]>
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; | |
8c3f9cd1 | 656 | u32 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, |
99bc4c38 | 693 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 694 | REQ_F_POLLED_BIT, |
bcda7baa | 695 | REQ_F_BUFFER_SELECTED_BIT, |
5b0bbee4 | 696 | REQ_F_NO_FILE_TABLE_BIT, |
900fad45 | 697 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 698 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 699 | REQ_F_REISSUE_BIT, |
7b29f92d JA |
700 | /* keep async read/write and isreg together and in order */ |
701 | REQ_F_ASYNC_READ_BIT, | |
702 | REQ_F_ASYNC_WRITE_BIT, | |
703 | REQ_F_ISREG_BIT, | |
84557871 JA |
704 | |
705 | /* not a real bit, just to check we're not overflowing the space */ | |
706 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
707 | }; |
708 | ||
709 | enum { | |
710 | /* ctx owns file */ | |
711 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
712 | /* drain existing IO first */ | |
713 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
714 | /* linked sqes */ | |
715 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
716 | /* doesn't sever on completion < 0 */ | |
717 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
718 | /* IOSQE_ASYNC */ | |
719 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
720 | /* IOSQE_BUFFER_SELECT */ |
721 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 722 | |
6b47ee6e PB |
723 | /* fail rest of links */ |
724 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 725 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
726 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
727 | /* read/write uses file position */ | |
728 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
729 | /* must not punt to workers */ | |
730 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 731 | /* has or had linked timeout */ |
6b47ee6e | 732 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
733 | /* needs cleanup */ |
734 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
735 | /* already went through poll handler */ |
736 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
737 | /* buffer already selected */ |
738 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
5b0bbee4 JA |
739 | /* doesn't need file table for this request */ |
740 | REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT), | |
900fad45 PB |
741 | /* linked timeout is active, i.e. prepared by link's head */ |
742 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
743 | /* completion is deferred through io_comp_state */ |
744 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
745 | /* caller should reissue async */ |
746 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
7b29f92d JA |
747 | /* supports async reads */ |
748 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
749 | /* supports async writes */ | |
750 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
751 | /* regular file */ | |
752 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
753 | }; |
754 | ||
755 | struct async_poll { | |
756 | struct io_poll_iocb poll; | |
807abcb0 | 757 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
758 | }; |
759 | ||
7cbf1722 JA |
760 | struct io_task_work { |
761 | struct io_wq_work_node node; | |
762 | task_work_func_t func; | |
763 | }; | |
764 | ||
09bb8394 JA |
765 | /* |
766 | * NOTE! Each of the iocb union members has the file pointer | |
767 | * as the first entry in their struct definition. So you can | |
768 | * access the file pointer through any of the sub-structs, | |
769 | * or directly as just 'ki_filp' in this struct. | |
770 | */ | |
2b188cc1 | 771 | struct io_kiocb { |
221c5eb2 | 772 | union { |
09bb8394 | 773 | struct file *file; |
9adbd45d | 774 | struct io_rw rw; |
221c5eb2 | 775 | struct io_poll_iocb poll; |
018043be | 776 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
777 | struct io_accept accept; |
778 | struct io_sync sync; | |
fbf23849 | 779 | struct io_cancel cancel; |
b29472ee | 780 | struct io_timeout timeout; |
0bdf7a2d | 781 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 782 | struct io_connect connect; |
e47293fd | 783 | struct io_sr_msg sr_msg; |
15b71abe | 784 | struct io_open open; |
b5dba59e | 785 | struct io_close close; |
269bbe5f | 786 | struct io_rsrc_update rsrc_update; |
4840e418 | 787 | struct io_fadvise fadvise; |
c1ca757b | 788 | struct io_madvise madvise; |
3e4827b0 | 789 | struct io_epoll epoll; |
7d67af2c | 790 | struct io_splice splice; |
ddf0322d | 791 | struct io_provide_buf pbuf; |
1d9e1288 | 792 | struct io_statx statx; |
36f4fa68 | 793 | struct io_shutdown shutdown; |
80a261fd | 794 | struct io_rename rename; |
14a1143b | 795 | struct io_unlink unlink; |
3ca405eb PB |
796 | /* use only after cleaning per-op data, see io_clean_op() */ |
797 | struct io_completion compl; | |
221c5eb2 | 798 | }; |
2b188cc1 | 799 | |
e8c2bc1f JA |
800 | /* opcode allocated if it needs to store data for async defer */ |
801 | void *async_data; | |
d625c6ee | 802 | u8 opcode; |
65a6543d XW |
803 | /* polled IO has completed */ |
804 | u8 iopoll_completed; | |
2b188cc1 | 805 | |
4f4eeba8 | 806 | u16 buf_index; |
9cf7c104 | 807 | u32 result; |
4f4eeba8 | 808 | |
010e8e6b PB |
809 | struct io_ring_ctx *ctx; |
810 | unsigned int flags; | |
abc54d63 | 811 | atomic_t refs; |
010e8e6b PB |
812 | struct task_struct *task; |
813 | u64 user_data; | |
d7718a9d | 814 | |
f2f87370 | 815 | struct io_kiocb *link; |
269bbe5f | 816 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 817 | |
d21ffe7e PB |
818 | /* |
819 | * 1. used with ctx->iopoll_list with reads/writes | |
820 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
821 | */ | |
010e8e6b | 822 | struct list_head inflight_entry; |
7cbf1722 JA |
823 | union { |
824 | struct io_task_work io_task_work; | |
825 | struct callback_head task_work; | |
826 | }; | |
010e8e6b PB |
827 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
828 | struct hlist_node hash_node; | |
829 | struct async_poll *apoll; | |
830 | struct io_wq_work work; | |
2b188cc1 | 831 | }; |
05589553 | 832 | |
13bf43f5 PB |
833 | struct io_tctx_node { |
834 | struct list_head ctx_node; | |
835 | struct task_struct *task; | |
13bf43f5 PB |
836 | struct io_ring_ctx *ctx; |
837 | }; | |
838 | ||
27dc8338 PB |
839 | struct io_defer_entry { |
840 | struct list_head list; | |
841 | struct io_kiocb *req; | |
9cf7c104 | 842 | u32 seq; |
2b188cc1 JA |
843 | }; |
844 | ||
d3656344 | 845 | struct io_op_def { |
d3656344 JA |
846 | /* needs req->file assigned */ |
847 | unsigned needs_file : 1; | |
d3656344 JA |
848 | /* hash wq insertion if file is a regular file */ |
849 | unsigned hash_reg_file : 1; | |
850 | /* unbound wq insertion if file is a non-regular file */ | |
851 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
852 | /* opcode is not supported by this kernel */ |
853 | unsigned not_supported : 1; | |
8a72758c JA |
854 | /* set if opcode supports polled "wait" */ |
855 | unsigned pollin : 1; | |
856 | unsigned pollout : 1; | |
bcda7baa JA |
857 | /* op supports buffer selection */ |
858 | unsigned buffer_select : 1; | |
26f0505a PB |
859 | /* do prep async if is going to be punted */ |
860 | unsigned needs_async_setup : 1; | |
27926b68 JA |
861 | /* should block plug */ |
862 | unsigned plug : 1; | |
e8c2bc1f JA |
863 | /* size of async data needed, if any */ |
864 | unsigned short async_size; | |
d3656344 JA |
865 | }; |
866 | ||
0918682b | 867 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
868 | [IORING_OP_NOP] = {}, |
869 | [IORING_OP_READV] = { | |
d3656344 JA |
870 | .needs_file = 1, |
871 | .unbound_nonreg_file = 1, | |
8a72758c | 872 | .pollin = 1, |
4d954c25 | 873 | .buffer_select = 1, |
26f0505a | 874 | .needs_async_setup = 1, |
27926b68 | 875 | .plug = 1, |
e8c2bc1f | 876 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 877 | }, |
0463b6c5 | 878 | [IORING_OP_WRITEV] = { |
d3656344 JA |
879 | .needs_file = 1, |
880 | .hash_reg_file = 1, | |
881 | .unbound_nonreg_file = 1, | |
8a72758c | 882 | .pollout = 1, |
26f0505a | 883 | .needs_async_setup = 1, |
27926b68 | 884 | .plug = 1, |
e8c2bc1f | 885 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 886 | }, |
0463b6c5 | 887 | [IORING_OP_FSYNC] = { |
d3656344 JA |
888 | .needs_file = 1, |
889 | }, | |
0463b6c5 | 890 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
891 | .needs_file = 1, |
892 | .unbound_nonreg_file = 1, | |
8a72758c | 893 | .pollin = 1, |
27926b68 | 894 | .plug = 1, |
e8c2bc1f | 895 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 896 | }, |
0463b6c5 | 897 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
898 | .needs_file = 1, |
899 | .hash_reg_file = 1, | |
900 | .unbound_nonreg_file = 1, | |
8a72758c | 901 | .pollout = 1, |
27926b68 | 902 | .plug = 1, |
e8c2bc1f | 903 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 904 | }, |
0463b6c5 | 905 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
906 | .needs_file = 1, |
907 | .unbound_nonreg_file = 1, | |
908 | }, | |
0463b6c5 PB |
909 | [IORING_OP_POLL_REMOVE] = {}, |
910 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
911 | .needs_file = 1, |
912 | }, | |
0463b6c5 | 913 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
914 | .needs_file = 1, |
915 | .unbound_nonreg_file = 1, | |
8a72758c | 916 | .pollout = 1, |
26f0505a | 917 | .needs_async_setup = 1, |
e8c2bc1f | 918 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 919 | }, |
0463b6c5 | 920 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
921 | .needs_file = 1, |
922 | .unbound_nonreg_file = 1, | |
8a72758c | 923 | .pollin = 1, |
52de1fe1 | 924 | .buffer_select = 1, |
26f0505a | 925 | .needs_async_setup = 1, |
e8c2bc1f | 926 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 927 | }, |
0463b6c5 | 928 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 929 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 930 | }, |
9c8e11b3 PB |
931 | [IORING_OP_TIMEOUT_REMOVE] = { |
932 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 933 | }, |
0463b6c5 | 934 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
935 | .needs_file = 1, |
936 | .unbound_nonreg_file = 1, | |
8a72758c | 937 | .pollin = 1, |
d3656344 | 938 | }, |
0463b6c5 PB |
939 | [IORING_OP_ASYNC_CANCEL] = {}, |
940 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 941 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_CONNECT] = { |
d3656344 JA |
944 | .needs_file = 1, |
945 | .unbound_nonreg_file = 1, | |
8a72758c | 946 | .pollout = 1, |
26f0505a | 947 | .needs_async_setup = 1, |
e8c2bc1f | 948 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 949 | }, |
0463b6c5 | 950 | [IORING_OP_FALLOCATE] = { |
d3656344 | 951 | .needs_file = 1, |
d3656344 | 952 | }, |
44526bed JA |
953 | [IORING_OP_OPENAT] = {}, |
954 | [IORING_OP_CLOSE] = {}, | |
955 | [IORING_OP_FILES_UPDATE] = {}, | |
956 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 957 | [IORING_OP_READ] = { |
3a6820f2 JA |
958 | .needs_file = 1, |
959 | .unbound_nonreg_file = 1, | |
8a72758c | 960 | .pollin = 1, |
bcda7baa | 961 | .buffer_select = 1, |
27926b68 | 962 | .plug = 1, |
e8c2bc1f | 963 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 964 | }, |
0463b6c5 | 965 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
966 | .needs_file = 1, |
967 | .unbound_nonreg_file = 1, | |
8a72758c | 968 | .pollout = 1, |
27926b68 | 969 | .plug = 1, |
e8c2bc1f | 970 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 971 | }, |
0463b6c5 | 972 | [IORING_OP_FADVISE] = { |
4840e418 | 973 | .needs_file = 1, |
c1ca757b | 974 | }, |
44526bed | 975 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 976 | [IORING_OP_SEND] = { |
fddaface JA |
977 | .needs_file = 1, |
978 | .unbound_nonreg_file = 1, | |
8a72758c | 979 | .pollout = 1, |
fddaface | 980 | }, |
0463b6c5 | 981 | [IORING_OP_RECV] = { |
fddaface JA |
982 | .needs_file = 1, |
983 | .unbound_nonreg_file = 1, | |
8a72758c | 984 | .pollin = 1, |
bcda7baa | 985 | .buffer_select = 1, |
fddaface | 986 | }, |
0463b6c5 | 987 | [IORING_OP_OPENAT2] = { |
cebdb986 | 988 | }, |
3e4827b0 JA |
989 | [IORING_OP_EPOLL_CTL] = { |
990 | .unbound_nonreg_file = 1, | |
3e4827b0 | 991 | }, |
7d67af2c PB |
992 | [IORING_OP_SPLICE] = { |
993 | .needs_file = 1, | |
994 | .hash_reg_file = 1, | |
995 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
996 | }, |
997 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 998 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
999 | [IORING_OP_TEE] = { |
1000 | .needs_file = 1, | |
1001 | .hash_reg_file = 1, | |
1002 | .unbound_nonreg_file = 1, | |
1003 | }, | |
36f4fa68 JA |
1004 | [IORING_OP_SHUTDOWN] = { |
1005 | .needs_file = 1, | |
1006 | }, | |
44526bed JA |
1007 | [IORING_OP_RENAMEAT] = {}, |
1008 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1009 | }; |
1010 | ||
7a612350 | 1011 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1012 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1013 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1014 | struct task_struct *task, | |
1015 | struct files_struct *files); | |
37d1e2e3 | 1016 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
269bbe5f | 1017 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node); |
bc9744cd | 1018 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
1ffc5422 | 1019 | struct io_ring_ctx *ctx); |
f2303b1f | 1020 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 1021 | |
23faba36 | 1022 | static bool io_rw_reissue(struct io_kiocb *req); |
78e19bbe | 1023 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1024 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1025 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c40f6379 | 1026 | static void io_double_put_req(struct io_kiocb *req); |
c7dae4ba JA |
1027 | static void io_dismantle_req(struct io_kiocb *req); |
1028 | static void io_put_task(struct task_struct *task, int nr); | |
1029 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 JA |
1030 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1031 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c | 1032 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1033 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1034 | unsigned nr_args); |
3ca405eb | 1035 | static void __io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1036 | static struct file *io_file_get(struct io_submit_state *state, |
1037 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1038 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1039 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1040 | |
907d1df3 | 1041 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1042 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1043 | struct io_ring_ctx *ctx); | |
179ae0d1 | 1044 | static int io_req_prep_async(struct io_kiocb *req); |
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 | ||
abc54d63 JA |
1479 | /* |
1480 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1481 | * see commit f958d7b528b1 for details. | |
1482 | */ | |
1483 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1484 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1485 | ||
de9b4cca JA |
1486 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1487 | { | |
abc54d63 | 1488 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1489 | } |
1490 | ||
1491 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1492 | { | |
abc54d63 JA |
1493 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1494 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1495 | } |
1496 | ||
1497 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1498 | { | |
abc54d63 JA |
1499 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1500 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1501 | } |
1502 | ||
1503 | static inline void req_ref_put(struct io_kiocb *req) | |
1504 | { | |
abc54d63 | 1505 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1506 | } |
1507 | ||
1508 | static inline void req_ref_get(struct io_kiocb *req) | |
1509 | { | |
abc54d63 JA |
1510 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1511 | atomic_inc(&req->refs); | |
de9b4cca JA |
1512 | } |
1513 | ||
8c3f9cd1 PB |
1514 | static void __io_cqring_fill_event(struct io_kiocb *req, long res, |
1515 | unsigned int cflags) | |
2b188cc1 | 1516 | { |
78e19bbe | 1517 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1518 | struct io_uring_cqe *cqe; |
1519 | ||
78e19bbe | 1520 | trace_io_uring_complete(ctx, req->user_data, res); |
51c3ff62 | 1521 | |
2b188cc1 JA |
1522 | /* |
1523 | * If we can't get a cq entry, userspace overflowed the | |
1524 | * submission (by quite a lot). Increment the overflow count in | |
1525 | * the ring. | |
1526 | */ | |
1527 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1528 | if (likely(cqe)) { |
78e19bbe | 1529 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1530 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1531 | WRITE_ONCE(cqe->flags, cflags); |
fdaf083c JA |
1532 | } else if (ctx->cq_overflow_flushed || |
1533 | atomic_read(&req->task->io_uring->in_idle)) { | |
0f212204 JA |
1534 | /* |
1535 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1536 | * then we cannot store the request for later flushing, we need | |
1537 | * to drop it on the floor. | |
1538 | */ | |
2c3bac6d PB |
1539 | ctx->cached_cq_overflow++; |
1540 | WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow); | |
1d7bb1d5 | 1541 | } else { |
ad3eb2c8 JA |
1542 | if (list_empty(&ctx->cq_overflow_list)) { |
1543 | set_bit(0, &ctx->sq_check_overflow); | |
1544 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1545 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1546 | } |
40d8ddd4 | 1547 | io_clean_op(req); |
1d7bb1d5 | 1548 | req->result = res; |
0f7e466b | 1549 | req->compl.cflags = cflags; |
de9b4cca | 1550 | req_ref_get(req); |
40d8ddd4 | 1551 | list_add_tail(&req->compl.list, &ctx->cq_overflow_list); |
2b188cc1 JA |
1552 | } |
1553 | } | |
1554 | ||
bcda7baa JA |
1555 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1556 | { | |
1557 | __io_cqring_fill_event(req, res, 0); | |
1558 | } | |
1559 | ||
7a612350 PB |
1560 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1561 | unsigned int cflags) | |
2b188cc1 | 1562 | { |
78e19bbe | 1563 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1564 | unsigned long flags; |
1565 | ||
1566 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1567 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1568 | /* |
1569 | * If we're the last reference to this request, add to our locked | |
1570 | * free_list cache. | |
1571 | */ | |
de9b4cca | 1572 | if (req_ref_put_and_test(req)) { |
c7dae4ba JA |
1573 | struct io_comp_state *cs = &ctx->submit_state.comp; |
1574 | ||
7a612350 PB |
1575 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1576 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1577 | io_disarm_next(req); | |
1578 | if (req->link) { | |
1579 | io_req_task_queue(req->link); | |
1580 | req->link = NULL; | |
1581 | } | |
1582 | } | |
c7dae4ba JA |
1583 | io_dismantle_req(req); |
1584 | io_put_task(req->task, 1); | |
1585 | list_add(&req->compl.list, &cs->locked_free_list); | |
1586 | cs->locked_free_nr++; | |
180f829f PB |
1587 | } else { |
1588 | if (!percpu_ref_tryget(&ctx->refs)) | |
1589 | req = NULL; | |
1590 | } | |
7a612350 | 1591 | io_commit_cqring(ctx); |
2b188cc1 | 1592 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1593 | |
180f829f PB |
1594 | if (req) { |
1595 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1596 | percpu_ref_put(&ctx->refs); |
180f829f | 1597 | } |
229a7b63 JA |
1598 | } |
1599 | ||
a38d68db | 1600 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1601 | unsigned int cflags) |
229a7b63 | 1602 | { |
a38d68db PB |
1603 | io_clean_op(req); |
1604 | req->result = res; | |
1605 | req->compl.cflags = cflags; | |
e342c807 | 1606 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1607 | } |
1608 | ||
889fca73 PB |
1609 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1610 | long res, unsigned cflags) | |
bcda7baa | 1611 | { |
889fca73 PB |
1612 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1613 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1614 | else |
c7dae4ba | 1615 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1616 | } |
1617 | ||
a38d68db | 1618 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1619 | { |
889fca73 | 1620 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1621 | } |
1622 | ||
f41db273 PB |
1623 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1624 | { | |
1625 | req_set_fail_links(req); | |
1626 | io_put_req(req); | |
1627 | io_req_complete_post(req, res, 0); | |
1628 | } | |
1629 | ||
c7dae4ba | 1630 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1631 | { |
c7dae4ba JA |
1632 | struct io_submit_state *state = &ctx->submit_state; |
1633 | struct io_comp_state *cs = &state->comp; | |
e5d1bc0a | 1634 | struct io_kiocb *req = NULL; |
0ddf92e8 | 1635 | |
c7dae4ba JA |
1636 | /* |
1637 | * If we have more than a batch's worth of requests in our IRQ side | |
1638 | * locked cache, grab the lock and move them over to our submission | |
1639 | * side cache. | |
1640 | */ | |
1641 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) { | |
1642 | spin_lock_irq(&ctx->completion_lock); | |
1643 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1644 | cs->locked_free_nr = 0; | |
1645 | spin_unlock_irq(&ctx->completion_lock); | |
1646 | } | |
0ddf92e8 | 1647 | |
c7dae4ba JA |
1648 | while (!list_empty(&cs->free_list)) { |
1649 | req = list_first_entry(&cs->free_list, struct io_kiocb, | |
1b4c351f JA |
1650 | compl.list); |
1651 | list_del(&req->compl.list); | |
e5d1bc0a PB |
1652 | state->reqs[state->free_reqs++] = req; |
1653 | if (state->free_reqs == ARRAY_SIZE(state->reqs)) | |
1654 | break; | |
1b4c351f JA |
1655 | } |
1656 | ||
e5d1bc0a | 1657 | return req != NULL; |
0ddf92e8 JA |
1658 | } |
1659 | ||
e5d1bc0a | 1660 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1661 | { |
e5d1bc0a PB |
1662 | struct io_submit_state *state = &ctx->submit_state; |
1663 | ||
1664 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1665 | ||
f6b6c7d6 | 1666 | if (!state->free_reqs) { |
291b2821 | 1667 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1668 | int ret; |
1669 | ||
c7dae4ba | 1670 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1671 | goto got_req; |
1672 | ||
bf019da7 PB |
1673 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1674 | state->reqs); | |
fd6fab2c JA |
1675 | |
1676 | /* | |
1677 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1678 | * retry single alloc to be on the safe side. | |
1679 | */ | |
1680 | if (unlikely(ret <= 0)) { | |
1681 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1682 | if (!state->reqs[0]) | |
3893f39f | 1683 | return NULL; |
fd6fab2c JA |
1684 | ret = 1; |
1685 | } | |
291b2821 | 1686 | state->free_reqs = ret; |
2b188cc1 | 1687 | } |
e5d1bc0a | 1688 | got_req: |
291b2821 PB |
1689 | state->free_reqs--; |
1690 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1691 | } |
1692 | ||
8da11c19 PB |
1693 | static inline void io_put_file(struct io_kiocb *req, struct file *file, |
1694 | bool fixed) | |
1695 | { | |
36f72fe2 | 1696 | if (!fixed) |
8da11c19 PB |
1697 | fput(file); |
1698 | } | |
1699 | ||
4edf20f9 | 1700 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1701 | { |
3ca405eb | 1702 | io_clean_op(req); |
929a3af9 | 1703 | |
e8c2bc1f JA |
1704 | if (req->async_data) |
1705 | kfree(req->async_data); | |
8da11c19 PB |
1706 | if (req->file) |
1707 | io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE)); | |
269bbe5f BM |
1708 | if (req->fixed_rsrc_refs) |
1709 | percpu_ref_put(req->fixed_rsrc_refs); | |
003e8dcc JA |
1710 | if (req->work.creds) { |
1711 | put_cred(req->work.creds); | |
1712 | req->work.creds = NULL; | |
1713 | } | |
f85c310a PB |
1714 | |
1715 | if (req->flags & REQ_F_INFLIGHT) { | |
1716 | struct io_ring_ctx *ctx = req->ctx; | |
f85c310a PB |
1717 | unsigned long flags; |
1718 | ||
1719 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1720 | list_del(&req->inflight_entry); | |
1721 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1722 | req->flags &= ~REQ_F_INFLIGHT; | |
f85c310a | 1723 | } |
e65ef56d JA |
1724 | } |
1725 | ||
b23fcf47 | 1726 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1727 | static inline void io_put_task(struct task_struct *task, int nr) |
1728 | { | |
1729 | struct io_uring_task *tctx = task->io_uring; | |
1730 | ||
1731 | percpu_counter_sub(&tctx->inflight, nr); | |
1732 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1733 | wake_up(&tctx->wait); | |
1734 | put_task_struct_many(task, nr); | |
1735 | } | |
1736 | ||
216578e5 | 1737 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1738 | { |
51a4cc11 | 1739 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1740 | |
216578e5 | 1741 | io_dismantle_req(req); |
7c660731 | 1742 | io_put_task(req->task, 1); |
c6ca97b3 | 1743 | |
3893f39f | 1744 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1745 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1746 | } |
1747 | ||
f2f87370 PB |
1748 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1749 | { | |
1750 | struct io_kiocb *nxt = req->link; | |
1751 | ||
1752 | req->link = nxt->link; | |
1753 | nxt->link = NULL; | |
1754 | } | |
1755 | ||
33cc89a9 PB |
1756 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1757 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1758 | { |
33cc89a9 | 1759 | struct io_kiocb *link = req->link; |
c9abd7ad | 1760 | bool cancelled = false; |
f2f87370 | 1761 | |
900fad45 PB |
1762 | /* |
1763 | * Can happen if a linked timeout fired and link had been like | |
1764 | * req -> link t-out -> link t-out [-> ...] | |
1765 | */ | |
c9abd7ad PB |
1766 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1767 | struct io_timeout_data *io = link->async_data; | |
1768 | int ret; | |
7c86ffee | 1769 | |
f2f87370 | 1770 | io_remove_next_linked(req); |
90cd7e42 | 1771 | link->timeout.head = NULL; |
c9abd7ad PB |
1772 | ret = hrtimer_try_to_cancel(&io->timer); |
1773 | if (ret != -1) { | |
1774 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1775 | io_put_req_deferred(link, 1); |
c9abd7ad PB |
1776 | cancelled = true; |
1777 | } | |
1778 | } | |
7c86ffee | 1779 | req->flags &= ~REQ_F_LINK_TIMEOUT; |
33cc89a9 | 1780 | return cancelled; |
7c86ffee PB |
1781 | } |
1782 | ||
d148ca4b | 1783 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1784 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1785 | { |
33cc89a9 | 1786 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1787 | |
f2f87370 | 1788 | req->link = NULL; |
f2f87370 PB |
1789 | while (link) { |
1790 | nxt = link->link; | |
1791 | link->link = NULL; | |
2665abfd | 1792 | |
f2f87370 | 1793 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1794 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1795 | io_put_req_deferred(link, 2); |
f2f87370 | 1796 | link = nxt; |
9e645e11 | 1797 | } |
33cc89a9 | 1798 | } |
9e645e11 | 1799 | |
33cc89a9 PB |
1800 | static bool io_disarm_next(struct io_kiocb *req) |
1801 | __must_hold(&req->ctx->completion_lock) | |
1802 | { | |
1803 | bool posted = false; | |
1804 | ||
1805 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1806 | posted = io_kill_linked_timeout(req); | |
1807 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1808 | posted |= (req->link != NULL); | |
1809 | io_fail_links(req); | |
1810 | } | |
1811 | return posted; | |
9e645e11 JA |
1812 | } |
1813 | ||
3fa5e0f3 | 1814 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1815 | { |
33cc89a9 | 1816 | struct io_kiocb *nxt; |
944e58bf | 1817 | |
9e645e11 JA |
1818 | /* |
1819 | * If LINK is set, we have dependent requests in this chain. If we | |
1820 | * didn't fail this request, queue the first one up, moving any other | |
1821 | * dependencies to the next request. In case of failure, fail the rest | |
1822 | * of the chain. | |
1823 | */ | |
33cc89a9 PB |
1824 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1825 | struct io_ring_ctx *ctx = req->ctx; | |
1826 | unsigned long flags; | |
1827 | bool posted; | |
1828 | ||
1829 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1830 | posted = io_disarm_next(req); | |
1831 | if (posted) | |
1832 | io_commit_cqring(req->ctx); | |
1833 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1834 | if (posted) | |
1835 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1836 | } |
33cc89a9 PB |
1837 | nxt = req->link; |
1838 | req->link = NULL; | |
1839 | return nxt; | |
4d7dd462 | 1840 | } |
9e645e11 | 1841 | |
f2f87370 | 1842 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1843 | { |
cdbff982 | 1844 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1845 | return NULL; |
1846 | return __io_req_find_next(req); | |
1847 | } | |
1848 | ||
2c32395d PB |
1849 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1850 | { | |
1851 | if (!ctx) | |
1852 | return; | |
1853 | if (ctx->submit_state.comp.nr) { | |
1854 | mutex_lock(&ctx->uring_lock); | |
1855 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1856 | mutex_unlock(&ctx->uring_lock); | |
1857 | } | |
1858 | percpu_ref_put(&ctx->refs); | |
1859 | } | |
1860 | ||
7cbf1722 | 1861 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1862 | { |
65453d1e | 1863 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1864 | struct io_wq_work_list list; |
1865 | struct io_wq_work_node *node; | |
c2c4c83c | 1866 | |
7cbf1722 JA |
1867 | if (wq_list_empty(&tctx->task_list)) |
1868 | return false; | |
6200b0ae | 1869 | |
0b81e80c | 1870 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1871 | list = tctx->task_list; |
1872 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1873 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1874 | |
7cbf1722 JA |
1875 | node = list.first; |
1876 | while (node) { | |
1877 | struct io_wq_work_node *next = node->next; | |
1878 | struct io_kiocb *req; | |
0ba9c9ed | 1879 | |
7cbf1722 | 1880 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1881 | if (req->ctx != ctx) { |
1882 | ctx_flush_and_put(ctx); | |
1883 | ctx = req->ctx; | |
1884 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1885 | } |
65453d1e | 1886 | |
2c32395d PB |
1887 | req->task_work.func(&req->task_work); |
1888 | node = next; | |
7cbf1722 JA |
1889 | } |
1890 | ||
2c32395d | 1891 | ctx_flush_and_put(ctx); |
7cbf1722 | 1892 | return list.first != NULL; |
c2c4c83c JA |
1893 | } |
1894 | ||
7cbf1722 | 1895 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1896 | { |
7cbf1722 | 1897 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1898 | |
1d5f360d JA |
1899 | clear_bit(0, &tctx->task_state); |
1900 | ||
7cbf1722 JA |
1901 | while (__tctx_task_work(tctx)) |
1902 | cond_resched(); | |
7cbf1722 JA |
1903 | } |
1904 | ||
1905 | static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req, | |
1906 | enum task_work_notify_mode notify) | |
1907 | { | |
1908 | struct io_uring_task *tctx = tsk->io_uring; | |
1909 | struct io_wq_work_node *node, *prev; | |
0b81e80c | 1910 | unsigned long flags; |
7cbf1722 JA |
1911 | int ret; |
1912 | ||
1913 | WARN_ON_ONCE(!tctx); | |
1914 | ||
0b81e80c | 1915 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1916 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1917 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1918 | |
1919 | /* task_work already pending, we're done */ | |
1920 | if (test_bit(0, &tctx->task_state) || | |
1921 | test_and_set_bit(0, &tctx->task_state)) | |
1922 | return 0; | |
1923 | ||
1924 | if (!task_work_add(tsk, &tctx->task_work, notify)) | |
1925 | return 0; | |
1926 | ||
1927 | /* | |
1928 | * Slow path - we failed, find and delete work. if the work is not | |
1929 | * in the list, it got run and we're fine. | |
1930 | */ | |
1931 | ret = 0; | |
0b81e80c | 1932 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1933 | wq_list_for_each(node, prev, &tctx->task_list) { |
1934 | if (&req->io_task_work.node == node) { | |
1935 | wq_list_del(&tctx->task_list, node, prev); | |
1936 | ret = 1; | |
1937 | break; | |
1938 | } | |
1939 | } | |
0b81e80c | 1940 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1941 | clear_bit(0, &tctx->task_state); |
1942 | return ret; | |
1943 | } | |
1944 | ||
355fb9e2 | 1945 | static int io_req_task_work_add(struct io_kiocb *req) |
c2c4c83c JA |
1946 | { |
1947 | struct task_struct *tsk = req->task; | |
1948 | struct io_ring_ctx *ctx = req->ctx; | |
91989c70 JA |
1949 | enum task_work_notify_mode notify; |
1950 | int ret; | |
c2c4c83c | 1951 | |
6200b0ae JA |
1952 | if (tsk->flags & PF_EXITING) |
1953 | return -ESRCH; | |
1954 | ||
c2c4c83c | 1955 | /* |
0ba9c9ed JA |
1956 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For |
1957 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1958 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1959 | * will do the job. | |
c2c4c83c | 1960 | */ |
91989c70 | 1961 | notify = TWA_NONE; |
355fb9e2 | 1962 | if (!(ctx->flags & IORING_SETUP_SQPOLL)) |
c2c4c83c JA |
1963 | notify = TWA_SIGNAL; |
1964 | ||
7cbf1722 | 1965 | ret = io_task_work_add(tsk, req, notify); |
c2c4c83c JA |
1966 | if (!ret) |
1967 | wake_up_process(tsk); | |
0ba9c9ed | 1968 | |
c2c4c83c JA |
1969 | return ret; |
1970 | } | |
1971 | ||
9b465711 PB |
1972 | static bool io_run_task_work_head(struct callback_head **work_head) |
1973 | { | |
1974 | struct callback_head *work, *next; | |
1975 | bool executed = false; | |
1976 | ||
1977 | do { | |
1978 | work = xchg(work_head, NULL); | |
1979 | if (!work) | |
1980 | break; | |
1981 | ||
1982 | do { | |
1983 | next = work->next; | |
1984 | work->func(work); | |
1985 | work = next; | |
1986 | cond_resched(); | |
1987 | } while (work); | |
1988 | executed = true; | |
1989 | } while (1); | |
1990 | ||
1991 | return executed; | |
1992 | } | |
1993 | ||
1994 | static void io_task_work_add_head(struct callback_head **work_head, | |
1995 | struct callback_head *task_work) | |
eab30c4d | 1996 | { |
7c25c0d1 | 1997 | struct callback_head *head; |
eab30c4d | 1998 | |
7c25c0d1 | 1999 | do { |
9b465711 PB |
2000 | head = READ_ONCE(*work_head); |
2001 | task_work->next = head; | |
2002 | } while (cmpxchg(work_head, head, task_work) != head); | |
2003 | } | |
2004 | ||
2005 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2006 | task_work_func_t cb) | |
2007 | { | |
2008 | init_task_work(&req->task_work, cb); | |
2009 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2010 | } |
2011 | ||
c40f6379 JA |
2012 | static void __io_req_task_cancel(struct io_kiocb *req, int error) |
2013 | { | |
2014 | struct io_ring_ctx *ctx = req->ctx; | |
2015 | ||
2016 | spin_lock_irq(&ctx->completion_lock); | |
2017 | io_cqring_fill_event(req, error); | |
2018 | io_commit_cqring(ctx); | |
2019 | spin_unlock_irq(&ctx->completion_lock); | |
2020 | ||
2021 | io_cqring_ev_posted(ctx); | |
2022 | req_set_fail_links(req); | |
2023 | io_double_put_req(req); | |
2024 | } | |
2025 | ||
2026 | static void io_req_task_cancel(struct callback_head *cb) | |
2027 | { | |
2028 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2029 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2030 | |
e83acd7d | 2031 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2032 | mutex_lock(&ctx->uring_lock); |
a3df7698 | 2033 | __io_req_task_cancel(req, req->result); |
792bb6eb | 2034 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2035 | } |
2036 | ||
2037 | static void __io_req_task_submit(struct io_kiocb *req) | |
2038 | { | |
2039 | struct io_ring_ctx *ctx = req->ctx; | |
2040 | ||
04fc6c80 | 2041 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2042 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2043 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2044 | __io_queue_sqe(req); |
81b6d05c | 2045 | else |
c40f6379 | 2046 | __io_req_task_cancel(req, -EFAULT); |
81b6d05c | 2047 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2048 | } |
2049 | ||
2050 | static void io_req_task_submit(struct callback_head *cb) | |
2051 | { | |
2052 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2053 | ||
2054 | __io_req_task_submit(req); | |
2055 | } | |
2056 | ||
2c4b8eb6 | 2057 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2058 | { |
2c4b8eb6 PB |
2059 | req->result = ret; |
2060 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2061 | |
2c4b8eb6 | 2062 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2063 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2064 | } |
2065 | ||
2c4b8eb6 | 2066 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2067 | { |
2c4b8eb6 | 2068 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2069 | |
2070 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2071 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2072 | } |
2073 | ||
f2f87370 | 2074 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2075 | { |
9b5f7bd9 | 2076 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2077 | |
2078 | if (nxt) | |
906a8c3f | 2079 | io_req_task_queue(nxt); |
c69f8dbe JL |
2080 | } |
2081 | ||
c3524383 | 2082 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2083 | { |
c3524383 PB |
2084 | io_queue_next(req); |
2085 | __io_free_req(req); | |
2086 | } | |
8766dd51 | 2087 | |
2d6500d4 | 2088 | struct req_batch { |
5af1d13e PB |
2089 | struct task_struct *task; |
2090 | int task_refs; | |
1b4c351f | 2091 | int ctx_refs; |
2d6500d4 PB |
2092 | }; |
2093 | ||
5af1d13e PB |
2094 | static inline void io_init_req_batch(struct req_batch *rb) |
2095 | { | |
5af1d13e | 2096 | rb->task_refs = 0; |
9ae72463 | 2097 | rb->ctx_refs = 0; |
5af1d13e PB |
2098 | rb->task = NULL; |
2099 | } | |
2100 | ||
2d6500d4 PB |
2101 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2102 | struct req_batch *rb) | |
2103 | { | |
6e833d53 | 2104 | if (rb->task) |
7c660731 | 2105 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2106 | if (rb->ctx_refs) |
2107 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2108 | } |
2109 | ||
6ff119a6 PB |
2110 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2111 | struct io_submit_state *state) | |
2d6500d4 | 2112 | { |
f2f87370 | 2113 | io_queue_next(req); |
96670657 | 2114 | io_dismantle_req(req); |
2d6500d4 | 2115 | |
e3bc8e9d | 2116 | if (req->task != rb->task) { |
7c660731 PB |
2117 | if (rb->task) |
2118 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2119 | rb->task = req->task; |
2120 | rb->task_refs = 0; | |
5af1d13e | 2121 | } |
e3bc8e9d | 2122 | rb->task_refs++; |
9ae72463 | 2123 | rb->ctx_refs++; |
5af1d13e | 2124 | |
bd759045 | 2125 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2126 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2127 | else |
2128 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2129 | } |
2130 | ||
905c172f PB |
2131 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2132 | struct io_ring_ctx *ctx) | |
2133 | { | |
2134 | int i, nr = cs->nr; | |
2135 | struct io_kiocb *req; | |
2136 | struct req_batch rb; | |
2137 | ||
2138 | io_init_req_batch(&rb); | |
2139 | spin_lock_irq(&ctx->completion_lock); | |
2140 | for (i = 0; i < nr; i++) { | |
2141 | req = cs->reqs[i]; | |
2142 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2143 | } | |
2144 | io_commit_cqring(ctx); | |
2145 | spin_unlock_irq(&ctx->completion_lock); | |
2146 | ||
2147 | io_cqring_ev_posted(ctx); | |
2148 | for (i = 0; i < nr; i++) { | |
2149 | req = cs->reqs[i]; | |
2150 | ||
2151 | /* submission and completion refs */ | |
de9b4cca | 2152 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2153 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2154 | } |
2155 | ||
2156 | io_req_free_batch_finish(ctx, &rb); | |
2157 | cs->nr = 0; | |
7a743e22 PB |
2158 | } |
2159 | ||
ba816ad6 JA |
2160 | /* |
2161 | * Drop reference to request, return next in chain (if there is one) if this | |
2162 | * was the last reference to this request. | |
2163 | */ | |
9b5f7bd9 | 2164 | static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2165 | { |
9b5f7bd9 PB |
2166 | struct io_kiocb *nxt = NULL; |
2167 | ||
de9b4cca | 2168 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2169 | nxt = io_req_find_next(req); |
4d7dd462 | 2170 | __io_free_req(req); |
2a44f467 | 2171 | } |
9b5f7bd9 | 2172 | return nxt; |
2b188cc1 JA |
2173 | } |
2174 | ||
e65ef56d JA |
2175 | static void io_put_req(struct io_kiocb *req) |
2176 | { | |
de9b4cca | 2177 | if (req_ref_put_and_test(req)) |
e65ef56d | 2178 | io_free_req(req); |
2b188cc1 JA |
2179 | } |
2180 | ||
216578e5 PB |
2181 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2182 | { | |
2183 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2184 | ||
2185 | io_free_req(req); | |
2186 | } | |
2187 | ||
2188 | static void io_free_req_deferred(struct io_kiocb *req) | |
2189 | { | |
2190 | int ret; | |
2191 | ||
7cbf1722 | 2192 | req->task_work.func = io_put_req_deferred_cb; |
355fb9e2 | 2193 | ret = io_req_task_work_add(req); |
eab30c4d PB |
2194 | if (unlikely(ret)) |
2195 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); | |
216578e5 PB |
2196 | } |
2197 | ||
2198 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2199 | { | |
de9b4cca | 2200 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2201 | io_free_req_deferred(req); |
2202 | } | |
2203 | ||
978db57e JA |
2204 | static void io_double_put_req(struct io_kiocb *req) |
2205 | { | |
2206 | /* drop both submit and complete references */ | |
de9b4cca | 2207 | if (req_ref_sub_and_test(req, 2)) |
978db57e JA |
2208 | io_free_req(req); |
2209 | } | |
2210 | ||
6c503150 | 2211 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2212 | { |
2213 | /* See comment at the top of this file */ | |
2214 | smp_rmb(); | |
e23de15f | 2215 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2216 | } |
2217 | ||
fb5ccc98 PB |
2218 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2219 | { | |
2220 | struct io_rings *rings = ctx->rings; | |
2221 | ||
2222 | /* make sure SQ entry isn't read before tail */ | |
2223 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2224 | } | |
2225 | ||
8ff069bf | 2226 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2227 | { |
8ff069bf | 2228 | unsigned int cflags; |
e94f141b | 2229 | |
bcda7baa JA |
2230 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2231 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2232 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2233 | kfree(kbuf); |
2234 | return cflags; | |
e94f141b JA |
2235 | } |
2236 | ||
8ff069bf | 2237 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2238 | { |
4d954c25 | 2239 | struct io_buffer *kbuf; |
bcda7baa | 2240 | |
4d954c25 | 2241 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2242 | return io_put_kbuf(req, kbuf); |
2243 | } | |
2244 | ||
4c6e277c JA |
2245 | static inline bool io_run_task_work(void) |
2246 | { | |
6200b0ae JA |
2247 | /* |
2248 | * Not safe to run on exiting task, and the task_work handling will | |
2249 | * not add work to such a task. | |
2250 | */ | |
2251 | if (unlikely(current->flags & PF_EXITING)) | |
2252 | return false; | |
4c6e277c JA |
2253 | if (current->task_works) { |
2254 | __set_current_state(TASK_RUNNING); | |
2255 | task_work_run(); | |
2256 | return true; | |
2257 | } | |
2258 | ||
2259 | return false; | |
bcda7baa JA |
2260 | } |
2261 | ||
def596e9 JA |
2262 | /* |
2263 | * Find and free completed poll iocbs | |
2264 | */ | |
2265 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2266 | struct list_head *done) | |
2267 | { | |
8237e045 | 2268 | struct req_batch rb; |
def596e9 | 2269 | struct io_kiocb *req; |
bbde017a XW |
2270 | |
2271 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2272 | smp_rmb(); | |
def596e9 | 2273 | |
5af1d13e | 2274 | io_init_req_batch(&rb); |
def596e9 | 2275 | while (!list_empty(done)) { |
bcda7baa JA |
2276 | int cflags = 0; |
2277 | ||
d21ffe7e | 2278 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2279 | list_del(&req->inflight_entry); |
2280 | ||
bbde017a XW |
2281 | if (READ_ONCE(req->result) == -EAGAIN) { |
2282 | req->iopoll_completed = 0; | |
23faba36 | 2283 | if (io_rw_reissue(req)) |
f161340d | 2284 | continue; |
bbde017a | 2285 | } |
def596e9 | 2286 | |
bcda7baa | 2287 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2288 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2289 | |
2290 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2291 | (*nr_events)++; |
2292 | ||
de9b4cca | 2293 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2294 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2295 | } |
def596e9 | 2296 | |
09bb8394 | 2297 | io_commit_cqring(ctx); |
80c18e4a | 2298 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2299 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2300 | } |
2301 | ||
def596e9 JA |
2302 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2303 | long min) | |
2304 | { | |
2305 | struct io_kiocb *req, *tmp; | |
2306 | LIST_HEAD(done); | |
2307 | bool spin; | |
2308 | int ret; | |
2309 | ||
2310 | /* | |
2311 | * Only spin for completions if we don't have multiple devices hanging | |
2312 | * off our complete list, and we're under the requested amount. | |
2313 | */ | |
2314 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2315 | ||
2316 | ret = 0; | |
d21ffe7e | 2317 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2318 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2319 | |
2320 | /* | |
581f9810 BM |
2321 | * Move completed and retryable entries to our local lists. |
2322 | * If we find a request that requires polling, break out | |
2323 | * and complete those lists first, if we have entries there. | |
def596e9 | 2324 | */ |
65a6543d | 2325 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2326 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2327 | continue; |
2328 | } | |
2329 | if (!list_empty(&done)) | |
2330 | break; | |
2331 | ||
2332 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2333 | if (ret < 0) | |
2334 | break; | |
2335 | ||
3aadc23e PB |
2336 | /* iopoll may have completed current req */ |
2337 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2338 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2339 | |
def596e9 JA |
2340 | if (ret && spin) |
2341 | spin = false; | |
2342 | ret = 0; | |
2343 | } | |
2344 | ||
2345 | if (!list_empty(&done)) | |
2346 | io_iopoll_complete(ctx, nr_events, &done); | |
2347 | ||
2348 | return ret; | |
2349 | } | |
2350 | ||
2351 | /* | |
d195a66e | 2352 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2353 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2354 | * as a non-spinning completion check. | |
2355 | */ | |
2356 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2357 | long min) | |
2358 | { | |
540e32a0 | 2359 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2360 | int ret; |
2361 | ||
2362 | ret = io_do_iopoll(ctx, nr_events, min); | |
2363 | if (ret < 0) | |
2364 | return ret; | |
eba0a4dd | 2365 | if (*nr_events >= min) |
def596e9 JA |
2366 | return 0; |
2367 | } | |
2368 | ||
2369 | return 1; | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * We can't just wait for polled events to come to us, we have to actively | |
2374 | * find and complete them. | |
2375 | */ | |
b2edc0a7 | 2376 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2377 | { |
2378 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2379 | return; | |
2380 | ||
2381 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2382 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2383 | unsigned int nr_events = 0; |
2384 | ||
b2edc0a7 | 2385 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2386 | |
b2edc0a7 PB |
2387 | /* let it sleep and repeat later if can't complete a request */ |
2388 | if (nr_events == 0) | |
2389 | break; | |
08f5439f JA |
2390 | /* |
2391 | * Ensure we allow local-to-the-cpu processing to take place, | |
2392 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2393 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2394 | */ |
3fcee5a6 PB |
2395 | if (need_resched()) { |
2396 | mutex_unlock(&ctx->uring_lock); | |
2397 | cond_resched(); | |
2398 | mutex_lock(&ctx->uring_lock); | |
2399 | } | |
def596e9 JA |
2400 | } |
2401 | mutex_unlock(&ctx->uring_lock); | |
2402 | } | |
2403 | ||
7668b92a | 2404 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2405 | { |
7668b92a | 2406 | unsigned int nr_events = 0; |
2b2ed975 | 2407 | int iters = 0, ret = 0; |
500f9fba | 2408 | |
c7849be9 XW |
2409 | /* |
2410 | * We disallow the app entering submit/complete with polling, but we | |
2411 | * still need to lock the ring to prevent racing with polled issue | |
2412 | * that got punted to a workqueue. | |
2413 | */ | |
2414 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2415 | do { |
a3a0e43f JA |
2416 | /* |
2417 | * Don't enter poll loop if we already have events pending. | |
2418 | * If we do, we can potentially be spinning for commands that | |
2419 | * already triggered a CQE (eg in error). | |
2420 | */ | |
6c503150 PB |
2421 | if (test_bit(0, &ctx->cq_check_overflow)) |
2422 | __io_cqring_overflow_flush(ctx, false, NULL, NULL); | |
2423 | if (io_cqring_events(ctx)) | |
a3a0e43f JA |
2424 | break; |
2425 | ||
500f9fba JA |
2426 | /* |
2427 | * If a submit got punted to a workqueue, we can have the | |
2428 | * application entering polling for a command before it gets | |
2429 | * issued. That app will hold the uring_lock for the duration | |
2430 | * of the poll right here, so we need to take a breather every | |
2431 | * now and then to ensure that the issue has a chance to add | |
2432 | * the poll to the issued list. Otherwise we can spin here | |
2433 | * forever, while the workqueue is stuck trying to acquire the | |
2434 | * very same mutex. | |
2435 | */ | |
2436 | if (!(++iters & 7)) { | |
2437 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2438 | io_run_task_work(); |
500f9fba JA |
2439 | mutex_lock(&ctx->uring_lock); |
2440 | } | |
2441 | ||
7668b92a | 2442 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2443 | if (ret <= 0) |
2444 | break; | |
2445 | ret = 0; | |
7668b92a | 2446 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2447 | |
500f9fba | 2448 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2449 | return ret; |
2450 | } | |
2451 | ||
491381ce | 2452 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2453 | { |
491381ce JA |
2454 | /* |
2455 | * Tell lockdep we inherited freeze protection from submission | |
2456 | * thread. | |
2457 | */ | |
2458 | if (req->flags & REQ_F_ISREG) { | |
2459 | struct inode *inode = file_inode(req->file); | |
2b188cc1 | 2460 | |
491381ce | 2461 | __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE); |
2b188cc1 | 2462 | } |
491381ce | 2463 | file_end_write(req->file); |
2b188cc1 JA |
2464 | } |
2465 | ||
b63534c4 | 2466 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2467 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2468 | { |
179ae0d1 PB |
2469 | /* either already prepared or successfully done */ |
2470 | return req->async_data || !io_req_prep_async(req); | |
b63534c4 | 2471 | } |
b63534c4 | 2472 | |
3e6a0d3c | 2473 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2474 | { |
355afaeb | 2475 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2476 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2477 | |
355afaeb JA |
2478 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2479 | return false; | |
3e6a0d3c JA |
2480 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2481 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2482 | return false; |
7c977a58 JA |
2483 | /* |
2484 | * If ref is dying, we might be running poll reap from the exit work. | |
2485 | * Don't attempt to reissue from that path, just let it fail with | |
2486 | * -EAGAIN. | |
2487 | */ | |
3e6a0d3c JA |
2488 | if (percpu_ref_is_dying(&ctx->refs)) |
2489 | return false; | |
2490 | return true; | |
2491 | } | |
e82ad485 JA |
2492 | #else |
2493 | static bool io_rw_should_reissue(struct io_kiocb *req) | |
2494 | { | |
2495 | return false; | |
2496 | } | |
3e6a0d3c JA |
2497 | #endif |
2498 | ||
2499 | static bool io_rw_reissue(struct io_kiocb *req) | |
2500 | { | |
2501 | #ifdef CONFIG_BLOCK | |
2502 | if (!io_rw_should_reissue(req)) | |
7c977a58 | 2503 | return false; |
b63534c4 | 2504 | |
55e6ac1e PB |
2505 | lockdep_assert_held(&req->ctx->uring_lock); |
2506 | ||
37d1e2e3 | 2507 | if (io_resubmit_prep(req)) { |
de9b4cca | 2508 | req_ref_get(req); |
fdee946d | 2509 | io_queue_async_work(req); |
b63534c4 | 2510 | return true; |
fdee946d | 2511 | } |
dc2a6e9a | 2512 | req_set_fail_links(req); |
b63534c4 JA |
2513 | #endif |
2514 | return false; | |
2515 | } | |
2516 | ||
a1d7c393 | 2517 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2518 | unsigned int issue_flags) |
a1d7c393 | 2519 | { |
2f8e45f1 PB |
2520 | int cflags = 0; |
2521 | ||
b65c128f PB |
2522 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2523 | kiocb_end_write(req); | |
230d50d4 JA |
2524 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_should_reissue(req)) { |
2525 | req->flags |= REQ_F_REISSUE; | |
23faba36 | 2526 | return; |
230d50d4 | 2527 | } |
2f8e45f1 PB |
2528 | if (res != req->result) |
2529 | req_set_fail_links(req); | |
2f8e45f1 PB |
2530 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2531 | cflags = io_put_rw_kbuf(req); | |
2532 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2533 | } |
2534 | ||
2535 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2536 | { | |
9adbd45d | 2537 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2538 | |
889fca73 | 2539 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2540 | } |
2541 | ||
def596e9 JA |
2542 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2543 | { | |
9adbd45d | 2544 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2545 | |
3e6a0d3c JA |
2546 | #ifdef CONFIG_BLOCK |
2547 | /* Rewind iter, if we have one. iopoll path resubmits as usual */ | |
2548 | if (res == -EAGAIN && io_rw_should_reissue(req)) { | |
2549 | struct io_async_rw *rw = req->async_data; | |
2550 | ||
2551 | if (rw) | |
2552 | iov_iter_revert(&rw->iter, | |
2553 | req->result - iov_iter_count(&rw->iter)); | |
2554 | else if (!io_resubmit_prep(req)) | |
2555 | res = -EIO; | |
2556 | } | |
2557 | #endif | |
2558 | ||
491381ce JA |
2559 | if (kiocb->ki_flags & IOCB_WRITE) |
2560 | kiocb_end_write(req); | |
def596e9 | 2561 | |
2d7d6792 | 2562 | if (res != -EAGAIN && res != req->result) |
4e88d6e7 | 2563 | req_set_fail_links(req); |
bbde017a XW |
2564 | |
2565 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2566 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2567 | smp_wmb(); |
2568 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2569 | } |
2570 | ||
2571 | /* | |
2572 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2573 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2574 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2575 | * accessing the kiocb cookie. | |
2576 | */ | |
2e9dbe90 | 2577 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2578 | { |
2579 | struct io_ring_ctx *ctx = req->ctx; | |
2580 | ||
2581 | /* | |
2582 | * Track whether we have multiple files in our lists. This will impact | |
2583 | * how we do polling eventually, not spinning if we're on potentially | |
2584 | * different devices. | |
2585 | */ | |
540e32a0 | 2586 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2587 | ctx->poll_multi_file = false; |
2588 | } else if (!ctx->poll_multi_file) { | |
2589 | struct io_kiocb *list_req; | |
2590 | ||
540e32a0 | 2591 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2592 | inflight_entry); |
9adbd45d | 2593 | if (list_req->file != req->file) |
def596e9 JA |
2594 | ctx->poll_multi_file = true; |
2595 | } | |
2596 | ||
2597 | /* | |
2598 | * For fast devices, IO may have already completed. If it has, add | |
2599 | * it to the front so we find it first. | |
2600 | */ | |
65a6543d | 2601 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2602 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2603 | else |
d21ffe7e | 2604 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2605 | |
2e9dbe90 XW |
2606 | /* |
2607 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2608 | * task context or in io worker task context. If current task context is | |
2609 | * sq thread, we don't need to check whether should wake up sq thread. | |
2610 | */ | |
2611 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2612 | wq_has_sleeper(&ctx->sq_data->wait)) |
2613 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2614 | } |
2615 | ||
9f13c35b PB |
2616 | static inline void io_state_file_put(struct io_submit_state *state) |
2617 | { | |
02b23a9a PB |
2618 | if (state->file_refs) { |
2619 | fput_many(state->file, state->file_refs); | |
2620 | state->file_refs = 0; | |
2621 | } | |
9a56a232 JA |
2622 | } |
2623 | ||
2624 | /* | |
2625 | * Get as many references to a file as we have IOs left in this submission, | |
2626 | * assuming most submissions are for one file, or at least that each file | |
2627 | * has more than one submission. | |
2628 | */ | |
8da11c19 | 2629 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2630 | { |
2631 | if (!state) | |
2632 | return fget(fd); | |
2633 | ||
6e1271e6 | 2634 | if (state->file_refs) { |
9a56a232 | 2635 | if (state->fd == fd) { |
6e1271e6 | 2636 | state->file_refs--; |
9a56a232 JA |
2637 | return state->file; |
2638 | } | |
02b23a9a | 2639 | io_state_file_put(state); |
9a56a232 JA |
2640 | } |
2641 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2642 | if (unlikely(!state->file)) |
9a56a232 JA |
2643 | return NULL; |
2644 | ||
2645 | state->fd = fd; | |
6e1271e6 | 2646 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2647 | return state->file; |
2648 | } | |
2649 | ||
4503b767 JA |
2650 | static bool io_bdev_nowait(struct block_device *bdev) |
2651 | { | |
9ba0d0c8 | 2652 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2653 | } |
2654 | ||
2b188cc1 JA |
2655 | /* |
2656 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2657 | * any file. For now, just ensure that anything potentially problematic is done | |
2658 | * inline. | |
2659 | */ | |
7b29f92d | 2660 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2661 | { |
2662 | umode_t mode = file_inode(file)->i_mode; | |
2663 | ||
4503b767 | 2664 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2665 | if (IS_ENABLED(CONFIG_BLOCK) && |
2666 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2667 | return true; |
2668 | return false; | |
2669 | } | |
2670 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2671 | return true; |
4503b767 | 2672 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2673 | if (IS_ENABLED(CONFIG_BLOCK) && |
2674 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2675 | file->f_op != &io_uring_fops) |
2676 | return true; | |
2677 | return false; | |
2678 | } | |
2b188cc1 | 2679 | |
c5b85625 JA |
2680 | /* any ->read/write should understand O_NONBLOCK */ |
2681 | if (file->f_flags & O_NONBLOCK) | |
2682 | return true; | |
2683 | ||
af197f50 JA |
2684 | if (!(file->f_mode & FMODE_NOWAIT)) |
2685 | return false; | |
2686 | ||
2687 | if (rw == READ) | |
2688 | return file->f_op->read_iter != NULL; | |
2689 | ||
2690 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2691 | } |
2692 | ||
7b29f92d JA |
2693 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2694 | { | |
2695 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2696 | return true; | |
2697 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2698 | return true; | |
2699 | ||
2700 | return __io_file_supports_async(req->file, rw); | |
2701 | } | |
2702 | ||
a88fc400 | 2703 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2704 | { |
def596e9 | 2705 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2706 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2707 | struct file *file = req->file; |
09bb8394 JA |
2708 | unsigned ioprio; |
2709 | int ret; | |
2b188cc1 | 2710 | |
7b29f92d | 2711 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2712 | req->flags |= REQ_F_ISREG; |
2713 | ||
2b188cc1 | 2714 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2715 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2716 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2717 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2718 | } |
2b188cc1 | 2719 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2720 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2721 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2722 | if (unlikely(ret)) | |
2723 | return ret; | |
2b188cc1 | 2724 | |
75c668cd PB |
2725 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2726 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2727 | req->flags |= REQ_F_NOWAIT; | |
2728 | ||
2b188cc1 JA |
2729 | ioprio = READ_ONCE(sqe->ioprio); |
2730 | if (ioprio) { | |
2731 | ret = ioprio_check_cap(ioprio); | |
2732 | if (ret) | |
09bb8394 | 2733 | return ret; |
2b188cc1 JA |
2734 | |
2735 | kiocb->ki_ioprio = ioprio; | |
2736 | } else | |
2737 | kiocb->ki_ioprio = get_current_ioprio(); | |
2738 | ||
def596e9 | 2739 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2740 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2741 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2742 | return -EOPNOTSUPP; |
2b188cc1 | 2743 | |
def596e9 JA |
2744 | kiocb->ki_flags |= IOCB_HIPRI; |
2745 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2746 | req->iopoll_completed = 0; |
def596e9 | 2747 | } else { |
09bb8394 JA |
2748 | if (kiocb->ki_flags & IOCB_HIPRI) |
2749 | return -EINVAL; | |
def596e9 JA |
2750 | kiocb->ki_complete = io_complete_rw; |
2751 | } | |
9adbd45d | 2752 | |
3529d8c2 JA |
2753 | req->rw.addr = READ_ONCE(sqe->addr); |
2754 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2755 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2756 | return 0; |
2b188cc1 JA |
2757 | } |
2758 | ||
2759 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2760 | { | |
2761 | switch (ret) { | |
2762 | case -EIOCBQUEUED: | |
2763 | break; | |
2764 | case -ERESTARTSYS: | |
2765 | case -ERESTARTNOINTR: | |
2766 | case -ERESTARTNOHAND: | |
2767 | case -ERESTART_RESTARTBLOCK: | |
2768 | /* | |
2769 | * We can't just restart the syscall, since previously | |
2770 | * submitted sqes may already be in progress. Just fail this | |
2771 | * IO with EINTR. | |
2772 | */ | |
2773 | ret = -EINTR; | |
df561f66 | 2774 | fallthrough; |
2b188cc1 JA |
2775 | default: |
2776 | kiocb->ki_complete(kiocb, ret, 0); | |
2777 | } | |
2778 | } | |
2779 | ||
a1d7c393 | 2780 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2781 | unsigned int issue_flags) |
ba816ad6 | 2782 | { |
ba04291e | 2783 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2784 | struct io_async_rw *io = req->async_data; |
97284637 | 2785 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2786 | |
227c0c96 | 2787 | /* add previously done IO, if any */ |
e8c2bc1f | 2788 | if (io && io->bytes_done > 0) { |
227c0c96 | 2789 | if (ret < 0) |
e8c2bc1f | 2790 | ret = io->bytes_done; |
227c0c96 | 2791 | else |
e8c2bc1f | 2792 | ret += io->bytes_done; |
227c0c96 JA |
2793 | } |
2794 | ||
ba04291e JA |
2795 | if (req->flags & REQ_F_CUR_POS) |
2796 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2797 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2798 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2799 | else |
2800 | io_rw_done(kiocb, ret); | |
97284637 PB |
2801 | |
2802 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2803 | req->flags &= ~REQ_F_REISSUE; | |
2804 | if (!io_rw_reissue(req)) { | |
2805 | int cflags = 0; | |
2806 | ||
2807 | req_set_fail_links(req); | |
2808 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2809 | cflags = io_put_rw_kbuf(req); | |
2810 | __io_req_complete(req, issue_flags, ret, cflags); | |
2811 | } | |
2812 | } | |
ba816ad6 JA |
2813 | } |
2814 | ||
847595de | 2815 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2816 | { |
9adbd45d JA |
2817 | struct io_ring_ctx *ctx = req->ctx; |
2818 | size_t len = req->rw.len; | |
edafccee | 2819 | struct io_mapped_ubuf *imu; |
4be1c615 | 2820 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2821 | size_t offset; |
2822 | u64 buf_addr; | |
2823 | ||
edafccee JA |
2824 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2825 | return -EFAULT; | |
edafccee JA |
2826 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2827 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2828 | buf_addr = req->rw.addr; |
edafccee JA |
2829 | |
2830 | /* overflow */ | |
2831 | if (buf_addr + len < buf_addr) | |
2832 | return -EFAULT; | |
2833 | /* not inside the mapped region */ | |
2834 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2835 | return -EFAULT; | |
2836 | ||
2837 | /* | |
2838 | * May not be a start of buffer, set size appropriately | |
2839 | * and advance us to the beginning. | |
2840 | */ | |
2841 | offset = buf_addr - imu->ubuf; | |
2842 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2843 | |
2844 | if (offset) { | |
2845 | /* | |
2846 | * Don't use iov_iter_advance() here, as it's really slow for | |
2847 | * using the latter parts of a big fixed buffer - it iterates | |
2848 | * over each segment manually. We can cheat a bit here, because | |
2849 | * we know that: | |
2850 | * | |
2851 | * 1) it's a BVEC iter, we set it up | |
2852 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2853 | * first and last bvec | |
2854 | * | |
2855 | * So just find our index, and adjust the iterator afterwards. | |
2856 | * If the offset is within the first bvec (or the whole first | |
2857 | * bvec, just use iov_iter_advance(). This makes it easier | |
2858 | * since we can just skip the first segment, which may not | |
2859 | * be PAGE_SIZE aligned. | |
2860 | */ | |
2861 | const struct bio_vec *bvec = imu->bvec; | |
2862 | ||
2863 | if (offset <= bvec->bv_len) { | |
2864 | iov_iter_advance(iter, offset); | |
2865 | } else { | |
2866 | unsigned long seg_skip; | |
2867 | ||
2868 | /* skip first vec */ | |
2869 | offset -= bvec->bv_len; | |
2870 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2871 | ||
2872 | iter->bvec = bvec + seg_skip; | |
2873 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2874 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2875 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2876 | } |
2877 | } | |
2878 | ||
847595de | 2879 | return 0; |
edafccee JA |
2880 | } |
2881 | ||
bcda7baa JA |
2882 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2883 | { | |
2884 | if (needs_lock) | |
2885 | mutex_unlock(&ctx->uring_lock); | |
2886 | } | |
2887 | ||
2888 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2889 | { | |
2890 | /* | |
2891 | * "Normal" inline submissions always hold the uring_lock, since we | |
2892 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2893 | * The only exception is when we've detached the request and issue it | |
2894 | * from an async worker thread, grab the lock for that case. | |
2895 | */ | |
2896 | if (needs_lock) | |
2897 | mutex_lock(&ctx->uring_lock); | |
2898 | } | |
2899 | ||
2900 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2901 | int bgid, struct io_buffer *kbuf, | |
2902 | bool needs_lock) | |
2903 | { | |
2904 | struct io_buffer *head; | |
2905 | ||
2906 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2907 | return kbuf; | |
2908 | ||
2909 | io_ring_submit_lock(req->ctx, needs_lock); | |
2910 | ||
2911 | lockdep_assert_held(&req->ctx->uring_lock); | |
2912 | ||
9e15c3a0 | 2913 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2914 | if (head) { |
2915 | if (!list_empty(&head->list)) { | |
2916 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2917 | list); | |
2918 | list_del(&kbuf->list); | |
2919 | } else { | |
2920 | kbuf = head; | |
9e15c3a0 | 2921 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2922 | } |
2923 | if (*len > kbuf->len) | |
2924 | *len = kbuf->len; | |
2925 | } else { | |
2926 | kbuf = ERR_PTR(-ENOBUFS); | |
2927 | } | |
2928 | ||
2929 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2930 | ||
2931 | return kbuf; | |
2932 | } | |
2933 | ||
4d954c25 JA |
2934 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2935 | bool needs_lock) | |
2936 | { | |
2937 | struct io_buffer *kbuf; | |
4f4eeba8 | 2938 | u16 bgid; |
4d954c25 JA |
2939 | |
2940 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2941 | bgid = req->buf_index; |
4d954c25 JA |
2942 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2943 | if (IS_ERR(kbuf)) | |
2944 | return kbuf; | |
2945 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2946 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2947 | return u64_to_user_ptr(kbuf->addr); | |
2948 | } | |
2949 | ||
2950 | #ifdef CONFIG_COMPAT | |
2951 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2952 | bool needs_lock) | |
2953 | { | |
2954 | struct compat_iovec __user *uiov; | |
2955 | compat_ssize_t clen; | |
2956 | void __user *buf; | |
2957 | ssize_t len; | |
2958 | ||
2959 | uiov = u64_to_user_ptr(req->rw.addr); | |
2960 | if (!access_ok(uiov, sizeof(*uiov))) | |
2961 | return -EFAULT; | |
2962 | if (__get_user(clen, &uiov->iov_len)) | |
2963 | return -EFAULT; | |
2964 | if (clen < 0) | |
2965 | return -EINVAL; | |
2966 | ||
2967 | len = clen; | |
2968 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2969 | if (IS_ERR(buf)) | |
2970 | return PTR_ERR(buf); | |
2971 | iov[0].iov_base = buf; | |
2972 | iov[0].iov_len = (compat_size_t) len; | |
2973 | return 0; | |
2974 | } | |
2975 | #endif | |
2976 | ||
2977 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2978 | bool needs_lock) | |
2979 | { | |
2980 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2981 | void __user *buf; | |
2982 | ssize_t len; | |
2983 | ||
2984 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2985 | return -EFAULT; | |
2986 | ||
2987 | len = iov[0].iov_len; | |
2988 | if (len < 0) | |
2989 | return -EINVAL; | |
2990 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2991 | if (IS_ERR(buf)) | |
2992 | return PTR_ERR(buf); | |
2993 | iov[0].iov_base = buf; | |
2994 | iov[0].iov_len = len; | |
2995 | return 0; | |
2996 | } | |
2997 | ||
2998 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2999 | bool needs_lock) | |
3000 | { | |
dddb3e26 JA |
3001 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
3002 | struct io_buffer *kbuf; | |
3003 | ||
3004 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
3005 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
3006 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 3007 | return 0; |
dddb3e26 | 3008 | } |
dd201662 | 3009 | if (req->rw.len != 1) |
4d954c25 JA |
3010 | return -EINVAL; |
3011 | ||
3012 | #ifdef CONFIG_COMPAT | |
3013 | if (req->ctx->compat) | |
3014 | return io_compat_import(req, iov, needs_lock); | |
3015 | #endif | |
3016 | ||
3017 | return __io_iov_buffer_select(req, iov, needs_lock); | |
3018 | } | |
3019 | ||
847595de PB |
3020 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
3021 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 3022 | { |
9adbd45d JA |
3023 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
3024 | size_t sqe_len = req->rw.len; | |
847595de | 3025 | u8 opcode = req->opcode; |
4d954c25 | 3026 | ssize_t ret; |
edafccee | 3027 | |
7d009165 | 3028 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 3029 | *iovec = NULL; |
9adbd45d | 3030 | return io_import_fixed(req, rw, iter); |
edafccee | 3031 | } |
2b188cc1 | 3032 | |
bcda7baa | 3033 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3034 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3035 | return -EINVAL; |
3036 | ||
3a6820f2 | 3037 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3038 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3039 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3040 | if (IS_ERR(buf)) |
4d954c25 | 3041 | return PTR_ERR(buf); |
3f9d6441 | 3042 | req->rw.len = sqe_len; |
bcda7baa JA |
3043 | } |
3044 | ||
3a6820f2 JA |
3045 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3046 | *iovec = NULL; | |
10fc72e4 | 3047 | return ret; |
3a6820f2 JA |
3048 | } |
3049 | ||
4d954c25 JA |
3050 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3051 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3052 | if (!ret) |
3053 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3054 | *iovec = NULL; |
3055 | return ret; | |
3056 | } | |
3057 | ||
89cd35c5 CH |
3058 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3059 | req->ctx->compat); | |
2b188cc1 JA |
3060 | } |
3061 | ||
0fef9483 JA |
3062 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3063 | { | |
5b09e37e | 3064 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3065 | } |
3066 | ||
31b51510 | 3067 | /* |
32960613 JA |
3068 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3069 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3070 | */ |
4017eb91 | 3071 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3072 | { |
4017eb91 JA |
3073 | struct kiocb *kiocb = &req->rw.kiocb; |
3074 | struct file *file = req->file; | |
32960613 JA |
3075 | ssize_t ret = 0; |
3076 | ||
3077 | /* | |
3078 | * Don't support polled IO through this interface, and we can't | |
3079 | * support non-blocking either. For the latter, this just causes | |
3080 | * the kiocb to be handled from an async context. | |
3081 | */ | |
3082 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3083 | return -EOPNOTSUPP; | |
3084 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3085 | return -EAGAIN; | |
3086 | ||
3087 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3088 | struct iovec iovec; |
32960613 JA |
3089 | ssize_t nr; |
3090 | ||
311ae9e1 PB |
3091 | if (!iov_iter_is_bvec(iter)) { |
3092 | iovec = iov_iter_iovec(iter); | |
3093 | } else { | |
4017eb91 JA |
3094 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3095 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3096 | } |
3097 | ||
32960613 JA |
3098 | if (rw == READ) { |
3099 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3100 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3101 | } else { |
3102 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3103 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3104 | } |
3105 | ||
3106 | if (nr < 0) { | |
3107 | if (!ret) | |
3108 | ret = nr; | |
3109 | break; | |
3110 | } | |
3111 | ret += nr; | |
3112 | if (nr != iovec.iov_len) | |
3113 | break; | |
4017eb91 JA |
3114 | req->rw.len -= nr; |
3115 | req->rw.addr += nr; | |
32960613 JA |
3116 | iov_iter_advance(iter, nr); |
3117 | } | |
3118 | ||
3119 | return ret; | |
3120 | } | |
3121 | ||
ff6165b2 JA |
3122 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3123 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3124 | { |
e8c2bc1f | 3125 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3126 | |
ff6165b2 | 3127 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3128 | rw->free_iovec = iovec; |
227c0c96 | 3129 | rw->bytes_done = 0; |
ff6165b2 | 3130 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3131 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3132 | return; |
b64e3444 | 3133 | if (!iovec) { |
ff6165b2 JA |
3134 | unsigned iov_off = 0; |
3135 | ||
3136 | rw->iter.iov = rw->fast_iov; | |
3137 | if (iter->iov != fast_iov) { | |
3138 | iov_off = iter->iov - fast_iov; | |
3139 | rw->iter.iov += iov_off; | |
3140 | } | |
3141 | if (rw->fast_iov != fast_iov) | |
3142 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3143 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3144 | } else { |
3145 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3146 | } |
3147 | } | |
3148 | ||
6cb78689 | 3149 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3150 | { |
e8c2bc1f JA |
3151 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3152 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3153 | return req->async_data == NULL; | |
3d9932a8 XW |
3154 | } |
3155 | ||
ff6165b2 JA |
3156 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3157 | const struct iovec *fast_iov, | |
227c0c96 | 3158 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3159 | { |
26f0505a | 3160 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3161 | return 0; |
e8c2bc1f | 3162 | if (!req->async_data) { |
6cb78689 | 3163 | if (io_alloc_async_data(req)) { |
6bf985dc | 3164 | kfree(iovec); |
5d204bcf | 3165 | return -ENOMEM; |
6bf985dc | 3166 | } |
b7bb4f7d | 3167 | |
ff6165b2 | 3168 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3169 | } |
b7bb4f7d | 3170 | return 0; |
f67676d1 JA |
3171 | } |
3172 | ||
73debe68 | 3173 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3174 | { |
e8c2bc1f | 3175 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3176 | struct iovec *iov = iorw->fast_iov; |
847595de | 3177 | int ret; |
c3e330a4 | 3178 | |
2846c481 | 3179 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3180 | if (unlikely(ret < 0)) |
3181 | return ret; | |
3182 | ||
ab0b196c PB |
3183 | iorw->bytes_done = 0; |
3184 | iorw->free_iovec = iov; | |
3185 | if (iov) | |
3186 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3187 | return 0; |
3188 | } | |
3189 | ||
73debe68 | 3190 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3191 | { |
3529d8c2 JA |
3192 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3193 | return -EBADF; | |
93642ef8 | 3194 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3195 | } |
3196 | ||
c1dd91d1 JA |
3197 | /* |
3198 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3199 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3200 | * This gets called when the page is unlocked, and we generally expect that to | |
3201 | * happen when the page IO is completed and the page is now uptodate. This will | |
3202 | * queue a task_work based retry of the operation, attempting to copy the data | |
3203 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3204 | * do a thread based blocking retry of the operation. That's the unexpected | |
3205 | * slow path. | |
3206 | */ | |
bcf5a063 JA |
3207 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3208 | int sync, void *arg) | |
3209 | { | |
3210 | struct wait_page_queue *wpq; | |
3211 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3212 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3213 | |
3214 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3215 | ||
cdc8fcb4 LT |
3216 | if (!wake_page_match(wpq, key)) |
3217 | return 0; | |
3218 | ||
c8d317aa | 3219 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3220 | list_del_init(&wait->entry); |
3221 | ||
bcf5a063 | 3222 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3223 | req_ref_get(req); |
921b9054 | 3224 | io_req_task_queue(req); |
bcf5a063 JA |
3225 | return 1; |
3226 | } | |
3227 | ||
c1dd91d1 JA |
3228 | /* |
3229 | * This controls whether a given IO request should be armed for async page | |
3230 | * based retry. If we return false here, the request is handed to the async | |
3231 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3232 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3233 | * will either succeed because the page is now uptodate and unlocked, or it | |
3234 | * will register a callback when the page is unlocked at IO completion. Through | |
3235 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3236 | * That retry will attempt the buffered read again. The retry will generally | |
3237 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3238 | * async worker threads for a blocking retry. | |
3239 | */ | |
227c0c96 | 3240 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3241 | { |
e8c2bc1f JA |
3242 | struct io_async_rw *rw = req->async_data; |
3243 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3244 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3245 | |
bcf5a063 JA |
3246 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3247 | if (req->flags & REQ_F_NOWAIT) | |
3248 | return false; | |
f67676d1 | 3249 | |
227c0c96 | 3250 | /* Only for buffered IO */ |
3b2a4439 | 3251 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3252 | return false; |
3b2a4439 | 3253 | |
bcf5a063 JA |
3254 | /* |
3255 | * just use poll if we can, and don't attempt if the fs doesn't | |
3256 | * support callback based unlocks | |
3257 | */ | |
3258 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3259 | return false; | |
f67676d1 | 3260 | |
3b2a4439 JA |
3261 | wait->wait.func = io_async_buf_func; |
3262 | wait->wait.private = req; | |
3263 | wait->wait.flags = 0; | |
3264 | INIT_LIST_HEAD(&wait->wait.entry); | |
3265 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3266 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3267 | kiocb->ki_waitq = wait; |
3b2a4439 | 3268 | return true; |
bcf5a063 JA |
3269 | } |
3270 | ||
3271 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3272 | { | |
3273 | if (req->file->f_op->read_iter) | |
3274 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3275 | else if (req->file->f_op->read) |
4017eb91 | 3276 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3277 | else |
3278 | return -EINVAL; | |
f67676d1 JA |
3279 | } |
3280 | ||
889fca73 | 3281 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3282 | { |
3283 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3284 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3285 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3286 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3287 | ssize_t io_size, ret, ret2; |
45d189c6 | 3288 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3289 | |
2846c481 | 3290 | if (rw) { |
e8c2bc1f | 3291 | iter = &rw->iter; |
2846c481 PB |
3292 | iovec = NULL; |
3293 | } else { | |
3294 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3295 | if (ret < 0) | |
3296 | return ret; | |
3297 | } | |
632546c4 | 3298 | io_size = iov_iter_count(iter); |
fa15bafb | 3299 | req->result = io_size; |
2b188cc1 | 3300 | |
fd6c2e4c JA |
3301 | /* Ensure we clear previously set non-block flag */ |
3302 | if (!force_nonblock) | |
29de5f6a | 3303 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3304 | else |
3305 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3306 | ||
24c74678 | 3307 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3308 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3309 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3310 | return ret ?: -EAGAIN; |
6713e7a6 | 3311 | } |
9e645e11 | 3312 | |
632546c4 | 3313 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3314 | if (unlikely(ret)) { |
3315 | kfree(iovec); | |
3316 | return ret; | |
3317 | } | |
2b188cc1 | 3318 | |
227c0c96 | 3319 | ret = io_iter_do_read(req, iter); |
32960613 | 3320 | |
230d50d4 | 3321 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3322 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3323 | /* IOPOLL retry should happen for io-wq threads */ |
3324 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3325 | goto done; |
75c668cd PB |
3326 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3327 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3328 | goto done; |
84216315 | 3329 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3330 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3331 | ret = 0; |
230d50d4 JA |
3332 | } else if (ret == -EIOCBQUEUED) { |
3333 | goto out_free; | |
7335e3bf | 3334 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3335 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3336 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3337 | goto done; |
227c0c96 JA |
3338 | } |
3339 | ||
227c0c96 | 3340 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3341 | if (ret2) |
3342 | return ret2; | |
3343 | ||
fe1cdd55 | 3344 | iovec = NULL; |
e8c2bc1f | 3345 | rw = req->async_data; |
227c0c96 | 3346 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3347 | iter = &rw->iter; |
227c0c96 | 3348 | |
b23df91b PB |
3349 | do { |
3350 | io_size -= ret; | |
3351 | rw->bytes_done += ret; | |
3352 | /* if we can retry, do so with the callbacks armed */ | |
3353 | if (!io_rw_should_retry(req)) { | |
3354 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3355 | return -EAGAIN; | |
3356 | } | |
3357 | ||
3358 | /* | |
3359 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3360 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3361 | * desired page gets unlocked. We can also get a partial read | |
3362 | * here, and if we do, then just retry at the new offset. | |
3363 | */ | |
3364 | ret = io_iter_do_read(req, iter); | |
3365 | if (ret == -EIOCBQUEUED) | |
3366 | return 0; | |
227c0c96 | 3367 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3368 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3369 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3370 | done: |
889fca73 | 3371 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3372 | out_free: |
3373 | /* it's faster to check here then delegate to kfree */ | |
3374 | if (iovec) | |
3375 | kfree(iovec); | |
5ea5dd45 | 3376 | return 0; |
2b188cc1 JA |
3377 | } |
3378 | ||
73debe68 | 3379 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3380 | { |
3529d8c2 JA |
3381 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3382 | return -EBADF; | |
93642ef8 | 3383 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3384 | } |
3385 | ||
889fca73 | 3386 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3387 | { |
3388 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3389 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3390 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3391 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3392 | ssize_t ret, ret2, io_size; |
45d189c6 | 3393 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3394 | |
2846c481 | 3395 | if (rw) { |
e8c2bc1f | 3396 | iter = &rw->iter; |
2846c481 PB |
3397 | iovec = NULL; |
3398 | } else { | |
3399 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3400 | if (ret < 0) | |
3401 | return ret; | |
3402 | } | |
632546c4 | 3403 | io_size = iov_iter_count(iter); |
fa15bafb | 3404 | req->result = io_size; |
2b188cc1 | 3405 | |
fd6c2e4c JA |
3406 | /* Ensure we clear previously set non-block flag */ |
3407 | if (!force_nonblock) | |
a88fc400 PB |
3408 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3409 | else | |
3410 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3411 | |
24c74678 | 3412 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3413 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3414 | goto copy_iov; |
31b51510 | 3415 | |
10d59345 JA |
3416 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3417 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3418 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3419 | goto copy_iov; |
31b51510 | 3420 | |
632546c4 | 3421 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3422 | if (unlikely(ret)) |
3423 | goto out_free; | |
4ed734b0 | 3424 | |
fa15bafb PB |
3425 | /* |
3426 | * Open-code file_start_write here to grab freeze protection, | |
3427 | * which will be released by another thread in | |
3428 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3429 | * released so that it doesn't complain about the held lock when | |
3430 | * we return to userspace. | |
3431 | */ | |
3432 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3433 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3434 | __sb_writers_release(file_inode(req->file)->i_sb, |
3435 | SB_FREEZE_WRITE); | |
3436 | } | |
3437 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3438 | |
fa15bafb | 3439 | if (req->file->f_op->write_iter) |
ff6165b2 | 3440 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3441 | else if (req->file->f_op->write) |
4017eb91 | 3442 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3443 | else |
3444 | ret2 = -EINVAL; | |
4ed734b0 | 3445 | |
6ad7f233 PB |
3446 | if (req->flags & REQ_F_REISSUE) { |
3447 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3448 | ret2 = -EAGAIN; |
6ad7f233 | 3449 | } |
230d50d4 | 3450 | |
fa15bafb PB |
3451 | /* |
3452 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3453 | * retry them without IOCB_NOWAIT. | |
3454 | */ | |
3455 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3456 | ret2 = -EAGAIN; | |
75c668cd PB |
3457 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3458 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3459 | goto done; |
fa15bafb | 3460 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3461 | /* IOPOLL retry should happen for io-wq threads */ |
3462 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3463 | goto copy_iov; | |
355afaeb | 3464 | done: |
889fca73 | 3465 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3466 | } else { |
f67676d1 | 3467 | copy_iov: |
84216315 | 3468 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3469 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3470 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3471 | return ret ?: -EAGAIN; |
2b188cc1 | 3472 | } |
31b51510 | 3473 | out_free: |
f261c168 | 3474 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3475 | if (iovec) |
6f2cc166 | 3476 | kfree(iovec); |
2b188cc1 JA |
3477 | return ret; |
3478 | } | |
3479 | ||
80a261fd JA |
3480 | static int io_renameat_prep(struct io_kiocb *req, |
3481 | const struct io_uring_sqe *sqe) | |
3482 | { | |
3483 | struct io_rename *ren = &req->rename; | |
3484 | const char __user *oldf, *newf; | |
3485 | ||
3486 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3487 | return -EBADF; | |
3488 | ||
3489 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3490 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3491 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3492 | ren->new_dfd = READ_ONCE(sqe->len); | |
3493 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3494 | ||
3495 | ren->oldpath = getname(oldf); | |
3496 | if (IS_ERR(ren->oldpath)) | |
3497 | return PTR_ERR(ren->oldpath); | |
3498 | ||
3499 | ren->newpath = getname(newf); | |
3500 | if (IS_ERR(ren->newpath)) { | |
3501 | putname(ren->oldpath); | |
3502 | return PTR_ERR(ren->newpath); | |
3503 | } | |
3504 | ||
3505 | req->flags |= REQ_F_NEED_CLEANUP; | |
3506 | return 0; | |
3507 | } | |
3508 | ||
45d189c6 | 3509 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3510 | { |
3511 | struct io_rename *ren = &req->rename; | |
3512 | int ret; | |
3513 | ||
45d189c6 | 3514 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3515 | return -EAGAIN; |
3516 | ||
3517 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3518 | ren->newpath, ren->flags); | |
3519 | ||
3520 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3521 | if (ret < 0) | |
3522 | req_set_fail_links(req); | |
3523 | io_req_complete(req, ret); | |
3524 | return 0; | |
3525 | } | |
3526 | ||
14a1143b JA |
3527 | static int io_unlinkat_prep(struct io_kiocb *req, |
3528 | const struct io_uring_sqe *sqe) | |
3529 | { | |
3530 | struct io_unlink *un = &req->unlink; | |
3531 | const char __user *fname; | |
3532 | ||
3533 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3534 | return -EBADF; | |
3535 | ||
3536 | un->dfd = READ_ONCE(sqe->fd); | |
3537 | ||
3538 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3539 | if (un->flags & ~AT_REMOVEDIR) | |
3540 | return -EINVAL; | |
3541 | ||
3542 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3543 | un->filename = getname(fname); | |
3544 | if (IS_ERR(un->filename)) | |
3545 | return PTR_ERR(un->filename); | |
3546 | ||
3547 | req->flags |= REQ_F_NEED_CLEANUP; | |
3548 | return 0; | |
3549 | } | |
3550 | ||
45d189c6 | 3551 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3552 | { |
3553 | struct io_unlink *un = &req->unlink; | |
3554 | int ret; | |
3555 | ||
45d189c6 | 3556 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3557 | return -EAGAIN; |
3558 | ||
3559 | if (un->flags & AT_REMOVEDIR) | |
3560 | ret = do_rmdir(un->dfd, un->filename); | |
3561 | else | |
3562 | ret = do_unlinkat(un->dfd, un->filename); | |
3563 | ||
3564 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3565 | if (ret < 0) | |
3566 | req_set_fail_links(req); | |
3567 | io_req_complete(req, ret); | |
3568 | return 0; | |
3569 | } | |
3570 | ||
36f4fa68 JA |
3571 | static int io_shutdown_prep(struct io_kiocb *req, |
3572 | const struct io_uring_sqe *sqe) | |
3573 | { | |
3574 | #if defined(CONFIG_NET) | |
3575 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3576 | return -EINVAL; | |
3577 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3578 | sqe->buf_index) | |
3579 | return -EINVAL; | |
3580 | ||
3581 | req->shutdown.how = READ_ONCE(sqe->len); | |
3582 | return 0; | |
3583 | #else | |
3584 | return -EOPNOTSUPP; | |
3585 | #endif | |
3586 | } | |
3587 | ||
45d189c6 | 3588 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3589 | { |
3590 | #if defined(CONFIG_NET) | |
3591 | struct socket *sock; | |
3592 | int ret; | |
3593 | ||
45d189c6 | 3594 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3595 | return -EAGAIN; |
3596 | ||
48aba79b | 3597 | sock = sock_from_file(req->file); |
36f4fa68 | 3598 | if (unlikely(!sock)) |
48aba79b | 3599 | return -ENOTSOCK; |
36f4fa68 JA |
3600 | |
3601 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3602 | if (ret < 0) |
3603 | req_set_fail_links(req); | |
36f4fa68 JA |
3604 | io_req_complete(req, ret); |
3605 | return 0; | |
3606 | #else | |
3607 | return -EOPNOTSUPP; | |
3608 | #endif | |
3609 | } | |
3610 | ||
f2a8d5c7 PB |
3611 | static int __io_splice_prep(struct io_kiocb *req, |
3612 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3613 | { |
3614 | struct io_splice* sp = &req->splice; | |
3615 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3616 | |
3232dd02 PB |
3617 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3618 | return -EINVAL; | |
7d67af2c PB |
3619 | |
3620 | sp->file_in = NULL; | |
7d67af2c PB |
3621 | sp->len = READ_ONCE(sqe->len); |
3622 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3623 | ||
3624 | if (unlikely(sp->flags & ~valid_flags)) | |
3625 | return -EINVAL; | |
3626 | ||
8371adf5 PB |
3627 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3628 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3629 | if (!sp->file_in) | |
3630 | return -EBADF; | |
7d67af2c PB |
3631 | req->flags |= REQ_F_NEED_CLEANUP; |
3632 | ||
7cdaf587 XW |
3633 | if (!S_ISREG(file_inode(sp->file_in)->i_mode)) { |
3634 | /* | |
3635 | * Splice operation will be punted aync, and here need to | |
3636 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
3637 | */ | |
7d67af2c | 3638 | req->work.flags |= IO_WQ_WORK_UNBOUND; |
7cdaf587 | 3639 | } |
7d67af2c PB |
3640 | |
3641 | return 0; | |
3642 | } | |
3643 | ||
f2a8d5c7 PB |
3644 | static int io_tee_prep(struct io_kiocb *req, |
3645 | const struct io_uring_sqe *sqe) | |
3646 | { | |
3647 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3648 | return -EINVAL; | |
3649 | return __io_splice_prep(req, sqe); | |
3650 | } | |
3651 | ||
45d189c6 | 3652 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3653 | { |
3654 | struct io_splice *sp = &req->splice; | |
3655 | struct file *in = sp->file_in; | |
3656 | struct file *out = sp->file_out; | |
3657 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3658 | long ret = 0; | |
3659 | ||
45d189c6 | 3660 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3661 | return -EAGAIN; |
3662 | if (sp->len) | |
3663 | ret = do_tee(in, out, sp->len, flags); | |
3664 | ||
3665 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3666 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3667 | ||
f2a8d5c7 PB |
3668 | if (ret != sp->len) |
3669 | req_set_fail_links(req); | |
e1e16097 | 3670 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3671 | return 0; |
3672 | } | |
3673 | ||
3674 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3675 | { | |
3676 | struct io_splice* sp = &req->splice; | |
3677 | ||
3678 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3679 | sp->off_out = READ_ONCE(sqe->off); | |
3680 | return __io_splice_prep(req, sqe); | |
3681 | } | |
3682 | ||
45d189c6 | 3683 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3684 | { |
3685 | struct io_splice *sp = &req->splice; | |
3686 | struct file *in = sp->file_in; | |
3687 | struct file *out = sp->file_out; | |
3688 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3689 | loff_t *poff_in, *poff_out; | |
c9687426 | 3690 | long ret = 0; |
7d67af2c | 3691 | |
45d189c6 | 3692 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3693 | return -EAGAIN; |
7d67af2c PB |
3694 | |
3695 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3696 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3697 | |
948a7749 | 3698 | if (sp->len) |
c9687426 | 3699 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c PB |
3700 | |
3701 | io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3702 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3703 | ||
7d67af2c PB |
3704 | if (ret != sp->len) |
3705 | req_set_fail_links(req); | |
e1e16097 | 3706 | io_req_complete(req, ret); |
7d67af2c PB |
3707 | return 0; |
3708 | } | |
3709 | ||
2b188cc1 JA |
3710 | /* |
3711 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3712 | */ | |
889fca73 | 3713 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3714 | { |
3715 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3716 | |
def596e9 JA |
3717 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3718 | return -EINVAL; | |
3719 | ||
889fca73 | 3720 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3721 | return 0; |
3722 | } | |
3723 | ||
1155c76a | 3724 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3725 | { |
6b06314c | 3726 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3727 | |
09bb8394 JA |
3728 | if (!req->file) |
3729 | return -EBADF; | |
c992fe29 | 3730 | |
6b06314c | 3731 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3732 | return -EINVAL; |
edafccee | 3733 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3734 | return -EINVAL; |
3735 | ||
8ed8d3c3 JA |
3736 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3737 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3738 | return -EINVAL; | |
3739 | ||
3740 | req->sync.off = READ_ONCE(sqe->off); | |
3741 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3742 | return 0; |
3743 | } | |
3744 | ||
45d189c6 | 3745 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3746 | { |
8ed8d3c3 | 3747 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3748 | int ret; |
3749 | ||
ac45abc0 | 3750 | /* fsync always requires a blocking context */ |
45d189c6 | 3751 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3752 | return -EAGAIN; |
3753 | ||
9adbd45d | 3754 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3755 | end > 0 ? end : LLONG_MAX, |
3756 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3757 | if (ret < 0) | |
3758 | req_set_fail_links(req); | |
e1e16097 | 3759 | io_req_complete(req, ret); |
c992fe29 CH |
3760 | return 0; |
3761 | } | |
3762 | ||
d63d1b5e JA |
3763 | static int io_fallocate_prep(struct io_kiocb *req, |
3764 | const struct io_uring_sqe *sqe) | |
3765 | { | |
3766 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3767 | return -EINVAL; | |
3232dd02 PB |
3768 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3769 | return -EINVAL; | |
d63d1b5e JA |
3770 | |
3771 | req->sync.off = READ_ONCE(sqe->off); | |
3772 | req->sync.len = READ_ONCE(sqe->addr); | |
3773 | req->sync.mode = READ_ONCE(sqe->len); | |
3774 | return 0; | |
3775 | } | |
3776 | ||
45d189c6 | 3777 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3778 | { |
ac45abc0 PB |
3779 | int ret; |
3780 | ||
d63d1b5e | 3781 | /* fallocate always requiring blocking context */ |
45d189c6 | 3782 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3783 | return -EAGAIN; |
ac45abc0 PB |
3784 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3785 | req->sync.len); | |
ac45abc0 PB |
3786 | if (ret < 0) |
3787 | req_set_fail_links(req); | |
e1e16097 | 3788 | io_req_complete(req, ret); |
5d17b4a4 JA |
3789 | return 0; |
3790 | } | |
3791 | ||
ec65fea5 | 3792 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3793 | { |
f8748881 | 3794 | const char __user *fname; |
15b71abe | 3795 | int ret; |
b7bb4f7d | 3796 | |
ec65fea5 | 3797 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3798 | return -EINVAL; |
ec65fea5 | 3799 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3800 | return -EBADF; |
03b1230c | 3801 | |
ec65fea5 PB |
3802 | /* open.how should be already initialised */ |
3803 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3804 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3805 | |
25e72d10 PB |
3806 | req->open.dfd = READ_ONCE(sqe->fd); |
3807 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3808 | req->open.filename = getname(fname); |
15b71abe JA |
3809 | if (IS_ERR(req->open.filename)) { |
3810 | ret = PTR_ERR(req->open.filename); | |
3811 | req->open.filename = NULL; | |
3812 | return ret; | |
3813 | } | |
4022e7af | 3814 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3815 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3816 | return 0; |
03b1230c JA |
3817 | } |
3818 | ||
ec65fea5 PB |
3819 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3820 | { | |
3821 | u64 flags, mode; | |
3822 | ||
14587a46 | 3823 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3824 | return -EINVAL; |
ec65fea5 PB |
3825 | mode = READ_ONCE(sqe->len); |
3826 | flags = READ_ONCE(sqe->open_flags); | |
3827 | req->open.how = build_open_how(flags, mode); | |
3828 | return __io_openat_prep(req, sqe); | |
3829 | } | |
3830 | ||
cebdb986 | 3831 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3832 | { |
cebdb986 | 3833 | struct open_how __user *how; |
cebdb986 | 3834 | size_t len; |
0fa03c62 JA |
3835 | int ret; |
3836 | ||
14587a46 | 3837 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3838 | return -EINVAL; |
cebdb986 JA |
3839 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3840 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3841 | if (len < OPEN_HOW_SIZE_VER0) |
3842 | return -EINVAL; | |
3529d8c2 | 3843 | |
cebdb986 JA |
3844 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3845 | len); | |
3846 | if (ret) | |
3847 | return ret; | |
3529d8c2 | 3848 | |
ec65fea5 | 3849 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3850 | } |
3851 | ||
45d189c6 | 3852 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3853 | { |
3854 | struct open_flags op; | |
15b71abe | 3855 | struct file *file; |
3a81fd02 JA |
3856 | bool nonblock_set; |
3857 | bool resolve_nonblock; | |
15b71abe JA |
3858 | int ret; |
3859 | ||
cebdb986 | 3860 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3861 | if (ret) |
3862 | goto err; | |
3a81fd02 JA |
3863 | nonblock_set = op.open_flag & O_NONBLOCK; |
3864 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3865 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3866 | /* |
3867 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3868 | * it'll always -EAGAIN | |
3869 | */ | |
3870 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3871 | return -EAGAIN; | |
3872 | op.lookup_flags |= LOOKUP_CACHED; | |
3873 | op.open_flag |= O_NONBLOCK; | |
3874 | } | |
15b71abe | 3875 | |
4022e7af | 3876 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3877 | if (ret < 0) |
3878 | goto err; | |
3879 | ||
3880 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3881 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3882 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3883 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3884 | /* |
3a81fd02 JA |
3885 | * We could hang on to this 'fd', but seems like marginal |
3886 | * gain for something that is now known to be a slower path. | |
3887 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3888 | */ |
3a81fd02 JA |
3889 | put_unused_fd(ret); |
3890 | return -EAGAIN; | |
3891 | } | |
3892 | ||
15b71abe JA |
3893 | if (IS_ERR(file)) { |
3894 | put_unused_fd(ret); | |
3895 | ret = PTR_ERR(file); | |
3896 | } else { | |
45d189c6 | 3897 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3898 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3899 | fsnotify_open(file); |
3900 | fd_install(ret, file); | |
3901 | } | |
3902 | err: | |
3903 | putname(req->open.filename); | |
8fef80bf | 3904 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3905 | if (ret < 0) |
3906 | req_set_fail_links(req); | |
e1e16097 | 3907 | io_req_complete(req, ret); |
15b71abe JA |
3908 | return 0; |
3909 | } | |
3910 | ||
45d189c6 | 3911 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3912 | { |
e45cff58 | 3913 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3914 | } |
3915 | ||
067524e9 JA |
3916 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3917 | const struct io_uring_sqe *sqe) | |
3918 | { | |
3919 | struct io_provide_buf *p = &req->pbuf; | |
3920 | u64 tmp; | |
3921 | ||
3922 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3923 | return -EINVAL; | |
3924 | ||
3925 | tmp = READ_ONCE(sqe->fd); | |
3926 | if (!tmp || tmp > USHRT_MAX) | |
3927 | return -EINVAL; | |
3928 | ||
3929 | memset(p, 0, sizeof(*p)); | |
3930 | p->nbufs = tmp; | |
3931 | p->bgid = READ_ONCE(sqe->buf_group); | |
3932 | return 0; | |
3933 | } | |
3934 | ||
3935 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3936 | int bgid, unsigned nbufs) | |
3937 | { | |
3938 | unsigned i = 0; | |
3939 | ||
3940 | /* shouldn't happen */ | |
3941 | if (!nbufs) | |
3942 | return 0; | |
3943 | ||
3944 | /* the head kbuf is the list itself */ | |
3945 | while (!list_empty(&buf->list)) { | |
3946 | struct io_buffer *nxt; | |
3947 | ||
3948 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3949 | list_del(&nxt->list); | |
3950 | kfree(nxt); | |
3951 | if (++i == nbufs) | |
3952 | return i; | |
3953 | } | |
3954 | i++; | |
3955 | kfree(buf); | |
9e15c3a0 | 3956 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3957 | |
3958 | return i; | |
3959 | } | |
3960 | ||
889fca73 | 3961 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3962 | { |
3963 | struct io_provide_buf *p = &req->pbuf; | |
3964 | struct io_ring_ctx *ctx = req->ctx; | |
3965 | struct io_buffer *head; | |
3966 | int ret = 0; | |
45d189c6 | 3967 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3968 | |
3969 | io_ring_submit_lock(ctx, !force_nonblock); | |
3970 | ||
3971 | lockdep_assert_held(&ctx->uring_lock); | |
3972 | ||
3973 | ret = -ENOENT; | |
9e15c3a0 | 3974 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3975 | if (head) |
3976 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3977 | if (ret < 0) |
3978 | req_set_fail_links(req); | |
067524e9 | 3979 | |
9fb8cb49 PB |
3980 | /* complete before unlock, IOPOLL may need the lock */ |
3981 | __io_req_complete(req, issue_flags, ret, 0); | |
3982 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3983 | return 0; |
3984 | } | |
3985 | ||
ddf0322d JA |
3986 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3987 | const struct io_uring_sqe *sqe) | |
3988 | { | |
d81269fe | 3989 | unsigned long size; |
ddf0322d JA |
3990 | struct io_provide_buf *p = &req->pbuf; |
3991 | u64 tmp; | |
3992 | ||
3993 | if (sqe->ioprio || sqe->rw_flags) | |
3994 | return -EINVAL; | |
3995 | ||
3996 | tmp = READ_ONCE(sqe->fd); | |
3997 | if (!tmp || tmp > USHRT_MAX) | |
3998 | return -E2BIG; | |
3999 | p->nbufs = tmp; | |
4000 | p->addr = READ_ONCE(sqe->addr); | |
4001 | p->len = READ_ONCE(sqe->len); | |
4002 | ||
d81269fe PB |
4003 | size = (unsigned long)p->len * p->nbufs; |
4004 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
4005 | return -EFAULT; |
4006 | ||
4007 | p->bgid = READ_ONCE(sqe->buf_group); | |
4008 | tmp = READ_ONCE(sqe->off); | |
4009 | if (tmp > USHRT_MAX) | |
4010 | return -E2BIG; | |
4011 | p->bid = tmp; | |
4012 | return 0; | |
4013 | } | |
4014 | ||
4015 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
4016 | { | |
4017 | struct io_buffer *buf; | |
4018 | u64 addr = pbuf->addr; | |
4019 | int i, bid = pbuf->bid; | |
4020 | ||
4021 | for (i = 0; i < pbuf->nbufs; i++) { | |
4022 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
4023 | if (!buf) | |
4024 | break; | |
4025 | ||
4026 | buf->addr = addr; | |
4027 | buf->len = pbuf->len; | |
4028 | buf->bid = bid; | |
4029 | addr += pbuf->len; | |
4030 | bid++; | |
4031 | if (!*head) { | |
4032 | INIT_LIST_HEAD(&buf->list); | |
4033 | *head = buf; | |
4034 | } else { | |
4035 | list_add_tail(&buf->list, &(*head)->list); | |
4036 | } | |
4037 | } | |
4038 | ||
4039 | return i ? i : -ENOMEM; | |
4040 | } | |
4041 | ||
889fca73 | 4042 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4043 | { |
4044 | struct io_provide_buf *p = &req->pbuf; | |
4045 | struct io_ring_ctx *ctx = req->ctx; | |
4046 | struct io_buffer *head, *list; | |
4047 | int ret = 0; | |
45d189c6 | 4048 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4049 | |
4050 | io_ring_submit_lock(ctx, !force_nonblock); | |
4051 | ||
4052 | lockdep_assert_held(&ctx->uring_lock); | |
4053 | ||
9e15c3a0 | 4054 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4055 | |
4056 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4057 | if (ret >= 0 && !list) { |
4058 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4059 | if (ret < 0) | |
067524e9 | 4060 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4061 | } |
ddf0322d JA |
4062 | if (ret < 0) |
4063 | req_set_fail_links(req); | |
9fb8cb49 PB |
4064 | /* complete before unlock, IOPOLL may need the lock */ |
4065 | __io_req_complete(req, issue_flags, ret, 0); | |
4066 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4067 | return 0; |
cebdb986 JA |
4068 | } |
4069 | ||
3e4827b0 JA |
4070 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4071 | const struct io_uring_sqe *sqe) | |
4072 | { | |
4073 | #if defined(CONFIG_EPOLL) | |
4074 | if (sqe->ioprio || sqe->buf_index) | |
4075 | return -EINVAL; | |
6ca56f84 | 4076 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4077 | return -EINVAL; |
3e4827b0 JA |
4078 | |
4079 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4080 | req->epoll.op = READ_ONCE(sqe->len); | |
4081 | req->epoll.fd = READ_ONCE(sqe->off); | |
4082 | ||
4083 | if (ep_op_has_event(req->epoll.op)) { | |
4084 | struct epoll_event __user *ev; | |
4085 | ||
4086 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4087 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4088 | return -EFAULT; | |
4089 | } | |
4090 | ||
4091 | return 0; | |
4092 | #else | |
4093 | return -EOPNOTSUPP; | |
4094 | #endif | |
4095 | } | |
4096 | ||
889fca73 | 4097 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4098 | { |
4099 | #if defined(CONFIG_EPOLL) | |
4100 | struct io_epoll *ie = &req->epoll; | |
4101 | int ret; | |
45d189c6 | 4102 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4103 | |
4104 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4105 | if (force_nonblock && ret == -EAGAIN) | |
4106 | return -EAGAIN; | |
4107 | ||
4108 | if (ret < 0) | |
4109 | req_set_fail_links(req); | |
889fca73 | 4110 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4111 | return 0; |
4112 | #else | |
4113 | return -EOPNOTSUPP; | |
4114 | #endif | |
4115 | } | |
4116 | ||
c1ca757b JA |
4117 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4118 | { | |
4119 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4120 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4121 | return -EINVAL; | |
3232dd02 PB |
4122 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4123 | return -EINVAL; | |
c1ca757b JA |
4124 | |
4125 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4126 | req->madvise.len = READ_ONCE(sqe->len); | |
4127 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4128 | return 0; | |
4129 | #else | |
4130 | return -EOPNOTSUPP; | |
4131 | #endif | |
4132 | } | |
4133 | ||
45d189c6 | 4134 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4135 | { |
4136 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4137 | struct io_madvise *ma = &req->madvise; | |
4138 | int ret; | |
4139 | ||
45d189c6 | 4140 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4141 | return -EAGAIN; |
4142 | ||
0726b01e | 4143 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4144 | if (ret < 0) |
4145 | req_set_fail_links(req); | |
e1e16097 | 4146 | io_req_complete(req, ret); |
c1ca757b JA |
4147 | return 0; |
4148 | #else | |
4149 | return -EOPNOTSUPP; | |
4150 | #endif | |
4151 | } | |
4152 | ||
4840e418 JA |
4153 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4154 | { | |
4155 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4156 | return -EINVAL; | |
3232dd02 PB |
4157 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4158 | return -EINVAL; | |
4840e418 JA |
4159 | |
4160 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4161 | req->fadvise.len = READ_ONCE(sqe->len); | |
4162 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4163 | return 0; | |
4164 | } | |
4165 | ||
45d189c6 | 4166 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4167 | { |
4168 | struct io_fadvise *fa = &req->fadvise; | |
4169 | int ret; | |
4170 | ||
45d189c6 | 4171 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4172 | switch (fa->advice) { |
4173 | case POSIX_FADV_NORMAL: | |
4174 | case POSIX_FADV_RANDOM: | |
4175 | case POSIX_FADV_SEQUENTIAL: | |
4176 | break; | |
4177 | default: | |
4178 | return -EAGAIN; | |
4179 | } | |
4180 | } | |
4840e418 JA |
4181 | |
4182 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4183 | if (ret < 0) | |
4184 | req_set_fail_links(req); | |
e1e16097 | 4185 | io_req_complete(req, ret); |
4840e418 JA |
4186 | return 0; |
4187 | } | |
4188 | ||
eddc7ef5 JA |
4189 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4190 | { | |
6ca56f84 | 4191 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4192 | return -EINVAL; |
eddc7ef5 JA |
4193 | if (sqe->ioprio || sqe->buf_index) |
4194 | return -EINVAL; | |
9c280f90 | 4195 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4196 | return -EBADF; |
eddc7ef5 | 4197 | |
1d9e1288 BM |
4198 | req->statx.dfd = READ_ONCE(sqe->fd); |
4199 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4200 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4201 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4202 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4203 | |
4204 | return 0; | |
4205 | } | |
4206 | ||
45d189c6 | 4207 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4208 | { |
1d9e1288 | 4209 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4210 | int ret; |
4211 | ||
45d189c6 | 4212 | if (issue_flags & IO_URING_F_NONBLOCK) { |
5b0bbee4 JA |
4213 | /* only need file table for an actual valid fd */ |
4214 | if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD) | |
4215 | req->flags |= REQ_F_NO_FILE_TABLE; | |
eddc7ef5 | 4216 | return -EAGAIN; |
5b0bbee4 | 4217 | } |
eddc7ef5 | 4218 | |
e62753e4 BM |
4219 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4220 | ctx->buffer); | |
eddc7ef5 | 4221 | |
eddc7ef5 JA |
4222 | if (ret < 0) |
4223 | req_set_fail_links(req); | |
e1e16097 | 4224 | io_req_complete(req, ret); |
eddc7ef5 JA |
4225 | return 0; |
4226 | } | |
4227 | ||
b5dba59e JA |
4228 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4229 | { | |
14587a46 | 4230 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4231 | return -EINVAL; |
b5dba59e JA |
4232 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4233 | sqe->rw_flags || sqe->buf_index) | |
4234 | return -EINVAL; | |
9c280f90 | 4235 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4236 | return -EBADF; |
b5dba59e JA |
4237 | |
4238 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4239 | return 0; |
b5dba59e JA |
4240 | } |
4241 | ||
889fca73 | 4242 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4243 | { |
9eac1904 | 4244 | struct files_struct *files = current->files; |
3af73b28 | 4245 | struct io_close *close = &req->close; |
9eac1904 JA |
4246 | struct fdtable *fdt; |
4247 | struct file *file; | |
b5dba59e JA |
4248 | int ret; |
4249 | ||
9eac1904 JA |
4250 | file = NULL; |
4251 | ret = -EBADF; | |
4252 | spin_lock(&files->file_lock); | |
4253 | fdt = files_fdtable(files); | |
4254 | if (close->fd >= fdt->max_fds) { | |
4255 | spin_unlock(&files->file_lock); | |
4256 | goto err; | |
4257 | } | |
4258 | file = fdt->fd[close->fd]; | |
4259 | if (!file) { | |
4260 | spin_unlock(&files->file_lock); | |
4261 | goto err; | |
4262 | } | |
4263 | ||
4264 | if (file->f_op == &io_uring_fops) { | |
4265 | spin_unlock(&files->file_lock); | |
4266 | file = NULL; | |
4267 | goto err; | |
3af73b28 | 4268 | } |
b5dba59e JA |
4269 | |
4270 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4271 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4272 | spin_unlock(&files->file_lock); |
0bf0eefd | 4273 | return -EAGAIN; |
a2100672 | 4274 | } |
b5dba59e | 4275 | |
9eac1904 JA |
4276 | ret = __close_fd_get_file(close->fd, &file); |
4277 | spin_unlock(&files->file_lock); | |
4278 | if (ret < 0) { | |
4279 | if (ret == -ENOENT) | |
4280 | ret = -EBADF; | |
4281 | goto err; | |
4282 | } | |
4283 | ||
3af73b28 | 4284 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4285 | ret = filp_close(file, current->files); |
4286 | err: | |
3af73b28 PB |
4287 | if (ret < 0) |
4288 | req_set_fail_links(req); | |
9eac1904 JA |
4289 | if (file) |
4290 | fput(file); | |
889fca73 | 4291 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4292 | return 0; |
b5dba59e JA |
4293 | } |
4294 | ||
1155c76a | 4295 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4296 | { |
4297 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4298 | |
5d17b4a4 JA |
4299 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4300 | return -EINVAL; | |
4301 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4302 | return -EINVAL; | |
4303 | ||
8ed8d3c3 JA |
4304 | req->sync.off = READ_ONCE(sqe->off); |
4305 | req->sync.len = READ_ONCE(sqe->len); | |
4306 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4307 | return 0; |
4308 | } | |
4309 | ||
45d189c6 | 4310 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4311 | { |
8ed8d3c3 JA |
4312 | int ret; |
4313 | ||
ac45abc0 | 4314 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4315 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4316 | return -EAGAIN; |
4317 | ||
9adbd45d | 4318 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4319 | req->sync.flags); |
4320 | if (ret < 0) | |
4321 | req_set_fail_links(req); | |
e1e16097 | 4322 | io_req_complete(req, ret); |
5d17b4a4 JA |
4323 | return 0; |
4324 | } | |
4325 | ||
469956e8 | 4326 | #if defined(CONFIG_NET) |
02d27d89 PB |
4327 | static int io_setup_async_msg(struct io_kiocb *req, |
4328 | struct io_async_msghdr *kmsg) | |
4329 | { | |
e8c2bc1f JA |
4330 | struct io_async_msghdr *async_msg = req->async_data; |
4331 | ||
4332 | if (async_msg) | |
02d27d89 | 4333 | return -EAGAIN; |
e8c2bc1f | 4334 | if (io_alloc_async_data(req)) { |
257e84a5 | 4335 | kfree(kmsg->free_iov); |
02d27d89 PB |
4336 | return -ENOMEM; |
4337 | } | |
e8c2bc1f | 4338 | async_msg = req->async_data; |
02d27d89 | 4339 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4340 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4341 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4342 | /* if were using fast_iov, set it to the new one */ |
4343 | if (!async_msg->free_iov) | |
4344 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4345 | ||
02d27d89 PB |
4346 | return -EAGAIN; |
4347 | } | |
4348 | ||
2ae523ed PB |
4349 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4350 | struct io_async_msghdr *iomsg) | |
4351 | { | |
2ae523ed | 4352 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4353 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4354 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4355 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4356 | } |
4357 | ||
93642ef8 PB |
4358 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4359 | { | |
4360 | int ret; | |
4361 | ||
93642ef8 PB |
4362 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4363 | if (!ret) | |
4364 | req->flags |= REQ_F_NEED_CLEANUP; | |
4365 | return ret; | |
4366 | } | |
4367 | ||
3529d8c2 | 4368 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4369 | { |
e47293fd | 4370 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4371 | |
d2b6f48b PB |
4372 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4373 | return -EINVAL; | |
4374 | ||
e47293fd | 4375 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4376 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4377 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4378 | |
d8768362 JA |
4379 | #ifdef CONFIG_COMPAT |
4380 | if (req->ctx->compat) | |
4381 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4382 | #endif | |
93642ef8 | 4383 | return 0; |
03b1230c JA |
4384 | } |
4385 | ||
889fca73 | 4386 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4387 | { |
6b754c8b | 4388 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4389 | struct socket *sock; |
7a7cacba | 4390 | unsigned flags; |
0031275d | 4391 | int min_ret = 0; |
0fa03c62 JA |
4392 | int ret; |
4393 | ||
dba4a925 | 4394 | sock = sock_from_file(req->file); |
7a7cacba | 4395 | if (unlikely(!sock)) |
dba4a925 | 4396 | return -ENOTSOCK; |
3529d8c2 | 4397 | |
257e84a5 PB |
4398 | kmsg = req->async_data; |
4399 | if (!kmsg) { | |
7a7cacba PB |
4400 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4401 | if (ret) | |
4402 | return ret; | |
4403 | kmsg = &iomsg; | |
0fa03c62 | 4404 | } |
0fa03c62 | 4405 | |
76cd979f | 4406 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4407 | if (flags & MSG_DONTWAIT) |
4408 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4409 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4410 | flags |= MSG_DONTWAIT; |
e47293fd | 4411 | |
0031275d SM |
4412 | if (flags & MSG_WAITALL) |
4413 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4414 | ||
7a7cacba | 4415 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4416 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4417 | return io_setup_async_msg(req, kmsg); |
4418 | if (ret == -ERESTARTSYS) | |
4419 | ret = -EINTR; | |
0fa03c62 | 4420 | |
257e84a5 PB |
4421 | /* fast path, check for non-NULL to avoid function call */ |
4422 | if (kmsg->free_iov) | |
4423 | kfree(kmsg->free_iov); | |
99bc4c38 | 4424 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4425 | if (ret < min_ret) |
4e88d6e7 | 4426 | req_set_fail_links(req); |
889fca73 | 4427 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4428 | return 0; |
03b1230c | 4429 | } |
aa1fa28f | 4430 | |
889fca73 | 4431 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4432 | { |
7a7cacba PB |
4433 | struct io_sr_msg *sr = &req->sr_msg; |
4434 | struct msghdr msg; | |
4435 | struct iovec iov; | |
fddaface | 4436 | struct socket *sock; |
7a7cacba | 4437 | unsigned flags; |
0031275d | 4438 | int min_ret = 0; |
fddaface JA |
4439 | int ret; |
4440 | ||
dba4a925 | 4441 | sock = sock_from_file(req->file); |
7a7cacba | 4442 | if (unlikely(!sock)) |
dba4a925 | 4443 | return -ENOTSOCK; |
fddaface | 4444 | |
7a7cacba PB |
4445 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4446 | if (unlikely(ret)) | |
14db8411 | 4447 | return ret; |
fddaface | 4448 | |
7a7cacba PB |
4449 | msg.msg_name = NULL; |
4450 | msg.msg_control = NULL; | |
4451 | msg.msg_controllen = 0; | |
4452 | msg.msg_namelen = 0; | |
fddaface | 4453 | |
76cd979f | 4454 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4455 | if (flags & MSG_DONTWAIT) |
4456 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4457 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4458 | flags |= MSG_DONTWAIT; |
fddaface | 4459 | |
0031275d SM |
4460 | if (flags & MSG_WAITALL) |
4461 | min_ret = iov_iter_count(&msg.msg_iter); | |
4462 | ||
7a7cacba PB |
4463 | msg.msg_flags = flags; |
4464 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4465 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4466 | return -EAGAIN; |
4467 | if (ret == -ERESTARTSYS) | |
4468 | ret = -EINTR; | |
fddaface | 4469 | |
0031275d | 4470 | if (ret < min_ret) |
fddaface | 4471 | req_set_fail_links(req); |
889fca73 | 4472 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4473 | return 0; |
fddaface JA |
4474 | } |
4475 | ||
1400e697 PB |
4476 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4477 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4478 | { |
4479 | struct io_sr_msg *sr = &req->sr_msg; | |
4480 | struct iovec __user *uiov; | |
4481 | size_t iov_len; | |
4482 | int ret; | |
4483 | ||
1400e697 PB |
4484 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4485 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4486 | if (ret) |
4487 | return ret; | |
4488 | ||
4489 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4490 | if (iov_len > 1) | |
4491 | return -EINVAL; | |
5476dfed | 4492 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4493 | return -EFAULT; |
5476dfed | 4494 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4495 | iomsg->free_iov = NULL; |
52de1fe1 | 4496 | } else { |
257e84a5 | 4497 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4498 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4499 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4500 | false); |
52de1fe1 JA |
4501 | if (ret > 0) |
4502 | ret = 0; | |
4503 | } | |
4504 | ||
4505 | return ret; | |
4506 | } | |
4507 | ||
4508 | #ifdef CONFIG_COMPAT | |
4509 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4510 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4511 | { |
4512 | struct compat_msghdr __user *msg_compat; | |
4513 | struct io_sr_msg *sr = &req->sr_msg; | |
4514 | struct compat_iovec __user *uiov; | |
4515 | compat_uptr_t ptr; | |
4516 | compat_size_t len; | |
4517 | int ret; | |
4518 | ||
270a5940 | 4519 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4520 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4521 | &ptr, &len); |
4522 | if (ret) | |
4523 | return ret; | |
4524 | ||
4525 | uiov = compat_ptr(ptr); | |
4526 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4527 | compat_ssize_t clen; | |
4528 | ||
4529 | if (len > 1) | |
4530 | return -EINVAL; | |
4531 | if (!access_ok(uiov, sizeof(*uiov))) | |
4532 | return -EFAULT; | |
4533 | if (__get_user(clen, &uiov->iov_len)) | |
4534 | return -EFAULT; | |
4535 | if (clen < 0) | |
4536 | return -EINVAL; | |
2d280bc8 | 4537 | sr->len = clen; |
257e84a5 | 4538 | iomsg->free_iov = NULL; |
52de1fe1 | 4539 | } else { |
257e84a5 | 4540 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4541 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4542 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4543 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4544 | if (ret < 0) |
4545 | return ret; | |
4546 | } | |
4547 | ||
4548 | return 0; | |
4549 | } | |
4550 | #endif | |
4551 | ||
1400e697 PB |
4552 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4553 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4554 | { |
1400e697 | 4555 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4556 | |
4557 | #ifdef CONFIG_COMPAT | |
4558 | if (req->ctx->compat) | |
1400e697 | 4559 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4560 | #endif |
52de1fe1 | 4561 | |
1400e697 | 4562 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4563 | } |
4564 | ||
bcda7baa | 4565 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4566 | bool needs_lock) |
bcda7baa JA |
4567 | { |
4568 | struct io_sr_msg *sr = &req->sr_msg; | |
4569 | struct io_buffer *kbuf; | |
4570 | ||
bcda7baa JA |
4571 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4572 | if (IS_ERR(kbuf)) | |
4573 | return kbuf; | |
4574 | ||
4575 | sr->kbuf = kbuf; | |
4576 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4577 | return kbuf; |
fddaface JA |
4578 | } |
4579 | ||
7fbb1b54 PB |
4580 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4581 | { | |
4582 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4583 | } | |
4584 | ||
93642ef8 | 4585 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4586 | { |
99bc4c38 | 4587 | int ret; |
3529d8c2 | 4588 | |
93642ef8 PB |
4589 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4590 | if (!ret) | |
4591 | req->flags |= REQ_F_NEED_CLEANUP; | |
4592 | return ret; | |
4593 | } | |
4594 | ||
4595 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4596 | { | |
4597 | struct io_sr_msg *sr = &req->sr_msg; | |
4598 | ||
d2b6f48b PB |
4599 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4600 | return -EINVAL; | |
4601 | ||
3529d8c2 | 4602 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4603 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4604 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4605 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4606 | |
d8768362 JA |
4607 | #ifdef CONFIG_COMPAT |
4608 | if (req->ctx->compat) | |
4609 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4610 | #endif | |
93642ef8 | 4611 | return 0; |
aa1fa28f JA |
4612 | } |
4613 | ||
889fca73 | 4614 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4615 | { |
6b754c8b | 4616 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4617 | struct socket *sock; |
7fbb1b54 | 4618 | struct io_buffer *kbuf; |
7a7cacba | 4619 | unsigned flags; |
0031275d | 4620 | int min_ret = 0; |
52de1fe1 | 4621 | int ret, cflags = 0; |
45d189c6 | 4622 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4623 | |
dba4a925 | 4624 | sock = sock_from_file(req->file); |
7a7cacba | 4625 | if (unlikely(!sock)) |
dba4a925 | 4626 | return -ENOTSOCK; |
3529d8c2 | 4627 | |
257e84a5 PB |
4628 | kmsg = req->async_data; |
4629 | if (!kmsg) { | |
7a7cacba PB |
4630 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4631 | if (ret) | |
681fda8d | 4632 | return ret; |
7a7cacba PB |
4633 | kmsg = &iomsg; |
4634 | } | |
03b1230c | 4635 | |
bc02ef33 | 4636 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4637 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4638 | if (IS_ERR(kbuf)) |
52de1fe1 | 4639 | return PTR_ERR(kbuf); |
7a7cacba | 4640 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4641 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4642 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4643 | 1, req->sr_msg.len); |
4644 | } | |
52de1fe1 | 4645 | |
76cd979f | 4646 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4647 | if (flags & MSG_DONTWAIT) |
4648 | req->flags |= REQ_F_NOWAIT; | |
4649 | else if (force_nonblock) | |
4650 | flags |= MSG_DONTWAIT; | |
e47293fd | 4651 | |
0031275d SM |
4652 | if (flags & MSG_WAITALL) |
4653 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4654 | ||
7a7cacba PB |
4655 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4656 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4657 | if (force_nonblock && ret == -EAGAIN) |
4658 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4659 | if (ret == -ERESTARTSYS) |
4660 | ret = -EINTR; | |
03b1230c | 4661 | |
7fbb1b54 PB |
4662 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4663 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4664 | /* fast path, check for non-NULL to avoid function call */ |
4665 | if (kmsg->free_iov) | |
4666 | kfree(kmsg->free_iov); | |
99bc4c38 | 4667 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4668 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4669 | req_set_fail_links(req); |
889fca73 | 4670 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4671 | return 0; |
0fa03c62 | 4672 | } |
5d17b4a4 | 4673 | |
889fca73 | 4674 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4675 | { |
6b754c8b | 4676 | struct io_buffer *kbuf; |
7a7cacba PB |
4677 | struct io_sr_msg *sr = &req->sr_msg; |
4678 | struct msghdr msg; | |
4679 | void __user *buf = sr->buf; | |
fddaface | 4680 | struct socket *sock; |
7a7cacba PB |
4681 | struct iovec iov; |
4682 | unsigned flags; | |
0031275d | 4683 | int min_ret = 0; |
bcda7baa | 4684 | int ret, cflags = 0; |
45d189c6 | 4685 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4686 | |
dba4a925 | 4687 | sock = sock_from_file(req->file); |
7a7cacba | 4688 | if (unlikely(!sock)) |
dba4a925 | 4689 | return -ENOTSOCK; |
fddaface | 4690 | |
bc02ef33 | 4691 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4692 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4693 | if (IS_ERR(kbuf)) |
4694 | return PTR_ERR(kbuf); | |
7a7cacba | 4695 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4696 | } |
bcda7baa | 4697 | |
7a7cacba | 4698 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4699 | if (unlikely(ret)) |
4700 | goto out_free; | |
fddaface | 4701 | |
7a7cacba PB |
4702 | msg.msg_name = NULL; |
4703 | msg.msg_control = NULL; | |
4704 | msg.msg_controllen = 0; | |
4705 | msg.msg_namelen = 0; | |
4706 | msg.msg_iocb = NULL; | |
4707 | msg.msg_flags = 0; | |
fddaface | 4708 | |
76cd979f | 4709 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4710 | if (flags & MSG_DONTWAIT) |
4711 | req->flags |= REQ_F_NOWAIT; | |
4712 | else if (force_nonblock) | |
4713 | flags |= MSG_DONTWAIT; | |
4714 | ||
0031275d SM |
4715 | if (flags & MSG_WAITALL) |
4716 | min_ret = iov_iter_count(&msg.msg_iter); | |
4717 | ||
7a7cacba PB |
4718 | ret = sock_recvmsg(sock, &msg, flags); |
4719 | if (force_nonblock && ret == -EAGAIN) | |
4720 | return -EAGAIN; | |
4721 | if (ret == -ERESTARTSYS) | |
4722 | ret = -EINTR; | |
14c32eee | 4723 | out_free: |
7fbb1b54 PB |
4724 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4725 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4726 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4727 | req_set_fail_links(req); |
889fca73 | 4728 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4729 | return 0; |
fddaface JA |
4730 | } |
4731 | ||
3529d8c2 | 4732 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4733 | { |
8ed8d3c3 JA |
4734 | struct io_accept *accept = &req->accept; |
4735 | ||
14587a46 | 4736 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4737 | return -EINVAL; |
8042d6ce | 4738 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4739 | return -EINVAL; |
4740 | ||
d55e5f5b JA |
4741 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4742 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4743 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4744 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4745 | return 0; |
8ed8d3c3 | 4746 | } |
17f2fe35 | 4747 | |
889fca73 | 4748 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4749 | { |
4750 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4751 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4752 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4753 | int ret; |
4754 | ||
e697deed JX |
4755 | if (req->file->f_flags & O_NONBLOCK) |
4756 | req->flags |= REQ_F_NOWAIT; | |
4757 | ||
8ed8d3c3 | 4758 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4759 | accept->addr_len, accept->flags, |
4760 | accept->nofile); | |
8ed8d3c3 | 4761 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4762 | return -EAGAIN; |
ac45abc0 PB |
4763 | if (ret < 0) { |
4764 | if (ret == -ERESTARTSYS) | |
4765 | ret = -EINTR; | |
4e88d6e7 | 4766 | req_set_fail_links(req); |
ac45abc0 | 4767 | } |
889fca73 | 4768 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4769 | return 0; |
8ed8d3c3 JA |
4770 | } |
4771 | ||
93642ef8 PB |
4772 | static int io_connect_prep_async(struct io_kiocb *req) |
4773 | { | |
4774 | struct io_async_connect *io = req->async_data; | |
4775 | struct io_connect *conn = &req->connect; | |
4776 | ||
4777 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4778 | } | |
4779 | ||
3529d8c2 | 4780 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4781 | { |
3529d8c2 | 4782 | struct io_connect *conn = &req->connect; |
f499a021 | 4783 | |
14587a46 | 4784 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4785 | return -EINVAL; |
4786 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4787 | return -EINVAL; | |
4788 | ||
3529d8c2 JA |
4789 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4790 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4791 | return 0; |
f499a021 JA |
4792 | } |
4793 | ||
889fca73 | 4794 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4795 | { |
e8c2bc1f | 4796 | struct io_async_connect __io, *io; |
f8e85cf2 | 4797 | unsigned file_flags; |
3fbb51c1 | 4798 | int ret; |
45d189c6 | 4799 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4800 | |
e8c2bc1f JA |
4801 | if (req->async_data) { |
4802 | io = req->async_data; | |
f499a021 | 4803 | } else { |
3529d8c2 JA |
4804 | ret = move_addr_to_kernel(req->connect.addr, |
4805 | req->connect.addr_len, | |
e8c2bc1f | 4806 | &__io.address); |
f499a021 JA |
4807 | if (ret) |
4808 | goto out; | |
4809 | io = &__io; | |
4810 | } | |
4811 | ||
3fbb51c1 JA |
4812 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4813 | ||
e8c2bc1f | 4814 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4815 | req->connect.addr_len, file_flags); |
87f80d62 | 4816 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4817 | if (req->async_data) |
b7bb4f7d | 4818 | return -EAGAIN; |
e8c2bc1f | 4819 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4820 | ret = -ENOMEM; |
4821 | goto out; | |
4822 | } | |
e8c2bc1f | 4823 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4824 | return -EAGAIN; |
f499a021 | 4825 | } |
f8e85cf2 JA |
4826 | if (ret == -ERESTARTSYS) |
4827 | ret = -EINTR; | |
f499a021 | 4828 | out: |
4e88d6e7 JA |
4829 | if (ret < 0) |
4830 | req_set_fail_links(req); | |
889fca73 | 4831 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4832 | return 0; |
469956e8 Y |
4833 | } |
4834 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4835 | #define IO_NETOP_FN(op) \ |
4836 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4837 | { \ | |
4838 | return -EOPNOTSUPP; \ | |
4839 | } | |
4840 | ||
4841 | #define IO_NETOP_PREP(op) \ | |
4842 | IO_NETOP_FN(op) \ | |
4843 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4844 | { \ | |
4845 | return -EOPNOTSUPP; \ | |
4846 | } \ | |
4847 | ||
4848 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4849 | IO_NETOP_PREP(op) \ | |
4850 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4851 | { \ | |
4852 | return -EOPNOTSUPP; \ | |
4853 | } | |
4854 | ||
4855 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4856 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4857 | IO_NETOP_PREP_ASYNC(connect); | |
4858 | IO_NETOP_PREP(accept); | |
4859 | IO_NETOP_FN(send); | |
4860 | IO_NETOP_FN(recv); | |
469956e8 | 4861 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4862 | |
d7718a9d JA |
4863 | struct io_poll_table { |
4864 | struct poll_table_struct pt; | |
4865 | struct io_kiocb *req; | |
4866 | int error; | |
4867 | }; | |
ce593a6c | 4868 | |
d7718a9d JA |
4869 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4870 | __poll_t mask, task_work_func_t func) | |
4871 | { | |
aa96bf8a | 4872 | int ret; |
d7718a9d JA |
4873 | |
4874 | /* for instances that support it check for an event match first: */ | |
4875 | if (mask && !(mask & poll->events)) | |
4876 | return 0; | |
4877 | ||
4878 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4879 | ||
4880 | list_del_init(&poll->wait.entry); | |
4881 | ||
d7718a9d | 4882 | req->result = mask; |
7cbf1722 | 4883 | req->task_work.func = func; |
6d816e08 | 4884 | |
d7718a9d | 4885 | /* |
e3aabf95 JA |
4886 | * If this fails, then the task is exiting. When a task exits, the |
4887 | * work gets canceled, so just cancel this request as well instead | |
4888 | * of executing it. We can't safely execute it anyway, as we may not | |
4889 | * have the needed state needed for it anyway. | |
d7718a9d | 4890 | */ |
355fb9e2 | 4891 | ret = io_req_task_work_add(req); |
aa96bf8a | 4892 | if (unlikely(ret)) { |
e3aabf95 | 4893 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4894 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4895 | } |
d7718a9d JA |
4896 | return 1; |
4897 | } | |
4898 | ||
74ce6ce4 JA |
4899 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4900 | __acquires(&req->ctx->completion_lock) | |
4901 | { | |
4902 | struct io_ring_ctx *ctx = req->ctx; | |
4903 | ||
4904 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4905 | struct poll_table_struct pt = { ._key = poll->events }; | |
4906 | ||
4907 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4908 | } | |
4909 | ||
4910 | spin_lock_irq(&ctx->completion_lock); | |
4911 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4912 | add_wait_queue(poll->head, &poll->wait); | |
4913 | return true; | |
4914 | } | |
4915 | ||
4916 | return false; | |
4917 | } | |
4918 | ||
d4e7cd36 | 4919 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4920 | { |
e8c2bc1f | 4921 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4922 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4923 | return req->async_data; |
d4e7cd36 JA |
4924 | return req->apoll->double_poll; |
4925 | } | |
4926 | ||
4927 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4928 | { | |
4929 | if (req->opcode == IORING_OP_POLL_ADD) | |
4930 | return &req->poll; | |
4931 | return &req->apoll->poll; | |
4932 | } | |
4933 | ||
4934 | static void io_poll_remove_double(struct io_kiocb *req) | |
4935 | { | |
4936 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4937 | |
4938 | lockdep_assert_held(&req->ctx->completion_lock); | |
4939 | ||
4940 | if (poll && poll->head) { | |
4941 | struct wait_queue_head *head = poll->head; | |
4942 | ||
4943 | spin_lock(&head->lock); | |
4944 | list_del_init(&poll->wait.entry); | |
4945 | if (poll->wait.private) | |
de9b4cca | 4946 | req_ref_put(req); |
18bceab1 JA |
4947 | poll->head = NULL; |
4948 | spin_unlock(&head->lock); | |
4949 | } | |
4950 | } | |
4951 | ||
4952 | static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) | |
4953 | { | |
4954 | struct io_ring_ctx *ctx = req->ctx; | |
4955 | ||
45ab03b1 JA |
4956 | if (!error && req->poll.canceled) |
4957 | error = -ECANCELED; | |
4958 | ||
d4e7cd36 | 4959 | io_poll_remove_double(req); |
18bceab1 JA |
4960 | req->poll.done = true; |
4961 | io_cqring_fill_event(req, error ? error : mangle_poll(mask)); | |
4962 | io_commit_cqring(ctx); | |
4963 | } | |
4964 | ||
dd221f46 | 4965 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4966 | { |
dd221f46 | 4967 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4968 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4969 | struct io_kiocb *nxt; |
18bceab1 JA |
4970 | |
4971 | if (io_poll_rewait(req, &req->poll)) { | |
4972 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 PB |
4973 | } else { |
4974 | hash_del(&req->hash_node); | |
4975 | io_poll_complete(req, req->result, 0); | |
4976 | spin_unlock_irq(&ctx->completion_lock); | |
18bceab1 | 4977 | |
dd221f46 PB |
4978 | nxt = io_put_req_find_next(req); |
4979 | io_cqring_ev_posted(ctx); | |
4980 | if (nxt) | |
4981 | __io_req_task_submit(nxt); | |
4982 | } | |
18bceab1 JA |
4983 | } |
4984 | ||
4985 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4986 | int sync, void *key) | |
4987 | { | |
4988 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4989 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4990 | __poll_t mask = key_to_poll(key); |
4991 | ||
4992 | /* for instances that support it check for an event match first: */ | |
4993 | if (mask && !(mask & poll->events)) | |
4994 | return 0; | |
4995 | ||
8706e04e JA |
4996 | list_del_init(&wait->entry); |
4997 | ||
807abcb0 | 4998 | if (poll && poll->head) { |
18bceab1 JA |
4999 | bool done; |
5000 | ||
807abcb0 JA |
5001 | spin_lock(&poll->head->lock); |
5002 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 5003 | if (!done) |
807abcb0 | 5004 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
5005 | /* make sure double remove sees this as being gone */ |
5006 | wait->private = NULL; | |
807abcb0 | 5007 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
5008 | if (!done) { |
5009 | /* use wait func handler, so it matches the rq type */ | |
5010 | poll->wait.func(&poll->wait, mode, sync, key); | |
5011 | } | |
18bceab1 | 5012 | } |
de9b4cca | 5013 | req_ref_put(req); |
18bceab1 JA |
5014 | return 1; |
5015 | } | |
5016 | ||
5017 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
5018 | wait_queue_func_t wake_func) | |
5019 | { | |
5020 | poll->head = NULL; | |
5021 | poll->done = false; | |
5022 | poll->canceled = false; | |
5023 | poll->events = events; | |
5024 | INIT_LIST_HEAD(&poll->wait.entry); | |
5025 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5026 | } | |
5027 | ||
5028 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5029 | struct wait_queue_head *head, |
5030 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5031 | { |
5032 | struct io_kiocb *req = pt->req; | |
5033 | ||
5034 | /* | |
5035 | * If poll->head is already set, it's because the file being polled | |
5036 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5037 | * for write). Setup a separate io_poll_iocb if this happens. | |
5038 | */ | |
5039 | if (unlikely(poll->head)) { | |
58852d4d PB |
5040 | struct io_poll_iocb *poll_one = poll; |
5041 | ||
18bceab1 | 5042 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5043 | if (*poll_ptr) { |
18bceab1 JA |
5044 | pt->error = -EINVAL; |
5045 | return; | |
5046 | } | |
1c3b3e65 JA |
5047 | /* double add on the same waitqueue head, ignore */ |
5048 | if (poll->head == head) | |
5049 | return; | |
18bceab1 JA |
5050 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5051 | if (!poll) { | |
5052 | pt->error = -ENOMEM; | |
5053 | return; | |
5054 | } | |
58852d4d | 5055 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5056 | req_ref_get(req); |
18bceab1 | 5057 | poll->wait.private = req; |
807abcb0 | 5058 | *poll_ptr = poll; |
18bceab1 JA |
5059 | } |
5060 | ||
5061 | pt->error = 0; | |
5062 | poll->head = head; | |
a31eb4a2 JX |
5063 | |
5064 | if (poll->events & EPOLLEXCLUSIVE) | |
5065 | add_wait_queue_exclusive(head, &poll->wait); | |
5066 | else | |
5067 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5068 | } |
5069 | ||
5070 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5071 | struct poll_table_struct *p) | |
5072 | { | |
5073 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5074 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5075 | |
807abcb0 | 5076 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5077 | } |
5078 | ||
d7718a9d JA |
5079 | static void io_async_task_func(struct callback_head *cb) |
5080 | { | |
5081 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5082 | struct async_poll *apoll = req->apoll; | |
5083 | struct io_ring_ctx *ctx = req->ctx; | |
5084 | ||
5085 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5086 | ||
74ce6ce4 | 5087 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5088 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5089 | return; |
d7718a9d JA |
5090 | } |
5091 | ||
31067255 | 5092 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5093 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5094 | hash_del(&req->hash_node); |
2bae047e | 5095 | |
d4e7cd36 | 5096 | io_poll_remove_double(req); |
74ce6ce4 JA |
5097 | spin_unlock_irq(&ctx->completion_lock); |
5098 | ||
0be0b0e3 PB |
5099 | if (!READ_ONCE(apoll->poll.canceled)) |
5100 | __io_req_task_submit(req); | |
5101 | else | |
5102 | __io_req_task_cancel(req, -ECANCELED); | |
aa340845 | 5103 | |
807abcb0 | 5104 | kfree(apoll->double_poll); |
31067255 | 5105 | kfree(apoll); |
d7718a9d JA |
5106 | } |
5107 | ||
5108 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5109 | void *key) | |
5110 | { | |
5111 | struct io_kiocb *req = wait->private; | |
5112 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5113 | ||
5114 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5115 | key_to_poll(key)); | |
5116 | ||
5117 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5118 | } | |
5119 | ||
5120 | static void io_poll_req_insert(struct io_kiocb *req) | |
5121 | { | |
5122 | struct io_ring_ctx *ctx = req->ctx; | |
5123 | struct hlist_head *list; | |
5124 | ||
5125 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5126 | hlist_add_head(&req->hash_node, list); | |
5127 | } | |
5128 | ||
5129 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5130 | struct io_poll_iocb *poll, | |
5131 | struct io_poll_table *ipt, __poll_t mask, | |
5132 | wait_queue_func_t wake_func) | |
5133 | __acquires(&ctx->completion_lock) | |
5134 | { | |
5135 | struct io_ring_ctx *ctx = req->ctx; | |
5136 | bool cancel = false; | |
5137 | ||
4d52f338 | 5138 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5139 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5140 | poll->file = req->file; |
18bceab1 | 5141 | poll->wait.private = req; |
d7718a9d JA |
5142 | |
5143 | ipt->pt._key = mask; | |
5144 | ipt->req = req; | |
5145 | ipt->error = -EINVAL; | |
5146 | ||
d7718a9d JA |
5147 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5148 | ||
5149 | spin_lock_irq(&ctx->completion_lock); | |
5150 | if (likely(poll->head)) { | |
5151 | spin_lock(&poll->head->lock); | |
5152 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5153 | if (ipt->error) | |
5154 | cancel = true; | |
5155 | ipt->error = 0; | |
5156 | mask = 0; | |
5157 | } | |
5158 | if (mask || ipt->error) | |
5159 | list_del_init(&poll->wait.entry); | |
5160 | else if (cancel) | |
5161 | WRITE_ONCE(poll->canceled, true); | |
5162 | else if (!poll->done) /* actually waiting for an event */ | |
5163 | io_poll_req_insert(req); | |
5164 | spin_unlock(&poll->head->lock); | |
5165 | } | |
5166 | ||
5167 | return mask; | |
5168 | } | |
5169 | ||
5170 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5171 | { | |
5172 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5173 | struct io_ring_ctx *ctx = req->ctx; | |
5174 | struct async_poll *apoll; | |
5175 | struct io_poll_table ipt; | |
5176 | __poll_t mask, ret; | |
9dab14b8 | 5177 | int rw; |
d7718a9d JA |
5178 | |
5179 | if (!req->file || !file_can_poll(req->file)) | |
5180 | return false; | |
24c74678 | 5181 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5182 | return false; |
9dab14b8 JA |
5183 | if (def->pollin) |
5184 | rw = READ; | |
5185 | else if (def->pollout) | |
5186 | rw = WRITE; | |
5187 | else | |
5188 | return false; | |
5189 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5190 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5191 | return false; |
5192 | ||
5193 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5194 | if (unlikely(!apoll)) | |
5195 | return false; | |
807abcb0 | 5196 | apoll->double_poll = NULL; |
d7718a9d JA |
5197 | |
5198 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5199 | req->apoll = apoll; |
d7718a9d | 5200 | |
8755d97a | 5201 | mask = 0; |
d7718a9d | 5202 | if (def->pollin) |
8755d97a | 5203 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5204 | if (def->pollout) |
5205 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5206 | |
5207 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5208 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5209 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5210 | mask &= ~POLLIN; | |
5211 | ||
d7718a9d JA |
5212 | mask |= POLLERR | POLLPRI; |
5213 | ||
5214 | ipt.pt._qproc = io_async_queue_proc; | |
5215 | ||
5216 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5217 | io_async_wake); | |
a36da65c | 5218 | if (ret || ipt.error) { |
d4e7cd36 | 5219 | io_poll_remove_double(req); |
d7718a9d | 5220 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5221 | kfree(apoll->double_poll); |
d7718a9d JA |
5222 | kfree(apoll); |
5223 | return false; | |
5224 | } | |
5225 | spin_unlock_irq(&ctx->completion_lock); | |
5226 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5227 | apoll->poll.events); | |
5228 | return true; | |
5229 | } | |
5230 | ||
5231 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
5232 | struct io_poll_iocb *poll) | |
221c5eb2 | 5233 | { |
b41e9852 | 5234 | bool do_complete = false; |
221c5eb2 JA |
5235 | |
5236 | spin_lock(&poll->head->lock); | |
5237 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5238 | if (!list_empty(&poll->wait.entry)) { |
5239 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5240 | do_complete = true; |
221c5eb2 JA |
5241 | } |
5242 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5243 | hash_del(&req->hash_node); |
d7718a9d JA |
5244 | return do_complete; |
5245 | } | |
5246 | ||
5247 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5248 | { | |
5249 | bool do_complete; | |
5250 | ||
d4e7cd36 JA |
5251 | io_poll_remove_double(req); |
5252 | ||
d7718a9d JA |
5253 | if (req->opcode == IORING_OP_POLL_ADD) { |
5254 | do_complete = __io_poll_remove_one(req, &req->poll); | |
5255 | } else { | |
3bfa5bcb JA |
5256 | struct async_poll *apoll = req->apoll; |
5257 | ||
d7718a9d | 5258 | /* non-poll requests have submit ref still */ |
3bfa5bcb JA |
5259 | do_complete = __io_poll_remove_one(req, &apoll->poll); |
5260 | if (do_complete) { | |
d7718a9d | 5261 | io_put_req(req); |
807abcb0 | 5262 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5263 | kfree(apoll); |
5264 | } | |
b1f573bd XW |
5265 | } |
5266 | ||
b41e9852 JA |
5267 | if (do_complete) { |
5268 | io_cqring_fill_event(req, -ECANCELED); | |
5269 | io_commit_cqring(req->ctx); | |
f254ac04 | 5270 | req_set_fail_links(req); |
216578e5 | 5271 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5272 | } |
5273 | ||
5274 | return do_complete; | |
221c5eb2 JA |
5275 | } |
5276 | ||
76e1b642 JA |
5277 | /* |
5278 | * Returns true if we found and killed one or more poll requests | |
5279 | */ | |
6b81928d PB |
5280 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5281 | struct files_struct *files) | |
221c5eb2 | 5282 | { |
78076bb6 | 5283 | struct hlist_node *tmp; |
221c5eb2 | 5284 | struct io_kiocb *req; |
8e2e1faf | 5285 | int posted = 0, i; |
221c5eb2 JA |
5286 | |
5287 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5288 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5289 | struct hlist_head *list; | |
5290 | ||
5291 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5292 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5293 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5294 | posted += io_poll_remove_one(req); |
5295 | } | |
221c5eb2 JA |
5296 | } |
5297 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5298 | |
8e2e1faf JA |
5299 | if (posted) |
5300 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5301 | |
5302 | return posted != 0; | |
221c5eb2 JA |
5303 | } |
5304 | ||
47f46768 JA |
5305 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) |
5306 | { | |
78076bb6 | 5307 | struct hlist_head *list; |
47f46768 JA |
5308 | struct io_kiocb *req; |
5309 | ||
78076bb6 JA |
5310 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5311 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5312 | if (sqe_addr != req->user_data) |
5313 | continue; | |
5314 | if (io_poll_remove_one(req)) | |
eac406c6 | 5315 | return 0; |
b41e9852 | 5316 | return -EALREADY; |
47f46768 JA |
5317 | } |
5318 | ||
5319 | return -ENOENT; | |
5320 | } | |
5321 | ||
3529d8c2 JA |
5322 | static int io_poll_remove_prep(struct io_kiocb *req, |
5323 | const struct io_uring_sqe *sqe) | |
0969e783 | 5324 | { |
0969e783 JA |
5325 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5326 | return -EINVAL; | |
5327 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5328 | sqe->poll_events) | |
5329 | return -EINVAL; | |
5330 | ||
018043be | 5331 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5332 | return 0; |
5333 | } | |
5334 | ||
221c5eb2 JA |
5335 | /* |
5336 | * Find a running poll command that matches one specified in sqe->addr, | |
5337 | * and remove it if found. | |
5338 | */ | |
61e98203 | 5339 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5340 | { |
5341 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5342 | int ret; |
221c5eb2 | 5343 | |
221c5eb2 | 5344 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5345 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5346 | spin_unlock_irq(&ctx->completion_lock); |
5347 | ||
4e88d6e7 JA |
5348 | if (ret < 0) |
5349 | req_set_fail_links(req); | |
e1e16097 | 5350 | io_req_complete(req, ret); |
221c5eb2 JA |
5351 | return 0; |
5352 | } | |
5353 | ||
221c5eb2 JA |
5354 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5355 | void *key) | |
5356 | { | |
c2f2eb7d JA |
5357 | struct io_kiocb *req = wait->private; |
5358 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5359 | |
d7718a9d | 5360 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5361 | } |
5362 | ||
221c5eb2 JA |
5363 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5364 | struct poll_table_struct *p) | |
5365 | { | |
5366 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5367 | ||
e8c2bc1f | 5368 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5369 | } |
5370 | ||
3529d8c2 | 5371 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5372 | { |
5373 | struct io_poll_iocb *poll = &req->poll; | |
5769a351 | 5374 | u32 events; |
221c5eb2 JA |
5375 | |
5376 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5377 | return -EINVAL; | |
5378 | if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index) | |
5379 | return -EINVAL; | |
5380 | ||
5769a351 JX |
5381 | events = READ_ONCE(sqe->poll32_events); |
5382 | #ifdef __BIG_ENDIAN | |
5383 | events = swahw32(events); | |
5384 | #endif | |
a31eb4a2 JX |
5385 | poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP | |
5386 | (events & EPOLLEXCLUSIVE); | |
0969e783 JA |
5387 | return 0; |
5388 | } | |
5389 | ||
61e98203 | 5390 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5391 | { |
5392 | struct io_poll_iocb *poll = &req->poll; | |
5393 | struct io_ring_ctx *ctx = req->ctx; | |
5394 | struct io_poll_table ipt; | |
0969e783 | 5395 | __poll_t mask; |
0969e783 | 5396 | |
d7718a9d | 5397 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5398 | |
d7718a9d JA |
5399 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5400 | io_poll_wake); | |
221c5eb2 | 5401 | |
8c838788 | 5402 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5403 | ipt.error = 0; |
b0dd8a41 | 5404 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5405 | } |
221c5eb2 JA |
5406 | spin_unlock_irq(&ctx->completion_lock); |
5407 | ||
8c838788 JA |
5408 | if (mask) { |
5409 | io_cqring_ev_posted(ctx); | |
014db007 | 5410 | io_put_req(req); |
221c5eb2 | 5411 | } |
8c838788 | 5412 | return ipt.error; |
221c5eb2 JA |
5413 | } |
5414 | ||
5262f567 JA |
5415 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5416 | { | |
ad8a48ac JA |
5417 | struct io_timeout_data *data = container_of(timer, |
5418 | struct io_timeout_data, timer); | |
5419 | struct io_kiocb *req = data->req; | |
5420 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5421 | unsigned long flags; |
5422 | ||
5262f567 | 5423 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5424 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5425 | atomic_set(&req->ctx->cq_timeouts, |
5426 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5427 | ||
78e19bbe | 5428 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5429 | io_commit_cqring(ctx); |
5430 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5431 | ||
5432 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5433 | req_set_fail_links(req); |
5262f567 JA |
5434 | io_put_req(req); |
5435 | return HRTIMER_NORESTART; | |
5436 | } | |
5437 | ||
fbd15848 PB |
5438 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5439 | __u64 user_data) | |
f254ac04 | 5440 | { |
fbd15848 | 5441 | struct io_timeout_data *io; |
47f46768 JA |
5442 | struct io_kiocb *req; |
5443 | int ret = -ENOENT; | |
f254ac04 | 5444 | |
135fcde8 | 5445 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5446 | if (user_data == req->user_data) { |
47f46768 JA |
5447 | ret = 0; |
5448 | break; | |
5449 | } | |
5450 | } | |
5451 | ||
5452 | if (ret == -ENOENT) | |
fbd15848 PB |
5453 | return ERR_PTR(ret); |
5454 | ||
5455 | io = req->async_data; | |
e8c2bc1f | 5456 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5457 | if (ret == -1) |
fbd15848 | 5458 | return ERR_PTR(-EALREADY); |
a71976f3 | 5459 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5460 | return req; |
5461 | } | |
47f46768 | 5462 | |
fbd15848 PB |
5463 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5464 | { | |
5465 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5466 | ||
5467 | if (IS_ERR(req)) | |
5468 | return PTR_ERR(req); | |
f254ac04 JA |
5469 | |
5470 | req_set_fail_links(req); | |
f254ac04 | 5471 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5472 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5473 | return 0; |
5474 | } | |
5475 | ||
9c8e11b3 PB |
5476 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5477 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5478 | { |
9c8e11b3 PB |
5479 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5480 | struct io_timeout_data *data; | |
47f46768 | 5481 | |
9c8e11b3 PB |
5482 | if (IS_ERR(req)) |
5483 | return PTR_ERR(req); | |
47f46768 | 5484 | |
9c8e11b3 PB |
5485 | req->timeout.off = 0; /* noseq */ |
5486 | data = req->async_data; | |
5487 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5488 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5489 | data->timer.function = io_timeout_fn; | |
5490 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5491 | return 0; | |
47f46768 JA |
5492 | } |
5493 | ||
3529d8c2 JA |
5494 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5495 | const struct io_uring_sqe *sqe) | |
b29472ee | 5496 | { |
9c8e11b3 PB |
5497 | struct io_timeout_rem *tr = &req->timeout_rem; |
5498 | ||
b29472ee JA |
5499 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5500 | return -EINVAL; | |
61710e43 DA |
5501 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5502 | return -EINVAL; | |
9c8e11b3 | 5503 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5504 | return -EINVAL; |
5505 | ||
9c8e11b3 PB |
5506 | tr->addr = READ_ONCE(sqe->addr); |
5507 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5508 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5509 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5510 | return -EINVAL; | |
5511 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5512 | return -EFAULT; | |
5513 | } else if (tr->flags) { | |
5514 | /* timeout removal doesn't support flags */ | |
b29472ee | 5515 | return -EINVAL; |
9c8e11b3 | 5516 | } |
b29472ee | 5517 | |
b29472ee JA |
5518 | return 0; |
5519 | } | |
5520 | ||
8662daec PB |
5521 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5522 | { | |
5523 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5524 | : HRTIMER_MODE_REL; | |
5525 | } | |
5526 | ||
11365043 JA |
5527 | /* |
5528 | * Remove or update an existing timeout command | |
5529 | */ | |
61e98203 | 5530 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5531 | { |
9c8e11b3 | 5532 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5533 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5534 | int ret; |
11365043 | 5535 | |
11365043 | 5536 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5537 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5538 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5539 | else |
5540 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5541 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5542 | |
47f46768 | 5543 | io_cqring_fill_event(req, ret); |
11365043 JA |
5544 | io_commit_cqring(ctx); |
5545 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5546 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5547 | if (ret < 0) |
5548 | req_set_fail_links(req); | |
ec9c02ad | 5549 | io_put_req(req); |
11365043 | 5550 | return 0; |
5262f567 JA |
5551 | } |
5552 | ||
3529d8c2 | 5553 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5554 | bool is_timeout_link) |
5262f567 | 5555 | { |
ad8a48ac | 5556 | struct io_timeout_data *data; |
a41525ab | 5557 | unsigned flags; |
56080b02 | 5558 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5559 | |
ad8a48ac | 5560 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5561 | return -EINVAL; |
ad8a48ac | 5562 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5563 | return -EINVAL; |
56080b02 | 5564 | if (off && is_timeout_link) |
2d28390a | 5565 | return -EINVAL; |
a41525ab JA |
5566 | flags = READ_ONCE(sqe->timeout_flags); |
5567 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5568 | return -EINVAL; |
bdf20073 | 5569 | |
bfe68a22 | 5570 | req->timeout.off = off; |
26a61679 | 5571 | |
e8c2bc1f | 5572 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5573 | return -ENOMEM; |
5574 | ||
e8c2bc1f | 5575 | data = req->async_data; |
ad8a48ac | 5576 | data->req = req; |
ad8a48ac JA |
5577 | |
5578 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5579 | return -EFAULT; |
5580 | ||
8662daec | 5581 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5582 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5583 | if (is_timeout_link) |
5584 | io_req_track_inflight(req); | |
ad8a48ac JA |
5585 | return 0; |
5586 | } | |
5587 | ||
61e98203 | 5588 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5589 | { |
ad8a48ac | 5590 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5591 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5592 | struct list_head *entry; |
bfe68a22 | 5593 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5594 | |
733f5c95 | 5595 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5596 | |
5262f567 JA |
5597 | /* |
5598 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5599 | * timeout event to be satisfied. If it isn't set, then this is |
5600 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5601 | */ |
8eb7e2d0 | 5602 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5603 | entry = ctx->timeout_list.prev; |
5604 | goto add; | |
5605 | } | |
5262f567 | 5606 | |
bfe68a22 PB |
5607 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5608 | req->timeout.target_seq = tail + off; | |
5262f567 | 5609 | |
f010505b MDG |
5610 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5611 | * This is safe because ->completion_lock is held, and submissions | |
5612 | * and completions are never mixed in the same ->completion_lock section. | |
5613 | */ | |
5614 | ctx->cq_last_tm_flush = tail; | |
5615 | ||
5262f567 JA |
5616 | /* |
5617 | * Insertion sort, ensuring the first entry in the list is always | |
5618 | * the one we need first. | |
5619 | */ | |
5262f567 | 5620 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5621 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5622 | timeout.list); | |
5262f567 | 5623 | |
8eb7e2d0 | 5624 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5625 | continue; |
bfe68a22 PB |
5626 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5627 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5628 | break; |
5629 | } | |
93bd25bb | 5630 | add: |
135fcde8 | 5631 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5632 | data->timer.function = io_timeout_fn; |
5633 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5634 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5635 | return 0; |
5636 | } | |
5262f567 | 5637 | |
f458dd84 PB |
5638 | struct io_cancel_data { |
5639 | struct io_ring_ctx *ctx; | |
5640 | u64 user_data; | |
5641 | }; | |
5642 | ||
62755e35 JA |
5643 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5644 | { | |
5645 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5646 | struct io_cancel_data *cd = data; |
62755e35 | 5647 | |
f458dd84 | 5648 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5649 | } |
5650 | ||
f458dd84 PB |
5651 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5652 | struct io_ring_ctx *ctx) | |
62755e35 | 5653 | { |
f458dd84 | 5654 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5655 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5656 | int ret = 0; |
5657 | ||
f458dd84 | 5658 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5659 | return -ENOENT; |
5660 | ||
f458dd84 | 5661 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5662 | switch (cancel_ret) { |
5663 | case IO_WQ_CANCEL_OK: | |
5664 | ret = 0; | |
5665 | break; | |
5666 | case IO_WQ_CANCEL_RUNNING: | |
5667 | ret = -EALREADY; | |
5668 | break; | |
5669 | case IO_WQ_CANCEL_NOTFOUND: | |
5670 | ret = -ENOENT; | |
5671 | break; | |
5672 | } | |
5673 | ||
e977d6d3 JA |
5674 | return ret; |
5675 | } | |
5676 | ||
47f46768 JA |
5677 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5678 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5679 | int success_ret) |
47f46768 JA |
5680 | { |
5681 | unsigned long flags; | |
5682 | int ret; | |
5683 | ||
f458dd84 | 5684 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5685 | if (ret != -ENOENT) { |
5686 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5687 | goto done; | |
5688 | } | |
5689 | ||
5690 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5691 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5692 | if (ret != -ENOENT) | |
5693 | goto done; | |
5694 | ret = io_poll_cancel(ctx, sqe_addr); | |
5695 | done: | |
b0dd8a41 JA |
5696 | if (!ret) |
5697 | ret = success_ret; | |
47f46768 JA |
5698 | io_cqring_fill_event(req, ret); |
5699 | io_commit_cqring(ctx); | |
5700 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5701 | io_cqring_ev_posted(ctx); | |
5702 | ||
4e88d6e7 JA |
5703 | if (ret < 0) |
5704 | req_set_fail_links(req); | |
014db007 | 5705 | io_put_req(req); |
47f46768 JA |
5706 | } |
5707 | ||
3529d8c2 JA |
5708 | static int io_async_cancel_prep(struct io_kiocb *req, |
5709 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5710 | { |
fbf23849 | 5711 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5712 | return -EINVAL; |
61710e43 DA |
5713 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5714 | return -EINVAL; | |
5715 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5716 | return -EINVAL; |
5717 | ||
fbf23849 JA |
5718 | req->cancel.addr = READ_ONCE(sqe->addr); |
5719 | return 0; | |
5720 | } | |
5721 | ||
61e98203 | 5722 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5723 | { |
5724 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5725 | u64 sqe_addr = req->cancel.addr; |
5726 | struct io_tctx_node *node; | |
5727 | int ret; | |
5728 | ||
5729 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5730 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5731 | spin_lock_irq(&ctx->completion_lock); | |
5732 | if (ret != -ENOENT) | |
5733 | goto done; | |
5734 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5735 | if (ret != -ENOENT) | |
5736 | goto done; | |
5737 | ret = io_poll_cancel(ctx, sqe_addr); | |
5738 | if (ret != -ENOENT) | |
5739 | goto done; | |
5740 | spin_unlock_irq(&ctx->completion_lock); | |
5741 | ||
5742 | /* slow path, try all io-wq's */ | |
5743 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5744 | ret = -ENOENT; | |
5745 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5746 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5747 | |
58f99373 PB |
5748 | if (!tctx || !tctx->io_wq) |
5749 | continue; | |
5750 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5751 | if (ret != -ENOENT) | |
5752 | break; | |
5753 | } | |
5754 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5755 | ||
5756 | spin_lock_irq(&ctx->completion_lock); | |
5757 | done: | |
5758 | io_cqring_fill_event(req, ret); | |
5759 | io_commit_cqring(ctx); | |
5760 | spin_unlock_irq(&ctx->completion_lock); | |
5761 | io_cqring_ev_posted(ctx); | |
5762 | ||
5763 | if (ret < 0) | |
5764 | req_set_fail_links(req); | |
5765 | io_put_req(req); | |
5262f567 JA |
5766 | return 0; |
5767 | } | |
5768 | ||
269bbe5f | 5769 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5770 | const struct io_uring_sqe *sqe) |
5771 | { | |
6ca56f84 JA |
5772 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5773 | return -EINVAL; | |
61710e43 DA |
5774 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5775 | return -EINVAL; | |
5776 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5777 | return -EINVAL; |
5778 | ||
269bbe5f BM |
5779 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5780 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5781 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5782 | return -EINVAL; |
269bbe5f | 5783 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5784 | return 0; |
5785 | } | |
5786 | ||
889fca73 | 5787 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5788 | { |
5789 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5790 | struct io_uring_rsrc_update up; |
05f3fb3c | 5791 | int ret; |
fbf23849 | 5792 | |
45d189c6 | 5793 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5794 | return -EAGAIN; |
05f3fb3c | 5795 | |
269bbe5f BM |
5796 | up.offset = req->rsrc_update.offset; |
5797 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5798 | |
5799 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5800 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5801 | mutex_unlock(&ctx->uring_lock); |
5802 | ||
5803 | if (ret < 0) | |
5804 | req_set_fail_links(req); | |
889fca73 | 5805 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5806 | return 0; |
5807 | } | |
5808 | ||
bfe76559 | 5809 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5810 | { |
d625c6ee | 5811 | switch (req->opcode) { |
e781573e | 5812 | case IORING_OP_NOP: |
bfe76559 | 5813 | return 0; |
f67676d1 JA |
5814 | case IORING_OP_READV: |
5815 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5816 | case IORING_OP_READ: |
bfe76559 | 5817 | return io_read_prep(req, sqe); |
f67676d1 JA |
5818 | case IORING_OP_WRITEV: |
5819 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5820 | case IORING_OP_WRITE: |
bfe76559 | 5821 | return io_write_prep(req, sqe); |
0969e783 | 5822 | case IORING_OP_POLL_ADD: |
bfe76559 | 5823 | return io_poll_add_prep(req, sqe); |
0969e783 | 5824 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5825 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5826 | case IORING_OP_FSYNC: |
1155c76a | 5827 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5828 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5829 | return io_sfr_prep(req, sqe); |
03b1230c | 5830 | case IORING_OP_SENDMSG: |
fddaface | 5831 | case IORING_OP_SEND: |
bfe76559 | 5832 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5833 | case IORING_OP_RECVMSG: |
fddaface | 5834 | case IORING_OP_RECV: |
bfe76559 | 5835 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5836 | case IORING_OP_CONNECT: |
bfe76559 | 5837 | return io_connect_prep(req, sqe); |
2d28390a | 5838 | case IORING_OP_TIMEOUT: |
bfe76559 | 5839 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5840 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5841 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5842 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5843 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5844 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5845 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5846 | case IORING_OP_ACCEPT: |
bfe76559 | 5847 | return io_accept_prep(req, sqe); |
d63d1b5e | 5848 | case IORING_OP_FALLOCATE: |
bfe76559 | 5849 | return io_fallocate_prep(req, sqe); |
15b71abe | 5850 | case IORING_OP_OPENAT: |
bfe76559 | 5851 | return io_openat_prep(req, sqe); |
b5dba59e | 5852 | case IORING_OP_CLOSE: |
bfe76559 | 5853 | return io_close_prep(req, sqe); |
05f3fb3c | 5854 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5855 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5856 | case IORING_OP_STATX: |
bfe76559 | 5857 | return io_statx_prep(req, sqe); |
4840e418 | 5858 | case IORING_OP_FADVISE: |
bfe76559 | 5859 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5860 | case IORING_OP_MADVISE: |
bfe76559 | 5861 | return io_madvise_prep(req, sqe); |
cebdb986 | 5862 | case IORING_OP_OPENAT2: |
bfe76559 | 5863 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5864 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5865 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5866 | case IORING_OP_SPLICE: |
bfe76559 | 5867 | return io_splice_prep(req, sqe); |
ddf0322d | 5868 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5869 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5870 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5871 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5872 | case IORING_OP_TEE: |
bfe76559 | 5873 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5874 | case IORING_OP_SHUTDOWN: |
5875 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5876 | case IORING_OP_RENAMEAT: |
5877 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5878 | case IORING_OP_UNLINKAT: |
5879 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5880 | } |
5881 | ||
bfe76559 PB |
5882 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5883 | req->opcode); | |
5884 | return-EINVAL; | |
5885 | } | |
5886 | ||
93642ef8 | 5887 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5888 | { |
b7e298d2 PB |
5889 | if (!io_op_defs[req->opcode].needs_async_setup) |
5890 | return 0; | |
5891 | if (WARN_ON_ONCE(req->async_data)) | |
5892 | return -EFAULT; | |
5893 | if (io_alloc_async_data(req)) | |
5894 | return -EAGAIN; | |
5895 | ||
93642ef8 PB |
5896 | switch (req->opcode) { |
5897 | case IORING_OP_READV: | |
93642ef8 PB |
5898 | return io_rw_prep_async(req, READ); |
5899 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5900 | return io_rw_prep_async(req, WRITE); |
5901 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5902 | return io_sendmsg_prep_async(req); |
5903 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5904 | return io_recvmsg_prep_async(req); |
5905 | case IORING_OP_CONNECT: | |
5906 | return io_connect_prep_async(req); | |
5907 | } | |
b7e298d2 PB |
5908 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5909 | req->opcode); | |
5910 | return -EFAULT; | |
f67676d1 JA |
5911 | } |
5912 | ||
9cf7c104 PB |
5913 | static u32 io_get_sequence(struct io_kiocb *req) |
5914 | { | |
5915 | struct io_kiocb *pos; | |
5916 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5917 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5918 | |
f2f87370 PB |
5919 | io_for_each_link(pos, req) |
5920 | nr_reqs++; | |
9cf7c104 PB |
5921 | |
5922 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5923 | return total_submitted - nr_reqs; | |
5924 | } | |
5925 | ||
be7053b7 | 5926 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5927 | { |
a197f664 | 5928 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5929 | struct io_defer_entry *de; |
f67676d1 | 5930 | int ret; |
9cf7c104 | 5931 | u32 seq; |
de0617e4 | 5932 | |
9d858b21 | 5933 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5934 | if (likely(list_empty_careful(&ctx->defer_list) && |
5935 | !(req->flags & REQ_F_IO_DRAIN))) | |
5936 | return 0; | |
5937 | ||
5938 | seq = io_get_sequence(req); | |
5939 | /* Still a chance to pass the sequence check */ | |
5940 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5941 | return 0; |
5942 | ||
b7e298d2 | 5943 | ret = io_req_prep_async(req); |
be7053b7 PB |
5944 | if (ret) |
5945 | return ret; | |
cbdcb435 | 5946 | io_prep_async_link(req); |
27dc8338 PB |
5947 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5948 | if (!de) | |
5949 | return -ENOMEM; | |
2d28390a | 5950 | |
de0617e4 | 5951 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 5952 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 5953 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 5954 | kfree(de); |
ae34817b PB |
5955 | io_queue_async_work(req); |
5956 | return -EIOCBQUEUED; | |
de0617e4 JA |
5957 | } |
5958 | ||
915967f6 | 5959 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 5960 | de->req = req; |
9cf7c104 | 5961 | de->seq = seq; |
27dc8338 | 5962 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
5963 | spin_unlock_irq(&ctx->completion_lock); |
5964 | return -EIOCBQUEUED; | |
5965 | } | |
5966 | ||
3ca405eb | 5967 | static void __io_clean_op(struct io_kiocb *req) |
99bc4c38 | 5968 | { |
0e1b6fe3 PB |
5969 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
5970 | switch (req->opcode) { | |
5971 | case IORING_OP_READV: | |
5972 | case IORING_OP_READ_FIXED: | |
5973 | case IORING_OP_READ: | |
bcda7baa | 5974 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
5975 | break; |
5976 | case IORING_OP_RECVMSG: | |
5977 | case IORING_OP_RECV: | |
bcda7baa | 5978 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
5979 | break; |
5980 | } | |
5981 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
5982 | } |
5983 | ||
0e1b6fe3 PB |
5984 | if (req->flags & REQ_F_NEED_CLEANUP) { |
5985 | switch (req->opcode) { | |
5986 | case IORING_OP_READV: | |
5987 | case IORING_OP_READ_FIXED: | |
5988 | case IORING_OP_READ: | |
5989 | case IORING_OP_WRITEV: | |
5990 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
5991 | case IORING_OP_WRITE: { |
5992 | struct io_async_rw *io = req->async_data; | |
5993 | if (io->free_iovec) | |
5994 | kfree(io->free_iovec); | |
0e1b6fe3 | 5995 | break; |
e8c2bc1f | 5996 | } |
0e1b6fe3 | 5997 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
5998 | case IORING_OP_SENDMSG: { |
5999 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6000 | |
6001 | kfree(io->free_iov); | |
0e1b6fe3 | 6002 | break; |
e8c2bc1f | 6003 | } |
0e1b6fe3 PB |
6004 | case IORING_OP_SPLICE: |
6005 | case IORING_OP_TEE: | |
6006 | io_put_file(req, req->splice.file_in, | |
6007 | (req->splice.flags & SPLICE_F_FD_IN_FIXED)); | |
6008 | break; | |
f3cd4850 JA |
6009 | case IORING_OP_OPENAT: |
6010 | case IORING_OP_OPENAT2: | |
6011 | if (req->open.filename) | |
6012 | putname(req->open.filename); | |
6013 | break; | |
80a261fd JA |
6014 | case IORING_OP_RENAMEAT: |
6015 | putname(req->rename.oldpath); | |
6016 | putname(req->rename.newpath); | |
6017 | break; | |
14a1143b JA |
6018 | case IORING_OP_UNLINKAT: |
6019 | putname(req->unlink.filename); | |
6020 | break; | |
0e1b6fe3 PB |
6021 | } |
6022 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6023 | } |
99bc4c38 PB |
6024 | } |
6025 | ||
889fca73 | 6026 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6027 | { |
a197f664 | 6028 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6029 | const struct cred *creds = NULL; |
d625c6ee | 6030 | int ret; |
2b188cc1 | 6031 | |
003e8dcc JA |
6032 | if (req->work.creds && req->work.creds != current_cred()) |
6033 | creds = override_creds(req->work.creds); | |
5730b27e | 6034 | |
d625c6ee | 6035 | switch (req->opcode) { |
2b188cc1 | 6036 | case IORING_OP_NOP: |
889fca73 | 6037 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6038 | break; |
6039 | case IORING_OP_READV: | |
edafccee | 6040 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6041 | case IORING_OP_READ: |
889fca73 | 6042 | ret = io_read(req, issue_flags); |
edafccee | 6043 | break; |
3529d8c2 | 6044 | case IORING_OP_WRITEV: |
edafccee | 6045 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6046 | case IORING_OP_WRITE: |
889fca73 | 6047 | ret = io_write(req, issue_flags); |
2b188cc1 | 6048 | break; |
c992fe29 | 6049 | case IORING_OP_FSYNC: |
45d189c6 | 6050 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6051 | break; |
221c5eb2 | 6052 | case IORING_OP_POLL_ADD: |
61e98203 | 6053 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6054 | break; |
6055 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6056 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6057 | break; |
5d17b4a4 | 6058 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6059 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6060 | break; |
0fa03c62 | 6061 | case IORING_OP_SENDMSG: |
889fca73 | 6062 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6063 | break; |
fddaface | 6064 | case IORING_OP_SEND: |
889fca73 | 6065 | ret = io_send(req, issue_flags); |
0fa03c62 | 6066 | break; |
aa1fa28f | 6067 | case IORING_OP_RECVMSG: |
889fca73 | 6068 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6069 | break; |
fddaface | 6070 | case IORING_OP_RECV: |
889fca73 | 6071 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6072 | break; |
5262f567 | 6073 | case IORING_OP_TIMEOUT: |
61e98203 | 6074 | ret = io_timeout(req, issue_flags); |
5262f567 | 6075 | break; |
11365043 | 6076 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6077 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6078 | break; |
17f2fe35 | 6079 | case IORING_OP_ACCEPT: |
889fca73 | 6080 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6081 | break; |
f8e85cf2 | 6082 | case IORING_OP_CONNECT: |
889fca73 | 6083 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6084 | break; |
62755e35 | 6085 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6086 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6087 | break; |
d63d1b5e | 6088 | case IORING_OP_FALLOCATE: |
45d189c6 | 6089 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6090 | break; |
15b71abe | 6091 | case IORING_OP_OPENAT: |
45d189c6 | 6092 | ret = io_openat(req, issue_flags); |
15b71abe | 6093 | break; |
b5dba59e | 6094 | case IORING_OP_CLOSE: |
889fca73 | 6095 | ret = io_close(req, issue_flags); |
b5dba59e | 6096 | break; |
05f3fb3c | 6097 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6098 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6099 | break; |
eddc7ef5 | 6100 | case IORING_OP_STATX: |
45d189c6 | 6101 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6102 | break; |
4840e418 | 6103 | case IORING_OP_FADVISE: |
45d189c6 | 6104 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6105 | break; |
c1ca757b | 6106 | case IORING_OP_MADVISE: |
45d189c6 | 6107 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6108 | break; |
cebdb986 | 6109 | case IORING_OP_OPENAT2: |
45d189c6 | 6110 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6111 | break; |
3e4827b0 | 6112 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6113 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6114 | break; |
7d67af2c | 6115 | case IORING_OP_SPLICE: |
45d189c6 | 6116 | ret = io_splice(req, issue_flags); |
7d67af2c | 6117 | break; |
ddf0322d | 6118 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6119 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6120 | break; |
067524e9 | 6121 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6122 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6123 | break; |
f2a8d5c7 | 6124 | case IORING_OP_TEE: |
45d189c6 | 6125 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6126 | break; |
36f4fa68 | 6127 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6128 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6129 | break; |
80a261fd | 6130 | case IORING_OP_RENAMEAT: |
45d189c6 | 6131 | ret = io_renameat(req, issue_flags); |
80a261fd | 6132 | break; |
14a1143b | 6133 | case IORING_OP_UNLINKAT: |
45d189c6 | 6134 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6135 | break; |
2b188cc1 JA |
6136 | default: |
6137 | ret = -EINVAL; | |
6138 | break; | |
6139 | } | |
6140 | ||
5730b27e JA |
6141 | if (creds) |
6142 | revert_creds(creds); | |
6143 | ||
def596e9 JA |
6144 | if (ret) |
6145 | return ret; | |
6146 | ||
b532576e JA |
6147 | /* If the op doesn't have a file, we're not polling for it */ |
6148 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6149 | const bool in_async = io_wq_current_is_worker(); |
6150 | ||
11ba820b JA |
6151 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6152 | if (in_async) | |
6153 | mutex_lock(&ctx->uring_lock); | |
6154 | ||
2e9dbe90 | 6155 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6156 | |
6157 | if (in_async) | |
6158 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6159 | } |
6160 | ||
6161 | return 0; | |
2b188cc1 JA |
6162 | } |
6163 | ||
5280f7e5 | 6164 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6165 | { |
6166 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6167 | struct io_kiocb *timeout; |
561fb04a | 6168 | int ret = 0; |
2b188cc1 | 6169 | |
6df1db6b PB |
6170 | timeout = io_prep_linked_timeout(req); |
6171 | if (timeout) | |
6172 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6173 | |
4014d943 | 6174 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6175 | ret = -ECANCELED; |
31b51510 | 6176 | |
561fb04a | 6177 | if (!ret) { |
561fb04a | 6178 | do { |
889fca73 | 6179 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6180 | /* |
6181 | * We can get EAGAIN for polled IO even though we're | |
6182 | * forcing a sync submission from here, since we can't | |
6183 | * wait for request slots on the block side. | |
6184 | */ | |
6185 | if (ret != -EAGAIN) | |
6186 | break; | |
6187 | cond_resched(); | |
6188 | } while (1); | |
6189 | } | |
31b51510 | 6190 | |
a3df7698 | 6191 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6192 | if (ret) { |
a3df7698 | 6193 | /* io-wq is going to take one down */ |
de9b4cca | 6194 | req_ref_get(req); |
a3df7698 | 6195 | io_req_task_queue_fail(req, ret); |
edafccee | 6196 | } |
2b188cc1 JA |
6197 | } |
6198 | ||
7b29f92d JA |
6199 | #define FFS_ASYNC_READ 0x1UL |
6200 | #define FFS_ASYNC_WRITE 0x2UL | |
6201 | #ifdef CONFIG_64BIT | |
6202 | #define FFS_ISREG 0x4UL | |
6203 | #else | |
6204 | #define FFS_ISREG 0x0UL | |
6205 | #endif | |
6206 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6207 | ||
dafecf19 PB |
6208 | static inline struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data, |
6209 | unsigned i) | |
65e19f54 | 6210 | { |
269bbe5f | 6211 | struct fixed_rsrc_table *table; |
65e19f54 | 6212 | |
dafecf19 PB |
6213 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
6214 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
6215 | } | |
6216 | ||
6217 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, | |
6218 | int index) | |
6219 | { | |
7b29f92d JA |
6220 | struct file **file_slot = io_fixed_file_slot(ctx->file_data, index); |
6221 | ||
6222 | return (struct file *) ((unsigned long) *file_slot & FFS_MASK); | |
65e19f54 JA |
6223 | } |
6224 | ||
8371adf5 PB |
6225 | static struct file *io_file_get(struct io_submit_state *state, |
6226 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6227 | { |
a197f664 | 6228 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6229 | struct file *file; |
09bb8394 | 6230 | |
8da11c19 | 6231 | if (fixed) { |
7b29f92d JA |
6232 | unsigned long file_ptr; |
6233 | ||
479f517b | 6234 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6235 | return NULL; |
b7620121 | 6236 | fd = array_index_nospec(fd, ctx->nr_user_files); |
7b29f92d JA |
6237 | file_ptr = (unsigned long) *io_fixed_file_slot(ctx->file_data, fd); |
6238 | file = (struct file *) (file_ptr & FFS_MASK); | |
6239 | file_ptr &= ~FFS_MASK; | |
6240 | /* mask in overlapping REQ_F and FFS bits */ | |
6241 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
36f72fe2 | 6242 | io_set_resource_node(req); |
09bb8394 | 6243 | } else { |
c826bd7a | 6244 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6245 | file = __io_file_get(state, fd); |
d44f554e JA |
6246 | |
6247 | /* we don't allow fixed io_uring files */ | |
6248 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6249 | io_req_track_inflight(req); | |
09bb8394 JA |
6250 | } |
6251 | ||
8371adf5 | 6252 | return file; |
09bb8394 JA |
6253 | } |
6254 | ||
2665abfd | 6255 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6256 | { |
ad8a48ac JA |
6257 | struct io_timeout_data *data = container_of(timer, |
6258 | struct io_timeout_data, timer); | |
90cd7e42 | 6259 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6260 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6261 | unsigned long flags; |
2665abfd JA |
6262 | |
6263 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6264 | prev = req->timeout.head; |
6265 | req->timeout.head = NULL; | |
2665abfd JA |
6266 | |
6267 | /* | |
6268 | * We don't expect the list to be empty, that will only happen if we | |
6269 | * race with the completion of the linked work. | |
6270 | */ | |
de9b4cca | 6271 | if (prev && req_ref_inc_not_zero(prev)) |
f2f87370 | 6272 | io_remove_next_linked(prev); |
90cd7e42 PB |
6273 | else |
6274 | prev = NULL; | |
2665abfd JA |
6275 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6276 | ||
6277 | if (prev) { | |
014db007 | 6278 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6279 | io_put_req_deferred(prev, 1); |
47f46768 | 6280 | } else { |
9ae1f8dd PB |
6281 | io_req_complete_post(req, -ETIME, 0); |
6282 | io_put_req_deferred(req, 1); | |
2665abfd | 6283 | } |
2665abfd JA |
6284 | return HRTIMER_NORESTART; |
6285 | } | |
6286 | ||
de968c18 | 6287 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6288 | { |
de968c18 PB |
6289 | struct io_ring_ctx *ctx = req->ctx; |
6290 | ||
6291 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6292 | /* |
f2f87370 PB |
6293 | * If the back reference is NULL, then our linked request finished |
6294 | * before we got a chance to setup the timer | |
76a46e06 | 6295 | */ |
90cd7e42 | 6296 | if (req->timeout.head) { |
e8c2bc1f | 6297 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6298 | |
ad8a48ac JA |
6299 | data->timer.function = io_link_timeout_fn; |
6300 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6301 | data->mode); | |
2665abfd | 6302 | } |
76a46e06 | 6303 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6304 | /* drop submission reference */ |
76a46e06 JA |
6305 | io_put_req(req); |
6306 | } | |
2665abfd | 6307 | |
ad8a48ac | 6308 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6309 | { |
f2f87370 | 6310 | struct io_kiocb *nxt = req->link; |
2665abfd | 6311 | |
f2f87370 PB |
6312 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6313 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6314 | return NULL; |
2665abfd | 6315 | |
90cd7e42 | 6316 | nxt->timeout.head = req; |
900fad45 | 6317 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6318 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6319 | return nxt; |
2665abfd JA |
6320 | } |
6321 | ||
c5eef2b9 | 6322 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6323 | { |
d3d7298d | 6324 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6325 | int ret; |
2b188cc1 | 6326 | |
c5eef2b9 | 6327 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6328 | |
491381ce JA |
6329 | /* |
6330 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6331 | * doesn't support non-blocking read/write attempts | |
6332 | */ | |
24c74678 | 6333 | if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
f063c547 | 6334 | if (!io_arm_poll_handler(req)) { |
f063c547 PB |
6335 | /* |
6336 | * Queued up for async execution, worker will release | |
6337 | * submit reference when the iocb is actually submitted. | |
6338 | */ | |
6339 | io_queue_async_work(req); | |
2b188cc1 | 6340 | } |
0d63c148 PB |
6341 | } else if (likely(!ret)) { |
6342 | /* drop submission reference */ | |
e342c807 | 6343 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6344 | struct io_ring_ctx *ctx = req->ctx; |
6345 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6346 | |
6dd0be1e | 6347 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6348 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6349 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6350 | } else { |
d3d7298d | 6351 | io_put_req(req); |
0d63c148 PB |
6352 | } |
6353 | } else { | |
f41db273 | 6354 | io_req_complete_failed(req, ret); |
9e645e11 | 6355 | } |
d3d7298d PB |
6356 | if (linked_timeout) |
6357 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6358 | } |
6359 | ||
be7053b7 | 6360 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6361 | { |
6362 | int ret; | |
6363 | ||
be7053b7 | 6364 | ret = io_req_defer(req); |
4fe2c963 JL |
6365 | if (ret) { |
6366 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6367 | fail_req: |
f41db273 | 6368 | io_req_complete_failed(req, ret); |
4fe2c963 | 6369 | } |
2550878f | 6370 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6371 | ret = io_req_prep_async(req); |
be7053b7 PB |
6372 | if (unlikely(ret)) |
6373 | goto fail_req; | |
ce35a47a JA |
6374 | io_queue_async_work(req); |
6375 | } else { | |
c5eef2b9 | 6376 | __io_queue_sqe(req); |
ce35a47a | 6377 | } |
4fe2c963 JL |
6378 | } |
6379 | ||
b16fed66 PB |
6380 | /* |
6381 | * Check SQE restrictions (opcode and flags). | |
6382 | * | |
6383 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6384 | */ | |
6385 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6386 | struct io_kiocb *req, | |
6387 | unsigned int sqe_flags) | |
4fe2c963 | 6388 | { |
b16fed66 PB |
6389 | if (!ctx->restricted) |
6390 | return true; | |
6391 | ||
6392 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6393 | return false; | |
6394 | ||
6395 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6396 | ctx->restrictions.sqe_flags_required) | |
6397 | return false; | |
6398 | ||
6399 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6400 | ctx->restrictions.sqe_flags_required)) | |
6401 | return false; | |
6402 | ||
6403 | return true; | |
4fe2c963 JL |
6404 | } |
6405 | ||
b16fed66 PB |
6406 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6407 | const struct io_uring_sqe *sqe) | |
6408 | { | |
6409 | struct io_submit_state *state; | |
6410 | unsigned int sqe_flags; | |
003e8dcc | 6411 | int personality, ret = 0; |
b16fed66 PB |
6412 | |
6413 | req->opcode = READ_ONCE(sqe->opcode); | |
6414 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6415 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6416 | req->user_data = READ_ONCE(sqe->user_data); | |
6417 | req->async_data = NULL; | |
6418 | req->file = NULL; | |
6419 | req->ctx = ctx; | |
6420 | req->link = NULL; | |
6421 | req->fixed_rsrc_refs = NULL; | |
6422 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6423 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6424 | req->task = current; |
6425 | req->result = 0; | |
93e68e03 JA |
6426 | req->work.list.next = NULL; |
6427 | req->work.creds = NULL; | |
6428 | req->work.flags = 0; | |
b16fed66 PB |
6429 | |
6430 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6431 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6432 | req->flags = 0; | |
b16fed66 | 6433 | return -EINVAL; |
ebf4a5db | 6434 | } |
b16fed66 PB |
6435 | |
6436 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6437 | return -EINVAL; | |
6438 | ||
b16fed66 PB |
6439 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6440 | return -EACCES; | |
6441 | ||
6442 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6443 | !io_op_defs[req->opcode].buffer_select) | |
6444 | return -EOPNOTSUPP; | |
863e0560 | 6445 | |
003e8dcc JA |
6446 | personality = READ_ONCE(sqe->personality); |
6447 | if (personality) { | |
61cf9370 | 6448 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6449 | if (!req->work.creds) |
6450 | return -EINVAL; | |
6451 | get_cred(req->work.creds); | |
003e8dcc | 6452 | } |
b16fed66 PB |
6453 | state = &ctx->submit_state; |
6454 | ||
6455 | /* | |
6456 | * Plug now if we have more than 1 IO left after this, and the target | |
6457 | * is potentially a read/write to block based storage. | |
6458 | */ | |
6459 | if (!state->plug_started && state->ios_left > 1 && | |
6460 | io_op_defs[req->opcode].plug) { | |
6461 | blk_start_plug(&state->plug); | |
6462 | state->plug_started = true; | |
6463 | } | |
6464 | ||
6465 | if (io_op_defs[req->opcode].needs_file) { | |
6466 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6467 | ||
6468 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6469 | if (unlikely(!req->file)) | |
6470 | ret = -EBADF; | |
6471 | } | |
6472 | ||
6473 | state->ios_left--; | |
6474 | return ret; | |
6475 | } | |
6476 | ||
a6b8cadc | 6477 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6478 | const struct io_uring_sqe *sqe) |
9e645e11 | 6479 | { |
a1ab7b35 | 6480 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6481 | int ret; |
9e645e11 | 6482 | |
a6b8cadc PB |
6483 | ret = io_init_req(ctx, req, sqe); |
6484 | if (unlikely(ret)) { | |
6485 | fail_req: | |
de59bc10 PB |
6486 | if (link->head) { |
6487 | /* fail even hard links since we don't submit */ | |
cf109604 | 6488 | link->head->flags |= REQ_F_FAIL_LINK; |
f41db273 | 6489 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6490 | link->head = NULL; |
6491 | } | |
f41db273 | 6492 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6493 | return ret; |
6494 | } | |
be7053b7 PB |
6495 | ret = io_req_prep(req, sqe); |
6496 | if (unlikely(ret)) | |
6497 | goto fail_req; | |
a6b8cadc | 6498 | |
be7053b7 | 6499 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6500 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6501 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6502 | ||
9e645e11 JA |
6503 | /* |
6504 | * If we already have a head request, queue this one for async | |
6505 | * submittal once the head completes. If we don't have a head but | |
6506 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6507 | * submitted sync once the chain is complete. If none of those | |
6508 | * conditions are true (normal request), then just queue it. | |
6509 | */ | |
863e0560 PB |
6510 | if (link->head) { |
6511 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6512 | |
8cdf2193 PB |
6513 | /* |
6514 | * Taking sequential execution of a link, draining both sides | |
6515 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6516 | * requests in the link. So, it drains the head and the | |
6517 | * next after the link request. The last one is done via | |
6518 | * drain_next flag to persist the effect across calls. | |
6519 | */ | |
ef4ff581 | 6520 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6521 | head->flags |= REQ_F_IO_DRAIN; |
6522 | ctx->drain_next = 1; | |
6523 | } | |
b7e298d2 | 6524 | ret = io_req_prep_async(req); |
cf109604 | 6525 | if (unlikely(ret)) |
a6b8cadc | 6526 | goto fail_req; |
9d76377f | 6527 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6528 | link->last->link = req; |
863e0560 | 6529 | link->last = req; |
32fe525b PB |
6530 | |
6531 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6532 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6533 | io_queue_sqe(head); |
863e0560 | 6534 | link->head = NULL; |
32fe525b | 6535 | } |
9e645e11 | 6536 | } else { |
711be031 PB |
6537 | if (unlikely(ctx->drain_next)) { |
6538 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6539 | ctx->drain_next = 0; |
711be031 | 6540 | } |
ef4ff581 | 6541 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6542 | link->head = req; |
6543 | link->last = req; | |
711be031 | 6544 | } else { |
be7053b7 | 6545 | io_queue_sqe(req); |
711be031 | 6546 | } |
9e645e11 | 6547 | } |
2e6e1fde | 6548 | |
1d4240cc | 6549 | return 0; |
9e645e11 JA |
6550 | } |
6551 | ||
9a56a232 JA |
6552 | /* |
6553 | * Batched submission is done, ensure local IO is flushed out. | |
6554 | */ | |
ba88ff11 PB |
6555 | static void io_submit_state_end(struct io_submit_state *state, |
6556 | struct io_ring_ctx *ctx) | |
9a56a232 | 6557 | { |
a1ab7b35 | 6558 | if (state->link.head) |
de59bc10 | 6559 | io_queue_sqe(state->link.head); |
6dd0be1e | 6560 | if (state->comp.nr) |
ba88ff11 | 6561 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6562 | if (state->plug_started) |
6563 | blk_finish_plug(&state->plug); | |
9f13c35b | 6564 | io_state_file_put(state); |
9a56a232 JA |
6565 | } |
6566 | ||
6567 | /* | |
6568 | * Start submission side cache. | |
6569 | */ | |
6570 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6571 | unsigned int max_ios) |
9a56a232 | 6572 | { |
27926b68 | 6573 | state->plug_started = false; |
9a56a232 | 6574 | state->ios_left = max_ios; |
a1ab7b35 PB |
6575 | /* set only head, no need to init link_last in advance */ |
6576 | state->link.head = NULL; | |
9a56a232 JA |
6577 | } |
6578 | ||
2b188cc1 JA |
6579 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6580 | { | |
75b28aff | 6581 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6582 | |
caf582c6 PB |
6583 | /* |
6584 | * Ensure any loads from the SQEs are done at this point, | |
6585 | * since once we write the new head, the application could | |
6586 | * write new data to them. | |
6587 | */ | |
6588 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6589 | } |
6590 | ||
2b188cc1 | 6591 | /* |
3529d8c2 | 6592 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6593 | * that is mapped by userspace. This means that care needs to be taken to |
6594 | * ensure that reads are stable, as we cannot rely on userspace always | |
6595 | * being a good citizen. If members of the sqe are validated and then later | |
6596 | * used, it's important that those reads are done through READ_ONCE() to | |
6597 | * prevent a re-load down the line. | |
6598 | */ | |
709b302f | 6599 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6600 | { |
75b28aff | 6601 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6602 | unsigned head; |
6603 | ||
6604 | /* | |
6605 | * The cached sq head (or cq tail) serves two purposes: | |
6606 | * | |
6607 | * 1) allows us to batch the cost of updating the user visible | |
6608 | * head updates. | |
6609 | * 2) allows the kernel side to track the head on its own, even | |
6610 | * though the application is the one updating it. | |
6611 | */ | |
4fccfcbb | 6612 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6613 | if (likely(head < ctx->sq_entries)) |
6614 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6615 | |
6616 | /* drop invalid entries */ | |
498ccd9e | 6617 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6618 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6619 | return NULL; |
6620 | } | |
6621 | ||
0f212204 | 6622 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6623 | { |
46c4e16a | 6624 | int submitted = 0; |
6c271ce2 | 6625 | |
c4a2ed72 | 6626 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6627 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c503150 | 6628 | if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL)) |
ad3eb2c8 JA |
6629 | return -EBUSY; |
6630 | } | |
6c271ce2 | 6631 | |
ee7d46d9 PB |
6632 | /* make sure SQ entry isn't read before tail */ |
6633 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6634 | |
2b85edfc PB |
6635 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6636 | return -EAGAIN; | |
6c271ce2 | 6637 | |
d8a6df10 | 6638 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6639 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6640 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6641 | |
46c4e16a | 6642 | while (submitted < nr) { |
3529d8c2 | 6643 | const struct io_uring_sqe *sqe; |
196be95c | 6644 | struct io_kiocb *req; |
fb5ccc98 | 6645 | |
258b29a9 | 6646 | req = io_alloc_req(ctx); |
196be95c PB |
6647 | if (unlikely(!req)) { |
6648 | if (!submitted) | |
6649 | submitted = -EAGAIN; | |
fb5ccc98 | 6650 | break; |
196be95c | 6651 | } |
4fccfcbb PB |
6652 | sqe = io_get_sqe(ctx); |
6653 | if (unlikely(!sqe)) { | |
6654 | kmem_cache_free(req_cachep, req); | |
6655 | break; | |
6656 | } | |
d3656344 JA |
6657 | /* will complete beyond this point, count as submitted */ |
6658 | submitted++; | |
a1ab7b35 | 6659 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6660 | break; |
6c271ce2 JA |
6661 | } |
6662 | ||
9466f437 PB |
6663 | if (unlikely(submitted != nr)) { |
6664 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6665 | struct io_uring_task *tctx = current->io_uring; |
6666 | int unused = nr - ref_used; | |
9466f437 | 6667 | |
d8a6df10 JA |
6668 | percpu_ref_put_many(&ctx->refs, unused); |
6669 | percpu_counter_sub(&tctx->inflight, unused); | |
6670 | put_task_struct_many(current, unused); | |
9466f437 | 6671 | } |
6c271ce2 | 6672 | |
a1ab7b35 | 6673 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6674 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6675 | io_commit_sqring(ctx); | |
6676 | ||
6c271ce2 JA |
6677 | return submitted; |
6678 | } | |
6679 | ||
23b3628e XW |
6680 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6681 | { | |
6682 | /* Tell userspace we may need a wakeup call */ | |
6683 | spin_lock_irq(&ctx->completion_lock); | |
6684 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6685 | spin_unlock_irq(&ctx->completion_lock); | |
6686 | } | |
6687 | ||
6688 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6689 | { | |
6690 | spin_lock_irq(&ctx->completion_lock); | |
6691 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6692 | spin_unlock_irq(&ctx->completion_lock); | |
6693 | } | |
6694 | ||
08369246 | 6695 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6696 | { |
c8d1ba58 | 6697 | unsigned int to_submit; |
bdcd3eab | 6698 | int ret = 0; |
6c271ce2 | 6699 | |
c8d1ba58 | 6700 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6701 | /* if we're handling multiple rings, cap submit size for fairness */ |
6702 | if (cap_entries && to_submit > 8) | |
6703 | to_submit = 8; | |
6704 | ||
906a3c6f | 6705 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6706 | unsigned nr_events = 0; |
a4c0b3de | 6707 | |
c8d1ba58 | 6708 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6709 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6710 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6711 | |
0298ef96 PB |
6712 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6713 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6714 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6715 | mutex_unlock(&ctx->uring_lock); |
6716 | } | |
6c271ce2 | 6717 | |
90554200 JA |
6718 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6719 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6720 | |
08369246 XW |
6721 | return ret; |
6722 | } | |
6c271ce2 | 6723 | |
08369246 XW |
6724 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6725 | { | |
6726 | struct io_ring_ctx *ctx; | |
6727 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6728 | |
c9dca27d PB |
6729 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6730 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6731 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6732 | } |
6c271ce2 | 6733 | |
c8d1ba58 JA |
6734 | static int io_sq_thread(void *data) |
6735 | { | |
69fb2131 JA |
6736 | struct io_sq_data *sqd = data; |
6737 | struct io_ring_ctx *ctx; | |
a0d9205f | 6738 | unsigned long timeout = 0; |
37d1e2e3 | 6739 | char buf[TASK_COMM_LEN]; |
08369246 | 6740 | DEFINE_WAIT(wait); |
6c271ce2 | 6741 | |
696ee88a | 6742 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6743 | set_task_comm(current, buf); |
37d1e2e3 JA |
6744 | current->pf_io_worker = NULL; |
6745 | ||
6746 | if (sqd->sq_cpu != -1) | |
6747 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6748 | else | |
6749 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6750 | current->flags |= PF_NO_SETAFFINITY; | |
6751 | ||
09a6f4ef | 6752 | mutex_lock(&sqd->lock); |
05962f95 | 6753 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6754 | int ret; |
6755 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6756 | |
82734c5b JA |
6757 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6758 | signal_pending(current)) { | |
6759 | bool did_sig = false; | |
6760 | ||
09a6f4ef | 6761 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6762 | if (signal_pending(current)) { |
6763 | struct ksignal ksig; | |
6764 | ||
6765 | did_sig = get_signal(&ksig); | |
6766 | } | |
05962f95 | 6767 | cond_resched(); |
09a6f4ef | 6768 | mutex_lock(&sqd->lock); |
82734c5b JA |
6769 | if (did_sig) |
6770 | break; | |
521d6a73 | 6771 | io_run_task_work(); |
b7f5a0bf | 6772 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6773 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6774 | continue; |
08369246 | 6775 | } |
08369246 | 6776 | sqt_spin = false; |
e95eee2d | 6777 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6778 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6779 | const struct cred *creds = NULL; |
6780 | ||
6781 | if (ctx->sq_creds != current_cred()) | |
6782 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6783 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6784 | if (creds) |
6785 | revert_creds(creds); | |
08369246 XW |
6786 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6787 | sqt_spin = true; | |
69fb2131 | 6788 | } |
6c271ce2 | 6789 | |
08369246 | 6790 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6791 | io_run_task_work(); |
6792 | cond_resched(); | |
08369246 XW |
6793 | if (sqt_spin) |
6794 | timeout = jiffies + sqd->sq_thread_idle; | |
6795 | continue; | |
6796 | } | |
6797 | ||
08369246 XW |
6798 | needs_sched = true; |
6799 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6800 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6801 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6802 | !list_empty_careful(&ctx->iopoll_list)) { | |
6803 | needs_sched = false; | |
6804 | break; | |
6805 | } | |
6806 | if (io_sqring_entries(ctx)) { | |
6807 | needs_sched = false; | |
6808 | break; | |
6809 | } | |
6810 | } | |
6811 | ||
05962f95 | 6812 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6813 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6814 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6815 | |
09a6f4ef | 6816 | mutex_unlock(&sqd->lock); |
69fb2131 | 6817 | schedule(); |
09a6f4ef | 6818 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6819 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6820 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6821 | } |
08369246 XW |
6822 | |
6823 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6824 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6825 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6826 | } |
28cea78a | 6827 | |
521d6a73 PB |
6828 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6829 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6830 | sqd->thread = NULL; |
05962f95 | 6831 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6832 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6833 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6834 | |
6835 | io_run_task_work(); | |
b7f5a0bf | 6836 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6837 | complete(&sqd->exited); |
6838 | do_exit(0); | |
6c271ce2 JA |
6839 | } |
6840 | ||
bda52162 JA |
6841 | struct io_wait_queue { |
6842 | struct wait_queue_entry wq; | |
6843 | struct io_ring_ctx *ctx; | |
6844 | unsigned to_wait; | |
6845 | unsigned nr_timeouts; | |
6846 | }; | |
6847 | ||
6c503150 | 6848 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6849 | { |
6850 | struct io_ring_ctx *ctx = iowq->ctx; | |
6851 | ||
6852 | /* | |
d195a66e | 6853 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6854 | * started waiting. For timeouts, we always want to return to userspace, |
6855 | * regardless of event count. | |
6856 | */ | |
6c503150 | 6857 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6858 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6859 | } | |
6860 | ||
6861 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6862 | int wake_flags, void *key) | |
6863 | { | |
6864 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6865 | wq); | |
6866 | ||
6c503150 PB |
6867 | /* |
6868 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6869 | * the task, and the next invocation will do it. | |
6870 | */ | |
6871 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6872 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6873 | return -1; | |
bda52162 JA |
6874 | } |
6875 | ||
af9c1a44 JA |
6876 | static int io_run_task_work_sig(void) |
6877 | { | |
6878 | if (io_run_task_work()) | |
6879 | return 1; | |
6880 | if (!signal_pending(current)) | |
6881 | return 0; | |
0b8cfa97 | 6882 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6883 | return -ERESTARTSYS; |
af9c1a44 JA |
6884 | return -EINTR; |
6885 | } | |
6886 | ||
eeb60b9a PB |
6887 | /* when returns >0, the caller should retry */ |
6888 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6889 | struct io_wait_queue *iowq, | |
6890 | signed long *timeout) | |
6891 | { | |
6892 | int ret; | |
6893 | ||
6894 | /* make sure we run task_work before checking for signals */ | |
6895 | ret = io_run_task_work_sig(); | |
6896 | if (ret || io_should_wake(iowq)) | |
6897 | return ret; | |
6898 | /* let the caller flush overflows, retry */ | |
6899 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6900 | return 1; | |
6901 | ||
6902 | *timeout = schedule_timeout(*timeout); | |
6903 | return !*timeout ? -ETIME : 1; | |
6904 | } | |
6905 | ||
2b188cc1 JA |
6906 | /* |
6907 | * Wait until events become available, if we don't already have some. The | |
6908 | * application must reap them itself, as they reside on the shared cq ring. | |
6909 | */ | |
6910 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6911 | const sigset_t __user *sig, size_t sigsz, |
6912 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6913 | { |
bda52162 JA |
6914 | struct io_wait_queue iowq = { |
6915 | .wq = { | |
6916 | .private = current, | |
6917 | .func = io_wake_function, | |
6918 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6919 | }, | |
6920 | .ctx = ctx, | |
6921 | .to_wait = min_events, | |
6922 | }; | |
75b28aff | 6923 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6924 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6925 | int ret; | |
2b188cc1 | 6926 | |
b41e9852 | 6927 | do { |
6c503150 PB |
6928 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
6929 | if (io_cqring_events(ctx) >= min_events) | |
b41e9852 | 6930 | return 0; |
4c6e277c | 6931 | if (!io_run_task_work()) |
b41e9852 | 6932 | break; |
b41e9852 | 6933 | } while (1); |
2b188cc1 JA |
6934 | |
6935 | if (sig) { | |
9e75ad5d AB |
6936 | #ifdef CONFIG_COMPAT |
6937 | if (in_compat_syscall()) | |
6938 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 6939 | sigsz); |
9e75ad5d AB |
6940 | else |
6941 | #endif | |
b772434b | 6942 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 6943 | |
2b188cc1 JA |
6944 | if (ret) |
6945 | return ret; | |
6946 | } | |
6947 | ||
c73ebb68 | 6948 | if (uts) { |
c1d5a224 PB |
6949 | struct timespec64 ts; |
6950 | ||
c73ebb68 HX |
6951 | if (get_timespec64(&ts, uts)) |
6952 | return -EFAULT; | |
6953 | timeout = timespec64_to_jiffies(&ts); | |
6954 | } | |
6955 | ||
bda52162 | 6956 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 6957 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 6958 | do { |
ca0a2651 JA |
6959 | /* if we can't even flush overflow, don't wait for more */ |
6960 | if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) { | |
6961 | ret = -EBUSY; | |
6962 | break; | |
6963 | } | |
bda52162 JA |
6964 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
6965 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
6966 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
6967 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 6968 | cond_resched(); |
eeb60b9a | 6969 | } while (ret > 0); |
bda52162 | 6970 | |
b7db41c9 | 6971 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 6972 | |
75b28aff | 6973 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
6974 | } |
6975 | ||
6b06314c JA |
6976 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
6977 | { | |
6978 | #if defined(CONFIG_UNIX) | |
6979 | if (ctx->ring_sock) { | |
6980 | struct sock *sock = ctx->ring_sock->sk; | |
6981 | struct sk_buff *skb; | |
6982 | ||
6983 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
6984 | kfree_skb(skb); | |
6985 | } | |
6986 | #else | |
6987 | int i; | |
6988 | ||
65e19f54 JA |
6989 | for (i = 0; i < ctx->nr_user_files; i++) { |
6990 | struct file *file; | |
6991 | ||
6992 | file = io_file_from_index(ctx, i); | |
6993 | if (file) | |
6994 | fput(file); | |
6995 | } | |
6b06314c JA |
6996 | #endif |
6997 | } | |
6998 | ||
00835dce | 6999 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7000 | { |
269bbe5f | 7001 | struct fixed_rsrc_data *data; |
05f3fb3c | 7002 | |
269bbe5f | 7003 | data = container_of(ref, struct fixed_rsrc_data, refs); |
05f3fb3c JA |
7004 | complete(&data->done); |
7005 | } | |
7006 | ||
2a63b2d9 | 7007 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7008 | { |
2a63b2d9 | 7009 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7010 | } |
7011 | ||
2a63b2d9 | 7012 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7013 | { |
2a63b2d9 BM |
7014 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7015 | } | |
65e19f54 | 7016 | |
d67d2263 BM |
7017 | static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx, |
7018 | struct fixed_rsrc_data *rsrc_data, | |
269bbe5f | 7019 | struct fixed_rsrc_ref_node *ref_node) |
1642b445 | 7020 | { |
2a63b2d9 | 7021 | io_rsrc_ref_lock(ctx); |
269bbe5f | 7022 | rsrc_data->node = ref_node; |
d67d2263 | 7023 | list_add_tail(&ref_node->node, &ctx->rsrc_ref_list); |
2a63b2d9 | 7024 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7025 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7026 | } |
7027 | ||
8bad28d8 | 7028 | static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data) |
6b06314c | 7029 | { |
8bad28d8 | 7030 | struct fixed_rsrc_ref_node *ref_node = NULL; |
6b06314c | 7031 | |
2a63b2d9 | 7032 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7033 | ref_node = data->node; |
e6cb007c | 7034 | data->node = NULL; |
2a63b2d9 | 7035 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7036 | if (ref_node) |
7037 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
7038 | } |
7039 | ||
7040 | static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data, | |
7041 | struct io_ring_ctx *ctx, | |
f2303b1f PB |
7042 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7043 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7044 | { |
f2303b1f | 7045 | struct fixed_rsrc_ref_node *backup_node; |
8bad28d8 | 7046 | int ret; |
05589553 | 7047 | |
8bad28d8 HX |
7048 | if (data->quiesce) |
7049 | return -ENXIO; | |
05589553 | 7050 | |
8bad28d8 | 7051 | data->quiesce = true; |
1ffc5422 | 7052 | do { |
f2303b1f PB |
7053 | ret = -ENOMEM; |
7054 | backup_node = alloc_fixed_rsrc_ref_node(ctx); | |
7055 | if (!backup_node) | |
7056 | break; | |
7057 | backup_node->rsrc_data = data; | |
7058 | backup_node->rsrc_put = rsrc_put; | |
7059 | ||
8bad28d8 HX |
7060 | io_sqe_rsrc_kill_node(ctx, data); |
7061 | percpu_ref_kill(&data->refs); | |
7062 | flush_delayed_work(&ctx->rsrc_put_work); | |
7063 | ||
1ffc5422 PB |
7064 | ret = wait_for_completion_interruptible(&data->done); |
7065 | if (!ret) | |
7066 | break; | |
8bad28d8 | 7067 | |
cb5e1b81 | 7068 | percpu_ref_resurrect(&data->refs); |
8bad28d8 HX |
7069 | io_sqe_rsrc_set_node(ctx, data, backup_node); |
7070 | backup_node = NULL; | |
cb5e1b81 | 7071 | reinit_completion(&data->done); |
8bad28d8 | 7072 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7073 | ret = io_run_task_work_sig(); |
8bad28d8 | 7074 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7075 | } while (ret >= 0); |
8bad28d8 | 7076 | data->quiesce = false; |
05f3fb3c | 7077 | |
8bad28d8 HX |
7078 | if (backup_node) |
7079 | destroy_fixed_rsrc_ref_node(backup_node); | |
7080 | return ret; | |
d7954b2b BM |
7081 | } |
7082 | ||
1ad555c6 BM |
7083 | static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx) |
7084 | { | |
7085 | struct fixed_rsrc_data *data; | |
7086 | ||
7087 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7088 | if (!data) | |
7089 | return NULL; | |
7090 | ||
00835dce | 7091 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7092 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7093 | kfree(data); | |
7094 | return NULL; | |
7095 | } | |
7096 | data->ctx = ctx; | |
7097 | init_completion(&data->done); | |
7098 | return data; | |
7099 | } | |
7100 | ||
7101 | static void free_fixed_rsrc_data(struct fixed_rsrc_data *data) | |
7102 | { | |
7103 | percpu_ref_exit(&data->refs); | |
7104 | kfree(data->table); | |
7105 | kfree(data); | |
7106 | } | |
7107 | ||
d7954b2b BM |
7108 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7109 | { | |
7110 | struct fixed_rsrc_data *data = ctx->file_data; | |
d7954b2b BM |
7111 | unsigned nr_tables, i; |
7112 | int ret; | |
7113 | ||
8bad28d8 HX |
7114 | /* |
7115 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7116 | * Since we possibly drop uring lock later in this function to | |
7117 | * run task work. | |
7118 | */ | |
7119 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7120 | return -ENXIO; |
f2303b1f | 7121 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7122 | if (ret) |
7123 | return ret; | |
7124 | ||
6b06314c | 7125 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7126 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7127 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7128 | kfree(data->table[i].files); |
1ad555c6 | 7129 | free_fixed_rsrc_data(data); |
05f3fb3c | 7130 | ctx->file_data = NULL; |
6b06314c JA |
7131 | ctx->nr_user_files = 0; |
7132 | return 0; | |
7133 | } | |
7134 | ||
37d1e2e3 | 7135 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7136 | __releases(&sqd->lock) |
37d1e2e3 | 7137 | { |
521d6a73 PB |
7138 | WARN_ON_ONCE(sqd->thread == current); |
7139 | ||
9e138a48 PB |
7140 | /* |
7141 | * Do the dance but not conditional clear_bit() because it'd race with | |
7142 | * other threads incrementing park_pending and setting the bit. | |
7143 | */ | |
37d1e2e3 | 7144 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7145 | if (atomic_dec_return(&sqd->park_pending)) |
7146 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7147 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7148 | } |
7149 | ||
86e0d676 | 7150 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7151 | __acquires(&sqd->lock) |
37d1e2e3 | 7152 | { |
521d6a73 PB |
7153 | WARN_ON_ONCE(sqd->thread == current); |
7154 | ||
9e138a48 | 7155 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7156 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7157 | mutex_lock(&sqd->lock); |
05962f95 | 7158 | if (sqd->thread) |
86e0d676 | 7159 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7160 | } |
7161 | ||
7162 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7163 | { | |
521d6a73 PB |
7164 | WARN_ON_ONCE(sqd->thread == current); |
7165 | ||
09a6f4ef | 7166 | mutex_lock(&sqd->lock); |
05962f95 | 7167 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7168 | if (sqd->thread) |
7169 | wake_up_process(sqd->thread); | |
09a6f4ef | 7170 | mutex_unlock(&sqd->lock); |
05962f95 | 7171 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7172 | } |
7173 | ||
534ca6d6 | 7174 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7175 | { |
534ca6d6 | 7176 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7177 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7178 | ||
37d1e2e3 JA |
7179 | io_sq_thread_stop(sqd); |
7180 | kfree(sqd); | |
7181 | } | |
7182 | } | |
7183 | ||
7184 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7185 | { | |
7186 | struct io_sq_data *sqd = ctx->sq_data; | |
7187 | ||
7188 | if (sqd) { | |
05962f95 | 7189 | io_sq_thread_park(sqd); |
521d6a73 | 7190 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7191 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7192 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7193 | |
7194 | io_put_sq_data(sqd); | |
7195 | ctx->sq_data = NULL; | |
7c30f36a SM |
7196 | if (ctx->sq_creds) |
7197 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7198 | } |
7199 | } | |
7200 | ||
aa06165d JA |
7201 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7202 | { | |
7203 | struct io_ring_ctx *ctx_attach; | |
7204 | struct io_sq_data *sqd; | |
7205 | struct fd f; | |
7206 | ||
7207 | f = fdget(p->wq_fd); | |
7208 | if (!f.file) | |
7209 | return ERR_PTR(-ENXIO); | |
7210 | if (f.file->f_op != &io_uring_fops) { | |
7211 | fdput(f); | |
7212 | return ERR_PTR(-EINVAL); | |
7213 | } | |
7214 | ||
7215 | ctx_attach = f.file->private_data; | |
7216 | sqd = ctx_attach->sq_data; | |
7217 | if (!sqd) { | |
7218 | fdput(f); | |
7219 | return ERR_PTR(-EINVAL); | |
7220 | } | |
5c2469e0 JA |
7221 | if (sqd->task_tgid != current->tgid) { |
7222 | fdput(f); | |
7223 | return ERR_PTR(-EPERM); | |
7224 | } | |
aa06165d JA |
7225 | |
7226 | refcount_inc(&sqd->refs); | |
7227 | fdput(f); | |
7228 | return sqd; | |
7229 | } | |
7230 | ||
26984fbf PB |
7231 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7232 | bool *attached) | |
534ca6d6 JA |
7233 | { |
7234 | struct io_sq_data *sqd; | |
7235 | ||
26984fbf | 7236 | *attached = false; |
5c2469e0 JA |
7237 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7238 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7239 | if (!IS_ERR(sqd)) { |
7240 | *attached = true; | |
5c2469e0 | 7241 | return sqd; |
26984fbf | 7242 | } |
5c2469e0 JA |
7243 | /* fall through for EPERM case, setup new sqd/task */ |
7244 | if (PTR_ERR(sqd) != -EPERM) | |
7245 | return sqd; | |
7246 | } | |
aa06165d | 7247 | |
534ca6d6 JA |
7248 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7249 | if (!sqd) | |
7250 | return ERR_PTR(-ENOMEM); | |
7251 | ||
9e138a48 | 7252 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7253 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7254 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7255 | mutex_init(&sqd->lock); |
534ca6d6 | 7256 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7257 | init_completion(&sqd->exited); |
534ca6d6 JA |
7258 | return sqd; |
7259 | } | |
7260 | ||
6b06314c | 7261 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7262 | /* |
7263 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7264 | * the io_uring can be safely unregistered on process exit, even if we have | |
7265 | * loops in the file referencing. | |
7266 | */ | |
7267 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7268 | { | |
7269 | struct sock *sk = ctx->ring_sock->sk; | |
7270 | struct scm_fp_list *fpl; | |
7271 | struct sk_buff *skb; | |
08a45173 | 7272 | int i, nr_files; |
6b06314c | 7273 | |
6b06314c JA |
7274 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7275 | if (!fpl) | |
7276 | return -ENOMEM; | |
7277 | ||
7278 | skb = alloc_skb(0, GFP_KERNEL); | |
7279 | if (!skb) { | |
7280 | kfree(fpl); | |
7281 | return -ENOMEM; | |
7282 | } | |
7283 | ||
7284 | skb->sk = sk; | |
6b06314c | 7285 | |
08a45173 | 7286 | nr_files = 0; |
62e398be | 7287 | fpl->user = get_uid(current_user()); |
6b06314c | 7288 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7289 | struct file *file = io_file_from_index(ctx, i + offset); |
7290 | ||
7291 | if (!file) | |
08a45173 | 7292 | continue; |
65e19f54 | 7293 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7294 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7295 | nr_files++; | |
6b06314c JA |
7296 | } |
7297 | ||
08a45173 JA |
7298 | if (nr_files) { |
7299 | fpl->max = SCM_MAX_FD; | |
7300 | fpl->count = nr_files; | |
7301 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7302 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7303 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7304 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7305 | |
08a45173 JA |
7306 | for (i = 0; i < nr_files; i++) |
7307 | fput(fpl->fp[i]); | |
7308 | } else { | |
7309 | kfree_skb(skb); | |
7310 | kfree(fpl); | |
7311 | } | |
6b06314c JA |
7312 | |
7313 | return 0; | |
7314 | } | |
7315 | ||
7316 | /* | |
7317 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7318 | * causes regular reference counting to break down. We rely on the UNIX | |
7319 | * garbage collection to take care of this problem for us. | |
7320 | */ | |
7321 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7322 | { | |
7323 | unsigned left, total; | |
7324 | int ret = 0; | |
7325 | ||
7326 | total = 0; | |
7327 | left = ctx->nr_user_files; | |
7328 | while (left) { | |
7329 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7330 | |
7331 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7332 | if (ret) | |
7333 | break; | |
7334 | left -= this_files; | |
7335 | total += this_files; | |
7336 | } | |
7337 | ||
7338 | if (!ret) | |
7339 | return 0; | |
7340 | ||
7341 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7342 | struct file *file = io_file_from_index(ctx, total); |
7343 | ||
7344 | if (file) | |
7345 | fput(file); | |
6b06314c JA |
7346 | total++; |
7347 | } | |
7348 | ||
7349 | return ret; | |
7350 | } | |
7351 | #else | |
7352 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7353 | { | |
7354 | return 0; | |
7355 | } | |
7356 | #endif | |
7357 | ||
269bbe5f | 7358 | static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data, |
5398ae69 | 7359 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7360 | { |
7361 | int i; | |
7362 | ||
7363 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7364 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7365 | unsigned this_files; |
7366 | ||
7367 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7368 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7369 | GFP_KERNEL); | |
7370 | if (!table->files) | |
7371 | break; | |
7372 | nr_files -= this_files; | |
7373 | } | |
7374 | ||
7375 | if (i == nr_tables) | |
7376 | return 0; | |
7377 | ||
7378 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7379 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7380 | kfree(table->files); |
7381 | } | |
7382 | return 1; | |
7383 | } | |
7384 | ||
50238531 | 7385 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7386 | { |
50238531 | 7387 | struct file *file = prsrc->file; |
05f3fb3c JA |
7388 | #if defined(CONFIG_UNIX) |
7389 | struct sock *sock = ctx->ring_sock->sk; | |
7390 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7391 | struct sk_buff *skb; | |
7392 | int i; | |
7393 | ||
7394 | __skb_queue_head_init(&list); | |
7395 | ||
7396 | /* | |
7397 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7398 | * remove this entry and rearrange the file array. | |
7399 | */ | |
7400 | skb = skb_dequeue(head); | |
7401 | while (skb) { | |
7402 | struct scm_fp_list *fp; | |
7403 | ||
7404 | fp = UNIXCB(skb).fp; | |
7405 | for (i = 0; i < fp->count; i++) { | |
7406 | int left; | |
7407 | ||
7408 | if (fp->fp[i] != file) | |
7409 | continue; | |
7410 | ||
7411 | unix_notinflight(fp->user, fp->fp[i]); | |
7412 | left = fp->count - 1 - i; | |
7413 | if (left) { | |
7414 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7415 | left * sizeof(struct file *)); | |
7416 | } | |
7417 | fp->count--; | |
7418 | if (!fp->count) { | |
7419 | kfree_skb(skb); | |
7420 | skb = NULL; | |
7421 | } else { | |
7422 | __skb_queue_tail(&list, skb); | |
7423 | } | |
7424 | fput(file); | |
7425 | file = NULL; | |
7426 | break; | |
7427 | } | |
7428 | ||
7429 | if (!file) | |
7430 | break; | |
7431 | ||
7432 | __skb_queue_tail(&list, skb); | |
7433 | ||
7434 | skb = skb_dequeue(head); | |
7435 | } | |
7436 | ||
7437 | if (skb_peek(&list)) { | |
7438 | spin_lock_irq(&head->lock); | |
7439 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7440 | __skb_queue_tail(head, skb); | |
7441 | spin_unlock_irq(&head->lock); | |
7442 | } | |
7443 | #else | |
7444 | fput(file); | |
7445 | #endif | |
7446 | } | |
7447 | ||
269bbe5f | 7448 | static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node) |
65e19f54 | 7449 | { |
269bbe5f BM |
7450 | struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data; |
7451 | struct io_ring_ctx *ctx = rsrc_data->ctx; | |
7452 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7453 | |
269bbe5f BM |
7454 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7455 | list_del(&prsrc->list); | |
50238531 | 7456 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7457 | kfree(prsrc); |
65e19f54 | 7458 | } |
05589553 | 7459 | |
05589553 XW |
7460 | percpu_ref_exit(&ref_node->refs); |
7461 | kfree(ref_node); | |
269bbe5f | 7462 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7463 | } |
65e19f54 | 7464 | |
269bbe5f | 7465 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7466 | { |
7467 | struct io_ring_ctx *ctx; | |
7468 | struct llist_node *node; | |
7469 | ||
269bbe5f BM |
7470 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7471 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7472 | |
7473 | while (node) { | |
269bbe5f | 7474 | struct fixed_rsrc_ref_node *ref_node; |
4a38aed2 JA |
7475 | struct llist_node *next = node->next; |
7476 | ||
269bbe5f BM |
7477 | ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist); |
7478 | __io_rsrc_put_work(ref_node); | |
4a38aed2 JA |
7479 | node = next; |
7480 | } | |
7481 | } | |
7482 | ||
00835dce | 7483 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7484 | { |
269bbe5f BM |
7485 | struct fixed_rsrc_ref_node *ref_node; |
7486 | struct fixed_rsrc_data *data; | |
4a38aed2 | 7487 | struct io_ring_ctx *ctx; |
e297822b | 7488 | bool first_add = false; |
4a38aed2 | 7489 | int delay = HZ; |
65e19f54 | 7490 | |
269bbe5f BM |
7491 | ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs); |
7492 | data = ref_node->rsrc_data; | |
e297822b PB |
7493 | ctx = data->ctx; |
7494 | ||
2a63b2d9 | 7495 | io_rsrc_ref_lock(ctx); |
e297822b PB |
7496 | ref_node->done = true; |
7497 | ||
d67d2263 BM |
7498 | while (!list_empty(&ctx->rsrc_ref_list)) { |
7499 | ref_node = list_first_entry(&ctx->rsrc_ref_list, | |
269bbe5f | 7500 | struct fixed_rsrc_ref_node, node); |
e297822b PB |
7501 | /* recycle ref nodes in order */ |
7502 | if (!ref_node->done) | |
7503 | break; | |
7504 | list_del(&ref_node->node); | |
269bbe5f | 7505 | first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist); |
e297822b | 7506 | } |
2a63b2d9 | 7507 | io_rsrc_ref_unlock(ctx); |
05589553 | 7508 | |
e297822b | 7509 | if (percpu_ref_is_dying(&data->refs)) |
4a38aed2 | 7510 | delay = 0; |
05589553 | 7511 | |
4a38aed2 | 7512 | if (!delay) |
269bbe5f | 7513 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0); |
4a38aed2 | 7514 | else if (first_add) |
269bbe5f | 7515 | queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay); |
05f3fb3c | 7516 | } |
65e19f54 | 7517 | |
6802535d | 7518 | static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node( |
05589553 | 7519 | struct io_ring_ctx *ctx) |
05f3fb3c | 7520 | { |
269bbe5f | 7521 | struct fixed_rsrc_ref_node *ref_node; |
05f3fb3c | 7522 | |
05589553 XW |
7523 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7524 | if (!ref_node) | |
3e2224c5 | 7525 | return NULL; |
05f3fb3c | 7526 | |
00835dce | 7527 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7528 | 0, GFP_KERNEL)) { |
7529 | kfree(ref_node); | |
3e2224c5 | 7530 | return NULL; |
05589553 XW |
7531 | } |
7532 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7533 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7534 | ref_node->done = false; |
05589553 | 7535 | return ref_node; |
05589553 XW |
7536 | } |
7537 | ||
bc9744cd PB |
7538 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
7539 | struct fixed_rsrc_ref_node *ref_node) | |
6802535d | 7540 | { |
269bbe5f | 7541 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7542 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7543 | } |
7544 | ||
269bbe5f | 7545 | static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node) |
05589553 XW |
7546 | { |
7547 | percpu_ref_exit(&ref_node->refs); | |
7548 | kfree(ref_node); | |
65e19f54 JA |
7549 | } |
7550 | ||
ea64ec02 | 7551 | |
6b06314c JA |
7552 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7553 | unsigned nr_args) | |
7554 | { | |
7555 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7556 | unsigned nr_tables, i; |
05f3fb3c | 7557 | struct file *file; |
600cf3f8 | 7558 | int fd, ret = -ENOMEM; |
269bbe5f BM |
7559 | struct fixed_rsrc_ref_node *ref_node; |
7560 | struct fixed_rsrc_data *file_data; | |
6b06314c | 7561 | |
05f3fb3c | 7562 | if (ctx->file_data) |
6b06314c JA |
7563 | return -EBUSY; |
7564 | if (!nr_args) | |
7565 | return -EINVAL; | |
7566 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7567 | return -EMFILE; | |
7568 | ||
1ad555c6 | 7569 | file_data = alloc_fixed_rsrc_data(ctx); |
5398ae69 | 7570 | if (!file_data) |
05f3fb3c | 7571 | return -ENOMEM; |
13770a71 | 7572 | ctx->file_data = file_data; |
05f3fb3c | 7573 | |
65e19f54 | 7574 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7575 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7576 | GFP_KERNEL); |
600cf3f8 PB |
7577 | if (!file_data->table) |
7578 | goto out_free; | |
05f3fb3c | 7579 | |
600cf3f8 | 7580 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7581 | goto out_free; |
65e19f54 | 7582 | |
08a45173 | 7583 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
7b29f92d JA |
7584 | unsigned long file_ptr; |
7585 | ||
600cf3f8 PB |
7586 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7587 | ret = -EFAULT; | |
7588 | goto out_fput; | |
7589 | } | |
08a45173 | 7590 | /* allow sparse sets */ |
600cf3f8 | 7591 | if (fd == -1) |
08a45173 | 7592 | continue; |
6b06314c | 7593 | |
05f3fb3c | 7594 | file = fget(fd); |
6b06314c | 7595 | ret = -EBADF; |
05f3fb3c | 7596 | if (!file) |
600cf3f8 | 7597 | goto out_fput; |
05f3fb3c | 7598 | |
6b06314c JA |
7599 | /* |
7600 | * Don't allow io_uring instances to be registered. If UNIX | |
7601 | * isn't enabled, then this causes a reference cycle and this | |
7602 | * instance can never get freed. If UNIX is enabled we'll | |
7603 | * handle it just fine, but there's still no point in allowing | |
7604 | * a ring fd as it doesn't support regular read/write anyway. | |
7605 | */ | |
05f3fb3c JA |
7606 | if (file->f_op == &io_uring_fops) { |
7607 | fput(file); | |
600cf3f8 | 7608 | goto out_fput; |
6b06314c | 7609 | } |
7b29f92d JA |
7610 | file_ptr = (unsigned long) file; |
7611 | if (__io_file_supports_async(file, READ)) | |
7612 | file_ptr |= FFS_ASYNC_READ; | |
7613 | if (__io_file_supports_async(file, WRITE)) | |
7614 | file_ptr |= FFS_ASYNC_WRITE; | |
7615 | if (S_ISREG(file_inode(file)->i_mode)) | |
7616 | file_ptr |= FFS_ISREG; | |
7617 | *io_fixed_file_slot(file_data, i) = (struct file *) file_ptr; | |
6b06314c JA |
7618 | } |
7619 | ||
6b06314c | 7620 | ret = io_sqe_files_scm(ctx); |
05589553 | 7621 | if (ret) { |
6b06314c | 7622 | io_sqe_files_unregister(ctx); |
05589553 XW |
7623 | return ret; |
7624 | } | |
6b06314c | 7625 | |
bc9744cd | 7626 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 | 7627 | if (!ref_node) { |
05589553 | 7628 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7629 | return -ENOMEM; |
05589553 | 7630 | } |
bc9744cd | 7631 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7632 | |
d67d2263 | 7633 | io_sqe_rsrc_set_node(ctx, file_data, ref_node); |
6b06314c | 7634 | return ret; |
600cf3f8 PB |
7635 | out_fput: |
7636 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7637 | file = io_file_from_index(ctx, i); | |
7638 | if (file) | |
7639 | fput(file); | |
7640 | } | |
7641 | for (i = 0; i < nr_tables; i++) | |
7642 | kfree(file_data->table[i].files); | |
7643 | ctx->nr_user_files = 0; | |
600cf3f8 | 7644 | out_free: |
1ad555c6 | 7645 | free_fixed_rsrc_data(ctx->file_data); |
55cbc256 | 7646 | ctx->file_data = NULL; |
6b06314c JA |
7647 | return ret; |
7648 | } | |
7649 | ||
c3a31e60 JA |
7650 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7651 | int index) | |
7652 | { | |
7653 | #if defined(CONFIG_UNIX) | |
7654 | struct sock *sock = ctx->ring_sock->sk; | |
7655 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7656 | struct sk_buff *skb; | |
7657 | ||
7658 | /* | |
7659 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7660 | * file set. If there's no room, fall back to allocating a new skb | |
7661 | * and filling it in. | |
7662 | */ | |
7663 | spin_lock_irq(&head->lock); | |
7664 | skb = skb_peek(head); | |
7665 | if (skb) { | |
7666 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7667 | ||
7668 | if (fpl->count < SCM_MAX_FD) { | |
7669 | __skb_unlink(skb, head); | |
7670 | spin_unlock_irq(&head->lock); | |
7671 | fpl->fp[fpl->count] = get_file(file); | |
7672 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7673 | fpl->count++; | |
7674 | spin_lock_irq(&head->lock); | |
7675 | __skb_queue_head(head, skb); | |
7676 | } else { | |
7677 | skb = NULL; | |
7678 | } | |
7679 | } | |
7680 | spin_unlock_irq(&head->lock); | |
7681 | ||
7682 | if (skb) { | |
7683 | fput(file); | |
7684 | return 0; | |
7685 | } | |
7686 | ||
7687 | return __io_sqe_files_scm(ctx, 1, index); | |
7688 | #else | |
7689 | return 0; | |
7690 | #endif | |
7691 | } | |
7692 | ||
50238531 | 7693 | static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc) |
05f3fb3c | 7694 | { |
269bbe5f BM |
7695 | struct io_rsrc_put *prsrc; |
7696 | struct fixed_rsrc_ref_node *ref_node = data->node; | |
05f3fb3c | 7697 | |
269bbe5f BM |
7698 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7699 | if (!prsrc) | |
a5318d3c | 7700 | return -ENOMEM; |
05f3fb3c | 7701 | |
50238531 | 7702 | prsrc->rsrc = rsrc; |
269bbe5f | 7703 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7704 | |
a5318d3c | 7705 | return 0; |
05f3fb3c JA |
7706 | } |
7707 | ||
269bbe5f BM |
7708 | static inline int io_queue_file_removal(struct fixed_rsrc_data *data, |
7709 | struct file *file) | |
7710 | { | |
50238531 | 7711 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7712 | } |
7713 | ||
05f3fb3c | 7714 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7715 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7716 | unsigned nr_args) |
7717 | { | |
269bbe5f BM |
7718 | struct fixed_rsrc_data *data = ctx->file_data; |
7719 | struct fixed_rsrc_ref_node *ref_node; | |
ea64ec02 | 7720 | struct file *file, **file_slot; |
c3a31e60 JA |
7721 | __s32 __user *fds; |
7722 | int fd, i, err; | |
7723 | __u32 done; | |
05589553 | 7724 | bool needs_switch = false; |
c3a31e60 | 7725 | |
05f3fb3c | 7726 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7727 | return -EOVERFLOW; |
7728 | if (done > ctx->nr_user_files) | |
7729 | return -EINVAL; | |
7730 | ||
bc9744cd | 7731 | ref_node = alloc_fixed_rsrc_ref_node(ctx); |
3e2224c5 MWO |
7732 | if (!ref_node) |
7733 | return -ENOMEM; | |
bc9744cd | 7734 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7735 | |
269bbe5f | 7736 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7737 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7738 | err = 0; |
7739 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7740 | err = -EFAULT; | |
7741 | break; | |
7742 | } | |
4e0377a1 | 7743 | if (fd == IORING_REGISTER_FILES_SKIP) |
7744 | continue; | |
7745 | ||
67973b93 | 7746 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7747 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7748 | ||
7749 | if (*file_slot) { | |
7b29f92d JA |
7750 | file = (struct file *) ((unsigned long) *file_slot & FFS_MASK); |
7751 | err = io_queue_file_removal(data, file); | |
a5318d3c HD |
7752 | if (err) |
7753 | break; | |
ea64ec02 | 7754 | *file_slot = NULL; |
05589553 | 7755 | needs_switch = true; |
c3a31e60 JA |
7756 | } |
7757 | if (fd != -1) { | |
c3a31e60 JA |
7758 | file = fget(fd); |
7759 | if (!file) { | |
7760 | err = -EBADF; | |
7761 | break; | |
7762 | } | |
7763 | /* | |
7764 | * Don't allow io_uring instances to be registered. If | |
7765 | * UNIX isn't enabled, then this causes a reference | |
7766 | * cycle and this instance can never get freed. If UNIX | |
7767 | * is enabled we'll handle it just fine, but there's | |
7768 | * still no point in allowing a ring fd as it doesn't | |
7769 | * support regular read/write anyway. | |
7770 | */ | |
7771 | if (file->f_op == &io_uring_fops) { | |
7772 | fput(file); | |
7773 | err = -EBADF; | |
7774 | break; | |
7775 | } | |
e68a3ff8 | 7776 | *file_slot = file; |
c3a31e60 | 7777 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7778 | if (err) { |
e68a3ff8 | 7779 | *file_slot = NULL; |
f3bd9dae | 7780 | fput(file); |
c3a31e60 | 7781 | break; |
f3bd9dae | 7782 | } |
c3a31e60 | 7783 | } |
05f3fb3c JA |
7784 | } |
7785 | ||
05589553 | 7786 | if (needs_switch) { |
b2e96852 | 7787 | percpu_ref_kill(&data->node->refs); |
d67d2263 | 7788 | io_sqe_rsrc_set_node(ctx, data, ref_node); |
05589553 | 7789 | } else |
269bbe5f | 7790 | destroy_fixed_rsrc_ref_node(ref_node); |
c3a31e60 JA |
7791 | |
7792 | return done ? done : err; | |
7793 | } | |
05589553 | 7794 | |
05f3fb3c JA |
7795 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7796 | unsigned nr_args) | |
7797 | { | |
269bbe5f | 7798 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7799 | |
7800 | if (!ctx->file_data) | |
7801 | return -ENXIO; | |
7802 | if (!nr_args) | |
7803 | return -EINVAL; | |
7804 | if (copy_from_user(&up, arg, sizeof(up))) | |
7805 | return -EFAULT; | |
7806 | if (up.resv) | |
7807 | return -EINVAL; | |
7808 | ||
7809 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7810 | } | |
c3a31e60 | 7811 | |
5280f7e5 | 7812 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7813 | { |
7814 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7815 | ||
5280f7e5 PB |
7816 | req = io_put_req_find_next(req); |
7817 | return req ? &req->work : NULL; | |
7d723065 JA |
7818 | } |
7819 | ||
5aa75ed5 | 7820 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx) |
24369c2e | 7821 | { |
e941894e | 7822 | struct io_wq_hash *hash; |
24369c2e | 7823 | struct io_wq_data data; |
24369c2e | 7824 | unsigned int concurrency; |
24369c2e | 7825 | |
e941894e JA |
7826 | hash = ctx->hash_map; |
7827 | if (!hash) { | |
7828 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7829 | if (!hash) | |
7830 | return ERR_PTR(-ENOMEM); | |
7831 | refcount_set(&hash->refs, 1); | |
7832 | init_waitqueue_head(&hash->wait); | |
7833 | ctx->hash_map = hash; | |
24369c2e PB |
7834 | } |
7835 | ||
e941894e | 7836 | data.hash = hash; |
e9fd9396 | 7837 | data.free_work = io_free_work; |
f5fa38c5 | 7838 | data.do_work = io_wq_submit_work; |
24369c2e | 7839 | |
d25e3a3d JA |
7840 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7841 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7842 | |
5aa75ed5 | 7843 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7844 | } |
7845 | ||
5aa75ed5 JA |
7846 | static int io_uring_alloc_task_context(struct task_struct *task, |
7847 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7848 | { |
7849 | struct io_uring_task *tctx; | |
d8a6df10 | 7850 | int ret; |
0f212204 JA |
7851 | |
7852 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7853 | if (unlikely(!tctx)) | |
7854 | return -ENOMEM; | |
7855 | ||
d8a6df10 JA |
7856 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7857 | if (unlikely(ret)) { | |
7858 | kfree(tctx); | |
7859 | return ret; | |
7860 | } | |
7861 | ||
5aa75ed5 JA |
7862 | tctx->io_wq = io_init_wq_offload(ctx); |
7863 | if (IS_ERR(tctx->io_wq)) { | |
7864 | ret = PTR_ERR(tctx->io_wq); | |
7865 | percpu_counter_destroy(&tctx->inflight); | |
7866 | kfree(tctx); | |
7867 | return ret; | |
7868 | } | |
7869 | ||
0f212204 JA |
7870 | xa_init(&tctx->xa); |
7871 | init_waitqueue_head(&tctx->wait); | |
7872 | tctx->last = NULL; | |
fdaf083c | 7873 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7874 | task->io_uring = tctx; |
7cbf1722 JA |
7875 | spin_lock_init(&tctx->task_lock); |
7876 | INIT_WQ_LIST(&tctx->task_list); | |
7877 | tctx->task_state = 0; | |
7878 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7879 | return 0; |
7880 | } | |
7881 | ||
7882 | void __io_uring_free(struct task_struct *tsk) | |
7883 | { | |
7884 | struct io_uring_task *tctx = tsk->io_uring; | |
7885 | ||
7886 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7887 | WARN_ON_ONCE(tctx->io_wq); |
7888 | ||
d8a6df10 | 7889 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7890 | kfree(tctx); |
7891 | tsk->io_uring = NULL; | |
7892 | } | |
7893 | ||
7e84e1c7 SG |
7894 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7895 | struct io_uring_params *p) | |
2b188cc1 JA |
7896 | { |
7897 | int ret; | |
7898 | ||
d25e3a3d JA |
7899 | /* Retain compatibility with failing for an invalid attach attempt */ |
7900 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7901 | IORING_SETUP_ATTACH_WQ) { | |
7902 | struct fd f; | |
7903 | ||
7904 | f = fdget(p->wq_fd); | |
7905 | if (!f.file) | |
7906 | return -ENXIO; | |
7907 | if (f.file->f_op != &io_uring_fops) { | |
7908 | fdput(f); | |
7909 | return -EINVAL; | |
7910 | } | |
7911 | fdput(f); | |
7912 | } | |
6c271ce2 | 7913 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7914 | struct task_struct *tsk; |
534ca6d6 | 7915 | struct io_sq_data *sqd; |
26984fbf | 7916 | bool attached; |
534ca6d6 | 7917 | |
3ec482d1 | 7918 | ret = -EPERM; |
ce59fc69 | 7919 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) |
3ec482d1 JA |
7920 | goto err; |
7921 | ||
26984fbf | 7922 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7923 | if (IS_ERR(sqd)) { |
7924 | ret = PTR_ERR(sqd); | |
7925 | goto err; | |
7926 | } | |
69fb2131 | 7927 | |
7c30f36a | 7928 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7929 | ctx->sq_data = sqd; |
917257da JA |
7930 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7931 | if (!ctx->sq_thread_idle) | |
7932 | ctx->sq_thread_idle = HZ; | |
7933 | ||
26984fbf | 7934 | ret = 0; |
78d7f6ba | 7935 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7936 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7937 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7938 | /* don't attach to a dying SQPOLL thread, would be racy */ |
de75a3d3 | 7939 | if (attached && !sqd->thread) |
26984fbf | 7940 | ret = -ENXIO; |
78d7f6ba PB |
7941 | io_sq_thread_unpark(sqd); |
7942 | ||
de75a3d3 PB |
7943 | if (ret < 0) |
7944 | goto err; | |
7945 | if (attached) | |
5aa75ed5 | 7946 | return 0; |
aa06165d | 7947 | |
6c271ce2 | 7948 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7949 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7950 | |
917257da | 7951 | ret = -EINVAL; |
44a9bd18 | 7952 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7953 | goto err_sqpoll; |
7889f44d | 7954 | if (!cpu_online(cpu)) |
e8f98f24 | 7955 | goto err_sqpoll; |
917257da | 7956 | |
37d1e2e3 | 7957 | sqd->sq_cpu = cpu; |
6c271ce2 | 7958 | } else { |
37d1e2e3 | 7959 | sqd->sq_cpu = -1; |
6c271ce2 | 7960 | } |
37d1e2e3 JA |
7961 | |
7962 | sqd->task_pid = current->pid; | |
5c2469e0 | 7963 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7964 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7965 | if (IS_ERR(tsk)) { | |
7966 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7967 | goto err_sqpoll; |
6c271ce2 | 7968 | } |
97a73a0f | 7969 | |
46fe18b1 | 7970 | sqd->thread = tsk; |
97a73a0f | 7971 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7972 | wake_up_new_task(tsk); |
0f212204 JA |
7973 | if (ret) |
7974 | goto err; | |
6c271ce2 JA |
7975 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7976 | /* Can't have SQ_AFF without SQPOLL */ | |
7977 | ret = -EINVAL; | |
7978 | goto err; | |
7979 | } | |
7980 | ||
2b188cc1 JA |
7981 | return 0; |
7982 | err: | |
37d1e2e3 | 7983 | io_sq_thread_finish(ctx); |
2b188cc1 | 7984 | return ret; |
e8f98f24 JA |
7985 | err_sqpoll: |
7986 | complete(&ctx->sq_data->exited); | |
7987 | goto err; | |
2b188cc1 JA |
7988 | } |
7989 | ||
a087e2b5 BM |
7990 | static inline void __io_unaccount_mem(struct user_struct *user, |
7991 | unsigned long nr_pages) | |
2b188cc1 JA |
7992 | { |
7993 | atomic_long_sub(nr_pages, &user->locked_vm); | |
7994 | } | |
7995 | ||
a087e2b5 BM |
7996 | static inline int __io_account_mem(struct user_struct *user, |
7997 | unsigned long nr_pages) | |
2b188cc1 JA |
7998 | { |
7999 | unsigned long page_limit, cur_pages, new_pages; | |
8000 | ||
8001 | /* Don't allow more pages than we can safely lock */ | |
8002 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8003 | ||
8004 | do { | |
8005 | cur_pages = atomic_long_read(&user->locked_vm); | |
8006 | new_pages = cur_pages + nr_pages; | |
8007 | if (new_pages > page_limit) | |
8008 | return -ENOMEM; | |
8009 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8010 | new_pages) != cur_pages); | |
8011 | ||
8012 | return 0; | |
8013 | } | |
8014 | ||
26bfa89e | 8015 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8016 | { |
62e398be | 8017 | if (ctx->user) |
a087e2b5 | 8018 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8019 | |
26bfa89e JA |
8020 | if (ctx->mm_account) |
8021 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8022 | } |
8023 | ||
26bfa89e | 8024 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8025 | { |
30975825 BM |
8026 | int ret; |
8027 | ||
62e398be | 8028 | if (ctx->user) { |
30975825 BM |
8029 | ret = __io_account_mem(ctx->user, nr_pages); |
8030 | if (ret) | |
8031 | return ret; | |
8032 | } | |
8033 | ||
26bfa89e JA |
8034 | if (ctx->mm_account) |
8035 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8036 | |
8037 | return 0; | |
8038 | } | |
8039 | ||
2b188cc1 JA |
8040 | static void io_mem_free(void *ptr) |
8041 | { | |
52e04ef4 MR |
8042 | struct page *page; |
8043 | ||
8044 | if (!ptr) | |
8045 | return; | |
2b188cc1 | 8046 | |
52e04ef4 | 8047 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8048 | if (put_page_testzero(page)) |
8049 | free_compound_page(page); | |
8050 | } | |
8051 | ||
8052 | static void *io_mem_alloc(size_t size) | |
8053 | { | |
8054 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8055 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8056 | |
8057 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8058 | } | |
8059 | ||
75b28aff HV |
8060 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8061 | size_t *sq_offset) | |
8062 | { | |
8063 | struct io_rings *rings; | |
8064 | size_t off, sq_array_size; | |
8065 | ||
8066 | off = struct_size(rings, cqes, cq_entries); | |
8067 | if (off == SIZE_MAX) | |
8068 | return SIZE_MAX; | |
8069 | ||
8070 | #ifdef CONFIG_SMP | |
8071 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8072 | if (off == 0) | |
8073 | return SIZE_MAX; | |
8074 | #endif | |
8075 | ||
b36200f5 DV |
8076 | if (sq_offset) |
8077 | *sq_offset = off; | |
8078 | ||
75b28aff HV |
8079 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8080 | if (sq_array_size == SIZE_MAX) | |
8081 | return SIZE_MAX; | |
8082 | ||
8083 | if (check_add_overflow(off, sq_array_size, &off)) | |
8084 | return SIZE_MAX; | |
8085 | ||
75b28aff HV |
8086 | return off; |
8087 | } | |
8088 | ||
0a96bbe4 | 8089 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8090 | { |
8091 | int i, j; | |
8092 | ||
8093 | if (!ctx->user_bufs) | |
8094 | return -ENXIO; | |
8095 | ||
8096 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8097 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8098 | ||
8099 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8100 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8101 | |
de293938 | 8102 | if (imu->acct_pages) |
26bfa89e | 8103 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8104 | kvfree(imu->bvec); |
edafccee JA |
8105 | imu->nr_bvecs = 0; |
8106 | } | |
8107 | ||
8108 | kfree(ctx->user_bufs); | |
8109 | ctx->user_bufs = NULL; | |
8110 | ctx->nr_user_bufs = 0; | |
8111 | return 0; | |
8112 | } | |
8113 | ||
8114 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8115 | void __user *arg, unsigned index) | |
8116 | { | |
8117 | struct iovec __user *src; | |
8118 | ||
8119 | #ifdef CONFIG_COMPAT | |
8120 | if (ctx->compat) { | |
8121 | struct compat_iovec __user *ciovs; | |
8122 | struct compat_iovec ciov; | |
8123 | ||
8124 | ciovs = (struct compat_iovec __user *) arg; | |
8125 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8126 | return -EFAULT; | |
8127 | ||
d55e5f5b | 8128 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8129 | dst->iov_len = ciov.iov_len; |
8130 | return 0; | |
8131 | } | |
8132 | #endif | |
8133 | src = (struct iovec __user *) arg; | |
8134 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8135 | return -EFAULT; | |
8136 | return 0; | |
8137 | } | |
8138 | ||
de293938 JA |
8139 | /* |
8140 | * Not super efficient, but this is just a registration time. And we do cache | |
8141 | * the last compound head, so generally we'll only do a full search if we don't | |
8142 | * match that one. | |
8143 | * | |
8144 | * We check if the given compound head page has already been accounted, to | |
8145 | * avoid double accounting it. This allows us to account the full size of the | |
8146 | * page, not just the constituent pages of a huge page. | |
8147 | */ | |
8148 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8149 | int nr_pages, struct page *hpage) | |
8150 | { | |
8151 | int i, j; | |
8152 | ||
8153 | /* check current page array */ | |
8154 | for (i = 0; i < nr_pages; i++) { | |
8155 | if (!PageCompound(pages[i])) | |
8156 | continue; | |
8157 | if (compound_head(pages[i]) == hpage) | |
8158 | return true; | |
8159 | } | |
8160 | ||
8161 | /* check previously registered pages */ | |
8162 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8163 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8164 | ||
8165 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8166 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8167 | continue; | |
8168 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8169 | return true; | |
8170 | } | |
8171 | } | |
8172 | ||
8173 | return false; | |
8174 | } | |
8175 | ||
8176 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8177 | int nr_pages, struct io_mapped_ubuf *imu, | |
8178 | struct page **last_hpage) | |
8179 | { | |
8180 | int i, ret; | |
8181 | ||
8182 | for (i = 0; i < nr_pages; i++) { | |
8183 | if (!PageCompound(pages[i])) { | |
8184 | imu->acct_pages++; | |
8185 | } else { | |
8186 | struct page *hpage; | |
8187 | ||
8188 | hpage = compound_head(pages[i]); | |
8189 | if (hpage == *last_hpage) | |
8190 | continue; | |
8191 | *last_hpage = hpage; | |
8192 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8193 | continue; | |
8194 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8195 | } | |
8196 | } | |
8197 | ||
8198 | if (!imu->acct_pages) | |
8199 | return 0; | |
8200 | ||
26bfa89e | 8201 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8202 | if (ret) |
8203 | imu->acct_pages = 0; | |
8204 | return ret; | |
8205 | } | |
8206 | ||
0a96bbe4 BM |
8207 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8208 | struct io_mapped_ubuf *imu, | |
8209 | struct page **last_hpage) | |
edafccee JA |
8210 | { |
8211 | struct vm_area_struct **vmas = NULL; | |
8212 | struct page **pages = NULL; | |
0a96bbe4 BM |
8213 | unsigned long off, start, end, ubuf; |
8214 | size_t size; | |
8215 | int ret, pret, nr_pages, i; | |
8216 | ||
8217 | ubuf = (unsigned long) iov->iov_base; | |
8218 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8219 | start = ubuf >> PAGE_SHIFT; | |
8220 | nr_pages = end - start; | |
8221 | ||
8222 | ret = -ENOMEM; | |
8223 | ||
8224 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8225 | if (!pages) | |
8226 | goto done; | |
8227 | ||
8228 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8229 | GFP_KERNEL); | |
8230 | if (!vmas) | |
8231 | goto done; | |
edafccee | 8232 | |
0a96bbe4 BM |
8233 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8234 | GFP_KERNEL); | |
8235 | if (!imu->bvec) | |
8236 | goto done; | |
8237 | ||
8238 | ret = 0; | |
8239 | mmap_read_lock(current->mm); | |
8240 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8241 | pages, vmas); | |
8242 | if (pret == nr_pages) { | |
8243 | /* don't support file backed memory */ | |
8244 | for (i = 0; i < nr_pages; i++) { | |
8245 | struct vm_area_struct *vma = vmas[i]; | |
8246 | ||
8247 | if (vma->vm_file && | |
8248 | !is_file_hugepages(vma->vm_file)) { | |
8249 | ret = -EOPNOTSUPP; | |
8250 | break; | |
8251 | } | |
8252 | } | |
8253 | } else { | |
8254 | ret = pret < 0 ? pret : -EFAULT; | |
8255 | } | |
8256 | mmap_read_unlock(current->mm); | |
8257 | if (ret) { | |
8258 | /* | |
8259 | * if we did partial map, or found file backed vmas, | |
8260 | * release any pages we did get | |
8261 | */ | |
8262 | if (pret > 0) | |
8263 | unpin_user_pages(pages, pret); | |
8264 | kvfree(imu->bvec); | |
8265 | goto done; | |
8266 | } | |
8267 | ||
8268 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8269 | if (ret) { | |
8270 | unpin_user_pages(pages, pret); | |
8271 | kvfree(imu->bvec); | |
8272 | goto done; | |
8273 | } | |
8274 | ||
8275 | off = ubuf & ~PAGE_MASK; | |
8276 | size = iov->iov_len; | |
8277 | for (i = 0; i < nr_pages; i++) { | |
8278 | size_t vec_len; | |
8279 | ||
8280 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8281 | imu->bvec[i].bv_page = pages[i]; | |
8282 | imu->bvec[i].bv_len = vec_len; | |
8283 | imu->bvec[i].bv_offset = off; | |
8284 | off = 0; | |
8285 | size -= vec_len; | |
8286 | } | |
8287 | /* store original address for later verification */ | |
8288 | imu->ubuf = ubuf; | |
8289 | imu->len = iov->iov_len; | |
8290 | imu->nr_bvecs = nr_pages; | |
8291 | ret = 0; | |
8292 | done: | |
8293 | kvfree(pages); | |
8294 | kvfree(vmas); | |
8295 | return ret; | |
8296 | } | |
8297 | ||
2b358604 | 8298 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8299 | { |
edafccee JA |
8300 | if (ctx->user_bufs) |
8301 | return -EBUSY; | |
8302 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8303 | return -EINVAL; | |
8304 | ||
8305 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8306 | GFP_KERNEL); | |
8307 | if (!ctx->user_bufs) | |
8308 | return -ENOMEM; | |
8309 | ||
2b358604 BM |
8310 | return 0; |
8311 | } | |
edafccee | 8312 | |
2b358604 BM |
8313 | static int io_buffer_validate(struct iovec *iov) |
8314 | { | |
8315 | /* | |
8316 | * Don't impose further limits on the size and buffer | |
8317 | * constraints here, we'll -EINVAL later when IO is | |
8318 | * submitted if they are wrong. | |
8319 | */ | |
8320 | if (!iov->iov_base || !iov->iov_len) | |
8321 | return -EFAULT; | |
edafccee | 8322 | |
2b358604 BM |
8323 | /* arbitrary limit, but we need something */ |
8324 | if (iov->iov_len > SZ_1G) | |
8325 | return -EFAULT; | |
edafccee | 8326 | |
2b358604 BM |
8327 | return 0; |
8328 | } | |
edafccee | 8329 | |
2b358604 BM |
8330 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8331 | unsigned int nr_args) | |
8332 | { | |
8333 | int i, ret; | |
8334 | struct iovec iov; | |
8335 | struct page *last_hpage = NULL; | |
edafccee | 8336 | |
2b358604 BM |
8337 | ret = io_buffers_map_alloc(ctx, nr_args); |
8338 | if (ret) | |
8339 | return ret; | |
edafccee | 8340 | |
edafccee JA |
8341 | for (i = 0; i < nr_args; i++) { |
8342 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8343 | |
edafccee JA |
8344 | ret = io_copy_iov(ctx, &iov, arg, i); |
8345 | if (ret) | |
0a96bbe4 | 8346 | break; |
de293938 | 8347 | |
2b358604 BM |
8348 | ret = io_buffer_validate(&iov); |
8349 | if (ret) | |
0a96bbe4 | 8350 | break; |
edafccee | 8351 | |
0a96bbe4 BM |
8352 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8353 | if (ret) | |
8354 | break; | |
edafccee JA |
8355 | |
8356 | ctx->nr_user_bufs++; | |
8357 | } | |
0a96bbe4 BM |
8358 | |
8359 | if (ret) | |
8360 | io_sqe_buffers_unregister(ctx); | |
8361 | ||
edafccee JA |
8362 | return ret; |
8363 | } | |
8364 | ||
9b402849 JA |
8365 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8366 | { | |
8367 | __s32 __user *fds = arg; | |
8368 | int fd; | |
8369 | ||
8370 | if (ctx->cq_ev_fd) | |
8371 | return -EBUSY; | |
8372 | ||
8373 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8374 | return -EFAULT; | |
8375 | ||
8376 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8377 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8378 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8379 | ctx->cq_ev_fd = NULL; | |
8380 | return ret; | |
8381 | } | |
8382 | ||
8383 | return 0; | |
8384 | } | |
8385 | ||
8386 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8387 | { | |
8388 | if (ctx->cq_ev_fd) { | |
8389 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8390 | ctx->cq_ev_fd = NULL; | |
8391 | return 0; | |
8392 | } | |
8393 | ||
8394 | return -ENXIO; | |
8395 | } | |
8396 | ||
5a2e745d JA |
8397 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8398 | { | |
9e15c3a0 JA |
8399 | struct io_buffer *buf; |
8400 | unsigned long index; | |
8401 | ||
8402 | xa_for_each(&ctx->io_buffers, index, buf) | |
8403 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8404 | } |
8405 | ||
68e68ee6 | 8406 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8407 | { |
68e68ee6 | 8408 | struct io_kiocb *req, *nxt; |
1b4c351f | 8409 | |
68e68ee6 JA |
8410 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8411 | if (tsk && req->task != tsk) | |
8412 | continue; | |
1b4c351f JA |
8413 | list_del(&req->compl.list); |
8414 | kmem_cache_free(req_cachep, req); | |
8415 | } | |
8416 | } | |
8417 | ||
4010fec4 | 8418 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8419 | { |
bf019da7 | 8420 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8421 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8422 | |
9a4fdbd8 JA |
8423 | mutex_lock(&ctx->uring_lock); |
8424 | ||
8e5c66c4 | 8425 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8426 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8427 | submit_state->reqs); | |
8e5c66c4 PB |
8428 | submit_state->free_reqs = 0; |
8429 | } | |
9a4fdbd8 JA |
8430 | |
8431 | spin_lock_irq(&ctx->completion_lock); | |
e5547d2c PB |
8432 | list_splice_init(&cs->locked_free_list, &cs->free_list); |
8433 | cs->locked_free_nr = 0; | |
9a4fdbd8 JA |
8434 | spin_unlock_irq(&ctx->completion_lock); |
8435 | ||
e5547d2c PB |
8436 | io_req_cache_free(&cs->free_list, NULL); |
8437 | ||
9a4fdbd8 JA |
8438 | mutex_unlock(&ctx->uring_lock); |
8439 | } | |
8440 | ||
2b188cc1 JA |
8441 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8442 | { | |
04fc6c80 PB |
8443 | /* |
8444 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8445 | * and they are free to do so while still holding uring_lock or |
8446 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8447 | */ |
8448 | mutex_lock(&ctx->uring_lock); | |
8449 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8450 | spin_lock_irq(&ctx->completion_lock); |
8451 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8452 | |
37d1e2e3 | 8453 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8454 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8455 | |
37d1e2e3 | 8456 | if (ctx->mm_account) { |
2aede0e4 JA |
8457 | mmdrop(ctx->mm_account); |
8458 | ctx->mm_account = NULL; | |
30975825 | 8459 | } |
def596e9 | 8460 | |
8bad28d8 | 8461 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8462 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8463 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8464 | io_eventfd_unregister(ctx); |
5a2e745d | 8465 | io_destroy_buffers(ctx); |
def596e9 | 8466 | |
2b188cc1 | 8467 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8468 | if (ctx->ring_sock) { |
8469 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8470 | sock_release(ctx->ring_sock); |
355e8d26 | 8471 | } |
2b188cc1 JA |
8472 | #endif |
8473 | ||
75b28aff | 8474 | io_mem_free(ctx->rings); |
2b188cc1 | 8475 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8476 | |
8477 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8478 | free_uid(ctx->user); |
4010fec4 | 8479 | io_req_caches_free(ctx); |
e941894e JA |
8480 | if (ctx->hash_map) |
8481 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8482 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8483 | kfree(ctx); |
8484 | } | |
8485 | ||
8486 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8487 | { | |
8488 | struct io_ring_ctx *ctx = file->private_data; | |
8489 | __poll_t mask = 0; | |
8490 | ||
8491 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8492 | /* |
8493 | * synchronizes with barrier from wq_has_sleeper call in | |
8494 | * io_commit_cqring | |
8495 | */ | |
2b188cc1 | 8496 | smp_rmb(); |
90554200 | 8497 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8498 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8499 | |
8500 | /* | |
8501 | * Don't flush cqring overflow list here, just do a simple check. | |
8502 | * Otherwise there could possible be ABBA deadlock: | |
8503 | * CPU0 CPU1 | |
8504 | * ---- ---- | |
8505 | * lock(&ctx->uring_lock); | |
8506 | * lock(&ep->mtx); | |
8507 | * lock(&ctx->uring_lock); | |
8508 | * lock(&ep->mtx); | |
8509 | * | |
8510 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8511 | * pushs them to do the flush. | |
8512 | */ | |
8513 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8514 | mask |= EPOLLIN | EPOLLRDNORM; |
8515 | ||
8516 | return mask; | |
8517 | } | |
8518 | ||
8519 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8520 | { | |
8521 | struct io_ring_ctx *ctx = file->private_data; | |
8522 | ||
8523 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8524 | } | |
8525 | ||
0bead8cd | 8526 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8527 | { |
4379bf8b | 8528 | const struct cred *creds; |
071698e1 | 8529 | |
61cf9370 | 8530 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8531 | if (creds) { |
8532 | put_cred(creds); | |
0bead8cd | 8533 | return 0; |
1e6fa521 | 8534 | } |
0bead8cd YD |
8535 | |
8536 | return -EINVAL; | |
8537 | } | |
8538 | ||
9b465711 | 8539 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8540 | { |
9b465711 | 8541 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8542 | } |
8543 | ||
d56d938b PB |
8544 | struct io_tctx_exit { |
8545 | struct callback_head task_work; | |
8546 | struct completion completion; | |
baf186c4 | 8547 | struct io_ring_ctx *ctx; |
d56d938b PB |
8548 | }; |
8549 | ||
8550 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8551 | { | |
8552 | struct io_uring_task *tctx = current->io_uring; | |
8553 | struct io_tctx_exit *work; | |
8554 | ||
8555 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8556 | /* | |
8557 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8558 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8559 | */ | |
8560 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8561 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8562 | complete(&work->completion); |
8563 | } | |
8564 | ||
85faa7b8 JA |
8565 | static void io_ring_exit_work(struct work_struct *work) |
8566 | { | |
d56d938b | 8567 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8568 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8569 | struct io_tctx_exit exit; |
8570 | struct io_tctx_node *node; | |
8571 | int ret; | |
85faa7b8 | 8572 | |
a185f1db PB |
8573 | /* prevent SQPOLL from submitting new requests */ |
8574 | if (ctx->sq_data) { | |
8575 | io_sq_thread_park(ctx->sq_data); | |
8576 | list_del_init(&ctx->sqd_list); | |
8577 | io_sqd_update_thread_idle(ctx->sq_data); | |
8578 | io_sq_thread_unpark(ctx->sq_data); | |
8579 | } | |
8580 | ||
56952e91 JA |
8581 | /* |
8582 | * If we're doing polled IO and end up having requests being | |
8583 | * submitted async (out-of-line), then completions can come in while | |
8584 | * we're waiting for refs to drop. We need to reap these manually, | |
8585 | * as nobody else will be looking for them. | |
8586 | */ | |
b2edc0a7 | 8587 | do { |
9936c7c2 | 8588 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8589 | |
8590 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8591 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8592 | |
8593 | mutex_lock(&ctx->uring_lock); | |
8594 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8595 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8596 | ||
d56d938b PB |
8597 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8598 | ctx_node); | |
baf186c4 | 8599 | exit.ctx = ctx; |
d56d938b PB |
8600 | init_completion(&exit.completion); |
8601 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8602 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8603 | if (WARN_ON_ONCE(ret)) | |
8604 | continue; | |
8605 | wake_up_process(node->task); | |
8606 | ||
8607 | mutex_unlock(&ctx->uring_lock); | |
8608 | wait_for_completion(&exit.completion); | |
8609 | cond_resched(); | |
8610 | mutex_lock(&ctx->uring_lock); | |
8611 | } | |
8612 | mutex_unlock(&ctx->uring_lock); | |
8613 | ||
85faa7b8 JA |
8614 | io_ring_ctx_free(ctx); |
8615 | } | |
8616 | ||
80c4cbdb PB |
8617 | /* Returns true if we found and killed one or more timeouts */ |
8618 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8619 | struct files_struct *files) | |
8620 | { | |
8621 | struct io_kiocb *req, *tmp; | |
8622 | int canceled = 0; | |
8623 | ||
8624 | spin_lock_irq(&ctx->completion_lock); | |
8625 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8626 | if (io_match_task(req, tsk, files)) { | |
8627 | io_kill_timeout(req, -ECANCELED); | |
8628 | canceled++; | |
8629 | } | |
8630 | } | |
51520426 PB |
8631 | if (canceled != 0) |
8632 | io_commit_cqring(ctx); | |
80c4cbdb | 8633 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8634 | if (canceled != 0) |
8635 | io_cqring_ev_posted(ctx); | |
8636 | return canceled != 0; | |
8637 | } | |
8638 | ||
2b188cc1 JA |
8639 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8640 | { | |
61cf9370 MWO |
8641 | unsigned long index; |
8642 | struct creds *creds; | |
8643 | ||
2b188cc1 JA |
8644 | mutex_lock(&ctx->uring_lock); |
8645 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8646 | /* if force is set, the ring is going away. always drop after that */ |
8647 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8648 | if (ctx->rings) |
6c503150 | 8649 | __io_cqring_overflow_flush(ctx, true, NULL, NULL); |
61cf9370 MWO |
8650 | xa_for_each(&ctx->personalities, index, creds) |
8651 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8652 | mutex_unlock(&ctx->uring_lock); |
8653 | ||
6b81928d PB |
8654 | io_kill_timeouts(ctx, NULL, NULL); |
8655 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8656 | |
15dff286 | 8657 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8658 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8659 | |
85faa7b8 | 8660 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8661 | /* |
8662 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8663 | * if we're exiting a ton of rings at the same time. It just adds | |
8664 | * noise and overhead, there's no discernable change in runtime | |
8665 | * over using system_wq. | |
8666 | */ | |
8667 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8668 | } |
8669 | ||
8670 | static int io_uring_release(struct inode *inode, struct file *file) | |
8671 | { | |
8672 | struct io_ring_ctx *ctx = file->private_data; | |
8673 | ||
8674 | file->private_data = NULL; | |
8675 | io_ring_ctx_wait_and_kill(ctx); | |
8676 | return 0; | |
8677 | } | |
8678 | ||
f6edbabb PB |
8679 | struct io_task_cancel { |
8680 | struct task_struct *task; | |
8681 | struct files_struct *files; | |
8682 | }; | |
f254ac04 | 8683 | |
f6edbabb | 8684 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8685 | { |
9a472ef7 | 8686 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8687 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8688 | bool ret; |
8689 | ||
f6edbabb | 8690 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8691 | unsigned long flags; |
8692 | struct io_ring_ctx *ctx = req->ctx; | |
8693 | ||
8694 | /* protect against races with linked timeouts */ | |
8695 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8696 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8697 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8698 | } else { | |
f6edbabb | 8699 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8700 | } |
8701 | return ret; | |
b711d4ea JA |
8702 | } |
8703 | ||
e1915f76 | 8704 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8705 | struct task_struct *task, |
b7ddce3c PB |
8706 | struct files_struct *files) |
8707 | { | |
e1915f76 | 8708 | struct io_defer_entry *de; |
b7ddce3c PB |
8709 | LIST_HEAD(list); |
8710 | ||
8711 | spin_lock_irq(&ctx->completion_lock); | |
8712 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8713 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8714 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8715 | break; | |
8716 | } | |
8717 | } | |
8718 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8719 | if (list_empty(&list)) |
8720 | return false; | |
b7ddce3c PB |
8721 | |
8722 | while (!list_empty(&list)) { | |
8723 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8724 | list_del_init(&de->list); | |
f41db273 | 8725 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8726 | kfree(de); |
8727 | } | |
e1915f76 | 8728 | return true; |
b7ddce3c PB |
8729 | } |
8730 | ||
1b00764f PB |
8731 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8732 | { | |
8733 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8734 | ||
8735 | return req->ctx == data; | |
8736 | } | |
8737 | ||
8738 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8739 | { | |
8740 | struct io_tctx_node *node; | |
8741 | enum io_wq_cancel cret; | |
8742 | bool ret = false; | |
8743 | ||
8744 | mutex_lock(&ctx->uring_lock); | |
8745 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8746 | struct io_uring_task *tctx = node->task->io_uring; | |
8747 | ||
8748 | /* | |
8749 | * io_wq will stay alive while we hold uring_lock, because it's | |
8750 | * killed after ctx nodes, which requires to take the lock. | |
8751 | */ | |
8752 | if (!tctx || !tctx->io_wq) | |
8753 | continue; | |
8754 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8755 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8756 | } | |
8757 | mutex_unlock(&ctx->uring_lock); | |
8758 | ||
8759 | return ret; | |
8760 | } | |
8761 | ||
9936c7c2 PB |
8762 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8763 | struct task_struct *task, | |
8764 | struct files_struct *files) | |
8765 | { | |
8766 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8767 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8768 | |
8769 | while (1) { | |
8770 | enum io_wq_cancel cret; | |
8771 | bool ret = false; | |
8772 | ||
1b00764f PB |
8773 | if (!task) { |
8774 | ret |= io_uring_try_cancel_iowq(ctx); | |
8775 | } else if (tctx && tctx->io_wq) { | |
8776 | /* | |
8777 | * Cancels requests of all rings, not only @ctx, but | |
8778 | * it's fine as the task is in exit/exec. | |
8779 | */ | |
5aa75ed5 | 8780 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8781 | &cancel, true); |
8782 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8783 | } | |
8784 | ||
8785 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8786 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8787 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8788 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8789 | io_iopoll_try_reap_events(ctx); | |
8790 | ret = true; | |
8791 | } | |
8792 | } | |
8793 | ||
e1915f76 | 8794 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8795 | ret |= io_poll_remove_all(ctx, task, files); |
8796 | ret |= io_kill_timeouts(ctx, task, files); | |
8797 | ret |= io_run_task_work(); | |
ba50a036 | 8798 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8799 | io_cqring_overflow_flush(ctx, true, task, files); |
8800 | if (!ret) | |
8801 | break; | |
8802 | cond_resched(); | |
8803 | } | |
8804 | } | |
8805 | ||
ca70f00b PB |
8806 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8807 | struct task_struct *task, | |
8808 | struct files_struct *files) | |
8809 | { | |
8810 | struct io_kiocb *req; | |
8811 | int cnt = 0; | |
8812 | ||
8813 | spin_lock_irq(&ctx->inflight_lock); | |
8814 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8815 | cnt += io_match_task(req, task, files); | |
8816 | spin_unlock_irq(&ctx->inflight_lock); | |
8817 | return cnt; | |
8818 | } | |
8819 | ||
b52fda00 | 8820 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8821 | struct task_struct *task, |
fcb323cc JA |
8822 | struct files_struct *files) |
8823 | { | |
fcb323cc | 8824 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8825 | DEFINE_WAIT(wait); |
ca70f00b | 8826 | int inflight; |
fcb323cc | 8827 | |
ca70f00b PB |
8828 | inflight = io_uring_count_inflight(ctx, task, files); |
8829 | if (!inflight) | |
fcb323cc | 8830 | break; |
f6edbabb | 8831 | |
9936c7c2 | 8832 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8833 | |
8834 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8835 | TASK_UNINTERRUPTIBLE); | |
8836 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8837 | schedule(); | |
c98de08c | 8838 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8839 | } |
0f212204 JA |
8840 | } |
8841 | ||
cf27f3b1 | 8842 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8843 | { |
236434c3 | 8844 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8845 | struct io_tctx_node *node; |
a528b04e | 8846 | int ret; |
236434c3 MWO |
8847 | |
8848 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8849 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8850 | if (unlikely(ret)) |
8851 | return ret; | |
236434c3 | 8852 | tctx = current->io_uring; |
0f212204 | 8853 | } |
cf27f3b1 PB |
8854 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8855 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8856 | if (!node) | |
8857 | return -ENOMEM; | |
8858 | node->ctx = ctx; | |
8859 | node->task = current; | |
13bf43f5 | 8860 | |
cf27f3b1 PB |
8861 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8862 | node, GFP_KERNEL)); | |
8863 | if (ret) { | |
8864 | kfree(node); | |
8865 | return ret; | |
0f212204 | 8866 | } |
cf27f3b1 PB |
8867 | |
8868 | mutex_lock(&ctx->uring_lock); | |
8869 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8870 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8871 | } |
cf27f3b1 | 8872 | tctx->last = ctx; |
0f212204 JA |
8873 | return 0; |
8874 | } | |
8875 | ||
cf27f3b1 PB |
8876 | /* |
8877 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8878 | */ | |
8879 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
8880 | { | |
8881 | struct io_uring_task *tctx = current->io_uring; | |
8882 | ||
8883 | if (likely(tctx && tctx->last == ctx)) | |
8884 | return 0; | |
8885 | return __io_uring_add_task_file(ctx); | |
8886 | } | |
8887 | ||
0f212204 JA |
8888 | /* |
8889 | * Remove this io_uring_file -> task mapping. | |
8890 | */ | |
2941267b | 8891 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8892 | { |
8893 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8894 | struct io_tctx_node *node; |
2941267b | 8895 | |
eebd2e37 PB |
8896 | if (!tctx) |
8897 | return; | |
13bf43f5 PB |
8898 | node = xa_erase(&tctx->xa, index); |
8899 | if (!node) | |
2941267b | 8900 | return; |
0f212204 | 8901 | |
13bf43f5 PB |
8902 | WARN_ON_ONCE(current != node->task); |
8903 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8904 | ||
8905 | mutex_lock(&node->ctx->uring_lock); | |
8906 | list_del(&node->ctx_node); | |
8907 | mutex_unlock(&node->ctx->uring_lock); | |
8908 | ||
baf186c4 | 8909 | if (tctx->last == node->ctx) |
0f212204 | 8910 | tctx->last = NULL; |
13bf43f5 | 8911 | kfree(node); |
0f212204 JA |
8912 | } |
8913 | ||
8452d4a6 | 8914 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8915 | { |
13bf43f5 | 8916 | struct io_tctx_node *node; |
de7f1d9e PB |
8917 | unsigned long index; |
8918 | ||
13bf43f5 | 8919 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8920 | io_uring_del_task_file(index); |
8452d4a6 PB |
8921 | if (tctx->io_wq) { |
8922 | io_wq_put_and_exit(tctx->io_wq); | |
8923 | tctx->io_wq = NULL; | |
8924 | } | |
de7f1d9e PB |
8925 | } |
8926 | ||
521d6a73 PB |
8927 | static s64 tctx_inflight(struct io_uring_task *tctx) |
8928 | { | |
8929 | return percpu_counter_sum(&tctx->inflight); | |
8930 | } | |
8931 | ||
8932 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8933 | { | |
8934 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8935 | struct io_ring_ctx *ctx = work->ctx; | |
8936 | struct io_sq_data *sqd = ctx->sq_data; | |
8937 | ||
8938 | if (sqd->thread) | |
8939 | io_uring_cancel_sqpoll(ctx); | |
8940 | complete(&work->completion); | |
8941 | } | |
8942 | ||
8943 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8944 | { | |
8945 | struct io_sq_data *sqd = ctx->sq_data; | |
8946 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8947 | struct task_struct *task; | |
8948 | ||
8949 | io_sq_thread_park(sqd); | |
8950 | list_del_init(&ctx->sqd_list); | |
8951 | io_sqd_update_thread_idle(sqd); | |
8952 | task = sqd->thread; | |
8953 | if (task) { | |
8954 | init_completion(&work.completion); | |
8955 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8956 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8957 | wake_up_process(task); |
8958 | } | |
8959 | io_sq_thread_unpark(sqd); | |
8960 | ||
8961 | if (task) | |
8962 | wait_for_completion(&work.completion); | |
8963 | } | |
8964 | ||
0f212204 JA |
8965 | void __io_uring_files_cancel(struct files_struct *files) |
8966 | { | |
8967 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8968 | struct io_tctx_node *node; |
ce765372 | 8969 | unsigned long index; |
0f212204 JA |
8970 | |
8971 | /* make sure overflow events are dropped */ | |
fdaf083c | 8972 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
8973 | xa_for_each(&tctx->xa, index, node) { |
8974 | struct io_ring_ctx *ctx = node->ctx; | |
8975 | ||
8976 | if (ctx->sq_data) { | |
8977 | io_sqpoll_cancel_sync(ctx); | |
8978 | continue; | |
8979 | } | |
8980 | io_uring_cancel_files(ctx, current, files); | |
8981 | if (!files) | |
8982 | io_uring_try_cancel_requests(ctx, current, NULL); | |
8983 | } | |
fdaf083c | 8984 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 8985 | |
8452d4a6 PB |
8986 | if (files) |
8987 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
8988 | } |
8989 | ||
521d6a73 | 8990 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8991 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8992 | { | |
37d1e2e3 | 8993 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8994 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8995 | s64 inflight; |
8996 | DEFINE_WAIT(wait); | |
fdaf083c | 8997 | |
521d6a73 PB |
8998 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8999 | ||
0e9ddb39 PB |
9000 | atomic_inc(&tctx->in_idle); |
9001 | do { | |
9002 | /* read completions before cancelations */ | |
9003 | inflight = tctx_inflight(tctx); | |
9004 | if (!inflight) | |
9005 | break; | |
521d6a73 | 9006 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 9007 | |
0e9ddb39 PB |
9008 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9009 | /* | |
9010 | * If we've seen completions, retry without waiting. This | |
9011 | * avoids a race where a completion comes in before we did | |
9012 | * prepare_to_wait(). | |
9013 | */ | |
9014 | if (inflight == tctx_inflight(tctx)) | |
9015 | schedule(); | |
9016 | finish_wait(&tctx->wait, &wait); | |
9017 | } while (1); | |
9018 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9019 | } |
9020 | ||
0f212204 JA |
9021 | /* |
9022 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9023 | * requests. | |
9024 | */ | |
9025 | void __io_uring_task_cancel(void) | |
9026 | { | |
9027 | struct io_uring_task *tctx = current->io_uring; | |
9028 | DEFINE_WAIT(wait); | |
d8a6df10 | 9029 | s64 inflight; |
0f212204 JA |
9030 | |
9031 | /* make sure overflow events are dropped */ | |
fdaf083c | 9032 | atomic_inc(&tctx->in_idle); |
5a978dcf PB |
9033 | __io_uring_files_cancel(NULL); |
9034 | ||
d8a6df10 | 9035 | do { |
0f212204 | 9036 | /* read completions before cancelations */ |
fdaf083c | 9037 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9038 | if (!inflight) |
9039 | break; | |
0f212204 JA |
9040 | __io_uring_files_cancel(NULL); |
9041 | ||
9042 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9043 | ||
9044 | /* | |
a1bb3cd5 PB |
9045 | * If we've seen completions, retry without waiting. This |
9046 | * avoids a race where a completion comes in before we did | |
9047 | * prepare_to_wait(). | |
0f212204 | 9048 | */ |
a1bb3cd5 PB |
9049 | if (inflight == tctx_inflight(tctx)) |
9050 | schedule(); | |
f57555ed | 9051 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9052 | } while (1); |
0f212204 | 9053 | |
fdaf083c | 9054 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9055 | |
8452d4a6 PB |
9056 | io_uring_clean_tctx(tctx); |
9057 | /* all current's requests should be gone, we can kill tctx */ | |
9058 | __io_uring_free(current); | |
44e728b8 PB |
9059 | } |
9060 | ||
6c5c240e RP |
9061 | static void *io_uring_validate_mmap_request(struct file *file, |
9062 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9063 | { |
2b188cc1 | 9064 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9065 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9066 | struct page *page; |
9067 | void *ptr; | |
9068 | ||
9069 | switch (offset) { | |
9070 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9071 | case IORING_OFF_CQ_RING: |
9072 | ptr = ctx->rings; | |
2b188cc1 JA |
9073 | break; |
9074 | case IORING_OFF_SQES: | |
9075 | ptr = ctx->sq_sqes; | |
9076 | break; | |
2b188cc1 | 9077 | default: |
6c5c240e | 9078 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9079 | } |
9080 | ||
9081 | page = virt_to_head_page(ptr); | |
a50b854e | 9082 | if (sz > page_size(page)) |
6c5c240e RP |
9083 | return ERR_PTR(-EINVAL); |
9084 | ||
9085 | return ptr; | |
9086 | } | |
9087 | ||
9088 | #ifdef CONFIG_MMU | |
9089 | ||
9090 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9091 | { | |
9092 | size_t sz = vma->vm_end - vma->vm_start; | |
9093 | unsigned long pfn; | |
9094 | void *ptr; | |
9095 | ||
9096 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9097 | if (IS_ERR(ptr)) | |
9098 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9099 | |
9100 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9101 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9102 | } | |
9103 | ||
6c5c240e RP |
9104 | #else /* !CONFIG_MMU */ |
9105 | ||
9106 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9107 | { | |
9108 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9109 | } | |
9110 | ||
9111 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9112 | { | |
9113 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9114 | } | |
9115 | ||
9116 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9117 | unsigned long addr, unsigned long len, | |
9118 | unsigned long pgoff, unsigned long flags) | |
9119 | { | |
9120 | void *ptr; | |
9121 | ||
9122 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9123 | if (IS_ERR(ptr)) | |
9124 | return PTR_ERR(ptr); | |
9125 | ||
9126 | return (unsigned long) ptr; | |
9127 | } | |
9128 | ||
9129 | #endif /* !CONFIG_MMU */ | |
9130 | ||
d9d05217 | 9131 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9132 | { |
9133 | DEFINE_WAIT(wait); | |
9134 | ||
9135 | do { | |
9136 | if (!io_sqring_full(ctx)) | |
9137 | break; | |
90554200 JA |
9138 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9139 | ||
9140 | if (!io_sqring_full(ctx)) | |
9141 | break; | |
90554200 JA |
9142 | schedule(); |
9143 | } while (!signal_pending(current)); | |
9144 | ||
9145 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9146 | return 0; |
90554200 JA |
9147 | } |
9148 | ||
c73ebb68 HX |
9149 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9150 | struct __kernel_timespec __user **ts, | |
9151 | const sigset_t __user **sig) | |
9152 | { | |
9153 | struct io_uring_getevents_arg arg; | |
9154 | ||
9155 | /* | |
9156 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9157 | * is just a pointer to the sigset_t. | |
9158 | */ | |
9159 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9160 | *sig = (const sigset_t __user *) argp; | |
9161 | *ts = NULL; | |
9162 | return 0; | |
9163 | } | |
9164 | ||
9165 | /* | |
9166 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9167 | * timespec and sigset_t pointers if good. | |
9168 | */ | |
9169 | if (*argsz != sizeof(arg)) | |
9170 | return -EINVAL; | |
9171 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9172 | return -EFAULT; | |
9173 | *sig = u64_to_user_ptr(arg.sigmask); | |
9174 | *argsz = arg.sigmask_sz; | |
9175 | *ts = u64_to_user_ptr(arg.ts); | |
9176 | return 0; | |
9177 | } | |
9178 | ||
2b188cc1 | 9179 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9180 | u32, min_complete, u32, flags, const void __user *, argp, |
9181 | size_t, argsz) | |
2b188cc1 JA |
9182 | { |
9183 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9184 | int submitted = 0; |
9185 | struct fd f; | |
33f993da | 9186 | long ret; |
2b188cc1 | 9187 | |
4c6e277c | 9188 | io_run_task_work(); |
b41e9852 | 9189 | |
33f993da PB |
9190 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9191 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9192 | return -EINVAL; |
9193 | ||
9194 | f = fdget(fd); | |
33f993da | 9195 | if (unlikely(!f.file)) |
2b188cc1 JA |
9196 | return -EBADF; |
9197 | ||
9198 | ret = -EOPNOTSUPP; | |
33f993da | 9199 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9200 | goto out_fput; |
9201 | ||
9202 | ret = -ENXIO; | |
9203 | ctx = f.file->private_data; | |
33f993da | 9204 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9205 | goto out_fput; |
9206 | ||
7e84e1c7 | 9207 | ret = -EBADFD; |
33f993da | 9208 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9209 | goto out; |
9210 | ||
6c271ce2 JA |
9211 | /* |
9212 | * For SQ polling, the thread will do all submissions and completions. | |
9213 | * Just return the requested submit count, and wake the thread if | |
9214 | * we were asked to. | |
9215 | */ | |
b2a9eada | 9216 | ret = 0; |
6c271ce2 | 9217 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c503150 | 9218 | io_cqring_overflow_flush(ctx, false, NULL, NULL); |
89448c47 | 9219 | |
d9d05217 | 9220 | ret = -EOWNERDEAD; |
04147488 SM |
9221 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9222 | goto out; | |
9223 | } | |
6c271ce2 | 9224 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9225 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9226 | if (flags & IORING_ENTER_SQ_WAIT) { |
9227 | ret = io_sqpoll_wait_sq(ctx); | |
9228 | if (ret) | |
9229 | goto out; | |
9230 | } | |
6c271ce2 | 9231 | submitted = to_submit; |
b2a9eada | 9232 | } else if (to_submit) { |
baf186c4 | 9233 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9234 | if (unlikely(ret)) |
9235 | goto out; | |
2b188cc1 | 9236 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9237 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9238 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9239 | |
9240 | if (submitted != to_submit) | |
9241 | goto out; | |
2b188cc1 JA |
9242 | } |
9243 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9244 | const sigset_t __user *sig; |
9245 | struct __kernel_timespec __user *ts; | |
9246 | ||
9247 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9248 | if (unlikely(ret)) | |
9249 | goto out; | |
9250 | ||
2b188cc1 JA |
9251 | min_complete = min(min_complete, ctx->cq_entries); |
9252 | ||
32b2244a XW |
9253 | /* |
9254 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9255 | * space applications don't need to do io completion events | |
9256 | * polling again, they can rely on io_sq_thread to do polling | |
9257 | * work, which can reduce cpu usage and uring_lock contention. | |
9258 | */ | |
9259 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9260 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9261 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9262 | } else { |
c73ebb68 | 9263 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9264 | } |
2b188cc1 JA |
9265 | } |
9266 | ||
7c504e65 | 9267 | out: |
6805b32e | 9268 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9269 | out_fput: |
9270 | fdput(f); | |
9271 | return submitted ? submitted : ret; | |
9272 | } | |
9273 | ||
bebdb65e | 9274 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9275 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9276 | const struct cred *cred) | |
87ce955b | 9277 | { |
87ce955b JA |
9278 | struct user_namespace *uns = seq_user_ns(m); |
9279 | struct group_info *gi; | |
9280 | kernel_cap_t cap; | |
9281 | unsigned __capi; | |
9282 | int g; | |
9283 | ||
9284 | seq_printf(m, "%5d\n", id); | |
9285 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9286 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9287 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9288 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9289 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9290 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9291 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9292 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9293 | seq_puts(m, "\n\tGroups:\t"); | |
9294 | gi = cred->group_info; | |
9295 | for (g = 0; g < gi->ngroups; g++) { | |
9296 | seq_put_decimal_ull(m, g ? " " : "", | |
9297 | from_kgid_munged(uns, gi->gid[g])); | |
9298 | } | |
9299 | seq_puts(m, "\n\tCapEff:\t"); | |
9300 | cap = cred->cap_effective; | |
9301 | CAP_FOR_EACH_U32(__capi) | |
9302 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9303 | seq_putc(m, '\n'); | |
9304 | return 0; | |
9305 | } | |
9306 | ||
9307 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9308 | { | |
dbbe9c64 | 9309 | struct io_sq_data *sq = NULL; |
fad8e0de | 9310 | bool has_lock; |
87ce955b JA |
9311 | int i; |
9312 | ||
fad8e0de JA |
9313 | /* |
9314 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9315 | * since fdinfo case grabs it in the opposite direction of normal use | |
9316 | * cases. If we fail to get the lock, we just don't iterate any | |
9317 | * structures that could be going away outside the io_uring mutex. | |
9318 | */ | |
9319 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9320 | ||
5f3f26f9 | 9321 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9322 | sq = ctx->sq_data; |
5f3f26f9 JA |
9323 | if (!sq->thread) |
9324 | sq = NULL; | |
9325 | } | |
dbbe9c64 JQ |
9326 | |
9327 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9328 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9329 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9330 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9331 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9332 | |
87ce955b JA |
9333 | if (f) |
9334 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9335 | else | |
9336 | seq_printf(m, "%5u: <none>\n", i); | |
9337 | } | |
9338 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9339 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9340 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9341 | ||
9342 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9343 | (unsigned int) buf->len); | |
9344 | } | |
61cf9370 MWO |
9345 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9346 | unsigned long index; | |
9347 | const struct cred *cred; | |
9348 | ||
87ce955b | 9349 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9350 | xa_for_each(&ctx->personalities, index, cred) |
9351 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9352 | } |
d7718a9d JA |
9353 | seq_printf(m, "PollList:\n"); |
9354 | spin_lock_irq(&ctx->completion_lock); | |
9355 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9356 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9357 | struct io_kiocb *req; | |
9358 | ||
9359 | hlist_for_each_entry(req, list, hash_node) | |
9360 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9361 | req->task->task_works != NULL); | |
9362 | } | |
9363 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9364 | if (has_lock) |
9365 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9366 | } |
9367 | ||
9368 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9369 | { | |
9370 | struct io_ring_ctx *ctx = f->private_data; | |
9371 | ||
9372 | if (percpu_ref_tryget(&ctx->refs)) { | |
9373 | __io_uring_show_fdinfo(ctx, m); | |
9374 | percpu_ref_put(&ctx->refs); | |
9375 | } | |
9376 | } | |
bebdb65e | 9377 | #endif |
87ce955b | 9378 | |
2b188cc1 JA |
9379 | static const struct file_operations io_uring_fops = { |
9380 | .release = io_uring_release, | |
9381 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9382 | #ifndef CONFIG_MMU |
9383 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9384 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9385 | #endif | |
2b188cc1 JA |
9386 | .poll = io_uring_poll, |
9387 | .fasync = io_uring_fasync, | |
bebdb65e | 9388 | #ifdef CONFIG_PROC_FS |
87ce955b | 9389 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9390 | #endif |
2b188cc1 JA |
9391 | }; |
9392 | ||
9393 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9394 | struct io_uring_params *p) | |
9395 | { | |
75b28aff HV |
9396 | struct io_rings *rings; |
9397 | size_t size, sq_array_offset; | |
2b188cc1 | 9398 | |
bd740481 JA |
9399 | /* make sure these are sane, as we already accounted them */ |
9400 | ctx->sq_entries = p->sq_entries; | |
9401 | ctx->cq_entries = p->cq_entries; | |
9402 | ||
75b28aff HV |
9403 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9404 | if (size == SIZE_MAX) | |
9405 | return -EOVERFLOW; | |
9406 | ||
9407 | rings = io_mem_alloc(size); | |
9408 | if (!rings) | |
2b188cc1 JA |
9409 | return -ENOMEM; |
9410 | ||
75b28aff HV |
9411 | ctx->rings = rings; |
9412 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9413 | rings->sq_ring_mask = p->sq_entries - 1; | |
9414 | rings->cq_ring_mask = p->cq_entries - 1; | |
9415 | rings->sq_ring_entries = p->sq_entries; | |
9416 | rings->cq_ring_entries = p->cq_entries; | |
9417 | ctx->sq_mask = rings->sq_ring_mask; | |
9418 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9419 | |
9420 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9421 | if (size == SIZE_MAX) { |
9422 | io_mem_free(ctx->rings); | |
9423 | ctx->rings = NULL; | |
2b188cc1 | 9424 | return -EOVERFLOW; |
eb065d30 | 9425 | } |
2b188cc1 JA |
9426 | |
9427 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9428 | if (!ctx->sq_sqes) { |
9429 | io_mem_free(ctx->rings); | |
9430 | ctx->rings = NULL; | |
2b188cc1 | 9431 | return -ENOMEM; |
eb065d30 | 9432 | } |
2b188cc1 | 9433 | |
2b188cc1 JA |
9434 | return 0; |
9435 | } | |
9436 | ||
9faadcc8 PB |
9437 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9438 | { | |
9439 | int ret, fd; | |
9440 | ||
9441 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9442 | if (fd < 0) | |
9443 | return fd; | |
9444 | ||
baf186c4 | 9445 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9446 | if (ret) { |
9447 | put_unused_fd(fd); | |
9448 | return ret; | |
9449 | } | |
9450 | fd_install(fd, file); | |
9451 | return fd; | |
9452 | } | |
9453 | ||
2b188cc1 JA |
9454 | /* |
9455 | * Allocate an anonymous fd, this is what constitutes the application | |
9456 | * visible backing of an io_uring instance. The application mmaps this | |
9457 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9458 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9459 | */ | |
9faadcc8 | 9460 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9461 | { |
9462 | struct file *file; | |
9faadcc8 | 9463 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9464 | int ret; |
9465 | ||
2b188cc1 JA |
9466 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9467 | &ctx->ring_sock); | |
9468 | if (ret) | |
9faadcc8 | 9469 | return ERR_PTR(ret); |
2b188cc1 JA |
9470 | #endif |
9471 | ||
2b188cc1 JA |
9472 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9473 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9474 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9475 | if (IS_ERR(file)) { |
9476 | sock_release(ctx->ring_sock); | |
9477 | ctx->ring_sock = NULL; | |
9478 | } else { | |
9479 | ctx->ring_sock->file = file; | |
0f212204 | 9480 | } |
2b188cc1 | 9481 | #endif |
9faadcc8 | 9482 | return file; |
2b188cc1 JA |
9483 | } |
9484 | ||
7f13657d XW |
9485 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9486 | struct io_uring_params __user *params) | |
2b188cc1 | 9487 | { |
2b188cc1 | 9488 | struct io_ring_ctx *ctx; |
9faadcc8 | 9489 | struct file *file; |
2b188cc1 JA |
9490 | int ret; |
9491 | ||
8110c1a6 | 9492 | if (!entries) |
2b188cc1 | 9493 | return -EINVAL; |
8110c1a6 JA |
9494 | if (entries > IORING_MAX_ENTRIES) { |
9495 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9496 | return -EINVAL; | |
9497 | entries = IORING_MAX_ENTRIES; | |
9498 | } | |
2b188cc1 JA |
9499 | |
9500 | /* | |
9501 | * Use twice as many entries for the CQ ring. It's possible for the | |
9502 | * application to drive a higher depth than the size of the SQ ring, | |
9503 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9504 | * some flexibility in overcommitting a bit. If the application has |
9505 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9506 | * of CQ ring entries manually. | |
2b188cc1 JA |
9507 | */ |
9508 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9509 | if (p->flags & IORING_SETUP_CQSIZE) { |
9510 | /* | |
9511 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9512 | * to a power-of-two, if it isn't already. We do NOT impose | |
9513 | * any cq vs sq ring sizing. | |
9514 | */ | |
eb2667b3 | 9515 | if (!p->cq_entries) |
33a107f0 | 9516 | return -EINVAL; |
8110c1a6 JA |
9517 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9518 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9519 | return -EINVAL; | |
9520 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9521 | } | |
eb2667b3 JQ |
9522 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9523 | if (p->cq_entries < p->sq_entries) | |
9524 | return -EINVAL; | |
33a107f0 JA |
9525 | } else { |
9526 | p->cq_entries = 2 * p->sq_entries; | |
9527 | } | |
2b188cc1 | 9528 | |
2b188cc1 | 9529 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9530 | if (!ctx) |
2b188cc1 | 9531 | return -ENOMEM; |
2b188cc1 | 9532 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9533 | if (!capable(CAP_IPC_LOCK)) |
9534 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9535 | |
9536 | /* | |
9537 | * This is just grabbed for accounting purposes. When a process exits, | |
9538 | * the mm is exited and dropped before the files, hence we need to hang | |
9539 | * on to this mm purely for the purposes of being able to unaccount | |
9540 | * memory (locked/pinned vm). It's not used for anything else. | |
9541 | */ | |
6b7898eb | 9542 | mmgrab(current->mm); |
2aede0e4 | 9543 | ctx->mm_account = current->mm; |
6b7898eb | 9544 | |
2b188cc1 JA |
9545 | ret = io_allocate_scq_urings(ctx, p); |
9546 | if (ret) | |
9547 | goto err; | |
9548 | ||
7e84e1c7 | 9549 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9550 | if (ret) |
9551 | goto err; | |
9552 | ||
2b188cc1 | 9553 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9554 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9555 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9556 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9557 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9558 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9559 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9560 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9561 | |
9562 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9563 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9564 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9565 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9566 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9567 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9568 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9569 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9570 | |
7f13657d XW |
9571 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9572 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9573 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9574 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9575 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9576 | |
9577 | if (copy_to_user(params, p, sizeof(*p))) { | |
9578 | ret = -EFAULT; | |
9579 | goto err; | |
9580 | } | |
d1719f70 | 9581 | |
9faadcc8 PB |
9582 | file = io_uring_get_file(ctx); |
9583 | if (IS_ERR(file)) { | |
9584 | ret = PTR_ERR(file); | |
9585 | goto err; | |
9586 | } | |
9587 | ||
044c1ab3 JA |
9588 | /* |
9589 | * Install ring fd as the very last thing, so we don't risk someone | |
9590 | * having closed it before we finish setup | |
9591 | */ | |
9faadcc8 PB |
9592 | ret = io_uring_install_fd(ctx, file); |
9593 | if (ret < 0) { | |
9594 | /* fput will clean it up */ | |
9595 | fput(file); | |
9596 | return ret; | |
9597 | } | |
044c1ab3 | 9598 | |
c826bd7a | 9599 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9600 | return ret; |
9601 | err: | |
9602 | io_ring_ctx_wait_and_kill(ctx); | |
9603 | return ret; | |
9604 | } | |
9605 | ||
9606 | /* | |
9607 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9608 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9609 | * params structure passed in. | |
9610 | */ | |
9611 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9612 | { | |
9613 | struct io_uring_params p; | |
2b188cc1 JA |
9614 | int i; |
9615 | ||
9616 | if (copy_from_user(&p, params, sizeof(p))) | |
9617 | return -EFAULT; | |
9618 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9619 | if (p.resv[i]) | |
9620 | return -EINVAL; | |
9621 | } | |
9622 | ||
6c271ce2 | 9623 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9624 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9625 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9626 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9627 | return -EINVAL; |
9628 | ||
7f13657d | 9629 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9630 | } |
9631 | ||
9632 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9633 | struct io_uring_params __user *, params) | |
9634 | { | |
9635 | return io_uring_setup(entries, params); | |
9636 | } | |
9637 | ||
66f4af93 JA |
9638 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9639 | { | |
9640 | struct io_uring_probe *p; | |
9641 | size_t size; | |
9642 | int i, ret; | |
9643 | ||
9644 | size = struct_size(p, ops, nr_args); | |
9645 | if (size == SIZE_MAX) | |
9646 | return -EOVERFLOW; | |
9647 | p = kzalloc(size, GFP_KERNEL); | |
9648 | if (!p) | |
9649 | return -ENOMEM; | |
9650 | ||
9651 | ret = -EFAULT; | |
9652 | if (copy_from_user(p, arg, size)) | |
9653 | goto out; | |
9654 | ret = -EINVAL; | |
9655 | if (memchr_inv(p, 0, size)) | |
9656 | goto out; | |
9657 | ||
9658 | p->last_op = IORING_OP_LAST - 1; | |
9659 | if (nr_args > IORING_OP_LAST) | |
9660 | nr_args = IORING_OP_LAST; | |
9661 | ||
9662 | for (i = 0; i < nr_args; i++) { | |
9663 | p->ops[i].op = i; | |
9664 | if (!io_op_defs[i].not_supported) | |
9665 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9666 | } | |
9667 | p->ops_len = i; | |
9668 | ||
9669 | ret = 0; | |
9670 | if (copy_to_user(arg, p, size)) | |
9671 | ret = -EFAULT; | |
9672 | out: | |
9673 | kfree(p); | |
9674 | return ret; | |
9675 | } | |
9676 | ||
071698e1 JA |
9677 | static int io_register_personality(struct io_ring_ctx *ctx) |
9678 | { | |
4379bf8b | 9679 | const struct cred *creds; |
61cf9370 | 9680 | u32 id; |
1e6fa521 | 9681 | int ret; |
071698e1 | 9682 | |
4379bf8b | 9683 | creds = get_current_cred(); |
1e6fa521 | 9684 | |
61cf9370 MWO |
9685 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9686 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9687 | if (!ret) | |
9688 | return id; | |
9689 | put_cred(creds); | |
1e6fa521 | 9690 | return ret; |
071698e1 JA |
9691 | } |
9692 | ||
21b55dbc SG |
9693 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9694 | unsigned int nr_args) | |
9695 | { | |
9696 | struct io_uring_restriction *res; | |
9697 | size_t size; | |
9698 | int i, ret; | |
9699 | ||
7e84e1c7 SG |
9700 | /* Restrictions allowed only if rings started disabled */ |
9701 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9702 | return -EBADFD; | |
9703 | ||
21b55dbc | 9704 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9705 | if (ctx->restrictions.registered) |
21b55dbc SG |
9706 | return -EBUSY; |
9707 | ||
9708 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9709 | return -EINVAL; | |
9710 | ||
9711 | size = array_size(nr_args, sizeof(*res)); | |
9712 | if (size == SIZE_MAX) | |
9713 | return -EOVERFLOW; | |
9714 | ||
9715 | res = memdup_user(arg, size); | |
9716 | if (IS_ERR(res)) | |
9717 | return PTR_ERR(res); | |
9718 | ||
9719 | ret = 0; | |
9720 | ||
9721 | for (i = 0; i < nr_args; i++) { | |
9722 | switch (res[i].opcode) { | |
9723 | case IORING_RESTRICTION_REGISTER_OP: | |
9724 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9725 | ret = -EINVAL; | |
9726 | goto out; | |
9727 | } | |
9728 | ||
9729 | __set_bit(res[i].register_op, | |
9730 | ctx->restrictions.register_op); | |
9731 | break; | |
9732 | case IORING_RESTRICTION_SQE_OP: | |
9733 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9734 | ret = -EINVAL; | |
9735 | goto out; | |
9736 | } | |
9737 | ||
9738 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9739 | break; | |
9740 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9741 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9742 | break; | |
9743 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9744 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9745 | break; | |
9746 | default: | |
9747 | ret = -EINVAL; | |
9748 | goto out; | |
9749 | } | |
9750 | } | |
9751 | ||
9752 | out: | |
9753 | /* Reset all restrictions if an error happened */ | |
9754 | if (ret != 0) | |
9755 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9756 | else | |
7e84e1c7 | 9757 | ctx->restrictions.registered = true; |
21b55dbc SG |
9758 | |
9759 | kfree(res); | |
9760 | return ret; | |
9761 | } | |
9762 | ||
7e84e1c7 SG |
9763 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9764 | { | |
9765 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9766 | return -EBADFD; | |
9767 | ||
9768 | if (ctx->restrictions.registered) | |
9769 | ctx->restricted = 1; | |
9770 | ||
0298ef96 PB |
9771 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9772 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9773 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9774 | return 0; |
9775 | } | |
9776 | ||
071698e1 JA |
9777 | static bool io_register_op_must_quiesce(int op) |
9778 | { | |
9779 | switch (op) { | |
9780 | case IORING_UNREGISTER_FILES: | |
9781 | case IORING_REGISTER_FILES_UPDATE: | |
9782 | case IORING_REGISTER_PROBE: | |
9783 | case IORING_REGISTER_PERSONALITY: | |
9784 | case IORING_UNREGISTER_PERSONALITY: | |
9785 | return false; | |
9786 | default: | |
9787 | return true; | |
9788 | } | |
9789 | } | |
9790 | ||
edafccee JA |
9791 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9792 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9793 | __releases(ctx->uring_lock) |
9794 | __acquires(ctx->uring_lock) | |
edafccee JA |
9795 | { |
9796 | int ret; | |
9797 | ||
35fa71a0 JA |
9798 | /* |
9799 | * We're inside the ring mutex, if the ref is already dying, then | |
9800 | * someone else killed the ctx or is already going through | |
9801 | * io_uring_register(). | |
9802 | */ | |
9803 | if (percpu_ref_is_dying(&ctx->refs)) | |
9804 | return -ENXIO; | |
9805 | ||
071698e1 | 9806 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9807 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9808 | |
05f3fb3c JA |
9809 | /* |
9810 | * Drop uring mutex before waiting for references to exit. If | |
9811 | * another thread is currently inside io_uring_enter() it might | |
9812 | * need to grab the uring_lock to make progress. If we hold it | |
9813 | * here across the drain wait, then we can deadlock. It's safe | |
9814 | * to drop the mutex here, since no new references will come in | |
9815 | * after we've killed the percpu ref. | |
9816 | */ | |
9817 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9818 | do { |
9819 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9820 | if (!ret) | |
9821 | break; | |
ed6930c9 JA |
9822 | ret = io_run_task_work_sig(); |
9823 | if (ret < 0) | |
9824 | break; | |
af9c1a44 JA |
9825 | } while (1); |
9826 | ||
05f3fb3c | 9827 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9828 | |
c150368b JA |
9829 | if (ret) { |
9830 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9831 | goto out_quiesce; |
9832 | } | |
9833 | } | |
9834 | ||
9835 | if (ctx->restricted) { | |
9836 | if (opcode >= IORING_REGISTER_LAST) { | |
9837 | ret = -EINVAL; | |
9838 | goto out; | |
9839 | } | |
9840 | ||
9841 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9842 | ret = -EACCES; | |
c150368b JA |
9843 | goto out; |
9844 | } | |
05f3fb3c | 9845 | } |
edafccee JA |
9846 | |
9847 | switch (opcode) { | |
9848 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9849 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9850 | break; |
9851 | case IORING_UNREGISTER_BUFFERS: | |
9852 | ret = -EINVAL; | |
9853 | if (arg || nr_args) | |
9854 | break; | |
0a96bbe4 | 9855 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9856 | break; |
6b06314c JA |
9857 | case IORING_REGISTER_FILES: |
9858 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9859 | break; | |
9860 | case IORING_UNREGISTER_FILES: | |
9861 | ret = -EINVAL; | |
9862 | if (arg || nr_args) | |
9863 | break; | |
9864 | ret = io_sqe_files_unregister(ctx); | |
9865 | break; | |
c3a31e60 JA |
9866 | case IORING_REGISTER_FILES_UPDATE: |
9867 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9868 | break; | |
9b402849 | 9869 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9870 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9871 | ret = -EINVAL; |
9872 | if (nr_args != 1) | |
9873 | break; | |
9874 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9875 | if (ret) |
9876 | break; | |
9877 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9878 | ctx->eventfd_async = 1; | |
9879 | else | |
9880 | ctx->eventfd_async = 0; | |
9b402849 JA |
9881 | break; |
9882 | case IORING_UNREGISTER_EVENTFD: | |
9883 | ret = -EINVAL; | |
9884 | if (arg || nr_args) | |
9885 | break; | |
9886 | ret = io_eventfd_unregister(ctx); | |
9887 | break; | |
66f4af93 JA |
9888 | case IORING_REGISTER_PROBE: |
9889 | ret = -EINVAL; | |
9890 | if (!arg || nr_args > 256) | |
9891 | break; | |
9892 | ret = io_probe(ctx, arg, nr_args); | |
9893 | break; | |
071698e1 JA |
9894 | case IORING_REGISTER_PERSONALITY: |
9895 | ret = -EINVAL; | |
9896 | if (arg || nr_args) | |
9897 | break; | |
9898 | ret = io_register_personality(ctx); | |
9899 | break; | |
9900 | case IORING_UNREGISTER_PERSONALITY: | |
9901 | ret = -EINVAL; | |
9902 | if (arg) | |
9903 | break; | |
9904 | ret = io_unregister_personality(ctx, nr_args); | |
9905 | break; | |
7e84e1c7 SG |
9906 | case IORING_REGISTER_ENABLE_RINGS: |
9907 | ret = -EINVAL; | |
9908 | if (arg || nr_args) | |
9909 | break; | |
9910 | ret = io_register_enable_rings(ctx); | |
9911 | break; | |
21b55dbc SG |
9912 | case IORING_REGISTER_RESTRICTIONS: |
9913 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9914 | break; | |
edafccee JA |
9915 | default: |
9916 | ret = -EINVAL; | |
9917 | break; | |
9918 | } | |
9919 | ||
21b55dbc | 9920 | out: |
071698e1 | 9921 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9922 | /* bring the ctx back to life */ |
05f3fb3c | 9923 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 9924 | out_quiesce: |
0f158b4c | 9925 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9926 | } |
edafccee JA |
9927 | return ret; |
9928 | } | |
9929 | ||
9930 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9931 | void __user *, arg, unsigned int, nr_args) | |
9932 | { | |
9933 | struct io_ring_ctx *ctx; | |
9934 | long ret = -EBADF; | |
9935 | struct fd f; | |
9936 | ||
9937 | f = fdget(fd); | |
9938 | if (!f.file) | |
9939 | return -EBADF; | |
9940 | ||
9941 | ret = -EOPNOTSUPP; | |
9942 | if (f.file->f_op != &io_uring_fops) | |
9943 | goto out_fput; | |
9944 | ||
9945 | ctx = f.file->private_data; | |
9946 | ||
b6c23dd5 PB |
9947 | io_run_task_work(); |
9948 | ||
edafccee JA |
9949 | mutex_lock(&ctx->uring_lock); |
9950 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9951 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9952 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9953 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9954 | out_fput: |
9955 | fdput(f); | |
9956 | return ret; | |
9957 | } | |
9958 | ||
2b188cc1 JA |
9959 | static int __init io_uring_init(void) |
9960 | { | |
d7f62e82 SM |
9961 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9962 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9963 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9964 | } while (0) | |
9965 | ||
9966 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9967 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9968 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9969 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9970 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9971 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9972 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9973 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9974 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9975 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9976 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9977 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9978 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9979 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9980 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9981 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9982 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9983 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9984 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9985 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9986 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9987 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9988 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9989 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9990 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9991 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9992 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9993 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9994 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9995 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9996 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9997 | |
d3656344 | 9998 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9999 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10000 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10001 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10002 | return 0; |
10003 | }; | |
10004 | __initcall(io_uring_init); |