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; | |
4751f53d | 197 | u64 ubuf_end; |
edafccee | 198 | unsigned int nr_bvecs; |
de293938 | 199 | unsigned long acct_pages; |
41edf1a5 | 200 | struct bio_vec bvec[]; |
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 | ||
a04b0ac0 PB |
210 | struct io_fixed_file { |
211 | /* file * with additional FFS_* flags */ | |
212 | unsigned long file_ptr; | |
213 | }; | |
214 | ||
269bbe5f BM |
215 | struct io_rsrc_put { |
216 | struct list_head list; | |
b60c8dce | 217 | u64 tag; |
50238531 BM |
218 | union { |
219 | void *rsrc; | |
220 | struct file *file; | |
bd54b6fe | 221 | struct io_mapped_ubuf *buf; |
50238531 | 222 | }; |
269bbe5f BM |
223 | }; |
224 | ||
aeca241b PB |
225 | struct io_file_table { |
226 | /* two level table */ | |
227 | struct io_fixed_file **files; | |
31b51510 JA |
228 | }; |
229 | ||
b895c9a6 | 230 | struct io_rsrc_node { |
05589553 XW |
231 | struct percpu_ref refs; |
232 | struct list_head node; | |
269bbe5f | 233 | struct list_head rsrc_list; |
b895c9a6 | 234 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 235 | struct llist_node llist; |
e297822b | 236 | bool done; |
05589553 XW |
237 | }; |
238 | ||
40ae0ff7 PB |
239 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
240 | ||
b895c9a6 | 241 | struct io_rsrc_data { |
05f3fb3c JA |
242 | struct io_ring_ctx *ctx; |
243 | ||
b60c8dce | 244 | u64 *tags; |
40ae0ff7 | 245 | rsrc_put_fn *do_put; |
3e942498 | 246 | atomic_t refs; |
05f3fb3c | 247 | struct completion done; |
8bad28d8 | 248 | bool quiesce; |
05f3fb3c JA |
249 | }; |
250 | ||
5a2e745d JA |
251 | struct io_buffer { |
252 | struct list_head list; | |
253 | __u64 addr; | |
254 | __s32 len; | |
255 | __u16 bid; | |
256 | }; | |
257 | ||
21b55dbc SG |
258 | struct io_restriction { |
259 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
260 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
261 | u8 sqe_flags_allowed; | |
262 | u8 sqe_flags_required; | |
7e84e1c7 | 263 | bool registered; |
21b55dbc SG |
264 | }; |
265 | ||
37d1e2e3 JA |
266 | enum { |
267 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
268 | IO_SQ_THREAD_SHOULD_PARK, | |
269 | }; | |
270 | ||
534ca6d6 JA |
271 | struct io_sq_data { |
272 | refcount_t refs; | |
9e138a48 | 273 | atomic_t park_pending; |
09a6f4ef | 274 | struct mutex lock; |
69fb2131 JA |
275 | |
276 | /* ctx's that are using this sqd */ | |
277 | struct list_head ctx_list; | |
69fb2131 | 278 | |
534ca6d6 JA |
279 | struct task_struct *thread; |
280 | struct wait_queue_head wait; | |
08369246 XW |
281 | |
282 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
283 | int sq_cpu; |
284 | pid_t task_pid; | |
5c2469e0 | 285 | pid_t task_tgid; |
37d1e2e3 JA |
286 | |
287 | unsigned long state; | |
37d1e2e3 | 288 | struct completion exited; |
b7f5a0bf | 289 | struct callback_head *park_task_work; |
534ca6d6 JA |
290 | }; |
291 | ||
258b29a9 | 292 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 293 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 294 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 295 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
296 | |
297 | struct io_comp_state { | |
6dd0be1e | 298 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 299 | unsigned int nr; |
c7dae4ba JA |
300 | unsigned int locked_free_nr; |
301 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 302 | struct list_head free_list; |
c7dae4ba JA |
303 | /* IRQ completion list, under ->completion_lock */ |
304 | struct list_head locked_free_list; | |
258b29a9 PB |
305 | }; |
306 | ||
a1ab7b35 PB |
307 | struct io_submit_link { |
308 | struct io_kiocb *head; | |
309 | struct io_kiocb *last; | |
310 | }; | |
311 | ||
258b29a9 PB |
312 | struct io_submit_state { |
313 | struct blk_plug plug; | |
a1ab7b35 | 314 | struct io_submit_link link; |
258b29a9 PB |
315 | |
316 | /* | |
317 | * io_kiocb alloc cache | |
318 | */ | |
bf019da7 | 319 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
320 | unsigned int free_reqs; |
321 | ||
322 | bool plug_started; | |
323 | ||
324 | /* | |
325 | * Batch completion logic | |
326 | */ | |
327 | struct io_comp_state comp; | |
328 | ||
329 | /* | |
330 | * File reference cache | |
331 | */ | |
332 | struct file *file; | |
333 | unsigned int fd; | |
334 | unsigned int file_refs; | |
335 | unsigned int ios_left; | |
336 | }; | |
337 | ||
2b188cc1 JA |
338 | struct io_ring_ctx { |
339 | struct { | |
340 | struct percpu_ref refs; | |
341 | } ____cacheline_aligned_in_smp; | |
342 | ||
343 | struct { | |
344 | unsigned int flags; | |
e1d85334 | 345 | unsigned int compat: 1; |
e1d85334 RD |
346 | unsigned int drain_next: 1; |
347 | unsigned int eventfd_async: 1; | |
21b55dbc | 348 | unsigned int restricted: 1; |
2b188cc1 | 349 | |
75b28aff HV |
350 | /* |
351 | * Ring buffer of indices into array of io_uring_sqe, which is | |
352 | * mmapped by the application using the IORING_OFF_SQES offset. | |
353 | * | |
354 | * This indirection could e.g. be used to assign fixed | |
355 | * io_uring_sqe entries to operations and only submit them to | |
356 | * the queue when needed. | |
357 | * | |
358 | * The kernel modifies neither the indices array nor the entries | |
359 | * array. | |
360 | */ | |
361 | u32 *sq_array; | |
2b188cc1 JA |
362 | unsigned cached_sq_head; |
363 | unsigned sq_entries; | |
364 | unsigned sq_mask; | |
6c271ce2 | 365 | unsigned sq_thread_idle; |
498ccd9e | 366 | unsigned cached_sq_dropped; |
2c3bac6d | 367 | unsigned cached_cq_overflow; |
ad3eb2c8 | 368 | unsigned long sq_check_overflow; |
de0617e4 | 369 | |
e941894e JA |
370 | /* hashed buffered write serialization */ |
371 | struct io_wq_hash *hash_map; | |
372 | ||
de0617e4 | 373 | struct list_head defer_list; |
5262f567 | 374 | struct list_head timeout_list; |
1d7bb1d5 | 375 | struct list_head cq_overflow_list; |
fcb323cc | 376 | |
ad3eb2c8 | 377 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
378 | } ____cacheline_aligned_in_smp; |
379 | ||
3c1a2ead JA |
380 | struct { |
381 | struct mutex uring_lock; | |
382 | wait_queue_head_t wait; | |
383 | } ____cacheline_aligned_in_smp; | |
384 | ||
385 | struct io_submit_state submit_state; | |
386 | ||
206aefde JA |
387 | struct io_rings *rings; |
388 | ||
2aede0e4 JA |
389 | /* Only used for accounting purposes */ |
390 | struct mm_struct *mm_account; | |
391 | ||
7c30f36a | 392 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
393 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
394 | ||
90554200 | 395 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 396 | struct list_head sqd_list; |
75b28aff | 397 | |
6b06314c JA |
398 | /* |
399 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
400 | * readers must ensure that ->refs is alive as long as the file* is | |
401 | * used. Only updated through io_uring_register(2). | |
402 | */ | |
b895c9a6 | 403 | struct io_rsrc_data *file_data; |
aeca241b | 404 | struct io_file_table file_table; |
6b06314c JA |
405 | unsigned nr_user_files; |
406 | ||
edafccee | 407 | /* if used, fixed mapped user buffers */ |
bd54b6fe | 408 | struct io_rsrc_data *buf_data; |
edafccee | 409 | unsigned nr_user_bufs; |
41edf1a5 | 410 | struct io_mapped_ubuf **user_bufs; |
edafccee | 411 | |
2b188cc1 JA |
412 | struct user_struct *user; |
413 | ||
0f158b4c | 414 | struct completion ref_comp; |
206aefde JA |
415 | |
416 | #if defined(CONFIG_UNIX) | |
417 | struct socket *ring_sock; | |
418 | #endif | |
419 | ||
9e15c3a0 | 420 | struct xarray io_buffers; |
5a2e745d | 421 | |
61cf9370 MWO |
422 | struct xarray personalities; |
423 | u32 pers_next; | |
071698e1 | 424 | |
206aefde JA |
425 | struct { |
426 | unsigned cached_cq_tail; | |
427 | unsigned cq_entries; | |
428 | unsigned cq_mask; | |
429 | atomic_t cq_timeouts; | |
f010505b | 430 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 431 | unsigned long cq_check_overflow; |
206aefde JA |
432 | struct wait_queue_head cq_wait; |
433 | struct fasync_struct *cq_fasync; | |
434 | struct eventfd_ctx *cq_ev_fd; | |
435 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 436 | |
2b188cc1 JA |
437 | struct { |
438 | spinlock_t completion_lock; | |
e94f141b | 439 | |
def596e9 | 440 | /* |
540e32a0 | 441 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
442 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
443 | * For SQPOLL, only the single threaded io_sq_thread() will | |
444 | * manipulate the list, hence no extra locking is needed there. | |
445 | */ | |
540e32a0 | 446 | struct list_head iopoll_list; |
78076bb6 JA |
447 | struct hlist_head *cancel_hash; |
448 | unsigned cancel_hash_bits; | |
e94f141b | 449 | bool poll_multi_file; |
2b188cc1 | 450 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 451 | |
269bbe5f BM |
452 | struct delayed_work rsrc_put_work; |
453 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
454 | struct list_head rsrc_ref_list; |
455 | spinlock_t rsrc_ref_lock; | |
a7f0ed5a | 456 | struct io_rsrc_node *rsrc_node; |
b895c9a6 | 457 | struct io_rsrc_node *rsrc_backup_node; |
4a38aed2 | 458 | |
21b55dbc | 459 | struct io_restriction restrictions; |
3c1a2ead | 460 | |
7c25c0d1 JA |
461 | /* exit task_work */ |
462 | struct callback_head *exit_task_work; | |
463 | ||
3c1a2ead JA |
464 | /* Keep this last, we don't need it for the fast path */ |
465 | struct work_struct exit_work; | |
13bf43f5 | 466 | struct list_head tctx_list; |
2b188cc1 JA |
467 | }; |
468 | ||
53e043b2 SM |
469 | struct io_uring_task { |
470 | /* submission side */ | |
471 | struct xarray xa; | |
472 | struct wait_queue_head wait; | |
ee53fb2b SM |
473 | const struct io_ring_ctx *last; |
474 | struct io_wq *io_wq; | |
53e043b2 | 475 | struct percpu_counter inflight; |
b303fe2e | 476 | atomic_t inflight_tracked; |
53e043b2 | 477 | atomic_t in_idle; |
53e043b2 SM |
478 | |
479 | spinlock_t task_lock; | |
480 | struct io_wq_work_list task_list; | |
481 | unsigned long task_state; | |
482 | struct callback_head task_work; | |
483 | }; | |
484 | ||
09bb8394 JA |
485 | /* |
486 | * First field must be the file pointer in all the | |
487 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
488 | */ | |
221c5eb2 JA |
489 | struct io_poll_iocb { |
490 | struct file *file; | |
018043be | 491 | struct wait_queue_head *head; |
221c5eb2 | 492 | __poll_t events; |
8c838788 | 493 | bool done; |
221c5eb2 | 494 | bool canceled; |
9d805892 PB |
495 | struct wait_queue_entry wait; |
496 | }; | |
497 | ||
498 | struct io_poll_update { | |
499 | struct file *file; | |
500 | u64 old_user_data; | |
501 | u64 new_user_data; | |
502 | __poll_t events; | |
b69de288 JA |
503 | bool update_events; |
504 | bool update_user_data; | |
221c5eb2 JA |
505 | }; |
506 | ||
b5dba59e JA |
507 | struct io_close { |
508 | struct file *file; | |
b5dba59e JA |
509 | int fd; |
510 | }; | |
511 | ||
ad8a48ac JA |
512 | struct io_timeout_data { |
513 | struct io_kiocb *req; | |
514 | struct hrtimer timer; | |
515 | struct timespec64 ts; | |
516 | enum hrtimer_mode mode; | |
517 | }; | |
518 | ||
8ed8d3c3 JA |
519 | struct io_accept { |
520 | struct file *file; | |
521 | struct sockaddr __user *addr; | |
522 | int __user *addr_len; | |
523 | int flags; | |
09952e3e | 524 | unsigned long nofile; |
8ed8d3c3 JA |
525 | }; |
526 | ||
527 | struct io_sync { | |
528 | struct file *file; | |
529 | loff_t len; | |
530 | loff_t off; | |
531 | int flags; | |
d63d1b5e | 532 | int mode; |
8ed8d3c3 JA |
533 | }; |
534 | ||
fbf23849 JA |
535 | struct io_cancel { |
536 | struct file *file; | |
537 | u64 addr; | |
538 | }; | |
539 | ||
b29472ee JA |
540 | struct io_timeout { |
541 | struct file *file; | |
bfe68a22 PB |
542 | u32 off; |
543 | u32 target_seq; | |
135fcde8 | 544 | struct list_head list; |
90cd7e42 PB |
545 | /* head of the link, used by linked timeouts only */ |
546 | struct io_kiocb *head; | |
b29472ee JA |
547 | }; |
548 | ||
0bdf7a2d PB |
549 | struct io_timeout_rem { |
550 | struct file *file; | |
551 | u64 addr; | |
9c8e11b3 PB |
552 | |
553 | /* timeout update */ | |
554 | struct timespec64 ts; | |
555 | u32 flags; | |
0bdf7a2d PB |
556 | }; |
557 | ||
9adbd45d JA |
558 | struct io_rw { |
559 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
560 | struct kiocb kiocb; | |
561 | u64 addr; | |
562 | u64 len; | |
563 | }; | |
564 | ||
3fbb51c1 JA |
565 | struct io_connect { |
566 | struct file *file; | |
567 | struct sockaddr __user *addr; | |
568 | int addr_len; | |
569 | }; | |
570 | ||
e47293fd JA |
571 | struct io_sr_msg { |
572 | struct file *file; | |
fddaface | 573 | union { |
4af3417a PB |
574 | struct compat_msghdr __user *umsg_compat; |
575 | struct user_msghdr __user *umsg; | |
576 | void __user *buf; | |
fddaface | 577 | }; |
e47293fd | 578 | int msg_flags; |
bcda7baa | 579 | int bgid; |
fddaface | 580 | size_t len; |
bcda7baa | 581 | struct io_buffer *kbuf; |
e47293fd JA |
582 | }; |
583 | ||
15b71abe JA |
584 | struct io_open { |
585 | struct file *file; | |
586 | int dfd; | |
15b71abe | 587 | struct filename *filename; |
c12cedf2 | 588 | struct open_how how; |
4022e7af | 589 | unsigned long nofile; |
15b71abe JA |
590 | }; |
591 | ||
269bbe5f | 592 | struct io_rsrc_update { |
05f3fb3c JA |
593 | struct file *file; |
594 | u64 arg; | |
595 | u32 nr_args; | |
596 | u32 offset; | |
597 | }; | |
598 | ||
4840e418 JA |
599 | struct io_fadvise { |
600 | struct file *file; | |
601 | u64 offset; | |
602 | u32 len; | |
603 | u32 advice; | |
604 | }; | |
605 | ||
c1ca757b JA |
606 | struct io_madvise { |
607 | struct file *file; | |
608 | u64 addr; | |
609 | u32 len; | |
610 | u32 advice; | |
611 | }; | |
612 | ||
3e4827b0 JA |
613 | struct io_epoll { |
614 | struct file *file; | |
615 | int epfd; | |
616 | int op; | |
617 | int fd; | |
618 | struct epoll_event event; | |
e47293fd JA |
619 | }; |
620 | ||
7d67af2c PB |
621 | struct io_splice { |
622 | struct file *file_out; | |
623 | struct file *file_in; | |
624 | loff_t off_out; | |
625 | loff_t off_in; | |
626 | u64 len; | |
627 | unsigned int flags; | |
628 | }; | |
629 | ||
ddf0322d JA |
630 | struct io_provide_buf { |
631 | struct file *file; | |
632 | __u64 addr; | |
38134ada | 633 | __u32 len; |
ddf0322d JA |
634 | __u32 bgid; |
635 | __u16 nbufs; | |
636 | __u16 bid; | |
637 | }; | |
638 | ||
1d9e1288 BM |
639 | struct io_statx { |
640 | struct file *file; | |
641 | int dfd; | |
642 | unsigned int mask; | |
643 | unsigned int flags; | |
e62753e4 | 644 | const char __user *filename; |
1d9e1288 BM |
645 | struct statx __user *buffer; |
646 | }; | |
647 | ||
36f4fa68 JA |
648 | struct io_shutdown { |
649 | struct file *file; | |
650 | int how; | |
651 | }; | |
652 | ||
80a261fd JA |
653 | struct io_rename { |
654 | struct file *file; | |
655 | int old_dfd; | |
656 | int new_dfd; | |
657 | struct filename *oldpath; | |
658 | struct filename *newpath; | |
659 | int flags; | |
660 | }; | |
661 | ||
14a1143b JA |
662 | struct io_unlink { |
663 | struct file *file; | |
664 | int dfd; | |
665 | int flags; | |
666 | struct filename *filename; | |
667 | }; | |
668 | ||
3ca405eb PB |
669 | struct io_completion { |
670 | struct file *file; | |
671 | struct list_head list; | |
8c3f9cd1 | 672 | u32 cflags; |
3ca405eb PB |
673 | }; |
674 | ||
f499a021 JA |
675 | struct io_async_connect { |
676 | struct sockaddr_storage address; | |
677 | }; | |
678 | ||
03b1230c JA |
679 | struct io_async_msghdr { |
680 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
681 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
682 | struct iovec *free_iov; | |
03b1230c JA |
683 | struct sockaddr __user *uaddr; |
684 | struct msghdr msg; | |
b537916c | 685 | struct sockaddr_storage addr; |
03b1230c JA |
686 | }; |
687 | ||
f67676d1 JA |
688 | struct io_async_rw { |
689 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
690 | const struct iovec *free_iovec; |
691 | struct iov_iter iter; | |
227c0c96 | 692 | size_t bytes_done; |
bcf5a063 | 693 | struct wait_page_queue wpq; |
f67676d1 JA |
694 | }; |
695 | ||
6b47ee6e PB |
696 | enum { |
697 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
698 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
699 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
700 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
701 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 702 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 703 | |
6b47ee6e PB |
704 | REQ_F_FAIL_LINK_BIT, |
705 | REQ_F_INFLIGHT_BIT, | |
706 | REQ_F_CUR_POS_BIT, | |
707 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 708 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 709 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 710 | REQ_F_POLLED_BIT, |
bcda7baa | 711 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 712 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 713 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 714 | REQ_F_REISSUE_BIT, |
8c130827 | 715 | REQ_F_DONT_REISSUE_BIT, |
7b29f92d JA |
716 | /* keep async read/write and isreg together and in order */ |
717 | REQ_F_ASYNC_READ_BIT, | |
718 | REQ_F_ASYNC_WRITE_BIT, | |
719 | REQ_F_ISREG_BIT, | |
84557871 JA |
720 | |
721 | /* not a real bit, just to check we're not overflowing the space */ | |
722 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
723 | }; |
724 | ||
725 | enum { | |
726 | /* ctx owns file */ | |
727 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
728 | /* drain existing IO first */ | |
729 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
730 | /* linked sqes */ | |
731 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
732 | /* doesn't sever on completion < 0 */ | |
733 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
734 | /* IOSQE_ASYNC */ | |
735 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
736 | /* IOSQE_BUFFER_SELECT */ |
737 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 738 | |
6b47ee6e PB |
739 | /* fail rest of links */ |
740 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 741 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
742 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
743 | /* read/write uses file position */ | |
744 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
745 | /* must not punt to workers */ | |
746 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 747 | /* has or had linked timeout */ |
6b47ee6e | 748 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
749 | /* needs cleanup */ |
750 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
751 | /* already went through poll handler */ |
752 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
753 | /* buffer already selected */ |
754 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
755 | /* linked timeout is active, i.e. prepared by link's head */ |
756 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
757 | /* completion is deferred through io_comp_state */ |
758 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
759 | /* caller should reissue async */ |
760 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
761 | /* don't attempt request reissue, see io_rw_reissue() */ |
762 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
7b29f92d JA |
763 | /* supports async reads */ |
764 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
765 | /* supports async writes */ | |
766 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
767 | /* regular file */ | |
768 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
769 | }; |
770 | ||
771 | struct async_poll { | |
772 | struct io_poll_iocb poll; | |
807abcb0 | 773 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
774 | }; |
775 | ||
7cbf1722 JA |
776 | struct io_task_work { |
777 | struct io_wq_work_node node; | |
778 | task_work_func_t func; | |
779 | }; | |
780 | ||
09bb8394 JA |
781 | /* |
782 | * NOTE! Each of the iocb union members has the file pointer | |
783 | * as the first entry in their struct definition. So you can | |
784 | * access the file pointer through any of the sub-structs, | |
785 | * or directly as just 'ki_filp' in this struct. | |
786 | */ | |
2b188cc1 | 787 | struct io_kiocb { |
221c5eb2 | 788 | union { |
09bb8394 | 789 | struct file *file; |
9adbd45d | 790 | struct io_rw rw; |
221c5eb2 | 791 | struct io_poll_iocb poll; |
9d805892 | 792 | struct io_poll_update poll_update; |
8ed8d3c3 JA |
793 | struct io_accept accept; |
794 | struct io_sync sync; | |
fbf23849 | 795 | struct io_cancel cancel; |
b29472ee | 796 | struct io_timeout timeout; |
0bdf7a2d | 797 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 798 | struct io_connect connect; |
e47293fd | 799 | struct io_sr_msg sr_msg; |
15b71abe | 800 | struct io_open open; |
b5dba59e | 801 | struct io_close close; |
269bbe5f | 802 | struct io_rsrc_update rsrc_update; |
4840e418 | 803 | struct io_fadvise fadvise; |
c1ca757b | 804 | struct io_madvise madvise; |
3e4827b0 | 805 | struct io_epoll epoll; |
7d67af2c | 806 | struct io_splice splice; |
ddf0322d | 807 | struct io_provide_buf pbuf; |
1d9e1288 | 808 | struct io_statx statx; |
36f4fa68 | 809 | struct io_shutdown shutdown; |
80a261fd | 810 | struct io_rename rename; |
14a1143b | 811 | struct io_unlink unlink; |
3ca405eb PB |
812 | /* use only after cleaning per-op data, see io_clean_op() */ |
813 | struct io_completion compl; | |
221c5eb2 | 814 | }; |
2b188cc1 | 815 | |
e8c2bc1f JA |
816 | /* opcode allocated if it needs to store data for async defer */ |
817 | void *async_data; | |
d625c6ee | 818 | u8 opcode; |
65a6543d XW |
819 | /* polled IO has completed */ |
820 | u8 iopoll_completed; | |
2b188cc1 | 821 | |
4f4eeba8 | 822 | u16 buf_index; |
9cf7c104 | 823 | u32 result; |
4f4eeba8 | 824 | |
010e8e6b PB |
825 | struct io_ring_ctx *ctx; |
826 | unsigned int flags; | |
abc54d63 | 827 | atomic_t refs; |
010e8e6b PB |
828 | struct task_struct *task; |
829 | u64 user_data; | |
d7718a9d | 830 | |
f2f87370 | 831 | struct io_kiocb *link; |
269bbe5f | 832 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 833 | |
b303fe2e | 834 | /* used with ctx->iopoll_list with reads/writes */ |
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; | |
eae071c9 PB |
844 | /* store used ubuf, so we can prevent reloading */ |
845 | struct io_mapped_ubuf *imu; | |
2b188cc1 | 846 | }; |
05589553 | 847 | |
13bf43f5 PB |
848 | struct io_tctx_node { |
849 | struct list_head ctx_node; | |
850 | struct task_struct *task; | |
13bf43f5 PB |
851 | struct io_ring_ctx *ctx; |
852 | }; | |
853 | ||
27dc8338 PB |
854 | struct io_defer_entry { |
855 | struct list_head list; | |
856 | struct io_kiocb *req; | |
9cf7c104 | 857 | u32 seq; |
2b188cc1 JA |
858 | }; |
859 | ||
d3656344 | 860 | struct io_op_def { |
d3656344 JA |
861 | /* needs req->file assigned */ |
862 | unsigned needs_file : 1; | |
d3656344 JA |
863 | /* hash wq insertion if file is a regular file */ |
864 | unsigned hash_reg_file : 1; | |
865 | /* unbound wq insertion if file is a non-regular file */ | |
866 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
867 | /* opcode is not supported by this kernel */ |
868 | unsigned not_supported : 1; | |
8a72758c JA |
869 | /* set if opcode supports polled "wait" */ |
870 | unsigned pollin : 1; | |
871 | unsigned pollout : 1; | |
bcda7baa JA |
872 | /* op supports buffer selection */ |
873 | unsigned buffer_select : 1; | |
26f0505a PB |
874 | /* do prep async if is going to be punted */ |
875 | unsigned needs_async_setup : 1; | |
27926b68 JA |
876 | /* should block plug */ |
877 | unsigned plug : 1; | |
e8c2bc1f JA |
878 | /* size of async data needed, if any */ |
879 | unsigned short async_size; | |
d3656344 JA |
880 | }; |
881 | ||
0918682b | 882 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
883 | [IORING_OP_NOP] = {}, |
884 | [IORING_OP_READV] = { | |
d3656344 JA |
885 | .needs_file = 1, |
886 | .unbound_nonreg_file = 1, | |
8a72758c | 887 | .pollin = 1, |
4d954c25 | 888 | .buffer_select = 1, |
26f0505a | 889 | .needs_async_setup = 1, |
27926b68 | 890 | .plug = 1, |
e8c2bc1f | 891 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 892 | }, |
0463b6c5 | 893 | [IORING_OP_WRITEV] = { |
d3656344 JA |
894 | .needs_file = 1, |
895 | .hash_reg_file = 1, | |
896 | .unbound_nonreg_file = 1, | |
8a72758c | 897 | .pollout = 1, |
26f0505a | 898 | .needs_async_setup = 1, |
27926b68 | 899 | .plug = 1, |
e8c2bc1f | 900 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 901 | }, |
0463b6c5 | 902 | [IORING_OP_FSYNC] = { |
d3656344 JA |
903 | .needs_file = 1, |
904 | }, | |
0463b6c5 | 905 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
906 | .needs_file = 1, |
907 | .unbound_nonreg_file = 1, | |
8a72758c | 908 | .pollin = 1, |
27926b68 | 909 | .plug = 1, |
e8c2bc1f | 910 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 911 | }, |
0463b6c5 | 912 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
913 | .needs_file = 1, |
914 | .hash_reg_file = 1, | |
915 | .unbound_nonreg_file = 1, | |
8a72758c | 916 | .pollout = 1, |
27926b68 | 917 | .plug = 1, |
e8c2bc1f | 918 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 919 | }, |
0463b6c5 | 920 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
921 | .needs_file = 1, |
922 | .unbound_nonreg_file = 1, | |
923 | }, | |
0463b6c5 PB |
924 | [IORING_OP_POLL_REMOVE] = {}, |
925 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
926 | .needs_file = 1, |
927 | }, | |
0463b6c5 | 928 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
929 | .needs_file = 1, |
930 | .unbound_nonreg_file = 1, | |
8a72758c | 931 | .pollout = 1, |
26f0505a | 932 | .needs_async_setup = 1, |
e8c2bc1f | 933 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 934 | }, |
0463b6c5 | 935 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
936 | .needs_file = 1, |
937 | .unbound_nonreg_file = 1, | |
8a72758c | 938 | .pollin = 1, |
52de1fe1 | 939 | .buffer_select = 1, |
26f0505a | 940 | .needs_async_setup = 1, |
e8c2bc1f | 941 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 942 | }, |
0463b6c5 | 943 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 944 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 945 | }, |
9c8e11b3 PB |
946 | [IORING_OP_TIMEOUT_REMOVE] = { |
947 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 948 | }, |
0463b6c5 | 949 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
950 | .needs_file = 1, |
951 | .unbound_nonreg_file = 1, | |
8a72758c | 952 | .pollin = 1, |
d3656344 | 953 | }, |
0463b6c5 PB |
954 | [IORING_OP_ASYNC_CANCEL] = {}, |
955 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 956 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 957 | }, |
0463b6c5 | 958 | [IORING_OP_CONNECT] = { |
d3656344 JA |
959 | .needs_file = 1, |
960 | .unbound_nonreg_file = 1, | |
8a72758c | 961 | .pollout = 1, |
26f0505a | 962 | .needs_async_setup = 1, |
e8c2bc1f | 963 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 964 | }, |
0463b6c5 | 965 | [IORING_OP_FALLOCATE] = { |
d3656344 | 966 | .needs_file = 1, |
d3656344 | 967 | }, |
44526bed JA |
968 | [IORING_OP_OPENAT] = {}, |
969 | [IORING_OP_CLOSE] = {}, | |
970 | [IORING_OP_FILES_UPDATE] = {}, | |
971 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 972 | [IORING_OP_READ] = { |
3a6820f2 JA |
973 | .needs_file = 1, |
974 | .unbound_nonreg_file = 1, | |
8a72758c | 975 | .pollin = 1, |
bcda7baa | 976 | .buffer_select = 1, |
27926b68 | 977 | .plug = 1, |
e8c2bc1f | 978 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 979 | }, |
0463b6c5 | 980 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
981 | .needs_file = 1, |
982 | .unbound_nonreg_file = 1, | |
8a72758c | 983 | .pollout = 1, |
27926b68 | 984 | .plug = 1, |
e8c2bc1f | 985 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 986 | }, |
0463b6c5 | 987 | [IORING_OP_FADVISE] = { |
4840e418 | 988 | .needs_file = 1, |
c1ca757b | 989 | }, |
44526bed | 990 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 991 | [IORING_OP_SEND] = { |
fddaface JA |
992 | .needs_file = 1, |
993 | .unbound_nonreg_file = 1, | |
8a72758c | 994 | .pollout = 1, |
fddaface | 995 | }, |
0463b6c5 | 996 | [IORING_OP_RECV] = { |
fddaface JA |
997 | .needs_file = 1, |
998 | .unbound_nonreg_file = 1, | |
8a72758c | 999 | .pollin = 1, |
bcda7baa | 1000 | .buffer_select = 1, |
fddaface | 1001 | }, |
0463b6c5 | 1002 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1003 | }, |
3e4827b0 JA |
1004 | [IORING_OP_EPOLL_CTL] = { |
1005 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1006 | }, |
7d67af2c PB |
1007 | [IORING_OP_SPLICE] = { |
1008 | .needs_file = 1, | |
1009 | .hash_reg_file = 1, | |
1010 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1011 | }, |
1012 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1013 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1014 | [IORING_OP_TEE] = { |
1015 | .needs_file = 1, | |
1016 | .hash_reg_file = 1, | |
1017 | .unbound_nonreg_file = 1, | |
1018 | }, | |
36f4fa68 JA |
1019 | [IORING_OP_SHUTDOWN] = { |
1020 | .needs_file = 1, | |
1021 | }, | |
44526bed JA |
1022 | [IORING_OP_RENAMEAT] = {}, |
1023 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1024 | }; |
1025 | ||
7a612350 | 1026 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1027 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1028 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1029 | struct task_struct *task, | |
1030 | struct files_struct *files); | |
734551df | 1031 | static void io_uring_cancel_sqpoll(struct io_sq_data *sqd); |
b895c9a6 | 1032 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1033 | |
d4d19c19 PB |
1034 | static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1035 | long res, unsigned int cflags); | |
ec9c02ad | 1036 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1037 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1038 | static void io_dismantle_req(struct io_kiocb *req); |
1039 | static void io_put_task(struct task_struct *task, int nr); | |
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); | |
fdecb662 | 1042 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 1043 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 | 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 | 1081 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1082 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1083 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1084 | } |
1085 | } | |
1086 | ||
f70865db PB |
1087 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1088 | { | |
1089 | bool got = percpu_ref_tryget(ref); | |
1090 | ||
1091 | /* already at zero, wait for ->release() */ | |
1092 | if (!got) | |
1093 | wait_for_completion(compl); | |
1094 | percpu_ref_resurrect(ref); | |
1095 | if (got) | |
1096 | percpu_ref_put(ref); | |
1097 | } | |
1098 | ||
08d23634 PB |
1099 | static bool io_match_task(struct io_kiocb *head, |
1100 | struct task_struct *task, | |
1101 | struct files_struct *files) | |
1102 | { | |
1103 | struct io_kiocb *req; | |
1104 | ||
68207680 | 1105 | if (task && head->task != task) |
08d23634 PB |
1106 | return false; |
1107 | if (!files) | |
1108 | return true; | |
1109 | ||
1110 | io_for_each_link(req, head) { | |
b05a1bcd | 1111 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1112 | return true; |
08d23634 PB |
1113 | } |
1114 | return false; | |
1115 | } | |
1116 | ||
c40f6379 JA |
1117 | static inline void req_set_fail_links(struct io_kiocb *req) |
1118 | { | |
e4335ed3 | 1119 | if (req->flags & REQ_F_LINK) |
c40f6379 JA |
1120 | req->flags |= REQ_F_FAIL_LINK; |
1121 | } | |
4a38aed2 | 1122 | |
2b188cc1 JA |
1123 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1124 | { | |
1125 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1126 | ||
0f158b4c | 1127 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1128 | } |
1129 | ||
8eb7e2d0 PB |
1130 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1131 | { | |
1132 | return !req->timeout.off; | |
1133 | } | |
1134 | ||
2b188cc1 JA |
1135 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1136 | { | |
1137 | struct io_ring_ctx *ctx; | |
78076bb6 | 1138 | int hash_bits; |
2b188cc1 JA |
1139 | |
1140 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1141 | if (!ctx) | |
1142 | return NULL; | |
1143 | ||
78076bb6 JA |
1144 | /* |
1145 | * Use 5 bits less than the max cq entries, that should give us around | |
1146 | * 32 entries per hash list if totally full and uniformly spread. | |
1147 | */ | |
1148 | hash_bits = ilog2(p->cq_entries); | |
1149 | hash_bits -= 5; | |
1150 | if (hash_bits <= 0) | |
1151 | hash_bits = 1; | |
1152 | ctx->cancel_hash_bits = hash_bits; | |
1153 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1154 | GFP_KERNEL); | |
1155 | if (!ctx->cancel_hash) | |
1156 | goto err; | |
1157 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1158 | ||
21482896 | 1159 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1160 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1161 | goto err; | |
2b188cc1 JA |
1162 | |
1163 | ctx->flags = p->flags; | |
90554200 | 1164 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1165 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1166 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1167 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1168 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1169 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1170 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1171 | mutex_init(&ctx->uring_lock); |
1172 | init_waitqueue_head(&ctx->wait); | |
1173 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1174 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1175 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1176 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1177 | spin_lock_init(&ctx->rsrc_ref_lock); |
1178 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1179 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1180 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1181 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1182 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1183 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1184 | return ctx; |
206aefde | 1185 | err: |
78076bb6 | 1186 | kfree(ctx->cancel_hash); |
206aefde JA |
1187 | kfree(ctx); |
1188 | return NULL; | |
2b188cc1 JA |
1189 | } |
1190 | ||
9cf7c104 | 1191 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1192 | { |
2bc9930e JA |
1193 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1194 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1195 | |
9cf7c104 | 1196 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1197 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1198 | } |
de0617e4 | 1199 | |
9d858b21 | 1200 | return false; |
de0617e4 JA |
1201 | } |
1202 | ||
ce3d5aae PB |
1203 | static void io_req_track_inflight(struct io_kiocb *req) |
1204 | { | |
ce3d5aae | 1205 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1206 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1207 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1208 | } |
1209 | } | |
1210 | ||
1e6fa521 JA |
1211 | static void io_prep_async_work(struct io_kiocb *req) |
1212 | { | |
1213 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1214 | struct io_ring_ctx *ctx = req->ctx; |
1215 | ||
003e8dcc JA |
1216 | if (!req->work.creds) |
1217 | req->work.creds = get_current_cred(); | |
1218 | ||
e1d675df PB |
1219 | req->work.list.next = NULL; |
1220 | req->work.flags = 0; | |
feaadc4f PB |
1221 | if (req->flags & REQ_F_FORCE_ASYNC) |
1222 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1223 | ||
1e6fa521 JA |
1224 | if (req->flags & REQ_F_ISREG) { |
1225 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1226 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1227 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1228 | if (def->unbound_nonreg_file) |
1229 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1230 | } | |
e1d675df PB |
1231 | |
1232 | switch (req->opcode) { | |
1233 | case IORING_OP_SPLICE: | |
1234 | case IORING_OP_TEE: | |
e1d675df PB |
1235 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1236 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1237 | break; | |
1238 | } | |
561fb04a | 1239 | } |
cccf0ee8 | 1240 | |
cbdcb435 | 1241 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1242 | { |
cbdcb435 | 1243 | struct io_kiocb *cur; |
54a91f3b | 1244 | |
f2f87370 PB |
1245 | io_for_each_link(cur, req) |
1246 | io_prep_async_work(cur); | |
561fb04a JA |
1247 | } |
1248 | ||
ebf93667 | 1249 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1250 | { |
a197f664 | 1251 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1252 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1253 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1254 | |
3bfe6106 JA |
1255 | BUG_ON(!tctx); |
1256 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1257 | |
cbdcb435 PB |
1258 | /* init ->work of the whole link before punting */ |
1259 | io_prep_async_link(req); | |
d07f1e8a PB |
1260 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1261 | &req->work, req->flags); | |
ebf93667 | 1262 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1263 | if (link) |
1264 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1265 | } |
1266 | ||
1ee4160c | 1267 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1268 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1269 | { |
e8c2bc1f | 1270 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1271 | |
fd9c7bc5 | 1272 | if (hrtimer_try_to_cancel(&io->timer) != -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); |
d4d19c19 | 1276 | io_cqring_fill_event(req->ctx, req->user_data, status, 0); |
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 | ||
8d13326e | 1354 | static inline struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 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 | { | |
44c769de | 1373 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1374 | return false; |
7e55a19c SG |
1375 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1376 | return false; | |
44c769de | 1377 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1378 | } |
1379 | ||
b41e9852 | 1380 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1381 | { |
b1445e59 PB |
1382 | /* see waitqueue_active() comment */ |
1383 | smp_mb(); | |
1384 | ||
1d7bb1d5 JA |
1385 | if (waitqueue_active(&ctx->wait)) |
1386 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1387 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1388 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1389 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1390 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1391 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1392 | wake_up_interruptible(&ctx->cq_wait); |
1393 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1394 | } | |
1d7bb1d5 JA |
1395 | } |
1396 | ||
80c18e4a PB |
1397 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1398 | { | |
b1445e59 PB |
1399 | /* see waitqueue_active() comment */ |
1400 | smp_mb(); | |
1401 | ||
80c18e4a PB |
1402 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1403 | if (waitqueue_active(&ctx->wait)) | |
1404 | wake_up(&ctx->wait); | |
1405 | } | |
1406 | if (io_should_trigger_evfd(ctx)) | |
1407 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1408 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1409 | wake_up_interruptible(&ctx->cq_wait); |
1410 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1411 | } | |
80c18e4a PB |
1412 | } |
1413 | ||
c4a2ed72 | 1414 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1415 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 JA |
1416 | { |
1417 | struct io_rings *rings = ctx->rings; | |
1d7bb1d5 | 1418 | unsigned long flags; |
b18032bb | 1419 | bool all_flushed, posted; |
1d7bb1d5 | 1420 | |
e23de15f PB |
1421 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1422 | return false; | |
1d7bb1d5 | 1423 | |
b18032bb | 1424 | posted = false; |
1d7bb1d5 | 1425 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae PB |
1426 | while (!list_empty(&ctx->cq_overflow_list)) { |
1427 | struct io_uring_cqe *cqe = io_get_cqring(ctx); | |
1428 | struct io_overflow_cqe *ocqe; | |
e6c8aa9a | 1429 | |
1d7bb1d5 JA |
1430 | if (!cqe && !force) |
1431 | break; | |
6c2450ae PB |
1432 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1433 | struct io_overflow_cqe, list); | |
1434 | if (cqe) | |
1435 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1436 | else | |
1d7bb1d5 | 1437 | WRITE_ONCE(ctx->rings->cq_overflow, |
6c2450ae | 1438 | ++ctx->cached_cq_overflow); |
b18032bb | 1439 | posted = true; |
6c2450ae PB |
1440 | list_del(&ocqe->list); |
1441 | kfree(ocqe); | |
1d7bb1d5 JA |
1442 | } |
1443 | ||
09e88404 PB |
1444 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1445 | if (all_flushed) { | |
1446 | clear_bit(0, &ctx->sq_check_overflow); | |
1447 | clear_bit(0, &ctx->cq_check_overflow); | |
1448 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1449 | } | |
46930143 | 1450 | |
b18032bb JA |
1451 | if (posted) |
1452 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1453 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1454 | if (posted) |
1455 | io_cqring_ev_posted(ctx); | |
09e88404 | 1456 | return all_flushed; |
1d7bb1d5 JA |
1457 | } |
1458 | ||
6c2450ae | 1459 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1460 | { |
ca0a2651 JA |
1461 | bool ret = true; |
1462 | ||
6c503150 PB |
1463 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1464 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1465 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1466 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1467 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1468 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1469 | mutex_unlock(&ctx->uring_lock); | |
1470 | } | |
ca0a2651 JA |
1471 | |
1472 | return ret; | |
6c503150 PB |
1473 | } |
1474 | ||
abc54d63 JA |
1475 | /* |
1476 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1477 | * see commit f958d7b528b1 for details. | |
1478 | */ | |
1479 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1480 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1481 | ||
de9b4cca JA |
1482 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1483 | { | |
abc54d63 | 1484 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1485 | } |
1486 | ||
1487 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1488 | { | |
abc54d63 JA |
1489 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1490 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1491 | } |
1492 | ||
1493 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1494 | { | |
abc54d63 JA |
1495 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1496 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1497 | } |
1498 | ||
1499 | static inline void req_ref_put(struct io_kiocb *req) | |
1500 | { | |
abc54d63 | 1501 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1502 | } |
1503 | ||
1504 | static inline void req_ref_get(struct io_kiocb *req) | |
1505 | { | |
abc54d63 JA |
1506 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1507 | atomic_inc(&req->refs); | |
de9b4cca JA |
1508 | } |
1509 | ||
d4d19c19 PB |
1510 | static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data, |
1511 | long res, unsigned int cflags) | |
2b188cc1 | 1512 | { |
cce4b8b0 | 1513 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1514 | |
cce4b8b0 PB |
1515 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1516 | if (!ocqe) { | |
1517 | /* | |
1518 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1519 | * or cannot allocate an overflow entry, then we need to drop it | |
1520 | * on the floor. | |
1521 | */ | |
1522 | WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow); | |
1523 | return false; | |
2b188cc1 | 1524 | } |
cce4b8b0 PB |
1525 | if (list_empty(&ctx->cq_overflow_list)) { |
1526 | set_bit(0, &ctx->sq_check_overflow); | |
1527 | set_bit(0, &ctx->cq_check_overflow); | |
1528 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; | |
1529 | } | |
d4d19c19 | 1530 | ocqe->cqe.user_data = user_data; |
cce4b8b0 PB |
1531 | ocqe->cqe.res = res; |
1532 | ocqe->cqe.flags = cflags; | |
1533 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1534 | return true; | |
2b188cc1 JA |
1535 | } |
1536 | ||
d4d19c19 PB |
1537 | static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1538 | long res, unsigned int cflags) | |
8d13326e | 1539 | { |
8d13326e PB |
1540 | struct io_uring_cqe *cqe; |
1541 | ||
d4d19c19 | 1542 | trace_io_uring_complete(ctx, user_data, res, cflags); |
8d13326e PB |
1543 | |
1544 | /* | |
1545 | * If we can't get a cq entry, userspace overflowed the | |
1546 | * submission (by quite a lot). Increment the overflow count in | |
1547 | * the ring. | |
1548 | */ | |
1549 | cqe = io_get_cqring(ctx); | |
1550 | if (likely(cqe)) { | |
d4d19c19 | 1551 | WRITE_ONCE(cqe->user_data, user_data); |
8d13326e PB |
1552 | WRITE_ONCE(cqe->res, res); |
1553 | WRITE_ONCE(cqe->flags, cflags); | |
1554 | return true; | |
1555 | } | |
d4d19c19 | 1556 | return io_cqring_event_overflow(ctx, user_data, res, cflags); |
8d13326e PB |
1557 | } |
1558 | ||
1559 | /* not as hot to bloat with inlining */ | |
d4d19c19 PB |
1560 | static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data, |
1561 | long res, unsigned int cflags) | |
8d13326e | 1562 | { |
d4d19c19 | 1563 | return __io_cqring_fill_event(ctx, user_data, res, cflags); |
8d13326e PB |
1564 | } |
1565 | ||
7a612350 PB |
1566 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1567 | unsigned int cflags) | |
2b188cc1 | 1568 | { |
78e19bbe | 1569 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1570 | unsigned long flags; |
1571 | ||
1572 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
d4d19c19 | 1573 | __io_cqring_fill_event(ctx, req->user_data, res, cflags); |
c7dae4ba JA |
1574 | /* |
1575 | * If we're the last reference to this request, add to our locked | |
1576 | * free_list cache. | |
1577 | */ | |
de9b4cca | 1578 | if (req_ref_put_and_test(req)) { |
c7dae4ba JA |
1579 | struct io_comp_state *cs = &ctx->submit_state.comp; |
1580 | ||
7a612350 PB |
1581 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1582 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1583 | io_disarm_next(req); | |
1584 | if (req->link) { | |
1585 | io_req_task_queue(req->link); | |
1586 | req->link = NULL; | |
1587 | } | |
1588 | } | |
c7dae4ba JA |
1589 | io_dismantle_req(req); |
1590 | io_put_task(req->task, 1); | |
1591 | list_add(&req->compl.list, &cs->locked_free_list); | |
1592 | cs->locked_free_nr++; | |
180f829f PB |
1593 | } else { |
1594 | if (!percpu_ref_tryget(&ctx->refs)) | |
1595 | req = NULL; | |
1596 | } | |
7a612350 | 1597 | io_commit_cqring(ctx); |
2b188cc1 | 1598 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1599 | |
180f829f PB |
1600 | if (req) { |
1601 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1602 | percpu_ref_put(&ctx->refs); |
180f829f | 1603 | } |
229a7b63 JA |
1604 | } |
1605 | ||
4e3d9ff9 JA |
1606 | static inline bool io_req_needs_clean(struct io_kiocb *req) |
1607 | { | |
75652a30 | 1608 | return req->flags & (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | |
3a0a6902 | 1609 | REQ_F_POLLED | REQ_F_INFLIGHT); |
4e3d9ff9 JA |
1610 | } |
1611 | ||
a38d68db | 1612 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1613 | unsigned int cflags) |
229a7b63 | 1614 | { |
4e3d9ff9 | 1615 | if (io_req_needs_clean(req)) |
68fb8979 | 1616 | io_clean_op(req); |
a38d68db PB |
1617 | req->result = res; |
1618 | req->compl.cflags = cflags; | |
e342c807 | 1619 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1620 | } |
1621 | ||
889fca73 PB |
1622 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1623 | long res, unsigned cflags) | |
bcda7baa | 1624 | { |
889fca73 PB |
1625 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1626 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1627 | else |
c7dae4ba | 1628 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1629 | } |
1630 | ||
a38d68db | 1631 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1632 | { |
889fca73 | 1633 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1634 | } |
1635 | ||
f41db273 PB |
1636 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1637 | { | |
1638 | req_set_fail_links(req); | |
1639 | io_put_req(req); | |
1640 | io_req_complete_post(req, res, 0); | |
1641 | } | |
1642 | ||
dac7a098 PB |
1643 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1644 | struct io_comp_state *cs) | |
1645 | { | |
1646 | spin_lock_irq(&ctx->completion_lock); | |
1647 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1648 | cs->locked_free_nr = 0; | |
1649 | spin_unlock_irq(&ctx->completion_lock); | |
1650 | } | |
1651 | ||
dd78f492 | 1652 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1653 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1654 | { |
c7dae4ba JA |
1655 | struct io_submit_state *state = &ctx->submit_state; |
1656 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1657 | int nr; |
0ddf92e8 | 1658 | |
c7dae4ba JA |
1659 | /* |
1660 | * If we have more than a batch's worth of requests in our IRQ side | |
1661 | * locked cache, grab the lock and move them over to our submission | |
1662 | * side cache. | |
1663 | */ | |
dac7a098 PB |
1664 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) |
1665 | io_flush_cached_locked_reqs(ctx, cs); | |
0ddf92e8 | 1666 | |
dd78f492 | 1667 | nr = state->free_reqs; |
c7dae4ba | 1668 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1669 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1670 | struct io_kiocb, compl.list); | |
1671 | ||
1b4c351f | 1672 | list_del(&req->compl.list); |
dd78f492 PB |
1673 | state->reqs[nr++] = req; |
1674 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1675 | break; |
1b4c351f JA |
1676 | } |
1677 | ||
dd78f492 PB |
1678 | state->free_reqs = nr; |
1679 | return nr != 0; | |
0ddf92e8 JA |
1680 | } |
1681 | ||
e5d1bc0a | 1682 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1683 | { |
e5d1bc0a PB |
1684 | struct io_submit_state *state = &ctx->submit_state; |
1685 | ||
1686 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1687 | ||
f6b6c7d6 | 1688 | if (!state->free_reqs) { |
291b2821 | 1689 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1690 | int ret; |
1691 | ||
c7dae4ba | 1692 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1693 | goto got_req; |
1694 | ||
bf019da7 PB |
1695 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1696 | state->reqs); | |
fd6fab2c JA |
1697 | |
1698 | /* | |
1699 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1700 | * retry single alloc to be on the safe side. | |
1701 | */ | |
1702 | if (unlikely(ret <= 0)) { | |
1703 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1704 | if (!state->reqs[0]) | |
3893f39f | 1705 | return NULL; |
fd6fab2c JA |
1706 | ret = 1; |
1707 | } | |
291b2821 | 1708 | state->free_reqs = ret; |
2b188cc1 | 1709 | } |
e5d1bc0a | 1710 | got_req: |
291b2821 PB |
1711 | state->free_reqs--; |
1712 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1713 | } |
1714 | ||
e1d767f0 | 1715 | static inline void io_put_file(struct file *file) |
8da11c19 | 1716 | { |
e1d767f0 | 1717 | if (file) |
8da11c19 PB |
1718 | fput(file); |
1719 | } | |
1720 | ||
4edf20f9 | 1721 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1722 | { |
094bae49 PB |
1723 | unsigned int flags = req->flags; |
1724 | ||
3a0a6902 PB |
1725 | if (io_req_needs_clean(req)) |
1726 | io_clean_op(req); | |
e1d767f0 PB |
1727 | if (!(flags & REQ_F_FIXED_FILE)) |
1728 | io_put_file(req->file); | |
269bbe5f BM |
1729 | if (req->fixed_rsrc_refs) |
1730 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1731 | if (req->async_data) |
1732 | kfree(req->async_data); | |
003e8dcc JA |
1733 | if (req->work.creds) { |
1734 | put_cred(req->work.creds); | |
1735 | req->work.creds = NULL; | |
1736 | } | |
e65ef56d JA |
1737 | } |
1738 | ||
b23fcf47 | 1739 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1740 | static inline void io_put_task(struct task_struct *task, int nr) |
1741 | { | |
1742 | struct io_uring_task *tctx = task->io_uring; | |
1743 | ||
1744 | percpu_counter_sub(&tctx->inflight, nr); | |
1745 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1746 | wake_up(&tctx->wait); | |
1747 | put_task_struct_many(task, nr); | |
1748 | } | |
1749 | ||
216578e5 | 1750 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1751 | { |
51a4cc11 | 1752 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1753 | |
216578e5 | 1754 | io_dismantle_req(req); |
7c660731 | 1755 | io_put_task(req->task, 1); |
c6ca97b3 | 1756 | |
3893f39f | 1757 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1758 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1759 | } |
1760 | ||
f2f87370 PB |
1761 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1762 | { | |
1763 | struct io_kiocb *nxt = req->link; | |
1764 | ||
1765 | req->link = nxt->link; | |
1766 | nxt->link = NULL; | |
1767 | } | |
1768 | ||
33cc89a9 PB |
1769 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1770 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1771 | { |
33cc89a9 | 1772 | struct io_kiocb *link = req->link; |
f2f87370 | 1773 | |
900fad45 PB |
1774 | /* |
1775 | * Can happen if a linked timeout fired and link had been like | |
1776 | * req -> link t-out -> link t-out [-> ...] | |
1777 | */ | |
c9abd7ad PB |
1778 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1779 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1780 | |
f2f87370 | 1781 | io_remove_next_linked(req); |
90cd7e42 | 1782 | link->timeout.head = NULL; |
fd9c7bc5 | 1783 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
d4d19c19 PB |
1784 | io_cqring_fill_event(link->ctx, link->user_data, |
1785 | -ECANCELED, 0); | |
33cc89a9 | 1786 | io_put_req_deferred(link, 1); |
d4729fbd | 1787 | return true; |
c9abd7ad PB |
1788 | } |
1789 | } | |
d4729fbd | 1790 | return false; |
7c86ffee PB |
1791 | } |
1792 | ||
d148ca4b | 1793 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1794 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1795 | { |
33cc89a9 | 1796 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1797 | |
f2f87370 | 1798 | req->link = NULL; |
f2f87370 PB |
1799 | while (link) { |
1800 | nxt = link->link; | |
1801 | link->link = NULL; | |
2665abfd | 1802 | |
f2f87370 | 1803 | trace_io_uring_fail_link(req, link); |
d4d19c19 | 1804 | io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0); |
1575f21a | 1805 | io_put_req_deferred(link, 2); |
f2f87370 | 1806 | link = nxt; |
9e645e11 | 1807 | } |
33cc89a9 | 1808 | } |
9e645e11 | 1809 | |
33cc89a9 PB |
1810 | static bool io_disarm_next(struct io_kiocb *req) |
1811 | __must_hold(&req->ctx->completion_lock) | |
1812 | { | |
1813 | bool posted = false; | |
1814 | ||
1815 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1816 | posted = io_kill_linked_timeout(req); | |
e4335ed3 PB |
1817 | if (unlikely((req->flags & REQ_F_FAIL_LINK) && |
1818 | !(req->flags & REQ_F_HARDLINK))) { | |
33cc89a9 PB |
1819 | posted |= (req->link != NULL); |
1820 | io_fail_links(req); | |
1821 | } | |
1822 | return posted; | |
9e645e11 JA |
1823 | } |
1824 | ||
3fa5e0f3 | 1825 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1826 | { |
33cc89a9 | 1827 | struct io_kiocb *nxt; |
944e58bf | 1828 | |
9e645e11 JA |
1829 | /* |
1830 | * If LINK is set, we have dependent requests in this chain. If we | |
1831 | * didn't fail this request, queue the first one up, moving any other | |
1832 | * dependencies to the next request. In case of failure, fail the rest | |
1833 | * of the chain. | |
1834 | */ | |
33cc89a9 PB |
1835 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1836 | struct io_ring_ctx *ctx = req->ctx; | |
1837 | unsigned long flags; | |
1838 | bool posted; | |
1839 | ||
1840 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1841 | posted = io_disarm_next(req); | |
1842 | if (posted) | |
1843 | io_commit_cqring(req->ctx); | |
1844 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1845 | if (posted) | |
1846 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1847 | } |
33cc89a9 PB |
1848 | nxt = req->link; |
1849 | req->link = NULL; | |
1850 | return nxt; | |
4d7dd462 | 1851 | } |
9e645e11 | 1852 | |
f2f87370 | 1853 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1854 | { |
cdbff982 | 1855 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1856 | return NULL; |
1857 | return __io_req_find_next(req); | |
1858 | } | |
1859 | ||
2c32395d PB |
1860 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1861 | { | |
1862 | if (!ctx) | |
1863 | return; | |
1864 | if (ctx->submit_state.comp.nr) { | |
1865 | mutex_lock(&ctx->uring_lock); | |
1866 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1867 | mutex_unlock(&ctx->uring_lock); | |
1868 | } | |
1869 | percpu_ref_put(&ctx->refs); | |
1870 | } | |
1871 | ||
7cbf1722 | 1872 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1873 | { |
65453d1e | 1874 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1875 | struct io_wq_work_list list; |
1876 | struct io_wq_work_node *node; | |
c2c4c83c | 1877 | |
7cbf1722 JA |
1878 | if (wq_list_empty(&tctx->task_list)) |
1879 | return false; | |
6200b0ae | 1880 | |
0b81e80c | 1881 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1882 | list = tctx->task_list; |
1883 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1884 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1885 | |
7cbf1722 JA |
1886 | node = list.first; |
1887 | while (node) { | |
1888 | struct io_wq_work_node *next = node->next; | |
1889 | struct io_kiocb *req; | |
0ba9c9ed | 1890 | |
7cbf1722 | 1891 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1892 | if (req->ctx != ctx) { |
1893 | ctx_flush_and_put(ctx); | |
1894 | ctx = req->ctx; | |
1895 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1896 | } |
65453d1e | 1897 | |
2c32395d PB |
1898 | req->task_work.func(&req->task_work); |
1899 | node = next; | |
7cbf1722 JA |
1900 | } |
1901 | ||
2c32395d | 1902 | ctx_flush_and_put(ctx); |
7cbf1722 | 1903 | return list.first != NULL; |
c2c4c83c JA |
1904 | } |
1905 | ||
7cbf1722 | 1906 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1907 | { |
7cbf1722 | 1908 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1909 | |
1d5f360d JA |
1910 | clear_bit(0, &tctx->task_state); |
1911 | ||
7cbf1722 JA |
1912 | while (__tctx_task_work(tctx)) |
1913 | cond_resched(); | |
7cbf1722 JA |
1914 | } |
1915 | ||
c15b79de | 1916 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1917 | { |
c15b79de | 1918 | struct task_struct *tsk = req->task; |
7cbf1722 | 1919 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1920 | enum task_work_notify_mode notify; |
7cbf1722 | 1921 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1922 | unsigned long flags; |
c15b79de PB |
1923 | int ret = 0; |
1924 | ||
1925 | if (unlikely(tsk->flags & PF_EXITING)) | |
1926 | return -ESRCH; | |
7cbf1722 JA |
1927 | |
1928 | WARN_ON_ONCE(!tctx); | |
1929 | ||
0b81e80c | 1930 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1931 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1932 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1933 | |
1934 | /* task_work already pending, we're done */ | |
1935 | if (test_bit(0, &tctx->task_state) || | |
1936 | test_and_set_bit(0, &tctx->task_state)) | |
1937 | return 0; | |
1938 | ||
c15b79de PB |
1939 | /* |
1940 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1941 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1942 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1943 | * will do the job. | |
1944 | */ | |
1945 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1946 | ||
1947 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1948 | wake_up_process(tsk); | |
7cbf1722 | 1949 | return 0; |
c15b79de | 1950 | } |
7cbf1722 JA |
1951 | |
1952 | /* | |
1953 | * Slow path - we failed, find and delete work. if the work is not | |
1954 | * in the list, it got run and we're fine. | |
1955 | */ | |
0b81e80c | 1956 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1957 | wq_list_for_each(node, prev, &tctx->task_list) { |
1958 | if (&req->io_task_work.node == node) { | |
1959 | wq_list_del(&tctx->task_list, node, prev); | |
1960 | ret = 1; | |
1961 | break; | |
1962 | } | |
1963 | } | |
0b81e80c | 1964 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1965 | clear_bit(0, &tctx->task_state); |
1966 | return ret; | |
1967 | } | |
1968 | ||
9b465711 PB |
1969 | static bool io_run_task_work_head(struct callback_head **work_head) |
1970 | { | |
1971 | struct callback_head *work, *next; | |
1972 | bool executed = false; | |
1973 | ||
1974 | do { | |
1975 | work = xchg(work_head, NULL); | |
1976 | if (!work) | |
1977 | break; | |
1978 | ||
1979 | do { | |
1980 | next = work->next; | |
1981 | work->func(work); | |
1982 | work = next; | |
1983 | cond_resched(); | |
1984 | } while (work); | |
1985 | executed = true; | |
1986 | } while (1); | |
1987 | ||
1988 | return executed; | |
1989 | } | |
1990 | ||
1991 | static void io_task_work_add_head(struct callback_head **work_head, | |
1992 | struct callback_head *task_work) | |
eab30c4d | 1993 | { |
7c25c0d1 | 1994 | struct callback_head *head; |
eab30c4d | 1995 | |
7c25c0d1 | 1996 | do { |
9b465711 PB |
1997 | head = READ_ONCE(*work_head); |
1998 | task_work->next = head; | |
1999 | } while (cmpxchg(work_head, head, task_work) != head); | |
2000 | } | |
2001 | ||
2002 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2003 | task_work_func_t cb) | |
2004 | { | |
2005 | init_task_work(&req->task_work, cb); | |
2006 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2007 | } |
2008 | ||
c40f6379 JA |
2009 | static void io_req_task_cancel(struct callback_head *cb) |
2010 | { | |
2011 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2012 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2013 | |
e83acd7d | 2014 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2015 | mutex_lock(&ctx->uring_lock); |
2593553a | 2016 | io_req_complete_failed(req, req->result); |
792bb6eb | 2017 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2018 | } |
2019 | ||
2020 | static void __io_req_task_submit(struct io_kiocb *req) | |
2021 | { | |
2022 | struct io_ring_ctx *ctx = req->ctx; | |
2023 | ||
04fc6c80 | 2024 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2025 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2026 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2027 | __io_queue_sqe(req); |
81b6d05c | 2028 | else |
2593553a | 2029 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2030 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2031 | } |
2032 | ||
2033 | static void io_req_task_submit(struct callback_head *cb) | |
2034 | { | |
2035 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2036 | ||
2037 | __io_req_task_submit(req); | |
2038 | } | |
2039 | ||
2c4b8eb6 | 2040 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2041 | { |
2c4b8eb6 PB |
2042 | req->result = ret; |
2043 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2044 | |
2c4b8eb6 | 2045 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2046 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2047 | } |
2048 | ||
2c4b8eb6 | 2049 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2050 | { |
2c4b8eb6 | 2051 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2052 | |
2053 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2054 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2055 | } |
2056 | ||
f2f87370 | 2057 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2058 | { |
9b5f7bd9 | 2059 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2060 | |
2061 | if (nxt) | |
906a8c3f | 2062 | io_req_task_queue(nxt); |
c69f8dbe JL |
2063 | } |
2064 | ||
c3524383 | 2065 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2066 | { |
c3524383 PB |
2067 | io_queue_next(req); |
2068 | __io_free_req(req); | |
2069 | } | |
8766dd51 | 2070 | |
2d6500d4 | 2071 | struct req_batch { |
5af1d13e PB |
2072 | struct task_struct *task; |
2073 | int task_refs; | |
1b4c351f | 2074 | int ctx_refs; |
2d6500d4 PB |
2075 | }; |
2076 | ||
5af1d13e PB |
2077 | static inline void io_init_req_batch(struct req_batch *rb) |
2078 | { | |
5af1d13e | 2079 | rb->task_refs = 0; |
9ae72463 | 2080 | rb->ctx_refs = 0; |
5af1d13e PB |
2081 | rb->task = NULL; |
2082 | } | |
2083 | ||
2d6500d4 PB |
2084 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2085 | struct req_batch *rb) | |
2086 | { | |
6e833d53 | 2087 | if (rb->task) |
7c660731 | 2088 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2089 | if (rb->ctx_refs) |
2090 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2091 | } |
2092 | ||
6ff119a6 PB |
2093 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2094 | struct io_submit_state *state) | |
2d6500d4 | 2095 | { |
f2f87370 | 2096 | io_queue_next(req); |
96670657 | 2097 | io_dismantle_req(req); |
2d6500d4 | 2098 | |
e3bc8e9d | 2099 | if (req->task != rb->task) { |
7c660731 PB |
2100 | if (rb->task) |
2101 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2102 | rb->task = req->task; |
2103 | rb->task_refs = 0; | |
5af1d13e | 2104 | } |
e3bc8e9d | 2105 | rb->task_refs++; |
9ae72463 | 2106 | rb->ctx_refs++; |
5af1d13e | 2107 | |
bd759045 | 2108 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2109 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2110 | else |
2111 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2112 | } |
2113 | ||
905c172f PB |
2114 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2115 | struct io_ring_ctx *ctx) | |
2116 | { | |
2117 | int i, nr = cs->nr; | |
2118 | struct io_kiocb *req; | |
2119 | struct req_batch rb; | |
2120 | ||
2121 | io_init_req_batch(&rb); | |
2122 | spin_lock_irq(&ctx->completion_lock); | |
2123 | for (i = 0; i < nr; i++) { | |
2124 | req = cs->reqs[i]; | |
d4d19c19 PB |
2125 | __io_cqring_fill_event(ctx, req->user_data, req->result, |
2126 | req->compl.cflags); | |
905c172f PB |
2127 | } |
2128 | io_commit_cqring(ctx); | |
2129 | spin_unlock_irq(&ctx->completion_lock); | |
2130 | ||
2131 | io_cqring_ev_posted(ctx); | |
2132 | for (i = 0; i < nr; i++) { | |
2133 | req = cs->reqs[i]; | |
2134 | ||
2135 | /* submission and completion refs */ | |
de9b4cca | 2136 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2137 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2138 | } |
2139 | ||
2140 | io_req_free_batch_finish(ctx, &rb); | |
2141 | cs->nr = 0; | |
7a743e22 PB |
2142 | } |
2143 | ||
ba816ad6 JA |
2144 | /* |
2145 | * Drop reference to request, return next in chain (if there is one) if this | |
2146 | * was the last reference to this request. | |
2147 | */ | |
0d85035a | 2148 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2149 | { |
9b5f7bd9 PB |
2150 | struct io_kiocb *nxt = NULL; |
2151 | ||
de9b4cca | 2152 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2153 | nxt = io_req_find_next(req); |
4d7dd462 | 2154 | __io_free_req(req); |
2a44f467 | 2155 | } |
9b5f7bd9 | 2156 | return nxt; |
2b188cc1 JA |
2157 | } |
2158 | ||
0d85035a | 2159 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2160 | { |
de9b4cca | 2161 | if (req_ref_put_and_test(req)) |
e65ef56d | 2162 | io_free_req(req); |
2b188cc1 JA |
2163 | } |
2164 | ||
216578e5 PB |
2165 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2166 | { | |
2167 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2168 | ||
2169 | io_free_req(req); | |
2170 | } | |
2171 | ||
2172 | static void io_free_req_deferred(struct io_kiocb *req) | |
2173 | { | |
7cbf1722 | 2174 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2175 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2176 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2177 | } |
2178 | ||
2179 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2180 | { | |
de9b4cca | 2181 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2182 | io_free_req_deferred(req); |
2183 | } | |
2184 | ||
6c503150 | 2185 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2186 | { |
2187 | /* See comment at the top of this file */ | |
2188 | smp_rmb(); | |
e23de15f | 2189 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2190 | } |
2191 | ||
fb5ccc98 PB |
2192 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2193 | { | |
2194 | struct io_rings *rings = ctx->rings; | |
2195 | ||
2196 | /* make sure SQ entry isn't read before tail */ | |
2197 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2198 | } | |
2199 | ||
8ff069bf | 2200 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2201 | { |
8ff069bf | 2202 | unsigned int cflags; |
e94f141b | 2203 | |
bcda7baa JA |
2204 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2205 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2206 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2207 | kfree(kbuf); |
2208 | return cflags; | |
e94f141b JA |
2209 | } |
2210 | ||
8ff069bf | 2211 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2212 | { |
4d954c25 | 2213 | struct io_buffer *kbuf; |
bcda7baa | 2214 | |
4d954c25 | 2215 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2216 | return io_put_kbuf(req, kbuf); |
2217 | } | |
2218 | ||
4c6e277c JA |
2219 | static inline bool io_run_task_work(void) |
2220 | { | |
6200b0ae JA |
2221 | /* |
2222 | * Not safe to run on exiting task, and the task_work handling will | |
2223 | * not add work to such a task. | |
2224 | */ | |
2225 | if (unlikely(current->flags & PF_EXITING)) | |
2226 | return false; | |
4c6e277c JA |
2227 | if (current->task_works) { |
2228 | __set_current_state(TASK_RUNNING); | |
2229 | task_work_run(); | |
2230 | return true; | |
2231 | } | |
2232 | ||
2233 | return false; | |
bcda7baa JA |
2234 | } |
2235 | ||
def596e9 JA |
2236 | /* |
2237 | * Find and free completed poll iocbs | |
2238 | */ | |
2239 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2240 | struct list_head *done) | |
2241 | { | |
8237e045 | 2242 | struct req_batch rb; |
def596e9 | 2243 | struct io_kiocb *req; |
bbde017a XW |
2244 | |
2245 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2246 | smp_rmb(); | |
def596e9 | 2247 | |
5af1d13e | 2248 | io_init_req_batch(&rb); |
def596e9 | 2249 | while (!list_empty(done)) { |
bcda7baa JA |
2250 | int cflags = 0; |
2251 | ||
d21ffe7e | 2252 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2253 | list_del(&req->inflight_entry); |
2254 | ||
8c130827 PB |
2255 | if (READ_ONCE(req->result) == -EAGAIN && |
2256 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2257 | req->iopoll_completed = 0; |
8c130827 PB |
2258 | req_ref_get(req); |
2259 | io_queue_async_work(req); | |
2260 | continue; | |
bbde017a | 2261 | } |
def596e9 | 2262 | |
bcda7baa | 2263 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2264 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2265 | |
d4d19c19 | 2266 | __io_cqring_fill_event(ctx, req->user_data, req->result, cflags); |
def596e9 JA |
2267 | (*nr_events)++; |
2268 | ||
de9b4cca | 2269 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2270 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2271 | } |
def596e9 | 2272 | |
09bb8394 | 2273 | io_commit_cqring(ctx); |
80c18e4a | 2274 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2275 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2276 | } |
2277 | ||
def596e9 JA |
2278 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2279 | long min) | |
2280 | { | |
2281 | struct io_kiocb *req, *tmp; | |
2282 | LIST_HEAD(done); | |
2283 | bool spin; | |
2284 | int ret; | |
2285 | ||
2286 | /* | |
2287 | * Only spin for completions if we don't have multiple devices hanging | |
2288 | * off our complete list, and we're under the requested amount. | |
2289 | */ | |
2290 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2291 | ||
2292 | ret = 0; | |
d21ffe7e | 2293 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2294 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2295 | |
2296 | /* | |
581f9810 BM |
2297 | * Move completed and retryable entries to our local lists. |
2298 | * If we find a request that requires polling, break out | |
2299 | * and complete those lists first, if we have entries there. | |
def596e9 | 2300 | */ |
65a6543d | 2301 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2302 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2303 | continue; |
2304 | } | |
2305 | if (!list_empty(&done)) | |
2306 | break; | |
2307 | ||
2308 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2309 | if (ret < 0) | |
2310 | break; | |
2311 | ||
3aadc23e PB |
2312 | /* iopoll may have completed current req */ |
2313 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2314 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2315 | |
def596e9 JA |
2316 | if (ret && spin) |
2317 | spin = false; | |
2318 | ret = 0; | |
2319 | } | |
2320 | ||
2321 | if (!list_empty(&done)) | |
2322 | io_iopoll_complete(ctx, nr_events, &done); | |
2323 | ||
2324 | return ret; | |
2325 | } | |
2326 | ||
def596e9 JA |
2327 | /* |
2328 | * We can't just wait for polled events to come to us, we have to actively | |
2329 | * find and complete them. | |
2330 | */ | |
b2edc0a7 | 2331 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2332 | { |
2333 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2334 | return; | |
2335 | ||
2336 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2337 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2338 | unsigned int nr_events = 0; |
2339 | ||
b2edc0a7 | 2340 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2341 | |
b2edc0a7 PB |
2342 | /* let it sleep and repeat later if can't complete a request */ |
2343 | if (nr_events == 0) | |
2344 | break; | |
08f5439f JA |
2345 | /* |
2346 | * Ensure we allow local-to-the-cpu processing to take place, | |
2347 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2348 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2349 | */ |
3fcee5a6 PB |
2350 | if (need_resched()) { |
2351 | mutex_unlock(&ctx->uring_lock); | |
2352 | cond_resched(); | |
2353 | mutex_lock(&ctx->uring_lock); | |
2354 | } | |
def596e9 JA |
2355 | } |
2356 | mutex_unlock(&ctx->uring_lock); | |
2357 | } | |
2358 | ||
7668b92a | 2359 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2360 | { |
7668b92a | 2361 | unsigned int nr_events = 0; |
e9979b36 | 2362 | int ret = 0; |
500f9fba | 2363 | |
c7849be9 XW |
2364 | /* |
2365 | * We disallow the app entering submit/complete with polling, but we | |
2366 | * still need to lock the ring to prevent racing with polled issue | |
2367 | * that got punted to a workqueue. | |
2368 | */ | |
2369 | mutex_lock(&ctx->uring_lock); | |
f39c8a5b PB |
2370 | /* |
2371 | * Don't enter poll loop if we already have events pending. | |
2372 | * If we do, we can potentially be spinning for commands that | |
2373 | * already triggered a CQE (eg in error). | |
2374 | */ | |
2375 | if (test_bit(0, &ctx->cq_check_overflow)) | |
2376 | __io_cqring_overflow_flush(ctx, false); | |
2377 | if (io_cqring_events(ctx)) | |
2378 | goto out; | |
def596e9 | 2379 | do { |
500f9fba JA |
2380 | /* |
2381 | * If a submit got punted to a workqueue, we can have the | |
2382 | * application entering polling for a command before it gets | |
2383 | * issued. That app will hold the uring_lock for the duration | |
2384 | * of the poll right here, so we need to take a breather every | |
2385 | * now and then to ensure that the issue has a chance to add | |
2386 | * the poll to the issued list. Otherwise we can spin here | |
2387 | * forever, while the workqueue is stuck trying to acquire the | |
2388 | * very same mutex. | |
2389 | */ | |
e9979b36 | 2390 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2391 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2392 | io_run_task_work(); |
500f9fba | 2393 | mutex_lock(&ctx->uring_lock); |
e9979b36 PB |
2394 | |
2395 | if (list_empty(&ctx->iopoll_list)) | |
2396 | break; | |
500f9fba | 2397 | } |
f39c8a5b PB |
2398 | ret = io_do_iopoll(ctx, &nr_events, min); |
2399 | } while (!ret && nr_events < min && !need_resched()); | |
2400 | out: | |
500f9fba | 2401 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2402 | return ret; |
2403 | } | |
2404 | ||
491381ce | 2405 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2406 | { |
491381ce JA |
2407 | /* |
2408 | * Tell lockdep we inherited freeze protection from submission | |
2409 | * thread. | |
2410 | */ | |
2411 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2412 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2413 | |
1c98679d PB |
2414 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2415 | sb_end_write(sb); | |
2b188cc1 JA |
2416 | } |
2417 | } | |
2418 | ||
b63534c4 | 2419 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2420 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2421 | { |
ab454438 PB |
2422 | struct io_async_rw *rw = req->async_data; |
2423 | ||
2424 | if (!rw) | |
2425 | return !io_req_prep_async(req); | |
2426 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2427 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2428 | return true; | |
b63534c4 | 2429 | } |
b63534c4 | 2430 | |
3e6a0d3c | 2431 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2432 | { |
355afaeb | 2433 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2434 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2435 | |
355afaeb JA |
2436 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2437 | return false; | |
3e6a0d3c JA |
2438 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2439 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2440 | return false; |
7c977a58 JA |
2441 | /* |
2442 | * If ref is dying, we might be running poll reap from the exit work. | |
2443 | * Don't attempt to reissue from that path, just let it fail with | |
2444 | * -EAGAIN. | |
2445 | */ | |
3e6a0d3c JA |
2446 | if (percpu_ref_is_dying(&ctx->refs)) |
2447 | return false; | |
2448 | return true; | |
2449 | } | |
e82ad485 | 2450 | #else |
a1ff1e3f JA |
2451 | static bool io_resubmit_prep(struct io_kiocb *req) |
2452 | { | |
2453 | return false; | |
2454 | } | |
e82ad485 JA |
2455 | static bool io_rw_should_reissue(struct io_kiocb *req) |
2456 | { | |
2457 | return false; | |
2458 | } | |
3e6a0d3c JA |
2459 | #endif |
2460 | ||
a1d7c393 | 2461 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2462 | unsigned int issue_flags) |
a1d7c393 | 2463 | { |
2f8e45f1 PB |
2464 | int cflags = 0; |
2465 | ||
b65c128f PB |
2466 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2467 | kiocb_end_write(req); | |
9532b99b PB |
2468 | if (res != req->result) { |
2469 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2470 | io_rw_should_reissue(req)) { | |
2471 | req->flags |= REQ_F_REISSUE; | |
2472 | return; | |
2473 | } | |
2f8e45f1 | 2474 | req_set_fail_links(req); |
9532b99b | 2475 | } |
2f8e45f1 PB |
2476 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2477 | cflags = io_put_rw_kbuf(req); | |
2478 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2479 | } |
2480 | ||
2481 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2482 | { | |
9adbd45d | 2483 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2484 | |
889fca73 | 2485 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2486 | } |
2487 | ||
def596e9 JA |
2488 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2489 | { | |
9adbd45d | 2490 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2491 | |
491381ce JA |
2492 | if (kiocb->ki_flags & IOCB_WRITE) |
2493 | kiocb_end_write(req); | |
9532b99b | 2494 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2495 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2496 | io_resubmit_prep(req))) { | |
9532b99b PB |
2497 | req_set_fail_links(req); |
2498 | req->flags |= REQ_F_DONT_REISSUE; | |
2499 | } | |
8c130827 | 2500 | } |
bbde017a XW |
2501 | |
2502 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2503 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2504 | smp_wmb(); |
2505 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2506 | } |
2507 | ||
2508 | /* | |
2509 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2510 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
f39c8a5b | 2511 | * find it from a io_do_iopoll() thread before the issuer is done |
def596e9 JA |
2512 | * accessing the kiocb cookie. |
2513 | */ | |
2e9dbe90 | 2514 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2515 | { |
2516 | struct io_ring_ctx *ctx = req->ctx; | |
2517 | ||
2518 | /* | |
2519 | * Track whether we have multiple files in our lists. This will impact | |
2520 | * how we do polling eventually, not spinning if we're on potentially | |
2521 | * different devices. | |
2522 | */ | |
540e32a0 | 2523 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2524 | ctx->poll_multi_file = false; |
2525 | } else if (!ctx->poll_multi_file) { | |
2526 | struct io_kiocb *list_req; | |
2527 | ||
540e32a0 | 2528 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2529 | inflight_entry); |
9adbd45d | 2530 | if (list_req->file != req->file) |
def596e9 JA |
2531 | ctx->poll_multi_file = true; |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * For fast devices, IO may have already completed. If it has, add | |
2536 | * it to the front so we find it first. | |
2537 | */ | |
65a6543d | 2538 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2539 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2540 | else |
d21ffe7e | 2541 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2542 | |
2e9dbe90 XW |
2543 | /* |
2544 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2545 | * task context or in io worker task context. If current task context is | |
2546 | * sq thread, we don't need to check whether should wake up sq thread. | |
2547 | */ | |
2548 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2549 | wq_has_sleeper(&ctx->sq_data->wait)) |
2550 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2551 | } |
2552 | ||
9f13c35b PB |
2553 | static inline void io_state_file_put(struct io_submit_state *state) |
2554 | { | |
02b23a9a PB |
2555 | if (state->file_refs) { |
2556 | fput_many(state->file, state->file_refs); | |
2557 | state->file_refs = 0; | |
2558 | } | |
9a56a232 JA |
2559 | } |
2560 | ||
2561 | /* | |
2562 | * Get as many references to a file as we have IOs left in this submission, | |
2563 | * assuming most submissions are for one file, or at least that each file | |
2564 | * has more than one submission. | |
2565 | */ | |
8da11c19 | 2566 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2567 | { |
2568 | if (!state) | |
2569 | return fget(fd); | |
2570 | ||
6e1271e6 | 2571 | if (state->file_refs) { |
9a56a232 | 2572 | if (state->fd == fd) { |
6e1271e6 | 2573 | state->file_refs--; |
9a56a232 JA |
2574 | return state->file; |
2575 | } | |
02b23a9a | 2576 | io_state_file_put(state); |
9a56a232 JA |
2577 | } |
2578 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2579 | if (unlikely(!state->file)) |
9a56a232 JA |
2580 | return NULL; |
2581 | ||
2582 | state->fd = fd; | |
6e1271e6 | 2583 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2584 | return state->file; |
2585 | } | |
2586 | ||
4503b767 JA |
2587 | static bool io_bdev_nowait(struct block_device *bdev) |
2588 | { | |
9ba0d0c8 | 2589 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2590 | } |
2591 | ||
2b188cc1 JA |
2592 | /* |
2593 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2594 | * any file. For now, just ensure that anything potentially problematic is done | |
2595 | * inline. | |
2596 | */ | |
7b29f92d | 2597 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2598 | { |
2599 | umode_t mode = file_inode(file)->i_mode; | |
2600 | ||
4503b767 | 2601 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2602 | if (IS_ENABLED(CONFIG_BLOCK) && |
2603 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2604 | return true; |
2605 | return false; | |
2606 | } | |
2607 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2608 | return true; |
4503b767 | 2609 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2610 | if (IS_ENABLED(CONFIG_BLOCK) && |
2611 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2612 | file->f_op != &io_uring_fops) |
2613 | return true; | |
2614 | return false; | |
2615 | } | |
2b188cc1 | 2616 | |
c5b85625 JA |
2617 | /* any ->read/write should understand O_NONBLOCK */ |
2618 | if (file->f_flags & O_NONBLOCK) | |
2619 | return true; | |
2620 | ||
af197f50 JA |
2621 | if (!(file->f_mode & FMODE_NOWAIT)) |
2622 | return false; | |
2623 | ||
2624 | if (rw == READ) | |
2625 | return file->f_op->read_iter != NULL; | |
2626 | ||
2627 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2628 | } |
2629 | ||
7b29f92d JA |
2630 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2631 | { | |
2632 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2633 | return true; | |
2634 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2635 | return true; | |
2636 | ||
2637 | return __io_file_supports_async(req->file, rw); | |
2638 | } | |
2639 | ||
a88fc400 | 2640 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2641 | { |
def596e9 | 2642 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2643 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2644 | struct file *file = req->file; |
09bb8394 JA |
2645 | unsigned ioprio; |
2646 | int ret; | |
2b188cc1 | 2647 | |
7b29f92d | 2648 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2649 | req->flags |= REQ_F_ISREG; |
2650 | ||
2b188cc1 | 2651 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2652 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2653 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2654 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2655 | } |
2b188cc1 | 2656 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2657 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2658 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2659 | if (unlikely(ret)) | |
2660 | return ret; | |
2b188cc1 | 2661 | |
75c668cd PB |
2662 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2663 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2664 | req->flags |= REQ_F_NOWAIT; | |
2665 | ||
2b188cc1 JA |
2666 | ioprio = READ_ONCE(sqe->ioprio); |
2667 | if (ioprio) { | |
2668 | ret = ioprio_check_cap(ioprio); | |
2669 | if (ret) | |
09bb8394 | 2670 | return ret; |
2b188cc1 JA |
2671 | |
2672 | kiocb->ki_ioprio = ioprio; | |
2673 | } else | |
2674 | kiocb->ki_ioprio = get_current_ioprio(); | |
2675 | ||
def596e9 | 2676 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2677 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2678 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2679 | return -EOPNOTSUPP; |
2b188cc1 | 2680 | |
def596e9 JA |
2681 | kiocb->ki_flags |= IOCB_HIPRI; |
2682 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2683 | req->iopoll_completed = 0; |
def596e9 | 2684 | } else { |
09bb8394 JA |
2685 | if (kiocb->ki_flags & IOCB_HIPRI) |
2686 | return -EINVAL; | |
def596e9 JA |
2687 | kiocb->ki_complete = io_complete_rw; |
2688 | } | |
9adbd45d | 2689 | |
eae071c9 PB |
2690 | if (req->opcode == IORING_OP_READ_FIXED || |
2691 | req->opcode == IORING_OP_WRITE_FIXED) { | |
2692 | req->imu = NULL; | |
2693 | io_req_set_rsrc_node(req); | |
2694 | } | |
2695 | ||
3529d8c2 JA |
2696 | req->rw.addr = READ_ONCE(sqe->addr); |
2697 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2698 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2699 | return 0; |
2b188cc1 JA |
2700 | } |
2701 | ||
2702 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2703 | { | |
2704 | switch (ret) { | |
2705 | case -EIOCBQUEUED: | |
2706 | break; | |
2707 | case -ERESTARTSYS: | |
2708 | case -ERESTARTNOINTR: | |
2709 | case -ERESTARTNOHAND: | |
2710 | case -ERESTART_RESTARTBLOCK: | |
2711 | /* | |
2712 | * We can't just restart the syscall, since previously | |
2713 | * submitted sqes may already be in progress. Just fail this | |
2714 | * IO with EINTR. | |
2715 | */ | |
2716 | ret = -EINTR; | |
df561f66 | 2717 | fallthrough; |
2b188cc1 JA |
2718 | default: |
2719 | kiocb->ki_complete(kiocb, ret, 0); | |
2720 | } | |
2721 | } | |
2722 | ||
a1d7c393 | 2723 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2724 | unsigned int issue_flags) |
ba816ad6 | 2725 | { |
ba04291e | 2726 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2727 | struct io_async_rw *io = req->async_data; |
97284637 | 2728 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2729 | |
227c0c96 | 2730 | /* add previously done IO, if any */ |
e8c2bc1f | 2731 | if (io && io->bytes_done > 0) { |
227c0c96 | 2732 | if (ret < 0) |
e8c2bc1f | 2733 | ret = io->bytes_done; |
227c0c96 | 2734 | else |
e8c2bc1f | 2735 | ret += io->bytes_done; |
227c0c96 JA |
2736 | } |
2737 | ||
ba04291e JA |
2738 | if (req->flags & REQ_F_CUR_POS) |
2739 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2740 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2741 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2742 | else |
2743 | io_rw_done(kiocb, ret); | |
97284637 PB |
2744 | |
2745 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2746 | req->flags &= ~REQ_F_REISSUE; | |
a7be7c23 | 2747 | if (io_resubmit_prep(req)) { |
8c130827 PB |
2748 | req_ref_get(req); |
2749 | io_queue_async_work(req); | |
2750 | } else { | |
97284637 PB |
2751 | int cflags = 0; |
2752 | ||
2753 | req_set_fail_links(req); | |
2754 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2755 | cflags = io_put_rw_kbuf(req); | |
2756 | __io_req_complete(req, issue_flags, ret, cflags); | |
2757 | } | |
2758 | } | |
ba816ad6 JA |
2759 | } |
2760 | ||
eae071c9 PB |
2761 | static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter, |
2762 | struct io_mapped_ubuf *imu) | |
edafccee | 2763 | { |
9adbd45d | 2764 | size_t len = req->rw.len; |
75769e3f | 2765 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2766 | size_t offset; |
edafccee | 2767 | |
75769e3f | 2768 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2769 | return -EFAULT; |
2770 | /* not inside the mapped region */ | |
4751f53d | 2771 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2772 | return -EFAULT; |
2773 | ||
2774 | /* | |
2775 | * May not be a start of buffer, set size appropriately | |
2776 | * and advance us to the beginning. | |
2777 | */ | |
2778 | offset = buf_addr - imu->ubuf; | |
2779 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2780 | |
2781 | if (offset) { | |
2782 | /* | |
2783 | * Don't use iov_iter_advance() here, as it's really slow for | |
2784 | * using the latter parts of a big fixed buffer - it iterates | |
2785 | * over each segment manually. We can cheat a bit here, because | |
2786 | * we know that: | |
2787 | * | |
2788 | * 1) it's a BVEC iter, we set it up | |
2789 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2790 | * first and last bvec | |
2791 | * | |
2792 | * So just find our index, and adjust the iterator afterwards. | |
2793 | * If the offset is within the first bvec (or the whole first | |
2794 | * bvec, just use iov_iter_advance(). This makes it easier | |
2795 | * since we can just skip the first segment, which may not | |
2796 | * be PAGE_SIZE aligned. | |
2797 | */ | |
2798 | const struct bio_vec *bvec = imu->bvec; | |
2799 | ||
2800 | if (offset <= bvec->bv_len) { | |
2801 | iov_iter_advance(iter, offset); | |
2802 | } else { | |
2803 | unsigned long seg_skip; | |
2804 | ||
2805 | /* skip first vec */ | |
2806 | offset -= bvec->bv_len; | |
2807 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2808 | ||
2809 | iter->bvec = bvec + seg_skip; | |
2810 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2811 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2812 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2813 | } |
2814 | } | |
2815 | ||
847595de | 2816 | return 0; |
edafccee JA |
2817 | } |
2818 | ||
eae071c9 PB |
2819 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
2820 | { | |
2821 | struct io_ring_ctx *ctx = req->ctx; | |
2822 | struct io_mapped_ubuf *imu = req->imu; | |
2823 | u16 index, buf_index = req->buf_index; | |
2824 | ||
2825 | if (likely(!imu)) { | |
2826 | if (unlikely(buf_index >= ctx->nr_user_bufs)) | |
2827 | return -EFAULT; | |
2828 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); | |
2829 | imu = READ_ONCE(ctx->user_bufs[index]); | |
2830 | req->imu = imu; | |
2831 | } | |
2832 | return __io_import_fixed(req, rw, iter, imu); | |
2833 | } | |
2834 | ||
bcda7baa JA |
2835 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2836 | { | |
2837 | if (needs_lock) | |
2838 | mutex_unlock(&ctx->uring_lock); | |
2839 | } | |
2840 | ||
2841 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2842 | { | |
2843 | /* | |
2844 | * "Normal" inline submissions always hold the uring_lock, since we | |
2845 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2846 | * The only exception is when we've detached the request and issue it | |
2847 | * from an async worker thread, grab the lock for that case. | |
2848 | */ | |
2849 | if (needs_lock) | |
2850 | mutex_lock(&ctx->uring_lock); | |
2851 | } | |
2852 | ||
2853 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2854 | int bgid, struct io_buffer *kbuf, | |
2855 | bool needs_lock) | |
2856 | { | |
2857 | struct io_buffer *head; | |
2858 | ||
2859 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2860 | return kbuf; | |
2861 | ||
2862 | io_ring_submit_lock(req->ctx, needs_lock); | |
2863 | ||
2864 | lockdep_assert_held(&req->ctx->uring_lock); | |
2865 | ||
9e15c3a0 | 2866 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2867 | if (head) { |
2868 | if (!list_empty(&head->list)) { | |
2869 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2870 | list); | |
2871 | list_del(&kbuf->list); | |
2872 | } else { | |
2873 | kbuf = head; | |
9e15c3a0 | 2874 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2875 | } |
2876 | if (*len > kbuf->len) | |
2877 | *len = kbuf->len; | |
2878 | } else { | |
2879 | kbuf = ERR_PTR(-ENOBUFS); | |
2880 | } | |
2881 | ||
2882 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2883 | ||
2884 | return kbuf; | |
2885 | } | |
2886 | ||
4d954c25 JA |
2887 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2888 | bool needs_lock) | |
2889 | { | |
2890 | struct io_buffer *kbuf; | |
4f4eeba8 | 2891 | u16 bgid; |
4d954c25 JA |
2892 | |
2893 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2894 | bgid = req->buf_index; |
4d954c25 JA |
2895 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2896 | if (IS_ERR(kbuf)) | |
2897 | return kbuf; | |
2898 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2899 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2900 | return u64_to_user_ptr(kbuf->addr); | |
2901 | } | |
2902 | ||
2903 | #ifdef CONFIG_COMPAT | |
2904 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2905 | bool needs_lock) | |
2906 | { | |
2907 | struct compat_iovec __user *uiov; | |
2908 | compat_ssize_t clen; | |
2909 | void __user *buf; | |
2910 | ssize_t len; | |
2911 | ||
2912 | uiov = u64_to_user_ptr(req->rw.addr); | |
2913 | if (!access_ok(uiov, sizeof(*uiov))) | |
2914 | return -EFAULT; | |
2915 | if (__get_user(clen, &uiov->iov_len)) | |
2916 | return -EFAULT; | |
2917 | if (clen < 0) | |
2918 | return -EINVAL; | |
2919 | ||
2920 | len = clen; | |
2921 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2922 | if (IS_ERR(buf)) | |
2923 | return PTR_ERR(buf); | |
2924 | iov[0].iov_base = buf; | |
2925 | iov[0].iov_len = (compat_size_t) len; | |
2926 | return 0; | |
2927 | } | |
2928 | #endif | |
2929 | ||
2930 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2931 | bool needs_lock) | |
2932 | { | |
2933 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2934 | void __user *buf; | |
2935 | ssize_t len; | |
2936 | ||
2937 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2938 | return -EFAULT; | |
2939 | ||
2940 | len = iov[0].iov_len; | |
2941 | if (len < 0) | |
2942 | return -EINVAL; | |
2943 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2944 | if (IS_ERR(buf)) | |
2945 | return PTR_ERR(buf); | |
2946 | iov[0].iov_base = buf; | |
2947 | iov[0].iov_len = len; | |
2948 | return 0; | |
2949 | } | |
2950 | ||
2951 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2952 | bool needs_lock) | |
2953 | { | |
dddb3e26 JA |
2954 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2955 | struct io_buffer *kbuf; | |
2956 | ||
2957 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2958 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2959 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2960 | return 0; |
dddb3e26 | 2961 | } |
dd201662 | 2962 | if (req->rw.len != 1) |
4d954c25 JA |
2963 | return -EINVAL; |
2964 | ||
2965 | #ifdef CONFIG_COMPAT | |
2966 | if (req->ctx->compat) | |
2967 | return io_compat_import(req, iov, needs_lock); | |
2968 | #endif | |
2969 | ||
2970 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2971 | } | |
2972 | ||
847595de PB |
2973 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2974 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2975 | { |
9adbd45d JA |
2976 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2977 | size_t sqe_len = req->rw.len; | |
847595de | 2978 | u8 opcode = req->opcode; |
4d954c25 | 2979 | ssize_t ret; |
edafccee | 2980 | |
7d009165 | 2981 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2982 | *iovec = NULL; |
9adbd45d | 2983 | return io_import_fixed(req, rw, iter); |
edafccee | 2984 | } |
2b188cc1 | 2985 | |
bcda7baa | 2986 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 2987 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
2988 | return -EINVAL; |
2989 | ||
3a6820f2 | 2990 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 2991 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 2992 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 2993 | if (IS_ERR(buf)) |
4d954c25 | 2994 | return PTR_ERR(buf); |
3f9d6441 | 2995 | req->rw.len = sqe_len; |
bcda7baa JA |
2996 | } |
2997 | ||
3a6820f2 JA |
2998 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
2999 | *iovec = NULL; | |
10fc72e4 | 3000 | return ret; |
3a6820f2 JA |
3001 | } |
3002 | ||
4d954c25 JA |
3003 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3004 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3005 | if (!ret) |
3006 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3007 | *iovec = NULL; |
3008 | return ret; | |
3009 | } | |
3010 | ||
89cd35c5 CH |
3011 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3012 | req->ctx->compat); | |
2b188cc1 JA |
3013 | } |
3014 | ||
0fef9483 JA |
3015 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3016 | { | |
5b09e37e | 3017 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3018 | } |
3019 | ||
31b51510 | 3020 | /* |
32960613 JA |
3021 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3022 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3023 | */ |
4017eb91 | 3024 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3025 | { |
4017eb91 JA |
3026 | struct kiocb *kiocb = &req->rw.kiocb; |
3027 | struct file *file = req->file; | |
32960613 JA |
3028 | ssize_t ret = 0; |
3029 | ||
3030 | /* | |
3031 | * Don't support polled IO through this interface, and we can't | |
3032 | * support non-blocking either. For the latter, this just causes | |
3033 | * the kiocb to be handled from an async context. | |
3034 | */ | |
3035 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3036 | return -EOPNOTSUPP; | |
3037 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3038 | return -EAGAIN; | |
3039 | ||
3040 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3041 | struct iovec iovec; |
32960613 JA |
3042 | ssize_t nr; |
3043 | ||
311ae9e1 PB |
3044 | if (!iov_iter_is_bvec(iter)) { |
3045 | iovec = iov_iter_iovec(iter); | |
3046 | } else { | |
4017eb91 JA |
3047 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3048 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3049 | } |
3050 | ||
32960613 JA |
3051 | if (rw == READ) { |
3052 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3053 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3054 | } else { |
3055 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3056 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3057 | } |
3058 | ||
3059 | if (nr < 0) { | |
3060 | if (!ret) | |
3061 | ret = nr; | |
3062 | break; | |
3063 | } | |
3064 | ret += nr; | |
3065 | if (nr != iovec.iov_len) | |
3066 | break; | |
4017eb91 JA |
3067 | req->rw.len -= nr; |
3068 | req->rw.addr += nr; | |
32960613 JA |
3069 | iov_iter_advance(iter, nr); |
3070 | } | |
3071 | ||
3072 | return ret; | |
3073 | } | |
3074 | ||
ff6165b2 JA |
3075 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3076 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3077 | { |
e8c2bc1f | 3078 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3079 | |
ff6165b2 | 3080 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3081 | rw->free_iovec = iovec; |
227c0c96 | 3082 | rw->bytes_done = 0; |
ff6165b2 | 3083 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3084 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3085 | return; |
b64e3444 | 3086 | if (!iovec) { |
ff6165b2 JA |
3087 | unsigned iov_off = 0; |
3088 | ||
3089 | rw->iter.iov = rw->fast_iov; | |
3090 | if (iter->iov != fast_iov) { | |
3091 | iov_off = iter->iov - fast_iov; | |
3092 | rw->iter.iov += iov_off; | |
3093 | } | |
3094 | if (rw->fast_iov != fast_iov) | |
3095 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3096 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3097 | } else { |
3098 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3099 | } |
3100 | } | |
3101 | ||
6cb78689 | 3102 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3103 | { |
e8c2bc1f JA |
3104 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3105 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3106 | return req->async_data == NULL; | |
3d9932a8 XW |
3107 | } |
3108 | ||
ff6165b2 JA |
3109 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3110 | const struct iovec *fast_iov, | |
227c0c96 | 3111 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3112 | { |
26f0505a | 3113 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3114 | return 0; |
e8c2bc1f | 3115 | if (!req->async_data) { |
6cb78689 | 3116 | if (io_alloc_async_data(req)) { |
6bf985dc | 3117 | kfree(iovec); |
5d204bcf | 3118 | return -ENOMEM; |
6bf985dc | 3119 | } |
b7bb4f7d | 3120 | |
ff6165b2 | 3121 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3122 | } |
b7bb4f7d | 3123 | return 0; |
f67676d1 JA |
3124 | } |
3125 | ||
73debe68 | 3126 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3127 | { |
e8c2bc1f | 3128 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3129 | struct iovec *iov = iorw->fast_iov; |
847595de | 3130 | int ret; |
c3e330a4 | 3131 | |
2846c481 | 3132 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3133 | if (unlikely(ret < 0)) |
3134 | return ret; | |
3135 | ||
ab0b196c PB |
3136 | iorw->bytes_done = 0; |
3137 | iorw->free_iovec = iov; | |
3138 | if (iov) | |
3139 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3140 | return 0; |
3141 | } | |
3142 | ||
73debe68 | 3143 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3144 | { |
3529d8c2 JA |
3145 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3146 | return -EBADF; | |
93642ef8 | 3147 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3148 | } |
3149 | ||
c1dd91d1 JA |
3150 | /* |
3151 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3152 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3153 | * This gets called when the page is unlocked, and we generally expect that to | |
3154 | * happen when the page IO is completed and the page is now uptodate. This will | |
3155 | * queue a task_work based retry of the operation, attempting to copy the data | |
3156 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3157 | * do a thread based blocking retry of the operation. That's the unexpected | |
3158 | * slow path. | |
3159 | */ | |
bcf5a063 JA |
3160 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3161 | int sync, void *arg) | |
3162 | { | |
3163 | struct wait_page_queue *wpq; | |
3164 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3165 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3166 | |
3167 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3168 | ||
cdc8fcb4 LT |
3169 | if (!wake_page_match(wpq, key)) |
3170 | return 0; | |
3171 | ||
c8d317aa | 3172 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3173 | list_del_init(&wait->entry); |
3174 | ||
bcf5a063 | 3175 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3176 | req_ref_get(req); |
921b9054 | 3177 | io_req_task_queue(req); |
bcf5a063 JA |
3178 | return 1; |
3179 | } | |
3180 | ||
c1dd91d1 JA |
3181 | /* |
3182 | * This controls whether a given IO request should be armed for async page | |
3183 | * based retry. If we return false here, the request is handed to the async | |
3184 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3185 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3186 | * will either succeed because the page is now uptodate and unlocked, or it | |
3187 | * will register a callback when the page is unlocked at IO completion. Through | |
3188 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3189 | * That retry will attempt the buffered read again. The retry will generally | |
3190 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3191 | * async worker threads for a blocking retry. | |
3192 | */ | |
227c0c96 | 3193 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3194 | { |
e8c2bc1f JA |
3195 | struct io_async_rw *rw = req->async_data; |
3196 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3197 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3198 | |
bcf5a063 JA |
3199 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3200 | if (req->flags & REQ_F_NOWAIT) | |
3201 | return false; | |
f67676d1 | 3202 | |
227c0c96 | 3203 | /* Only for buffered IO */ |
3b2a4439 | 3204 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3205 | return false; |
3b2a4439 | 3206 | |
bcf5a063 JA |
3207 | /* |
3208 | * just use poll if we can, and don't attempt if the fs doesn't | |
3209 | * support callback based unlocks | |
3210 | */ | |
3211 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3212 | return false; | |
f67676d1 | 3213 | |
3b2a4439 JA |
3214 | wait->wait.func = io_async_buf_func; |
3215 | wait->wait.private = req; | |
3216 | wait->wait.flags = 0; | |
3217 | INIT_LIST_HEAD(&wait->wait.entry); | |
3218 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3219 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3220 | kiocb->ki_waitq = wait; |
3b2a4439 | 3221 | return true; |
bcf5a063 JA |
3222 | } |
3223 | ||
3224 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3225 | { | |
3226 | if (req->file->f_op->read_iter) | |
3227 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3228 | else if (req->file->f_op->read) |
4017eb91 | 3229 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3230 | else |
3231 | return -EINVAL; | |
f67676d1 JA |
3232 | } |
3233 | ||
889fca73 | 3234 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3235 | { |
3236 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3237 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3238 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3239 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3240 | ssize_t io_size, ret, ret2; |
45d189c6 | 3241 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3242 | |
2846c481 | 3243 | if (rw) { |
e8c2bc1f | 3244 | iter = &rw->iter; |
2846c481 PB |
3245 | iovec = NULL; |
3246 | } else { | |
3247 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3248 | if (ret < 0) | |
3249 | return ret; | |
3250 | } | |
632546c4 | 3251 | io_size = iov_iter_count(iter); |
fa15bafb | 3252 | req->result = io_size; |
2b188cc1 | 3253 | |
fd6c2e4c JA |
3254 | /* Ensure we clear previously set non-block flag */ |
3255 | if (!force_nonblock) | |
29de5f6a | 3256 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3257 | else |
3258 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3259 | ||
24c74678 | 3260 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3261 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3262 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3263 | return ret ?: -EAGAIN; |
6713e7a6 | 3264 | } |
9e645e11 | 3265 | |
632546c4 | 3266 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3267 | if (unlikely(ret)) { |
3268 | kfree(iovec); | |
3269 | return ret; | |
3270 | } | |
2b188cc1 | 3271 | |
227c0c96 | 3272 | ret = io_iter_do_read(req, iter); |
32960613 | 3273 | |
230d50d4 | 3274 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3275 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3276 | /* IOPOLL retry should happen for io-wq threads */ |
3277 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3278 | goto done; |
75c668cd PB |
3279 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3280 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3281 | goto done; |
84216315 | 3282 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3283 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3284 | ret = 0; |
230d50d4 JA |
3285 | } else if (ret == -EIOCBQUEUED) { |
3286 | goto out_free; | |
7335e3bf | 3287 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3288 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3289 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3290 | goto done; |
227c0c96 JA |
3291 | } |
3292 | ||
227c0c96 | 3293 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3294 | if (ret2) |
3295 | return ret2; | |
3296 | ||
fe1cdd55 | 3297 | iovec = NULL; |
e8c2bc1f | 3298 | rw = req->async_data; |
227c0c96 | 3299 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3300 | iter = &rw->iter; |
227c0c96 | 3301 | |
b23df91b PB |
3302 | do { |
3303 | io_size -= ret; | |
3304 | rw->bytes_done += ret; | |
3305 | /* if we can retry, do so with the callbacks armed */ | |
3306 | if (!io_rw_should_retry(req)) { | |
3307 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3308 | return -EAGAIN; | |
3309 | } | |
3310 | ||
3311 | /* | |
3312 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3313 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3314 | * desired page gets unlocked. We can also get a partial read | |
3315 | * here, and if we do, then just retry at the new offset. | |
3316 | */ | |
3317 | ret = io_iter_do_read(req, iter); | |
3318 | if (ret == -EIOCBQUEUED) | |
3319 | return 0; | |
227c0c96 | 3320 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3321 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3322 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3323 | done: |
889fca73 | 3324 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3325 | out_free: |
3326 | /* it's faster to check here then delegate to kfree */ | |
3327 | if (iovec) | |
3328 | kfree(iovec); | |
5ea5dd45 | 3329 | return 0; |
2b188cc1 JA |
3330 | } |
3331 | ||
73debe68 | 3332 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3333 | { |
3529d8c2 JA |
3334 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3335 | return -EBADF; | |
93642ef8 | 3336 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3337 | } |
3338 | ||
889fca73 | 3339 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3340 | { |
3341 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3342 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3343 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3344 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3345 | ssize_t ret, ret2, io_size; |
45d189c6 | 3346 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3347 | |
2846c481 | 3348 | if (rw) { |
e8c2bc1f | 3349 | iter = &rw->iter; |
2846c481 PB |
3350 | iovec = NULL; |
3351 | } else { | |
3352 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3353 | if (ret < 0) | |
3354 | return ret; | |
3355 | } | |
632546c4 | 3356 | io_size = iov_iter_count(iter); |
fa15bafb | 3357 | req->result = io_size; |
2b188cc1 | 3358 | |
fd6c2e4c JA |
3359 | /* Ensure we clear previously set non-block flag */ |
3360 | if (!force_nonblock) | |
a88fc400 PB |
3361 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3362 | else | |
3363 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3364 | |
24c74678 | 3365 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3366 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3367 | goto copy_iov; |
31b51510 | 3368 | |
10d59345 JA |
3369 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3370 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3371 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3372 | goto copy_iov; |
31b51510 | 3373 | |
632546c4 | 3374 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3375 | if (unlikely(ret)) |
3376 | goto out_free; | |
4ed734b0 | 3377 | |
fa15bafb PB |
3378 | /* |
3379 | * Open-code file_start_write here to grab freeze protection, | |
3380 | * which will be released by another thread in | |
3381 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3382 | * released so that it doesn't complain about the held lock when | |
3383 | * we return to userspace. | |
3384 | */ | |
3385 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3386 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3387 | __sb_writers_release(file_inode(req->file)->i_sb, |
3388 | SB_FREEZE_WRITE); | |
3389 | } | |
3390 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3391 | |
fa15bafb | 3392 | if (req->file->f_op->write_iter) |
ff6165b2 | 3393 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3394 | else if (req->file->f_op->write) |
4017eb91 | 3395 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3396 | else |
3397 | ret2 = -EINVAL; | |
4ed734b0 | 3398 | |
6ad7f233 PB |
3399 | if (req->flags & REQ_F_REISSUE) { |
3400 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3401 | ret2 = -EAGAIN; |
6ad7f233 | 3402 | } |
230d50d4 | 3403 | |
fa15bafb PB |
3404 | /* |
3405 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3406 | * retry them without IOCB_NOWAIT. | |
3407 | */ | |
3408 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3409 | ret2 = -EAGAIN; | |
75c668cd PB |
3410 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3411 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3412 | goto done; |
fa15bafb | 3413 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3414 | /* IOPOLL retry should happen for io-wq threads */ |
3415 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3416 | goto copy_iov; | |
355afaeb | 3417 | done: |
889fca73 | 3418 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3419 | } else { |
f67676d1 | 3420 | copy_iov: |
84216315 | 3421 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3422 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3423 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3424 | return ret ?: -EAGAIN; |
2b188cc1 | 3425 | } |
31b51510 | 3426 | out_free: |
f261c168 | 3427 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3428 | if (iovec) |
6f2cc166 | 3429 | kfree(iovec); |
2b188cc1 JA |
3430 | return ret; |
3431 | } | |
3432 | ||
80a261fd JA |
3433 | static int io_renameat_prep(struct io_kiocb *req, |
3434 | const struct io_uring_sqe *sqe) | |
3435 | { | |
3436 | struct io_rename *ren = &req->rename; | |
3437 | const char __user *oldf, *newf; | |
3438 | ||
3439 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3440 | return -EBADF; | |
3441 | ||
3442 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3443 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3444 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3445 | ren->new_dfd = READ_ONCE(sqe->len); | |
3446 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3447 | ||
3448 | ren->oldpath = getname(oldf); | |
3449 | if (IS_ERR(ren->oldpath)) | |
3450 | return PTR_ERR(ren->oldpath); | |
3451 | ||
3452 | ren->newpath = getname(newf); | |
3453 | if (IS_ERR(ren->newpath)) { | |
3454 | putname(ren->oldpath); | |
3455 | return PTR_ERR(ren->newpath); | |
3456 | } | |
3457 | ||
3458 | req->flags |= REQ_F_NEED_CLEANUP; | |
3459 | return 0; | |
3460 | } | |
3461 | ||
45d189c6 | 3462 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3463 | { |
3464 | struct io_rename *ren = &req->rename; | |
3465 | int ret; | |
3466 | ||
45d189c6 | 3467 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3468 | return -EAGAIN; |
3469 | ||
3470 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3471 | ren->newpath, ren->flags); | |
3472 | ||
3473 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3474 | if (ret < 0) | |
3475 | req_set_fail_links(req); | |
3476 | io_req_complete(req, ret); | |
3477 | return 0; | |
3478 | } | |
3479 | ||
14a1143b JA |
3480 | static int io_unlinkat_prep(struct io_kiocb *req, |
3481 | const struct io_uring_sqe *sqe) | |
3482 | { | |
3483 | struct io_unlink *un = &req->unlink; | |
3484 | const char __user *fname; | |
3485 | ||
3486 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3487 | return -EBADF; | |
3488 | ||
3489 | un->dfd = READ_ONCE(sqe->fd); | |
3490 | ||
3491 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3492 | if (un->flags & ~AT_REMOVEDIR) | |
3493 | return -EINVAL; | |
3494 | ||
3495 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3496 | un->filename = getname(fname); | |
3497 | if (IS_ERR(un->filename)) | |
3498 | return PTR_ERR(un->filename); | |
3499 | ||
3500 | req->flags |= REQ_F_NEED_CLEANUP; | |
3501 | return 0; | |
3502 | } | |
3503 | ||
45d189c6 | 3504 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3505 | { |
3506 | struct io_unlink *un = &req->unlink; | |
3507 | int ret; | |
3508 | ||
45d189c6 | 3509 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3510 | return -EAGAIN; |
3511 | ||
3512 | if (un->flags & AT_REMOVEDIR) | |
3513 | ret = do_rmdir(un->dfd, un->filename); | |
3514 | else | |
3515 | ret = do_unlinkat(un->dfd, un->filename); | |
3516 | ||
3517 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3518 | if (ret < 0) | |
3519 | req_set_fail_links(req); | |
3520 | io_req_complete(req, ret); | |
3521 | return 0; | |
3522 | } | |
3523 | ||
36f4fa68 JA |
3524 | static int io_shutdown_prep(struct io_kiocb *req, |
3525 | const struct io_uring_sqe *sqe) | |
3526 | { | |
3527 | #if defined(CONFIG_NET) | |
3528 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3529 | return -EINVAL; | |
3530 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3531 | sqe->buf_index) | |
3532 | return -EINVAL; | |
3533 | ||
3534 | req->shutdown.how = READ_ONCE(sqe->len); | |
3535 | return 0; | |
3536 | #else | |
3537 | return -EOPNOTSUPP; | |
3538 | #endif | |
3539 | } | |
3540 | ||
45d189c6 | 3541 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3542 | { |
3543 | #if defined(CONFIG_NET) | |
3544 | struct socket *sock; | |
3545 | int ret; | |
3546 | ||
45d189c6 | 3547 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3548 | return -EAGAIN; |
3549 | ||
48aba79b | 3550 | sock = sock_from_file(req->file); |
36f4fa68 | 3551 | if (unlikely(!sock)) |
48aba79b | 3552 | return -ENOTSOCK; |
36f4fa68 JA |
3553 | |
3554 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3555 | if (ret < 0) |
3556 | req_set_fail_links(req); | |
36f4fa68 JA |
3557 | io_req_complete(req, ret); |
3558 | return 0; | |
3559 | #else | |
3560 | return -EOPNOTSUPP; | |
3561 | #endif | |
3562 | } | |
3563 | ||
f2a8d5c7 PB |
3564 | static int __io_splice_prep(struct io_kiocb *req, |
3565 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3566 | { |
3567 | struct io_splice* sp = &req->splice; | |
3568 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3569 | |
3232dd02 PB |
3570 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3571 | return -EINVAL; | |
7d67af2c PB |
3572 | |
3573 | sp->file_in = NULL; | |
7d67af2c PB |
3574 | sp->len = READ_ONCE(sqe->len); |
3575 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3576 | ||
3577 | if (unlikely(sp->flags & ~valid_flags)) | |
3578 | return -EINVAL; | |
3579 | ||
8371adf5 PB |
3580 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3581 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3582 | if (!sp->file_in) | |
3583 | return -EBADF; | |
7d67af2c | 3584 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3585 | return 0; |
3586 | } | |
3587 | ||
f2a8d5c7 PB |
3588 | static int io_tee_prep(struct io_kiocb *req, |
3589 | const struct io_uring_sqe *sqe) | |
3590 | { | |
3591 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3592 | return -EINVAL; | |
3593 | return __io_splice_prep(req, sqe); | |
3594 | } | |
3595 | ||
45d189c6 | 3596 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3597 | { |
3598 | struct io_splice *sp = &req->splice; | |
3599 | struct file *in = sp->file_in; | |
3600 | struct file *out = sp->file_out; | |
3601 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3602 | long ret = 0; | |
3603 | ||
45d189c6 | 3604 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3605 | return -EAGAIN; |
3606 | if (sp->len) | |
3607 | ret = do_tee(in, out, sp->len, flags); | |
3608 | ||
e1d767f0 PB |
3609 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3610 | io_put_file(in); | |
f2a8d5c7 PB |
3611 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3612 | ||
f2a8d5c7 PB |
3613 | if (ret != sp->len) |
3614 | req_set_fail_links(req); | |
e1e16097 | 3615 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3616 | return 0; |
3617 | } | |
3618 | ||
3619 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3620 | { | |
3621 | struct io_splice* sp = &req->splice; | |
3622 | ||
3623 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3624 | sp->off_out = READ_ONCE(sqe->off); | |
3625 | return __io_splice_prep(req, sqe); | |
3626 | } | |
3627 | ||
45d189c6 | 3628 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3629 | { |
3630 | struct io_splice *sp = &req->splice; | |
3631 | struct file *in = sp->file_in; | |
3632 | struct file *out = sp->file_out; | |
3633 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3634 | loff_t *poff_in, *poff_out; | |
c9687426 | 3635 | long ret = 0; |
7d67af2c | 3636 | |
45d189c6 | 3637 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3638 | return -EAGAIN; |
7d67af2c PB |
3639 | |
3640 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3641 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3642 | |
948a7749 | 3643 | if (sp->len) |
c9687426 | 3644 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3645 | |
e1d767f0 PB |
3646 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3647 | io_put_file(in); | |
7d67af2c PB |
3648 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3649 | ||
7d67af2c PB |
3650 | if (ret != sp->len) |
3651 | req_set_fail_links(req); | |
e1e16097 | 3652 | io_req_complete(req, ret); |
7d67af2c PB |
3653 | return 0; |
3654 | } | |
3655 | ||
2b188cc1 JA |
3656 | /* |
3657 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3658 | */ | |
889fca73 | 3659 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3660 | { |
3661 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3662 | |
def596e9 JA |
3663 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3664 | return -EINVAL; | |
3665 | ||
889fca73 | 3666 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3667 | return 0; |
3668 | } | |
3669 | ||
1155c76a | 3670 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3671 | { |
6b06314c | 3672 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3673 | |
09bb8394 JA |
3674 | if (!req->file) |
3675 | return -EBADF; | |
c992fe29 | 3676 | |
6b06314c | 3677 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3678 | return -EINVAL; |
edafccee | 3679 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3680 | return -EINVAL; |
3681 | ||
8ed8d3c3 JA |
3682 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3683 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3684 | return -EINVAL; | |
3685 | ||
3686 | req->sync.off = READ_ONCE(sqe->off); | |
3687 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3688 | return 0; |
3689 | } | |
3690 | ||
45d189c6 | 3691 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3692 | { |
8ed8d3c3 | 3693 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3694 | int ret; |
3695 | ||
ac45abc0 | 3696 | /* fsync always requires a blocking context */ |
45d189c6 | 3697 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3698 | return -EAGAIN; |
3699 | ||
9adbd45d | 3700 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3701 | end > 0 ? end : LLONG_MAX, |
3702 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3703 | if (ret < 0) | |
3704 | req_set_fail_links(req); | |
e1e16097 | 3705 | io_req_complete(req, ret); |
c992fe29 CH |
3706 | return 0; |
3707 | } | |
3708 | ||
d63d1b5e JA |
3709 | static int io_fallocate_prep(struct io_kiocb *req, |
3710 | const struct io_uring_sqe *sqe) | |
3711 | { | |
3712 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3713 | return -EINVAL; | |
3232dd02 PB |
3714 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3715 | return -EINVAL; | |
d63d1b5e JA |
3716 | |
3717 | req->sync.off = READ_ONCE(sqe->off); | |
3718 | req->sync.len = READ_ONCE(sqe->addr); | |
3719 | req->sync.mode = READ_ONCE(sqe->len); | |
3720 | return 0; | |
3721 | } | |
3722 | ||
45d189c6 | 3723 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3724 | { |
ac45abc0 PB |
3725 | int ret; |
3726 | ||
d63d1b5e | 3727 | /* fallocate always requiring blocking context */ |
45d189c6 | 3728 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3729 | return -EAGAIN; |
ac45abc0 PB |
3730 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3731 | req->sync.len); | |
ac45abc0 PB |
3732 | if (ret < 0) |
3733 | req_set_fail_links(req); | |
e1e16097 | 3734 | io_req_complete(req, ret); |
5d17b4a4 JA |
3735 | return 0; |
3736 | } | |
3737 | ||
ec65fea5 | 3738 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3739 | { |
f8748881 | 3740 | const char __user *fname; |
15b71abe | 3741 | int ret; |
b7bb4f7d | 3742 | |
ec65fea5 | 3743 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3744 | return -EINVAL; |
ec65fea5 | 3745 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3746 | return -EBADF; |
03b1230c | 3747 | |
ec65fea5 PB |
3748 | /* open.how should be already initialised */ |
3749 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3750 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3751 | |
25e72d10 PB |
3752 | req->open.dfd = READ_ONCE(sqe->fd); |
3753 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3754 | req->open.filename = getname(fname); |
15b71abe JA |
3755 | if (IS_ERR(req->open.filename)) { |
3756 | ret = PTR_ERR(req->open.filename); | |
3757 | req->open.filename = NULL; | |
3758 | return ret; | |
3759 | } | |
4022e7af | 3760 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3761 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3762 | return 0; |
03b1230c JA |
3763 | } |
3764 | ||
ec65fea5 PB |
3765 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3766 | { | |
3767 | u64 flags, mode; | |
3768 | ||
14587a46 | 3769 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3770 | return -EINVAL; |
ec65fea5 PB |
3771 | mode = READ_ONCE(sqe->len); |
3772 | flags = READ_ONCE(sqe->open_flags); | |
3773 | req->open.how = build_open_how(flags, mode); | |
3774 | return __io_openat_prep(req, sqe); | |
3775 | } | |
3776 | ||
cebdb986 | 3777 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3778 | { |
cebdb986 | 3779 | struct open_how __user *how; |
cebdb986 | 3780 | size_t len; |
0fa03c62 JA |
3781 | int ret; |
3782 | ||
14587a46 | 3783 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3784 | return -EINVAL; |
cebdb986 JA |
3785 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3786 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3787 | if (len < OPEN_HOW_SIZE_VER0) |
3788 | return -EINVAL; | |
3529d8c2 | 3789 | |
cebdb986 JA |
3790 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3791 | len); | |
3792 | if (ret) | |
3793 | return ret; | |
3529d8c2 | 3794 | |
ec65fea5 | 3795 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3796 | } |
3797 | ||
45d189c6 | 3798 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3799 | { |
3800 | struct open_flags op; | |
15b71abe | 3801 | struct file *file; |
3a81fd02 JA |
3802 | bool nonblock_set; |
3803 | bool resolve_nonblock; | |
15b71abe JA |
3804 | int ret; |
3805 | ||
cebdb986 | 3806 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3807 | if (ret) |
3808 | goto err; | |
3a81fd02 JA |
3809 | nonblock_set = op.open_flag & O_NONBLOCK; |
3810 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3811 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3812 | /* |
3813 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3814 | * it'll always -EAGAIN | |
3815 | */ | |
3816 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3817 | return -EAGAIN; | |
3818 | op.lookup_flags |= LOOKUP_CACHED; | |
3819 | op.open_flag |= O_NONBLOCK; | |
3820 | } | |
15b71abe | 3821 | |
4022e7af | 3822 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3823 | if (ret < 0) |
3824 | goto err; | |
3825 | ||
3826 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3827 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3828 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3829 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3830 | /* |
3a81fd02 JA |
3831 | * We could hang on to this 'fd', but seems like marginal |
3832 | * gain for something that is now known to be a slower path. | |
3833 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3834 | */ |
3a81fd02 JA |
3835 | put_unused_fd(ret); |
3836 | return -EAGAIN; | |
3837 | } | |
3838 | ||
15b71abe JA |
3839 | if (IS_ERR(file)) { |
3840 | put_unused_fd(ret); | |
3841 | ret = PTR_ERR(file); | |
3842 | } else { | |
45d189c6 | 3843 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3844 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3845 | fsnotify_open(file); |
3846 | fd_install(ret, file); | |
3847 | } | |
3848 | err: | |
3849 | putname(req->open.filename); | |
8fef80bf | 3850 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3851 | if (ret < 0) |
3852 | req_set_fail_links(req); | |
0bdf3398 | 3853 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3854 | return 0; |
3855 | } | |
3856 | ||
45d189c6 | 3857 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3858 | { |
e45cff58 | 3859 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3860 | } |
3861 | ||
067524e9 JA |
3862 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3863 | const struct io_uring_sqe *sqe) | |
3864 | { | |
3865 | struct io_provide_buf *p = &req->pbuf; | |
3866 | u64 tmp; | |
3867 | ||
3868 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3869 | return -EINVAL; | |
3870 | ||
3871 | tmp = READ_ONCE(sqe->fd); | |
3872 | if (!tmp || tmp > USHRT_MAX) | |
3873 | return -EINVAL; | |
3874 | ||
3875 | memset(p, 0, sizeof(*p)); | |
3876 | p->nbufs = tmp; | |
3877 | p->bgid = READ_ONCE(sqe->buf_group); | |
3878 | return 0; | |
3879 | } | |
3880 | ||
3881 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3882 | int bgid, unsigned nbufs) | |
3883 | { | |
3884 | unsigned i = 0; | |
3885 | ||
3886 | /* shouldn't happen */ | |
3887 | if (!nbufs) | |
3888 | return 0; | |
3889 | ||
3890 | /* the head kbuf is the list itself */ | |
3891 | while (!list_empty(&buf->list)) { | |
3892 | struct io_buffer *nxt; | |
3893 | ||
3894 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3895 | list_del(&nxt->list); | |
3896 | kfree(nxt); | |
3897 | if (++i == nbufs) | |
3898 | return i; | |
3899 | } | |
3900 | i++; | |
3901 | kfree(buf); | |
9e15c3a0 | 3902 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3903 | |
3904 | return i; | |
3905 | } | |
3906 | ||
889fca73 | 3907 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3908 | { |
3909 | struct io_provide_buf *p = &req->pbuf; | |
3910 | struct io_ring_ctx *ctx = req->ctx; | |
3911 | struct io_buffer *head; | |
3912 | int ret = 0; | |
45d189c6 | 3913 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3914 | |
3915 | io_ring_submit_lock(ctx, !force_nonblock); | |
3916 | ||
3917 | lockdep_assert_held(&ctx->uring_lock); | |
3918 | ||
3919 | ret = -ENOENT; | |
9e15c3a0 | 3920 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3921 | if (head) |
3922 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3923 | if (ret < 0) |
3924 | req_set_fail_links(req); | |
067524e9 | 3925 | |
9fb8cb49 PB |
3926 | /* complete before unlock, IOPOLL may need the lock */ |
3927 | __io_req_complete(req, issue_flags, ret, 0); | |
3928 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3929 | return 0; |
3930 | } | |
3931 | ||
ddf0322d JA |
3932 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3933 | const struct io_uring_sqe *sqe) | |
3934 | { | |
38134ada | 3935 | unsigned long size, tmp_check; |
ddf0322d JA |
3936 | struct io_provide_buf *p = &req->pbuf; |
3937 | u64 tmp; | |
3938 | ||
3939 | if (sqe->ioprio || sqe->rw_flags) | |
3940 | return -EINVAL; | |
3941 | ||
3942 | tmp = READ_ONCE(sqe->fd); | |
3943 | if (!tmp || tmp > USHRT_MAX) | |
3944 | return -E2BIG; | |
3945 | p->nbufs = tmp; | |
3946 | p->addr = READ_ONCE(sqe->addr); | |
3947 | p->len = READ_ONCE(sqe->len); | |
3948 | ||
38134ada PB |
3949 | if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, |
3950 | &size)) | |
3951 | return -EOVERFLOW; | |
3952 | if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) | |
3953 | return -EOVERFLOW; | |
3954 | ||
d81269fe PB |
3955 | size = (unsigned long)p->len * p->nbufs; |
3956 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3957 | return -EFAULT; |
3958 | ||
3959 | p->bgid = READ_ONCE(sqe->buf_group); | |
3960 | tmp = READ_ONCE(sqe->off); | |
3961 | if (tmp > USHRT_MAX) | |
3962 | return -E2BIG; | |
3963 | p->bid = tmp; | |
3964 | return 0; | |
3965 | } | |
3966 | ||
3967 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3968 | { | |
3969 | struct io_buffer *buf; | |
3970 | u64 addr = pbuf->addr; | |
3971 | int i, bid = pbuf->bid; | |
3972 | ||
3973 | for (i = 0; i < pbuf->nbufs; i++) { | |
3974 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3975 | if (!buf) | |
3976 | break; | |
3977 | ||
3978 | buf->addr = addr; | |
3979 | buf->len = pbuf->len; | |
3980 | buf->bid = bid; | |
3981 | addr += pbuf->len; | |
3982 | bid++; | |
3983 | if (!*head) { | |
3984 | INIT_LIST_HEAD(&buf->list); | |
3985 | *head = buf; | |
3986 | } else { | |
3987 | list_add_tail(&buf->list, &(*head)->list); | |
3988 | } | |
3989 | } | |
3990 | ||
3991 | return i ? i : -ENOMEM; | |
3992 | } | |
3993 | ||
889fca73 | 3994 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
3995 | { |
3996 | struct io_provide_buf *p = &req->pbuf; | |
3997 | struct io_ring_ctx *ctx = req->ctx; | |
3998 | struct io_buffer *head, *list; | |
3999 | int ret = 0; | |
45d189c6 | 4000 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4001 | |
4002 | io_ring_submit_lock(ctx, !force_nonblock); | |
4003 | ||
4004 | lockdep_assert_held(&ctx->uring_lock); | |
4005 | ||
9e15c3a0 | 4006 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4007 | |
4008 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4009 | if (ret >= 0 && !list) { |
4010 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4011 | if (ret < 0) | |
067524e9 | 4012 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4013 | } |
ddf0322d JA |
4014 | if (ret < 0) |
4015 | req_set_fail_links(req); | |
9fb8cb49 PB |
4016 | /* complete before unlock, IOPOLL may need the lock */ |
4017 | __io_req_complete(req, issue_flags, ret, 0); | |
4018 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4019 | return 0; |
cebdb986 JA |
4020 | } |
4021 | ||
3e4827b0 JA |
4022 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4023 | const struct io_uring_sqe *sqe) | |
4024 | { | |
4025 | #if defined(CONFIG_EPOLL) | |
4026 | if (sqe->ioprio || sqe->buf_index) | |
4027 | return -EINVAL; | |
6ca56f84 | 4028 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4029 | return -EINVAL; |
3e4827b0 JA |
4030 | |
4031 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4032 | req->epoll.op = READ_ONCE(sqe->len); | |
4033 | req->epoll.fd = READ_ONCE(sqe->off); | |
4034 | ||
4035 | if (ep_op_has_event(req->epoll.op)) { | |
4036 | struct epoll_event __user *ev; | |
4037 | ||
4038 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4039 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4040 | return -EFAULT; | |
4041 | } | |
4042 | ||
4043 | return 0; | |
4044 | #else | |
4045 | return -EOPNOTSUPP; | |
4046 | #endif | |
4047 | } | |
4048 | ||
889fca73 | 4049 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4050 | { |
4051 | #if defined(CONFIG_EPOLL) | |
4052 | struct io_epoll *ie = &req->epoll; | |
4053 | int ret; | |
45d189c6 | 4054 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4055 | |
4056 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4057 | if (force_nonblock && ret == -EAGAIN) | |
4058 | return -EAGAIN; | |
4059 | ||
4060 | if (ret < 0) | |
4061 | req_set_fail_links(req); | |
889fca73 | 4062 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4063 | return 0; |
4064 | #else | |
4065 | return -EOPNOTSUPP; | |
4066 | #endif | |
4067 | } | |
4068 | ||
c1ca757b JA |
4069 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4070 | { | |
4071 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4072 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4073 | return -EINVAL; | |
3232dd02 PB |
4074 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4075 | return -EINVAL; | |
c1ca757b JA |
4076 | |
4077 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4078 | req->madvise.len = READ_ONCE(sqe->len); | |
4079 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4080 | return 0; | |
4081 | #else | |
4082 | return -EOPNOTSUPP; | |
4083 | #endif | |
4084 | } | |
4085 | ||
45d189c6 | 4086 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4087 | { |
4088 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4089 | struct io_madvise *ma = &req->madvise; | |
4090 | int ret; | |
4091 | ||
45d189c6 | 4092 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4093 | return -EAGAIN; |
4094 | ||
0726b01e | 4095 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4096 | if (ret < 0) |
4097 | req_set_fail_links(req); | |
e1e16097 | 4098 | io_req_complete(req, ret); |
c1ca757b JA |
4099 | return 0; |
4100 | #else | |
4101 | return -EOPNOTSUPP; | |
4102 | #endif | |
4103 | } | |
4104 | ||
4840e418 JA |
4105 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4106 | { | |
4107 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4108 | return -EINVAL; | |
3232dd02 PB |
4109 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4110 | return -EINVAL; | |
4840e418 JA |
4111 | |
4112 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4113 | req->fadvise.len = READ_ONCE(sqe->len); | |
4114 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4115 | return 0; | |
4116 | } | |
4117 | ||
45d189c6 | 4118 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4119 | { |
4120 | struct io_fadvise *fa = &req->fadvise; | |
4121 | int ret; | |
4122 | ||
45d189c6 | 4123 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4124 | switch (fa->advice) { |
4125 | case POSIX_FADV_NORMAL: | |
4126 | case POSIX_FADV_RANDOM: | |
4127 | case POSIX_FADV_SEQUENTIAL: | |
4128 | break; | |
4129 | default: | |
4130 | return -EAGAIN; | |
4131 | } | |
4132 | } | |
4840e418 JA |
4133 | |
4134 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4135 | if (ret < 0) | |
4136 | req_set_fail_links(req); | |
0bdf3398 | 4137 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4138 | return 0; |
4139 | } | |
4140 | ||
eddc7ef5 JA |
4141 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4142 | { | |
6ca56f84 | 4143 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4144 | return -EINVAL; |
eddc7ef5 JA |
4145 | if (sqe->ioprio || sqe->buf_index) |
4146 | return -EINVAL; | |
9c280f90 | 4147 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4148 | return -EBADF; |
eddc7ef5 | 4149 | |
1d9e1288 BM |
4150 | req->statx.dfd = READ_ONCE(sqe->fd); |
4151 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4152 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4153 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4154 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4155 | |
4156 | return 0; | |
4157 | } | |
4158 | ||
45d189c6 | 4159 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4160 | { |
1d9e1288 | 4161 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4162 | int ret; |
4163 | ||
59d70013 | 4164 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4165 | return -EAGAIN; |
4166 | ||
e62753e4 BM |
4167 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4168 | ctx->buffer); | |
eddc7ef5 | 4169 | |
eddc7ef5 JA |
4170 | if (ret < 0) |
4171 | req_set_fail_links(req); | |
e1e16097 | 4172 | io_req_complete(req, ret); |
eddc7ef5 JA |
4173 | return 0; |
4174 | } | |
4175 | ||
b5dba59e JA |
4176 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4177 | { | |
14587a46 | 4178 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4179 | return -EINVAL; |
b5dba59e JA |
4180 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4181 | sqe->rw_flags || sqe->buf_index) | |
4182 | return -EINVAL; | |
9c280f90 | 4183 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4184 | return -EBADF; |
b5dba59e JA |
4185 | |
4186 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4187 | return 0; |
b5dba59e JA |
4188 | } |
4189 | ||
889fca73 | 4190 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4191 | { |
9eac1904 | 4192 | struct files_struct *files = current->files; |
3af73b28 | 4193 | struct io_close *close = &req->close; |
9eac1904 | 4194 | struct fdtable *fdt; |
a1fde923 PB |
4195 | struct file *file = NULL; |
4196 | int ret = -EBADF; | |
b5dba59e | 4197 | |
9eac1904 JA |
4198 | spin_lock(&files->file_lock); |
4199 | fdt = files_fdtable(files); | |
4200 | if (close->fd >= fdt->max_fds) { | |
4201 | spin_unlock(&files->file_lock); | |
4202 | goto err; | |
4203 | } | |
4204 | file = fdt->fd[close->fd]; | |
a1fde923 | 4205 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4206 | spin_unlock(&files->file_lock); |
4207 | file = NULL; | |
4208 | goto err; | |
3af73b28 | 4209 | } |
b5dba59e JA |
4210 | |
4211 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4212 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4213 | spin_unlock(&files->file_lock); |
0bf0eefd | 4214 | return -EAGAIN; |
a2100672 | 4215 | } |
b5dba59e | 4216 | |
9eac1904 JA |
4217 | ret = __close_fd_get_file(close->fd, &file); |
4218 | spin_unlock(&files->file_lock); | |
4219 | if (ret < 0) { | |
4220 | if (ret == -ENOENT) | |
4221 | ret = -EBADF; | |
4222 | goto err; | |
4223 | } | |
4224 | ||
3af73b28 | 4225 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4226 | ret = filp_close(file, current->files); |
4227 | err: | |
3af73b28 PB |
4228 | if (ret < 0) |
4229 | req_set_fail_links(req); | |
9eac1904 JA |
4230 | if (file) |
4231 | fput(file); | |
889fca73 | 4232 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4233 | return 0; |
b5dba59e JA |
4234 | } |
4235 | ||
1155c76a | 4236 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4237 | { |
4238 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4239 | |
5d17b4a4 JA |
4240 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4241 | return -EINVAL; | |
4242 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4243 | return -EINVAL; | |
4244 | ||
8ed8d3c3 JA |
4245 | req->sync.off = READ_ONCE(sqe->off); |
4246 | req->sync.len = READ_ONCE(sqe->len); | |
4247 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4248 | return 0; |
4249 | } | |
4250 | ||
45d189c6 | 4251 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4252 | { |
8ed8d3c3 JA |
4253 | int ret; |
4254 | ||
ac45abc0 | 4255 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4256 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4257 | return -EAGAIN; |
4258 | ||
9adbd45d | 4259 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4260 | req->sync.flags); |
4261 | if (ret < 0) | |
4262 | req_set_fail_links(req); | |
e1e16097 | 4263 | io_req_complete(req, ret); |
5d17b4a4 JA |
4264 | return 0; |
4265 | } | |
4266 | ||
469956e8 | 4267 | #if defined(CONFIG_NET) |
02d27d89 PB |
4268 | static int io_setup_async_msg(struct io_kiocb *req, |
4269 | struct io_async_msghdr *kmsg) | |
4270 | { | |
e8c2bc1f JA |
4271 | struct io_async_msghdr *async_msg = req->async_data; |
4272 | ||
4273 | if (async_msg) | |
02d27d89 | 4274 | return -EAGAIN; |
e8c2bc1f | 4275 | if (io_alloc_async_data(req)) { |
257e84a5 | 4276 | kfree(kmsg->free_iov); |
02d27d89 PB |
4277 | return -ENOMEM; |
4278 | } | |
e8c2bc1f | 4279 | async_msg = req->async_data; |
02d27d89 | 4280 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4281 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4282 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4283 | /* if were using fast_iov, set it to the new one */ |
4284 | if (!async_msg->free_iov) | |
4285 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4286 | ||
02d27d89 PB |
4287 | return -EAGAIN; |
4288 | } | |
4289 | ||
2ae523ed PB |
4290 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4291 | struct io_async_msghdr *iomsg) | |
4292 | { | |
2ae523ed | 4293 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4294 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4295 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4296 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4297 | } |
4298 | ||
93642ef8 PB |
4299 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4300 | { | |
4301 | int ret; | |
4302 | ||
93642ef8 PB |
4303 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4304 | if (!ret) | |
4305 | req->flags |= REQ_F_NEED_CLEANUP; | |
4306 | return ret; | |
4307 | } | |
4308 | ||
3529d8c2 | 4309 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4310 | { |
e47293fd | 4311 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4312 | |
d2b6f48b PB |
4313 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4314 | return -EINVAL; | |
4315 | ||
270a5940 | 4316 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4317 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4318 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4319 | if (sr->msg_flags & MSG_DONTWAIT) | |
4320 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4321 | |
d8768362 JA |
4322 | #ifdef CONFIG_COMPAT |
4323 | if (req->ctx->compat) | |
4324 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4325 | #endif | |
93642ef8 | 4326 | return 0; |
03b1230c JA |
4327 | } |
4328 | ||
889fca73 | 4329 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4330 | { |
6b754c8b | 4331 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4332 | struct socket *sock; |
7a7cacba | 4333 | unsigned flags; |
0031275d | 4334 | int min_ret = 0; |
0fa03c62 JA |
4335 | int ret; |
4336 | ||
dba4a925 | 4337 | sock = sock_from_file(req->file); |
7a7cacba | 4338 | if (unlikely(!sock)) |
dba4a925 | 4339 | return -ENOTSOCK; |
3529d8c2 | 4340 | |
257e84a5 PB |
4341 | kmsg = req->async_data; |
4342 | if (!kmsg) { | |
7a7cacba PB |
4343 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4344 | if (ret) | |
4345 | return ret; | |
4346 | kmsg = &iomsg; | |
0fa03c62 | 4347 | } |
0fa03c62 | 4348 | |
04411806 PB |
4349 | flags = req->sr_msg.msg_flags; |
4350 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4351 | flags |= MSG_DONTWAIT; |
0031275d SM |
4352 | if (flags & MSG_WAITALL) |
4353 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4354 | ||
7a7cacba | 4355 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4356 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4357 | return io_setup_async_msg(req, kmsg); |
4358 | if (ret == -ERESTARTSYS) | |
4359 | ret = -EINTR; | |
0fa03c62 | 4360 | |
257e84a5 PB |
4361 | /* fast path, check for non-NULL to avoid function call */ |
4362 | if (kmsg->free_iov) | |
4363 | kfree(kmsg->free_iov); | |
99bc4c38 | 4364 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4365 | if (ret < min_ret) |
4e88d6e7 | 4366 | req_set_fail_links(req); |
889fca73 | 4367 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4368 | return 0; |
03b1230c | 4369 | } |
aa1fa28f | 4370 | |
889fca73 | 4371 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4372 | { |
7a7cacba PB |
4373 | struct io_sr_msg *sr = &req->sr_msg; |
4374 | struct msghdr msg; | |
4375 | struct iovec iov; | |
fddaface | 4376 | struct socket *sock; |
7a7cacba | 4377 | unsigned flags; |
0031275d | 4378 | int min_ret = 0; |
fddaface JA |
4379 | int ret; |
4380 | ||
dba4a925 | 4381 | sock = sock_from_file(req->file); |
7a7cacba | 4382 | if (unlikely(!sock)) |
dba4a925 | 4383 | return -ENOTSOCK; |
fddaface | 4384 | |
7a7cacba PB |
4385 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4386 | if (unlikely(ret)) | |
14db8411 | 4387 | return ret; |
fddaface | 4388 | |
7a7cacba PB |
4389 | msg.msg_name = NULL; |
4390 | msg.msg_control = NULL; | |
4391 | msg.msg_controllen = 0; | |
4392 | msg.msg_namelen = 0; | |
fddaface | 4393 | |
04411806 PB |
4394 | flags = req->sr_msg.msg_flags; |
4395 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4396 | flags |= MSG_DONTWAIT; |
0031275d SM |
4397 | if (flags & MSG_WAITALL) |
4398 | min_ret = iov_iter_count(&msg.msg_iter); | |
4399 | ||
7a7cacba PB |
4400 | msg.msg_flags = flags; |
4401 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4402 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4403 | return -EAGAIN; |
4404 | if (ret == -ERESTARTSYS) | |
4405 | ret = -EINTR; | |
fddaface | 4406 | |
0031275d | 4407 | if (ret < min_ret) |
fddaface | 4408 | req_set_fail_links(req); |
889fca73 | 4409 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4410 | return 0; |
fddaface JA |
4411 | } |
4412 | ||
1400e697 PB |
4413 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4414 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4415 | { |
4416 | struct io_sr_msg *sr = &req->sr_msg; | |
4417 | struct iovec __user *uiov; | |
4418 | size_t iov_len; | |
4419 | int ret; | |
4420 | ||
1400e697 PB |
4421 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4422 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4423 | if (ret) |
4424 | return ret; | |
4425 | ||
4426 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4427 | if (iov_len > 1) | |
4428 | return -EINVAL; | |
5476dfed | 4429 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4430 | return -EFAULT; |
5476dfed | 4431 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4432 | iomsg->free_iov = NULL; |
52de1fe1 | 4433 | } else { |
257e84a5 | 4434 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4435 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4436 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4437 | false); |
52de1fe1 JA |
4438 | if (ret > 0) |
4439 | ret = 0; | |
4440 | } | |
4441 | ||
4442 | return ret; | |
4443 | } | |
4444 | ||
4445 | #ifdef CONFIG_COMPAT | |
4446 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4447 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4448 | { |
52de1fe1 JA |
4449 | struct io_sr_msg *sr = &req->sr_msg; |
4450 | struct compat_iovec __user *uiov; | |
4451 | compat_uptr_t ptr; | |
4452 | compat_size_t len; | |
4453 | int ret; | |
4454 | ||
4af3417a PB |
4455 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4456 | &ptr, &len); | |
52de1fe1 JA |
4457 | if (ret) |
4458 | return ret; | |
4459 | ||
4460 | uiov = compat_ptr(ptr); | |
4461 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4462 | compat_ssize_t clen; | |
4463 | ||
4464 | if (len > 1) | |
4465 | return -EINVAL; | |
4466 | if (!access_ok(uiov, sizeof(*uiov))) | |
4467 | return -EFAULT; | |
4468 | if (__get_user(clen, &uiov->iov_len)) | |
4469 | return -EFAULT; | |
4470 | if (clen < 0) | |
4471 | return -EINVAL; | |
2d280bc8 | 4472 | sr->len = clen; |
257e84a5 | 4473 | iomsg->free_iov = NULL; |
52de1fe1 | 4474 | } else { |
257e84a5 | 4475 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4476 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4477 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4478 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4479 | if (ret < 0) |
4480 | return ret; | |
4481 | } | |
4482 | ||
4483 | return 0; | |
4484 | } | |
4485 | #endif | |
4486 | ||
1400e697 PB |
4487 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4488 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4489 | { |
1400e697 | 4490 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4491 | |
4492 | #ifdef CONFIG_COMPAT | |
4493 | if (req->ctx->compat) | |
1400e697 | 4494 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4495 | #endif |
52de1fe1 | 4496 | |
1400e697 | 4497 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4498 | } |
4499 | ||
bcda7baa | 4500 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4501 | bool needs_lock) |
bcda7baa JA |
4502 | { |
4503 | struct io_sr_msg *sr = &req->sr_msg; | |
4504 | struct io_buffer *kbuf; | |
4505 | ||
bcda7baa JA |
4506 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4507 | if (IS_ERR(kbuf)) | |
4508 | return kbuf; | |
4509 | ||
4510 | sr->kbuf = kbuf; | |
4511 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4512 | return kbuf; |
fddaface JA |
4513 | } |
4514 | ||
7fbb1b54 PB |
4515 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4516 | { | |
4517 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4518 | } | |
4519 | ||
93642ef8 | 4520 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4521 | { |
99bc4c38 | 4522 | int ret; |
3529d8c2 | 4523 | |
93642ef8 PB |
4524 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4525 | if (!ret) | |
4526 | req->flags |= REQ_F_NEED_CLEANUP; | |
4527 | return ret; | |
4528 | } | |
4529 | ||
4530 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4531 | { | |
4532 | struct io_sr_msg *sr = &req->sr_msg; | |
4533 | ||
d2b6f48b PB |
4534 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4535 | return -EINVAL; | |
4536 | ||
270a5940 | 4537 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4538 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4539 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4540 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4541 | if (sr->msg_flags & MSG_DONTWAIT) | |
4542 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4543 | |
d8768362 JA |
4544 | #ifdef CONFIG_COMPAT |
4545 | if (req->ctx->compat) | |
4546 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4547 | #endif | |
93642ef8 | 4548 | return 0; |
aa1fa28f JA |
4549 | } |
4550 | ||
889fca73 | 4551 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4552 | { |
6b754c8b | 4553 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4554 | struct socket *sock; |
7fbb1b54 | 4555 | struct io_buffer *kbuf; |
7a7cacba | 4556 | unsigned flags; |
0031275d | 4557 | int min_ret = 0; |
52de1fe1 | 4558 | int ret, cflags = 0; |
45d189c6 | 4559 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4560 | |
dba4a925 | 4561 | sock = sock_from_file(req->file); |
7a7cacba | 4562 | if (unlikely(!sock)) |
dba4a925 | 4563 | return -ENOTSOCK; |
3529d8c2 | 4564 | |
257e84a5 PB |
4565 | kmsg = req->async_data; |
4566 | if (!kmsg) { | |
7a7cacba PB |
4567 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4568 | if (ret) | |
681fda8d | 4569 | return ret; |
7a7cacba PB |
4570 | kmsg = &iomsg; |
4571 | } | |
03b1230c | 4572 | |
bc02ef33 | 4573 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4574 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4575 | if (IS_ERR(kbuf)) |
52de1fe1 | 4576 | return PTR_ERR(kbuf); |
7a7cacba | 4577 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4578 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4579 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4580 | 1, req->sr_msg.len); |
4581 | } | |
52de1fe1 | 4582 | |
04411806 PB |
4583 | flags = req->sr_msg.msg_flags; |
4584 | if (force_nonblock) | |
7a7cacba | 4585 | flags |= MSG_DONTWAIT; |
0031275d SM |
4586 | if (flags & MSG_WAITALL) |
4587 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4588 | ||
7a7cacba PB |
4589 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4590 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4591 | if (force_nonblock && ret == -EAGAIN) |
4592 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4593 | if (ret == -ERESTARTSYS) |
4594 | ret = -EINTR; | |
03b1230c | 4595 | |
7fbb1b54 PB |
4596 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4597 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4598 | /* fast path, check for non-NULL to avoid function call */ |
4599 | if (kmsg->free_iov) | |
4600 | kfree(kmsg->free_iov); | |
99bc4c38 | 4601 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4602 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4603 | req_set_fail_links(req); |
889fca73 | 4604 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4605 | return 0; |
0fa03c62 | 4606 | } |
5d17b4a4 | 4607 | |
889fca73 | 4608 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4609 | { |
6b754c8b | 4610 | struct io_buffer *kbuf; |
7a7cacba PB |
4611 | struct io_sr_msg *sr = &req->sr_msg; |
4612 | struct msghdr msg; | |
4613 | void __user *buf = sr->buf; | |
fddaface | 4614 | struct socket *sock; |
7a7cacba PB |
4615 | struct iovec iov; |
4616 | unsigned flags; | |
0031275d | 4617 | int min_ret = 0; |
bcda7baa | 4618 | int ret, cflags = 0; |
45d189c6 | 4619 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4620 | |
dba4a925 | 4621 | sock = sock_from_file(req->file); |
7a7cacba | 4622 | if (unlikely(!sock)) |
dba4a925 | 4623 | return -ENOTSOCK; |
fddaface | 4624 | |
bc02ef33 | 4625 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4626 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4627 | if (IS_ERR(kbuf)) |
4628 | return PTR_ERR(kbuf); | |
7a7cacba | 4629 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4630 | } |
bcda7baa | 4631 | |
7a7cacba | 4632 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4633 | if (unlikely(ret)) |
4634 | goto out_free; | |
fddaface | 4635 | |
7a7cacba PB |
4636 | msg.msg_name = NULL; |
4637 | msg.msg_control = NULL; | |
4638 | msg.msg_controllen = 0; | |
4639 | msg.msg_namelen = 0; | |
4640 | msg.msg_iocb = NULL; | |
4641 | msg.msg_flags = 0; | |
fddaface | 4642 | |
04411806 PB |
4643 | flags = req->sr_msg.msg_flags; |
4644 | if (force_nonblock) | |
7a7cacba | 4645 | flags |= MSG_DONTWAIT; |
0031275d SM |
4646 | if (flags & MSG_WAITALL) |
4647 | min_ret = iov_iter_count(&msg.msg_iter); | |
4648 | ||
7a7cacba PB |
4649 | ret = sock_recvmsg(sock, &msg, flags); |
4650 | if (force_nonblock && ret == -EAGAIN) | |
4651 | return -EAGAIN; | |
4652 | if (ret == -ERESTARTSYS) | |
4653 | ret = -EINTR; | |
14c32eee | 4654 | out_free: |
7fbb1b54 PB |
4655 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4656 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4657 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4658 | req_set_fail_links(req); |
889fca73 | 4659 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4660 | return 0; |
fddaface JA |
4661 | } |
4662 | ||
3529d8c2 | 4663 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4664 | { |
8ed8d3c3 JA |
4665 | struct io_accept *accept = &req->accept; |
4666 | ||
14587a46 | 4667 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4668 | return -EINVAL; |
8042d6ce | 4669 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4670 | return -EINVAL; |
4671 | ||
d55e5f5b JA |
4672 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4673 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4674 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4675 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4676 | return 0; |
8ed8d3c3 | 4677 | } |
17f2fe35 | 4678 | |
889fca73 | 4679 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4680 | { |
4681 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4682 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4683 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4684 | int ret; |
4685 | ||
e697deed JX |
4686 | if (req->file->f_flags & O_NONBLOCK) |
4687 | req->flags |= REQ_F_NOWAIT; | |
4688 | ||
8ed8d3c3 | 4689 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4690 | accept->addr_len, accept->flags, |
4691 | accept->nofile); | |
8ed8d3c3 | 4692 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4693 | return -EAGAIN; |
ac45abc0 PB |
4694 | if (ret < 0) { |
4695 | if (ret == -ERESTARTSYS) | |
4696 | ret = -EINTR; | |
4e88d6e7 | 4697 | req_set_fail_links(req); |
ac45abc0 | 4698 | } |
889fca73 | 4699 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4700 | return 0; |
8ed8d3c3 JA |
4701 | } |
4702 | ||
93642ef8 PB |
4703 | static int io_connect_prep_async(struct io_kiocb *req) |
4704 | { | |
4705 | struct io_async_connect *io = req->async_data; | |
4706 | struct io_connect *conn = &req->connect; | |
4707 | ||
4708 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4709 | } | |
4710 | ||
3529d8c2 | 4711 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4712 | { |
3529d8c2 | 4713 | struct io_connect *conn = &req->connect; |
f499a021 | 4714 | |
14587a46 | 4715 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4716 | return -EINVAL; |
4717 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4718 | return -EINVAL; | |
4719 | ||
3529d8c2 JA |
4720 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4721 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4722 | return 0; |
f499a021 JA |
4723 | } |
4724 | ||
889fca73 | 4725 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4726 | { |
e8c2bc1f | 4727 | struct io_async_connect __io, *io; |
f8e85cf2 | 4728 | unsigned file_flags; |
3fbb51c1 | 4729 | int ret; |
45d189c6 | 4730 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4731 | |
e8c2bc1f JA |
4732 | if (req->async_data) { |
4733 | io = req->async_data; | |
f499a021 | 4734 | } else { |
3529d8c2 JA |
4735 | ret = move_addr_to_kernel(req->connect.addr, |
4736 | req->connect.addr_len, | |
e8c2bc1f | 4737 | &__io.address); |
f499a021 JA |
4738 | if (ret) |
4739 | goto out; | |
4740 | io = &__io; | |
4741 | } | |
4742 | ||
3fbb51c1 JA |
4743 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4744 | ||
e8c2bc1f | 4745 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4746 | req->connect.addr_len, file_flags); |
87f80d62 | 4747 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4748 | if (req->async_data) |
b7bb4f7d | 4749 | return -EAGAIN; |
e8c2bc1f | 4750 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4751 | ret = -ENOMEM; |
4752 | goto out; | |
4753 | } | |
e8c2bc1f | 4754 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4755 | return -EAGAIN; |
f499a021 | 4756 | } |
f8e85cf2 JA |
4757 | if (ret == -ERESTARTSYS) |
4758 | ret = -EINTR; | |
f499a021 | 4759 | out: |
4e88d6e7 JA |
4760 | if (ret < 0) |
4761 | req_set_fail_links(req); | |
889fca73 | 4762 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4763 | return 0; |
469956e8 Y |
4764 | } |
4765 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4766 | #define IO_NETOP_FN(op) \ |
4767 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4768 | { \ | |
4769 | return -EOPNOTSUPP; \ | |
4770 | } | |
4771 | ||
4772 | #define IO_NETOP_PREP(op) \ | |
4773 | IO_NETOP_FN(op) \ | |
4774 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4775 | { \ | |
4776 | return -EOPNOTSUPP; \ | |
4777 | } \ | |
4778 | ||
4779 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4780 | IO_NETOP_PREP(op) \ | |
4781 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4782 | { \ | |
4783 | return -EOPNOTSUPP; \ | |
4784 | } | |
4785 | ||
4786 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4787 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4788 | IO_NETOP_PREP_ASYNC(connect); | |
4789 | IO_NETOP_PREP(accept); | |
4790 | IO_NETOP_FN(send); | |
4791 | IO_NETOP_FN(recv); | |
469956e8 | 4792 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4793 | |
d7718a9d JA |
4794 | struct io_poll_table { |
4795 | struct poll_table_struct pt; | |
4796 | struct io_kiocb *req; | |
4797 | int error; | |
4798 | }; | |
ce593a6c | 4799 | |
d7718a9d JA |
4800 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4801 | __poll_t mask, task_work_func_t func) | |
4802 | { | |
aa96bf8a | 4803 | int ret; |
d7718a9d JA |
4804 | |
4805 | /* for instances that support it check for an event match first: */ | |
4806 | if (mask && !(mask & poll->events)) | |
4807 | return 0; | |
4808 | ||
4809 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4810 | ||
4811 | list_del_init(&poll->wait.entry); | |
4812 | ||
d7718a9d | 4813 | req->result = mask; |
7cbf1722 | 4814 | req->task_work.func = func; |
6d816e08 | 4815 | |
d7718a9d | 4816 | /* |
e3aabf95 JA |
4817 | * If this fails, then the task is exiting. When a task exits, the |
4818 | * work gets canceled, so just cancel this request as well instead | |
4819 | * of executing it. We can't safely execute it anyway, as we may not | |
4820 | * have the needed state needed for it anyway. | |
d7718a9d | 4821 | */ |
355fb9e2 | 4822 | ret = io_req_task_work_add(req); |
aa96bf8a | 4823 | if (unlikely(ret)) { |
e3aabf95 | 4824 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4825 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4826 | } |
d7718a9d JA |
4827 | return 1; |
4828 | } | |
4829 | ||
74ce6ce4 JA |
4830 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4831 | __acquires(&req->ctx->completion_lock) | |
4832 | { | |
4833 | struct io_ring_ctx *ctx = req->ctx; | |
4834 | ||
4835 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4836 | struct poll_table_struct pt = { ._key = poll->events }; | |
4837 | ||
4838 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4839 | } | |
4840 | ||
4841 | spin_lock_irq(&ctx->completion_lock); | |
4842 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4843 | add_wait_queue(poll->head, &poll->wait); | |
4844 | return true; | |
4845 | } | |
4846 | ||
4847 | return false; | |
4848 | } | |
4849 | ||
d4e7cd36 | 4850 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4851 | { |
e8c2bc1f | 4852 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4853 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4854 | return req->async_data; |
d4e7cd36 JA |
4855 | return req->apoll->double_poll; |
4856 | } | |
4857 | ||
4858 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4859 | { | |
4860 | if (req->opcode == IORING_OP_POLL_ADD) | |
4861 | return &req->poll; | |
4862 | return &req->apoll->poll; | |
4863 | } | |
4864 | ||
4865 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4866 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4867 | { |
4868 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4869 | |
4870 | lockdep_assert_held(&req->ctx->completion_lock); | |
4871 | ||
4872 | if (poll && poll->head) { | |
4873 | struct wait_queue_head *head = poll->head; | |
4874 | ||
4875 | spin_lock(&head->lock); | |
4876 | list_del_init(&poll->wait.entry); | |
4877 | if (poll->wait.private) | |
de9b4cca | 4878 | req_ref_put(req); |
18bceab1 JA |
4879 | poll->head = NULL; |
4880 | spin_unlock(&head->lock); | |
4881 | } | |
4882 | } | |
4883 | ||
e27414be | 4884 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4885 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4886 | { |
4887 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4888 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4889 | int error; |
18bceab1 | 4890 | |
e27414be | 4891 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4892 | error = -ECANCELED; |
88e41cf9 | 4893 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4894 | } else { |
5082620f | 4895 | error = mangle_poll(mask); |
e27414be | 4896 | } |
b69de288 JA |
4897 | if (req->poll.events & EPOLLONESHOT) |
4898 | flags = 0; | |
d4d19c19 | 4899 | if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) { |
5082620f | 4900 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4901 | req->poll.done = true; |
4902 | flags = 0; | |
4903 | } | |
18bceab1 | 4904 | io_commit_cqring(ctx); |
88e41cf9 | 4905 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4906 | } |
4907 | ||
dd221f46 | 4908 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4909 | { |
dd221f46 | 4910 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4911 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4912 | struct io_kiocb *nxt; |
18bceab1 JA |
4913 | |
4914 | if (io_poll_rewait(req, &req->poll)) { | |
4915 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4916 | } else { |
f40b964a | 4917 | bool done; |
88e41cf9 | 4918 | |
e27414be | 4919 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4920 | if (done) { |
4921 | hash_del(&req->hash_node); | |
f40b964a | 4922 | } else { |
88e41cf9 JA |
4923 | req->result = 0; |
4924 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4925 | } | |
dd221f46 | 4926 | spin_unlock_irq(&ctx->completion_lock); |
f40b964a | 4927 | io_cqring_ev_posted(ctx); |
18bceab1 | 4928 | |
88e41cf9 JA |
4929 | if (done) { |
4930 | nxt = io_put_req_find_next(req); | |
4931 | if (nxt) | |
4932 | __io_req_task_submit(nxt); | |
4933 | } | |
dd221f46 | 4934 | } |
18bceab1 JA |
4935 | } |
4936 | ||
4937 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4938 | int sync, void *key) | |
4939 | { | |
4940 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4941 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4942 | __poll_t mask = key_to_poll(key); |
4943 | ||
4944 | /* for instances that support it check for an event match first: */ | |
4945 | if (mask && !(mask & poll->events)) | |
4946 | return 0; | |
88e41cf9 JA |
4947 | if (!(poll->events & EPOLLONESHOT)) |
4948 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4949 | |
8706e04e JA |
4950 | list_del_init(&wait->entry); |
4951 | ||
807abcb0 | 4952 | if (poll && poll->head) { |
18bceab1 JA |
4953 | bool done; |
4954 | ||
807abcb0 JA |
4955 | spin_lock(&poll->head->lock); |
4956 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4957 | if (!done) |
807abcb0 | 4958 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4959 | /* make sure double remove sees this as being gone */ |
4960 | wait->private = NULL; | |
807abcb0 | 4961 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4962 | if (!done) { |
4963 | /* use wait func handler, so it matches the rq type */ | |
4964 | poll->wait.func(&poll->wait, mode, sync, key); | |
4965 | } | |
18bceab1 | 4966 | } |
de9b4cca | 4967 | req_ref_put(req); |
18bceab1 JA |
4968 | return 1; |
4969 | } | |
4970 | ||
4971 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4972 | wait_queue_func_t wake_func) | |
4973 | { | |
4974 | poll->head = NULL; | |
4975 | poll->done = false; | |
4976 | poll->canceled = false; | |
464dca61 JA |
4977 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
4978 | /* mask in events that we always want/need */ | |
4979 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
4980 | INIT_LIST_HEAD(&poll->wait.entry); |
4981 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4982 | } | |
4983 | ||
4984 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4985 | struct wait_queue_head *head, |
4986 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4987 | { |
4988 | struct io_kiocb *req = pt->req; | |
4989 | ||
4990 | /* | |
4991 | * If poll->head is already set, it's because the file being polled | |
4992 | * uses multiple waitqueues for poll handling (eg one for read, one | |
4993 | * for write). Setup a separate io_poll_iocb if this happens. | |
4994 | */ | |
4995 | if (unlikely(poll->head)) { | |
58852d4d PB |
4996 | struct io_poll_iocb *poll_one = poll; |
4997 | ||
18bceab1 | 4998 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 4999 | if (*poll_ptr) { |
18bceab1 JA |
5000 | pt->error = -EINVAL; |
5001 | return; | |
5002 | } | |
ea6a693d JA |
5003 | /* |
5004 | * Can't handle multishot for double wait for now, turn it | |
5005 | * into one-shot mode. | |
5006 | */ | |
5007 | if (!(req->poll.events & EPOLLONESHOT)) | |
5008 | req->poll.events |= EPOLLONESHOT; | |
1c3b3e65 JA |
5009 | /* double add on the same waitqueue head, ignore */ |
5010 | if (poll->head == head) | |
5011 | return; | |
18bceab1 JA |
5012 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5013 | if (!poll) { | |
5014 | pt->error = -ENOMEM; | |
5015 | return; | |
5016 | } | |
58852d4d | 5017 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5018 | req_ref_get(req); |
18bceab1 | 5019 | poll->wait.private = req; |
807abcb0 | 5020 | *poll_ptr = poll; |
18bceab1 JA |
5021 | } |
5022 | ||
5023 | pt->error = 0; | |
5024 | poll->head = head; | |
a31eb4a2 JX |
5025 | |
5026 | if (poll->events & EPOLLEXCLUSIVE) | |
5027 | add_wait_queue_exclusive(head, &poll->wait); | |
5028 | else | |
5029 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5030 | } |
5031 | ||
5032 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5033 | struct poll_table_struct *p) | |
5034 | { | |
5035 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5036 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5037 | |
807abcb0 | 5038 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5039 | } |
5040 | ||
d7718a9d JA |
5041 | static void io_async_task_func(struct callback_head *cb) |
5042 | { | |
5043 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5044 | struct async_poll *apoll = req->apoll; | |
5045 | struct io_ring_ctx *ctx = req->ctx; | |
5046 | ||
5047 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5048 | ||
74ce6ce4 | 5049 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5050 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5051 | return; |
d7718a9d JA |
5052 | } |
5053 | ||
0ea13b44 | 5054 | hash_del(&req->hash_node); |
d4e7cd36 | 5055 | io_poll_remove_double(req); |
74ce6ce4 JA |
5056 | spin_unlock_irq(&ctx->completion_lock); |
5057 | ||
0be0b0e3 PB |
5058 | if (!READ_ONCE(apoll->poll.canceled)) |
5059 | __io_req_task_submit(req); | |
5060 | else | |
2593553a | 5061 | io_req_complete_failed(req, -ECANCELED); |
d7718a9d JA |
5062 | } |
5063 | ||
5064 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5065 | void *key) | |
5066 | { | |
5067 | struct io_kiocb *req = wait->private; | |
5068 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5069 | ||
5070 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5071 | key_to_poll(key)); | |
5072 | ||
5073 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5074 | } | |
5075 | ||
5076 | static void io_poll_req_insert(struct io_kiocb *req) | |
5077 | { | |
5078 | struct io_ring_ctx *ctx = req->ctx; | |
5079 | struct hlist_head *list; | |
5080 | ||
5081 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5082 | hlist_add_head(&req->hash_node, list); | |
5083 | } | |
5084 | ||
5085 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5086 | struct io_poll_iocb *poll, | |
5087 | struct io_poll_table *ipt, __poll_t mask, | |
5088 | wait_queue_func_t wake_func) | |
5089 | __acquires(&ctx->completion_lock) | |
5090 | { | |
5091 | struct io_ring_ctx *ctx = req->ctx; | |
5092 | bool cancel = false; | |
5093 | ||
4d52f338 | 5094 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5095 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5096 | poll->file = req->file; |
18bceab1 | 5097 | poll->wait.private = req; |
d7718a9d JA |
5098 | |
5099 | ipt->pt._key = mask; | |
5100 | ipt->req = req; | |
5101 | ipt->error = -EINVAL; | |
5102 | ||
d7718a9d JA |
5103 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5104 | ||
5105 | spin_lock_irq(&ctx->completion_lock); | |
5106 | if (likely(poll->head)) { | |
5107 | spin_lock(&poll->head->lock); | |
5108 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5109 | if (ipt->error) | |
5110 | cancel = true; | |
5111 | ipt->error = 0; | |
5112 | mask = 0; | |
5113 | } | |
88e41cf9 | 5114 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5115 | list_del_init(&poll->wait.entry); |
5116 | else if (cancel) | |
5117 | WRITE_ONCE(poll->canceled, true); | |
5118 | else if (!poll->done) /* actually waiting for an event */ | |
5119 | io_poll_req_insert(req); | |
5120 | spin_unlock(&poll->head->lock); | |
5121 | } | |
5122 | ||
5123 | return mask; | |
5124 | } | |
5125 | ||
5126 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5127 | { | |
5128 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5129 | struct io_ring_ctx *ctx = req->ctx; | |
5130 | struct async_poll *apoll; | |
5131 | struct io_poll_table ipt; | |
5132 | __poll_t mask, ret; | |
9dab14b8 | 5133 | int rw; |
d7718a9d JA |
5134 | |
5135 | if (!req->file || !file_can_poll(req->file)) | |
5136 | return false; | |
24c74678 | 5137 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5138 | return false; |
9dab14b8 JA |
5139 | if (def->pollin) |
5140 | rw = READ; | |
5141 | else if (def->pollout) | |
5142 | rw = WRITE; | |
5143 | else | |
5144 | return false; | |
5145 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5146 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5147 | return false; |
5148 | ||
5149 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5150 | if (unlikely(!apoll)) | |
5151 | return false; | |
807abcb0 | 5152 | apoll->double_poll = NULL; |
d7718a9d JA |
5153 | |
5154 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5155 | req->apoll = apoll; |
d7718a9d | 5156 | |
88e41cf9 | 5157 | mask = EPOLLONESHOT; |
d7718a9d | 5158 | if (def->pollin) |
8755d97a | 5159 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5160 | if (def->pollout) |
5161 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5162 | |
5163 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5164 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5165 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5166 | mask &= ~POLLIN; | |
5167 | ||
d7718a9d JA |
5168 | mask |= POLLERR | POLLPRI; |
5169 | ||
5170 | ipt.pt._qproc = io_async_queue_proc; | |
5171 | ||
5172 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5173 | io_async_wake); | |
a36da65c | 5174 | if (ret || ipt.error) { |
d4e7cd36 | 5175 | io_poll_remove_double(req); |
d7718a9d | 5176 | spin_unlock_irq(&ctx->completion_lock); |
d7718a9d JA |
5177 | return false; |
5178 | } | |
5179 | spin_unlock_irq(&ctx->completion_lock); | |
5180 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5181 | apoll->poll.events); | |
5182 | return true; | |
5183 | } | |
5184 | ||
5185 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5186 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5187 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5188 | { |
b41e9852 | 5189 | bool do_complete = false; |
221c5eb2 | 5190 | |
5082620f JA |
5191 | if (!poll->head) |
5192 | return false; | |
221c5eb2 | 5193 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5194 | if (do_cancel) |
5195 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5196 | if (!list_empty(&poll->wait.entry)) { |
5197 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5198 | do_complete = true; |
221c5eb2 JA |
5199 | } |
5200 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5201 | hash_del(&req->hash_node); |
d7718a9d JA |
5202 | return do_complete; |
5203 | } | |
5204 | ||
b2c3f7e1 | 5205 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5206 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5207 | { |
5208 | bool do_complete; | |
5209 | ||
d4e7cd36 | 5210 | io_poll_remove_double(req); |
e31001a3 | 5211 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5212 | |
e31001a3 | 5213 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
d7718a9d | 5214 | /* non-poll requests have submit ref still */ |
e31001a3 | 5215 | req_ref_put(req); |
b1f573bd | 5216 | } |
b2c3f7e1 JA |
5217 | return do_complete; |
5218 | } | |
5219 | ||
5220 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5221 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5222 | { |
5223 | bool do_complete; | |
5224 | ||
5225 | do_complete = io_poll_remove_waitqs(req); | |
b41e9852 | 5226 | if (do_complete) { |
d4d19c19 | 5227 | io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0); |
b41e9852 | 5228 | io_commit_cqring(req->ctx); |
f254ac04 | 5229 | req_set_fail_links(req); |
216578e5 | 5230 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5231 | } |
5232 | ||
5233 | return do_complete; | |
221c5eb2 JA |
5234 | } |
5235 | ||
76e1b642 JA |
5236 | /* |
5237 | * Returns true if we found and killed one or more poll requests | |
5238 | */ | |
6b81928d PB |
5239 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5240 | struct files_struct *files) | |
221c5eb2 | 5241 | { |
78076bb6 | 5242 | struct hlist_node *tmp; |
221c5eb2 | 5243 | struct io_kiocb *req; |
8e2e1faf | 5244 | int posted = 0, i; |
221c5eb2 JA |
5245 | |
5246 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5247 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5248 | struct hlist_head *list; | |
5249 | ||
5250 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5251 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5252 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5253 | posted += io_poll_remove_one(req); |
5254 | } | |
221c5eb2 JA |
5255 | } |
5256 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5257 | |
8e2e1faf JA |
5258 | if (posted) |
5259 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5260 | |
5261 | return posted != 0; | |
221c5eb2 JA |
5262 | } |
5263 | ||
9ba5fac8 PB |
5264 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5265 | bool poll_only) | |
e07785b0 | 5266 | __must_hold(&ctx->completion_lock) |
47f46768 | 5267 | { |
78076bb6 | 5268 | struct hlist_head *list; |
47f46768 JA |
5269 | struct io_kiocb *req; |
5270 | ||
78076bb6 JA |
5271 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5272 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5273 | if (sqe_addr != req->user_data) |
5274 | continue; | |
9ba5fac8 PB |
5275 | if (poll_only && req->opcode != IORING_OP_POLL_ADD) |
5276 | continue; | |
b2cb805f | 5277 | return req; |
47f46768 | 5278 | } |
b2cb805f JA |
5279 | return NULL; |
5280 | } | |
5281 | ||
9ba5fac8 PB |
5282 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr, |
5283 | bool poll_only) | |
e07785b0 | 5284 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5285 | { |
5286 | struct io_kiocb *req; | |
5287 | ||
9ba5fac8 | 5288 | req = io_poll_find(ctx, sqe_addr, poll_only); |
b2cb805f JA |
5289 | if (!req) |
5290 | return -ENOENT; | |
5291 | if (io_poll_remove_one(req)) | |
5292 | return 0; | |
5293 | ||
5294 | return -EALREADY; | |
47f46768 JA |
5295 | } |
5296 | ||
9096af3e PB |
5297 | static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe, |
5298 | unsigned int flags) | |
5299 | { | |
5300 | u32 events; | |
5301 | ||
5302 | events = READ_ONCE(sqe->poll32_events); | |
5303 | #ifdef __BIG_ENDIAN | |
5304 | events = swahw32(events); | |
5305 | #endif | |
5306 | if (!(flags & IORING_POLL_ADD_MULTI)) | |
5307 | events |= EPOLLONESHOT; | |
5308 | return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
5309 | } | |
5310 | ||
c5de0036 | 5311 | static int io_poll_update_prep(struct io_kiocb *req, |
3529d8c2 | 5312 | const struct io_uring_sqe *sqe) |
0969e783 | 5313 | { |
c5de0036 PB |
5314 | struct io_poll_update *upd = &req->poll_update; |
5315 | u32 flags; | |
5316 | ||
0969e783 JA |
5317 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5318 | return -EINVAL; | |
c5de0036 PB |
5319 | if (sqe->ioprio || sqe->buf_index) |
5320 | return -EINVAL; | |
5321 | flags = READ_ONCE(sqe->len); | |
5322 | if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA | | |
5323 | IORING_POLL_ADD_MULTI)) | |
5324 | return -EINVAL; | |
5325 | /* meaningless without update */ | |
5326 | if (flags == IORING_POLL_ADD_MULTI) | |
0969e783 JA |
5327 | return -EINVAL; |
5328 | ||
c5de0036 PB |
5329 | upd->old_user_data = READ_ONCE(sqe->addr); |
5330 | upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5331 | upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
221c5eb2 | 5332 | |
c5de0036 PB |
5333 | upd->new_user_data = READ_ONCE(sqe->off); |
5334 | if (!upd->update_user_data && upd->new_user_data) | |
5335 | return -EINVAL; | |
5336 | if (upd->update_events) | |
5337 | upd->events = io_poll_parse_events(sqe, flags); | |
5338 | else if (sqe->poll32_events) | |
5339 | return -EINVAL; | |
221c5eb2 | 5340 | |
221c5eb2 JA |
5341 | return 0; |
5342 | } | |
5343 | ||
221c5eb2 JA |
5344 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5345 | void *key) | |
5346 | { | |
c2f2eb7d JA |
5347 | struct io_kiocb *req = wait->private; |
5348 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5349 | |
d7718a9d | 5350 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5351 | } |
5352 | ||
221c5eb2 JA |
5353 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5354 | struct poll_table_struct *p) | |
5355 | { | |
5356 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5357 | ||
e8c2bc1f | 5358 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5359 | } |
5360 | ||
3529d8c2 | 5361 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 | 5362 | { |
c5de0036 PB |
5363 | struct io_poll_iocb *poll = &req->poll; |
5364 | u32 flags; | |
221c5eb2 JA |
5365 | |
5366 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5367 | return -EINVAL; | |
c5de0036 | 5368 | if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) |
88e41cf9 JA |
5369 | return -EINVAL; |
5370 | flags = READ_ONCE(sqe->len); | |
c5de0036 | 5371 | if (flags & ~IORING_POLL_ADD_MULTI) |
221c5eb2 | 5372 | return -EINVAL; |
9096af3e | 5373 | |
c5de0036 | 5374 | poll->events = io_poll_parse_events(sqe, flags); |
0969e783 JA |
5375 | return 0; |
5376 | } | |
5377 | ||
c5de0036 | 5378 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) |
0969e783 JA |
5379 | { |
5380 | struct io_poll_iocb *poll = &req->poll; | |
5381 | struct io_ring_ctx *ctx = req->ctx; | |
5382 | struct io_poll_table ipt; | |
0969e783 | 5383 | __poll_t mask; |
0969e783 | 5384 | |
d7718a9d | 5385 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5386 | |
d7718a9d JA |
5387 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5388 | io_poll_wake); | |
221c5eb2 | 5389 | |
8c838788 | 5390 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5391 | ipt.error = 0; |
e27414be | 5392 | io_poll_complete(req, mask); |
221c5eb2 | 5393 | } |
221c5eb2 JA |
5394 | spin_unlock_irq(&ctx->completion_lock); |
5395 | ||
8c838788 JA |
5396 | if (mask) { |
5397 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5398 | if (poll->events & EPOLLONESHOT) |
5399 | io_put_req(req); | |
221c5eb2 | 5400 | } |
8c838788 | 5401 | return ipt.error; |
221c5eb2 JA |
5402 | } |
5403 | ||
c5de0036 | 5404 | static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags) |
b69de288 JA |
5405 | { |
5406 | struct io_ring_ctx *ctx = req->ctx; | |
5407 | struct io_kiocb *preq; | |
cb3b200e | 5408 | bool completing; |
b69de288 JA |
5409 | int ret; |
5410 | ||
5411 | spin_lock_irq(&ctx->completion_lock); | |
9ba5fac8 | 5412 | preq = io_poll_find(ctx, req->poll_update.old_user_data, true); |
b69de288 JA |
5413 | if (!preq) { |
5414 | ret = -ENOENT; | |
5415 | goto err; | |
b69de288 | 5416 | } |
cb3b200e | 5417 | |
c5de0036 PB |
5418 | if (!req->poll_update.update_events && !req->poll_update.update_user_data) { |
5419 | completing = true; | |
5420 | ret = io_poll_remove_one(preq) ? 0 : -EALREADY; | |
5421 | goto err; | |
5422 | } | |
5423 | ||
cb3b200e JA |
5424 | /* |
5425 | * Don't allow racy completion with singleshot, as we cannot safely | |
5426 | * update those. For multishot, if we're racing with completion, just | |
5427 | * let completion re-add it. | |
5428 | */ | |
5429 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5430 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5431 | ret = -EALREADY; | |
5432 | goto err; | |
b69de288 JA |
5433 | } |
5434 | /* we now have a detached poll request. reissue. */ | |
5435 | ret = 0; | |
5436 | err: | |
b69de288 | 5437 | if (ret < 0) { |
cb3b200e | 5438 | spin_unlock_irq(&ctx->completion_lock); |
b69de288 JA |
5439 | req_set_fail_links(req); |
5440 | io_req_complete(req, ret); | |
5441 | return 0; | |
5442 | } | |
5443 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5444 | if (req->poll_update.update_events) { |
b69de288 | 5445 | preq->poll.events &= ~0xffff; |
9d805892 | 5446 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5447 | preq->poll.events |= IO_POLL_UNMASK; |
5448 | } | |
9d805892 PB |
5449 | if (req->poll_update.update_user_data) |
5450 | preq->user_data = req->poll_update.new_user_data; | |
cb3b200e JA |
5451 | spin_unlock_irq(&ctx->completion_lock); |
5452 | ||
b69de288 JA |
5453 | /* complete update request, we're done with it */ |
5454 | io_req_complete(req, ret); | |
5455 | ||
cb3b200e | 5456 | if (!completing) { |
c5de0036 | 5457 | ret = io_poll_add(preq, issue_flags); |
cb3b200e JA |
5458 | if (ret < 0) { |
5459 | req_set_fail_links(preq); | |
5460 | io_req_complete(preq, ret); | |
5461 | } | |
b69de288 JA |
5462 | } |
5463 | return 0; | |
5464 | } | |
5465 | ||
5262f567 JA |
5466 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5467 | { | |
ad8a48ac JA |
5468 | struct io_timeout_data *data = container_of(timer, |
5469 | struct io_timeout_data, timer); | |
5470 | struct io_kiocb *req = data->req; | |
5471 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5472 | unsigned long flags; |
5473 | ||
5262f567 | 5474 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5475 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5476 | atomic_set(&req->ctx->cq_timeouts, |
5477 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5478 | ||
d4d19c19 | 5479 | io_cqring_fill_event(ctx, req->user_data, -ETIME, 0); |
5262f567 JA |
5480 | io_commit_cqring(ctx); |
5481 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5482 | ||
5483 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5484 | req_set_fail_links(req); |
5262f567 JA |
5485 | io_put_req(req); |
5486 | return HRTIMER_NORESTART; | |
5487 | } | |
5488 | ||
fbd15848 PB |
5489 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5490 | __u64 user_data) | |
e07785b0 | 5491 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5492 | { |
fbd15848 | 5493 | struct io_timeout_data *io; |
47f46768 | 5494 | struct io_kiocb *req; |
fd9c7bc5 | 5495 | bool found = false; |
f254ac04 | 5496 | |
135fcde8 | 5497 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5498 | found = user_data == req->user_data; |
5499 | if (found) | |
47f46768 | 5500 | break; |
47f46768 | 5501 | } |
fd9c7bc5 PB |
5502 | if (!found) |
5503 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5504 | |
5505 | io = req->async_data; | |
fd9c7bc5 | 5506 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5507 | return ERR_PTR(-EALREADY); |
a71976f3 | 5508 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5509 | return req; |
5510 | } | |
47f46768 | 5511 | |
fbd15848 | 5512 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5513 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5514 | { |
5515 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5516 | ||
5517 | if (IS_ERR(req)) | |
5518 | return PTR_ERR(req); | |
f254ac04 JA |
5519 | |
5520 | req_set_fail_links(req); | |
d4d19c19 | 5521 | io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0); |
216578e5 | 5522 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5523 | return 0; |
5524 | } | |
5525 | ||
9c8e11b3 PB |
5526 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5527 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5528 | __must_hold(&ctx->completion_lock) |
47f46768 | 5529 | { |
9c8e11b3 PB |
5530 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5531 | struct io_timeout_data *data; | |
47f46768 | 5532 | |
9c8e11b3 PB |
5533 | if (IS_ERR(req)) |
5534 | return PTR_ERR(req); | |
47f46768 | 5535 | |
9c8e11b3 PB |
5536 | req->timeout.off = 0; /* noseq */ |
5537 | data = req->async_data; | |
5538 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5539 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5540 | data->timer.function = io_timeout_fn; | |
5541 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5542 | return 0; | |
47f46768 JA |
5543 | } |
5544 | ||
3529d8c2 JA |
5545 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5546 | const struct io_uring_sqe *sqe) | |
b29472ee | 5547 | { |
9c8e11b3 PB |
5548 | struct io_timeout_rem *tr = &req->timeout_rem; |
5549 | ||
b29472ee JA |
5550 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5551 | return -EINVAL; | |
61710e43 DA |
5552 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5553 | return -EINVAL; | |
9c8e11b3 | 5554 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5555 | return -EINVAL; |
5556 | ||
9c8e11b3 PB |
5557 | tr->addr = READ_ONCE(sqe->addr); |
5558 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5559 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5560 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5561 | return -EINVAL; | |
5562 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5563 | return -EFAULT; | |
5564 | } else if (tr->flags) { | |
5565 | /* timeout removal doesn't support flags */ | |
b29472ee | 5566 | return -EINVAL; |
9c8e11b3 | 5567 | } |
b29472ee | 5568 | |
b29472ee JA |
5569 | return 0; |
5570 | } | |
5571 | ||
8662daec PB |
5572 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5573 | { | |
5574 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5575 | : HRTIMER_MODE_REL; | |
5576 | } | |
5577 | ||
11365043 JA |
5578 | /* |
5579 | * Remove or update an existing timeout command | |
5580 | */ | |
61e98203 | 5581 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5582 | { |
9c8e11b3 | 5583 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5584 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5585 | int ret; |
11365043 | 5586 | |
11365043 | 5587 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5588 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5589 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5590 | else |
5591 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5592 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5593 | |
d4d19c19 | 5594 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
11365043 JA |
5595 | io_commit_cqring(ctx); |
5596 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5597 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5598 | if (ret < 0) |
5599 | req_set_fail_links(req); | |
ec9c02ad | 5600 | io_put_req(req); |
11365043 | 5601 | return 0; |
5262f567 JA |
5602 | } |
5603 | ||
3529d8c2 | 5604 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5605 | bool is_timeout_link) |
5262f567 | 5606 | { |
ad8a48ac | 5607 | struct io_timeout_data *data; |
a41525ab | 5608 | unsigned flags; |
56080b02 | 5609 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5610 | |
ad8a48ac | 5611 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5612 | return -EINVAL; |
ad8a48ac | 5613 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5614 | return -EINVAL; |
56080b02 | 5615 | if (off && is_timeout_link) |
2d28390a | 5616 | return -EINVAL; |
a41525ab JA |
5617 | flags = READ_ONCE(sqe->timeout_flags); |
5618 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5619 | return -EINVAL; |
bdf20073 | 5620 | |
bfe68a22 | 5621 | req->timeout.off = off; |
26a61679 | 5622 | |
e8c2bc1f | 5623 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5624 | return -ENOMEM; |
5625 | ||
e8c2bc1f | 5626 | data = req->async_data; |
ad8a48ac | 5627 | data->req = req; |
ad8a48ac JA |
5628 | |
5629 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5630 | return -EFAULT; |
5631 | ||
8662daec | 5632 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5633 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5634 | if (is_timeout_link) |
5635 | io_req_track_inflight(req); | |
ad8a48ac JA |
5636 | return 0; |
5637 | } | |
5638 | ||
61e98203 | 5639 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5640 | { |
ad8a48ac | 5641 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5642 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5643 | struct list_head *entry; |
bfe68a22 | 5644 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5645 | |
733f5c95 | 5646 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5647 | |
5262f567 JA |
5648 | /* |
5649 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5650 | * timeout event to be satisfied. If it isn't set, then this is |
5651 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5652 | */ |
8eb7e2d0 | 5653 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5654 | entry = ctx->timeout_list.prev; |
5655 | goto add; | |
5656 | } | |
5262f567 | 5657 | |
bfe68a22 PB |
5658 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5659 | req->timeout.target_seq = tail + off; | |
5262f567 | 5660 | |
f010505b MDG |
5661 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5662 | * This is safe because ->completion_lock is held, and submissions | |
5663 | * and completions are never mixed in the same ->completion_lock section. | |
5664 | */ | |
5665 | ctx->cq_last_tm_flush = tail; | |
5666 | ||
5262f567 JA |
5667 | /* |
5668 | * Insertion sort, ensuring the first entry in the list is always | |
5669 | * the one we need first. | |
5670 | */ | |
5262f567 | 5671 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5672 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5673 | timeout.list); | |
5262f567 | 5674 | |
8eb7e2d0 | 5675 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5676 | continue; |
bfe68a22 PB |
5677 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5678 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5679 | break; |
5680 | } | |
93bd25bb | 5681 | add: |
135fcde8 | 5682 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5683 | data->timer.function = io_timeout_fn; |
5684 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5685 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5686 | return 0; |
5687 | } | |
5262f567 | 5688 | |
f458dd84 PB |
5689 | struct io_cancel_data { |
5690 | struct io_ring_ctx *ctx; | |
5691 | u64 user_data; | |
5692 | }; | |
5693 | ||
62755e35 JA |
5694 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5695 | { | |
5696 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5697 | struct io_cancel_data *cd = data; |
62755e35 | 5698 | |
f458dd84 | 5699 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5700 | } |
5701 | ||
f458dd84 PB |
5702 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5703 | struct io_ring_ctx *ctx) | |
62755e35 | 5704 | { |
f458dd84 | 5705 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5706 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5707 | int ret = 0; |
5708 | ||
f458dd84 | 5709 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5710 | return -ENOENT; |
5711 | ||
f458dd84 | 5712 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5713 | switch (cancel_ret) { |
5714 | case IO_WQ_CANCEL_OK: | |
5715 | ret = 0; | |
5716 | break; | |
5717 | case IO_WQ_CANCEL_RUNNING: | |
5718 | ret = -EALREADY; | |
5719 | break; | |
5720 | case IO_WQ_CANCEL_NOTFOUND: | |
5721 | ret = -ENOENT; | |
5722 | break; | |
5723 | } | |
5724 | ||
e977d6d3 JA |
5725 | return ret; |
5726 | } | |
5727 | ||
47f46768 JA |
5728 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5729 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5730 | int success_ret) |
47f46768 JA |
5731 | { |
5732 | unsigned long flags; | |
5733 | int ret; | |
5734 | ||
f458dd84 | 5735 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5736 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5737 | if (ret != -ENOENT) |
5738 | goto done; | |
47f46768 JA |
5739 | ret = io_timeout_cancel(ctx, sqe_addr); |
5740 | if (ret != -ENOENT) | |
5741 | goto done; | |
9ba5fac8 | 5742 | ret = io_poll_cancel(ctx, sqe_addr, false); |
47f46768 | 5743 | done: |
b0dd8a41 JA |
5744 | if (!ret) |
5745 | ret = success_ret; | |
d4d19c19 | 5746 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
47f46768 JA |
5747 | io_commit_cqring(ctx); |
5748 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5749 | io_cqring_ev_posted(ctx); | |
5750 | ||
4e88d6e7 JA |
5751 | if (ret < 0) |
5752 | req_set_fail_links(req); | |
47f46768 JA |
5753 | } |
5754 | ||
3529d8c2 JA |
5755 | static int io_async_cancel_prep(struct io_kiocb *req, |
5756 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5757 | { |
fbf23849 | 5758 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5759 | return -EINVAL; |
61710e43 DA |
5760 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5761 | return -EINVAL; | |
5762 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5763 | return -EINVAL; |
5764 | ||
fbf23849 JA |
5765 | req->cancel.addr = READ_ONCE(sqe->addr); |
5766 | return 0; | |
5767 | } | |
5768 | ||
61e98203 | 5769 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5770 | { |
5771 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5772 | u64 sqe_addr = req->cancel.addr; |
5773 | struct io_tctx_node *node; | |
5774 | int ret; | |
5775 | ||
5776 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5777 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5778 | spin_lock_irq(&ctx->completion_lock); | |
5779 | if (ret != -ENOENT) | |
5780 | goto done; | |
5781 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5782 | if (ret != -ENOENT) | |
5783 | goto done; | |
9ba5fac8 | 5784 | ret = io_poll_cancel(ctx, sqe_addr, false); |
58f99373 PB |
5785 | if (ret != -ENOENT) |
5786 | goto done; | |
5787 | spin_unlock_irq(&ctx->completion_lock); | |
5788 | ||
5789 | /* slow path, try all io-wq's */ | |
5790 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5791 | ret = -ENOENT; | |
5792 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5793 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5794 | |
58f99373 PB |
5795 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5796 | if (ret != -ENOENT) | |
5797 | break; | |
5798 | } | |
5799 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5800 | ||
5801 | spin_lock_irq(&ctx->completion_lock); | |
5802 | done: | |
d4d19c19 | 5803 | io_cqring_fill_event(ctx, req->user_data, ret, 0); |
58f99373 PB |
5804 | io_commit_cqring(ctx); |
5805 | spin_unlock_irq(&ctx->completion_lock); | |
5806 | io_cqring_ev_posted(ctx); | |
5807 | ||
5808 | if (ret < 0) | |
5809 | req_set_fail_links(req); | |
5810 | io_put_req(req); | |
5262f567 JA |
5811 | return 0; |
5812 | } | |
5813 | ||
269bbe5f | 5814 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5815 | const struct io_uring_sqe *sqe) |
5816 | { | |
6ca56f84 JA |
5817 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5818 | return -EINVAL; | |
61710e43 DA |
5819 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5820 | return -EINVAL; | |
5821 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5822 | return -EINVAL; |
5823 | ||
269bbe5f BM |
5824 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5825 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5826 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5827 | return -EINVAL; |
269bbe5f | 5828 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5829 | return 0; |
5830 | } | |
5831 | ||
889fca73 | 5832 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5833 | { |
5834 | struct io_ring_ctx *ctx = req->ctx; | |
c3bdad02 | 5835 | struct io_uring_rsrc_update2 up; |
05f3fb3c | 5836 | int ret; |
fbf23849 | 5837 | |
45d189c6 | 5838 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5839 | return -EAGAIN; |
05f3fb3c | 5840 | |
269bbe5f BM |
5841 | up.offset = req->rsrc_update.offset; |
5842 | up.data = req->rsrc_update.arg; | |
c3bdad02 PB |
5843 | up.nr = 0; |
5844 | up.tags = 0; | |
05f3fb3c JA |
5845 | |
5846 | mutex_lock(&ctx->uring_lock); | |
fdecb662 | 5847 | ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, |
98f0b3b4 | 5848 | &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5849 | mutex_unlock(&ctx->uring_lock); |
5850 | ||
5851 | if (ret < 0) | |
5852 | req_set_fail_links(req); | |
889fca73 | 5853 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5854 | return 0; |
5855 | } | |
5856 | ||
bfe76559 | 5857 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5858 | { |
d625c6ee | 5859 | switch (req->opcode) { |
e781573e | 5860 | case IORING_OP_NOP: |
bfe76559 | 5861 | return 0; |
f67676d1 JA |
5862 | case IORING_OP_READV: |
5863 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5864 | case IORING_OP_READ: |
bfe76559 | 5865 | return io_read_prep(req, sqe); |
f67676d1 JA |
5866 | case IORING_OP_WRITEV: |
5867 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5868 | case IORING_OP_WRITE: |
bfe76559 | 5869 | return io_write_prep(req, sqe); |
0969e783 | 5870 | case IORING_OP_POLL_ADD: |
bfe76559 | 5871 | return io_poll_add_prep(req, sqe); |
0969e783 | 5872 | case IORING_OP_POLL_REMOVE: |
c5de0036 | 5873 | return io_poll_update_prep(req, sqe); |
8ed8d3c3 | 5874 | case IORING_OP_FSYNC: |
1155c76a | 5875 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5876 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5877 | return io_sfr_prep(req, sqe); |
03b1230c | 5878 | case IORING_OP_SENDMSG: |
fddaface | 5879 | case IORING_OP_SEND: |
bfe76559 | 5880 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5881 | case IORING_OP_RECVMSG: |
fddaface | 5882 | case IORING_OP_RECV: |
bfe76559 | 5883 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5884 | case IORING_OP_CONNECT: |
bfe76559 | 5885 | return io_connect_prep(req, sqe); |
2d28390a | 5886 | case IORING_OP_TIMEOUT: |
bfe76559 | 5887 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5888 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5889 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5890 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5891 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5892 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5893 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5894 | case IORING_OP_ACCEPT: |
bfe76559 | 5895 | return io_accept_prep(req, sqe); |
d63d1b5e | 5896 | case IORING_OP_FALLOCATE: |
bfe76559 | 5897 | return io_fallocate_prep(req, sqe); |
15b71abe | 5898 | case IORING_OP_OPENAT: |
bfe76559 | 5899 | return io_openat_prep(req, sqe); |
b5dba59e | 5900 | case IORING_OP_CLOSE: |
bfe76559 | 5901 | return io_close_prep(req, sqe); |
05f3fb3c | 5902 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5903 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5904 | case IORING_OP_STATX: |
bfe76559 | 5905 | return io_statx_prep(req, sqe); |
4840e418 | 5906 | case IORING_OP_FADVISE: |
bfe76559 | 5907 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5908 | case IORING_OP_MADVISE: |
bfe76559 | 5909 | return io_madvise_prep(req, sqe); |
cebdb986 | 5910 | case IORING_OP_OPENAT2: |
bfe76559 | 5911 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5912 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5913 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5914 | case IORING_OP_SPLICE: |
bfe76559 | 5915 | return io_splice_prep(req, sqe); |
ddf0322d | 5916 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5917 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5918 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5919 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5920 | case IORING_OP_TEE: |
bfe76559 | 5921 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5922 | case IORING_OP_SHUTDOWN: |
5923 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5924 | case IORING_OP_RENAMEAT: |
5925 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5926 | case IORING_OP_UNLINKAT: |
5927 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5928 | } |
5929 | ||
bfe76559 PB |
5930 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5931 | req->opcode); | |
bd54b6fe | 5932 | return -EINVAL; |
bfe76559 PB |
5933 | } |
5934 | ||
93642ef8 | 5935 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5936 | { |
b7e298d2 PB |
5937 | if (!io_op_defs[req->opcode].needs_async_setup) |
5938 | return 0; | |
5939 | if (WARN_ON_ONCE(req->async_data)) | |
5940 | return -EFAULT; | |
5941 | if (io_alloc_async_data(req)) | |
5942 | return -EAGAIN; | |
5943 | ||
93642ef8 PB |
5944 | switch (req->opcode) { |
5945 | case IORING_OP_READV: | |
93642ef8 PB |
5946 | return io_rw_prep_async(req, READ); |
5947 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5948 | return io_rw_prep_async(req, WRITE); |
5949 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5950 | return io_sendmsg_prep_async(req); |
5951 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5952 | return io_recvmsg_prep_async(req); |
5953 | case IORING_OP_CONNECT: | |
5954 | return io_connect_prep_async(req); | |
5955 | } | |
b7e298d2 PB |
5956 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5957 | req->opcode); | |
5958 | return -EFAULT; | |
f67676d1 JA |
5959 | } |
5960 | ||
9cf7c104 PB |
5961 | static u32 io_get_sequence(struct io_kiocb *req) |
5962 | { | |
5963 | struct io_kiocb *pos; | |
5964 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5965 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5966 | |
f2f87370 PB |
5967 | io_for_each_link(pos, req) |
5968 | nr_reqs++; | |
9cf7c104 PB |
5969 | |
5970 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5971 | return total_submitted - nr_reqs; | |
5972 | } | |
5973 | ||
be7053b7 | 5974 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5975 | { |
a197f664 | 5976 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 5977 | struct io_defer_entry *de; |
f67676d1 | 5978 | int ret; |
9cf7c104 | 5979 | u32 seq; |
de0617e4 | 5980 | |
9d858b21 | 5981 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
5982 | if (likely(list_empty_careful(&ctx->defer_list) && |
5983 | !(req->flags & REQ_F_IO_DRAIN))) | |
5984 | return 0; | |
5985 | ||
5986 | seq = io_get_sequence(req); | |
5987 | /* Still a chance to pass the sequence check */ | |
5988 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
5989 | return 0; |
5990 | ||
b7e298d2 | 5991 | ret = io_req_prep_async(req); |
be7053b7 PB |
5992 | if (ret) |
5993 | return ret; | |
cbdcb435 | 5994 | io_prep_async_link(req); |
27dc8338 PB |
5995 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
5996 | if (!de) | |
5997 | return -ENOMEM; | |
2d28390a | 5998 | |
de0617e4 | 5999 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6000 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6001 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6002 | kfree(de); |
ae34817b PB |
6003 | io_queue_async_work(req); |
6004 | return -EIOCBQUEUED; | |
de0617e4 JA |
6005 | } |
6006 | ||
915967f6 | 6007 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6008 | de->req = req; |
9cf7c104 | 6009 | de->seq = seq; |
27dc8338 | 6010 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6011 | spin_unlock_irq(&ctx->completion_lock); |
6012 | return -EIOCBQUEUED; | |
6013 | } | |
6014 | ||
68fb8979 | 6015 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6016 | { |
0e1b6fe3 PB |
6017 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6018 | switch (req->opcode) { | |
6019 | case IORING_OP_READV: | |
6020 | case IORING_OP_READ_FIXED: | |
6021 | case IORING_OP_READ: | |
bcda7baa | 6022 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6023 | break; |
6024 | case IORING_OP_RECVMSG: | |
6025 | case IORING_OP_RECV: | |
bcda7baa | 6026 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6027 | break; |
6028 | } | |
6029 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6030 | } |
6031 | ||
0e1b6fe3 PB |
6032 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6033 | switch (req->opcode) { | |
6034 | case IORING_OP_READV: | |
6035 | case IORING_OP_READ_FIXED: | |
6036 | case IORING_OP_READ: | |
6037 | case IORING_OP_WRITEV: | |
6038 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6039 | case IORING_OP_WRITE: { |
6040 | struct io_async_rw *io = req->async_data; | |
6041 | if (io->free_iovec) | |
6042 | kfree(io->free_iovec); | |
0e1b6fe3 | 6043 | break; |
e8c2bc1f | 6044 | } |
0e1b6fe3 | 6045 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6046 | case IORING_OP_SENDMSG: { |
6047 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6048 | |
6049 | kfree(io->free_iov); | |
0e1b6fe3 | 6050 | break; |
e8c2bc1f | 6051 | } |
0e1b6fe3 PB |
6052 | case IORING_OP_SPLICE: |
6053 | case IORING_OP_TEE: | |
e1d767f0 PB |
6054 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6055 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6056 | break; |
f3cd4850 JA |
6057 | case IORING_OP_OPENAT: |
6058 | case IORING_OP_OPENAT2: | |
6059 | if (req->open.filename) | |
6060 | putname(req->open.filename); | |
6061 | break; | |
80a261fd JA |
6062 | case IORING_OP_RENAMEAT: |
6063 | putname(req->rename.oldpath); | |
6064 | putname(req->rename.newpath); | |
6065 | break; | |
14a1143b JA |
6066 | case IORING_OP_UNLINKAT: |
6067 | putname(req->unlink.filename); | |
6068 | break; | |
0e1b6fe3 PB |
6069 | } |
6070 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6071 | } |
75652a30 JA |
6072 | if ((req->flags & REQ_F_POLLED) && req->apoll) { |
6073 | kfree(req->apoll->double_poll); | |
6074 | kfree(req->apoll); | |
6075 | req->apoll = NULL; | |
6076 | } | |
3a0a6902 PB |
6077 | if (req->flags & REQ_F_INFLIGHT) { |
6078 | struct io_uring_task *tctx = req->task->io_uring; | |
6079 | ||
6080 | atomic_dec(&tctx->inflight_tracked); | |
6081 | req->flags &= ~REQ_F_INFLIGHT; | |
6082 | } | |
99bc4c38 PB |
6083 | } |
6084 | ||
889fca73 | 6085 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6086 | { |
a197f664 | 6087 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6088 | const struct cred *creds = NULL; |
d625c6ee | 6089 | int ret; |
2b188cc1 | 6090 | |
003e8dcc JA |
6091 | if (req->work.creds && req->work.creds != current_cred()) |
6092 | creds = override_creds(req->work.creds); | |
5730b27e | 6093 | |
d625c6ee | 6094 | switch (req->opcode) { |
2b188cc1 | 6095 | case IORING_OP_NOP: |
889fca73 | 6096 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6097 | break; |
6098 | case IORING_OP_READV: | |
edafccee | 6099 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6100 | case IORING_OP_READ: |
889fca73 | 6101 | ret = io_read(req, issue_flags); |
edafccee | 6102 | break; |
3529d8c2 | 6103 | case IORING_OP_WRITEV: |
edafccee | 6104 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6105 | case IORING_OP_WRITE: |
889fca73 | 6106 | ret = io_write(req, issue_flags); |
2b188cc1 | 6107 | break; |
c992fe29 | 6108 | case IORING_OP_FSYNC: |
45d189c6 | 6109 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6110 | break; |
221c5eb2 | 6111 | case IORING_OP_POLL_ADD: |
61e98203 | 6112 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6113 | break; |
6114 | case IORING_OP_POLL_REMOVE: | |
c5de0036 | 6115 | ret = io_poll_update(req, issue_flags); |
221c5eb2 | 6116 | break; |
5d17b4a4 | 6117 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6118 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6119 | break; |
0fa03c62 | 6120 | case IORING_OP_SENDMSG: |
889fca73 | 6121 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6122 | break; |
fddaface | 6123 | case IORING_OP_SEND: |
889fca73 | 6124 | ret = io_send(req, issue_flags); |
0fa03c62 | 6125 | break; |
aa1fa28f | 6126 | case IORING_OP_RECVMSG: |
889fca73 | 6127 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6128 | break; |
fddaface | 6129 | case IORING_OP_RECV: |
889fca73 | 6130 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6131 | break; |
5262f567 | 6132 | case IORING_OP_TIMEOUT: |
61e98203 | 6133 | ret = io_timeout(req, issue_flags); |
5262f567 | 6134 | break; |
11365043 | 6135 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6136 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6137 | break; |
17f2fe35 | 6138 | case IORING_OP_ACCEPT: |
889fca73 | 6139 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6140 | break; |
f8e85cf2 | 6141 | case IORING_OP_CONNECT: |
889fca73 | 6142 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6143 | break; |
62755e35 | 6144 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6145 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6146 | break; |
d63d1b5e | 6147 | case IORING_OP_FALLOCATE: |
45d189c6 | 6148 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6149 | break; |
15b71abe | 6150 | case IORING_OP_OPENAT: |
45d189c6 | 6151 | ret = io_openat(req, issue_flags); |
15b71abe | 6152 | break; |
b5dba59e | 6153 | case IORING_OP_CLOSE: |
889fca73 | 6154 | ret = io_close(req, issue_flags); |
b5dba59e | 6155 | break; |
05f3fb3c | 6156 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6157 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6158 | break; |
eddc7ef5 | 6159 | case IORING_OP_STATX: |
45d189c6 | 6160 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6161 | break; |
4840e418 | 6162 | case IORING_OP_FADVISE: |
45d189c6 | 6163 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6164 | break; |
c1ca757b | 6165 | case IORING_OP_MADVISE: |
45d189c6 | 6166 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6167 | break; |
cebdb986 | 6168 | case IORING_OP_OPENAT2: |
45d189c6 | 6169 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6170 | break; |
3e4827b0 | 6171 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6172 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6173 | break; |
7d67af2c | 6174 | case IORING_OP_SPLICE: |
45d189c6 | 6175 | ret = io_splice(req, issue_flags); |
7d67af2c | 6176 | break; |
ddf0322d | 6177 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6178 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6179 | break; |
067524e9 | 6180 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6181 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6182 | break; |
f2a8d5c7 | 6183 | case IORING_OP_TEE: |
45d189c6 | 6184 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6185 | break; |
36f4fa68 | 6186 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6187 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6188 | break; |
80a261fd | 6189 | case IORING_OP_RENAMEAT: |
45d189c6 | 6190 | ret = io_renameat(req, issue_flags); |
80a261fd | 6191 | break; |
14a1143b | 6192 | case IORING_OP_UNLINKAT: |
45d189c6 | 6193 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6194 | break; |
2b188cc1 JA |
6195 | default: |
6196 | ret = -EINVAL; | |
6197 | break; | |
6198 | } | |
6199 | ||
5730b27e JA |
6200 | if (creds) |
6201 | revert_creds(creds); | |
6202 | ||
def596e9 JA |
6203 | if (ret) |
6204 | return ret; | |
6205 | ||
b532576e JA |
6206 | /* If the op doesn't have a file, we're not polling for it */ |
6207 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6208 | const bool in_async = io_wq_current_is_worker(); |
6209 | ||
11ba820b JA |
6210 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6211 | if (in_async) | |
6212 | mutex_lock(&ctx->uring_lock); | |
6213 | ||
2e9dbe90 | 6214 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6215 | |
6216 | if (in_async) | |
6217 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6218 | } |
6219 | ||
6220 | return 0; | |
2b188cc1 JA |
6221 | } |
6222 | ||
5280f7e5 | 6223 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6224 | { |
6225 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6226 | struct io_kiocb *timeout; |
561fb04a | 6227 | int ret = 0; |
2b188cc1 | 6228 | |
6df1db6b PB |
6229 | timeout = io_prep_linked_timeout(req); |
6230 | if (timeout) | |
6231 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6232 | |
4014d943 | 6233 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6234 | ret = -ECANCELED; |
31b51510 | 6235 | |
561fb04a | 6236 | if (!ret) { |
561fb04a | 6237 | do { |
889fca73 | 6238 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6239 | /* |
6240 | * We can get EAGAIN for polled IO even though we're | |
6241 | * forcing a sync submission from here, since we can't | |
6242 | * wait for request slots on the block side. | |
6243 | */ | |
6244 | if (ret != -EAGAIN) | |
6245 | break; | |
6246 | cond_resched(); | |
6247 | } while (1); | |
6248 | } | |
31b51510 | 6249 | |
a3df7698 | 6250 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6251 | if (ret) { |
a3df7698 | 6252 | /* io-wq is going to take one down */ |
de9b4cca | 6253 | req_ref_get(req); |
a3df7698 | 6254 | io_req_task_queue_fail(req, ret); |
edafccee | 6255 | } |
2b188cc1 JA |
6256 | } |
6257 | ||
7b29f92d JA |
6258 | #define FFS_ASYNC_READ 0x1UL |
6259 | #define FFS_ASYNC_WRITE 0x2UL | |
6260 | #ifdef CONFIG_64BIT | |
6261 | #define FFS_ISREG 0x4UL | |
6262 | #else | |
6263 | #define FFS_ISREG 0x0UL | |
6264 | #endif | |
6265 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6266 | ||
aeca241b | 6267 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6268 | unsigned i) |
65e19f54 | 6269 | { |
aeca241b | 6270 | struct io_fixed_file *table_l2; |
65e19f54 | 6271 | |
aeca241b PB |
6272 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6273 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6274 | } |
6275 | ||
6276 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, | |
6277 | int index) | |
6278 | { | |
aeca241b | 6279 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
7b29f92d | 6280 | |
a04b0ac0 | 6281 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6282 | } |
6283 | ||
a04b0ac0 | 6284 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6285 | { |
6286 | unsigned long file_ptr = (unsigned long) file; | |
6287 | ||
6288 | if (__io_file_supports_async(file, READ)) | |
6289 | file_ptr |= FFS_ASYNC_READ; | |
6290 | if (__io_file_supports_async(file, WRITE)) | |
6291 | file_ptr |= FFS_ASYNC_WRITE; | |
6292 | if (S_ISREG(file_inode(file)->i_mode)) | |
6293 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6294 | file_slot->file_ptr = file_ptr; |
9a321c98 PB |
6295 | } |
6296 | ||
8371adf5 PB |
6297 | static struct file *io_file_get(struct io_submit_state *state, |
6298 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6299 | { |
a197f664 | 6300 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6301 | struct file *file; |
09bb8394 | 6302 | |
8da11c19 | 6303 | if (fixed) { |
7b29f92d JA |
6304 | unsigned long file_ptr; |
6305 | ||
479f517b | 6306 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6307 | return NULL; |
b7620121 | 6308 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6309 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6310 | file = (struct file *) (file_ptr & FFS_MASK); |
6311 | file_ptr &= ~FFS_MASK; | |
6312 | /* mask in overlapping REQ_F and FFS bits */ | |
6313 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6314 | io_req_set_rsrc_node(req); |
09bb8394 | 6315 | } else { |
c826bd7a | 6316 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6317 | file = __io_file_get(state, fd); |
d44f554e JA |
6318 | |
6319 | /* we don't allow fixed io_uring files */ | |
6320 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6321 | io_req_track_inflight(req); | |
09bb8394 JA |
6322 | } |
6323 | ||
8371adf5 | 6324 | return file; |
09bb8394 JA |
6325 | } |
6326 | ||
2665abfd | 6327 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6328 | { |
ad8a48ac JA |
6329 | struct io_timeout_data *data = container_of(timer, |
6330 | struct io_timeout_data, timer); | |
90cd7e42 | 6331 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6332 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6333 | unsigned long flags; |
2665abfd JA |
6334 | |
6335 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6336 | prev = req->timeout.head; |
6337 | req->timeout.head = NULL; | |
2665abfd JA |
6338 | |
6339 | /* | |
6340 | * We don't expect the list to be empty, that will only happen if we | |
6341 | * race with the completion of the linked work. | |
6342 | */ | |
de9b4cca | 6343 | if (prev && req_ref_inc_not_zero(prev)) |
f2f87370 | 6344 | io_remove_next_linked(prev); |
90cd7e42 PB |
6345 | else |
6346 | prev = NULL; | |
2665abfd JA |
6347 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6348 | ||
6349 | if (prev) { | |
014db007 | 6350 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6351 | io_put_req_deferred(prev, 1); |
47f46768 | 6352 | } else { |
9ae1f8dd | 6353 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6354 | } |
df9727af | 6355 | io_put_req_deferred(req, 1); |
2665abfd JA |
6356 | return HRTIMER_NORESTART; |
6357 | } | |
6358 | ||
de968c18 | 6359 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6360 | { |
de968c18 PB |
6361 | struct io_ring_ctx *ctx = req->ctx; |
6362 | ||
6363 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6364 | /* |
f2f87370 PB |
6365 | * If the back reference is NULL, then our linked request finished |
6366 | * before we got a chance to setup the timer | |
76a46e06 | 6367 | */ |
90cd7e42 | 6368 | if (req->timeout.head) { |
e8c2bc1f | 6369 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6370 | |
ad8a48ac JA |
6371 | data->timer.function = io_link_timeout_fn; |
6372 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6373 | data->mode); | |
2665abfd | 6374 | } |
76a46e06 | 6375 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6376 | /* drop submission reference */ |
76a46e06 JA |
6377 | io_put_req(req); |
6378 | } | |
2665abfd | 6379 | |
ad8a48ac | 6380 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6381 | { |
f2f87370 | 6382 | struct io_kiocb *nxt = req->link; |
2665abfd | 6383 | |
f2f87370 PB |
6384 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6385 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6386 | return NULL; |
2665abfd | 6387 | |
90cd7e42 | 6388 | nxt->timeout.head = req; |
900fad45 | 6389 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6390 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6391 | return nxt; |
2665abfd JA |
6392 | } |
6393 | ||
c5eef2b9 | 6394 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6395 | { |
d3d7298d | 6396 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6397 | int ret; |
2b188cc1 | 6398 | |
c5eef2b9 | 6399 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6400 | |
491381ce JA |
6401 | /* |
6402 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6403 | * doesn't support non-blocking read/write attempts | |
6404 | */ | |
1840038e | 6405 | if (likely(!ret)) { |
0d63c148 | 6406 | /* drop submission reference */ |
e342c807 | 6407 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6408 | struct io_ring_ctx *ctx = req->ctx; |
6409 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6410 | |
6dd0be1e | 6411 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6412 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6413 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6414 | } else { |
d3d7298d | 6415 | io_put_req(req); |
0d63c148 | 6416 | } |
1840038e PB |
6417 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6418 | if (!io_arm_poll_handler(req)) { | |
6419 | /* | |
6420 | * Queued up for async execution, worker will release | |
6421 | * submit reference when the iocb is actually submitted. | |
6422 | */ | |
6423 | io_queue_async_work(req); | |
6424 | } | |
0d63c148 | 6425 | } else { |
f41db273 | 6426 | io_req_complete_failed(req, ret); |
9e645e11 | 6427 | } |
d3d7298d PB |
6428 | if (linked_timeout) |
6429 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6430 | } |
6431 | ||
be7053b7 | 6432 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6433 | { |
6434 | int ret; | |
6435 | ||
be7053b7 | 6436 | ret = io_req_defer(req); |
4fe2c963 JL |
6437 | if (ret) { |
6438 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6439 | fail_req: |
f41db273 | 6440 | io_req_complete_failed(req, ret); |
4fe2c963 | 6441 | } |
2550878f | 6442 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6443 | ret = io_req_prep_async(req); |
be7053b7 PB |
6444 | if (unlikely(ret)) |
6445 | goto fail_req; | |
ce35a47a JA |
6446 | io_queue_async_work(req); |
6447 | } else { | |
c5eef2b9 | 6448 | __io_queue_sqe(req); |
ce35a47a | 6449 | } |
4fe2c963 JL |
6450 | } |
6451 | ||
b16fed66 PB |
6452 | /* |
6453 | * Check SQE restrictions (opcode and flags). | |
6454 | * | |
6455 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6456 | */ | |
6457 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6458 | struct io_kiocb *req, | |
6459 | unsigned int sqe_flags) | |
4fe2c963 | 6460 | { |
b16fed66 PB |
6461 | if (!ctx->restricted) |
6462 | return true; | |
6463 | ||
6464 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6465 | return false; | |
6466 | ||
6467 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6468 | ctx->restrictions.sqe_flags_required) | |
6469 | return false; | |
6470 | ||
6471 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6472 | ctx->restrictions.sqe_flags_required)) | |
6473 | return false; | |
6474 | ||
6475 | return true; | |
4fe2c963 JL |
6476 | } |
6477 | ||
b16fed66 PB |
6478 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6479 | const struct io_uring_sqe *sqe) | |
6480 | { | |
6481 | struct io_submit_state *state; | |
6482 | unsigned int sqe_flags; | |
003e8dcc | 6483 | int personality, ret = 0; |
b16fed66 PB |
6484 | |
6485 | req->opcode = READ_ONCE(sqe->opcode); | |
6486 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6487 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6488 | req->user_data = READ_ONCE(sqe->user_data); | |
6489 | req->async_data = NULL; | |
6490 | req->file = NULL; | |
6491 | req->ctx = ctx; | |
6492 | req->link = NULL; | |
6493 | req->fixed_rsrc_refs = NULL; | |
6494 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6495 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6496 | req->task = current; |
6497 | req->result = 0; | |
93e68e03 | 6498 | req->work.creds = NULL; |
b16fed66 PB |
6499 | |
6500 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6501 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6502 | req->flags = 0; | |
b16fed66 | 6503 | return -EINVAL; |
ebf4a5db | 6504 | } |
b16fed66 PB |
6505 | |
6506 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6507 | return -EINVAL; | |
6508 | ||
b16fed66 PB |
6509 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6510 | return -EACCES; | |
6511 | ||
6512 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6513 | !io_op_defs[req->opcode].buffer_select) | |
6514 | return -EOPNOTSUPP; | |
863e0560 | 6515 | |
003e8dcc JA |
6516 | personality = READ_ONCE(sqe->personality); |
6517 | if (personality) { | |
61cf9370 | 6518 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6519 | if (!req->work.creds) |
6520 | return -EINVAL; | |
6521 | get_cred(req->work.creds); | |
003e8dcc | 6522 | } |
b16fed66 PB |
6523 | state = &ctx->submit_state; |
6524 | ||
6525 | /* | |
6526 | * Plug now if we have more than 1 IO left after this, and the target | |
6527 | * is potentially a read/write to block based storage. | |
6528 | */ | |
6529 | if (!state->plug_started && state->ios_left > 1 && | |
6530 | io_op_defs[req->opcode].plug) { | |
6531 | blk_start_plug(&state->plug); | |
6532 | state->plug_started = true; | |
6533 | } | |
6534 | ||
6535 | if (io_op_defs[req->opcode].needs_file) { | |
6536 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6537 | ||
6538 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6539 | if (unlikely(!req->file)) | |
6540 | ret = -EBADF; | |
6541 | } | |
6542 | ||
6543 | state->ios_left--; | |
6544 | return ret; | |
6545 | } | |
6546 | ||
a6b8cadc | 6547 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6548 | const struct io_uring_sqe *sqe) |
9e645e11 | 6549 | { |
a1ab7b35 | 6550 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6551 | int ret; |
9e645e11 | 6552 | |
a6b8cadc PB |
6553 | ret = io_init_req(ctx, req, sqe); |
6554 | if (unlikely(ret)) { | |
6555 | fail_req: | |
de59bc10 PB |
6556 | if (link->head) { |
6557 | /* fail even hard links since we don't submit */ | |
cf109604 | 6558 | link->head->flags |= REQ_F_FAIL_LINK; |
f41db273 | 6559 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6560 | link->head = NULL; |
6561 | } | |
f41db273 | 6562 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6563 | return ret; |
6564 | } | |
be7053b7 PB |
6565 | ret = io_req_prep(req, sqe); |
6566 | if (unlikely(ret)) | |
6567 | goto fail_req; | |
a6b8cadc | 6568 | |
be7053b7 | 6569 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6570 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6571 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6572 | ||
9e645e11 JA |
6573 | /* |
6574 | * If we already have a head request, queue this one for async | |
6575 | * submittal once the head completes. If we don't have a head but | |
6576 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6577 | * submitted sync once the chain is complete. If none of those | |
6578 | * conditions are true (normal request), then just queue it. | |
6579 | */ | |
863e0560 PB |
6580 | if (link->head) { |
6581 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6582 | |
8cdf2193 PB |
6583 | /* |
6584 | * Taking sequential execution of a link, draining both sides | |
6585 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6586 | * requests in the link. So, it drains the head and the | |
6587 | * next after the link request. The last one is done via | |
6588 | * drain_next flag to persist the effect across calls. | |
6589 | */ | |
ef4ff581 | 6590 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6591 | head->flags |= REQ_F_IO_DRAIN; |
6592 | ctx->drain_next = 1; | |
6593 | } | |
b7e298d2 | 6594 | ret = io_req_prep_async(req); |
cf109604 | 6595 | if (unlikely(ret)) |
a6b8cadc | 6596 | goto fail_req; |
9d76377f | 6597 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6598 | link->last->link = req; |
863e0560 | 6599 | link->last = req; |
32fe525b PB |
6600 | |
6601 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6602 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6603 | io_queue_sqe(head); |
863e0560 | 6604 | link->head = NULL; |
32fe525b | 6605 | } |
9e645e11 | 6606 | } else { |
711be031 PB |
6607 | if (unlikely(ctx->drain_next)) { |
6608 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6609 | ctx->drain_next = 0; |
711be031 | 6610 | } |
ef4ff581 | 6611 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6612 | link->head = req; |
6613 | link->last = req; | |
711be031 | 6614 | } else { |
be7053b7 | 6615 | io_queue_sqe(req); |
711be031 | 6616 | } |
9e645e11 | 6617 | } |
2e6e1fde | 6618 | |
1d4240cc | 6619 | return 0; |
9e645e11 JA |
6620 | } |
6621 | ||
9a56a232 JA |
6622 | /* |
6623 | * Batched submission is done, ensure local IO is flushed out. | |
6624 | */ | |
ba88ff11 PB |
6625 | static void io_submit_state_end(struct io_submit_state *state, |
6626 | struct io_ring_ctx *ctx) | |
9a56a232 | 6627 | { |
a1ab7b35 | 6628 | if (state->link.head) |
de59bc10 | 6629 | io_queue_sqe(state->link.head); |
6dd0be1e | 6630 | if (state->comp.nr) |
ba88ff11 | 6631 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6632 | if (state->plug_started) |
6633 | blk_finish_plug(&state->plug); | |
9f13c35b | 6634 | io_state_file_put(state); |
9a56a232 JA |
6635 | } |
6636 | ||
6637 | /* | |
6638 | * Start submission side cache. | |
6639 | */ | |
6640 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6641 | unsigned int max_ios) |
9a56a232 | 6642 | { |
27926b68 | 6643 | state->plug_started = false; |
9a56a232 | 6644 | state->ios_left = max_ios; |
a1ab7b35 PB |
6645 | /* set only head, no need to init link_last in advance */ |
6646 | state->link.head = NULL; | |
9a56a232 JA |
6647 | } |
6648 | ||
2b188cc1 JA |
6649 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6650 | { | |
75b28aff | 6651 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6652 | |
caf582c6 PB |
6653 | /* |
6654 | * Ensure any loads from the SQEs are done at this point, | |
6655 | * since once we write the new head, the application could | |
6656 | * write new data to them. | |
6657 | */ | |
6658 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6659 | } |
6660 | ||
2b188cc1 | 6661 | /* |
3529d8c2 | 6662 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6663 | * that is mapped by userspace. This means that care needs to be taken to |
6664 | * ensure that reads are stable, as we cannot rely on userspace always | |
6665 | * being a good citizen. If members of the sqe are validated and then later | |
6666 | * used, it's important that those reads are done through READ_ONCE() to | |
6667 | * prevent a re-load down the line. | |
6668 | */ | |
709b302f | 6669 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6670 | { |
75b28aff | 6671 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6672 | unsigned head; |
6673 | ||
6674 | /* | |
6675 | * The cached sq head (or cq tail) serves two purposes: | |
6676 | * | |
6677 | * 1) allows us to batch the cost of updating the user visible | |
6678 | * head updates. | |
6679 | * 2) allows the kernel side to track the head on its own, even | |
6680 | * though the application is the one updating it. | |
6681 | */ | |
4fccfcbb | 6682 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6683 | if (likely(head < ctx->sq_entries)) |
6684 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6685 | |
6686 | /* drop invalid entries */ | |
498ccd9e | 6687 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6688 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6689 | return NULL; |
6690 | } | |
6691 | ||
0f212204 | 6692 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6693 | { |
46c4e16a | 6694 | int submitted = 0; |
6c271ce2 | 6695 | |
ee7d46d9 PB |
6696 | /* make sure SQ entry isn't read before tail */ |
6697 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6698 | |
2b85edfc PB |
6699 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6700 | return -EAGAIN; | |
6c271ce2 | 6701 | |
d8a6df10 | 6702 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6703 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6704 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6705 | |
46c4e16a | 6706 | while (submitted < nr) { |
3529d8c2 | 6707 | const struct io_uring_sqe *sqe; |
196be95c | 6708 | struct io_kiocb *req; |
fb5ccc98 | 6709 | |
258b29a9 | 6710 | req = io_alloc_req(ctx); |
196be95c PB |
6711 | if (unlikely(!req)) { |
6712 | if (!submitted) | |
6713 | submitted = -EAGAIN; | |
fb5ccc98 | 6714 | break; |
196be95c | 6715 | } |
4fccfcbb PB |
6716 | sqe = io_get_sqe(ctx); |
6717 | if (unlikely(!sqe)) { | |
6718 | kmem_cache_free(req_cachep, req); | |
6719 | break; | |
6720 | } | |
d3656344 JA |
6721 | /* will complete beyond this point, count as submitted */ |
6722 | submitted++; | |
a1ab7b35 | 6723 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6724 | break; |
6c271ce2 JA |
6725 | } |
6726 | ||
9466f437 PB |
6727 | if (unlikely(submitted != nr)) { |
6728 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6729 | struct io_uring_task *tctx = current->io_uring; |
6730 | int unused = nr - ref_used; | |
9466f437 | 6731 | |
d8a6df10 JA |
6732 | percpu_ref_put_many(&ctx->refs, unused); |
6733 | percpu_counter_sub(&tctx->inflight, unused); | |
6734 | put_task_struct_many(current, unused); | |
9466f437 | 6735 | } |
6c271ce2 | 6736 | |
a1ab7b35 | 6737 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6738 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6739 | io_commit_sqring(ctx); | |
6740 | ||
6c271ce2 JA |
6741 | return submitted; |
6742 | } | |
6743 | ||
23b3628e XW |
6744 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6745 | { | |
6746 | /* Tell userspace we may need a wakeup call */ | |
6747 | spin_lock_irq(&ctx->completion_lock); | |
6748 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6749 | spin_unlock_irq(&ctx->completion_lock); | |
6750 | } | |
6751 | ||
6752 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6753 | { | |
6754 | spin_lock_irq(&ctx->completion_lock); | |
6755 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6756 | spin_unlock_irq(&ctx->completion_lock); | |
6757 | } | |
6758 | ||
08369246 | 6759 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6760 | { |
c8d1ba58 | 6761 | unsigned int to_submit; |
bdcd3eab | 6762 | int ret = 0; |
6c271ce2 | 6763 | |
c8d1ba58 | 6764 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6765 | /* if we're handling multiple rings, cap submit size for fairness */ |
6766 | if (cap_entries && to_submit > 8) | |
6767 | to_submit = 8; | |
6768 | ||
906a3c6f | 6769 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6770 | unsigned nr_events = 0; |
a4c0b3de | 6771 | |
c8d1ba58 | 6772 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6773 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6774 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6775 | |
3b763ba1 PB |
6776 | /* |
6777 | * Don't submit if refs are dying, good for io_uring_register(), | |
6778 | * but also it is relied upon by io_ring_exit_work() | |
6779 | */ | |
0298ef96 PB |
6780 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6781 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6782 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6783 | mutex_unlock(&ctx->uring_lock); |
6784 | } | |
6c271ce2 | 6785 | |
90554200 JA |
6786 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6787 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6788 | |
08369246 XW |
6789 | return ret; |
6790 | } | |
6c271ce2 | 6791 | |
08369246 XW |
6792 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6793 | { | |
6794 | struct io_ring_ctx *ctx; | |
6795 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6796 | |
c9dca27d PB |
6797 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6798 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6799 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6800 | } |
6c271ce2 | 6801 | |
c8d1ba58 JA |
6802 | static int io_sq_thread(void *data) |
6803 | { | |
69fb2131 JA |
6804 | struct io_sq_data *sqd = data; |
6805 | struct io_ring_ctx *ctx; | |
a0d9205f | 6806 | unsigned long timeout = 0; |
37d1e2e3 | 6807 | char buf[TASK_COMM_LEN]; |
08369246 | 6808 | DEFINE_WAIT(wait); |
6c271ce2 | 6809 | |
696ee88a | 6810 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6811 | set_task_comm(current, buf); |
37d1e2e3 JA |
6812 | current->pf_io_worker = NULL; |
6813 | ||
6814 | if (sqd->sq_cpu != -1) | |
6815 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6816 | else | |
6817 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6818 | current->flags |= PF_NO_SETAFFINITY; | |
6819 | ||
09a6f4ef | 6820 | mutex_lock(&sqd->lock); |
05962f95 | 6821 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6822 | int ret; |
6823 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6824 | |
82734c5b JA |
6825 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6826 | signal_pending(current)) { | |
6827 | bool did_sig = false; | |
6828 | ||
09a6f4ef | 6829 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6830 | if (signal_pending(current)) { |
6831 | struct ksignal ksig; | |
6832 | ||
6833 | did_sig = get_signal(&ksig); | |
6834 | } | |
05962f95 | 6835 | cond_resched(); |
09a6f4ef | 6836 | mutex_lock(&sqd->lock); |
82734c5b JA |
6837 | if (did_sig) |
6838 | break; | |
521d6a73 | 6839 | io_run_task_work(); |
b7f5a0bf | 6840 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6841 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6842 | continue; |
08369246 | 6843 | } |
08369246 | 6844 | sqt_spin = false; |
e95eee2d | 6845 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6846 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6847 | const struct cred *creds = NULL; |
6848 | ||
6849 | if (ctx->sq_creds != current_cred()) | |
6850 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6851 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6852 | if (creds) |
6853 | revert_creds(creds); | |
08369246 XW |
6854 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6855 | sqt_spin = true; | |
69fb2131 | 6856 | } |
6c271ce2 | 6857 | |
08369246 | 6858 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6859 | io_run_task_work(); |
6860 | cond_resched(); | |
08369246 XW |
6861 | if (sqt_spin) |
6862 | timeout = jiffies + sqd->sq_thread_idle; | |
6863 | continue; | |
6864 | } | |
6865 | ||
08369246 | 6866 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); |
724cb4f9 | 6867 | if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6868 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6869 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6870 | |
724cb4f9 HX |
6871 | needs_sched = true; |
6872 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6873 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6874 | !list_empty_careful(&ctx->iopoll_list)) { | |
6875 | needs_sched = false; | |
6876 | break; | |
6877 | } | |
6878 | if (io_sqring_entries(ctx)) { | |
6879 | needs_sched = false; | |
6880 | break; | |
6881 | } | |
6882 | } | |
6883 | ||
6884 | if (needs_sched) { | |
6885 | mutex_unlock(&sqd->lock); | |
6886 | schedule(); | |
6887 | mutex_lock(&sqd->lock); | |
6888 | } | |
69fb2131 JA |
6889 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6890 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6891 | } |
08369246 XW |
6892 | |
6893 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6894 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6895 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6896 | } |
28cea78a | 6897 | |
734551df | 6898 | io_uring_cancel_sqpoll(sqd); |
37d1e2e3 | 6899 | sqd->thread = NULL; |
05962f95 | 6900 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6901 | io_ring_set_wakeup_flag(ctx); |
521d6a73 | 6902 | io_run_task_work(); |
b7f5a0bf | 6903 | io_run_task_work_head(&sqd->park_task_work); |
734551df PB |
6904 | mutex_unlock(&sqd->lock); |
6905 | ||
37d1e2e3 JA |
6906 | complete(&sqd->exited); |
6907 | do_exit(0); | |
6c271ce2 JA |
6908 | } |
6909 | ||
bda52162 JA |
6910 | struct io_wait_queue { |
6911 | struct wait_queue_entry wq; | |
6912 | struct io_ring_ctx *ctx; | |
6913 | unsigned to_wait; | |
6914 | unsigned nr_timeouts; | |
6915 | }; | |
6916 | ||
6c503150 | 6917 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6918 | { |
6919 | struct io_ring_ctx *ctx = iowq->ctx; | |
6920 | ||
6921 | /* | |
d195a66e | 6922 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6923 | * started waiting. For timeouts, we always want to return to userspace, |
6924 | * regardless of event count. | |
6925 | */ | |
6c503150 | 6926 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6927 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6928 | } | |
6929 | ||
6930 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6931 | int wake_flags, void *key) | |
6932 | { | |
6933 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6934 | wq); | |
6935 | ||
6c503150 PB |
6936 | /* |
6937 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6938 | * the task, and the next invocation will do it. | |
6939 | */ | |
6940 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6941 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6942 | return -1; | |
bda52162 JA |
6943 | } |
6944 | ||
af9c1a44 JA |
6945 | static int io_run_task_work_sig(void) |
6946 | { | |
6947 | if (io_run_task_work()) | |
6948 | return 1; | |
6949 | if (!signal_pending(current)) | |
6950 | return 0; | |
0b8cfa97 | 6951 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6952 | return -ERESTARTSYS; |
af9c1a44 JA |
6953 | return -EINTR; |
6954 | } | |
6955 | ||
eeb60b9a PB |
6956 | /* when returns >0, the caller should retry */ |
6957 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6958 | struct io_wait_queue *iowq, | |
6959 | signed long *timeout) | |
6960 | { | |
6961 | int ret; | |
6962 | ||
6963 | /* make sure we run task_work before checking for signals */ | |
6964 | ret = io_run_task_work_sig(); | |
6965 | if (ret || io_should_wake(iowq)) | |
6966 | return ret; | |
6967 | /* let the caller flush overflows, retry */ | |
6968 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6969 | return 1; | |
6970 | ||
6971 | *timeout = schedule_timeout(*timeout); | |
6972 | return !*timeout ? -ETIME : 1; | |
6973 | } | |
6974 | ||
2b188cc1 JA |
6975 | /* |
6976 | * Wait until events become available, if we don't already have some. The | |
6977 | * application must reap them itself, as they reside on the shared cq ring. | |
6978 | */ | |
6979 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6980 | const sigset_t __user *sig, size_t sigsz, |
6981 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6982 | { |
bda52162 JA |
6983 | struct io_wait_queue iowq = { |
6984 | .wq = { | |
6985 | .private = current, | |
6986 | .func = io_wake_function, | |
6987 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
6988 | }, | |
6989 | .ctx = ctx, | |
6990 | .to_wait = min_events, | |
6991 | }; | |
75b28aff | 6992 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
6993 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
6994 | int ret; | |
2b188cc1 | 6995 | |
b41e9852 | 6996 | do { |
6c2450ae | 6997 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 6998 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 6999 | return 0; |
4c6e277c | 7000 | if (!io_run_task_work()) |
b41e9852 | 7001 | break; |
b41e9852 | 7002 | } while (1); |
2b188cc1 JA |
7003 | |
7004 | if (sig) { | |
9e75ad5d AB |
7005 | #ifdef CONFIG_COMPAT |
7006 | if (in_compat_syscall()) | |
7007 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7008 | sigsz); |
9e75ad5d AB |
7009 | else |
7010 | #endif | |
b772434b | 7011 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7012 | |
2b188cc1 JA |
7013 | if (ret) |
7014 | return ret; | |
7015 | } | |
7016 | ||
c73ebb68 | 7017 | if (uts) { |
c1d5a224 PB |
7018 | struct timespec64 ts; |
7019 | ||
c73ebb68 HX |
7020 | if (get_timespec64(&ts, uts)) |
7021 | return -EFAULT; | |
7022 | timeout = timespec64_to_jiffies(&ts); | |
7023 | } | |
7024 | ||
bda52162 | 7025 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7026 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7027 | do { |
ca0a2651 | 7028 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7029 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7030 | ret = -EBUSY; |
7031 | break; | |
7032 | } | |
bda52162 JA |
7033 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7034 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7035 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7036 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 7037 | cond_resched(); |
eeb60b9a | 7038 | } while (ret > 0); |
bda52162 | 7039 | |
b7db41c9 | 7040 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7041 | |
75b28aff | 7042 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7043 | } |
7044 | ||
aeca241b | 7045 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) |
846a4ef2 PB |
7046 | { |
7047 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); | |
7048 | ||
7049 | for (i = 0; i < nr_tables; i++) | |
aeca241b PB |
7050 | kfree(table->files[i]); |
7051 | kfree(table->files); | |
7052 | table->files = NULL; | |
846a4ef2 PB |
7053 | } |
7054 | ||
2a63b2d9 | 7055 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7056 | { |
2a63b2d9 | 7057 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7058 | } |
7059 | ||
2a63b2d9 | 7060 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7061 | { |
2a63b2d9 BM |
7062 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7063 | } | |
65e19f54 | 7064 | |
28a9fe25 PB |
7065 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
7066 | { | |
7067 | percpu_ref_exit(&ref_node->refs); | |
7068 | kfree(ref_node); | |
7069 | } | |
7070 | ||
a7f0ed5a PB |
7071 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7072 | struct io_rsrc_data *data_to_kill) | |
1642b445 | 7073 | { |
a7f0ed5a PB |
7074 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7075 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
82fbcfa9 | 7076 | |
a7f0ed5a PB |
7077 | if (data_to_kill) { |
7078 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7079 | |
a7f0ed5a PB |
7080 | rsrc_node->rsrc_data = data_to_kill; |
7081 | io_rsrc_ref_lock(ctx); | |
7082 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7083 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7084 | |
3e942498 | 7085 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7086 | percpu_ref_kill(&rsrc_node->refs); |
7087 | ctx->rsrc_node = NULL; | |
7088 | } | |
6b06314c | 7089 | |
a7f0ed5a PB |
7090 | if (!ctx->rsrc_node) { |
7091 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7092 | ctx->rsrc_backup_node = NULL; | |
7093 | } | |
8bad28d8 HX |
7094 | } |
7095 | ||
a7f0ed5a | 7096 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7097 | { |
7098 | if (ctx->rsrc_backup_node) | |
7099 | return 0; | |
b895c9a6 | 7100 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe PB |
7101 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
7102 | } | |
7103 | ||
40ae0ff7 | 7104 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7105 | { |
7106 | int ret; | |
05589553 | 7107 | |
215c3902 | 7108 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7109 | if (data->quiesce) |
7110 | return -ENXIO; | |
05589553 | 7111 | |
8bad28d8 | 7112 | data->quiesce = true; |
1ffc5422 | 7113 | do { |
a7f0ed5a | 7114 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7115 | if (ret) |
f2303b1f | 7116 | break; |
a7f0ed5a PB |
7117 | io_rsrc_node_switch(ctx, data); |
7118 | ||
3e942498 PB |
7119 | /* kill initial ref, already quiesced if zero */ |
7120 | if (atomic_dec_and_test(&data->refs)) | |
7121 | break; | |
8bad28d8 | 7122 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7123 | ret = wait_for_completion_interruptible(&data->done); |
7124 | if (!ret) | |
7125 | break; | |
8bad28d8 | 7126 | |
3e942498 PB |
7127 | atomic_inc(&data->refs); |
7128 | /* wait for all works potentially completing data->done */ | |
7129 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7130 | reinit_completion(&data->done); |
8dd03afe | 7131 | |
8bad28d8 | 7132 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7133 | ret = io_run_task_work_sig(); |
8bad28d8 | 7134 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7135 | } while (ret >= 0); |
8bad28d8 | 7136 | data->quiesce = false; |
05f3fb3c | 7137 | |
8bad28d8 | 7138 | return ret; |
d7954b2b BM |
7139 | } |
7140 | ||
44b31f2f PB |
7141 | static void io_rsrc_data_free(struct io_rsrc_data *data) |
7142 | { | |
b60c8dce | 7143 | kvfree(data->tags); |
44b31f2f PB |
7144 | kfree(data); |
7145 | } | |
7146 | ||
40ae0ff7 | 7147 | static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx, |
b60c8dce PB |
7148 | rsrc_put_fn *do_put, |
7149 | unsigned nr) | |
1ad555c6 | 7150 | { |
b895c9a6 | 7151 | struct io_rsrc_data *data; |
1ad555c6 BM |
7152 | |
7153 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7154 | if (!data) | |
7155 | return NULL; | |
7156 | ||
b60c8dce PB |
7157 | data->tags = kvcalloc(nr, sizeof(*data->tags), GFP_KERNEL); |
7158 | if (!data->tags) { | |
7159 | kfree(data); | |
7160 | return NULL; | |
7161 | } | |
7162 | ||
3e942498 | 7163 | atomic_set(&data->refs, 1); |
1ad555c6 | 7164 | data->ctx = ctx; |
40ae0ff7 | 7165 | data->do_put = do_put; |
1ad555c6 BM |
7166 | init_completion(&data->done); |
7167 | return data; | |
7168 | } | |
7169 | ||
fff4db76 PB |
7170 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7171 | { | |
7172 | #if defined(CONFIG_UNIX) | |
7173 | if (ctx->ring_sock) { | |
7174 | struct sock *sock = ctx->ring_sock->sk; | |
7175 | struct sk_buff *skb; | |
7176 | ||
7177 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7178 | kfree_skb(skb); | |
7179 | } | |
7180 | #else | |
7181 | int i; | |
7182 | ||
7183 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7184 | struct file *file; | |
7185 | ||
7186 | file = io_file_from_index(ctx, i); | |
7187 | if (file) | |
7188 | fput(file); | |
7189 | } | |
7190 | #endif | |
7191 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); | |
44b31f2f | 7192 | io_rsrc_data_free(ctx->file_data); |
fff4db76 PB |
7193 | ctx->file_data = NULL; |
7194 | ctx->nr_user_files = 0; | |
7195 | } | |
7196 | ||
d7954b2b BM |
7197 | static int io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7198 | { | |
d7954b2b BM |
7199 | int ret; |
7200 | ||
08480400 | 7201 | if (!ctx->file_data) |
d7954b2b | 7202 | return -ENXIO; |
08480400 PB |
7203 | ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); |
7204 | if (!ret) | |
7205 | __io_sqe_files_unregister(ctx); | |
7206 | return ret; | |
6b06314c JA |
7207 | } |
7208 | ||
37d1e2e3 | 7209 | static void io_sq_thread_unpark(struct io_sq_data *sqd) |
09a6f4ef | 7210 | __releases(&sqd->lock) |
37d1e2e3 | 7211 | { |
521d6a73 PB |
7212 | WARN_ON_ONCE(sqd->thread == current); |
7213 | ||
9e138a48 PB |
7214 | /* |
7215 | * Do the dance but not conditional clear_bit() because it'd race with | |
7216 | * other threads incrementing park_pending and setting the bit. | |
7217 | */ | |
37d1e2e3 | 7218 | clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
9e138a48 PB |
7219 | if (atomic_dec_return(&sqd->park_pending)) |
7220 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); | |
09a6f4ef | 7221 | mutex_unlock(&sqd->lock); |
37d1e2e3 JA |
7222 | } |
7223 | ||
86e0d676 | 7224 | static void io_sq_thread_park(struct io_sq_data *sqd) |
09a6f4ef | 7225 | __acquires(&sqd->lock) |
37d1e2e3 | 7226 | { |
521d6a73 PB |
7227 | WARN_ON_ONCE(sqd->thread == current); |
7228 | ||
9e138a48 | 7229 | atomic_inc(&sqd->park_pending); |
86e0d676 | 7230 | set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); |
09a6f4ef | 7231 | mutex_lock(&sqd->lock); |
05962f95 | 7232 | if (sqd->thread) |
86e0d676 | 7233 | wake_up_process(sqd->thread); |
37d1e2e3 JA |
7234 | } |
7235 | ||
7236 | static void io_sq_thread_stop(struct io_sq_data *sqd) | |
7237 | { | |
521d6a73 | 7238 | WARN_ON_ONCE(sqd->thread == current); |
88885f66 | 7239 | WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); |
521d6a73 | 7240 | |
05962f95 | 7241 | set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); |
88885f66 | 7242 | mutex_lock(&sqd->lock); |
e8f98f24 JA |
7243 | if (sqd->thread) |
7244 | wake_up_process(sqd->thread); | |
09a6f4ef | 7245 | mutex_unlock(&sqd->lock); |
05962f95 | 7246 | wait_for_completion(&sqd->exited); |
37d1e2e3 JA |
7247 | } |
7248 | ||
534ca6d6 | 7249 | static void io_put_sq_data(struct io_sq_data *sqd) |
6c271ce2 | 7250 | { |
534ca6d6 | 7251 | if (refcount_dec_and_test(&sqd->refs)) { |
9e138a48 PB |
7252 | WARN_ON_ONCE(atomic_read(&sqd->park_pending)); |
7253 | ||
37d1e2e3 JA |
7254 | io_sq_thread_stop(sqd); |
7255 | kfree(sqd); | |
7256 | } | |
7257 | } | |
7258 | ||
7259 | static void io_sq_thread_finish(struct io_ring_ctx *ctx) | |
7260 | { | |
7261 | struct io_sq_data *sqd = ctx->sq_data; | |
7262 | ||
7263 | if (sqd) { | |
05962f95 | 7264 | io_sq_thread_park(sqd); |
521d6a73 | 7265 | list_del_init(&ctx->sqd_list); |
37d1e2e3 | 7266 | io_sqd_update_thread_idle(sqd); |
05962f95 | 7267 | io_sq_thread_unpark(sqd); |
37d1e2e3 JA |
7268 | |
7269 | io_put_sq_data(sqd); | |
7270 | ctx->sq_data = NULL; | |
534ca6d6 JA |
7271 | } |
7272 | } | |
7273 | ||
aa06165d JA |
7274 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7275 | { | |
7276 | struct io_ring_ctx *ctx_attach; | |
7277 | struct io_sq_data *sqd; | |
7278 | struct fd f; | |
7279 | ||
7280 | f = fdget(p->wq_fd); | |
7281 | if (!f.file) | |
7282 | return ERR_PTR(-ENXIO); | |
7283 | if (f.file->f_op != &io_uring_fops) { | |
7284 | fdput(f); | |
7285 | return ERR_PTR(-EINVAL); | |
7286 | } | |
7287 | ||
7288 | ctx_attach = f.file->private_data; | |
7289 | sqd = ctx_attach->sq_data; | |
7290 | if (!sqd) { | |
7291 | fdput(f); | |
7292 | return ERR_PTR(-EINVAL); | |
7293 | } | |
5c2469e0 JA |
7294 | if (sqd->task_tgid != current->tgid) { |
7295 | fdput(f); | |
7296 | return ERR_PTR(-EPERM); | |
7297 | } | |
aa06165d JA |
7298 | |
7299 | refcount_inc(&sqd->refs); | |
7300 | fdput(f); | |
7301 | return sqd; | |
7302 | } | |
7303 | ||
26984fbf PB |
7304 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7305 | bool *attached) | |
534ca6d6 JA |
7306 | { |
7307 | struct io_sq_data *sqd; | |
7308 | ||
26984fbf | 7309 | *attached = false; |
5c2469e0 JA |
7310 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7311 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7312 | if (!IS_ERR(sqd)) { |
7313 | *attached = true; | |
5c2469e0 | 7314 | return sqd; |
26984fbf | 7315 | } |
5c2469e0 JA |
7316 | /* fall through for EPERM case, setup new sqd/task */ |
7317 | if (PTR_ERR(sqd) != -EPERM) | |
7318 | return sqd; | |
7319 | } | |
aa06165d | 7320 | |
534ca6d6 JA |
7321 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7322 | if (!sqd) | |
7323 | return ERR_PTR(-ENOMEM); | |
7324 | ||
9e138a48 | 7325 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7326 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7327 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7328 | mutex_init(&sqd->lock); |
534ca6d6 | 7329 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7330 | init_completion(&sqd->exited); |
534ca6d6 JA |
7331 | return sqd; |
7332 | } | |
7333 | ||
6b06314c | 7334 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7335 | /* |
7336 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7337 | * the io_uring can be safely unregistered on process exit, even if we have | |
7338 | * loops in the file referencing. | |
7339 | */ | |
7340 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7341 | { | |
7342 | struct sock *sk = ctx->ring_sock->sk; | |
7343 | struct scm_fp_list *fpl; | |
7344 | struct sk_buff *skb; | |
08a45173 | 7345 | int i, nr_files; |
6b06314c | 7346 | |
6b06314c JA |
7347 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7348 | if (!fpl) | |
7349 | return -ENOMEM; | |
7350 | ||
7351 | skb = alloc_skb(0, GFP_KERNEL); | |
7352 | if (!skb) { | |
7353 | kfree(fpl); | |
7354 | return -ENOMEM; | |
7355 | } | |
7356 | ||
7357 | skb->sk = sk; | |
6b06314c | 7358 | |
08a45173 | 7359 | nr_files = 0; |
62e398be | 7360 | fpl->user = get_uid(current_user()); |
6b06314c | 7361 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7362 | struct file *file = io_file_from_index(ctx, i + offset); |
7363 | ||
7364 | if (!file) | |
08a45173 | 7365 | continue; |
65e19f54 | 7366 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7367 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7368 | nr_files++; | |
6b06314c JA |
7369 | } |
7370 | ||
08a45173 JA |
7371 | if (nr_files) { |
7372 | fpl->max = SCM_MAX_FD; | |
7373 | fpl->count = nr_files; | |
7374 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7375 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7376 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7377 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7378 | |
08a45173 JA |
7379 | for (i = 0; i < nr_files; i++) |
7380 | fput(fpl->fp[i]); | |
7381 | } else { | |
7382 | kfree_skb(skb); | |
7383 | kfree(fpl); | |
7384 | } | |
6b06314c JA |
7385 | |
7386 | return 0; | |
7387 | } | |
7388 | ||
7389 | /* | |
7390 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7391 | * causes regular reference counting to break down. We rely on the UNIX | |
7392 | * garbage collection to take care of this problem for us. | |
7393 | */ | |
7394 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7395 | { | |
7396 | unsigned left, total; | |
7397 | int ret = 0; | |
7398 | ||
7399 | total = 0; | |
7400 | left = ctx->nr_user_files; | |
7401 | while (left) { | |
7402 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7403 | |
7404 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7405 | if (ret) | |
7406 | break; | |
7407 | left -= this_files; | |
7408 | total += this_files; | |
7409 | } | |
7410 | ||
7411 | if (!ret) | |
7412 | return 0; | |
7413 | ||
7414 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7415 | struct file *file = io_file_from_index(ctx, total); |
7416 | ||
7417 | if (file) | |
7418 | fput(file); | |
6b06314c JA |
7419 | total++; |
7420 | } | |
7421 | ||
7422 | return ret; | |
7423 | } | |
7424 | #else | |
7425 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7426 | { | |
7427 | return 0; | |
7428 | } | |
7429 | #endif | |
7430 | ||
aeca241b | 7431 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
65e19f54 | 7432 | { |
846a4ef2 PB |
7433 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); |
7434 | ||
aeca241b PB |
7435 | table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL); |
7436 | if (!table->files) | |
846a4ef2 | 7437 | return false; |
65e19f54 JA |
7438 | |
7439 | for (i = 0; i < nr_tables; i++) { | |
846a4ef2 | 7440 | unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE); |
65e19f54 | 7441 | |
aeca241b | 7442 | table->files[i] = kcalloc(this_files, sizeof(*table->files[i]), |
65e19f54 | 7443 | GFP_KERNEL); |
aeca241b | 7444 | if (!table->files[i]) |
65e19f54 JA |
7445 | break; |
7446 | nr_files -= this_files; | |
7447 | } | |
7448 | ||
7449 | if (i == nr_tables) | |
846a4ef2 | 7450 | return true; |
65e19f54 | 7451 | |
aeca241b | 7452 | io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE); |
846a4ef2 | 7453 | return false; |
65e19f54 JA |
7454 | } |
7455 | ||
47e90392 | 7456 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7457 | { |
50238531 | 7458 | struct file *file = prsrc->file; |
05f3fb3c JA |
7459 | #if defined(CONFIG_UNIX) |
7460 | struct sock *sock = ctx->ring_sock->sk; | |
7461 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7462 | struct sk_buff *skb; | |
7463 | int i; | |
7464 | ||
7465 | __skb_queue_head_init(&list); | |
7466 | ||
7467 | /* | |
7468 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7469 | * remove this entry and rearrange the file array. | |
7470 | */ | |
7471 | skb = skb_dequeue(head); | |
7472 | while (skb) { | |
7473 | struct scm_fp_list *fp; | |
7474 | ||
7475 | fp = UNIXCB(skb).fp; | |
7476 | for (i = 0; i < fp->count; i++) { | |
7477 | int left; | |
7478 | ||
7479 | if (fp->fp[i] != file) | |
7480 | continue; | |
7481 | ||
7482 | unix_notinflight(fp->user, fp->fp[i]); | |
7483 | left = fp->count - 1 - i; | |
7484 | if (left) { | |
7485 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7486 | left * sizeof(struct file *)); | |
7487 | } | |
7488 | fp->count--; | |
7489 | if (!fp->count) { | |
7490 | kfree_skb(skb); | |
7491 | skb = NULL; | |
7492 | } else { | |
7493 | __skb_queue_tail(&list, skb); | |
7494 | } | |
7495 | fput(file); | |
7496 | file = NULL; | |
7497 | break; | |
7498 | } | |
7499 | ||
7500 | if (!file) | |
7501 | break; | |
7502 | ||
7503 | __skb_queue_tail(&list, skb); | |
7504 | ||
7505 | skb = skb_dequeue(head); | |
7506 | } | |
7507 | ||
7508 | if (skb_peek(&list)) { | |
7509 | spin_lock_irq(&head->lock); | |
7510 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7511 | __skb_queue_tail(head, skb); | |
7512 | spin_unlock_irq(&head->lock); | |
7513 | } | |
7514 | #else | |
7515 | fput(file); | |
7516 | #endif | |
7517 | } | |
7518 | ||
b895c9a6 | 7519 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7520 | { |
b895c9a6 | 7521 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7522 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7523 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7524 | |
269bbe5f BM |
7525 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7526 | list_del(&prsrc->list); | |
b60c8dce PB |
7527 | |
7528 | if (prsrc->tag) { | |
7529 | bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; | |
7530 | unsigned long flags; | |
7531 | ||
7532 | io_ring_submit_lock(ctx, lock_ring); | |
7533 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
7534 | io_cqring_fill_event(ctx, prsrc->tag, 0, 0); | |
7535 | io_commit_cqring(ctx); | |
7536 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
7537 | io_cqring_ev_posted(ctx); | |
7538 | io_ring_submit_unlock(ctx, lock_ring); | |
7539 | } | |
7540 | ||
40ae0ff7 | 7541 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7542 | kfree(prsrc); |
65e19f54 | 7543 | } |
05589553 | 7544 | |
28a9fe25 | 7545 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7546 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7547 | complete(&rsrc_data->done); | |
2faf852d | 7548 | } |
65e19f54 | 7549 | |
269bbe5f | 7550 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7551 | { |
7552 | struct io_ring_ctx *ctx; | |
7553 | struct llist_node *node; | |
7554 | ||
269bbe5f BM |
7555 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7556 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7557 | |
7558 | while (node) { | |
b895c9a6 | 7559 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7560 | struct llist_node *next = node->next; |
7561 | ||
b895c9a6 | 7562 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7563 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7564 | node = next; |
7565 | } | |
7566 | } | |
7567 | ||
00835dce | 7568 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7569 | { |
b895c9a6 | 7570 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7571 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7572 | bool first_add = false; |
65e19f54 | 7573 | |
2a63b2d9 | 7574 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7575 | node->done = true; |
e297822b | 7576 | |
d67d2263 | 7577 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7578 | node = list_first_entry(&ctx->rsrc_ref_list, |
7579 | struct io_rsrc_node, node); | |
e297822b | 7580 | /* recycle ref nodes in order */ |
b895c9a6 | 7581 | if (!node->done) |
e297822b | 7582 | break; |
b895c9a6 PB |
7583 | list_del(&node->node); |
7584 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7585 | } |
2a63b2d9 | 7586 | io_rsrc_ref_unlock(ctx); |
05589553 | 7587 | |
3e942498 PB |
7588 | if (first_add) |
7589 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7590 | } |
65e19f54 | 7591 | |
b895c9a6 | 7592 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7593 | { |
b895c9a6 | 7594 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7595 | |
05589553 XW |
7596 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7597 | if (!ref_node) | |
3e2224c5 | 7598 | return NULL; |
05f3fb3c | 7599 | |
00835dce | 7600 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7601 | 0, GFP_KERNEL)) { |
7602 | kfree(ref_node); | |
3e2224c5 | 7603 | return NULL; |
05589553 XW |
7604 | } |
7605 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7606 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7607 | ref_node->done = false; |
05589553 | 7608 | return ref_node; |
05589553 XW |
7609 | } |
7610 | ||
6b06314c | 7611 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
792e3582 | 7612 | unsigned nr_args, u64 __user *tags) |
6b06314c JA |
7613 | { |
7614 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7615 | struct file *file; |
f3baed39 | 7616 | int fd, ret; |
846a4ef2 | 7617 | unsigned i; |
b895c9a6 | 7618 | struct io_rsrc_data *file_data; |
6b06314c | 7619 | |
05f3fb3c | 7620 | if (ctx->file_data) |
6b06314c JA |
7621 | return -EBUSY; |
7622 | if (!nr_args) | |
7623 | return -EINVAL; | |
7624 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7625 | return -EMFILE; | |
a7f0ed5a | 7626 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7627 | if (ret) |
7628 | return ret; | |
6b06314c | 7629 | |
b60c8dce | 7630 | file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put, nr_args); |
5398ae69 | 7631 | if (!file_data) |
05f3fb3c | 7632 | return -ENOMEM; |
13770a71 | 7633 | ctx->file_data = file_data; |
f3baed39 | 7634 | ret = -ENOMEM; |
aeca241b | 7635 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7636 | goto out_free; |
65e19f54 | 7637 | |
08a45173 | 7638 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
792e3582 PB |
7639 | u64 tag = 0; |
7640 | ||
7641 | if ((tags && copy_from_user(&tag, &tags[i], sizeof(tag))) || | |
7642 | copy_from_user(&fd, &fds[i], sizeof(fd))) { | |
600cf3f8 PB |
7643 | ret = -EFAULT; |
7644 | goto out_fput; | |
7645 | } | |
08a45173 | 7646 | /* allow sparse sets */ |
792e3582 PB |
7647 | if (fd == -1) { |
7648 | ret = -EINVAL; | |
7649 | if (unlikely(tag)) | |
7650 | goto out_fput; | |
08a45173 | 7651 | continue; |
792e3582 | 7652 | } |
6b06314c | 7653 | |
05f3fb3c | 7654 | file = fget(fd); |
6b06314c | 7655 | ret = -EBADF; |
792e3582 | 7656 | if (unlikely(!file)) |
600cf3f8 | 7657 | goto out_fput; |
05f3fb3c | 7658 | |
6b06314c JA |
7659 | /* |
7660 | * Don't allow io_uring instances to be registered. If UNIX | |
7661 | * isn't enabled, then this causes a reference cycle and this | |
7662 | * instance can never get freed. If UNIX is enabled we'll | |
7663 | * handle it just fine, but there's still no point in allowing | |
7664 | * a ring fd as it doesn't support regular read/write anyway. | |
7665 | */ | |
05f3fb3c JA |
7666 | if (file->f_op == &io_uring_fops) { |
7667 | fput(file); | |
600cf3f8 | 7668 | goto out_fput; |
6b06314c | 7669 | } |
792e3582 | 7670 | ctx->file_data->tags[i] = tag; |
aeca241b | 7671 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7672 | } |
7673 | ||
6b06314c | 7674 | ret = io_sqe_files_scm(ctx); |
05589553 | 7675 | if (ret) { |
08480400 | 7676 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7677 | return ret; |
7678 | } | |
6b06314c | 7679 | |
a7f0ed5a | 7680 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7681 | return ret; |
600cf3f8 PB |
7682 | out_fput: |
7683 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7684 | file = io_file_from_index(ctx, i); | |
7685 | if (file) | |
7686 | fput(file); | |
7687 | } | |
aeca241b | 7688 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7689 | ctx->nr_user_files = 0; |
600cf3f8 | 7690 | out_free: |
44b31f2f | 7691 | io_rsrc_data_free(ctx->file_data); |
55cbc256 | 7692 | ctx->file_data = NULL; |
6b06314c JA |
7693 | return ret; |
7694 | } | |
7695 | ||
c3a31e60 JA |
7696 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7697 | int index) | |
7698 | { | |
7699 | #if defined(CONFIG_UNIX) | |
7700 | struct sock *sock = ctx->ring_sock->sk; | |
7701 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7702 | struct sk_buff *skb; | |
7703 | ||
7704 | /* | |
7705 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7706 | * file set. If there's no room, fall back to allocating a new skb | |
7707 | * and filling it in. | |
7708 | */ | |
7709 | spin_lock_irq(&head->lock); | |
7710 | skb = skb_peek(head); | |
7711 | if (skb) { | |
7712 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7713 | ||
7714 | if (fpl->count < SCM_MAX_FD) { | |
7715 | __skb_unlink(skb, head); | |
7716 | spin_unlock_irq(&head->lock); | |
7717 | fpl->fp[fpl->count] = get_file(file); | |
7718 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7719 | fpl->count++; | |
7720 | spin_lock_irq(&head->lock); | |
7721 | __skb_queue_head(head, skb); | |
7722 | } else { | |
7723 | skb = NULL; | |
7724 | } | |
7725 | } | |
7726 | spin_unlock_irq(&head->lock); | |
7727 | ||
7728 | if (skb) { | |
7729 | fput(file); | |
7730 | return 0; | |
7731 | } | |
7732 | ||
7733 | return __io_sqe_files_scm(ctx, 1, index); | |
7734 | #else | |
7735 | return 0; | |
7736 | #endif | |
7737 | } | |
7738 | ||
b60c8dce | 7739 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx, |
e7c78371 | 7740 | struct io_rsrc_node *node, void *rsrc) |
05f3fb3c | 7741 | { |
269bbe5f | 7742 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7743 | |
269bbe5f BM |
7744 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7745 | if (!prsrc) | |
a5318d3c | 7746 | return -ENOMEM; |
05f3fb3c | 7747 | |
b60c8dce | 7748 | prsrc->tag = data->tags[idx]; |
50238531 | 7749 | prsrc->rsrc = rsrc; |
e7c78371 | 7750 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7751 | return 0; |
05f3fb3c JA |
7752 | } |
7753 | ||
7754 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
c3bdad02 | 7755 | struct io_uring_rsrc_update2 *up, |
05f3fb3c JA |
7756 | unsigned nr_args) |
7757 | { | |
c3bdad02 | 7758 | u64 __user *tags = u64_to_user_ptr(up->tags); |
98f0b3b4 | 7759 | __s32 __user *fds = u64_to_user_ptr(up->data); |
b895c9a6 | 7760 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7761 | struct io_fixed_file *file_slot; |
7762 | struct file *file; | |
98f0b3b4 PB |
7763 | int fd, i, err = 0; |
7764 | unsigned int done; | |
05589553 | 7765 | bool needs_switch = false; |
c3a31e60 | 7766 | |
98f0b3b4 PB |
7767 | if (!ctx->file_data) |
7768 | return -ENXIO; | |
7769 | if (up->offset + nr_args > ctx->nr_user_files) | |
c3a31e60 | 7770 | return -EINVAL; |
05589553 | 7771 | |
67973b93 | 7772 | for (done = 0; done < nr_args; done++) { |
c3bdad02 PB |
7773 | u64 tag = 0; |
7774 | ||
7775 | if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) || | |
7776 | copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
c3a31e60 JA |
7777 | err = -EFAULT; |
7778 | break; | |
7779 | } | |
c3bdad02 PB |
7780 | if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { |
7781 | err = -EINVAL; | |
7782 | break; | |
7783 | } | |
4e0377a1 | 7784 | if (fd == IORING_REGISTER_FILES_SKIP) |
7785 | continue; | |
7786 | ||
67973b93 | 7787 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7788 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7789 | |
a04b0ac0 PB |
7790 | if (file_slot->file_ptr) { |
7791 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
b60c8dce PB |
7792 | err = io_queue_rsrc_removal(data, up->offset + done, |
7793 | ctx->rsrc_node, file); | |
a5318d3c HD |
7794 | if (err) |
7795 | break; | |
a04b0ac0 | 7796 | file_slot->file_ptr = 0; |
05589553 | 7797 | needs_switch = true; |
c3a31e60 JA |
7798 | } |
7799 | if (fd != -1) { | |
c3a31e60 JA |
7800 | file = fget(fd); |
7801 | if (!file) { | |
7802 | err = -EBADF; | |
7803 | break; | |
7804 | } | |
7805 | /* | |
7806 | * Don't allow io_uring instances to be registered. If | |
7807 | * UNIX isn't enabled, then this causes a reference | |
7808 | * cycle and this instance can never get freed. If UNIX | |
7809 | * is enabled we'll handle it just fine, but there's | |
7810 | * still no point in allowing a ring fd as it doesn't | |
7811 | * support regular read/write anyway. | |
7812 | */ | |
7813 | if (file->f_op == &io_uring_fops) { | |
7814 | fput(file); | |
7815 | err = -EBADF; | |
7816 | break; | |
7817 | } | |
c3bdad02 | 7818 | data->tags[up->offset + done] = tag; |
9a321c98 | 7819 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7820 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7821 | if (err) { |
a04b0ac0 | 7822 | file_slot->file_ptr = 0; |
f3bd9dae | 7823 | fput(file); |
c3a31e60 | 7824 | break; |
f3bd9dae | 7825 | } |
c3a31e60 | 7826 | } |
05f3fb3c JA |
7827 | } |
7828 | ||
a7f0ed5a PB |
7829 | if (needs_switch) |
7830 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7831 | return done ? done : err; |
7832 | } | |
05589553 | 7833 | |
5280f7e5 | 7834 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7835 | { |
7836 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7837 | ||
5280f7e5 PB |
7838 | req = io_put_req_find_next(req); |
7839 | return req ? &req->work : NULL; | |
7d723065 JA |
7840 | } |
7841 | ||
685fe7fe JA |
7842 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7843 | struct task_struct *task) | |
24369c2e | 7844 | { |
e941894e | 7845 | struct io_wq_hash *hash; |
24369c2e | 7846 | struct io_wq_data data; |
24369c2e | 7847 | unsigned int concurrency; |
24369c2e | 7848 | |
e941894e JA |
7849 | hash = ctx->hash_map; |
7850 | if (!hash) { | |
7851 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7852 | if (!hash) | |
7853 | return ERR_PTR(-ENOMEM); | |
7854 | refcount_set(&hash->refs, 1); | |
7855 | init_waitqueue_head(&hash->wait); | |
7856 | ctx->hash_map = hash; | |
24369c2e PB |
7857 | } |
7858 | ||
e941894e | 7859 | data.hash = hash; |
685fe7fe | 7860 | data.task = task; |
e9fd9396 | 7861 | data.free_work = io_free_work; |
f5fa38c5 | 7862 | data.do_work = io_wq_submit_work; |
24369c2e | 7863 | |
d25e3a3d JA |
7864 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7865 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7866 | |
5aa75ed5 | 7867 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7868 | } |
7869 | ||
5aa75ed5 JA |
7870 | static int io_uring_alloc_task_context(struct task_struct *task, |
7871 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7872 | { |
7873 | struct io_uring_task *tctx; | |
d8a6df10 | 7874 | int ret; |
0f212204 JA |
7875 | |
7876 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7877 | if (unlikely(!tctx)) | |
7878 | return -ENOMEM; | |
7879 | ||
d8a6df10 JA |
7880 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7881 | if (unlikely(ret)) { | |
7882 | kfree(tctx); | |
7883 | return ret; | |
7884 | } | |
7885 | ||
685fe7fe | 7886 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7887 | if (IS_ERR(tctx->io_wq)) { |
7888 | ret = PTR_ERR(tctx->io_wq); | |
7889 | percpu_counter_destroy(&tctx->inflight); | |
7890 | kfree(tctx); | |
7891 | return ret; | |
7892 | } | |
7893 | ||
0f212204 JA |
7894 | xa_init(&tctx->xa); |
7895 | init_waitqueue_head(&tctx->wait); | |
7896 | tctx->last = NULL; | |
fdaf083c | 7897 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7898 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7899 | task->io_uring = tctx; |
7cbf1722 JA |
7900 | spin_lock_init(&tctx->task_lock); |
7901 | INIT_WQ_LIST(&tctx->task_list); | |
7902 | tctx->task_state = 0; | |
7903 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7904 | return 0; |
7905 | } | |
7906 | ||
7907 | void __io_uring_free(struct task_struct *tsk) | |
7908 | { | |
7909 | struct io_uring_task *tctx = tsk->io_uring; | |
7910 | ||
7911 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7912 | WARN_ON_ONCE(tctx->io_wq); |
7913 | ||
d8a6df10 | 7914 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7915 | kfree(tctx); |
7916 | tsk->io_uring = NULL; | |
7917 | } | |
7918 | ||
7e84e1c7 SG |
7919 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7920 | struct io_uring_params *p) | |
2b188cc1 JA |
7921 | { |
7922 | int ret; | |
7923 | ||
d25e3a3d JA |
7924 | /* Retain compatibility with failing for an invalid attach attempt */ |
7925 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7926 | IORING_SETUP_ATTACH_WQ) { | |
7927 | struct fd f; | |
7928 | ||
7929 | f = fdget(p->wq_fd); | |
7930 | if (!f.file) | |
7931 | return -ENXIO; | |
d25e3a3d | 7932 | fdput(f); |
f2a48dd0 PB |
7933 | if (f.file->f_op != &io_uring_fops) |
7934 | return -EINVAL; | |
d25e3a3d | 7935 | } |
6c271ce2 | 7936 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7937 | struct task_struct *tsk; |
534ca6d6 | 7938 | struct io_sq_data *sqd; |
26984fbf | 7939 | bool attached; |
534ca6d6 | 7940 | |
26984fbf | 7941 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7942 | if (IS_ERR(sqd)) { |
7943 | ret = PTR_ERR(sqd); | |
7944 | goto err; | |
7945 | } | |
69fb2131 | 7946 | |
7c30f36a | 7947 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7948 | ctx->sq_data = sqd; |
917257da JA |
7949 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7950 | if (!ctx->sq_thread_idle) | |
7951 | ctx->sq_thread_idle = HZ; | |
7952 | ||
78d7f6ba | 7953 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7954 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7955 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7956 | /* don't attach to a dying SQPOLL thread, would be racy */ |
f2a48dd0 | 7957 | ret = (attached && !sqd->thread) ? -ENXIO : 0; |
78d7f6ba PB |
7958 | io_sq_thread_unpark(sqd); |
7959 | ||
de75a3d3 PB |
7960 | if (ret < 0) |
7961 | goto err; | |
7962 | if (attached) | |
5aa75ed5 | 7963 | return 0; |
aa06165d | 7964 | |
6c271ce2 | 7965 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7966 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7967 | |
917257da | 7968 | ret = -EINVAL; |
f2a48dd0 | 7969 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
e8f98f24 | 7970 | goto err_sqpoll; |
37d1e2e3 | 7971 | sqd->sq_cpu = cpu; |
6c271ce2 | 7972 | } else { |
37d1e2e3 | 7973 | sqd->sq_cpu = -1; |
6c271ce2 | 7974 | } |
37d1e2e3 JA |
7975 | |
7976 | sqd->task_pid = current->pid; | |
5c2469e0 | 7977 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7978 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7979 | if (IS_ERR(tsk)) { | |
7980 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7981 | goto err_sqpoll; |
6c271ce2 | 7982 | } |
97a73a0f | 7983 | |
46fe18b1 | 7984 | sqd->thread = tsk; |
97a73a0f | 7985 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7986 | wake_up_new_task(tsk); |
0f212204 JA |
7987 | if (ret) |
7988 | goto err; | |
6c271ce2 JA |
7989 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7990 | /* Can't have SQ_AFF without SQPOLL */ | |
7991 | ret = -EINVAL; | |
7992 | goto err; | |
7993 | } | |
7994 | ||
2b188cc1 | 7995 | return 0; |
f2a48dd0 PB |
7996 | err_sqpoll: |
7997 | complete(&ctx->sq_data->exited); | |
2b188cc1 | 7998 | err: |
37d1e2e3 | 7999 | io_sq_thread_finish(ctx); |
2b188cc1 JA |
8000 | return ret; |
8001 | } | |
8002 | ||
a087e2b5 BM |
8003 | static inline void __io_unaccount_mem(struct user_struct *user, |
8004 | unsigned long nr_pages) | |
2b188cc1 JA |
8005 | { |
8006 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8007 | } | |
8008 | ||
a087e2b5 BM |
8009 | static inline int __io_account_mem(struct user_struct *user, |
8010 | unsigned long nr_pages) | |
2b188cc1 JA |
8011 | { |
8012 | unsigned long page_limit, cur_pages, new_pages; | |
8013 | ||
8014 | /* Don't allow more pages than we can safely lock */ | |
8015 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8016 | ||
8017 | do { | |
8018 | cur_pages = atomic_long_read(&user->locked_vm); | |
8019 | new_pages = cur_pages + nr_pages; | |
8020 | if (new_pages > page_limit) | |
8021 | return -ENOMEM; | |
8022 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8023 | new_pages) != cur_pages); | |
8024 | ||
8025 | return 0; | |
8026 | } | |
8027 | ||
26bfa89e | 8028 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8029 | { |
62e398be | 8030 | if (ctx->user) |
a087e2b5 | 8031 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8032 | |
26bfa89e JA |
8033 | if (ctx->mm_account) |
8034 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8035 | } |
8036 | ||
26bfa89e | 8037 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8038 | { |
30975825 BM |
8039 | int ret; |
8040 | ||
62e398be | 8041 | if (ctx->user) { |
30975825 BM |
8042 | ret = __io_account_mem(ctx->user, nr_pages); |
8043 | if (ret) | |
8044 | return ret; | |
8045 | } | |
8046 | ||
26bfa89e JA |
8047 | if (ctx->mm_account) |
8048 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8049 | |
8050 | return 0; | |
8051 | } | |
8052 | ||
2b188cc1 JA |
8053 | static void io_mem_free(void *ptr) |
8054 | { | |
52e04ef4 MR |
8055 | struct page *page; |
8056 | ||
8057 | if (!ptr) | |
8058 | return; | |
2b188cc1 | 8059 | |
52e04ef4 | 8060 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8061 | if (put_page_testzero(page)) |
8062 | free_compound_page(page); | |
8063 | } | |
8064 | ||
8065 | static void *io_mem_alloc(size_t size) | |
8066 | { | |
8067 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8068 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8069 | |
8070 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8071 | } | |
8072 | ||
75b28aff HV |
8073 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8074 | size_t *sq_offset) | |
8075 | { | |
8076 | struct io_rings *rings; | |
8077 | size_t off, sq_array_size; | |
8078 | ||
8079 | off = struct_size(rings, cqes, cq_entries); | |
8080 | if (off == SIZE_MAX) | |
8081 | return SIZE_MAX; | |
8082 | ||
8083 | #ifdef CONFIG_SMP | |
8084 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8085 | if (off == 0) | |
8086 | return SIZE_MAX; | |
8087 | #endif | |
8088 | ||
b36200f5 DV |
8089 | if (sq_offset) |
8090 | *sq_offset = off; | |
8091 | ||
75b28aff HV |
8092 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8093 | if (sq_array_size == SIZE_MAX) | |
8094 | return SIZE_MAX; | |
8095 | ||
8096 | if (check_add_overflow(off, sq_array_size, &off)) | |
8097 | return SIZE_MAX; | |
8098 | ||
75b28aff HV |
8099 | return off; |
8100 | } | |
8101 | ||
41edf1a5 | 8102 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot) |
7f61a1e9 | 8103 | { |
41edf1a5 | 8104 | struct io_mapped_ubuf *imu = *slot; |
7f61a1e9 PB |
8105 | unsigned int i; |
8106 | ||
8107 | for (i = 0; i < imu->nr_bvecs; i++) | |
8108 | unpin_user_page(imu->bvec[i].bv_page); | |
8109 | if (imu->acct_pages) | |
8110 | io_unaccount_mem(ctx, imu->acct_pages); | |
41edf1a5 PB |
8111 | kvfree(imu); |
8112 | *slot = NULL; | |
7f61a1e9 PB |
8113 | } |
8114 | ||
bd54b6fe | 8115 | static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
edafccee | 8116 | { |
bd54b6fe BM |
8117 | /* no updates yet, so not used */ |
8118 | WARN_ON_ONCE(1); | |
8119 | } | |
edafccee | 8120 | |
bd54b6fe BM |
8121 | static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
8122 | { | |
8123 | unsigned int i; | |
edafccee | 8124 | |
7f61a1e9 PB |
8125 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8126 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee | 8127 | kfree(ctx->user_bufs); |
bd54b6fe | 8128 | kfree(ctx->buf_data); |
edafccee | 8129 | ctx->user_bufs = NULL; |
bd54b6fe | 8130 | ctx->buf_data = NULL; |
edafccee | 8131 | ctx->nr_user_bufs = 0; |
bd54b6fe BM |
8132 | } |
8133 | ||
8134 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) | |
8135 | { | |
8136 | int ret; | |
8137 | ||
8138 | if (!ctx->buf_data) | |
8139 | return -ENXIO; | |
8140 | ||
8141 | ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); | |
8142 | if (!ret) | |
8143 | __io_sqe_buffers_unregister(ctx); | |
8144 | return ret; | |
edafccee JA |
8145 | } |
8146 | ||
8147 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8148 | void __user *arg, unsigned index) | |
8149 | { | |
8150 | struct iovec __user *src; | |
8151 | ||
8152 | #ifdef CONFIG_COMPAT | |
8153 | if (ctx->compat) { | |
8154 | struct compat_iovec __user *ciovs; | |
8155 | struct compat_iovec ciov; | |
8156 | ||
8157 | ciovs = (struct compat_iovec __user *) arg; | |
8158 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8159 | return -EFAULT; | |
8160 | ||
d55e5f5b | 8161 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8162 | dst->iov_len = ciov.iov_len; |
8163 | return 0; | |
8164 | } | |
8165 | #endif | |
8166 | src = (struct iovec __user *) arg; | |
8167 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8168 | return -EFAULT; | |
8169 | return 0; | |
8170 | } | |
8171 | ||
de293938 JA |
8172 | /* |
8173 | * Not super efficient, but this is just a registration time. And we do cache | |
8174 | * the last compound head, so generally we'll only do a full search if we don't | |
8175 | * match that one. | |
8176 | * | |
8177 | * We check if the given compound head page has already been accounted, to | |
8178 | * avoid double accounting it. This allows us to account the full size of the | |
8179 | * page, not just the constituent pages of a huge page. | |
8180 | */ | |
8181 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8182 | int nr_pages, struct page *hpage) | |
8183 | { | |
8184 | int i, j; | |
8185 | ||
8186 | /* check current page array */ | |
8187 | for (i = 0; i < nr_pages; i++) { | |
8188 | if (!PageCompound(pages[i])) | |
8189 | continue; | |
8190 | if (compound_head(pages[i]) == hpage) | |
8191 | return true; | |
8192 | } | |
8193 | ||
8194 | /* check previously registered pages */ | |
8195 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
41edf1a5 | 8196 | struct io_mapped_ubuf *imu = ctx->user_bufs[i]; |
de293938 JA |
8197 | |
8198 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8199 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8200 | continue; | |
8201 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8202 | return true; | |
8203 | } | |
8204 | } | |
8205 | ||
8206 | return false; | |
8207 | } | |
8208 | ||
8209 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8210 | int nr_pages, struct io_mapped_ubuf *imu, | |
8211 | struct page **last_hpage) | |
8212 | { | |
8213 | int i, ret; | |
8214 | ||
8215 | for (i = 0; i < nr_pages; i++) { | |
8216 | if (!PageCompound(pages[i])) { | |
8217 | imu->acct_pages++; | |
8218 | } else { | |
8219 | struct page *hpage; | |
8220 | ||
8221 | hpage = compound_head(pages[i]); | |
8222 | if (hpage == *last_hpage) | |
8223 | continue; | |
8224 | *last_hpage = hpage; | |
8225 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8226 | continue; | |
8227 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8228 | } | |
8229 | } | |
8230 | ||
8231 | if (!imu->acct_pages) | |
8232 | return 0; | |
8233 | ||
26bfa89e | 8234 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8235 | if (ret) |
8236 | imu->acct_pages = 0; | |
8237 | return ret; | |
8238 | } | |
8239 | ||
0a96bbe4 | 8240 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
41edf1a5 | 8241 | struct io_mapped_ubuf **pimu, |
0a96bbe4 | 8242 | struct page **last_hpage) |
edafccee | 8243 | { |
41edf1a5 | 8244 | struct io_mapped_ubuf *imu = NULL; |
edafccee JA |
8245 | struct vm_area_struct **vmas = NULL; |
8246 | struct page **pages = NULL; | |
0a96bbe4 BM |
8247 | unsigned long off, start, end, ubuf; |
8248 | size_t size; | |
8249 | int ret, pret, nr_pages, i; | |
8250 | ||
8251 | ubuf = (unsigned long) iov->iov_base; | |
8252 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8253 | start = ubuf >> PAGE_SHIFT; | |
8254 | nr_pages = end - start; | |
8255 | ||
41edf1a5 | 8256 | *pimu = NULL; |
0a96bbe4 BM |
8257 | ret = -ENOMEM; |
8258 | ||
8259 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8260 | if (!pages) | |
8261 | goto done; | |
8262 | ||
8263 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8264 | GFP_KERNEL); | |
8265 | if (!vmas) | |
8266 | goto done; | |
edafccee | 8267 | |
41edf1a5 | 8268 | imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL); |
0a96bbe4 BM |
8269 | if (!imu->bvec) |
8270 | goto done; | |
8271 | ||
8272 | ret = 0; | |
8273 | mmap_read_lock(current->mm); | |
8274 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8275 | pages, vmas); | |
8276 | if (pret == nr_pages) { | |
8277 | /* don't support file backed memory */ | |
8278 | for (i = 0; i < nr_pages; i++) { | |
8279 | struct vm_area_struct *vma = vmas[i]; | |
8280 | ||
8281 | if (vma->vm_file && | |
8282 | !is_file_hugepages(vma->vm_file)) { | |
8283 | ret = -EOPNOTSUPP; | |
8284 | break; | |
8285 | } | |
8286 | } | |
8287 | } else { | |
8288 | ret = pret < 0 ? pret : -EFAULT; | |
8289 | } | |
8290 | mmap_read_unlock(current->mm); | |
8291 | if (ret) { | |
8292 | /* | |
8293 | * if we did partial map, or found file backed vmas, | |
8294 | * release any pages we did get | |
8295 | */ | |
8296 | if (pret > 0) | |
8297 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8298 | goto done; |
8299 | } | |
8300 | ||
8301 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8302 | if (ret) { | |
8303 | unpin_user_pages(pages, pret); | |
0a96bbe4 BM |
8304 | goto done; |
8305 | } | |
8306 | ||
8307 | off = ubuf & ~PAGE_MASK; | |
8308 | size = iov->iov_len; | |
8309 | for (i = 0; i < nr_pages; i++) { | |
8310 | size_t vec_len; | |
8311 | ||
8312 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8313 | imu->bvec[i].bv_page = pages[i]; | |
8314 | imu->bvec[i].bv_len = vec_len; | |
8315 | imu->bvec[i].bv_offset = off; | |
8316 | off = 0; | |
8317 | size -= vec_len; | |
8318 | } | |
8319 | /* store original address for later verification */ | |
8320 | imu->ubuf = ubuf; | |
4751f53d | 8321 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 | 8322 | imu->nr_bvecs = nr_pages; |
41edf1a5 | 8323 | *pimu = imu; |
0a96bbe4 BM |
8324 | ret = 0; |
8325 | done: | |
41edf1a5 PB |
8326 | if (ret) |
8327 | kvfree(imu); | |
0a96bbe4 BM |
8328 | kvfree(pages); |
8329 | kvfree(vmas); | |
8330 | return ret; | |
8331 | } | |
8332 | ||
2b358604 | 8333 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8334 | { |
87094465 PB |
8335 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8336 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8337 | } |
edafccee | 8338 | |
2b358604 BM |
8339 | static int io_buffer_validate(struct iovec *iov) |
8340 | { | |
50e96989 PB |
8341 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8342 | ||
2b358604 BM |
8343 | /* |
8344 | * Don't impose further limits on the size and buffer | |
8345 | * constraints here, we'll -EINVAL later when IO is | |
8346 | * submitted if they are wrong. | |
8347 | */ | |
8348 | if (!iov->iov_base || !iov->iov_len) | |
8349 | return -EFAULT; | |
edafccee | 8350 | |
2b358604 BM |
8351 | /* arbitrary limit, but we need something */ |
8352 | if (iov->iov_len > SZ_1G) | |
8353 | return -EFAULT; | |
edafccee | 8354 | |
50e96989 PB |
8355 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8356 | return -EOVERFLOW; | |
8357 | ||
2b358604 BM |
8358 | return 0; |
8359 | } | |
edafccee | 8360 | |
2b358604 BM |
8361 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8362 | unsigned int nr_args) | |
8363 | { | |
bd54b6fe BM |
8364 | struct page *last_hpage = NULL; |
8365 | struct io_rsrc_data *data; | |
2b358604 BM |
8366 | int i, ret; |
8367 | struct iovec iov; | |
edafccee | 8368 | |
87094465 PB |
8369 | if (ctx->user_bufs) |
8370 | return -EBUSY; | |
8371 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8372 | return -EINVAL; | |
bd54b6fe | 8373 | ret = io_rsrc_node_switch_start(ctx); |
2b358604 BM |
8374 | if (ret) |
8375 | return ret; | |
bd54b6fe BM |
8376 | data = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, nr_args); |
8377 | if (!data) | |
8378 | return -ENOMEM; | |
8379 | ret = io_buffers_map_alloc(ctx, nr_args); | |
8380 | if (ret) { | |
8381 | kfree(data); | |
8382 | return ret; | |
8383 | } | |
edafccee | 8384 | |
87094465 | 8385 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee JA |
8386 | ret = io_copy_iov(ctx, &iov, arg, i); |
8387 | if (ret) | |
0a96bbe4 | 8388 | break; |
2b358604 BM |
8389 | ret = io_buffer_validate(&iov); |
8390 | if (ret) | |
0a96bbe4 | 8391 | break; |
41edf1a5 PB |
8392 | |
8393 | ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], | |
8394 | &last_hpage); | |
0a96bbe4 BM |
8395 | if (ret) |
8396 | break; | |
edafccee | 8397 | } |
0a96bbe4 | 8398 | |
bd54b6fe | 8399 | WARN_ON_ONCE(ctx->buf_data); |
0a96bbe4 | 8400 | |
bd54b6fe BM |
8401 | ctx->buf_data = data; |
8402 | if (ret) | |
8403 | __io_sqe_buffers_unregister(ctx); | |
8404 | else | |
8405 | io_rsrc_node_switch(ctx, NULL); | |
edafccee JA |
8406 | return ret; |
8407 | } | |
8408 | ||
9b402849 JA |
8409 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8410 | { | |
8411 | __s32 __user *fds = arg; | |
8412 | int fd; | |
8413 | ||
8414 | if (ctx->cq_ev_fd) | |
8415 | return -EBUSY; | |
8416 | ||
8417 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8418 | return -EFAULT; | |
8419 | ||
8420 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8421 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8422 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8423 | ctx->cq_ev_fd = NULL; | |
8424 | return ret; | |
8425 | } | |
8426 | ||
8427 | return 0; | |
8428 | } | |
8429 | ||
8430 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8431 | { | |
8432 | if (ctx->cq_ev_fd) { | |
8433 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8434 | ctx->cq_ev_fd = NULL; | |
8435 | return 0; | |
8436 | } | |
8437 | ||
8438 | return -ENXIO; | |
8439 | } | |
8440 | ||
5a2e745d JA |
8441 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8442 | { | |
9e15c3a0 JA |
8443 | struct io_buffer *buf; |
8444 | unsigned long index; | |
8445 | ||
8446 | xa_for_each(&ctx->io_buffers, index, buf) | |
8447 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8448 | } |
8449 | ||
68e68ee6 | 8450 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8451 | { |
68e68ee6 | 8452 | struct io_kiocb *req, *nxt; |
1b4c351f | 8453 | |
68e68ee6 JA |
8454 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8455 | if (tsk && req->task != tsk) | |
8456 | continue; | |
1b4c351f JA |
8457 | list_del(&req->compl.list); |
8458 | kmem_cache_free(req_cachep, req); | |
8459 | } | |
8460 | } | |
8461 | ||
4010fec4 | 8462 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8463 | { |
bf019da7 | 8464 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8465 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8466 | |
9a4fdbd8 JA |
8467 | mutex_lock(&ctx->uring_lock); |
8468 | ||
8e5c66c4 | 8469 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8470 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8471 | submit_state->reqs); | |
8e5c66c4 PB |
8472 | submit_state->free_reqs = 0; |
8473 | } | |
9a4fdbd8 | 8474 | |
dac7a098 | 8475 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8476 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8477 | mutex_unlock(&ctx->uring_lock); |
8478 | } | |
8479 | ||
bd54b6fe BM |
8480 | static bool io_wait_rsrc_data(struct io_rsrc_data *data) |
8481 | { | |
8482 | if (!data) | |
8483 | return false; | |
8484 | if (!atomic_dec_and_test(&data->refs)) | |
8485 | wait_for_completion(&data->done); | |
8486 | return true; | |
8487 | } | |
8488 | ||
2b188cc1 JA |
8489 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8490 | { | |
37d1e2e3 | 8491 | io_sq_thread_finish(ctx); |
2aede0e4 | 8492 | |
37d1e2e3 | 8493 | if (ctx->mm_account) { |
2aede0e4 JA |
8494 | mmdrop(ctx->mm_account); |
8495 | ctx->mm_account = NULL; | |
30975825 | 8496 | } |
def596e9 | 8497 | |
8bad28d8 | 8498 | mutex_lock(&ctx->uring_lock); |
bd54b6fe BM |
8499 | if (io_wait_rsrc_data(ctx->buf_data)) |
8500 | __io_sqe_buffers_unregister(ctx); | |
8501 | if (io_wait_rsrc_data(ctx->file_data)) | |
08480400 | 8502 | __io_sqe_files_unregister(ctx); |
c4ea060e PB |
8503 | if (ctx->rings) |
8504 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8505 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8506 | io_eventfd_unregister(ctx); |
5a2e745d | 8507 | io_destroy_buffers(ctx); |
07db298a PB |
8508 | if (ctx->sq_creds) |
8509 | put_cred(ctx->sq_creds); | |
def596e9 | 8510 | |
a7f0ed5a PB |
8511 | /* there are no registered resources left, nobody uses it */ |
8512 | if (ctx->rsrc_node) | |
8513 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8514 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8515 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8516 | flush_delayed_work(&ctx->rsrc_put_work); |
8517 | ||
8518 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8519 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
8dd03afe | 8520 | |
2b188cc1 | 8521 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8522 | if (ctx->ring_sock) { |
8523 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8524 | sock_release(ctx->ring_sock); |
355e8d26 | 8525 | } |
2b188cc1 JA |
8526 | #endif |
8527 | ||
75b28aff | 8528 | io_mem_free(ctx->rings); |
2b188cc1 | 8529 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8530 | |
8531 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8532 | free_uid(ctx->user); |
4010fec4 | 8533 | io_req_caches_free(ctx); |
e941894e JA |
8534 | if (ctx->hash_map) |
8535 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8536 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8537 | kfree(ctx); |
8538 | } | |
8539 | ||
8540 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8541 | { | |
8542 | struct io_ring_ctx *ctx = file->private_data; | |
8543 | __poll_t mask = 0; | |
8544 | ||
8545 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8546 | /* |
8547 | * synchronizes with barrier from wq_has_sleeper call in | |
8548 | * io_commit_cqring | |
8549 | */ | |
2b188cc1 | 8550 | smp_rmb(); |
90554200 | 8551 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8552 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8553 | |
8554 | /* | |
8555 | * Don't flush cqring overflow list here, just do a simple check. | |
8556 | * Otherwise there could possible be ABBA deadlock: | |
8557 | * CPU0 CPU1 | |
8558 | * ---- ---- | |
8559 | * lock(&ctx->uring_lock); | |
8560 | * lock(&ep->mtx); | |
8561 | * lock(&ctx->uring_lock); | |
8562 | * lock(&ep->mtx); | |
8563 | * | |
8564 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8565 | * pushs them to do the flush. | |
8566 | */ | |
8567 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8568 | mask |= EPOLLIN | EPOLLRDNORM; |
8569 | ||
8570 | return mask; | |
8571 | } | |
8572 | ||
8573 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8574 | { | |
8575 | struct io_ring_ctx *ctx = file->private_data; | |
8576 | ||
8577 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8578 | } | |
8579 | ||
0bead8cd | 8580 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8581 | { |
4379bf8b | 8582 | const struct cred *creds; |
071698e1 | 8583 | |
61cf9370 | 8584 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8585 | if (creds) { |
8586 | put_cred(creds); | |
0bead8cd | 8587 | return 0; |
1e6fa521 | 8588 | } |
0bead8cd YD |
8589 | |
8590 | return -EINVAL; | |
8591 | } | |
8592 | ||
9b465711 | 8593 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8594 | { |
9b465711 | 8595 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8596 | } |
8597 | ||
d56d938b PB |
8598 | struct io_tctx_exit { |
8599 | struct callback_head task_work; | |
8600 | struct completion completion; | |
baf186c4 | 8601 | struct io_ring_ctx *ctx; |
d56d938b PB |
8602 | }; |
8603 | ||
8604 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8605 | { | |
8606 | struct io_uring_task *tctx = current->io_uring; | |
8607 | struct io_tctx_exit *work; | |
8608 | ||
8609 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8610 | /* | |
8611 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8612 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8613 | */ | |
8614 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8615 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8616 | complete(&work->completion); |
8617 | } | |
8618 | ||
85faa7b8 JA |
8619 | static void io_ring_exit_work(struct work_struct *work) |
8620 | { | |
d56d938b | 8621 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8622 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8623 | struct io_tctx_exit exit; |
8624 | struct io_tctx_node *node; | |
8625 | int ret; | |
85faa7b8 | 8626 | |
56952e91 JA |
8627 | /* |
8628 | * If we're doing polled IO and end up having requests being | |
8629 | * submitted async (out-of-line), then completions can come in while | |
8630 | * we're waiting for refs to drop. We need to reap these manually, | |
8631 | * as nobody else will be looking for them. | |
8632 | */ | |
b2edc0a7 | 8633 | do { |
9936c7c2 | 8634 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8635 | |
8636 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8637 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8638 | |
7f00651a PB |
8639 | init_completion(&exit.completion); |
8640 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8641 | exit.ctx = ctx; | |
89b5066e PB |
8642 | /* |
8643 | * Some may use context even when all refs and requests have been put, | |
8644 | * and they are free to do so while still holding uring_lock or | |
8645 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8646 | * this lock/unlock section also waits them to finish. | |
8647 | */ | |
d56d938b PB |
8648 | mutex_lock(&ctx->uring_lock); |
8649 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8650 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8651 | ||
d56d938b PB |
8652 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8653 | ctx_node); | |
7f00651a PB |
8654 | /* don't spin on a single task if cancellation failed */ |
8655 | list_rotate_left(&ctx->tctx_list); | |
d56d938b PB |
8656 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); |
8657 | if (WARN_ON_ONCE(ret)) | |
8658 | continue; | |
8659 | wake_up_process(node->task); | |
8660 | ||
8661 | mutex_unlock(&ctx->uring_lock); | |
8662 | wait_for_completion(&exit.completion); | |
d56d938b PB |
8663 | mutex_lock(&ctx->uring_lock); |
8664 | } | |
8665 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8666 | spin_lock_irq(&ctx->completion_lock); |
8667 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8668 | |
85faa7b8 JA |
8669 | io_ring_ctx_free(ctx); |
8670 | } | |
8671 | ||
80c4cbdb PB |
8672 | /* Returns true if we found and killed one or more timeouts */ |
8673 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8674 | struct files_struct *files) | |
8675 | { | |
8676 | struct io_kiocb *req, *tmp; | |
8677 | int canceled = 0; | |
8678 | ||
8679 | spin_lock_irq(&ctx->completion_lock); | |
8680 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8681 | if (io_match_task(req, tsk, files)) { | |
8682 | io_kill_timeout(req, -ECANCELED); | |
8683 | canceled++; | |
8684 | } | |
8685 | } | |
51520426 PB |
8686 | if (canceled != 0) |
8687 | io_commit_cqring(ctx); | |
80c4cbdb | 8688 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8689 | if (canceled != 0) |
8690 | io_cqring_ev_posted(ctx); | |
8691 | return canceled != 0; | |
8692 | } | |
8693 | ||
2b188cc1 JA |
8694 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8695 | { | |
61cf9370 MWO |
8696 | unsigned long index; |
8697 | struct creds *creds; | |
8698 | ||
2b188cc1 JA |
8699 | mutex_lock(&ctx->uring_lock); |
8700 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8701 | if (ctx->rings) |
6c2450ae | 8702 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8703 | xa_for_each(&ctx->personalities, index, creds) |
8704 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8705 | mutex_unlock(&ctx->uring_lock); |
8706 | ||
6b81928d PB |
8707 | io_kill_timeouts(ctx, NULL, NULL); |
8708 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8709 | |
15dff286 | 8710 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8711 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8712 | |
85faa7b8 | 8713 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8714 | /* |
8715 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8716 | * if we're exiting a ton of rings at the same time. It just adds | |
8717 | * noise and overhead, there's no discernable change in runtime | |
8718 | * over using system_wq. | |
8719 | */ | |
8720 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8721 | } |
8722 | ||
8723 | static int io_uring_release(struct inode *inode, struct file *file) | |
8724 | { | |
8725 | struct io_ring_ctx *ctx = file->private_data; | |
8726 | ||
8727 | file->private_data = NULL; | |
8728 | io_ring_ctx_wait_and_kill(ctx); | |
8729 | return 0; | |
8730 | } | |
8731 | ||
f6edbabb PB |
8732 | struct io_task_cancel { |
8733 | struct task_struct *task; | |
8734 | struct files_struct *files; | |
8735 | }; | |
f254ac04 | 8736 | |
f6edbabb | 8737 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8738 | { |
9a472ef7 | 8739 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8740 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8741 | bool ret; |
8742 | ||
f6edbabb | 8743 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8744 | unsigned long flags; |
8745 | struct io_ring_ctx *ctx = req->ctx; | |
8746 | ||
8747 | /* protect against races with linked timeouts */ | |
8748 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8749 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8750 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8751 | } else { | |
f6edbabb | 8752 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8753 | } |
8754 | return ret; | |
b711d4ea JA |
8755 | } |
8756 | ||
e1915f76 | 8757 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8758 | struct task_struct *task, |
b7ddce3c PB |
8759 | struct files_struct *files) |
8760 | { | |
e1915f76 | 8761 | struct io_defer_entry *de; |
b7ddce3c PB |
8762 | LIST_HEAD(list); |
8763 | ||
8764 | spin_lock_irq(&ctx->completion_lock); | |
8765 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8766 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8767 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8768 | break; | |
8769 | } | |
8770 | } | |
8771 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8772 | if (list_empty(&list)) |
8773 | return false; | |
b7ddce3c PB |
8774 | |
8775 | while (!list_empty(&list)) { | |
8776 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8777 | list_del_init(&de->list); | |
f41db273 | 8778 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8779 | kfree(de); |
8780 | } | |
e1915f76 | 8781 | return true; |
b7ddce3c PB |
8782 | } |
8783 | ||
1b00764f PB |
8784 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8785 | { | |
8786 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8787 | ||
8788 | return req->ctx == data; | |
8789 | } | |
8790 | ||
8791 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8792 | { | |
8793 | struct io_tctx_node *node; | |
8794 | enum io_wq_cancel cret; | |
8795 | bool ret = false; | |
8796 | ||
8797 | mutex_lock(&ctx->uring_lock); | |
8798 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8799 | struct io_uring_task *tctx = node->task->io_uring; | |
8800 | ||
8801 | /* | |
8802 | * io_wq will stay alive while we hold uring_lock, because it's | |
8803 | * killed after ctx nodes, which requires to take the lock. | |
8804 | */ | |
8805 | if (!tctx || !tctx->io_wq) | |
8806 | continue; | |
8807 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8808 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8809 | } | |
8810 | mutex_unlock(&ctx->uring_lock); | |
8811 | ||
8812 | return ret; | |
8813 | } | |
8814 | ||
9936c7c2 PB |
8815 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8816 | struct task_struct *task, | |
8817 | struct files_struct *files) | |
8818 | { | |
8819 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8820 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8821 | |
8822 | while (1) { | |
8823 | enum io_wq_cancel cret; | |
8824 | bool ret = false; | |
8825 | ||
1b00764f PB |
8826 | if (!task) { |
8827 | ret |= io_uring_try_cancel_iowq(ctx); | |
8828 | } else if (tctx && tctx->io_wq) { | |
8829 | /* | |
8830 | * Cancels requests of all rings, not only @ctx, but | |
8831 | * it's fine as the task is in exit/exec. | |
8832 | */ | |
5aa75ed5 | 8833 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8834 | &cancel, true); |
8835 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8836 | } | |
8837 | ||
8838 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8839 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8840 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8841 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8842 | io_iopoll_try_reap_events(ctx); | |
8843 | ret = true; | |
8844 | } | |
8845 | } | |
8846 | ||
e1915f76 | 8847 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8848 | ret |= io_poll_remove_all(ctx, task, files); |
8849 | ret |= io_kill_timeouts(ctx, task, files); | |
8850 | ret |= io_run_task_work(); | |
ba50a036 | 8851 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8852 | if (!ret) |
8853 | break; | |
8854 | cond_resched(); | |
8855 | } | |
8856 | } | |
8857 | ||
cf27f3b1 | 8858 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8859 | { |
236434c3 | 8860 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8861 | struct io_tctx_node *node; |
a528b04e | 8862 | int ret; |
236434c3 MWO |
8863 | |
8864 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8865 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8866 | if (unlikely(ret)) |
8867 | return ret; | |
236434c3 | 8868 | tctx = current->io_uring; |
0f212204 | 8869 | } |
cf27f3b1 PB |
8870 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8871 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8872 | if (!node) | |
8873 | return -ENOMEM; | |
8874 | node->ctx = ctx; | |
8875 | node->task = current; | |
13bf43f5 | 8876 | |
cf27f3b1 PB |
8877 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8878 | node, GFP_KERNEL)); | |
8879 | if (ret) { | |
8880 | kfree(node); | |
8881 | return ret; | |
0f212204 | 8882 | } |
cf27f3b1 PB |
8883 | |
8884 | mutex_lock(&ctx->uring_lock); | |
8885 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8886 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8887 | } |
cf27f3b1 | 8888 | tctx->last = ctx; |
0f212204 JA |
8889 | return 0; |
8890 | } | |
8891 | ||
cf27f3b1 PB |
8892 | /* |
8893 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8894 | */ | |
8895 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
8896 | { | |
8897 | struct io_uring_task *tctx = current->io_uring; | |
8898 | ||
8899 | if (likely(tctx && tctx->last == ctx)) | |
8900 | return 0; | |
8901 | return __io_uring_add_task_file(ctx); | |
8902 | } | |
8903 | ||
0f212204 JA |
8904 | /* |
8905 | * Remove this io_uring_file -> task mapping. | |
8906 | */ | |
2941267b | 8907 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8908 | { |
8909 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8910 | struct io_tctx_node *node; |
2941267b | 8911 | |
eebd2e37 PB |
8912 | if (!tctx) |
8913 | return; | |
13bf43f5 PB |
8914 | node = xa_erase(&tctx->xa, index); |
8915 | if (!node) | |
2941267b | 8916 | return; |
0f212204 | 8917 | |
13bf43f5 PB |
8918 | WARN_ON_ONCE(current != node->task); |
8919 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8920 | ||
8921 | mutex_lock(&node->ctx->uring_lock); | |
8922 | list_del(&node->ctx_node); | |
8923 | mutex_unlock(&node->ctx->uring_lock); | |
8924 | ||
baf186c4 | 8925 | if (tctx->last == node->ctx) |
0f212204 | 8926 | tctx->last = NULL; |
13bf43f5 | 8927 | kfree(node); |
0f212204 JA |
8928 | } |
8929 | ||
8452d4a6 | 8930 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8931 | { |
13bf43f5 | 8932 | struct io_tctx_node *node; |
de7f1d9e PB |
8933 | unsigned long index; |
8934 | ||
13bf43f5 | 8935 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8936 | io_uring_del_task_file(index); |
8452d4a6 PB |
8937 | if (tctx->io_wq) { |
8938 | io_wq_put_and_exit(tctx->io_wq); | |
8939 | tctx->io_wq = NULL; | |
8940 | } | |
de7f1d9e PB |
8941 | } |
8942 | ||
3f48cf18 | 8943 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 8944 | { |
3f48cf18 PB |
8945 | if (tracked) |
8946 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
8947 | return percpu_counter_sum(&tctx->inflight); |
8948 | } | |
8949 | ||
8950 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8951 | { | |
8952 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
734551df | 8953 | struct io_sq_data *sqd = work->ctx->sq_data; |
521d6a73 PB |
8954 | |
8955 | if (sqd->thread) | |
734551df PB |
8956 | io_uring_cancel_sqpoll(sqd); |
8957 | list_del_init(&work->ctx->sqd_list); | |
521d6a73 PB |
8958 | complete(&work->completion); |
8959 | } | |
8960 | ||
8961 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8962 | { | |
8963 | struct io_sq_data *sqd = ctx->sq_data; | |
8964 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8965 | struct task_struct *task; | |
8966 | ||
8967 | io_sq_thread_park(sqd); | |
521d6a73 PB |
8968 | io_sqd_update_thread_idle(sqd); |
8969 | task = sqd->thread; | |
8970 | if (task) { | |
8971 | init_completion(&work.completion); | |
8972 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8973 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 | 8974 | wake_up_process(task); |
734551df PB |
8975 | } else { |
8976 | list_del_init(&ctx->sqd_list); | |
521d6a73 PB |
8977 | } |
8978 | io_sq_thread_unpark(sqd); | |
8979 | ||
8980 | if (task) | |
8981 | wait_for_completion(&work.completion); | |
8982 | } | |
8983 | ||
368b2080 | 8984 | static void io_uring_try_cancel(struct files_struct *files) |
0f212204 JA |
8985 | { |
8986 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8987 | struct io_tctx_node *node; |
ce765372 | 8988 | unsigned long index; |
0f212204 | 8989 | |
521d6a73 PB |
8990 | xa_for_each(&tctx->xa, index, node) { |
8991 | struct io_ring_ctx *ctx = node->ctx; | |
8992 | ||
8993 | if (ctx->sq_data) { | |
8994 | io_sqpoll_cancel_sync(ctx); | |
8995 | continue; | |
8996 | } | |
368b2080 | 8997 | io_uring_try_cancel_requests(ctx, current, files); |
521d6a73 | 8998 | } |
fdaf083c JA |
8999 | } |
9000 | ||
521d6a73 | 9001 | /* should only be called by SQPOLL task */ |
734551df | 9002 | static void io_uring_cancel_sqpoll(struct io_sq_data *sqd) |
0e9ddb39 | 9003 | { |
521d6a73 | 9004 | struct io_uring_task *tctx = current->io_uring; |
734551df | 9005 | struct io_ring_ctx *ctx; |
0e9ddb39 PB |
9006 | s64 inflight; |
9007 | DEFINE_WAIT(wait); | |
fdaf083c | 9008 | |
734551df | 9009 | WARN_ON_ONCE(!sqd || sqd->thread != current); |
521d6a73 | 9010 | |
0e9ddb39 PB |
9011 | atomic_inc(&tctx->in_idle); |
9012 | do { | |
9013 | /* read completions before cancelations */ | |
3f48cf18 | 9014 | inflight = tctx_inflight(tctx, false); |
0e9ddb39 PB |
9015 | if (!inflight) |
9016 | break; | |
734551df PB |
9017 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
9018 | io_uring_try_cancel_requests(ctx, current, NULL); | |
fdaf083c | 9019 | |
0e9ddb39 PB |
9020 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9021 | /* | |
9022 | * If we've seen completions, retry without waiting. This | |
9023 | * avoids a race where a completion comes in before we did | |
9024 | * prepare_to_wait(). | |
9025 | */ | |
3f48cf18 | 9026 | if (inflight == tctx_inflight(tctx, false)) |
0e9ddb39 PB |
9027 | schedule(); |
9028 | finish_wait(&tctx->wait, &wait); | |
9029 | } while (1); | |
9030 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
9031 | } |
9032 | ||
0f212204 JA |
9033 | /* |
9034 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
9035 | * requests. | |
9036 | */ | |
3f48cf18 | 9037 | void __io_uring_cancel(struct files_struct *files) |
0f212204 JA |
9038 | { |
9039 | struct io_uring_task *tctx = current->io_uring; | |
9040 | DEFINE_WAIT(wait); | |
d8a6df10 | 9041 | s64 inflight; |
0f212204 JA |
9042 | |
9043 | /* make sure overflow events are dropped */ | |
fdaf083c | 9044 | atomic_inc(&tctx->in_idle); |
3f48cf18 | 9045 | io_uring_try_cancel(files); |
5a978dcf | 9046 | |
d8a6df10 | 9047 | do { |
0f212204 | 9048 | /* read completions before cancelations */ |
3f48cf18 | 9049 | inflight = tctx_inflight(tctx, !!files); |
d8a6df10 JA |
9050 | if (!inflight) |
9051 | break; | |
3f48cf18 | 9052 | io_uring_try_cancel(files); |
0f212204 JA |
9053 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9054 | ||
9055 | /* | |
a1bb3cd5 PB |
9056 | * If we've seen completions, retry without waiting. This |
9057 | * avoids a race where a completion comes in before we did | |
9058 | * prepare_to_wait(). | |
0f212204 | 9059 | */ |
3f48cf18 | 9060 | if (inflight == tctx_inflight(tctx, !!files)) |
a1bb3cd5 | 9061 | schedule(); |
f57555ed | 9062 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9063 | } while (1); |
fdaf083c | 9064 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9065 | |
8452d4a6 | 9066 | io_uring_clean_tctx(tctx); |
3f48cf18 PB |
9067 | if (!files) { |
9068 | /* for exec all current's requests should be gone, kill tctx */ | |
9069 | __io_uring_free(current); | |
9070 | } | |
44e728b8 PB |
9071 | } |
9072 | ||
6c5c240e RP |
9073 | static void *io_uring_validate_mmap_request(struct file *file, |
9074 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9075 | { |
2b188cc1 | 9076 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9077 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9078 | struct page *page; |
9079 | void *ptr; | |
9080 | ||
9081 | switch (offset) { | |
9082 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9083 | case IORING_OFF_CQ_RING: |
9084 | ptr = ctx->rings; | |
2b188cc1 JA |
9085 | break; |
9086 | case IORING_OFF_SQES: | |
9087 | ptr = ctx->sq_sqes; | |
9088 | break; | |
2b188cc1 | 9089 | default: |
6c5c240e | 9090 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9091 | } |
9092 | ||
9093 | page = virt_to_head_page(ptr); | |
a50b854e | 9094 | if (sz > page_size(page)) |
6c5c240e RP |
9095 | return ERR_PTR(-EINVAL); |
9096 | ||
9097 | return ptr; | |
9098 | } | |
9099 | ||
9100 | #ifdef CONFIG_MMU | |
9101 | ||
9102 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9103 | { | |
9104 | size_t sz = vma->vm_end - vma->vm_start; | |
9105 | unsigned long pfn; | |
9106 | void *ptr; | |
9107 | ||
9108 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9109 | if (IS_ERR(ptr)) | |
9110 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9111 | |
9112 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9113 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9114 | } | |
9115 | ||
6c5c240e RP |
9116 | #else /* !CONFIG_MMU */ |
9117 | ||
9118 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9119 | { | |
9120 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9121 | } | |
9122 | ||
9123 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9124 | { | |
9125 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9126 | } | |
9127 | ||
9128 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9129 | unsigned long addr, unsigned long len, | |
9130 | unsigned long pgoff, unsigned long flags) | |
9131 | { | |
9132 | void *ptr; | |
9133 | ||
9134 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9135 | if (IS_ERR(ptr)) | |
9136 | return PTR_ERR(ptr); | |
9137 | ||
9138 | return (unsigned long) ptr; | |
9139 | } | |
9140 | ||
9141 | #endif /* !CONFIG_MMU */ | |
9142 | ||
d9d05217 | 9143 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9144 | { |
9145 | DEFINE_WAIT(wait); | |
9146 | ||
9147 | do { | |
9148 | if (!io_sqring_full(ctx)) | |
9149 | break; | |
90554200 JA |
9150 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9151 | ||
9152 | if (!io_sqring_full(ctx)) | |
9153 | break; | |
90554200 JA |
9154 | schedule(); |
9155 | } while (!signal_pending(current)); | |
9156 | ||
9157 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9158 | return 0; |
90554200 JA |
9159 | } |
9160 | ||
c73ebb68 HX |
9161 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9162 | struct __kernel_timespec __user **ts, | |
9163 | const sigset_t __user **sig) | |
9164 | { | |
9165 | struct io_uring_getevents_arg arg; | |
9166 | ||
9167 | /* | |
9168 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9169 | * is just a pointer to the sigset_t. | |
9170 | */ | |
9171 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9172 | *sig = (const sigset_t __user *) argp; | |
9173 | *ts = NULL; | |
9174 | return 0; | |
9175 | } | |
9176 | ||
9177 | /* | |
9178 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9179 | * timespec and sigset_t pointers if good. | |
9180 | */ | |
9181 | if (*argsz != sizeof(arg)) | |
9182 | return -EINVAL; | |
9183 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9184 | return -EFAULT; | |
9185 | *sig = u64_to_user_ptr(arg.sigmask); | |
9186 | *argsz = arg.sigmask_sz; | |
9187 | *ts = u64_to_user_ptr(arg.ts); | |
9188 | return 0; | |
9189 | } | |
9190 | ||
2b188cc1 | 9191 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9192 | u32, min_complete, u32, flags, const void __user *, argp, |
9193 | size_t, argsz) | |
2b188cc1 JA |
9194 | { |
9195 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9196 | int submitted = 0; |
9197 | struct fd f; | |
33f993da | 9198 | long ret; |
2b188cc1 | 9199 | |
4c6e277c | 9200 | io_run_task_work(); |
b41e9852 | 9201 | |
33f993da PB |
9202 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9203 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9204 | return -EINVAL; |
9205 | ||
9206 | f = fdget(fd); | |
33f993da | 9207 | if (unlikely(!f.file)) |
2b188cc1 JA |
9208 | return -EBADF; |
9209 | ||
9210 | ret = -EOPNOTSUPP; | |
33f993da | 9211 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9212 | goto out_fput; |
9213 | ||
9214 | ret = -ENXIO; | |
9215 | ctx = f.file->private_data; | |
33f993da | 9216 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9217 | goto out_fput; |
9218 | ||
7e84e1c7 | 9219 | ret = -EBADFD; |
33f993da | 9220 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9221 | goto out; |
9222 | ||
6c271ce2 JA |
9223 | /* |
9224 | * For SQ polling, the thread will do all submissions and completions. | |
9225 | * Just return the requested submit count, and wake the thread if | |
9226 | * we were asked to. | |
9227 | */ | |
b2a9eada | 9228 | ret = 0; |
6c271ce2 | 9229 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9230 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9231 | |
d9d05217 | 9232 | ret = -EOWNERDEAD; |
04147488 SM |
9233 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9234 | goto out; | |
9235 | } | |
6c271ce2 | 9236 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9237 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9238 | if (flags & IORING_ENTER_SQ_WAIT) { |
9239 | ret = io_sqpoll_wait_sq(ctx); | |
9240 | if (ret) | |
9241 | goto out; | |
9242 | } | |
6c271ce2 | 9243 | submitted = to_submit; |
b2a9eada | 9244 | } else if (to_submit) { |
baf186c4 | 9245 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9246 | if (unlikely(ret)) |
9247 | goto out; | |
2b188cc1 | 9248 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9249 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9250 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9251 | |
9252 | if (submitted != to_submit) | |
9253 | goto out; | |
2b188cc1 JA |
9254 | } |
9255 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9256 | const sigset_t __user *sig; |
9257 | struct __kernel_timespec __user *ts; | |
9258 | ||
9259 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9260 | if (unlikely(ret)) | |
9261 | goto out; | |
9262 | ||
2b188cc1 JA |
9263 | min_complete = min(min_complete, ctx->cq_entries); |
9264 | ||
32b2244a XW |
9265 | /* |
9266 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9267 | * space applications don't need to do io completion events | |
9268 | * polling again, they can rely on io_sq_thread to do polling | |
9269 | * work, which can reduce cpu usage and uring_lock contention. | |
9270 | */ | |
9271 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9272 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9273 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9274 | } else { |
c73ebb68 | 9275 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9276 | } |
2b188cc1 JA |
9277 | } |
9278 | ||
7c504e65 | 9279 | out: |
6805b32e | 9280 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9281 | out_fput: |
9282 | fdput(f); | |
9283 | return submitted ? submitted : ret; | |
9284 | } | |
9285 | ||
bebdb65e | 9286 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9287 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9288 | const struct cred *cred) | |
87ce955b | 9289 | { |
87ce955b JA |
9290 | struct user_namespace *uns = seq_user_ns(m); |
9291 | struct group_info *gi; | |
9292 | kernel_cap_t cap; | |
9293 | unsigned __capi; | |
9294 | int g; | |
9295 | ||
9296 | seq_printf(m, "%5d\n", id); | |
9297 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9298 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9299 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9300 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9301 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9302 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9303 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9304 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9305 | seq_puts(m, "\n\tGroups:\t"); | |
9306 | gi = cred->group_info; | |
9307 | for (g = 0; g < gi->ngroups; g++) { | |
9308 | seq_put_decimal_ull(m, g ? " " : "", | |
9309 | from_kgid_munged(uns, gi->gid[g])); | |
9310 | } | |
9311 | seq_puts(m, "\n\tCapEff:\t"); | |
9312 | cap = cred->cap_effective; | |
9313 | CAP_FOR_EACH_U32(__capi) | |
9314 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9315 | seq_putc(m, '\n'); | |
9316 | return 0; | |
9317 | } | |
9318 | ||
9319 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9320 | { | |
dbbe9c64 | 9321 | struct io_sq_data *sq = NULL; |
fad8e0de | 9322 | bool has_lock; |
87ce955b JA |
9323 | int i; |
9324 | ||
fad8e0de JA |
9325 | /* |
9326 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9327 | * since fdinfo case grabs it in the opposite direction of normal use | |
9328 | * cases. If we fail to get the lock, we just don't iterate any | |
9329 | * structures that could be going away outside the io_uring mutex. | |
9330 | */ | |
9331 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9332 | ||
5f3f26f9 | 9333 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9334 | sq = ctx->sq_data; |
5f3f26f9 JA |
9335 | if (!sq->thread) |
9336 | sq = NULL; | |
9337 | } | |
dbbe9c64 JQ |
9338 | |
9339 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9340 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9341 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9342 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9343 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9344 | |
87ce955b JA |
9345 | if (f) |
9346 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9347 | else | |
9348 | seq_printf(m, "%5u: <none>\n", i); | |
9349 | } | |
9350 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9351 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
41edf1a5 | 9352 | struct io_mapped_ubuf *buf = ctx->user_bufs[i]; |
4751f53d | 9353 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9354 | |
4751f53d | 9355 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9356 | } |
61cf9370 MWO |
9357 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9358 | unsigned long index; | |
9359 | const struct cred *cred; | |
9360 | ||
87ce955b | 9361 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9362 | xa_for_each(&ctx->personalities, index, cred) |
9363 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9364 | } |
d7718a9d JA |
9365 | seq_printf(m, "PollList:\n"); |
9366 | spin_lock_irq(&ctx->completion_lock); | |
9367 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9368 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9369 | struct io_kiocb *req; | |
9370 | ||
9371 | hlist_for_each_entry(req, list, hash_node) | |
9372 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9373 | req->task->task_works != NULL); | |
9374 | } | |
9375 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9376 | if (has_lock) |
9377 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9378 | } |
9379 | ||
9380 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9381 | { | |
9382 | struct io_ring_ctx *ctx = f->private_data; | |
9383 | ||
9384 | if (percpu_ref_tryget(&ctx->refs)) { | |
9385 | __io_uring_show_fdinfo(ctx, m); | |
9386 | percpu_ref_put(&ctx->refs); | |
9387 | } | |
9388 | } | |
bebdb65e | 9389 | #endif |
87ce955b | 9390 | |
2b188cc1 JA |
9391 | static const struct file_operations io_uring_fops = { |
9392 | .release = io_uring_release, | |
9393 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9394 | #ifndef CONFIG_MMU |
9395 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9396 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9397 | #endif | |
2b188cc1 JA |
9398 | .poll = io_uring_poll, |
9399 | .fasync = io_uring_fasync, | |
bebdb65e | 9400 | #ifdef CONFIG_PROC_FS |
87ce955b | 9401 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9402 | #endif |
2b188cc1 JA |
9403 | }; |
9404 | ||
9405 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9406 | struct io_uring_params *p) | |
9407 | { | |
75b28aff HV |
9408 | struct io_rings *rings; |
9409 | size_t size, sq_array_offset; | |
2b188cc1 | 9410 | |
bd740481 JA |
9411 | /* make sure these are sane, as we already accounted them */ |
9412 | ctx->sq_entries = p->sq_entries; | |
9413 | ctx->cq_entries = p->cq_entries; | |
9414 | ||
75b28aff HV |
9415 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9416 | if (size == SIZE_MAX) | |
9417 | return -EOVERFLOW; | |
9418 | ||
9419 | rings = io_mem_alloc(size); | |
9420 | if (!rings) | |
2b188cc1 JA |
9421 | return -ENOMEM; |
9422 | ||
75b28aff HV |
9423 | ctx->rings = rings; |
9424 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9425 | rings->sq_ring_mask = p->sq_entries - 1; | |
9426 | rings->cq_ring_mask = p->cq_entries - 1; | |
9427 | rings->sq_ring_entries = p->sq_entries; | |
9428 | rings->cq_ring_entries = p->cq_entries; | |
9429 | ctx->sq_mask = rings->sq_ring_mask; | |
9430 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9431 | |
9432 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9433 | if (size == SIZE_MAX) { |
9434 | io_mem_free(ctx->rings); | |
9435 | ctx->rings = NULL; | |
2b188cc1 | 9436 | return -EOVERFLOW; |
eb065d30 | 9437 | } |
2b188cc1 JA |
9438 | |
9439 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9440 | if (!ctx->sq_sqes) { |
9441 | io_mem_free(ctx->rings); | |
9442 | ctx->rings = NULL; | |
2b188cc1 | 9443 | return -ENOMEM; |
eb065d30 | 9444 | } |
2b188cc1 | 9445 | |
2b188cc1 JA |
9446 | return 0; |
9447 | } | |
9448 | ||
9faadcc8 PB |
9449 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9450 | { | |
9451 | int ret, fd; | |
9452 | ||
9453 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9454 | if (fd < 0) | |
9455 | return fd; | |
9456 | ||
baf186c4 | 9457 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9458 | if (ret) { |
9459 | put_unused_fd(fd); | |
9460 | return ret; | |
9461 | } | |
9462 | fd_install(fd, file); | |
9463 | return fd; | |
9464 | } | |
9465 | ||
2b188cc1 JA |
9466 | /* |
9467 | * Allocate an anonymous fd, this is what constitutes the application | |
9468 | * visible backing of an io_uring instance. The application mmaps this | |
9469 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9470 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9471 | */ | |
9faadcc8 | 9472 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9473 | { |
9474 | struct file *file; | |
9faadcc8 | 9475 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9476 | int ret; |
9477 | ||
2b188cc1 JA |
9478 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9479 | &ctx->ring_sock); | |
9480 | if (ret) | |
9faadcc8 | 9481 | return ERR_PTR(ret); |
2b188cc1 JA |
9482 | #endif |
9483 | ||
2b188cc1 JA |
9484 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9485 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9486 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9487 | if (IS_ERR(file)) { |
9488 | sock_release(ctx->ring_sock); | |
9489 | ctx->ring_sock = NULL; | |
9490 | } else { | |
9491 | ctx->ring_sock->file = file; | |
0f212204 | 9492 | } |
2b188cc1 | 9493 | #endif |
9faadcc8 | 9494 | return file; |
2b188cc1 JA |
9495 | } |
9496 | ||
7f13657d XW |
9497 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9498 | struct io_uring_params __user *params) | |
2b188cc1 | 9499 | { |
2b188cc1 | 9500 | struct io_ring_ctx *ctx; |
9faadcc8 | 9501 | struct file *file; |
2b188cc1 JA |
9502 | int ret; |
9503 | ||
8110c1a6 | 9504 | if (!entries) |
2b188cc1 | 9505 | return -EINVAL; |
8110c1a6 JA |
9506 | if (entries > IORING_MAX_ENTRIES) { |
9507 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9508 | return -EINVAL; | |
9509 | entries = IORING_MAX_ENTRIES; | |
9510 | } | |
2b188cc1 JA |
9511 | |
9512 | /* | |
9513 | * Use twice as many entries for the CQ ring. It's possible for the | |
9514 | * application to drive a higher depth than the size of the SQ ring, | |
9515 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9516 | * some flexibility in overcommitting a bit. If the application has |
9517 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9518 | * of CQ ring entries manually. | |
2b188cc1 JA |
9519 | */ |
9520 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9521 | if (p->flags & IORING_SETUP_CQSIZE) { |
9522 | /* | |
9523 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9524 | * to a power-of-two, if it isn't already. We do NOT impose | |
9525 | * any cq vs sq ring sizing. | |
9526 | */ | |
eb2667b3 | 9527 | if (!p->cq_entries) |
33a107f0 | 9528 | return -EINVAL; |
8110c1a6 JA |
9529 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9530 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9531 | return -EINVAL; | |
9532 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9533 | } | |
eb2667b3 JQ |
9534 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9535 | if (p->cq_entries < p->sq_entries) | |
9536 | return -EINVAL; | |
33a107f0 JA |
9537 | } else { |
9538 | p->cq_entries = 2 * p->sq_entries; | |
9539 | } | |
2b188cc1 | 9540 | |
2b188cc1 | 9541 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9542 | if (!ctx) |
2b188cc1 | 9543 | return -ENOMEM; |
2b188cc1 | 9544 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9545 | if (!capable(CAP_IPC_LOCK)) |
9546 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9547 | |
9548 | /* | |
9549 | * This is just grabbed for accounting purposes. When a process exits, | |
9550 | * the mm is exited and dropped before the files, hence we need to hang | |
9551 | * on to this mm purely for the purposes of being able to unaccount | |
9552 | * memory (locked/pinned vm). It's not used for anything else. | |
9553 | */ | |
6b7898eb | 9554 | mmgrab(current->mm); |
2aede0e4 | 9555 | ctx->mm_account = current->mm; |
6b7898eb | 9556 | |
2b188cc1 JA |
9557 | ret = io_allocate_scq_urings(ctx, p); |
9558 | if (ret) | |
9559 | goto err; | |
9560 | ||
7e84e1c7 | 9561 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9562 | if (ret) |
9563 | goto err; | |
eae071c9 PB |
9564 | /* always set a rsrc node */ |
9565 | io_rsrc_node_switch_start(ctx); | |
9566 | io_rsrc_node_switch(ctx, NULL); | |
2b188cc1 | 9567 | |
2b188cc1 | 9568 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9569 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9570 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9571 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9572 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9573 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9574 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9575 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9576 | |
9577 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9578 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9579 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9580 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9581 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9582 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9583 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9584 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9585 | |
7f13657d XW |
9586 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9587 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9588 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9589 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9590 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9591 | |
9592 | if (copy_to_user(params, p, sizeof(*p))) { | |
9593 | ret = -EFAULT; | |
9594 | goto err; | |
9595 | } | |
d1719f70 | 9596 | |
9faadcc8 PB |
9597 | file = io_uring_get_file(ctx); |
9598 | if (IS_ERR(file)) { | |
9599 | ret = PTR_ERR(file); | |
9600 | goto err; | |
9601 | } | |
9602 | ||
044c1ab3 JA |
9603 | /* |
9604 | * Install ring fd as the very last thing, so we don't risk someone | |
9605 | * having closed it before we finish setup | |
9606 | */ | |
9faadcc8 PB |
9607 | ret = io_uring_install_fd(ctx, file); |
9608 | if (ret < 0) { | |
9609 | /* fput will clean it up */ | |
9610 | fput(file); | |
9611 | return ret; | |
9612 | } | |
044c1ab3 | 9613 | |
c826bd7a | 9614 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9615 | return ret; |
9616 | err: | |
9617 | io_ring_ctx_wait_and_kill(ctx); | |
9618 | return ret; | |
9619 | } | |
9620 | ||
9621 | /* | |
9622 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9623 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9624 | * params structure passed in. | |
9625 | */ | |
9626 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9627 | { | |
9628 | struct io_uring_params p; | |
2b188cc1 JA |
9629 | int i; |
9630 | ||
9631 | if (copy_from_user(&p, params, sizeof(p))) | |
9632 | return -EFAULT; | |
9633 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9634 | if (p.resv[i]) | |
9635 | return -EINVAL; | |
9636 | } | |
9637 | ||
6c271ce2 | 9638 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9639 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9640 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9641 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9642 | return -EINVAL; |
9643 | ||
7f13657d | 9644 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9645 | } |
9646 | ||
9647 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9648 | struct io_uring_params __user *, params) | |
9649 | { | |
9650 | return io_uring_setup(entries, params); | |
9651 | } | |
9652 | ||
66f4af93 JA |
9653 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9654 | { | |
9655 | struct io_uring_probe *p; | |
9656 | size_t size; | |
9657 | int i, ret; | |
9658 | ||
9659 | size = struct_size(p, ops, nr_args); | |
9660 | if (size == SIZE_MAX) | |
9661 | return -EOVERFLOW; | |
9662 | p = kzalloc(size, GFP_KERNEL); | |
9663 | if (!p) | |
9664 | return -ENOMEM; | |
9665 | ||
9666 | ret = -EFAULT; | |
9667 | if (copy_from_user(p, arg, size)) | |
9668 | goto out; | |
9669 | ret = -EINVAL; | |
9670 | if (memchr_inv(p, 0, size)) | |
9671 | goto out; | |
9672 | ||
9673 | p->last_op = IORING_OP_LAST - 1; | |
9674 | if (nr_args > IORING_OP_LAST) | |
9675 | nr_args = IORING_OP_LAST; | |
9676 | ||
9677 | for (i = 0; i < nr_args; i++) { | |
9678 | p->ops[i].op = i; | |
9679 | if (!io_op_defs[i].not_supported) | |
9680 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9681 | } | |
9682 | p->ops_len = i; | |
9683 | ||
9684 | ret = 0; | |
9685 | if (copy_to_user(arg, p, size)) | |
9686 | ret = -EFAULT; | |
9687 | out: | |
9688 | kfree(p); | |
9689 | return ret; | |
9690 | } | |
9691 | ||
071698e1 JA |
9692 | static int io_register_personality(struct io_ring_ctx *ctx) |
9693 | { | |
4379bf8b | 9694 | const struct cred *creds; |
61cf9370 | 9695 | u32 id; |
1e6fa521 | 9696 | int ret; |
071698e1 | 9697 | |
4379bf8b | 9698 | creds = get_current_cred(); |
1e6fa521 | 9699 | |
61cf9370 MWO |
9700 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9701 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9702 | if (!ret) | |
9703 | return id; | |
9704 | put_cred(creds); | |
1e6fa521 | 9705 | return ret; |
071698e1 JA |
9706 | } |
9707 | ||
21b55dbc SG |
9708 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9709 | unsigned int nr_args) | |
9710 | { | |
9711 | struct io_uring_restriction *res; | |
9712 | size_t size; | |
9713 | int i, ret; | |
9714 | ||
7e84e1c7 SG |
9715 | /* Restrictions allowed only if rings started disabled */ |
9716 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9717 | return -EBADFD; | |
9718 | ||
21b55dbc | 9719 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9720 | if (ctx->restrictions.registered) |
21b55dbc SG |
9721 | return -EBUSY; |
9722 | ||
9723 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9724 | return -EINVAL; | |
9725 | ||
9726 | size = array_size(nr_args, sizeof(*res)); | |
9727 | if (size == SIZE_MAX) | |
9728 | return -EOVERFLOW; | |
9729 | ||
9730 | res = memdup_user(arg, size); | |
9731 | if (IS_ERR(res)) | |
9732 | return PTR_ERR(res); | |
9733 | ||
9734 | ret = 0; | |
9735 | ||
9736 | for (i = 0; i < nr_args; i++) { | |
9737 | switch (res[i].opcode) { | |
9738 | case IORING_RESTRICTION_REGISTER_OP: | |
9739 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9740 | ret = -EINVAL; | |
9741 | goto out; | |
9742 | } | |
9743 | ||
9744 | __set_bit(res[i].register_op, | |
9745 | ctx->restrictions.register_op); | |
9746 | break; | |
9747 | case IORING_RESTRICTION_SQE_OP: | |
9748 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9749 | ret = -EINVAL; | |
9750 | goto out; | |
9751 | } | |
9752 | ||
9753 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9754 | break; | |
9755 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9756 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9757 | break; | |
9758 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9759 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9760 | break; | |
9761 | default: | |
9762 | ret = -EINVAL; | |
9763 | goto out; | |
9764 | } | |
9765 | } | |
9766 | ||
9767 | out: | |
9768 | /* Reset all restrictions if an error happened */ | |
9769 | if (ret != 0) | |
9770 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9771 | else | |
7e84e1c7 | 9772 | ctx->restrictions.registered = true; |
21b55dbc SG |
9773 | |
9774 | kfree(res); | |
9775 | return ret; | |
9776 | } | |
9777 | ||
7e84e1c7 SG |
9778 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9779 | { | |
9780 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9781 | return -EBADFD; | |
9782 | ||
9783 | if (ctx->restrictions.registered) | |
9784 | ctx->restricted = 1; | |
9785 | ||
0298ef96 PB |
9786 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9787 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9788 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9789 | return 0; |
9790 | } | |
9791 | ||
fdecb662 | 9792 | static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type, |
c3bdad02 | 9793 | struct io_uring_rsrc_update2 *up, |
98f0b3b4 PB |
9794 | unsigned nr_args) |
9795 | { | |
9796 | __u32 tmp; | |
9797 | int err; | |
9798 | ||
c3bdad02 PB |
9799 | if (up->resv) |
9800 | return -EINVAL; | |
98f0b3b4 PB |
9801 | if (check_add_overflow(up->offset, nr_args, &tmp)) |
9802 | return -EOVERFLOW; | |
9803 | err = io_rsrc_node_switch_start(ctx); | |
9804 | if (err) | |
9805 | return err; | |
9806 | ||
fdecb662 PB |
9807 | switch (type) { |
9808 | case IORING_RSRC_FILE: | |
98f0b3b4 PB |
9809 | return __io_sqe_files_update(ctx, up, nr_args); |
9810 | } | |
9811 | return -EINVAL; | |
9812 | } | |
9813 | ||
c3bdad02 PB |
9814 | static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg, |
9815 | unsigned nr_args) | |
98f0b3b4 | 9816 | { |
c3bdad02 | 9817 | struct io_uring_rsrc_update2 up; |
98f0b3b4 PB |
9818 | |
9819 | if (!nr_args) | |
9820 | return -EINVAL; | |
c3bdad02 PB |
9821 | memset(&up, 0, sizeof(up)); |
9822 | if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update))) | |
9823 | return -EFAULT; | |
9824 | return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args); | |
9825 | } | |
9826 | ||
9827 | static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg, | |
9828 | unsigned size) | |
9829 | { | |
9830 | struct io_uring_rsrc_update2 up; | |
9831 | ||
9832 | if (size != sizeof(up)) | |
9833 | return -EINVAL; | |
98f0b3b4 PB |
9834 | if (copy_from_user(&up, arg, sizeof(up))) |
9835 | return -EFAULT; | |
c3bdad02 | 9836 | if (!up.nr) |
98f0b3b4 | 9837 | return -EINVAL; |
c3bdad02 | 9838 | return __io_register_rsrc_update(ctx, up.type, &up, up.nr); |
98f0b3b4 PB |
9839 | } |
9840 | ||
792e3582 PB |
9841 | static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, |
9842 | unsigned int size) | |
9843 | { | |
9844 | struct io_uring_rsrc_register rr; | |
9845 | ||
9846 | /* keep it extendible */ | |
9847 | if (size != sizeof(rr)) | |
9848 | return -EINVAL; | |
9849 | ||
9850 | memset(&rr, 0, sizeof(rr)); | |
9851 | if (copy_from_user(&rr, arg, size)) | |
9852 | return -EFAULT; | |
9853 | if (!rr.nr) | |
9854 | return -EINVAL; | |
9855 | ||
9856 | switch (rr.type) { | |
9857 | case IORING_RSRC_FILE: | |
9858 | return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data), | |
9859 | rr.nr, u64_to_user_ptr(rr.tags)); | |
9860 | } | |
9861 | return -EINVAL; | |
9862 | } | |
9863 | ||
071698e1 JA |
9864 | static bool io_register_op_must_quiesce(int op) |
9865 | { | |
9866 | switch (op) { | |
bd54b6fe BM |
9867 | case IORING_REGISTER_BUFFERS: |
9868 | case IORING_UNREGISTER_BUFFERS: | |
f4f7d21c | 9869 | case IORING_REGISTER_FILES: |
071698e1 JA |
9870 | case IORING_UNREGISTER_FILES: |
9871 | case IORING_REGISTER_FILES_UPDATE: | |
9872 | case IORING_REGISTER_PROBE: | |
9873 | case IORING_REGISTER_PERSONALITY: | |
9874 | case IORING_UNREGISTER_PERSONALITY: | |
792e3582 | 9875 | case IORING_REGISTER_RSRC: |
c3bdad02 | 9876 | case IORING_REGISTER_RSRC_UPDATE: |
071698e1 JA |
9877 | return false; |
9878 | default: | |
9879 | return true; | |
9880 | } | |
9881 | } | |
9882 | ||
edafccee JA |
9883 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9884 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9885 | __releases(ctx->uring_lock) |
9886 | __acquires(ctx->uring_lock) | |
edafccee JA |
9887 | { |
9888 | int ret; | |
9889 | ||
35fa71a0 JA |
9890 | /* |
9891 | * We're inside the ring mutex, if the ref is already dying, then | |
9892 | * someone else killed the ctx or is already going through | |
9893 | * io_uring_register(). | |
9894 | */ | |
9895 | if (percpu_ref_is_dying(&ctx->refs)) | |
9896 | return -ENXIO; | |
9897 | ||
75c4021a PB |
9898 | if (ctx->restricted) { |
9899 | if (opcode >= IORING_REGISTER_LAST) | |
9900 | return -EINVAL; | |
9901 | opcode = array_index_nospec(opcode, IORING_REGISTER_LAST); | |
9902 | if (!test_bit(opcode, ctx->restrictions.register_op)) | |
9903 | return -EACCES; | |
9904 | } | |
9905 | ||
071698e1 | 9906 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9907 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9908 | |
05f3fb3c JA |
9909 | /* |
9910 | * Drop uring mutex before waiting for references to exit. If | |
9911 | * another thread is currently inside io_uring_enter() it might | |
9912 | * need to grab the uring_lock to make progress. If we hold it | |
9913 | * here across the drain wait, then we can deadlock. It's safe | |
9914 | * to drop the mutex here, since no new references will come in | |
9915 | * after we've killed the percpu ref. | |
9916 | */ | |
9917 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9918 | do { |
9919 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9920 | if (!ret) | |
9921 | break; | |
ed6930c9 JA |
9922 | ret = io_run_task_work_sig(); |
9923 | if (ret < 0) | |
9924 | break; | |
af9c1a44 | 9925 | } while (1); |
05f3fb3c | 9926 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9927 | |
c150368b | 9928 | if (ret) { |
f70865db PB |
9929 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
9930 | return ret; | |
21b55dbc SG |
9931 | } |
9932 | } | |
9933 | ||
edafccee JA |
9934 | switch (opcode) { |
9935 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9936 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9937 | break; |
9938 | case IORING_UNREGISTER_BUFFERS: | |
9939 | ret = -EINVAL; | |
9940 | if (arg || nr_args) | |
9941 | break; | |
0a96bbe4 | 9942 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9943 | break; |
6b06314c | 9944 | case IORING_REGISTER_FILES: |
792e3582 | 9945 | ret = io_sqe_files_register(ctx, arg, nr_args, NULL); |
6b06314c JA |
9946 | break; |
9947 | case IORING_UNREGISTER_FILES: | |
9948 | ret = -EINVAL; | |
9949 | if (arg || nr_args) | |
9950 | break; | |
9951 | ret = io_sqe_files_unregister(ctx); | |
9952 | break; | |
c3a31e60 | 9953 | case IORING_REGISTER_FILES_UPDATE: |
c3bdad02 | 9954 | ret = io_register_files_update(ctx, arg, nr_args); |
c3a31e60 | 9955 | break; |
9b402849 | 9956 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9957 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9958 | ret = -EINVAL; |
9959 | if (nr_args != 1) | |
9960 | break; | |
9961 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9962 | if (ret) |
9963 | break; | |
9964 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9965 | ctx->eventfd_async = 1; | |
9966 | else | |
9967 | ctx->eventfd_async = 0; | |
9b402849 JA |
9968 | break; |
9969 | case IORING_UNREGISTER_EVENTFD: | |
9970 | ret = -EINVAL; | |
9971 | if (arg || nr_args) | |
9972 | break; | |
9973 | ret = io_eventfd_unregister(ctx); | |
9974 | break; | |
66f4af93 JA |
9975 | case IORING_REGISTER_PROBE: |
9976 | ret = -EINVAL; | |
9977 | if (!arg || nr_args > 256) | |
9978 | break; | |
9979 | ret = io_probe(ctx, arg, nr_args); | |
9980 | break; | |
071698e1 JA |
9981 | case IORING_REGISTER_PERSONALITY: |
9982 | ret = -EINVAL; | |
9983 | if (arg || nr_args) | |
9984 | break; | |
9985 | ret = io_register_personality(ctx); | |
9986 | break; | |
9987 | case IORING_UNREGISTER_PERSONALITY: | |
9988 | ret = -EINVAL; | |
9989 | if (arg) | |
9990 | break; | |
9991 | ret = io_unregister_personality(ctx, nr_args); | |
9992 | break; | |
7e84e1c7 SG |
9993 | case IORING_REGISTER_ENABLE_RINGS: |
9994 | ret = -EINVAL; | |
9995 | if (arg || nr_args) | |
9996 | break; | |
9997 | ret = io_register_enable_rings(ctx); | |
9998 | break; | |
21b55dbc SG |
9999 | case IORING_REGISTER_RESTRICTIONS: |
10000 | ret = io_register_restrictions(ctx, arg, nr_args); | |
10001 | break; | |
792e3582 PB |
10002 | case IORING_REGISTER_RSRC: |
10003 | ret = io_register_rsrc(ctx, arg, nr_args); | |
10004 | break; | |
c3bdad02 PB |
10005 | case IORING_REGISTER_RSRC_UPDATE: |
10006 | ret = io_register_rsrc_update(ctx, arg, nr_args); | |
10007 | break; | |
edafccee JA |
10008 | default: |
10009 | ret = -EINVAL; | |
10010 | break; | |
10011 | } | |
10012 | ||
071698e1 | 10013 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 10014 | /* bring the ctx back to life */ |
05f3fb3c | 10015 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 10016 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 10017 | } |
edafccee JA |
10018 | return ret; |
10019 | } | |
10020 | ||
10021 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
10022 | void __user *, arg, unsigned int, nr_args) | |
10023 | { | |
10024 | struct io_ring_ctx *ctx; | |
10025 | long ret = -EBADF; | |
10026 | struct fd f; | |
10027 | ||
10028 | f = fdget(fd); | |
10029 | if (!f.file) | |
10030 | return -EBADF; | |
10031 | ||
10032 | ret = -EOPNOTSUPP; | |
10033 | if (f.file->f_op != &io_uring_fops) | |
10034 | goto out_fput; | |
10035 | ||
10036 | ctx = f.file->private_data; | |
10037 | ||
b6c23dd5 PB |
10038 | io_run_task_work(); |
10039 | ||
edafccee JA |
10040 | mutex_lock(&ctx->uring_lock); |
10041 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
10042 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
10043 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
10044 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
10045 | out_fput: |
10046 | fdput(f); | |
10047 | return ret; | |
10048 | } | |
10049 | ||
2b188cc1 JA |
10050 | static int __init io_uring_init(void) |
10051 | { | |
d7f62e82 SM |
10052 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
10053 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
10054 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
10055 | } while (0) | |
10056 | ||
10057 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
10058 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
10059 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
10060 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
10061 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
10062 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
10063 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
10064 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
10065 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
10066 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 10067 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
10068 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
10069 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
10070 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
10071 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
10072 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
10073 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
10074 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
10075 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
10076 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
10077 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
10078 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
10079 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
10080 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
10081 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
10082 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 10083 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
10084 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
10085 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
10086 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 10087 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 10088 | |
d3656344 | 10089 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 10090 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
10091 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
10092 | SLAB_ACCOUNT); | |
2b188cc1 JA |
10093 | return 0; |
10094 | }; | |
10095 | __initcall(io_uring_init); |