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 | ||
6c2450ae PB |
205 | struct io_overflow_cqe { |
206 | struct io_uring_cqe cqe; | |
207 | struct list_head list; | |
208 | }; | |
209 | ||
269bbe5f BM |
210 | struct io_rsrc_put { |
211 | struct list_head list; | |
50238531 BM |
212 | union { |
213 | void *rsrc; | |
214 | struct file *file; | |
215 | }; | |
269bbe5f BM |
216 | }; |
217 | ||
218 | struct fixed_rsrc_table { | |
65e19f54 | 219 | struct file **files; |
31b51510 JA |
220 | }; |
221 | ||
b895c9a6 | 222 | struct io_rsrc_node { |
05589553 XW |
223 | struct percpu_ref refs; |
224 | struct list_head node; | |
269bbe5f | 225 | struct list_head rsrc_list; |
b895c9a6 | 226 | struct io_rsrc_data *rsrc_data; |
50238531 BM |
227 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
228 | struct io_rsrc_put *prsrc); | |
4a38aed2 | 229 | struct llist_node llist; |
e297822b | 230 | bool done; |
05589553 XW |
231 | }; |
232 | ||
b895c9a6 | 233 | struct io_rsrc_data { |
269bbe5f | 234 | struct fixed_rsrc_table *table; |
05f3fb3c JA |
235 | struct io_ring_ctx *ctx; |
236 | ||
b895c9a6 | 237 | struct io_rsrc_node *node; |
05f3fb3c | 238 | struct percpu_ref refs; |
05f3fb3c | 239 | struct completion done; |
8bad28d8 | 240 | bool quiesce; |
05f3fb3c JA |
241 | }; |
242 | ||
5a2e745d JA |
243 | struct io_buffer { |
244 | struct list_head list; | |
245 | __u64 addr; | |
246 | __s32 len; | |
247 | __u16 bid; | |
248 | }; | |
249 | ||
21b55dbc SG |
250 | struct io_restriction { |
251 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
252 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
253 | u8 sqe_flags_allowed; | |
254 | u8 sqe_flags_required; | |
7e84e1c7 | 255 | bool registered; |
21b55dbc SG |
256 | }; |
257 | ||
37d1e2e3 JA |
258 | enum { |
259 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
260 | IO_SQ_THREAD_SHOULD_PARK, | |
261 | }; | |
262 | ||
534ca6d6 JA |
263 | struct io_sq_data { |
264 | refcount_t refs; | |
9e138a48 | 265 | atomic_t park_pending; |
09a6f4ef | 266 | struct mutex lock; |
69fb2131 JA |
267 | |
268 | /* ctx's that are using this sqd */ | |
269 | struct list_head ctx_list; | |
69fb2131 | 270 | |
534ca6d6 JA |
271 | struct task_struct *thread; |
272 | struct wait_queue_head wait; | |
08369246 XW |
273 | |
274 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
275 | int sq_cpu; |
276 | pid_t task_pid; | |
5c2469e0 | 277 | pid_t task_tgid; |
37d1e2e3 JA |
278 | |
279 | unsigned long state; | |
37d1e2e3 | 280 | struct completion exited; |
b7f5a0bf | 281 | struct callback_head *park_task_work; |
534ca6d6 JA |
282 | }; |
283 | ||
258b29a9 | 284 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 285 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 286 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 287 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
288 | |
289 | struct io_comp_state { | |
6dd0be1e | 290 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 291 | unsigned int nr; |
c7dae4ba JA |
292 | unsigned int locked_free_nr; |
293 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 294 | struct list_head free_list; |
c7dae4ba JA |
295 | /* IRQ completion list, under ->completion_lock */ |
296 | struct list_head locked_free_list; | |
258b29a9 PB |
297 | }; |
298 | ||
a1ab7b35 PB |
299 | struct io_submit_link { |
300 | struct io_kiocb *head; | |
301 | struct io_kiocb *last; | |
302 | }; | |
303 | ||
258b29a9 PB |
304 | struct io_submit_state { |
305 | struct blk_plug plug; | |
a1ab7b35 | 306 | struct io_submit_link link; |
258b29a9 PB |
307 | |
308 | /* | |
309 | * io_kiocb alloc cache | |
310 | */ | |
bf019da7 | 311 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
312 | unsigned int free_reqs; |
313 | ||
314 | bool plug_started; | |
315 | ||
316 | /* | |
317 | * Batch completion logic | |
318 | */ | |
319 | struct io_comp_state comp; | |
320 | ||
321 | /* | |
322 | * File reference cache | |
323 | */ | |
324 | struct file *file; | |
325 | unsigned int fd; | |
326 | unsigned int file_refs; | |
327 | unsigned int ios_left; | |
328 | }; | |
329 | ||
2b188cc1 JA |
330 | struct io_ring_ctx { |
331 | struct { | |
332 | struct percpu_ref refs; | |
333 | } ____cacheline_aligned_in_smp; | |
334 | ||
335 | struct { | |
336 | unsigned int flags; | |
e1d85334 | 337 | unsigned int compat: 1; |
e1d85334 RD |
338 | unsigned int cq_overflow_flushed: 1; |
339 | unsigned int drain_next: 1; | |
340 | unsigned int eventfd_async: 1; | |
21b55dbc | 341 | unsigned int restricted: 1; |
2b188cc1 | 342 | |
75b28aff HV |
343 | /* |
344 | * Ring buffer of indices into array of io_uring_sqe, which is | |
345 | * mmapped by the application using the IORING_OFF_SQES offset. | |
346 | * | |
347 | * This indirection could e.g. be used to assign fixed | |
348 | * io_uring_sqe entries to operations and only submit them to | |
349 | * the queue when needed. | |
350 | * | |
351 | * The kernel modifies neither the indices array nor the entries | |
352 | * array. | |
353 | */ | |
354 | u32 *sq_array; | |
2b188cc1 JA |
355 | unsigned cached_sq_head; |
356 | unsigned sq_entries; | |
357 | unsigned sq_mask; | |
6c271ce2 | 358 | unsigned sq_thread_idle; |
498ccd9e | 359 | unsigned cached_sq_dropped; |
2c3bac6d | 360 | unsigned cached_cq_overflow; |
ad3eb2c8 | 361 | unsigned long sq_check_overflow; |
de0617e4 | 362 | |
e941894e JA |
363 | /* hashed buffered write serialization */ |
364 | struct io_wq_hash *hash_map; | |
365 | ||
de0617e4 | 366 | struct list_head defer_list; |
5262f567 | 367 | struct list_head timeout_list; |
1d7bb1d5 | 368 | struct list_head cq_overflow_list; |
fcb323cc | 369 | |
ad3eb2c8 | 370 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
371 | } ____cacheline_aligned_in_smp; |
372 | ||
3c1a2ead JA |
373 | struct { |
374 | struct mutex uring_lock; | |
375 | wait_queue_head_t wait; | |
376 | } ____cacheline_aligned_in_smp; | |
377 | ||
378 | struct io_submit_state submit_state; | |
379 | ||
206aefde JA |
380 | struct io_rings *rings; |
381 | ||
2aede0e4 JA |
382 | /* Only used for accounting purposes */ |
383 | struct mm_struct *mm_account; | |
384 | ||
7c30f36a | 385 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
386 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
387 | ||
90554200 | 388 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 389 | struct list_head sqd_list; |
75b28aff | 390 | |
6b06314c JA |
391 | /* |
392 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
393 | * readers must ensure that ->refs is alive as long as the file* is | |
394 | * used. Only updated through io_uring_register(2). | |
395 | */ | |
b895c9a6 | 396 | struct io_rsrc_data *file_data; |
6b06314c JA |
397 | unsigned nr_user_files; |
398 | ||
edafccee JA |
399 | /* if used, fixed mapped user buffers */ |
400 | unsigned nr_user_bufs; | |
401 | struct io_mapped_ubuf *user_bufs; | |
402 | ||
2b188cc1 JA |
403 | struct user_struct *user; |
404 | ||
0f158b4c | 405 | struct completion ref_comp; |
206aefde JA |
406 | |
407 | #if defined(CONFIG_UNIX) | |
408 | struct socket *ring_sock; | |
409 | #endif | |
410 | ||
9e15c3a0 | 411 | struct xarray io_buffers; |
5a2e745d | 412 | |
61cf9370 MWO |
413 | struct xarray personalities; |
414 | u32 pers_next; | |
071698e1 | 415 | |
206aefde JA |
416 | struct { |
417 | unsigned cached_cq_tail; | |
418 | unsigned cq_entries; | |
419 | unsigned cq_mask; | |
420 | atomic_t cq_timeouts; | |
f010505b | 421 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 422 | unsigned long cq_check_overflow; |
206aefde JA |
423 | struct wait_queue_head cq_wait; |
424 | struct fasync_struct *cq_fasync; | |
425 | struct eventfd_ctx *cq_ev_fd; | |
426 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 427 | |
2b188cc1 JA |
428 | struct { |
429 | spinlock_t completion_lock; | |
e94f141b | 430 | |
def596e9 | 431 | /* |
540e32a0 | 432 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
433 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
434 | * For SQPOLL, only the single threaded io_sq_thread() will | |
435 | * manipulate the list, hence no extra locking is needed there. | |
436 | */ | |
540e32a0 | 437 | struct list_head iopoll_list; |
78076bb6 JA |
438 | struct hlist_head *cancel_hash; |
439 | unsigned cancel_hash_bits; | |
e94f141b | 440 | bool poll_multi_file; |
31b51510 | 441 | |
fcb323cc JA |
442 | spinlock_t inflight_lock; |
443 | struct list_head inflight_list; | |
2b188cc1 | 444 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 445 | |
269bbe5f BM |
446 | struct delayed_work rsrc_put_work; |
447 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
448 | struct list_head rsrc_ref_list; |
449 | spinlock_t rsrc_ref_lock; | |
b895c9a6 | 450 | struct io_rsrc_node *rsrc_backup_node; |
4a38aed2 | 451 | |
21b55dbc | 452 | struct io_restriction restrictions; |
3c1a2ead | 453 | |
7c25c0d1 JA |
454 | /* exit task_work */ |
455 | struct callback_head *exit_task_work; | |
456 | ||
e941894e JA |
457 | struct wait_queue_head hash_wait; |
458 | ||
3c1a2ead JA |
459 | /* Keep this last, we don't need it for the fast path */ |
460 | struct work_struct exit_work; | |
13bf43f5 | 461 | struct list_head tctx_list; |
2b188cc1 JA |
462 | }; |
463 | ||
53e043b2 SM |
464 | struct io_uring_task { |
465 | /* submission side */ | |
466 | struct xarray xa; | |
467 | struct wait_queue_head wait; | |
ee53fb2b SM |
468 | const struct io_ring_ctx *last; |
469 | struct io_wq *io_wq; | |
53e043b2 SM |
470 | struct percpu_counter inflight; |
471 | atomic_t in_idle; | |
53e043b2 SM |
472 | |
473 | spinlock_t task_lock; | |
474 | struct io_wq_work_list task_list; | |
475 | unsigned long task_state; | |
476 | struct callback_head task_work; | |
477 | }; | |
478 | ||
09bb8394 JA |
479 | /* |
480 | * First field must be the file pointer in all the | |
481 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
482 | */ | |
221c5eb2 JA |
483 | struct io_poll_iocb { |
484 | struct file *file; | |
018043be | 485 | struct wait_queue_head *head; |
221c5eb2 | 486 | __poll_t events; |
8c838788 | 487 | bool done; |
221c5eb2 | 488 | bool canceled; |
b69de288 JA |
489 | bool update_events; |
490 | bool update_user_data; | |
491 | union { | |
492 | struct wait_queue_entry wait; | |
493 | struct { | |
494 | u64 old_user_data; | |
495 | u64 new_user_data; | |
496 | }; | |
497 | }; | |
221c5eb2 JA |
498 | }; |
499 | ||
018043be PB |
500 | struct io_poll_remove { |
501 | struct file *file; | |
502 | u64 addr; | |
503 | }; | |
504 | ||
b5dba59e JA |
505 | struct io_close { |
506 | struct file *file; | |
b5dba59e JA |
507 | int fd; |
508 | }; | |
509 | ||
ad8a48ac JA |
510 | struct io_timeout_data { |
511 | struct io_kiocb *req; | |
512 | struct hrtimer timer; | |
513 | struct timespec64 ts; | |
514 | enum hrtimer_mode mode; | |
515 | }; | |
516 | ||
8ed8d3c3 JA |
517 | struct io_accept { |
518 | struct file *file; | |
519 | struct sockaddr __user *addr; | |
520 | int __user *addr_len; | |
521 | int flags; | |
09952e3e | 522 | unsigned long nofile; |
8ed8d3c3 JA |
523 | }; |
524 | ||
525 | struct io_sync { | |
526 | struct file *file; | |
527 | loff_t len; | |
528 | loff_t off; | |
529 | int flags; | |
d63d1b5e | 530 | int mode; |
8ed8d3c3 JA |
531 | }; |
532 | ||
fbf23849 JA |
533 | struct io_cancel { |
534 | struct file *file; | |
535 | u64 addr; | |
536 | }; | |
537 | ||
b29472ee JA |
538 | struct io_timeout { |
539 | struct file *file; | |
bfe68a22 PB |
540 | u32 off; |
541 | u32 target_seq; | |
135fcde8 | 542 | struct list_head list; |
90cd7e42 PB |
543 | /* head of the link, used by linked timeouts only */ |
544 | struct io_kiocb *head; | |
b29472ee JA |
545 | }; |
546 | ||
0bdf7a2d PB |
547 | struct io_timeout_rem { |
548 | struct file *file; | |
549 | u64 addr; | |
9c8e11b3 PB |
550 | |
551 | /* timeout update */ | |
552 | struct timespec64 ts; | |
553 | u32 flags; | |
0bdf7a2d PB |
554 | }; |
555 | ||
9adbd45d JA |
556 | struct io_rw { |
557 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
558 | struct kiocb kiocb; | |
559 | u64 addr; | |
560 | u64 len; | |
561 | }; | |
562 | ||
3fbb51c1 JA |
563 | struct io_connect { |
564 | struct file *file; | |
565 | struct sockaddr __user *addr; | |
566 | int addr_len; | |
567 | }; | |
568 | ||
e47293fd JA |
569 | struct io_sr_msg { |
570 | struct file *file; | |
fddaface | 571 | union { |
270a5940 | 572 | struct user_msghdr __user *umsg; |
fddaface JA |
573 | void __user *buf; |
574 | }; | |
e47293fd | 575 | int msg_flags; |
bcda7baa | 576 | int bgid; |
fddaface | 577 | size_t len; |
bcda7baa | 578 | struct io_buffer *kbuf; |
e47293fd JA |
579 | }; |
580 | ||
15b71abe JA |
581 | struct io_open { |
582 | struct file *file; | |
583 | int dfd; | |
15b71abe | 584 | struct filename *filename; |
c12cedf2 | 585 | struct open_how how; |
4022e7af | 586 | unsigned long nofile; |
15b71abe JA |
587 | }; |
588 | ||
269bbe5f | 589 | struct io_rsrc_update { |
05f3fb3c JA |
590 | struct file *file; |
591 | u64 arg; | |
592 | u32 nr_args; | |
593 | u32 offset; | |
594 | }; | |
595 | ||
4840e418 JA |
596 | struct io_fadvise { |
597 | struct file *file; | |
598 | u64 offset; | |
599 | u32 len; | |
600 | u32 advice; | |
601 | }; | |
602 | ||
c1ca757b JA |
603 | struct io_madvise { |
604 | struct file *file; | |
605 | u64 addr; | |
606 | u32 len; | |
607 | u32 advice; | |
608 | }; | |
609 | ||
3e4827b0 JA |
610 | struct io_epoll { |
611 | struct file *file; | |
612 | int epfd; | |
613 | int op; | |
614 | int fd; | |
615 | struct epoll_event event; | |
e47293fd JA |
616 | }; |
617 | ||
7d67af2c PB |
618 | struct io_splice { |
619 | struct file *file_out; | |
620 | struct file *file_in; | |
621 | loff_t off_out; | |
622 | loff_t off_in; | |
623 | u64 len; | |
624 | unsigned int flags; | |
625 | }; | |
626 | ||
ddf0322d JA |
627 | struct io_provide_buf { |
628 | struct file *file; | |
629 | __u64 addr; | |
630 | __s32 len; | |
631 | __u32 bgid; | |
632 | __u16 nbufs; | |
633 | __u16 bid; | |
634 | }; | |
635 | ||
1d9e1288 BM |
636 | struct io_statx { |
637 | struct file *file; | |
638 | int dfd; | |
639 | unsigned int mask; | |
640 | unsigned int flags; | |
e62753e4 | 641 | const char __user *filename; |
1d9e1288 BM |
642 | struct statx __user *buffer; |
643 | }; | |
644 | ||
36f4fa68 JA |
645 | struct io_shutdown { |
646 | struct file *file; | |
647 | int how; | |
648 | }; | |
649 | ||
80a261fd JA |
650 | struct io_rename { |
651 | struct file *file; | |
652 | int old_dfd; | |
653 | int new_dfd; | |
654 | struct filename *oldpath; | |
655 | struct filename *newpath; | |
656 | int flags; | |
657 | }; | |
658 | ||
14a1143b JA |
659 | struct io_unlink { |
660 | struct file *file; | |
661 | int dfd; | |
662 | int flags; | |
663 | struct filename *filename; | |
664 | }; | |
665 | ||
3ca405eb PB |
666 | struct io_completion { |
667 | struct file *file; | |
668 | struct list_head list; | |
8c3f9cd1 | 669 | u32 cflags; |
3ca405eb PB |
670 | }; |
671 | ||
f499a021 JA |
672 | struct io_async_connect { |
673 | struct sockaddr_storage address; | |
674 | }; | |
675 | ||
03b1230c JA |
676 | struct io_async_msghdr { |
677 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
678 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
679 | struct iovec *free_iov; | |
03b1230c JA |
680 | struct sockaddr __user *uaddr; |
681 | struct msghdr msg; | |
b537916c | 682 | struct sockaddr_storage addr; |
03b1230c JA |
683 | }; |
684 | ||
f67676d1 JA |
685 | struct io_async_rw { |
686 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
687 | const struct iovec *free_iovec; |
688 | struct iov_iter iter; | |
227c0c96 | 689 | size_t bytes_done; |
bcf5a063 | 690 | struct wait_page_queue wpq; |
f67676d1 JA |
691 | }; |
692 | ||
6b47ee6e PB |
693 | enum { |
694 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
695 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
696 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
697 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
698 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 699 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 700 | |
6b47ee6e PB |
701 | REQ_F_FAIL_LINK_BIT, |
702 | REQ_F_INFLIGHT_BIT, | |
703 | REQ_F_CUR_POS_BIT, | |
704 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 705 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 706 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 707 | REQ_F_POLLED_BIT, |
bcda7baa | 708 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 709 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 710 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 711 | REQ_F_REISSUE_BIT, |
8c130827 | 712 | REQ_F_DONT_REISSUE_BIT, |
7b29f92d JA |
713 | /* keep async read/write and isreg together and in order */ |
714 | REQ_F_ASYNC_READ_BIT, | |
715 | REQ_F_ASYNC_WRITE_BIT, | |
716 | REQ_F_ISREG_BIT, | |
84557871 JA |
717 | |
718 | /* not a real bit, just to check we're not overflowing the space */ | |
719 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
720 | }; |
721 | ||
722 | enum { | |
723 | /* ctx owns file */ | |
724 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
725 | /* drain existing IO first */ | |
726 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
727 | /* linked sqes */ | |
728 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
729 | /* doesn't sever on completion < 0 */ | |
730 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
731 | /* IOSQE_ASYNC */ | |
732 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
733 | /* IOSQE_BUFFER_SELECT */ |
734 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 735 | |
6b47ee6e PB |
736 | /* fail rest of links */ |
737 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 738 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
739 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
740 | /* read/write uses file position */ | |
741 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
742 | /* must not punt to workers */ | |
743 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 744 | /* has or had linked timeout */ |
6b47ee6e | 745 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
746 | /* needs cleanup */ |
747 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
748 | /* already went through poll handler */ |
749 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
750 | /* buffer already selected */ |
751 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
752 | /* linked timeout is active, i.e. prepared by link's head */ |
753 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
754 | /* completion is deferred through io_comp_state */ |
755 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
756 | /* caller should reissue async */ |
757 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
758 | /* don't attempt request reissue, see io_rw_reissue() */ |
759 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
760 | /* supports async reads */ |
761 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
762 | /* supports async writes */ | |
763 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
764 | /* regular file */ | |
765 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
766 | }; |
767 | ||
768 | struct async_poll { | |
769 | struct io_poll_iocb poll; | |
807abcb0 | 770 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
771 | }; |
772 | ||
7cbf1722 JA |
773 | struct io_task_work { |
774 | struct io_wq_work_node node; | |
775 | task_work_func_t func; | |
776 | }; | |
777 | ||
09bb8394 JA |
778 | /* |
779 | * NOTE! Each of the iocb union members has the file pointer | |
780 | * as the first entry in their struct definition. So you can | |
781 | * access the file pointer through any of the sub-structs, | |
782 | * or directly as just 'ki_filp' in this struct. | |
783 | */ | |
2b188cc1 | 784 | struct io_kiocb { |
221c5eb2 | 785 | union { |
09bb8394 | 786 | struct file *file; |
9adbd45d | 787 | struct io_rw rw; |
221c5eb2 | 788 | struct io_poll_iocb poll; |
018043be | 789 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
790 | struct io_accept accept; |
791 | struct io_sync sync; | |
fbf23849 | 792 | struct io_cancel cancel; |
b29472ee | 793 | struct io_timeout timeout; |
0bdf7a2d | 794 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 795 | struct io_connect connect; |
e47293fd | 796 | struct io_sr_msg sr_msg; |
15b71abe | 797 | struct io_open open; |
b5dba59e | 798 | struct io_close close; |
269bbe5f | 799 | struct io_rsrc_update rsrc_update; |
4840e418 | 800 | struct io_fadvise fadvise; |
c1ca757b | 801 | struct io_madvise madvise; |
3e4827b0 | 802 | struct io_epoll epoll; |
7d67af2c | 803 | struct io_splice splice; |
ddf0322d | 804 | struct io_provide_buf pbuf; |
1d9e1288 | 805 | struct io_statx statx; |
36f4fa68 | 806 | struct io_shutdown shutdown; |
80a261fd | 807 | struct io_rename rename; |
14a1143b | 808 | struct io_unlink unlink; |
3ca405eb PB |
809 | /* use only after cleaning per-op data, see io_clean_op() */ |
810 | struct io_completion compl; | |
221c5eb2 | 811 | }; |
2b188cc1 | 812 | |
e8c2bc1f JA |
813 | /* opcode allocated if it needs to store data for async defer */ |
814 | void *async_data; | |
d625c6ee | 815 | u8 opcode; |
65a6543d XW |
816 | /* polled IO has completed */ |
817 | u8 iopoll_completed; | |
2b188cc1 | 818 | |
4f4eeba8 | 819 | u16 buf_index; |
9cf7c104 | 820 | u32 result; |
4f4eeba8 | 821 | |
010e8e6b PB |
822 | struct io_ring_ctx *ctx; |
823 | unsigned int flags; | |
abc54d63 | 824 | atomic_t refs; |
010e8e6b PB |
825 | struct task_struct *task; |
826 | u64 user_data; | |
d7718a9d | 827 | |
f2f87370 | 828 | struct io_kiocb *link; |
269bbe5f | 829 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 830 | |
d21ffe7e PB |
831 | /* |
832 | * 1. used with ctx->iopoll_list with reads/writes | |
833 | * 2. to track reqs with ->files (see io_op_def::file_table) | |
834 | */ | |
010e8e6b | 835 | struct list_head inflight_entry; |
7cbf1722 JA |
836 | union { |
837 | struct io_task_work io_task_work; | |
838 | struct callback_head task_work; | |
839 | }; | |
010e8e6b PB |
840 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
841 | struct hlist_node hash_node; | |
842 | struct async_poll *apoll; | |
843 | struct io_wq_work work; | |
2b188cc1 | 844 | }; |
05589553 | 845 | |
13bf43f5 PB |
846 | struct io_tctx_node { |
847 | struct list_head ctx_node; | |
848 | struct task_struct *task; | |
13bf43f5 PB |
849 | struct io_ring_ctx *ctx; |
850 | }; | |
851 | ||
27dc8338 PB |
852 | struct io_defer_entry { |
853 | struct list_head list; | |
854 | struct io_kiocb *req; | |
9cf7c104 | 855 | u32 seq; |
2b188cc1 JA |
856 | }; |
857 | ||
d3656344 | 858 | struct io_op_def { |
d3656344 JA |
859 | /* needs req->file assigned */ |
860 | unsigned needs_file : 1; | |
d3656344 JA |
861 | /* hash wq insertion if file is a regular file */ |
862 | unsigned hash_reg_file : 1; | |
863 | /* unbound wq insertion if file is a non-regular file */ | |
864 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
865 | /* opcode is not supported by this kernel */ |
866 | unsigned not_supported : 1; | |
8a72758c JA |
867 | /* set if opcode supports polled "wait" */ |
868 | unsigned pollin : 1; | |
869 | unsigned pollout : 1; | |
bcda7baa JA |
870 | /* op supports buffer selection */ |
871 | unsigned buffer_select : 1; | |
26f0505a PB |
872 | /* do prep async if is going to be punted */ |
873 | unsigned needs_async_setup : 1; | |
27926b68 JA |
874 | /* should block plug */ |
875 | unsigned plug : 1; | |
e8c2bc1f JA |
876 | /* size of async data needed, if any */ |
877 | unsigned short async_size; | |
d3656344 JA |
878 | }; |
879 | ||
0918682b | 880 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
881 | [IORING_OP_NOP] = {}, |
882 | [IORING_OP_READV] = { | |
d3656344 JA |
883 | .needs_file = 1, |
884 | .unbound_nonreg_file = 1, | |
8a72758c | 885 | .pollin = 1, |
4d954c25 | 886 | .buffer_select = 1, |
26f0505a | 887 | .needs_async_setup = 1, |
27926b68 | 888 | .plug = 1, |
e8c2bc1f | 889 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 890 | }, |
0463b6c5 | 891 | [IORING_OP_WRITEV] = { |
d3656344 JA |
892 | .needs_file = 1, |
893 | .hash_reg_file = 1, | |
894 | .unbound_nonreg_file = 1, | |
8a72758c | 895 | .pollout = 1, |
26f0505a | 896 | .needs_async_setup = 1, |
27926b68 | 897 | .plug = 1, |
e8c2bc1f | 898 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 899 | }, |
0463b6c5 | 900 | [IORING_OP_FSYNC] = { |
d3656344 JA |
901 | .needs_file = 1, |
902 | }, | |
0463b6c5 | 903 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
904 | .needs_file = 1, |
905 | .unbound_nonreg_file = 1, | |
8a72758c | 906 | .pollin = 1, |
27926b68 | 907 | .plug = 1, |
e8c2bc1f | 908 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 909 | }, |
0463b6c5 | 910 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
911 | .needs_file = 1, |
912 | .hash_reg_file = 1, | |
913 | .unbound_nonreg_file = 1, | |
8a72758c | 914 | .pollout = 1, |
27926b68 | 915 | .plug = 1, |
e8c2bc1f | 916 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 917 | }, |
0463b6c5 | 918 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
919 | .needs_file = 1, |
920 | .unbound_nonreg_file = 1, | |
921 | }, | |
0463b6c5 PB |
922 | [IORING_OP_POLL_REMOVE] = {}, |
923 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
924 | .needs_file = 1, |
925 | }, | |
0463b6c5 | 926 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
927 | .needs_file = 1, |
928 | .unbound_nonreg_file = 1, | |
8a72758c | 929 | .pollout = 1, |
26f0505a | 930 | .needs_async_setup = 1, |
e8c2bc1f | 931 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 932 | }, |
0463b6c5 | 933 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
934 | .needs_file = 1, |
935 | .unbound_nonreg_file = 1, | |
8a72758c | 936 | .pollin = 1, |
52de1fe1 | 937 | .buffer_select = 1, |
26f0505a | 938 | .needs_async_setup = 1, |
e8c2bc1f | 939 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 940 | }, |
0463b6c5 | 941 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 942 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 943 | }, |
9c8e11b3 PB |
944 | [IORING_OP_TIMEOUT_REMOVE] = { |
945 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 946 | }, |
0463b6c5 | 947 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
948 | .needs_file = 1, |
949 | .unbound_nonreg_file = 1, | |
8a72758c | 950 | .pollin = 1, |
d3656344 | 951 | }, |
0463b6c5 PB |
952 | [IORING_OP_ASYNC_CANCEL] = {}, |
953 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 954 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 955 | }, |
0463b6c5 | 956 | [IORING_OP_CONNECT] = { |
d3656344 JA |
957 | .needs_file = 1, |
958 | .unbound_nonreg_file = 1, | |
8a72758c | 959 | .pollout = 1, |
26f0505a | 960 | .needs_async_setup = 1, |
e8c2bc1f | 961 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 962 | }, |
0463b6c5 | 963 | [IORING_OP_FALLOCATE] = { |
d3656344 | 964 | .needs_file = 1, |
d3656344 | 965 | }, |
44526bed JA |
966 | [IORING_OP_OPENAT] = {}, |
967 | [IORING_OP_CLOSE] = {}, | |
968 | [IORING_OP_FILES_UPDATE] = {}, | |
969 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 970 | [IORING_OP_READ] = { |
3a6820f2 JA |
971 | .needs_file = 1, |
972 | .unbound_nonreg_file = 1, | |
8a72758c | 973 | .pollin = 1, |
bcda7baa | 974 | .buffer_select = 1, |
27926b68 | 975 | .plug = 1, |
e8c2bc1f | 976 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 977 | }, |
0463b6c5 | 978 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
979 | .needs_file = 1, |
980 | .unbound_nonreg_file = 1, | |
8a72758c | 981 | .pollout = 1, |
27926b68 | 982 | .plug = 1, |
e8c2bc1f | 983 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 984 | }, |
0463b6c5 | 985 | [IORING_OP_FADVISE] = { |
4840e418 | 986 | .needs_file = 1, |
c1ca757b | 987 | }, |
44526bed | 988 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 989 | [IORING_OP_SEND] = { |
fddaface JA |
990 | .needs_file = 1, |
991 | .unbound_nonreg_file = 1, | |
8a72758c | 992 | .pollout = 1, |
fddaface | 993 | }, |
0463b6c5 | 994 | [IORING_OP_RECV] = { |
fddaface JA |
995 | .needs_file = 1, |
996 | .unbound_nonreg_file = 1, | |
8a72758c | 997 | .pollin = 1, |
bcda7baa | 998 | .buffer_select = 1, |
fddaface | 999 | }, |
0463b6c5 | 1000 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1001 | }, |
3e4827b0 JA |
1002 | [IORING_OP_EPOLL_CTL] = { |
1003 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1004 | }, |
7d67af2c PB |
1005 | [IORING_OP_SPLICE] = { |
1006 | .needs_file = 1, | |
1007 | .hash_reg_file = 1, | |
1008 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1009 | }, |
1010 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1011 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1012 | [IORING_OP_TEE] = { |
1013 | .needs_file = 1, | |
1014 | .hash_reg_file = 1, | |
1015 | .unbound_nonreg_file = 1, | |
1016 | }, | |
36f4fa68 JA |
1017 | [IORING_OP_SHUTDOWN] = { |
1018 | .needs_file = 1, | |
1019 | }, | |
44526bed JA |
1020 | [IORING_OP_RENAMEAT] = {}, |
1021 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1022 | }; |
1023 | ||
7a612350 | 1024 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1025 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1026 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1027 | struct task_struct *task, | |
1028 | struct files_struct *files); | |
37d1e2e3 | 1029 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
b895c9a6 PB |
1030 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node); |
1031 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); | |
f2303b1f | 1032 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
1ffc5422 | 1033 | |
78e19bbe | 1034 | static void io_cqring_fill_event(struct io_kiocb *req, long res); |
ec9c02ad | 1035 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1036 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1037 | static void io_dismantle_req(struct io_kiocb *req); |
1038 | static void io_put_task(struct task_struct *task, int nr); | |
1039 | static void io_queue_next(struct io_kiocb *req); | |
94ae5e77 JA |
1040 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1041 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c | 1042 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1043 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1044 | unsigned nr_args); |
68fb8979 | 1045 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1046 | static struct file *io_file_get(struct io_submit_state *state, |
1047 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1048 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1049 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1050 | |
907d1df3 | 1051 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1052 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1053 | struct io_ring_ctx *ctx); | |
5082620f | 1054 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1055 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1056 | |
2b188cc1 JA |
1057 | static struct kmem_cache *req_cachep; |
1058 | ||
0918682b | 1059 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1060 | |
1061 | struct sock *io_uring_get_socket(struct file *file) | |
1062 | { | |
1063 | #if defined(CONFIG_UNIX) | |
1064 | if (file->f_op == &io_uring_fops) { | |
1065 | struct io_ring_ctx *ctx = file->private_data; | |
1066 | ||
1067 | return ctx->ring_sock->sk; | |
1068 | } | |
1069 | #endif | |
1070 | return NULL; | |
1071 | } | |
1072 | EXPORT_SYMBOL(io_uring_get_socket); | |
1073 | ||
f2f87370 PB |
1074 | #define io_for_each_link(pos, head) \ |
1075 | for (pos = (head); pos; pos = pos->link) | |
1076 | ||
b895c9a6 | 1077 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1078 | { |
1079 | struct io_ring_ctx *ctx = req->ctx; | |
1080 | ||
269bbe5f BM |
1081 | if (!req->fixed_rsrc_refs) { |
1082 | req->fixed_rsrc_refs = &ctx->file_data->node->refs; | |
1083 | percpu_ref_get(req->fixed_rsrc_refs); | |
36f72fe2 PB |
1084 | } |
1085 | } | |
1086 | ||
08d23634 PB |
1087 | static bool io_match_task(struct io_kiocb *head, |
1088 | struct task_struct *task, | |
1089 | struct files_struct *files) | |
1090 | { | |
1091 | struct io_kiocb *req; | |
1092 | ||
68207680 | 1093 | if (task && head->task != task) |
08d23634 PB |
1094 | return false; |
1095 | if (!files) | |
1096 | return true; | |
1097 | ||
1098 | io_for_each_link(req, head) { | |
b05a1bcd | 1099 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1100 | return true; |
08d23634 PB |
1101 | } |
1102 | return false; | |
1103 | } | |
1104 | ||
c40f6379 JA |
1105 | static inline void req_set_fail_links(struct io_kiocb *req) |
1106 | { | |
1107 | if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK) | |
1108 | req->flags |= REQ_F_FAIL_LINK; | |
1109 | } | |
4a38aed2 | 1110 | |
2b188cc1 JA |
1111 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1112 | { | |
1113 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1114 | ||
0f158b4c | 1115 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1116 | } |
1117 | ||
8eb7e2d0 PB |
1118 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1119 | { | |
1120 | return !req->timeout.off; | |
1121 | } | |
1122 | ||
2b188cc1 JA |
1123 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1124 | { | |
1125 | struct io_ring_ctx *ctx; | |
78076bb6 | 1126 | int hash_bits; |
2b188cc1 JA |
1127 | |
1128 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1129 | if (!ctx) | |
1130 | return NULL; | |
1131 | ||
78076bb6 JA |
1132 | /* |
1133 | * Use 5 bits less than the max cq entries, that should give us around | |
1134 | * 32 entries per hash list if totally full and uniformly spread. | |
1135 | */ | |
1136 | hash_bits = ilog2(p->cq_entries); | |
1137 | hash_bits -= 5; | |
1138 | if (hash_bits <= 0) | |
1139 | hash_bits = 1; | |
1140 | ctx->cancel_hash_bits = hash_bits; | |
1141 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1142 | GFP_KERNEL); | |
1143 | if (!ctx->cancel_hash) | |
1144 | goto err; | |
1145 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1146 | ||
21482896 | 1147 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1148 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1149 | goto err; | |
2b188cc1 JA |
1150 | |
1151 | ctx->flags = p->flags; | |
90554200 | 1152 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1153 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1154 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1155 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1156 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1157 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1158 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1159 | mutex_init(&ctx->uring_lock); |
1160 | init_waitqueue_head(&ctx->wait); | |
1161 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1162 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1163 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1164 | INIT_LIST_HEAD(&ctx->timeout_list); |
fcb323cc JA |
1165 | spin_lock_init(&ctx->inflight_lock); |
1166 | INIT_LIST_HEAD(&ctx->inflight_list); | |
d67d2263 BM |
1167 | spin_lock_init(&ctx->rsrc_ref_lock); |
1168 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1169 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1170 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1171 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1172 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1173 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1174 | return ctx; |
206aefde | 1175 | err: |
78076bb6 | 1176 | kfree(ctx->cancel_hash); |
206aefde JA |
1177 | kfree(ctx); |
1178 | return NULL; | |
2b188cc1 JA |
1179 | } |
1180 | ||
9cf7c104 | 1181 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1182 | { |
2bc9930e JA |
1183 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1184 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1185 | |
9cf7c104 | 1186 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1187 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1188 | } |
de0617e4 | 1189 | |
9d858b21 | 1190 | return false; |
de0617e4 JA |
1191 | } |
1192 | ||
ce3d5aae PB |
1193 | static void io_req_track_inflight(struct io_kiocb *req) |
1194 | { | |
1195 | struct io_ring_ctx *ctx = req->ctx; | |
1196 | ||
1197 | if (!(req->flags & REQ_F_INFLIGHT)) { | |
ce3d5aae PB |
1198 | req->flags |= REQ_F_INFLIGHT; |
1199 | ||
1200 | spin_lock_irq(&ctx->inflight_lock); | |
1201 | list_add(&req->inflight_entry, &ctx->inflight_list); | |
1202 | spin_unlock_irq(&ctx->inflight_lock); | |
1203 | } | |
1204 | } | |
1205 | ||
1e6fa521 JA |
1206 | static void io_prep_async_work(struct io_kiocb *req) |
1207 | { | |
1208 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1209 | struct io_ring_ctx *ctx = req->ctx; |
1210 | ||
003e8dcc JA |
1211 | if (!req->work.creds) |
1212 | req->work.creds = get_current_cred(); | |
1213 | ||
e1d675df PB |
1214 | req->work.list.next = NULL; |
1215 | req->work.flags = 0; | |
feaadc4f PB |
1216 | if (req->flags & REQ_F_FORCE_ASYNC) |
1217 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1218 | ||
1e6fa521 JA |
1219 | if (req->flags & REQ_F_ISREG) { |
1220 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1221 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1222 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1223 | if (def->unbound_nonreg_file) |
1224 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1225 | } | |
e1d675df PB |
1226 | |
1227 | switch (req->opcode) { | |
1228 | case IORING_OP_SPLICE: | |
1229 | case IORING_OP_TEE: | |
1230 | /* | |
1231 | * Splice operation will be punted aync, and here need to | |
1232 | * modify io_wq_work.flags, so initialize io_wq_work firstly. | |
1233 | */ | |
1234 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) | |
1235 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1236 | break; | |
1237 | } | |
561fb04a | 1238 | } |
cccf0ee8 | 1239 | |
cbdcb435 | 1240 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1241 | { |
cbdcb435 | 1242 | struct io_kiocb *cur; |
54a91f3b | 1243 | |
f2f87370 PB |
1244 | io_for_each_link(cur, req) |
1245 | io_prep_async_work(cur); | |
561fb04a JA |
1246 | } |
1247 | ||
ebf93667 | 1248 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1249 | { |
a197f664 | 1250 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1251 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1252 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1253 | |
3bfe6106 JA |
1254 | BUG_ON(!tctx); |
1255 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1256 | |
cbdcb435 PB |
1257 | /* init ->work of the whole link before punting */ |
1258 | io_prep_async_link(req); | |
d07f1e8a PB |
1259 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1260 | &req->work, req->flags); | |
ebf93667 | 1261 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1262 | if (link) |
1263 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1264 | } |
1265 | ||
1ee4160c | 1266 | static void io_kill_timeout(struct io_kiocb *req, int status) |
5262f567 | 1267 | { |
e8c2bc1f | 1268 | struct io_timeout_data *io = req->async_data; |
5262f567 JA |
1269 | int ret; |
1270 | ||
e8c2bc1f | 1271 | ret = hrtimer_try_to_cancel(&io->timer); |
5262f567 | 1272 | if (ret != -1) { |
01cec8c1 PB |
1273 | atomic_set(&req->ctx->cq_timeouts, |
1274 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1275 | list_del_init(&req->timeout.list); |
1ee4160c | 1276 | io_cqring_fill_event(req, status); |
216578e5 | 1277 | io_put_req_deferred(req, 1); |
5262f567 JA |
1278 | } |
1279 | } | |
1280 | ||
04518945 | 1281 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1282 | { |
04518945 | 1283 | do { |
27dc8338 PB |
1284 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1285 | struct io_defer_entry, list); | |
de0617e4 | 1286 | |
9cf7c104 | 1287 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1288 | break; |
27dc8338 | 1289 | list_del_init(&de->list); |
907d1df3 | 1290 | io_req_task_queue(de->req); |
27dc8338 | 1291 | kfree(de); |
04518945 PB |
1292 | } while (!list_empty(&ctx->defer_list)); |
1293 | } | |
1294 | ||
360428f8 | 1295 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1296 | { |
f010505b MDG |
1297 | u32 seq; |
1298 | ||
1299 | if (list_empty(&ctx->timeout_list)) | |
1300 | return; | |
1301 | ||
1302 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1303 | ||
1304 | do { | |
1305 | u32 events_needed, events_got; | |
360428f8 | 1306 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1307 | struct io_kiocb, timeout.list); |
de0617e4 | 1308 | |
8eb7e2d0 | 1309 | if (io_is_timeout_noseq(req)) |
360428f8 | 1310 | break; |
f010505b MDG |
1311 | |
1312 | /* | |
1313 | * Since seq can easily wrap around over time, subtract | |
1314 | * the last seq at which timeouts were flushed before comparing. | |
1315 | * Assuming not more than 2^31-1 events have happened since, | |
1316 | * these subtractions won't have wrapped, so we can check if | |
1317 | * target is in [last_seq, current_seq] by comparing the two. | |
1318 | */ | |
1319 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1320 | events_got = seq - ctx->cq_last_tm_flush; | |
1321 | if (events_got < events_needed) | |
360428f8 | 1322 | break; |
bfe68a22 | 1323 | |
135fcde8 | 1324 | list_del_init(&req->timeout.list); |
1ee4160c | 1325 | io_kill_timeout(req, 0); |
f010505b MDG |
1326 | } while (!list_empty(&ctx->timeout_list)); |
1327 | ||
1328 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1329 | } |
5262f567 | 1330 | |
360428f8 PB |
1331 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1332 | { | |
1333 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1334 | |
1335 | /* order cqe stores with ring update */ | |
1336 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1337 | |
04518945 PB |
1338 | if (unlikely(!list_empty(&ctx->defer_list))) |
1339 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1340 | } |
1341 | ||
90554200 JA |
1342 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1343 | { | |
1344 | struct io_rings *r = ctx->rings; | |
1345 | ||
1346 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1347 | } | |
1348 | ||
888aae2e PB |
1349 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1350 | { | |
1351 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1352 | } | |
1353 | ||
2b188cc1 JA |
1354 | static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
1355 | { | |
75b28aff | 1356 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1357 | unsigned tail; |
1358 | ||
115e12e5 SB |
1359 | /* |
1360 | * writes to the cq entry need to come after reading head; the | |
1361 | * control dependency is enough as we're using WRITE_ONCE to | |
1362 | * fill the cq entry | |
1363 | */ | |
888aae2e | 1364 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1365 | return NULL; |
1366 | ||
888aae2e | 1367 | tail = ctx->cached_cq_tail++; |
75b28aff | 1368 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1369 | } |
1370 | ||
f2842ab5 JA |
1371 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1372 | { | |
f0b493e6 JA |
1373 | if (!ctx->cq_ev_fd) |
1374 | return false; | |
7e55a19c SG |
1375 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1376 | return false; | |
f2842ab5 JA |
1377 | if (!ctx->eventfd_async) |
1378 | return true; | |
b41e9852 | 1379 | return io_wq_current_is_worker(); |
f2842ab5 JA |
1380 | } |
1381 | ||
b41e9852 | 1382 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1383 | { |
b1445e59 PB |
1384 | /* see waitqueue_active() comment */ |
1385 | smp_mb(); | |
1386 | ||
1d7bb1d5 JA |
1387 | if (waitqueue_active(&ctx->wait)) |
1388 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1389 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1390 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1391 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1392 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1393 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1394 | wake_up_interruptible(&ctx->cq_wait); |
1395 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1396 | } | |
1d7bb1d5 JA |
1397 | } |
1398 | ||
80c18e4a PB |
1399 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1400 | { | |
b1445e59 PB |
1401 | /* see waitqueue_active() comment */ |
1402 | smp_mb(); | |
1403 | ||
80c18e4a PB |
1404 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1405 | if (waitqueue_active(&ctx->wait)) | |
1406 | wake_up(&ctx->wait); | |
1407 | } | |
1408 | if (io_should_trigger_evfd(ctx)) | |
1409 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1410 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1411 | wake_up_interruptible(&ctx->cq_wait); |
1412 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1413 | } | |
80c18e4a PB |
1414 | } |
1415 | ||
c4a2ed72 | 1416 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1417 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 JA |
1418 | { |
1419 | struct io_rings *rings = ctx->rings; | |
1d7bb1d5 | 1420 | unsigned long flags; |
b18032bb | 1421 | bool all_flushed, posted; |
1d7bb1d5 | 1422 | |
e23de15f PB |
1423 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1424 | return false; | |
1d7bb1d5 | 1425 | |
b18032bb | 1426 | posted = false; |
1d7bb1d5 | 1427 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae PB |
1428 | while (!list_empty(&ctx->cq_overflow_list)) { |
1429 | struct io_uring_cqe *cqe = io_get_cqring(ctx); | |
1430 | struct io_overflow_cqe *ocqe; | |
e6c8aa9a | 1431 | |
1d7bb1d5 JA |
1432 | if (!cqe && !force) |
1433 | break; | |
6c2450ae PB |
1434 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1435 | struct io_overflow_cqe, list); | |
1436 | if (cqe) | |
1437 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1438 | else | |
1d7bb1d5 | 1439 | WRITE_ONCE(ctx->rings->cq_overflow, |
6c2450ae | 1440 | ++ctx->cached_cq_overflow); |
b18032bb | 1441 | posted = true; |
6c2450ae PB |
1442 | list_del(&ocqe->list); |
1443 | kfree(ocqe); | |
1d7bb1d5 JA |
1444 | } |
1445 | ||
09e88404 PB |
1446 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1447 | if (all_flushed) { | |
1448 | clear_bit(0, &ctx->sq_check_overflow); | |
1449 | clear_bit(0, &ctx->cq_check_overflow); | |
1450 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1451 | } | |
46930143 | 1452 | |
b18032bb JA |
1453 | if (posted) |
1454 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1455 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1456 | if (posted) |
1457 | io_cqring_ev_posted(ctx); | |
09e88404 | 1458 | return all_flushed; |
1d7bb1d5 JA |
1459 | } |
1460 | ||
6c2450ae | 1461 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1462 | { |
ca0a2651 JA |
1463 | bool ret = true; |
1464 | ||
6c503150 PB |
1465 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1466 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1467 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1468 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1469 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1470 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1471 | mutex_unlock(&ctx->uring_lock); | |
1472 | } | |
ca0a2651 JA |
1473 | |
1474 | return ret; | |
6c503150 PB |
1475 | } |
1476 | ||
abc54d63 JA |
1477 | /* |
1478 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1479 | * see commit f958d7b528b1 for details. | |
1480 | */ | |
1481 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1482 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1483 | ||
de9b4cca JA |
1484 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1485 | { | |
abc54d63 | 1486 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1487 | } |
1488 | ||
1489 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1490 | { | |
abc54d63 JA |
1491 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1492 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1493 | } |
1494 | ||
1495 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1496 | { | |
abc54d63 JA |
1497 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1498 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1499 | } |
1500 | ||
1501 | static inline void req_ref_put(struct io_kiocb *req) | |
1502 | { | |
abc54d63 | 1503 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1504 | } |
1505 | ||
1506 | static inline void req_ref_get(struct io_kiocb *req) | |
1507 | { | |
abc54d63 JA |
1508 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1509 | atomic_inc(&req->refs); | |
de9b4cca JA |
1510 | } |
1511 | ||
5082620f | 1512 | static bool __io_cqring_fill_event(struct io_kiocb *req, long res, |
8c3f9cd1 | 1513 | unsigned int cflags) |
2b188cc1 | 1514 | { |
78e19bbe | 1515 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1516 | struct io_uring_cqe *cqe; |
1517 | ||
7471e1af | 1518 | trace_io_uring_complete(ctx, req->user_data, res, cflags); |
51c3ff62 | 1519 | |
2b188cc1 JA |
1520 | /* |
1521 | * If we can't get a cq entry, userspace overflowed the | |
1522 | * submission (by quite a lot). Increment the overflow count in | |
1523 | * the ring. | |
1524 | */ | |
1525 | cqe = io_get_cqring(ctx); | |
1d7bb1d5 | 1526 | if (likely(cqe)) { |
78e19bbe | 1527 | WRITE_ONCE(cqe->user_data, req->user_data); |
2b188cc1 | 1528 | WRITE_ONCE(cqe->res, res); |
bcda7baa | 1529 | WRITE_ONCE(cqe->flags, cflags); |
5082620f | 1530 | return true; |
6c2450ae PB |
1531 | } |
1532 | if (!ctx->cq_overflow_flushed && | |
1533 | !atomic_read(&req->task->io_uring->in_idle)) { | |
1534 | struct io_overflow_cqe *ocqe; | |
1535 | ||
1536 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); | |
1537 | if (!ocqe) | |
1538 | goto overflow; | |
ad3eb2c8 JA |
1539 | if (list_empty(&ctx->cq_overflow_list)) { |
1540 | set_bit(0, &ctx->sq_check_overflow); | |
1541 | set_bit(0, &ctx->cq_check_overflow); | |
6d5f9049 | 1542 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; |
ad3eb2c8 | 1543 | } |
6c2450ae PB |
1544 | ocqe->cqe.user_data = req->user_data; |
1545 | ocqe->cqe.res = res; | |
1546 | ocqe->cqe.flags = cflags; | |
1547 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
5082620f | 1548 | return true; |
2b188cc1 | 1549 | } |
6c2450ae PB |
1550 | overflow: |
1551 | /* | |
1552 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1553 | * or cannot allocate an overflow entry, then we need to drop it | |
1554 | * on the floor. | |
1555 | */ | |
1556 | WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow); | |
5082620f | 1557 | return false; |
2b188cc1 JA |
1558 | } |
1559 | ||
bcda7baa JA |
1560 | static void io_cqring_fill_event(struct io_kiocb *req, long res) |
1561 | { | |
1562 | __io_cqring_fill_event(req, res, 0); | |
1563 | } | |
1564 | ||
7a612350 PB |
1565 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1566 | unsigned int cflags) | |
2b188cc1 | 1567 | { |
78e19bbe | 1568 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1569 | unsigned long flags; |
1570 | ||
1571 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
bcda7baa | 1572 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1573 | /* |
1574 | * If we're the last reference to this request, add to our locked | |
1575 | * free_list cache. | |
1576 | */ | |
de9b4cca | 1577 | if (req_ref_put_and_test(req)) { |
c7dae4ba JA |
1578 | struct io_comp_state *cs = &ctx->submit_state.comp; |
1579 | ||
7a612350 PB |
1580 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1581 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1582 | io_disarm_next(req); | |
1583 | if (req->link) { | |
1584 | io_req_task_queue(req->link); | |
1585 | req->link = NULL; | |
1586 | } | |
1587 | } | |
c7dae4ba JA |
1588 | io_dismantle_req(req); |
1589 | io_put_task(req->task, 1); | |
1590 | list_add(&req->compl.list, &cs->locked_free_list); | |
1591 | cs->locked_free_nr++; | |
180f829f PB |
1592 | } else { |
1593 | if (!percpu_ref_tryget(&ctx->refs)) | |
1594 | req = NULL; | |
1595 | } | |
7a612350 | 1596 | io_commit_cqring(ctx); |
2b188cc1 | 1597 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1598 | |
180f829f PB |
1599 | if (req) { |
1600 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1601 | percpu_ref_put(&ctx->refs); |
180f829f | 1602 | } |
229a7b63 JA |
1603 | } |
1604 | ||
a38d68db | 1605 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1606 | unsigned int cflags) |
229a7b63 | 1607 | { |
68fb8979 PB |
1608 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
1609 | io_clean_op(req); | |
a38d68db PB |
1610 | req->result = res; |
1611 | req->compl.cflags = cflags; | |
e342c807 | 1612 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1613 | } |
1614 | ||
889fca73 PB |
1615 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1616 | long res, unsigned cflags) | |
bcda7baa | 1617 | { |
889fca73 PB |
1618 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1619 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1620 | else |
c7dae4ba | 1621 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1622 | } |
1623 | ||
a38d68db | 1624 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1625 | { |
889fca73 | 1626 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1627 | } |
1628 | ||
f41db273 PB |
1629 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1630 | { | |
1631 | req_set_fail_links(req); | |
1632 | io_put_req(req); | |
1633 | io_req_complete_post(req, res, 0); | |
1634 | } | |
1635 | ||
dac7a098 PB |
1636 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1637 | struct io_comp_state *cs) | |
1638 | { | |
1639 | spin_lock_irq(&ctx->completion_lock); | |
1640 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1641 | cs->locked_free_nr = 0; | |
1642 | spin_unlock_irq(&ctx->completion_lock); | |
1643 | } | |
1644 | ||
dd78f492 | 1645 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1646 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1647 | { |
c7dae4ba JA |
1648 | struct io_submit_state *state = &ctx->submit_state; |
1649 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1650 | int nr; |
0ddf92e8 | 1651 | |
c7dae4ba JA |
1652 | /* |
1653 | * If we have more than a batch's worth of requests in our IRQ side | |
1654 | * locked cache, grab the lock and move them over to our submission | |
1655 | * side cache. | |
1656 | */ | |
dac7a098 PB |
1657 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) |
1658 | io_flush_cached_locked_reqs(ctx, cs); | |
0ddf92e8 | 1659 | |
dd78f492 | 1660 | nr = state->free_reqs; |
c7dae4ba | 1661 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1662 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1663 | struct io_kiocb, compl.list); | |
1664 | ||
1b4c351f | 1665 | list_del(&req->compl.list); |
dd78f492 PB |
1666 | state->reqs[nr++] = req; |
1667 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1668 | break; |
1b4c351f JA |
1669 | } |
1670 | ||
dd78f492 PB |
1671 | state->free_reqs = nr; |
1672 | return nr != 0; | |
0ddf92e8 JA |
1673 | } |
1674 | ||
e5d1bc0a | 1675 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1676 | { |
e5d1bc0a PB |
1677 | struct io_submit_state *state = &ctx->submit_state; |
1678 | ||
1679 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1680 | ||
f6b6c7d6 | 1681 | if (!state->free_reqs) { |
291b2821 | 1682 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1683 | int ret; |
1684 | ||
c7dae4ba | 1685 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1686 | goto got_req; |
1687 | ||
bf019da7 PB |
1688 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1689 | state->reqs); | |
fd6fab2c JA |
1690 | |
1691 | /* | |
1692 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1693 | * retry single alloc to be on the safe side. | |
1694 | */ | |
1695 | if (unlikely(ret <= 0)) { | |
1696 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1697 | if (!state->reqs[0]) | |
3893f39f | 1698 | return NULL; |
fd6fab2c JA |
1699 | ret = 1; |
1700 | } | |
291b2821 | 1701 | state->free_reqs = ret; |
2b188cc1 | 1702 | } |
e5d1bc0a | 1703 | got_req: |
291b2821 PB |
1704 | state->free_reqs--; |
1705 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1706 | } |
1707 | ||
e1d767f0 | 1708 | static inline void io_put_file(struct file *file) |
8da11c19 | 1709 | { |
e1d767f0 | 1710 | if (file) |
8da11c19 PB |
1711 | fput(file); |
1712 | } | |
1713 | ||
4edf20f9 | 1714 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1715 | { |
094bae49 PB |
1716 | unsigned int flags = req->flags; |
1717 | ||
e1d767f0 PB |
1718 | if (!(flags & REQ_F_FIXED_FILE)) |
1719 | io_put_file(req->file); | |
094bae49 PB |
1720 | if (flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
1721 | REQ_F_INFLIGHT)) { | |
1722 | io_clean_op(req); | |
1723 | ||
1724 | if (req->flags & REQ_F_INFLIGHT) { | |
1725 | struct io_ring_ctx *ctx = req->ctx; | |
1726 | unsigned long flags; | |
1727 | ||
1728 | spin_lock_irqsave(&ctx->inflight_lock, flags); | |
1729 | list_del(&req->inflight_entry); | |
1730 | spin_unlock_irqrestore(&ctx->inflight_lock, flags); | |
1731 | req->flags &= ~REQ_F_INFLIGHT; | |
1732 | } | |
1733 | } | |
269bbe5f BM |
1734 | if (req->fixed_rsrc_refs) |
1735 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1736 | if (req->async_data) |
1737 | kfree(req->async_data); | |
003e8dcc JA |
1738 | if (req->work.creds) { |
1739 | put_cred(req->work.creds); | |
1740 | req->work.creds = NULL; | |
1741 | } | |
e65ef56d JA |
1742 | } |
1743 | ||
b23fcf47 | 1744 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1745 | static inline void io_put_task(struct task_struct *task, int nr) |
1746 | { | |
1747 | struct io_uring_task *tctx = task->io_uring; | |
1748 | ||
1749 | percpu_counter_sub(&tctx->inflight, nr); | |
1750 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1751 | wake_up(&tctx->wait); | |
1752 | put_task_struct_many(task, nr); | |
1753 | } | |
1754 | ||
216578e5 | 1755 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1756 | { |
51a4cc11 | 1757 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1758 | |
216578e5 | 1759 | io_dismantle_req(req); |
7c660731 | 1760 | io_put_task(req->task, 1); |
c6ca97b3 | 1761 | |
3893f39f | 1762 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1763 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1764 | } |
1765 | ||
f2f87370 PB |
1766 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1767 | { | |
1768 | struct io_kiocb *nxt = req->link; | |
1769 | ||
1770 | req->link = nxt->link; | |
1771 | nxt->link = NULL; | |
1772 | } | |
1773 | ||
33cc89a9 PB |
1774 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1775 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1776 | { |
33cc89a9 | 1777 | struct io_kiocb *link = req->link; |
f2f87370 | 1778 | |
900fad45 PB |
1779 | /* |
1780 | * Can happen if a linked timeout fired and link had been like | |
1781 | * req -> link t-out -> link t-out [-> ...] | |
1782 | */ | |
c9abd7ad PB |
1783 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1784 | struct io_timeout_data *io = link->async_data; | |
1785 | int ret; | |
7c86ffee | 1786 | |
f2f87370 | 1787 | io_remove_next_linked(req); |
90cd7e42 | 1788 | link->timeout.head = NULL; |
c9abd7ad PB |
1789 | ret = hrtimer_try_to_cancel(&io->timer); |
1790 | if (ret != -1) { | |
1791 | io_cqring_fill_event(link, -ECANCELED); | |
33cc89a9 | 1792 | io_put_req_deferred(link, 1); |
d4729fbd | 1793 | return true; |
c9abd7ad PB |
1794 | } |
1795 | } | |
d4729fbd | 1796 | return false; |
7c86ffee PB |
1797 | } |
1798 | ||
d148ca4b | 1799 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1800 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1801 | { |
33cc89a9 | 1802 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1803 | |
f2f87370 | 1804 | req->link = NULL; |
f2f87370 PB |
1805 | while (link) { |
1806 | nxt = link->link; | |
1807 | link->link = NULL; | |
2665abfd | 1808 | |
f2f87370 | 1809 | trace_io_uring_fail_link(req, link); |
7c86ffee | 1810 | io_cqring_fill_event(link, -ECANCELED); |
1575f21a | 1811 | io_put_req_deferred(link, 2); |
f2f87370 | 1812 | link = nxt; |
9e645e11 | 1813 | } |
33cc89a9 | 1814 | } |
9e645e11 | 1815 | |
33cc89a9 PB |
1816 | static bool io_disarm_next(struct io_kiocb *req) |
1817 | __must_hold(&req->ctx->completion_lock) | |
1818 | { | |
1819 | bool posted = false; | |
1820 | ||
1821 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1822 | posted = io_kill_linked_timeout(req); | |
1823 | if (unlikely(req->flags & REQ_F_FAIL_LINK)) { | |
1824 | posted |= (req->link != NULL); | |
1825 | io_fail_links(req); | |
1826 | } | |
1827 | return posted; | |
9e645e11 JA |
1828 | } |
1829 | ||
3fa5e0f3 | 1830 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1831 | { |
33cc89a9 | 1832 | struct io_kiocb *nxt; |
944e58bf | 1833 | |
9e645e11 JA |
1834 | /* |
1835 | * If LINK is set, we have dependent requests in this chain. If we | |
1836 | * didn't fail this request, queue the first one up, moving any other | |
1837 | * dependencies to the next request. In case of failure, fail the rest | |
1838 | * of the chain. | |
1839 | */ | |
33cc89a9 PB |
1840 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1841 | struct io_ring_ctx *ctx = req->ctx; | |
1842 | unsigned long flags; | |
1843 | bool posted; | |
1844 | ||
1845 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1846 | posted = io_disarm_next(req); | |
1847 | if (posted) | |
1848 | io_commit_cqring(req->ctx); | |
1849 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1850 | if (posted) | |
1851 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1852 | } |
33cc89a9 PB |
1853 | nxt = req->link; |
1854 | req->link = NULL; | |
1855 | return nxt; | |
4d7dd462 | 1856 | } |
9e645e11 | 1857 | |
f2f87370 | 1858 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1859 | { |
cdbff982 | 1860 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1861 | return NULL; |
1862 | return __io_req_find_next(req); | |
1863 | } | |
1864 | ||
2c32395d PB |
1865 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1866 | { | |
1867 | if (!ctx) | |
1868 | return; | |
1869 | if (ctx->submit_state.comp.nr) { | |
1870 | mutex_lock(&ctx->uring_lock); | |
1871 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1872 | mutex_unlock(&ctx->uring_lock); | |
1873 | } | |
1874 | percpu_ref_put(&ctx->refs); | |
1875 | } | |
1876 | ||
7cbf1722 | 1877 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1878 | { |
65453d1e | 1879 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1880 | struct io_wq_work_list list; |
1881 | struct io_wq_work_node *node; | |
c2c4c83c | 1882 | |
7cbf1722 JA |
1883 | if (wq_list_empty(&tctx->task_list)) |
1884 | return false; | |
6200b0ae | 1885 | |
0b81e80c | 1886 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1887 | list = tctx->task_list; |
1888 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1889 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1890 | |
7cbf1722 JA |
1891 | node = list.first; |
1892 | while (node) { | |
1893 | struct io_wq_work_node *next = node->next; | |
1894 | struct io_kiocb *req; | |
0ba9c9ed | 1895 | |
7cbf1722 | 1896 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1897 | if (req->ctx != ctx) { |
1898 | ctx_flush_and_put(ctx); | |
1899 | ctx = req->ctx; | |
1900 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1901 | } |
65453d1e | 1902 | |
2c32395d PB |
1903 | req->task_work.func(&req->task_work); |
1904 | node = next; | |
7cbf1722 JA |
1905 | } |
1906 | ||
2c32395d | 1907 | ctx_flush_and_put(ctx); |
7cbf1722 | 1908 | return list.first != NULL; |
c2c4c83c JA |
1909 | } |
1910 | ||
7cbf1722 | 1911 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1912 | { |
7cbf1722 | 1913 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1914 | |
1d5f360d JA |
1915 | clear_bit(0, &tctx->task_state); |
1916 | ||
7cbf1722 JA |
1917 | while (__tctx_task_work(tctx)) |
1918 | cond_resched(); | |
7cbf1722 JA |
1919 | } |
1920 | ||
c15b79de | 1921 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1922 | { |
c15b79de | 1923 | struct task_struct *tsk = req->task; |
7cbf1722 | 1924 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1925 | enum task_work_notify_mode notify; |
7cbf1722 | 1926 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1927 | unsigned long flags; |
c15b79de PB |
1928 | int ret = 0; |
1929 | ||
1930 | if (unlikely(tsk->flags & PF_EXITING)) | |
1931 | return -ESRCH; | |
7cbf1722 JA |
1932 | |
1933 | WARN_ON_ONCE(!tctx); | |
1934 | ||
0b81e80c | 1935 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1936 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1937 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1938 | |
1939 | /* task_work already pending, we're done */ | |
1940 | if (test_bit(0, &tctx->task_state) || | |
1941 | test_and_set_bit(0, &tctx->task_state)) | |
1942 | return 0; | |
1943 | ||
c15b79de PB |
1944 | /* |
1945 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1946 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1947 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1948 | * will do the job. | |
1949 | */ | |
1950 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1951 | ||
1952 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1953 | wake_up_process(tsk); | |
7cbf1722 | 1954 | return 0; |
c15b79de | 1955 | } |
7cbf1722 JA |
1956 | |
1957 | /* | |
1958 | * Slow path - we failed, find and delete work. if the work is not | |
1959 | * in the list, it got run and we're fine. | |
1960 | */ | |
0b81e80c | 1961 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1962 | wq_list_for_each(node, prev, &tctx->task_list) { |
1963 | if (&req->io_task_work.node == node) { | |
1964 | wq_list_del(&tctx->task_list, node, prev); | |
1965 | ret = 1; | |
1966 | break; | |
1967 | } | |
1968 | } | |
0b81e80c | 1969 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1970 | clear_bit(0, &tctx->task_state); |
1971 | return ret; | |
1972 | } | |
1973 | ||
9b465711 PB |
1974 | static bool io_run_task_work_head(struct callback_head **work_head) |
1975 | { | |
1976 | struct callback_head *work, *next; | |
1977 | bool executed = false; | |
1978 | ||
1979 | do { | |
1980 | work = xchg(work_head, NULL); | |
1981 | if (!work) | |
1982 | break; | |
1983 | ||
1984 | do { | |
1985 | next = work->next; | |
1986 | work->func(work); | |
1987 | work = next; | |
1988 | cond_resched(); | |
1989 | } while (work); | |
1990 | executed = true; | |
1991 | } while (1); | |
1992 | ||
1993 | return executed; | |
1994 | } | |
1995 | ||
1996 | static void io_task_work_add_head(struct callback_head **work_head, | |
1997 | struct callback_head *task_work) | |
eab30c4d | 1998 | { |
7c25c0d1 | 1999 | struct callback_head *head; |
eab30c4d | 2000 | |
7c25c0d1 | 2001 | do { |
9b465711 PB |
2002 | head = READ_ONCE(*work_head); |
2003 | task_work->next = head; | |
2004 | } while (cmpxchg(work_head, head, task_work) != head); | |
2005 | } | |
2006 | ||
2007 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2008 | task_work_func_t cb) | |
2009 | { | |
2010 | init_task_work(&req->task_work, cb); | |
2011 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2012 | } |
2013 | ||
c40f6379 JA |
2014 | static void io_req_task_cancel(struct callback_head *cb) |
2015 | { | |
2016 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2017 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2018 | |
e83acd7d | 2019 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2020 | mutex_lock(&ctx->uring_lock); |
2593553a | 2021 | io_req_complete_failed(req, req->result); |
792bb6eb | 2022 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2023 | } |
2024 | ||
2025 | static void __io_req_task_submit(struct io_kiocb *req) | |
2026 | { | |
2027 | struct io_ring_ctx *ctx = req->ctx; | |
2028 | ||
04fc6c80 | 2029 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2030 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2031 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2032 | __io_queue_sqe(req); |
81b6d05c | 2033 | else |
2593553a | 2034 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2035 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2036 | } |
2037 | ||
2038 | static void io_req_task_submit(struct callback_head *cb) | |
2039 | { | |
2040 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2041 | ||
2042 | __io_req_task_submit(req); | |
2043 | } | |
2044 | ||
2c4b8eb6 | 2045 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2046 | { |
2c4b8eb6 PB |
2047 | req->result = ret; |
2048 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2049 | |
2c4b8eb6 | 2050 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2051 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2052 | } |
2053 | ||
2c4b8eb6 | 2054 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2055 | { |
2c4b8eb6 | 2056 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2057 | |
2058 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2059 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2060 | } |
2061 | ||
f2f87370 | 2062 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2063 | { |
9b5f7bd9 | 2064 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2065 | |
2066 | if (nxt) | |
906a8c3f | 2067 | io_req_task_queue(nxt); |
c69f8dbe JL |
2068 | } |
2069 | ||
c3524383 | 2070 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2071 | { |
c3524383 PB |
2072 | io_queue_next(req); |
2073 | __io_free_req(req); | |
2074 | } | |
8766dd51 | 2075 | |
2d6500d4 | 2076 | struct req_batch { |
5af1d13e PB |
2077 | struct task_struct *task; |
2078 | int task_refs; | |
1b4c351f | 2079 | int ctx_refs; |
2d6500d4 PB |
2080 | }; |
2081 | ||
5af1d13e PB |
2082 | static inline void io_init_req_batch(struct req_batch *rb) |
2083 | { | |
5af1d13e | 2084 | rb->task_refs = 0; |
9ae72463 | 2085 | rb->ctx_refs = 0; |
5af1d13e PB |
2086 | rb->task = NULL; |
2087 | } | |
2088 | ||
2d6500d4 PB |
2089 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2090 | struct req_batch *rb) | |
2091 | { | |
6e833d53 | 2092 | if (rb->task) |
7c660731 | 2093 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2094 | if (rb->ctx_refs) |
2095 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2096 | } |
2097 | ||
6ff119a6 PB |
2098 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2099 | struct io_submit_state *state) | |
2d6500d4 | 2100 | { |
f2f87370 | 2101 | io_queue_next(req); |
96670657 | 2102 | io_dismantle_req(req); |
2d6500d4 | 2103 | |
e3bc8e9d | 2104 | if (req->task != rb->task) { |
7c660731 PB |
2105 | if (rb->task) |
2106 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2107 | rb->task = req->task; |
2108 | rb->task_refs = 0; | |
5af1d13e | 2109 | } |
e3bc8e9d | 2110 | rb->task_refs++; |
9ae72463 | 2111 | rb->ctx_refs++; |
5af1d13e | 2112 | |
bd759045 | 2113 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2114 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2115 | else |
2116 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2117 | } |
2118 | ||
905c172f PB |
2119 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2120 | struct io_ring_ctx *ctx) | |
2121 | { | |
2122 | int i, nr = cs->nr; | |
2123 | struct io_kiocb *req; | |
2124 | struct req_batch rb; | |
2125 | ||
2126 | io_init_req_batch(&rb); | |
2127 | spin_lock_irq(&ctx->completion_lock); | |
2128 | for (i = 0; i < nr; i++) { | |
2129 | req = cs->reqs[i]; | |
2130 | __io_cqring_fill_event(req, req->result, req->compl.cflags); | |
2131 | } | |
2132 | io_commit_cqring(ctx); | |
2133 | spin_unlock_irq(&ctx->completion_lock); | |
2134 | ||
2135 | io_cqring_ev_posted(ctx); | |
2136 | for (i = 0; i < nr; i++) { | |
2137 | req = cs->reqs[i]; | |
2138 | ||
2139 | /* submission and completion refs */ | |
de9b4cca | 2140 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2141 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2142 | } |
2143 | ||
2144 | io_req_free_batch_finish(ctx, &rb); | |
2145 | cs->nr = 0; | |
7a743e22 PB |
2146 | } |
2147 | ||
ba816ad6 JA |
2148 | /* |
2149 | * Drop reference to request, return next in chain (if there is one) if this | |
2150 | * was the last reference to this request. | |
2151 | */ | |
0d85035a | 2152 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2153 | { |
9b5f7bd9 PB |
2154 | struct io_kiocb *nxt = NULL; |
2155 | ||
de9b4cca | 2156 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2157 | nxt = io_req_find_next(req); |
4d7dd462 | 2158 | __io_free_req(req); |
2a44f467 | 2159 | } |
9b5f7bd9 | 2160 | return nxt; |
2b188cc1 JA |
2161 | } |
2162 | ||
0d85035a | 2163 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2164 | { |
de9b4cca | 2165 | if (req_ref_put_and_test(req)) |
e65ef56d | 2166 | io_free_req(req); |
2b188cc1 JA |
2167 | } |
2168 | ||
216578e5 PB |
2169 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2170 | { | |
2171 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2172 | ||
2173 | io_free_req(req); | |
2174 | } | |
2175 | ||
2176 | static void io_free_req_deferred(struct io_kiocb *req) | |
2177 | { | |
7cbf1722 | 2178 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2179 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2180 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2181 | } |
2182 | ||
2183 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2184 | { | |
de9b4cca | 2185 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2186 | io_free_req_deferred(req); |
2187 | } | |
2188 | ||
6c503150 | 2189 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2190 | { |
2191 | /* See comment at the top of this file */ | |
2192 | smp_rmb(); | |
e23de15f | 2193 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2194 | } |
2195 | ||
fb5ccc98 PB |
2196 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2197 | { | |
2198 | struct io_rings *rings = ctx->rings; | |
2199 | ||
2200 | /* make sure SQ entry isn't read before tail */ | |
2201 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2202 | } | |
2203 | ||
8ff069bf | 2204 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2205 | { |
8ff069bf | 2206 | unsigned int cflags; |
e94f141b | 2207 | |
bcda7baa JA |
2208 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2209 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2210 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2211 | kfree(kbuf); |
2212 | return cflags; | |
e94f141b JA |
2213 | } |
2214 | ||
8ff069bf | 2215 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2216 | { |
4d954c25 | 2217 | struct io_buffer *kbuf; |
bcda7baa | 2218 | |
4d954c25 | 2219 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2220 | return io_put_kbuf(req, kbuf); |
2221 | } | |
2222 | ||
4c6e277c JA |
2223 | static inline bool io_run_task_work(void) |
2224 | { | |
6200b0ae JA |
2225 | /* |
2226 | * Not safe to run on exiting task, and the task_work handling will | |
2227 | * not add work to such a task. | |
2228 | */ | |
2229 | if (unlikely(current->flags & PF_EXITING)) | |
2230 | return false; | |
4c6e277c JA |
2231 | if (current->task_works) { |
2232 | __set_current_state(TASK_RUNNING); | |
2233 | task_work_run(); | |
2234 | return true; | |
2235 | } | |
2236 | ||
2237 | return false; | |
bcda7baa JA |
2238 | } |
2239 | ||
def596e9 JA |
2240 | /* |
2241 | * Find and free completed poll iocbs | |
2242 | */ | |
2243 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2244 | struct list_head *done) | |
2245 | { | |
8237e045 | 2246 | struct req_batch rb; |
def596e9 | 2247 | struct io_kiocb *req; |
bbde017a XW |
2248 | |
2249 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2250 | smp_rmb(); | |
def596e9 | 2251 | |
5af1d13e | 2252 | io_init_req_batch(&rb); |
def596e9 | 2253 | while (!list_empty(done)) { |
bcda7baa JA |
2254 | int cflags = 0; |
2255 | ||
d21ffe7e | 2256 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2257 | list_del(&req->inflight_entry); |
2258 | ||
8c130827 PB |
2259 | if (READ_ONCE(req->result) == -EAGAIN && |
2260 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2261 | req->iopoll_completed = 0; |
8c130827 PB |
2262 | req_ref_get(req); |
2263 | io_queue_async_work(req); | |
2264 | continue; | |
bbde017a | 2265 | } |
def596e9 | 2266 | |
bcda7baa | 2267 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2268 | cflags = io_put_rw_kbuf(req); |
bcda7baa JA |
2269 | |
2270 | __io_cqring_fill_event(req, req->result, cflags); | |
def596e9 JA |
2271 | (*nr_events)++; |
2272 | ||
de9b4cca | 2273 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2274 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2275 | } |
def596e9 | 2276 | |
09bb8394 | 2277 | io_commit_cqring(ctx); |
80c18e4a | 2278 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2279 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2280 | } |
2281 | ||
def596e9 JA |
2282 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2283 | long min) | |
2284 | { | |
2285 | struct io_kiocb *req, *tmp; | |
2286 | LIST_HEAD(done); | |
2287 | bool spin; | |
2288 | int ret; | |
2289 | ||
2290 | /* | |
2291 | * Only spin for completions if we don't have multiple devices hanging | |
2292 | * off our complete list, and we're under the requested amount. | |
2293 | */ | |
2294 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2295 | ||
2296 | ret = 0; | |
d21ffe7e | 2297 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2298 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2299 | |
2300 | /* | |
581f9810 BM |
2301 | * Move completed and retryable entries to our local lists. |
2302 | * If we find a request that requires polling, break out | |
2303 | * and complete those lists first, if we have entries there. | |
def596e9 | 2304 | */ |
65a6543d | 2305 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2306 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2307 | continue; |
2308 | } | |
2309 | if (!list_empty(&done)) | |
2310 | break; | |
2311 | ||
2312 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2313 | if (ret < 0) | |
2314 | break; | |
2315 | ||
3aadc23e PB |
2316 | /* iopoll may have completed current req */ |
2317 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2318 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2319 | |
def596e9 JA |
2320 | if (ret && spin) |
2321 | spin = false; | |
2322 | ret = 0; | |
2323 | } | |
2324 | ||
2325 | if (!list_empty(&done)) | |
2326 | io_iopoll_complete(ctx, nr_events, &done); | |
2327 | ||
2328 | return ret; | |
2329 | } | |
2330 | ||
2331 | /* | |
d195a66e | 2332 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2333 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2334 | * as a non-spinning completion check. | |
2335 | */ | |
2336 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2337 | long min) | |
2338 | { | |
540e32a0 | 2339 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2340 | int ret; |
2341 | ||
2342 | ret = io_do_iopoll(ctx, nr_events, min); | |
2343 | if (ret < 0) | |
2344 | return ret; | |
eba0a4dd | 2345 | if (*nr_events >= min) |
def596e9 JA |
2346 | return 0; |
2347 | } | |
2348 | ||
2349 | return 1; | |
2350 | } | |
2351 | ||
2352 | /* | |
2353 | * We can't just wait for polled events to come to us, we have to actively | |
2354 | * find and complete them. | |
2355 | */ | |
b2edc0a7 | 2356 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2357 | { |
2358 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2359 | return; | |
2360 | ||
2361 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2362 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2363 | unsigned int nr_events = 0; |
2364 | ||
b2edc0a7 | 2365 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2366 | |
b2edc0a7 PB |
2367 | /* let it sleep and repeat later if can't complete a request */ |
2368 | if (nr_events == 0) | |
2369 | break; | |
08f5439f JA |
2370 | /* |
2371 | * Ensure we allow local-to-the-cpu processing to take place, | |
2372 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2373 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2374 | */ |
3fcee5a6 PB |
2375 | if (need_resched()) { |
2376 | mutex_unlock(&ctx->uring_lock); | |
2377 | cond_resched(); | |
2378 | mutex_lock(&ctx->uring_lock); | |
2379 | } | |
def596e9 JA |
2380 | } |
2381 | mutex_unlock(&ctx->uring_lock); | |
2382 | } | |
2383 | ||
7668b92a | 2384 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2385 | { |
7668b92a | 2386 | unsigned int nr_events = 0; |
2b2ed975 | 2387 | int iters = 0, ret = 0; |
500f9fba | 2388 | |
c7849be9 XW |
2389 | /* |
2390 | * We disallow the app entering submit/complete with polling, but we | |
2391 | * still need to lock the ring to prevent racing with polled issue | |
2392 | * that got punted to a workqueue. | |
2393 | */ | |
2394 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2395 | do { |
a3a0e43f JA |
2396 | /* |
2397 | * Don't enter poll loop if we already have events pending. | |
2398 | * If we do, we can potentially be spinning for commands that | |
2399 | * already triggered a CQE (eg in error). | |
2400 | */ | |
6c503150 | 2401 | if (test_bit(0, &ctx->cq_check_overflow)) |
6c2450ae | 2402 | __io_cqring_overflow_flush(ctx, false); |
6c503150 | 2403 | if (io_cqring_events(ctx)) |
a3a0e43f JA |
2404 | break; |
2405 | ||
500f9fba JA |
2406 | /* |
2407 | * If a submit got punted to a workqueue, we can have the | |
2408 | * application entering polling for a command before it gets | |
2409 | * issued. That app will hold the uring_lock for the duration | |
2410 | * of the poll right here, so we need to take a breather every | |
2411 | * now and then to ensure that the issue has a chance to add | |
2412 | * the poll to the issued list. Otherwise we can spin here | |
2413 | * forever, while the workqueue is stuck trying to acquire the | |
2414 | * very same mutex. | |
2415 | */ | |
2416 | if (!(++iters & 7)) { | |
2417 | mutex_unlock(&ctx->uring_lock); | |
4c6e277c | 2418 | io_run_task_work(); |
500f9fba JA |
2419 | mutex_lock(&ctx->uring_lock); |
2420 | } | |
2421 | ||
7668b92a | 2422 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2423 | if (ret <= 0) |
2424 | break; | |
2425 | ret = 0; | |
7668b92a | 2426 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2427 | |
500f9fba | 2428 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2429 | return ret; |
2430 | } | |
2431 | ||
491381ce | 2432 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2433 | { |
491381ce JA |
2434 | /* |
2435 | * Tell lockdep we inherited freeze protection from submission | |
2436 | * thread. | |
2437 | */ | |
2438 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2439 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2440 | |
1c98679d PB |
2441 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2442 | sb_end_write(sb); | |
2b188cc1 JA |
2443 | } |
2444 | } | |
2445 | ||
b63534c4 | 2446 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2447 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2448 | { |
ab454438 PB |
2449 | struct io_async_rw *rw = req->async_data; |
2450 | ||
2451 | if (!rw) | |
2452 | return !io_req_prep_async(req); | |
2453 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2454 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2455 | return true; | |
b63534c4 | 2456 | } |
b63534c4 | 2457 | |
3e6a0d3c | 2458 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2459 | { |
355afaeb | 2460 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2461 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2462 | |
355afaeb JA |
2463 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2464 | return false; | |
3e6a0d3c JA |
2465 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2466 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2467 | return false; |
7c977a58 JA |
2468 | /* |
2469 | * If ref is dying, we might be running poll reap from the exit work. | |
2470 | * Don't attempt to reissue from that path, just let it fail with | |
2471 | * -EAGAIN. | |
2472 | */ | |
3e6a0d3c JA |
2473 | if (percpu_ref_is_dying(&ctx->refs)) |
2474 | return false; | |
2475 | return true; | |
2476 | } | |
e82ad485 JA |
2477 | #else |
2478 | static bool io_rw_should_reissue(struct io_kiocb *req) | |
2479 | { | |
2480 | return false; | |
2481 | } | |
3e6a0d3c JA |
2482 | #endif |
2483 | ||
a1d7c393 | 2484 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2485 | unsigned int issue_flags) |
a1d7c393 | 2486 | { |
2f8e45f1 PB |
2487 | int cflags = 0; |
2488 | ||
b65c128f PB |
2489 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2490 | kiocb_end_write(req); | |
9532b99b PB |
2491 | if (res != req->result) { |
2492 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2493 | io_rw_should_reissue(req)) { | |
2494 | req->flags |= REQ_F_REISSUE; | |
2495 | return; | |
2496 | } | |
2f8e45f1 | 2497 | req_set_fail_links(req); |
9532b99b | 2498 | } |
2f8e45f1 PB |
2499 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2500 | cflags = io_put_rw_kbuf(req); | |
2501 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2502 | } |
2503 | ||
2504 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2505 | { | |
9adbd45d | 2506 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2507 | |
889fca73 | 2508 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2509 | } |
2510 | ||
def596e9 JA |
2511 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2512 | { | |
9adbd45d | 2513 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2514 | |
491381ce JA |
2515 | if (kiocb->ki_flags & IOCB_WRITE) |
2516 | kiocb_end_write(req); | |
9532b99b PB |
2517 | if (unlikely(res != req->result)) { |
2518 | bool fail = true; | |
def596e9 | 2519 | |
9532b99b PB |
2520 | #ifdef CONFIG_BLOCK |
2521 | if (res == -EAGAIN && io_rw_should_reissue(req) && | |
2522 | io_resubmit_prep(req)) | |
2523 | fail = false; | |
2524 | #endif | |
2525 | if (fail) { | |
2526 | req_set_fail_links(req); | |
2527 | req->flags |= REQ_F_DONT_REISSUE; | |
2528 | } | |
8c130827 | 2529 | } |
bbde017a XW |
2530 | |
2531 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2532 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2533 | smp_wmb(); |
2534 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2535 | } |
2536 | ||
2537 | /* | |
2538 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2539 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2540 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2541 | * accessing the kiocb cookie. | |
2542 | */ | |
2e9dbe90 | 2543 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2544 | { |
2545 | struct io_ring_ctx *ctx = req->ctx; | |
2546 | ||
2547 | /* | |
2548 | * Track whether we have multiple files in our lists. This will impact | |
2549 | * how we do polling eventually, not spinning if we're on potentially | |
2550 | * different devices. | |
2551 | */ | |
540e32a0 | 2552 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2553 | ctx->poll_multi_file = false; |
2554 | } else if (!ctx->poll_multi_file) { | |
2555 | struct io_kiocb *list_req; | |
2556 | ||
540e32a0 | 2557 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2558 | inflight_entry); |
9adbd45d | 2559 | if (list_req->file != req->file) |
def596e9 JA |
2560 | ctx->poll_multi_file = true; |
2561 | } | |
2562 | ||
2563 | /* | |
2564 | * For fast devices, IO may have already completed. If it has, add | |
2565 | * it to the front so we find it first. | |
2566 | */ | |
65a6543d | 2567 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2568 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2569 | else |
d21ffe7e | 2570 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2571 | |
2e9dbe90 XW |
2572 | /* |
2573 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2574 | * task context or in io worker task context. If current task context is | |
2575 | * sq thread, we don't need to check whether should wake up sq thread. | |
2576 | */ | |
2577 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2578 | wq_has_sleeper(&ctx->sq_data->wait)) |
2579 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2580 | } |
2581 | ||
9f13c35b PB |
2582 | static inline void io_state_file_put(struct io_submit_state *state) |
2583 | { | |
02b23a9a PB |
2584 | if (state->file_refs) { |
2585 | fput_many(state->file, state->file_refs); | |
2586 | state->file_refs = 0; | |
2587 | } | |
9a56a232 JA |
2588 | } |
2589 | ||
2590 | /* | |
2591 | * Get as many references to a file as we have IOs left in this submission, | |
2592 | * assuming most submissions are for one file, or at least that each file | |
2593 | * has more than one submission. | |
2594 | */ | |
8da11c19 | 2595 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2596 | { |
2597 | if (!state) | |
2598 | return fget(fd); | |
2599 | ||
6e1271e6 | 2600 | if (state->file_refs) { |
9a56a232 | 2601 | if (state->fd == fd) { |
6e1271e6 | 2602 | state->file_refs--; |
9a56a232 JA |
2603 | return state->file; |
2604 | } | |
02b23a9a | 2605 | io_state_file_put(state); |
9a56a232 JA |
2606 | } |
2607 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2608 | if (unlikely(!state->file)) |
9a56a232 JA |
2609 | return NULL; |
2610 | ||
2611 | state->fd = fd; | |
6e1271e6 | 2612 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2613 | return state->file; |
2614 | } | |
2615 | ||
4503b767 JA |
2616 | static bool io_bdev_nowait(struct block_device *bdev) |
2617 | { | |
9ba0d0c8 | 2618 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2619 | } |
2620 | ||
2b188cc1 JA |
2621 | /* |
2622 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2623 | * any file. For now, just ensure that anything potentially problematic is done | |
2624 | * inline. | |
2625 | */ | |
7b29f92d | 2626 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2627 | { |
2628 | umode_t mode = file_inode(file)->i_mode; | |
2629 | ||
4503b767 | 2630 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2631 | if (IS_ENABLED(CONFIG_BLOCK) && |
2632 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2633 | return true; |
2634 | return false; | |
2635 | } | |
2636 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2637 | return true; |
4503b767 | 2638 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2639 | if (IS_ENABLED(CONFIG_BLOCK) && |
2640 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2641 | file->f_op != &io_uring_fops) |
2642 | return true; | |
2643 | return false; | |
2644 | } | |
2b188cc1 | 2645 | |
c5b85625 JA |
2646 | /* any ->read/write should understand O_NONBLOCK */ |
2647 | if (file->f_flags & O_NONBLOCK) | |
2648 | return true; | |
2649 | ||
af197f50 JA |
2650 | if (!(file->f_mode & FMODE_NOWAIT)) |
2651 | return false; | |
2652 | ||
2653 | if (rw == READ) | |
2654 | return file->f_op->read_iter != NULL; | |
2655 | ||
2656 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2657 | } |
2658 | ||
7b29f92d JA |
2659 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2660 | { | |
2661 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2662 | return true; | |
2663 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2664 | return true; | |
2665 | ||
2666 | return __io_file_supports_async(req->file, rw); | |
2667 | } | |
2668 | ||
a88fc400 | 2669 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2670 | { |
def596e9 | 2671 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2672 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2673 | struct file *file = req->file; |
09bb8394 JA |
2674 | unsigned ioprio; |
2675 | int ret; | |
2b188cc1 | 2676 | |
7b29f92d | 2677 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2678 | req->flags |= REQ_F_ISREG; |
2679 | ||
2b188cc1 | 2680 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2681 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2682 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2683 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2684 | } |
2b188cc1 | 2685 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2686 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2687 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2688 | if (unlikely(ret)) | |
2689 | return ret; | |
2b188cc1 | 2690 | |
75c668cd PB |
2691 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2692 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2693 | req->flags |= REQ_F_NOWAIT; | |
2694 | ||
2b188cc1 JA |
2695 | ioprio = READ_ONCE(sqe->ioprio); |
2696 | if (ioprio) { | |
2697 | ret = ioprio_check_cap(ioprio); | |
2698 | if (ret) | |
09bb8394 | 2699 | return ret; |
2b188cc1 JA |
2700 | |
2701 | kiocb->ki_ioprio = ioprio; | |
2702 | } else | |
2703 | kiocb->ki_ioprio = get_current_ioprio(); | |
2704 | ||
def596e9 | 2705 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2706 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2707 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2708 | return -EOPNOTSUPP; |
2b188cc1 | 2709 | |
def596e9 JA |
2710 | kiocb->ki_flags |= IOCB_HIPRI; |
2711 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2712 | req->iopoll_completed = 0; |
def596e9 | 2713 | } else { |
09bb8394 JA |
2714 | if (kiocb->ki_flags & IOCB_HIPRI) |
2715 | return -EINVAL; | |
def596e9 JA |
2716 | kiocb->ki_complete = io_complete_rw; |
2717 | } | |
9adbd45d | 2718 | |
3529d8c2 JA |
2719 | req->rw.addr = READ_ONCE(sqe->addr); |
2720 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2721 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2722 | return 0; |
2b188cc1 JA |
2723 | } |
2724 | ||
2725 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2726 | { | |
2727 | switch (ret) { | |
2728 | case -EIOCBQUEUED: | |
2729 | break; | |
2730 | case -ERESTARTSYS: | |
2731 | case -ERESTARTNOINTR: | |
2732 | case -ERESTARTNOHAND: | |
2733 | case -ERESTART_RESTARTBLOCK: | |
2734 | /* | |
2735 | * We can't just restart the syscall, since previously | |
2736 | * submitted sqes may already be in progress. Just fail this | |
2737 | * IO with EINTR. | |
2738 | */ | |
2739 | ret = -EINTR; | |
df561f66 | 2740 | fallthrough; |
2b188cc1 JA |
2741 | default: |
2742 | kiocb->ki_complete(kiocb, ret, 0); | |
2743 | } | |
2744 | } | |
2745 | ||
a1d7c393 | 2746 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2747 | unsigned int issue_flags) |
ba816ad6 | 2748 | { |
ba04291e | 2749 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2750 | struct io_async_rw *io = req->async_data; |
97284637 | 2751 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2752 | |
227c0c96 | 2753 | /* add previously done IO, if any */ |
e8c2bc1f | 2754 | if (io && io->bytes_done > 0) { |
227c0c96 | 2755 | if (ret < 0) |
e8c2bc1f | 2756 | ret = io->bytes_done; |
227c0c96 | 2757 | else |
e8c2bc1f | 2758 | ret += io->bytes_done; |
227c0c96 JA |
2759 | } |
2760 | ||
ba04291e JA |
2761 | if (req->flags & REQ_F_CUR_POS) |
2762 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2763 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2764 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2765 | else |
2766 | io_rw_done(kiocb, ret); | |
97284637 PB |
2767 | |
2768 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2769 | req->flags &= ~REQ_F_REISSUE; | |
8c130827 PB |
2770 | if (!io_resubmit_prep(req)) { |
2771 | req_ref_get(req); | |
2772 | io_queue_async_work(req); | |
2773 | } else { | |
97284637 PB |
2774 | int cflags = 0; |
2775 | ||
2776 | req_set_fail_links(req); | |
2777 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2778 | cflags = io_put_rw_kbuf(req); | |
2779 | __io_req_complete(req, issue_flags, ret, cflags); | |
2780 | } | |
2781 | } | |
ba816ad6 JA |
2782 | } |
2783 | ||
847595de | 2784 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2785 | { |
9adbd45d JA |
2786 | struct io_ring_ctx *ctx = req->ctx; |
2787 | size_t len = req->rw.len; | |
edafccee | 2788 | struct io_mapped_ubuf *imu; |
4be1c615 | 2789 | u16 index, buf_index = req->buf_index; |
edafccee JA |
2790 | size_t offset; |
2791 | u64 buf_addr; | |
2792 | ||
edafccee JA |
2793 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2794 | return -EFAULT; | |
edafccee JA |
2795 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2796 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2797 | buf_addr = req->rw.addr; |
edafccee JA |
2798 | |
2799 | /* overflow */ | |
2800 | if (buf_addr + len < buf_addr) | |
2801 | return -EFAULT; | |
2802 | /* not inside the mapped region */ | |
2803 | if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len) | |
2804 | return -EFAULT; | |
2805 | ||
2806 | /* | |
2807 | * May not be a start of buffer, set size appropriately | |
2808 | * and advance us to the beginning. | |
2809 | */ | |
2810 | offset = buf_addr - imu->ubuf; | |
2811 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2812 | |
2813 | if (offset) { | |
2814 | /* | |
2815 | * Don't use iov_iter_advance() here, as it's really slow for | |
2816 | * using the latter parts of a big fixed buffer - it iterates | |
2817 | * over each segment manually. We can cheat a bit here, because | |
2818 | * we know that: | |
2819 | * | |
2820 | * 1) it's a BVEC iter, we set it up | |
2821 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2822 | * first and last bvec | |
2823 | * | |
2824 | * So just find our index, and adjust the iterator afterwards. | |
2825 | * If the offset is within the first bvec (or the whole first | |
2826 | * bvec, just use iov_iter_advance(). This makes it easier | |
2827 | * since we can just skip the first segment, which may not | |
2828 | * be PAGE_SIZE aligned. | |
2829 | */ | |
2830 | const struct bio_vec *bvec = imu->bvec; | |
2831 | ||
2832 | if (offset <= bvec->bv_len) { | |
2833 | iov_iter_advance(iter, offset); | |
2834 | } else { | |
2835 | unsigned long seg_skip; | |
2836 | ||
2837 | /* skip first vec */ | |
2838 | offset -= bvec->bv_len; | |
2839 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2840 | ||
2841 | iter->bvec = bvec + seg_skip; | |
2842 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2843 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2844 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2845 | } |
2846 | } | |
2847 | ||
847595de | 2848 | return 0; |
edafccee JA |
2849 | } |
2850 | ||
bcda7baa JA |
2851 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2852 | { | |
2853 | if (needs_lock) | |
2854 | mutex_unlock(&ctx->uring_lock); | |
2855 | } | |
2856 | ||
2857 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2858 | { | |
2859 | /* | |
2860 | * "Normal" inline submissions always hold the uring_lock, since we | |
2861 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2862 | * The only exception is when we've detached the request and issue it | |
2863 | * from an async worker thread, grab the lock for that case. | |
2864 | */ | |
2865 | if (needs_lock) | |
2866 | mutex_lock(&ctx->uring_lock); | |
2867 | } | |
2868 | ||
2869 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2870 | int bgid, struct io_buffer *kbuf, | |
2871 | bool needs_lock) | |
2872 | { | |
2873 | struct io_buffer *head; | |
2874 | ||
2875 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2876 | return kbuf; | |
2877 | ||
2878 | io_ring_submit_lock(req->ctx, needs_lock); | |
2879 | ||
2880 | lockdep_assert_held(&req->ctx->uring_lock); | |
2881 | ||
9e15c3a0 | 2882 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2883 | if (head) { |
2884 | if (!list_empty(&head->list)) { | |
2885 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2886 | list); | |
2887 | list_del(&kbuf->list); | |
2888 | } else { | |
2889 | kbuf = head; | |
9e15c3a0 | 2890 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2891 | } |
2892 | if (*len > kbuf->len) | |
2893 | *len = kbuf->len; | |
2894 | } else { | |
2895 | kbuf = ERR_PTR(-ENOBUFS); | |
2896 | } | |
2897 | ||
2898 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2899 | ||
2900 | return kbuf; | |
2901 | } | |
2902 | ||
4d954c25 JA |
2903 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2904 | bool needs_lock) | |
2905 | { | |
2906 | struct io_buffer *kbuf; | |
4f4eeba8 | 2907 | u16 bgid; |
4d954c25 JA |
2908 | |
2909 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2910 | bgid = req->buf_index; |
4d954c25 JA |
2911 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2912 | if (IS_ERR(kbuf)) | |
2913 | return kbuf; | |
2914 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2915 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2916 | return u64_to_user_ptr(kbuf->addr); | |
2917 | } | |
2918 | ||
2919 | #ifdef CONFIG_COMPAT | |
2920 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2921 | bool needs_lock) | |
2922 | { | |
2923 | struct compat_iovec __user *uiov; | |
2924 | compat_ssize_t clen; | |
2925 | void __user *buf; | |
2926 | ssize_t len; | |
2927 | ||
2928 | uiov = u64_to_user_ptr(req->rw.addr); | |
2929 | if (!access_ok(uiov, sizeof(*uiov))) | |
2930 | return -EFAULT; | |
2931 | if (__get_user(clen, &uiov->iov_len)) | |
2932 | return -EFAULT; | |
2933 | if (clen < 0) | |
2934 | return -EINVAL; | |
2935 | ||
2936 | len = clen; | |
2937 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2938 | if (IS_ERR(buf)) | |
2939 | return PTR_ERR(buf); | |
2940 | iov[0].iov_base = buf; | |
2941 | iov[0].iov_len = (compat_size_t) len; | |
2942 | return 0; | |
2943 | } | |
2944 | #endif | |
2945 | ||
2946 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2947 | bool needs_lock) | |
2948 | { | |
2949 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2950 | void __user *buf; | |
2951 | ssize_t len; | |
2952 | ||
2953 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2954 | return -EFAULT; | |
2955 | ||
2956 | len = iov[0].iov_len; | |
2957 | if (len < 0) | |
2958 | return -EINVAL; | |
2959 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2960 | if (IS_ERR(buf)) | |
2961 | return PTR_ERR(buf); | |
2962 | iov[0].iov_base = buf; | |
2963 | iov[0].iov_len = len; | |
2964 | return 0; | |
2965 | } | |
2966 | ||
2967 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2968 | bool needs_lock) | |
2969 | { | |
dddb3e26 JA |
2970 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2971 | struct io_buffer *kbuf; | |
2972 | ||
2973 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2974 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2975 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2976 | return 0; |
dddb3e26 | 2977 | } |
dd201662 | 2978 | if (req->rw.len != 1) |
4d954c25 JA |
2979 | return -EINVAL; |
2980 | ||
2981 | #ifdef CONFIG_COMPAT | |
2982 | if (req->ctx->compat) | |
2983 | return io_compat_import(req, iov, needs_lock); | |
2984 | #endif | |
2985 | ||
2986 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2987 | } | |
2988 | ||
847595de PB |
2989 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2990 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2991 | { |
9adbd45d JA |
2992 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2993 | size_t sqe_len = req->rw.len; | |
847595de | 2994 | u8 opcode = req->opcode; |
4d954c25 | 2995 | ssize_t ret; |
edafccee | 2996 | |
7d009165 | 2997 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2998 | *iovec = NULL; |
9adbd45d | 2999 | return io_import_fixed(req, rw, iter); |
edafccee | 3000 | } |
2b188cc1 | 3001 | |
bcda7baa | 3002 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3003 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3004 | return -EINVAL; |
3005 | ||
3a6820f2 | 3006 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3007 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3008 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3009 | if (IS_ERR(buf)) |
4d954c25 | 3010 | return PTR_ERR(buf); |
3f9d6441 | 3011 | req->rw.len = sqe_len; |
bcda7baa JA |
3012 | } |
3013 | ||
3a6820f2 JA |
3014 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3015 | *iovec = NULL; | |
10fc72e4 | 3016 | return ret; |
3a6820f2 JA |
3017 | } |
3018 | ||
4d954c25 JA |
3019 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3020 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3021 | if (!ret) |
3022 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3023 | *iovec = NULL; |
3024 | return ret; | |
3025 | } | |
3026 | ||
89cd35c5 CH |
3027 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3028 | req->ctx->compat); | |
2b188cc1 JA |
3029 | } |
3030 | ||
0fef9483 JA |
3031 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3032 | { | |
5b09e37e | 3033 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3034 | } |
3035 | ||
31b51510 | 3036 | /* |
32960613 JA |
3037 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3038 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3039 | */ |
4017eb91 | 3040 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3041 | { |
4017eb91 JA |
3042 | struct kiocb *kiocb = &req->rw.kiocb; |
3043 | struct file *file = req->file; | |
32960613 JA |
3044 | ssize_t ret = 0; |
3045 | ||
3046 | /* | |
3047 | * Don't support polled IO through this interface, and we can't | |
3048 | * support non-blocking either. For the latter, this just causes | |
3049 | * the kiocb to be handled from an async context. | |
3050 | */ | |
3051 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3052 | return -EOPNOTSUPP; | |
3053 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3054 | return -EAGAIN; | |
3055 | ||
3056 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3057 | struct iovec iovec; |
32960613 JA |
3058 | ssize_t nr; |
3059 | ||
311ae9e1 PB |
3060 | if (!iov_iter_is_bvec(iter)) { |
3061 | iovec = iov_iter_iovec(iter); | |
3062 | } else { | |
4017eb91 JA |
3063 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3064 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3065 | } |
3066 | ||
32960613 JA |
3067 | if (rw == READ) { |
3068 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3069 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3070 | } else { |
3071 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3072 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3073 | } |
3074 | ||
3075 | if (nr < 0) { | |
3076 | if (!ret) | |
3077 | ret = nr; | |
3078 | break; | |
3079 | } | |
3080 | ret += nr; | |
3081 | if (nr != iovec.iov_len) | |
3082 | break; | |
4017eb91 JA |
3083 | req->rw.len -= nr; |
3084 | req->rw.addr += nr; | |
32960613 JA |
3085 | iov_iter_advance(iter, nr); |
3086 | } | |
3087 | ||
3088 | return ret; | |
3089 | } | |
3090 | ||
ff6165b2 JA |
3091 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3092 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3093 | { |
e8c2bc1f | 3094 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3095 | |
ff6165b2 | 3096 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3097 | rw->free_iovec = iovec; |
227c0c96 | 3098 | rw->bytes_done = 0; |
ff6165b2 | 3099 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3100 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3101 | return; |
b64e3444 | 3102 | if (!iovec) { |
ff6165b2 JA |
3103 | unsigned iov_off = 0; |
3104 | ||
3105 | rw->iter.iov = rw->fast_iov; | |
3106 | if (iter->iov != fast_iov) { | |
3107 | iov_off = iter->iov - fast_iov; | |
3108 | rw->iter.iov += iov_off; | |
3109 | } | |
3110 | if (rw->fast_iov != fast_iov) | |
3111 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3112 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3113 | } else { |
3114 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3115 | } |
3116 | } | |
3117 | ||
6cb78689 | 3118 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3119 | { |
e8c2bc1f JA |
3120 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3121 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3122 | return req->async_data == NULL; | |
3d9932a8 XW |
3123 | } |
3124 | ||
ff6165b2 JA |
3125 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3126 | const struct iovec *fast_iov, | |
227c0c96 | 3127 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3128 | { |
26f0505a | 3129 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3130 | return 0; |
e8c2bc1f | 3131 | if (!req->async_data) { |
6cb78689 | 3132 | if (io_alloc_async_data(req)) { |
6bf985dc | 3133 | kfree(iovec); |
5d204bcf | 3134 | return -ENOMEM; |
6bf985dc | 3135 | } |
b7bb4f7d | 3136 | |
ff6165b2 | 3137 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3138 | } |
b7bb4f7d | 3139 | return 0; |
f67676d1 JA |
3140 | } |
3141 | ||
73debe68 | 3142 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3143 | { |
e8c2bc1f | 3144 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3145 | struct iovec *iov = iorw->fast_iov; |
847595de | 3146 | int ret; |
c3e330a4 | 3147 | |
2846c481 | 3148 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3149 | if (unlikely(ret < 0)) |
3150 | return ret; | |
3151 | ||
ab0b196c PB |
3152 | iorw->bytes_done = 0; |
3153 | iorw->free_iovec = iov; | |
3154 | if (iov) | |
3155 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3156 | return 0; |
3157 | } | |
3158 | ||
73debe68 | 3159 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3160 | { |
3529d8c2 JA |
3161 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3162 | return -EBADF; | |
93642ef8 | 3163 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3164 | } |
3165 | ||
c1dd91d1 JA |
3166 | /* |
3167 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3168 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3169 | * This gets called when the page is unlocked, and we generally expect that to | |
3170 | * happen when the page IO is completed and the page is now uptodate. This will | |
3171 | * queue a task_work based retry of the operation, attempting to copy the data | |
3172 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3173 | * do a thread based blocking retry of the operation. That's the unexpected | |
3174 | * slow path. | |
3175 | */ | |
bcf5a063 JA |
3176 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3177 | int sync, void *arg) | |
3178 | { | |
3179 | struct wait_page_queue *wpq; | |
3180 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3181 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3182 | |
3183 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3184 | ||
cdc8fcb4 LT |
3185 | if (!wake_page_match(wpq, key)) |
3186 | return 0; | |
3187 | ||
c8d317aa | 3188 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3189 | list_del_init(&wait->entry); |
3190 | ||
bcf5a063 | 3191 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3192 | req_ref_get(req); |
921b9054 | 3193 | io_req_task_queue(req); |
bcf5a063 JA |
3194 | return 1; |
3195 | } | |
3196 | ||
c1dd91d1 JA |
3197 | /* |
3198 | * This controls whether a given IO request should be armed for async page | |
3199 | * based retry. If we return false here, the request is handed to the async | |
3200 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3201 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3202 | * will either succeed because the page is now uptodate and unlocked, or it | |
3203 | * will register a callback when the page is unlocked at IO completion. Through | |
3204 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3205 | * That retry will attempt the buffered read again. The retry will generally | |
3206 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3207 | * async worker threads for a blocking retry. | |
3208 | */ | |
227c0c96 | 3209 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3210 | { |
e8c2bc1f JA |
3211 | struct io_async_rw *rw = req->async_data; |
3212 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3213 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3214 | |
bcf5a063 JA |
3215 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3216 | if (req->flags & REQ_F_NOWAIT) | |
3217 | return false; | |
f67676d1 | 3218 | |
227c0c96 | 3219 | /* Only for buffered IO */ |
3b2a4439 | 3220 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3221 | return false; |
3b2a4439 | 3222 | |
bcf5a063 JA |
3223 | /* |
3224 | * just use poll if we can, and don't attempt if the fs doesn't | |
3225 | * support callback based unlocks | |
3226 | */ | |
3227 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3228 | return false; | |
f67676d1 | 3229 | |
3b2a4439 JA |
3230 | wait->wait.func = io_async_buf_func; |
3231 | wait->wait.private = req; | |
3232 | wait->wait.flags = 0; | |
3233 | INIT_LIST_HEAD(&wait->wait.entry); | |
3234 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3235 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3236 | kiocb->ki_waitq = wait; |
3b2a4439 | 3237 | return true; |
bcf5a063 JA |
3238 | } |
3239 | ||
3240 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3241 | { | |
3242 | if (req->file->f_op->read_iter) | |
3243 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3244 | else if (req->file->f_op->read) |
4017eb91 | 3245 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3246 | else |
3247 | return -EINVAL; | |
f67676d1 JA |
3248 | } |
3249 | ||
889fca73 | 3250 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3251 | { |
3252 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3253 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3254 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3255 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3256 | ssize_t io_size, ret, ret2; |
45d189c6 | 3257 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3258 | |
2846c481 | 3259 | if (rw) { |
e8c2bc1f | 3260 | iter = &rw->iter; |
2846c481 PB |
3261 | iovec = NULL; |
3262 | } else { | |
3263 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3264 | if (ret < 0) | |
3265 | return ret; | |
3266 | } | |
632546c4 | 3267 | io_size = iov_iter_count(iter); |
fa15bafb | 3268 | req->result = io_size; |
2b188cc1 | 3269 | |
fd6c2e4c JA |
3270 | /* Ensure we clear previously set non-block flag */ |
3271 | if (!force_nonblock) | |
29de5f6a | 3272 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3273 | else |
3274 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3275 | ||
24c74678 | 3276 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3277 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3278 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3279 | return ret ?: -EAGAIN; |
6713e7a6 | 3280 | } |
9e645e11 | 3281 | |
632546c4 | 3282 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3283 | if (unlikely(ret)) { |
3284 | kfree(iovec); | |
3285 | return ret; | |
3286 | } | |
2b188cc1 | 3287 | |
227c0c96 | 3288 | ret = io_iter_do_read(req, iter); |
32960613 | 3289 | |
230d50d4 | 3290 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3291 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3292 | /* IOPOLL retry should happen for io-wq threads */ |
3293 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3294 | goto done; |
75c668cd PB |
3295 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3296 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3297 | goto done; |
84216315 | 3298 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3299 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3300 | ret = 0; |
230d50d4 JA |
3301 | } else if (ret == -EIOCBQUEUED) { |
3302 | goto out_free; | |
7335e3bf | 3303 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3304 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3305 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3306 | goto done; |
227c0c96 JA |
3307 | } |
3308 | ||
227c0c96 | 3309 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3310 | if (ret2) |
3311 | return ret2; | |
3312 | ||
fe1cdd55 | 3313 | iovec = NULL; |
e8c2bc1f | 3314 | rw = req->async_data; |
227c0c96 | 3315 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3316 | iter = &rw->iter; |
227c0c96 | 3317 | |
b23df91b PB |
3318 | do { |
3319 | io_size -= ret; | |
3320 | rw->bytes_done += ret; | |
3321 | /* if we can retry, do so with the callbacks armed */ | |
3322 | if (!io_rw_should_retry(req)) { | |
3323 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3324 | return -EAGAIN; | |
3325 | } | |
3326 | ||
3327 | /* | |
3328 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3329 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3330 | * desired page gets unlocked. We can also get a partial read | |
3331 | * here, and if we do, then just retry at the new offset. | |
3332 | */ | |
3333 | ret = io_iter_do_read(req, iter); | |
3334 | if (ret == -EIOCBQUEUED) | |
3335 | return 0; | |
227c0c96 | 3336 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3337 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3338 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3339 | done: |
889fca73 | 3340 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3341 | out_free: |
3342 | /* it's faster to check here then delegate to kfree */ | |
3343 | if (iovec) | |
3344 | kfree(iovec); | |
5ea5dd45 | 3345 | return 0; |
2b188cc1 JA |
3346 | } |
3347 | ||
73debe68 | 3348 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3349 | { |
3529d8c2 JA |
3350 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3351 | return -EBADF; | |
93642ef8 | 3352 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3353 | } |
3354 | ||
889fca73 | 3355 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3356 | { |
3357 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3358 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3359 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3360 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3361 | ssize_t ret, ret2, io_size; |
45d189c6 | 3362 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3363 | |
2846c481 | 3364 | if (rw) { |
e8c2bc1f | 3365 | iter = &rw->iter; |
2846c481 PB |
3366 | iovec = NULL; |
3367 | } else { | |
3368 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3369 | if (ret < 0) | |
3370 | return ret; | |
3371 | } | |
632546c4 | 3372 | io_size = iov_iter_count(iter); |
fa15bafb | 3373 | req->result = io_size; |
2b188cc1 | 3374 | |
fd6c2e4c JA |
3375 | /* Ensure we clear previously set non-block flag */ |
3376 | if (!force_nonblock) | |
a88fc400 PB |
3377 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3378 | else | |
3379 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3380 | |
24c74678 | 3381 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3382 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3383 | goto copy_iov; |
31b51510 | 3384 | |
10d59345 JA |
3385 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3386 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3387 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3388 | goto copy_iov; |
31b51510 | 3389 | |
632546c4 | 3390 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3391 | if (unlikely(ret)) |
3392 | goto out_free; | |
4ed734b0 | 3393 | |
fa15bafb PB |
3394 | /* |
3395 | * Open-code file_start_write here to grab freeze protection, | |
3396 | * which will be released by another thread in | |
3397 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3398 | * released so that it doesn't complain about the held lock when | |
3399 | * we return to userspace. | |
3400 | */ | |
3401 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3402 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3403 | __sb_writers_release(file_inode(req->file)->i_sb, |
3404 | SB_FREEZE_WRITE); | |
3405 | } | |
3406 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3407 | |
fa15bafb | 3408 | if (req->file->f_op->write_iter) |
ff6165b2 | 3409 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3410 | else if (req->file->f_op->write) |
4017eb91 | 3411 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3412 | else |
3413 | ret2 = -EINVAL; | |
4ed734b0 | 3414 | |
6ad7f233 PB |
3415 | if (req->flags & REQ_F_REISSUE) { |
3416 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3417 | ret2 = -EAGAIN; |
6ad7f233 | 3418 | } |
230d50d4 | 3419 | |
fa15bafb PB |
3420 | /* |
3421 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3422 | * retry them without IOCB_NOWAIT. | |
3423 | */ | |
3424 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3425 | ret2 = -EAGAIN; | |
75c668cd PB |
3426 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3427 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3428 | goto done; |
fa15bafb | 3429 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3430 | /* IOPOLL retry should happen for io-wq threads */ |
3431 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3432 | goto copy_iov; | |
355afaeb | 3433 | done: |
889fca73 | 3434 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3435 | } else { |
f67676d1 | 3436 | copy_iov: |
84216315 | 3437 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3438 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3439 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3440 | return ret ?: -EAGAIN; |
2b188cc1 | 3441 | } |
31b51510 | 3442 | out_free: |
f261c168 | 3443 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3444 | if (iovec) |
6f2cc166 | 3445 | kfree(iovec); |
2b188cc1 JA |
3446 | return ret; |
3447 | } | |
3448 | ||
80a261fd JA |
3449 | static int io_renameat_prep(struct io_kiocb *req, |
3450 | const struct io_uring_sqe *sqe) | |
3451 | { | |
3452 | struct io_rename *ren = &req->rename; | |
3453 | const char __user *oldf, *newf; | |
3454 | ||
3455 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3456 | return -EBADF; | |
3457 | ||
3458 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3459 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3460 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3461 | ren->new_dfd = READ_ONCE(sqe->len); | |
3462 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3463 | ||
3464 | ren->oldpath = getname(oldf); | |
3465 | if (IS_ERR(ren->oldpath)) | |
3466 | return PTR_ERR(ren->oldpath); | |
3467 | ||
3468 | ren->newpath = getname(newf); | |
3469 | if (IS_ERR(ren->newpath)) { | |
3470 | putname(ren->oldpath); | |
3471 | return PTR_ERR(ren->newpath); | |
3472 | } | |
3473 | ||
3474 | req->flags |= REQ_F_NEED_CLEANUP; | |
3475 | return 0; | |
3476 | } | |
3477 | ||
45d189c6 | 3478 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3479 | { |
3480 | struct io_rename *ren = &req->rename; | |
3481 | int ret; | |
3482 | ||
45d189c6 | 3483 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3484 | return -EAGAIN; |
3485 | ||
3486 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3487 | ren->newpath, ren->flags); | |
3488 | ||
3489 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3490 | if (ret < 0) | |
3491 | req_set_fail_links(req); | |
3492 | io_req_complete(req, ret); | |
3493 | return 0; | |
3494 | } | |
3495 | ||
14a1143b JA |
3496 | static int io_unlinkat_prep(struct io_kiocb *req, |
3497 | const struct io_uring_sqe *sqe) | |
3498 | { | |
3499 | struct io_unlink *un = &req->unlink; | |
3500 | const char __user *fname; | |
3501 | ||
3502 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3503 | return -EBADF; | |
3504 | ||
3505 | un->dfd = READ_ONCE(sqe->fd); | |
3506 | ||
3507 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3508 | if (un->flags & ~AT_REMOVEDIR) | |
3509 | return -EINVAL; | |
3510 | ||
3511 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3512 | un->filename = getname(fname); | |
3513 | if (IS_ERR(un->filename)) | |
3514 | return PTR_ERR(un->filename); | |
3515 | ||
3516 | req->flags |= REQ_F_NEED_CLEANUP; | |
3517 | return 0; | |
3518 | } | |
3519 | ||
45d189c6 | 3520 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3521 | { |
3522 | struct io_unlink *un = &req->unlink; | |
3523 | int ret; | |
3524 | ||
45d189c6 | 3525 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3526 | return -EAGAIN; |
3527 | ||
3528 | if (un->flags & AT_REMOVEDIR) | |
3529 | ret = do_rmdir(un->dfd, un->filename); | |
3530 | else | |
3531 | ret = do_unlinkat(un->dfd, un->filename); | |
3532 | ||
3533 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3534 | if (ret < 0) | |
3535 | req_set_fail_links(req); | |
3536 | io_req_complete(req, ret); | |
3537 | return 0; | |
3538 | } | |
3539 | ||
36f4fa68 JA |
3540 | static int io_shutdown_prep(struct io_kiocb *req, |
3541 | const struct io_uring_sqe *sqe) | |
3542 | { | |
3543 | #if defined(CONFIG_NET) | |
3544 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3545 | return -EINVAL; | |
3546 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3547 | sqe->buf_index) | |
3548 | return -EINVAL; | |
3549 | ||
3550 | req->shutdown.how = READ_ONCE(sqe->len); | |
3551 | return 0; | |
3552 | #else | |
3553 | return -EOPNOTSUPP; | |
3554 | #endif | |
3555 | } | |
3556 | ||
45d189c6 | 3557 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3558 | { |
3559 | #if defined(CONFIG_NET) | |
3560 | struct socket *sock; | |
3561 | int ret; | |
3562 | ||
45d189c6 | 3563 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3564 | return -EAGAIN; |
3565 | ||
48aba79b | 3566 | sock = sock_from_file(req->file); |
36f4fa68 | 3567 | if (unlikely(!sock)) |
48aba79b | 3568 | return -ENOTSOCK; |
36f4fa68 JA |
3569 | |
3570 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3571 | if (ret < 0) |
3572 | req_set_fail_links(req); | |
36f4fa68 JA |
3573 | io_req_complete(req, ret); |
3574 | return 0; | |
3575 | #else | |
3576 | return -EOPNOTSUPP; | |
3577 | #endif | |
3578 | } | |
3579 | ||
f2a8d5c7 PB |
3580 | static int __io_splice_prep(struct io_kiocb *req, |
3581 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3582 | { |
3583 | struct io_splice* sp = &req->splice; | |
3584 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3585 | |
3232dd02 PB |
3586 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3587 | return -EINVAL; | |
7d67af2c PB |
3588 | |
3589 | sp->file_in = NULL; | |
7d67af2c PB |
3590 | sp->len = READ_ONCE(sqe->len); |
3591 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3592 | ||
3593 | if (unlikely(sp->flags & ~valid_flags)) | |
3594 | return -EINVAL; | |
3595 | ||
8371adf5 PB |
3596 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3597 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3598 | if (!sp->file_in) | |
3599 | return -EBADF; | |
7d67af2c | 3600 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3601 | return 0; |
3602 | } | |
3603 | ||
f2a8d5c7 PB |
3604 | static int io_tee_prep(struct io_kiocb *req, |
3605 | const struct io_uring_sqe *sqe) | |
3606 | { | |
3607 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3608 | return -EINVAL; | |
3609 | return __io_splice_prep(req, sqe); | |
3610 | } | |
3611 | ||
45d189c6 | 3612 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3613 | { |
3614 | struct io_splice *sp = &req->splice; | |
3615 | struct file *in = sp->file_in; | |
3616 | struct file *out = sp->file_out; | |
3617 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3618 | long ret = 0; | |
3619 | ||
45d189c6 | 3620 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3621 | return -EAGAIN; |
3622 | if (sp->len) | |
3623 | ret = do_tee(in, out, sp->len, flags); | |
3624 | ||
e1d767f0 PB |
3625 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3626 | io_put_file(in); | |
f2a8d5c7 PB |
3627 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3628 | ||
f2a8d5c7 PB |
3629 | if (ret != sp->len) |
3630 | req_set_fail_links(req); | |
e1e16097 | 3631 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3632 | return 0; |
3633 | } | |
3634 | ||
3635 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3636 | { | |
3637 | struct io_splice* sp = &req->splice; | |
3638 | ||
3639 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3640 | sp->off_out = READ_ONCE(sqe->off); | |
3641 | return __io_splice_prep(req, sqe); | |
3642 | } | |
3643 | ||
45d189c6 | 3644 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3645 | { |
3646 | struct io_splice *sp = &req->splice; | |
3647 | struct file *in = sp->file_in; | |
3648 | struct file *out = sp->file_out; | |
3649 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3650 | loff_t *poff_in, *poff_out; | |
c9687426 | 3651 | long ret = 0; |
7d67af2c | 3652 | |
45d189c6 | 3653 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3654 | return -EAGAIN; |
7d67af2c PB |
3655 | |
3656 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3657 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3658 | |
948a7749 | 3659 | if (sp->len) |
c9687426 | 3660 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3661 | |
e1d767f0 PB |
3662 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3663 | io_put_file(in); | |
7d67af2c PB |
3664 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3665 | ||
7d67af2c PB |
3666 | if (ret != sp->len) |
3667 | req_set_fail_links(req); | |
e1e16097 | 3668 | io_req_complete(req, ret); |
7d67af2c PB |
3669 | return 0; |
3670 | } | |
3671 | ||
2b188cc1 JA |
3672 | /* |
3673 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3674 | */ | |
889fca73 | 3675 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3676 | { |
3677 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3678 | |
def596e9 JA |
3679 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3680 | return -EINVAL; | |
3681 | ||
889fca73 | 3682 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3683 | return 0; |
3684 | } | |
3685 | ||
1155c76a | 3686 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3687 | { |
6b06314c | 3688 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3689 | |
09bb8394 JA |
3690 | if (!req->file) |
3691 | return -EBADF; | |
c992fe29 | 3692 | |
6b06314c | 3693 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3694 | return -EINVAL; |
edafccee | 3695 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3696 | return -EINVAL; |
3697 | ||
8ed8d3c3 JA |
3698 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3699 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3700 | return -EINVAL; | |
3701 | ||
3702 | req->sync.off = READ_ONCE(sqe->off); | |
3703 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3704 | return 0; |
3705 | } | |
3706 | ||
45d189c6 | 3707 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3708 | { |
8ed8d3c3 | 3709 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3710 | int ret; |
3711 | ||
ac45abc0 | 3712 | /* fsync always requires a blocking context */ |
45d189c6 | 3713 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3714 | return -EAGAIN; |
3715 | ||
9adbd45d | 3716 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3717 | end > 0 ? end : LLONG_MAX, |
3718 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3719 | if (ret < 0) | |
3720 | req_set_fail_links(req); | |
e1e16097 | 3721 | io_req_complete(req, ret); |
c992fe29 CH |
3722 | return 0; |
3723 | } | |
3724 | ||
d63d1b5e JA |
3725 | static int io_fallocate_prep(struct io_kiocb *req, |
3726 | const struct io_uring_sqe *sqe) | |
3727 | { | |
3728 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3729 | return -EINVAL; | |
3232dd02 PB |
3730 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3731 | return -EINVAL; | |
d63d1b5e JA |
3732 | |
3733 | req->sync.off = READ_ONCE(sqe->off); | |
3734 | req->sync.len = READ_ONCE(sqe->addr); | |
3735 | req->sync.mode = READ_ONCE(sqe->len); | |
3736 | return 0; | |
3737 | } | |
3738 | ||
45d189c6 | 3739 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3740 | { |
ac45abc0 PB |
3741 | int ret; |
3742 | ||
d63d1b5e | 3743 | /* fallocate always requiring blocking context */ |
45d189c6 | 3744 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3745 | return -EAGAIN; |
ac45abc0 PB |
3746 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3747 | req->sync.len); | |
ac45abc0 PB |
3748 | if (ret < 0) |
3749 | req_set_fail_links(req); | |
e1e16097 | 3750 | io_req_complete(req, ret); |
5d17b4a4 JA |
3751 | return 0; |
3752 | } | |
3753 | ||
ec65fea5 | 3754 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3755 | { |
f8748881 | 3756 | const char __user *fname; |
15b71abe | 3757 | int ret; |
b7bb4f7d | 3758 | |
ec65fea5 | 3759 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3760 | return -EINVAL; |
ec65fea5 | 3761 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3762 | return -EBADF; |
03b1230c | 3763 | |
ec65fea5 PB |
3764 | /* open.how should be already initialised */ |
3765 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3766 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3767 | |
25e72d10 PB |
3768 | req->open.dfd = READ_ONCE(sqe->fd); |
3769 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3770 | req->open.filename = getname(fname); |
15b71abe JA |
3771 | if (IS_ERR(req->open.filename)) { |
3772 | ret = PTR_ERR(req->open.filename); | |
3773 | req->open.filename = NULL; | |
3774 | return ret; | |
3775 | } | |
4022e7af | 3776 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3777 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3778 | return 0; |
03b1230c JA |
3779 | } |
3780 | ||
ec65fea5 PB |
3781 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3782 | { | |
3783 | u64 flags, mode; | |
3784 | ||
14587a46 | 3785 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3786 | return -EINVAL; |
ec65fea5 PB |
3787 | mode = READ_ONCE(sqe->len); |
3788 | flags = READ_ONCE(sqe->open_flags); | |
3789 | req->open.how = build_open_how(flags, mode); | |
3790 | return __io_openat_prep(req, sqe); | |
3791 | } | |
3792 | ||
cebdb986 | 3793 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3794 | { |
cebdb986 | 3795 | struct open_how __user *how; |
cebdb986 | 3796 | size_t len; |
0fa03c62 JA |
3797 | int ret; |
3798 | ||
14587a46 | 3799 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3800 | return -EINVAL; |
cebdb986 JA |
3801 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3802 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3803 | if (len < OPEN_HOW_SIZE_VER0) |
3804 | return -EINVAL; | |
3529d8c2 | 3805 | |
cebdb986 JA |
3806 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3807 | len); | |
3808 | if (ret) | |
3809 | return ret; | |
3529d8c2 | 3810 | |
ec65fea5 | 3811 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3812 | } |
3813 | ||
45d189c6 | 3814 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3815 | { |
3816 | struct open_flags op; | |
15b71abe | 3817 | struct file *file; |
3a81fd02 JA |
3818 | bool nonblock_set; |
3819 | bool resolve_nonblock; | |
15b71abe JA |
3820 | int ret; |
3821 | ||
cebdb986 | 3822 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3823 | if (ret) |
3824 | goto err; | |
3a81fd02 JA |
3825 | nonblock_set = op.open_flag & O_NONBLOCK; |
3826 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3827 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3828 | /* |
3829 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3830 | * it'll always -EAGAIN | |
3831 | */ | |
3832 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3833 | return -EAGAIN; | |
3834 | op.lookup_flags |= LOOKUP_CACHED; | |
3835 | op.open_flag |= O_NONBLOCK; | |
3836 | } | |
15b71abe | 3837 | |
4022e7af | 3838 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3839 | if (ret < 0) |
3840 | goto err; | |
3841 | ||
3842 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3843 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3844 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3845 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3846 | /* |
3a81fd02 JA |
3847 | * We could hang on to this 'fd', but seems like marginal |
3848 | * gain for something that is now known to be a slower path. | |
3849 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3850 | */ |
3a81fd02 JA |
3851 | put_unused_fd(ret); |
3852 | return -EAGAIN; | |
3853 | } | |
3854 | ||
15b71abe JA |
3855 | if (IS_ERR(file)) { |
3856 | put_unused_fd(ret); | |
3857 | ret = PTR_ERR(file); | |
3858 | } else { | |
45d189c6 | 3859 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3860 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3861 | fsnotify_open(file); |
3862 | fd_install(ret, file); | |
3863 | } | |
3864 | err: | |
3865 | putname(req->open.filename); | |
8fef80bf | 3866 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3867 | if (ret < 0) |
3868 | req_set_fail_links(req); | |
e1e16097 | 3869 | io_req_complete(req, ret); |
15b71abe JA |
3870 | return 0; |
3871 | } | |
3872 | ||
45d189c6 | 3873 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3874 | { |
e45cff58 | 3875 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3876 | } |
3877 | ||
067524e9 JA |
3878 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3879 | const struct io_uring_sqe *sqe) | |
3880 | { | |
3881 | struct io_provide_buf *p = &req->pbuf; | |
3882 | u64 tmp; | |
3883 | ||
3884 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3885 | return -EINVAL; | |
3886 | ||
3887 | tmp = READ_ONCE(sqe->fd); | |
3888 | if (!tmp || tmp > USHRT_MAX) | |
3889 | return -EINVAL; | |
3890 | ||
3891 | memset(p, 0, sizeof(*p)); | |
3892 | p->nbufs = tmp; | |
3893 | p->bgid = READ_ONCE(sqe->buf_group); | |
3894 | return 0; | |
3895 | } | |
3896 | ||
3897 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3898 | int bgid, unsigned nbufs) | |
3899 | { | |
3900 | unsigned i = 0; | |
3901 | ||
3902 | /* shouldn't happen */ | |
3903 | if (!nbufs) | |
3904 | return 0; | |
3905 | ||
3906 | /* the head kbuf is the list itself */ | |
3907 | while (!list_empty(&buf->list)) { | |
3908 | struct io_buffer *nxt; | |
3909 | ||
3910 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3911 | list_del(&nxt->list); | |
3912 | kfree(nxt); | |
3913 | if (++i == nbufs) | |
3914 | return i; | |
3915 | } | |
3916 | i++; | |
3917 | kfree(buf); | |
9e15c3a0 | 3918 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3919 | |
3920 | return i; | |
3921 | } | |
3922 | ||
889fca73 | 3923 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3924 | { |
3925 | struct io_provide_buf *p = &req->pbuf; | |
3926 | struct io_ring_ctx *ctx = req->ctx; | |
3927 | struct io_buffer *head; | |
3928 | int ret = 0; | |
45d189c6 | 3929 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3930 | |
3931 | io_ring_submit_lock(ctx, !force_nonblock); | |
3932 | ||
3933 | lockdep_assert_held(&ctx->uring_lock); | |
3934 | ||
3935 | ret = -ENOENT; | |
9e15c3a0 | 3936 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3937 | if (head) |
3938 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3939 | if (ret < 0) |
3940 | req_set_fail_links(req); | |
067524e9 | 3941 | |
9fb8cb49 PB |
3942 | /* complete before unlock, IOPOLL may need the lock */ |
3943 | __io_req_complete(req, issue_flags, ret, 0); | |
3944 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3945 | return 0; |
3946 | } | |
3947 | ||
ddf0322d JA |
3948 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3949 | const struct io_uring_sqe *sqe) | |
3950 | { | |
d81269fe | 3951 | unsigned long size; |
ddf0322d JA |
3952 | struct io_provide_buf *p = &req->pbuf; |
3953 | u64 tmp; | |
3954 | ||
3955 | if (sqe->ioprio || sqe->rw_flags) | |
3956 | return -EINVAL; | |
3957 | ||
3958 | tmp = READ_ONCE(sqe->fd); | |
3959 | if (!tmp || tmp > USHRT_MAX) | |
3960 | return -E2BIG; | |
3961 | p->nbufs = tmp; | |
3962 | p->addr = READ_ONCE(sqe->addr); | |
3963 | p->len = READ_ONCE(sqe->len); | |
3964 | ||
d81269fe PB |
3965 | size = (unsigned long)p->len * p->nbufs; |
3966 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3967 | return -EFAULT; |
3968 | ||
3969 | p->bgid = READ_ONCE(sqe->buf_group); | |
3970 | tmp = READ_ONCE(sqe->off); | |
3971 | if (tmp > USHRT_MAX) | |
3972 | return -E2BIG; | |
3973 | p->bid = tmp; | |
3974 | return 0; | |
3975 | } | |
3976 | ||
3977 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3978 | { | |
3979 | struct io_buffer *buf; | |
3980 | u64 addr = pbuf->addr; | |
3981 | int i, bid = pbuf->bid; | |
3982 | ||
3983 | for (i = 0; i < pbuf->nbufs; i++) { | |
3984 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3985 | if (!buf) | |
3986 | break; | |
3987 | ||
3988 | buf->addr = addr; | |
3989 | buf->len = pbuf->len; | |
3990 | buf->bid = bid; | |
3991 | addr += pbuf->len; | |
3992 | bid++; | |
3993 | if (!*head) { | |
3994 | INIT_LIST_HEAD(&buf->list); | |
3995 | *head = buf; | |
3996 | } else { | |
3997 | list_add_tail(&buf->list, &(*head)->list); | |
3998 | } | |
3999 | } | |
4000 | ||
4001 | return i ? i : -ENOMEM; | |
4002 | } | |
4003 | ||
889fca73 | 4004 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4005 | { |
4006 | struct io_provide_buf *p = &req->pbuf; | |
4007 | struct io_ring_ctx *ctx = req->ctx; | |
4008 | struct io_buffer *head, *list; | |
4009 | int ret = 0; | |
45d189c6 | 4010 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4011 | |
4012 | io_ring_submit_lock(ctx, !force_nonblock); | |
4013 | ||
4014 | lockdep_assert_held(&ctx->uring_lock); | |
4015 | ||
9e15c3a0 | 4016 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4017 | |
4018 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4019 | if (ret >= 0 && !list) { |
4020 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4021 | if (ret < 0) | |
067524e9 | 4022 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4023 | } |
ddf0322d JA |
4024 | if (ret < 0) |
4025 | req_set_fail_links(req); | |
9fb8cb49 PB |
4026 | /* complete before unlock, IOPOLL may need the lock */ |
4027 | __io_req_complete(req, issue_flags, ret, 0); | |
4028 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4029 | return 0; |
cebdb986 JA |
4030 | } |
4031 | ||
3e4827b0 JA |
4032 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4033 | const struct io_uring_sqe *sqe) | |
4034 | { | |
4035 | #if defined(CONFIG_EPOLL) | |
4036 | if (sqe->ioprio || sqe->buf_index) | |
4037 | return -EINVAL; | |
6ca56f84 | 4038 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4039 | return -EINVAL; |
3e4827b0 JA |
4040 | |
4041 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4042 | req->epoll.op = READ_ONCE(sqe->len); | |
4043 | req->epoll.fd = READ_ONCE(sqe->off); | |
4044 | ||
4045 | if (ep_op_has_event(req->epoll.op)) { | |
4046 | struct epoll_event __user *ev; | |
4047 | ||
4048 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4049 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4050 | return -EFAULT; | |
4051 | } | |
4052 | ||
4053 | return 0; | |
4054 | #else | |
4055 | return -EOPNOTSUPP; | |
4056 | #endif | |
4057 | } | |
4058 | ||
889fca73 | 4059 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4060 | { |
4061 | #if defined(CONFIG_EPOLL) | |
4062 | struct io_epoll *ie = &req->epoll; | |
4063 | int ret; | |
45d189c6 | 4064 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4065 | |
4066 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4067 | if (force_nonblock && ret == -EAGAIN) | |
4068 | return -EAGAIN; | |
4069 | ||
4070 | if (ret < 0) | |
4071 | req_set_fail_links(req); | |
889fca73 | 4072 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4073 | return 0; |
4074 | #else | |
4075 | return -EOPNOTSUPP; | |
4076 | #endif | |
4077 | } | |
4078 | ||
c1ca757b JA |
4079 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4080 | { | |
4081 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4082 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4083 | return -EINVAL; | |
3232dd02 PB |
4084 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4085 | return -EINVAL; | |
c1ca757b JA |
4086 | |
4087 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4088 | req->madvise.len = READ_ONCE(sqe->len); | |
4089 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4090 | return 0; | |
4091 | #else | |
4092 | return -EOPNOTSUPP; | |
4093 | #endif | |
4094 | } | |
4095 | ||
45d189c6 | 4096 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4097 | { |
4098 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4099 | struct io_madvise *ma = &req->madvise; | |
4100 | int ret; | |
4101 | ||
45d189c6 | 4102 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4103 | return -EAGAIN; |
4104 | ||
0726b01e | 4105 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4106 | if (ret < 0) |
4107 | req_set_fail_links(req); | |
e1e16097 | 4108 | io_req_complete(req, ret); |
c1ca757b JA |
4109 | return 0; |
4110 | #else | |
4111 | return -EOPNOTSUPP; | |
4112 | #endif | |
4113 | } | |
4114 | ||
4840e418 JA |
4115 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4116 | { | |
4117 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4118 | return -EINVAL; | |
3232dd02 PB |
4119 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4120 | return -EINVAL; | |
4840e418 JA |
4121 | |
4122 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4123 | req->fadvise.len = READ_ONCE(sqe->len); | |
4124 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4125 | return 0; | |
4126 | } | |
4127 | ||
45d189c6 | 4128 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4129 | { |
4130 | struct io_fadvise *fa = &req->fadvise; | |
4131 | int ret; | |
4132 | ||
45d189c6 | 4133 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4134 | switch (fa->advice) { |
4135 | case POSIX_FADV_NORMAL: | |
4136 | case POSIX_FADV_RANDOM: | |
4137 | case POSIX_FADV_SEQUENTIAL: | |
4138 | break; | |
4139 | default: | |
4140 | return -EAGAIN; | |
4141 | } | |
4142 | } | |
4840e418 JA |
4143 | |
4144 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4145 | if (ret < 0) | |
4146 | req_set_fail_links(req); | |
e1e16097 | 4147 | io_req_complete(req, ret); |
4840e418 JA |
4148 | return 0; |
4149 | } | |
4150 | ||
eddc7ef5 JA |
4151 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4152 | { | |
6ca56f84 | 4153 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4154 | return -EINVAL; |
eddc7ef5 JA |
4155 | if (sqe->ioprio || sqe->buf_index) |
4156 | return -EINVAL; | |
9c280f90 | 4157 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4158 | return -EBADF; |
eddc7ef5 | 4159 | |
1d9e1288 BM |
4160 | req->statx.dfd = READ_ONCE(sqe->fd); |
4161 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4162 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4163 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4164 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4165 | |
4166 | return 0; | |
4167 | } | |
4168 | ||
45d189c6 | 4169 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4170 | { |
1d9e1288 | 4171 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4172 | int ret; |
4173 | ||
59d70013 | 4174 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4175 | return -EAGAIN; |
4176 | ||
e62753e4 BM |
4177 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4178 | ctx->buffer); | |
eddc7ef5 | 4179 | |
eddc7ef5 JA |
4180 | if (ret < 0) |
4181 | req_set_fail_links(req); | |
e1e16097 | 4182 | io_req_complete(req, ret); |
eddc7ef5 JA |
4183 | return 0; |
4184 | } | |
4185 | ||
b5dba59e JA |
4186 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4187 | { | |
14587a46 | 4188 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4189 | return -EINVAL; |
b5dba59e JA |
4190 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4191 | sqe->rw_flags || sqe->buf_index) | |
4192 | return -EINVAL; | |
9c280f90 | 4193 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4194 | return -EBADF; |
b5dba59e JA |
4195 | |
4196 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4197 | return 0; |
b5dba59e JA |
4198 | } |
4199 | ||
889fca73 | 4200 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4201 | { |
9eac1904 | 4202 | struct files_struct *files = current->files; |
3af73b28 | 4203 | struct io_close *close = &req->close; |
9eac1904 JA |
4204 | struct fdtable *fdt; |
4205 | struct file *file; | |
b5dba59e JA |
4206 | int ret; |
4207 | ||
9eac1904 JA |
4208 | file = NULL; |
4209 | ret = -EBADF; | |
4210 | spin_lock(&files->file_lock); | |
4211 | fdt = files_fdtable(files); | |
4212 | if (close->fd >= fdt->max_fds) { | |
4213 | spin_unlock(&files->file_lock); | |
4214 | goto err; | |
4215 | } | |
4216 | file = fdt->fd[close->fd]; | |
4217 | if (!file) { | |
4218 | spin_unlock(&files->file_lock); | |
4219 | goto err; | |
4220 | } | |
4221 | ||
4222 | if (file->f_op == &io_uring_fops) { | |
4223 | spin_unlock(&files->file_lock); | |
4224 | file = NULL; | |
4225 | goto err; | |
3af73b28 | 4226 | } |
b5dba59e JA |
4227 | |
4228 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4229 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4230 | spin_unlock(&files->file_lock); |
0bf0eefd | 4231 | return -EAGAIN; |
a2100672 | 4232 | } |
b5dba59e | 4233 | |
9eac1904 JA |
4234 | ret = __close_fd_get_file(close->fd, &file); |
4235 | spin_unlock(&files->file_lock); | |
4236 | if (ret < 0) { | |
4237 | if (ret == -ENOENT) | |
4238 | ret = -EBADF; | |
4239 | goto err; | |
4240 | } | |
4241 | ||
3af73b28 | 4242 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4243 | ret = filp_close(file, current->files); |
4244 | err: | |
3af73b28 PB |
4245 | if (ret < 0) |
4246 | req_set_fail_links(req); | |
9eac1904 JA |
4247 | if (file) |
4248 | fput(file); | |
889fca73 | 4249 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4250 | return 0; |
b5dba59e JA |
4251 | } |
4252 | ||
1155c76a | 4253 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4254 | { |
4255 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4256 | |
5d17b4a4 JA |
4257 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4258 | return -EINVAL; | |
4259 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4260 | return -EINVAL; | |
4261 | ||
8ed8d3c3 JA |
4262 | req->sync.off = READ_ONCE(sqe->off); |
4263 | req->sync.len = READ_ONCE(sqe->len); | |
4264 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4265 | return 0; |
4266 | } | |
4267 | ||
45d189c6 | 4268 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4269 | { |
8ed8d3c3 JA |
4270 | int ret; |
4271 | ||
ac45abc0 | 4272 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4273 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4274 | return -EAGAIN; |
4275 | ||
9adbd45d | 4276 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4277 | req->sync.flags); |
4278 | if (ret < 0) | |
4279 | req_set_fail_links(req); | |
e1e16097 | 4280 | io_req_complete(req, ret); |
5d17b4a4 JA |
4281 | return 0; |
4282 | } | |
4283 | ||
469956e8 | 4284 | #if defined(CONFIG_NET) |
02d27d89 PB |
4285 | static int io_setup_async_msg(struct io_kiocb *req, |
4286 | struct io_async_msghdr *kmsg) | |
4287 | { | |
e8c2bc1f JA |
4288 | struct io_async_msghdr *async_msg = req->async_data; |
4289 | ||
4290 | if (async_msg) | |
02d27d89 | 4291 | return -EAGAIN; |
e8c2bc1f | 4292 | if (io_alloc_async_data(req)) { |
257e84a5 | 4293 | kfree(kmsg->free_iov); |
02d27d89 PB |
4294 | return -ENOMEM; |
4295 | } | |
e8c2bc1f | 4296 | async_msg = req->async_data; |
02d27d89 | 4297 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4298 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4299 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4300 | /* if were using fast_iov, set it to the new one */ |
4301 | if (!async_msg->free_iov) | |
4302 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4303 | ||
02d27d89 PB |
4304 | return -EAGAIN; |
4305 | } | |
4306 | ||
2ae523ed PB |
4307 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4308 | struct io_async_msghdr *iomsg) | |
4309 | { | |
2ae523ed | 4310 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4311 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4312 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4313 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4314 | } |
4315 | ||
93642ef8 PB |
4316 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4317 | { | |
4318 | int ret; | |
4319 | ||
93642ef8 PB |
4320 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4321 | if (!ret) | |
4322 | req->flags |= REQ_F_NEED_CLEANUP; | |
4323 | return ret; | |
4324 | } | |
4325 | ||
3529d8c2 | 4326 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4327 | { |
e47293fd | 4328 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4329 | |
d2b6f48b PB |
4330 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4331 | return -EINVAL; | |
4332 | ||
e47293fd | 4333 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4334 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4335 | sr->len = READ_ONCE(sqe->len); |
3529d8c2 | 4336 | |
d8768362 JA |
4337 | #ifdef CONFIG_COMPAT |
4338 | if (req->ctx->compat) | |
4339 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4340 | #endif | |
93642ef8 | 4341 | return 0; |
03b1230c JA |
4342 | } |
4343 | ||
889fca73 | 4344 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4345 | { |
6b754c8b | 4346 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4347 | struct socket *sock; |
7a7cacba | 4348 | unsigned flags; |
0031275d | 4349 | int min_ret = 0; |
0fa03c62 JA |
4350 | int ret; |
4351 | ||
dba4a925 | 4352 | sock = sock_from_file(req->file); |
7a7cacba | 4353 | if (unlikely(!sock)) |
dba4a925 | 4354 | return -ENOTSOCK; |
3529d8c2 | 4355 | |
257e84a5 PB |
4356 | kmsg = req->async_data; |
4357 | if (!kmsg) { | |
7a7cacba PB |
4358 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4359 | if (ret) | |
4360 | return ret; | |
4361 | kmsg = &iomsg; | |
0fa03c62 | 4362 | } |
0fa03c62 | 4363 | |
76cd979f | 4364 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4365 | if (flags & MSG_DONTWAIT) |
4366 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4367 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4368 | flags |= MSG_DONTWAIT; |
e47293fd | 4369 | |
0031275d SM |
4370 | if (flags & MSG_WAITALL) |
4371 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4372 | ||
7a7cacba | 4373 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4374 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4375 | return io_setup_async_msg(req, kmsg); |
4376 | if (ret == -ERESTARTSYS) | |
4377 | ret = -EINTR; | |
0fa03c62 | 4378 | |
257e84a5 PB |
4379 | /* fast path, check for non-NULL to avoid function call */ |
4380 | if (kmsg->free_iov) | |
4381 | kfree(kmsg->free_iov); | |
99bc4c38 | 4382 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4383 | if (ret < min_ret) |
4e88d6e7 | 4384 | req_set_fail_links(req); |
889fca73 | 4385 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4386 | return 0; |
03b1230c | 4387 | } |
aa1fa28f | 4388 | |
889fca73 | 4389 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4390 | { |
7a7cacba PB |
4391 | struct io_sr_msg *sr = &req->sr_msg; |
4392 | struct msghdr msg; | |
4393 | struct iovec iov; | |
fddaface | 4394 | struct socket *sock; |
7a7cacba | 4395 | unsigned flags; |
0031275d | 4396 | int min_ret = 0; |
fddaface JA |
4397 | int ret; |
4398 | ||
dba4a925 | 4399 | sock = sock_from_file(req->file); |
7a7cacba | 4400 | if (unlikely(!sock)) |
dba4a925 | 4401 | return -ENOTSOCK; |
fddaface | 4402 | |
7a7cacba PB |
4403 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4404 | if (unlikely(ret)) | |
14db8411 | 4405 | return ret; |
fddaface | 4406 | |
7a7cacba PB |
4407 | msg.msg_name = NULL; |
4408 | msg.msg_control = NULL; | |
4409 | msg.msg_controllen = 0; | |
4410 | msg.msg_namelen = 0; | |
fddaface | 4411 | |
76cd979f | 4412 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4413 | if (flags & MSG_DONTWAIT) |
4414 | req->flags |= REQ_F_NOWAIT; | |
45d189c6 | 4415 | else if (issue_flags & IO_URING_F_NONBLOCK) |
7a7cacba | 4416 | flags |= MSG_DONTWAIT; |
fddaface | 4417 | |
0031275d SM |
4418 | if (flags & MSG_WAITALL) |
4419 | min_ret = iov_iter_count(&msg.msg_iter); | |
4420 | ||
7a7cacba PB |
4421 | msg.msg_flags = flags; |
4422 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4423 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4424 | return -EAGAIN; |
4425 | if (ret == -ERESTARTSYS) | |
4426 | ret = -EINTR; | |
fddaface | 4427 | |
0031275d | 4428 | if (ret < min_ret) |
fddaface | 4429 | req_set_fail_links(req); |
889fca73 | 4430 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4431 | return 0; |
fddaface JA |
4432 | } |
4433 | ||
1400e697 PB |
4434 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4435 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4436 | { |
4437 | struct io_sr_msg *sr = &req->sr_msg; | |
4438 | struct iovec __user *uiov; | |
4439 | size_t iov_len; | |
4440 | int ret; | |
4441 | ||
1400e697 PB |
4442 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4443 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4444 | if (ret) |
4445 | return ret; | |
4446 | ||
4447 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4448 | if (iov_len > 1) | |
4449 | return -EINVAL; | |
5476dfed | 4450 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4451 | return -EFAULT; |
5476dfed | 4452 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4453 | iomsg->free_iov = NULL; |
52de1fe1 | 4454 | } else { |
257e84a5 | 4455 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4456 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4457 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4458 | false); |
52de1fe1 JA |
4459 | if (ret > 0) |
4460 | ret = 0; | |
4461 | } | |
4462 | ||
4463 | return ret; | |
4464 | } | |
4465 | ||
4466 | #ifdef CONFIG_COMPAT | |
4467 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4468 | struct io_async_msghdr *iomsg) |
52de1fe1 JA |
4469 | { |
4470 | struct compat_msghdr __user *msg_compat; | |
4471 | struct io_sr_msg *sr = &req->sr_msg; | |
4472 | struct compat_iovec __user *uiov; | |
4473 | compat_uptr_t ptr; | |
4474 | compat_size_t len; | |
4475 | int ret; | |
4476 | ||
270a5940 | 4477 | msg_compat = (struct compat_msghdr __user *) sr->umsg; |
1400e697 | 4478 | ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr, |
52de1fe1 JA |
4479 | &ptr, &len); |
4480 | if (ret) | |
4481 | return ret; | |
4482 | ||
4483 | uiov = compat_ptr(ptr); | |
4484 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4485 | compat_ssize_t clen; | |
4486 | ||
4487 | if (len > 1) | |
4488 | return -EINVAL; | |
4489 | if (!access_ok(uiov, sizeof(*uiov))) | |
4490 | return -EFAULT; | |
4491 | if (__get_user(clen, &uiov->iov_len)) | |
4492 | return -EFAULT; | |
4493 | if (clen < 0) | |
4494 | return -EINVAL; | |
2d280bc8 | 4495 | sr->len = clen; |
257e84a5 | 4496 | iomsg->free_iov = NULL; |
52de1fe1 | 4497 | } else { |
257e84a5 | 4498 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4499 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4500 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4501 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4502 | if (ret < 0) |
4503 | return ret; | |
4504 | } | |
4505 | ||
4506 | return 0; | |
4507 | } | |
4508 | #endif | |
4509 | ||
1400e697 PB |
4510 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4511 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4512 | { |
1400e697 | 4513 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4514 | |
4515 | #ifdef CONFIG_COMPAT | |
4516 | if (req->ctx->compat) | |
1400e697 | 4517 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4518 | #endif |
52de1fe1 | 4519 | |
1400e697 | 4520 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4521 | } |
4522 | ||
bcda7baa | 4523 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4524 | bool needs_lock) |
bcda7baa JA |
4525 | { |
4526 | struct io_sr_msg *sr = &req->sr_msg; | |
4527 | struct io_buffer *kbuf; | |
4528 | ||
bcda7baa JA |
4529 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4530 | if (IS_ERR(kbuf)) | |
4531 | return kbuf; | |
4532 | ||
4533 | sr->kbuf = kbuf; | |
4534 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4535 | return kbuf; |
fddaface JA |
4536 | } |
4537 | ||
7fbb1b54 PB |
4538 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4539 | { | |
4540 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4541 | } | |
4542 | ||
93642ef8 | 4543 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4544 | { |
99bc4c38 | 4545 | int ret; |
3529d8c2 | 4546 | |
93642ef8 PB |
4547 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4548 | if (!ret) | |
4549 | req->flags |= REQ_F_NEED_CLEANUP; | |
4550 | return ret; | |
4551 | } | |
4552 | ||
4553 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4554 | { | |
4555 | struct io_sr_msg *sr = &req->sr_msg; | |
4556 | ||
d2b6f48b PB |
4557 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4558 | return -EINVAL; | |
4559 | ||
3529d8c2 | 4560 | sr->msg_flags = READ_ONCE(sqe->msg_flags); |
270a5940 | 4561 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4562 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4563 | sr->bgid = READ_ONCE(sqe->buf_group); |
06b76d44 | 4564 | |
d8768362 JA |
4565 | #ifdef CONFIG_COMPAT |
4566 | if (req->ctx->compat) | |
4567 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4568 | #endif | |
93642ef8 | 4569 | return 0; |
aa1fa28f JA |
4570 | } |
4571 | ||
889fca73 | 4572 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4573 | { |
6b754c8b | 4574 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4575 | struct socket *sock; |
7fbb1b54 | 4576 | struct io_buffer *kbuf; |
7a7cacba | 4577 | unsigned flags; |
0031275d | 4578 | int min_ret = 0; |
52de1fe1 | 4579 | int ret, cflags = 0; |
45d189c6 | 4580 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4581 | |
dba4a925 | 4582 | sock = sock_from_file(req->file); |
7a7cacba | 4583 | if (unlikely(!sock)) |
dba4a925 | 4584 | return -ENOTSOCK; |
3529d8c2 | 4585 | |
257e84a5 PB |
4586 | kmsg = req->async_data; |
4587 | if (!kmsg) { | |
7a7cacba PB |
4588 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4589 | if (ret) | |
681fda8d | 4590 | return ret; |
7a7cacba PB |
4591 | kmsg = &iomsg; |
4592 | } | |
03b1230c | 4593 | |
bc02ef33 | 4594 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4595 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4596 | if (IS_ERR(kbuf)) |
52de1fe1 | 4597 | return PTR_ERR(kbuf); |
7a7cacba | 4598 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4599 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4600 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4601 | 1, req->sr_msg.len); |
4602 | } | |
52de1fe1 | 4603 | |
76cd979f | 4604 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4605 | if (flags & MSG_DONTWAIT) |
4606 | req->flags |= REQ_F_NOWAIT; | |
4607 | else if (force_nonblock) | |
4608 | flags |= MSG_DONTWAIT; | |
e47293fd | 4609 | |
0031275d SM |
4610 | if (flags & MSG_WAITALL) |
4611 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4612 | ||
7a7cacba PB |
4613 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4614 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4615 | if (force_nonblock && ret == -EAGAIN) |
4616 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4617 | if (ret == -ERESTARTSYS) |
4618 | ret = -EINTR; | |
03b1230c | 4619 | |
7fbb1b54 PB |
4620 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4621 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4622 | /* fast path, check for non-NULL to avoid function call */ |
4623 | if (kmsg->free_iov) | |
4624 | kfree(kmsg->free_iov); | |
99bc4c38 | 4625 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4626 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4627 | req_set_fail_links(req); |
889fca73 | 4628 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4629 | return 0; |
0fa03c62 | 4630 | } |
5d17b4a4 | 4631 | |
889fca73 | 4632 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4633 | { |
6b754c8b | 4634 | struct io_buffer *kbuf; |
7a7cacba PB |
4635 | struct io_sr_msg *sr = &req->sr_msg; |
4636 | struct msghdr msg; | |
4637 | void __user *buf = sr->buf; | |
fddaface | 4638 | struct socket *sock; |
7a7cacba PB |
4639 | struct iovec iov; |
4640 | unsigned flags; | |
0031275d | 4641 | int min_ret = 0; |
bcda7baa | 4642 | int ret, cflags = 0; |
45d189c6 | 4643 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4644 | |
dba4a925 | 4645 | sock = sock_from_file(req->file); |
7a7cacba | 4646 | if (unlikely(!sock)) |
dba4a925 | 4647 | return -ENOTSOCK; |
fddaface | 4648 | |
bc02ef33 | 4649 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4650 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4651 | if (IS_ERR(kbuf)) |
4652 | return PTR_ERR(kbuf); | |
7a7cacba | 4653 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4654 | } |
bcda7baa | 4655 | |
7a7cacba | 4656 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4657 | if (unlikely(ret)) |
4658 | goto out_free; | |
fddaface | 4659 | |
7a7cacba PB |
4660 | msg.msg_name = NULL; |
4661 | msg.msg_control = NULL; | |
4662 | msg.msg_controllen = 0; | |
4663 | msg.msg_namelen = 0; | |
4664 | msg.msg_iocb = NULL; | |
4665 | msg.msg_flags = 0; | |
fddaface | 4666 | |
76cd979f | 4667 | flags = req->sr_msg.msg_flags | MSG_NOSIGNAL; |
7a7cacba PB |
4668 | if (flags & MSG_DONTWAIT) |
4669 | req->flags |= REQ_F_NOWAIT; | |
4670 | else if (force_nonblock) | |
4671 | flags |= MSG_DONTWAIT; | |
4672 | ||
0031275d SM |
4673 | if (flags & MSG_WAITALL) |
4674 | min_ret = iov_iter_count(&msg.msg_iter); | |
4675 | ||
7a7cacba PB |
4676 | ret = sock_recvmsg(sock, &msg, flags); |
4677 | if (force_nonblock && ret == -EAGAIN) | |
4678 | return -EAGAIN; | |
4679 | if (ret == -ERESTARTSYS) | |
4680 | ret = -EINTR; | |
14c32eee | 4681 | out_free: |
7fbb1b54 PB |
4682 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4683 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4684 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4685 | req_set_fail_links(req); |
889fca73 | 4686 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4687 | return 0; |
fddaface JA |
4688 | } |
4689 | ||
3529d8c2 | 4690 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4691 | { |
8ed8d3c3 JA |
4692 | struct io_accept *accept = &req->accept; |
4693 | ||
14587a46 | 4694 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4695 | return -EINVAL; |
8042d6ce | 4696 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4697 | return -EINVAL; |
4698 | ||
d55e5f5b JA |
4699 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4700 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4701 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4702 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4703 | return 0; |
8ed8d3c3 | 4704 | } |
17f2fe35 | 4705 | |
889fca73 | 4706 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4707 | { |
4708 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4709 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4710 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4711 | int ret; |
4712 | ||
e697deed JX |
4713 | if (req->file->f_flags & O_NONBLOCK) |
4714 | req->flags |= REQ_F_NOWAIT; | |
4715 | ||
8ed8d3c3 | 4716 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4717 | accept->addr_len, accept->flags, |
4718 | accept->nofile); | |
8ed8d3c3 | 4719 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4720 | return -EAGAIN; |
ac45abc0 PB |
4721 | if (ret < 0) { |
4722 | if (ret == -ERESTARTSYS) | |
4723 | ret = -EINTR; | |
4e88d6e7 | 4724 | req_set_fail_links(req); |
ac45abc0 | 4725 | } |
889fca73 | 4726 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4727 | return 0; |
8ed8d3c3 JA |
4728 | } |
4729 | ||
93642ef8 PB |
4730 | static int io_connect_prep_async(struct io_kiocb *req) |
4731 | { | |
4732 | struct io_async_connect *io = req->async_data; | |
4733 | struct io_connect *conn = &req->connect; | |
4734 | ||
4735 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4736 | } | |
4737 | ||
3529d8c2 | 4738 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4739 | { |
3529d8c2 | 4740 | struct io_connect *conn = &req->connect; |
f499a021 | 4741 | |
14587a46 | 4742 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4743 | return -EINVAL; |
4744 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4745 | return -EINVAL; | |
4746 | ||
3529d8c2 JA |
4747 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4748 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4749 | return 0; |
f499a021 JA |
4750 | } |
4751 | ||
889fca73 | 4752 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4753 | { |
e8c2bc1f | 4754 | struct io_async_connect __io, *io; |
f8e85cf2 | 4755 | unsigned file_flags; |
3fbb51c1 | 4756 | int ret; |
45d189c6 | 4757 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4758 | |
e8c2bc1f JA |
4759 | if (req->async_data) { |
4760 | io = req->async_data; | |
f499a021 | 4761 | } else { |
3529d8c2 JA |
4762 | ret = move_addr_to_kernel(req->connect.addr, |
4763 | req->connect.addr_len, | |
e8c2bc1f | 4764 | &__io.address); |
f499a021 JA |
4765 | if (ret) |
4766 | goto out; | |
4767 | io = &__io; | |
4768 | } | |
4769 | ||
3fbb51c1 JA |
4770 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4771 | ||
e8c2bc1f | 4772 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4773 | req->connect.addr_len, file_flags); |
87f80d62 | 4774 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4775 | if (req->async_data) |
b7bb4f7d | 4776 | return -EAGAIN; |
e8c2bc1f | 4777 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4778 | ret = -ENOMEM; |
4779 | goto out; | |
4780 | } | |
e8c2bc1f | 4781 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4782 | return -EAGAIN; |
f499a021 | 4783 | } |
f8e85cf2 JA |
4784 | if (ret == -ERESTARTSYS) |
4785 | ret = -EINTR; | |
f499a021 | 4786 | out: |
4e88d6e7 JA |
4787 | if (ret < 0) |
4788 | req_set_fail_links(req); | |
889fca73 | 4789 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4790 | return 0; |
469956e8 Y |
4791 | } |
4792 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4793 | #define IO_NETOP_FN(op) \ |
4794 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4795 | { \ | |
4796 | return -EOPNOTSUPP; \ | |
4797 | } | |
4798 | ||
4799 | #define IO_NETOP_PREP(op) \ | |
4800 | IO_NETOP_FN(op) \ | |
4801 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4802 | { \ | |
4803 | return -EOPNOTSUPP; \ | |
4804 | } \ | |
4805 | ||
4806 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4807 | IO_NETOP_PREP(op) \ | |
4808 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4809 | { \ | |
4810 | return -EOPNOTSUPP; \ | |
4811 | } | |
4812 | ||
4813 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4814 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4815 | IO_NETOP_PREP_ASYNC(connect); | |
4816 | IO_NETOP_PREP(accept); | |
4817 | IO_NETOP_FN(send); | |
4818 | IO_NETOP_FN(recv); | |
469956e8 | 4819 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4820 | |
d7718a9d JA |
4821 | struct io_poll_table { |
4822 | struct poll_table_struct pt; | |
4823 | struct io_kiocb *req; | |
4824 | int error; | |
4825 | }; | |
ce593a6c | 4826 | |
d7718a9d JA |
4827 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4828 | __poll_t mask, task_work_func_t func) | |
4829 | { | |
aa96bf8a | 4830 | int ret; |
d7718a9d JA |
4831 | |
4832 | /* for instances that support it check for an event match first: */ | |
4833 | if (mask && !(mask & poll->events)) | |
4834 | return 0; | |
4835 | ||
4836 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4837 | ||
4838 | list_del_init(&poll->wait.entry); | |
4839 | ||
d7718a9d | 4840 | req->result = mask; |
7cbf1722 | 4841 | req->task_work.func = func; |
6d816e08 | 4842 | |
d7718a9d | 4843 | /* |
e3aabf95 JA |
4844 | * If this fails, then the task is exiting. When a task exits, the |
4845 | * work gets canceled, so just cancel this request as well instead | |
4846 | * of executing it. We can't safely execute it anyway, as we may not | |
4847 | * have the needed state needed for it anyway. | |
d7718a9d | 4848 | */ |
355fb9e2 | 4849 | ret = io_req_task_work_add(req); |
aa96bf8a | 4850 | if (unlikely(ret)) { |
e3aabf95 | 4851 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4852 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4853 | } |
d7718a9d JA |
4854 | return 1; |
4855 | } | |
4856 | ||
74ce6ce4 JA |
4857 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4858 | __acquires(&req->ctx->completion_lock) | |
4859 | { | |
4860 | struct io_ring_ctx *ctx = req->ctx; | |
4861 | ||
4862 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4863 | struct poll_table_struct pt = { ._key = poll->events }; | |
4864 | ||
4865 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4866 | } | |
4867 | ||
4868 | spin_lock_irq(&ctx->completion_lock); | |
4869 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4870 | add_wait_queue(poll->head, &poll->wait); | |
4871 | return true; | |
4872 | } | |
4873 | ||
4874 | return false; | |
4875 | } | |
4876 | ||
d4e7cd36 | 4877 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4878 | { |
e8c2bc1f | 4879 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4880 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4881 | return req->async_data; |
d4e7cd36 JA |
4882 | return req->apoll->double_poll; |
4883 | } | |
4884 | ||
4885 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4886 | { | |
4887 | if (req->opcode == IORING_OP_POLL_ADD) | |
4888 | return &req->poll; | |
4889 | return &req->apoll->poll; | |
4890 | } | |
4891 | ||
4892 | static void io_poll_remove_double(struct io_kiocb *req) | |
4893 | { | |
4894 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4895 | |
4896 | lockdep_assert_held(&req->ctx->completion_lock); | |
4897 | ||
4898 | if (poll && poll->head) { | |
4899 | struct wait_queue_head *head = poll->head; | |
4900 | ||
4901 | spin_lock(&head->lock); | |
4902 | list_del_init(&poll->wait.entry); | |
4903 | if (poll->wait.private) | |
de9b4cca | 4904 | req_ref_put(req); |
18bceab1 JA |
4905 | poll->head = NULL; |
4906 | spin_unlock(&head->lock); | |
4907 | } | |
4908 | } | |
4909 | ||
88e41cf9 | 4910 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask, int error) |
18bceab1 JA |
4911 | { |
4912 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4913 | unsigned flags = IORING_CQE_F_MORE; |
18bceab1 | 4914 | |
88e41cf9 | 4915 | if (!error && req->poll.canceled) { |
45ab03b1 | 4916 | error = -ECANCELED; |
88e41cf9 JA |
4917 | req->poll.events |= EPOLLONESHOT; |
4918 | } | |
5082620f JA |
4919 | if (!error) |
4920 | error = mangle_poll(mask); | |
b69de288 JA |
4921 | if (req->poll.events & EPOLLONESHOT) |
4922 | flags = 0; | |
4923 | if (!__io_cqring_fill_event(req, error, flags)) { | |
5082620f | 4924 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4925 | req->poll.done = true; |
4926 | flags = 0; | |
4927 | } | |
18bceab1 | 4928 | io_commit_cqring(ctx); |
88e41cf9 | 4929 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4930 | } |
4931 | ||
dd221f46 | 4932 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4933 | { |
dd221f46 | 4934 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4935 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4936 | struct io_kiocb *nxt; |
18bceab1 JA |
4937 | |
4938 | if (io_poll_rewait(req, &req->poll)) { | |
4939 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4940 | } else { |
88e41cf9 JA |
4941 | bool done, post_ev; |
4942 | ||
4943 | post_ev = done = io_poll_complete(req, req->result, 0); | |
4944 | if (done) { | |
4945 | hash_del(&req->hash_node); | |
4946 | } else if (!(req->poll.events & EPOLLONESHOT)) { | |
4947 | post_ev = true; | |
4948 | req->result = 0; | |
4949 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4950 | } | |
dd221f46 | 4951 | spin_unlock_irq(&ctx->completion_lock); |
18bceab1 | 4952 | |
88e41cf9 JA |
4953 | if (post_ev) |
4954 | io_cqring_ev_posted(ctx); | |
4955 | if (done) { | |
4956 | nxt = io_put_req_find_next(req); | |
4957 | if (nxt) | |
4958 | __io_req_task_submit(nxt); | |
4959 | } | |
dd221f46 | 4960 | } |
18bceab1 JA |
4961 | } |
4962 | ||
4963 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4964 | int sync, void *key) | |
4965 | { | |
4966 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4967 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4968 | __poll_t mask = key_to_poll(key); |
4969 | ||
4970 | /* for instances that support it check for an event match first: */ | |
4971 | if (mask && !(mask & poll->events)) | |
4972 | return 0; | |
88e41cf9 JA |
4973 | if (!(poll->events & EPOLLONESHOT)) |
4974 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4975 | |
8706e04e JA |
4976 | list_del_init(&wait->entry); |
4977 | ||
807abcb0 | 4978 | if (poll && poll->head) { |
18bceab1 JA |
4979 | bool done; |
4980 | ||
807abcb0 JA |
4981 | spin_lock(&poll->head->lock); |
4982 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4983 | if (!done) |
807abcb0 | 4984 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4985 | /* make sure double remove sees this as being gone */ |
4986 | wait->private = NULL; | |
807abcb0 | 4987 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4988 | if (!done) { |
4989 | /* use wait func handler, so it matches the rq type */ | |
4990 | poll->wait.func(&poll->wait, mode, sync, key); | |
4991 | } | |
18bceab1 | 4992 | } |
de9b4cca | 4993 | req_ref_put(req); |
18bceab1 JA |
4994 | return 1; |
4995 | } | |
4996 | ||
4997 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4998 | wait_queue_func_t wake_func) | |
4999 | { | |
5000 | poll->head = NULL; | |
5001 | poll->done = false; | |
5002 | poll->canceled = false; | |
b69de288 | 5003 | poll->update_events = poll->update_user_data = false; |
464dca61 JA |
5004 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
5005 | /* mask in events that we always want/need */ | |
5006 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
5007 | INIT_LIST_HEAD(&poll->wait.entry); |
5008 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
5009 | } | |
5010 | ||
5011 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
5012 | struct wait_queue_head *head, |
5013 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
5014 | { |
5015 | struct io_kiocb *req = pt->req; | |
5016 | ||
5017 | /* | |
5018 | * If poll->head is already set, it's because the file being polled | |
5019 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5020 | * for write). Setup a separate io_poll_iocb if this happens. | |
5021 | */ | |
5022 | if (unlikely(poll->head)) { | |
58852d4d PB |
5023 | struct io_poll_iocb *poll_one = poll; |
5024 | ||
18bceab1 | 5025 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5026 | if (*poll_ptr) { |
18bceab1 JA |
5027 | pt->error = -EINVAL; |
5028 | return; | |
5029 | } | |
1c3b3e65 JA |
5030 | /* double add on the same waitqueue head, ignore */ |
5031 | if (poll->head == head) | |
5032 | return; | |
18bceab1 JA |
5033 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5034 | if (!poll) { | |
5035 | pt->error = -ENOMEM; | |
5036 | return; | |
5037 | } | |
58852d4d | 5038 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5039 | req_ref_get(req); |
18bceab1 | 5040 | poll->wait.private = req; |
807abcb0 | 5041 | *poll_ptr = poll; |
18bceab1 JA |
5042 | } |
5043 | ||
5044 | pt->error = 0; | |
5045 | poll->head = head; | |
a31eb4a2 JX |
5046 | |
5047 | if (poll->events & EPOLLEXCLUSIVE) | |
5048 | add_wait_queue_exclusive(head, &poll->wait); | |
5049 | else | |
5050 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5051 | } |
5052 | ||
5053 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5054 | struct poll_table_struct *p) | |
5055 | { | |
5056 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5057 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5058 | |
807abcb0 | 5059 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5060 | } |
5061 | ||
d7718a9d JA |
5062 | static void io_async_task_func(struct callback_head *cb) |
5063 | { | |
5064 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5065 | struct async_poll *apoll = req->apoll; | |
5066 | struct io_ring_ctx *ctx = req->ctx; | |
5067 | ||
5068 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5069 | ||
74ce6ce4 | 5070 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5071 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5072 | return; |
d7718a9d JA |
5073 | } |
5074 | ||
31067255 | 5075 | /* If req is still hashed, it cannot have been canceled. Don't check. */ |
0be0b0e3 | 5076 | if (hash_hashed(&req->hash_node)) |
74ce6ce4 | 5077 | hash_del(&req->hash_node); |
2bae047e | 5078 | |
d4e7cd36 | 5079 | io_poll_remove_double(req); |
74ce6ce4 JA |
5080 | spin_unlock_irq(&ctx->completion_lock); |
5081 | ||
0be0b0e3 PB |
5082 | if (!READ_ONCE(apoll->poll.canceled)) |
5083 | __io_req_task_submit(req); | |
5084 | else | |
2593553a | 5085 | io_req_complete_failed(req, -ECANCELED); |
aa340845 | 5086 | |
807abcb0 | 5087 | kfree(apoll->double_poll); |
31067255 | 5088 | kfree(apoll); |
d7718a9d JA |
5089 | } |
5090 | ||
5091 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5092 | void *key) | |
5093 | { | |
5094 | struct io_kiocb *req = wait->private; | |
5095 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5096 | ||
5097 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5098 | key_to_poll(key)); | |
5099 | ||
5100 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5101 | } | |
5102 | ||
5103 | static void io_poll_req_insert(struct io_kiocb *req) | |
5104 | { | |
5105 | struct io_ring_ctx *ctx = req->ctx; | |
5106 | struct hlist_head *list; | |
5107 | ||
5108 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5109 | hlist_add_head(&req->hash_node, list); | |
5110 | } | |
5111 | ||
5112 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5113 | struct io_poll_iocb *poll, | |
5114 | struct io_poll_table *ipt, __poll_t mask, | |
5115 | wait_queue_func_t wake_func) | |
5116 | __acquires(&ctx->completion_lock) | |
5117 | { | |
5118 | struct io_ring_ctx *ctx = req->ctx; | |
5119 | bool cancel = false; | |
5120 | ||
4d52f338 | 5121 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5122 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5123 | poll->file = req->file; |
18bceab1 | 5124 | poll->wait.private = req; |
d7718a9d JA |
5125 | |
5126 | ipt->pt._key = mask; | |
5127 | ipt->req = req; | |
5128 | ipt->error = -EINVAL; | |
5129 | ||
d7718a9d JA |
5130 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5131 | ||
5132 | spin_lock_irq(&ctx->completion_lock); | |
5133 | if (likely(poll->head)) { | |
5134 | spin_lock(&poll->head->lock); | |
5135 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5136 | if (ipt->error) | |
5137 | cancel = true; | |
5138 | ipt->error = 0; | |
5139 | mask = 0; | |
5140 | } | |
88e41cf9 | 5141 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5142 | list_del_init(&poll->wait.entry); |
5143 | else if (cancel) | |
5144 | WRITE_ONCE(poll->canceled, true); | |
5145 | else if (!poll->done) /* actually waiting for an event */ | |
5146 | io_poll_req_insert(req); | |
5147 | spin_unlock(&poll->head->lock); | |
5148 | } | |
5149 | ||
5150 | return mask; | |
5151 | } | |
5152 | ||
5153 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5154 | { | |
5155 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5156 | struct io_ring_ctx *ctx = req->ctx; | |
5157 | struct async_poll *apoll; | |
5158 | struct io_poll_table ipt; | |
5159 | __poll_t mask, ret; | |
9dab14b8 | 5160 | int rw; |
d7718a9d JA |
5161 | |
5162 | if (!req->file || !file_can_poll(req->file)) | |
5163 | return false; | |
24c74678 | 5164 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5165 | return false; |
9dab14b8 JA |
5166 | if (def->pollin) |
5167 | rw = READ; | |
5168 | else if (def->pollout) | |
5169 | rw = WRITE; | |
5170 | else | |
5171 | return false; | |
5172 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5173 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5174 | return false; |
5175 | ||
5176 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5177 | if (unlikely(!apoll)) | |
5178 | return false; | |
807abcb0 | 5179 | apoll->double_poll = NULL; |
d7718a9d JA |
5180 | |
5181 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5182 | req->apoll = apoll; |
d7718a9d | 5183 | |
88e41cf9 | 5184 | mask = EPOLLONESHOT; |
d7718a9d | 5185 | if (def->pollin) |
8755d97a | 5186 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5187 | if (def->pollout) |
5188 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5189 | |
5190 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5191 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5192 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5193 | mask &= ~POLLIN; | |
5194 | ||
d7718a9d JA |
5195 | mask |= POLLERR | POLLPRI; |
5196 | ||
5197 | ipt.pt._qproc = io_async_queue_proc; | |
5198 | ||
5199 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5200 | io_async_wake); | |
a36da65c | 5201 | if (ret || ipt.error) { |
d4e7cd36 | 5202 | io_poll_remove_double(req); |
d7718a9d | 5203 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5204 | kfree(apoll->double_poll); |
d7718a9d JA |
5205 | kfree(apoll); |
5206 | return false; | |
5207 | } | |
5208 | spin_unlock_irq(&ctx->completion_lock); | |
5209 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5210 | apoll->poll.events); | |
5211 | return true; | |
5212 | } | |
5213 | ||
5214 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5215 | struct io_poll_iocb *poll, bool do_cancel) |
221c5eb2 | 5216 | { |
b41e9852 | 5217 | bool do_complete = false; |
221c5eb2 | 5218 | |
5082620f JA |
5219 | if (!poll->head) |
5220 | return false; | |
221c5eb2 | 5221 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5222 | if (do_cancel) |
5223 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5224 | if (!list_empty(&poll->wait.entry)) { |
5225 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5226 | do_complete = true; |
221c5eb2 JA |
5227 | } |
5228 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5229 | hash_del(&req->hash_node); |
d7718a9d JA |
5230 | return do_complete; |
5231 | } | |
5232 | ||
b2c3f7e1 | 5233 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
d7718a9d JA |
5234 | { |
5235 | bool do_complete; | |
5236 | ||
d4e7cd36 JA |
5237 | io_poll_remove_double(req); |
5238 | ||
d7718a9d | 5239 | if (req->opcode == IORING_OP_POLL_ADD) { |
b2e720ac | 5240 | do_complete = __io_poll_remove_one(req, &req->poll, true); |
d7718a9d | 5241 | } else { |
3bfa5bcb JA |
5242 | struct async_poll *apoll = req->apoll; |
5243 | ||
d7718a9d | 5244 | /* non-poll requests have submit ref still */ |
b2e720ac | 5245 | do_complete = __io_poll_remove_one(req, &apoll->poll, true); |
3bfa5bcb | 5246 | if (do_complete) { |
d7718a9d | 5247 | io_put_req(req); |
807abcb0 | 5248 | kfree(apoll->double_poll); |
3bfa5bcb JA |
5249 | kfree(apoll); |
5250 | } | |
b1f573bd XW |
5251 | } |
5252 | ||
b2c3f7e1 JA |
5253 | return do_complete; |
5254 | } | |
5255 | ||
5256 | static bool io_poll_remove_one(struct io_kiocb *req) | |
5257 | { | |
5258 | bool do_complete; | |
5259 | ||
5260 | do_complete = io_poll_remove_waitqs(req); | |
b41e9852 JA |
5261 | if (do_complete) { |
5262 | io_cqring_fill_event(req, -ECANCELED); | |
5263 | io_commit_cqring(req->ctx); | |
f254ac04 | 5264 | req_set_fail_links(req); |
216578e5 | 5265 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5266 | } |
5267 | ||
5268 | return do_complete; | |
221c5eb2 JA |
5269 | } |
5270 | ||
76e1b642 JA |
5271 | /* |
5272 | * Returns true if we found and killed one or more poll requests | |
5273 | */ | |
6b81928d PB |
5274 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5275 | struct files_struct *files) | |
221c5eb2 | 5276 | { |
78076bb6 | 5277 | struct hlist_node *tmp; |
221c5eb2 | 5278 | struct io_kiocb *req; |
8e2e1faf | 5279 | int posted = 0, i; |
221c5eb2 JA |
5280 | |
5281 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5282 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5283 | struct hlist_head *list; | |
5284 | ||
5285 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5286 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5287 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5288 | posted += io_poll_remove_one(req); |
5289 | } | |
221c5eb2 JA |
5290 | } |
5291 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5292 | |
8e2e1faf JA |
5293 | if (posted) |
5294 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5295 | |
5296 | return posted != 0; | |
221c5eb2 JA |
5297 | } |
5298 | ||
b2cb805f | 5299 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr) |
47f46768 | 5300 | { |
78076bb6 | 5301 | struct hlist_head *list; |
47f46768 JA |
5302 | struct io_kiocb *req; |
5303 | ||
78076bb6 JA |
5304 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5305 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5306 | if (sqe_addr != req->user_data) |
5307 | continue; | |
b2cb805f | 5308 | return req; |
47f46768 JA |
5309 | } |
5310 | ||
b2cb805f JA |
5311 | return NULL; |
5312 | } | |
5313 | ||
5314 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) | |
5315 | { | |
5316 | struct io_kiocb *req; | |
5317 | ||
5318 | req = io_poll_find(ctx, sqe_addr); | |
5319 | if (!req) | |
5320 | return -ENOENT; | |
5321 | if (io_poll_remove_one(req)) | |
5322 | return 0; | |
5323 | ||
5324 | return -EALREADY; | |
47f46768 JA |
5325 | } |
5326 | ||
3529d8c2 JA |
5327 | static int io_poll_remove_prep(struct io_kiocb *req, |
5328 | const struct io_uring_sqe *sqe) | |
0969e783 | 5329 | { |
0969e783 JA |
5330 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5331 | return -EINVAL; | |
5332 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5333 | sqe->poll_events) | |
5334 | return -EINVAL; | |
5335 | ||
018043be | 5336 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5337 | return 0; |
5338 | } | |
5339 | ||
221c5eb2 JA |
5340 | /* |
5341 | * Find a running poll command that matches one specified in sqe->addr, | |
5342 | * and remove it if found. | |
5343 | */ | |
61e98203 | 5344 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5345 | { |
5346 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5347 | int ret; |
221c5eb2 | 5348 | |
221c5eb2 | 5349 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5350 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5351 | spin_unlock_irq(&ctx->completion_lock); |
5352 | ||
4e88d6e7 JA |
5353 | if (ret < 0) |
5354 | req_set_fail_links(req); | |
e1e16097 | 5355 | io_req_complete(req, ret); |
221c5eb2 JA |
5356 | return 0; |
5357 | } | |
5358 | ||
221c5eb2 JA |
5359 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5360 | void *key) | |
5361 | { | |
c2f2eb7d JA |
5362 | struct io_kiocb *req = wait->private; |
5363 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5364 | |
d7718a9d | 5365 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5366 | } |
5367 | ||
221c5eb2 JA |
5368 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5369 | struct poll_table_struct *p) | |
5370 | { | |
5371 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5372 | ||
e8c2bc1f | 5373 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5374 | } |
5375 | ||
3529d8c2 | 5376 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 JA |
5377 | { |
5378 | struct io_poll_iocb *poll = &req->poll; | |
88e41cf9 | 5379 | u32 events, flags; |
221c5eb2 JA |
5380 | |
5381 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5382 | return -EINVAL; | |
b69de288 | 5383 | if (sqe->ioprio || sqe->buf_index) |
88e41cf9 JA |
5384 | return -EINVAL; |
5385 | flags = READ_ONCE(sqe->len); | |
b69de288 JA |
5386 | if (flags & ~(IORING_POLL_ADD_MULTI | IORING_POLL_UPDATE_EVENTS | |
5387 | IORING_POLL_UPDATE_USER_DATA)) | |
221c5eb2 | 5388 | return -EINVAL; |
5769a351 JX |
5389 | events = READ_ONCE(sqe->poll32_events); |
5390 | #ifdef __BIG_ENDIAN | |
5391 | events = swahw32(events); | |
5392 | #endif | |
b69de288 | 5393 | if (!(flags & IORING_POLL_ADD_MULTI)) |
88e41cf9 | 5394 | events |= EPOLLONESHOT; |
b69de288 JA |
5395 | poll->update_events = poll->update_user_data = false; |
5396 | if (flags & IORING_POLL_UPDATE_EVENTS) { | |
5397 | poll->update_events = true; | |
5398 | poll->old_user_data = READ_ONCE(sqe->addr); | |
5399 | } | |
5400 | if (flags & IORING_POLL_UPDATE_USER_DATA) { | |
5401 | poll->update_user_data = true; | |
5402 | poll->new_user_data = READ_ONCE(sqe->off); | |
5403 | } | |
5404 | if (!(poll->update_events || poll->update_user_data) && | |
5405 | (sqe->off || sqe->addr)) | |
5406 | return -EINVAL; | |
88e41cf9 JA |
5407 | poll->events = demangle_poll(events) | |
5408 | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
0969e783 JA |
5409 | return 0; |
5410 | } | |
5411 | ||
b69de288 | 5412 | static int __io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5413 | { |
5414 | struct io_poll_iocb *poll = &req->poll; | |
5415 | struct io_ring_ctx *ctx = req->ctx; | |
5416 | struct io_poll_table ipt; | |
0969e783 | 5417 | __poll_t mask; |
0969e783 | 5418 | |
d7718a9d | 5419 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5420 | |
d7718a9d JA |
5421 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5422 | io_poll_wake); | |
221c5eb2 | 5423 | |
8c838788 | 5424 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5425 | ipt.error = 0; |
b0dd8a41 | 5426 | io_poll_complete(req, mask, 0); |
221c5eb2 | 5427 | } |
221c5eb2 JA |
5428 | spin_unlock_irq(&ctx->completion_lock); |
5429 | ||
8c838788 JA |
5430 | if (mask) { |
5431 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5432 | if (poll->events & EPOLLONESHOT) |
5433 | io_put_req(req); | |
221c5eb2 | 5434 | } |
8c838788 | 5435 | return ipt.error; |
221c5eb2 JA |
5436 | } |
5437 | ||
b69de288 JA |
5438 | static int io_poll_update(struct io_kiocb *req) |
5439 | { | |
5440 | struct io_ring_ctx *ctx = req->ctx; | |
5441 | struct io_kiocb *preq; | |
5442 | int ret; | |
5443 | ||
5444 | spin_lock_irq(&ctx->completion_lock); | |
5445 | preq = io_poll_find(ctx, req->poll.old_user_data); | |
5446 | if (!preq) { | |
5447 | ret = -ENOENT; | |
5448 | goto err; | |
5449 | } else if (preq->opcode != IORING_OP_POLL_ADD) { | |
5450 | /* don't allow internal poll updates */ | |
5451 | ret = -EACCES; | |
5452 | goto err; | |
5453 | } | |
b2e720ac JA |
5454 | if (!__io_poll_remove_one(preq, &preq->poll, false)) { |
5455 | if (preq->poll.events & EPOLLONESHOT) { | |
5456 | ret = -EALREADY; | |
5457 | goto err; | |
5458 | } | |
b69de288 JA |
5459 | } |
5460 | /* we now have a detached poll request. reissue. */ | |
5461 | ret = 0; | |
5462 | err: | |
5463 | spin_unlock_irq(&ctx->completion_lock); | |
5464 | if (ret < 0) { | |
5465 | req_set_fail_links(req); | |
5466 | io_req_complete(req, ret); | |
5467 | return 0; | |
5468 | } | |
5469 | /* only mask one event flags, keep behavior flags */ | |
5470 | if (req->poll.update_events) { | |
5471 | preq->poll.events &= ~0xffff; | |
5472 | preq->poll.events |= req->poll.events & 0xffff; | |
5473 | preq->poll.events |= IO_POLL_UNMASK; | |
5474 | } | |
5475 | if (req->poll.update_user_data) | |
5476 | preq->user_data = req->poll.new_user_data; | |
5477 | ||
5478 | /* complete update request, we're done with it */ | |
5479 | io_req_complete(req, ret); | |
5480 | ||
5481 | ret = __io_poll_add(preq); | |
5482 | if (ret < 0) { | |
5483 | req_set_fail_links(preq); | |
5484 | io_req_complete(preq, ret); | |
5485 | } | |
5486 | return 0; | |
5487 | } | |
5488 | ||
5489 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) | |
5490 | { | |
5491 | if (!req->poll.update_events && !req->poll.update_user_data) | |
5492 | return __io_poll_add(req); | |
5493 | return io_poll_update(req); | |
5494 | } | |
5495 | ||
5262f567 JA |
5496 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5497 | { | |
ad8a48ac JA |
5498 | struct io_timeout_data *data = container_of(timer, |
5499 | struct io_timeout_data, timer); | |
5500 | struct io_kiocb *req = data->req; | |
5501 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5502 | unsigned long flags; |
5503 | ||
5262f567 | 5504 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5505 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5506 | atomic_set(&req->ctx->cq_timeouts, |
5507 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5508 | ||
78e19bbe | 5509 | io_cqring_fill_event(req, -ETIME); |
5262f567 JA |
5510 | io_commit_cqring(ctx); |
5511 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5512 | ||
5513 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5514 | req_set_fail_links(req); |
5262f567 JA |
5515 | io_put_req(req); |
5516 | return HRTIMER_NORESTART; | |
5517 | } | |
5518 | ||
fbd15848 PB |
5519 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5520 | __u64 user_data) | |
f254ac04 | 5521 | { |
fbd15848 | 5522 | struct io_timeout_data *io; |
47f46768 JA |
5523 | struct io_kiocb *req; |
5524 | int ret = -ENOENT; | |
f254ac04 | 5525 | |
135fcde8 | 5526 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
47f46768 | 5527 | if (user_data == req->user_data) { |
47f46768 JA |
5528 | ret = 0; |
5529 | break; | |
5530 | } | |
5531 | } | |
5532 | ||
5533 | if (ret == -ENOENT) | |
fbd15848 PB |
5534 | return ERR_PTR(ret); |
5535 | ||
5536 | io = req->async_data; | |
e8c2bc1f | 5537 | ret = hrtimer_try_to_cancel(&io->timer); |
f254ac04 | 5538 | if (ret == -1) |
fbd15848 | 5539 | return ERR_PTR(-EALREADY); |
a71976f3 | 5540 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5541 | return req; |
5542 | } | |
47f46768 | 5543 | |
fbd15848 PB |
5544 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
5545 | { | |
5546 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5547 | ||
5548 | if (IS_ERR(req)) | |
5549 | return PTR_ERR(req); | |
f254ac04 JA |
5550 | |
5551 | req_set_fail_links(req); | |
f254ac04 | 5552 | io_cqring_fill_event(req, -ECANCELED); |
216578e5 | 5553 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5554 | return 0; |
5555 | } | |
5556 | ||
9c8e11b3 PB |
5557 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5558 | struct timespec64 *ts, enum hrtimer_mode mode) | |
47f46768 | 5559 | { |
9c8e11b3 PB |
5560 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5561 | struct io_timeout_data *data; | |
47f46768 | 5562 | |
9c8e11b3 PB |
5563 | if (IS_ERR(req)) |
5564 | return PTR_ERR(req); | |
47f46768 | 5565 | |
9c8e11b3 PB |
5566 | req->timeout.off = 0; /* noseq */ |
5567 | data = req->async_data; | |
5568 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5569 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5570 | data->timer.function = io_timeout_fn; | |
5571 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5572 | return 0; | |
47f46768 JA |
5573 | } |
5574 | ||
3529d8c2 JA |
5575 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5576 | const struct io_uring_sqe *sqe) | |
b29472ee | 5577 | { |
9c8e11b3 PB |
5578 | struct io_timeout_rem *tr = &req->timeout_rem; |
5579 | ||
b29472ee JA |
5580 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5581 | return -EINVAL; | |
61710e43 DA |
5582 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5583 | return -EINVAL; | |
9c8e11b3 | 5584 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5585 | return -EINVAL; |
5586 | ||
9c8e11b3 PB |
5587 | tr->addr = READ_ONCE(sqe->addr); |
5588 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5589 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5590 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5591 | return -EINVAL; | |
5592 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5593 | return -EFAULT; | |
5594 | } else if (tr->flags) { | |
5595 | /* timeout removal doesn't support flags */ | |
b29472ee | 5596 | return -EINVAL; |
9c8e11b3 | 5597 | } |
b29472ee | 5598 | |
b29472ee JA |
5599 | return 0; |
5600 | } | |
5601 | ||
8662daec PB |
5602 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5603 | { | |
5604 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5605 | : HRTIMER_MODE_REL; | |
5606 | } | |
5607 | ||
11365043 JA |
5608 | /* |
5609 | * Remove or update an existing timeout command | |
5610 | */ | |
61e98203 | 5611 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5612 | { |
9c8e11b3 | 5613 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5614 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5615 | int ret; |
11365043 | 5616 | |
11365043 | 5617 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5618 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5619 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5620 | else |
5621 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5622 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5623 | |
47f46768 | 5624 | io_cqring_fill_event(req, ret); |
11365043 JA |
5625 | io_commit_cqring(ctx); |
5626 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5627 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5628 | if (ret < 0) |
5629 | req_set_fail_links(req); | |
ec9c02ad | 5630 | io_put_req(req); |
11365043 | 5631 | return 0; |
5262f567 JA |
5632 | } |
5633 | ||
3529d8c2 | 5634 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5635 | bool is_timeout_link) |
5262f567 | 5636 | { |
ad8a48ac | 5637 | struct io_timeout_data *data; |
a41525ab | 5638 | unsigned flags; |
56080b02 | 5639 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5640 | |
ad8a48ac | 5641 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5642 | return -EINVAL; |
ad8a48ac | 5643 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5644 | return -EINVAL; |
56080b02 | 5645 | if (off && is_timeout_link) |
2d28390a | 5646 | return -EINVAL; |
a41525ab JA |
5647 | flags = READ_ONCE(sqe->timeout_flags); |
5648 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5649 | return -EINVAL; |
bdf20073 | 5650 | |
bfe68a22 | 5651 | req->timeout.off = off; |
26a61679 | 5652 | |
e8c2bc1f | 5653 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5654 | return -ENOMEM; |
5655 | ||
e8c2bc1f | 5656 | data = req->async_data; |
ad8a48ac | 5657 | data->req = req; |
ad8a48ac JA |
5658 | |
5659 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5660 | return -EFAULT; |
5661 | ||
8662daec | 5662 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5663 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5664 | if (is_timeout_link) |
5665 | io_req_track_inflight(req); | |
ad8a48ac JA |
5666 | return 0; |
5667 | } | |
5668 | ||
61e98203 | 5669 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5670 | { |
ad8a48ac | 5671 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5672 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5673 | struct list_head *entry; |
bfe68a22 | 5674 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5675 | |
733f5c95 | 5676 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5677 | |
5262f567 JA |
5678 | /* |
5679 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5680 | * timeout event to be satisfied. If it isn't set, then this is |
5681 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5682 | */ |
8eb7e2d0 | 5683 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5684 | entry = ctx->timeout_list.prev; |
5685 | goto add; | |
5686 | } | |
5262f567 | 5687 | |
bfe68a22 PB |
5688 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5689 | req->timeout.target_seq = tail + off; | |
5262f567 | 5690 | |
f010505b MDG |
5691 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5692 | * This is safe because ->completion_lock is held, and submissions | |
5693 | * and completions are never mixed in the same ->completion_lock section. | |
5694 | */ | |
5695 | ctx->cq_last_tm_flush = tail; | |
5696 | ||
5262f567 JA |
5697 | /* |
5698 | * Insertion sort, ensuring the first entry in the list is always | |
5699 | * the one we need first. | |
5700 | */ | |
5262f567 | 5701 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5702 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5703 | timeout.list); | |
5262f567 | 5704 | |
8eb7e2d0 | 5705 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5706 | continue; |
bfe68a22 PB |
5707 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5708 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5709 | break; |
5710 | } | |
93bd25bb | 5711 | add: |
135fcde8 | 5712 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5713 | data->timer.function = io_timeout_fn; |
5714 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5715 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5716 | return 0; |
5717 | } | |
5262f567 | 5718 | |
f458dd84 PB |
5719 | struct io_cancel_data { |
5720 | struct io_ring_ctx *ctx; | |
5721 | u64 user_data; | |
5722 | }; | |
5723 | ||
62755e35 JA |
5724 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5725 | { | |
5726 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5727 | struct io_cancel_data *cd = data; |
62755e35 | 5728 | |
f458dd84 | 5729 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5730 | } |
5731 | ||
f458dd84 PB |
5732 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5733 | struct io_ring_ctx *ctx) | |
62755e35 | 5734 | { |
f458dd84 | 5735 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5736 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5737 | int ret = 0; |
5738 | ||
f458dd84 | 5739 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5740 | return -ENOENT; |
5741 | ||
f458dd84 | 5742 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5743 | switch (cancel_ret) { |
5744 | case IO_WQ_CANCEL_OK: | |
5745 | ret = 0; | |
5746 | break; | |
5747 | case IO_WQ_CANCEL_RUNNING: | |
5748 | ret = -EALREADY; | |
5749 | break; | |
5750 | case IO_WQ_CANCEL_NOTFOUND: | |
5751 | ret = -ENOENT; | |
5752 | break; | |
5753 | } | |
5754 | ||
e977d6d3 JA |
5755 | return ret; |
5756 | } | |
5757 | ||
47f46768 JA |
5758 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5759 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5760 | int success_ret) |
47f46768 JA |
5761 | { |
5762 | unsigned long flags; | |
5763 | int ret; | |
5764 | ||
f458dd84 | 5765 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 JA |
5766 | if (ret != -ENOENT) { |
5767 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5768 | goto done; | |
5769 | } | |
5770 | ||
5771 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
5772 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5773 | if (ret != -ENOENT) | |
5774 | goto done; | |
5775 | ret = io_poll_cancel(ctx, sqe_addr); | |
5776 | done: | |
b0dd8a41 JA |
5777 | if (!ret) |
5778 | ret = success_ret; | |
47f46768 JA |
5779 | io_cqring_fill_event(req, ret); |
5780 | io_commit_cqring(ctx); | |
5781 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5782 | io_cqring_ev_posted(ctx); | |
5783 | ||
4e88d6e7 JA |
5784 | if (ret < 0) |
5785 | req_set_fail_links(req); | |
014db007 | 5786 | io_put_req(req); |
47f46768 JA |
5787 | } |
5788 | ||
3529d8c2 JA |
5789 | static int io_async_cancel_prep(struct io_kiocb *req, |
5790 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5791 | { |
fbf23849 | 5792 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5793 | return -EINVAL; |
61710e43 DA |
5794 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5795 | return -EINVAL; | |
5796 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5797 | return -EINVAL; |
5798 | ||
fbf23849 JA |
5799 | req->cancel.addr = READ_ONCE(sqe->addr); |
5800 | return 0; | |
5801 | } | |
5802 | ||
61e98203 | 5803 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5804 | { |
5805 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5806 | u64 sqe_addr = req->cancel.addr; |
5807 | struct io_tctx_node *node; | |
5808 | int ret; | |
5809 | ||
5810 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5811 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5812 | spin_lock_irq(&ctx->completion_lock); | |
5813 | if (ret != -ENOENT) | |
5814 | goto done; | |
5815 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5816 | if (ret != -ENOENT) | |
5817 | goto done; | |
5818 | ret = io_poll_cancel(ctx, sqe_addr); | |
5819 | if (ret != -ENOENT) | |
5820 | goto done; | |
5821 | spin_unlock_irq(&ctx->completion_lock); | |
5822 | ||
5823 | /* slow path, try all io-wq's */ | |
5824 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5825 | ret = -ENOENT; | |
5826 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5827 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5828 | |
58f99373 PB |
5829 | if (!tctx || !tctx->io_wq) |
5830 | continue; | |
5831 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); | |
5832 | if (ret != -ENOENT) | |
5833 | break; | |
5834 | } | |
5835 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5836 | ||
5837 | spin_lock_irq(&ctx->completion_lock); | |
5838 | done: | |
5839 | io_cqring_fill_event(req, ret); | |
5840 | io_commit_cqring(ctx); | |
5841 | spin_unlock_irq(&ctx->completion_lock); | |
5842 | io_cqring_ev_posted(ctx); | |
5843 | ||
5844 | if (ret < 0) | |
5845 | req_set_fail_links(req); | |
5846 | io_put_req(req); | |
5262f567 JA |
5847 | return 0; |
5848 | } | |
5849 | ||
269bbe5f | 5850 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5851 | const struct io_uring_sqe *sqe) |
5852 | { | |
6ca56f84 JA |
5853 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5854 | return -EINVAL; | |
61710e43 DA |
5855 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5856 | return -EINVAL; | |
5857 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5858 | return -EINVAL; |
5859 | ||
269bbe5f BM |
5860 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5861 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5862 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5863 | return -EINVAL; |
269bbe5f | 5864 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5865 | return 0; |
5866 | } | |
5867 | ||
889fca73 | 5868 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5869 | { |
5870 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5871 | struct io_uring_rsrc_update up; |
05f3fb3c | 5872 | int ret; |
fbf23849 | 5873 | |
45d189c6 | 5874 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5875 | return -EAGAIN; |
05f3fb3c | 5876 | |
269bbe5f BM |
5877 | up.offset = req->rsrc_update.offset; |
5878 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5879 | |
5880 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5881 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5882 | mutex_unlock(&ctx->uring_lock); |
5883 | ||
5884 | if (ret < 0) | |
5885 | req_set_fail_links(req); | |
889fca73 | 5886 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5887 | return 0; |
5888 | } | |
5889 | ||
bfe76559 | 5890 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5891 | { |
d625c6ee | 5892 | switch (req->opcode) { |
e781573e | 5893 | case IORING_OP_NOP: |
bfe76559 | 5894 | return 0; |
f67676d1 JA |
5895 | case IORING_OP_READV: |
5896 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5897 | case IORING_OP_READ: |
bfe76559 | 5898 | return io_read_prep(req, sqe); |
f67676d1 JA |
5899 | case IORING_OP_WRITEV: |
5900 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5901 | case IORING_OP_WRITE: |
bfe76559 | 5902 | return io_write_prep(req, sqe); |
0969e783 | 5903 | case IORING_OP_POLL_ADD: |
bfe76559 | 5904 | return io_poll_add_prep(req, sqe); |
0969e783 | 5905 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5906 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5907 | case IORING_OP_FSYNC: |
1155c76a | 5908 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5909 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5910 | return io_sfr_prep(req, sqe); |
03b1230c | 5911 | case IORING_OP_SENDMSG: |
fddaface | 5912 | case IORING_OP_SEND: |
bfe76559 | 5913 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5914 | case IORING_OP_RECVMSG: |
fddaface | 5915 | case IORING_OP_RECV: |
bfe76559 | 5916 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5917 | case IORING_OP_CONNECT: |
bfe76559 | 5918 | return io_connect_prep(req, sqe); |
2d28390a | 5919 | case IORING_OP_TIMEOUT: |
bfe76559 | 5920 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5921 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5922 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5923 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5924 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5925 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5926 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5927 | case IORING_OP_ACCEPT: |
bfe76559 | 5928 | return io_accept_prep(req, sqe); |
d63d1b5e | 5929 | case IORING_OP_FALLOCATE: |
bfe76559 | 5930 | return io_fallocate_prep(req, sqe); |
15b71abe | 5931 | case IORING_OP_OPENAT: |
bfe76559 | 5932 | return io_openat_prep(req, sqe); |
b5dba59e | 5933 | case IORING_OP_CLOSE: |
bfe76559 | 5934 | return io_close_prep(req, sqe); |
05f3fb3c | 5935 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5936 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5937 | case IORING_OP_STATX: |
bfe76559 | 5938 | return io_statx_prep(req, sqe); |
4840e418 | 5939 | case IORING_OP_FADVISE: |
bfe76559 | 5940 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5941 | case IORING_OP_MADVISE: |
bfe76559 | 5942 | return io_madvise_prep(req, sqe); |
cebdb986 | 5943 | case IORING_OP_OPENAT2: |
bfe76559 | 5944 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5945 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5946 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5947 | case IORING_OP_SPLICE: |
bfe76559 | 5948 | return io_splice_prep(req, sqe); |
ddf0322d | 5949 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5950 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5951 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5952 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5953 | case IORING_OP_TEE: |
bfe76559 | 5954 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5955 | case IORING_OP_SHUTDOWN: |
5956 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5957 | case IORING_OP_RENAMEAT: |
5958 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5959 | case IORING_OP_UNLINKAT: |
5960 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5961 | } |
5962 | ||
bfe76559 PB |
5963 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5964 | req->opcode); | |
5965 | return-EINVAL; | |
5966 | } | |
5967 | ||
93642ef8 | 5968 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5969 | { |
b7e298d2 PB |
5970 | if (!io_op_defs[req->opcode].needs_async_setup) |
5971 | return 0; | |
5972 | if (WARN_ON_ONCE(req->async_data)) | |
5973 | return -EFAULT; | |
5974 | if (io_alloc_async_data(req)) | |
5975 | return -EAGAIN; | |
5976 | ||
93642ef8 PB |
5977 | switch (req->opcode) { |
5978 | case IORING_OP_READV: | |
93642ef8 PB |
5979 | return io_rw_prep_async(req, READ); |
5980 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5981 | return io_rw_prep_async(req, WRITE); |
5982 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5983 | return io_sendmsg_prep_async(req); |
5984 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5985 | return io_recvmsg_prep_async(req); |
5986 | case IORING_OP_CONNECT: | |
5987 | return io_connect_prep_async(req); | |
5988 | } | |
b7e298d2 PB |
5989 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5990 | req->opcode); | |
5991 | return -EFAULT; | |
f67676d1 JA |
5992 | } |
5993 | ||
9cf7c104 PB |
5994 | static u32 io_get_sequence(struct io_kiocb *req) |
5995 | { | |
5996 | struct io_kiocb *pos; | |
5997 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5998 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5999 | |
f2f87370 PB |
6000 | io_for_each_link(pos, req) |
6001 | nr_reqs++; | |
9cf7c104 PB |
6002 | |
6003 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
6004 | return total_submitted - nr_reqs; | |
6005 | } | |
6006 | ||
be7053b7 | 6007 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 6008 | { |
a197f664 | 6009 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6010 | struct io_defer_entry *de; |
f67676d1 | 6011 | int ret; |
9cf7c104 | 6012 | u32 seq; |
de0617e4 | 6013 | |
9d858b21 | 6014 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6015 | if (likely(list_empty_careful(&ctx->defer_list) && |
6016 | !(req->flags & REQ_F_IO_DRAIN))) | |
6017 | return 0; | |
6018 | ||
6019 | seq = io_get_sequence(req); | |
6020 | /* Still a chance to pass the sequence check */ | |
6021 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6022 | return 0; |
6023 | ||
b7e298d2 | 6024 | ret = io_req_prep_async(req); |
be7053b7 PB |
6025 | if (ret) |
6026 | return ret; | |
cbdcb435 | 6027 | io_prep_async_link(req); |
27dc8338 PB |
6028 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6029 | if (!de) | |
6030 | return -ENOMEM; | |
2d28390a | 6031 | |
de0617e4 | 6032 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6033 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6034 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6035 | kfree(de); |
ae34817b PB |
6036 | io_queue_async_work(req); |
6037 | return -EIOCBQUEUED; | |
de0617e4 JA |
6038 | } |
6039 | ||
915967f6 | 6040 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6041 | de->req = req; |
9cf7c104 | 6042 | de->seq = seq; |
27dc8338 | 6043 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6044 | spin_unlock_irq(&ctx->completion_lock); |
6045 | return -EIOCBQUEUED; | |
6046 | } | |
6047 | ||
68fb8979 | 6048 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6049 | { |
0e1b6fe3 PB |
6050 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6051 | switch (req->opcode) { | |
6052 | case IORING_OP_READV: | |
6053 | case IORING_OP_READ_FIXED: | |
6054 | case IORING_OP_READ: | |
bcda7baa | 6055 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6056 | break; |
6057 | case IORING_OP_RECVMSG: | |
6058 | case IORING_OP_RECV: | |
bcda7baa | 6059 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6060 | break; |
6061 | } | |
6062 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6063 | } |
6064 | ||
0e1b6fe3 PB |
6065 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6066 | switch (req->opcode) { | |
6067 | case IORING_OP_READV: | |
6068 | case IORING_OP_READ_FIXED: | |
6069 | case IORING_OP_READ: | |
6070 | case IORING_OP_WRITEV: | |
6071 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6072 | case IORING_OP_WRITE: { |
6073 | struct io_async_rw *io = req->async_data; | |
6074 | if (io->free_iovec) | |
6075 | kfree(io->free_iovec); | |
0e1b6fe3 | 6076 | break; |
e8c2bc1f | 6077 | } |
0e1b6fe3 | 6078 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6079 | case IORING_OP_SENDMSG: { |
6080 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6081 | |
6082 | kfree(io->free_iov); | |
0e1b6fe3 | 6083 | break; |
e8c2bc1f | 6084 | } |
0e1b6fe3 PB |
6085 | case IORING_OP_SPLICE: |
6086 | case IORING_OP_TEE: | |
e1d767f0 PB |
6087 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6088 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6089 | break; |
f3cd4850 JA |
6090 | case IORING_OP_OPENAT: |
6091 | case IORING_OP_OPENAT2: | |
6092 | if (req->open.filename) | |
6093 | putname(req->open.filename); | |
6094 | break; | |
80a261fd JA |
6095 | case IORING_OP_RENAMEAT: |
6096 | putname(req->rename.oldpath); | |
6097 | putname(req->rename.newpath); | |
6098 | break; | |
14a1143b JA |
6099 | case IORING_OP_UNLINKAT: |
6100 | putname(req->unlink.filename); | |
6101 | break; | |
0e1b6fe3 PB |
6102 | } |
6103 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6104 | } |
99bc4c38 PB |
6105 | } |
6106 | ||
889fca73 | 6107 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6108 | { |
a197f664 | 6109 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6110 | const struct cred *creds = NULL; |
d625c6ee | 6111 | int ret; |
2b188cc1 | 6112 | |
003e8dcc JA |
6113 | if (req->work.creds && req->work.creds != current_cred()) |
6114 | creds = override_creds(req->work.creds); | |
5730b27e | 6115 | |
d625c6ee | 6116 | switch (req->opcode) { |
2b188cc1 | 6117 | case IORING_OP_NOP: |
889fca73 | 6118 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6119 | break; |
6120 | case IORING_OP_READV: | |
edafccee | 6121 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6122 | case IORING_OP_READ: |
889fca73 | 6123 | ret = io_read(req, issue_flags); |
edafccee | 6124 | break; |
3529d8c2 | 6125 | case IORING_OP_WRITEV: |
edafccee | 6126 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6127 | case IORING_OP_WRITE: |
889fca73 | 6128 | ret = io_write(req, issue_flags); |
2b188cc1 | 6129 | break; |
c992fe29 | 6130 | case IORING_OP_FSYNC: |
45d189c6 | 6131 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6132 | break; |
221c5eb2 | 6133 | case IORING_OP_POLL_ADD: |
61e98203 | 6134 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6135 | break; |
6136 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6137 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6138 | break; |
5d17b4a4 | 6139 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6140 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6141 | break; |
0fa03c62 | 6142 | case IORING_OP_SENDMSG: |
889fca73 | 6143 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6144 | break; |
fddaface | 6145 | case IORING_OP_SEND: |
889fca73 | 6146 | ret = io_send(req, issue_flags); |
0fa03c62 | 6147 | break; |
aa1fa28f | 6148 | case IORING_OP_RECVMSG: |
889fca73 | 6149 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6150 | break; |
fddaface | 6151 | case IORING_OP_RECV: |
889fca73 | 6152 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6153 | break; |
5262f567 | 6154 | case IORING_OP_TIMEOUT: |
61e98203 | 6155 | ret = io_timeout(req, issue_flags); |
5262f567 | 6156 | break; |
11365043 | 6157 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6158 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6159 | break; |
17f2fe35 | 6160 | case IORING_OP_ACCEPT: |
889fca73 | 6161 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6162 | break; |
f8e85cf2 | 6163 | case IORING_OP_CONNECT: |
889fca73 | 6164 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6165 | break; |
62755e35 | 6166 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6167 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6168 | break; |
d63d1b5e | 6169 | case IORING_OP_FALLOCATE: |
45d189c6 | 6170 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6171 | break; |
15b71abe | 6172 | case IORING_OP_OPENAT: |
45d189c6 | 6173 | ret = io_openat(req, issue_flags); |
15b71abe | 6174 | break; |
b5dba59e | 6175 | case IORING_OP_CLOSE: |
889fca73 | 6176 | ret = io_close(req, issue_flags); |
b5dba59e | 6177 | break; |
05f3fb3c | 6178 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6179 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6180 | break; |
eddc7ef5 | 6181 | case IORING_OP_STATX: |
45d189c6 | 6182 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6183 | break; |
4840e418 | 6184 | case IORING_OP_FADVISE: |
45d189c6 | 6185 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6186 | break; |
c1ca757b | 6187 | case IORING_OP_MADVISE: |
45d189c6 | 6188 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6189 | break; |
cebdb986 | 6190 | case IORING_OP_OPENAT2: |
45d189c6 | 6191 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6192 | break; |
3e4827b0 | 6193 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6194 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6195 | break; |
7d67af2c | 6196 | case IORING_OP_SPLICE: |
45d189c6 | 6197 | ret = io_splice(req, issue_flags); |
7d67af2c | 6198 | break; |
ddf0322d | 6199 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6200 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6201 | break; |
067524e9 | 6202 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6203 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6204 | break; |
f2a8d5c7 | 6205 | case IORING_OP_TEE: |
45d189c6 | 6206 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6207 | break; |
36f4fa68 | 6208 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6209 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6210 | break; |
80a261fd | 6211 | case IORING_OP_RENAMEAT: |
45d189c6 | 6212 | ret = io_renameat(req, issue_flags); |
80a261fd | 6213 | break; |
14a1143b | 6214 | case IORING_OP_UNLINKAT: |
45d189c6 | 6215 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6216 | break; |
2b188cc1 JA |
6217 | default: |
6218 | ret = -EINVAL; | |
6219 | break; | |
6220 | } | |
6221 | ||
5730b27e JA |
6222 | if (creds) |
6223 | revert_creds(creds); | |
6224 | ||
def596e9 JA |
6225 | if (ret) |
6226 | return ret; | |
6227 | ||
b532576e JA |
6228 | /* If the op doesn't have a file, we're not polling for it */ |
6229 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6230 | const bool in_async = io_wq_current_is_worker(); |
6231 | ||
11ba820b JA |
6232 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6233 | if (in_async) | |
6234 | mutex_lock(&ctx->uring_lock); | |
6235 | ||
2e9dbe90 | 6236 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6237 | |
6238 | if (in_async) | |
6239 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6240 | } |
6241 | ||
6242 | return 0; | |
2b188cc1 JA |
6243 | } |
6244 | ||
5280f7e5 | 6245 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6246 | { |
6247 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6248 | struct io_kiocb *timeout; |
561fb04a | 6249 | int ret = 0; |
2b188cc1 | 6250 | |
6df1db6b PB |
6251 | timeout = io_prep_linked_timeout(req); |
6252 | if (timeout) | |
6253 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6254 | |
4014d943 | 6255 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6256 | ret = -ECANCELED; |
31b51510 | 6257 | |
561fb04a | 6258 | if (!ret) { |
561fb04a | 6259 | do { |
889fca73 | 6260 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6261 | /* |
6262 | * We can get EAGAIN for polled IO even though we're | |
6263 | * forcing a sync submission from here, since we can't | |
6264 | * wait for request slots on the block side. | |
6265 | */ | |
6266 | if (ret != -EAGAIN) | |
6267 | break; | |
6268 | cond_resched(); | |
6269 | } while (1); | |
6270 | } | |
31b51510 | 6271 | |
a3df7698 | 6272 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6273 | if (ret) { |
a3df7698 | 6274 | /* io-wq is going to take one down */ |
de9b4cca | 6275 | req_ref_get(req); |
a3df7698 | 6276 | io_req_task_queue_fail(req, ret); |
edafccee | 6277 | } |
2b188cc1 JA |
6278 | } |
6279 | ||
7b29f92d JA |
6280 | #define FFS_ASYNC_READ 0x1UL |
6281 | #define FFS_ASYNC_WRITE 0x2UL | |
6282 | #ifdef CONFIG_64BIT | |
6283 | #define FFS_ISREG 0x4UL | |
6284 | #else | |
6285 | #define FFS_ISREG 0x0UL | |
6286 | #endif | |
6287 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6288 | ||
b895c9a6 | 6289 | static inline struct file **io_fixed_file_slot(struct io_rsrc_data *file_data, |
dafecf19 | 6290 | unsigned i) |
65e19f54 | 6291 | { |
269bbe5f | 6292 | struct fixed_rsrc_table *table; |
65e19f54 | 6293 | |
dafecf19 PB |
6294 | table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT]; |
6295 | return &table->files[i & IORING_FILE_TABLE_MASK]; | |
6296 | } | |
6297 | ||
6298 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, | |
6299 | int index) | |
6300 | { | |
7b29f92d JA |
6301 | struct file **file_slot = io_fixed_file_slot(ctx->file_data, index); |
6302 | ||
6303 | return (struct file *) ((unsigned long) *file_slot & FFS_MASK); | |
65e19f54 JA |
6304 | } |
6305 | ||
8371adf5 PB |
6306 | static struct file *io_file_get(struct io_submit_state *state, |
6307 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6308 | { |
a197f664 | 6309 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6310 | struct file *file; |
09bb8394 | 6311 | |
8da11c19 | 6312 | if (fixed) { |
7b29f92d JA |
6313 | unsigned long file_ptr; |
6314 | ||
479f517b | 6315 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6316 | return NULL; |
b7620121 | 6317 | fd = array_index_nospec(fd, ctx->nr_user_files); |
7b29f92d JA |
6318 | file_ptr = (unsigned long) *io_fixed_file_slot(ctx->file_data, fd); |
6319 | file = (struct file *) (file_ptr & FFS_MASK); | |
6320 | file_ptr &= ~FFS_MASK; | |
6321 | /* mask in overlapping REQ_F and FFS bits */ | |
6322 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6323 | io_req_set_rsrc_node(req); |
09bb8394 | 6324 | } else { |
c826bd7a | 6325 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6326 | file = __io_file_get(state, fd); |
d44f554e JA |
6327 | |
6328 | /* we don't allow fixed io_uring files */ | |
6329 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6330 | io_req_track_inflight(req); | |
09bb8394 JA |
6331 | } |
6332 | ||
8371adf5 | 6333 | return file; |
09bb8394 JA |
6334 | } |
6335 | ||
2665abfd | 6336 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6337 | { |
ad8a48ac JA |
6338 | struct io_timeout_data *data = container_of(timer, |
6339 | struct io_timeout_data, timer); | |
90cd7e42 | 6340 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6341 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6342 | unsigned long flags; |
2665abfd JA |
6343 | |
6344 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6345 | prev = req->timeout.head; |
6346 | req->timeout.head = NULL; | |
2665abfd JA |
6347 | |
6348 | /* | |
6349 | * We don't expect the list to be empty, that will only happen if we | |
6350 | * race with the completion of the linked work. | |
6351 | */ | |
de9b4cca | 6352 | if (prev && req_ref_inc_not_zero(prev)) |
f2f87370 | 6353 | io_remove_next_linked(prev); |
90cd7e42 PB |
6354 | else |
6355 | prev = NULL; | |
2665abfd JA |
6356 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6357 | ||
6358 | if (prev) { | |
014db007 | 6359 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6360 | io_put_req_deferred(prev, 1); |
47f46768 | 6361 | } else { |
9ae1f8dd PB |
6362 | io_req_complete_post(req, -ETIME, 0); |
6363 | io_put_req_deferred(req, 1); | |
2665abfd | 6364 | } |
2665abfd JA |
6365 | return HRTIMER_NORESTART; |
6366 | } | |
6367 | ||
de968c18 | 6368 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6369 | { |
de968c18 PB |
6370 | struct io_ring_ctx *ctx = req->ctx; |
6371 | ||
6372 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6373 | /* |
f2f87370 PB |
6374 | * If the back reference is NULL, then our linked request finished |
6375 | * before we got a chance to setup the timer | |
76a46e06 | 6376 | */ |
90cd7e42 | 6377 | if (req->timeout.head) { |
e8c2bc1f | 6378 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6379 | |
ad8a48ac JA |
6380 | data->timer.function = io_link_timeout_fn; |
6381 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6382 | data->mode); | |
2665abfd | 6383 | } |
76a46e06 | 6384 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6385 | /* drop submission reference */ |
76a46e06 JA |
6386 | io_put_req(req); |
6387 | } | |
2665abfd | 6388 | |
ad8a48ac | 6389 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6390 | { |
f2f87370 | 6391 | struct io_kiocb *nxt = req->link; |
2665abfd | 6392 | |
f2f87370 PB |
6393 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6394 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6395 | return NULL; |
2665abfd | 6396 | |
90cd7e42 | 6397 | nxt->timeout.head = req; |
900fad45 | 6398 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6399 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6400 | return nxt; |
2665abfd JA |
6401 | } |
6402 | ||
c5eef2b9 | 6403 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6404 | { |
d3d7298d | 6405 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6406 | int ret; |
2b188cc1 | 6407 | |
c5eef2b9 | 6408 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6409 | |
491381ce JA |
6410 | /* |
6411 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6412 | * doesn't support non-blocking read/write attempts | |
6413 | */ | |
1840038e | 6414 | if (likely(!ret)) { |
0d63c148 | 6415 | /* drop submission reference */ |
e342c807 | 6416 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6417 | struct io_ring_ctx *ctx = req->ctx; |
6418 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6419 | |
6dd0be1e | 6420 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6421 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6422 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6423 | } else { |
d3d7298d | 6424 | io_put_req(req); |
0d63c148 | 6425 | } |
1840038e PB |
6426 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6427 | if (!io_arm_poll_handler(req)) { | |
6428 | /* | |
6429 | * Queued up for async execution, worker will release | |
6430 | * submit reference when the iocb is actually submitted. | |
6431 | */ | |
6432 | io_queue_async_work(req); | |
6433 | } | |
0d63c148 | 6434 | } else { |
f41db273 | 6435 | io_req_complete_failed(req, ret); |
9e645e11 | 6436 | } |
d3d7298d PB |
6437 | if (linked_timeout) |
6438 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6439 | } |
6440 | ||
be7053b7 | 6441 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6442 | { |
6443 | int ret; | |
6444 | ||
be7053b7 | 6445 | ret = io_req_defer(req); |
4fe2c963 JL |
6446 | if (ret) { |
6447 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6448 | fail_req: |
f41db273 | 6449 | io_req_complete_failed(req, ret); |
4fe2c963 | 6450 | } |
2550878f | 6451 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6452 | ret = io_req_prep_async(req); |
be7053b7 PB |
6453 | if (unlikely(ret)) |
6454 | goto fail_req; | |
ce35a47a JA |
6455 | io_queue_async_work(req); |
6456 | } else { | |
c5eef2b9 | 6457 | __io_queue_sqe(req); |
ce35a47a | 6458 | } |
4fe2c963 JL |
6459 | } |
6460 | ||
b16fed66 PB |
6461 | /* |
6462 | * Check SQE restrictions (opcode and flags). | |
6463 | * | |
6464 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6465 | */ | |
6466 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6467 | struct io_kiocb *req, | |
6468 | unsigned int sqe_flags) | |
4fe2c963 | 6469 | { |
b16fed66 PB |
6470 | if (!ctx->restricted) |
6471 | return true; | |
6472 | ||
6473 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6474 | return false; | |
6475 | ||
6476 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6477 | ctx->restrictions.sqe_flags_required) | |
6478 | return false; | |
6479 | ||
6480 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6481 | ctx->restrictions.sqe_flags_required)) | |
6482 | return false; | |
6483 | ||
6484 | return true; | |
4fe2c963 JL |
6485 | } |
6486 | ||
b16fed66 PB |
6487 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6488 | const struct io_uring_sqe *sqe) | |
6489 | { | |
6490 | struct io_submit_state *state; | |
6491 | unsigned int sqe_flags; | |
003e8dcc | 6492 | int personality, ret = 0; |
b16fed66 PB |
6493 | |
6494 | req->opcode = READ_ONCE(sqe->opcode); | |
6495 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6496 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6497 | req->user_data = READ_ONCE(sqe->user_data); | |
6498 | req->async_data = NULL; | |
6499 | req->file = NULL; | |
6500 | req->ctx = ctx; | |
6501 | req->link = NULL; | |
6502 | req->fixed_rsrc_refs = NULL; | |
6503 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6504 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6505 | req->task = current; |
6506 | req->result = 0; | |
93e68e03 | 6507 | req->work.creds = NULL; |
b16fed66 PB |
6508 | |
6509 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6510 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6511 | req->flags = 0; | |
b16fed66 | 6512 | return -EINVAL; |
ebf4a5db | 6513 | } |
b16fed66 PB |
6514 | |
6515 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6516 | return -EINVAL; | |
6517 | ||
b16fed66 PB |
6518 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6519 | return -EACCES; | |
6520 | ||
6521 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6522 | !io_op_defs[req->opcode].buffer_select) | |
6523 | return -EOPNOTSUPP; | |
863e0560 | 6524 | |
003e8dcc JA |
6525 | personality = READ_ONCE(sqe->personality); |
6526 | if (personality) { | |
61cf9370 | 6527 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6528 | if (!req->work.creds) |
6529 | return -EINVAL; | |
6530 | get_cred(req->work.creds); | |
003e8dcc | 6531 | } |
b16fed66 PB |
6532 | state = &ctx->submit_state; |
6533 | ||
6534 | /* | |
6535 | * Plug now if we have more than 1 IO left after this, and the target | |
6536 | * is potentially a read/write to block based storage. | |
6537 | */ | |
6538 | if (!state->plug_started && state->ios_left > 1 && | |
6539 | io_op_defs[req->opcode].plug) { | |
6540 | blk_start_plug(&state->plug); | |
6541 | state->plug_started = true; | |
6542 | } | |
6543 | ||
6544 | if (io_op_defs[req->opcode].needs_file) { | |
6545 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6546 | ||
6547 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6548 | if (unlikely(!req->file)) | |
6549 | ret = -EBADF; | |
6550 | } | |
6551 | ||
6552 | state->ios_left--; | |
6553 | return ret; | |
6554 | } | |
6555 | ||
a6b8cadc | 6556 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6557 | const struct io_uring_sqe *sqe) |
9e645e11 | 6558 | { |
a1ab7b35 | 6559 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6560 | int ret; |
9e645e11 | 6561 | |
a6b8cadc PB |
6562 | ret = io_init_req(ctx, req, sqe); |
6563 | if (unlikely(ret)) { | |
6564 | fail_req: | |
de59bc10 PB |
6565 | if (link->head) { |
6566 | /* fail even hard links since we don't submit */ | |
cf109604 | 6567 | link->head->flags |= REQ_F_FAIL_LINK; |
f41db273 | 6568 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6569 | link->head = NULL; |
6570 | } | |
f41db273 | 6571 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6572 | return ret; |
6573 | } | |
be7053b7 PB |
6574 | ret = io_req_prep(req, sqe); |
6575 | if (unlikely(ret)) | |
6576 | goto fail_req; | |
a6b8cadc | 6577 | |
be7053b7 | 6578 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6579 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6580 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6581 | ||
9e645e11 JA |
6582 | /* |
6583 | * If we already have a head request, queue this one for async | |
6584 | * submittal once the head completes. If we don't have a head but | |
6585 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6586 | * submitted sync once the chain is complete. If none of those | |
6587 | * conditions are true (normal request), then just queue it. | |
6588 | */ | |
863e0560 PB |
6589 | if (link->head) { |
6590 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6591 | |
8cdf2193 PB |
6592 | /* |
6593 | * Taking sequential execution of a link, draining both sides | |
6594 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6595 | * requests in the link. So, it drains the head and the | |
6596 | * next after the link request. The last one is done via | |
6597 | * drain_next flag to persist the effect across calls. | |
6598 | */ | |
ef4ff581 | 6599 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6600 | head->flags |= REQ_F_IO_DRAIN; |
6601 | ctx->drain_next = 1; | |
6602 | } | |
b7e298d2 | 6603 | ret = io_req_prep_async(req); |
cf109604 | 6604 | if (unlikely(ret)) |
a6b8cadc | 6605 | goto fail_req; |
9d76377f | 6606 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6607 | link->last->link = req; |
863e0560 | 6608 | link->last = req; |
32fe525b PB |
6609 | |
6610 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6611 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6612 | io_queue_sqe(head); |
863e0560 | 6613 | link->head = NULL; |
32fe525b | 6614 | } |
9e645e11 | 6615 | } else { |
711be031 PB |
6616 | if (unlikely(ctx->drain_next)) { |
6617 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6618 | ctx->drain_next = 0; |
711be031 | 6619 | } |
ef4ff581 | 6620 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6621 | link->head = req; |
6622 | link->last = req; | |
711be031 | 6623 | } else { |
be7053b7 | 6624 | io_queue_sqe(req); |
711be031 | 6625 | } |
9e645e11 | 6626 | } |
2e6e1fde | 6627 | |
1d4240cc | 6628 | return 0; |
9e645e11 JA |
6629 | } |
6630 | ||
9a56a232 JA |
6631 | /* |
6632 | * Batched submission is done, ensure local IO is flushed out. | |
6633 | */ | |
ba88ff11 PB |
6634 | static void io_submit_state_end(struct io_submit_state *state, |
6635 | struct io_ring_ctx *ctx) | |
9a56a232 | 6636 | { |
a1ab7b35 | 6637 | if (state->link.head) |
de59bc10 | 6638 | io_queue_sqe(state->link.head); |
6dd0be1e | 6639 | if (state->comp.nr) |
ba88ff11 | 6640 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6641 | if (state->plug_started) |
6642 | blk_finish_plug(&state->plug); | |
9f13c35b | 6643 | io_state_file_put(state); |
9a56a232 JA |
6644 | } |
6645 | ||
6646 | /* | |
6647 | * Start submission side cache. | |
6648 | */ | |
6649 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6650 | unsigned int max_ios) |
9a56a232 | 6651 | { |
27926b68 | 6652 | state->plug_started = false; |
9a56a232 | 6653 | state->ios_left = max_ios; |
a1ab7b35 PB |
6654 | /* set only head, no need to init link_last in advance */ |
6655 | state->link.head = NULL; | |
9a56a232 JA |
6656 | } |
6657 | ||
2b188cc1 JA |
6658 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6659 | { | |
75b28aff | 6660 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6661 | |
caf582c6 PB |
6662 | /* |
6663 | * Ensure any loads from the SQEs are done at this point, | |
6664 | * since once we write the new head, the application could | |
6665 | * write new data to them. | |
6666 | */ | |
6667 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6668 | } |
6669 | ||
2b188cc1 | 6670 | /* |
3529d8c2 | 6671 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6672 | * that is mapped by userspace. This means that care needs to be taken to |
6673 | * ensure that reads are stable, as we cannot rely on userspace always | |
6674 | * being a good citizen. If members of the sqe are validated and then later | |
6675 | * used, it's important that those reads are done through READ_ONCE() to | |
6676 | * prevent a re-load down the line. | |
6677 | */ | |
709b302f | 6678 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6679 | { |
75b28aff | 6680 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6681 | unsigned head; |
6682 | ||
6683 | /* | |
6684 | * The cached sq head (or cq tail) serves two purposes: | |
6685 | * | |
6686 | * 1) allows us to batch the cost of updating the user visible | |
6687 | * head updates. | |
6688 | * 2) allows the kernel side to track the head on its own, even | |
6689 | * though the application is the one updating it. | |
6690 | */ | |
4fccfcbb | 6691 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6692 | if (likely(head < ctx->sq_entries)) |
6693 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6694 | |
6695 | /* drop invalid entries */ | |
498ccd9e | 6696 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6697 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6698 | return NULL; |
6699 | } | |
6700 | ||
0f212204 | 6701 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6702 | { |
46c4e16a | 6703 | int submitted = 0; |
6c271ce2 | 6704 | |
c4a2ed72 | 6705 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6706 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c2450ae | 6707 | if (!__io_cqring_overflow_flush(ctx, false)) |
ad3eb2c8 JA |
6708 | return -EBUSY; |
6709 | } | |
6c271ce2 | 6710 | |
ee7d46d9 PB |
6711 | /* make sure SQ entry isn't read before tail */ |
6712 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6713 | |
2b85edfc PB |
6714 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6715 | return -EAGAIN; | |
6c271ce2 | 6716 | |
d8a6df10 | 6717 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6718 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6719 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6720 | |
46c4e16a | 6721 | while (submitted < nr) { |
3529d8c2 | 6722 | const struct io_uring_sqe *sqe; |
196be95c | 6723 | struct io_kiocb *req; |
fb5ccc98 | 6724 | |
258b29a9 | 6725 | req = io_alloc_req(ctx); |
196be95c PB |
6726 | if (unlikely(!req)) { |
6727 | if (!submitted) | |
6728 | submitted = -EAGAIN; | |
fb5ccc98 | 6729 | break; |
196be95c | 6730 | } |
4fccfcbb PB |
6731 | sqe = io_get_sqe(ctx); |
6732 | if (unlikely(!sqe)) { | |
6733 | kmem_cache_free(req_cachep, req); | |
6734 | break; | |
6735 | } | |
d3656344 JA |
6736 | /* will complete beyond this point, count as submitted */ |
6737 | submitted++; | |
a1ab7b35 | 6738 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6739 | break; |
6c271ce2 JA |
6740 | } |
6741 | ||
9466f437 PB |
6742 | if (unlikely(submitted != nr)) { |
6743 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6744 | struct io_uring_task *tctx = current->io_uring; |
6745 | int unused = nr - ref_used; | |
9466f437 | 6746 | |
d8a6df10 JA |
6747 | percpu_ref_put_many(&ctx->refs, unused); |
6748 | percpu_counter_sub(&tctx->inflight, unused); | |
6749 | put_task_struct_many(current, unused); | |
9466f437 | 6750 | } |
6c271ce2 | 6751 | |
a1ab7b35 | 6752 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6753 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6754 | io_commit_sqring(ctx); | |
6755 | ||
6c271ce2 JA |
6756 | return submitted; |
6757 | } | |
6758 | ||
23b3628e XW |
6759 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6760 | { | |
6761 | /* Tell userspace we may need a wakeup call */ | |
6762 | spin_lock_irq(&ctx->completion_lock); | |
6763 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6764 | spin_unlock_irq(&ctx->completion_lock); | |
6765 | } | |
6766 | ||
6767 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6768 | { | |
6769 | spin_lock_irq(&ctx->completion_lock); | |
6770 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6771 | spin_unlock_irq(&ctx->completion_lock); | |
6772 | } | |
6773 | ||
08369246 | 6774 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6775 | { |
c8d1ba58 | 6776 | unsigned int to_submit; |
bdcd3eab | 6777 | int ret = 0; |
6c271ce2 | 6778 | |
c8d1ba58 | 6779 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6780 | /* if we're handling multiple rings, cap submit size for fairness */ |
6781 | if (cap_entries && to_submit > 8) | |
6782 | to_submit = 8; | |
6783 | ||
906a3c6f | 6784 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6785 | unsigned nr_events = 0; |
a4c0b3de | 6786 | |
c8d1ba58 | 6787 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6788 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6789 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6790 | |
0298ef96 PB |
6791 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6792 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6793 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6794 | mutex_unlock(&ctx->uring_lock); |
6795 | } | |
6c271ce2 | 6796 | |
90554200 JA |
6797 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6798 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6799 | |
08369246 XW |
6800 | return ret; |
6801 | } | |
6c271ce2 | 6802 | |
08369246 XW |
6803 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6804 | { | |
6805 | struct io_ring_ctx *ctx; | |
6806 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6807 | |
c9dca27d PB |
6808 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6809 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6810 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6811 | } |
6c271ce2 | 6812 | |
c8d1ba58 JA |
6813 | static int io_sq_thread(void *data) |
6814 | { | |
69fb2131 JA |
6815 | struct io_sq_data *sqd = data; |
6816 | struct io_ring_ctx *ctx; | |
a0d9205f | 6817 | unsigned long timeout = 0; |
37d1e2e3 | 6818 | char buf[TASK_COMM_LEN]; |
08369246 | 6819 | DEFINE_WAIT(wait); |
6c271ce2 | 6820 | |
696ee88a | 6821 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6822 | set_task_comm(current, buf); |
37d1e2e3 JA |
6823 | current->pf_io_worker = NULL; |
6824 | ||
6825 | if (sqd->sq_cpu != -1) | |
6826 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6827 | else | |
6828 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6829 | current->flags |= PF_NO_SETAFFINITY; | |
6830 | ||
09a6f4ef | 6831 | mutex_lock(&sqd->lock); |
05962f95 | 6832 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6833 | int ret; |
6834 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6835 | |
82734c5b JA |
6836 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6837 | signal_pending(current)) { | |
6838 | bool did_sig = false; | |
6839 | ||
09a6f4ef | 6840 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6841 | if (signal_pending(current)) { |
6842 | struct ksignal ksig; | |
6843 | ||
6844 | did_sig = get_signal(&ksig); | |
6845 | } | |
05962f95 | 6846 | cond_resched(); |
09a6f4ef | 6847 | mutex_lock(&sqd->lock); |
82734c5b JA |
6848 | if (did_sig) |
6849 | break; | |
521d6a73 | 6850 | io_run_task_work(); |
b7f5a0bf | 6851 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6852 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6853 | continue; |
08369246 | 6854 | } |
08369246 | 6855 | sqt_spin = false; |
e95eee2d | 6856 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6857 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6858 | const struct cred *creds = NULL; |
6859 | ||
6860 | if (ctx->sq_creds != current_cred()) | |
6861 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6862 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6863 | if (creds) |
6864 | revert_creds(creds); | |
08369246 XW |
6865 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6866 | sqt_spin = true; | |
69fb2131 | 6867 | } |
6c271ce2 | 6868 | |
08369246 | 6869 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6870 | io_run_task_work(); |
6871 | cond_resched(); | |
08369246 XW |
6872 | if (sqt_spin) |
6873 | timeout = jiffies + sqd->sq_thread_idle; | |
6874 | continue; | |
6875 | } | |
6876 | ||
08369246 XW |
6877 | needs_sched = true; |
6878 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6879 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6880 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6881 | !list_empty_careful(&ctx->iopoll_list)) { | |
6882 | needs_sched = false; | |
6883 | break; | |
6884 | } | |
6885 | if (io_sqring_entries(ctx)) { | |
6886 | needs_sched = false; | |
6887 | break; | |
6888 | } | |
6889 | } | |
6890 | ||
05962f95 | 6891 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6892 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6893 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6894 | |
09a6f4ef | 6895 | mutex_unlock(&sqd->lock); |
69fb2131 | 6896 | schedule(); |
09a6f4ef | 6897 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6898 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6899 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6900 | } |
08369246 XW |
6901 | |
6902 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6903 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6904 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6905 | } |
28cea78a | 6906 | |
521d6a73 PB |
6907 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6908 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6909 | sqd->thread = NULL; |
05962f95 | 6910 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6911 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6912 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6913 | |
6914 | io_run_task_work(); | |
b7f5a0bf | 6915 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6916 | complete(&sqd->exited); |
6917 | do_exit(0); | |
6c271ce2 JA |
6918 | } |
6919 | ||
bda52162 JA |
6920 | struct io_wait_queue { |
6921 | struct wait_queue_entry wq; | |
6922 | struct io_ring_ctx *ctx; | |
6923 | unsigned to_wait; | |
6924 | unsigned nr_timeouts; | |
6925 | }; | |
6926 | ||
6c503150 | 6927 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6928 | { |
6929 | struct io_ring_ctx *ctx = iowq->ctx; | |
6930 | ||
6931 | /* | |
d195a66e | 6932 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6933 | * started waiting. For timeouts, we always want to return to userspace, |
6934 | * regardless of event count. | |
6935 | */ | |
6c503150 | 6936 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6937 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6938 | } | |
6939 | ||
6940 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6941 | int wake_flags, void *key) | |
6942 | { | |
6943 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6944 | wq); | |
6945 | ||
6c503150 PB |
6946 | /* |
6947 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6948 | * the task, and the next invocation will do it. | |
6949 | */ | |
6950 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6951 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6952 | return -1; | |
bda52162 JA |
6953 | } |
6954 | ||
af9c1a44 JA |
6955 | static int io_run_task_work_sig(void) |
6956 | { | |
6957 | if (io_run_task_work()) | |
6958 | return 1; | |
6959 | if (!signal_pending(current)) | |
6960 | return 0; | |
0b8cfa97 | 6961 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6962 | return -ERESTARTSYS; |
af9c1a44 JA |
6963 | return -EINTR; |
6964 | } | |
6965 | ||
eeb60b9a PB |
6966 | /* when returns >0, the caller should retry */ |
6967 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6968 | struct io_wait_queue *iowq, | |
6969 | signed long *timeout) | |
6970 | { | |
6971 | int ret; | |
6972 | ||
6973 | /* make sure we run task_work before checking for signals */ | |
6974 | ret = io_run_task_work_sig(); | |
6975 | if (ret || io_should_wake(iowq)) | |
6976 | return ret; | |
6977 | /* let the caller flush overflows, retry */ | |
6978 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6979 | return 1; | |
6980 | ||
6981 | *timeout = schedule_timeout(*timeout); | |
6982 | return !*timeout ? -ETIME : 1; | |
6983 | } | |
6984 | ||
2b188cc1 JA |
6985 | /* |
6986 | * Wait until events become available, if we don't already have some. The | |
6987 | * application must reap them itself, as they reside on the shared cq ring. | |
6988 | */ | |
6989 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6990 | const sigset_t __user *sig, size_t sigsz, |
6991 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6992 | { |
bda52162 JA |
6993 | struct io_wait_queue iowq = { |
6994 | .wq = { | |
6995 | .private = current, | |
6996 | .func = io_wake_function, | |
6997 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6998 | }, | |
6999 | .ctx = ctx, | |
7000 | .to_wait = min_events, | |
7001 | }; | |
75b28aff | 7002 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7003 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7004 | int ret; | |
2b188cc1 | 7005 | |
b41e9852 | 7006 | do { |
6c2450ae | 7007 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7008 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7009 | return 0; |
4c6e277c | 7010 | if (!io_run_task_work()) |
b41e9852 | 7011 | break; |
b41e9852 | 7012 | } while (1); |
2b188cc1 JA |
7013 | |
7014 | if (sig) { | |
9e75ad5d AB |
7015 | #ifdef CONFIG_COMPAT |
7016 | if (in_compat_syscall()) | |
7017 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7018 | sigsz); |
9e75ad5d AB |
7019 | else |
7020 | #endif | |
b772434b | 7021 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7022 | |
2b188cc1 JA |
7023 | if (ret) |
7024 | return ret; | |
7025 | } | |
7026 | ||
c73ebb68 | 7027 | if (uts) { |
c1d5a224 PB |
7028 | struct timespec64 ts; |
7029 | ||
c73ebb68 HX |
7030 | if (get_timespec64(&ts, uts)) |
7031 | return -EFAULT; | |
7032 | timeout = timespec64_to_jiffies(&ts); | |
7033 | } | |
7034 | ||
bda52162 | 7035 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7036 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7037 | do { |
ca0a2651 | 7038 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7039 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7040 | ret = -EBUSY; |
7041 | break; | |
7042 | } | |
bda52162 JA |
7043 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7044 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7045 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7046 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 7047 | cond_resched(); |
eeb60b9a | 7048 | } while (ret > 0); |
bda52162 | 7049 | |
b7db41c9 | 7050 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7051 | |
75b28aff | 7052 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7053 | } |
7054 | ||
6b06314c JA |
7055 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7056 | { | |
7057 | #if defined(CONFIG_UNIX) | |
7058 | if (ctx->ring_sock) { | |
7059 | struct sock *sock = ctx->ring_sock->sk; | |
7060 | struct sk_buff *skb; | |
7061 | ||
7062 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7063 | kfree_skb(skb); | |
7064 | } | |
7065 | #else | |
7066 | int i; | |
7067 | ||
65e19f54 JA |
7068 | for (i = 0; i < ctx->nr_user_files; i++) { |
7069 | struct file *file; | |
7070 | ||
7071 | file = io_file_from_index(ctx, i); | |
7072 | if (file) | |
7073 | fput(file); | |
7074 | } | |
6b06314c JA |
7075 | #endif |
7076 | } | |
7077 | ||
00835dce | 7078 | static void io_rsrc_data_ref_zero(struct percpu_ref *ref) |
05f3fb3c | 7079 | { |
b895c9a6 | 7080 | struct io_rsrc_data *data = container_of(ref, struct io_rsrc_data, refs); |
05f3fb3c | 7081 | |
05f3fb3c JA |
7082 | complete(&data->done); |
7083 | } | |
7084 | ||
2a63b2d9 | 7085 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7086 | { |
2a63b2d9 | 7087 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7088 | } |
7089 | ||
2a63b2d9 | 7090 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7091 | { |
2a63b2d9 BM |
7092 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7093 | } | |
65e19f54 | 7094 | |
b895c9a6 PB |
7095 | static void io_rsrc_node_set(struct io_ring_ctx *ctx, |
7096 | struct io_rsrc_data *rsrc_data, | |
7097 | struct io_rsrc_node *rsrc_node) | |
1642b445 | 7098 | { |
2a63b2d9 | 7099 | io_rsrc_ref_lock(ctx); |
b895c9a6 PB |
7100 | rsrc_data->node = rsrc_node; |
7101 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
2a63b2d9 | 7102 | io_rsrc_ref_unlock(ctx); |
269bbe5f | 7103 | percpu_ref_get(&rsrc_data->refs); |
1642b445 PB |
7104 | } |
7105 | ||
b895c9a6 | 7106 | static void io_rsrc_node_kill(struct io_ring_ctx *ctx, struct io_rsrc_data *data) |
6b06314c | 7107 | { |
b895c9a6 | 7108 | struct io_rsrc_node *ref_node = NULL; |
6b06314c | 7109 | |
2a63b2d9 | 7110 | io_rsrc_ref_lock(ctx); |
1e5d770b | 7111 | ref_node = data->node; |
e6cb007c | 7112 | data->node = NULL; |
2a63b2d9 | 7113 | io_rsrc_ref_unlock(ctx); |
05589553 XW |
7114 | if (ref_node) |
7115 | percpu_ref_kill(&ref_node->refs); | |
8bad28d8 HX |
7116 | } |
7117 | ||
b895c9a6 | 7118 | static int io_rsrc_node_prealloc(struct io_ring_ctx *ctx) |
8dd03afe PB |
7119 | { |
7120 | if (ctx->rsrc_backup_node) | |
7121 | return 0; | |
b895c9a6 | 7122 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe PB |
7123 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
7124 | } | |
7125 | ||
b895c9a6 PB |
7126 | static struct io_rsrc_node * |
7127 | io_rsrc_node_get(struct io_ring_ctx *ctx, | |
7128 | struct io_rsrc_data *rsrc_data, | |
7129 | void (*rsrc_put)(struct io_ring_ctx *ctx, | |
7130 | struct io_rsrc_put *prsrc)) | |
8dd03afe | 7131 | { |
b895c9a6 | 7132 | struct io_rsrc_node *node = ctx->rsrc_backup_node; |
8dd03afe PB |
7133 | |
7134 | WARN_ON_ONCE(!node); | |
7135 | ||
7136 | ctx->rsrc_backup_node = NULL; | |
7137 | node->rsrc_data = rsrc_data; | |
7138 | node->rsrc_put = rsrc_put; | |
7139 | return node; | |
7140 | } | |
7141 | ||
b895c9a6 | 7142 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, |
8bad28d8 | 7143 | struct io_ring_ctx *ctx, |
f2303b1f PB |
7144 | void (*rsrc_put)(struct io_ring_ctx *ctx, |
7145 | struct io_rsrc_put *prsrc)) | |
8bad28d8 | 7146 | { |
b895c9a6 | 7147 | struct io_rsrc_node *node; |
8bad28d8 | 7148 | int ret; |
05589553 | 7149 | |
8bad28d8 HX |
7150 | if (data->quiesce) |
7151 | return -ENXIO; | |
05589553 | 7152 | |
8bad28d8 | 7153 | data->quiesce = true; |
1ffc5422 | 7154 | do { |
b895c9a6 | 7155 | ret = io_rsrc_node_prealloc(ctx); |
8dd03afe | 7156 | if (ret) |
f2303b1f | 7157 | break; |
b895c9a6 | 7158 | io_rsrc_node_kill(ctx, data); |
8bad28d8 HX |
7159 | percpu_ref_kill(&data->refs); |
7160 | flush_delayed_work(&ctx->rsrc_put_work); | |
7161 | ||
1ffc5422 PB |
7162 | ret = wait_for_completion_interruptible(&data->done); |
7163 | if (!ret) | |
7164 | break; | |
8bad28d8 | 7165 | |
cb5e1b81 | 7166 | percpu_ref_resurrect(&data->refs); |
b895c9a6 PB |
7167 | node = io_rsrc_node_get(ctx, data, rsrc_put); |
7168 | io_rsrc_node_set(ctx, data, node); | |
cb5e1b81 | 7169 | reinit_completion(&data->done); |
8dd03afe | 7170 | |
8bad28d8 | 7171 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7172 | ret = io_run_task_work_sig(); |
8bad28d8 | 7173 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7174 | } while (ret >= 0); |
8bad28d8 | 7175 | data->quiesce = false; |
05f3fb3c | 7176 | |
8bad28d8 | 7177 | return ret; |
d7954b2b BM |
7178 | } |
7179 | ||
b895c9a6 | 7180 | static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx) |
1ad555c6 | 7181 | { |
b895c9a6 | 7182 | struct io_rsrc_data *data; |
1ad555c6 BM |
7183 | |
7184 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7185 | if (!data) | |
7186 | return NULL; | |
7187 | ||
00835dce | 7188 | if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero, |
1ad555c6 BM |
7189 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) { |
7190 | kfree(data); | |
7191 | return NULL; | |
7192 | } | |
7193 | data->ctx = ctx; | |
7194 | init_completion(&data->done); | |
7195 | return data; | |
7196 | } | |
7197 | ||
b895c9a6 | 7198 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
1ad555c6 BM |
7199 | { |
7200 | percpu_ref_exit(&data->refs); | |
7201 | kfree(data->table); | |
7202 | kfree(data); | |
7203 | } | |
7204 | ||
d7954b2b BM |
7205 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7206 | { | |
b895c9a6 | 7207 | struct io_rsrc_data *data = ctx->file_data; |
d7954b2b BM |
7208 | unsigned nr_tables, i; |
7209 | int ret; | |
7210 | ||
8bad28d8 HX |
7211 | /* |
7212 | * percpu_ref_is_dying() is to stop parallel files unregister | |
7213 | * Since we possibly drop uring lock later in this function to | |
7214 | * run task work. | |
7215 | */ | |
7216 | if (!data || percpu_ref_is_dying(&data->refs)) | |
d7954b2b | 7217 | return -ENXIO; |
f2303b1f | 7218 | ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put); |
d7954b2b BM |
7219 | if (ret) |
7220 | return ret; | |
7221 | ||
6b06314c | 7222 | __io_sqe_files_unregister(ctx); |
65e19f54 JA |
7223 | nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE); |
7224 | for (i = 0; i < nr_tables; i++) | |
05f3fb3c | 7225 | kfree(data->table[i].files); |
b895c9a6 | 7226 | io_rsrc_data_free(data); |
05f3fb3c | 7227 | ctx->file_data = NULL; |
6b06314c JA |
7228 | ctx->nr_user_files = 0; |
7229 | return 0; | |
7230 | } | |
7231 | ||
37d1e2e3 | 7232 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7233 | __releases(&sqd->lock) |
37d1e2e3 | 7234 | { |
521d6a73 PB |
7235 | WARN_ON_ONCE(sqd->thread == current); |
7236 | ||
9e138a48 PB |
7237 | /* |
7238 | * Do the dance but not conditional clear_bit() because it'd race with | |
7239 | * other threads incrementing park_pending and setting the bit. | |
7240 | */ | |
37d1e2e3 | 7241 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7242 | if (atomic_dec_return(&sqd->park_pending)) |
7243 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7244 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7245 | } |
7246 | ||
86e0d676 | 7247 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7248 | __acquires(&sqd->lock) |
37d1e2e3 | 7249 | { |
521d6a73 PB |
7250 | WARN_ON_ONCE(sqd->thread == current); |
7251 | ||
9e138a48 | 7252 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7253 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7254 | mutex_lock(&sqd->lock); |
05962f95 | 7255 | if (sqd->thread) |
86e0d676 | 7256 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7257 | } |
7258 | ||
7259 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7260 | { | |
521d6a73 PB |
7261 | WARN_ON_ONCE(sqd->thread == current); |
7262 | ||
09a6f4ef | 7263 | mutex_lock(&sqd->lock); |
05962f95 | 7264 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
e8f98f24 JA |
7265 | if (sqd->thread) |
7266 | wake_up_process(sqd->thread); | |
09a6f4ef | 7267 | mutex_unlock(&sqd->lock); |
05962f95 | 7268 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7269 | } |
7270 | ||
534ca6d6 | 7271 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7272 | { |
534ca6d6 | 7273 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7274 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7275 | ||
37d1e2e3 JA |
7276 | io_sq_thread_stop(sqd); |
7277 | kfree(sqd); | |
7278 | } | |
7279 | } | |
7280 | ||
7281 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7282 | { | |
7283 | struct io_sq_data *sqd = ctx->sq_data; | |
7284 | ||
7285 | if (sqd) { | |
05962f95 | 7286 | io_sq_thread_park(sqd); |
521d6a73 | 7287 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7288 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7289 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7290 | |
7291 | io_put_sq_data(sqd); | |
7292 | ctx->sq_data = NULL; | |
7c30f36a SM |
7293 | if (ctx->sq_creds) |
7294 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7295 | } |
7296 | } | |
7297 | ||
aa06165d JA |
7298 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7299 | { | |
7300 | struct io_ring_ctx *ctx_attach; | |
7301 | struct io_sq_data *sqd; | |
7302 | struct fd f; | |
7303 | ||
7304 | f = fdget(p->wq_fd); | |
7305 | if (!f.file) | |
7306 | return ERR_PTR(-ENXIO); | |
7307 | if (f.file->f_op != &io_uring_fops) { | |
7308 | fdput(f); | |
7309 | return ERR_PTR(-EINVAL); | |
7310 | } | |
7311 | ||
7312 | ctx_attach = f.file->private_data; | |
7313 | sqd = ctx_attach->sq_data; | |
7314 | if (!sqd) { | |
7315 | fdput(f); | |
7316 | return ERR_PTR(-EINVAL); | |
7317 | } | |
5c2469e0 JA |
7318 | if (sqd->task_tgid != current->tgid) { |
7319 | fdput(f); | |
7320 | return ERR_PTR(-EPERM); | |
7321 | } | |
aa06165d JA |
7322 | |
7323 | refcount_inc(&sqd->refs); | |
7324 | fdput(f); | |
7325 | return sqd; | |
7326 | } | |
7327 | ||
26984fbf PB |
7328 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7329 | bool *attached) | |
534ca6d6 JA |
7330 | { |
7331 | struct io_sq_data *sqd; | |
7332 | ||
26984fbf | 7333 | *attached = false; |
5c2469e0 JA |
7334 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7335 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7336 | if (!IS_ERR(sqd)) { |
7337 | *attached = true; | |
5c2469e0 | 7338 | return sqd; |
26984fbf | 7339 | } |
5c2469e0 JA |
7340 | /* fall through for EPERM case, setup new sqd/task */ |
7341 | if (PTR_ERR(sqd) != -EPERM) | |
7342 | return sqd; | |
7343 | } | |
aa06165d | 7344 | |
534ca6d6 JA |
7345 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7346 | if (!sqd) | |
7347 | return ERR_PTR(-ENOMEM); | |
7348 | ||
9e138a48 | 7349 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7350 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7351 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7352 | mutex_init(&sqd->lock); |
534ca6d6 | 7353 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7354 | init_completion(&sqd->exited); |
534ca6d6 JA |
7355 | return sqd; |
7356 | } | |
7357 | ||
6b06314c | 7358 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7359 | /* |
7360 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7361 | * the io_uring can be safely unregistered on process exit, even if we have | |
7362 | * loops in the file referencing. | |
7363 | */ | |
7364 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7365 | { | |
7366 | struct sock *sk = ctx->ring_sock->sk; | |
7367 | struct scm_fp_list *fpl; | |
7368 | struct sk_buff *skb; | |
08a45173 | 7369 | int i, nr_files; |
6b06314c | 7370 | |
6b06314c JA |
7371 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7372 | if (!fpl) | |
7373 | return -ENOMEM; | |
7374 | ||
7375 | skb = alloc_skb(0, GFP_KERNEL); | |
7376 | if (!skb) { | |
7377 | kfree(fpl); | |
7378 | return -ENOMEM; | |
7379 | } | |
7380 | ||
7381 | skb->sk = sk; | |
6b06314c | 7382 | |
08a45173 | 7383 | nr_files = 0; |
62e398be | 7384 | fpl->user = get_uid(current_user()); |
6b06314c | 7385 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7386 | struct file *file = io_file_from_index(ctx, i + offset); |
7387 | ||
7388 | if (!file) | |
08a45173 | 7389 | continue; |
65e19f54 | 7390 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7391 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7392 | nr_files++; | |
6b06314c JA |
7393 | } |
7394 | ||
08a45173 JA |
7395 | if (nr_files) { |
7396 | fpl->max = SCM_MAX_FD; | |
7397 | fpl->count = nr_files; | |
7398 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7399 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7400 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7401 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7402 | |
08a45173 JA |
7403 | for (i = 0; i < nr_files; i++) |
7404 | fput(fpl->fp[i]); | |
7405 | } else { | |
7406 | kfree_skb(skb); | |
7407 | kfree(fpl); | |
7408 | } | |
6b06314c JA |
7409 | |
7410 | return 0; | |
7411 | } | |
7412 | ||
7413 | /* | |
7414 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7415 | * causes regular reference counting to break down. We rely on the UNIX | |
7416 | * garbage collection to take care of this problem for us. | |
7417 | */ | |
7418 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7419 | { | |
7420 | unsigned left, total; | |
7421 | int ret = 0; | |
7422 | ||
7423 | total = 0; | |
7424 | left = ctx->nr_user_files; | |
7425 | while (left) { | |
7426 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7427 | |
7428 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7429 | if (ret) | |
7430 | break; | |
7431 | left -= this_files; | |
7432 | total += this_files; | |
7433 | } | |
7434 | ||
7435 | if (!ret) | |
7436 | return 0; | |
7437 | ||
7438 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7439 | struct file *file = io_file_from_index(ctx, total); |
7440 | ||
7441 | if (file) | |
7442 | fput(file); | |
6b06314c JA |
7443 | total++; |
7444 | } | |
7445 | ||
7446 | return ret; | |
7447 | } | |
7448 | #else | |
7449 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7450 | { | |
7451 | return 0; | |
7452 | } | |
7453 | #endif | |
7454 | ||
b895c9a6 | 7455 | static int io_sqe_alloc_file_tables(struct io_rsrc_data *file_data, |
5398ae69 | 7456 | unsigned nr_tables, unsigned nr_files) |
65e19f54 JA |
7457 | { |
7458 | int i; | |
7459 | ||
7460 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7461 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7462 | unsigned this_files; |
7463 | ||
7464 | this_files = min(nr_files, IORING_MAX_FILES_TABLE); | |
7465 | table->files = kcalloc(this_files, sizeof(struct file *), | |
7466 | GFP_KERNEL); | |
7467 | if (!table->files) | |
7468 | break; | |
7469 | nr_files -= this_files; | |
7470 | } | |
7471 | ||
7472 | if (i == nr_tables) | |
7473 | return 0; | |
7474 | ||
7475 | for (i = 0; i < nr_tables; i++) { | |
269bbe5f | 7476 | struct fixed_rsrc_table *table = &file_data->table[i]; |
65e19f54 JA |
7477 | kfree(table->files); |
7478 | } | |
7479 | return 1; | |
7480 | } | |
7481 | ||
50238531 | 7482 | static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7483 | { |
50238531 | 7484 | struct file *file = prsrc->file; |
05f3fb3c JA |
7485 | #if defined(CONFIG_UNIX) |
7486 | struct sock *sock = ctx->ring_sock->sk; | |
7487 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7488 | struct sk_buff *skb; | |
7489 | int i; | |
7490 | ||
7491 | __skb_queue_head_init(&list); | |
7492 | ||
7493 | /* | |
7494 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7495 | * remove this entry and rearrange the file array. | |
7496 | */ | |
7497 | skb = skb_dequeue(head); | |
7498 | while (skb) { | |
7499 | struct scm_fp_list *fp; | |
7500 | ||
7501 | fp = UNIXCB(skb).fp; | |
7502 | for (i = 0; i < fp->count; i++) { | |
7503 | int left; | |
7504 | ||
7505 | if (fp->fp[i] != file) | |
7506 | continue; | |
7507 | ||
7508 | unix_notinflight(fp->user, fp->fp[i]); | |
7509 | left = fp->count - 1 - i; | |
7510 | if (left) { | |
7511 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7512 | left * sizeof(struct file *)); | |
7513 | } | |
7514 | fp->count--; | |
7515 | if (!fp->count) { | |
7516 | kfree_skb(skb); | |
7517 | skb = NULL; | |
7518 | } else { | |
7519 | __skb_queue_tail(&list, skb); | |
7520 | } | |
7521 | fput(file); | |
7522 | file = NULL; | |
7523 | break; | |
7524 | } | |
7525 | ||
7526 | if (!file) | |
7527 | break; | |
7528 | ||
7529 | __skb_queue_tail(&list, skb); | |
7530 | ||
7531 | skb = skb_dequeue(head); | |
7532 | } | |
7533 | ||
7534 | if (skb_peek(&list)) { | |
7535 | spin_lock_irq(&head->lock); | |
7536 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7537 | __skb_queue_tail(head, skb); | |
7538 | spin_unlock_irq(&head->lock); | |
7539 | } | |
7540 | #else | |
7541 | fput(file); | |
7542 | #endif | |
7543 | } | |
7544 | ||
b895c9a6 | 7545 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7546 | { |
b895c9a6 | 7547 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7548 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7549 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7550 | |
269bbe5f BM |
7551 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7552 | list_del(&prsrc->list); | |
50238531 | 7553 | ref_node->rsrc_put(ctx, prsrc); |
269bbe5f | 7554 | kfree(prsrc); |
65e19f54 | 7555 | } |
05589553 | 7556 | |
05589553 XW |
7557 | percpu_ref_exit(&ref_node->refs); |
7558 | kfree(ref_node); | |
269bbe5f | 7559 | percpu_ref_put(&rsrc_data->refs); |
2faf852d | 7560 | } |
65e19f54 | 7561 | |
269bbe5f | 7562 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7563 | { |
7564 | struct io_ring_ctx *ctx; | |
7565 | struct llist_node *node; | |
7566 | ||
269bbe5f BM |
7567 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7568 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7569 | |
7570 | while (node) { | |
b895c9a6 | 7571 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7572 | struct llist_node *next = node->next; |
7573 | ||
b895c9a6 | 7574 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7575 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7576 | node = next; |
7577 | } | |
7578 | } | |
7579 | ||
00835dce | 7580 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7581 | { |
b895c9a6 PB |
7582 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
7583 | struct io_rsrc_data *data = node->rsrc_data; | |
7584 | struct io_ring_ctx *ctx = data->ctx; | |
e297822b | 7585 | bool first_add = false; |
221aa924 | 7586 | int delay; |
65e19f54 | 7587 | |
2a63b2d9 | 7588 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7589 | node->done = true; |
e297822b | 7590 | |
d67d2263 | 7591 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7592 | node = list_first_entry(&ctx->rsrc_ref_list, |
7593 | struct io_rsrc_node, node); | |
e297822b | 7594 | /* recycle ref nodes in order */ |
b895c9a6 | 7595 | if (!node->done) |
e297822b | 7596 | break; |
b895c9a6 PB |
7597 | list_del(&node->node); |
7598 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7599 | } |
2a63b2d9 | 7600 | io_rsrc_ref_unlock(ctx); |
05589553 | 7601 | |
221aa924 PB |
7602 | delay = percpu_ref_is_dying(&data->refs) ? 0 : HZ; |
7603 | if (first_add || !delay) | |
7604 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, delay); | |
05f3fb3c | 7605 | } |
65e19f54 | 7606 | |
b895c9a6 | 7607 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7608 | { |
b895c9a6 | 7609 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7610 | |
05589553 XW |
7611 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7612 | if (!ref_node) | |
3e2224c5 | 7613 | return NULL; |
05f3fb3c | 7614 | |
00835dce | 7615 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7616 | 0, GFP_KERNEL)) { |
7617 | kfree(ref_node); | |
3e2224c5 | 7618 | return NULL; |
05589553 XW |
7619 | } |
7620 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7621 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7622 | ref_node->done = false; |
05589553 | 7623 | return ref_node; |
05589553 XW |
7624 | } |
7625 | ||
bc9744cd | 7626 | static void init_fixed_file_ref_node(struct io_ring_ctx *ctx, |
b895c9a6 | 7627 | struct io_rsrc_node *ref_node) |
6802535d | 7628 | { |
269bbe5f | 7629 | ref_node->rsrc_data = ctx->file_data; |
50238531 | 7630 | ref_node->rsrc_put = io_ring_file_put; |
05589553 XW |
7631 | } |
7632 | ||
b895c9a6 | 7633 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
05589553 XW |
7634 | { |
7635 | percpu_ref_exit(&ref_node->refs); | |
7636 | kfree(ref_node); | |
65e19f54 JA |
7637 | } |
7638 | ||
6b06314c JA |
7639 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7640 | unsigned nr_args) | |
7641 | { | |
7642 | __s32 __user *fds = (__s32 __user *) arg; | |
600cf3f8 | 7643 | unsigned nr_tables, i; |
05f3fb3c | 7644 | struct file *file; |
600cf3f8 | 7645 | int fd, ret = -ENOMEM; |
b895c9a6 PB |
7646 | struct io_rsrc_node *ref_node; |
7647 | struct io_rsrc_data *file_data; | |
6b06314c | 7648 | |
05f3fb3c | 7649 | if (ctx->file_data) |
6b06314c JA |
7650 | return -EBUSY; |
7651 | if (!nr_args) | |
7652 | return -EINVAL; | |
7653 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7654 | return -EMFILE; | |
7655 | ||
b895c9a6 | 7656 | file_data = io_rsrc_data_alloc(ctx); |
5398ae69 | 7657 | if (!file_data) |
05f3fb3c | 7658 | return -ENOMEM; |
13770a71 | 7659 | ctx->file_data = file_data; |
05f3fb3c | 7660 | |
65e19f54 | 7661 | nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE); |
035fbafc | 7662 | file_data->table = kcalloc(nr_tables, sizeof(*file_data->table), |
5398ae69 | 7663 | GFP_KERNEL); |
600cf3f8 PB |
7664 | if (!file_data->table) |
7665 | goto out_free; | |
05f3fb3c | 7666 | |
600cf3f8 | 7667 | if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args)) |
1ad555c6 | 7668 | goto out_free; |
65e19f54 | 7669 | |
08a45173 | 7670 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
7b29f92d JA |
7671 | unsigned long file_ptr; |
7672 | ||
600cf3f8 PB |
7673 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7674 | ret = -EFAULT; | |
7675 | goto out_fput; | |
7676 | } | |
08a45173 | 7677 | /* allow sparse sets */ |
600cf3f8 | 7678 | if (fd == -1) |
08a45173 | 7679 | continue; |
6b06314c | 7680 | |
05f3fb3c | 7681 | file = fget(fd); |
6b06314c | 7682 | ret = -EBADF; |
05f3fb3c | 7683 | if (!file) |
600cf3f8 | 7684 | goto out_fput; |
05f3fb3c | 7685 | |
6b06314c JA |
7686 | /* |
7687 | * Don't allow io_uring instances to be registered. If UNIX | |
7688 | * isn't enabled, then this causes a reference cycle and this | |
7689 | * instance can never get freed. If UNIX is enabled we'll | |
7690 | * handle it just fine, but there's still no point in allowing | |
7691 | * a ring fd as it doesn't support regular read/write anyway. | |
7692 | */ | |
05f3fb3c JA |
7693 | if (file->f_op == &io_uring_fops) { |
7694 | fput(file); | |
600cf3f8 | 7695 | goto out_fput; |
6b06314c | 7696 | } |
7b29f92d JA |
7697 | file_ptr = (unsigned long) file; |
7698 | if (__io_file_supports_async(file, READ)) | |
7699 | file_ptr |= FFS_ASYNC_READ; | |
7700 | if (__io_file_supports_async(file, WRITE)) | |
7701 | file_ptr |= FFS_ASYNC_WRITE; | |
7702 | if (S_ISREG(file_inode(file)->i_mode)) | |
7703 | file_ptr |= FFS_ISREG; | |
7704 | *io_fixed_file_slot(file_data, i) = (struct file *) file_ptr; | |
6b06314c JA |
7705 | } |
7706 | ||
6b06314c | 7707 | ret = io_sqe_files_scm(ctx); |
05589553 | 7708 | if (ret) { |
6b06314c | 7709 | io_sqe_files_unregister(ctx); |
05589553 XW |
7710 | return ret; |
7711 | } | |
6b06314c | 7712 | |
b895c9a6 | 7713 | ref_node = io_rsrc_node_alloc(ctx); |
3e2224c5 | 7714 | if (!ref_node) { |
05589553 | 7715 | io_sqe_files_unregister(ctx); |
3e2224c5 | 7716 | return -ENOMEM; |
05589553 | 7717 | } |
bc9744cd | 7718 | init_fixed_file_ref_node(ctx, ref_node); |
05589553 | 7719 | |
b895c9a6 | 7720 | io_rsrc_node_set(ctx, file_data, ref_node); |
6b06314c | 7721 | return ret; |
600cf3f8 PB |
7722 | out_fput: |
7723 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7724 | file = io_file_from_index(ctx, i); | |
7725 | if (file) | |
7726 | fput(file); | |
7727 | } | |
7728 | for (i = 0; i < nr_tables; i++) | |
7729 | kfree(file_data->table[i].files); | |
7730 | ctx->nr_user_files = 0; | |
600cf3f8 | 7731 | out_free: |
b895c9a6 | 7732 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7733 | ctx->file_data = NULL; |
6b06314c JA |
7734 | return ret; |
7735 | } | |
7736 | ||
c3a31e60 JA |
7737 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7738 | int index) | |
7739 | { | |
7740 | #if defined(CONFIG_UNIX) | |
7741 | struct sock *sock = ctx->ring_sock->sk; | |
7742 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7743 | struct sk_buff *skb; | |
7744 | ||
7745 | /* | |
7746 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7747 | * file set. If there's no room, fall back to allocating a new skb | |
7748 | * and filling it in. | |
7749 | */ | |
7750 | spin_lock_irq(&head->lock); | |
7751 | skb = skb_peek(head); | |
7752 | if (skb) { | |
7753 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7754 | ||
7755 | if (fpl->count < SCM_MAX_FD) { | |
7756 | __skb_unlink(skb, head); | |
7757 | spin_unlock_irq(&head->lock); | |
7758 | fpl->fp[fpl->count] = get_file(file); | |
7759 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7760 | fpl->count++; | |
7761 | spin_lock_irq(&head->lock); | |
7762 | __skb_queue_head(head, skb); | |
7763 | } else { | |
7764 | skb = NULL; | |
7765 | } | |
7766 | } | |
7767 | spin_unlock_irq(&head->lock); | |
7768 | ||
7769 | if (skb) { | |
7770 | fput(file); | |
7771 | return 0; | |
7772 | } | |
7773 | ||
7774 | return __io_sqe_files_scm(ctx, 1, index); | |
7775 | #else | |
7776 | return 0; | |
7777 | #endif | |
7778 | } | |
7779 | ||
b895c9a6 | 7780 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, void *rsrc) |
05f3fb3c | 7781 | { |
269bbe5f | 7782 | struct io_rsrc_put *prsrc; |
b895c9a6 | 7783 | struct io_rsrc_node *ref_node = data->node; |
05f3fb3c | 7784 | |
269bbe5f BM |
7785 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7786 | if (!prsrc) | |
a5318d3c | 7787 | return -ENOMEM; |
05f3fb3c | 7788 | |
50238531 | 7789 | prsrc->rsrc = rsrc; |
269bbe5f | 7790 | list_add(&prsrc->list, &ref_node->rsrc_list); |
05589553 | 7791 | |
a5318d3c | 7792 | return 0; |
05f3fb3c JA |
7793 | } |
7794 | ||
b895c9a6 | 7795 | static inline int io_queue_file_removal(struct io_rsrc_data *data, |
269bbe5f BM |
7796 | struct file *file) |
7797 | { | |
50238531 | 7798 | return io_queue_rsrc_removal(data, (void *)file); |
269bbe5f BM |
7799 | } |
7800 | ||
05f3fb3c | 7801 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 7802 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7803 | unsigned nr_args) |
7804 | { | |
b895c9a6 PB |
7805 | struct io_rsrc_data *data = ctx->file_data; |
7806 | struct io_rsrc_node *ref_node; | |
ea64ec02 | 7807 | struct file *file, **file_slot; |
c3a31e60 JA |
7808 | __s32 __user *fds; |
7809 | int fd, i, err; | |
7810 | __u32 done; | |
05589553 | 7811 | bool needs_switch = false; |
c3a31e60 | 7812 | |
05f3fb3c | 7813 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7814 | return -EOVERFLOW; |
7815 | if (done > ctx->nr_user_files) | |
7816 | return -EINVAL; | |
b895c9a6 | 7817 | err = io_rsrc_node_prealloc(ctx); |
8dd03afe PB |
7818 | if (err) |
7819 | return err; | |
05589553 | 7820 | |
269bbe5f | 7821 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7822 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7823 | err = 0; |
7824 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7825 | err = -EFAULT; | |
7826 | break; | |
7827 | } | |
4e0377a1 | 7828 | if (fd == IORING_REGISTER_FILES_SKIP) |
7829 | continue; | |
7830 | ||
67973b93 | 7831 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
ea64ec02 PB |
7832 | file_slot = io_fixed_file_slot(ctx->file_data, i); |
7833 | ||
7834 | if (*file_slot) { | |
7b29f92d JA |
7835 | file = (struct file *) ((unsigned long) *file_slot & FFS_MASK); |
7836 | err = io_queue_file_removal(data, file); | |
a5318d3c HD |
7837 | if (err) |
7838 | break; | |
ea64ec02 | 7839 | *file_slot = NULL; |
05589553 | 7840 | needs_switch = true; |
c3a31e60 JA |
7841 | } |
7842 | if (fd != -1) { | |
c3a31e60 JA |
7843 | file = fget(fd); |
7844 | if (!file) { | |
7845 | err = -EBADF; | |
7846 | break; | |
7847 | } | |
7848 | /* | |
7849 | * Don't allow io_uring instances to be registered. If | |
7850 | * UNIX isn't enabled, then this causes a reference | |
7851 | * cycle and this instance can never get freed. If UNIX | |
7852 | * is enabled we'll handle it just fine, but there's | |
7853 | * still no point in allowing a ring fd as it doesn't | |
7854 | * support regular read/write anyway. | |
7855 | */ | |
7856 | if (file->f_op == &io_uring_fops) { | |
7857 | fput(file); | |
7858 | err = -EBADF; | |
7859 | break; | |
7860 | } | |
e68a3ff8 | 7861 | *file_slot = file; |
c3a31e60 | 7862 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7863 | if (err) { |
e68a3ff8 | 7864 | *file_slot = NULL; |
f3bd9dae | 7865 | fput(file); |
c3a31e60 | 7866 | break; |
f3bd9dae | 7867 | } |
c3a31e60 | 7868 | } |
05f3fb3c JA |
7869 | } |
7870 | ||
05589553 | 7871 | if (needs_switch) { |
b2e96852 | 7872 | percpu_ref_kill(&data->node->refs); |
b895c9a6 PB |
7873 | ref_node = io_rsrc_node_get(ctx, data, io_ring_file_put); |
7874 | io_rsrc_node_set(ctx, data, ref_node); | |
8dd03afe | 7875 | } |
c3a31e60 JA |
7876 | return done ? done : err; |
7877 | } | |
05589553 | 7878 | |
05f3fb3c JA |
7879 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7880 | unsigned nr_args) | |
7881 | { | |
269bbe5f | 7882 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7883 | |
7884 | if (!ctx->file_data) | |
7885 | return -ENXIO; | |
7886 | if (!nr_args) | |
7887 | return -EINVAL; | |
7888 | if (copy_from_user(&up, arg, sizeof(up))) | |
7889 | return -EFAULT; | |
7890 | if (up.resv) | |
7891 | return -EINVAL; | |
7892 | ||
7893 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7894 | } | |
c3a31e60 | 7895 | |
5280f7e5 | 7896 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7897 | { |
7898 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7899 | ||
5280f7e5 PB |
7900 | req = io_put_req_find_next(req); |
7901 | return req ? &req->work : NULL; | |
7d723065 JA |
7902 | } |
7903 | ||
685fe7fe JA |
7904 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7905 | struct task_struct *task) | |
24369c2e | 7906 | { |
e941894e | 7907 | struct io_wq_hash *hash; |
24369c2e | 7908 | struct io_wq_data data; |
24369c2e | 7909 | unsigned int concurrency; |
24369c2e | 7910 | |
e941894e JA |
7911 | hash = ctx->hash_map; |
7912 | if (!hash) { | |
7913 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7914 | if (!hash) | |
7915 | return ERR_PTR(-ENOMEM); | |
7916 | refcount_set(&hash->refs, 1); | |
7917 | init_waitqueue_head(&hash->wait); | |
7918 | ctx->hash_map = hash; | |
24369c2e PB |
7919 | } |
7920 | ||
e941894e | 7921 | data.hash = hash; |
685fe7fe | 7922 | data.task = task; |
e9fd9396 | 7923 | data.free_work = io_free_work; |
f5fa38c5 | 7924 | data.do_work = io_wq_submit_work; |
24369c2e | 7925 | |
d25e3a3d JA |
7926 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7927 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7928 | |
5aa75ed5 | 7929 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7930 | } |
7931 | ||
5aa75ed5 JA |
7932 | static int io_uring_alloc_task_context(struct task_struct *task, |
7933 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7934 | { |
7935 | struct io_uring_task *tctx; | |
d8a6df10 | 7936 | int ret; |
0f212204 JA |
7937 | |
7938 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7939 | if (unlikely(!tctx)) | |
7940 | return -ENOMEM; | |
7941 | ||
d8a6df10 JA |
7942 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7943 | if (unlikely(ret)) { | |
7944 | kfree(tctx); | |
7945 | return ret; | |
7946 | } | |
7947 | ||
685fe7fe | 7948 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7949 | if (IS_ERR(tctx->io_wq)) { |
7950 | ret = PTR_ERR(tctx->io_wq); | |
7951 | percpu_counter_destroy(&tctx->inflight); | |
7952 | kfree(tctx); | |
7953 | return ret; | |
7954 | } | |
7955 | ||
0f212204 JA |
7956 | xa_init(&tctx->xa); |
7957 | init_waitqueue_head(&tctx->wait); | |
7958 | tctx->last = NULL; | |
fdaf083c | 7959 | atomic_set(&tctx->in_idle, 0); |
0f212204 | 7960 | task->io_uring = tctx; |
7cbf1722 JA |
7961 | spin_lock_init(&tctx->task_lock); |
7962 | INIT_WQ_LIST(&tctx->task_list); | |
7963 | tctx->task_state = 0; | |
7964 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7965 | return 0; |
7966 | } | |
7967 | ||
7968 | void __io_uring_free(struct task_struct *tsk) | |
7969 | { | |
7970 | struct io_uring_task *tctx = tsk->io_uring; | |
7971 | ||
7972 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7973 | WARN_ON_ONCE(tctx->io_wq); |
7974 | ||
d8a6df10 | 7975 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7976 | kfree(tctx); |
7977 | tsk->io_uring = NULL; | |
7978 | } | |
7979 | ||
7e84e1c7 SG |
7980 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7981 | struct io_uring_params *p) | |
2b188cc1 JA |
7982 | { |
7983 | int ret; | |
7984 | ||
d25e3a3d JA |
7985 | /* Retain compatibility with failing for an invalid attach attempt */ |
7986 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7987 | IORING_SETUP_ATTACH_WQ) { | |
7988 | struct fd f; | |
7989 | ||
7990 | f = fdget(p->wq_fd); | |
7991 | if (!f.file) | |
7992 | return -ENXIO; | |
7993 | if (f.file->f_op != &io_uring_fops) { | |
7994 | fdput(f); | |
7995 | return -EINVAL; | |
7996 | } | |
7997 | fdput(f); | |
7998 | } | |
6c271ce2 | 7999 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 8000 | struct task_struct *tsk; |
534ca6d6 | 8001 | struct io_sq_data *sqd; |
26984fbf | 8002 | bool attached; |
534ca6d6 | 8003 | |
26984fbf | 8004 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
8005 | if (IS_ERR(sqd)) { |
8006 | ret = PTR_ERR(sqd); | |
8007 | goto err; | |
8008 | } | |
69fb2131 | 8009 | |
7c30f36a | 8010 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 8011 | ctx->sq_data = sqd; |
917257da JA |
8012 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
8013 | if (!ctx->sq_thread_idle) | |
8014 | ctx->sq_thread_idle = HZ; | |
8015 | ||
26984fbf | 8016 | ret = 0; |
78d7f6ba | 8017 | io_sq_thread_park(sqd); |
de75a3d3 PB |
8018 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
8019 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 8020 | /* don't attach to a dying SQPOLL thread, would be racy */ |
de75a3d3 | 8021 | if (attached && !sqd->thread) |
26984fbf | 8022 | ret = -ENXIO; |
78d7f6ba PB |
8023 | io_sq_thread_unpark(sqd); |
8024 | ||
de75a3d3 PB |
8025 | if (ret < 0) |
8026 | goto err; | |
8027 | if (attached) | |
5aa75ed5 | 8028 | return 0; |
aa06165d | 8029 | |
6c271ce2 | 8030 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 8031 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 8032 | |
917257da | 8033 | ret = -EINVAL; |
44a9bd18 | 8034 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 8035 | goto err_sqpoll; |
7889f44d | 8036 | if (!cpu_online(cpu)) |
e8f98f24 | 8037 | goto err_sqpoll; |
917257da | 8038 | |
37d1e2e3 | 8039 | sqd->sq_cpu = cpu; |
6c271ce2 | 8040 | } else { |
37d1e2e3 | 8041 | sqd->sq_cpu = -1; |
6c271ce2 | 8042 | } |
37d1e2e3 JA |
8043 | |
8044 | sqd->task_pid = current->pid; | |
5c2469e0 | 8045 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
8046 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
8047 | if (IS_ERR(tsk)) { | |
8048 | ret = PTR_ERR(tsk); | |
e8f98f24 | 8049 | goto err_sqpoll; |
6c271ce2 | 8050 | } |
97a73a0f | 8051 | |
46fe18b1 | 8052 | sqd->thread = tsk; |
97a73a0f | 8053 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 8054 | wake_up_new_task(tsk); |
0f212204 JA |
8055 | if (ret) |
8056 | goto err; | |
6c271ce2 JA |
8057 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
8058 | /* Can't have SQ_AFF without SQPOLL */ | |
8059 | ret = -EINVAL; | |
8060 | goto err; | |
8061 | } | |
8062 | ||
2b188cc1 JA |
8063 | return 0; |
8064 | err: | |
37d1e2e3 | 8065 | io_sq_thread_finish(ctx); |
2b188cc1 | 8066 | return ret; |
e8f98f24 JA |
8067 | err_sqpoll: |
8068 | complete(&ctx->sq_data->exited); | |
8069 | goto err; | |
2b188cc1 JA |
8070 | } |
8071 | ||
a087e2b5 BM |
8072 | static inline void __io_unaccount_mem(struct user_struct *user, |
8073 | unsigned long nr_pages) | |
2b188cc1 JA |
8074 | { |
8075 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8076 | } | |
8077 | ||
a087e2b5 BM |
8078 | static inline int __io_account_mem(struct user_struct *user, |
8079 | unsigned long nr_pages) | |
2b188cc1 JA |
8080 | { |
8081 | unsigned long page_limit, cur_pages, new_pages; | |
8082 | ||
8083 | /* Don't allow more pages than we can safely lock */ | |
8084 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8085 | ||
8086 | do { | |
8087 | cur_pages = atomic_long_read(&user->locked_vm); | |
8088 | new_pages = cur_pages + nr_pages; | |
8089 | if (new_pages > page_limit) | |
8090 | return -ENOMEM; | |
8091 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8092 | new_pages) != cur_pages); | |
8093 | ||
8094 | return 0; | |
8095 | } | |
8096 | ||
26bfa89e | 8097 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8098 | { |
62e398be | 8099 | if (ctx->user) |
a087e2b5 | 8100 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8101 | |
26bfa89e JA |
8102 | if (ctx->mm_account) |
8103 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8104 | } |
8105 | ||
26bfa89e | 8106 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8107 | { |
30975825 BM |
8108 | int ret; |
8109 | ||
62e398be | 8110 | if (ctx->user) { |
30975825 BM |
8111 | ret = __io_account_mem(ctx->user, nr_pages); |
8112 | if (ret) | |
8113 | return ret; | |
8114 | } | |
8115 | ||
26bfa89e JA |
8116 | if (ctx->mm_account) |
8117 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8118 | |
8119 | return 0; | |
8120 | } | |
8121 | ||
2b188cc1 JA |
8122 | static void io_mem_free(void *ptr) |
8123 | { | |
52e04ef4 MR |
8124 | struct page *page; |
8125 | ||
8126 | if (!ptr) | |
8127 | return; | |
2b188cc1 | 8128 | |
52e04ef4 | 8129 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8130 | if (put_page_testzero(page)) |
8131 | free_compound_page(page); | |
8132 | } | |
8133 | ||
8134 | static void *io_mem_alloc(size_t size) | |
8135 | { | |
8136 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8137 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8138 | |
8139 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8140 | } | |
8141 | ||
75b28aff HV |
8142 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8143 | size_t *sq_offset) | |
8144 | { | |
8145 | struct io_rings *rings; | |
8146 | size_t off, sq_array_size; | |
8147 | ||
8148 | off = struct_size(rings, cqes, cq_entries); | |
8149 | if (off == SIZE_MAX) | |
8150 | return SIZE_MAX; | |
8151 | ||
8152 | #ifdef CONFIG_SMP | |
8153 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8154 | if (off == 0) | |
8155 | return SIZE_MAX; | |
8156 | #endif | |
8157 | ||
b36200f5 DV |
8158 | if (sq_offset) |
8159 | *sq_offset = off; | |
8160 | ||
75b28aff HV |
8161 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8162 | if (sq_array_size == SIZE_MAX) | |
8163 | return SIZE_MAX; | |
8164 | ||
8165 | if (check_add_overflow(off, sq_array_size, &off)) | |
8166 | return SIZE_MAX; | |
8167 | ||
75b28aff HV |
8168 | return off; |
8169 | } | |
8170 | ||
0a96bbe4 | 8171 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee JA |
8172 | { |
8173 | int i, j; | |
8174 | ||
8175 | if (!ctx->user_bufs) | |
8176 | return -ENXIO; | |
8177 | ||
8178 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8179 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8180 | ||
8181 | for (j = 0; j < imu->nr_bvecs; j++) | |
f1f6a7dd | 8182 | unpin_user_page(imu->bvec[j].bv_page); |
edafccee | 8183 | |
de293938 | 8184 | if (imu->acct_pages) |
26bfa89e | 8185 | io_unaccount_mem(ctx, imu->acct_pages); |
d4ef6475 | 8186 | kvfree(imu->bvec); |
edafccee JA |
8187 | imu->nr_bvecs = 0; |
8188 | } | |
8189 | ||
8190 | kfree(ctx->user_bufs); | |
8191 | ctx->user_bufs = NULL; | |
8192 | ctx->nr_user_bufs = 0; | |
8193 | return 0; | |
8194 | } | |
8195 | ||
8196 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8197 | void __user *arg, unsigned index) | |
8198 | { | |
8199 | struct iovec __user *src; | |
8200 | ||
8201 | #ifdef CONFIG_COMPAT | |
8202 | if (ctx->compat) { | |
8203 | struct compat_iovec __user *ciovs; | |
8204 | struct compat_iovec ciov; | |
8205 | ||
8206 | ciovs = (struct compat_iovec __user *) arg; | |
8207 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8208 | return -EFAULT; | |
8209 | ||
d55e5f5b | 8210 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8211 | dst->iov_len = ciov.iov_len; |
8212 | return 0; | |
8213 | } | |
8214 | #endif | |
8215 | src = (struct iovec __user *) arg; | |
8216 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8217 | return -EFAULT; | |
8218 | return 0; | |
8219 | } | |
8220 | ||
de293938 JA |
8221 | /* |
8222 | * Not super efficient, but this is just a registration time. And we do cache | |
8223 | * the last compound head, so generally we'll only do a full search if we don't | |
8224 | * match that one. | |
8225 | * | |
8226 | * We check if the given compound head page has already been accounted, to | |
8227 | * avoid double accounting it. This allows us to account the full size of the | |
8228 | * page, not just the constituent pages of a huge page. | |
8229 | */ | |
8230 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8231 | int nr_pages, struct page *hpage) | |
8232 | { | |
8233 | int i, j; | |
8234 | ||
8235 | /* check current page array */ | |
8236 | for (i = 0; i < nr_pages; i++) { | |
8237 | if (!PageCompound(pages[i])) | |
8238 | continue; | |
8239 | if (compound_head(pages[i]) == hpage) | |
8240 | return true; | |
8241 | } | |
8242 | ||
8243 | /* check previously registered pages */ | |
8244 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8245 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8246 | ||
8247 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8248 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8249 | continue; | |
8250 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8251 | return true; | |
8252 | } | |
8253 | } | |
8254 | ||
8255 | return false; | |
8256 | } | |
8257 | ||
8258 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8259 | int nr_pages, struct io_mapped_ubuf *imu, | |
8260 | struct page **last_hpage) | |
8261 | { | |
8262 | int i, ret; | |
8263 | ||
8264 | for (i = 0; i < nr_pages; i++) { | |
8265 | if (!PageCompound(pages[i])) { | |
8266 | imu->acct_pages++; | |
8267 | } else { | |
8268 | struct page *hpage; | |
8269 | ||
8270 | hpage = compound_head(pages[i]); | |
8271 | if (hpage == *last_hpage) | |
8272 | continue; | |
8273 | *last_hpage = hpage; | |
8274 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8275 | continue; | |
8276 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8277 | } | |
8278 | } | |
8279 | ||
8280 | if (!imu->acct_pages) | |
8281 | return 0; | |
8282 | ||
26bfa89e | 8283 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8284 | if (ret) |
8285 | imu->acct_pages = 0; | |
8286 | return ret; | |
8287 | } | |
8288 | ||
0a96bbe4 BM |
8289 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8290 | struct io_mapped_ubuf *imu, | |
8291 | struct page **last_hpage) | |
edafccee JA |
8292 | { |
8293 | struct vm_area_struct **vmas = NULL; | |
8294 | struct page **pages = NULL; | |
0a96bbe4 BM |
8295 | unsigned long off, start, end, ubuf; |
8296 | size_t size; | |
8297 | int ret, pret, nr_pages, i; | |
8298 | ||
8299 | ubuf = (unsigned long) iov->iov_base; | |
8300 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8301 | start = ubuf >> PAGE_SHIFT; | |
8302 | nr_pages = end - start; | |
8303 | ||
8304 | ret = -ENOMEM; | |
8305 | ||
8306 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8307 | if (!pages) | |
8308 | goto done; | |
8309 | ||
8310 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8311 | GFP_KERNEL); | |
8312 | if (!vmas) | |
8313 | goto done; | |
edafccee | 8314 | |
0a96bbe4 BM |
8315 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8316 | GFP_KERNEL); | |
8317 | if (!imu->bvec) | |
8318 | goto done; | |
8319 | ||
8320 | ret = 0; | |
8321 | mmap_read_lock(current->mm); | |
8322 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8323 | pages, vmas); | |
8324 | if (pret == nr_pages) { | |
8325 | /* don't support file backed memory */ | |
8326 | for (i = 0; i < nr_pages; i++) { | |
8327 | struct vm_area_struct *vma = vmas[i]; | |
8328 | ||
8329 | if (vma->vm_file && | |
8330 | !is_file_hugepages(vma->vm_file)) { | |
8331 | ret = -EOPNOTSUPP; | |
8332 | break; | |
8333 | } | |
8334 | } | |
8335 | } else { | |
8336 | ret = pret < 0 ? pret : -EFAULT; | |
8337 | } | |
8338 | mmap_read_unlock(current->mm); | |
8339 | if (ret) { | |
8340 | /* | |
8341 | * if we did partial map, or found file backed vmas, | |
8342 | * release any pages we did get | |
8343 | */ | |
8344 | if (pret > 0) | |
8345 | unpin_user_pages(pages, pret); | |
8346 | kvfree(imu->bvec); | |
8347 | goto done; | |
8348 | } | |
8349 | ||
8350 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8351 | if (ret) { | |
8352 | unpin_user_pages(pages, pret); | |
8353 | kvfree(imu->bvec); | |
8354 | goto done; | |
8355 | } | |
8356 | ||
8357 | off = ubuf & ~PAGE_MASK; | |
8358 | size = iov->iov_len; | |
8359 | for (i = 0; i < nr_pages; i++) { | |
8360 | size_t vec_len; | |
8361 | ||
8362 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8363 | imu->bvec[i].bv_page = pages[i]; | |
8364 | imu->bvec[i].bv_len = vec_len; | |
8365 | imu->bvec[i].bv_offset = off; | |
8366 | off = 0; | |
8367 | size -= vec_len; | |
8368 | } | |
8369 | /* store original address for later verification */ | |
8370 | imu->ubuf = ubuf; | |
8371 | imu->len = iov->iov_len; | |
8372 | imu->nr_bvecs = nr_pages; | |
8373 | ret = 0; | |
8374 | done: | |
8375 | kvfree(pages); | |
8376 | kvfree(vmas); | |
8377 | return ret; | |
8378 | } | |
8379 | ||
2b358604 | 8380 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8381 | { |
edafccee JA |
8382 | if (ctx->user_bufs) |
8383 | return -EBUSY; | |
8384 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8385 | return -EINVAL; | |
8386 | ||
8387 | ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf), | |
8388 | GFP_KERNEL); | |
8389 | if (!ctx->user_bufs) | |
8390 | return -ENOMEM; | |
8391 | ||
2b358604 BM |
8392 | return 0; |
8393 | } | |
edafccee | 8394 | |
2b358604 BM |
8395 | static int io_buffer_validate(struct iovec *iov) |
8396 | { | |
50e96989 PB |
8397 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8398 | ||
2b358604 BM |
8399 | /* |
8400 | * Don't impose further limits on the size and buffer | |
8401 | * constraints here, we'll -EINVAL later when IO is | |
8402 | * submitted if they are wrong. | |
8403 | */ | |
8404 | if (!iov->iov_base || !iov->iov_len) | |
8405 | return -EFAULT; | |
edafccee | 8406 | |
2b358604 BM |
8407 | /* arbitrary limit, but we need something */ |
8408 | if (iov->iov_len > SZ_1G) | |
8409 | return -EFAULT; | |
edafccee | 8410 | |
50e96989 PB |
8411 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8412 | return -EOVERFLOW; | |
8413 | ||
2b358604 BM |
8414 | return 0; |
8415 | } | |
edafccee | 8416 | |
2b358604 BM |
8417 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8418 | unsigned int nr_args) | |
8419 | { | |
8420 | int i, ret; | |
8421 | struct iovec iov; | |
8422 | struct page *last_hpage = NULL; | |
edafccee | 8423 | |
2b358604 BM |
8424 | ret = io_buffers_map_alloc(ctx, nr_args); |
8425 | if (ret) | |
8426 | return ret; | |
edafccee | 8427 | |
edafccee JA |
8428 | for (i = 0; i < nr_args; i++) { |
8429 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
edafccee | 8430 | |
edafccee JA |
8431 | ret = io_copy_iov(ctx, &iov, arg, i); |
8432 | if (ret) | |
0a96bbe4 | 8433 | break; |
de293938 | 8434 | |
2b358604 BM |
8435 | ret = io_buffer_validate(&iov); |
8436 | if (ret) | |
0a96bbe4 | 8437 | break; |
edafccee | 8438 | |
0a96bbe4 BM |
8439 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8440 | if (ret) | |
8441 | break; | |
edafccee JA |
8442 | |
8443 | ctx->nr_user_bufs++; | |
8444 | } | |
0a96bbe4 BM |
8445 | |
8446 | if (ret) | |
8447 | io_sqe_buffers_unregister(ctx); | |
8448 | ||
edafccee JA |
8449 | return ret; |
8450 | } | |
8451 | ||
9b402849 JA |
8452 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8453 | { | |
8454 | __s32 __user *fds = arg; | |
8455 | int fd; | |
8456 | ||
8457 | if (ctx->cq_ev_fd) | |
8458 | return -EBUSY; | |
8459 | ||
8460 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8461 | return -EFAULT; | |
8462 | ||
8463 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8464 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8465 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8466 | ctx->cq_ev_fd = NULL; | |
8467 | return ret; | |
8468 | } | |
8469 | ||
8470 | return 0; | |
8471 | } | |
8472 | ||
8473 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8474 | { | |
8475 | if (ctx->cq_ev_fd) { | |
8476 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8477 | ctx->cq_ev_fd = NULL; | |
8478 | return 0; | |
8479 | } | |
8480 | ||
8481 | return -ENXIO; | |
8482 | } | |
8483 | ||
5a2e745d JA |
8484 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8485 | { | |
9e15c3a0 JA |
8486 | struct io_buffer *buf; |
8487 | unsigned long index; | |
8488 | ||
8489 | xa_for_each(&ctx->io_buffers, index, buf) | |
8490 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8491 | } |
8492 | ||
68e68ee6 | 8493 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8494 | { |
68e68ee6 | 8495 | struct io_kiocb *req, *nxt; |
1b4c351f | 8496 | |
68e68ee6 JA |
8497 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8498 | if (tsk && req->task != tsk) | |
8499 | continue; | |
1b4c351f JA |
8500 | list_del(&req->compl.list); |
8501 | kmem_cache_free(req_cachep, req); | |
8502 | } | |
8503 | } | |
8504 | ||
4010fec4 | 8505 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8506 | { |
bf019da7 | 8507 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8508 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8509 | |
9a4fdbd8 JA |
8510 | mutex_lock(&ctx->uring_lock); |
8511 | ||
8e5c66c4 | 8512 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8513 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8514 | submit_state->reqs); | |
8e5c66c4 PB |
8515 | submit_state->free_reqs = 0; |
8516 | } | |
9a4fdbd8 | 8517 | |
dac7a098 | 8518 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8519 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8520 | mutex_unlock(&ctx->uring_lock); |
8521 | } | |
8522 | ||
2b188cc1 JA |
8523 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8524 | { | |
04fc6c80 PB |
8525 | /* |
8526 | * Some may use context even when all refs and requests have been put, | |
180f829f PB |
8527 | * and they are free to do so while still holding uring_lock or |
8528 | * completion_lock, see __io_req_task_submit(). Wait for them to finish. | |
04fc6c80 PB |
8529 | */ |
8530 | mutex_lock(&ctx->uring_lock); | |
8531 | mutex_unlock(&ctx->uring_lock); | |
180f829f PB |
8532 | spin_lock_irq(&ctx->completion_lock); |
8533 | spin_unlock_irq(&ctx->completion_lock); | |
04fc6c80 | 8534 | |
37d1e2e3 | 8535 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8536 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8537 | |
37d1e2e3 | 8538 | if (ctx->mm_account) { |
2aede0e4 JA |
8539 | mmdrop(ctx->mm_account); |
8540 | ctx->mm_account = NULL; | |
30975825 | 8541 | } |
def596e9 | 8542 | |
8bad28d8 | 8543 | mutex_lock(&ctx->uring_lock); |
6b06314c | 8544 | io_sqe_files_unregister(ctx); |
8bad28d8 | 8545 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8546 | io_eventfd_unregister(ctx); |
5a2e745d | 8547 | io_destroy_buffers(ctx); |
def596e9 | 8548 | |
8dd03afe | 8549 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8550 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
8dd03afe | 8551 | |
2b188cc1 | 8552 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8553 | if (ctx->ring_sock) { |
8554 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8555 | sock_release(ctx->ring_sock); |
355e8d26 | 8556 | } |
2b188cc1 JA |
8557 | #endif |
8558 | ||
75b28aff | 8559 | io_mem_free(ctx->rings); |
2b188cc1 | 8560 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8561 | |
8562 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8563 | free_uid(ctx->user); |
4010fec4 | 8564 | io_req_caches_free(ctx); |
e941894e JA |
8565 | if (ctx->hash_map) |
8566 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8567 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8568 | kfree(ctx); |
8569 | } | |
8570 | ||
8571 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8572 | { | |
8573 | struct io_ring_ctx *ctx = file->private_data; | |
8574 | __poll_t mask = 0; | |
8575 | ||
8576 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8577 | /* |
8578 | * synchronizes with barrier from wq_has_sleeper call in | |
8579 | * io_commit_cqring | |
8580 | */ | |
2b188cc1 | 8581 | smp_rmb(); |
90554200 | 8582 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8583 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8584 | |
8585 | /* | |
8586 | * Don't flush cqring overflow list here, just do a simple check. | |
8587 | * Otherwise there could possible be ABBA deadlock: | |
8588 | * CPU0 CPU1 | |
8589 | * ---- ---- | |
8590 | * lock(&ctx->uring_lock); | |
8591 | * lock(&ep->mtx); | |
8592 | * lock(&ctx->uring_lock); | |
8593 | * lock(&ep->mtx); | |
8594 | * | |
8595 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8596 | * pushs them to do the flush. | |
8597 | */ | |
8598 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8599 | mask |= EPOLLIN | EPOLLRDNORM; |
8600 | ||
8601 | return mask; | |
8602 | } | |
8603 | ||
8604 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8605 | { | |
8606 | struct io_ring_ctx *ctx = file->private_data; | |
8607 | ||
8608 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8609 | } | |
8610 | ||
0bead8cd | 8611 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8612 | { |
4379bf8b | 8613 | const struct cred *creds; |
071698e1 | 8614 | |
61cf9370 | 8615 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8616 | if (creds) { |
8617 | put_cred(creds); | |
0bead8cd | 8618 | return 0; |
1e6fa521 | 8619 | } |
0bead8cd YD |
8620 | |
8621 | return -EINVAL; | |
8622 | } | |
8623 | ||
9b465711 | 8624 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8625 | { |
9b465711 | 8626 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8627 | } |
8628 | ||
d56d938b PB |
8629 | struct io_tctx_exit { |
8630 | struct callback_head task_work; | |
8631 | struct completion completion; | |
baf186c4 | 8632 | struct io_ring_ctx *ctx; |
d56d938b PB |
8633 | }; |
8634 | ||
8635 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8636 | { | |
8637 | struct io_uring_task *tctx = current->io_uring; | |
8638 | struct io_tctx_exit *work; | |
8639 | ||
8640 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8641 | /* | |
8642 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8643 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8644 | */ | |
8645 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8646 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8647 | complete(&work->completion); |
8648 | } | |
8649 | ||
85faa7b8 JA |
8650 | static void io_ring_exit_work(struct work_struct *work) |
8651 | { | |
d56d938b | 8652 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8653 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8654 | struct io_tctx_exit exit; |
8655 | struct io_tctx_node *node; | |
8656 | int ret; | |
85faa7b8 | 8657 | |
a185f1db PB |
8658 | /* prevent SQPOLL from submitting new requests */ |
8659 | if (ctx->sq_data) { | |
8660 | io_sq_thread_park(ctx->sq_data); | |
8661 | list_del_init(&ctx->sqd_list); | |
8662 | io_sqd_update_thread_idle(ctx->sq_data); | |
8663 | io_sq_thread_unpark(ctx->sq_data); | |
8664 | } | |
8665 | ||
56952e91 JA |
8666 | /* |
8667 | * If we're doing polled IO and end up having requests being | |
8668 | * submitted async (out-of-line), then completions can come in while | |
8669 | * we're waiting for refs to drop. We need to reap these manually, | |
8670 | * as nobody else will be looking for them. | |
8671 | */ | |
b2edc0a7 | 8672 | do { |
9936c7c2 | 8673 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8674 | |
8675 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8676 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b PB |
8677 | |
8678 | mutex_lock(&ctx->uring_lock); | |
8679 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8680 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8681 | ||
d56d938b PB |
8682 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8683 | ctx_node); | |
baf186c4 | 8684 | exit.ctx = ctx; |
d56d938b PB |
8685 | init_completion(&exit.completion); |
8686 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8687 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8688 | if (WARN_ON_ONCE(ret)) | |
8689 | continue; | |
8690 | wake_up_process(node->task); | |
8691 | ||
8692 | mutex_unlock(&ctx->uring_lock); | |
8693 | wait_for_completion(&exit.completion); | |
8694 | cond_resched(); | |
8695 | mutex_lock(&ctx->uring_lock); | |
8696 | } | |
8697 | mutex_unlock(&ctx->uring_lock); | |
8698 | ||
85faa7b8 JA |
8699 | io_ring_ctx_free(ctx); |
8700 | } | |
8701 | ||
80c4cbdb PB |
8702 | /* Returns true if we found and killed one or more timeouts */ |
8703 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8704 | struct files_struct *files) | |
8705 | { | |
8706 | struct io_kiocb *req, *tmp; | |
8707 | int canceled = 0; | |
8708 | ||
8709 | spin_lock_irq(&ctx->completion_lock); | |
8710 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8711 | if (io_match_task(req, tsk, files)) { | |
8712 | io_kill_timeout(req, -ECANCELED); | |
8713 | canceled++; | |
8714 | } | |
8715 | } | |
51520426 PB |
8716 | if (canceled != 0) |
8717 | io_commit_cqring(ctx); | |
80c4cbdb | 8718 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8719 | if (canceled != 0) |
8720 | io_cqring_ev_posted(ctx); | |
8721 | return canceled != 0; | |
8722 | } | |
8723 | ||
2b188cc1 JA |
8724 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8725 | { | |
61cf9370 MWO |
8726 | unsigned long index; |
8727 | struct creds *creds; | |
8728 | ||
2b188cc1 JA |
8729 | mutex_lock(&ctx->uring_lock); |
8730 | percpu_ref_kill(&ctx->refs); | |
cda286f0 PB |
8731 | /* if force is set, the ring is going away. always drop after that */ |
8732 | ctx->cq_overflow_flushed = 1; | |
634578f8 | 8733 | if (ctx->rings) |
6c2450ae | 8734 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8735 | xa_for_each(&ctx->personalities, index, creds) |
8736 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8737 | mutex_unlock(&ctx->uring_lock); |
8738 | ||
6b81928d PB |
8739 | io_kill_timeouts(ctx, NULL, NULL); |
8740 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8741 | |
15dff286 | 8742 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8743 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8744 | |
85faa7b8 | 8745 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8746 | /* |
8747 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8748 | * if we're exiting a ton of rings at the same time. It just adds | |
8749 | * noise and overhead, there's no discernable change in runtime | |
8750 | * over using system_wq. | |
8751 | */ | |
8752 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8753 | } |
8754 | ||
8755 | static int io_uring_release(struct inode *inode, struct file *file) | |
8756 | { | |
8757 | struct io_ring_ctx *ctx = file->private_data; | |
8758 | ||
8759 | file->private_data = NULL; | |
8760 | io_ring_ctx_wait_and_kill(ctx); | |
8761 | return 0; | |
8762 | } | |
8763 | ||
f6edbabb PB |
8764 | struct io_task_cancel { |
8765 | struct task_struct *task; | |
8766 | struct files_struct *files; | |
8767 | }; | |
f254ac04 | 8768 | |
f6edbabb | 8769 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8770 | { |
9a472ef7 | 8771 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8772 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8773 | bool ret; |
8774 | ||
f6edbabb | 8775 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8776 | unsigned long flags; |
8777 | struct io_ring_ctx *ctx = req->ctx; | |
8778 | ||
8779 | /* protect against races with linked timeouts */ | |
8780 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8781 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8782 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8783 | } else { | |
f6edbabb | 8784 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8785 | } |
8786 | return ret; | |
b711d4ea JA |
8787 | } |
8788 | ||
e1915f76 | 8789 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8790 | struct task_struct *task, |
b7ddce3c PB |
8791 | struct files_struct *files) |
8792 | { | |
e1915f76 | 8793 | struct io_defer_entry *de; |
b7ddce3c PB |
8794 | LIST_HEAD(list); |
8795 | ||
8796 | spin_lock_irq(&ctx->completion_lock); | |
8797 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8798 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8799 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8800 | break; | |
8801 | } | |
8802 | } | |
8803 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8804 | if (list_empty(&list)) |
8805 | return false; | |
b7ddce3c PB |
8806 | |
8807 | while (!list_empty(&list)) { | |
8808 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8809 | list_del_init(&de->list); | |
f41db273 | 8810 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8811 | kfree(de); |
8812 | } | |
e1915f76 | 8813 | return true; |
b7ddce3c PB |
8814 | } |
8815 | ||
1b00764f PB |
8816 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8817 | { | |
8818 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8819 | ||
8820 | return req->ctx == data; | |
8821 | } | |
8822 | ||
8823 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8824 | { | |
8825 | struct io_tctx_node *node; | |
8826 | enum io_wq_cancel cret; | |
8827 | bool ret = false; | |
8828 | ||
8829 | mutex_lock(&ctx->uring_lock); | |
8830 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8831 | struct io_uring_task *tctx = node->task->io_uring; | |
8832 | ||
8833 | /* | |
8834 | * io_wq will stay alive while we hold uring_lock, because it's | |
8835 | * killed after ctx nodes, which requires to take the lock. | |
8836 | */ | |
8837 | if (!tctx || !tctx->io_wq) | |
8838 | continue; | |
8839 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8840 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8841 | } | |
8842 | mutex_unlock(&ctx->uring_lock); | |
8843 | ||
8844 | return ret; | |
8845 | } | |
8846 | ||
9936c7c2 PB |
8847 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8848 | struct task_struct *task, | |
8849 | struct files_struct *files) | |
8850 | { | |
8851 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8852 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8853 | |
8854 | while (1) { | |
8855 | enum io_wq_cancel cret; | |
8856 | bool ret = false; | |
8857 | ||
1b00764f PB |
8858 | if (!task) { |
8859 | ret |= io_uring_try_cancel_iowq(ctx); | |
8860 | } else if (tctx && tctx->io_wq) { | |
8861 | /* | |
8862 | * Cancels requests of all rings, not only @ctx, but | |
8863 | * it's fine as the task is in exit/exec. | |
8864 | */ | |
5aa75ed5 | 8865 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8866 | &cancel, true); |
8867 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8868 | } | |
8869 | ||
8870 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8871 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8872 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8873 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8874 | io_iopoll_try_reap_events(ctx); | |
8875 | ret = true; | |
8876 | } | |
8877 | } | |
8878 | ||
e1915f76 | 8879 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8880 | ret |= io_poll_remove_all(ctx, task, files); |
8881 | ret |= io_kill_timeouts(ctx, task, files); | |
8882 | ret |= io_run_task_work(); | |
ba50a036 | 8883 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8884 | if (!ret) |
8885 | break; | |
8886 | cond_resched(); | |
8887 | } | |
8888 | } | |
8889 | ||
ca70f00b PB |
8890 | static int io_uring_count_inflight(struct io_ring_ctx *ctx, |
8891 | struct task_struct *task, | |
8892 | struct files_struct *files) | |
8893 | { | |
8894 | struct io_kiocb *req; | |
8895 | int cnt = 0; | |
8896 | ||
8897 | spin_lock_irq(&ctx->inflight_lock); | |
8898 | list_for_each_entry(req, &ctx->inflight_list, inflight_entry) | |
8899 | cnt += io_match_task(req, task, files); | |
8900 | spin_unlock_irq(&ctx->inflight_lock); | |
8901 | return cnt; | |
8902 | } | |
8903 | ||
b52fda00 | 8904 | static void io_uring_cancel_files(struct io_ring_ctx *ctx, |
df9923f9 | 8905 | struct task_struct *task, |
fcb323cc JA |
8906 | struct files_struct *files) |
8907 | { | |
fcb323cc | 8908 | while (!list_empty_careful(&ctx->inflight_list)) { |
d8f1b971 | 8909 | DEFINE_WAIT(wait); |
ca70f00b | 8910 | int inflight; |
fcb323cc | 8911 | |
ca70f00b PB |
8912 | inflight = io_uring_count_inflight(ctx, task, files); |
8913 | if (!inflight) | |
fcb323cc | 8914 | break; |
f6edbabb | 8915 | |
9936c7c2 | 8916 | io_uring_try_cancel_requests(ctx, task, files); |
ca70f00b PB |
8917 | |
8918 | prepare_to_wait(&task->io_uring->wait, &wait, | |
8919 | TASK_UNINTERRUPTIBLE); | |
8920 | if (inflight == io_uring_count_inflight(ctx, task, files)) | |
8921 | schedule(); | |
c98de08c | 8922 | finish_wait(&task->io_uring->wait, &wait); |
0f212204 | 8923 | } |
0f212204 JA |
8924 | } |
8925 | ||
cf27f3b1 | 8926 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8927 | { |
236434c3 | 8928 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8929 | struct io_tctx_node *node; |
a528b04e | 8930 | int ret; |
236434c3 MWO |
8931 | |
8932 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8933 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8934 | if (unlikely(ret)) |
8935 | return ret; | |
236434c3 | 8936 | tctx = current->io_uring; |
0f212204 | 8937 | } |
cf27f3b1 PB |
8938 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8939 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8940 | if (!node) | |
8941 | return -ENOMEM; | |
8942 | node->ctx = ctx; | |
8943 | node->task = current; | |
13bf43f5 | 8944 | |
cf27f3b1 PB |
8945 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8946 | node, GFP_KERNEL)); | |
8947 | if (ret) { | |
8948 | kfree(node); | |
8949 | return ret; | |
0f212204 | 8950 | } |
cf27f3b1 PB |
8951 | |
8952 | mutex_lock(&ctx->uring_lock); | |
8953 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8954 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8955 | } |
cf27f3b1 | 8956 | tctx->last = ctx; |
0f212204 JA |
8957 | return 0; |
8958 | } | |
8959 | ||
cf27f3b1 PB |
8960 | /* |
8961 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8962 | */ | |
8963 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
8964 | { | |
8965 | struct io_uring_task *tctx = current->io_uring; | |
8966 | ||
8967 | if (likely(tctx && tctx->last == ctx)) | |
8968 | return 0; | |
8969 | return __io_uring_add_task_file(ctx); | |
8970 | } | |
8971 | ||
0f212204 JA |
8972 | /* |
8973 | * Remove this io_uring_file -> task mapping. | |
8974 | */ | |
2941267b | 8975 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8976 | { |
8977 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8978 | struct io_tctx_node *node; |
2941267b | 8979 | |
eebd2e37 PB |
8980 | if (!tctx) |
8981 | return; | |
13bf43f5 PB |
8982 | node = xa_erase(&tctx->xa, index); |
8983 | if (!node) | |
2941267b | 8984 | return; |
0f212204 | 8985 | |
13bf43f5 PB |
8986 | WARN_ON_ONCE(current != node->task); |
8987 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8988 | ||
8989 | mutex_lock(&node->ctx->uring_lock); | |
8990 | list_del(&node->ctx_node); | |
8991 | mutex_unlock(&node->ctx->uring_lock); | |
8992 | ||
baf186c4 | 8993 | if (tctx->last == node->ctx) |
0f212204 | 8994 | tctx->last = NULL; |
13bf43f5 | 8995 | kfree(node); |
0f212204 JA |
8996 | } |
8997 | ||
8452d4a6 | 8998 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8999 | { |
13bf43f5 | 9000 | struct io_tctx_node *node; |
de7f1d9e PB |
9001 | unsigned long index; |
9002 | ||
13bf43f5 | 9003 | xa_for_each(&tctx->xa, index, node) |
2941267b | 9004 | io_uring_del_task_file(index); |
8452d4a6 PB |
9005 | if (tctx->io_wq) { |
9006 | io_wq_put_and_exit(tctx->io_wq); | |
9007 | tctx->io_wq = NULL; | |
9008 | } | |
de7f1d9e PB |
9009 | } |
9010 | ||
521d6a73 PB |
9011 | static s64 tctx_inflight(struct io_uring_task *tctx) |
9012 | { | |
9013 | return percpu_counter_sum(&tctx->inflight); | |
9014 | } | |
9015 | ||
9016 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
9017 | { | |
9018 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
9019 | struct io_ring_ctx *ctx = work->ctx; | |
9020 | struct io_sq_data *sqd = ctx->sq_data; | |
9021 | ||
9022 | if (sqd->thread) | |
9023 | io_uring_cancel_sqpoll(ctx); | |
9024 | complete(&work->completion); | |
9025 | } | |
9026 | ||
9027 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
9028 | { | |
9029 | struct io_sq_data *sqd = ctx->sq_data; | |
9030 | struct io_tctx_exit work = { .ctx = ctx, }; | |
9031 | struct task_struct *task; | |
9032 | ||
9033 | io_sq_thread_park(sqd); | |
9034 | list_del_init(&ctx->sqd_list); | |
9035 | io_sqd_update_thread_idle(sqd); | |
9036 | task = sqd->thread; | |
9037 | if (task) { | |
9038 | init_completion(&work.completion); | |
9039 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 9040 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
9041 | wake_up_process(task); |
9042 | } | |
9043 | io_sq_thread_unpark(sqd); | |
9044 | ||
9045 | if (task) | |
9046 | wait_for_completion(&work.completion); | |
9047 | } | |
9048 | ||
0f212204 JA |
9049 | void __io_uring_files_cancel(struct files_struct *files) |
9050 | { | |
9051 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 9052 | struct io_tctx_node *node; |
ce765372 | 9053 | unsigned long index; |
0f212204 JA |
9054 | |
9055 | /* make sure overflow events are dropped */ | |
fdaf083c | 9056 | atomic_inc(&tctx->in_idle); |
521d6a73 PB |
9057 | xa_for_each(&tctx->xa, index, node) { |
9058 | struct io_ring_ctx *ctx = node->ctx; | |
9059 | ||
9060 | if (ctx->sq_data) { | |
9061 | io_sqpoll_cancel_sync(ctx); | |
9062 | continue; | |
9063 | } | |
9064 | io_uring_cancel_files(ctx, current, files); | |
9065 | if (!files) | |
9066 | io_uring_try_cancel_requests(ctx, current, NULL); | |
9067 | } | |
fdaf083c | 9068 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9069 | |
8452d4a6 PB |
9070 | if (files) |
9071 | io_uring_clean_tctx(tctx); | |
fdaf083c JA |
9072 | } |
9073 | ||
521d6a73 | 9074 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
9075 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
9076 | { | |
37d1e2e3 | 9077 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 9078 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
9079 | s64 inflight; |
9080 | DEFINE_WAIT(wait); | |
fdaf083c | 9081 | |
521d6a73 PB |
9082 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
9083 | ||
0e9ddb39 PB |
9084 | atomic_inc(&tctx->in_idle); |
9085 | do { | |
9086 | /* read completions before cancelations */ | |
9087 | inflight = tctx_inflight(tctx); | |
9088 | if (!inflight) | |
9089 | break; | |
521d6a73 | 9090 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 9091 | |
0e9ddb39 PB |
9092 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9093 | /* | |
9094 | * If we've seen completions, retry without waiting. This | |
9095 | * avoids a race where a completion comes in before we did | |
9096 | * prepare_to_wait(). | |
9097 | */ | |
9098 | if (inflight == tctx_inflight(tctx)) | |
9099 | schedule(); | |
9100 | finish_wait(&tctx->wait, &wait); | |
9101 | } while (1); | |
9102 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9103 | } |
9104 | ||
0f212204 JA |
9105 | /* |
9106 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9107 | * requests. | |
9108 | */ | |
9109 | void __io_uring_task_cancel(void) | |
9110 | { | |
9111 | struct io_uring_task *tctx = current->io_uring; | |
9112 | DEFINE_WAIT(wait); | |
d8a6df10 | 9113 | s64 inflight; |
0f212204 JA |
9114 | |
9115 | /* make sure overflow events are dropped */ | |
fdaf083c | 9116 | atomic_inc(&tctx->in_idle); |
5a978dcf PB |
9117 | __io_uring_files_cancel(NULL); |
9118 | ||
d8a6df10 | 9119 | do { |
0f212204 | 9120 | /* read completions before cancelations */ |
fdaf083c | 9121 | inflight = tctx_inflight(tctx); |
d8a6df10 JA |
9122 | if (!inflight) |
9123 | break; | |
0f212204 JA |
9124 | __io_uring_files_cancel(NULL); |
9125 | ||
9126 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
9127 | ||
9128 | /* | |
a1bb3cd5 PB |
9129 | * If we've seen completions, retry without waiting. This |
9130 | * avoids a race where a completion comes in before we did | |
9131 | * prepare_to_wait(). | |
0f212204 | 9132 | */ |
a1bb3cd5 PB |
9133 | if (inflight == tctx_inflight(tctx)) |
9134 | schedule(); | |
f57555ed | 9135 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9136 | } while (1); |
0f212204 | 9137 | |
fdaf083c | 9138 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9139 | |
8452d4a6 PB |
9140 | io_uring_clean_tctx(tctx); |
9141 | /* all current's requests should be gone, we can kill tctx */ | |
9142 | __io_uring_free(current); | |
44e728b8 PB |
9143 | } |
9144 | ||
6c5c240e RP |
9145 | static void *io_uring_validate_mmap_request(struct file *file, |
9146 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9147 | { |
2b188cc1 | 9148 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9149 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9150 | struct page *page; |
9151 | void *ptr; | |
9152 | ||
9153 | switch (offset) { | |
9154 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9155 | case IORING_OFF_CQ_RING: |
9156 | ptr = ctx->rings; | |
2b188cc1 JA |
9157 | break; |
9158 | case IORING_OFF_SQES: | |
9159 | ptr = ctx->sq_sqes; | |
9160 | break; | |
2b188cc1 | 9161 | default: |
6c5c240e | 9162 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9163 | } |
9164 | ||
9165 | page = virt_to_head_page(ptr); | |
a50b854e | 9166 | if (sz > page_size(page)) |
6c5c240e RP |
9167 | return ERR_PTR(-EINVAL); |
9168 | ||
9169 | return ptr; | |
9170 | } | |
9171 | ||
9172 | #ifdef CONFIG_MMU | |
9173 | ||
9174 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9175 | { | |
9176 | size_t sz = vma->vm_end - vma->vm_start; | |
9177 | unsigned long pfn; | |
9178 | void *ptr; | |
9179 | ||
9180 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9181 | if (IS_ERR(ptr)) | |
9182 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9183 | |
9184 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9185 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9186 | } | |
9187 | ||
6c5c240e RP |
9188 | #else /* !CONFIG_MMU */ |
9189 | ||
9190 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9191 | { | |
9192 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9193 | } | |
9194 | ||
9195 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9196 | { | |
9197 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9198 | } | |
9199 | ||
9200 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9201 | unsigned long addr, unsigned long len, | |
9202 | unsigned long pgoff, unsigned long flags) | |
9203 | { | |
9204 | void *ptr; | |
9205 | ||
9206 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9207 | if (IS_ERR(ptr)) | |
9208 | return PTR_ERR(ptr); | |
9209 | ||
9210 | return (unsigned long) ptr; | |
9211 | } | |
9212 | ||
9213 | #endif /* !CONFIG_MMU */ | |
9214 | ||
d9d05217 | 9215 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9216 | { |
9217 | DEFINE_WAIT(wait); | |
9218 | ||
9219 | do { | |
9220 | if (!io_sqring_full(ctx)) | |
9221 | break; | |
90554200 JA |
9222 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9223 | ||
9224 | if (!io_sqring_full(ctx)) | |
9225 | break; | |
90554200 JA |
9226 | schedule(); |
9227 | } while (!signal_pending(current)); | |
9228 | ||
9229 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9230 | return 0; |
90554200 JA |
9231 | } |
9232 | ||
c73ebb68 HX |
9233 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9234 | struct __kernel_timespec __user **ts, | |
9235 | const sigset_t __user **sig) | |
9236 | { | |
9237 | struct io_uring_getevents_arg arg; | |
9238 | ||
9239 | /* | |
9240 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9241 | * is just a pointer to the sigset_t. | |
9242 | */ | |
9243 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9244 | *sig = (const sigset_t __user *) argp; | |
9245 | *ts = NULL; | |
9246 | return 0; | |
9247 | } | |
9248 | ||
9249 | /* | |
9250 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9251 | * timespec and sigset_t pointers if good. | |
9252 | */ | |
9253 | if (*argsz != sizeof(arg)) | |
9254 | return -EINVAL; | |
9255 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9256 | return -EFAULT; | |
9257 | *sig = u64_to_user_ptr(arg.sigmask); | |
9258 | *argsz = arg.sigmask_sz; | |
9259 | *ts = u64_to_user_ptr(arg.ts); | |
9260 | return 0; | |
9261 | } | |
9262 | ||
2b188cc1 | 9263 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9264 | u32, min_complete, u32, flags, const void __user *, argp, |
9265 | size_t, argsz) | |
2b188cc1 JA |
9266 | { |
9267 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9268 | int submitted = 0; |
9269 | struct fd f; | |
33f993da | 9270 | long ret; |
2b188cc1 | 9271 | |
4c6e277c | 9272 | io_run_task_work(); |
b41e9852 | 9273 | |
33f993da PB |
9274 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9275 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9276 | return -EINVAL; |
9277 | ||
9278 | f = fdget(fd); | |
33f993da | 9279 | if (unlikely(!f.file)) |
2b188cc1 JA |
9280 | return -EBADF; |
9281 | ||
9282 | ret = -EOPNOTSUPP; | |
33f993da | 9283 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9284 | goto out_fput; |
9285 | ||
9286 | ret = -ENXIO; | |
9287 | ctx = f.file->private_data; | |
33f993da | 9288 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9289 | goto out_fput; |
9290 | ||
7e84e1c7 | 9291 | ret = -EBADFD; |
33f993da | 9292 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9293 | goto out; |
9294 | ||
6c271ce2 JA |
9295 | /* |
9296 | * For SQ polling, the thread will do all submissions and completions. | |
9297 | * Just return the requested submit count, and wake the thread if | |
9298 | * we were asked to. | |
9299 | */ | |
b2a9eada | 9300 | ret = 0; |
6c271ce2 | 9301 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9302 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9303 | |
d9d05217 | 9304 | ret = -EOWNERDEAD; |
04147488 SM |
9305 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9306 | goto out; | |
9307 | } | |
6c271ce2 | 9308 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9309 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9310 | if (flags & IORING_ENTER_SQ_WAIT) { |
9311 | ret = io_sqpoll_wait_sq(ctx); | |
9312 | if (ret) | |
9313 | goto out; | |
9314 | } | |
6c271ce2 | 9315 | submitted = to_submit; |
b2a9eada | 9316 | } else if (to_submit) { |
baf186c4 | 9317 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9318 | if (unlikely(ret)) |
9319 | goto out; | |
2b188cc1 | 9320 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9321 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9322 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9323 | |
9324 | if (submitted != to_submit) | |
9325 | goto out; | |
2b188cc1 JA |
9326 | } |
9327 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9328 | const sigset_t __user *sig; |
9329 | struct __kernel_timespec __user *ts; | |
9330 | ||
9331 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9332 | if (unlikely(ret)) | |
9333 | goto out; | |
9334 | ||
2b188cc1 JA |
9335 | min_complete = min(min_complete, ctx->cq_entries); |
9336 | ||
32b2244a XW |
9337 | /* |
9338 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9339 | * space applications don't need to do io completion events | |
9340 | * polling again, they can rely on io_sq_thread to do polling | |
9341 | * work, which can reduce cpu usage and uring_lock contention. | |
9342 | */ | |
9343 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9344 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9345 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9346 | } else { |
c73ebb68 | 9347 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9348 | } |
2b188cc1 JA |
9349 | } |
9350 | ||
7c504e65 | 9351 | out: |
6805b32e | 9352 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9353 | out_fput: |
9354 | fdput(f); | |
9355 | return submitted ? submitted : ret; | |
9356 | } | |
9357 | ||
bebdb65e | 9358 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9359 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9360 | const struct cred *cred) | |
87ce955b | 9361 | { |
87ce955b JA |
9362 | struct user_namespace *uns = seq_user_ns(m); |
9363 | struct group_info *gi; | |
9364 | kernel_cap_t cap; | |
9365 | unsigned __capi; | |
9366 | int g; | |
9367 | ||
9368 | seq_printf(m, "%5d\n", id); | |
9369 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9370 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9371 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9372 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9373 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9374 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9375 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9376 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9377 | seq_puts(m, "\n\tGroups:\t"); | |
9378 | gi = cred->group_info; | |
9379 | for (g = 0; g < gi->ngroups; g++) { | |
9380 | seq_put_decimal_ull(m, g ? " " : "", | |
9381 | from_kgid_munged(uns, gi->gid[g])); | |
9382 | } | |
9383 | seq_puts(m, "\n\tCapEff:\t"); | |
9384 | cap = cred->cap_effective; | |
9385 | CAP_FOR_EACH_U32(__capi) | |
9386 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9387 | seq_putc(m, '\n'); | |
9388 | return 0; | |
9389 | } | |
9390 | ||
9391 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9392 | { | |
dbbe9c64 | 9393 | struct io_sq_data *sq = NULL; |
fad8e0de | 9394 | bool has_lock; |
87ce955b JA |
9395 | int i; |
9396 | ||
fad8e0de JA |
9397 | /* |
9398 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9399 | * since fdinfo case grabs it in the opposite direction of normal use | |
9400 | * cases. If we fail to get the lock, we just don't iterate any | |
9401 | * structures that could be going away outside the io_uring mutex. | |
9402 | */ | |
9403 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9404 | ||
5f3f26f9 | 9405 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9406 | sq = ctx->sq_data; |
5f3f26f9 JA |
9407 | if (!sq->thread) |
9408 | sq = NULL; | |
9409 | } | |
dbbe9c64 JQ |
9410 | |
9411 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9412 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9413 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9414 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9415 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9416 | |
87ce955b JA |
9417 | if (f) |
9418 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9419 | else | |
9420 | seq_printf(m, "%5u: <none>\n", i); | |
9421 | } | |
9422 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9423 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b JA |
9424 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
9425 | ||
9426 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, | |
9427 | (unsigned int) buf->len); | |
9428 | } | |
61cf9370 MWO |
9429 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9430 | unsigned long index; | |
9431 | const struct cred *cred; | |
9432 | ||
87ce955b | 9433 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9434 | xa_for_each(&ctx->personalities, index, cred) |
9435 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9436 | } |
d7718a9d JA |
9437 | seq_printf(m, "PollList:\n"); |
9438 | spin_lock_irq(&ctx->completion_lock); | |
9439 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9440 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9441 | struct io_kiocb *req; | |
9442 | ||
9443 | hlist_for_each_entry(req, list, hash_node) | |
9444 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9445 | req->task->task_works != NULL); | |
9446 | } | |
9447 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9448 | if (has_lock) |
9449 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9450 | } |
9451 | ||
9452 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9453 | { | |
9454 | struct io_ring_ctx *ctx = f->private_data; | |
9455 | ||
9456 | if (percpu_ref_tryget(&ctx->refs)) { | |
9457 | __io_uring_show_fdinfo(ctx, m); | |
9458 | percpu_ref_put(&ctx->refs); | |
9459 | } | |
9460 | } | |
bebdb65e | 9461 | #endif |
87ce955b | 9462 | |
2b188cc1 JA |
9463 | static const struct file_operations io_uring_fops = { |
9464 | .release = io_uring_release, | |
9465 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9466 | #ifndef CONFIG_MMU |
9467 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9468 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9469 | #endif | |
2b188cc1 JA |
9470 | .poll = io_uring_poll, |
9471 | .fasync = io_uring_fasync, | |
bebdb65e | 9472 | #ifdef CONFIG_PROC_FS |
87ce955b | 9473 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9474 | #endif |
2b188cc1 JA |
9475 | }; |
9476 | ||
9477 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9478 | struct io_uring_params *p) | |
9479 | { | |
75b28aff HV |
9480 | struct io_rings *rings; |
9481 | size_t size, sq_array_offset; | |
2b188cc1 | 9482 | |
bd740481 JA |
9483 | /* make sure these are sane, as we already accounted them */ |
9484 | ctx->sq_entries = p->sq_entries; | |
9485 | ctx->cq_entries = p->cq_entries; | |
9486 | ||
75b28aff HV |
9487 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9488 | if (size == SIZE_MAX) | |
9489 | return -EOVERFLOW; | |
9490 | ||
9491 | rings = io_mem_alloc(size); | |
9492 | if (!rings) | |
2b188cc1 JA |
9493 | return -ENOMEM; |
9494 | ||
75b28aff HV |
9495 | ctx->rings = rings; |
9496 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9497 | rings->sq_ring_mask = p->sq_entries - 1; | |
9498 | rings->cq_ring_mask = p->cq_entries - 1; | |
9499 | rings->sq_ring_entries = p->sq_entries; | |
9500 | rings->cq_ring_entries = p->cq_entries; | |
9501 | ctx->sq_mask = rings->sq_ring_mask; | |
9502 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9503 | |
9504 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9505 | if (size == SIZE_MAX) { |
9506 | io_mem_free(ctx->rings); | |
9507 | ctx->rings = NULL; | |
2b188cc1 | 9508 | return -EOVERFLOW; |
eb065d30 | 9509 | } |
2b188cc1 JA |
9510 | |
9511 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9512 | if (!ctx->sq_sqes) { |
9513 | io_mem_free(ctx->rings); | |
9514 | ctx->rings = NULL; | |
2b188cc1 | 9515 | return -ENOMEM; |
eb065d30 | 9516 | } |
2b188cc1 | 9517 | |
2b188cc1 JA |
9518 | return 0; |
9519 | } | |
9520 | ||
9faadcc8 PB |
9521 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9522 | { | |
9523 | int ret, fd; | |
9524 | ||
9525 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9526 | if (fd < 0) | |
9527 | return fd; | |
9528 | ||
baf186c4 | 9529 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9530 | if (ret) { |
9531 | put_unused_fd(fd); | |
9532 | return ret; | |
9533 | } | |
9534 | fd_install(fd, file); | |
9535 | return fd; | |
9536 | } | |
9537 | ||
2b188cc1 JA |
9538 | /* |
9539 | * Allocate an anonymous fd, this is what constitutes the application | |
9540 | * visible backing of an io_uring instance. The application mmaps this | |
9541 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9542 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9543 | */ | |
9faadcc8 | 9544 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9545 | { |
9546 | struct file *file; | |
9faadcc8 | 9547 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9548 | int ret; |
9549 | ||
2b188cc1 JA |
9550 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9551 | &ctx->ring_sock); | |
9552 | if (ret) | |
9faadcc8 | 9553 | return ERR_PTR(ret); |
2b188cc1 JA |
9554 | #endif |
9555 | ||
2b188cc1 JA |
9556 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9557 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9558 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9559 | if (IS_ERR(file)) { |
9560 | sock_release(ctx->ring_sock); | |
9561 | ctx->ring_sock = NULL; | |
9562 | } else { | |
9563 | ctx->ring_sock->file = file; | |
0f212204 | 9564 | } |
2b188cc1 | 9565 | #endif |
9faadcc8 | 9566 | return file; |
2b188cc1 JA |
9567 | } |
9568 | ||
7f13657d XW |
9569 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9570 | struct io_uring_params __user *params) | |
2b188cc1 | 9571 | { |
2b188cc1 | 9572 | struct io_ring_ctx *ctx; |
9faadcc8 | 9573 | struct file *file; |
2b188cc1 JA |
9574 | int ret; |
9575 | ||
8110c1a6 | 9576 | if (!entries) |
2b188cc1 | 9577 | return -EINVAL; |
8110c1a6 JA |
9578 | if (entries > IORING_MAX_ENTRIES) { |
9579 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9580 | return -EINVAL; | |
9581 | entries = IORING_MAX_ENTRIES; | |
9582 | } | |
2b188cc1 JA |
9583 | |
9584 | /* | |
9585 | * Use twice as many entries for the CQ ring. It's possible for the | |
9586 | * application to drive a higher depth than the size of the SQ ring, | |
9587 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9588 | * some flexibility in overcommitting a bit. If the application has |
9589 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9590 | * of CQ ring entries manually. | |
2b188cc1 JA |
9591 | */ |
9592 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9593 | if (p->flags & IORING_SETUP_CQSIZE) { |
9594 | /* | |
9595 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9596 | * to a power-of-two, if it isn't already. We do NOT impose | |
9597 | * any cq vs sq ring sizing. | |
9598 | */ | |
eb2667b3 | 9599 | if (!p->cq_entries) |
33a107f0 | 9600 | return -EINVAL; |
8110c1a6 JA |
9601 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9602 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9603 | return -EINVAL; | |
9604 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9605 | } | |
eb2667b3 JQ |
9606 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9607 | if (p->cq_entries < p->sq_entries) | |
9608 | return -EINVAL; | |
33a107f0 JA |
9609 | } else { |
9610 | p->cq_entries = 2 * p->sq_entries; | |
9611 | } | |
2b188cc1 | 9612 | |
2b188cc1 | 9613 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9614 | if (!ctx) |
2b188cc1 | 9615 | return -ENOMEM; |
2b188cc1 | 9616 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9617 | if (!capable(CAP_IPC_LOCK)) |
9618 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9619 | |
9620 | /* | |
9621 | * This is just grabbed for accounting purposes. When a process exits, | |
9622 | * the mm is exited and dropped before the files, hence we need to hang | |
9623 | * on to this mm purely for the purposes of being able to unaccount | |
9624 | * memory (locked/pinned vm). It's not used for anything else. | |
9625 | */ | |
6b7898eb | 9626 | mmgrab(current->mm); |
2aede0e4 | 9627 | ctx->mm_account = current->mm; |
6b7898eb | 9628 | |
2b188cc1 JA |
9629 | ret = io_allocate_scq_urings(ctx, p); |
9630 | if (ret) | |
9631 | goto err; | |
9632 | ||
7e84e1c7 | 9633 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9634 | if (ret) |
9635 | goto err; | |
9636 | ||
2b188cc1 | 9637 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9638 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9639 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9640 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9641 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9642 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9643 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9644 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9645 | |
9646 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9647 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9648 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9649 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9650 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9651 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9652 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9653 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9654 | |
7f13657d XW |
9655 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9656 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9657 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9658 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9659 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9660 | |
9661 | if (copy_to_user(params, p, sizeof(*p))) { | |
9662 | ret = -EFAULT; | |
9663 | goto err; | |
9664 | } | |
d1719f70 | 9665 | |
9faadcc8 PB |
9666 | file = io_uring_get_file(ctx); |
9667 | if (IS_ERR(file)) { | |
9668 | ret = PTR_ERR(file); | |
9669 | goto err; | |
9670 | } | |
9671 | ||
044c1ab3 JA |
9672 | /* |
9673 | * Install ring fd as the very last thing, so we don't risk someone | |
9674 | * having closed it before we finish setup | |
9675 | */ | |
9faadcc8 PB |
9676 | ret = io_uring_install_fd(ctx, file); |
9677 | if (ret < 0) { | |
9678 | /* fput will clean it up */ | |
9679 | fput(file); | |
9680 | return ret; | |
9681 | } | |
044c1ab3 | 9682 | |
c826bd7a | 9683 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9684 | return ret; |
9685 | err: | |
9686 | io_ring_ctx_wait_and_kill(ctx); | |
9687 | return ret; | |
9688 | } | |
9689 | ||
9690 | /* | |
9691 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9692 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9693 | * params structure passed in. | |
9694 | */ | |
9695 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9696 | { | |
9697 | struct io_uring_params p; | |
2b188cc1 JA |
9698 | int i; |
9699 | ||
9700 | if (copy_from_user(&p, params, sizeof(p))) | |
9701 | return -EFAULT; | |
9702 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9703 | if (p.resv[i]) | |
9704 | return -EINVAL; | |
9705 | } | |
9706 | ||
6c271ce2 | 9707 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9708 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9709 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9710 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9711 | return -EINVAL; |
9712 | ||
7f13657d | 9713 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9714 | } |
9715 | ||
9716 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9717 | struct io_uring_params __user *, params) | |
9718 | { | |
9719 | return io_uring_setup(entries, params); | |
9720 | } | |
9721 | ||
66f4af93 JA |
9722 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9723 | { | |
9724 | struct io_uring_probe *p; | |
9725 | size_t size; | |
9726 | int i, ret; | |
9727 | ||
9728 | size = struct_size(p, ops, nr_args); | |
9729 | if (size == SIZE_MAX) | |
9730 | return -EOVERFLOW; | |
9731 | p = kzalloc(size, GFP_KERNEL); | |
9732 | if (!p) | |
9733 | return -ENOMEM; | |
9734 | ||
9735 | ret = -EFAULT; | |
9736 | if (copy_from_user(p, arg, size)) | |
9737 | goto out; | |
9738 | ret = -EINVAL; | |
9739 | if (memchr_inv(p, 0, size)) | |
9740 | goto out; | |
9741 | ||
9742 | p->last_op = IORING_OP_LAST - 1; | |
9743 | if (nr_args > IORING_OP_LAST) | |
9744 | nr_args = IORING_OP_LAST; | |
9745 | ||
9746 | for (i = 0; i < nr_args; i++) { | |
9747 | p->ops[i].op = i; | |
9748 | if (!io_op_defs[i].not_supported) | |
9749 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9750 | } | |
9751 | p->ops_len = i; | |
9752 | ||
9753 | ret = 0; | |
9754 | if (copy_to_user(arg, p, size)) | |
9755 | ret = -EFAULT; | |
9756 | out: | |
9757 | kfree(p); | |
9758 | return ret; | |
9759 | } | |
9760 | ||
071698e1 JA |
9761 | static int io_register_personality(struct io_ring_ctx *ctx) |
9762 | { | |
4379bf8b | 9763 | const struct cred *creds; |
61cf9370 | 9764 | u32 id; |
1e6fa521 | 9765 | int ret; |
071698e1 | 9766 | |
4379bf8b | 9767 | creds = get_current_cred(); |
1e6fa521 | 9768 | |
61cf9370 MWO |
9769 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9770 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9771 | if (!ret) | |
9772 | return id; | |
9773 | put_cred(creds); | |
1e6fa521 | 9774 | return ret; |
071698e1 JA |
9775 | } |
9776 | ||
21b55dbc SG |
9777 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9778 | unsigned int nr_args) | |
9779 | { | |
9780 | struct io_uring_restriction *res; | |
9781 | size_t size; | |
9782 | int i, ret; | |
9783 | ||
7e84e1c7 SG |
9784 | /* Restrictions allowed only if rings started disabled */ |
9785 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9786 | return -EBADFD; | |
9787 | ||
21b55dbc | 9788 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9789 | if (ctx->restrictions.registered) |
21b55dbc SG |
9790 | return -EBUSY; |
9791 | ||
9792 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9793 | return -EINVAL; | |
9794 | ||
9795 | size = array_size(nr_args, sizeof(*res)); | |
9796 | if (size == SIZE_MAX) | |
9797 | return -EOVERFLOW; | |
9798 | ||
9799 | res = memdup_user(arg, size); | |
9800 | if (IS_ERR(res)) | |
9801 | return PTR_ERR(res); | |
9802 | ||
9803 | ret = 0; | |
9804 | ||
9805 | for (i = 0; i < nr_args; i++) { | |
9806 | switch (res[i].opcode) { | |
9807 | case IORING_RESTRICTION_REGISTER_OP: | |
9808 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9809 | ret = -EINVAL; | |
9810 | goto out; | |
9811 | } | |
9812 | ||
9813 | __set_bit(res[i].register_op, | |
9814 | ctx->restrictions.register_op); | |
9815 | break; | |
9816 | case IORING_RESTRICTION_SQE_OP: | |
9817 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9818 | ret = -EINVAL; | |
9819 | goto out; | |
9820 | } | |
9821 | ||
9822 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9823 | break; | |
9824 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9825 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9826 | break; | |
9827 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9828 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9829 | break; | |
9830 | default: | |
9831 | ret = -EINVAL; | |
9832 | goto out; | |
9833 | } | |
9834 | } | |
9835 | ||
9836 | out: | |
9837 | /* Reset all restrictions if an error happened */ | |
9838 | if (ret != 0) | |
9839 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9840 | else | |
7e84e1c7 | 9841 | ctx->restrictions.registered = true; |
21b55dbc SG |
9842 | |
9843 | kfree(res); | |
9844 | return ret; | |
9845 | } | |
9846 | ||
7e84e1c7 SG |
9847 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9848 | { | |
9849 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9850 | return -EBADFD; | |
9851 | ||
9852 | if (ctx->restrictions.registered) | |
9853 | ctx->restricted = 1; | |
9854 | ||
0298ef96 PB |
9855 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9856 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9857 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9858 | return 0; |
9859 | } | |
9860 | ||
071698e1 JA |
9861 | static bool io_register_op_must_quiesce(int op) |
9862 | { | |
9863 | switch (op) { | |
9864 | case IORING_UNREGISTER_FILES: | |
9865 | case IORING_REGISTER_FILES_UPDATE: | |
9866 | case IORING_REGISTER_PROBE: | |
9867 | case IORING_REGISTER_PERSONALITY: | |
9868 | case IORING_UNREGISTER_PERSONALITY: | |
9869 | return false; | |
9870 | default: | |
9871 | return true; | |
9872 | } | |
9873 | } | |
9874 | ||
edafccee JA |
9875 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9876 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9877 | __releases(ctx->uring_lock) |
9878 | __acquires(ctx->uring_lock) | |
edafccee JA |
9879 | { |
9880 | int ret; | |
9881 | ||
35fa71a0 JA |
9882 | /* |
9883 | * We're inside the ring mutex, if the ref is already dying, then | |
9884 | * someone else killed the ctx or is already going through | |
9885 | * io_uring_register(). | |
9886 | */ | |
9887 | if (percpu_ref_is_dying(&ctx->refs)) | |
9888 | return -ENXIO; | |
9889 | ||
071698e1 | 9890 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9891 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9892 | |
05f3fb3c JA |
9893 | /* |
9894 | * Drop uring mutex before waiting for references to exit. If | |
9895 | * another thread is currently inside io_uring_enter() it might | |
9896 | * need to grab the uring_lock to make progress. If we hold it | |
9897 | * here across the drain wait, then we can deadlock. It's safe | |
9898 | * to drop the mutex here, since no new references will come in | |
9899 | * after we've killed the percpu ref. | |
9900 | */ | |
9901 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9902 | do { |
9903 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9904 | if (!ret) | |
9905 | break; | |
ed6930c9 JA |
9906 | ret = io_run_task_work_sig(); |
9907 | if (ret < 0) | |
9908 | break; | |
af9c1a44 JA |
9909 | } while (1); |
9910 | ||
05f3fb3c | 9911 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9912 | |
c150368b JA |
9913 | if (ret) { |
9914 | percpu_ref_resurrect(&ctx->refs); | |
21b55dbc SG |
9915 | goto out_quiesce; |
9916 | } | |
9917 | } | |
9918 | ||
9919 | if (ctx->restricted) { | |
9920 | if (opcode >= IORING_REGISTER_LAST) { | |
9921 | ret = -EINVAL; | |
9922 | goto out; | |
9923 | } | |
9924 | ||
9925 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9926 | ret = -EACCES; | |
c150368b JA |
9927 | goto out; |
9928 | } | |
05f3fb3c | 9929 | } |
edafccee JA |
9930 | |
9931 | switch (opcode) { | |
9932 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9933 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9934 | break; |
9935 | case IORING_UNREGISTER_BUFFERS: | |
9936 | ret = -EINVAL; | |
9937 | if (arg || nr_args) | |
9938 | break; | |
0a96bbe4 | 9939 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9940 | break; |
6b06314c JA |
9941 | case IORING_REGISTER_FILES: |
9942 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9943 | break; | |
9944 | case IORING_UNREGISTER_FILES: | |
9945 | ret = -EINVAL; | |
9946 | if (arg || nr_args) | |
9947 | break; | |
9948 | ret = io_sqe_files_unregister(ctx); | |
9949 | break; | |
c3a31e60 JA |
9950 | case IORING_REGISTER_FILES_UPDATE: |
9951 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9952 | break; | |
9b402849 | 9953 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9954 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9955 | ret = -EINVAL; |
9956 | if (nr_args != 1) | |
9957 | break; | |
9958 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9959 | if (ret) |
9960 | break; | |
9961 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9962 | ctx->eventfd_async = 1; | |
9963 | else | |
9964 | ctx->eventfd_async = 0; | |
9b402849 JA |
9965 | break; |
9966 | case IORING_UNREGISTER_EVENTFD: | |
9967 | ret = -EINVAL; | |
9968 | if (arg || nr_args) | |
9969 | break; | |
9970 | ret = io_eventfd_unregister(ctx); | |
9971 | break; | |
66f4af93 JA |
9972 | case IORING_REGISTER_PROBE: |
9973 | ret = -EINVAL; | |
9974 | if (!arg || nr_args > 256) | |
9975 | break; | |
9976 | ret = io_probe(ctx, arg, nr_args); | |
9977 | break; | |
071698e1 JA |
9978 | case IORING_REGISTER_PERSONALITY: |
9979 | ret = -EINVAL; | |
9980 | if (arg || nr_args) | |
9981 | break; | |
9982 | ret = io_register_personality(ctx); | |
9983 | break; | |
9984 | case IORING_UNREGISTER_PERSONALITY: | |
9985 | ret = -EINVAL; | |
9986 | if (arg) | |
9987 | break; | |
9988 | ret = io_unregister_personality(ctx, nr_args); | |
9989 | break; | |
7e84e1c7 SG |
9990 | case IORING_REGISTER_ENABLE_RINGS: |
9991 | ret = -EINVAL; | |
9992 | if (arg || nr_args) | |
9993 | break; | |
9994 | ret = io_register_enable_rings(ctx); | |
9995 | break; | |
21b55dbc SG |
9996 | case IORING_REGISTER_RESTRICTIONS: |
9997 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9998 | break; | |
edafccee JA |
9999 | default: |
10000 | ret = -EINVAL; | |
10001 | break; | |
10002 | } | |
10003 | ||
21b55dbc | 10004 | out: |
071698e1 | 10005 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10006 | /* bring the ctx back to life */ |
05f3fb3c | 10007 | percpu_ref_reinit(&ctx->refs); |
21b55dbc | 10008 | out_quiesce: |
0f158b4c | 10009 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10010 | } |
edafccee JA |
10011 | return ret; |
10012 | } | |
10013 | ||
10014 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10015 | void __user *, arg, unsigned int, nr_args) | |
10016 | { | |
10017 | struct io_ring_ctx *ctx; | |
10018 | long ret = -EBADF; | |
10019 | struct fd f; | |
10020 | ||
10021 | f = fdget(fd); | |
10022 | if (!f.file) | |
10023 | return -EBADF; | |
10024 | ||
10025 | ret = -EOPNOTSUPP; | |
10026 | if (f.file->f_op != &io_uring_fops) | |
10027 | goto out_fput; | |
10028 | ||
10029 | ctx = f.file->private_data; | |
10030 | ||
b6c23dd5 PB |
10031 | io_run_task_work(); |
10032 | ||
edafccee JA |
10033 | mutex_lock(&ctx->uring_lock); |
10034 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10035 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10036 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10037 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10038 | out_fput: |
10039 | fdput(f); | |
10040 | return ret; | |
10041 | } | |
10042 | ||
2b188cc1 JA |
10043 | static int __init io_uring_init(void) |
10044 | { | |
d7f62e82 SM |
10045 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10046 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10047 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10048 | } while (0) | |
10049 | ||
10050 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10051 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10052 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10053 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10054 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10055 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10056 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10057 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10058 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10059 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10060 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10061 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10062 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10063 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10064 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10065 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10066 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10067 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10068 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10069 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10070 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10071 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10072 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10073 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10074 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10075 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10076 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10077 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10078 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10079 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10080 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10081 | |
d3656344 | 10082 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10083 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10084 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10085 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10086 | return 0; |
10087 | }; | |
10088 | __initcall(io_uring_init); |