
9839 | if (opcode >= IORING_REGISTER_LAST) { | |
9840 | ret = -EINVAL; | |
9841 | goto out; | |
9842 | } | |
9843 | ||
9844 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9845 | ret = -EACCES; | |
c150368b JA |
9846 | goto out; |
9847 | } | |
05f3fb3c | 9848 | } |
edafccee JA |
9849 | |
9850 | switch (opcode) { | |
9851 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9852 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9853 | break; |
9854 | case IORING_UNREGISTER_BUFFERS: | |
9855 | ret = -EINVAL; | |
9856 | if (arg || nr_args) | |
9857 | break; | |
0a96bbe4 | 9858 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9859 | break; |
6b06314c JA |
9860 | case IORING_REGISTER_FILES: |
9861 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9862 | break; | |
9863 | case IORING_UNREGISTER_FILES: | |
9864 | ret = -EINVAL; | |
9865 | if (arg || nr_args) | |
9866 | break; | |
9867 | ret = io_sqe_files_unregister(ctx); | |
9868 | break; | |
c3a31e60 JA |
9869 | case IORING_REGISTER_FILES_UPDATE: |
9870 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9871 | break; | |
9b402849 | 9872 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9873 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9874 | ret = -EINVAL; |
9875 | if (nr_args != 1) | |
9876 | break; | |
9877 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9878 | if (ret) |
9879 | break; | |
9880 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9881 | ctx->eventfd_async = 1; | |
9882 | else | |
9883 | ctx->eventfd_async = 0; | |
9b402849 JA |
9884 | break; |
9885 | case IORING_UNREGISTER_EVENTFD: | |
9886 | ret = -EINVAL; | |
9887 | if (arg || nr_args) | |
9888 | break; | |
9889 | ret = io_eventfd_unregister(ctx); | |
9890 | break; | |
66f4af93 JA |
9891 | case IORING_REGISTER_PROBE: |
9892 | ret = -EINVAL; | |
9893 | if (!arg || nr_args > 256) | |
9894 | break; | |
9895 | ret = io_probe(ctx, arg, nr_args); | |
9896 | break; | |
071698e1 JA |
9897 | case IORING_REGISTER_PERSONALITY: |
9898 | ret = -EINVAL; | |
9899 | if (arg || nr_args) | |
9900 | break; | |
9901 | ret = io_register_personality(ctx); | |
9902 | break; | |
9903 | case IORING_UNREGISTER_PERSONALITY: | |
9904 | ret = -EINVAL; | |
9905 | if (arg) | |
9906 | break; | |
9907 | ret = io_unregister_personality(ctx, nr_args); | |
9908 | break; | |
7e84e1c7 SG |
9909 | case IORING_REGISTER_ENABLE_RINGS: |
9910 | ret = -EINVAL; | |
9911 | if (arg || nr_args) | |
9912 | break; | |
9913 | ret = io_register_enable_rings(ctx); | |
9914 | break; | |
21b55dbc SG |
9915 | case IORING_REGISTER_RESTRICTIONS: |
9916 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9917 | break; | |
edafccee JA |
9918 | default: |
9919 | ret = -EINVAL; | |
9920 | break; | |
9921 | } | |
9922 | ||
21b55dbc | 9923 | out: |
071698e1 | 9924 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9925 | /* bring the ctx back to life */ |
05f3fb3c | 9926 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9927 | out_quiesce: |
0f158b4c | 9928 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9929 | } |
edafccee JA |
9930 | return ret; |
9931 | } | |
9932 | ||
9933 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9934 | void __user *, arg, unsigned int, nr_args) | |
9935 | { | |
9936 | struct io_ring_ctx *ctx; | |
9937 | long ret = -EBADF; | |
9938 | struct fd f; | |
9939 | ||
9940 | f = fdget(fd); | |
9941 | if (!f.file) | |
9942 | return -EBADF; | |
9943 | ||
9944 | ret = -EOPNOTSUPP; | |
9945 | if (f.file->f_op != &io_uring_fops) | |
9946 | goto out_fput; | |
9947 | ||
9948 | ctx = f.file->private_data; | |
9949 | ||
b6c23dd5 PB |
9950 | io_run_task_work(); |
9951 | ||
edafccee JA |
9952 | mutex_lock(&ctx->uring_lock); |
9953 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9954 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9955 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9956 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9957 | out_fput: |
9958 | fdput(f); | |
9959 | return ret; | |
9960 | } | |
9961 | ||
2b188cc1 JA |
9962 | static int __init io_uring_init(void) |
9963 | { | |
d7f62e82 SM |
9964 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9965 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9966 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9967 | } while (0) | |
9968 | ||
9969 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9970 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9971 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9972 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9973 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9974 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9975 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9976 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9977 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9978 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9979 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9980 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9981 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9982 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9983 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9984 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9985 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9986 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9987 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9988 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9989 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9990 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9991 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9992 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9993 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9994 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9995 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9996 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9997 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9998 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9999 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10000 | |
d3656344 | 10001 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10002 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10003 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10004 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10005 | return 0; |
10006 | }; | |
10007 | __initcall(io_uring_init); |