]>
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 JA |
198 | struct bio_vec *bvec; |
199 | unsigned int nr_bvecs; | |
de293938 | 200 | unsigned long acct_pages; |
edafccee JA |
201 | }; |
202 | ||
50238531 BM |
203 | struct io_ring_ctx; |
204 | ||
6c2450ae PB |
205 | struct io_overflow_cqe { |
206 | struct io_uring_cqe cqe; | |
207 | struct list_head list; | |
208 | }; | |
209 | ||
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; | |
50238531 BM |
217 | union { |
218 | void *rsrc; | |
219 | struct file *file; | |
220 | }; | |
269bbe5f BM |
221 | }; |
222 | ||
aeca241b PB |
223 | struct io_file_table { |
224 | /* two level table */ | |
225 | struct io_fixed_file **files; | |
31b51510 JA |
226 | }; |
227 | ||
b895c9a6 | 228 | struct io_rsrc_node { |
05589553 XW |
229 | struct percpu_ref refs; |
230 | struct list_head node; | |
269bbe5f | 231 | struct list_head rsrc_list; |
b895c9a6 | 232 | struct io_rsrc_data *rsrc_data; |
4a38aed2 | 233 | struct llist_node llist; |
e297822b | 234 | bool done; |
05589553 XW |
235 | }; |
236 | ||
40ae0ff7 PB |
237 | typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc); |
238 | ||
b895c9a6 | 239 | struct io_rsrc_data { |
05f3fb3c JA |
240 | struct io_ring_ctx *ctx; |
241 | ||
40ae0ff7 | 242 | rsrc_put_fn *do_put; |
3e942498 | 243 | atomic_t refs; |
05f3fb3c | 244 | struct completion done; |
8bad28d8 | 245 | bool quiesce; |
05f3fb3c JA |
246 | }; |
247 | ||
5a2e745d JA |
248 | struct io_buffer { |
249 | struct list_head list; | |
250 | __u64 addr; | |
251 | __s32 len; | |
252 | __u16 bid; | |
253 | }; | |
254 | ||
21b55dbc SG |
255 | struct io_restriction { |
256 | DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); | |
257 | DECLARE_BITMAP(sqe_op, IORING_OP_LAST); | |
258 | u8 sqe_flags_allowed; | |
259 | u8 sqe_flags_required; | |
7e84e1c7 | 260 | bool registered; |
21b55dbc SG |
261 | }; |
262 | ||
37d1e2e3 JA |
263 | enum { |
264 | IO_SQ_THREAD_SHOULD_STOP = 0, | |
265 | IO_SQ_THREAD_SHOULD_PARK, | |
266 | }; | |
267 | ||
534ca6d6 JA |
268 | struct io_sq_data { |
269 | refcount_t refs; | |
9e138a48 | 270 | atomic_t park_pending; |
09a6f4ef | 271 | struct mutex lock; |
69fb2131 JA |
272 | |
273 | /* ctx's that are using this sqd */ | |
274 | struct list_head ctx_list; | |
69fb2131 | 275 | |
534ca6d6 JA |
276 | struct task_struct *thread; |
277 | struct wait_queue_head wait; | |
08369246 XW |
278 | |
279 | unsigned sq_thread_idle; | |
37d1e2e3 JA |
280 | int sq_cpu; |
281 | pid_t task_pid; | |
5c2469e0 | 282 | pid_t task_tgid; |
37d1e2e3 JA |
283 | |
284 | unsigned long state; | |
37d1e2e3 | 285 | struct completion exited; |
b7f5a0bf | 286 | struct callback_head *park_task_work; |
534ca6d6 JA |
287 | }; |
288 | ||
258b29a9 | 289 | #define IO_IOPOLL_BATCH 8 |
6dd0be1e | 290 | #define IO_COMPL_BATCH 32 |
6ff119a6 | 291 | #define IO_REQ_CACHE_SIZE 32 |
bf019da7 | 292 | #define IO_REQ_ALLOC_BATCH 8 |
258b29a9 PB |
293 | |
294 | struct io_comp_state { | |
6dd0be1e | 295 | struct io_kiocb *reqs[IO_COMPL_BATCH]; |
1b4c351f | 296 | unsigned int nr; |
c7dae4ba JA |
297 | unsigned int locked_free_nr; |
298 | /* inline/task_work completion list, under ->uring_lock */ | |
1b4c351f | 299 | struct list_head free_list; |
c7dae4ba JA |
300 | /* IRQ completion list, under ->completion_lock */ |
301 | struct list_head locked_free_list; | |
258b29a9 PB |
302 | }; |
303 | ||
a1ab7b35 PB |
304 | struct io_submit_link { |
305 | struct io_kiocb *head; | |
306 | struct io_kiocb *last; | |
307 | }; | |
308 | ||
258b29a9 PB |
309 | struct io_submit_state { |
310 | struct blk_plug plug; | |
a1ab7b35 | 311 | struct io_submit_link link; |
258b29a9 PB |
312 | |
313 | /* | |
314 | * io_kiocb alloc cache | |
315 | */ | |
bf019da7 | 316 | void *reqs[IO_REQ_CACHE_SIZE]; |
258b29a9 PB |
317 | unsigned int free_reqs; |
318 | ||
319 | bool plug_started; | |
320 | ||
321 | /* | |
322 | * Batch completion logic | |
323 | */ | |
324 | struct io_comp_state comp; | |
325 | ||
326 | /* | |
327 | * File reference cache | |
328 | */ | |
329 | struct file *file; | |
330 | unsigned int fd; | |
331 | unsigned int file_refs; | |
332 | unsigned int ios_left; | |
333 | }; | |
334 | ||
2b188cc1 JA |
335 | struct io_ring_ctx { |
336 | struct { | |
337 | struct percpu_ref refs; | |
338 | } ____cacheline_aligned_in_smp; | |
339 | ||
340 | struct { | |
341 | unsigned int flags; | |
e1d85334 | 342 | unsigned int compat: 1; |
e1d85334 RD |
343 | unsigned int drain_next: 1; |
344 | unsigned int eventfd_async: 1; | |
21b55dbc | 345 | unsigned int restricted: 1; |
2b188cc1 | 346 | |
75b28aff HV |
347 | /* |
348 | * Ring buffer of indices into array of io_uring_sqe, which is | |
349 | * mmapped by the application using the IORING_OFF_SQES offset. | |
350 | * | |
351 | * This indirection could e.g. be used to assign fixed | |
352 | * io_uring_sqe entries to operations and only submit them to | |
353 | * the queue when needed. | |
354 | * | |
355 | * The kernel modifies neither the indices array nor the entries | |
356 | * array. | |
357 | */ | |
358 | u32 *sq_array; | |
2b188cc1 JA |
359 | unsigned cached_sq_head; |
360 | unsigned sq_entries; | |
361 | unsigned sq_mask; | |
6c271ce2 | 362 | unsigned sq_thread_idle; |
498ccd9e | 363 | unsigned cached_sq_dropped; |
2c3bac6d | 364 | unsigned cached_cq_overflow; |
ad3eb2c8 | 365 | unsigned long sq_check_overflow; |
de0617e4 | 366 | |
e941894e JA |
367 | /* hashed buffered write serialization */ |
368 | struct io_wq_hash *hash_map; | |
369 | ||
de0617e4 | 370 | struct list_head defer_list; |
5262f567 | 371 | struct list_head timeout_list; |
1d7bb1d5 | 372 | struct list_head cq_overflow_list; |
fcb323cc | 373 | |
ad3eb2c8 | 374 | struct io_uring_sqe *sq_sqes; |
2b188cc1 JA |
375 | } ____cacheline_aligned_in_smp; |
376 | ||
3c1a2ead JA |
377 | struct { |
378 | struct mutex uring_lock; | |
379 | wait_queue_head_t wait; | |
380 | } ____cacheline_aligned_in_smp; | |
381 | ||
382 | struct io_submit_state submit_state; | |
383 | ||
206aefde JA |
384 | struct io_rings *rings; |
385 | ||
2aede0e4 JA |
386 | /* Only used for accounting purposes */ |
387 | struct mm_struct *mm_account; | |
388 | ||
7c30f36a | 389 | const struct cred *sq_creds; /* cred used for __io_sq_thread() */ |
534ca6d6 JA |
390 | struct io_sq_data *sq_data; /* if using sq thread polling */ |
391 | ||
90554200 | 392 | struct wait_queue_head sqo_sq_wait; |
69fb2131 | 393 | struct list_head sqd_list; |
75b28aff | 394 | |
6b06314c JA |
395 | /* |
396 | * If used, fixed file set. Writers must ensure that ->refs is dead, | |
397 | * readers must ensure that ->refs is alive as long as the file* is | |
398 | * used. Only updated through io_uring_register(2). | |
399 | */ | |
b895c9a6 | 400 | struct io_rsrc_data *file_data; |
aeca241b | 401 | struct io_file_table file_table; |
6b06314c JA |
402 | unsigned nr_user_files; |
403 | ||
edafccee JA |
404 | /* if used, fixed mapped user buffers */ |
405 | unsigned nr_user_bufs; | |
406 | struct io_mapped_ubuf *user_bufs; | |
407 | ||
2b188cc1 JA |
408 | struct user_struct *user; |
409 | ||
0f158b4c | 410 | struct completion ref_comp; |
206aefde JA |
411 | |
412 | #if defined(CONFIG_UNIX) | |
413 | struct socket *ring_sock; | |
414 | #endif | |
415 | ||
9e15c3a0 | 416 | struct xarray io_buffers; |
5a2e745d | 417 | |
61cf9370 MWO |
418 | struct xarray personalities; |
419 | u32 pers_next; | |
071698e1 | 420 | |
206aefde JA |
421 | struct { |
422 | unsigned cached_cq_tail; | |
423 | unsigned cq_entries; | |
424 | unsigned cq_mask; | |
425 | atomic_t cq_timeouts; | |
f010505b | 426 | unsigned cq_last_tm_flush; |
ad3eb2c8 | 427 | unsigned long cq_check_overflow; |
206aefde JA |
428 | struct wait_queue_head cq_wait; |
429 | struct fasync_struct *cq_fasync; | |
430 | struct eventfd_ctx *cq_ev_fd; | |
431 | } ____cacheline_aligned_in_smp; | |
2b188cc1 | 432 | |
2b188cc1 JA |
433 | struct { |
434 | spinlock_t completion_lock; | |
e94f141b | 435 | |
def596e9 | 436 | /* |
540e32a0 | 437 | * ->iopoll_list is protected by the ctx->uring_lock for |
def596e9 JA |
438 | * io_uring instances that don't use IORING_SETUP_SQPOLL. |
439 | * For SQPOLL, only the single threaded io_sq_thread() will | |
440 | * manipulate the list, hence no extra locking is needed there. | |
441 | */ | |
540e32a0 | 442 | struct list_head iopoll_list; |
78076bb6 JA |
443 | struct hlist_head *cancel_hash; |
444 | unsigned cancel_hash_bits; | |
e94f141b | 445 | bool poll_multi_file; |
2b188cc1 | 446 | } ____cacheline_aligned_in_smp; |
85faa7b8 | 447 | |
269bbe5f BM |
448 | struct delayed_work rsrc_put_work; |
449 | struct llist_head rsrc_put_llist; | |
d67d2263 BM |
450 | struct list_head rsrc_ref_list; |
451 | spinlock_t rsrc_ref_lock; | |
a7f0ed5a | 452 | struct io_rsrc_node *rsrc_node; |
b895c9a6 | 453 | struct io_rsrc_node *rsrc_backup_node; |
4a38aed2 | 454 | |
21b55dbc | 455 | struct io_restriction restrictions; |
3c1a2ead | 456 | |
7c25c0d1 JA |
457 | /* exit task_work */ |
458 | struct callback_head *exit_task_work; | |
459 | ||
3c1a2ead JA |
460 | /* Keep this last, we don't need it for the fast path */ |
461 | struct work_struct exit_work; | |
13bf43f5 | 462 | struct list_head tctx_list; |
2b188cc1 JA |
463 | }; |
464 | ||
53e043b2 SM |
465 | struct io_uring_task { |
466 | /* submission side */ | |
467 | struct xarray xa; | |
468 | struct wait_queue_head wait; | |
ee53fb2b SM |
469 | const struct io_ring_ctx *last; |
470 | struct io_wq *io_wq; | |
53e043b2 | 471 | struct percpu_counter inflight; |
b303fe2e | 472 | atomic_t inflight_tracked; |
53e043b2 | 473 | atomic_t in_idle; |
53e043b2 SM |
474 | |
475 | spinlock_t task_lock; | |
476 | struct io_wq_work_list task_list; | |
477 | unsigned long task_state; | |
478 | struct callback_head task_work; | |
479 | }; | |
480 | ||
09bb8394 JA |
481 | /* |
482 | * First field must be the file pointer in all the | |
483 | * iocb unions! See also 'struct kiocb' in <linux/fs.h> | |
484 | */ | |
221c5eb2 JA |
485 | struct io_poll_iocb { |
486 | struct file *file; | |
018043be | 487 | struct wait_queue_head *head; |
221c5eb2 | 488 | __poll_t events; |
8c838788 | 489 | bool done; |
221c5eb2 | 490 | bool canceled; |
9d805892 PB |
491 | struct wait_queue_entry wait; |
492 | }; | |
493 | ||
494 | struct io_poll_update { | |
495 | struct file *file; | |
496 | u64 old_user_data; | |
497 | u64 new_user_data; | |
498 | __poll_t events; | |
b69de288 JA |
499 | bool update_events; |
500 | bool update_user_data; | |
221c5eb2 JA |
501 | }; |
502 | ||
018043be PB |
503 | struct io_poll_remove { |
504 | struct file *file; | |
505 | u64 addr; | |
506 | }; | |
507 | ||
b5dba59e JA |
508 | struct io_close { |
509 | struct file *file; | |
b5dba59e JA |
510 | int fd; |
511 | }; | |
512 | ||
ad8a48ac JA |
513 | struct io_timeout_data { |
514 | struct io_kiocb *req; | |
515 | struct hrtimer timer; | |
516 | struct timespec64 ts; | |
517 | enum hrtimer_mode mode; | |
518 | }; | |
519 | ||
8ed8d3c3 JA |
520 | struct io_accept { |
521 | struct file *file; | |
522 | struct sockaddr __user *addr; | |
523 | int __user *addr_len; | |
524 | int flags; | |
09952e3e | 525 | unsigned long nofile; |
8ed8d3c3 JA |
526 | }; |
527 | ||
528 | struct io_sync { | |
529 | struct file *file; | |
530 | loff_t len; | |
531 | loff_t off; | |
532 | int flags; | |
d63d1b5e | 533 | int mode; |
8ed8d3c3 JA |
534 | }; |
535 | ||
fbf23849 JA |
536 | struct io_cancel { |
537 | struct file *file; | |
538 | u64 addr; | |
539 | }; | |
540 | ||
b29472ee JA |
541 | struct io_timeout { |
542 | struct file *file; | |
bfe68a22 PB |
543 | u32 off; |
544 | u32 target_seq; | |
135fcde8 | 545 | struct list_head list; |
90cd7e42 PB |
546 | /* head of the link, used by linked timeouts only */ |
547 | struct io_kiocb *head; | |
b29472ee JA |
548 | }; |
549 | ||
0bdf7a2d PB |
550 | struct io_timeout_rem { |
551 | struct file *file; | |
552 | u64 addr; | |
9c8e11b3 PB |
553 | |
554 | /* timeout update */ | |
555 | struct timespec64 ts; | |
556 | u32 flags; | |
0bdf7a2d PB |
557 | }; |
558 | ||
9adbd45d JA |
559 | struct io_rw { |
560 | /* NOTE: kiocb has the file as the first member, so don't do it here */ | |
561 | struct kiocb kiocb; | |
562 | u64 addr; | |
563 | u64 len; | |
564 | }; | |
565 | ||
3fbb51c1 JA |
566 | struct io_connect { |
567 | struct file *file; | |
568 | struct sockaddr __user *addr; | |
569 | int addr_len; | |
570 | }; | |
571 | ||
e47293fd JA |
572 | struct io_sr_msg { |
573 | struct file *file; | |
fddaface | 574 | union { |
4af3417a PB |
575 | struct compat_msghdr __user *umsg_compat; |
576 | struct user_msghdr __user *umsg; | |
577 | void __user *buf; | |
fddaface | 578 | }; |
e47293fd | 579 | int msg_flags; |
bcda7baa | 580 | int bgid; |
fddaface | 581 | size_t len; |
bcda7baa | 582 | struct io_buffer *kbuf; |
e47293fd JA |
583 | }; |
584 | ||
15b71abe JA |
585 | struct io_open { |
586 | struct file *file; | |
587 | int dfd; | |
15b71abe | 588 | struct filename *filename; |
c12cedf2 | 589 | struct open_how how; |
4022e7af | 590 | unsigned long nofile; |
15b71abe JA |
591 | }; |
592 | ||
269bbe5f | 593 | struct io_rsrc_update { |
05f3fb3c JA |
594 | struct file *file; |
595 | u64 arg; | |
596 | u32 nr_args; | |
597 | u32 offset; | |
598 | }; | |
599 | ||
4840e418 JA |
600 | struct io_fadvise { |
601 | struct file *file; | |
602 | u64 offset; | |
603 | u32 len; | |
604 | u32 advice; | |
605 | }; | |
606 | ||
c1ca757b JA |
607 | struct io_madvise { |
608 | struct file *file; | |
609 | u64 addr; | |
610 | u32 len; | |
611 | u32 advice; | |
612 | }; | |
613 | ||
3e4827b0 JA |
614 | struct io_epoll { |
615 | struct file *file; | |
616 | int epfd; | |
617 | int op; | |
618 | int fd; | |
619 | struct epoll_event event; | |
e47293fd JA |
620 | }; |
621 | ||
7d67af2c PB |
622 | struct io_splice { |
623 | struct file *file_out; | |
624 | struct file *file_in; | |
625 | loff_t off_out; | |
626 | loff_t off_in; | |
627 | u64 len; | |
628 | unsigned int flags; | |
629 | }; | |
630 | ||
ddf0322d JA |
631 | struct io_provide_buf { |
632 | struct file *file; | |
633 | __u64 addr; | |
634 | __s32 len; | |
635 | __u32 bgid; | |
636 | __u16 nbufs; | |
637 | __u16 bid; | |
638 | }; | |
639 | ||
1d9e1288 BM |
640 | struct io_statx { |
641 | struct file *file; | |
642 | int dfd; | |
643 | unsigned int mask; | |
644 | unsigned int flags; | |
e62753e4 | 645 | const char __user *filename; |
1d9e1288 BM |
646 | struct statx __user *buffer; |
647 | }; | |
648 | ||
36f4fa68 JA |
649 | struct io_shutdown { |
650 | struct file *file; | |
651 | int how; | |
652 | }; | |
653 | ||
80a261fd JA |
654 | struct io_rename { |
655 | struct file *file; | |
656 | int old_dfd; | |
657 | int new_dfd; | |
658 | struct filename *oldpath; | |
659 | struct filename *newpath; | |
660 | int flags; | |
661 | }; | |
662 | ||
14a1143b JA |
663 | struct io_unlink { |
664 | struct file *file; | |
665 | int dfd; | |
666 | int flags; | |
667 | struct filename *filename; | |
668 | }; | |
669 | ||
3ca405eb PB |
670 | struct io_completion { |
671 | struct file *file; | |
672 | struct list_head list; | |
8c3f9cd1 | 673 | u32 cflags; |
3ca405eb PB |
674 | }; |
675 | ||
f499a021 JA |
676 | struct io_async_connect { |
677 | struct sockaddr_storage address; | |
678 | }; | |
679 | ||
03b1230c JA |
680 | struct io_async_msghdr { |
681 | struct iovec fast_iov[UIO_FASTIOV]; | |
257e84a5 PB |
682 | /* points to an allocated iov, if NULL we use fast_iov instead */ |
683 | struct iovec *free_iov; | |
03b1230c JA |
684 | struct sockaddr __user *uaddr; |
685 | struct msghdr msg; | |
b537916c | 686 | struct sockaddr_storage addr; |
03b1230c JA |
687 | }; |
688 | ||
f67676d1 JA |
689 | struct io_async_rw { |
690 | struct iovec fast_iov[UIO_FASTIOV]; | |
ff6165b2 JA |
691 | const struct iovec *free_iovec; |
692 | struct iov_iter iter; | |
227c0c96 | 693 | size_t bytes_done; |
bcf5a063 | 694 | struct wait_page_queue wpq; |
f67676d1 JA |
695 | }; |
696 | ||
6b47ee6e PB |
697 | enum { |
698 | REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, | |
699 | REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, | |
700 | REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, | |
701 | REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, | |
702 | REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, | |
bcda7baa | 703 | REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, |
6b47ee6e | 704 | |
6b47ee6e PB |
705 | REQ_F_FAIL_LINK_BIT, |
706 | REQ_F_INFLIGHT_BIT, | |
707 | REQ_F_CUR_POS_BIT, | |
708 | REQ_F_NOWAIT_BIT, | |
6b47ee6e | 709 | REQ_F_LINK_TIMEOUT_BIT, |
99bc4c38 | 710 | REQ_F_NEED_CLEANUP_BIT, |
d7718a9d | 711 | REQ_F_POLLED_BIT, |
bcda7baa | 712 | REQ_F_BUFFER_SELECTED_BIT, |
900fad45 | 713 | REQ_F_LTIMEOUT_ACTIVE_BIT, |
e342c807 | 714 | REQ_F_COMPLETE_INLINE_BIT, |
230d50d4 | 715 | REQ_F_REISSUE_BIT, |
8c130827 | 716 | REQ_F_DONT_REISSUE_BIT, |
9d805892 | 717 | REQ_F_POLL_UPDATE_BIT, |
7b29f92d JA |
718 | /* keep async read/write and isreg together and in order */ |
719 | REQ_F_ASYNC_READ_BIT, | |
720 | REQ_F_ASYNC_WRITE_BIT, | |
721 | REQ_F_ISREG_BIT, | |
84557871 JA |
722 | |
723 | /* not a real bit, just to check we're not overflowing the space */ | |
724 | __REQ_F_LAST_BIT, | |
6b47ee6e PB |
725 | }; |
726 | ||
727 | enum { | |
728 | /* ctx owns file */ | |
729 | REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT), | |
730 | /* drain existing IO first */ | |
731 | REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT), | |
732 | /* linked sqes */ | |
733 | REQ_F_LINK = BIT(REQ_F_LINK_BIT), | |
734 | /* doesn't sever on completion < 0 */ | |
735 | REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT), | |
736 | /* IOSQE_ASYNC */ | |
737 | REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT), | |
bcda7baa JA |
738 | /* IOSQE_BUFFER_SELECT */ |
739 | REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT), | |
6b47ee6e | 740 | |
6b47ee6e PB |
741 | /* fail rest of links */ |
742 | REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT), | |
b05a1bcd | 743 | /* on inflight list, should be cancelled and waited on exit reliably */ |
6b47ee6e PB |
744 | REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT), |
745 | /* read/write uses file position */ | |
746 | REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT), | |
747 | /* must not punt to workers */ | |
748 | REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT), | |
900fad45 | 749 | /* has or had linked timeout */ |
6b47ee6e | 750 | REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT), |
99bc4c38 PB |
751 | /* needs cleanup */ |
752 | REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT), | |
d7718a9d JA |
753 | /* already went through poll handler */ |
754 | REQ_F_POLLED = BIT(REQ_F_POLLED_BIT), | |
bcda7baa JA |
755 | /* buffer already selected */ |
756 | REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT), | |
900fad45 PB |
757 | /* linked timeout is active, i.e. prepared by link's head */ |
758 | REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT), | |
e342c807 PB |
759 | /* completion is deferred through io_comp_state */ |
760 | REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT), | |
230d50d4 JA |
761 | /* caller should reissue async */ |
762 | REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT), | |
8c130827 PB |
763 | /* don't attempt request reissue, see io_rw_reissue() */ |
764 | REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT), | |
9d805892 PB |
765 | /* switches between poll and poll update */ |
766 | REQ_F_POLL_UPDATE = BIT(REQ_F_POLL_UPDATE_BIT), | |
7b29f92d JA |
767 | /* supports async reads */ |
768 | REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT), | |
769 | /* supports async writes */ | |
770 | REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT), | |
771 | /* regular file */ | |
772 | REQ_F_ISREG = BIT(REQ_F_ISREG_BIT), | |
d7718a9d JA |
773 | }; |
774 | ||
775 | struct async_poll { | |
776 | struct io_poll_iocb poll; | |
807abcb0 | 777 | struct io_poll_iocb *double_poll; |
6b47ee6e PB |
778 | }; |
779 | ||
7cbf1722 JA |
780 | struct io_task_work { |
781 | struct io_wq_work_node node; | |
782 | task_work_func_t func; | |
783 | }; | |
784 | ||
09bb8394 JA |
785 | /* |
786 | * NOTE! Each of the iocb union members has the file pointer | |
787 | * as the first entry in their struct definition. So you can | |
788 | * access the file pointer through any of the sub-structs, | |
789 | * or directly as just 'ki_filp' in this struct. | |
790 | */ | |
2b188cc1 | 791 | struct io_kiocb { |
221c5eb2 | 792 | union { |
09bb8394 | 793 | struct file *file; |
9adbd45d | 794 | struct io_rw rw; |
221c5eb2 | 795 | struct io_poll_iocb poll; |
9d805892 | 796 | struct io_poll_update poll_update; |
018043be | 797 | struct io_poll_remove poll_remove; |
8ed8d3c3 JA |
798 | struct io_accept accept; |
799 | struct io_sync sync; | |
fbf23849 | 800 | struct io_cancel cancel; |
b29472ee | 801 | struct io_timeout timeout; |
0bdf7a2d | 802 | struct io_timeout_rem timeout_rem; |
3fbb51c1 | 803 | struct io_connect connect; |
e47293fd | 804 | struct io_sr_msg sr_msg; |
15b71abe | 805 | struct io_open open; |
b5dba59e | 806 | struct io_close close; |
269bbe5f | 807 | struct io_rsrc_update rsrc_update; |
4840e418 | 808 | struct io_fadvise fadvise; |
c1ca757b | 809 | struct io_madvise madvise; |
3e4827b0 | 810 | struct io_epoll epoll; |
7d67af2c | 811 | struct io_splice splice; |
ddf0322d | 812 | struct io_provide_buf pbuf; |
1d9e1288 | 813 | struct io_statx statx; |
36f4fa68 | 814 | struct io_shutdown shutdown; |
80a261fd | 815 | struct io_rename rename; |
14a1143b | 816 | struct io_unlink unlink; |
3ca405eb PB |
817 | /* use only after cleaning per-op data, see io_clean_op() */ |
818 | struct io_completion compl; | |
221c5eb2 | 819 | }; |
2b188cc1 | 820 | |
e8c2bc1f JA |
821 | /* opcode allocated if it needs to store data for async defer */ |
822 | void *async_data; | |
d625c6ee | 823 | u8 opcode; |
65a6543d XW |
824 | /* polled IO has completed */ |
825 | u8 iopoll_completed; | |
2b188cc1 | 826 | |
4f4eeba8 | 827 | u16 buf_index; |
9cf7c104 | 828 | u32 result; |
4f4eeba8 | 829 | |
010e8e6b PB |
830 | struct io_ring_ctx *ctx; |
831 | unsigned int flags; | |
abc54d63 | 832 | atomic_t refs; |
010e8e6b PB |
833 | struct task_struct *task; |
834 | u64 user_data; | |
d7718a9d | 835 | |
f2f87370 | 836 | struct io_kiocb *link; |
269bbe5f | 837 | struct percpu_ref *fixed_rsrc_refs; |
fcb323cc | 838 | |
b303fe2e | 839 | /* used with ctx->iopoll_list with reads/writes */ |
010e8e6b | 840 | struct list_head inflight_entry; |
7cbf1722 JA |
841 | union { |
842 | struct io_task_work io_task_work; | |
843 | struct callback_head task_work; | |
844 | }; | |
010e8e6b PB |
845 | /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ |
846 | struct hlist_node hash_node; | |
847 | struct async_poll *apoll; | |
848 | struct io_wq_work work; | |
2b188cc1 | 849 | }; |
05589553 | 850 | |
13bf43f5 PB |
851 | struct io_tctx_node { |
852 | struct list_head ctx_node; | |
853 | struct task_struct *task; | |
13bf43f5 PB |
854 | struct io_ring_ctx *ctx; |
855 | }; | |
856 | ||
27dc8338 PB |
857 | struct io_defer_entry { |
858 | struct list_head list; | |
859 | struct io_kiocb *req; | |
9cf7c104 | 860 | u32 seq; |
2b188cc1 JA |
861 | }; |
862 | ||
d3656344 | 863 | struct io_op_def { |
d3656344 JA |
864 | /* needs req->file assigned */ |
865 | unsigned needs_file : 1; | |
d3656344 JA |
866 | /* hash wq insertion if file is a regular file */ |
867 | unsigned hash_reg_file : 1; | |
868 | /* unbound wq insertion if file is a non-regular file */ | |
869 | unsigned unbound_nonreg_file : 1; | |
66f4af93 JA |
870 | /* opcode is not supported by this kernel */ |
871 | unsigned not_supported : 1; | |
8a72758c JA |
872 | /* set if opcode supports polled "wait" */ |
873 | unsigned pollin : 1; | |
874 | unsigned pollout : 1; | |
bcda7baa JA |
875 | /* op supports buffer selection */ |
876 | unsigned buffer_select : 1; | |
26f0505a PB |
877 | /* do prep async if is going to be punted */ |
878 | unsigned needs_async_setup : 1; | |
27926b68 JA |
879 | /* should block plug */ |
880 | unsigned plug : 1; | |
e8c2bc1f JA |
881 | /* size of async data needed, if any */ |
882 | unsigned short async_size; | |
d3656344 JA |
883 | }; |
884 | ||
0918682b | 885 | static const struct io_op_def io_op_defs[] = { |
0463b6c5 PB |
886 | [IORING_OP_NOP] = {}, |
887 | [IORING_OP_READV] = { | |
d3656344 JA |
888 | .needs_file = 1, |
889 | .unbound_nonreg_file = 1, | |
8a72758c | 890 | .pollin = 1, |
4d954c25 | 891 | .buffer_select = 1, |
26f0505a | 892 | .needs_async_setup = 1, |
27926b68 | 893 | .plug = 1, |
e8c2bc1f | 894 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 895 | }, |
0463b6c5 | 896 | [IORING_OP_WRITEV] = { |
d3656344 JA |
897 | .needs_file = 1, |
898 | .hash_reg_file = 1, | |
899 | .unbound_nonreg_file = 1, | |
8a72758c | 900 | .pollout = 1, |
26f0505a | 901 | .needs_async_setup = 1, |
27926b68 | 902 | .plug = 1, |
e8c2bc1f | 903 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 904 | }, |
0463b6c5 | 905 | [IORING_OP_FSYNC] = { |
d3656344 JA |
906 | .needs_file = 1, |
907 | }, | |
0463b6c5 | 908 | [IORING_OP_READ_FIXED] = { |
d3656344 JA |
909 | .needs_file = 1, |
910 | .unbound_nonreg_file = 1, | |
8a72758c | 911 | .pollin = 1, |
27926b68 | 912 | .plug = 1, |
e8c2bc1f | 913 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 914 | }, |
0463b6c5 | 915 | [IORING_OP_WRITE_FIXED] = { |
d3656344 JA |
916 | .needs_file = 1, |
917 | .hash_reg_file = 1, | |
918 | .unbound_nonreg_file = 1, | |
8a72758c | 919 | .pollout = 1, |
27926b68 | 920 | .plug = 1, |
e8c2bc1f | 921 | .async_size = sizeof(struct io_async_rw), |
d3656344 | 922 | }, |
0463b6c5 | 923 | [IORING_OP_POLL_ADD] = { |
d3656344 JA |
924 | .needs_file = 1, |
925 | .unbound_nonreg_file = 1, | |
926 | }, | |
0463b6c5 PB |
927 | [IORING_OP_POLL_REMOVE] = {}, |
928 | [IORING_OP_SYNC_FILE_RANGE] = { | |
d3656344 JA |
929 | .needs_file = 1, |
930 | }, | |
0463b6c5 | 931 | [IORING_OP_SENDMSG] = { |
d3656344 JA |
932 | .needs_file = 1, |
933 | .unbound_nonreg_file = 1, | |
8a72758c | 934 | .pollout = 1, |
26f0505a | 935 | .needs_async_setup = 1, |
e8c2bc1f | 936 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 937 | }, |
0463b6c5 | 938 | [IORING_OP_RECVMSG] = { |
d3656344 JA |
939 | .needs_file = 1, |
940 | .unbound_nonreg_file = 1, | |
8a72758c | 941 | .pollin = 1, |
52de1fe1 | 942 | .buffer_select = 1, |
26f0505a | 943 | .needs_async_setup = 1, |
e8c2bc1f | 944 | .async_size = sizeof(struct io_async_msghdr), |
d3656344 | 945 | }, |
0463b6c5 | 946 | [IORING_OP_TIMEOUT] = { |
e8c2bc1f | 947 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 948 | }, |
9c8e11b3 PB |
949 | [IORING_OP_TIMEOUT_REMOVE] = { |
950 | /* used by timeout updates' prep() */ | |
9c8e11b3 | 951 | }, |
0463b6c5 | 952 | [IORING_OP_ACCEPT] = { |
d3656344 JA |
953 | .needs_file = 1, |
954 | .unbound_nonreg_file = 1, | |
8a72758c | 955 | .pollin = 1, |
d3656344 | 956 | }, |
0463b6c5 PB |
957 | [IORING_OP_ASYNC_CANCEL] = {}, |
958 | [IORING_OP_LINK_TIMEOUT] = { | |
e8c2bc1f | 959 | .async_size = sizeof(struct io_timeout_data), |
d3656344 | 960 | }, |
0463b6c5 | 961 | [IORING_OP_CONNECT] = { |
d3656344 JA |
962 | .needs_file = 1, |
963 | .unbound_nonreg_file = 1, | |
8a72758c | 964 | .pollout = 1, |
26f0505a | 965 | .needs_async_setup = 1, |
e8c2bc1f | 966 | .async_size = sizeof(struct io_async_connect), |
d3656344 | 967 | }, |
0463b6c5 | 968 | [IORING_OP_FALLOCATE] = { |
d3656344 | 969 | .needs_file = 1, |
d3656344 | 970 | }, |
44526bed JA |
971 | [IORING_OP_OPENAT] = {}, |
972 | [IORING_OP_CLOSE] = {}, | |
973 | [IORING_OP_FILES_UPDATE] = {}, | |
974 | [IORING_OP_STATX] = {}, | |
0463b6c5 | 975 | [IORING_OP_READ] = { |
3a6820f2 JA |
976 | .needs_file = 1, |
977 | .unbound_nonreg_file = 1, | |
8a72758c | 978 | .pollin = 1, |
bcda7baa | 979 | .buffer_select = 1, |
27926b68 | 980 | .plug = 1, |
e8c2bc1f | 981 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 982 | }, |
0463b6c5 | 983 | [IORING_OP_WRITE] = { |
3a6820f2 JA |
984 | .needs_file = 1, |
985 | .unbound_nonreg_file = 1, | |
8a72758c | 986 | .pollout = 1, |
27926b68 | 987 | .plug = 1, |
e8c2bc1f | 988 | .async_size = sizeof(struct io_async_rw), |
3a6820f2 | 989 | }, |
0463b6c5 | 990 | [IORING_OP_FADVISE] = { |
4840e418 | 991 | .needs_file = 1, |
c1ca757b | 992 | }, |
44526bed | 993 | [IORING_OP_MADVISE] = {}, |
0463b6c5 | 994 | [IORING_OP_SEND] = { |
fddaface JA |
995 | .needs_file = 1, |
996 | .unbound_nonreg_file = 1, | |
8a72758c | 997 | .pollout = 1, |
fddaface | 998 | }, |
0463b6c5 | 999 | [IORING_OP_RECV] = { |
fddaface JA |
1000 | .needs_file = 1, |
1001 | .unbound_nonreg_file = 1, | |
8a72758c | 1002 | .pollin = 1, |
bcda7baa | 1003 | .buffer_select = 1, |
fddaface | 1004 | }, |
0463b6c5 | 1005 | [IORING_OP_OPENAT2] = { |
cebdb986 | 1006 | }, |
3e4827b0 JA |
1007 | [IORING_OP_EPOLL_CTL] = { |
1008 | .unbound_nonreg_file = 1, | |
3e4827b0 | 1009 | }, |
7d67af2c PB |
1010 | [IORING_OP_SPLICE] = { |
1011 | .needs_file = 1, | |
1012 | .hash_reg_file = 1, | |
1013 | .unbound_nonreg_file = 1, | |
ddf0322d JA |
1014 | }, |
1015 | [IORING_OP_PROVIDE_BUFFERS] = {}, | |
067524e9 | 1016 | [IORING_OP_REMOVE_BUFFERS] = {}, |
f2a8d5c7 PB |
1017 | [IORING_OP_TEE] = { |
1018 | .needs_file = 1, | |
1019 | .hash_reg_file = 1, | |
1020 | .unbound_nonreg_file = 1, | |
1021 | }, | |
36f4fa68 JA |
1022 | [IORING_OP_SHUTDOWN] = { |
1023 | .needs_file = 1, | |
1024 | }, | |
44526bed JA |
1025 | [IORING_OP_RENAMEAT] = {}, |
1026 | [IORING_OP_UNLINKAT] = {}, | |
d3656344 JA |
1027 | }; |
1028 | ||
7a612350 | 1029 | static bool io_disarm_next(struct io_kiocb *req); |
d56d938b | 1030 | static void io_uring_del_task_file(unsigned long index); |
9936c7c2 PB |
1031 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
1032 | struct task_struct *task, | |
1033 | struct files_struct *files); | |
37d1e2e3 | 1034 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx); |
b895c9a6 | 1035 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx); |
1ffc5422 | 1036 | |
ff642164 | 1037 | static bool io_cqring_fill_event(struct io_kiocb *req, long res, unsigned cflags); |
ec9c02ad | 1038 | static void io_put_req(struct io_kiocb *req); |
216578e5 | 1039 | static void io_put_req_deferred(struct io_kiocb *req, int nr); |
c7dae4ba JA |
1040 | static void io_dismantle_req(struct io_kiocb *req); |
1041 | static void io_put_task(struct task_struct *task, int nr); | |
94ae5e77 JA |
1042 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req); |
1043 | static void io_queue_linked_timeout(struct io_kiocb *req); | |
05f3fb3c | 1044 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, |
269bbe5f | 1045 | struct io_uring_rsrc_update *ip, |
05f3fb3c | 1046 | unsigned nr_args); |
68fb8979 | 1047 | static void io_clean_op(struct io_kiocb *req); |
8371adf5 PB |
1048 | static struct file *io_file_get(struct io_submit_state *state, |
1049 | struct io_kiocb *req, int fd, bool fixed); | |
c5eef2b9 | 1050 | static void __io_queue_sqe(struct io_kiocb *req); |
269bbe5f | 1051 | static void io_rsrc_put_work(struct work_struct *work); |
de0617e4 | 1052 | |
907d1df3 | 1053 | static void io_req_task_queue(struct io_kiocb *req); |
65453d1e JA |
1054 | static void io_submit_flush_completions(struct io_comp_state *cs, |
1055 | struct io_ring_ctx *ctx); | |
5082620f | 1056 | static bool io_poll_remove_waitqs(struct io_kiocb *req); |
179ae0d1 | 1057 | static int io_req_prep_async(struct io_kiocb *req); |
de0617e4 | 1058 | |
2b188cc1 JA |
1059 | static struct kmem_cache *req_cachep; |
1060 | ||
0918682b | 1061 | static const struct file_operations io_uring_fops; |
2b188cc1 JA |
1062 | |
1063 | struct sock *io_uring_get_socket(struct file *file) | |
1064 | { | |
1065 | #if defined(CONFIG_UNIX) | |
1066 | if (file->f_op == &io_uring_fops) { | |
1067 | struct io_ring_ctx *ctx = file->private_data; | |
1068 | ||
1069 | return ctx->ring_sock->sk; | |
1070 | } | |
1071 | #endif | |
1072 | return NULL; | |
1073 | } | |
1074 | EXPORT_SYMBOL(io_uring_get_socket); | |
1075 | ||
f2f87370 PB |
1076 | #define io_for_each_link(pos, head) \ |
1077 | for (pos = (head); pos; pos = pos->link) | |
1078 | ||
b895c9a6 | 1079 | static inline void io_req_set_rsrc_node(struct io_kiocb *req) |
36f72fe2 PB |
1080 | { |
1081 | struct io_ring_ctx *ctx = req->ctx; | |
1082 | ||
269bbe5f | 1083 | if (!req->fixed_rsrc_refs) { |
a7f0ed5a | 1084 | req->fixed_rsrc_refs = &ctx->rsrc_node->refs; |
269bbe5f | 1085 | percpu_ref_get(req->fixed_rsrc_refs); |
36f72fe2 PB |
1086 | } |
1087 | } | |
1088 | ||
f70865db PB |
1089 | static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl) |
1090 | { | |
1091 | bool got = percpu_ref_tryget(ref); | |
1092 | ||
1093 | /* already at zero, wait for ->release() */ | |
1094 | if (!got) | |
1095 | wait_for_completion(compl); | |
1096 | percpu_ref_resurrect(ref); | |
1097 | if (got) | |
1098 | percpu_ref_put(ref); | |
1099 | } | |
1100 | ||
08d23634 PB |
1101 | static bool io_match_task(struct io_kiocb *head, |
1102 | struct task_struct *task, | |
1103 | struct files_struct *files) | |
1104 | { | |
1105 | struct io_kiocb *req; | |
1106 | ||
68207680 | 1107 | if (task && head->task != task) |
08d23634 PB |
1108 | return false; |
1109 | if (!files) | |
1110 | return true; | |
1111 | ||
1112 | io_for_each_link(req, head) { | |
b05a1bcd | 1113 | if (req->flags & REQ_F_INFLIGHT) |
02a13674 | 1114 | return true; |
08d23634 PB |
1115 | } |
1116 | return false; | |
1117 | } | |
1118 | ||
c40f6379 JA |
1119 | static inline void req_set_fail_links(struct io_kiocb *req) |
1120 | { | |
e4335ed3 | 1121 | if (req->flags & REQ_F_LINK) |
c40f6379 JA |
1122 | req->flags |= REQ_F_FAIL_LINK; |
1123 | } | |
4a38aed2 | 1124 | |
2b188cc1 JA |
1125 | static void io_ring_ctx_ref_free(struct percpu_ref *ref) |
1126 | { | |
1127 | struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs); | |
1128 | ||
0f158b4c | 1129 | complete(&ctx->ref_comp); |
2b188cc1 JA |
1130 | } |
1131 | ||
8eb7e2d0 PB |
1132 | static inline bool io_is_timeout_noseq(struct io_kiocb *req) |
1133 | { | |
1134 | return !req->timeout.off; | |
1135 | } | |
1136 | ||
2b188cc1 JA |
1137 | static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) |
1138 | { | |
1139 | struct io_ring_ctx *ctx; | |
78076bb6 | 1140 | int hash_bits; |
2b188cc1 JA |
1141 | |
1142 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1143 | if (!ctx) | |
1144 | return NULL; | |
1145 | ||
78076bb6 JA |
1146 | /* |
1147 | * Use 5 bits less than the max cq entries, that should give us around | |
1148 | * 32 entries per hash list if totally full and uniformly spread. | |
1149 | */ | |
1150 | hash_bits = ilog2(p->cq_entries); | |
1151 | hash_bits -= 5; | |
1152 | if (hash_bits <= 0) | |
1153 | hash_bits = 1; | |
1154 | ctx->cancel_hash_bits = hash_bits; | |
1155 | ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), | |
1156 | GFP_KERNEL); | |
1157 | if (!ctx->cancel_hash) | |
1158 | goto err; | |
1159 | __hash_init(ctx->cancel_hash, 1U << hash_bits); | |
1160 | ||
21482896 | 1161 | if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, |
206aefde JA |
1162 | PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) |
1163 | goto err; | |
2b188cc1 JA |
1164 | |
1165 | ctx->flags = p->flags; | |
90554200 | 1166 | init_waitqueue_head(&ctx->sqo_sq_wait); |
69fb2131 | 1167 | INIT_LIST_HEAD(&ctx->sqd_list); |
2b188cc1 | 1168 | init_waitqueue_head(&ctx->cq_wait); |
1d7bb1d5 | 1169 | INIT_LIST_HEAD(&ctx->cq_overflow_list); |
0f158b4c | 1170 | init_completion(&ctx->ref_comp); |
9e15c3a0 | 1171 | xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); |
61cf9370 | 1172 | xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); |
2b188cc1 JA |
1173 | mutex_init(&ctx->uring_lock); |
1174 | init_waitqueue_head(&ctx->wait); | |
1175 | spin_lock_init(&ctx->completion_lock); | |
540e32a0 | 1176 | INIT_LIST_HEAD(&ctx->iopoll_list); |
de0617e4 | 1177 | INIT_LIST_HEAD(&ctx->defer_list); |
5262f567 | 1178 | INIT_LIST_HEAD(&ctx->timeout_list); |
d67d2263 BM |
1179 | spin_lock_init(&ctx->rsrc_ref_lock); |
1180 | INIT_LIST_HEAD(&ctx->rsrc_ref_list); | |
269bbe5f BM |
1181 | INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); |
1182 | init_llist_head(&ctx->rsrc_put_llist); | |
13bf43f5 | 1183 | INIT_LIST_HEAD(&ctx->tctx_list); |
1b4c351f | 1184 | INIT_LIST_HEAD(&ctx->submit_state.comp.free_list); |
c7dae4ba | 1185 | INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list); |
2b188cc1 | 1186 | return ctx; |
206aefde | 1187 | err: |
78076bb6 | 1188 | kfree(ctx->cancel_hash); |
206aefde JA |
1189 | kfree(ctx); |
1190 | return NULL; | |
2b188cc1 JA |
1191 | } |
1192 | ||
9cf7c104 | 1193 | static bool req_need_defer(struct io_kiocb *req, u32 seq) |
7adf4eaf | 1194 | { |
2bc9930e JA |
1195 | if (unlikely(req->flags & REQ_F_IO_DRAIN)) { |
1196 | struct io_ring_ctx *ctx = req->ctx; | |
a197f664 | 1197 | |
9cf7c104 | 1198 | return seq != ctx->cached_cq_tail |
2c3bac6d | 1199 | + READ_ONCE(ctx->cached_cq_overflow); |
2bc9930e | 1200 | } |
de0617e4 | 1201 | |
9d858b21 | 1202 | return false; |
de0617e4 JA |
1203 | } |
1204 | ||
ce3d5aae PB |
1205 | static void io_req_track_inflight(struct io_kiocb *req) |
1206 | { | |
ce3d5aae | 1207 | if (!(req->flags & REQ_F_INFLIGHT)) { |
ce3d5aae | 1208 | req->flags |= REQ_F_INFLIGHT; |
b303fe2e | 1209 | atomic_inc(¤t->io_uring->inflight_tracked); |
ce3d5aae PB |
1210 | } |
1211 | } | |
1212 | ||
1e6fa521 JA |
1213 | static void io_prep_async_work(struct io_kiocb *req) |
1214 | { | |
1215 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
1e6fa521 JA |
1216 | struct io_ring_ctx *ctx = req->ctx; |
1217 | ||
003e8dcc JA |
1218 | if (!req->work.creds) |
1219 | req->work.creds = get_current_cred(); | |
1220 | ||
e1d675df PB |
1221 | req->work.list.next = NULL; |
1222 | req->work.flags = 0; | |
feaadc4f PB |
1223 | if (req->flags & REQ_F_FORCE_ASYNC) |
1224 | req->work.flags |= IO_WQ_WORK_CONCURRENT; | |
1225 | ||
1e6fa521 JA |
1226 | if (req->flags & REQ_F_ISREG) { |
1227 | if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) | |
1228 | io_wq_hash_work(&req->work, file_inode(req->file)); | |
4b982bd0 | 1229 | } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { |
1e6fa521 JA |
1230 | if (def->unbound_nonreg_file) |
1231 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1232 | } | |
e1d675df PB |
1233 | |
1234 | switch (req->opcode) { | |
1235 | case IORING_OP_SPLICE: | |
1236 | case IORING_OP_TEE: | |
e1d675df PB |
1237 | if (!S_ISREG(file_inode(req->splice.file_in)->i_mode)) |
1238 | req->work.flags |= IO_WQ_WORK_UNBOUND; | |
1239 | break; | |
1240 | } | |
561fb04a | 1241 | } |
cccf0ee8 | 1242 | |
cbdcb435 | 1243 | static void io_prep_async_link(struct io_kiocb *req) |
561fb04a | 1244 | { |
cbdcb435 | 1245 | struct io_kiocb *cur; |
54a91f3b | 1246 | |
f2f87370 PB |
1247 | io_for_each_link(cur, req) |
1248 | io_prep_async_work(cur); | |
561fb04a JA |
1249 | } |
1250 | ||
ebf93667 | 1251 | static void io_queue_async_work(struct io_kiocb *req) |
561fb04a | 1252 | { |
a197f664 | 1253 | struct io_ring_ctx *ctx = req->ctx; |
cbdcb435 | 1254 | struct io_kiocb *link = io_prep_linked_timeout(req); |
5aa75ed5 | 1255 | struct io_uring_task *tctx = req->task->io_uring; |
561fb04a | 1256 | |
3bfe6106 JA |
1257 | BUG_ON(!tctx); |
1258 | BUG_ON(!tctx->io_wq); | |
561fb04a | 1259 | |
cbdcb435 PB |
1260 | /* init ->work of the whole link before punting */ |
1261 | io_prep_async_link(req); | |
d07f1e8a PB |
1262 | trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, |
1263 | &req->work, req->flags); | |
ebf93667 | 1264 | io_wq_enqueue(tctx->io_wq, &req->work); |
7271ef3a JA |
1265 | if (link) |
1266 | io_queue_linked_timeout(link); | |
cbdcb435 PB |
1267 | } |
1268 | ||
1ee4160c | 1269 | static void io_kill_timeout(struct io_kiocb *req, int status) |
8c855885 | 1270 | __must_hold(&req->ctx->completion_lock) |
5262f567 | 1271 | { |
e8c2bc1f | 1272 | struct io_timeout_data *io = req->async_data; |
5262f567 | 1273 | |
fd9c7bc5 | 1274 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
01cec8c1 PB |
1275 | atomic_set(&req->ctx->cq_timeouts, |
1276 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
135fcde8 | 1277 | list_del_init(&req->timeout.list); |
ff642164 | 1278 | io_cqring_fill_event(req, status, 0); |
216578e5 | 1279 | io_put_req_deferred(req, 1); |
5262f567 JA |
1280 | } |
1281 | } | |
1282 | ||
04518945 | 1283 | static void __io_queue_deferred(struct io_ring_ctx *ctx) |
de0617e4 | 1284 | { |
04518945 | 1285 | do { |
27dc8338 PB |
1286 | struct io_defer_entry *de = list_first_entry(&ctx->defer_list, |
1287 | struct io_defer_entry, list); | |
de0617e4 | 1288 | |
9cf7c104 | 1289 | if (req_need_defer(de->req, de->seq)) |
04518945 | 1290 | break; |
27dc8338 | 1291 | list_del_init(&de->list); |
907d1df3 | 1292 | io_req_task_queue(de->req); |
27dc8338 | 1293 | kfree(de); |
04518945 PB |
1294 | } while (!list_empty(&ctx->defer_list)); |
1295 | } | |
1296 | ||
360428f8 | 1297 | static void io_flush_timeouts(struct io_ring_ctx *ctx) |
de0617e4 | 1298 | { |
f010505b MDG |
1299 | u32 seq; |
1300 | ||
1301 | if (list_empty(&ctx->timeout_list)) | |
1302 | return; | |
1303 | ||
1304 | seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); | |
1305 | ||
1306 | do { | |
1307 | u32 events_needed, events_got; | |
360428f8 | 1308 | struct io_kiocb *req = list_first_entry(&ctx->timeout_list, |
135fcde8 | 1309 | struct io_kiocb, timeout.list); |
de0617e4 | 1310 | |
8eb7e2d0 | 1311 | if (io_is_timeout_noseq(req)) |
360428f8 | 1312 | break; |
f010505b MDG |
1313 | |
1314 | /* | |
1315 | * Since seq can easily wrap around over time, subtract | |
1316 | * the last seq at which timeouts were flushed before comparing. | |
1317 | * Assuming not more than 2^31-1 events have happened since, | |
1318 | * these subtractions won't have wrapped, so we can check if | |
1319 | * target is in [last_seq, current_seq] by comparing the two. | |
1320 | */ | |
1321 | events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; | |
1322 | events_got = seq - ctx->cq_last_tm_flush; | |
1323 | if (events_got < events_needed) | |
360428f8 | 1324 | break; |
bfe68a22 | 1325 | |
135fcde8 | 1326 | list_del_init(&req->timeout.list); |
1ee4160c | 1327 | io_kill_timeout(req, 0); |
f010505b MDG |
1328 | } while (!list_empty(&ctx->timeout_list)); |
1329 | ||
1330 | ctx->cq_last_tm_flush = seq; | |
360428f8 | 1331 | } |
5262f567 | 1332 | |
360428f8 PB |
1333 | static void io_commit_cqring(struct io_ring_ctx *ctx) |
1334 | { | |
1335 | io_flush_timeouts(ctx); | |
ec30e04b PB |
1336 | |
1337 | /* order cqe stores with ring update */ | |
1338 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
de0617e4 | 1339 | |
04518945 PB |
1340 | if (unlikely(!list_empty(&ctx->defer_list))) |
1341 | __io_queue_deferred(ctx); | |
de0617e4 JA |
1342 | } |
1343 | ||
90554200 JA |
1344 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
1345 | { | |
1346 | struct io_rings *r = ctx->rings; | |
1347 | ||
1348 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries; | |
1349 | } | |
1350 | ||
888aae2e PB |
1351 | static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx) |
1352 | { | |
1353 | return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); | |
1354 | } | |
1355 | ||
8d13326e | 1356 | static inline struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx) |
2b188cc1 | 1357 | { |
75b28aff | 1358 | struct io_rings *rings = ctx->rings; |
2b188cc1 JA |
1359 | unsigned tail; |
1360 | ||
115e12e5 SB |
1361 | /* |
1362 | * writes to the cq entry need to come after reading head; the | |
1363 | * control dependency is enough as we're using WRITE_ONCE to | |
1364 | * fill the cq entry | |
1365 | */ | |
888aae2e | 1366 | if (__io_cqring_events(ctx) == rings->cq_ring_entries) |
2b188cc1 JA |
1367 | return NULL; |
1368 | ||
888aae2e | 1369 | tail = ctx->cached_cq_tail++; |
75b28aff | 1370 | return &rings->cqes[tail & ctx->cq_mask]; |
2b188cc1 JA |
1371 | } |
1372 | ||
f2842ab5 JA |
1373 | static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx) |
1374 | { | |
44c769de | 1375 | if (likely(!ctx->cq_ev_fd)) |
f0b493e6 | 1376 | return false; |
7e55a19c SG |
1377 | if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) |
1378 | return false; | |
44c769de | 1379 | return !ctx->eventfd_async || io_wq_current_is_worker(); |
f2842ab5 JA |
1380 | } |
1381 | ||
b41e9852 | 1382 | static void io_cqring_ev_posted(struct io_ring_ctx *ctx) |
1d7bb1d5 | 1383 | { |
b1445e59 PB |
1384 | /* see waitqueue_active() comment */ |
1385 | smp_mb(); | |
1386 | ||
1d7bb1d5 JA |
1387 | if (waitqueue_active(&ctx->wait)) |
1388 | wake_up(&ctx->wait); | |
534ca6d6 JA |
1389 | if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) |
1390 | wake_up(&ctx->sq_data->wait); | |
b41e9852 | 1391 | if (io_should_trigger_evfd(ctx)) |
1d7bb1d5 | 1392 | eventfd_signal(ctx->cq_ev_fd, 1); |
b1445e59 | 1393 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1394 | wake_up_interruptible(&ctx->cq_wait); |
1395 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1396 | } | |
1d7bb1d5 JA |
1397 | } |
1398 | ||
80c18e4a PB |
1399 | static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx) |
1400 | { | |
b1445e59 PB |
1401 | /* see waitqueue_active() comment */ |
1402 | smp_mb(); | |
1403 | ||
80c18e4a PB |
1404 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
1405 | if (waitqueue_active(&ctx->wait)) | |
1406 | wake_up(&ctx->wait); | |
1407 | } | |
1408 | if (io_should_trigger_evfd(ctx)) | |
1409 | eventfd_signal(ctx->cq_ev_fd, 1); | |
b1445e59 | 1410 | if (waitqueue_active(&ctx->cq_wait)) { |
4aa84f2f PB |
1411 | wake_up_interruptible(&ctx->cq_wait); |
1412 | kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); | |
1413 | } | |
80c18e4a PB |
1414 | } |
1415 | ||
c4a2ed72 | 1416 | /* Returns true if there are no backlogged entries after the flush */ |
6c2450ae | 1417 | static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
1d7bb1d5 JA |
1418 | { |
1419 | struct io_rings *rings = ctx->rings; | |
1d7bb1d5 | 1420 | unsigned long flags; |
b18032bb | 1421 | bool all_flushed, posted; |
1d7bb1d5 | 1422 | |
e23de15f PB |
1423 | if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries) |
1424 | return false; | |
1d7bb1d5 | 1425 | |
b18032bb | 1426 | posted = false; |
1d7bb1d5 | 1427 | spin_lock_irqsave(&ctx->completion_lock, flags); |
6c2450ae PB |
1428 | while (!list_empty(&ctx->cq_overflow_list)) { |
1429 | struct io_uring_cqe *cqe = io_get_cqring(ctx); | |
1430 | struct io_overflow_cqe *ocqe; | |
e6c8aa9a | 1431 | |
1d7bb1d5 JA |
1432 | if (!cqe && !force) |
1433 | break; | |
6c2450ae PB |
1434 | ocqe = list_first_entry(&ctx->cq_overflow_list, |
1435 | struct io_overflow_cqe, list); | |
1436 | if (cqe) | |
1437 | memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); | |
1438 | else | |
1d7bb1d5 | 1439 | WRITE_ONCE(ctx->rings->cq_overflow, |
6c2450ae | 1440 | ++ctx->cached_cq_overflow); |
b18032bb | 1441 | posted = true; |
6c2450ae PB |
1442 | list_del(&ocqe->list); |
1443 | kfree(ocqe); | |
1d7bb1d5 JA |
1444 | } |
1445 | ||
09e88404 PB |
1446 | all_flushed = list_empty(&ctx->cq_overflow_list); |
1447 | if (all_flushed) { | |
1448 | clear_bit(0, &ctx->sq_check_overflow); | |
1449 | clear_bit(0, &ctx->cq_check_overflow); | |
1450 | ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; | |
1451 | } | |
46930143 | 1452 | |
b18032bb JA |
1453 | if (posted) |
1454 | io_commit_cqring(ctx); | |
1d7bb1d5 | 1455 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
b18032bb JA |
1456 | if (posted) |
1457 | io_cqring_ev_posted(ctx); | |
09e88404 | 1458 | return all_flushed; |
1d7bb1d5 JA |
1459 | } |
1460 | ||
6c2450ae | 1461 | static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) |
6c503150 | 1462 | { |
ca0a2651 JA |
1463 | bool ret = true; |
1464 | ||
6c503150 PB |
1465 | if (test_bit(0, &ctx->cq_check_overflow)) { |
1466 | /* iopoll syncs against uring_lock, not completion_lock */ | |
1467 | if (ctx->flags & IORING_SETUP_IOPOLL) | |
1468 | mutex_lock(&ctx->uring_lock); | |
6c2450ae | 1469 | ret = __io_cqring_overflow_flush(ctx, force); |
6c503150 PB |
1470 | if (ctx->flags & IORING_SETUP_IOPOLL) |
1471 | mutex_unlock(&ctx->uring_lock); | |
1472 | } | |
ca0a2651 JA |
1473 | |
1474 | return ret; | |
6c503150 PB |
1475 | } |
1476 | ||
abc54d63 JA |
1477 | /* |
1478 | * Shamelessly stolen from the mm implementation of page reference checking, | |
1479 | * see commit f958d7b528b1 for details. | |
1480 | */ | |
1481 | #define req_ref_zero_or_close_to_overflow(req) \ | |
1482 | ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u) | |
1483 | ||
de9b4cca JA |
1484 | static inline bool req_ref_inc_not_zero(struct io_kiocb *req) |
1485 | { | |
abc54d63 | 1486 | return atomic_inc_not_zero(&req->refs); |
de9b4cca JA |
1487 | } |
1488 | ||
1489 | static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs) | |
1490 | { | |
abc54d63 JA |
1491 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1492 | return atomic_sub_and_test(refs, &req->refs); | |
de9b4cca JA |
1493 | } |
1494 | ||
1495 | static inline bool req_ref_put_and_test(struct io_kiocb *req) | |
1496 | { | |
abc54d63 JA |
1497 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1498 | return atomic_dec_and_test(&req->refs); | |
de9b4cca JA |
1499 | } |
1500 | ||
1501 | static inline void req_ref_put(struct io_kiocb *req) | |
1502 | { | |
abc54d63 | 1503 | WARN_ON_ONCE(req_ref_put_and_test(req)); |
de9b4cca JA |
1504 | } |
1505 | ||
1506 | static inline void req_ref_get(struct io_kiocb *req) | |
1507 | { | |
abc54d63 JA |
1508 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
1509 | atomic_inc(&req->refs); | |
de9b4cca JA |
1510 | } |
1511 | ||
8d13326e PB |
1512 | static bool io_cqring_event_overflow(struct io_kiocb *req, long res, |
1513 | unsigned int cflags) | |
2b188cc1 | 1514 | { |
78e19bbe | 1515 | struct io_ring_ctx *ctx = req->ctx; |
cce4b8b0 | 1516 | struct io_overflow_cqe *ocqe; |
2b188cc1 | 1517 | |
cce4b8b0 PB |
1518 | ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT); |
1519 | if (!ocqe) { | |
1520 | /* | |
1521 | * If we're in ring overflow flush mode, or in task cancel mode, | |
1522 | * or cannot allocate an overflow entry, then we need to drop it | |
1523 | * on the floor. | |
1524 | */ | |
1525 | WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow); | |
1526 | return false; | |
2b188cc1 | 1527 | } |
cce4b8b0 PB |
1528 | if (list_empty(&ctx->cq_overflow_list)) { |
1529 | set_bit(0, &ctx->sq_check_overflow); | |
1530 | set_bit(0, &ctx->cq_check_overflow); | |
1531 | ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; | |
1532 | } | |
1533 | ocqe->cqe.user_data = req->user_data; | |
1534 | ocqe->cqe.res = res; | |
1535 | ocqe->cqe.flags = cflags; | |
1536 | list_add_tail(&ocqe->list, &ctx->cq_overflow_list); | |
1537 | return true; | |
2b188cc1 JA |
1538 | } |
1539 | ||
8d13326e PB |
1540 | static inline bool __io_cqring_fill_event(struct io_kiocb *req, long res, |
1541 | unsigned int cflags) | |
1542 | { | |
1543 | struct io_ring_ctx *ctx = req->ctx; | |
1544 | struct io_uring_cqe *cqe; | |
1545 | ||
1546 | trace_io_uring_complete(ctx, req->user_data, res, cflags); | |
1547 | ||
1548 | /* | |
1549 | * If we can't get a cq entry, userspace overflowed the | |
1550 | * submission (by quite a lot). Increment the overflow count in | |
1551 | * the ring. | |
1552 | */ | |
1553 | cqe = io_get_cqring(ctx); | |
1554 | if (likely(cqe)) { | |
1555 | WRITE_ONCE(cqe->user_data, req->user_data); | |
1556 | WRITE_ONCE(cqe->res, res); | |
1557 | WRITE_ONCE(cqe->flags, cflags); | |
1558 | return true; | |
1559 | } | |
1560 | return io_cqring_event_overflow(req, res, cflags); | |
1561 | } | |
1562 | ||
1563 | /* not as hot to bloat with inlining */ | |
1564 | static noinline bool io_cqring_fill_event(struct io_kiocb *req, long res, | |
1565 | unsigned int cflags) | |
1566 | { | |
1567 | return __io_cqring_fill_event(req, res, cflags); | |
1568 | } | |
1569 | ||
7a612350 PB |
1570 | static void io_req_complete_post(struct io_kiocb *req, long res, |
1571 | unsigned int cflags) | |
2b188cc1 | 1572 | { |
78e19bbe | 1573 | struct io_ring_ctx *ctx = req->ctx; |
2b188cc1 JA |
1574 | unsigned long flags; |
1575 | ||
1576 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
8d13326e | 1577 | __io_cqring_fill_event(req, res, cflags); |
c7dae4ba JA |
1578 | /* |
1579 | * If we're the last reference to this request, add to our locked | |
1580 | * free_list cache. | |
1581 | */ | |
de9b4cca | 1582 | if (req_ref_put_and_test(req)) { |
c7dae4ba JA |
1583 | struct io_comp_state *cs = &ctx->submit_state.comp; |
1584 | ||
7a612350 PB |
1585 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
1586 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) | |
1587 | io_disarm_next(req); | |
1588 | if (req->link) { | |
1589 | io_req_task_queue(req->link); | |
1590 | req->link = NULL; | |
1591 | } | |
1592 | } | |
c7dae4ba JA |
1593 | io_dismantle_req(req); |
1594 | io_put_task(req->task, 1); | |
1595 | list_add(&req->compl.list, &cs->locked_free_list); | |
1596 | cs->locked_free_nr++; | |
180f829f PB |
1597 | } else { |
1598 | if (!percpu_ref_tryget(&ctx->refs)) | |
1599 | req = NULL; | |
1600 | } | |
7a612350 | 1601 | io_commit_cqring(ctx); |
2b188cc1 | 1602 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
7a612350 | 1603 | |
180f829f PB |
1604 | if (req) { |
1605 | io_cqring_ev_posted(ctx); | |
c7dae4ba | 1606 | percpu_ref_put(&ctx->refs); |
180f829f | 1607 | } |
229a7b63 JA |
1608 | } |
1609 | ||
a38d68db | 1610 | static void io_req_complete_state(struct io_kiocb *req, long res, |
889fca73 | 1611 | unsigned int cflags) |
229a7b63 | 1612 | { |
68fb8979 PB |
1613 | if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED)) |
1614 | io_clean_op(req); | |
a38d68db PB |
1615 | req->result = res; |
1616 | req->compl.cflags = cflags; | |
e342c807 | 1617 | req->flags |= REQ_F_COMPLETE_INLINE; |
e1e16097 JA |
1618 | } |
1619 | ||
889fca73 PB |
1620 | static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags, |
1621 | long res, unsigned cflags) | |
bcda7baa | 1622 | { |
889fca73 PB |
1623 | if (issue_flags & IO_URING_F_COMPLETE_DEFER) |
1624 | io_req_complete_state(req, res, cflags); | |
a38d68db | 1625 | else |
c7dae4ba | 1626 | io_req_complete_post(req, res, cflags); |
bcda7baa JA |
1627 | } |
1628 | ||
a38d68db | 1629 | static inline void io_req_complete(struct io_kiocb *req, long res) |
0ddf92e8 | 1630 | { |
889fca73 | 1631 | __io_req_complete(req, 0, res, 0); |
0ddf92e8 JA |
1632 | } |
1633 | ||
f41db273 PB |
1634 | static void io_req_complete_failed(struct io_kiocb *req, long res) |
1635 | { | |
1636 | req_set_fail_links(req); | |
1637 | io_put_req(req); | |
1638 | io_req_complete_post(req, res, 0); | |
1639 | } | |
1640 | ||
dac7a098 PB |
1641 | static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx, |
1642 | struct io_comp_state *cs) | |
1643 | { | |
1644 | spin_lock_irq(&ctx->completion_lock); | |
1645 | list_splice_init(&cs->locked_free_list, &cs->free_list); | |
1646 | cs->locked_free_nr = 0; | |
1647 | spin_unlock_irq(&ctx->completion_lock); | |
1648 | } | |
1649 | ||
dd78f492 | 1650 | /* Returns true IFF there are requests in the cache */ |
c7dae4ba | 1651 | static bool io_flush_cached_reqs(struct io_ring_ctx *ctx) |
0ddf92e8 | 1652 | { |
c7dae4ba JA |
1653 | struct io_submit_state *state = &ctx->submit_state; |
1654 | struct io_comp_state *cs = &state->comp; | |
dd78f492 | 1655 | int nr; |
0ddf92e8 | 1656 | |
c7dae4ba JA |
1657 | /* |
1658 | * If we have more than a batch's worth of requests in our IRQ side | |
1659 | * locked cache, grab the lock and move them over to our submission | |
1660 | * side cache. | |
1661 | */ | |
dac7a098 PB |
1662 | if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) |
1663 | io_flush_cached_locked_reqs(ctx, cs); | |
0ddf92e8 | 1664 | |
dd78f492 | 1665 | nr = state->free_reqs; |
c7dae4ba | 1666 | while (!list_empty(&cs->free_list)) { |
dd78f492 PB |
1667 | struct io_kiocb *req = list_first_entry(&cs->free_list, |
1668 | struct io_kiocb, compl.list); | |
1669 | ||
1b4c351f | 1670 | list_del(&req->compl.list); |
dd78f492 PB |
1671 | state->reqs[nr++] = req; |
1672 | if (nr == ARRAY_SIZE(state->reqs)) | |
e5d1bc0a | 1673 | break; |
1b4c351f JA |
1674 | } |
1675 | ||
dd78f492 PB |
1676 | state->free_reqs = nr; |
1677 | return nr != 0; | |
0ddf92e8 JA |
1678 | } |
1679 | ||
e5d1bc0a | 1680 | static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) |
2b188cc1 | 1681 | { |
e5d1bc0a PB |
1682 | struct io_submit_state *state = &ctx->submit_state; |
1683 | ||
1684 | BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs)); | |
1685 | ||
f6b6c7d6 | 1686 | if (!state->free_reqs) { |
291b2821 | 1687 | gfp_t gfp = GFP_KERNEL | __GFP_NOWARN; |
2579f913 JA |
1688 | int ret; |
1689 | ||
c7dae4ba | 1690 | if (io_flush_cached_reqs(ctx)) |
e5d1bc0a PB |
1691 | goto got_req; |
1692 | ||
bf019da7 PB |
1693 | ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH, |
1694 | state->reqs); | |
fd6fab2c JA |
1695 | |
1696 | /* | |
1697 | * Bulk alloc is all-or-nothing. If we fail to get a batch, | |
1698 | * retry single alloc to be on the safe side. | |
1699 | */ | |
1700 | if (unlikely(ret <= 0)) { | |
1701 | state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); | |
1702 | if (!state->reqs[0]) | |
3893f39f | 1703 | return NULL; |
fd6fab2c JA |
1704 | ret = 1; |
1705 | } | |
291b2821 | 1706 | state->free_reqs = ret; |
2b188cc1 | 1707 | } |
e5d1bc0a | 1708 | got_req: |
291b2821 PB |
1709 | state->free_reqs--; |
1710 | return state->reqs[state->free_reqs]; | |
2b188cc1 JA |
1711 | } |
1712 | ||
e1d767f0 | 1713 | static inline void io_put_file(struct file *file) |
8da11c19 | 1714 | { |
e1d767f0 | 1715 | if (file) |
8da11c19 PB |
1716 | fput(file); |
1717 | } | |
1718 | ||
4edf20f9 | 1719 | static void io_dismantle_req(struct io_kiocb *req) |
2b188cc1 | 1720 | { |
094bae49 PB |
1721 | unsigned int flags = req->flags; |
1722 | ||
e1d767f0 PB |
1723 | if (!(flags & REQ_F_FIXED_FILE)) |
1724 | io_put_file(req->file); | |
094bae49 PB |
1725 | if (flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED | |
1726 | REQ_F_INFLIGHT)) { | |
1727 | io_clean_op(req); | |
1728 | ||
1729 | if (req->flags & REQ_F_INFLIGHT) { | |
b303fe2e | 1730 | struct io_uring_task *tctx = req->task->io_uring; |
094bae49 | 1731 | |
b303fe2e | 1732 | atomic_dec(&tctx->inflight_tracked); |
094bae49 PB |
1733 | req->flags &= ~REQ_F_INFLIGHT; |
1734 | } | |
1735 | } | |
269bbe5f BM |
1736 | if (req->fixed_rsrc_refs) |
1737 | percpu_ref_put(req->fixed_rsrc_refs); | |
094bae49 PB |
1738 | if (req->async_data) |
1739 | kfree(req->async_data); | |
003e8dcc JA |
1740 | if (req->work.creds) { |
1741 | put_cred(req->work.creds); | |
1742 | req->work.creds = NULL; | |
1743 | } | |
e65ef56d JA |
1744 | } |
1745 | ||
b23fcf47 | 1746 | /* must to be called somewhat shortly after putting a request */ |
7c660731 PB |
1747 | static inline void io_put_task(struct task_struct *task, int nr) |
1748 | { | |
1749 | struct io_uring_task *tctx = task->io_uring; | |
1750 | ||
1751 | percpu_counter_sub(&tctx->inflight, nr); | |
1752 | if (unlikely(atomic_read(&tctx->in_idle))) | |
1753 | wake_up(&tctx->wait); | |
1754 | put_task_struct_many(task, nr); | |
1755 | } | |
1756 | ||
216578e5 | 1757 | static void __io_free_req(struct io_kiocb *req) |
c6ca97b3 | 1758 | { |
51a4cc11 | 1759 | struct io_ring_ctx *ctx = req->ctx; |
c6ca97b3 | 1760 | |
216578e5 | 1761 | io_dismantle_req(req); |
7c660731 | 1762 | io_put_task(req->task, 1); |
c6ca97b3 | 1763 | |
3893f39f | 1764 | kmem_cache_free(req_cachep, req); |
ecfc5177 | 1765 | percpu_ref_put(&ctx->refs); |
e65ef56d JA |
1766 | } |
1767 | ||
f2f87370 PB |
1768 | static inline void io_remove_next_linked(struct io_kiocb *req) |
1769 | { | |
1770 | struct io_kiocb *nxt = req->link; | |
1771 | ||
1772 | req->link = nxt->link; | |
1773 | nxt->link = NULL; | |
1774 | } | |
1775 | ||
33cc89a9 PB |
1776 | static bool io_kill_linked_timeout(struct io_kiocb *req) |
1777 | __must_hold(&req->ctx->completion_lock) | |
2665abfd | 1778 | { |
33cc89a9 | 1779 | struct io_kiocb *link = req->link; |
f2f87370 | 1780 | |
900fad45 PB |
1781 | /* |
1782 | * Can happen if a linked timeout fired and link had been like | |
1783 | * req -> link t-out -> link t-out [-> ...] | |
1784 | */ | |
c9abd7ad PB |
1785 | if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) { |
1786 | struct io_timeout_data *io = link->async_data; | |
7c86ffee | 1787 | |
f2f87370 | 1788 | io_remove_next_linked(req); |
90cd7e42 | 1789 | link->timeout.head = NULL; |
fd9c7bc5 | 1790 | if (hrtimer_try_to_cancel(&io->timer) != -1) { |
ff642164 | 1791 | io_cqring_fill_event(link, -ECANCELED, 0); |
33cc89a9 | 1792 | io_put_req_deferred(link, 1); |
d4729fbd | 1793 | return true; |
c9abd7ad PB |
1794 | } |
1795 | } | |
d4729fbd | 1796 | return false; |
7c86ffee PB |
1797 | } |
1798 | ||
d148ca4b | 1799 | static void io_fail_links(struct io_kiocb *req) |
33cc89a9 | 1800 | __must_hold(&req->ctx->completion_lock) |
9e645e11 | 1801 | { |
33cc89a9 | 1802 | struct io_kiocb *nxt, *link = req->link; |
9e645e11 | 1803 | |
f2f87370 | 1804 | req->link = NULL; |
f2f87370 PB |
1805 | while (link) { |
1806 | nxt = link->link; | |
1807 | link->link = NULL; | |
2665abfd | 1808 | |
f2f87370 | 1809 | trace_io_uring_fail_link(req, link); |
ff642164 | 1810 | io_cqring_fill_event(link, -ECANCELED, 0); |
1575f21a | 1811 | io_put_req_deferred(link, 2); |
f2f87370 | 1812 | link = nxt; |
9e645e11 | 1813 | } |
33cc89a9 | 1814 | } |
9e645e11 | 1815 | |
33cc89a9 PB |
1816 | static bool io_disarm_next(struct io_kiocb *req) |
1817 | __must_hold(&req->ctx->completion_lock) | |
1818 | { | |
1819 | bool posted = false; | |
1820 | ||
1821 | if (likely(req->flags & REQ_F_LINK_TIMEOUT)) | |
1822 | posted = io_kill_linked_timeout(req); | |
e4335ed3 PB |
1823 | if (unlikely((req->flags & REQ_F_FAIL_LINK) && |
1824 | !(req->flags & REQ_F_HARDLINK))) { | |
33cc89a9 PB |
1825 | posted |= (req->link != NULL); |
1826 | io_fail_links(req); | |
1827 | } | |
1828 | return posted; | |
9e645e11 JA |
1829 | } |
1830 | ||
3fa5e0f3 | 1831 | static struct io_kiocb *__io_req_find_next(struct io_kiocb *req) |
c69f8dbe | 1832 | { |
33cc89a9 | 1833 | struct io_kiocb *nxt; |
944e58bf | 1834 | |
9e645e11 JA |
1835 | /* |
1836 | * If LINK is set, we have dependent requests in this chain. If we | |
1837 | * didn't fail this request, queue the first one up, moving any other | |
1838 | * dependencies to the next request. In case of failure, fail the rest | |
1839 | * of the chain. | |
1840 | */ | |
33cc89a9 PB |
1841 | if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) { |
1842 | struct io_ring_ctx *ctx = req->ctx; | |
1843 | unsigned long flags; | |
1844 | bool posted; | |
1845 | ||
1846 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
1847 | posted = io_disarm_next(req); | |
1848 | if (posted) | |
1849 | io_commit_cqring(req->ctx); | |
1850 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
1851 | if (posted) | |
1852 | io_cqring_ev_posted(ctx); | |
f2f87370 | 1853 | } |
33cc89a9 PB |
1854 | nxt = req->link; |
1855 | req->link = NULL; | |
1856 | return nxt; | |
4d7dd462 | 1857 | } |
9e645e11 | 1858 | |
f2f87370 | 1859 | static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req) |
3fa5e0f3 | 1860 | { |
cdbff982 | 1861 | if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) |
3fa5e0f3 PB |
1862 | return NULL; |
1863 | return __io_req_find_next(req); | |
1864 | } | |
1865 | ||
2c32395d PB |
1866 | static void ctx_flush_and_put(struct io_ring_ctx *ctx) |
1867 | { | |
1868 | if (!ctx) | |
1869 | return; | |
1870 | if (ctx->submit_state.comp.nr) { | |
1871 | mutex_lock(&ctx->uring_lock); | |
1872 | io_submit_flush_completions(&ctx->submit_state.comp, ctx); | |
1873 | mutex_unlock(&ctx->uring_lock); | |
1874 | } | |
1875 | percpu_ref_put(&ctx->refs); | |
1876 | } | |
1877 | ||
7cbf1722 | 1878 | static bool __tctx_task_work(struct io_uring_task *tctx) |
c2c4c83c | 1879 | { |
65453d1e | 1880 | struct io_ring_ctx *ctx = NULL; |
7cbf1722 JA |
1881 | struct io_wq_work_list list; |
1882 | struct io_wq_work_node *node; | |
c2c4c83c | 1883 | |
7cbf1722 JA |
1884 | if (wq_list_empty(&tctx->task_list)) |
1885 | return false; | |
6200b0ae | 1886 | |
0b81e80c | 1887 | spin_lock_irq(&tctx->task_lock); |
7cbf1722 JA |
1888 | list = tctx->task_list; |
1889 | INIT_WQ_LIST(&tctx->task_list); | |
0b81e80c | 1890 | spin_unlock_irq(&tctx->task_lock); |
c2c4c83c | 1891 | |
7cbf1722 JA |
1892 | node = list.first; |
1893 | while (node) { | |
1894 | struct io_wq_work_node *next = node->next; | |
1895 | struct io_kiocb *req; | |
0ba9c9ed | 1896 | |
7cbf1722 | 1897 | req = container_of(node, struct io_kiocb, io_task_work.node); |
2c32395d PB |
1898 | if (req->ctx != ctx) { |
1899 | ctx_flush_and_put(ctx); | |
1900 | ctx = req->ctx; | |
1901 | percpu_ref_get(&ctx->refs); | |
65453d1e | 1902 | } |
65453d1e | 1903 | |
2c32395d PB |
1904 | req->task_work.func(&req->task_work); |
1905 | node = next; | |
7cbf1722 JA |
1906 | } |
1907 | ||
2c32395d | 1908 | ctx_flush_and_put(ctx); |
7cbf1722 | 1909 | return list.first != NULL; |
c2c4c83c JA |
1910 | } |
1911 | ||
7cbf1722 | 1912 | static void tctx_task_work(struct callback_head *cb) |
c40f6379 | 1913 | { |
7cbf1722 | 1914 | struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work); |
c40f6379 | 1915 | |
1d5f360d JA |
1916 | clear_bit(0, &tctx->task_state); |
1917 | ||
7cbf1722 JA |
1918 | while (__tctx_task_work(tctx)) |
1919 | cond_resched(); | |
7cbf1722 JA |
1920 | } |
1921 | ||
c15b79de | 1922 | static int io_req_task_work_add(struct io_kiocb *req) |
7cbf1722 | 1923 | { |
c15b79de | 1924 | struct task_struct *tsk = req->task; |
7cbf1722 | 1925 | struct io_uring_task *tctx = tsk->io_uring; |
c15b79de | 1926 | enum task_work_notify_mode notify; |
7cbf1722 | 1927 | struct io_wq_work_node *node, *prev; |
0b81e80c | 1928 | unsigned long flags; |
c15b79de PB |
1929 | int ret = 0; |
1930 | ||
1931 | if (unlikely(tsk->flags & PF_EXITING)) | |
1932 | return -ESRCH; | |
7cbf1722 JA |
1933 | |
1934 | WARN_ON_ONCE(!tctx); | |
1935 | ||
0b81e80c | 1936 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 | 1937 | wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); |
0b81e80c | 1938 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1939 | |
1940 | /* task_work already pending, we're done */ | |
1941 | if (test_bit(0, &tctx->task_state) || | |
1942 | test_and_set_bit(0, &tctx->task_state)) | |
1943 | return 0; | |
1944 | ||
c15b79de PB |
1945 | /* |
1946 | * SQPOLL kernel thread doesn't need notification, just a wakeup. For | |
1947 | * all other cases, use TWA_SIGNAL unconditionally to ensure we're | |
1948 | * processing task_work. There's no reliable way to tell if TWA_RESUME | |
1949 | * will do the job. | |
1950 | */ | |
1951 | notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; | |
1952 | ||
1953 | if (!task_work_add(tsk, &tctx->task_work, notify)) { | |
1954 | wake_up_process(tsk); | |
7cbf1722 | 1955 | return 0; |
c15b79de | 1956 | } |
7cbf1722 JA |
1957 | |
1958 | /* | |
1959 | * Slow path - we failed, find and delete work. if the work is not | |
1960 | * in the list, it got run and we're fine. | |
1961 | */ | |
0b81e80c | 1962 | spin_lock_irqsave(&tctx->task_lock, flags); |
7cbf1722 JA |
1963 | wq_list_for_each(node, prev, &tctx->task_list) { |
1964 | if (&req->io_task_work.node == node) { | |
1965 | wq_list_del(&tctx->task_list, node, prev); | |
1966 | ret = 1; | |
1967 | break; | |
1968 | } | |
1969 | } | |
0b81e80c | 1970 | spin_unlock_irqrestore(&tctx->task_lock, flags); |
7cbf1722 JA |
1971 | clear_bit(0, &tctx->task_state); |
1972 | return ret; | |
1973 | } | |
1974 | ||
9b465711 PB |
1975 | static bool io_run_task_work_head(struct callback_head **work_head) |
1976 | { | |
1977 | struct callback_head *work, *next; | |
1978 | bool executed = false; | |
1979 | ||
1980 | do { | |
1981 | work = xchg(work_head, NULL); | |
1982 | if (!work) | |
1983 | break; | |
1984 | ||
1985 | do { | |
1986 | next = work->next; | |
1987 | work->func(work); | |
1988 | work = next; | |
1989 | cond_resched(); | |
1990 | } while (work); | |
1991 | executed = true; | |
1992 | } while (1); | |
1993 | ||
1994 | return executed; | |
1995 | } | |
1996 | ||
1997 | static void io_task_work_add_head(struct callback_head **work_head, | |
1998 | struct callback_head *task_work) | |
eab30c4d | 1999 | { |
7c25c0d1 | 2000 | struct callback_head *head; |
eab30c4d | 2001 | |
7c25c0d1 | 2002 | do { |
9b465711 PB |
2003 | head = READ_ONCE(*work_head); |
2004 | task_work->next = head; | |
2005 | } while (cmpxchg(work_head, head, task_work) != head); | |
2006 | } | |
2007 | ||
2008 | static void io_req_task_work_add_fallback(struct io_kiocb *req, | |
2009 | task_work_func_t cb) | |
2010 | { | |
2011 | init_task_work(&req->task_work, cb); | |
2012 | io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work); | |
eab30c4d PB |
2013 | } |
2014 | ||
c40f6379 JA |
2015 | static void io_req_task_cancel(struct callback_head *cb) |
2016 | { | |
2017 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
87ceb6a6 | 2018 | struct io_ring_ctx *ctx = req->ctx; |
c40f6379 | 2019 | |
e83acd7d | 2020 | /* ctx is guaranteed to stay alive while we hold uring_lock */ |
792bb6eb | 2021 | mutex_lock(&ctx->uring_lock); |
2593553a | 2022 | io_req_complete_failed(req, req->result); |
792bb6eb | 2023 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2024 | } |
2025 | ||
2026 | static void __io_req_task_submit(struct io_kiocb *req) | |
2027 | { | |
2028 | struct io_ring_ctx *ctx = req->ctx; | |
2029 | ||
04fc6c80 | 2030 | /* ctx stays valid until unlock, even if we drop all ours ctx->refs */ |
81b6d05c | 2031 | mutex_lock(&ctx->uring_lock); |
70aacfe6 | 2032 | if (!(current->flags & PF_EXITING) && !current->in_execve) |
c5eef2b9 | 2033 | __io_queue_sqe(req); |
81b6d05c | 2034 | else |
2593553a | 2035 | io_req_complete_failed(req, -EFAULT); |
81b6d05c | 2036 | mutex_unlock(&ctx->uring_lock); |
c40f6379 JA |
2037 | } |
2038 | ||
2039 | static void io_req_task_submit(struct callback_head *cb) | |
2040 | { | |
2041 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2042 | ||
2043 | __io_req_task_submit(req); | |
2044 | } | |
2045 | ||
2c4b8eb6 | 2046 | static void io_req_task_queue_fail(struct io_kiocb *req, int ret) |
c40f6379 | 2047 | { |
2c4b8eb6 PB |
2048 | req->result = ret; |
2049 | req->task_work.func = io_req_task_cancel; | |
c40f6379 | 2050 | |
2c4b8eb6 | 2051 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2052 | io_req_task_work_add_fallback(req, io_req_task_cancel); |
c40f6379 JA |
2053 | } |
2054 | ||
2c4b8eb6 | 2055 | static void io_req_task_queue(struct io_kiocb *req) |
a3df7698 | 2056 | { |
2c4b8eb6 | 2057 | req->task_work.func = io_req_task_submit; |
a3df7698 PB |
2058 | |
2059 | if (unlikely(io_req_task_work_add(req))) | |
2c4b8eb6 | 2060 | io_req_task_queue_fail(req, -ECANCELED); |
a3df7698 PB |
2061 | } |
2062 | ||
f2f87370 | 2063 | static inline void io_queue_next(struct io_kiocb *req) |
c69f8dbe | 2064 | { |
9b5f7bd9 | 2065 | struct io_kiocb *nxt = io_req_find_next(req); |
944e58bf PB |
2066 | |
2067 | if (nxt) | |
906a8c3f | 2068 | io_req_task_queue(nxt); |
c69f8dbe JL |
2069 | } |
2070 | ||
c3524383 | 2071 | static void io_free_req(struct io_kiocb *req) |
7a743e22 | 2072 | { |
c3524383 PB |
2073 | io_queue_next(req); |
2074 | __io_free_req(req); | |
2075 | } | |
8766dd51 | 2076 | |
2d6500d4 | 2077 | struct req_batch { |
5af1d13e PB |
2078 | struct task_struct *task; |
2079 | int task_refs; | |
1b4c351f | 2080 | int ctx_refs; |
2d6500d4 PB |
2081 | }; |
2082 | ||
5af1d13e PB |
2083 | static inline void io_init_req_batch(struct req_batch *rb) |
2084 | { | |
5af1d13e | 2085 | rb->task_refs = 0; |
9ae72463 | 2086 | rb->ctx_refs = 0; |
5af1d13e PB |
2087 | rb->task = NULL; |
2088 | } | |
2089 | ||
2d6500d4 PB |
2090 | static void io_req_free_batch_finish(struct io_ring_ctx *ctx, |
2091 | struct req_batch *rb) | |
2092 | { | |
6e833d53 | 2093 | if (rb->task) |
7c660731 | 2094 | io_put_task(rb->task, rb->task_refs); |
9ae72463 PB |
2095 | if (rb->ctx_refs) |
2096 | percpu_ref_put_many(&ctx->refs, rb->ctx_refs); | |
2d6500d4 PB |
2097 | } |
2098 | ||
6ff119a6 PB |
2099 | static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req, |
2100 | struct io_submit_state *state) | |
2d6500d4 | 2101 | { |
f2f87370 | 2102 | io_queue_next(req); |
96670657 | 2103 | io_dismantle_req(req); |
2d6500d4 | 2104 | |
e3bc8e9d | 2105 | if (req->task != rb->task) { |
7c660731 PB |
2106 | if (rb->task) |
2107 | io_put_task(rb->task, rb->task_refs); | |
e3bc8e9d JA |
2108 | rb->task = req->task; |
2109 | rb->task_refs = 0; | |
5af1d13e | 2110 | } |
e3bc8e9d | 2111 | rb->task_refs++; |
9ae72463 | 2112 | rb->ctx_refs++; |
5af1d13e | 2113 | |
bd759045 | 2114 | if (state->free_reqs != ARRAY_SIZE(state->reqs)) |
6ff119a6 | 2115 | state->reqs[state->free_reqs++] = req; |
bd759045 PB |
2116 | else |
2117 | list_add(&req->compl.list, &state->comp.free_list); | |
7a743e22 PB |
2118 | } |
2119 | ||
905c172f PB |
2120 | static void io_submit_flush_completions(struct io_comp_state *cs, |
2121 | struct io_ring_ctx *ctx) | |
2122 | { | |
2123 | int i, nr = cs->nr; | |
2124 | struct io_kiocb *req; | |
2125 | struct req_batch rb; | |
2126 | ||
2127 | io_init_req_batch(&rb); | |
2128 | spin_lock_irq(&ctx->completion_lock); | |
2129 | for (i = 0; i < nr; i++) { | |
2130 | req = cs->reqs[i]; | |
8d13326e | 2131 | __io_cqring_fill_event(req, req->result, req->compl.cflags); |
905c172f PB |
2132 | } |
2133 | io_commit_cqring(ctx); | |
2134 | spin_unlock_irq(&ctx->completion_lock); | |
2135 | ||
2136 | io_cqring_ev_posted(ctx); | |
2137 | for (i = 0; i < nr; i++) { | |
2138 | req = cs->reqs[i]; | |
2139 | ||
2140 | /* submission and completion refs */ | |
de9b4cca | 2141 | if (req_ref_sub_and_test(req, 2)) |
6ff119a6 | 2142 | io_req_free_batch(&rb, req, &ctx->submit_state); |
905c172f PB |
2143 | } |
2144 | ||
2145 | io_req_free_batch_finish(ctx, &rb); | |
2146 | cs->nr = 0; | |
7a743e22 PB |
2147 | } |
2148 | ||
ba816ad6 JA |
2149 | /* |
2150 | * Drop reference to request, return next in chain (if there is one) if this | |
2151 | * was the last reference to this request. | |
2152 | */ | |
0d85035a | 2153 | static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req) |
e65ef56d | 2154 | { |
9b5f7bd9 PB |
2155 | struct io_kiocb *nxt = NULL; |
2156 | ||
de9b4cca | 2157 | if (req_ref_put_and_test(req)) { |
9b5f7bd9 | 2158 | nxt = io_req_find_next(req); |
4d7dd462 | 2159 | __io_free_req(req); |
2a44f467 | 2160 | } |
9b5f7bd9 | 2161 | return nxt; |
2b188cc1 JA |
2162 | } |
2163 | ||
0d85035a | 2164 | static inline void io_put_req(struct io_kiocb *req) |
e65ef56d | 2165 | { |
de9b4cca | 2166 | if (req_ref_put_and_test(req)) |
e65ef56d | 2167 | io_free_req(req); |
2b188cc1 JA |
2168 | } |
2169 | ||
216578e5 PB |
2170 | static void io_put_req_deferred_cb(struct callback_head *cb) |
2171 | { | |
2172 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
2173 | ||
2174 | io_free_req(req); | |
2175 | } | |
2176 | ||
2177 | static void io_free_req_deferred(struct io_kiocb *req) | |
2178 | { | |
7cbf1722 | 2179 | req->task_work.func = io_put_req_deferred_cb; |
a05432fb | 2180 | if (unlikely(io_req_task_work_add(req))) |
eab30c4d | 2181 | io_req_task_work_add_fallback(req, io_put_req_deferred_cb); |
216578e5 PB |
2182 | } |
2183 | ||
2184 | static inline void io_put_req_deferred(struct io_kiocb *req, int refs) | |
2185 | { | |
de9b4cca | 2186 | if (req_ref_sub_and_test(req, refs)) |
216578e5 PB |
2187 | io_free_req_deferred(req); |
2188 | } | |
2189 | ||
6c503150 | 2190 | static unsigned io_cqring_events(struct io_ring_ctx *ctx) |
a3a0e43f JA |
2191 | { |
2192 | /* See comment at the top of this file */ | |
2193 | smp_rmb(); | |
e23de15f | 2194 | return __io_cqring_events(ctx); |
a3a0e43f JA |
2195 | } |
2196 | ||
fb5ccc98 PB |
2197 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) |
2198 | { | |
2199 | struct io_rings *rings = ctx->rings; | |
2200 | ||
2201 | /* make sure SQ entry isn't read before tail */ | |
2202 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
2203 | } | |
2204 | ||
8ff069bf | 2205 | static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf) |
e94f141b | 2206 | { |
8ff069bf | 2207 | unsigned int cflags; |
e94f141b | 2208 | |
bcda7baa JA |
2209 | cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; |
2210 | cflags |= IORING_CQE_F_BUFFER; | |
0e1b6fe3 | 2211 | req->flags &= ~REQ_F_BUFFER_SELECTED; |
bcda7baa JA |
2212 | kfree(kbuf); |
2213 | return cflags; | |
e94f141b JA |
2214 | } |
2215 | ||
8ff069bf | 2216 | static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req) |
bcda7baa | 2217 | { |
4d954c25 | 2218 | struct io_buffer *kbuf; |
bcda7baa | 2219 | |
4d954c25 | 2220 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; |
8ff069bf PB |
2221 | return io_put_kbuf(req, kbuf); |
2222 | } | |
2223 | ||
4c6e277c JA |
2224 | static inline bool io_run_task_work(void) |
2225 | { | |
6200b0ae JA |
2226 | /* |
2227 | * Not safe to run on exiting task, and the task_work handling will | |
2228 | * not add work to such a task. | |
2229 | */ | |
2230 | if (unlikely(current->flags & PF_EXITING)) | |
2231 | return false; | |
4c6e277c JA |
2232 | if (current->task_works) { |
2233 | __set_current_state(TASK_RUNNING); | |
2234 | task_work_run(); | |
2235 | return true; | |
2236 | } | |
2237 | ||
2238 | return false; | |
bcda7baa JA |
2239 | } |
2240 | ||
def596e9 JA |
2241 | /* |
2242 | * Find and free completed poll iocbs | |
2243 | */ | |
2244 | static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2245 | struct list_head *done) | |
2246 | { | |
8237e045 | 2247 | struct req_batch rb; |
def596e9 | 2248 | struct io_kiocb *req; |
bbde017a XW |
2249 | |
2250 | /* order with ->result store in io_complete_rw_iopoll() */ | |
2251 | smp_rmb(); | |
def596e9 | 2252 | |
5af1d13e | 2253 | io_init_req_batch(&rb); |
def596e9 | 2254 | while (!list_empty(done)) { |
bcda7baa JA |
2255 | int cflags = 0; |
2256 | ||
d21ffe7e | 2257 | req = list_first_entry(done, struct io_kiocb, inflight_entry); |
f161340d PB |
2258 | list_del(&req->inflight_entry); |
2259 | ||
8c130827 PB |
2260 | if (READ_ONCE(req->result) == -EAGAIN && |
2261 | !(req->flags & REQ_F_DONT_REISSUE)) { | |
bbde017a | 2262 | req->iopoll_completed = 0; |
8c130827 PB |
2263 | req_ref_get(req); |
2264 | io_queue_async_work(req); | |
2265 | continue; | |
bbde017a | 2266 | } |
def596e9 | 2267 | |
bcda7baa | 2268 | if (req->flags & REQ_F_BUFFER_SELECTED) |
8ff069bf | 2269 | cflags = io_put_rw_kbuf(req); |
bcda7baa | 2270 | |
8d13326e | 2271 | __io_cqring_fill_event(req, req->result, cflags); |
def596e9 JA |
2272 | (*nr_events)++; |
2273 | ||
de9b4cca | 2274 | if (req_ref_put_and_test(req)) |
6ff119a6 | 2275 | io_req_free_batch(&rb, req, &ctx->submit_state); |
def596e9 | 2276 | } |
def596e9 | 2277 | |
09bb8394 | 2278 | io_commit_cqring(ctx); |
80c18e4a | 2279 | io_cqring_ev_posted_iopoll(ctx); |
2d6500d4 | 2280 | io_req_free_batch_finish(ctx, &rb); |
581f9810 BM |
2281 | } |
2282 | ||
def596e9 JA |
2283 | static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events, |
2284 | long min) | |
2285 | { | |
2286 | struct io_kiocb *req, *tmp; | |
2287 | LIST_HEAD(done); | |
2288 | bool spin; | |
2289 | int ret; | |
2290 | ||
2291 | /* | |
2292 | * Only spin for completions if we don't have multiple devices hanging | |
2293 | * off our complete list, and we're under the requested amount. | |
2294 | */ | |
2295 | spin = !ctx->poll_multi_file && *nr_events < min; | |
2296 | ||
2297 | ret = 0; | |
d21ffe7e | 2298 | list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { |
9adbd45d | 2299 | struct kiocb *kiocb = &req->rw.kiocb; |
def596e9 JA |
2300 | |
2301 | /* | |
581f9810 BM |
2302 | * Move completed and retryable entries to our local lists. |
2303 | * If we find a request that requires polling, break out | |
2304 | * and complete those lists first, if we have entries there. | |
def596e9 | 2305 | */ |
65a6543d | 2306 | if (READ_ONCE(req->iopoll_completed)) { |
d21ffe7e | 2307 | list_move_tail(&req->inflight_entry, &done); |
def596e9 JA |
2308 | continue; |
2309 | } | |
2310 | if (!list_empty(&done)) | |
2311 | break; | |
2312 | ||
2313 | ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); | |
2314 | if (ret < 0) | |
2315 | break; | |
2316 | ||
3aadc23e PB |
2317 | /* iopoll may have completed current req */ |
2318 | if (READ_ONCE(req->iopoll_completed)) | |
d21ffe7e | 2319 | list_move_tail(&req->inflight_entry, &done); |
3aadc23e | 2320 | |
def596e9 JA |
2321 | if (ret && spin) |
2322 | spin = false; | |
2323 | ret = 0; | |
2324 | } | |
2325 | ||
2326 | if (!list_empty(&done)) | |
2327 | io_iopoll_complete(ctx, nr_events, &done); | |
2328 | ||
2329 | return ret; | |
2330 | } | |
2331 | ||
2332 | /* | |
d195a66e | 2333 | * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a |
def596e9 JA |
2334 | * non-spinning poll check - we'll still enter the driver poll loop, but only |
2335 | * as a non-spinning completion check. | |
2336 | */ | |
2337 | static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events, | |
2338 | long min) | |
2339 | { | |
540e32a0 | 2340 | while (!list_empty(&ctx->iopoll_list) && !need_resched()) { |
def596e9 JA |
2341 | int ret; |
2342 | ||
2343 | ret = io_do_iopoll(ctx, nr_events, min); | |
2344 | if (ret < 0) | |
2345 | return ret; | |
eba0a4dd | 2346 | if (*nr_events >= min) |
def596e9 JA |
2347 | return 0; |
2348 | } | |
2349 | ||
2350 | return 1; | |
2351 | } | |
2352 | ||
2353 | /* | |
2354 | * We can't just wait for polled events to come to us, we have to actively | |
2355 | * find and complete them. | |
2356 | */ | |
b2edc0a7 | 2357 | static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx) |
def596e9 JA |
2358 | { |
2359 | if (!(ctx->flags & IORING_SETUP_IOPOLL)) | |
2360 | return; | |
2361 | ||
2362 | mutex_lock(&ctx->uring_lock); | |
540e32a0 | 2363 | while (!list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2364 | unsigned int nr_events = 0; |
2365 | ||
b2edc0a7 | 2366 | io_do_iopoll(ctx, &nr_events, 0); |
08f5439f | 2367 | |
b2edc0a7 PB |
2368 | /* let it sleep and repeat later if can't complete a request */ |
2369 | if (nr_events == 0) | |
2370 | break; | |
08f5439f JA |
2371 | /* |
2372 | * Ensure we allow local-to-the-cpu processing to take place, | |
2373 | * in this case we need to ensure that we reap all events. | |
3fcee5a6 | 2374 | * Also let task_work, etc. to progress by releasing the mutex |
08f5439f | 2375 | */ |
3fcee5a6 PB |
2376 | if (need_resched()) { |
2377 | mutex_unlock(&ctx->uring_lock); | |
2378 | cond_resched(); | |
2379 | mutex_lock(&ctx->uring_lock); | |
2380 | } | |
def596e9 JA |
2381 | } |
2382 | mutex_unlock(&ctx->uring_lock); | |
2383 | } | |
2384 | ||
7668b92a | 2385 | static int io_iopoll_check(struct io_ring_ctx *ctx, long min) |
def596e9 | 2386 | { |
7668b92a | 2387 | unsigned int nr_events = 0; |
e9979b36 | 2388 | int ret = 0; |
500f9fba | 2389 | |
c7849be9 XW |
2390 | /* |
2391 | * We disallow the app entering submit/complete with polling, but we | |
2392 | * still need to lock the ring to prevent racing with polled issue | |
2393 | * that got punted to a workqueue. | |
2394 | */ | |
2395 | mutex_lock(&ctx->uring_lock); | |
def596e9 | 2396 | do { |
a3a0e43f JA |
2397 | /* |
2398 | * Don't enter poll loop if we already have events pending. | |
2399 | * If we do, we can potentially be spinning for commands that | |
2400 | * already triggered a CQE (eg in error). | |
2401 | */ | |
6c503150 | 2402 | if (test_bit(0, &ctx->cq_check_overflow)) |
6c2450ae | 2403 | __io_cqring_overflow_flush(ctx, false); |
6c503150 | 2404 | if (io_cqring_events(ctx)) |
a3a0e43f JA |
2405 | break; |
2406 | ||
500f9fba JA |
2407 | /* |
2408 | * If a submit got punted to a workqueue, we can have the | |
2409 | * application entering polling for a command before it gets | |
2410 | * issued. That app will hold the uring_lock for the duration | |
2411 | * of the poll right here, so we need to take a breather every | |
2412 | * now and then to ensure that the issue has a chance to add | |
2413 | * the poll to the issued list. Otherwise we can spin here | |
2414 | * forever, while the workqueue is stuck trying to acquire the | |
2415 | * very same mutex. | |
2416 | */ | |
e9979b36 | 2417 | if (list_empty(&ctx->iopoll_list)) { |
500f9fba | 2418 | mutex_unlock(&ctx->uring_lock); |
4c6e277c | 2419 | io_run_task_work(); |
500f9fba | 2420 | mutex_lock(&ctx->uring_lock); |
e9979b36 PB |
2421 | |
2422 | if (list_empty(&ctx->iopoll_list)) | |
2423 | break; | |
500f9fba JA |
2424 | } |
2425 | ||
7668b92a | 2426 | ret = io_iopoll_getevents(ctx, &nr_events, min); |
def596e9 JA |
2427 | if (ret <= 0) |
2428 | break; | |
2429 | ret = 0; | |
7668b92a | 2430 | } while (min && !nr_events && !need_resched()); |
def596e9 | 2431 | |
500f9fba | 2432 | mutex_unlock(&ctx->uring_lock); |
def596e9 JA |
2433 | return ret; |
2434 | } | |
2435 | ||
491381ce | 2436 | static void kiocb_end_write(struct io_kiocb *req) |
2b188cc1 | 2437 | { |
491381ce JA |
2438 | /* |
2439 | * Tell lockdep we inherited freeze protection from submission | |
2440 | * thread. | |
2441 | */ | |
2442 | if (req->flags & REQ_F_ISREG) { | |
1c98679d | 2443 | struct super_block *sb = file_inode(req->file)->i_sb; |
2b188cc1 | 2444 | |
1c98679d PB |
2445 | __sb_writers_acquired(sb, SB_FREEZE_WRITE); |
2446 | sb_end_write(sb); | |
2b188cc1 JA |
2447 | } |
2448 | } | |
2449 | ||
b63534c4 | 2450 | #ifdef CONFIG_BLOCK |
dc2a6e9a | 2451 | static bool io_resubmit_prep(struct io_kiocb *req) |
b63534c4 | 2452 | { |
ab454438 PB |
2453 | struct io_async_rw *rw = req->async_data; |
2454 | ||
2455 | if (!rw) | |
2456 | return !io_req_prep_async(req); | |
2457 | /* may have left rw->iter inconsistent on -EIOCBQUEUED */ | |
2458 | iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter)); | |
2459 | return true; | |
b63534c4 | 2460 | } |
b63534c4 | 2461 | |
3e6a0d3c | 2462 | static bool io_rw_should_reissue(struct io_kiocb *req) |
b63534c4 | 2463 | { |
355afaeb | 2464 | umode_t mode = file_inode(req->file)->i_mode; |
3e6a0d3c | 2465 | struct io_ring_ctx *ctx = req->ctx; |
b63534c4 | 2466 | |
355afaeb JA |
2467 | if (!S_ISBLK(mode) && !S_ISREG(mode)) |
2468 | return false; | |
3e6a0d3c JA |
2469 | if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && |
2470 | !(ctx->flags & IORING_SETUP_IOPOLL))) | |
b63534c4 | 2471 | return false; |
7c977a58 JA |
2472 | /* |
2473 | * If ref is dying, we might be running poll reap from the exit work. | |
2474 | * Don't attempt to reissue from that path, just let it fail with | |
2475 | * -EAGAIN. | |
2476 | */ | |
3e6a0d3c JA |
2477 | if (percpu_ref_is_dying(&ctx->refs)) |
2478 | return false; | |
2479 | return true; | |
2480 | } | |
e82ad485 | 2481 | #else |
a1ff1e3f JA |
2482 | static bool io_resubmit_prep(struct io_kiocb *req) |
2483 | { | |
2484 | return false; | |
2485 | } | |
e82ad485 JA |
2486 | static bool io_rw_should_reissue(struct io_kiocb *req) |
2487 | { | |
2488 | return false; | |
2489 | } | |
3e6a0d3c JA |
2490 | #endif |
2491 | ||
a1d7c393 | 2492 | static void __io_complete_rw(struct io_kiocb *req, long res, long res2, |
889fca73 | 2493 | unsigned int issue_flags) |
a1d7c393 | 2494 | { |
2f8e45f1 PB |
2495 | int cflags = 0; |
2496 | ||
b65c128f PB |
2497 | if (req->rw.kiocb.ki_flags & IOCB_WRITE) |
2498 | kiocb_end_write(req); | |
9532b99b PB |
2499 | if (res != req->result) { |
2500 | if ((res == -EAGAIN || res == -EOPNOTSUPP) && | |
2501 | io_rw_should_reissue(req)) { | |
2502 | req->flags |= REQ_F_REISSUE; | |
2503 | return; | |
2504 | } | |
2f8e45f1 | 2505 | req_set_fail_links(req); |
9532b99b | 2506 | } |
2f8e45f1 PB |
2507 | if (req->flags & REQ_F_BUFFER_SELECTED) |
2508 | cflags = io_put_rw_kbuf(req); | |
2509 | __io_req_complete(req, issue_flags, res, cflags); | |
ba816ad6 JA |
2510 | } |
2511 | ||
2512 | static void io_complete_rw(struct kiocb *kiocb, long res, long res2) | |
2513 | { | |
9adbd45d | 2514 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
ba816ad6 | 2515 | |
889fca73 | 2516 | __io_complete_rw(req, res, res2, 0); |
2b188cc1 JA |
2517 | } |
2518 | ||
def596e9 JA |
2519 | static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2) |
2520 | { | |
9adbd45d | 2521 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
def596e9 | 2522 | |
491381ce JA |
2523 | if (kiocb->ki_flags & IOCB_WRITE) |
2524 | kiocb_end_write(req); | |
9532b99b | 2525 | if (unlikely(res != req->result)) { |
a1ff1e3f JA |
2526 | if (!(res == -EAGAIN && io_rw_should_reissue(req) && |
2527 | io_resubmit_prep(req))) { | |
9532b99b PB |
2528 | req_set_fail_links(req); |
2529 | req->flags |= REQ_F_DONT_REISSUE; | |
2530 | } | |
8c130827 | 2531 | } |
bbde017a XW |
2532 | |
2533 | WRITE_ONCE(req->result, res); | |
b9b0e0d3 | 2534 | /* order with io_iopoll_complete() checking ->result */ |
cd664b0e PB |
2535 | smp_wmb(); |
2536 | WRITE_ONCE(req->iopoll_completed, 1); | |
def596e9 JA |
2537 | } |
2538 | ||
2539 | /* | |
2540 | * After the iocb has been issued, it's safe to be found on the poll list. | |
2541 | * Adding the kiocb to the list AFTER submission ensures that we don't | |
2542 | * find it from a io_iopoll_getevents() thread before the issuer is done | |
2543 | * accessing the kiocb cookie. | |
2544 | */ | |
2e9dbe90 | 2545 | static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async) |
def596e9 JA |
2546 | { |
2547 | struct io_ring_ctx *ctx = req->ctx; | |
2548 | ||
2549 | /* | |
2550 | * Track whether we have multiple files in our lists. This will impact | |
2551 | * how we do polling eventually, not spinning if we're on potentially | |
2552 | * different devices. | |
2553 | */ | |
540e32a0 | 2554 | if (list_empty(&ctx->iopoll_list)) { |
def596e9 JA |
2555 | ctx->poll_multi_file = false; |
2556 | } else if (!ctx->poll_multi_file) { | |
2557 | struct io_kiocb *list_req; | |
2558 | ||
540e32a0 | 2559 | list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, |
d21ffe7e | 2560 | inflight_entry); |
9adbd45d | 2561 | if (list_req->file != req->file) |
def596e9 JA |
2562 | ctx->poll_multi_file = true; |
2563 | } | |
2564 | ||
2565 | /* | |
2566 | * For fast devices, IO may have already completed. If it has, add | |
2567 | * it to the front so we find it first. | |
2568 | */ | |
65a6543d | 2569 | if (READ_ONCE(req->iopoll_completed)) |
d21ffe7e | 2570 | list_add(&req->inflight_entry, &ctx->iopoll_list); |
def596e9 | 2571 | else |
d21ffe7e | 2572 | list_add_tail(&req->inflight_entry, &ctx->iopoll_list); |
bdcd3eab | 2573 | |
2e9dbe90 XW |
2574 | /* |
2575 | * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread | |
2576 | * task context or in io worker task context. If current task context is | |
2577 | * sq thread, we don't need to check whether should wake up sq thread. | |
2578 | */ | |
2579 | if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) && | |
534ca6d6 JA |
2580 | wq_has_sleeper(&ctx->sq_data->wait)) |
2581 | wake_up(&ctx->sq_data->wait); | |
def596e9 JA |
2582 | } |
2583 | ||
9f13c35b PB |
2584 | static inline void io_state_file_put(struct io_submit_state *state) |
2585 | { | |
02b23a9a PB |
2586 | if (state->file_refs) { |
2587 | fput_many(state->file, state->file_refs); | |
2588 | state->file_refs = 0; | |
2589 | } | |
9a56a232 JA |
2590 | } |
2591 | ||
2592 | /* | |
2593 | * Get as many references to a file as we have IOs left in this submission, | |
2594 | * assuming most submissions are for one file, or at least that each file | |
2595 | * has more than one submission. | |
2596 | */ | |
8da11c19 | 2597 | static struct file *__io_file_get(struct io_submit_state *state, int fd) |
9a56a232 JA |
2598 | { |
2599 | if (!state) | |
2600 | return fget(fd); | |
2601 | ||
6e1271e6 | 2602 | if (state->file_refs) { |
9a56a232 | 2603 | if (state->fd == fd) { |
6e1271e6 | 2604 | state->file_refs--; |
9a56a232 JA |
2605 | return state->file; |
2606 | } | |
02b23a9a | 2607 | io_state_file_put(state); |
9a56a232 JA |
2608 | } |
2609 | state->file = fget_many(fd, state->ios_left); | |
6e1271e6 | 2610 | if (unlikely(!state->file)) |
9a56a232 JA |
2611 | return NULL; |
2612 | ||
2613 | state->fd = fd; | |
6e1271e6 | 2614 | state->file_refs = state->ios_left - 1; |
9a56a232 JA |
2615 | return state->file; |
2616 | } | |
2617 | ||
4503b767 JA |
2618 | static bool io_bdev_nowait(struct block_device *bdev) |
2619 | { | |
9ba0d0c8 | 2620 | return !bdev || blk_queue_nowait(bdev_get_queue(bdev)); |
4503b767 JA |
2621 | } |
2622 | ||
2b188cc1 JA |
2623 | /* |
2624 | * If we tracked the file through the SCM inflight mechanism, we could support | |
2625 | * any file. For now, just ensure that anything potentially problematic is done | |
2626 | * inline. | |
2627 | */ | |
7b29f92d | 2628 | static bool __io_file_supports_async(struct file *file, int rw) |
2b188cc1 JA |
2629 | { |
2630 | umode_t mode = file_inode(file)->i_mode; | |
2631 | ||
4503b767 | 2632 | if (S_ISBLK(mode)) { |
4e7b5671 CH |
2633 | if (IS_ENABLED(CONFIG_BLOCK) && |
2634 | io_bdev_nowait(I_BDEV(file->f_mapping->host))) | |
4503b767 JA |
2635 | return true; |
2636 | return false; | |
2637 | } | |
2638 | if (S_ISCHR(mode) || S_ISSOCK(mode)) | |
2b188cc1 | 2639 | return true; |
4503b767 | 2640 | if (S_ISREG(mode)) { |
4e7b5671 CH |
2641 | if (IS_ENABLED(CONFIG_BLOCK) && |
2642 | io_bdev_nowait(file->f_inode->i_sb->s_bdev) && | |
4503b767 JA |
2643 | file->f_op != &io_uring_fops) |
2644 | return true; | |
2645 | return false; | |
2646 | } | |
2b188cc1 | 2647 | |
c5b85625 JA |
2648 | /* any ->read/write should understand O_NONBLOCK */ |
2649 | if (file->f_flags & O_NONBLOCK) | |
2650 | return true; | |
2651 | ||
af197f50 JA |
2652 | if (!(file->f_mode & FMODE_NOWAIT)) |
2653 | return false; | |
2654 | ||
2655 | if (rw == READ) | |
2656 | return file->f_op->read_iter != NULL; | |
2657 | ||
2658 | return file->f_op->write_iter != NULL; | |
2b188cc1 JA |
2659 | } |
2660 | ||
7b29f92d JA |
2661 | static bool io_file_supports_async(struct io_kiocb *req, int rw) |
2662 | { | |
2663 | if (rw == READ && (req->flags & REQ_F_ASYNC_READ)) | |
2664 | return true; | |
2665 | else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE)) | |
2666 | return true; | |
2667 | ||
2668 | return __io_file_supports_async(req->file, rw); | |
2669 | } | |
2670 | ||
a88fc400 | 2671 | static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
2b188cc1 | 2672 | { |
def596e9 | 2673 | struct io_ring_ctx *ctx = req->ctx; |
9adbd45d | 2674 | struct kiocb *kiocb = &req->rw.kiocb; |
75c668cd | 2675 | struct file *file = req->file; |
09bb8394 JA |
2676 | unsigned ioprio; |
2677 | int ret; | |
2b188cc1 | 2678 | |
7b29f92d | 2679 | if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode)) |
491381ce JA |
2680 | req->flags |= REQ_F_ISREG; |
2681 | ||
2b188cc1 | 2682 | kiocb->ki_pos = READ_ONCE(sqe->off); |
75c668cd | 2683 | if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) { |
ba04291e | 2684 | req->flags |= REQ_F_CUR_POS; |
75c668cd | 2685 | kiocb->ki_pos = file->f_pos; |
ba04291e | 2686 | } |
2b188cc1 | 2687 | kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); |
3e577dcd PB |
2688 | kiocb->ki_flags = iocb_flags(kiocb->ki_filp); |
2689 | ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); | |
2690 | if (unlikely(ret)) | |
2691 | return ret; | |
2b188cc1 | 2692 | |
75c668cd PB |
2693 | /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */ |
2694 | if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK)) | |
2695 | req->flags |= REQ_F_NOWAIT; | |
2696 | ||
2b188cc1 JA |
2697 | ioprio = READ_ONCE(sqe->ioprio); |
2698 | if (ioprio) { | |
2699 | ret = ioprio_check_cap(ioprio); | |
2700 | if (ret) | |
09bb8394 | 2701 | return ret; |
2b188cc1 JA |
2702 | |
2703 | kiocb->ki_ioprio = ioprio; | |
2704 | } else | |
2705 | kiocb->ki_ioprio = get_current_ioprio(); | |
2706 | ||
def596e9 | 2707 | if (ctx->flags & IORING_SETUP_IOPOLL) { |
def596e9 JA |
2708 | if (!(kiocb->ki_flags & IOCB_DIRECT) || |
2709 | !kiocb->ki_filp->f_op->iopoll) | |
09bb8394 | 2710 | return -EOPNOTSUPP; |
2b188cc1 | 2711 | |
def596e9 JA |
2712 | kiocb->ki_flags |= IOCB_HIPRI; |
2713 | kiocb->ki_complete = io_complete_rw_iopoll; | |
65a6543d | 2714 | req->iopoll_completed = 0; |
def596e9 | 2715 | } else { |
09bb8394 JA |
2716 | if (kiocb->ki_flags & IOCB_HIPRI) |
2717 | return -EINVAL; | |
def596e9 JA |
2718 | kiocb->ki_complete = io_complete_rw; |
2719 | } | |
9adbd45d | 2720 | |
3529d8c2 JA |
2721 | req->rw.addr = READ_ONCE(sqe->addr); |
2722 | req->rw.len = READ_ONCE(sqe->len); | |
4f4eeba8 | 2723 | req->buf_index = READ_ONCE(sqe->buf_index); |
2b188cc1 | 2724 | return 0; |
2b188cc1 JA |
2725 | } |
2726 | ||
2727 | static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret) | |
2728 | { | |
2729 | switch (ret) { | |
2730 | case -EIOCBQUEUED: | |
2731 | break; | |
2732 | case -ERESTARTSYS: | |
2733 | case -ERESTARTNOINTR: | |
2734 | case -ERESTARTNOHAND: | |
2735 | case -ERESTART_RESTARTBLOCK: | |
2736 | /* | |
2737 | * We can't just restart the syscall, since previously | |
2738 | * submitted sqes may already be in progress. Just fail this | |
2739 | * IO with EINTR. | |
2740 | */ | |
2741 | ret = -EINTR; | |
df561f66 | 2742 | fallthrough; |
2b188cc1 JA |
2743 | default: |
2744 | kiocb->ki_complete(kiocb, ret, 0); | |
2745 | } | |
2746 | } | |
2747 | ||
a1d7c393 | 2748 | static void kiocb_done(struct kiocb *kiocb, ssize_t ret, |
889fca73 | 2749 | unsigned int issue_flags) |
ba816ad6 | 2750 | { |
ba04291e | 2751 | struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb); |
e8c2bc1f | 2752 | struct io_async_rw *io = req->async_data; |
97284637 | 2753 | bool check_reissue = kiocb->ki_complete == io_complete_rw; |
ba04291e | 2754 | |
227c0c96 | 2755 | /* add previously done IO, if any */ |
e8c2bc1f | 2756 | if (io && io->bytes_done > 0) { |
227c0c96 | 2757 | if (ret < 0) |
e8c2bc1f | 2758 | ret = io->bytes_done; |
227c0c96 | 2759 | else |
e8c2bc1f | 2760 | ret += io->bytes_done; |
227c0c96 JA |
2761 | } |
2762 | ||
ba04291e JA |
2763 | if (req->flags & REQ_F_CUR_POS) |
2764 | req->file->f_pos = kiocb->ki_pos; | |
bcaec089 | 2765 | if (ret >= 0 && kiocb->ki_complete == io_complete_rw) |
889fca73 | 2766 | __io_complete_rw(req, ret, 0, issue_flags); |
ba816ad6 JA |
2767 | else |
2768 | io_rw_done(kiocb, ret); | |
97284637 PB |
2769 | |
2770 | if (check_reissue && req->flags & REQ_F_REISSUE) { | |
2771 | req->flags &= ~REQ_F_REISSUE; | |
8c130827 PB |
2772 | if (!io_resubmit_prep(req)) { |
2773 | req_ref_get(req); | |
2774 | io_queue_async_work(req); | |
2775 | } else { | |
97284637 PB |
2776 | int cflags = 0; |
2777 | ||
2778 | req_set_fail_links(req); | |
2779 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2780 | cflags = io_put_rw_kbuf(req); | |
2781 | __io_req_complete(req, issue_flags, ret, cflags); | |
2782 | } | |
2783 | } | |
ba816ad6 JA |
2784 | } |
2785 | ||
847595de | 2786 | static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter) |
edafccee | 2787 | { |
9adbd45d JA |
2788 | struct io_ring_ctx *ctx = req->ctx; |
2789 | size_t len = req->rw.len; | |
edafccee | 2790 | struct io_mapped_ubuf *imu; |
4be1c615 | 2791 | u16 index, buf_index = req->buf_index; |
75769e3f | 2792 | u64 buf_end, buf_addr = req->rw.addr; |
edafccee | 2793 | size_t offset; |
edafccee | 2794 | |
edafccee JA |
2795 | if (unlikely(buf_index >= ctx->nr_user_bufs)) |
2796 | return -EFAULT; | |
edafccee JA |
2797 | index = array_index_nospec(buf_index, ctx->nr_user_bufs); |
2798 | imu = &ctx->user_bufs[index]; | |
9adbd45d | 2799 | buf_addr = req->rw.addr; |
edafccee | 2800 | |
75769e3f | 2801 | if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end))) |
edafccee JA |
2802 | return -EFAULT; |
2803 | /* not inside the mapped region */ | |
4751f53d | 2804 | if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) |
edafccee JA |
2805 | return -EFAULT; |
2806 | ||
2807 | /* | |
2808 | * May not be a start of buffer, set size appropriately | |
2809 | * and advance us to the beginning. | |
2810 | */ | |
2811 | offset = buf_addr - imu->ubuf; | |
2812 | iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); | |
bd11b3a3 JA |
2813 | |
2814 | if (offset) { | |
2815 | /* | |
2816 | * Don't use iov_iter_advance() here, as it's really slow for | |
2817 | * using the latter parts of a big fixed buffer - it iterates | |
2818 | * over each segment manually. We can cheat a bit here, because | |
2819 | * we know that: | |
2820 | * | |
2821 | * 1) it's a BVEC iter, we set it up | |
2822 | * 2) all bvecs are PAGE_SIZE in size, except potentially the | |
2823 | * first and last bvec | |
2824 | * | |
2825 | * So just find our index, and adjust the iterator afterwards. | |
2826 | * If the offset is within the first bvec (or the whole first | |
2827 | * bvec, just use iov_iter_advance(). This makes it easier | |
2828 | * since we can just skip the first segment, which may not | |
2829 | * be PAGE_SIZE aligned. | |
2830 | */ | |
2831 | const struct bio_vec *bvec = imu->bvec; | |
2832 | ||
2833 | if (offset <= bvec->bv_len) { | |
2834 | iov_iter_advance(iter, offset); | |
2835 | } else { | |
2836 | unsigned long seg_skip; | |
2837 | ||
2838 | /* skip first vec */ | |
2839 | offset -= bvec->bv_len; | |
2840 | seg_skip = 1 + (offset >> PAGE_SHIFT); | |
2841 | ||
2842 | iter->bvec = bvec + seg_skip; | |
2843 | iter->nr_segs -= seg_skip; | |
99c79f66 | 2844 | iter->count -= bvec->bv_len + offset; |
bd11b3a3 | 2845 | iter->iov_offset = offset & ~PAGE_MASK; |
bd11b3a3 JA |
2846 | } |
2847 | } | |
2848 | ||
847595de | 2849 | return 0; |
edafccee JA |
2850 | } |
2851 | ||
bcda7baa JA |
2852 | static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock) |
2853 | { | |
2854 | if (needs_lock) | |
2855 | mutex_unlock(&ctx->uring_lock); | |
2856 | } | |
2857 | ||
2858 | static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock) | |
2859 | { | |
2860 | /* | |
2861 | * "Normal" inline submissions always hold the uring_lock, since we | |
2862 | * grab it from the system call. Same is true for the SQPOLL offload. | |
2863 | * The only exception is when we've detached the request and issue it | |
2864 | * from an async worker thread, grab the lock for that case. | |
2865 | */ | |
2866 | if (needs_lock) | |
2867 | mutex_lock(&ctx->uring_lock); | |
2868 | } | |
2869 | ||
2870 | static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len, | |
2871 | int bgid, struct io_buffer *kbuf, | |
2872 | bool needs_lock) | |
2873 | { | |
2874 | struct io_buffer *head; | |
2875 | ||
2876 | if (req->flags & REQ_F_BUFFER_SELECTED) | |
2877 | return kbuf; | |
2878 | ||
2879 | io_ring_submit_lock(req->ctx, needs_lock); | |
2880 | ||
2881 | lockdep_assert_held(&req->ctx->uring_lock); | |
2882 | ||
9e15c3a0 | 2883 | head = xa_load(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2884 | if (head) { |
2885 | if (!list_empty(&head->list)) { | |
2886 | kbuf = list_last_entry(&head->list, struct io_buffer, | |
2887 | list); | |
2888 | list_del(&kbuf->list); | |
2889 | } else { | |
2890 | kbuf = head; | |
9e15c3a0 | 2891 | xa_erase(&req->ctx->io_buffers, bgid); |
bcda7baa JA |
2892 | } |
2893 | if (*len > kbuf->len) | |
2894 | *len = kbuf->len; | |
2895 | } else { | |
2896 | kbuf = ERR_PTR(-ENOBUFS); | |
2897 | } | |
2898 | ||
2899 | io_ring_submit_unlock(req->ctx, needs_lock); | |
2900 | ||
2901 | return kbuf; | |
2902 | } | |
2903 | ||
4d954c25 JA |
2904 | static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len, |
2905 | bool needs_lock) | |
2906 | { | |
2907 | struct io_buffer *kbuf; | |
4f4eeba8 | 2908 | u16 bgid; |
4d954c25 JA |
2909 | |
2910 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
4f4eeba8 | 2911 | bgid = req->buf_index; |
4d954c25 JA |
2912 | kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock); |
2913 | if (IS_ERR(kbuf)) | |
2914 | return kbuf; | |
2915 | req->rw.addr = (u64) (unsigned long) kbuf; | |
2916 | req->flags |= REQ_F_BUFFER_SELECTED; | |
2917 | return u64_to_user_ptr(kbuf->addr); | |
2918 | } | |
2919 | ||
2920 | #ifdef CONFIG_COMPAT | |
2921 | static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov, | |
2922 | bool needs_lock) | |
2923 | { | |
2924 | struct compat_iovec __user *uiov; | |
2925 | compat_ssize_t clen; | |
2926 | void __user *buf; | |
2927 | ssize_t len; | |
2928 | ||
2929 | uiov = u64_to_user_ptr(req->rw.addr); | |
2930 | if (!access_ok(uiov, sizeof(*uiov))) | |
2931 | return -EFAULT; | |
2932 | if (__get_user(clen, &uiov->iov_len)) | |
2933 | return -EFAULT; | |
2934 | if (clen < 0) | |
2935 | return -EINVAL; | |
2936 | ||
2937 | len = clen; | |
2938 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2939 | if (IS_ERR(buf)) | |
2940 | return PTR_ERR(buf); | |
2941 | iov[0].iov_base = buf; | |
2942 | iov[0].iov_len = (compat_size_t) len; | |
2943 | return 0; | |
2944 | } | |
2945 | #endif | |
2946 | ||
2947 | static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2948 | bool needs_lock) | |
2949 | { | |
2950 | struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); | |
2951 | void __user *buf; | |
2952 | ssize_t len; | |
2953 | ||
2954 | if (copy_from_user(iov, uiov, sizeof(*uiov))) | |
2955 | return -EFAULT; | |
2956 | ||
2957 | len = iov[0].iov_len; | |
2958 | if (len < 0) | |
2959 | return -EINVAL; | |
2960 | buf = io_rw_buffer_select(req, &len, needs_lock); | |
2961 | if (IS_ERR(buf)) | |
2962 | return PTR_ERR(buf); | |
2963 | iov[0].iov_base = buf; | |
2964 | iov[0].iov_len = len; | |
2965 | return 0; | |
2966 | } | |
2967 | ||
2968 | static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov, | |
2969 | bool needs_lock) | |
2970 | { | |
dddb3e26 JA |
2971 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
2972 | struct io_buffer *kbuf; | |
2973 | ||
2974 | kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; | |
2975 | iov[0].iov_base = u64_to_user_ptr(kbuf->addr); | |
2976 | iov[0].iov_len = kbuf->len; | |
4d954c25 | 2977 | return 0; |
dddb3e26 | 2978 | } |
dd201662 | 2979 | if (req->rw.len != 1) |
4d954c25 JA |
2980 | return -EINVAL; |
2981 | ||
2982 | #ifdef CONFIG_COMPAT | |
2983 | if (req->ctx->compat) | |
2984 | return io_compat_import(req, iov, needs_lock); | |
2985 | #endif | |
2986 | ||
2987 | return __io_iov_buffer_select(req, iov, needs_lock); | |
2988 | } | |
2989 | ||
847595de PB |
2990 | static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec, |
2991 | struct iov_iter *iter, bool needs_lock) | |
2b188cc1 | 2992 | { |
9adbd45d JA |
2993 | void __user *buf = u64_to_user_ptr(req->rw.addr); |
2994 | size_t sqe_len = req->rw.len; | |
847595de | 2995 | u8 opcode = req->opcode; |
4d954c25 | 2996 | ssize_t ret; |
edafccee | 2997 | |
7d009165 | 2998 | if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { |
edafccee | 2999 | *iovec = NULL; |
9adbd45d | 3000 | return io_import_fixed(req, rw, iter); |
edafccee | 3001 | } |
2b188cc1 | 3002 | |
bcda7baa | 3003 | /* buffer index only valid with fixed read/write, or buffer select */ |
4f4eeba8 | 3004 | if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) |
9adbd45d JA |
3005 | return -EINVAL; |
3006 | ||
3a6820f2 | 3007 | if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) { |
bcda7baa | 3008 | if (req->flags & REQ_F_BUFFER_SELECT) { |
4d954c25 | 3009 | buf = io_rw_buffer_select(req, &sqe_len, needs_lock); |
867a23ea | 3010 | if (IS_ERR(buf)) |
4d954c25 | 3011 | return PTR_ERR(buf); |
3f9d6441 | 3012 | req->rw.len = sqe_len; |
bcda7baa JA |
3013 | } |
3014 | ||
3a6820f2 JA |
3015 | ret = import_single_range(rw, buf, sqe_len, *iovec, iter); |
3016 | *iovec = NULL; | |
10fc72e4 | 3017 | return ret; |
3a6820f2 JA |
3018 | } |
3019 | ||
4d954c25 JA |
3020 | if (req->flags & REQ_F_BUFFER_SELECT) { |
3021 | ret = io_iov_buffer_select(req, *iovec, needs_lock); | |
847595de PB |
3022 | if (!ret) |
3023 | iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); | |
4d954c25 JA |
3024 | *iovec = NULL; |
3025 | return ret; | |
3026 | } | |
3027 | ||
89cd35c5 CH |
3028 | return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter, |
3029 | req->ctx->compat); | |
2b188cc1 JA |
3030 | } |
3031 | ||
0fef9483 JA |
3032 | static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb) |
3033 | { | |
5b09e37e | 3034 | return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; |
0fef9483 JA |
3035 | } |
3036 | ||
31b51510 | 3037 | /* |
32960613 JA |
3038 | * For files that don't have ->read_iter() and ->write_iter(), handle them |
3039 | * by looping over ->read() or ->write() manually. | |
31b51510 | 3040 | */ |
4017eb91 | 3041 | static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter) |
32960613 | 3042 | { |
4017eb91 JA |
3043 | struct kiocb *kiocb = &req->rw.kiocb; |
3044 | struct file *file = req->file; | |
32960613 JA |
3045 | ssize_t ret = 0; |
3046 | ||
3047 | /* | |
3048 | * Don't support polled IO through this interface, and we can't | |
3049 | * support non-blocking either. For the latter, this just causes | |
3050 | * the kiocb to be handled from an async context. | |
3051 | */ | |
3052 | if (kiocb->ki_flags & IOCB_HIPRI) | |
3053 | return -EOPNOTSUPP; | |
3054 | if (kiocb->ki_flags & IOCB_NOWAIT) | |
3055 | return -EAGAIN; | |
3056 | ||
3057 | while (iov_iter_count(iter)) { | |
311ae9e1 | 3058 | struct iovec iovec; |
32960613 JA |
3059 | ssize_t nr; |
3060 | ||
311ae9e1 PB |
3061 | if (!iov_iter_is_bvec(iter)) { |
3062 | iovec = iov_iter_iovec(iter); | |
3063 | } else { | |
4017eb91 JA |
3064 | iovec.iov_base = u64_to_user_ptr(req->rw.addr); |
3065 | iovec.iov_len = req->rw.len; | |
311ae9e1 PB |
3066 | } |
3067 | ||
32960613 JA |
3068 | if (rw == READ) { |
3069 | nr = file->f_op->read(file, iovec.iov_base, | |
0fef9483 | 3070 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3071 | } else { |
3072 | nr = file->f_op->write(file, iovec.iov_base, | |
0fef9483 | 3073 | iovec.iov_len, io_kiocb_ppos(kiocb)); |
32960613 JA |
3074 | } |
3075 | ||
3076 | if (nr < 0) { | |
3077 | if (!ret) | |
3078 | ret = nr; | |
3079 | break; | |
3080 | } | |
3081 | ret += nr; | |
3082 | if (nr != iovec.iov_len) | |
3083 | break; | |
4017eb91 JA |
3084 | req->rw.len -= nr; |
3085 | req->rw.addr += nr; | |
32960613 JA |
3086 | iov_iter_advance(iter, nr); |
3087 | } | |
3088 | ||
3089 | return ret; | |
3090 | } | |
3091 | ||
ff6165b2 JA |
3092 | static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec, |
3093 | const struct iovec *fast_iov, struct iov_iter *iter) | |
f67676d1 | 3094 | { |
e8c2bc1f | 3095 | struct io_async_rw *rw = req->async_data; |
b64e3444 | 3096 | |
ff6165b2 | 3097 | memcpy(&rw->iter, iter, sizeof(*iter)); |
afb87658 | 3098 | rw->free_iovec = iovec; |
227c0c96 | 3099 | rw->bytes_done = 0; |
ff6165b2 | 3100 | /* can only be fixed buffers, no need to do anything */ |
9c3a205c | 3101 | if (iov_iter_is_bvec(iter)) |
ff6165b2 | 3102 | return; |
b64e3444 | 3103 | if (!iovec) { |
ff6165b2 JA |
3104 | unsigned iov_off = 0; |
3105 | ||
3106 | rw->iter.iov = rw->fast_iov; | |
3107 | if (iter->iov != fast_iov) { | |
3108 | iov_off = iter->iov - fast_iov; | |
3109 | rw->iter.iov += iov_off; | |
3110 | } | |
3111 | if (rw->fast_iov != fast_iov) | |
3112 | memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, | |
45097dae | 3113 | sizeof(struct iovec) * iter->nr_segs); |
99bc4c38 PB |
3114 | } else { |
3115 | req->flags |= REQ_F_NEED_CLEANUP; | |
f67676d1 JA |
3116 | } |
3117 | } | |
3118 | ||
6cb78689 | 3119 | static inline int io_alloc_async_data(struct io_kiocb *req) |
3d9932a8 | 3120 | { |
e8c2bc1f JA |
3121 | WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); |
3122 | req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); | |
3123 | return req->async_data == NULL; | |
3d9932a8 XW |
3124 | } |
3125 | ||
ff6165b2 JA |
3126 | static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec, |
3127 | const struct iovec *fast_iov, | |
227c0c96 | 3128 | struct iov_iter *iter, bool force) |
b7bb4f7d | 3129 | { |
26f0505a | 3130 | if (!force && !io_op_defs[req->opcode].needs_async_setup) |
74566df3 | 3131 | return 0; |
e8c2bc1f | 3132 | if (!req->async_data) { |
6cb78689 | 3133 | if (io_alloc_async_data(req)) { |
6bf985dc | 3134 | kfree(iovec); |
5d204bcf | 3135 | return -ENOMEM; |
6bf985dc | 3136 | } |
b7bb4f7d | 3137 | |
ff6165b2 | 3138 | io_req_map_rw(req, iovec, fast_iov, iter); |
5d204bcf | 3139 | } |
b7bb4f7d | 3140 | return 0; |
f67676d1 JA |
3141 | } |
3142 | ||
73debe68 | 3143 | static inline int io_rw_prep_async(struct io_kiocb *req, int rw) |
c3e330a4 | 3144 | { |
e8c2bc1f | 3145 | struct io_async_rw *iorw = req->async_data; |
f4bff104 | 3146 | struct iovec *iov = iorw->fast_iov; |
847595de | 3147 | int ret; |
c3e330a4 | 3148 | |
2846c481 | 3149 | ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); |
c3e330a4 PB |
3150 | if (unlikely(ret < 0)) |
3151 | return ret; | |
3152 | ||
ab0b196c PB |
3153 | iorw->bytes_done = 0; |
3154 | iorw->free_iovec = iov; | |
3155 | if (iov) | |
3156 | req->flags |= REQ_F_NEED_CLEANUP; | |
c3e330a4 PB |
3157 | return 0; |
3158 | } | |
3159 | ||
73debe68 | 3160 | static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3161 | { |
3529d8c2 JA |
3162 | if (unlikely(!(req->file->f_mode & FMODE_READ))) |
3163 | return -EBADF; | |
93642ef8 | 3164 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3165 | } |
3166 | ||
c1dd91d1 JA |
3167 | /* |
3168 | * This is our waitqueue callback handler, registered through lock_page_async() | |
3169 | * when we initially tried to do the IO with the iocb armed our waitqueue. | |
3170 | * This gets called when the page is unlocked, and we generally expect that to | |
3171 | * happen when the page IO is completed and the page is now uptodate. This will | |
3172 | * queue a task_work based retry of the operation, attempting to copy the data | |
3173 | * again. If the latter fails because the page was NOT uptodate, then we will | |
3174 | * do a thread based blocking retry of the operation. That's the unexpected | |
3175 | * slow path. | |
3176 | */ | |
bcf5a063 JA |
3177 | static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode, |
3178 | int sync, void *arg) | |
3179 | { | |
3180 | struct wait_page_queue *wpq; | |
3181 | struct io_kiocb *req = wait->private; | |
bcf5a063 | 3182 | struct wait_page_key *key = arg; |
bcf5a063 JA |
3183 | |
3184 | wpq = container_of(wait, struct wait_page_queue, wait); | |
3185 | ||
cdc8fcb4 LT |
3186 | if (!wake_page_match(wpq, key)) |
3187 | return 0; | |
3188 | ||
c8d317aa | 3189 | req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; |
bcf5a063 JA |
3190 | list_del_init(&wait->entry); |
3191 | ||
bcf5a063 | 3192 | /* submit ref gets dropped, acquire a new one */ |
de9b4cca | 3193 | req_ref_get(req); |
921b9054 | 3194 | io_req_task_queue(req); |
bcf5a063 JA |
3195 | return 1; |
3196 | } | |
3197 | ||
c1dd91d1 JA |
3198 | /* |
3199 | * This controls whether a given IO request should be armed for async page | |
3200 | * based retry. If we return false here, the request is handed to the async | |
3201 | * worker threads for retry. If we're doing buffered reads on a regular file, | |
3202 | * we prepare a private wait_page_queue entry and retry the operation. This | |
3203 | * will either succeed because the page is now uptodate and unlocked, or it | |
3204 | * will register a callback when the page is unlocked at IO completion. Through | |
3205 | * that callback, io_uring uses task_work to setup a retry of the operation. | |
3206 | * That retry will attempt the buffered read again. The retry will generally | |
3207 | * succeed, or in rare cases where it fails, we then fall back to using the | |
3208 | * async worker threads for a blocking retry. | |
3209 | */ | |
227c0c96 | 3210 | static bool io_rw_should_retry(struct io_kiocb *req) |
f67676d1 | 3211 | { |
e8c2bc1f JA |
3212 | struct io_async_rw *rw = req->async_data; |
3213 | struct wait_page_queue *wait = &rw->wpq; | |
bcf5a063 | 3214 | struct kiocb *kiocb = &req->rw.kiocb; |
f67676d1 | 3215 | |
bcf5a063 JA |
3216 | /* never retry for NOWAIT, we just complete with -EAGAIN */ |
3217 | if (req->flags & REQ_F_NOWAIT) | |
3218 | return false; | |
f67676d1 | 3219 | |
227c0c96 | 3220 | /* Only for buffered IO */ |
3b2a4439 | 3221 | if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) |
bcf5a063 | 3222 | return false; |
3b2a4439 | 3223 | |
bcf5a063 JA |
3224 | /* |
3225 | * just use poll if we can, and don't attempt if the fs doesn't | |
3226 | * support callback based unlocks | |
3227 | */ | |
3228 | if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) | |
3229 | return false; | |
f67676d1 | 3230 | |
3b2a4439 JA |
3231 | wait->wait.func = io_async_buf_func; |
3232 | wait->wait.private = req; | |
3233 | wait->wait.flags = 0; | |
3234 | INIT_LIST_HEAD(&wait->wait.entry); | |
3235 | kiocb->ki_flags |= IOCB_WAITQ; | |
c8d317aa | 3236 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3b2a4439 | 3237 | kiocb->ki_waitq = wait; |
3b2a4439 | 3238 | return true; |
bcf5a063 JA |
3239 | } |
3240 | ||
3241 | static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter) | |
3242 | { | |
3243 | if (req->file->f_op->read_iter) | |
3244 | return call_read_iter(req->file, &req->rw.kiocb, iter); | |
2dd2111d | 3245 | else if (req->file->f_op->read) |
4017eb91 | 3246 | return loop_rw_iter(READ, req, iter); |
2dd2111d GH |
3247 | else |
3248 | return -EINVAL; | |
f67676d1 JA |
3249 | } |
3250 | ||
889fca73 | 3251 | static int io_read(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3252 | { |
3253 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3254 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3255 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3256 | struct io_async_rw *rw = req->async_data; |
227c0c96 | 3257 | ssize_t io_size, ret, ret2; |
45d189c6 | 3258 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ff6165b2 | 3259 | |
2846c481 | 3260 | if (rw) { |
e8c2bc1f | 3261 | iter = &rw->iter; |
2846c481 PB |
3262 | iovec = NULL; |
3263 | } else { | |
3264 | ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock); | |
3265 | if (ret < 0) | |
3266 | return ret; | |
3267 | } | |
632546c4 | 3268 | io_size = iov_iter_count(iter); |
fa15bafb | 3269 | req->result = io_size; |
2b188cc1 | 3270 | |
fd6c2e4c JA |
3271 | /* Ensure we clear previously set non-block flag */ |
3272 | if (!force_nonblock) | |
29de5f6a | 3273 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
a88fc400 PB |
3274 | else |
3275 | kiocb->ki_flags |= IOCB_NOWAIT; | |
3276 | ||
24c74678 | 3277 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3278 | if (force_nonblock && !io_file_supports_async(req, READ)) { |
6713e7a6 | 3279 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc | 3280 | return ret ?: -EAGAIN; |
6713e7a6 | 3281 | } |
9e645e11 | 3282 | |
632546c4 | 3283 | ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size); |
5ea5dd45 PB |
3284 | if (unlikely(ret)) { |
3285 | kfree(iovec); | |
3286 | return ret; | |
3287 | } | |
2b188cc1 | 3288 | |
227c0c96 | 3289 | ret = io_iter_do_read(req, iter); |
32960613 | 3290 | |
230d50d4 | 3291 | if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { |
6ad7f233 | 3292 | req->flags &= ~REQ_F_REISSUE; |
eefdf30f JA |
3293 | /* IOPOLL retry should happen for io-wq threads */ |
3294 | if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
f91daf56 | 3295 | goto done; |
75c668cd PB |
3296 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3297 | if (req->flags & REQ_F_NOWAIT) | |
355afaeb | 3298 | goto done; |
84216315 | 3299 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3300 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
f38c7e3a | 3301 | ret = 0; |
230d50d4 JA |
3302 | } else if (ret == -EIOCBQUEUED) { |
3303 | goto out_free; | |
7335e3bf | 3304 | } else if (ret <= 0 || ret == io_size || !force_nonblock || |
75c668cd | 3305 | (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) { |
7335e3bf | 3306 | /* read all, failed, already did sync or don't want to retry */ |
00d23d51 | 3307 | goto done; |
227c0c96 JA |
3308 | } |
3309 | ||
227c0c96 | 3310 | ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true); |
6bf985dc PB |
3311 | if (ret2) |
3312 | return ret2; | |
3313 | ||
fe1cdd55 | 3314 | iovec = NULL; |
e8c2bc1f | 3315 | rw = req->async_data; |
227c0c96 | 3316 | /* now use our persistent iterator, if we aren't already */ |
e8c2bc1f | 3317 | iter = &rw->iter; |
227c0c96 | 3318 | |
b23df91b PB |
3319 | do { |
3320 | io_size -= ret; | |
3321 | rw->bytes_done += ret; | |
3322 | /* if we can retry, do so with the callbacks armed */ | |
3323 | if (!io_rw_should_retry(req)) { | |
3324 | kiocb->ki_flags &= ~IOCB_WAITQ; | |
3325 | return -EAGAIN; | |
3326 | } | |
3327 | ||
3328 | /* | |
3329 | * Now retry read with the IOCB_WAITQ parts set in the iocb. If | |
3330 | * we get -EIOCBQUEUED, then we'll get a notification when the | |
3331 | * desired page gets unlocked. We can also get a partial read | |
3332 | * here, and if we do, then just retry at the new offset. | |
3333 | */ | |
3334 | ret = io_iter_do_read(req, iter); | |
3335 | if (ret == -EIOCBQUEUED) | |
3336 | return 0; | |
227c0c96 | 3337 | /* we got some bytes, but not all. retry. */ |
b5b0ecb7 | 3338 | kiocb->ki_flags &= ~IOCB_WAITQ; |
b23df91b | 3339 | } while (ret > 0 && ret < io_size); |
227c0c96 | 3340 | done: |
889fca73 | 3341 | kiocb_done(kiocb, ret, issue_flags); |
fe1cdd55 PB |
3342 | out_free: |
3343 | /* it's faster to check here then delegate to kfree */ | |
3344 | if (iovec) | |
3345 | kfree(iovec); | |
5ea5dd45 | 3346 | return 0; |
2b188cc1 JA |
3347 | } |
3348 | ||
73debe68 | 3349 | static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 3350 | { |
3529d8c2 JA |
3351 | if (unlikely(!(req->file->f_mode & FMODE_WRITE))) |
3352 | return -EBADF; | |
93642ef8 | 3353 | return io_prep_rw(req, sqe); |
f67676d1 JA |
3354 | } |
3355 | ||
889fca73 | 3356 | static int io_write(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3357 | { |
3358 | struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs; | |
9adbd45d | 3359 | struct kiocb *kiocb = &req->rw.kiocb; |
ff6165b2 | 3360 | struct iov_iter __iter, *iter = &__iter; |
e8c2bc1f | 3361 | struct io_async_rw *rw = req->async_data; |
fa15bafb | 3362 | ssize_t ret, ret2, io_size; |
45d189c6 | 3363 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
2b188cc1 | 3364 | |
2846c481 | 3365 | if (rw) { |
e8c2bc1f | 3366 | iter = &rw->iter; |
2846c481 PB |
3367 | iovec = NULL; |
3368 | } else { | |
3369 | ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock); | |
3370 | if (ret < 0) | |
3371 | return ret; | |
3372 | } | |
632546c4 | 3373 | io_size = iov_iter_count(iter); |
fa15bafb | 3374 | req->result = io_size; |
2b188cc1 | 3375 | |
fd6c2e4c JA |
3376 | /* Ensure we clear previously set non-block flag */ |
3377 | if (!force_nonblock) | |
a88fc400 PB |
3378 | kiocb->ki_flags &= ~IOCB_NOWAIT; |
3379 | else | |
3380 | kiocb->ki_flags |= IOCB_NOWAIT; | |
fd6c2e4c | 3381 | |
24c74678 | 3382 | /* If the file doesn't support async, just async punt */ |
7b29f92d | 3383 | if (force_nonblock && !io_file_supports_async(req, WRITE)) |
f67676d1 | 3384 | goto copy_iov; |
31b51510 | 3385 | |
10d59345 JA |
3386 | /* file path doesn't support NOWAIT for non-direct_IO */ |
3387 | if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && | |
3388 | (req->flags & REQ_F_ISREG)) | |
f67676d1 | 3389 | goto copy_iov; |
31b51510 | 3390 | |
632546c4 | 3391 | ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size); |
fa15bafb PB |
3392 | if (unlikely(ret)) |
3393 | goto out_free; | |
4ed734b0 | 3394 | |
fa15bafb PB |
3395 | /* |
3396 | * Open-code file_start_write here to grab freeze protection, | |
3397 | * which will be released by another thread in | |
3398 | * io_complete_rw(). Fool lockdep by telling it the lock got | |
3399 | * released so that it doesn't complain about the held lock when | |
3400 | * we return to userspace. | |
3401 | */ | |
3402 | if (req->flags & REQ_F_ISREG) { | |
8a3c84b6 | 3403 | sb_start_write(file_inode(req->file)->i_sb); |
fa15bafb PB |
3404 | __sb_writers_release(file_inode(req->file)->i_sb, |
3405 | SB_FREEZE_WRITE); | |
3406 | } | |
3407 | kiocb->ki_flags |= IOCB_WRITE; | |
4ed734b0 | 3408 | |
fa15bafb | 3409 | if (req->file->f_op->write_iter) |
ff6165b2 | 3410 | ret2 = call_write_iter(req->file, kiocb, iter); |
2dd2111d | 3411 | else if (req->file->f_op->write) |
4017eb91 | 3412 | ret2 = loop_rw_iter(WRITE, req, iter); |
2dd2111d GH |
3413 | else |
3414 | ret2 = -EINVAL; | |
4ed734b0 | 3415 | |
6ad7f233 PB |
3416 | if (req->flags & REQ_F_REISSUE) { |
3417 | req->flags &= ~REQ_F_REISSUE; | |
230d50d4 | 3418 | ret2 = -EAGAIN; |
6ad7f233 | 3419 | } |
230d50d4 | 3420 | |
fa15bafb PB |
3421 | /* |
3422 | * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just | |
3423 | * retry them without IOCB_NOWAIT. | |
3424 | */ | |
3425 | if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) | |
3426 | ret2 = -EAGAIN; | |
75c668cd PB |
3427 | /* no retry on NONBLOCK nor RWF_NOWAIT */ |
3428 | if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) | |
355afaeb | 3429 | goto done; |
fa15bafb | 3430 | if (!force_nonblock || ret2 != -EAGAIN) { |
eefdf30f JA |
3431 | /* IOPOLL retry should happen for io-wq threads */ |
3432 | if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) | |
3433 | goto copy_iov; | |
355afaeb | 3434 | done: |
889fca73 | 3435 | kiocb_done(kiocb, ret2, issue_flags); |
fa15bafb | 3436 | } else { |
f67676d1 | 3437 | copy_iov: |
84216315 | 3438 | /* some cases will consume bytes even on error returns */ |
632546c4 | 3439 | iov_iter_revert(iter, io_size - iov_iter_count(iter)); |
227c0c96 | 3440 | ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false); |
6bf985dc | 3441 | return ret ?: -EAGAIN; |
2b188cc1 | 3442 | } |
31b51510 | 3443 | out_free: |
f261c168 | 3444 | /* it's reportedly faster than delegating the null check to kfree() */ |
252917c3 | 3445 | if (iovec) |
6f2cc166 | 3446 | kfree(iovec); |
2b188cc1 JA |
3447 | return ret; |
3448 | } | |
3449 | ||
80a261fd JA |
3450 | static int io_renameat_prep(struct io_kiocb *req, |
3451 | const struct io_uring_sqe *sqe) | |
3452 | { | |
3453 | struct io_rename *ren = &req->rename; | |
3454 | const char __user *oldf, *newf; | |
3455 | ||
3456 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3457 | return -EBADF; | |
3458 | ||
3459 | ren->old_dfd = READ_ONCE(sqe->fd); | |
3460 | oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3461 | newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
3462 | ren->new_dfd = READ_ONCE(sqe->len); | |
3463 | ren->flags = READ_ONCE(sqe->rename_flags); | |
3464 | ||
3465 | ren->oldpath = getname(oldf); | |
3466 | if (IS_ERR(ren->oldpath)) | |
3467 | return PTR_ERR(ren->oldpath); | |
3468 | ||
3469 | ren->newpath = getname(newf); | |
3470 | if (IS_ERR(ren->newpath)) { | |
3471 | putname(ren->oldpath); | |
3472 | return PTR_ERR(ren->newpath); | |
3473 | } | |
3474 | ||
3475 | req->flags |= REQ_F_NEED_CLEANUP; | |
3476 | return 0; | |
3477 | } | |
3478 | ||
45d189c6 | 3479 | static int io_renameat(struct io_kiocb *req, unsigned int issue_flags) |
80a261fd JA |
3480 | { |
3481 | struct io_rename *ren = &req->rename; | |
3482 | int ret; | |
3483 | ||
45d189c6 | 3484 | if (issue_flags & IO_URING_F_NONBLOCK) |
80a261fd JA |
3485 | return -EAGAIN; |
3486 | ||
3487 | ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, | |
3488 | ren->newpath, ren->flags); | |
3489 | ||
3490 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3491 | if (ret < 0) | |
3492 | req_set_fail_links(req); | |
3493 | io_req_complete(req, ret); | |
3494 | return 0; | |
3495 | } | |
3496 | ||
14a1143b JA |
3497 | static int io_unlinkat_prep(struct io_kiocb *req, |
3498 | const struct io_uring_sqe *sqe) | |
3499 | { | |
3500 | struct io_unlink *un = &req->unlink; | |
3501 | const char __user *fname; | |
3502 | ||
3503 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) | |
3504 | return -EBADF; | |
3505 | ||
3506 | un->dfd = READ_ONCE(sqe->fd); | |
3507 | ||
3508 | un->flags = READ_ONCE(sqe->unlink_flags); | |
3509 | if (un->flags & ~AT_REMOVEDIR) | |
3510 | return -EINVAL; | |
3511 | ||
3512 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
3513 | un->filename = getname(fname); | |
3514 | if (IS_ERR(un->filename)) | |
3515 | return PTR_ERR(un->filename); | |
3516 | ||
3517 | req->flags |= REQ_F_NEED_CLEANUP; | |
3518 | return 0; | |
3519 | } | |
3520 | ||
45d189c6 | 3521 | static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags) |
14a1143b JA |
3522 | { |
3523 | struct io_unlink *un = &req->unlink; | |
3524 | int ret; | |
3525 | ||
45d189c6 | 3526 | if (issue_flags & IO_URING_F_NONBLOCK) |
14a1143b JA |
3527 | return -EAGAIN; |
3528 | ||
3529 | if (un->flags & AT_REMOVEDIR) | |
3530 | ret = do_rmdir(un->dfd, un->filename); | |
3531 | else | |
3532 | ret = do_unlinkat(un->dfd, un->filename); | |
3533 | ||
3534 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
3535 | if (ret < 0) | |
3536 | req_set_fail_links(req); | |
3537 | io_req_complete(req, ret); | |
3538 | return 0; | |
3539 | } | |
3540 | ||
36f4fa68 JA |
3541 | static int io_shutdown_prep(struct io_kiocb *req, |
3542 | const struct io_uring_sqe *sqe) | |
3543 | { | |
3544 | #if defined(CONFIG_NET) | |
3545 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
3546 | return -EINVAL; | |
3547 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || | |
3548 | sqe->buf_index) | |
3549 | return -EINVAL; | |
3550 | ||
3551 | req->shutdown.how = READ_ONCE(sqe->len); | |
3552 | return 0; | |
3553 | #else | |
3554 | return -EOPNOTSUPP; | |
3555 | #endif | |
3556 | } | |
3557 | ||
45d189c6 | 3558 | static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags) |
36f4fa68 JA |
3559 | { |
3560 | #if defined(CONFIG_NET) | |
3561 | struct socket *sock; | |
3562 | int ret; | |
3563 | ||
45d189c6 | 3564 | if (issue_flags & IO_URING_F_NONBLOCK) |
36f4fa68 JA |
3565 | return -EAGAIN; |
3566 | ||
48aba79b | 3567 | sock = sock_from_file(req->file); |
36f4fa68 | 3568 | if (unlikely(!sock)) |
48aba79b | 3569 | return -ENOTSOCK; |
36f4fa68 JA |
3570 | |
3571 | ret = __sys_shutdown_sock(sock, req->shutdown.how); | |
a146468d JA |
3572 | if (ret < 0) |
3573 | req_set_fail_links(req); | |
36f4fa68 JA |
3574 | io_req_complete(req, ret); |
3575 | return 0; | |
3576 | #else | |
3577 | return -EOPNOTSUPP; | |
3578 | #endif | |
3579 | } | |
3580 | ||
f2a8d5c7 PB |
3581 | static int __io_splice_prep(struct io_kiocb *req, |
3582 | const struct io_uring_sqe *sqe) | |
7d67af2c PB |
3583 | { |
3584 | struct io_splice* sp = &req->splice; | |
3585 | unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL; | |
7d67af2c | 3586 | |
3232dd02 PB |
3587 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3588 | return -EINVAL; | |
7d67af2c PB |
3589 | |
3590 | sp->file_in = NULL; | |
7d67af2c PB |
3591 | sp->len = READ_ONCE(sqe->len); |
3592 | sp->flags = READ_ONCE(sqe->splice_flags); | |
3593 | ||
3594 | if (unlikely(sp->flags & ~valid_flags)) | |
3595 | return -EINVAL; | |
3596 | ||
8371adf5 PB |
3597 | sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in), |
3598 | (sp->flags & SPLICE_F_FD_IN_FIXED)); | |
3599 | if (!sp->file_in) | |
3600 | return -EBADF; | |
7d67af2c | 3601 | req->flags |= REQ_F_NEED_CLEANUP; |
7d67af2c PB |
3602 | return 0; |
3603 | } | |
3604 | ||
f2a8d5c7 PB |
3605 | static int io_tee_prep(struct io_kiocb *req, |
3606 | const struct io_uring_sqe *sqe) | |
3607 | { | |
3608 | if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) | |
3609 | return -EINVAL; | |
3610 | return __io_splice_prep(req, sqe); | |
3611 | } | |
3612 | ||
45d189c6 | 3613 | static int io_tee(struct io_kiocb *req, unsigned int issue_flags) |
f2a8d5c7 PB |
3614 | { |
3615 | struct io_splice *sp = &req->splice; | |
3616 | struct file *in = sp->file_in; | |
3617 | struct file *out = sp->file_out; | |
3618 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3619 | long ret = 0; | |
3620 | ||
45d189c6 | 3621 | if (issue_flags & IO_URING_F_NONBLOCK) |
f2a8d5c7 PB |
3622 | return -EAGAIN; |
3623 | if (sp->len) | |
3624 | ret = do_tee(in, out, sp->len, flags); | |
3625 | ||
e1d767f0 PB |
3626 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3627 | io_put_file(in); | |
f2a8d5c7 PB |
3628 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3629 | ||
f2a8d5c7 PB |
3630 | if (ret != sp->len) |
3631 | req_set_fail_links(req); | |
e1e16097 | 3632 | io_req_complete(req, ret); |
f2a8d5c7 PB |
3633 | return 0; |
3634 | } | |
3635 | ||
3636 | static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
3637 | { | |
3638 | struct io_splice* sp = &req->splice; | |
3639 | ||
3640 | sp->off_in = READ_ONCE(sqe->splice_off_in); | |
3641 | sp->off_out = READ_ONCE(sqe->off); | |
3642 | return __io_splice_prep(req, sqe); | |
3643 | } | |
3644 | ||
45d189c6 | 3645 | static int io_splice(struct io_kiocb *req, unsigned int issue_flags) |
7d67af2c PB |
3646 | { |
3647 | struct io_splice *sp = &req->splice; | |
3648 | struct file *in = sp->file_in; | |
3649 | struct file *out = sp->file_out; | |
3650 | unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; | |
3651 | loff_t *poff_in, *poff_out; | |
c9687426 | 3652 | long ret = 0; |
7d67af2c | 3653 | |
45d189c6 | 3654 | if (issue_flags & IO_URING_F_NONBLOCK) |
2fb3e822 | 3655 | return -EAGAIN; |
7d67af2c PB |
3656 | |
3657 | poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; | |
3658 | poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; | |
c9687426 | 3659 | |
948a7749 | 3660 | if (sp->len) |
c9687426 | 3661 | ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); |
7d67af2c | 3662 | |
e1d767f0 PB |
3663 | if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) |
3664 | io_put_file(in); | |
7d67af2c PB |
3665 | req->flags &= ~REQ_F_NEED_CLEANUP; |
3666 | ||
7d67af2c PB |
3667 | if (ret != sp->len) |
3668 | req_set_fail_links(req); | |
e1e16097 | 3669 | io_req_complete(req, ret); |
7d67af2c PB |
3670 | return 0; |
3671 | } | |
3672 | ||
2b188cc1 JA |
3673 | /* |
3674 | * IORING_OP_NOP just posts a completion event, nothing else. | |
3675 | */ | |
889fca73 | 3676 | static int io_nop(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 JA |
3677 | { |
3678 | struct io_ring_ctx *ctx = req->ctx; | |
2b188cc1 | 3679 | |
def596e9 JA |
3680 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
3681 | return -EINVAL; | |
3682 | ||
889fca73 | 3683 | __io_req_complete(req, issue_flags, 0, 0); |
2b188cc1 JA |
3684 | return 0; |
3685 | } | |
3686 | ||
1155c76a | 3687 | static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
c992fe29 | 3688 | { |
6b06314c | 3689 | struct io_ring_ctx *ctx = req->ctx; |
c992fe29 | 3690 | |
09bb8394 JA |
3691 | if (!req->file) |
3692 | return -EBADF; | |
c992fe29 | 3693 | |
6b06314c | 3694 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
def596e9 | 3695 | return -EINVAL; |
edafccee | 3696 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) |
c992fe29 CH |
3697 | return -EINVAL; |
3698 | ||
8ed8d3c3 JA |
3699 | req->sync.flags = READ_ONCE(sqe->fsync_flags); |
3700 | if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) | |
3701 | return -EINVAL; | |
3702 | ||
3703 | req->sync.off = READ_ONCE(sqe->off); | |
3704 | req->sync.len = READ_ONCE(sqe->len); | |
c992fe29 CH |
3705 | return 0; |
3706 | } | |
3707 | ||
45d189c6 | 3708 | static int io_fsync(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 3709 | { |
8ed8d3c3 | 3710 | loff_t end = req->sync.off + req->sync.len; |
8ed8d3c3 JA |
3711 | int ret; |
3712 | ||
ac45abc0 | 3713 | /* fsync always requires a blocking context */ |
45d189c6 | 3714 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
3715 | return -EAGAIN; |
3716 | ||
9adbd45d | 3717 | ret = vfs_fsync_range(req->file, req->sync.off, |
8ed8d3c3 JA |
3718 | end > 0 ? end : LLONG_MAX, |
3719 | req->sync.flags & IORING_FSYNC_DATASYNC); | |
3720 | if (ret < 0) | |
3721 | req_set_fail_links(req); | |
e1e16097 | 3722 | io_req_complete(req, ret); |
c992fe29 CH |
3723 | return 0; |
3724 | } | |
3725 | ||
d63d1b5e JA |
3726 | static int io_fallocate_prep(struct io_kiocb *req, |
3727 | const struct io_uring_sqe *sqe) | |
3728 | { | |
3729 | if (sqe->ioprio || sqe->buf_index || sqe->rw_flags) | |
3730 | return -EINVAL; | |
3232dd02 PB |
3731 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3732 | return -EINVAL; | |
d63d1b5e JA |
3733 | |
3734 | req->sync.off = READ_ONCE(sqe->off); | |
3735 | req->sync.len = READ_ONCE(sqe->addr); | |
3736 | req->sync.mode = READ_ONCE(sqe->len); | |
3737 | return 0; | |
3738 | } | |
3739 | ||
45d189c6 | 3740 | static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags) |
5d17b4a4 | 3741 | { |
ac45abc0 PB |
3742 | int ret; |
3743 | ||
d63d1b5e | 3744 | /* fallocate always requiring blocking context */ |
45d189c6 | 3745 | if (issue_flags & IO_URING_F_NONBLOCK) |
5d17b4a4 | 3746 | return -EAGAIN; |
ac45abc0 PB |
3747 | ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, |
3748 | req->sync.len); | |
ac45abc0 PB |
3749 | if (ret < 0) |
3750 | req_set_fail_links(req); | |
e1e16097 | 3751 | io_req_complete(req, ret); |
5d17b4a4 JA |
3752 | return 0; |
3753 | } | |
3754 | ||
ec65fea5 | 3755 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
b7bb4f7d | 3756 | { |
f8748881 | 3757 | const char __user *fname; |
15b71abe | 3758 | int ret; |
b7bb4f7d | 3759 | |
ec65fea5 | 3760 | if (unlikely(sqe->ioprio || sqe->buf_index)) |
15b71abe | 3761 | return -EINVAL; |
ec65fea5 | 3762 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
cf3040ca | 3763 | return -EBADF; |
03b1230c | 3764 | |
ec65fea5 PB |
3765 | /* open.how should be already initialised */ |
3766 | if (!(req->open.how.flags & O_PATH) && force_o_largefile()) | |
08a1d26e | 3767 | req->open.how.flags |= O_LARGEFILE; |
3529d8c2 | 3768 | |
25e72d10 PB |
3769 | req->open.dfd = READ_ONCE(sqe->fd); |
3770 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
f8748881 | 3771 | req->open.filename = getname(fname); |
15b71abe JA |
3772 | if (IS_ERR(req->open.filename)) { |
3773 | ret = PTR_ERR(req->open.filename); | |
3774 | req->open.filename = NULL; | |
3775 | return ret; | |
3776 | } | |
4022e7af | 3777 | req->open.nofile = rlimit(RLIMIT_NOFILE); |
8fef80bf | 3778 | req->flags |= REQ_F_NEED_CLEANUP; |
15b71abe | 3779 | return 0; |
03b1230c JA |
3780 | } |
3781 | ||
ec65fea5 PB |
3782 | static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
3783 | { | |
3784 | u64 flags, mode; | |
3785 | ||
14587a46 | 3786 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3787 | return -EINVAL; |
ec65fea5 PB |
3788 | mode = READ_ONCE(sqe->len); |
3789 | flags = READ_ONCE(sqe->open_flags); | |
3790 | req->open.how = build_open_how(flags, mode); | |
3791 | return __io_openat_prep(req, sqe); | |
3792 | } | |
3793 | ||
cebdb986 | 3794 | static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
aa1fa28f | 3795 | { |
cebdb986 | 3796 | struct open_how __user *how; |
cebdb986 | 3797 | size_t len; |
0fa03c62 JA |
3798 | int ret; |
3799 | ||
14587a46 | 3800 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4eb8dded | 3801 | return -EINVAL; |
cebdb986 JA |
3802 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
3803 | len = READ_ONCE(sqe->len); | |
cebdb986 JA |
3804 | if (len < OPEN_HOW_SIZE_VER0) |
3805 | return -EINVAL; | |
3529d8c2 | 3806 | |
cebdb986 JA |
3807 | ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, |
3808 | len); | |
3809 | if (ret) | |
3810 | return ret; | |
3529d8c2 | 3811 | |
ec65fea5 | 3812 | return __io_openat_prep(req, sqe); |
cebdb986 JA |
3813 | } |
3814 | ||
45d189c6 | 3815 | static int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
15b71abe JA |
3816 | { |
3817 | struct open_flags op; | |
15b71abe | 3818 | struct file *file; |
3a81fd02 JA |
3819 | bool nonblock_set; |
3820 | bool resolve_nonblock; | |
15b71abe JA |
3821 | int ret; |
3822 | ||
cebdb986 | 3823 | ret = build_open_flags(&req->open.how, &op); |
15b71abe JA |
3824 | if (ret) |
3825 | goto err; | |
3a81fd02 JA |
3826 | nonblock_set = op.open_flag & O_NONBLOCK; |
3827 | resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; | |
45d189c6 | 3828 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3a81fd02 JA |
3829 | /* |
3830 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, | |
3831 | * it'll always -EAGAIN | |
3832 | */ | |
3833 | if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) | |
3834 | return -EAGAIN; | |
3835 | op.lookup_flags |= LOOKUP_CACHED; | |
3836 | op.open_flag |= O_NONBLOCK; | |
3837 | } | |
15b71abe | 3838 | |
4022e7af | 3839 | ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); |
15b71abe JA |
3840 | if (ret < 0) |
3841 | goto err; | |
3842 | ||
3843 | file = do_filp_open(req->open.dfd, req->open.filename, &op); | |
3a81fd02 | 3844 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
45d189c6 PB |
3845 | if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) && |
3846 | file == ERR_PTR(-EAGAIN)) { | |
944d1444 | 3847 | /* |
3a81fd02 JA |
3848 | * We could hang on to this 'fd', but seems like marginal |
3849 | * gain for something that is now known to be a slower path. | |
3850 | * So just put it, and we'll get a new one when we retry. | |
944d1444 | 3851 | */ |
3a81fd02 JA |
3852 | put_unused_fd(ret); |
3853 | return -EAGAIN; | |
3854 | } | |
3855 | ||
15b71abe JA |
3856 | if (IS_ERR(file)) { |
3857 | put_unused_fd(ret); | |
3858 | ret = PTR_ERR(file); | |
3859 | } else { | |
45d189c6 | 3860 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
3a81fd02 | 3861 | file->f_flags &= ~O_NONBLOCK; |
15b71abe JA |
3862 | fsnotify_open(file); |
3863 | fd_install(ret, file); | |
3864 | } | |
3865 | err: | |
3866 | putname(req->open.filename); | |
8fef80bf | 3867 | req->flags &= ~REQ_F_NEED_CLEANUP; |
15b71abe JA |
3868 | if (ret < 0) |
3869 | req_set_fail_links(req); | |
0bdf3398 | 3870 | __io_req_complete(req, issue_flags, ret, 0); |
15b71abe JA |
3871 | return 0; |
3872 | } | |
3873 | ||
45d189c6 | 3874 | static int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
cebdb986 | 3875 | { |
e45cff58 | 3876 | return io_openat2(req, issue_flags); |
cebdb986 JA |
3877 | } |
3878 | ||
067524e9 JA |
3879 | static int io_remove_buffers_prep(struct io_kiocb *req, |
3880 | const struct io_uring_sqe *sqe) | |
3881 | { | |
3882 | struct io_provide_buf *p = &req->pbuf; | |
3883 | u64 tmp; | |
3884 | ||
3885 | if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off) | |
3886 | return -EINVAL; | |
3887 | ||
3888 | tmp = READ_ONCE(sqe->fd); | |
3889 | if (!tmp || tmp > USHRT_MAX) | |
3890 | return -EINVAL; | |
3891 | ||
3892 | memset(p, 0, sizeof(*p)); | |
3893 | p->nbufs = tmp; | |
3894 | p->bgid = READ_ONCE(sqe->buf_group); | |
3895 | return 0; | |
3896 | } | |
3897 | ||
3898 | static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf, | |
3899 | int bgid, unsigned nbufs) | |
3900 | { | |
3901 | unsigned i = 0; | |
3902 | ||
3903 | /* shouldn't happen */ | |
3904 | if (!nbufs) | |
3905 | return 0; | |
3906 | ||
3907 | /* the head kbuf is the list itself */ | |
3908 | while (!list_empty(&buf->list)) { | |
3909 | struct io_buffer *nxt; | |
3910 | ||
3911 | nxt = list_first_entry(&buf->list, struct io_buffer, list); | |
3912 | list_del(&nxt->list); | |
3913 | kfree(nxt); | |
3914 | if (++i == nbufs) | |
3915 | return i; | |
3916 | } | |
3917 | i++; | |
3918 | kfree(buf); | |
9e15c3a0 | 3919 | xa_erase(&ctx->io_buffers, bgid); |
067524e9 JA |
3920 | |
3921 | return i; | |
3922 | } | |
3923 | ||
889fca73 | 3924 | static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags) |
067524e9 JA |
3925 | { |
3926 | struct io_provide_buf *p = &req->pbuf; | |
3927 | struct io_ring_ctx *ctx = req->ctx; | |
3928 | struct io_buffer *head; | |
3929 | int ret = 0; | |
45d189c6 | 3930 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
067524e9 JA |
3931 | |
3932 | io_ring_submit_lock(ctx, !force_nonblock); | |
3933 | ||
3934 | lockdep_assert_held(&ctx->uring_lock); | |
3935 | ||
3936 | ret = -ENOENT; | |
9e15c3a0 | 3937 | head = xa_load(&ctx->io_buffers, p->bgid); |
067524e9 JA |
3938 | if (head) |
3939 | ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); | |
067524e9 JA |
3940 | if (ret < 0) |
3941 | req_set_fail_links(req); | |
067524e9 | 3942 | |
9fb8cb49 PB |
3943 | /* complete before unlock, IOPOLL may need the lock */ |
3944 | __io_req_complete(req, issue_flags, ret, 0); | |
3945 | io_ring_submit_unlock(ctx, !force_nonblock); | |
067524e9 JA |
3946 | return 0; |
3947 | } | |
3948 | ||
ddf0322d JA |
3949 | static int io_provide_buffers_prep(struct io_kiocb *req, |
3950 | const struct io_uring_sqe *sqe) | |
3951 | { | |
d81269fe | 3952 | unsigned long size; |
ddf0322d JA |
3953 | struct io_provide_buf *p = &req->pbuf; |
3954 | u64 tmp; | |
3955 | ||
3956 | if (sqe->ioprio || sqe->rw_flags) | |
3957 | return -EINVAL; | |
3958 | ||
3959 | tmp = READ_ONCE(sqe->fd); | |
3960 | if (!tmp || tmp > USHRT_MAX) | |
3961 | return -E2BIG; | |
3962 | p->nbufs = tmp; | |
3963 | p->addr = READ_ONCE(sqe->addr); | |
3964 | p->len = READ_ONCE(sqe->len); | |
3965 | ||
d81269fe PB |
3966 | size = (unsigned long)p->len * p->nbufs; |
3967 | if (!access_ok(u64_to_user_ptr(p->addr), size)) | |
ddf0322d JA |
3968 | return -EFAULT; |
3969 | ||
3970 | p->bgid = READ_ONCE(sqe->buf_group); | |
3971 | tmp = READ_ONCE(sqe->off); | |
3972 | if (tmp > USHRT_MAX) | |
3973 | return -E2BIG; | |
3974 | p->bid = tmp; | |
3975 | return 0; | |
3976 | } | |
3977 | ||
3978 | static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head) | |
3979 | { | |
3980 | struct io_buffer *buf; | |
3981 | u64 addr = pbuf->addr; | |
3982 | int i, bid = pbuf->bid; | |
3983 | ||
3984 | for (i = 0; i < pbuf->nbufs; i++) { | |
3985 | buf = kmalloc(sizeof(*buf), GFP_KERNEL); | |
3986 | if (!buf) | |
3987 | break; | |
3988 | ||
3989 | buf->addr = addr; | |
3990 | buf->len = pbuf->len; | |
3991 | buf->bid = bid; | |
3992 | addr += pbuf->len; | |
3993 | bid++; | |
3994 | if (!*head) { | |
3995 | INIT_LIST_HEAD(&buf->list); | |
3996 | *head = buf; | |
3997 | } else { | |
3998 | list_add_tail(&buf->list, &(*head)->list); | |
3999 | } | |
4000 | } | |
4001 | ||
4002 | return i ? i : -ENOMEM; | |
4003 | } | |
4004 | ||
889fca73 | 4005 | static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags) |
ddf0322d JA |
4006 | { |
4007 | struct io_provide_buf *p = &req->pbuf; | |
4008 | struct io_ring_ctx *ctx = req->ctx; | |
4009 | struct io_buffer *head, *list; | |
4010 | int ret = 0; | |
45d189c6 | 4011 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ddf0322d JA |
4012 | |
4013 | io_ring_submit_lock(ctx, !force_nonblock); | |
4014 | ||
4015 | lockdep_assert_held(&ctx->uring_lock); | |
4016 | ||
9e15c3a0 | 4017 | list = head = xa_load(&ctx->io_buffers, p->bgid); |
ddf0322d JA |
4018 | |
4019 | ret = io_add_buffers(p, &head); | |
9e15c3a0 JA |
4020 | if (ret >= 0 && !list) { |
4021 | ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL); | |
4022 | if (ret < 0) | |
067524e9 | 4023 | __io_remove_buffers(ctx, head, p->bgid, -1U); |
ddf0322d | 4024 | } |
ddf0322d JA |
4025 | if (ret < 0) |
4026 | req_set_fail_links(req); | |
9fb8cb49 PB |
4027 | /* complete before unlock, IOPOLL may need the lock */ |
4028 | __io_req_complete(req, issue_flags, ret, 0); | |
4029 | io_ring_submit_unlock(ctx, !force_nonblock); | |
ddf0322d | 4030 | return 0; |
cebdb986 JA |
4031 | } |
4032 | ||
3e4827b0 JA |
4033 | static int io_epoll_ctl_prep(struct io_kiocb *req, |
4034 | const struct io_uring_sqe *sqe) | |
4035 | { | |
4036 | #if defined(CONFIG_EPOLL) | |
4037 | if (sqe->ioprio || sqe->buf_index) | |
4038 | return -EINVAL; | |
6ca56f84 | 4039 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4040 | return -EINVAL; |
3e4827b0 JA |
4041 | |
4042 | req->epoll.epfd = READ_ONCE(sqe->fd); | |
4043 | req->epoll.op = READ_ONCE(sqe->len); | |
4044 | req->epoll.fd = READ_ONCE(sqe->off); | |
4045 | ||
4046 | if (ep_op_has_event(req->epoll.op)) { | |
4047 | struct epoll_event __user *ev; | |
4048 | ||
4049 | ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
4050 | if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) | |
4051 | return -EFAULT; | |
4052 | } | |
4053 | ||
4054 | return 0; | |
4055 | #else | |
4056 | return -EOPNOTSUPP; | |
4057 | #endif | |
4058 | } | |
4059 | ||
889fca73 | 4060 | static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags) |
3e4827b0 JA |
4061 | { |
4062 | #if defined(CONFIG_EPOLL) | |
4063 | struct io_epoll *ie = &req->epoll; | |
4064 | int ret; | |
45d189c6 | 4065 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
3e4827b0 JA |
4066 | |
4067 | ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); | |
4068 | if (force_nonblock && ret == -EAGAIN) | |
4069 | return -EAGAIN; | |
4070 | ||
4071 | if (ret < 0) | |
4072 | req_set_fail_links(req); | |
889fca73 | 4073 | __io_req_complete(req, issue_flags, ret, 0); |
3e4827b0 JA |
4074 | return 0; |
4075 | #else | |
4076 | return -EOPNOTSUPP; | |
4077 | #endif | |
4078 | } | |
4079 | ||
c1ca757b JA |
4080 | static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4081 | { | |
4082 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4083 | if (sqe->ioprio || sqe->buf_index || sqe->off) | |
4084 | return -EINVAL; | |
3232dd02 PB |
4085 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4086 | return -EINVAL; | |
c1ca757b JA |
4087 | |
4088 | req->madvise.addr = READ_ONCE(sqe->addr); | |
4089 | req->madvise.len = READ_ONCE(sqe->len); | |
4090 | req->madvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4091 | return 0; | |
4092 | #else | |
4093 | return -EOPNOTSUPP; | |
4094 | #endif | |
4095 | } | |
4096 | ||
45d189c6 | 4097 | static int io_madvise(struct io_kiocb *req, unsigned int issue_flags) |
c1ca757b JA |
4098 | { |
4099 | #if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU) | |
4100 | struct io_madvise *ma = &req->madvise; | |
4101 | int ret; | |
4102 | ||
45d189c6 | 4103 | if (issue_flags & IO_URING_F_NONBLOCK) |
c1ca757b JA |
4104 | return -EAGAIN; |
4105 | ||
0726b01e | 4106 | ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); |
c1ca757b JA |
4107 | if (ret < 0) |
4108 | req_set_fail_links(req); | |
e1e16097 | 4109 | io_req_complete(req, ret); |
c1ca757b JA |
4110 | return 0; |
4111 | #else | |
4112 | return -EOPNOTSUPP; | |
4113 | #endif | |
4114 | } | |
4115 | ||
4840e418 JA |
4116 | static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4117 | { | |
4118 | if (sqe->ioprio || sqe->buf_index || sqe->addr) | |
4119 | return -EINVAL; | |
3232dd02 PB |
4120 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4121 | return -EINVAL; | |
4840e418 JA |
4122 | |
4123 | req->fadvise.offset = READ_ONCE(sqe->off); | |
4124 | req->fadvise.len = READ_ONCE(sqe->len); | |
4125 | req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); | |
4126 | return 0; | |
4127 | } | |
4128 | ||
45d189c6 | 4129 | static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags) |
4840e418 JA |
4130 | { |
4131 | struct io_fadvise *fa = &req->fadvise; | |
4132 | int ret; | |
4133 | ||
45d189c6 | 4134 | if (issue_flags & IO_URING_F_NONBLOCK) { |
3e69426d JA |
4135 | switch (fa->advice) { |
4136 | case POSIX_FADV_NORMAL: | |
4137 | case POSIX_FADV_RANDOM: | |
4138 | case POSIX_FADV_SEQUENTIAL: | |
4139 | break; | |
4140 | default: | |
4141 | return -EAGAIN; | |
4142 | } | |
4143 | } | |
4840e418 JA |
4144 | |
4145 | ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); | |
4146 | if (ret < 0) | |
4147 | req_set_fail_links(req); | |
0bdf3398 | 4148 | __io_req_complete(req, issue_flags, ret, 0); |
4840e418 JA |
4149 | return 0; |
4150 | } | |
4151 | ||
eddc7ef5 JA |
4152 | static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4153 | { | |
6ca56f84 | 4154 | if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL))) |
3232dd02 | 4155 | return -EINVAL; |
eddc7ef5 JA |
4156 | if (sqe->ioprio || sqe->buf_index) |
4157 | return -EINVAL; | |
9c280f90 | 4158 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4159 | return -EBADF; |
eddc7ef5 | 4160 | |
1d9e1288 BM |
4161 | req->statx.dfd = READ_ONCE(sqe->fd); |
4162 | req->statx.mask = READ_ONCE(sqe->len); | |
e62753e4 | 4163 | req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
1d9e1288 BM |
4164 | req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
4165 | req->statx.flags = READ_ONCE(sqe->statx_flags); | |
eddc7ef5 JA |
4166 | |
4167 | return 0; | |
4168 | } | |
4169 | ||
45d189c6 | 4170 | static int io_statx(struct io_kiocb *req, unsigned int issue_flags) |
eddc7ef5 | 4171 | { |
1d9e1288 | 4172 | struct io_statx *ctx = &req->statx; |
eddc7ef5 JA |
4173 | int ret; |
4174 | ||
59d70013 | 4175 | if (issue_flags & IO_URING_F_NONBLOCK) |
eddc7ef5 JA |
4176 | return -EAGAIN; |
4177 | ||
e62753e4 BM |
4178 | ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, |
4179 | ctx->buffer); | |
eddc7ef5 | 4180 | |
eddc7ef5 JA |
4181 | if (ret < 0) |
4182 | req_set_fail_links(req); | |
e1e16097 | 4183 | io_req_complete(req, ret); |
eddc7ef5 JA |
4184 | return 0; |
4185 | } | |
4186 | ||
b5dba59e JA |
4187 | static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
4188 | { | |
14587a46 | 4189 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3232dd02 | 4190 | return -EINVAL; |
b5dba59e JA |
4191 | if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || |
4192 | sqe->rw_flags || sqe->buf_index) | |
4193 | return -EINVAL; | |
9c280f90 | 4194 | if (req->flags & REQ_F_FIXED_FILE) |
cf3040ca | 4195 | return -EBADF; |
b5dba59e JA |
4196 | |
4197 | req->close.fd = READ_ONCE(sqe->fd); | |
b5dba59e | 4198 | return 0; |
b5dba59e JA |
4199 | } |
4200 | ||
889fca73 | 4201 | static int io_close(struct io_kiocb *req, unsigned int issue_flags) |
b5dba59e | 4202 | { |
9eac1904 | 4203 | struct files_struct *files = current->files; |
3af73b28 | 4204 | struct io_close *close = &req->close; |
9eac1904 | 4205 | struct fdtable *fdt; |
a1fde923 PB |
4206 | struct file *file = NULL; |
4207 | int ret = -EBADF; | |
b5dba59e | 4208 | |
9eac1904 JA |
4209 | spin_lock(&files->file_lock); |
4210 | fdt = files_fdtable(files); | |
4211 | if (close->fd >= fdt->max_fds) { | |
4212 | spin_unlock(&files->file_lock); | |
4213 | goto err; | |
4214 | } | |
4215 | file = fdt->fd[close->fd]; | |
a1fde923 | 4216 | if (!file || file->f_op == &io_uring_fops) { |
9eac1904 JA |
4217 | spin_unlock(&files->file_lock); |
4218 | file = NULL; | |
4219 | goto err; | |
3af73b28 | 4220 | } |
b5dba59e JA |
4221 | |
4222 | /* if the file has a flush method, be safe and punt to async */ | |
45d189c6 | 4223 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
9eac1904 | 4224 | spin_unlock(&files->file_lock); |
0bf0eefd | 4225 | return -EAGAIN; |
a2100672 | 4226 | } |
b5dba59e | 4227 | |
9eac1904 JA |
4228 | ret = __close_fd_get_file(close->fd, &file); |
4229 | spin_unlock(&files->file_lock); | |
4230 | if (ret < 0) { | |
4231 | if (ret == -ENOENT) | |
4232 | ret = -EBADF; | |
4233 | goto err; | |
4234 | } | |
4235 | ||
3af73b28 | 4236 | /* No ->flush() or already async, safely close from here */ |
9eac1904 JA |
4237 | ret = filp_close(file, current->files); |
4238 | err: | |
3af73b28 PB |
4239 | if (ret < 0) |
4240 | req_set_fail_links(req); | |
9eac1904 JA |
4241 | if (file) |
4242 | fput(file); | |
889fca73 | 4243 | __io_req_complete(req, issue_flags, ret, 0); |
1a417f4e | 4244 | return 0; |
b5dba59e JA |
4245 | } |
4246 | ||
1155c76a | 4247 | static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
5d17b4a4 JA |
4248 | { |
4249 | struct io_ring_ctx *ctx = req->ctx; | |
5d17b4a4 | 4250 | |
5d17b4a4 JA |
4251 | if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) |
4252 | return -EINVAL; | |
4253 | if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index)) | |
4254 | return -EINVAL; | |
4255 | ||
8ed8d3c3 JA |
4256 | req->sync.off = READ_ONCE(sqe->off); |
4257 | req->sync.len = READ_ONCE(sqe->len); | |
4258 | req->sync.flags = READ_ONCE(sqe->sync_range_flags); | |
8ed8d3c3 JA |
4259 | return 0; |
4260 | } | |
4261 | ||
45d189c6 | 4262 | static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 | 4263 | { |
8ed8d3c3 JA |
4264 | int ret; |
4265 | ||
ac45abc0 | 4266 | /* sync_file_range always requires a blocking context */ |
45d189c6 | 4267 | if (issue_flags & IO_URING_F_NONBLOCK) |
ac45abc0 PB |
4268 | return -EAGAIN; |
4269 | ||
9adbd45d | 4270 | ret = sync_file_range(req->file, req->sync.off, req->sync.len, |
8ed8d3c3 JA |
4271 | req->sync.flags); |
4272 | if (ret < 0) | |
4273 | req_set_fail_links(req); | |
e1e16097 | 4274 | io_req_complete(req, ret); |
5d17b4a4 JA |
4275 | return 0; |
4276 | } | |
4277 | ||
469956e8 | 4278 | #if defined(CONFIG_NET) |
02d27d89 PB |
4279 | static int io_setup_async_msg(struct io_kiocb *req, |
4280 | struct io_async_msghdr *kmsg) | |
4281 | { | |
e8c2bc1f JA |
4282 | struct io_async_msghdr *async_msg = req->async_data; |
4283 | ||
4284 | if (async_msg) | |
02d27d89 | 4285 | return -EAGAIN; |
e8c2bc1f | 4286 | if (io_alloc_async_data(req)) { |
257e84a5 | 4287 | kfree(kmsg->free_iov); |
02d27d89 PB |
4288 | return -ENOMEM; |
4289 | } | |
e8c2bc1f | 4290 | async_msg = req->async_data; |
02d27d89 | 4291 | req->flags |= REQ_F_NEED_CLEANUP; |
e8c2bc1f | 4292 | memcpy(async_msg, kmsg, sizeof(*kmsg)); |
2a780802 | 4293 | async_msg->msg.msg_name = &async_msg->addr; |
257e84a5 PB |
4294 | /* if were using fast_iov, set it to the new one */ |
4295 | if (!async_msg->free_iov) | |
4296 | async_msg->msg.msg_iter.iov = async_msg->fast_iov; | |
4297 | ||
02d27d89 PB |
4298 | return -EAGAIN; |
4299 | } | |
4300 | ||
2ae523ed PB |
4301 | static int io_sendmsg_copy_hdr(struct io_kiocb *req, |
4302 | struct io_async_msghdr *iomsg) | |
4303 | { | |
2ae523ed | 4304 | iomsg->msg.msg_name = &iomsg->addr; |
257e84a5 | 4305 | iomsg->free_iov = iomsg->fast_iov; |
2ae523ed | 4306 | return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, |
257e84a5 | 4307 | req->sr_msg.msg_flags, &iomsg->free_iov); |
2ae523ed PB |
4308 | } |
4309 | ||
93642ef8 PB |
4310 | static int io_sendmsg_prep_async(struct io_kiocb *req) |
4311 | { | |
4312 | int ret; | |
4313 | ||
93642ef8 PB |
4314 | ret = io_sendmsg_copy_hdr(req, req->async_data); |
4315 | if (!ret) | |
4316 | req->flags |= REQ_F_NEED_CLEANUP; | |
4317 | return ret; | |
4318 | } | |
4319 | ||
3529d8c2 | 4320 | static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
03b1230c | 4321 | { |
e47293fd | 4322 | struct io_sr_msg *sr = &req->sr_msg; |
03b1230c | 4323 | |
d2b6f48b PB |
4324 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4325 | return -EINVAL; | |
4326 | ||
270a5940 | 4327 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
fddaface | 4328 | sr->len = READ_ONCE(sqe->len); |
04411806 PB |
4329 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4330 | if (sr->msg_flags & MSG_DONTWAIT) | |
4331 | req->flags |= REQ_F_NOWAIT; | |
3529d8c2 | 4332 | |
d8768362 JA |
4333 | #ifdef CONFIG_COMPAT |
4334 | if (req->ctx->compat) | |
4335 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4336 | #endif | |
93642ef8 | 4337 | return 0; |
03b1230c JA |
4338 | } |
4339 | ||
889fca73 | 4340 | static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4341 | { |
6b754c8b | 4342 | struct io_async_msghdr iomsg, *kmsg; |
0fa03c62 | 4343 | struct socket *sock; |
7a7cacba | 4344 | unsigned flags; |
0031275d | 4345 | int min_ret = 0; |
0fa03c62 JA |
4346 | int ret; |
4347 | ||
dba4a925 | 4348 | sock = sock_from_file(req->file); |
7a7cacba | 4349 | if (unlikely(!sock)) |
dba4a925 | 4350 | return -ENOTSOCK; |
3529d8c2 | 4351 | |
257e84a5 PB |
4352 | kmsg = req->async_data; |
4353 | if (!kmsg) { | |
7a7cacba PB |
4354 | ret = io_sendmsg_copy_hdr(req, &iomsg); |
4355 | if (ret) | |
4356 | return ret; | |
4357 | kmsg = &iomsg; | |
0fa03c62 | 4358 | } |
0fa03c62 | 4359 | |
04411806 PB |
4360 | flags = req->sr_msg.msg_flags; |
4361 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4362 | flags |= MSG_DONTWAIT; |
0031275d SM |
4363 | if (flags & MSG_WAITALL) |
4364 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4365 | ||
7a7cacba | 4366 | ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); |
45d189c6 | 4367 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4368 | return io_setup_async_msg(req, kmsg); |
4369 | if (ret == -ERESTARTSYS) | |
4370 | ret = -EINTR; | |
0fa03c62 | 4371 | |
257e84a5 PB |
4372 | /* fast path, check for non-NULL to avoid function call */ |
4373 | if (kmsg->free_iov) | |
4374 | kfree(kmsg->free_iov); | |
99bc4c38 | 4375 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4376 | if (ret < min_ret) |
4e88d6e7 | 4377 | req_set_fail_links(req); |
889fca73 | 4378 | __io_req_complete(req, issue_flags, ret, 0); |
5d17b4a4 | 4379 | return 0; |
03b1230c | 4380 | } |
aa1fa28f | 4381 | |
889fca73 | 4382 | static int io_send(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4383 | { |
7a7cacba PB |
4384 | struct io_sr_msg *sr = &req->sr_msg; |
4385 | struct msghdr msg; | |
4386 | struct iovec iov; | |
fddaface | 4387 | struct socket *sock; |
7a7cacba | 4388 | unsigned flags; |
0031275d | 4389 | int min_ret = 0; |
fddaface JA |
4390 | int ret; |
4391 | ||
dba4a925 | 4392 | sock = sock_from_file(req->file); |
7a7cacba | 4393 | if (unlikely(!sock)) |
dba4a925 | 4394 | return -ENOTSOCK; |
fddaface | 4395 | |
7a7cacba PB |
4396 | ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); |
4397 | if (unlikely(ret)) | |
14db8411 | 4398 | return ret; |
fddaface | 4399 | |
7a7cacba PB |
4400 | msg.msg_name = NULL; |
4401 | msg.msg_control = NULL; | |
4402 | msg.msg_controllen = 0; | |
4403 | msg.msg_namelen = 0; | |
fddaface | 4404 | |
04411806 PB |
4405 | flags = req->sr_msg.msg_flags; |
4406 | if (issue_flags & IO_URING_F_NONBLOCK) | |
7a7cacba | 4407 | flags |= MSG_DONTWAIT; |
0031275d SM |
4408 | if (flags & MSG_WAITALL) |
4409 | min_ret = iov_iter_count(&msg.msg_iter); | |
4410 | ||
7a7cacba PB |
4411 | msg.msg_flags = flags; |
4412 | ret = sock_sendmsg(sock, &msg); | |
45d189c6 | 4413 | if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) |
7a7cacba PB |
4414 | return -EAGAIN; |
4415 | if (ret == -ERESTARTSYS) | |
4416 | ret = -EINTR; | |
fddaface | 4417 | |
0031275d | 4418 | if (ret < min_ret) |
fddaface | 4419 | req_set_fail_links(req); |
889fca73 | 4420 | __io_req_complete(req, issue_flags, ret, 0); |
fddaface | 4421 | return 0; |
fddaface JA |
4422 | } |
4423 | ||
1400e697 PB |
4424 | static int __io_recvmsg_copy_hdr(struct io_kiocb *req, |
4425 | struct io_async_msghdr *iomsg) | |
52de1fe1 JA |
4426 | { |
4427 | struct io_sr_msg *sr = &req->sr_msg; | |
4428 | struct iovec __user *uiov; | |
4429 | size_t iov_len; | |
4430 | int ret; | |
4431 | ||
1400e697 PB |
4432 | ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, |
4433 | &iomsg->uaddr, &uiov, &iov_len); | |
52de1fe1 JA |
4434 | if (ret) |
4435 | return ret; | |
4436 | ||
4437 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4438 | if (iov_len > 1) | |
4439 | return -EINVAL; | |
5476dfed | 4440 | if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) |
52de1fe1 | 4441 | return -EFAULT; |
5476dfed | 4442 | sr->len = iomsg->fast_iov[0].iov_len; |
257e84a5 | 4443 | iomsg->free_iov = NULL; |
52de1fe1 | 4444 | } else { |
257e84a5 | 4445 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4446 | ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV, |
257e84a5 | 4447 | &iomsg->free_iov, &iomsg->msg.msg_iter, |
89cd35c5 | 4448 | false); |
52de1fe1 JA |
4449 | if (ret > 0) |
4450 | ret = 0; | |
4451 | } | |
4452 | ||
4453 | return ret; | |
4454 | } | |
4455 | ||
4456 | #ifdef CONFIG_COMPAT | |
4457 | static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req, | |
1400e697 | 4458 | struct io_async_msghdr *iomsg) |
52de1fe1 | 4459 | { |
52de1fe1 JA |
4460 | struct io_sr_msg *sr = &req->sr_msg; |
4461 | struct compat_iovec __user *uiov; | |
4462 | compat_uptr_t ptr; | |
4463 | compat_size_t len; | |
4464 | int ret; | |
4465 | ||
4af3417a PB |
4466 | ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, |
4467 | &ptr, &len); | |
52de1fe1 JA |
4468 | if (ret) |
4469 | return ret; | |
4470 | ||
4471 | uiov = compat_ptr(ptr); | |
4472 | if (req->flags & REQ_F_BUFFER_SELECT) { | |
4473 | compat_ssize_t clen; | |
4474 | ||
4475 | if (len > 1) | |
4476 | return -EINVAL; | |
4477 | if (!access_ok(uiov, sizeof(*uiov))) | |
4478 | return -EFAULT; | |
4479 | if (__get_user(clen, &uiov->iov_len)) | |
4480 | return -EFAULT; | |
4481 | if (clen < 0) | |
4482 | return -EINVAL; | |
2d280bc8 | 4483 | sr->len = clen; |
257e84a5 | 4484 | iomsg->free_iov = NULL; |
52de1fe1 | 4485 | } else { |
257e84a5 | 4486 | iomsg->free_iov = iomsg->fast_iov; |
89cd35c5 | 4487 | ret = __import_iovec(READ, (struct iovec __user *)uiov, len, |
257e84a5 | 4488 | UIO_FASTIOV, &iomsg->free_iov, |
89cd35c5 | 4489 | &iomsg->msg.msg_iter, true); |
52de1fe1 JA |
4490 | if (ret < 0) |
4491 | return ret; | |
4492 | } | |
4493 | ||
4494 | return 0; | |
4495 | } | |
4496 | #endif | |
4497 | ||
1400e697 PB |
4498 | static int io_recvmsg_copy_hdr(struct io_kiocb *req, |
4499 | struct io_async_msghdr *iomsg) | |
52de1fe1 | 4500 | { |
1400e697 | 4501 | iomsg->msg.msg_name = &iomsg->addr; |
52de1fe1 JA |
4502 | |
4503 | #ifdef CONFIG_COMPAT | |
4504 | if (req->ctx->compat) | |
1400e697 | 4505 | return __io_compat_recvmsg_copy_hdr(req, iomsg); |
fddaface | 4506 | #endif |
52de1fe1 | 4507 | |
1400e697 | 4508 | return __io_recvmsg_copy_hdr(req, iomsg); |
52de1fe1 JA |
4509 | } |
4510 | ||
bcda7baa | 4511 | static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req, |
7fbb1b54 | 4512 | bool needs_lock) |
bcda7baa JA |
4513 | { |
4514 | struct io_sr_msg *sr = &req->sr_msg; | |
4515 | struct io_buffer *kbuf; | |
4516 | ||
bcda7baa JA |
4517 | kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); |
4518 | if (IS_ERR(kbuf)) | |
4519 | return kbuf; | |
4520 | ||
4521 | sr->kbuf = kbuf; | |
4522 | req->flags |= REQ_F_BUFFER_SELECTED; | |
bcda7baa | 4523 | return kbuf; |
fddaface JA |
4524 | } |
4525 | ||
7fbb1b54 PB |
4526 | static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req) |
4527 | { | |
4528 | return io_put_kbuf(req, req->sr_msg.kbuf); | |
4529 | } | |
4530 | ||
93642ef8 | 4531 | static int io_recvmsg_prep_async(struct io_kiocb *req) |
aa1fa28f | 4532 | { |
99bc4c38 | 4533 | int ret; |
3529d8c2 | 4534 | |
93642ef8 PB |
4535 | ret = io_recvmsg_copy_hdr(req, req->async_data); |
4536 | if (!ret) | |
4537 | req->flags |= REQ_F_NEED_CLEANUP; | |
4538 | return ret; | |
4539 | } | |
4540 | ||
4541 | static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
4542 | { | |
4543 | struct io_sr_msg *sr = &req->sr_msg; | |
4544 | ||
d2b6f48b PB |
4545 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
4546 | return -EINVAL; | |
4547 | ||
270a5940 | 4548 | sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
0b7b21e4 | 4549 | sr->len = READ_ONCE(sqe->len); |
bcda7baa | 4550 | sr->bgid = READ_ONCE(sqe->buf_group); |
04411806 PB |
4551 | sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; |
4552 | if (sr->msg_flags & MSG_DONTWAIT) | |
4553 | req->flags |= REQ_F_NOWAIT; | |
06b76d44 | 4554 | |
d8768362 JA |
4555 | #ifdef CONFIG_COMPAT |
4556 | if (req->ctx->compat) | |
4557 | sr->msg_flags |= MSG_CMSG_COMPAT; | |
4558 | #endif | |
93642ef8 | 4559 | return 0; |
aa1fa28f JA |
4560 | } |
4561 | ||
889fca73 | 4562 | static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) |
aa1fa28f | 4563 | { |
6b754c8b | 4564 | struct io_async_msghdr iomsg, *kmsg; |
03b1230c | 4565 | struct socket *sock; |
7fbb1b54 | 4566 | struct io_buffer *kbuf; |
7a7cacba | 4567 | unsigned flags; |
0031275d | 4568 | int min_ret = 0; |
52de1fe1 | 4569 | int ret, cflags = 0; |
45d189c6 | 4570 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
03b1230c | 4571 | |
dba4a925 | 4572 | sock = sock_from_file(req->file); |
7a7cacba | 4573 | if (unlikely(!sock)) |
dba4a925 | 4574 | return -ENOTSOCK; |
3529d8c2 | 4575 | |
257e84a5 PB |
4576 | kmsg = req->async_data; |
4577 | if (!kmsg) { | |
7a7cacba PB |
4578 | ret = io_recvmsg_copy_hdr(req, &iomsg); |
4579 | if (ret) | |
681fda8d | 4580 | return ret; |
7a7cacba PB |
4581 | kmsg = &iomsg; |
4582 | } | |
03b1230c | 4583 | |
bc02ef33 | 4584 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4585 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bc02ef33 | 4586 | if (IS_ERR(kbuf)) |
52de1fe1 | 4587 | return PTR_ERR(kbuf); |
7a7cacba | 4588 | kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); |
5476dfed PB |
4589 | kmsg->fast_iov[0].iov_len = req->sr_msg.len; |
4590 | iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, | |
7a7cacba PB |
4591 | 1, req->sr_msg.len); |
4592 | } | |
52de1fe1 | 4593 | |
04411806 PB |
4594 | flags = req->sr_msg.msg_flags; |
4595 | if (force_nonblock) | |
7a7cacba | 4596 | flags |= MSG_DONTWAIT; |
0031275d SM |
4597 | if (flags & MSG_WAITALL) |
4598 | min_ret = iov_iter_count(&kmsg->msg.msg_iter); | |
4599 | ||
7a7cacba PB |
4600 | ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, |
4601 | kmsg->uaddr, flags); | |
0e1b6fe3 PB |
4602 | if (force_nonblock && ret == -EAGAIN) |
4603 | return io_setup_async_msg(req, kmsg); | |
7a7cacba PB |
4604 | if (ret == -ERESTARTSYS) |
4605 | ret = -EINTR; | |
03b1230c | 4606 | |
7fbb1b54 PB |
4607 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4608 | cflags = io_put_recv_kbuf(req); | |
257e84a5 PB |
4609 | /* fast path, check for non-NULL to avoid function call */ |
4610 | if (kmsg->free_iov) | |
4611 | kfree(kmsg->free_iov); | |
99bc4c38 | 4612 | req->flags &= ~REQ_F_NEED_CLEANUP; |
0031275d | 4613 | if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
4e88d6e7 | 4614 | req_set_fail_links(req); |
889fca73 | 4615 | __io_req_complete(req, issue_flags, ret, cflags); |
03b1230c | 4616 | return 0; |
0fa03c62 | 4617 | } |
5d17b4a4 | 4618 | |
889fca73 | 4619 | static int io_recv(struct io_kiocb *req, unsigned int issue_flags) |
fddaface | 4620 | { |
6b754c8b | 4621 | struct io_buffer *kbuf; |
7a7cacba PB |
4622 | struct io_sr_msg *sr = &req->sr_msg; |
4623 | struct msghdr msg; | |
4624 | void __user *buf = sr->buf; | |
fddaface | 4625 | struct socket *sock; |
7a7cacba PB |
4626 | struct iovec iov; |
4627 | unsigned flags; | |
0031275d | 4628 | int min_ret = 0; |
bcda7baa | 4629 | int ret, cflags = 0; |
45d189c6 | 4630 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
fddaface | 4631 | |
dba4a925 | 4632 | sock = sock_from_file(req->file); |
7a7cacba | 4633 | if (unlikely(!sock)) |
dba4a925 | 4634 | return -ENOTSOCK; |
fddaface | 4635 | |
bc02ef33 | 4636 | if (req->flags & REQ_F_BUFFER_SELECT) { |
7fbb1b54 | 4637 | kbuf = io_recv_buffer_select(req, !force_nonblock); |
bcda7baa JA |
4638 | if (IS_ERR(kbuf)) |
4639 | return PTR_ERR(kbuf); | |
7a7cacba | 4640 | buf = u64_to_user_ptr(kbuf->addr); |
bc02ef33 | 4641 | } |
bcda7baa | 4642 | |
7a7cacba | 4643 | ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); |
14c32eee PB |
4644 | if (unlikely(ret)) |
4645 | goto out_free; | |
fddaface | 4646 | |
7a7cacba PB |
4647 | msg.msg_name = NULL; |
4648 | msg.msg_control = NULL; | |
4649 | msg.msg_controllen = 0; | |
4650 | msg.msg_namelen = 0; | |
4651 | msg.msg_iocb = NULL; | |
4652 | msg.msg_flags = 0; | |
fddaface | 4653 | |
04411806 PB |
4654 | flags = req->sr_msg.msg_flags; |
4655 | if (force_nonblock) | |
7a7cacba | 4656 | flags |= MSG_DONTWAIT; |
0031275d SM |
4657 | if (flags & MSG_WAITALL) |
4658 | min_ret = iov_iter_count(&msg.msg_iter); | |
4659 | ||
7a7cacba PB |
4660 | ret = sock_recvmsg(sock, &msg, flags); |
4661 | if (force_nonblock && ret == -EAGAIN) | |
4662 | return -EAGAIN; | |
4663 | if (ret == -ERESTARTSYS) | |
4664 | ret = -EINTR; | |
14c32eee | 4665 | out_free: |
7fbb1b54 PB |
4666 | if (req->flags & REQ_F_BUFFER_SELECTED) |
4667 | cflags = io_put_recv_kbuf(req); | |
0031275d | 4668 | if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) |
fddaface | 4669 | req_set_fail_links(req); |
889fca73 | 4670 | __io_req_complete(req, issue_flags, ret, cflags); |
fddaface | 4671 | return 0; |
fddaface JA |
4672 | } |
4673 | ||
3529d8c2 | 4674 | static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
17f2fe35 | 4675 | { |
8ed8d3c3 JA |
4676 | struct io_accept *accept = &req->accept; |
4677 | ||
14587a46 | 4678 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
17f2fe35 | 4679 | return -EINVAL; |
8042d6ce | 4680 | if (sqe->ioprio || sqe->len || sqe->buf_index) |
17f2fe35 JA |
4681 | return -EINVAL; |
4682 | ||
d55e5f5b JA |
4683 | accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4684 | accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); | |
8ed8d3c3 | 4685 | accept->flags = READ_ONCE(sqe->accept_flags); |
09952e3e | 4686 | accept->nofile = rlimit(RLIMIT_NOFILE); |
8ed8d3c3 | 4687 | return 0; |
8ed8d3c3 | 4688 | } |
17f2fe35 | 4689 | |
889fca73 | 4690 | static int io_accept(struct io_kiocb *req, unsigned int issue_flags) |
8ed8d3c3 JA |
4691 | { |
4692 | struct io_accept *accept = &req->accept; | |
45d189c6 | 4693 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
ac45abc0 | 4694 | unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0; |
8ed8d3c3 JA |
4695 | int ret; |
4696 | ||
e697deed JX |
4697 | if (req->file->f_flags & O_NONBLOCK) |
4698 | req->flags |= REQ_F_NOWAIT; | |
4699 | ||
8ed8d3c3 | 4700 | ret = __sys_accept4_file(req->file, file_flags, accept->addr, |
09952e3e JA |
4701 | accept->addr_len, accept->flags, |
4702 | accept->nofile); | |
8ed8d3c3 | 4703 | if (ret == -EAGAIN && force_nonblock) |
17f2fe35 | 4704 | return -EAGAIN; |
ac45abc0 PB |
4705 | if (ret < 0) { |
4706 | if (ret == -ERESTARTSYS) | |
4707 | ret = -EINTR; | |
4e88d6e7 | 4708 | req_set_fail_links(req); |
ac45abc0 | 4709 | } |
889fca73 | 4710 | __io_req_complete(req, issue_flags, ret, 0); |
17f2fe35 | 4711 | return 0; |
8ed8d3c3 JA |
4712 | } |
4713 | ||
93642ef8 PB |
4714 | static int io_connect_prep_async(struct io_kiocb *req) |
4715 | { | |
4716 | struct io_async_connect *io = req->async_data; | |
4717 | struct io_connect *conn = &req->connect; | |
4718 | ||
4719 | return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); | |
4720 | } | |
4721 | ||
3529d8c2 | 4722 | static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f499a021 | 4723 | { |
3529d8c2 | 4724 | struct io_connect *conn = &req->connect; |
f499a021 | 4725 | |
14587a46 | 4726 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
3fbb51c1 JA |
4727 | return -EINVAL; |
4728 | if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags) | |
4729 | return -EINVAL; | |
4730 | ||
3529d8c2 JA |
4731 | conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
4732 | conn->addr_len = READ_ONCE(sqe->addr2); | |
93642ef8 | 4733 | return 0; |
f499a021 JA |
4734 | } |
4735 | ||
889fca73 | 4736 | static int io_connect(struct io_kiocb *req, unsigned int issue_flags) |
f8e85cf2 | 4737 | { |
e8c2bc1f | 4738 | struct io_async_connect __io, *io; |
f8e85cf2 | 4739 | unsigned file_flags; |
3fbb51c1 | 4740 | int ret; |
45d189c6 | 4741 | bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK; |
f8e85cf2 | 4742 | |
e8c2bc1f JA |
4743 | if (req->async_data) { |
4744 | io = req->async_data; | |
f499a021 | 4745 | } else { |
3529d8c2 JA |
4746 | ret = move_addr_to_kernel(req->connect.addr, |
4747 | req->connect.addr_len, | |
e8c2bc1f | 4748 | &__io.address); |
f499a021 JA |
4749 | if (ret) |
4750 | goto out; | |
4751 | io = &__io; | |
4752 | } | |
4753 | ||
3fbb51c1 JA |
4754 | file_flags = force_nonblock ? O_NONBLOCK : 0; |
4755 | ||
e8c2bc1f | 4756 | ret = __sys_connect_file(req->file, &io->address, |
3fbb51c1 | 4757 | req->connect.addr_len, file_flags); |
87f80d62 | 4758 | if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { |
e8c2bc1f | 4759 | if (req->async_data) |
b7bb4f7d | 4760 | return -EAGAIN; |
e8c2bc1f | 4761 | if (io_alloc_async_data(req)) { |
f499a021 JA |
4762 | ret = -ENOMEM; |
4763 | goto out; | |
4764 | } | |
e8c2bc1f | 4765 | memcpy(req->async_data, &__io, sizeof(__io)); |
f8e85cf2 | 4766 | return -EAGAIN; |
f499a021 | 4767 | } |
f8e85cf2 JA |
4768 | if (ret == -ERESTARTSYS) |
4769 | ret = -EINTR; | |
f499a021 | 4770 | out: |
4e88d6e7 JA |
4771 | if (ret < 0) |
4772 | req_set_fail_links(req); | |
889fca73 | 4773 | __io_req_complete(req, issue_flags, ret, 0); |
f8e85cf2 | 4774 | return 0; |
469956e8 Y |
4775 | } |
4776 | #else /* !CONFIG_NET */ | |
99a10081 JA |
4777 | #define IO_NETOP_FN(op) \ |
4778 | static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \ | |
4779 | { \ | |
4780 | return -EOPNOTSUPP; \ | |
4781 | } | |
4782 | ||
4783 | #define IO_NETOP_PREP(op) \ | |
4784 | IO_NETOP_FN(op) \ | |
4785 | static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \ | |
4786 | { \ | |
4787 | return -EOPNOTSUPP; \ | |
4788 | } \ | |
4789 | ||
4790 | #define IO_NETOP_PREP_ASYNC(op) \ | |
4791 | IO_NETOP_PREP(op) \ | |
4792 | static int io_##op##_prep_async(struct io_kiocb *req) \ | |
4793 | { \ | |
4794 | return -EOPNOTSUPP; \ | |
4795 | } | |
4796 | ||
4797 | IO_NETOP_PREP_ASYNC(sendmsg); | |
4798 | IO_NETOP_PREP_ASYNC(recvmsg); | |
4799 | IO_NETOP_PREP_ASYNC(connect); | |
4800 | IO_NETOP_PREP(accept); | |
4801 | IO_NETOP_FN(send); | |
4802 | IO_NETOP_FN(recv); | |
469956e8 | 4803 | #endif /* CONFIG_NET */ |
f8e85cf2 | 4804 | |
d7718a9d JA |
4805 | struct io_poll_table { |
4806 | struct poll_table_struct pt; | |
4807 | struct io_kiocb *req; | |
4808 | int error; | |
4809 | }; | |
ce593a6c | 4810 | |
d7718a9d JA |
4811 | static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll, |
4812 | __poll_t mask, task_work_func_t func) | |
4813 | { | |
aa96bf8a | 4814 | int ret; |
d7718a9d JA |
4815 | |
4816 | /* for instances that support it check for an event match first: */ | |
4817 | if (mask && !(mask & poll->events)) | |
4818 | return 0; | |
4819 | ||
4820 | trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); | |
4821 | ||
4822 | list_del_init(&poll->wait.entry); | |
4823 | ||
d7718a9d | 4824 | req->result = mask; |
7cbf1722 | 4825 | req->task_work.func = func; |
6d816e08 | 4826 | |
d7718a9d | 4827 | /* |
e3aabf95 JA |
4828 | * If this fails, then the task is exiting. When a task exits, the |
4829 | * work gets canceled, so just cancel this request as well instead | |
4830 | * of executing it. We can't safely execute it anyway, as we may not | |
4831 | * have the needed state needed for it anyway. | |
d7718a9d | 4832 | */ |
355fb9e2 | 4833 | ret = io_req_task_work_add(req); |
aa96bf8a | 4834 | if (unlikely(ret)) { |
e3aabf95 | 4835 | WRITE_ONCE(poll->canceled, true); |
eab30c4d | 4836 | io_req_task_work_add_fallback(req, func); |
aa96bf8a | 4837 | } |
d7718a9d JA |
4838 | return 1; |
4839 | } | |
4840 | ||
74ce6ce4 JA |
4841 | static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll) |
4842 | __acquires(&req->ctx->completion_lock) | |
4843 | { | |
4844 | struct io_ring_ctx *ctx = req->ctx; | |
4845 | ||
4846 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4847 | struct poll_table_struct pt = { ._key = poll->events }; | |
4848 | ||
4849 | req->result = vfs_poll(req->file, &pt) & poll->events; | |
4850 | } | |
4851 | ||
4852 | spin_lock_irq(&ctx->completion_lock); | |
4853 | if (!req->result && !READ_ONCE(poll->canceled)) { | |
4854 | add_wait_queue(poll->head, &poll->wait); | |
4855 | return true; | |
4856 | } | |
4857 | ||
4858 | return false; | |
4859 | } | |
4860 | ||
d4e7cd36 | 4861 | static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req) |
18bceab1 | 4862 | { |
e8c2bc1f | 4863 | /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ |
d4e7cd36 | 4864 | if (req->opcode == IORING_OP_POLL_ADD) |
e8c2bc1f | 4865 | return req->async_data; |
d4e7cd36 JA |
4866 | return req->apoll->double_poll; |
4867 | } | |
4868 | ||
4869 | static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req) | |
4870 | { | |
4871 | if (req->opcode == IORING_OP_POLL_ADD) | |
4872 | return &req->poll; | |
4873 | return &req->apoll->poll; | |
4874 | } | |
4875 | ||
4876 | static void io_poll_remove_double(struct io_kiocb *req) | |
e07785b0 | 4877 | __must_hold(&req->ctx->completion_lock) |
d4e7cd36 JA |
4878 | { |
4879 | struct io_poll_iocb *poll = io_poll_get_double(req); | |
18bceab1 JA |
4880 | |
4881 | lockdep_assert_held(&req->ctx->completion_lock); | |
4882 | ||
4883 | if (poll && poll->head) { | |
4884 | struct wait_queue_head *head = poll->head; | |
4885 | ||
4886 | spin_lock(&head->lock); | |
4887 | list_del_init(&poll->wait.entry); | |
4888 | if (poll->wait.private) | |
de9b4cca | 4889 | req_ref_put(req); |
18bceab1 JA |
4890 | poll->head = NULL; |
4891 | spin_unlock(&head->lock); | |
4892 | } | |
4893 | } | |
4894 | ||
e27414be | 4895 | static bool io_poll_complete(struct io_kiocb *req, __poll_t mask) |
e07785b0 | 4896 | __must_hold(&req->ctx->completion_lock) |
18bceab1 JA |
4897 | { |
4898 | struct io_ring_ctx *ctx = req->ctx; | |
88e41cf9 | 4899 | unsigned flags = IORING_CQE_F_MORE; |
e27414be | 4900 | int error; |
18bceab1 | 4901 | |
e27414be | 4902 | if (READ_ONCE(req->poll.canceled)) { |
45ab03b1 | 4903 | error = -ECANCELED; |
88e41cf9 | 4904 | req->poll.events |= EPOLLONESHOT; |
e27414be | 4905 | } else { |
5082620f | 4906 | error = mangle_poll(mask); |
e27414be | 4907 | } |
b69de288 JA |
4908 | if (req->poll.events & EPOLLONESHOT) |
4909 | flags = 0; | |
ff642164 | 4910 | if (!io_cqring_fill_event(req, error, flags)) { |
5082620f | 4911 | io_poll_remove_waitqs(req); |
88e41cf9 JA |
4912 | req->poll.done = true; |
4913 | flags = 0; | |
4914 | } | |
18bceab1 | 4915 | io_commit_cqring(ctx); |
88e41cf9 | 4916 | return !(flags & IORING_CQE_F_MORE); |
18bceab1 JA |
4917 | } |
4918 | ||
dd221f46 | 4919 | static void io_poll_task_func(struct callback_head *cb) |
18bceab1 | 4920 | { |
dd221f46 | 4921 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); |
18bceab1 | 4922 | struct io_ring_ctx *ctx = req->ctx; |
dd221f46 | 4923 | struct io_kiocb *nxt; |
18bceab1 JA |
4924 | |
4925 | if (io_poll_rewait(req, &req->poll)) { | |
4926 | spin_unlock_irq(&ctx->completion_lock); | |
dd221f46 | 4927 | } else { |
f40b964a | 4928 | bool done; |
88e41cf9 | 4929 | |
e27414be | 4930 | done = io_poll_complete(req, req->result); |
88e41cf9 JA |
4931 | if (done) { |
4932 | hash_del(&req->hash_node); | |
f40b964a | 4933 | } else { |
88e41cf9 JA |
4934 | req->result = 0; |
4935 | add_wait_queue(req->poll.head, &req->poll.wait); | |
4936 | } | |
dd221f46 | 4937 | spin_unlock_irq(&ctx->completion_lock); |
f40b964a | 4938 | io_cqring_ev_posted(ctx); |
18bceab1 | 4939 | |
88e41cf9 JA |
4940 | if (done) { |
4941 | nxt = io_put_req_find_next(req); | |
4942 | if (nxt) | |
4943 | __io_req_task_submit(nxt); | |
4944 | } | |
dd221f46 | 4945 | } |
18bceab1 JA |
4946 | } |
4947 | ||
4948 | static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode, | |
4949 | int sync, void *key) | |
4950 | { | |
4951 | struct io_kiocb *req = wait->private; | |
d4e7cd36 | 4952 | struct io_poll_iocb *poll = io_poll_get_single(req); |
18bceab1 JA |
4953 | __poll_t mask = key_to_poll(key); |
4954 | ||
4955 | /* for instances that support it check for an event match first: */ | |
4956 | if (mask && !(mask & poll->events)) | |
4957 | return 0; | |
88e41cf9 JA |
4958 | if (!(poll->events & EPOLLONESHOT)) |
4959 | return poll->wait.func(&poll->wait, mode, sync, key); | |
18bceab1 | 4960 | |
8706e04e JA |
4961 | list_del_init(&wait->entry); |
4962 | ||
807abcb0 | 4963 | if (poll && poll->head) { |
18bceab1 JA |
4964 | bool done; |
4965 | ||
807abcb0 JA |
4966 | spin_lock(&poll->head->lock); |
4967 | done = list_empty(&poll->wait.entry); | |
18bceab1 | 4968 | if (!done) |
807abcb0 | 4969 | list_del_init(&poll->wait.entry); |
d4e7cd36 JA |
4970 | /* make sure double remove sees this as being gone */ |
4971 | wait->private = NULL; | |
807abcb0 | 4972 | spin_unlock(&poll->head->lock); |
c8b5e260 JA |
4973 | if (!done) { |
4974 | /* use wait func handler, so it matches the rq type */ | |
4975 | poll->wait.func(&poll->wait, mode, sync, key); | |
4976 | } | |
18bceab1 | 4977 | } |
de9b4cca | 4978 | req_ref_put(req); |
18bceab1 JA |
4979 | return 1; |
4980 | } | |
4981 | ||
4982 | static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, | |
4983 | wait_queue_func_t wake_func) | |
4984 | { | |
4985 | poll->head = NULL; | |
4986 | poll->done = false; | |
4987 | poll->canceled = false; | |
464dca61 JA |
4988 | #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP) |
4989 | /* mask in events that we always want/need */ | |
4990 | poll->events = events | IO_POLL_UNMASK; | |
18bceab1 JA |
4991 | INIT_LIST_HEAD(&poll->wait.entry); |
4992 | init_waitqueue_func_entry(&poll->wait, wake_func); | |
4993 | } | |
4994 | ||
4995 | static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, | |
807abcb0 JA |
4996 | struct wait_queue_head *head, |
4997 | struct io_poll_iocb **poll_ptr) | |
18bceab1 JA |
4998 | { |
4999 | struct io_kiocb *req = pt->req; | |
5000 | ||
5001 | /* | |
5002 | * If poll->head is already set, it's because the file being polled | |
5003 | * uses multiple waitqueues for poll handling (eg one for read, one | |
5004 | * for write). Setup a separate io_poll_iocb if this happens. | |
5005 | */ | |
5006 | if (unlikely(poll->head)) { | |
58852d4d PB |
5007 | struct io_poll_iocb *poll_one = poll; |
5008 | ||
18bceab1 | 5009 | /* already have a 2nd entry, fail a third attempt */ |
807abcb0 | 5010 | if (*poll_ptr) { |
18bceab1 JA |
5011 | pt->error = -EINVAL; |
5012 | return; | |
5013 | } | |
1c3b3e65 JA |
5014 | /* double add on the same waitqueue head, ignore */ |
5015 | if (poll->head == head) | |
5016 | return; | |
18bceab1 JA |
5017 | poll = kmalloc(sizeof(*poll), GFP_ATOMIC); |
5018 | if (!poll) { | |
5019 | pt->error = -ENOMEM; | |
5020 | return; | |
5021 | } | |
58852d4d | 5022 | io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake); |
de9b4cca | 5023 | req_ref_get(req); |
18bceab1 | 5024 | poll->wait.private = req; |
807abcb0 | 5025 | *poll_ptr = poll; |
18bceab1 JA |
5026 | } |
5027 | ||
5028 | pt->error = 0; | |
5029 | poll->head = head; | |
a31eb4a2 JX |
5030 | |
5031 | if (poll->events & EPOLLEXCLUSIVE) | |
5032 | add_wait_queue_exclusive(head, &poll->wait); | |
5033 | else | |
5034 | add_wait_queue(head, &poll->wait); | |
18bceab1 JA |
5035 | } |
5036 | ||
5037 | static void io_async_queue_proc(struct file *file, struct wait_queue_head *head, | |
5038 | struct poll_table_struct *p) | |
5039 | { | |
5040 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
807abcb0 | 5041 | struct async_poll *apoll = pt->req->apoll; |
18bceab1 | 5042 | |
807abcb0 | 5043 | __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); |
18bceab1 JA |
5044 | } |
5045 | ||
d7718a9d JA |
5046 | static void io_async_task_func(struct callback_head *cb) |
5047 | { | |
5048 | struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work); | |
5049 | struct async_poll *apoll = req->apoll; | |
5050 | struct io_ring_ctx *ctx = req->ctx; | |
5051 | ||
5052 | trace_io_uring_task_run(req->ctx, req->opcode, req->user_data); | |
5053 | ||
74ce6ce4 | 5054 | if (io_poll_rewait(req, &apoll->poll)) { |
d7718a9d | 5055 | spin_unlock_irq(&ctx->completion_lock); |
74ce6ce4 | 5056 | return; |
d7718a9d JA |
5057 | } |
5058 | ||
0ea13b44 | 5059 | hash_del(&req->hash_node); |
d4e7cd36 | 5060 | io_poll_remove_double(req); |
74ce6ce4 JA |
5061 | spin_unlock_irq(&ctx->completion_lock); |
5062 | ||
0be0b0e3 PB |
5063 | if (!READ_ONCE(apoll->poll.canceled)) |
5064 | __io_req_task_submit(req); | |
5065 | else | |
2593553a | 5066 | io_req_complete_failed(req, -ECANCELED); |
aa340845 | 5067 | |
807abcb0 | 5068 | kfree(apoll->double_poll); |
31067255 | 5069 | kfree(apoll); |
d7718a9d JA |
5070 | } |
5071 | ||
5072 | static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync, | |
5073 | void *key) | |
5074 | { | |
5075 | struct io_kiocb *req = wait->private; | |
5076 | struct io_poll_iocb *poll = &req->apoll->poll; | |
5077 | ||
5078 | trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data, | |
5079 | key_to_poll(key)); | |
5080 | ||
5081 | return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func); | |
5082 | } | |
5083 | ||
5084 | static void io_poll_req_insert(struct io_kiocb *req) | |
5085 | { | |
5086 | struct io_ring_ctx *ctx = req->ctx; | |
5087 | struct hlist_head *list; | |
5088 | ||
5089 | list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; | |
5090 | hlist_add_head(&req->hash_node, list); | |
5091 | } | |
5092 | ||
5093 | static __poll_t __io_arm_poll_handler(struct io_kiocb *req, | |
5094 | struct io_poll_iocb *poll, | |
5095 | struct io_poll_table *ipt, __poll_t mask, | |
5096 | wait_queue_func_t wake_func) | |
5097 | __acquires(&ctx->completion_lock) | |
5098 | { | |
5099 | struct io_ring_ctx *ctx = req->ctx; | |
5100 | bool cancel = false; | |
5101 | ||
4d52f338 | 5102 | INIT_HLIST_NODE(&req->hash_node); |
18bceab1 | 5103 | io_init_poll_iocb(poll, mask, wake_func); |
b90cd197 | 5104 | poll->file = req->file; |
18bceab1 | 5105 | poll->wait.private = req; |
d7718a9d JA |
5106 | |
5107 | ipt->pt._key = mask; | |
5108 | ipt->req = req; | |
5109 | ipt->error = -EINVAL; | |
5110 | ||
d7718a9d JA |
5111 | mask = vfs_poll(req->file, &ipt->pt) & poll->events; |
5112 | ||
5113 | spin_lock_irq(&ctx->completion_lock); | |
5114 | if (likely(poll->head)) { | |
5115 | spin_lock(&poll->head->lock); | |
5116 | if (unlikely(list_empty(&poll->wait.entry))) { | |
5117 | if (ipt->error) | |
5118 | cancel = true; | |
5119 | ipt->error = 0; | |
5120 | mask = 0; | |
5121 | } | |
88e41cf9 | 5122 | if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error) |
d7718a9d JA |
5123 | list_del_init(&poll->wait.entry); |
5124 | else if (cancel) | |
5125 | WRITE_ONCE(poll->canceled, true); | |
5126 | else if (!poll->done) /* actually waiting for an event */ | |
5127 | io_poll_req_insert(req); | |
5128 | spin_unlock(&poll->head->lock); | |
5129 | } | |
5130 | ||
5131 | return mask; | |
5132 | } | |
5133 | ||
5134 | static bool io_arm_poll_handler(struct io_kiocb *req) | |
5135 | { | |
5136 | const struct io_op_def *def = &io_op_defs[req->opcode]; | |
5137 | struct io_ring_ctx *ctx = req->ctx; | |
5138 | struct async_poll *apoll; | |
5139 | struct io_poll_table ipt; | |
5140 | __poll_t mask, ret; | |
9dab14b8 | 5141 | int rw; |
d7718a9d JA |
5142 | |
5143 | if (!req->file || !file_can_poll(req->file)) | |
5144 | return false; | |
24c74678 | 5145 | if (req->flags & REQ_F_POLLED) |
d7718a9d | 5146 | return false; |
9dab14b8 JA |
5147 | if (def->pollin) |
5148 | rw = READ; | |
5149 | else if (def->pollout) | |
5150 | rw = WRITE; | |
5151 | else | |
5152 | return false; | |
5153 | /* if we can't nonblock try, then no point in arming a poll handler */ | |
7b29f92d | 5154 | if (!io_file_supports_async(req, rw)) |
d7718a9d JA |
5155 | return false; |
5156 | ||
5157 | apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC); | |
5158 | if (unlikely(!apoll)) | |
5159 | return false; | |
807abcb0 | 5160 | apoll->double_poll = NULL; |
d7718a9d JA |
5161 | |
5162 | req->flags |= REQ_F_POLLED; | |
d7718a9d | 5163 | req->apoll = apoll; |
d7718a9d | 5164 | |
88e41cf9 | 5165 | mask = EPOLLONESHOT; |
d7718a9d | 5166 | if (def->pollin) |
8755d97a | 5167 | mask |= POLLIN | POLLRDNORM; |
d7718a9d JA |
5168 | if (def->pollout) |
5169 | mask |= POLLOUT | POLLWRNORM; | |
901341bb LH |
5170 | |
5171 | /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */ | |
5172 | if ((req->opcode == IORING_OP_RECVMSG) && | |
5173 | (req->sr_msg.msg_flags & MSG_ERRQUEUE)) | |
5174 | mask &= ~POLLIN; | |
5175 | ||
d7718a9d JA |
5176 | mask |= POLLERR | POLLPRI; |
5177 | ||
5178 | ipt.pt._qproc = io_async_queue_proc; | |
5179 | ||
5180 | ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, | |
5181 | io_async_wake); | |
a36da65c | 5182 | if (ret || ipt.error) { |
d4e7cd36 | 5183 | io_poll_remove_double(req); |
d7718a9d | 5184 | spin_unlock_irq(&ctx->completion_lock); |
807abcb0 | 5185 | kfree(apoll->double_poll); |
d7718a9d JA |
5186 | kfree(apoll); |
5187 | return false; | |
5188 | } | |
5189 | spin_unlock_irq(&ctx->completion_lock); | |
5190 | trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask, | |
5191 | apoll->poll.events); | |
5192 | return true; | |
5193 | } | |
5194 | ||
5195 | static bool __io_poll_remove_one(struct io_kiocb *req, | |
b2e720ac | 5196 | struct io_poll_iocb *poll, bool do_cancel) |
e07785b0 | 5197 | __must_hold(&req->ctx->completion_lock) |
221c5eb2 | 5198 | { |
b41e9852 | 5199 | bool do_complete = false; |
221c5eb2 | 5200 | |
5082620f JA |
5201 | if (!poll->head) |
5202 | return false; | |
221c5eb2 | 5203 | spin_lock(&poll->head->lock); |
b2e720ac JA |
5204 | if (do_cancel) |
5205 | WRITE_ONCE(poll->canceled, true); | |
392edb45 JA |
5206 | if (!list_empty(&poll->wait.entry)) { |
5207 | list_del_init(&poll->wait.entry); | |
b41e9852 | 5208 | do_complete = true; |
221c5eb2 JA |
5209 | } |
5210 | spin_unlock(&poll->head->lock); | |
3bfa5bcb | 5211 | hash_del(&req->hash_node); |
d7718a9d JA |
5212 | return do_complete; |
5213 | } | |
5214 | ||
b2c3f7e1 | 5215 | static bool io_poll_remove_waitqs(struct io_kiocb *req) |
e07785b0 | 5216 | __must_hold(&req->ctx->completion_lock) |
d7718a9d JA |
5217 | { |
5218 | bool do_complete; | |
5219 | ||
d4e7cd36 | 5220 | io_poll_remove_double(req); |
e31001a3 | 5221 | do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true); |
d4e7cd36 | 5222 | |
e31001a3 | 5223 | if (req->opcode != IORING_OP_POLL_ADD && do_complete) { |
3bfa5bcb JA |
5224 | struct async_poll *apoll = req->apoll; |
5225 | ||
d7718a9d | 5226 | /* non-poll requests have submit ref still */ |
e31001a3 PB |
5227 | req_ref_put(req); |
5228 | kfree(apoll->double_poll); | |
5229 | kfree(apoll); | |
b1f573bd | 5230 | } |
b2c3f7e1 JA |
5231 | return do_complete; |
5232 | } | |
5233 | ||
5234 | static bool io_poll_remove_one(struct io_kiocb *req) | |
e07785b0 | 5235 | __must_hold(&req->ctx->completion_lock) |
b2c3f7e1 JA |
5236 | { |
5237 | bool do_complete; | |
5238 | ||
5239 | do_complete = io_poll_remove_waitqs(req); | |
b41e9852 | 5240 | if (do_complete) { |
ff642164 | 5241 | io_cqring_fill_event(req, -ECANCELED, 0); |
b41e9852 | 5242 | io_commit_cqring(req->ctx); |
f254ac04 | 5243 | req_set_fail_links(req); |
216578e5 | 5244 | io_put_req_deferred(req, 1); |
b41e9852 JA |
5245 | } |
5246 | ||
5247 | return do_complete; | |
221c5eb2 JA |
5248 | } |
5249 | ||
76e1b642 JA |
5250 | /* |
5251 | * Returns true if we found and killed one or more poll requests | |
5252 | */ | |
6b81928d PB |
5253 | static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk, |
5254 | struct files_struct *files) | |
221c5eb2 | 5255 | { |
78076bb6 | 5256 | struct hlist_node *tmp; |
221c5eb2 | 5257 | struct io_kiocb *req; |
8e2e1faf | 5258 | int posted = 0, i; |
221c5eb2 JA |
5259 | |
5260 | spin_lock_irq(&ctx->completion_lock); | |
78076bb6 JA |
5261 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { |
5262 | struct hlist_head *list; | |
5263 | ||
5264 | list = &ctx->cancel_hash[i]; | |
f3606e3a | 5265 | hlist_for_each_entry_safe(req, tmp, list, hash_node) { |
6b81928d | 5266 | if (io_match_task(req, tsk, files)) |
f3606e3a JA |
5267 | posted += io_poll_remove_one(req); |
5268 | } | |
221c5eb2 JA |
5269 | } |
5270 | spin_unlock_irq(&ctx->completion_lock); | |
b41e9852 | 5271 | |
8e2e1faf JA |
5272 | if (posted) |
5273 | io_cqring_ev_posted(ctx); | |
76e1b642 JA |
5274 | |
5275 | return posted != 0; | |
221c5eb2 JA |
5276 | } |
5277 | ||
b2cb805f | 5278 | static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr) |
e07785b0 | 5279 | __must_hold(&ctx->completion_lock) |
47f46768 | 5280 | { |
78076bb6 | 5281 | struct hlist_head *list; |
47f46768 JA |
5282 | struct io_kiocb *req; |
5283 | ||
78076bb6 JA |
5284 | list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; |
5285 | hlist_for_each_entry(req, list, hash_node) { | |
b41e9852 JA |
5286 | if (sqe_addr != req->user_data) |
5287 | continue; | |
b2cb805f | 5288 | return req; |
47f46768 JA |
5289 | } |
5290 | ||
b2cb805f JA |
5291 | return NULL; |
5292 | } | |
5293 | ||
5294 | static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr) | |
e07785b0 | 5295 | __must_hold(&ctx->completion_lock) |
b2cb805f JA |
5296 | { |
5297 | struct io_kiocb *req; | |
5298 | ||
5299 | req = io_poll_find(ctx, sqe_addr); | |
5300 | if (!req) | |
5301 | return -ENOENT; | |
5302 | if (io_poll_remove_one(req)) | |
5303 | return 0; | |
5304 | ||
5305 | return -EALREADY; | |
47f46768 JA |
5306 | } |
5307 | ||
3529d8c2 JA |
5308 | static int io_poll_remove_prep(struct io_kiocb *req, |
5309 | const struct io_uring_sqe *sqe) | |
0969e783 | 5310 | { |
0969e783 JA |
5311 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5312 | return -EINVAL; | |
5313 | if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || | |
5314 | sqe->poll_events) | |
5315 | return -EINVAL; | |
5316 | ||
018043be | 5317 | req->poll_remove.addr = READ_ONCE(sqe->addr); |
0969e783 JA |
5318 | return 0; |
5319 | } | |
5320 | ||
221c5eb2 JA |
5321 | /* |
5322 | * Find a running poll command that matches one specified in sqe->addr, | |
5323 | * and remove it if found. | |
5324 | */ | |
61e98203 | 5325 | static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags) |
221c5eb2 JA |
5326 | { |
5327 | struct io_ring_ctx *ctx = req->ctx; | |
47f46768 | 5328 | int ret; |
221c5eb2 | 5329 | |
221c5eb2 | 5330 | spin_lock_irq(&ctx->completion_lock); |
018043be | 5331 | ret = io_poll_cancel(ctx, req->poll_remove.addr); |
221c5eb2 JA |
5332 | spin_unlock_irq(&ctx->completion_lock); |
5333 | ||
4e88d6e7 JA |
5334 | if (ret < 0) |
5335 | req_set_fail_links(req); | |
0bdf3398 | 5336 | __io_req_complete(req, issue_flags, ret, 0); |
221c5eb2 JA |
5337 | return 0; |
5338 | } | |
5339 | ||
221c5eb2 JA |
5340 | static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync, |
5341 | void *key) | |
5342 | { | |
c2f2eb7d JA |
5343 | struct io_kiocb *req = wait->private; |
5344 | struct io_poll_iocb *poll = &req->poll; | |
221c5eb2 | 5345 | |
d7718a9d | 5346 | return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func); |
221c5eb2 JA |
5347 | } |
5348 | ||
221c5eb2 JA |
5349 | static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head, |
5350 | struct poll_table_struct *p) | |
5351 | { | |
5352 | struct io_poll_table *pt = container_of(p, struct io_poll_table, pt); | |
5353 | ||
e8c2bc1f | 5354 | __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data); |
eac406c6 JA |
5355 | } |
5356 | ||
3529d8c2 | 5357 | static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
221c5eb2 | 5358 | { |
88e41cf9 | 5359 | u32 events, flags; |
221c5eb2 JA |
5360 | |
5361 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) | |
5362 | return -EINVAL; | |
b69de288 | 5363 | if (sqe->ioprio || sqe->buf_index) |
88e41cf9 JA |
5364 | return -EINVAL; |
5365 | flags = READ_ONCE(sqe->len); | |
b69de288 JA |
5366 | if (flags & ~(IORING_POLL_ADD_MULTI | IORING_POLL_UPDATE_EVENTS | |
5367 | IORING_POLL_UPDATE_USER_DATA)) | |
221c5eb2 | 5368 | return -EINVAL; |
5769a351 JX |
5369 | events = READ_ONCE(sqe->poll32_events); |
5370 | #ifdef __BIG_ENDIAN | |
5371 | events = swahw32(events); | |
5372 | #endif | |
b69de288 | 5373 | if (!(flags & IORING_POLL_ADD_MULTI)) |
88e41cf9 | 5374 | events |= EPOLLONESHOT; |
9d805892 PB |
5375 | events = demangle_poll(events) | |
5376 | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT)); | |
66d2d00d PB |
5377 | |
5378 | if (flags & (IORING_POLL_UPDATE_EVENTS|IORING_POLL_UPDATE_USER_DATA)) { | |
9d805892 PB |
5379 | struct io_poll_update *poll_upd = &req->poll_update; |
5380 | ||
5381 | req->flags |= REQ_F_POLL_UPDATE; | |
5382 | poll_upd->events = events; | |
5383 | poll_upd->old_user_data = READ_ONCE(sqe->addr); | |
5384 | poll_upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; | |
5385 | poll_upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; | |
5386 | if (poll_upd->update_user_data) | |
5387 | poll_upd->new_user_data = READ_ONCE(sqe->off); | |
66d2d00d | 5388 | } else { |
9d805892 PB |
5389 | struct io_poll_iocb *poll = &req->poll; |
5390 | ||
5391 | poll->events = events; | |
66d2d00d PB |
5392 | if (sqe->off || sqe->addr) |
5393 | return -EINVAL; | |
b69de288 | 5394 | } |
0969e783 JA |
5395 | return 0; |
5396 | } | |
5397 | ||
b69de288 | 5398 | static int __io_poll_add(struct io_kiocb *req) |
0969e783 JA |
5399 | { |
5400 | struct io_poll_iocb *poll = &req->poll; | |
5401 | struct io_ring_ctx *ctx = req->ctx; | |
5402 | struct io_poll_table ipt; | |
0969e783 | 5403 | __poll_t mask; |
0969e783 | 5404 | |
d7718a9d | 5405 | ipt.pt._qproc = io_poll_queue_proc; |
36703247 | 5406 | |
d7718a9d JA |
5407 | mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events, |
5408 | io_poll_wake); | |
221c5eb2 | 5409 | |
8c838788 | 5410 | if (mask) { /* no async, we'd stolen it */ |
221c5eb2 | 5411 | ipt.error = 0; |
e27414be | 5412 | io_poll_complete(req, mask); |
221c5eb2 | 5413 | } |
221c5eb2 JA |
5414 | spin_unlock_irq(&ctx->completion_lock); |
5415 | ||
8c838788 JA |
5416 | if (mask) { |
5417 | io_cqring_ev_posted(ctx); | |
88e41cf9 JA |
5418 | if (poll->events & EPOLLONESHOT) |
5419 | io_put_req(req); | |
221c5eb2 | 5420 | } |
8c838788 | 5421 | return ipt.error; |
221c5eb2 JA |
5422 | } |
5423 | ||
b69de288 JA |
5424 | static int io_poll_update(struct io_kiocb *req) |
5425 | { | |
5426 | struct io_ring_ctx *ctx = req->ctx; | |
5427 | struct io_kiocb *preq; | |
cb3b200e | 5428 | bool completing; |
b69de288 JA |
5429 | int ret; |
5430 | ||
5431 | spin_lock_irq(&ctx->completion_lock); | |
9d805892 | 5432 | preq = io_poll_find(ctx, req->poll_update.old_user_data); |
b69de288 JA |
5433 | if (!preq) { |
5434 | ret = -ENOENT; | |
5435 | goto err; | |
5436 | } else if (preq->opcode != IORING_OP_POLL_ADD) { | |
5437 | /* don't allow internal poll updates */ | |
5438 | ret = -EACCES; | |
5439 | goto err; | |
5440 | } | |
cb3b200e JA |
5441 | |
5442 | /* | |
5443 | * Don't allow racy completion with singleshot, as we cannot safely | |
5444 | * update those. For multishot, if we're racing with completion, just | |
5445 | * let completion re-add it. | |
5446 | */ | |
5447 | completing = !__io_poll_remove_one(preq, &preq->poll, false); | |
5448 | if (completing && (preq->poll.events & EPOLLONESHOT)) { | |
5449 | ret = -EALREADY; | |
5450 | goto err; | |
b69de288 JA |
5451 | } |
5452 | /* we now have a detached poll request. reissue. */ | |
5453 | ret = 0; | |
5454 | err: | |
b69de288 | 5455 | if (ret < 0) { |
cb3b200e | 5456 | spin_unlock_irq(&ctx->completion_lock); |
b69de288 JA |
5457 | req_set_fail_links(req); |
5458 | io_req_complete(req, ret); | |
5459 | return 0; | |
5460 | } | |
5461 | /* only mask one event flags, keep behavior flags */ | |
9d805892 | 5462 | if (req->poll_update.update_events) { |
b69de288 | 5463 | preq->poll.events &= ~0xffff; |
9d805892 | 5464 | preq->poll.events |= req->poll_update.events & 0xffff; |
b69de288 JA |
5465 | preq->poll.events |= IO_POLL_UNMASK; |
5466 | } | |
9d805892 PB |
5467 | if (req->poll_update.update_user_data) |
5468 | preq->user_data = req->poll_update.new_user_data; | |
b69de288 | 5469 | |
cb3b200e JA |
5470 | spin_unlock_irq(&ctx->completion_lock); |
5471 | ||
b69de288 JA |
5472 | /* complete update request, we're done with it */ |
5473 | io_req_complete(req, ret); | |
5474 | ||
cb3b200e JA |
5475 | if (!completing) { |
5476 | ret = __io_poll_add(preq); | |
5477 | if (ret < 0) { | |
5478 | req_set_fail_links(preq); | |
5479 | io_req_complete(preq, ret); | |
5480 | } | |
b69de288 JA |
5481 | } |
5482 | return 0; | |
5483 | } | |
5484 | ||
5485 | static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags) | |
5486 | { | |
9d805892 | 5487 | if (!(req->flags & REQ_F_POLL_UPDATE)) |
b69de288 JA |
5488 | return __io_poll_add(req); |
5489 | return io_poll_update(req); | |
5490 | } | |
5491 | ||
5262f567 JA |
5492 | static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer) |
5493 | { | |
ad8a48ac JA |
5494 | struct io_timeout_data *data = container_of(timer, |
5495 | struct io_timeout_data, timer); | |
5496 | struct io_kiocb *req = data->req; | |
5497 | struct io_ring_ctx *ctx = req->ctx; | |
5262f567 JA |
5498 | unsigned long flags; |
5499 | ||
5262f567 | 5500 | spin_lock_irqsave(&ctx->completion_lock, flags); |
a71976f3 | 5501 | list_del_init(&req->timeout.list); |
01cec8c1 PB |
5502 | atomic_set(&req->ctx->cq_timeouts, |
5503 | atomic_read(&req->ctx->cq_timeouts) + 1); | |
5504 | ||
ff642164 | 5505 | io_cqring_fill_event(req, -ETIME, 0); |
5262f567 JA |
5506 | io_commit_cqring(ctx); |
5507 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5508 | ||
5509 | io_cqring_ev_posted(ctx); | |
4e88d6e7 | 5510 | req_set_fail_links(req); |
5262f567 JA |
5511 | io_put_req(req); |
5512 | return HRTIMER_NORESTART; | |
5513 | } | |
5514 | ||
fbd15848 PB |
5515 | static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx, |
5516 | __u64 user_data) | |
e07785b0 | 5517 | __must_hold(&ctx->completion_lock) |
f254ac04 | 5518 | { |
fbd15848 | 5519 | struct io_timeout_data *io; |
47f46768 | 5520 | struct io_kiocb *req; |
fd9c7bc5 | 5521 | bool found = false; |
f254ac04 | 5522 | |
135fcde8 | 5523 | list_for_each_entry(req, &ctx->timeout_list, timeout.list) { |
fd9c7bc5 PB |
5524 | found = user_data == req->user_data; |
5525 | if (found) | |
47f46768 | 5526 | break; |
47f46768 | 5527 | } |
fd9c7bc5 PB |
5528 | if (!found) |
5529 | return ERR_PTR(-ENOENT); | |
fbd15848 PB |
5530 | |
5531 | io = req->async_data; | |
fd9c7bc5 | 5532 | if (hrtimer_try_to_cancel(&io->timer) == -1) |
fbd15848 | 5533 | return ERR_PTR(-EALREADY); |
a71976f3 | 5534 | list_del_init(&req->timeout.list); |
fbd15848 PB |
5535 | return req; |
5536 | } | |
47f46768 | 5537 | |
fbd15848 | 5538 | static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data) |
e07785b0 | 5539 | __must_hold(&ctx->completion_lock) |
fbd15848 PB |
5540 | { |
5541 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); | |
5542 | ||
5543 | if (IS_ERR(req)) | |
5544 | return PTR_ERR(req); | |
f254ac04 JA |
5545 | |
5546 | req_set_fail_links(req); | |
ff642164 | 5547 | io_cqring_fill_event(req, -ECANCELED, 0); |
216578e5 | 5548 | io_put_req_deferred(req, 1); |
f254ac04 JA |
5549 | return 0; |
5550 | } | |
5551 | ||
9c8e11b3 PB |
5552 | static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data, |
5553 | struct timespec64 *ts, enum hrtimer_mode mode) | |
e07785b0 | 5554 | __must_hold(&ctx->completion_lock) |
47f46768 | 5555 | { |
9c8e11b3 PB |
5556 | struct io_kiocb *req = io_timeout_extract(ctx, user_data); |
5557 | struct io_timeout_data *data; | |
47f46768 | 5558 | |
9c8e11b3 PB |
5559 | if (IS_ERR(req)) |
5560 | return PTR_ERR(req); | |
47f46768 | 5561 | |
9c8e11b3 PB |
5562 | req->timeout.off = 0; /* noseq */ |
5563 | data = req->async_data; | |
5564 | list_add_tail(&req->timeout.list, &ctx->timeout_list); | |
5565 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode); | |
5566 | data->timer.function = io_timeout_fn; | |
5567 | hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); | |
5568 | return 0; | |
47f46768 JA |
5569 | } |
5570 | ||
3529d8c2 JA |
5571 | static int io_timeout_remove_prep(struct io_kiocb *req, |
5572 | const struct io_uring_sqe *sqe) | |
b29472ee | 5573 | { |
9c8e11b3 PB |
5574 | struct io_timeout_rem *tr = &req->timeout_rem; |
5575 | ||
b29472ee JA |
5576 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5577 | return -EINVAL; | |
61710e43 DA |
5578 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5579 | return -EINVAL; | |
9c8e11b3 | 5580 | if (sqe->ioprio || sqe->buf_index || sqe->len) |
b29472ee JA |
5581 | return -EINVAL; |
5582 | ||
9c8e11b3 PB |
5583 | tr->addr = READ_ONCE(sqe->addr); |
5584 | tr->flags = READ_ONCE(sqe->timeout_flags); | |
5585 | if (tr->flags & IORING_TIMEOUT_UPDATE) { | |
5586 | if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS)) | |
5587 | return -EINVAL; | |
5588 | if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) | |
5589 | return -EFAULT; | |
5590 | } else if (tr->flags) { | |
5591 | /* timeout removal doesn't support flags */ | |
b29472ee | 5592 | return -EINVAL; |
9c8e11b3 | 5593 | } |
b29472ee | 5594 | |
b29472ee JA |
5595 | return 0; |
5596 | } | |
5597 | ||
8662daec PB |
5598 | static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags) |
5599 | { | |
5600 | return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS | |
5601 | : HRTIMER_MODE_REL; | |
5602 | } | |
5603 | ||
11365043 JA |
5604 | /* |
5605 | * Remove or update an existing timeout command | |
5606 | */ | |
61e98203 | 5607 | static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags) |
11365043 | 5608 | { |
9c8e11b3 | 5609 | struct io_timeout_rem *tr = &req->timeout_rem; |
11365043 | 5610 | struct io_ring_ctx *ctx = req->ctx; |
47f46768 | 5611 | int ret; |
11365043 | 5612 | |
11365043 | 5613 | spin_lock_irq(&ctx->completion_lock); |
8662daec | 5614 | if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) |
9c8e11b3 | 5615 | ret = io_timeout_cancel(ctx, tr->addr); |
8662daec PB |
5616 | else |
5617 | ret = io_timeout_update(ctx, tr->addr, &tr->ts, | |
5618 | io_translate_timeout_mode(tr->flags)); | |
11365043 | 5619 | |
ff642164 | 5620 | io_cqring_fill_event(req, ret, 0); |
11365043 JA |
5621 | io_commit_cqring(ctx); |
5622 | spin_unlock_irq(&ctx->completion_lock); | |
5262f567 | 5623 | io_cqring_ev_posted(ctx); |
4e88d6e7 JA |
5624 | if (ret < 0) |
5625 | req_set_fail_links(req); | |
ec9c02ad | 5626 | io_put_req(req); |
11365043 | 5627 | return 0; |
5262f567 JA |
5628 | } |
5629 | ||
3529d8c2 | 5630 | static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe, |
2d28390a | 5631 | bool is_timeout_link) |
5262f567 | 5632 | { |
ad8a48ac | 5633 | struct io_timeout_data *data; |
a41525ab | 5634 | unsigned flags; |
56080b02 | 5635 | u32 off = READ_ONCE(sqe->off); |
5262f567 | 5636 | |
ad8a48ac | 5637 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
5262f567 | 5638 | return -EINVAL; |
ad8a48ac | 5639 | if (sqe->ioprio || sqe->buf_index || sqe->len != 1) |
a41525ab | 5640 | return -EINVAL; |
56080b02 | 5641 | if (off && is_timeout_link) |
2d28390a | 5642 | return -EINVAL; |
a41525ab JA |
5643 | flags = READ_ONCE(sqe->timeout_flags); |
5644 | if (flags & ~IORING_TIMEOUT_ABS) | |
5262f567 | 5645 | return -EINVAL; |
bdf20073 | 5646 | |
bfe68a22 | 5647 | req->timeout.off = off; |
26a61679 | 5648 | |
e8c2bc1f | 5649 | if (!req->async_data && io_alloc_async_data(req)) |
26a61679 JA |
5650 | return -ENOMEM; |
5651 | ||
e8c2bc1f | 5652 | data = req->async_data; |
ad8a48ac | 5653 | data->req = req; |
ad8a48ac JA |
5654 | |
5655 | if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) | |
5262f567 JA |
5656 | return -EFAULT; |
5657 | ||
8662daec | 5658 | data->mode = io_translate_timeout_mode(flags); |
ad8a48ac | 5659 | hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode); |
2482b58f PB |
5660 | if (is_timeout_link) |
5661 | io_req_track_inflight(req); | |
ad8a48ac JA |
5662 | return 0; |
5663 | } | |
5664 | ||
61e98203 | 5665 | static int io_timeout(struct io_kiocb *req, unsigned int issue_flags) |
ad8a48ac | 5666 | { |
ad8a48ac | 5667 | struct io_ring_ctx *ctx = req->ctx; |
e8c2bc1f | 5668 | struct io_timeout_data *data = req->async_data; |
ad8a48ac | 5669 | struct list_head *entry; |
bfe68a22 | 5670 | u32 tail, off = req->timeout.off; |
ad8a48ac | 5671 | |
733f5c95 | 5672 | spin_lock_irq(&ctx->completion_lock); |
93bd25bb | 5673 | |
5262f567 JA |
5674 | /* |
5675 | * sqe->off holds how many events that need to occur for this | |
93bd25bb JA |
5676 | * timeout event to be satisfied. If it isn't set, then this is |
5677 | * a pure timeout request, sequence isn't used. | |
5262f567 | 5678 | */ |
8eb7e2d0 | 5679 | if (io_is_timeout_noseq(req)) { |
93bd25bb JA |
5680 | entry = ctx->timeout_list.prev; |
5681 | goto add; | |
5682 | } | |
5262f567 | 5683 | |
bfe68a22 PB |
5684 | tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); |
5685 | req->timeout.target_seq = tail + off; | |
5262f567 | 5686 | |
f010505b MDG |
5687 | /* Update the last seq here in case io_flush_timeouts() hasn't. |
5688 | * This is safe because ->completion_lock is held, and submissions | |
5689 | * and completions are never mixed in the same ->completion_lock section. | |
5690 | */ | |
5691 | ctx->cq_last_tm_flush = tail; | |
5692 | ||
5262f567 JA |
5693 | /* |
5694 | * Insertion sort, ensuring the first entry in the list is always | |
5695 | * the one we need first. | |
5696 | */ | |
5262f567 | 5697 | list_for_each_prev(entry, &ctx->timeout_list) { |
135fcde8 PB |
5698 | struct io_kiocb *nxt = list_entry(entry, struct io_kiocb, |
5699 | timeout.list); | |
5262f567 | 5700 | |
8eb7e2d0 | 5701 | if (io_is_timeout_noseq(nxt)) |
93bd25bb | 5702 | continue; |
bfe68a22 PB |
5703 | /* nxt.seq is behind @tail, otherwise would've been completed */ |
5704 | if (off >= nxt->timeout.target_seq - tail) | |
5262f567 JA |
5705 | break; |
5706 | } | |
93bd25bb | 5707 | add: |
135fcde8 | 5708 | list_add(&req->timeout.list, entry); |
ad8a48ac JA |
5709 | data->timer.function = io_timeout_fn; |
5710 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); | |
5262f567 | 5711 | spin_unlock_irq(&ctx->completion_lock); |
5262f567 JA |
5712 | return 0; |
5713 | } | |
5262f567 | 5714 | |
f458dd84 PB |
5715 | struct io_cancel_data { |
5716 | struct io_ring_ctx *ctx; | |
5717 | u64 user_data; | |
5718 | }; | |
5719 | ||
62755e35 JA |
5720 | static bool io_cancel_cb(struct io_wq_work *work, void *data) |
5721 | { | |
5722 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
f458dd84 | 5723 | struct io_cancel_data *cd = data; |
62755e35 | 5724 | |
f458dd84 | 5725 | return req->ctx == cd->ctx && req->user_data == cd->user_data; |
62755e35 JA |
5726 | } |
5727 | ||
f458dd84 PB |
5728 | static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data, |
5729 | struct io_ring_ctx *ctx) | |
62755e35 | 5730 | { |
f458dd84 | 5731 | struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, }; |
62755e35 | 5732 | enum io_wq_cancel cancel_ret; |
62755e35 JA |
5733 | int ret = 0; |
5734 | ||
f458dd84 | 5735 | if (!tctx || !tctx->io_wq) |
5aa75ed5 JA |
5736 | return -ENOENT; |
5737 | ||
f458dd84 | 5738 | cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); |
62755e35 JA |
5739 | switch (cancel_ret) { |
5740 | case IO_WQ_CANCEL_OK: | |
5741 | ret = 0; | |
5742 | break; | |
5743 | case IO_WQ_CANCEL_RUNNING: | |
5744 | ret = -EALREADY; | |
5745 | break; | |
5746 | case IO_WQ_CANCEL_NOTFOUND: | |
5747 | ret = -ENOENT; | |
5748 | break; | |
5749 | } | |
5750 | ||
e977d6d3 JA |
5751 | return ret; |
5752 | } | |
5753 | ||
47f46768 JA |
5754 | static void io_async_find_and_cancel(struct io_ring_ctx *ctx, |
5755 | struct io_kiocb *req, __u64 sqe_addr, | |
014db007 | 5756 | int success_ret) |
47f46768 JA |
5757 | { |
5758 | unsigned long flags; | |
5759 | int ret; | |
5760 | ||
f458dd84 | 5761 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); |
47f46768 | 5762 | spin_lock_irqsave(&ctx->completion_lock, flags); |
df9727af PB |
5763 | if (ret != -ENOENT) |
5764 | goto done; | |
47f46768 JA |
5765 | ret = io_timeout_cancel(ctx, sqe_addr); |
5766 | if (ret != -ENOENT) | |
5767 | goto done; | |
5768 | ret = io_poll_cancel(ctx, sqe_addr); | |
5769 | done: | |
b0dd8a41 JA |
5770 | if (!ret) |
5771 | ret = success_ret; | |
ff642164 | 5772 | io_cqring_fill_event(req, ret, 0); |
47f46768 JA |
5773 | io_commit_cqring(ctx); |
5774 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
5775 | io_cqring_ev_posted(ctx); | |
5776 | ||
4e88d6e7 JA |
5777 | if (ret < 0) |
5778 | req_set_fail_links(req); | |
47f46768 JA |
5779 | } |
5780 | ||
3529d8c2 JA |
5781 | static int io_async_cancel_prep(struct io_kiocb *req, |
5782 | const struct io_uring_sqe *sqe) | |
e977d6d3 | 5783 | { |
fbf23849 | 5784 | if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) |
e977d6d3 | 5785 | return -EINVAL; |
61710e43 DA |
5786 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5787 | return -EINVAL; | |
5788 | if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags) | |
e977d6d3 JA |
5789 | return -EINVAL; |
5790 | ||
fbf23849 JA |
5791 | req->cancel.addr = READ_ONCE(sqe->addr); |
5792 | return 0; | |
5793 | } | |
5794 | ||
61e98203 | 5795 | static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5796 | { |
5797 | struct io_ring_ctx *ctx = req->ctx; | |
58f99373 PB |
5798 | u64 sqe_addr = req->cancel.addr; |
5799 | struct io_tctx_node *node; | |
5800 | int ret; | |
5801 | ||
5802 | /* tasks should wait for their io-wq threads, so safe w/o sync */ | |
5803 | ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); | |
5804 | spin_lock_irq(&ctx->completion_lock); | |
5805 | if (ret != -ENOENT) | |
5806 | goto done; | |
5807 | ret = io_timeout_cancel(ctx, sqe_addr); | |
5808 | if (ret != -ENOENT) | |
5809 | goto done; | |
5810 | ret = io_poll_cancel(ctx, sqe_addr); | |
5811 | if (ret != -ENOENT) | |
5812 | goto done; | |
5813 | spin_unlock_irq(&ctx->completion_lock); | |
5814 | ||
5815 | /* slow path, try all io-wq's */ | |
5816 | io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5817 | ret = -ENOENT; | |
5818 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
5819 | struct io_uring_task *tctx = node->task->io_uring; | |
fbf23849 | 5820 | |
58f99373 PB |
5821 | ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); |
5822 | if (ret != -ENOENT) | |
5823 | break; | |
5824 | } | |
5825 | io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK)); | |
5826 | ||
5827 | spin_lock_irq(&ctx->completion_lock); | |
5828 | done: | |
ff642164 | 5829 | io_cqring_fill_event(req, ret, 0); |
58f99373 PB |
5830 | io_commit_cqring(ctx); |
5831 | spin_unlock_irq(&ctx->completion_lock); | |
5832 | io_cqring_ev_posted(ctx); | |
5833 | ||
5834 | if (ret < 0) | |
5835 | req_set_fail_links(req); | |
5836 | io_put_req(req); | |
5262f567 JA |
5837 | return 0; |
5838 | } | |
5839 | ||
269bbe5f | 5840 | static int io_rsrc_update_prep(struct io_kiocb *req, |
05f3fb3c JA |
5841 | const struct io_uring_sqe *sqe) |
5842 | { | |
6ca56f84 JA |
5843 | if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL)) |
5844 | return -EINVAL; | |
61710e43 DA |
5845 | if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) |
5846 | return -EINVAL; | |
5847 | if (sqe->ioprio || sqe->rw_flags) | |
05f3fb3c JA |
5848 | return -EINVAL; |
5849 | ||
269bbe5f BM |
5850 | req->rsrc_update.offset = READ_ONCE(sqe->off); |
5851 | req->rsrc_update.nr_args = READ_ONCE(sqe->len); | |
5852 | if (!req->rsrc_update.nr_args) | |
05f3fb3c | 5853 | return -EINVAL; |
269bbe5f | 5854 | req->rsrc_update.arg = READ_ONCE(sqe->addr); |
05f3fb3c JA |
5855 | return 0; |
5856 | } | |
5857 | ||
889fca73 | 5858 | static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) |
fbf23849 JA |
5859 | { |
5860 | struct io_ring_ctx *ctx = req->ctx; | |
269bbe5f | 5861 | struct io_uring_rsrc_update up; |
05f3fb3c | 5862 | int ret; |
fbf23849 | 5863 | |
45d189c6 | 5864 | if (issue_flags & IO_URING_F_NONBLOCK) |
05f3fb3c | 5865 | return -EAGAIN; |
05f3fb3c | 5866 | |
269bbe5f BM |
5867 | up.offset = req->rsrc_update.offset; |
5868 | up.data = req->rsrc_update.arg; | |
05f3fb3c JA |
5869 | |
5870 | mutex_lock(&ctx->uring_lock); | |
269bbe5f | 5871 | ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args); |
05f3fb3c JA |
5872 | mutex_unlock(&ctx->uring_lock); |
5873 | ||
5874 | if (ret < 0) | |
5875 | req_set_fail_links(req); | |
889fca73 | 5876 | __io_req_complete(req, issue_flags, ret, 0); |
5262f567 JA |
5877 | return 0; |
5878 | } | |
5879 | ||
bfe76559 | 5880 | static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
f67676d1 | 5881 | { |
d625c6ee | 5882 | switch (req->opcode) { |
e781573e | 5883 | case IORING_OP_NOP: |
bfe76559 | 5884 | return 0; |
f67676d1 JA |
5885 | case IORING_OP_READV: |
5886 | case IORING_OP_READ_FIXED: | |
3a6820f2 | 5887 | case IORING_OP_READ: |
bfe76559 | 5888 | return io_read_prep(req, sqe); |
f67676d1 JA |
5889 | case IORING_OP_WRITEV: |
5890 | case IORING_OP_WRITE_FIXED: | |
3a6820f2 | 5891 | case IORING_OP_WRITE: |
bfe76559 | 5892 | return io_write_prep(req, sqe); |
0969e783 | 5893 | case IORING_OP_POLL_ADD: |
bfe76559 | 5894 | return io_poll_add_prep(req, sqe); |
0969e783 | 5895 | case IORING_OP_POLL_REMOVE: |
bfe76559 | 5896 | return io_poll_remove_prep(req, sqe); |
8ed8d3c3 | 5897 | case IORING_OP_FSYNC: |
1155c76a | 5898 | return io_fsync_prep(req, sqe); |
8ed8d3c3 | 5899 | case IORING_OP_SYNC_FILE_RANGE: |
1155c76a | 5900 | return io_sfr_prep(req, sqe); |
03b1230c | 5901 | case IORING_OP_SENDMSG: |
fddaface | 5902 | case IORING_OP_SEND: |
bfe76559 | 5903 | return io_sendmsg_prep(req, sqe); |
03b1230c | 5904 | case IORING_OP_RECVMSG: |
fddaface | 5905 | case IORING_OP_RECV: |
bfe76559 | 5906 | return io_recvmsg_prep(req, sqe); |
f499a021 | 5907 | case IORING_OP_CONNECT: |
bfe76559 | 5908 | return io_connect_prep(req, sqe); |
2d28390a | 5909 | case IORING_OP_TIMEOUT: |
bfe76559 | 5910 | return io_timeout_prep(req, sqe, false); |
b29472ee | 5911 | case IORING_OP_TIMEOUT_REMOVE: |
bfe76559 | 5912 | return io_timeout_remove_prep(req, sqe); |
fbf23849 | 5913 | case IORING_OP_ASYNC_CANCEL: |
bfe76559 | 5914 | return io_async_cancel_prep(req, sqe); |
2d28390a | 5915 | case IORING_OP_LINK_TIMEOUT: |
bfe76559 | 5916 | return io_timeout_prep(req, sqe, true); |
8ed8d3c3 | 5917 | case IORING_OP_ACCEPT: |
bfe76559 | 5918 | return io_accept_prep(req, sqe); |
d63d1b5e | 5919 | case IORING_OP_FALLOCATE: |
bfe76559 | 5920 | return io_fallocate_prep(req, sqe); |
15b71abe | 5921 | case IORING_OP_OPENAT: |
bfe76559 | 5922 | return io_openat_prep(req, sqe); |
b5dba59e | 5923 | case IORING_OP_CLOSE: |
bfe76559 | 5924 | return io_close_prep(req, sqe); |
05f3fb3c | 5925 | case IORING_OP_FILES_UPDATE: |
269bbe5f | 5926 | return io_rsrc_update_prep(req, sqe); |
eddc7ef5 | 5927 | case IORING_OP_STATX: |
bfe76559 | 5928 | return io_statx_prep(req, sqe); |
4840e418 | 5929 | case IORING_OP_FADVISE: |
bfe76559 | 5930 | return io_fadvise_prep(req, sqe); |
c1ca757b | 5931 | case IORING_OP_MADVISE: |
bfe76559 | 5932 | return io_madvise_prep(req, sqe); |
cebdb986 | 5933 | case IORING_OP_OPENAT2: |
bfe76559 | 5934 | return io_openat2_prep(req, sqe); |
3e4827b0 | 5935 | case IORING_OP_EPOLL_CTL: |
bfe76559 | 5936 | return io_epoll_ctl_prep(req, sqe); |
7d67af2c | 5937 | case IORING_OP_SPLICE: |
bfe76559 | 5938 | return io_splice_prep(req, sqe); |
ddf0322d | 5939 | case IORING_OP_PROVIDE_BUFFERS: |
bfe76559 | 5940 | return io_provide_buffers_prep(req, sqe); |
067524e9 | 5941 | case IORING_OP_REMOVE_BUFFERS: |
bfe76559 | 5942 | return io_remove_buffers_prep(req, sqe); |
f2a8d5c7 | 5943 | case IORING_OP_TEE: |
bfe76559 | 5944 | return io_tee_prep(req, sqe); |
36f4fa68 JA |
5945 | case IORING_OP_SHUTDOWN: |
5946 | return io_shutdown_prep(req, sqe); | |
80a261fd JA |
5947 | case IORING_OP_RENAMEAT: |
5948 | return io_renameat_prep(req, sqe); | |
14a1143b JA |
5949 | case IORING_OP_UNLINKAT: |
5950 | return io_unlinkat_prep(req, sqe); | |
f67676d1 JA |
5951 | } |
5952 | ||
bfe76559 PB |
5953 | printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n", |
5954 | req->opcode); | |
5955 | return-EINVAL; | |
5956 | } | |
5957 | ||
93642ef8 | 5958 | static int io_req_prep_async(struct io_kiocb *req) |
bfe76559 | 5959 | { |
b7e298d2 PB |
5960 | if (!io_op_defs[req->opcode].needs_async_setup) |
5961 | return 0; | |
5962 | if (WARN_ON_ONCE(req->async_data)) | |
5963 | return -EFAULT; | |
5964 | if (io_alloc_async_data(req)) | |
5965 | return -EAGAIN; | |
5966 | ||
93642ef8 PB |
5967 | switch (req->opcode) { |
5968 | case IORING_OP_READV: | |
93642ef8 PB |
5969 | return io_rw_prep_async(req, READ); |
5970 | case IORING_OP_WRITEV: | |
93642ef8 PB |
5971 | return io_rw_prep_async(req, WRITE); |
5972 | case IORING_OP_SENDMSG: | |
93642ef8 PB |
5973 | return io_sendmsg_prep_async(req); |
5974 | case IORING_OP_RECVMSG: | |
93642ef8 PB |
5975 | return io_recvmsg_prep_async(req); |
5976 | case IORING_OP_CONNECT: | |
5977 | return io_connect_prep_async(req); | |
5978 | } | |
b7e298d2 PB |
5979 | printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n", |
5980 | req->opcode); | |
5981 | return -EFAULT; | |
f67676d1 JA |
5982 | } |
5983 | ||
9cf7c104 PB |
5984 | static u32 io_get_sequence(struct io_kiocb *req) |
5985 | { | |
5986 | struct io_kiocb *pos; | |
5987 | struct io_ring_ctx *ctx = req->ctx; | |
f2f87370 | 5988 | u32 total_submitted, nr_reqs = 0; |
9cf7c104 | 5989 | |
f2f87370 PB |
5990 | io_for_each_link(pos, req) |
5991 | nr_reqs++; | |
9cf7c104 PB |
5992 | |
5993 | total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped; | |
5994 | return total_submitted - nr_reqs; | |
5995 | } | |
5996 | ||
be7053b7 | 5997 | static int io_req_defer(struct io_kiocb *req) |
de0617e4 | 5998 | { |
a197f664 | 5999 | struct io_ring_ctx *ctx = req->ctx; |
27dc8338 | 6000 | struct io_defer_entry *de; |
f67676d1 | 6001 | int ret; |
9cf7c104 | 6002 | u32 seq; |
de0617e4 | 6003 | |
9d858b21 | 6004 | /* Still need defer if there is pending req in defer list. */ |
9cf7c104 PB |
6005 | if (likely(list_empty_careful(&ctx->defer_list) && |
6006 | !(req->flags & REQ_F_IO_DRAIN))) | |
6007 | return 0; | |
6008 | ||
6009 | seq = io_get_sequence(req); | |
6010 | /* Still a chance to pass the sequence check */ | |
6011 | if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) | |
de0617e4 JA |
6012 | return 0; |
6013 | ||
b7e298d2 | 6014 | ret = io_req_prep_async(req); |
be7053b7 PB |
6015 | if (ret) |
6016 | return ret; | |
cbdcb435 | 6017 | io_prep_async_link(req); |
27dc8338 PB |
6018 | de = kmalloc(sizeof(*de), GFP_KERNEL); |
6019 | if (!de) | |
6020 | return -ENOMEM; | |
2d28390a | 6021 | |
de0617e4 | 6022 | spin_lock_irq(&ctx->completion_lock); |
9cf7c104 | 6023 | if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { |
de0617e4 | 6024 | spin_unlock_irq(&ctx->completion_lock); |
27dc8338 | 6025 | kfree(de); |
ae34817b PB |
6026 | io_queue_async_work(req); |
6027 | return -EIOCBQUEUED; | |
de0617e4 JA |
6028 | } |
6029 | ||
915967f6 | 6030 | trace_io_uring_defer(ctx, req, req->user_data); |
27dc8338 | 6031 | de->req = req; |
9cf7c104 | 6032 | de->seq = seq; |
27dc8338 | 6033 | list_add_tail(&de->list, &ctx->defer_list); |
de0617e4 JA |
6034 | spin_unlock_irq(&ctx->completion_lock); |
6035 | return -EIOCBQUEUED; | |
6036 | } | |
6037 | ||
68fb8979 | 6038 | static void io_clean_op(struct io_kiocb *req) |
99bc4c38 | 6039 | { |
0e1b6fe3 PB |
6040 | if (req->flags & REQ_F_BUFFER_SELECTED) { |
6041 | switch (req->opcode) { | |
6042 | case IORING_OP_READV: | |
6043 | case IORING_OP_READ_FIXED: | |
6044 | case IORING_OP_READ: | |
bcda7baa | 6045 | kfree((void *)(unsigned long)req->rw.addr); |
0e1b6fe3 PB |
6046 | break; |
6047 | case IORING_OP_RECVMSG: | |
6048 | case IORING_OP_RECV: | |
bcda7baa | 6049 | kfree(req->sr_msg.kbuf); |
0e1b6fe3 PB |
6050 | break; |
6051 | } | |
6052 | req->flags &= ~REQ_F_BUFFER_SELECTED; | |
99bc4c38 PB |
6053 | } |
6054 | ||
0e1b6fe3 PB |
6055 | if (req->flags & REQ_F_NEED_CLEANUP) { |
6056 | switch (req->opcode) { | |
6057 | case IORING_OP_READV: | |
6058 | case IORING_OP_READ_FIXED: | |
6059 | case IORING_OP_READ: | |
6060 | case IORING_OP_WRITEV: | |
6061 | case IORING_OP_WRITE_FIXED: | |
e8c2bc1f JA |
6062 | case IORING_OP_WRITE: { |
6063 | struct io_async_rw *io = req->async_data; | |
6064 | if (io->free_iovec) | |
6065 | kfree(io->free_iovec); | |
0e1b6fe3 | 6066 | break; |
e8c2bc1f | 6067 | } |
0e1b6fe3 | 6068 | case IORING_OP_RECVMSG: |
e8c2bc1f JA |
6069 | case IORING_OP_SENDMSG: { |
6070 | struct io_async_msghdr *io = req->async_data; | |
257e84a5 PB |
6071 | |
6072 | kfree(io->free_iov); | |
0e1b6fe3 | 6073 | break; |
e8c2bc1f | 6074 | } |
0e1b6fe3 PB |
6075 | case IORING_OP_SPLICE: |
6076 | case IORING_OP_TEE: | |
e1d767f0 PB |
6077 | if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED)) |
6078 | io_put_file(req->splice.file_in); | |
0e1b6fe3 | 6079 | break; |
f3cd4850 JA |
6080 | case IORING_OP_OPENAT: |
6081 | case IORING_OP_OPENAT2: | |
6082 | if (req->open.filename) | |
6083 | putname(req->open.filename); | |
6084 | break; | |
80a261fd JA |
6085 | case IORING_OP_RENAMEAT: |
6086 | putname(req->rename.oldpath); | |
6087 | putname(req->rename.newpath); | |
6088 | break; | |
14a1143b JA |
6089 | case IORING_OP_UNLINKAT: |
6090 | putname(req->unlink.filename); | |
6091 | break; | |
0e1b6fe3 PB |
6092 | } |
6093 | req->flags &= ~REQ_F_NEED_CLEANUP; | |
99bc4c38 | 6094 | } |
99bc4c38 PB |
6095 | } |
6096 | ||
889fca73 | 6097 | static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags) |
2b188cc1 | 6098 | { |
a197f664 | 6099 | struct io_ring_ctx *ctx = req->ctx; |
5730b27e | 6100 | const struct cred *creds = NULL; |
d625c6ee | 6101 | int ret; |
2b188cc1 | 6102 | |
003e8dcc JA |
6103 | if (req->work.creds && req->work.creds != current_cred()) |
6104 | creds = override_creds(req->work.creds); | |
5730b27e | 6105 | |
d625c6ee | 6106 | switch (req->opcode) { |
2b188cc1 | 6107 | case IORING_OP_NOP: |
889fca73 | 6108 | ret = io_nop(req, issue_flags); |
2b188cc1 JA |
6109 | break; |
6110 | case IORING_OP_READV: | |
edafccee | 6111 | case IORING_OP_READ_FIXED: |
3a6820f2 | 6112 | case IORING_OP_READ: |
889fca73 | 6113 | ret = io_read(req, issue_flags); |
edafccee | 6114 | break; |
3529d8c2 | 6115 | case IORING_OP_WRITEV: |
edafccee | 6116 | case IORING_OP_WRITE_FIXED: |
3a6820f2 | 6117 | case IORING_OP_WRITE: |
889fca73 | 6118 | ret = io_write(req, issue_flags); |
2b188cc1 | 6119 | break; |
c992fe29 | 6120 | case IORING_OP_FSYNC: |
45d189c6 | 6121 | ret = io_fsync(req, issue_flags); |
c992fe29 | 6122 | break; |
221c5eb2 | 6123 | case IORING_OP_POLL_ADD: |
61e98203 | 6124 | ret = io_poll_add(req, issue_flags); |
221c5eb2 JA |
6125 | break; |
6126 | case IORING_OP_POLL_REMOVE: | |
61e98203 | 6127 | ret = io_poll_remove(req, issue_flags); |
221c5eb2 | 6128 | break; |
5d17b4a4 | 6129 | case IORING_OP_SYNC_FILE_RANGE: |
45d189c6 | 6130 | ret = io_sync_file_range(req, issue_flags); |
5d17b4a4 | 6131 | break; |
0fa03c62 | 6132 | case IORING_OP_SENDMSG: |
889fca73 | 6133 | ret = io_sendmsg(req, issue_flags); |
062d04d7 | 6134 | break; |
fddaface | 6135 | case IORING_OP_SEND: |
889fca73 | 6136 | ret = io_send(req, issue_flags); |
0fa03c62 | 6137 | break; |
aa1fa28f | 6138 | case IORING_OP_RECVMSG: |
889fca73 | 6139 | ret = io_recvmsg(req, issue_flags); |
062d04d7 | 6140 | break; |
fddaface | 6141 | case IORING_OP_RECV: |
889fca73 | 6142 | ret = io_recv(req, issue_flags); |
aa1fa28f | 6143 | break; |
5262f567 | 6144 | case IORING_OP_TIMEOUT: |
61e98203 | 6145 | ret = io_timeout(req, issue_flags); |
5262f567 | 6146 | break; |
11365043 | 6147 | case IORING_OP_TIMEOUT_REMOVE: |
61e98203 | 6148 | ret = io_timeout_remove(req, issue_flags); |
11365043 | 6149 | break; |
17f2fe35 | 6150 | case IORING_OP_ACCEPT: |
889fca73 | 6151 | ret = io_accept(req, issue_flags); |
17f2fe35 | 6152 | break; |
f8e85cf2 | 6153 | case IORING_OP_CONNECT: |
889fca73 | 6154 | ret = io_connect(req, issue_flags); |
f8e85cf2 | 6155 | break; |
62755e35 | 6156 | case IORING_OP_ASYNC_CANCEL: |
61e98203 | 6157 | ret = io_async_cancel(req, issue_flags); |
62755e35 | 6158 | break; |
d63d1b5e | 6159 | case IORING_OP_FALLOCATE: |
45d189c6 | 6160 | ret = io_fallocate(req, issue_flags); |
d63d1b5e | 6161 | break; |
15b71abe | 6162 | case IORING_OP_OPENAT: |
45d189c6 | 6163 | ret = io_openat(req, issue_flags); |
15b71abe | 6164 | break; |
b5dba59e | 6165 | case IORING_OP_CLOSE: |
889fca73 | 6166 | ret = io_close(req, issue_flags); |
b5dba59e | 6167 | break; |
05f3fb3c | 6168 | case IORING_OP_FILES_UPDATE: |
889fca73 | 6169 | ret = io_files_update(req, issue_flags); |
05f3fb3c | 6170 | break; |
eddc7ef5 | 6171 | case IORING_OP_STATX: |
45d189c6 | 6172 | ret = io_statx(req, issue_flags); |
eddc7ef5 | 6173 | break; |
4840e418 | 6174 | case IORING_OP_FADVISE: |
45d189c6 | 6175 | ret = io_fadvise(req, issue_flags); |
4840e418 | 6176 | break; |
c1ca757b | 6177 | case IORING_OP_MADVISE: |
45d189c6 | 6178 | ret = io_madvise(req, issue_flags); |
c1ca757b | 6179 | break; |
cebdb986 | 6180 | case IORING_OP_OPENAT2: |
45d189c6 | 6181 | ret = io_openat2(req, issue_flags); |
cebdb986 | 6182 | break; |
3e4827b0 | 6183 | case IORING_OP_EPOLL_CTL: |
889fca73 | 6184 | ret = io_epoll_ctl(req, issue_flags); |
3e4827b0 | 6185 | break; |
7d67af2c | 6186 | case IORING_OP_SPLICE: |
45d189c6 | 6187 | ret = io_splice(req, issue_flags); |
7d67af2c | 6188 | break; |
ddf0322d | 6189 | case IORING_OP_PROVIDE_BUFFERS: |
889fca73 | 6190 | ret = io_provide_buffers(req, issue_flags); |
ddf0322d | 6191 | break; |
067524e9 | 6192 | case IORING_OP_REMOVE_BUFFERS: |
889fca73 | 6193 | ret = io_remove_buffers(req, issue_flags); |
3e4827b0 | 6194 | break; |
f2a8d5c7 | 6195 | case IORING_OP_TEE: |
45d189c6 | 6196 | ret = io_tee(req, issue_flags); |
f2a8d5c7 | 6197 | break; |
36f4fa68 | 6198 | case IORING_OP_SHUTDOWN: |
45d189c6 | 6199 | ret = io_shutdown(req, issue_flags); |
36f4fa68 | 6200 | break; |
80a261fd | 6201 | case IORING_OP_RENAMEAT: |
45d189c6 | 6202 | ret = io_renameat(req, issue_flags); |
80a261fd | 6203 | break; |
14a1143b | 6204 | case IORING_OP_UNLINKAT: |
45d189c6 | 6205 | ret = io_unlinkat(req, issue_flags); |
14a1143b | 6206 | break; |
2b188cc1 JA |
6207 | default: |
6208 | ret = -EINVAL; | |
6209 | break; | |
6210 | } | |
6211 | ||
5730b27e JA |
6212 | if (creds) |
6213 | revert_creds(creds); | |
6214 | ||
def596e9 JA |
6215 | if (ret) |
6216 | return ret; | |
6217 | ||
b532576e JA |
6218 | /* If the op doesn't have a file, we're not polling for it */ |
6219 | if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) { | |
11ba820b JA |
6220 | const bool in_async = io_wq_current_is_worker(); |
6221 | ||
11ba820b JA |
6222 | /* workqueue context doesn't hold uring_lock, grab it now */ |
6223 | if (in_async) | |
6224 | mutex_lock(&ctx->uring_lock); | |
6225 | ||
2e9dbe90 | 6226 | io_iopoll_req_issued(req, in_async); |
11ba820b JA |
6227 | |
6228 | if (in_async) | |
6229 | mutex_unlock(&ctx->uring_lock); | |
def596e9 JA |
6230 | } |
6231 | ||
6232 | return 0; | |
2b188cc1 JA |
6233 | } |
6234 | ||
5280f7e5 | 6235 | static void io_wq_submit_work(struct io_wq_work *work) |
2b188cc1 JA |
6236 | { |
6237 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
6df1db6b | 6238 | struct io_kiocb *timeout; |
561fb04a | 6239 | int ret = 0; |
2b188cc1 | 6240 | |
6df1db6b PB |
6241 | timeout = io_prep_linked_timeout(req); |
6242 | if (timeout) | |
6243 | io_queue_linked_timeout(timeout); | |
d4c81f38 | 6244 | |
4014d943 | 6245 | if (work->flags & IO_WQ_WORK_CANCEL) |
561fb04a | 6246 | ret = -ECANCELED; |
31b51510 | 6247 | |
561fb04a | 6248 | if (!ret) { |
561fb04a | 6249 | do { |
889fca73 | 6250 | ret = io_issue_sqe(req, 0); |
561fb04a JA |
6251 | /* |
6252 | * We can get EAGAIN for polled IO even though we're | |
6253 | * forcing a sync submission from here, since we can't | |
6254 | * wait for request slots on the block side. | |
6255 | */ | |
6256 | if (ret != -EAGAIN) | |
6257 | break; | |
6258 | cond_resched(); | |
6259 | } while (1); | |
6260 | } | |
31b51510 | 6261 | |
a3df7698 | 6262 | /* avoid locking problems by failing it from a clean context */ |
561fb04a | 6263 | if (ret) { |
a3df7698 | 6264 | /* io-wq is going to take one down */ |
de9b4cca | 6265 | req_ref_get(req); |
a3df7698 | 6266 | io_req_task_queue_fail(req, ret); |
edafccee | 6267 | } |
2b188cc1 JA |
6268 | } |
6269 | ||
7b29f92d JA |
6270 | #define FFS_ASYNC_READ 0x1UL |
6271 | #define FFS_ASYNC_WRITE 0x2UL | |
6272 | #ifdef CONFIG_64BIT | |
6273 | #define FFS_ISREG 0x4UL | |
6274 | #else | |
6275 | #define FFS_ISREG 0x0UL | |
6276 | #endif | |
6277 | #define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG) | |
6278 | ||
aeca241b | 6279 | static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table, |
a04b0ac0 | 6280 | unsigned i) |
65e19f54 | 6281 | { |
aeca241b | 6282 | struct io_fixed_file *table_l2; |
65e19f54 | 6283 | |
aeca241b PB |
6284 | table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT]; |
6285 | return &table_l2[i & IORING_FILE_TABLE_MASK]; | |
dafecf19 PB |
6286 | } |
6287 | ||
6288 | static inline struct file *io_file_from_index(struct io_ring_ctx *ctx, | |
6289 | int index) | |
6290 | { | |
aeca241b | 6291 | struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); |
7b29f92d | 6292 | |
a04b0ac0 | 6293 | return (struct file *) (slot->file_ptr & FFS_MASK); |
65e19f54 JA |
6294 | } |
6295 | ||
a04b0ac0 | 6296 | static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file) |
9a321c98 PB |
6297 | { |
6298 | unsigned long file_ptr = (unsigned long) file; | |
6299 | ||
6300 | if (__io_file_supports_async(file, READ)) | |
6301 | file_ptr |= FFS_ASYNC_READ; | |
6302 | if (__io_file_supports_async(file, WRITE)) | |
6303 | file_ptr |= FFS_ASYNC_WRITE; | |
6304 | if (S_ISREG(file_inode(file)->i_mode)) | |
6305 | file_ptr |= FFS_ISREG; | |
a04b0ac0 | 6306 | file_slot->file_ptr = file_ptr; |
9a321c98 PB |
6307 | } |
6308 | ||
8371adf5 PB |
6309 | static struct file *io_file_get(struct io_submit_state *state, |
6310 | struct io_kiocb *req, int fd, bool fixed) | |
09bb8394 | 6311 | { |
a197f664 | 6312 | struct io_ring_ctx *ctx = req->ctx; |
8da11c19 | 6313 | struct file *file; |
09bb8394 | 6314 | |
8da11c19 | 6315 | if (fixed) { |
7b29f92d JA |
6316 | unsigned long file_ptr; |
6317 | ||
479f517b | 6318 | if (unlikely((unsigned int)fd >= ctx->nr_user_files)) |
8371adf5 | 6319 | return NULL; |
b7620121 | 6320 | fd = array_index_nospec(fd, ctx->nr_user_files); |
aeca241b | 6321 | file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; |
7b29f92d JA |
6322 | file = (struct file *) (file_ptr & FFS_MASK); |
6323 | file_ptr &= ~FFS_MASK; | |
6324 | /* mask in overlapping REQ_F and FFS bits */ | |
6325 | req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT); | |
b895c9a6 | 6326 | io_req_set_rsrc_node(req); |
09bb8394 | 6327 | } else { |
c826bd7a | 6328 | trace_io_uring_file_get(ctx, fd); |
8da11c19 | 6329 | file = __io_file_get(state, fd); |
d44f554e JA |
6330 | |
6331 | /* we don't allow fixed io_uring files */ | |
6332 | if (file && unlikely(file->f_op == &io_uring_fops)) | |
6333 | io_req_track_inflight(req); | |
09bb8394 JA |
6334 | } |
6335 | ||
8371adf5 | 6336 | return file; |
09bb8394 JA |
6337 | } |
6338 | ||
2665abfd | 6339 | static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer) |
2b188cc1 | 6340 | { |
ad8a48ac JA |
6341 | struct io_timeout_data *data = container_of(timer, |
6342 | struct io_timeout_data, timer); | |
90cd7e42 | 6343 | struct io_kiocb *prev, *req = data->req; |
2665abfd | 6344 | struct io_ring_ctx *ctx = req->ctx; |
2665abfd | 6345 | unsigned long flags; |
2665abfd JA |
6346 | |
6347 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
90cd7e42 PB |
6348 | prev = req->timeout.head; |
6349 | req->timeout.head = NULL; | |
2665abfd JA |
6350 | |
6351 | /* | |
6352 | * We don't expect the list to be empty, that will only happen if we | |
6353 | * race with the completion of the linked work. | |
6354 | */ | |
de9b4cca | 6355 | if (prev && req_ref_inc_not_zero(prev)) |
f2f87370 | 6356 | io_remove_next_linked(prev); |
90cd7e42 PB |
6357 | else |
6358 | prev = NULL; | |
2665abfd JA |
6359 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
6360 | ||
6361 | if (prev) { | |
014db007 | 6362 | io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME); |
9ae1f8dd | 6363 | io_put_req_deferred(prev, 1); |
47f46768 | 6364 | } else { |
9ae1f8dd | 6365 | io_req_complete_post(req, -ETIME, 0); |
2665abfd | 6366 | } |
df9727af | 6367 | io_put_req_deferred(req, 1); |
2665abfd JA |
6368 | return HRTIMER_NORESTART; |
6369 | } | |
6370 | ||
de968c18 | 6371 | static void io_queue_linked_timeout(struct io_kiocb *req) |
2665abfd | 6372 | { |
de968c18 PB |
6373 | struct io_ring_ctx *ctx = req->ctx; |
6374 | ||
6375 | spin_lock_irq(&ctx->completion_lock); | |
76a46e06 | 6376 | /* |
f2f87370 PB |
6377 | * If the back reference is NULL, then our linked request finished |
6378 | * before we got a chance to setup the timer | |
76a46e06 | 6379 | */ |
90cd7e42 | 6380 | if (req->timeout.head) { |
e8c2bc1f | 6381 | struct io_timeout_data *data = req->async_data; |
94ae5e77 | 6382 | |
ad8a48ac JA |
6383 | data->timer.function = io_link_timeout_fn; |
6384 | hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), | |
6385 | data->mode); | |
2665abfd | 6386 | } |
76a46e06 | 6387 | spin_unlock_irq(&ctx->completion_lock); |
2665abfd | 6388 | /* drop submission reference */ |
76a46e06 JA |
6389 | io_put_req(req); |
6390 | } | |
2665abfd | 6391 | |
ad8a48ac | 6392 | static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req) |
2665abfd | 6393 | { |
f2f87370 | 6394 | struct io_kiocb *nxt = req->link; |
2665abfd | 6395 | |
f2f87370 PB |
6396 | if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) || |
6397 | nxt->opcode != IORING_OP_LINK_TIMEOUT) | |
76a46e06 | 6398 | return NULL; |
2665abfd | 6399 | |
90cd7e42 | 6400 | nxt->timeout.head = req; |
900fad45 | 6401 | nxt->flags |= REQ_F_LTIMEOUT_ACTIVE; |
76a46e06 | 6402 | req->flags |= REQ_F_LINK_TIMEOUT; |
76a46e06 | 6403 | return nxt; |
2665abfd JA |
6404 | } |
6405 | ||
c5eef2b9 | 6406 | static void __io_queue_sqe(struct io_kiocb *req) |
2b188cc1 | 6407 | { |
d3d7298d | 6408 | struct io_kiocb *linked_timeout = io_prep_linked_timeout(req); |
e0c5c576 | 6409 | int ret; |
2b188cc1 | 6410 | |
c5eef2b9 | 6411 | ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER); |
193155c8 | 6412 | |
491381ce JA |
6413 | /* |
6414 | * We async punt it if the file wasn't marked NOWAIT, or if the file | |
6415 | * doesn't support non-blocking read/write attempts | |
6416 | */ | |
1840038e | 6417 | if (likely(!ret)) { |
0d63c148 | 6418 | /* drop submission reference */ |
e342c807 | 6419 | if (req->flags & REQ_F_COMPLETE_INLINE) { |
c5eef2b9 PB |
6420 | struct io_ring_ctx *ctx = req->ctx; |
6421 | struct io_comp_state *cs = &ctx->submit_state.comp; | |
e65ef56d | 6422 | |
6dd0be1e | 6423 | cs->reqs[cs->nr++] = req; |
d3d7298d | 6424 | if (cs->nr == ARRAY_SIZE(cs->reqs)) |
c5eef2b9 | 6425 | io_submit_flush_completions(cs, ctx); |
9affd664 | 6426 | } else { |
d3d7298d | 6427 | io_put_req(req); |
0d63c148 | 6428 | } |
1840038e PB |
6429 | } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { |
6430 | if (!io_arm_poll_handler(req)) { | |
6431 | /* | |
6432 | * Queued up for async execution, worker will release | |
6433 | * submit reference when the iocb is actually submitted. | |
6434 | */ | |
6435 | io_queue_async_work(req); | |
6436 | } | |
0d63c148 | 6437 | } else { |
f41db273 | 6438 | io_req_complete_failed(req, ret); |
9e645e11 | 6439 | } |
d3d7298d PB |
6440 | if (linked_timeout) |
6441 | io_queue_linked_timeout(linked_timeout); | |
2b188cc1 JA |
6442 | } |
6443 | ||
be7053b7 | 6444 | static void io_queue_sqe(struct io_kiocb *req) |
4fe2c963 JL |
6445 | { |
6446 | int ret; | |
6447 | ||
be7053b7 | 6448 | ret = io_req_defer(req); |
4fe2c963 JL |
6449 | if (ret) { |
6450 | if (ret != -EIOCBQUEUED) { | |
1118591a | 6451 | fail_req: |
f41db273 | 6452 | io_req_complete_failed(req, ret); |
4fe2c963 | 6453 | } |
2550878f | 6454 | } else if (req->flags & REQ_F_FORCE_ASYNC) { |
b7e298d2 | 6455 | ret = io_req_prep_async(req); |
be7053b7 PB |
6456 | if (unlikely(ret)) |
6457 | goto fail_req; | |
ce35a47a JA |
6458 | io_queue_async_work(req); |
6459 | } else { | |
c5eef2b9 | 6460 | __io_queue_sqe(req); |
ce35a47a | 6461 | } |
4fe2c963 JL |
6462 | } |
6463 | ||
b16fed66 PB |
6464 | /* |
6465 | * Check SQE restrictions (opcode and flags). | |
6466 | * | |
6467 | * Returns 'true' if SQE is allowed, 'false' otherwise. | |
6468 | */ | |
6469 | static inline bool io_check_restriction(struct io_ring_ctx *ctx, | |
6470 | struct io_kiocb *req, | |
6471 | unsigned int sqe_flags) | |
4fe2c963 | 6472 | { |
b16fed66 PB |
6473 | if (!ctx->restricted) |
6474 | return true; | |
6475 | ||
6476 | if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) | |
6477 | return false; | |
6478 | ||
6479 | if ((sqe_flags & ctx->restrictions.sqe_flags_required) != | |
6480 | ctx->restrictions.sqe_flags_required) | |
6481 | return false; | |
6482 | ||
6483 | if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | | |
6484 | ctx->restrictions.sqe_flags_required)) | |
6485 | return false; | |
6486 | ||
6487 | return true; | |
4fe2c963 JL |
6488 | } |
6489 | ||
b16fed66 PB |
6490 | static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, |
6491 | const struct io_uring_sqe *sqe) | |
6492 | { | |
6493 | struct io_submit_state *state; | |
6494 | unsigned int sqe_flags; | |
003e8dcc | 6495 | int personality, ret = 0; |
b16fed66 PB |
6496 | |
6497 | req->opcode = READ_ONCE(sqe->opcode); | |
6498 | /* same numerical values with corresponding REQ_F_*, safe to copy */ | |
6499 | req->flags = sqe_flags = READ_ONCE(sqe->flags); | |
6500 | req->user_data = READ_ONCE(sqe->user_data); | |
6501 | req->async_data = NULL; | |
6502 | req->file = NULL; | |
6503 | req->ctx = ctx; | |
6504 | req->link = NULL; | |
6505 | req->fixed_rsrc_refs = NULL; | |
6506 | /* one is dropped after submission, the other at completion */ | |
abc54d63 | 6507 | atomic_set(&req->refs, 2); |
b16fed66 PB |
6508 | req->task = current; |
6509 | req->result = 0; | |
93e68e03 | 6510 | req->work.creds = NULL; |
b16fed66 PB |
6511 | |
6512 | /* enforce forwards compatibility on users */ | |
ebf4a5db PB |
6513 | if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) { |
6514 | req->flags = 0; | |
b16fed66 | 6515 | return -EINVAL; |
ebf4a5db | 6516 | } |
b16fed66 PB |
6517 | |
6518 | if (unlikely(req->opcode >= IORING_OP_LAST)) | |
6519 | return -EINVAL; | |
6520 | ||
b16fed66 PB |
6521 | if (unlikely(!io_check_restriction(ctx, req, sqe_flags))) |
6522 | return -EACCES; | |
6523 | ||
6524 | if ((sqe_flags & IOSQE_BUFFER_SELECT) && | |
6525 | !io_op_defs[req->opcode].buffer_select) | |
6526 | return -EOPNOTSUPP; | |
863e0560 | 6527 | |
003e8dcc JA |
6528 | personality = READ_ONCE(sqe->personality); |
6529 | if (personality) { | |
61cf9370 | 6530 | req->work.creds = xa_load(&ctx->personalities, personality); |
003e8dcc JA |
6531 | if (!req->work.creds) |
6532 | return -EINVAL; | |
6533 | get_cred(req->work.creds); | |
003e8dcc | 6534 | } |
b16fed66 PB |
6535 | state = &ctx->submit_state; |
6536 | ||
6537 | /* | |
6538 | * Plug now if we have more than 1 IO left after this, and the target | |
6539 | * is potentially a read/write to block based storage. | |
6540 | */ | |
6541 | if (!state->plug_started && state->ios_left > 1 && | |
6542 | io_op_defs[req->opcode].plug) { | |
6543 | blk_start_plug(&state->plug); | |
6544 | state->plug_started = true; | |
6545 | } | |
6546 | ||
6547 | if (io_op_defs[req->opcode].needs_file) { | |
6548 | bool fixed = req->flags & REQ_F_FIXED_FILE; | |
6549 | ||
6550 | req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed); | |
6551 | if (unlikely(!req->file)) | |
6552 | ret = -EBADF; | |
6553 | } | |
6554 | ||
6555 | state->ios_left--; | |
6556 | return ret; | |
6557 | } | |
6558 | ||
a6b8cadc | 6559 | static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, |
a1ab7b35 | 6560 | const struct io_uring_sqe *sqe) |
9e645e11 | 6561 | { |
a1ab7b35 | 6562 | struct io_submit_link *link = &ctx->submit_state.link; |
ef4ff581 | 6563 | int ret; |
9e645e11 | 6564 | |
a6b8cadc PB |
6565 | ret = io_init_req(ctx, req, sqe); |
6566 | if (unlikely(ret)) { | |
6567 | fail_req: | |
de59bc10 PB |
6568 | if (link->head) { |
6569 | /* fail even hard links since we don't submit */ | |
cf109604 | 6570 | link->head->flags |= REQ_F_FAIL_LINK; |
f41db273 | 6571 | io_req_complete_failed(link->head, -ECANCELED); |
de59bc10 PB |
6572 | link->head = NULL; |
6573 | } | |
f41db273 | 6574 | io_req_complete_failed(req, ret); |
a6b8cadc PB |
6575 | return ret; |
6576 | } | |
be7053b7 PB |
6577 | ret = io_req_prep(req, sqe); |
6578 | if (unlikely(ret)) | |
6579 | goto fail_req; | |
a6b8cadc | 6580 | |
be7053b7 | 6581 | /* don't need @sqe from now on */ |
a6b8cadc PB |
6582 | trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data, |
6583 | true, ctx->flags & IORING_SETUP_SQPOLL); | |
6584 | ||
9e645e11 JA |
6585 | /* |
6586 | * If we already have a head request, queue this one for async | |
6587 | * submittal once the head completes. If we don't have a head but | |
6588 | * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be | |
6589 | * submitted sync once the chain is complete. If none of those | |
6590 | * conditions are true (normal request), then just queue it. | |
6591 | */ | |
863e0560 PB |
6592 | if (link->head) { |
6593 | struct io_kiocb *head = link->head; | |
4e88d6e7 | 6594 | |
8cdf2193 PB |
6595 | /* |
6596 | * Taking sequential execution of a link, draining both sides | |
6597 | * of the link also fullfils IOSQE_IO_DRAIN semantics for all | |
6598 | * requests in the link. So, it drains the head and the | |
6599 | * next after the link request. The last one is done via | |
6600 | * drain_next flag to persist the effect across calls. | |
6601 | */ | |
ef4ff581 | 6602 | if (req->flags & REQ_F_IO_DRAIN) { |
711be031 PB |
6603 | head->flags |= REQ_F_IO_DRAIN; |
6604 | ctx->drain_next = 1; | |
6605 | } | |
b7e298d2 | 6606 | ret = io_req_prep_async(req); |
cf109604 | 6607 | if (unlikely(ret)) |
a6b8cadc | 6608 | goto fail_req; |
9d76377f | 6609 | trace_io_uring_link(ctx, req, head); |
f2f87370 | 6610 | link->last->link = req; |
863e0560 | 6611 | link->last = req; |
32fe525b PB |
6612 | |
6613 | /* last request of a link, enqueue the link */ | |
ef4ff581 | 6614 | if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { |
de59bc10 | 6615 | io_queue_sqe(head); |
863e0560 | 6616 | link->head = NULL; |
32fe525b | 6617 | } |
9e645e11 | 6618 | } else { |
711be031 PB |
6619 | if (unlikely(ctx->drain_next)) { |
6620 | req->flags |= REQ_F_IO_DRAIN; | |
ef4ff581 | 6621 | ctx->drain_next = 0; |
711be031 | 6622 | } |
ef4ff581 | 6623 | if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { |
863e0560 PB |
6624 | link->head = req; |
6625 | link->last = req; | |
711be031 | 6626 | } else { |
be7053b7 | 6627 | io_queue_sqe(req); |
711be031 | 6628 | } |
9e645e11 | 6629 | } |
2e6e1fde | 6630 | |
1d4240cc | 6631 | return 0; |
9e645e11 JA |
6632 | } |
6633 | ||
9a56a232 JA |
6634 | /* |
6635 | * Batched submission is done, ensure local IO is flushed out. | |
6636 | */ | |
ba88ff11 PB |
6637 | static void io_submit_state_end(struct io_submit_state *state, |
6638 | struct io_ring_ctx *ctx) | |
9a56a232 | 6639 | { |
a1ab7b35 | 6640 | if (state->link.head) |
de59bc10 | 6641 | io_queue_sqe(state->link.head); |
6dd0be1e | 6642 | if (state->comp.nr) |
ba88ff11 | 6643 | io_submit_flush_completions(&state->comp, ctx); |
27926b68 JA |
6644 | if (state->plug_started) |
6645 | blk_finish_plug(&state->plug); | |
9f13c35b | 6646 | io_state_file_put(state); |
9a56a232 JA |
6647 | } |
6648 | ||
6649 | /* | |
6650 | * Start submission side cache. | |
6651 | */ | |
6652 | static void io_submit_state_start(struct io_submit_state *state, | |
ba88ff11 | 6653 | unsigned int max_ios) |
9a56a232 | 6654 | { |
27926b68 | 6655 | state->plug_started = false; |
9a56a232 | 6656 | state->ios_left = max_ios; |
a1ab7b35 PB |
6657 | /* set only head, no need to init link_last in advance */ |
6658 | state->link.head = NULL; | |
9a56a232 JA |
6659 | } |
6660 | ||
2b188cc1 JA |
6661 | static void io_commit_sqring(struct io_ring_ctx *ctx) |
6662 | { | |
75b28aff | 6663 | struct io_rings *rings = ctx->rings; |
2b188cc1 | 6664 | |
caf582c6 PB |
6665 | /* |
6666 | * Ensure any loads from the SQEs are done at this point, | |
6667 | * since once we write the new head, the application could | |
6668 | * write new data to them. | |
6669 | */ | |
6670 | smp_store_release(&rings->sq.head, ctx->cached_sq_head); | |
2b188cc1 JA |
6671 | } |
6672 | ||
2b188cc1 | 6673 | /* |
3529d8c2 | 6674 | * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory |
2b188cc1 JA |
6675 | * that is mapped by userspace. This means that care needs to be taken to |
6676 | * ensure that reads are stable, as we cannot rely on userspace always | |
6677 | * being a good citizen. If members of the sqe are validated and then later | |
6678 | * used, it's important that those reads are done through READ_ONCE() to | |
6679 | * prevent a re-load down the line. | |
6680 | */ | |
709b302f | 6681 | static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx) |
2b188cc1 | 6682 | { |
75b28aff | 6683 | u32 *sq_array = ctx->sq_array; |
2b188cc1 JA |
6684 | unsigned head; |
6685 | ||
6686 | /* | |
6687 | * The cached sq head (or cq tail) serves two purposes: | |
6688 | * | |
6689 | * 1) allows us to batch the cost of updating the user visible | |
6690 | * head updates. | |
6691 | * 2) allows the kernel side to track the head on its own, even | |
6692 | * though the application is the one updating it. | |
6693 | */ | |
4fccfcbb | 6694 | head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]); |
709b302f PB |
6695 | if (likely(head < ctx->sq_entries)) |
6696 | return &ctx->sq_sqes[head]; | |
2b188cc1 JA |
6697 | |
6698 | /* drop invalid entries */ | |
498ccd9e | 6699 | ctx->cached_sq_dropped++; |
ee7d46d9 | 6700 | WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped); |
709b302f PB |
6701 | return NULL; |
6702 | } | |
6703 | ||
0f212204 | 6704 | static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) |
6c271ce2 | 6705 | { |
46c4e16a | 6706 | int submitted = 0; |
6c271ce2 | 6707 | |
c4a2ed72 | 6708 | /* if we have a backlog and couldn't flush it all, return BUSY */ |
ad3eb2c8 | 6709 | if (test_bit(0, &ctx->sq_check_overflow)) { |
6c2450ae | 6710 | if (!__io_cqring_overflow_flush(ctx, false)) |
ad3eb2c8 JA |
6711 | return -EBUSY; |
6712 | } | |
6c271ce2 | 6713 | |
ee7d46d9 PB |
6714 | /* make sure SQ entry isn't read before tail */ |
6715 | nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); | |
9ef4f124 | 6716 | |
2b85edfc PB |
6717 | if (!percpu_ref_tryget_many(&ctx->refs, nr)) |
6718 | return -EAGAIN; | |
6c271ce2 | 6719 | |
d8a6df10 | 6720 | percpu_counter_add(¤t->io_uring->inflight, nr); |
faf7b51c | 6721 | refcount_add(nr, ¤t->usage); |
ba88ff11 | 6722 | io_submit_state_start(&ctx->submit_state, nr); |
b14cca0c | 6723 | |
46c4e16a | 6724 | while (submitted < nr) { |
3529d8c2 | 6725 | const struct io_uring_sqe *sqe; |
196be95c | 6726 | struct io_kiocb *req; |
fb5ccc98 | 6727 | |
258b29a9 | 6728 | req = io_alloc_req(ctx); |
196be95c PB |
6729 | if (unlikely(!req)) { |
6730 | if (!submitted) | |
6731 | submitted = -EAGAIN; | |
fb5ccc98 | 6732 | break; |
196be95c | 6733 | } |
4fccfcbb PB |
6734 | sqe = io_get_sqe(ctx); |
6735 | if (unlikely(!sqe)) { | |
6736 | kmem_cache_free(req_cachep, req); | |
6737 | break; | |
6738 | } | |
d3656344 JA |
6739 | /* will complete beyond this point, count as submitted */ |
6740 | submitted++; | |
a1ab7b35 | 6741 | if (io_submit_sqe(ctx, req, sqe)) |
196be95c | 6742 | break; |
6c271ce2 JA |
6743 | } |
6744 | ||
9466f437 PB |
6745 | if (unlikely(submitted != nr)) { |
6746 | int ref_used = (submitted == -EAGAIN) ? 0 : submitted; | |
d8a6df10 JA |
6747 | struct io_uring_task *tctx = current->io_uring; |
6748 | int unused = nr - ref_used; | |
9466f437 | 6749 | |
d8a6df10 JA |
6750 | percpu_ref_put_many(&ctx->refs, unused); |
6751 | percpu_counter_sub(&tctx->inflight, unused); | |
6752 | put_task_struct_many(current, unused); | |
9466f437 | 6753 | } |
6c271ce2 | 6754 | |
a1ab7b35 | 6755 | io_submit_state_end(&ctx->submit_state, ctx); |
ae9428ca PB |
6756 | /* Commit SQ ring head once we've consumed and submitted all SQEs */ |
6757 | io_commit_sqring(ctx); | |
6758 | ||
6c271ce2 JA |
6759 | return submitted; |
6760 | } | |
6761 | ||
23b3628e XW |
6762 | static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx) |
6763 | { | |
6764 | /* Tell userspace we may need a wakeup call */ | |
6765 | spin_lock_irq(&ctx->completion_lock); | |
6766 | ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; | |
6767 | spin_unlock_irq(&ctx->completion_lock); | |
6768 | } | |
6769 | ||
6770 | static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx) | |
6771 | { | |
6772 | spin_lock_irq(&ctx->completion_lock); | |
6773 | ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; | |
6774 | spin_unlock_irq(&ctx->completion_lock); | |
6775 | } | |
6776 | ||
08369246 | 6777 | static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries) |
6c271ce2 | 6778 | { |
c8d1ba58 | 6779 | unsigned int to_submit; |
bdcd3eab | 6780 | int ret = 0; |
6c271ce2 | 6781 | |
c8d1ba58 | 6782 | to_submit = io_sqring_entries(ctx); |
e95eee2d JA |
6783 | /* if we're handling multiple rings, cap submit size for fairness */ |
6784 | if (cap_entries && to_submit > 8) | |
6785 | to_submit = 8; | |
6786 | ||
906a3c6f | 6787 | if (!list_empty(&ctx->iopoll_list) || to_submit) { |
c8d1ba58 | 6788 | unsigned nr_events = 0; |
a4c0b3de | 6789 | |
c8d1ba58 | 6790 | mutex_lock(&ctx->uring_lock); |
906a3c6f | 6791 | if (!list_empty(&ctx->iopoll_list)) |
c8d1ba58 | 6792 | io_do_iopoll(ctx, &nr_events, 0); |
906a3c6f | 6793 | |
0298ef96 PB |
6794 | if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && |
6795 | !(ctx->flags & IORING_SETUP_R_DISABLED)) | |
08369246 | 6796 | ret = io_submit_sqes(ctx, to_submit); |
c8d1ba58 JA |
6797 | mutex_unlock(&ctx->uring_lock); |
6798 | } | |
6c271ce2 | 6799 | |
90554200 JA |
6800 | if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait)) |
6801 | wake_up(&ctx->sqo_sq_wait); | |
6c271ce2 | 6802 | |
08369246 XW |
6803 | return ret; |
6804 | } | |
6c271ce2 | 6805 | |
08369246 XW |
6806 | static void io_sqd_update_thread_idle(struct io_sq_data *sqd) |
6807 | { | |
6808 | struct io_ring_ctx *ctx; | |
6809 | unsigned sq_thread_idle = 0; | |
6c271ce2 | 6810 | |
c9dca27d PB |
6811 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6812 | sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); | |
08369246 | 6813 | sqd->sq_thread_idle = sq_thread_idle; |
c8d1ba58 | 6814 | } |
6c271ce2 | 6815 | |
c8d1ba58 JA |
6816 | static int io_sq_thread(void *data) |
6817 | { | |
69fb2131 JA |
6818 | struct io_sq_data *sqd = data; |
6819 | struct io_ring_ctx *ctx; | |
a0d9205f | 6820 | unsigned long timeout = 0; |
37d1e2e3 | 6821 | char buf[TASK_COMM_LEN]; |
08369246 | 6822 | DEFINE_WAIT(wait); |
6c271ce2 | 6823 | |
696ee88a | 6824 | snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); |
37d1e2e3 | 6825 | set_task_comm(current, buf); |
37d1e2e3 JA |
6826 | current->pf_io_worker = NULL; |
6827 | ||
6828 | if (sqd->sq_cpu != -1) | |
6829 | set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); | |
6830 | else | |
6831 | set_cpus_allowed_ptr(current, cpu_online_mask); | |
6832 | current->flags |= PF_NO_SETAFFINITY; | |
6833 | ||
09a6f4ef | 6834 | mutex_lock(&sqd->lock); |
05962f95 | 6835 | while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) { |
08369246 XW |
6836 | int ret; |
6837 | bool cap_entries, sqt_spin, needs_sched; | |
c1edbf5f | 6838 | |
82734c5b JA |
6839 | if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || |
6840 | signal_pending(current)) { | |
6841 | bool did_sig = false; | |
6842 | ||
09a6f4ef | 6843 | mutex_unlock(&sqd->lock); |
82734c5b JA |
6844 | if (signal_pending(current)) { |
6845 | struct ksignal ksig; | |
6846 | ||
6847 | did_sig = get_signal(&ksig); | |
6848 | } | |
05962f95 | 6849 | cond_resched(); |
09a6f4ef | 6850 | mutex_lock(&sqd->lock); |
82734c5b JA |
6851 | if (did_sig) |
6852 | break; | |
521d6a73 | 6853 | io_run_task_work(); |
b7f5a0bf | 6854 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6855 | timeout = jiffies + sqd->sq_thread_idle; |
7d41e854 | 6856 | continue; |
08369246 | 6857 | } |
08369246 | 6858 | sqt_spin = false; |
e95eee2d | 6859 | cap_entries = !list_is_singular(&sqd->ctx_list); |
69fb2131 | 6860 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { |
7c30f36a SM |
6861 | const struct cred *creds = NULL; |
6862 | ||
6863 | if (ctx->sq_creds != current_cred()) | |
6864 | creds = override_creds(ctx->sq_creds); | |
08369246 | 6865 | ret = __io_sq_thread(ctx, cap_entries); |
7c30f36a SM |
6866 | if (creds) |
6867 | revert_creds(creds); | |
08369246 XW |
6868 | if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) |
6869 | sqt_spin = true; | |
69fb2131 | 6870 | } |
6c271ce2 | 6871 | |
08369246 | 6872 | if (sqt_spin || !time_after(jiffies, timeout)) { |
c8d1ba58 JA |
6873 | io_run_task_work(); |
6874 | cond_resched(); | |
08369246 XW |
6875 | if (sqt_spin) |
6876 | timeout = jiffies + sqd->sq_thread_idle; | |
6877 | continue; | |
6878 | } | |
6879 | ||
08369246 XW |
6880 | needs_sched = true; |
6881 | prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); | |
6882 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { | |
6883 | if ((ctx->flags & IORING_SETUP_IOPOLL) && | |
6884 | !list_empty_careful(&ctx->iopoll_list)) { | |
6885 | needs_sched = false; | |
6886 | break; | |
6887 | } | |
6888 | if (io_sqring_entries(ctx)) { | |
6889 | needs_sched = false; | |
6890 | break; | |
6891 | } | |
6892 | } | |
6893 | ||
05962f95 | 6894 | if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) { |
69fb2131 JA |
6895 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6896 | io_ring_set_wakeup_flag(ctx); | |
08369246 | 6897 | |
09a6f4ef | 6898 | mutex_unlock(&sqd->lock); |
69fb2131 | 6899 | schedule(); |
09a6f4ef | 6900 | mutex_lock(&sqd->lock); |
69fb2131 JA |
6901 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6902 | io_ring_clear_wakeup_flag(ctx); | |
6c271ce2 | 6903 | } |
08369246 XW |
6904 | |
6905 | finish_wait(&sqd->wait, &wait); | |
b7f5a0bf | 6906 | io_run_task_work_head(&sqd->park_task_work); |
08369246 | 6907 | timeout = jiffies + sqd->sq_thread_idle; |
6c271ce2 | 6908 | } |
28cea78a | 6909 | |
521d6a73 PB |
6910 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
6911 | io_uring_cancel_sqpoll(ctx); | |
37d1e2e3 | 6912 | sqd->thread = NULL; |
05962f95 | 6913 | list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) |
5f3f26f9 | 6914 | io_ring_set_wakeup_flag(ctx); |
09a6f4ef | 6915 | mutex_unlock(&sqd->lock); |
521d6a73 PB |
6916 | |
6917 | io_run_task_work(); | |
b7f5a0bf | 6918 | io_run_task_work_head(&sqd->park_task_work); |
37d1e2e3 JA |
6919 | complete(&sqd->exited); |
6920 | do_exit(0); | |
6c271ce2 JA |
6921 | } |
6922 | ||
bda52162 JA |
6923 | struct io_wait_queue { |
6924 | struct wait_queue_entry wq; | |
6925 | struct io_ring_ctx *ctx; | |
6926 | unsigned to_wait; | |
6927 | unsigned nr_timeouts; | |
6928 | }; | |
6929 | ||
6c503150 | 6930 | static inline bool io_should_wake(struct io_wait_queue *iowq) |
bda52162 JA |
6931 | { |
6932 | struct io_ring_ctx *ctx = iowq->ctx; | |
6933 | ||
6934 | /* | |
d195a66e | 6935 | * Wake up if we have enough events, or if a timeout occurred since we |
bda52162 JA |
6936 | * started waiting. For timeouts, we always want to return to userspace, |
6937 | * regardless of event count. | |
6938 | */ | |
6c503150 | 6939 | return io_cqring_events(ctx) >= iowq->to_wait || |
bda52162 JA |
6940 | atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; |
6941 | } | |
6942 | ||
6943 | static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode, | |
6944 | int wake_flags, void *key) | |
6945 | { | |
6946 | struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, | |
6947 | wq); | |
6948 | ||
6c503150 PB |
6949 | /* |
6950 | * Cannot safely flush overflowed CQEs from here, ensure we wake up | |
6951 | * the task, and the next invocation will do it. | |
6952 | */ | |
6953 | if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow)) | |
6954 | return autoremove_wake_function(curr, mode, wake_flags, key); | |
6955 | return -1; | |
bda52162 JA |
6956 | } |
6957 | ||
af9c1a44 JA |
6958 | static int io_run_task_work_sig(void) |
6959 | { | |
6960 | if (io_run_task_work()) | |
6961 | return 1; | |
6962 | if (!signal_pending(current)) | |
6963 | return 0; | |
0b8cfa97 | 6964 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) |
792ee0f6 | 6965 | return -ERESTARTSYS; |
af9c1a44 JA |
6966 | return -EINTR; |
6967 | } | |
6968 | ||
eeb60b9a PB |
6969 | /* when returns >0, the caller should retry */ |
6970 | static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx, | |
6971 | struct io_wait_queue *iowq, | |
6972 | signed long *timeout) | |
6973 | { | |
6974 | int ret; | |
6975 | ||
6976 | /* make sure we run task_work before checking for signals */ | |
6977 | ret = io_run_task_work_sig(); | |
6978 | if (ret || io_should_wake(iowq)) | |
6979 | return ret; | |
6980 | /* let the caller flush overflows, retry */ | |
6981 | if (test_bit(0, &ctx->cq_check_overflow)) | |
6982 | return 1; | |
6983 | ||
6984 | *timeout = schedule_timeout(*timeout); | |
6985 | return !*timeout ? -ETIME : 1; | |
6986 | } | |
6987 | ||
2b188cc1 JA |
6988 | /* |
6989 | * Wait until events become available, if we don't already have some. The | |
6990 | * application must reap them itself, as they reside on the shared cq ring. | |
6991 | */ | |
6992 | static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events, | |
c73ebb68 HX |
6993 | const sigset_t __user *sig, size_t sigsz, |
6994 | struct __kernel_timespec __user *uts) | |
2b188cc1 | 6995 | { |
bda52162 JA |
6996 | struct io_wait_queue iowq = { |
6997 | .wq = { | |
6998 | .private = current, | |
6999 | .func = io_wake_function, | |
7000 | .entry = LIST_HEAD_INIT(iowq.wq.entry), | |
7001 | }, | |
7002 | .ctx = ctx, | |
7003 | .to_wait = min_events, | |
7004 | }; | |
75b28aff | 7005 | struct io_rings *rings = ctx->rings; |
c1d5a224 PB |
7006 | signed long timeout = MAX_SCHEDULE_TIMEOUT; |
7007 | int ret; | |
2b188cc1 | 7008 | |
b41e9852 | 7009 | do { |
6c2450ae | 7010 | io_cqring_overflow_flush(ctx, false); |
6c503150 | 7011 | if (io_cqring_events(ctx) >= min_events) |
b41e9852 | 7012 | return 0; |
4c6e277c | 7013 | if (!io_run_task_work()) |
b41e9852 | 7014 | break; |
b41e9852 | 7015 | } while (1); |
2b188cc1 JA |
7016 | |
7017 | if (sig) { | |
9e75ad5d AB |
7018 | #ifdef CONFIG_COMPAT |
7019 | if (in_compat_syscall()) | |
7020 | ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig, | |
b772434b | 7021 | sigsz); |
9e75ad5d AB |
7022 | else |
7023 | #endif | |
b772434b | 7024 | ret = set_user_sigmask(sig, sigsz); |
9e75ad5d | 7025 | |
2b188cc1 JA |
7026 | if (ret) |
7027 | return ret; | |
7028 | } | |
7029 | ||
c73ebb68 | 7030 | if (uts) { |
c1d5a224 PB |
7031 | struct timespec64 ts; |
7032 | ||
c73ebb68 HX |
7033 | if (get_timespec64(&ts, uts)) |
7034 | return -EFAULT; | |
7035 | timeout = timespec64_to_jiffies(&ts); | |
7036 | } | |
7037 | ||
bda52162 | 7038 | iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); |
c826bd7a | 7039 | trace_io_uring_cqring_wait(ctx, min_events); |
bda52162 | 7040 | do { |
ca0a2651 | 7041 | /* if we can't even flush overflow, don't wait for more */ |
6c2450ae | 7042 | if (!io_cqring_overflow_flush(ctx, false)) { |
ca0a2651 JA |
7043 | ret = -EBUSY; |
7044 | break; | |
7045 | } | |
bda52162 JA |
7046 | prepare_to_wait_exclusive(&ctx->wait, &iowq.wq, |
7047 | TASK_INTERRUPTIBLE); | |
eeb60b9a PB |
7048 | ret = io_cqring_wait_schedule(ctx, &iowq, &timeout); |
7049 | finish_wait(&ctx->wait, &iowq.wq); | |
ca0a2651 | 7050 | cond_resched(); |
eeb60b9a | 7051 | } while (ret > 0); |
bda52162 | 7052 | |
b7db41c9 | 7053 | restore_saved_sigmask_unless(ret == -EINTR); |
2b188cc1 | 7054 | |
75b28aff | 7055 | return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; |
2b188cc1 JA |
7056 | } |
7057 | ||
aeca241b | 7058 | static void io_free_file_tables(struct io_file_table *table, unsigned nr_files) |
846a4ef2 PB |
7059 | { |
7060 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); | |
7061 | ||
7062 | for (i = 0; i < nr_tables; i++) | |
aeca241b PB |
7063 | kfree(table->files[i]); |
7064 | kfree(table->files); | |
7065 | table->files = NULL; | |
846a4ef2 PB |
7066 | } |
7067 | ||
6b06314c JA |
7068 | static void __io_sqe_files_unregister(struct io_ring_ctx *ctx) |
7069 | { | |
7070 | #if defined(CONFIG_UNIX) | |
7071 | if (ctx->ring_sock) { | |
7072 | struct sock *sock = ctx->ring_sock->sk; | |
7073 | struct sk_buff *skb; | |
7074 | ||
7075 | while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) | |
7076 | kfree_skb(skb); | |
7077 | } | |
7078 | #else | |
7079 | int i; | |
7080 | ||
65e19f54 JA |
7081 | for (i = 0; i < ctx->nr_user_files; i++) { |
7082 | struct file *file; | |
7083 | ||
7084 | file = io_file_from_index(ctx, i); | |
7085 | if (file) | |
7086 | fput(file); | |
7087 | } | |
6b06314c | 7088 | #endif |
08480400 PB |
7089 | io_free_file_tables(&ctx->file_table, ctx->nr_user_files); |
7090 | kfree(ctx->file_data); | |
7091 | ctx->file_data = NULL; | |
7092 | ctx->nr_user_files = 0; | |
6b06314c JA |
7093 | } |
7094 | ||
2a63b2d9 | 7095 | static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx) |
1642b445 | 7096 | { |
2a63b2d9 | 7097 | spin_lock_bh(&ctx->rsrc_ref_lock); |
1642b445 PB |
7098 | } |
7099 | ||
2a63b2d9 | 7100 | static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx) |
6b06314c | 7101 | { |
2a63b2d9 BM |
7102 | spin_unlock_bh(&ctx->rsrc_ref_lock); |
7103 | } | |
65e19f54 | 7104 | |
28a9fe25 PB |
7105 | static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node) |
7106 | { | |
7107 | percpu_ref_exit(&ref_node->refs); | |
7108 | kfree(ref_node); | |
7109 | } | |
7110 | ||
a7f0ed5a PB |
7111 | static void io_rsrc_node_switch(struct io_ring_ctx *ctx, |
7112 | struct io_rsrc_data *data_to_kill) | |
1642b445 | 7113 | { |
a7f0ed5a PB |
7114 | WARN_ON_ONCE(!ctx->rsrc_backup_node); |
7115 | WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); | |
82fbcfa9 | 7116 | |
a7f0ed5a PB |
7117 | if (data_to_kill) { |
7118 | struct io_rsrc_node *rsrc_node = ctx->rsrc_node; | |
82fbcfa9 | 7119 | |
a7f0ed5a PB |
7120 | rsrc_node->rsrc_data = data_to_kill; |
7121 | io_rsrc_ref_lock(ctx); | |
7122 | list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); | |
7123 | io_rsrc_ref_unlock(ctx); | |
82fbcfa9 | 7124 | |
3e942498 | 7125 | atomic_inc(&data_to_kill->refs); |
a7f0ed5a PB |
7126 | percpu_ref_kill(&rsrc_node->refs); |
7127 | ctx->rsrc_node = NULL; | |
7128 | } | |
6b06314c | 7129 | |
a7f0ed5a PB |
7130 | if (!ctx->rsrc_node) { |
7131 | ctx->rsrc_node = ctx->rsrc_backup_node; | |
7132 | ctx->rsrc_backup_node = NULL; | |
7133 | } | |
8bad28d8 HX |
7134 | } |
7135 | ||
a7f0ed5a | 7136 | static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx) |
8dd03afe PB |
7137 | { |
7138 | if (ctx->rsrc_backup_node) | |
7139 | return 0; | |
b895c9a6 | 7140 | ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); |
8dd03afe PB |
7141 | return ctx->rsrc_backup_node ? 0 : -ENOMEM; |
7142 | } | |
7143 | ||
40ae0ff7 | 7144 | static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx) |
8bad28d8 HX |
7145 | { |
7146 | int ret; | |
05589553 | 7147 | |
215c3902 | 7148 | /* As we may drop ->uring_lock, other task may have started quiesce */ |
8bad28d8 HX |
7149 | if (data->quiesce) |
7150 | return -ENXIO; | |
05589553 | 7151 | |
8bad28d8 | 7152 | data->quiesce = true; |
1ffc5422 | 7153 | do { |
a7f0ed5a | 7154 | ret = io_rsrc_node_switch_start(ctx); |
8dd03afe | 7155 | if (ret) |
f2303b1f | 7156 | break; |
a7f0ed5a PB |
7157 | io_rsrc_node_switch(ctx, data); |
7158 | ||
3e942498 PB |
7159 | /* kill initial ref, already quiesced if zero */ |
7160 | if (atomic_dec_and_test(&data->refs)) | |
7161 | break; | |
8bad28d8 | 7162 | flush_delayed_work(&ctx->rsrc_put_work); |
1ffc5422 PB |
7163 | ret = wait_for_completion_interruptible(&data->done); |
7164 | if (!ret) | |
7165 | break; | |
8bad28d8 | 7166 | |
3e942498 PB |
7167 | atomic_inc(&data->refs); |
7168 | /* wait for all works potentially completing data->done */ | |
7169 | flush_delayed_work(&ctx->rsrc_put_work); | |
cb5e1b81 | 7170 | reinit_completion(&data->done); |
8dd03afe | 7171 | |
8bad28d8 | 7172 | mutex_unlock(&ctx->uring_lock); |
1ffc5422 | 7173 | ret = io_run_task_work_sig(); |
8bad28d8 | 7174 | mutex_lock(&ctx->uring_lock); |
f2303b1f | 7175 | } while (ret >= 0); |
8bad28d8 | 7176 | data->quiesce = false; |
05f3fb3c | 7177 | |
8bad28d8 | 7178 | return ret; |
d7954b2b BM |
7179 | } |
7180 | ||
40ae0ff7 PB |
7181 | static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx, |
7182 | rsrc_put_fn *do_put) | |
1ad555c6 | 7183 | { |
b895c9a6 | 7184 | struct io_rsrc_data *data; |
1ad555c6 BM |
7185 | |
7186 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
7187 | if (!data) | |
7188 | return NULL; | |
7189 | ||
3e942498 | 7190 | atomic_set(&data->refs, 1); |
1ad555c6 | 7191 | data->ctx = ctx; |
40ae0ff7 | 7192 | data->do_put = do_put; |
1ad555c6 BM |
7193 | init_completion(&data->done); |
7194 | return data; | |
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; | |
7c30f36a SM |
7271 | if (ctx->sq_creds) |
7272 | put_cred(ctx->sq_creds); | |
534ca6d6 JA |
7273 | } |
7274 | } | |
7275 | ||
aa06165d JA |
7276 | static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p) |
7277 | { | |
7278 | struct io_ring_ctx *ctx_attach; | |
7279 | struct io_sq_data *sqd; | |
7280 | struct fd f; | |
7281 | ||
7282 | f = fdget(p->wq_fd); | |
7283 | if (!f.file) | |
7284 | return ERR_PTR(-ENXIO); | |
7285 | if (f.file->f_op != &io_uring_fops) { | |
7286 | fdput(f); | |
7287 | return ERR_PTR(-EINVAL); | |
7288 | } | |
7289 | ||
7290 | ctx_attach = f.file->private_data; | |
7291 | sqd = ctx_attach->sq_data; | |
7292 | if (!sqd) { | |
7293 | fdput(f); | |
7294 | return ERR_PTR(-EINVAL); | |
7295 | } | |
5c2469e0 JA |
7296 | if (sqd->task_tgid != current->tgid) { |
7297 | fdput(f); | |
7298 | return ERR_PTR(-EPERM); | |
7299 | } | |
aa06165d JA |
7300 | |
7301 | refcount_inc(&sqd->refs); | |
7302 | fdput(f); | |
7303 | return sqd; | |
7304 | } | |
7305 | ||
26984fbf PB |
7306 | static struct io_sq_data *io_get_sq_data(struct io_uring_params *p, |
7307 | bool *attached) | |
534ca6d6 JA |
7308 | { |
7309 | struct io_sq_data *sqd; | |
7310 | ||
26984fbf | 7311 | *attached = false; |
5c2469e0 JA |
7312 | if (p->flags & IORING_SETUP_ATTACH_WQ) { |
7313 | sqd = io_attach_sq_data(p); | |
26984fbf PB |
7314 | if (!IS_ERR(sqd)) { |
7315 | *attached = true; | |
5c2469e0 | 7316 | return sqd; |
26984fbf | 7317 | } |
5c2469e0 JA |
7318 | /* fall through for EPERM case, setup new sqd/task */ |
7319 | if (PTR_ERR(sqd) != -EPERM) | |
7320 | return sqd; | |
7321 | } | |
aa06165d | 7322 | |
534ca6d6 JA |
7323 | sqd = kzalloc(sizeof(*sqd), GFP_KERNEL); |
7324 | if (!sqd) | |
7325 | return ERR_PTR(-ENOMEM); | |
7326 | ||
9e138a48 | 7327 | atomic_set(&sqd->park_pending, 0); |
534ca6d6 | 7328 | refcount_set(&sqd->refs, 1); |
69fb2131 | 7329 | INIT_LIST_HEAD(&sqd->ctx_list); |
09a6f4ef | 7330 | mutex_init(&sqd->lock); |
534ca6d6 | 7331 | init_waitqueue_head(&sqd->wait); |
37d1e2e3 | 7332 | init_completion(&sqd->exited); |
534ca6d6 JA |
7333 | return sqd; |
7334 | } | |
7335 | ||
6b06314c | 7336 | #if defined(CONFIG_UNIX) |
6b06314c JA |
7337 | /* |
7338 | * Ensure the UNIX gc is aware of our file set, so we are certain that | |
7339 | * the io_uring can be safely unregistered on process exit, even if we have | |
7340 | * loops in the file referencing. | |
7341 | */ | |
7342 | static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset) | |
7343 | { | |
7344 | struct sock *sk = ctx->ring_sock->sk; | |
7345 | struct scm_fp_list *fpl; | |
7346 | struct sk_buff *skb; | |
08a45173 | 7347 | int i, nr_files; |
6b06314c | 7348 | |
6b06314c JA |
7349 | fpl = kzalloc(sizeof(*fpl), GFP_KERNEL); |
7350 | if (!fpl) | |
7351 | return -ENOMEM; | |
7352 | ||
7353 | skb = alloc_skb(0, GFP_KERNEL); | |
7354 | if (!skb) { | |
7355 | kfree(fpl); | |
7356 | return -ENOMEM; | |
7357 | } | |
7358 | ||
7359 | skb->sk = sk; | |
6b06314c | 7360 | |
08a45173 | 7361 | nr_files = 0; |
62e398be | 7362 | fpl->user = get_uid(current_user()); |
6b06314c | 7363 | for (i = 0; i < nr; i++) { |
65e19f54 JA |
7364 | struct file *file = io_file_from_index(ctx, i + offset); |
7365 | ||
7366 | if (!file) | |
08a45173 | 7367 | continue; |
65e19f54 | 7368 | fpl->fp[nr_files] = get_file(file); |
08a45173 JA |
7369 | unix_inflight(fpl->user, fpl->fp[nr_files]); |
7370 | nr_files++; | |
6b06314c JA |
7371 | } |
7372 | ||
08a45173 JA |
7373 | if (nr_files) { |
7374 | fpl->max = SCM_MAX_FD; | |
7375 | fpl->count = nr_files; | |
7376 | UNIXCB(skb).fp = fpl; | |
05f3fb3c | 7377 | skb->destructor = unix_destruct_scm; |
08a45173 JA |
7378 | refcount_add(skb->truesize, &sk->sk_wmem_alloc); |
7379 | skb_queue_head(&sk->sk_receive_queue, skb); | |
6b06314c | 7380 | |
08a45173 JA |
7381 | for (i = 0; i < nr_files; i++) |
7382 | fput(fpl->fp[i]); | |
7383 | } else { | |
7384 | kfree_skb(skb); | |
7385 | kfree(fpl); | |
7386 | } | |
6b06314c JA |
7387 | |
7388 | return 0; | |
7389 | } | |
7390 | ||
7391 | /* | |
7392 | * If UNIX sockets are enabled, fd passing can cause a reference cycle which | |
7393 | * causes regular reference counting to break down. We rely on the UNIX | |
7394 | * garbage collection to take care of this problem for us. | |
7395 | */ | |
7396 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7397 | { | |
7398 | unsigned left, total; | |
7399 | int ret = 0; | |
7400 | ||
7401 | total = 0; | |
7402 | left = ctx->nr_user_files; | |
7403 | while (left) { | |
7404 | unsigned this_files = min_t(unsigned, left, SCM_MAX_FD); | |
6b06314c JA |
7405 | |
7406 | ret = __io_sqe_files_scm(ctx, this_files, total); | |
7407 | if (ret) | |
7408 | break; | |
7409 | left -= this_files; | |
7410 | total += this_files; | |
7411 | } | |
7412 | ||
7413 | if (!ret) | |
7414 | return 0; | |
7415 | ||
7416 | while (total < ctx->nr_user_files) { | |
65e19f54 JA |
7417 | struct file *file = io_file_from_index(ctx, total); |
7418 | ||
7419 | if (file) | |
7420 | fput(file); | |
6b06314c JA |
7421 | total++; |
7422 | } | |
7423 | ||
7424 | return ret; | |
7425 | } | |
7426 | #else | |
7427 | static int io_sqe_files_scm(struct io_ring_ctx *ctx) | |
7428 | { | |
7429 | return 0; | |
7430 | } | |
7431 | #endif | |
7432 | ||
aeca241b | 7433 | static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files) |
65e19f54 | 7434 | { |
846a4ef2 PB |
7435 | unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE); |
7436 | ||
aeca241b PB |
7437 | table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL); |
7438 | if (!table->files) | |
846a4ef2 | 7439 | return false; |
65e19f54 JA |
7440 | |
7441 | for (i = 0; i < nr_tables; i++) { | |
846a4ef2 | 7442 | unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE); |
65e19f54 | 7443 | |
aeca241b | 7444 | table->files[i] = kcalloc(this_files, sizeof(*table->files[i]), |
65e19f54 | 7445 | GFP_KERNEL); |
aeca241b | 7446 | if (!table->files[i]) |
65e19f54 JA |
7447 | break; |
7448 | nr_files -= this_files; | |
7449 | } | |
7450 | ||
7451 | if (i == nr_tables) | |
846a4ef2 | 7452 | return true; |
65e19f54 | 7453 | |
aeca241b | 7454 | io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE); |
846a4ef2 | 7455 | return false; |
65e19f54 JA |
7456 | } |
7457 | ||
47e90392 | 7458 | static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc) |
05f3fb3c | 7459 | { |
50238531 | 7460 | struct file *file = prsrc->file; |
05f3fb3c JA |
7461 | #if defined(CONFIG_UNIX) |
7462 | struct sock *sock = ctx->ring_sock->sk; | |
7463 | struct sk_buff_head list, *head = &sock->sk_receive_queue; | |
7464 | struct sk_buff *skb; | |
7465 | int i; | |
7466 | ||
7467 | __skb_queue_head_init(&list); | |
7468 | ||
7469 | /* | |
7470 | * Find the skb that holds this file in its SCM_RIGHTS. When found, | |
7471 | * remove this entry and rearrange the file array. | |
7472 | */ | |
7473 | skb = skb_dequeue(head); | |
7474 | while (skb) { | |
7475 | struct scm_fp_list *fp; | |
7476 | ||
7477 | fp = UNIXCB(skb).fp; | |
7478 | for (i = 0; i < fp->count; i++) { | |
7479 | int left; | |
7480 | ||
7481 | if (fp->fp[i] != file) | |
7482 | continue; | |
7483 | ||
7484 | unix_notinflight(fp->user, fp->fp[i]); | |
7485 | left = fp->count - 1 - i; | |
7486 | if (left) { | |
7487 | memmove(&fp->fp[i], &fp->fp[i + 1], | |
7488 | left * sizeof(struct file *)); | |
7489 | } | |
7490 | fp->count--; | |
7491 | if (!fp->count) { | |
7492 | kfree_skb(skb); | |
7493 | skb = NULL; | |
7494 | } else { | |
7495 | __skb_queue_tail(&list, skb); | |
7496 | } | |
7497 | fput(file); | |
7498 | file = NULL; | |
7499 | break; | |
7500 | } | |
7501 | ||
7502 | if (!file) | |
7503 | break; | |
7504 | ||
7505 | __skb_queue_tail(&list, skb); | |
7506 | ||
7507 | skb = skb_dequeue(head); | |
7508 | } | |
7509 | ||
7510 | if (skb_peek(&list)) { | |
7511 | spin_lock_irq(&head->lock); | |
7512 | while ((skb = __skb_dequeue(&list)) != NULL) | |
7513 | __skb_queue_tail(head, skb); | |
7514 | spin_unlock_irq(&head->lock); | |
7515 | } | |
7516 | #else | |
7517 | fput(file); | |
7518 | #endif | |
7519 | } | |
7520 | ||
b895c9a6 | 7521 | static void __io_rsrc_put_work(struct io_rsrc_node *ref_node) |
65e19f54 | 7522 | { |
b895c9a6 | 7523 | struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; |
269bbe5f BM |
7524 | struct io_ring_ctx *ctx = rsrc_data->ctx; |
7525 | struct io_rsrc_put *prsrc, *tmp; | |
05589553 | 7526 | |
269bbe5f BM |
7527 | list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { |
7528 | list_del(&prsrc->list); | |
40ae0ff7 | 7529 | rsrc_data->do_put(ctx, prsrc); |
269bbe5f | 7530 | kfree(prsrc); |
65e19f54 | 7531 | } |
05589553 | 7532 | |
28a9fe25 | 7533 | io_rsrc_node_destroy(ref_node); |
3e942498 PB |
7534 | if (atomic_dec_and_test(&rsrc_data->refs)) |
7535 | complete(&rsrc_data->done); | |
2faf852d | 7536 | } |
65e19f54 | 7537 | |
269bbe5f | 7538 | static void io_rsrc_put_work(struct work_struct *work) |
4a38aed2 JA |
7539 | { |
7540 | struct io_ring_ctx *ctx; | |
7541 | struct llist_node *node; | |
7542 | ||
269bbe5f BM |
7543 | ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work); |
7544 | node = llist_del_all(&ctx->rsrc_put_llist); | |
4a38aed2 JA |
7545 | |
7546 | while (node) { | |
b895c9a6 | 7547 | struct io_rsrc_node *ref_node; |
4a38aed2 JA |
7548 | struct llist_node *next = node->next; |
7549 | ||
b895c9a6 | 7550 | ref_node = llist_entry(node, struct io_rsrc_node, llist); |
269bbe5f | 7551 | __io_rsrc_put_work(ref_node); |
4a38aed2 JA |
7552 | node = next; |
7553 | } | |
7554 | } | |
7555 | ||
00835dce | 7556 | static void io_rsrc_node_ref_zero(struct percpu_ref *ref) |
2faf852d | 7557 | { |
b895c9a6 | 7558 | struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs); |
3e942498 | 7559 | struct io_ring_ctx *ctx = node->rsrc_data->ctx; |
e297822b | 7560 | bool first_add = false; |
65e19f54 | 7561 | |
2a63b2d9 | 7562 | io_rsrc_ref_lock(ctx); |
b895c9a6 | 7563 | node->done = true; |
e297822b | 7564 | |
d67d2263 | 7565 | while (!list_empty(&ctx->rsrc_ref_list)) { |
b895c9a6 PB |
7566 | node = list_first_entry(&ctx->rsrc_ref_list, |
7567 | struct io_rsrc_node, node); | |
e297822b | 7568 | /* recycle ref nodes in order */ |
b895c9a6 | 7569 | if (!node->done) |
e297822b | 7570 | break; |
b895c9a6 PB |
7571 | list_del(&node->node); |
7572 | first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); | |
e297822b | 7573 | } |
2a63b2d9 | 7574 | io_rsrc_ref_unlock(ctx); |
05589553 | 7575 | |
3e942498 PB |
7576 | if (first_add) |
7577 | mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ); | |
05f3fb3c | 7578 | } |
65e19f54 | 7579 | |
b895c9a6 | 7580 | static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx) |
05f3fb3c | 7581 | { |
b895c9a6 | 7582 | struct io_rsrc_node *ref_node; |
05f3fb3c | 7583 | |
05589553 XW |
7584 | ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL); |
7585 | if (!ref_node) | |
3e2224c5 | 7586 | return NULL; |
05f3fb3c | 7587 | |
00835dce | 7588 | if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, |
05589553 XW |
7589 | 0, GFP_KERNEL)) { |
7590 | kfree(ref_node); | |
3e2224c5 | 7591 | return NULL; |
05589553 XW |
7592 | } |
7593 | INIT_LIST_HEAD(&ref_node->node); | |
269bbe5f | 7594 | INIT_LIST_HEAD(&ref_node->rsrc_list); |
e297822b | 7595 | ref_node->done = false; |
05589553 | 7596 | return ref_node; |
05589553 XW |
7597 | } |
7598 | ||
6b06314c JA |
7599 | static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, |
7600 | unsigned nr_args) | |
7601 | { | |
7602 | __s32 __user *fds = (__s32 __user *) arg; | |
05f3fb3c | 7603 | struct file *file; |
f3baed39 | 7604 | int fd, ret; |
846a4ef2 | 7605 | unsigned i; |
b895c9a6 | 7606 | struct io_rsrc_data *file_data; |
6b06314c | 7607 | |
05f3fb3c | 7608 | if (ctx->file_data) |
6b06314c JA |
7609 | return -EBUSY; |
7610 | if (!nr_args) | |
7611 | return -EINVAL; | |
7612 | if (nr_args > IORING_MAX_FIXED_FILES) | |
7613 | return -EMFILE; | |
a7f0ed5a | 7614 | ret = io_rsrc_node_switch_start(ctx); |
f3baed39 PB |
7615 | if (ret) |
7616 | return ret; | |
6b06314c | 7617 | |
47e90392 | 7618 | file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put); |
5398ae69 | 7619 | if (!file_data) |
05f3fb3c | 7620 | return -ENOMEM; |
13770a71 | 7621 | ctx->file_data = file_data; |
f3baed39 | 7622 | ret = -ENOMEM; |
aeca241b | 7623 | if (!io_alloc_file_tables(&ctx->file_table, nr_args)) |
1ad555c6 | 7624 | goto out_free; |
65e19f54 | 7625 | |
08a45173 | 7626 | for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { |
600cf3f8 PB |
7627 | if (copy_from_user(&fd, &fds[i], sizeof(fd))) { |
7628 | ret = -EFAULT; | |
7629 | goto out_fput; | |
7630 | } | |
08a45173 | 7631 | /* allow sparse sets */ |
600cf3f8 | 7632 | if (fd == -1) |
08a45173 | 7633 | continue; |
6b06314c | 7634 | |
05f3fb3c | 7635 | file = fget(fd); |
6b06314c | 7636 | ret = -EBADF; |
05f3fb3c | 7637 | if (!file) |
600cf3f8 | 7638 | goto out_fput; |
05f3fb3c | 7639 | |
6b06314c JA |
7640 | /* |
7641 | * Don't allow io_uring instances to be registered. If UNIX | |
7642 | * isn't enabled, then this causes a reference cycle and this | |
7643 | * instance can never get freed. If UNIX is enabled we'll | |
7644 | * handle it just fine, but there's still no point in allowing | |
7645 | * a ring fd as it doesn't support regular read/write anyway. | |
7646 | */ | |
05f3fb3c JA |
7647 | if (file->f_op == &io_uring_fops) { |
7648 | fput(file); | |
600cf3f8 | 7649 | goto out_fput; |
6b06314c | 7650 | } |
aeca241b | 7651 | io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); |
6b06314c JA |
7652 | } |
7653 | ||
6b06314c | 7654 | ret = io_sqe_files_scm(ctx); |
05589553 | 7655 | if (ret) { |
08480400 | 7656 | __io_sqe_files_unregister(ctx); |
05589553 XW |
7657 | return ret; |
7658 | } | |
6b06314c | 7659 | |
a7f0ed5a | 7660 | io_rsrc_node_switch(ctx, NULL); |
6b06314c | 7661 | return ret; |
600cf3f8 PB |
7662 | out_fput: |
7663 | for (i = 0; i < ctx->nr_user_files; i++) { | |
7664 | file = io_file_from_index(ctx, i); | |
7665 | if (file) | |
7666 | fput(file); | |
7667 | } | |
aeca241b | 7668 | io_free_file_tables(&ctx->file_table, nr_args); |
600cf3f8 | 7669 | ctx->nr_user_files = 0; |
600cf3f8 | 7670 | out_free: |
3e942498 | 7671 | kfree(ctx->file_data); |
55cbc256 | 7672 | ctx->file_data = NULL; |
6b06314c JA |
7673 | return ret; |
7674 | } | |
7675 | ||
c3a31e60 JA |
7676 | static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file, |
7677 | int index) | |
7678 | { | |
7679 | #if defined(CONFIG_UNIX) | |
7680 | struct sock *sock = ctx->ring_sock->sk; | |
7681 | struct sk_buff_head *head = &sock->sk_receive_queue; | |
7682 | struct sk_buff *skb; | |
7683 | ||
7684 | /* | |
7685 | * See if we can merge this file into an existing skb SCM_RIGHTS | |
7686 | * file set. If there's no room, fall back to allocating a new skb | |
7687 | * and filling it in. | |
7688 | */ | |
7689 | spin_lock_irq(&head->lock); | |
7690 | skb = skb_peek(head); | |
7691 | if (skb) { | |
7692 | struct scm_fp_list *fpl = UNIXCB(skb).fp; | |
7693 | ||
7694 | if (fpl->count < SCM_MAX_FD) { | |
7695 | __skb_unlink(skb, head); | |
7696 | spin_unlock_irq(&head->lock); | |
7697 | fpl->fp[fpl->count] = get_file(file); | |
7698 | unix_inflight(fpl->user, fpl->fp[fpl->count]); | |
7699 | fpl->count++; | |
7700 | spin_lock_irq(&head->lock); | |
7701 | __skb_queue_head(head, skb); | |
7702 | } else { | |
7703 | skb = NULL; | |
7704 | } | |
7705 | } | |
7706 | spin_unlock_irq(&head->lock); | |
7707 | ||
7708 | if (skb) { | |
7709 | fput(file); | |
7710 | return 0; | |
7711 | } | |
7712 | ||
7713 | return __io_sqe_files_scm(ctx, 1, index); | |
7714 | #else | |
7715 | return 0; | |
7716 | #endif | |
7717 | } | |
7718 | ||
e7c78371 PB |
7719 | static int io_queue_rsrc_removal(struct io_rsrc_data *data, |
7720 | struct io_rsrc_node *node, void *rsrc) | |
05f3fb3c | 7721 | { |
269bbe5f | 7722 | struct io_rsrc_put *prsrc; |
05f3fb3c | 7723 | |
269bbe5f BM |
7724 | prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL); |
7725 | if (!prsrc) | |
a5318d3c | 7726 | return -ENOMEM; |
05f3fb3c | 7727 | |
50238531 | 7728 | prsrc->rsrc = rsrc; |
e7c78371 | 7729 | list_add(&prsrc->list, &node->rsrc_list); |
a5318d3c | 7730 | return 0; |
05f3fb3c JA |
7731 | } |
7732 | ||
7733 | static int __io_sqe_files_update(struct io_ring_ctx *ctx, | |
269bbe5f | 7734 | struct io_uring_rsrc_update *up, |
05f3fb3c JA |
7735 | unsigned nr_args) |
7736 | { | |
b895c9a6 | 7737 | struct io_rsrc_data *data = ctx->file_data; |
a04b0ac0 PB |
7738 | struct io_fixed_file *file_slot; |
7739 | struct file *file; | |
c3a31e60 JA |
7740 | __s32 __user *fds; |
7741 | int fd, i, err; | |
7742 | __u32 done; | |
05589553 | 7743 | bool needs_switch = false; |
c3a31e60 | 7744 | |
05f3fb3c | 7745 | if (check_add_overflow(up->offset, nr_args, &done)) |
c3a31e60 JA |
7746 | return -EOVERFLOW; |
7747 | if (done > ctx->nr_user_files) | |
7748 | return -EINVAL; | |
a7f0ed5a | 7749 | err = io_rsrc_node_switch_start(ctx); |
8dd03afe PB |
7750 | if (err) |
7751 | return err; | |
05589553 | 7752 | |
269bbe5f | 7753 | fds = u64_to_user_ptr(up->data); |
67973b93 | 7754 | for (done = 0; done < nr_args; done++) { |
c3a31e60 JA |
7755 | err = 0; |
7756 | if (copy_from_user(&fd, &fds[done], sizeof(fd))) { | |
7757 | err = -EFAULT; | |
7758 | break; | |
7759 | } | |
4e0377a1 | 7760 | if (fd == IORING_REGISTER_FILES_SKIP) |
7761 | continue; | |
7762 | ||
67973b93 | 7763 | i = array_index_nospec(up->offset + done, ctx->nr_user_files); |
aeca241b | 7764 | file_slot = io_fixed_file_slot(&ctx->file_table, i); |
ea64ec02 | 7765 | |
a04b0ac0 PB |
7766 | if (file_slot->file_ptr) { |
7767 | file = (struct file *)(file_slot->file_ptr & FFS_MASK); | |
a7f0ed5a | 7768 | err = io_queue_rsrc_removal(data, ctx->rsrc_node, file); |
a5318d3c HD |
7769 | if (err) |
7770 | break; | |
a04b0ac0 | 7771 | file_slot->file_ptr = 0; |
05589553 | 7772 | needs_switch = true; |
c3a31e60 JA |
7773 | } |
7774 | if (fd != -1) { | |
c3a31e60 JA |
7775 | file = fget(fd); |
7776 | if (!file) { | |
7777 | err = -EBADF; | |
7778 | break; | |
7779 | } | |
7780 | /* | |
7781 | * Don't allow io_uring instances to be registered. If | |
7782 | * UNIX isn't enabled, then this causes a reference | |
7783 | * cycle and this instance can never get freed. If UNIX | |
7784 | * is enabled we'll handle it just fine, but there's | |
7785 | * still no point in allowing a ring fd as it doesn't | |
7786 | * support regular read/write anyway. | |
7787 | */ | |
7788 | if (file->f_op == &io_uring_fops) { | |
7789 | fput(file); | |
7790 | err = -EBADF; | |
7791 | break; | |
7792 | } | |
9a321c98 | 7793 | io_fixed_file_set(file_slot, file); |
c3a31e60 | 7794 | err = io_sqe_file_register(ctx, file, i); |
f3bd9dae | 7795 | if (err) { |
a04b0ac0 | 7796 | file_slot->file_ptr = 0; |
f3bd9dae | 7797 | fput(file); |
c3a31e60 | 7798 | break; |
f3bd9dae | 7799 | } |
c3a31e60 | 7800 | } |
05f3fb3c JA |
7801 | } |
7802 | ||
a7f0ed5a PB |
7803 | if (needs_switch) |
7804 | io_rsrc_node_switch(ctx, data); | |
c3a31e60 JA |
7805 | return done ? done : err; |
7806 | } | |
05589553 | 7807 | |
05f3fb3c JA |
7808 | static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg, |
7809 | unsigned nr_args) | |
7810 | { | |
269bbe5f | 7811 | struct io_uring_rsrc_update up; |
05f3fb3c JA |
7812 | |
7813 | if (!ctx->file_data) | |
7814 | return -ENXIO; | |
7815 | if (!nr_args) | |
7816 | return -EINVAL; | |
7817 | if (copy_from_user(&up, arg, sizeof(up))) | |
7818 | return -EFAULT; | |
7819 | if (up.resv) | |
7820 | return -EINVAL; | |
7821 | ||
7822 | return __io_sqe_files_update(ctx, &up, nr_args); | |
7823 | } | |
c3a31e60 | 7824 | |
5280f7e5 | 7825 | static struct io_wq_work *io_free_work(struct io_wq_work *work) |
7d723065 JA |
7826 | { |
7827 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
7828 | ||
5280f7e5 PB |
7829 | req = io_put_req_find_next(req); |
7830 | return req ? &req->work : NULL; | |
7d723065 JA |
7831 | } |
7832 | ||
685fe7fe JA |
7833 | static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx, |
7834 | struct task_struct *task) | |
24369c2e | 7835 | { |
e941894e | 7836 | struct io_wq_hash *hash; |
24369c2e | 7837 | struct io_wq_data data; |
24369c2e | 7838 | unsigned int concurrency; |
24369c2e | 7839 | |
e941894e JA |
7840 | hash = ctx->hash_map; |
7841 | if (!hash) { | |
7842 | hash = kzalloc(sizeof(*hash), GFP_KERNEL); | |
7843 | if (!hash) | |
7844 | return ERR_PTR(-ENOMEM); | |
7845 | refcount_set(&hash->refs, 1); | |
7846 | init_waitqueue_head(&hash->wait); | |
7847 | ctx->hash_map = hash; | |
24369c2e PB |
7848 | } |
7849 | ||
e941894e | 7850 | data.hash = hash; |
685fe7fe | 7851 | data.task = task; |
e9fd9396 | 7852 | data.free_work = io_free_work; |
f5fa38c5 | 7853 | data.do_work = io_wq_submit_work; |
24369c2e | 7854 | |
d25e3a3d JA |
7855 | /* Do QD, or 4 * CPUS, whatever is smallest */ |
7856 | concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); | |
24369c2e | 7857 | |
5aa75ed5 | 7858 | return io_wq_create(concurrency, &data); |
24369c2e PB |
7859 | } |
7860 | ||
5aa75ed5 JA |
7861 | static int io_uring_alloc_task_context(struct task_struct *task, |
7862 | struct io_ring_ctx *ctx) | |
0f212204 JA |
7863 | { |
7864 | struct io_uring_task *tctx; | |
d8a6df10 | 7865 | int ret; |
0f212204 JA |
7866 | |
7867 | tctx = kmalloc(sizeof(*tctx), GFP_KERNEL); | |
7868 | if (unlikely(!tctx)) | |
7869 | return -ENOMEM; | |
7870 | ||
d8a6df10 JA |
7871 | ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); |
7872 | if (unlikely(ret)) { | |
7873 | kfree(tctx); | |
7874 | return ret; | |
7875 | } | |
7876 | ||
685fe7fe | 7877 | tctx->io_wq = io_init_wq_offload(ctx, task); |
5aa75ed5 JA |
7878 | if (IS_ERR(tctx->io_wq)) { |
7879 | ret = PTR_ERR(tctx->io_wq); | |
7880 | percpu_counter_destroy(&tctx->inflight); | |
7881 | kfree(tctx); | |
7882 | return ret; | |
7883 | } | |
7884 | ||
0f212204 JA |
7885 | xa_init(&tctx->xa); |
7886 | init_waitqueue_head(&tctx->wait); | |
7887 | tctx->last = NULL; | |
fdaf083c | 7888 | atomic_set(&tctx->in_idle, 0); |
b303fe2e | 7889 | atomic_set(&tctx->inflight_tracked, 0); |
0f212204 | 7890 | task->io_uring = tctx; |
7cbf1722 JA |
7891 | spin_lock_init(&tctx->task_lock); |
7892 | INIT_WQ_LIST(&tctx->task_list); | |
7893 | tctx->task_state = 0; | |
7894 | init_task_work(&tctx->task_work, tctx_task_work); | |
0f212204 JA |
7895 | return 0; |
7896 | } | |
7897 | ||
7898 | void __io_uring_free(struct task_struct *tsk) | |
7899 | { | |
7900 | struct io_uring_task *tctx = tsk->io_uring; | |
7901 | ||
7902 | WARN_ON_ONCE(!xa_empty(&tctx->xa)); | |
ef8eaa4e PB |
7903 | WARN_ON_ONCE(tctx->io_wq); |
7904 | ||
d8a6df10 | 7905 | percpu_counter_destroy(&tctx->inflight); |
0f212204 JA |
7906 | kfree(tctx); |
7907 | tsk->io_uring = NULL; | |
7908 | } | |
7909 | ||
7e84e1c7 SG |
7910 | static int io_sq_offload_create(struct io_ring_ctx *ctx, |
7911 | struct io_uring_params *p) | |
2b188cc1 JA |
7912 | { |
7913 | int ret; | |
7914 | ||
d25e3a3d JA |
7915 | /* Retain compatibility with failing for an invalid attach attempt */ |
7916 | if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == | |
7917 | IORING_SETUP_ATTACH_WQ) { | |
7918 | struct fd f; | |
7919 | ||
7920 | f = fdget(p->wq_fd); | |
7921 | if (!f.file) | |
7922 | return -ENXIO; | |
7923 | if (f.file->f_op != &io_uring_fops) { | |
7924 | fdput(f); | |
7925 | return -EINVAL; | |
7926 | } | |
7927 | fdput(f); | |
7928 | } | |
6c271ce2 | 7929 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
46fe18b1 | 7930 | struct task_struct *tsk; |
534ca6d6 | 7931 | struct io_sq_data *sqd; |
26984fbf | 7932 | bool attached; |
534ca6d6 | 7933 | |
26984fbf | 7934 | sqd = io_get_sq_data(p, &attached); |
534ca6d6 JA |
7935 | if (IS_ERR(sqd)) { |
7936 | ret = PTR_ERR(sqd); | |
7937 | goto err; | |
7938 | } | |
69fb2131 | 7939 | |
7c30f36a | 7940 | ctx->sq_creds = get_current_cred(); |
534ca6d6 | 7941 | ctx->sq_data = sqd; |
917257da JA |
7942 | ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); |
7943 | if (!ctx->sq_thread_idle) | |
7944 | ctx->sq_thread_idle = HZ; | |
7945 | ||
26984fbf | 7946 | ret = 0; |
78d7f6ba | 7947 | io_sq_thread_park(sqd); |
de75a3d3 PB |
7948 | list_add(&ctx->sqd_list, &sqd->ctx_list); |
7949 | io_sqd_update_thread_idle(sqd); | |
26984fbf | 7950 | /* don't attach to a dying SQPOLL thread, would be racy */ |
de75a3d3 | 7951 | if (attached && !sqd->thread) |
26984fbf | 7952 | ret = -ENXIO; |
78d7f6ba PB |
7953 | io_sq_thread_unpark(sqd); |
7954 | ||
de75a3d3 PB |
7955 | if (ret < 0) |
7956 | goto err; | |
7957 | if (attached) | |
5aa75ed5 | 7958 | return 0; |
aa06165d | 7959 | |
6c271ce2 | 7960 | if (p->flags & IORING_SETUP_SQ_AFF) { |
44a9bd18 | 7961 | int cpu = p->sq_thread_cpu; |
6c271ce2 | 7962 | |
917257da | 7963 | ret = -EINVAL; |
44a9bd18 | 7964 | if (cpu >= nr_cpu_ids) |
e8f98f24 | 7965 | goto err_sqpoll; |
7889f44d | 7966 | if (!cpu_online(cpu)) |
e8f98f24 | 7967 | goto err_sqpoll; |
917257da | 7968 | |
37d1e2e3 | 7969 | sqd->sq_cpu = cpu; |
6c271ce2 | 7970 | } else { |
37d1e2e3 | 7971 | sqd->sq_cpu = -1; |
6c271ce2 | 7972 | } |
37d1e2e3 JA |
7973 | |
7974 | sqd->task_pid = current->pid; | |
5c2469e0 | 7975 | sqd->task_tgid = current->tgid; |
46fe18b1 JA |
7976 | tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE); |
7977 | if (IS_ERR(tsk)) { | |
7978 | ret = PTR_ERR(tsk); | |
e8f98f24 | 7979 | goto err_sqpoll; |
6c271ce2 | 7980 | } |
97a73a0f | 7981 | |
46fe18b1 | 7982 | sqd->thread = tsk; |
97a73a0f | 7983 | ret = io_uring_alloc_task_context(tsk, ctx); |
46fe18b1 | 7984 | wake_up_new_task(tsk); |
0f212204 JA |
7985 | if (ret) |
7986 | goto err; | |
6c271ce2 JA |
7987 | } else if (p->flags & IORING_SETUP_SQ_AFF) { |
7988 | /* Can't have SQ_AFF without SQPOLL */ | |
7989 | ret = -EINVAL; | |
7990 | goto err; | |
7991 | } | |
7992 | ||
2b188cc1 JA |
7993 | return 0; |
7994 | err: | |
37d1e2e3 | 7995 | io_sq_thread_finish(ctx); |
2b188cc1 | 7996 | return ret; |
e8f98f24 JA |
7997 | err_sqpoll: |
7998 | complete(&ctx->sq_data->exited); | |
7999 | goto err; | |
2b188cc1 JA |
8000 | } |
8001 | ||
a087e2b5 BM |
8002 | static inline void __io_unaccount_mem(struct user_struct *user, |
8003 | unsigned long nr_pages) | |
2b188cc1 JA |
8004 | { |
8005 | atomic_long_sub(nr_pages, &user->locked_vm); | |
8006 | } | |
8007 | ||
a087e2b5 BM |
8008 | static inline int __io_account_mem(struct user_struct *user, |
8009 | unsigned long nr_pages) | |
2b188cc1 JA |
8010 | { |
8011 | unsigned long page_limit, cur_pages, new_pages; | |
8012 | ||
8013 | /* Don't allow more pages than we can safely lock */ | |
8014 | page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
8015 | ||
8016 | do { | |
8017 | cur_pages = atomic_long_read(&user->locked_vm); | |
8018 | new_pages = cur_pages + nr_pages; | |
8019 | if (new_pages > page_limit) | |
8020 | return -ENOMEM; | |
8021 | } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, | |
8022 | new_pages) != cur_pages); | |
8023 | ||
8024 | return 0; | |
8025 | } | |
8026 | ||
26bfa89e | 8027 | static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8028 | { |
62e398be | 8029 | if (ctx->user) |
a087e2b5 | 8030 | __io_unaccount_mem(ctx->user, nr_pages); |
30975825 | 8031 | |
26bfa89e JA |
8032 | if (ctx->mm_account) |
8033 | atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8034 | } |
8035 | ||
26bfa89e | 8036 | static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages) |
a087e2b5 | 8037 | { |
30975825 BM |
8038 | int ret; |
8039 | ||
62e398be | 8040 | if (ctx->user) { |
30975825 BM |
8041 | ret = __io_account_mem(ctx->user, nr_pages); |
8042 | if (ret) | |
8043 | return ret; | |
8044 | } | |
8045 | ||
26bfa89e JA |
8046 | if (ctx->mm_account) |
8047 | atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); | |
a087e2b5 BM |
8048 | |
8049 | return 0; | |
8050 | } | |
8051 | ||
2b188cc1 JA |
8052 | static void io_mem_free(void *ptr) |
8053 | { | |
52e04ef4 MR |
8054 | struct page *page; |
8055 | ||
8056 | if (!ptr) | |
8057 | return; | |
2b188cc1 | 8058 | |
52e04ef4 | 8059 | page = virt_to_head_page(ptr); |
2b188cc1 JA |
8060 | if (put_page_testzero(page)) |
8061 | free_compound_page(page); | |
8062 | } | |
8063 | ||
8064 | static void *io_mem_alloc(size_t size) | |
8065 | { | |
8066 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP | | |
26bfa89e | 8067 | __GFP_NORETRY | __GFP_ACCOUNT; |
2b188cc1 JA |
8068 | |
8069 | return (void *) __get_free_pages(gfp_flags, get_order(size)); | |
8070 | } | |
8071 | ||
75b28aff HV |
8072 | static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries, |
8073 | size_t *sq_offset) | |
8074 | { | |
8075 | struct io_rings *rings; | |
8076 | size_t off, sq_array_size; | |
8077 | ||
8078 | off = struct_size(rings, cqes, cq_entries); | |
8079 | if (off == SIZE_MAX) | |
8080 | return SIZE_MAX; | |
8081 | ||
8082 | #ifdef CONFIG_SMP | |
8083 | off = ALIGN(off, SMP_CACHE_BYTES); | |
8084 | if (off == 0) | |
8085 | return SIZE_MAX; | |
8086 | #endif | |
8087 | ||
b36200f5 DV |
8088 | if (sq_offset) |
8089 | *sq_offset = off; | |
8090 | ||
75b28aff HV |
8091 | sq_array_size = array_size(sizeof(u32), sq_entries); |
8092 | if (sq_array_size == SIZE_MAX) | |
8093 | return SIZE_MAX; | |
8094 | ||
8095 | if (check_add_overflow(off, sq_array_size, &off)) | |
8096 | return SIZE_MAX; | |
8097 | ||
75b28aff HV |
8098 | return off; |
8099 | } | |
8100 | ||
7f61a1e9 PB |
8101 | static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf *imu) |
8102 | { | |
8103 | unsigned int i; | |
8104 | ||
8105 | for (i = 0; i < imu->nr_bvecs; i++) | |
8106 | unpin_user_page(imu->bvec[i].bv_page); | |
8107 | if (imu->acct_pages) | |
8108 | io_unaccount_mem(ctx, imu->acct_pages); | |
8109 | kvfree(imu->bvec); | |
8110 | imu->nr_bvecs = 0; | |
8111 | } | |
8112 | ||
0a96bbe4 | 8113 | static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx) |
edafccee | 8114 | { |
7f61a1e9 | 8115 | unsigned int i; |
edafccee JA |
8116 | |
8117 | if (!ctx->user_bufs) | |
8118 | return -ENXIO; | |
8119 | ||
7f61a1e9 PB |
8120 | for (i = 0; i < ctx->nr_user_bufs; i++) |
8121 | io_buffer_unmap(ctx, &ctx->user_bufs[i]); | |
edafccee JA |
8122 | kfree(ctx->user_bufs); |
8123 | ctx->user_bufs = NULL; | |
8124 | ctx->nr_user_bufs = 0; | |
8125 | return 0; | |
8126 | } | |
8127 | ||
8128 | static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst, | |
8129 | void __user *arg, unsigned index) | |
8130 | { | |
8131 | struct iovec __user *src; | |
8132 | ||
8133 | #ifdef CONFIG_COMPAT | |
8134 | if (ctx->compat) { | |
8135 | struct compat_iovec __user *ciovs; | |
8136 | struct compat_iovec ciov; | |
8137 | ||
8138 | ciovs = (struct compat_iovec __user *) arg; | |
8139 | if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov))) | |
8140 | return -EFAULT; | |
8141 | ||
d55e5f5b | 8142 | dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); |
edafccee JA |
8143 | dst->iov_len = ciov.iov_len; |
8144 | return 0; | |
8145 | } | |
8146 | #endif | |
8147 | src = (struct iovec __user *) arg; | |
8148 | if (copy_from_user(dst, &src[index], sizeof(*dst))) | |
8149 | return -EFAULT; | |
8150 | return 0; | |
8151 | } | |
8152 | ||
de293938 JA |
8153 | /* |
8154 | * Not super efficient, but this is just a registration time. And we do cache | |
8155 | * the last compound head, so generally we'll only do a full search if we don't | |
8156 | * match that one. | |
8157 | * | |
8158 | * We check if the given compound head page has already been accounted, to | |
8159 | * avoid double accounting it. This allows us to account the full size of the | |
8160 | * page, not just the constituent pages of a huge page. | |
8161 | */ | |
8162 | static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages, | |
8163 | int nr_pages, struct page *hpage) | |
8164 | { | |
8165 | int i, j; | |
8166 | ||
8167 | /* check current page array */ | |
8168 | for (i = 0; i < nr_pages; i++) { | |
8169 | if (!PageCompound(pages[i])) | |
8170 | continue; | |
8171 | if (compound_head(pages[i]) == hpage) | |
8172 | return true; | |
8173 | } | |
8174 | ||
8175 | /* check previously registered pages */ | |
8176 | for (i = 0; i < ctx->nr_user_bufs; i++) { | |
8177 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; | |
8178 | ||
8179 | for (j = 0; j < imu->nr_bvecs; j++) { | |
8180 | if (!PageCompound(imu->bvec[j].bv_page)) | |
8181 | continue; | |
8182 | if (compound_head(imu->bvec[j].bv_page) == hpage) | |
8183 | return true; | |
8184 | } | |
8185 | } | |
8186 | ||
8187 | return false; | |
8188 | } | |
8189 | ||
8190 | static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages, | |
8191 | int nr_pages, struct io_mapped_ubuf *imu, | |
8192 | struct page **last_hpage) | |
8193 | { | |
8194 | int i, ret; | |
8195 | ||
8196 | for (i = 0; i < nr_pages; i++) { | |
8197 | if (!PageCompound(pages[i])) { | |
8198 | imu->acct_pages++; | |
8199 | } else { | |
8200 | struct page *hpage; | |
8201 | ||
8202 | hpage = compound_head(pages[i]); | |
8203 | if (hpage == *last_hpage) | |
8204 | continue; | |
8205 | *last_hpage = hpage; | |
8206 | if (headpage_already_acct(ctx, pages, i, hpage)) | |
8207 | continue; | |
8208 | imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; | |
8209 | } | |
8210 | } | |
8211 | ||
8212 | if (!imu->acct_pages) | |
8213 | return 0; | |
8214 | ||
26bfa89e | 8215 | ret = io_account_mem(ctx, imu->acct_pages); |
de293938 JA |
8216 | if (ret) |
8217 | imu->acct_pages = 0; | |
8218 | return ret; | |
8219 | } | |
8220 | ||
0a96bbe4 BM |
8221 | static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov, |
8222 | struct io_mapped_ubuf *imu, | |
8223 | struct page **last_hpage) | |
edafccee JA |
8224 | { |
8225 | struct vm_area_struct **vmas = NULL; | |
8226 | struct page **pages = NULL; | |
0a96bbe4 BM |
8227 | unsigned long off, start, end, ubuf; |
8228 | size_t size; | |
8229 | int ret, pret, nr_pages, i; | |
8230 | ||
8231 | ubuf = (unsigned long) iov->iov_base; | |
8232 | end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
8233 | start = ubuf >> PAGE_SHIFT; | |
8234 | nr_pages = end - start; | |
8235 | ||
8236 | ret = -ENOMEM; | |
8237 | ||
8238 | pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
8239 | if (!pages) | |
8240 | goto done; | |
8241 | ||
8242 | vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *), | |
8243 | GFP_KERNEL); | |
8244 | if (!vmas) | |
8245 | goto done; | |
edafccee | 8246 | |
0a96bbe4 BM |
8247 | imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec), |
8248 | GFP_KERNEL); | |
8249 | if (!imu->bvec) | |
8250 | goto done; | |
8251 | ||
8252 | ret = 0; | |
8253 | mmap_read_lock(current->mm); | |
8254 | pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM, | |
8255 | pages, vmas); | |
8256 | if (pret == nr_pages) { | |
8257 | /* don't support file backed memory */ | |
8258 | for (i = 0; i < nr_pages; i++) { | |
8259 | struct vm_area_struct *vma = vmas[i]; | |
8260 | ||
8261 | if (vma->vm_file && | |
8262 | !is_file_hugepages(vma->vm_file)) { | |
8263 | ret = -EOPNOTSUPP; | |
8264 | break; | |
8265 | } | |
8266 | } | |
8267 | } else { | |
8268 | ret = pret < 0 ? pret : -EFAULT; | |
8269 | } | |
8270 | mmap_read_unlock(current->mm); | |
8271 | if (ret) { | |
8272 | /* | |
8273 | * if we did partial map, or found file backed vmas, | |
8274 | * release any pages we did get | |
8275 | */ | |
8276 | if (pret > 0) | |
8277 | unpin_user_pages(pages, pret); | |
8278 | kvfree(imu->bvec); | |
8279 | goto done; | |
8280 | } | |
8281 | ||
8282 | ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage); | |
8283 | if (ret) { | |
8284 | unpin_user_pages(pages, pret); | |
8285 | kvfree(imu->bvec); | |
8286 | goto done; | |
8287 | } | |
8288 | ||
8289 | off = ubuf & ~PAGE_MASK; | |
8290 | size = iov->iov_len; | |
8291 | for (i = 0; i < nr_pages; i++) { | |
8292 | size_t vec_len; | |
8293 | ||
8294 | vec_len = min_t(size_t, size, PAGE_SIZE - off); | |
8295 | imu->bvec[i].bv_page = pages[i]; | |
8296 | imu->bvec[i].bv_len = vec_len; | |
8297 | imu->bvec[i].bv_offset = off; | |
8298 | off = 0; | |
8299 | size -= vec_len; | |
8300 | } | |
8301 | /* store original address for later verification */ | |
8302 | imu->ubuf = ubuf; | |
4751f53d | 8303 | imu->ubuf_end = ubuf + iov->iov_len; |
0a96bbe4 BM |
8304 | imu->nr_bvecs = nr_pages; |
8305 | ret = 0; | |
8306 | done: | |
8307 | kvfree(pages); | |
8308 | kvfree(vmas); | |
8309 | return ret; | |
8310 | } | |
8311 | ||
2b358604 | 8312 | static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args) |
0a96bbe4 | 8313 | { |
87094465 PB |
8314 | ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); |
8315 | return ctx->user_bufs ? 0 : -ENOMEM; | |
2b358604 | 8316 | } |
edafccee | 8317 | |
2b358604 BM |
8318 | static int io_buffer_validate(struct iovec *iov) |
8319 | { | |
50e96989 PB |
8320 | unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); |
8321 | ||
2b358604 BM |
8322 | /* |
8323 | * Don't impose further limits on the size and buffer | |
8324 | * constraints here, we'll -EINVAL later when IO is | |
8325 | * submitted if they are wrong. | |
8326 | */ | |
8327 | if (!iov->iov_base || !iov->iov_len) | |
8328 | return -EFAULT; | |
edafccee | 8329 | |
2b358604 BM |
8330 | /* arbitrary limit, but we need something */ |
8331 | if (iov->iov_len > SZ_1G) | |
8332 | return -EFAULT; | |
edafccee | 8333 | |
50e96989 PB |
8334 | if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) |
8335 | return -EOVERFLOW; | |
8336 | ||
2b358604 BM |
8337 | return 0; |
8338 | } | |
edafccee | 8339 | |
2b358604 BM |
8340 | static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg, |
8341 | unsigned int nr_args) | |
8342 | { | |
8343 | int i, ret; | |
8344 | struct iovec iov; | |
8345 | struct page *last_hpage = NULL; | |
edafccee | 8346 | |
87094465 PB |
8347 | if (ctx->user_bufs) |
8348 | return -EBUSY; | |
8349 | if (!nr_args || nr_args > UIO_MAXIOV) | |
8350 | return -EINVAL; | |
2b358604 BM |
8351 | ret = io_buffers_map_alloc(ctx, nr_args); |
8352 | if (ret) | |
8353 | return ret; | |
edafccee | 8354 | |
87094465 | 8355 | for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { |
edafccee | 8356 | struct io_mapped_ubuf *imu = &ctx->user_bufs[i]; |
edafccee | 8357 | |
edafccee JA |
8358 | ret = io_copy_iov(ctx, &iov, arg, i); |
8359 | if (ret) | |
0a96bbe4 | 8360 | break; |
2b358604 BM |
8361 | ret = io_buffer_validate(&iov); |
8362 | if (ret) | |
0a96bbe4 | 8363 | break; |
0a96bbe4 BM |
8364 | ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage); |
8365 | if (ret) | |
8366 | break; | |
edafccee | 8367 | } |
0a96bbe4 BM |
8368 | |
8369 | if (ret) | |
8370 | io_sqe_buffers_unregister(ctx); | |
8371 | ||
edafccee JA |
8372 | return ret; |
8373 | } | |
8374 | ||
9b402849 JA |
8375 | static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg) |
8376 | { | |
8377 | __s32 __user *fds = arg; | |
8378 | int fd; | |
8379 | ||
8380 | if (ctx->cq_ev_fd) | |
8381 | return -EBUSY; | |
8382 | ||
8383 | if (copy_from_user(&fd, fds, sizeof(*fds))) | |
8384 | return -EFAULT; | |
8385 | ||
8386 | ctx->cq_ev_fd = eventfd_ctx_fdget(fd); | |
8387 | if (IS_ERR(ctx->cq_ev_fd)) { | |
8388 | int ret = PTR_ERR(ctx->cq_ev_fd); | |
8389 | ctx->cq_ev_fd = NULL; | |
8390 | return ret; | |
8391 | } | |
8392 | ||
8393 | return 0; | |
8394 | } | |
8395 | ||
8396 | static int io_eventfd_unregister(struct io_ring_ctx *ctx) | |
8397 | { | |
8398 | if (ctx->cq_ev_fd) { | |
8399 | eventfd_ctx_put(ctx->cq_ev_fd); | |
8400 | ctx->cq_ev_fd = NULL; | |
8401 | return 0; | |
8402 | } | |
8403 | ||
8404 | return -ENXIO; | |
8405 | } | |
8406 | ||
5a2e745d JA |
8407 | static void io_destroy_buffers(struct io_ring_ctx *ctx) |
8408 | { | |
9e15c3a0 JA |
8409 | struct io_buffer *buf; |
8410 | unsigned long index; | |
8411 | ||
8412 | xa_for_each(&ctx->io_buffers, index, buf) | |
8413 | __io_remove_buffers(ctx, buf, index, -1U); | |
5a2e745d JA |
8414 | } |
8415 | ||
68e68ee6 | 8416 | static void io_req_cache_free(struct list_head *list, struct task_struct *tsk) |
1b4c351f | 8417 | { |
68e68ee6 | 8418 | struct io_kiocb *req, *nxt; |
1b4c351f | 8419 | |
68e68ee6 JA |
8420 | list_for_each_entry_safe(req, nxt, list, compl.list) { |
8421 | if (tsk && req->task != tsk) | |
8422 | continue; | |
1b4c351f JA |
8423 | list_del(&req->compl.list); |
8424 | kmem_cache_free(req_cachep, req); | |
8425 | } | |
8426 | } | |
8427 | ||
4010fec4 | 8428 | static void io_req_caches_free(struct io_ring_ctx *ctx) |
2b188cc1 | 8429 | { |
bf019da7 | 8430 | struct io_submit_state *submit_state = &ctx->submit_state; |
e5547d2c | 8431 | struct io_comp_state *cs = &ctx->submit_state.comp; |
bf019da7 | 8432 | |
9a4fdbd8 JA |
8433 | mutex_lock(&ctx->uring_lock); |
8434 | ||
8e5c66c4 | 8435 | if (submit_state->free_reqs) { |
9a4fdbd8 JA |
8436 | kmem_cache_free_bulk(req_cachep, submit_state->free_reqs, |
8437 | submit_state->reqs); | |
8e5c66c4 PB |
8438 | submit_state->free_reqs = 0; |
8439 | } | |
9a4fdbd8 | 8440 | |
dac7a098 | 8441 | io_flush_cached_locked_reqs(ctx, cs); |
e5547d2c | 8442 | io_req_cache_free(&cs->free_list, NULL); |
9a4fdbd8 JA |
8443 | mutex_unlock(&ctx->uring_lock); |
8444 | } | |
8445 | ||
2b188cc1 JA |
8446 | static void io_ring_ctx_free(struct io_ring_ctx *ctx) |
8447 | { | |
37d1e2e3 | 8448 | io_sq_thread_finish(ctx); |
0a96bbe4 | 8449 | io_sqe_buffers_unregister(ctx); |
2aede0e4 | 8450 | |
37d1e2e3 | 8451 | if (ctx->mm_account) { |
2aede0e4 JA |
8452 | mmdrop(ctx->mm_account); |
8453 | ctx->mm_account = NULL; | |
30975825 | 8454 | } |
def596e9 | 8455 | |
8bad28d8 | 8456 | mutex_lock(&ctx->uring_lock); |
08480400 PB |
8457 | if (ctx->file_data) { |
8458 | if (!atomic_dec_and_test(&ctx->file_data->refs)) | |
8459 | wait_for_completion(&ctx->file_data->done); | |
8460 | __io_sqe_files_unregister(ctx); | |
8461 | } | |
c4ea060e PB |
8462 | if (ctx->rings) |
8463 | __io_cqring_overflow_flush(ctx, true); | |
8bad28d8 | 8464 | mutex_unlock(&ctx->uring_lock); |
9b402849 | 8465 | io_eventfd_unregister(ctx); |
5a2e745d | 8466 | io_destroy_buffers(ctx); |
def596e9 | 8467 | |
a7f0ed5a PB |
8468 | /* there are no registered resources left, nobody uses it */ |
8469 | if (ctx->rsrc_node) | |
8470 | io_rsrc_node_destroy(ctx->rsrc_node); | |
8dd03afe | 8471 | if (ctx->rsrc_backup_node) |
b895c9a6 | 8472 | io_rsrc_node_destroy(ctx->rsrc_backup_node); |
a7f0ed5a PB |
8473 | flush_delayed_work(&ctx->rsrc_put_work); |
8474 | ||
8475 | WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); | |
8476 | WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); | |
8dd03afe | 8477 | |
2b188cc1 | 8478 | #if defined(CONFIG_UNIX) |
355e8d26 EB |
8479 | if (ctx->ring_sock) { |
8480 | ctx->ring_sock->file = NULL; /* so that iput() is called */ | |
2b188cc1 | 8481 | sock_release(ctx->ring_sock); |
355e8d26 | 8482 | } |
2b188cc1 JA |
8483 | #endif |
8484 | ||
75b28aff | 8485 | io_mem_free(ctx->rings); |
2b188cc1 | 8486 | io_mem_free(ctx->sq_sqes); |
2b188cc1 JA |
8487 | |
8488 | percpu_ref_exit(&ctx->refs); | |
2b188cc1 | 8489 | free_uid(ctx->user); |
4010fec4 | 8490 | io_req_caches_free(ctx); |
e941894e JA |
8491 | if (ctx->hash_map) |
8492 | io_wq_put_hash(ctx->hash_map); | |
78076bb6 | 8493 | kfree(ctx->cancel_hash); |
2b188cc1 JA |
8494 | kfree(ctx); |
8495 | } | |
8496 | ||
8497 | static __poll_t io_uring_poll(struct file *file, poll_table *wait) | |
8498 | { | |
8499 | struct io_ring_ctx *ctx = file->private_data; | |
8500 | __poll_t mask = 0; | |
8501 | ||
8502 | poll_wait(file, &ctx->cq_wait, wait); | |
4f7067c3 SB |
8503 | /* |
8504 | * synchronizes with barrier from wq_has_sleeper call in | |
8505 | * io_commit_cqring | |
8506 | */ | |
2b188cc1 | 8507 | smp_rmb(); |
90554200 | 8508 | if (!io_sqring_full(ctx)) |
2b188cc1 | 8509 | mask |= EPOLLOUT | EPOLLWRNORM; |
ed670c3f HX |
8510 | |
8511 | /* | |
8512 | * Don't flush cqring overflow list here, just do a simple check. | |
8513 | * Otherwise there could possible be ABBA deadlock: | |
8514 | * CPU0 CPU1 | |
8515 | * ---- ---- | |
8516 | * lock(&ctx->uring_lock); | |
8517 | * lock(&ep->mtx); | |
8518 | * lock(&ctx->uring_lock); | |
8519 | * lock(&ep->mtx); | |
8520 | * | |
8521 | * Users may get EPOLLIN meanwhile seeing nothing in cqring, this | |
8522 | * pushs them to do the flush. | |
8523 | */ | |
8524 | if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow)) | |
2b188cc1 JA |
8525 | mask |= EPOLLIN | EPOLLRDNORM; |
8526 | ||
8527 | return mask; | |
8528 | } | |
8529 | ||
8530 | static int io_uring_fasync(int fd, struct file *file, int on) | |
8531 | { | |
8532 | struct io_ring_ctx *ctx = file->private_data; | |
8533 | ||
8534 | return fasync_helper(fd, file, on, &ctx->cq_fasync); | |
8535 | } | |
8536 | ||
0bead8cd | 8537 | static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id) |
071698e1 | 8538 | { |
4379bf8b | 8539 | const struct cred *creds; |
071698e1 | 8540 | |
61cf9370 | 8541 | creds = xa_erase(&ctx->personalities, id); |
4379bf8b JA |
8542 | if (creds) { |
8543 | put_cred(creds); | |
0bead8cd | 8544 | return 0; |
1e6fa521 | 8545 | } |
0bead8cd YD |
8546 | |
8547 | return -EINVAL; | |
8548 | } | |
8549 | ||
9b465711 | 8550 | static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx) |
7c25c0d1 | 8551 | { |
9b465711 | 8552 | return io_run_task_work_head(&ctx->exit_task_work); |
7c25c0d1 JA |
8553 | } |
8554 | ||
d56d938b PB |
8555 | struct io_tctx_exit { |
8556 | struct callback_head task_work; | |
8557 | struct completion completion; | |
baf186c4 | 8558 | struct io_ring_ctx *ctx; |
d56d938b PB |
8559 | }; |
8560 | ||
8561 | static void io_tctx_exit_cb(struct callback_head *cb) | |
8562 | { | |
8563 | struct io_uring_task *tctx = current->io_uring; | |
8564 | struct io_tctx_exit *work; | |
8565 | ||
8566 | work = container_of(cb, struct io_tctx_exit, task_work); | |
8567 | /* | |
8568 | * When @in_idle, we're in cancellation and it's racy to remove the | |
8569 | * node. It'll be removed by the end of cancellation, just ignore it. | |
8570 | */ | |
8571 | if (!atomic_read(&tctx->in_idle)) | |
baf186c4 | 8572 | io_uring_del_task_file((unsigned long)work->ctx); |
d56d938b PB |
8573 | complete(&work->completion); |
8574 | } | |
8575 | ||
85faa7b8 JA |
8576 | static void io_ring_exit_work(struct work_struct *work) |
8577 | { | |
d56d938b | 8578 | struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work); |
b5bb3a24 | 8579 | unsigned long timeout = jiffies + HZ * 60 * 5; |
d56d938b PB |
8580 | struct io_tctx_exit exit; |
8581 | struct io_tctx_node *node; | |
8582 | int ret; | |
85faa7b8 | 8583 | |
a185f1db PB |
8584 | /* prevent SQPOLL from submitting new requests */ |
8585 | if (ctx->sq_data) { | |
8586 | io_sq_thread_park(ctx->sq_data); | |
8587 | list_del_init(&ctx->sqd_list); | |
8588 | io_sqd_update_thread_idle(ctx->sq_data); | |
8589 | io_sq_thread_unpark(ctx->sq_data); | |
8590 | } | |
8591 | ||
56952e91 JA |
8592 | /* |
8593 | * If we're doing polled IO and end up having requests being | |
8594 | * submitted async (out-of-line), then completions can come in while | |
8595 | * we're waiting for refs to drop. We need to reap these manually, | |
8596 | * as nobody else will be looking for them. | |
8597 | */ | |
b2edc0a7 | 8598 | do { |
9936c7c2 | 8599 | io_uring_try_cancel_requests(ctx, NULL, NULL); |
b5bb3a24 PB |
8600 | |
8601 | WARN_ON_ONCE(time_after(jiffies, timeout)); | |
b2edc0a7 | 8602 | } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20)); |
d56d938b | 8603 | |
89b5066e PB |
8604 | /* |
8605 | * Some may use context even when all refs and requests have been put, | |
8606 | * and they are free to do so while still holding uring_lock or | |
8607 | * completion_lock, see __io_req_task_submit(). Apart from other work, | |
8608 | * this lock/unlock section also waits them to finish. | |
8609 | */ | |
d56d938b PB |
8610 | mutex_lock(&ctx->uring_lock); |
8611 | while (!list_empty(&ctx->tctx_list)) { | |
b5bb3a24 PB |
8612 | WARN_ON_ONCE(time_after(jiffies, timeout)); |
8613 | ||
d56d938b PB |
8614 | node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, |
8615 | ctx_node); | |
baf186c4 | 8616 | exit.ctx = ctx; |
d56d938b PB |
8617 | init_completion(&exit.completion); |
8618 | init_task_work(&exit.task_work, io_tctx_exit_cb); | |
8619 | ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); | |
8620 | if (WARN_ON_ONCE(ret)) | |
8621 | continue; | |
8622 | wake_up_process(node->task); | |
8623 | ||
8624 | mutex_unlock(&ctx->uring_lock); | |
8625 | wait_for_completion(&exit.completion); | |
8626 | cond_resched(); | |
8627 | mutex_lock(&ctx->uring_lock); | |
8628 | } | |
8629 | mutex_unlock(&ctx->uring_lock); | |
89b5066e PB |
8630 | spin_lock_irq(&ctx->completion_lock); |
8631 | spin_unlock_irq(&ctx->completion_lock); | |
d56d938b | 8632 | |
85faa7b8 JA |
8633 | io_ring_ctx_free(ctx); |
8634 | } | |
8635 | ||
80c4cbdb PB |
8636 | /* Returns true if we found and killed one or more timeouts */ |
8637 | static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk, | |
8638 | struct files_struct *files) | |
8639 | { | |
8640 | struct io_kiocb *req, *tmp; | |
8641 | int canceled = 0; | |
8642 | ||
8643 | spin_lock_irq(&ctx->completion_lock); | |
8644 | list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { | |
8645 | if (io_match_task(req, tsk, files)) { | |
8646 | io_kill_timeout(req, -ECANCELED); | |
8647 | canceled++; | |
8648 | } | |
8649 | } | |
51520426 PB |
8650 | if (canceled != 0) |
8651 | io_commit_cqring(ctx); | |
80c4cbdb | 8652 | spin_unlock_irq(&ctx->completion_lock); |
80c4cbdb PB |
8653 | if (canceled != 0) |
8654 | io_cqring_ev_posted(ctx); | |
8655 | return canceled != 0; | |
8656 | } | |
8657 | ||
2b188cc1 JA |
8658 | static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx) |
8659 | { | |
61cf9370 MWO |
8660 | unsigned long index; |
8661 | struct creds *creds; | |
8662 | ||
2b188cc1 JA |
8663 | mutex_lock(&ctx->uring_lock); |
8664 | percpu_ref_kill(&ctx->refs); | |
634578f8 | 8665 | if (ctx->rings) |
6c2450ae | 8666 | __io_cqring_overflow_flush(ctx, true); |
61cf9370 MWO |
8667 | xa_for_each(&ctx->personalities, index, creds) |
8668 | io_unregister_personality(ctx, index); | |
2b188cc1 JA |
8669 | mutex_unlock(&ctx->uring_lock); |
8670 | ||
6b81928d PB |
8671 | io_kill_timeouts(ctx, NULL, NULL); |
8672 | io_poll_remove_all(ctx, NULL, NULL); | |
561fb04a | 8673 | |
15dff286 | 8674 | /* if we failed setting up the ctx, we might not have any rings */ |
b2edc0a7 | 8675 | io_iopoll_try_reap_events(ctx); |
309fc03a | 8676 | |
85faa7b8 | 8677 | INIT_WORK(&ctx->exit_work, io_ring_exit_work); |
fc666777 JA |
8678 | /* |
8679 | * Use system_unbound_wq to avoid spawning tons of event kworkers | |
8680 | * if we're exiting a ton of rings at the same time. It just adds | |
8681 | * noise and overhead, there's no discernable change in runtime | |
8682 | * over using system_wq. | |
8683 | */ | |
8684 | queue_work(system_unbound_wq, &ctx->exit_work); | |
2b188cc1 JA |
8685 | } |
8686 | ||
8687 | static int io_uring_release(struct inode *inode, struct file *file) | |
8688 | { | |
8689 | struct io_ring_ctx *ctx = file->private_data; | |
8690 | ||
8691 | file->private_data = NULL; | |
8692 | io_ring_ctx_wait_and_kill(ctx); | |
8693 | return 0; | |
8694 | } | |
8695 | ||
f6edbabb PB |
8696 | struct io_task_cancel { |
8697 | struct task_struct *task; | |
8698 | struct files_struct *files; | |
8699 | }; | |
f254ac04 | 8700 | |
f6edbabb | 8701 | static bool io_cancel_task_cb(struct io_wq_work *work, void *data) |
b711d4ea | 8702 | { |
9a472ef7 | 8703 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); |
f6edbabb | 8704 | struct io_task_cancel *cancel = data; |
9a472ef7 PB |
8705 | bool ret; |
8706 | ||
f6edbabb | 8707 | if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) { |
9a472ef7 PB |
8708 | unsigned long flags; |
8709 | struct io_ring_ctx *ctx = req->ctx; | |
8710 | ||
8711 | /* protect against races with linked timeouts */ | |
8712 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
f6edbabb | 8713 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8714 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
8715 | } else { | |
f6edbabb | 8716 | ret = io_match_task(req, cancel->task, cancel->files); |
9a472ef7 PB |
8717 | } |
8718 | return ret; | |
b711d4ea JA |
8719 | } |
8720 | ||
e1915f76 | 8721 | static bool io_cancel_defer_files(struct io_ring_ctx *ctx, |
ef9865a4 | 8722 | struct task_struct *task, |
b7ddce3c PB |
8723 | struct files_struct *files) |
8724 | { | |
e1915f76 | 8725 | struct io_defer_entry *de; |
b7ddce3c PB |
8726 | LIST_HEAD(list); |
8727 | ||
8728 | spin_lock_irq(&ctx->completion_lock); | |
8729 | list_for_each_entry_reverse(de, &ctx->defer_list, list) { | |
08d23634 | 8730 | if (io_match_task(de->req, task, files)) { |
b7ddce3c PB |
8731 | list_cut_position(&list, &ctx->defer_list, &de->list); |
8732 | break; | |
8733 | } | |
8734 | } | |
8735 | spin_unlock_irq(&ctx->completion_lock); | |
e1915f76 PB |
8736 | if (list_empty(&list)) |
8737 | return false; | |
b7ddce3c PB |
8738 | |
8739 | while (!list_empty(&list)) { | |
8740 | de = list_first_entry(&list, struct io_defer_entry, list); | |
8741 | list_del_init(&de->list); | |
f41db273 | 8742 | io_req_complete_failed(de->req, -ECANCELED); |
b7ddce3c PB |
8743 | kfree(de); |
8744 | } | |
e1915f76 | 8745 | return true; |
b7ddce3c PB |
8746 | } |
8747 | ||
1b00764f PB |
8748 | static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data) |
8749 | { | |
8750 | struct io_kiocb *req = container_of(work, struct io_kiocb, work); | |
8751 | ||
8752 | return req->ctx == data; | |
8753 | } | |
8754 | ||
8755 | static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx) | |
8756 | { | |
8757 | struct io_tctx_node *node; | |
8758 | enum io_wq_cancel cret; | |
8759 | bool ret = false; | |
8760 | ||
8761 | mutex_lock(&ctx->uring_lock); | |
8762 | list_for_each_entry(node, &ctx->tctx_list, ctx_node) { | |
8763 | struct io_uring_task *tctx = node->task->io_uring; | |
8764 | ||
8765 | /* | |
8766 | * io_wq will stay alive while we hold uring_lock, because it's | |
8767 | * killed after ctx nodes, which requires to take the lock. | |
8768 | */ | |
8769 | if (!tctx || !tctx->io_wq) | |
8770 | continue; | |
8771 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); | |
8772 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8773 | } | |
8774 | mutex_unlock(&ctx->uring_lock); | |
8775 | ||
8776 | return ret; | |
8777 | } | |
8778 | ||
9936c7c2 PB |
8779 | static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx, |
8780 | struct task_struct *task, | |
8781 | struct files_struct *files) | |
8782 | { | |
8783 | struct io_task_cancel cancel = { .task = task, .files = files, }; | |
1b00764f | 8784 | struct io_uring_task *tctx = task ? task->io_uring : NULL; |
9936c7c2 PB |
8785 | |
8786 | while (1) { | |
8787 | enum io_wq_cancel cret; | |
8788 | bool ret = false; | |
8789 | ||
1b00764f PB |
8790 | if (!task) { |
8791 | ret |= io_uring_try_cancel_iowq(ctx); | |
8792 | } else if (tctx && tctx->io_wq) { | |
8793 | /* | |
8794 | * Cancels requests of all rings, not only @ctx, but | |
8795 | * it's fine as the task is in exit/exec. | |
8796 | */ | |
5aa75ed5 | 8797 | cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, |
9936c7c2 PB |
8798 | &cancel, true); |
8799 | ret |= (cret != IO_WQ_CANCEL_NOTFOUND); | |
8800 | } | |
8801 | ||
8802 | /* SQPOLL thread does its own polling */ | |
d052d1d6 JA |
8803 | if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) || |
8804 | (ctx->sq_data && ctx->sq_data->thread == current)) { | |
9936c7c2 PB |
8805 | while (!list_empty_careful(&ctx->iopoll_list)) { |
8806 | io_iopoll_try_reap_events(ctx); | |
8807 | ret = true; | |
8808 | } | |
8809 | } | |
8810 | ||
e1915f76 | 8811 | ret |= io_cancel_defer_files(ctx, task, files); |
9936c7c2 PB |
8812 | ret |= io_poll_remove_all(ctx, task, files); |
8813 | ret |= io_kill_timeouts(ctx, task, files); | |
8814 | ret |= io_run_task_work(); | |
ba50a036 | 8815 | ret |= io_run_ctx_fallback(ctx); |
9936c7c2 PB |
8816 | if (!ret) |
8817 | break; | |
8818 | cond_resched(); | |
8819 | } | |
8820 | } | |
8821 | ||
cf27f3b1 | 8822 | static int __io_uring_add_task_file(struct io_ring_ctx *ctx) |
0f212204 | 8823 | { |
236434c3 | 8824 | struct io_uring_task *tctx = current->io_uring; |
13bf43f5 | 8825 | struct io_tctx_node *node; |
a528b04e | 8826 | int ret; |
236434c3 MWO |
8827 | |
8828 | if (unlikely(!tctx)) { | |
5aa75ed5 | 8829 | ret = io_uring_alloc_task_context(current, ctx); |
0f212204 JA |
8830 | if (unlikely(ret)) |
8831 | return ret; | |
236434c3 | 8832 | tctx = current->io_uring; |
0f212204 | 8833 | } |
cf27f3b1 PB |
8834 | if (!xa_load(&tctx->xa, (unsigned long)ctx)) { |
8835 | node = kmalloc(sizeof(*node), GFP_KERNEL); | |
8836 | if (!node) | |
8837 | return -ENOMEM; | |
8838 | node->ctx = ctx; | |
8839 | node->task = current; | |
13bf43f5 | 8840 | |
cf27f3b1 PB |
8841 | ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, |
8842 | node, GFP_KERNEL)); | |
8843 | if (ret) { | |
8844 | kfree(node); | |
8845 | return ret; | |
0f212204 | 8846 | } |
cf27f3b1 PB |
8847 | |
8848 | mutex_lock(&ctx->uring_lock); | |
8849 | list_add(&node->ctx_node, &ctx->tctx_list); | |
8850 | mutex_unlock(&ctx->uring_lock); | |
0f212204 | 8851 | } |
cf27f3b1 | 8852 | tctx->last = ctx; |
0f212204 JA |
8853 | return 0; |
8854 | } | |
8855 | ||
cf27f3b1 PB |
8856 | /* |
8857 | * Note that this task has used io_uring. We use it for cancelation purposes. | |
8858 | */ | |
8859 | static inline int io_uring_add_task_file(struct io_ring_ctx *ctx) | |
8860 | { | |
8861 | struct io_uring_task *tctx = current->io_uring; | |
8862 | ||
8863 | if (likely(tctx && tctx->last == ctx)) | |
8864 | return 0; | |
8865 | return __io_uring_add_task_file(ctx); | |
8866 | } | |
8867 | ||
0f212204 JA |
8868 | /* |
8869 | * Remove this io_uring_file -> task mapping. | |
8870 | */ | |
2941267b | 8871 | static void io_uring_del_task_file(unsigned long index) |
0f212204 JA |
8872 | { |
8873 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8874 | struct io_tctx_node *node; |
2941267b | 8875 | |
eebd2e37 PB |
8876 | if (!tctx) |
8877 | return; | |
13bf43f5 PB |
8878 | node = xa_erase(&tctx->xa, index); |
8879 | if (!node) | |
2941267b | 8880 | return; |
0f212204 | 8881 | |
13bf43f5 PB |
8882 | WARN_ON_ONCE(current != node->task); |
8883 | WARN_ON_ONCE(list_empty(&node->ctx_node)); | |
8884 | ||
8885 | mutex_lock(&node->ctx->uring_lock); | |
8886 | list_del(&node->ctx_node); | |
8887 | mutex_unlock(&node->ctx->uring_lock); | |
8888 | ||
baf186c4 | 8889 | if (tctx->last == node->ctx) |
0f212204 | 8890 | tctx->last = NULL; |
13bf43f5 | 8891 | kfree(node); |
0f212204 JA |
8892 | } |
8893 | ||
8452d4a6 | 8894 | static void io_uring_clean_tctx(struct io_uring_task *tctx) |
de7f1d9e | 8895 | { |
13bf43f5 | 8896 | struct io_tctx_node *node; |
de7f1d9e PB |
8897 | unsigned long index; |
8898 | ||
13bf43f5 | 8899 | xa_for_each(&tctx->xa, index, node) |
2941267b | 8900 | io_uring_del_task_file(index); |
8452d4a6 PB |
8901 | if (tctx->io_wq) { |
8902 | io_wq_put_and_exit(tctx->io_wq); | |
8903 | tctx->io_wq = NULL; | |
8904 | } | |
de7f1d9e PB |
8905 | } |
8906 | ||
3f48cf18 | 8907 | static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked) |
521d6a73 | 8908 | { |
3f48cf18 PB |
8909 | if (tracked) |
8910 | return atomic_read(&tctx->inflight_tracked); | |
521d6a73 PB |
8911 | return percpu_counter_sum(&tctx->inflight); |
8912 | } | |
8913 | ||
8914 | static void io_sqpoll_cancel_cb(struct callback_head *cb) | |
8915 | { | |
8916 | struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work); | |
8917 | struct io_ring_ctx *ctx = work->ctx; | |
8918 | struct io_sq_data *sqd = ctx->sq_data; | |
8919 | ||
8920 | if (sqd->thread) | |
8921 | io_uring_cancel_sqpoll(ctx); | |
8922 | complete(&work->completion); | |
8923 | } | |
8924 | ||
8925 | static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx) | |
8926 | { | |
8927 | struct io_sq_data *sqd = ctx->sq_data; | |
8928 | struct io_tctx_exit work = { .ctx = ctx, }; | |
8929 | struct task_struct *task; | |
8930 | ||
8931 | io_sq_thread_park(sqd); | |
8932 | list_del_init(&ctx->sqd_list); | |
8933 | io_sqd_update_thread_idle(sqd); | |
8934 | task = sqd->thread; | |
8935 | if (task) { | |
8936 | init_completion(&work.completion); | |
8937 | init_task_work(&work.task_work, io_sqpoll_cancel_cb); | |
b7f5a0bf | 8938 | io_task_work_add_head(&sqd->park_task_work, &work.task_work); |
521d6a73 PB |
8939 | wake_up_process(task); |
8940 | } | |
8941 | io_sq_thread_unpark(sqd); | |
8942 | ||
8943 | if (task) | |
8944 | wait_for_completion(&work.completion); | |
8945 | } | |
8946 | ||
368b2080 | 8947 | static void io_uring_try_cancel(struct files_struct *files) |
0f212204 JA |
8948 | { |
8949 | struct io_uring_task *tctx = current->io_uring; | |
13bf43f5 | 8950 | struct io_tctx_node *node; |
ce765372 | 8951 | unsigned long index; |
0f212204 | 8952 | |
521d6a73 PB |
8953 | xa_for_each(&tctx->xa, index, node) { |
8954 | struct io_ring_ctx *ctx = node->ctx; | |
8955 | ||
8956 | if (ctx->sq_data) { | |
8957 | io_sqpoll_cancel_sync(ctx); | |
8958 | continue; | |
8959 | } | |
368b2080 | 8960 | io_uring_try_cancel_requests(ctx, current, files); |
521d6a73 | 8961 | } |
fdaf083c JA |
8962 | } |
8963 | ||
521d6a73 | 8964 | /* should only be called by SQPOLL task */ |
0e9ddb39 PB |
8965 | static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx) |
8966 | { | |
37d1e2e3 | 8967 | struct io_sq_data *sqd = ctx->sq_data; |
521d6a73 | 8968 | struct io_uring_task *tctx = current->io_uring; |
0e9ddb39 PB |
8969 | s64 inflight; |
8970 | DEFINE_WAIT(wait); | |
fdaf083c | 8971 | |
521d6a73 PB |
8972 | WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current); |
8973 | ||
0e9ddb39 PB |
8974 | atomic_inc(&tctx->in_idle); |
8975 | do { | |
8976 | /* read completions before cancelations */ | |
3f48cf18 | 8977 | inflight = tctx_inflight(tctx, false); |
0e9ddb39 PB |
8978 | if (!inflight) |
8979 | break; | |
521d6a73 | 8980 | io_uring_try_cancel_requests(ctx, current, NULL); |
fdaf083c | 8981 | |
0e9ddb39 PB |
8982 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
8983 | /* | |
8984 | * If we've seen completions, retry without waiting. This | |
8985 | * avoids a race where a completion comes in before we did | |
8986 | * prepare_to_wait(). | |
8987 | */ | |
3f48cf18 | 8988 | if (inflight == tctx_inflight(tctx, false)) |
0e9ddb39 PB |
8989 | schedule(); |
8990 | finish_wait(&tctx->wait, &wait); | |
8991 | } while (1); | |
8992 | atomic_dec(&tctx->in_idle); | |
0f212204 JA |
8993 | } |
8994 | ||
0f212204 JA |
8995 | /* |
8996 | * Find any io_uring fd that this task has registered or done IO on, and cancel | |
8997 | * requests. | |
8998 | */ | |
3f48cf18 | 8999 | void __io_uring_cancel(struct files_struct *files) |
0f212204 JA |
9000 | { |
9001 | struct io_uring_task *tctx = current->io_uring; | |
9002 | DEFINE_WAIT(wait); | |
d8a6df10 | 9003 | s64 inflight; |
0f212204 JA |
9004 | |
9005 | /* make sure overflow events are dropped */ | |
fdaf083c | 9006 | atomic_inc(&tctx->in_idle); |
3f48cf18 | 9007 | io_uring_try_cancel(files); |
5a978dcf | 9008 | |
d8a6df10 | 9009 | do { |
0f212204 | 9010 | /* read completions before cancelations */ |
3f48cf18 | 9011 | inflight = tctx_inflight(tctx, !!files); |
d8a6df10 JA |
9012 | if (!inflight) |
9013 | break; | |
3f48cf18 | 9014 | io_uring_try_cancel(files); |
0f212204 JA |
9015 | prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE); |
9016 | ||
9017 | /* | |
a1bb3cd5 PB |
9018 | * If we've seen completions, retry without waiting. This |
9019 | * avoids a race where a completion comes in before we did | |
9020 | * prepare_to_wait(). | |
0f212204 | 9021 | */ |
3f48cf18 | 9022 | if (inflight == tctx_inflight(tctx, !!files)) |
a1bb3cd5 | 9023 | schedule(); |
f57555ed | 9024 | finish_wait(&tctx->wait, &wait); |
d8a6df10 | 9025 | } while (1); |
fdaf083c | 9026 | atomic_dec(&tctx->in_idle); |
de7f1d9e | 9027 | |
8452d4a6 | 9028 | io_uring_clean_tctx(tctx); |
3f48cf18 PB |
9029 | if (!files) { |
9030 | /* for exec all current's requests should be gone, kill tctx */ | |
9031 | __io_uring_free(current); | |
9032 | } | |
44e728b8 PB |
9033 | } |
9034 | ||
6c5c240e RP |
9035 | static void *io_uring_validate_mmap_request(struct file *file, |
9036 | loff_t pgoff, size_t sz) | |
2b188cc1 | 9037 | { |
2b188cc1 | 9038 | struct io_ring_ctx *ctx = file->private_data; |
6c5c240e | 9039 | loff_t offset = pgoff << PAGE_SHIFT; |
2b188cc1 JA |
9040 | struct page *page; |
9041 | void *ptr; | |
9042 | ||
9043 | switch (offset) { | |
9044 | case IORING_OFF_SQ_RING: | |
75b28aff HV |
9045 | case IORING_OFF_CQ_RING: |
9046 | ptr = ctx->rings; | |
2b188cc1 JA |
9047 | break; |
9048 | case IORING_OFF_SQES: | |
9049 | ptr = ctx->sq_sqes; | |
9050 | break; | |
2b188cc1 | 9051 | default: |
6c5c240e | 9052 | return ERR_PTR(-EINVAL); |
2b188cc1 JA |
9053 | } |
9054 | ||
9055 | page = virt_to_head_page(ptr); | |
a50b854e | 9056 | if (sz > page_size(page)) |
6c5c240e RP |
9057 | return ERR_PTR(-EINVAL); |
9058 | ||
9059 | return ptr; | |
9060 | } | |
9061 | ||
9062 | #ifdef CONFIG_MMU | |
9063 | ||
9064 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9065 | { | |
9066 | size_t sz = vma->vm_end - vma->vm_start; | |
9067 | unsigned long pfn; | |
9068 | void *ptr; | |
9069 | ||
9070 | ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); | |
9071 | if (IS_ERR(ptr)) | |
9072 | return PTR_ERR(ptr); | |
2b188cc1 JA |
9073 | |
9074 | pfn = virt_to_phys(ptr) >> PAGE_SHIFT; | |
9075 | return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); | |
9076 | } | |
9077 | ||
6c5c240e RP |
9078 | #else /* !CONFIG_MMU */ |
9079 | ||
9080 | static int io_uring_mmap(struct file *file, struct vm_area_struct *vma) | |
9081 | { | |
9082 | return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; | |
9083 | } | |
9084 | ||
9085 | static unsigned int io_uring_nommu_mmap_capabilities(struct file *file) | |
9086 | { | |
9087 | return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE; | |
9088 | } | |
9089 | ||
9090 | static unsigned long io_uring_nommu_get_unmapped_area(struct file *file, | |
9091 | unsigned long addr, unsigned long len, | |
9092 | unsigned long pgoff, unsigned long flags) | |
9093 | { | |
9094 | void *ptr; | |
9095 | ||
9096 | ptr = io_uring_validate_mmap_request(file, pgoff, len); | |
9097 | if (IS_ERR(ptr)) | |
9098 | return PTR_ERR(ptr); | |
9099 | ||
9100 | return (unsigned long) ptr; | |
9101 | } | |
9102 | ||
9103 | #endif /* !CONFIG_MMU */ | |
9104 | ||
d9d05217 | 9105 | static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx) |
90554200 JA |
9106 | { |
9107 | DEFINE_WAIT(wait); | |
9108 | ||
9109 | do { | |
9110 | if (!io_sqring_full(ctx)) | |
9111 | break; | |
90554200 JA |
9112 | prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); |
9113 | ||
9114 | if (!io_sqring_full(ctx)) | |
9115 | break; | |
90554200 JA |
9116 | schedule(); |
9117 | } while (!signal_pending(current)); | |
9118 | ||
9119 | finish_wait(&ctx->sqo_sq_wait, &wait); | |
5199328a | 9120 | return 0; |
90554200 JA |
9121 | } |
9122 | ||
c73ebb68 HX |
9123 | static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz, |
9124 | struct __kernel_timespec __user **ts, | |
9125 | const sigset_t __user **sig) | |
9126 | { | |
9127 | struct io_uring_getevents_arg arg; | |
9128 | ||
9129 | /* | |
9130 | * If EXT_ARG isn't set, then we have no timespec and the argp pointer | |
9131 | * is just a pointer to the sigset_t. | |
9132 | */ | |
9133 | if (!(flags & IORING_ENTER_EXT_ARG)) { | |
9134 | *sig = (const sigset_t __user *) argp; | |
9135 | *ts = NULL; | |
9136 | return 0; | |
9137 | } | |
9138 | ||
9139 | /* | |
9140 | * EXT_ARG is set - ensure we agree on the size of it and copy in our | |
9141 | * timespec and sigset_t pointers if good. | |
9142 | */ | |
9143 | if (*argsz != sizeof(arg)) | |
9144 | return -EINVAL; | |
9145 | if (copy_from_user(&arg, argp, sizeof(arg))) | |
9146 | return -EFAULT; | |
9147 | *sig = u64_to_user_ptr(arg.sigmask); | |
9148 | *argsz = arg.sigmask_sz; | |
9149 | *ts = u64_to_user_ptr(arg.ts); | |
9150 | return 0; | |
9151 | } | |
9152 | ||
2b188cc1 | 9153 | SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit, |
c73ebb68 HX |
9154 | u32, min_complete, u32, flags, const void __user *, argp, |
9155 | size_t, argsz) | |
2b188cc1 JA |
9156 | { |
9157 | struct io_ring_ctx *ctx; | |
2b188cc1 JA |
9158 | int submitted = 0; |
9159 | struct fd f; | |
33f993da | 9160 | long ret; |
2b188cc1 | 9161 | |
4c6e277c | 9162 | io_run_task_work(); |
b41e9852 | 9163 | |
33f993da PB |
9164 | if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP | |
9165 | IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))) | |
2b188cc1 JA |
9166 | return -EINVAL; |
9167 | ||
9168 | f = fdget(fd); | |
33f993da | 9169 | if (unlikely(!f.file)) |
2b188cc1 JA |
9170 | return -EBADF; |
9171 | ||
9172 | ret = -EOPNOTSUPP; | |
33f993da | 9173 | if (unlikely(f.file->f_op != &io_uring_fops)) |
2b188cc1 JA |
9174 | goto out_fput; |
9175 | ||
9176 | ret = -ENXIO; | |
9177 | ctx = f.file->private_data; | |
33f993da | 9178 | if (unlikely(!percpu_ref_tryget(&ctx->refs))) |
2b188cc1 JA |
9179 | goto out_fput; |
9180 | ||
7e84e1c7 | 9181 | ret = -EBADFD; |
33f993da | 9182 | if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) |
7e84e1c7 SG |
9183 | goto out; |
9184 | ||
6c271ce2 JA |
9185 | /* |
9186 | * For SQ polling, the thread will do all submissions and completions. | |
9187 | * Just return the requested submit count, and wake the thread if | |
9188 | * we were asked to. | |
9189 | */ | |
b2a9eada | 9190 | ret = 0; |
6c271ce2 | 9191 | if (ctx->flags & IORING_SETUP_SQPOLL) { |
6c2450ae | 9192 | io_cqring_overflow_flush(ctx, false); |
89448c47 | 9193 | |
d9d05217 | 9194 | ret = -EOWNERDEAD; |
04147488 SM |
9195 | if (unlikely(ctx->sq_data->thread == NULL)) { |
9196 | goto out; | |
9197 | } | |
6c271ce2 | 9198 | if (flags & IORING_ENTER_SQ_WAKEUP) |
534ca6d6 | 9199 | wake_up(&ctx->sq_data->wait); |
d9d05217 PB |
9200 | if (flags & IORING_ENTER_SQ_WAIT) { |
9201 | ret = io_sqpoll_wait_sq(ctx); | |
9202 | if (ret) | |
9203 | goto out; | |
9204 | } | |
6c271ce2 | 9205 | submitted = to_submit; |
b2a9eada | 9206 | } else if (to_submit) { |
baf186c4 | 9207 | ret = io_uring_add_task_file(ctx); |
0f212204 JA |
9208 | if (unlikely(ret)) |
9209 | goto out; | |
2b188cc1 | 9210 | mutex_lock(&ctx->uring_lock); |
0f212204 | 9211 | submitted = io_submit_sqes(ctx, to_submit); |
2b188cc1 | 9212 | mutex_unlock(&ctx->uring_lock); |
7c504e65 PB |
9213 | |
9214 | if (submitted != to_submit) | |
9215 | goto out; | |
2b188cc1 JA |
9216 | } |
9217 | if (flags & IORING_ENTER_GETEVENTS) { | |
c73ebb68 HX |
9218 | const sigset_t __user *sig; |
9219 | struct __kernel_timespec __user *ts; | |
9220 | ||
9221 | ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig); | |
9222 | if (unlikely(ret)) | |
9223 | goto out; | |
9224 | ||
2b188cc1 JA |
9225 | min_complete = min(min_complete, ctx->cq_entries); |
9226 | ||
32b2244a XW |
9227 | /* |
9228 | * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user | |
9229 | * space applications don't need to do io completion events | |
9230 | * polling again, they can rely on io_sq_thread to do polling | |
9231 | * work, which can reduce cpu usage and uring_lock contention. | |
9232 | */ | |
9233 | if (ctx->flags & IORING_SETUP_IOPOLL && | |
9234 | !(ctx->flags & IORING_SETUP_SQPOLL)) { | |
7668b92a | 9235 | ret = io_iopoll_check(ctx, min_complete); |
def596e9 | 9236 | } else { |
c73ebb68 | 9237 | ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts); |
def596e9 | 9238 | } |
2b188cc1 JA |
9239 | } |
9240 | ||
7c504e65 | 9241 | out: |
6805b32e | 9242 | percpu_ref_put(&ctx->refs); |
2b188cc1 JA |
9243 | out_fput: |
9244 | fdput(f); | |
9245 | return submitted ? submitted : ret; | |
9246 | } | |
9247 | ||
bebdb65e | 9248 | #ifdef CONFIG_PROC_FS |
61cf9370 MWO |
9249 | static int io_uring_show_cred(struct seq_file *m, unsigned int id, |
9250 | const struct cred *cred) | |
87ce955b | 9251 | { |
87ce955b JA |
9252 | struct user_namespace *uns = seq_user_ns(m); |
9253 | struct group_info *gi; | |
9254 | kernel_cap_t cap; | |
9255 | unsigned __capi; | |
9256 | int g; | |
9257 | ||
9258 | seq_printf(m, "%5d\n", id); | |
9259 | seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); | |
9260 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); | |
9261 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); | |
9262 | seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); | |
9263 | seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); | |
9264 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); | |
9265 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); | |
9266 | seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); | |
9267 | seq_puts(m, "\n\tGroups:\t"); | |
9268 | gi = cred->group_info; | |
9269 | for (g = 0; g < gi->ngroups; g++) { | |
9270 | seq_put_decimal_ull(m, g ? " " : "", | |
9271 | from_kgid_munged(uns, gi->gid[g])); | |
9272 | } | |
9273 | seq_puts(m, "\n\tCapEff:\t"); | |
9274 | cap = cred->cap_effective; | |
9275 | CAP_FOR_EACH_U32(__capi) | |
9276 | seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); | |
9277 | seq_putc(m, '\n'); | |
9278 | return 0; | |
9279 | } | |
9280 | ||
9281 | static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m) | |
9282 | { | |
dbbe9c64 | 9283 | struct io_sq_data *sq = NULL; |
fad8e0de | 9284 | bool has_lock; |
87ce955b JA |
9285 | int i; |
9286 | ||
fad8e0de JA |
9287 | /* |
9288 | * Avoid ABBA deadlock between the seq lock and the io_uring mutex, | |
9289 | * since fdinfo case grabs it in the opposite direction of normal use | |
9290 | * cases. If we fail to get the lock, we just don't iterate any | |
9291 | * structures that could be going away outside the io_uring mutex. | |
9292 | */ | |
9293 | has_lock = mutex_trylock(&ctx->uring_lock); | |
9294 | ||
5f3f26f9 | 9295 | if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { |
dbbe9c64 | 9296 | sq = ctx->sq_data; |
5f3f26f9 JA |
9297 | if (!sq->thread) |
9298 | sq = NULL; | |
9299 | } | |
dbbe9c64 JQ |
9300 | |
9301 | seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); | |
9302 | seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); | |
87ce955b | 9303 | seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); |
fad8e0de | 9304 | for (i = 0; has_lock && i < ctx->nr_user_files; i++) { |
7b29f92d | 9305 | struct file *f = io_file_from_index(ctx, i); |
87ce955b | 9306 | |
87ce955b JA |
9307 | if (f) |
9308 | seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); | |
9309 | else | |
9310 | seq_printf(m, "%5u: <none>\n", i); | |
9311 | } | |
9312 | seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); | |
fad8e0de | 9313 | for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { |
87ce955b | 9314 | struct io_mapped_ubuf *buf = &ctx->user_bufs[i]; |
4751f53d | 9315 | unsigned int len = buf->ubuf_end - buf->ubuf; |
87ce955b | 9316 | |
4751f53d | 9317 | seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); |
87ce955b | 9318 | } |
61cf9370 MWO |
9319 | if (has_lock && !xa_empty(&ctx->personalities)) { |
9320 | unsigned long index; | |
9321 | const struct cred *cred; | |
9322 | ||
87ce955b | 9323 | seq_printf(m, "Personalities:\n"); |
61cf9370 MWO |
9324 | xa_for_each(&ctx->personalities, index, cred) |
9325 | io_uring_show_cred(m, index, cred); | |
87ce955b | 9326 | } |
d7718a9d JA |
9327 | seq_printf(m, "PollList:\n"); |
9328 | spin_lock_irq(&ctx->completion_lock); | |
9329 | for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { | |
9330 | struct hlist_head *list = &ctx->cancel_hash[i]; | |
9331 | struct io_kiocb *req; | |
9332 | ||
9333 | hlist_for_each_entry(req, list, hash_node) | |
9334 | seq_printf(m, " op=%d, task_works=%d\n", req->opcode, | |
9335 | req->task->task_works != NULL); | |
9336 | } | |
9337 | spin_unlock_irq(&ctx->completion_lock); | |
fad8e0de JA |
9338 | if (has_lock) |
9339 | mutex_unlock(&ctx->uring_lock); | |
87ce955b JA |
9340 | } |
9341 | ||
9342 | static void io_uring_show_fdinfo(struct seq_file *m, struct file *f) | |
9343 | { | |
9344 | struct io_ring_ctx *ctx = f->private_data; | |
9345 | ||
9346 | if (percpu_ref_tryget(&ctx->refs)) { | |
9347 | __io_uring_show_fdinfo(ctx, m); | |
9348 | percpu_ref_put(&ctx->refs); | |
9349 | } | |
9350 | } | |
bebdb65e | 9351 | #endif |
87ce955b | 9352 | |
2b188cc1 JA |
9353 | static const struct file_operations io_uring_fops = { |
9354 | .release = io_uring_release, | |
9355 | .mmap = io_uring_mmap, | |
6c5c240e RP |
9356 | #ifndef CONFIG_MMU |
9357 | .get_unmapped_area = io_uring_nommu_get_unmapped_area, | |
9358 | .mmap_capabilities = io_uring_nommu_mmap_capabilities, | |
9359 | #endif | |
2b188cc1 JA |
9360 | .poll = io_uring_poll, |
9361 | .fasync = io_uring_fasync, | |
bebdb65e | 9362 | #ifdef CONFIG_PROC_FS |
87ce955b | 9363 | .show_fdinfo = io_uring_show_fdinfo, |
bebdb65e | 9364 | #endif |
2b188cc1 JA |
9365 | }; |
9366 | ||
9367 | static int io_allocate_scq_urings(struct io_ring_ctx *ctx, | |
9368 | struct io_uring_params *p) | |
9369 | { | |
75b28aff HV |
9370 | struct io_rings *rings; |
9371 | size_t size, sq_array_offset; | |
2b188cc1 | 9372 | |
bd740481 JA |
9373 | /* make sure these are sane, as we already accounted them */ |
9374 | ctx->sq_entries = p->sq_entries; | |
9375 | ctx->cq_entries = p->cq_entries; | |
9376 | ||
75b28aff HV |
9377 | size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); |
9378 | if (size == SIZE_MAX) | |
9379 | return -EOVERFLOW; | |
9380 | ||
9381 | rings = io_mem_alloc(size); | |
9382 | if (!rings) | |
2b188cc1 JA |
9383 | return -ENOMEM; |
9384 | ||
75b28aff HV |
9385 | ctx->rings = rings; |
9386 | ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); | |
9387 | rings->sq_ring_mask = p->sq_entries - 1; | |
9388 | rings->cq_ring_mask = p->cq_entries - 1; | |
9389 | rings->sq_ring_entries = p->sq_entries; | |
9390 | rings->cq_ring_entries = p->cq_entries; | |
9391 | ctx->sq_mask = rings->sq_ring_mask; | |
9392 | ctx->cq_mask = rings->cq_ring_mask; | |
2b188cc1 JA |
9393 | |
9394 | size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); | |
eb065d30 JA |
9395 | if (size == SIZE_MAX) { |
9396 | io_mem_free(ctx->rings); | |
9397 | ctx->rings = NULL; | |
2b188cc1 | 9398 | return -EOVERFLOW; |
eb065d30 | 9399 | } |
2b188cc1 JA |
9400 | |
9401 | ctx->sq_sqes = io_mem_alloc(size); | |
eb065d30 JA |
9402 | if (!ctx->sq_sqes) { |
9403 | io_mem_free(ctx->rings); | |
9404 | ctx->rings = NULL; | |
2b188cc1 | 9405 | return -ENOMEM; |
eb065d30 | 9406 | } |
2b188cc1 | 9407 | |
2b188cc1 JA |
9408 | return 0; |
9409 | } | |
9410 | ||
9faadcc8 PB |
9411 | static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file) |
9412 | { | |
9413 | int ret, fd; | |
9414 | ||
9415 | fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC); | |
9416 | if (fd < 0) | |
9417 | return fd; | |
9418 | ||
baf186c4 | 9419 | ret = io_uring_add_task_file(ctx); |
9faadcc8 PB |
9420 | if (ret) { |
9421 | put_unused_fd(fd); | |
9422 | return ret; | |
9423 | } | |
9424 | fd_install(fd, file); | |
9425 | return fd; | |
9426 | } | |
9427 | ||
2b188cc1 JA |
9428 | /* |
9429 | * Allocate an anonymous fd, this is what constitutes the application | |
9430 | * visible backing of an io_uring instance. The application mmaps this | |
9431 | * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled, | |
9432 | * we have to tie this fd to a socket for file garbage collection purposes. | |
9433 | */ | |
9faadcc8 | 9434 | static struct file *io_uring_get_file(struct io_ring_ctx *ctx) |
2b188cc1 JA |
9435 | { |
9436 | struct file *file; | |
9faadcc8 | 9437 | #if defined(CONFIG_UNIX) |
2b188cc1 JA |
9438 | int ret; |
9439 | ||
2b188cc1 JA |
9440 | ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP, |
9441 | &ctx->ring_sock); | |
9442 | if (ret) | |
9faadcc8 | 9443 | return ERR_PTR(ret); |
2b188cc1 JA |
9444 | #endif |
9445 | ||
2b188cc1 JA |
9446 | file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx, |
9447 | O_RDWR | O_CLOEXEC); | |
2b188cc1 | 9448 | #if defined(CONFIG_UNIX) |
9faadcc8 PB |
9449 | if (IS_ERR(file)) { |
9450 | sock_release(ctx->ring_sock); | |
9451 | ctx->ring_sock = NULL; | |
9452 | } else { | |
9453 | ctx->ring_sock->file = file; | |
0f212204 | 9454 | } |
2b188cc1 | 9455 | #endif |
9faadcc8 | 9456 | return file; |
2b188cc1 JA |
9457 | } |
9458 | ||
7f13657d XW |
9459 | static int io_uring_create(unsigned entries, struct io_uring_params *p, |
9460 | struct io_uring_params __user *params) | |
2b188cc1 | 9461 | { |
2b188cc1 | 9462 | struct io_ring_ctx *ctx; |
9faadcc8 | 9463 | struct file *file; |
2b188cc1 JA |
9464 | int ret; |
9465 | ||
8110c1a6 | 9466 | if (!entries) |
2b188cc1 | 9467 | return -EINVAL; |
8110c1a6 JA |
9468 | if (entries > IORING_MAX_ENTRIES) { |
9469 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9470 | return -EINVAL; | |
9471 | entries = IORING_MAX_ENTRIES; | |
9472 | } | |
2b188cc1 JA |
9473 | |
9474 | /* | |
9475 | * Use twice as many entries for the CQ ring. It's possible for the | |
9476 | * application to drive a higher depth than the size of the SQ ring, | |
9477 | * since the sqes are only used at submission time. This allows for | |
33a107f0 JA |
9478 | * some flexibility in overcommitting a bit. If the application has |
9479 | * set IORING_SETUP_CQSIZE, it will have passed in the desired number | |
9480 | * of CQ ring entries manually. | |
2b188cc1 JA |
9481 | */ |
9482 | p->sq_entries = roundup_pow_of_two(entries); | |
33a107f0 JA |
9483 | if (p->flags & IORING_SETUP_CQSIZE) { |
9484 | /* | |
9485 | * If IORING_SETUP_CQSIZE is set, we do the same roundup | |
9486 | * to a power-of-two, if it isn't already. We do NOT impose | |
9487 | * any cq vs sq ring sizing. | |
9488 | */ | |
eb2667b3 | 9489 | if (!p->cq_entries) |
33a107f0 | 9490 | return -EINVAL; |
8110c1a6 JA |
9491 | if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { |
9492 | if (!(p->flags & IORING_SETUP_CLAMP)) | |
9493 | return -EINVAL; | |
9494 | p->cq_entries = IORING_MAX_CQ_ENTRIES; | |
9495 | } | |
eb2667b3 JQ |
9496 | p->cq_entries = roundup_pow_of_two(p->cq_entries); |
9497 | if (p->cq_entries < p->sq_entries) | |
9498 | return -EINVAL; | |
33a107f0 JA |
9499 | } else { |
9500 | p->cq_entries = 2 * p->sq_entries; | |
9501 | } | |
2b188cc1 | 9502 | |
2b188cc1 | 9503 | ctx = io_ring_ctx_alloc(p); |
62e398be | 9504 | if (!ctx) |
2b188cc1 | 9505 | return -ENOMEM; |
2b188cc1 | 9506 | ctx->compat = in_compat_syscall(); |
62e398be JA |
9507 | if (!capable(CAP_IPC_LOCK)) |
9508 | ctx->user = get_uid(current_user()); | |
2aede0e4 JA |
9509 | |
9510 | /* | |
9511 | * This is just grabbed for accounting purposes. When a process exits, | |
9512 | * the mm is exited and dropped before the files, hence we need to hang | |
9513 | * on to this mm purely for the purposes of being able to unaccount | |
9514 | * memory (locked/pinned vm). It's not used for anything else. | |
9515 | */ | |
6b7898eb | 9516 | mmgrab(current->mm); |
2aede0e4 | 9517 | ctx->mm_account = current->mm; |
6b7898eb | 9518 | |
2b188cc1 JA |
9519 | ret = io_allocate_scq_urings(ctx, p); |
9520 | if (ret) | |
9521 | goto err; | |
9522 | ||
7e84e1c7 | 9523 | ret = io_sq_offload_create(ctx, p); |
2b188cc1 JA |
9524 | if (ret) |
9525 | goto err; | |
9526 | ||
2b188cc1 | 9527 | memset(&p->sq_off, 0, sizeof(p->sq_off)); |
75b28aff HV |
9528 | p->sq_off.head = offsetof(struct io_rings, sq.head); |
9529 | p->sq_off.tail = offsetof(struct io_rings, sq.tail); | |
9530 | p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); | |
9531 | p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); | |
9532 | p->sq_off.flags = offsetof(struct io_rings, sq_flags); | |
9533 | p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); | |
9534 | p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; | |
2b188cc1 JA |
9535 | |
9536 | memset(&p->cq_off, 0, sizeof(p->cq_off)); | |
75b28aff HV |
9537 | p->cq_off.head = offsetof(struct io_rings, cq.head); |
9538 | p->cq_off.tail = offsetof(struct io_rings, cq.tail); | |
9539 | p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); | |
9540 | p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); | |
9541 | p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); | |
9542 | p->cq_off.cqes = offsetof(struct io_rings, cqes); | |
0d9b5b3a | 9543 | p->cq_off.flags = offsetof(struct io_rings, cq_flags); |
ac90f249 | 9544 | |
7f13657d XW |
9545 | p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | |
9546 | IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS | | |
5769a351 | 9547 | IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL | |
c73ebb68 | 9548 | IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED | |
1c0aa1fa | 9549 | IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS; |
7f13657d XW |
9550 | |
9551 | if (copy_to_user(params, p, sizeof(*p))) { | |
9552 | ret = -EFAULT; | |
9553 | goto err; | |
9554 | } | |
d1719f70 | 9555 | |
9faadcc8 PB |
9556 | file = io_uring_get_file(ctx); |
9557 | if (IS_ERR(file)) { | |
9558 | ret = PTR_ERR(file); | |
9559 | goto err; | |
9560 | } | |
9561 | ||
044c1ab3 JA |
9562 | /* |
9563 | * Install ring fd as the very last thing, so we don't risk someone | |
9564 | * having closed it before we finish setup | |
9565 | */ | |
9faadcc8 PB |
9566 | ret = io_uring_install_fd(ctx, file); |
9567 | if (ret < 0) { | |
9568 | /* fput will clean it up */ | |
9569 | fput(file); | |
9570 | return ret; | |
9571 | } | |
044c1ab3 | 9572 | |
c826bd7a | 9573 | trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); |
2b188cc1 JA |
9574 | return ret; |
9575 | err: | |
9576 | io_ring_ctx_wait_and_kill(ctx); | |
9577 | return ret; | |
9578 | } | |
9579 | ||
9580 | /* | |
9581 | * Sets up an aio uring context, and returns the fd. Applications asks for a | |
9582 | * ring size, we return the actual sq/cq ring sizes (among other things) in the | |
9583 | * params structure passed in. | |
9584 | */ | |
9585 | static long io_uring_setup(u32 entries, struct io_uring_params __user *params) | |
9586 | { | |
9587 | struct io_uring_params p; | |
2b188cc1 JA |
9588 | int i; |
9589 | ||
9590 | if (copy_from_user(&p, params, sizeof(p))) | |
9591 | return -EFAULT; | |
9592 | for (i = 0; i < ARRAY_SIZE(p.resv); i++) { | |
9593 | if (p.resv[i]) | |
9594 | return -EINVAL; | |
9595 | } | |
9596 | ||
6c271ce2 | 9597 | if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL | |
8110c1a6 | 9598 | IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE | |
7e84e1c7 SG |
9599 | IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ | |
9600 | IORING_SETUP_R_DISABLED)) | |
2b188cc1 JA |
9601 | return -EINVAL; |
9602 | ||
7f13657d | 9603 | return io_uring_create(entries, &p, params); |
2b188cc1 JA |
9604 | } |
9605 | ||
9606 | SYSCALL_DEFINE2(io_uring_setup, u32, entries, | |
9607 | struct io_uring_params __user *, params) | |
9608 | { | |
9609 | return io_uring_setup(entries, params); | |
9610 | } | |
9611 | ||
66f4af93 JA |
9612 | static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args) |
9613 | { | |
9614 | struct io_uring_probe *p; | |
9615 | size_t size; | |
9616 | int i, ret; | |
9617 | ||
9618 | size = struct_size(p, ops, nr_args); | |
9619 | if (size == SIZE_MAX) | |
9620 | return -EOVERFLOW; | |
9621 | p = kzalloc(size, GFP_KERNEL); | |
9622 | if (!p) | |
9623 | return -ENOMEM; | |
9624 | ||
9625 | ret = -EFAULT; | |
9626 | if (copy_from_user(p, arg, size)) | |
9627 | goto out; | |
9628 | ret = -EINVAL; | |
9629 | if (memchr_inv(p, 0, size)) | |
9630 | goto out; | |
9631 | ||
9632 | p->last_op = IORING_OP_LAST - 1; | |
9633 | if (nr_args > IORING_OP_LAST) | |
9634 | nr_args = IORING_OP_LAST; | |
9635 | ||
9636 | for (i = 0; i < nr_args; i++) { | |
9637 | p->ops[i].op = i; | |
9638 | if (!io_op_defs[i].not_supported) | |
9639 | p->ops[i].flags = IO_URING_OP_SUPPORTED; | |
9640 | } | |
9641 | p->ops_len = i; | |
9642 | ||
9643 | ret = 0; | |
9644 | if (copy_to_user(arg, p, size)) | |
9645 | ret = -EFAULT; | |
9646 | out: | |
9647 | kfree(p); | |
9648 | return ret; | |
9649 | } | |
9650 | ||
071698e1 JA |
9651 | static int io_register_personality(struct io_ring_ctx *ctx) |
9652 | { | |
4379bf8b | 9653 | const struct cred *creds; |
61cf9370 | 9654 | u32 id; |
1e6fa521 | 9655 | int ret; |
071698e1 | 9656 | |
4379bf8b | 9657 | creds = get_current_cred(); |
1e6fa521 | 9658 | |
61cf9370 MWO |
9659 | ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, |
9660 | XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); | |
9661 | if (!ret) | |
9662 | return id; | |
9663 | put_cred(creds); | |
1e6fa521 | 9664 | return ret; |
071698e1 JA |
9665 | } |
9666 | ||
21b55dbc SG |
9667 | static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg, |
9668 | unsigned int nr_args) | |
9669 | { | |
9670 | struct io_uring_restriction *res; | |
9671 | size_t size; | |
9672 | int i, ret; | |
9673 | ||
7e84e1c7 SG |
9674 | /* Restrictions allowed only if rings started disabled */ |
9675 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9676 | return -EBADFD; | |
9677 | ||
21b55dbc | 9678 | /* We allow only a single restrictions registration */ |
7e84e1c7 | 9679 | if (ctx->restrictions.registered) |
21b55dbc SG |
9680 | return -EBUSY; |
9681 | ||
9682 | if (!arg || nr_args > IORING_MAX_RESTRICTIONS) | |
9683 | return -EINVAL; | |
9684 | ||
9685 | size = array_size(nr_args, sizeof(*res)); | |
9686 | if (size == SIZE_MAX) | |
9687 | return -EOVERFLOW; | |
9688 | ||
9689 | res = memdup_user(arg, size); | |
9690 | if (IS_ERR(res)) | |
9691 | return PTR_ERR(res); | |
9692 | ||
9693 | ret = 0; | |
9694 | ||
9695 | for (i = 0; i < nr_args; i++) { | |
9696 | switch (res[i].opcode) { | |
9697 | case IORING_RESTRICTION_REGISTER_OP: | |
9698 | if (res[i].register_op >= IORING_REGISTER_LAST) { | |
9699 | ret = -EINVAL; | |
9700 | goto out; | |
9701 | } | |
9702 | ||
9703 | __set_bit(res[i].register_op, | |
9704 | ctx->restrictions.register_op); | |
9705 | break; | |
9706 | case IORING_RESTRICTION_SQE_OP: | |
9707 | if (res[i].sqe_op >= IORING_OP_LAST) { | |
9708 | ret = -EINVAL; | |
9709 | goto out; | |
9710 | } | |
9711 | ||
9712 | __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); | |
9713 | break; | |
9714 | case IORING_RESTRICTION_SQE_FLAGS_ALLOWED: | |
9715 | ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; | |
9716 | break; | |
9717 | case IORING_RESTRICTION_SQE_FLAGS_REQUIRED: | |
9718 | ctx->restrictions.sqe_flags_required = res[i].sqe_flags; | |
9719 | break; | |
9720 | default: | |
9721 | ret = -EINVAL; | |
9722 | goto out; | |
9723 | } | |
9724 | } | |
9725 | ||
9726 | out: | |
9727 | /* Reset all restrictions if an error happened */ | |
9728 | if (ret != 0) | |
9729 | memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); | |
9730 | else | |
7e84e1c7 | 9731 | ctx->restrictions.registered = true; |
21b55dbc SG |
9732 | |
9733 | kfree(res); | |
9734 | return ret; | |
9735 | } | |
9736 | ||
7e84e1c7 SG |
9737 | static int io_register_enable_rings(struct io_ring_ctx *ctx) |
9738 | { | |
9739 | if (!(ctx->flags & IORING_SETUP_R_DISABLED)) | |
9740 | return -EBADFD; | |
9741 | ||
9742 | if (ctx->restrictions.registered) | |
9743 | ctx->restricted = 1; | |
9744 | ||
0298ef96 PB |
9745 | ctx->flags &= ~IORING_SETUP_R_DISABLED; |
9746 | if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) | |
9747 | wake_up(&ctx->sq_data->wait); | |
7e84e1c7 SG |
9748 | return 0; |
9749 | } | |
9750 | ||
071698e1 JA |
9751 | static bool io_register_op_must_quiesce(int op) |
9752 | { | |
9753 | switch (op) { | |
f4f7d21c | 9754 | case IORING_REGISTER_FILES: |
071698e1 JA |
9755 | case IORING_UNREGISTER_FILES: |
9756 | case IORING_REGISTER_FILES_UPDATE: | |
9757 | case IORING_REGISTER_PROBE: | |
9758 | case IORING_REGISTER_PERSONALITY: | |
9759 | case IORING_UNREGISTER_PERSONALITY: | |
9760 | return false; | |
9761 | default: | |
9762 | return true; | |
9763 | } | |
9764 | } | |
9765 | ||
edafccee JA |
9766 | static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, |
9767 | void __user *arg, unsigned nr_args) | |
b19062a5 JA |
9768 | __releases(ctx->uring_lock) |
9769 | __acquires(ctx->uring_lock) | |
edafccee JA |
9770 | { |
9771 | int ret; | |
9772 | ||
35fa71a0 JA |
9773 | /* |
9774 | * We're inside the ring mutex, if the ref is already dying, then | |
9775 | * someone else killed the ctx or is already going through | |
9776 | * io_uring_register(). | |
9777 | */ | |
9778 | if (percpu_ref_is_dying(&ctx->refs)) | |
9779 | return -ENXIO; | |
9780 | ||
071698e1 | 9781 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9782 | percpu_ref_kill(&ctx->refs); |
b19062a5 | 9783 | |
05f3fb3c JA |
9784 | /* |
9785 | * Drop uring mutex before waiting for references to exit. If | |
9786 | * another thread is currently inside io_uring_enter() it might | |
9787 | * need to grab the uring_lock to make progress. If we hold it | |
9788 | * here across the drain wait, then we can deadlock. It's safe | |
9789 | * to drop the mutex here, since no new references will come in | |
9790 | * after we've killed the percpu ref. | |
9791 | */ | |
9792 | mutex_unlock(&ctx->uring_lock); | |
af9c1a44 JA |
9793 | do { |
9794 | ret = wait_for_completion_interruptible(&ctx->ref_comp); | |
9795 | if (!ret) | |
9796 | break; | |
ed6930c9 JA |
9797 | ret = io_run_task_work_sig(); |
9798 | if (ret < 0) | |
9799 | break; | |
af9c1a44 | 9800 | } while (1); |
05f3fb3c | 9801 | mutex_lock(&ctx->uring_lock); |
af9c1a44 | 9802 | |
c150368b | 9803 | if (ret) { |
f70865db PB |
9804 | io_refs_resurrect(&ctx->refs, &ctx->ref_comp); |
9805 | return ret; | |
21b55dbc SG |
9806 | } |
9807 | } | |
9808 | ||
9809 | if (ctx->restricted) { | |
9810 | if (opcode >= IORING_REGISTER_LAST) { | |
9811 | ret = -EINVAL; | |
9812 | goto out; | |
9813 | } | |
9814 | ||
9815 | if (!test_bit(opcode, ctx->restrictions.register_op)) { | |
9816 | ret = -EACCES; | |
c150368b JA |
9817 | goto out; |
9818 | } | |
05f3fb3c | 9819 | } |
edafccee JA |
9820 | |
9821 | switch (opcode) { | |
9822 | case IORING_REGISTER_BUFFERS: | |
0a96bbe4 | 9823 | ret = io_sqe_buffers_register(ctx, arg, nr_args); |
edafccee JA |
9824 | break; |
9825 | case IORING_UNREGISTER_BUFFERS: | |
9826 | ret = -EINVAL; | |
9827 | if (arg || nr_args) | |
9828 | break; | |
0a96bbe4 | 9829 | ret = io_sqe_buffers_unregister(ctx); |
edafccee | 9830 | break; |
6b06314c JA |
9831 | case IORING_REGISTER_FILES: |
9832 | ret = io_sqe_files_register(ctx, arg, nr_args); | |
9833 | break; | |
9834 | case IORING_UNREGISTER_FILES: | |
9835 | ret = -EINVAL; | |
9836 | if (arg || nr_args) | |
9837 | break; | |
9838 | ret = io_sqe_files_unregister(ctx); | |
9839 | break; | |
c3a31e60 JA |
9840 | case IORING_REGISTER_FILES_UPDATE: |
9841 | ret = io_sqe_files_update(ctx, arg, nr_args); | |
9842 | break; | |
9b402849 | 9843 | case IORING_REGISTER_EVENTFD: |
f2842ab5 | 9844 | case IORING_REGISTER_EVENTFD_ASYNC: |
9b402849 JA |
9845 | ret = -EINVAL; |
9846 | if (nr_args != 1) | |
9847 | break; | |
9848 | ret = io_eventfd_register(ctx, arg); | |
f2842ab5 JA |
9849 | if (ret) |
9850 | break; | |
9851 | if (opcode == IORING_REGISTER_EVENTFD_ASYNC) | |
9852 | ctx->eventfd_async = 1; | |
9853 | else | |
9854 | ctx->eventfd_async = 0; | |
9b402849 JA |
9855 | break; |
9856 | case IORING_UNREGISTER_EVENTFD: | |
9857 | ret = -EINVAL; | |
9858 | if (arg || nr_args) | |
9859 | break; | |
9860 | ret = io_eventfd_unregister(ctx); | |
9861 | break; | |
66f4af93 JA |
9862 | case IORING_REGISTER_PROBE: |
9863 | ret = -EINVAL; | |
9864 | if (!arg || nr_args > 256) | |
9865 | break; | |
9866 | ret = io_probe(ctx, arg, nr_args); | |
9867 | break; | |
071698e1 JA |
9868 | case IORING_REGISTER_PERSONALITY: |
9869 | ret = -EINVAL; | |
9870 | if (arg || nr_args) | |
9871 | break; | |
9872 | ret = io_register_personality(ctx); | |
9873 | break; | |
9874 | case IORING_UNREGISTER_PERSONALITY: | |
9875 | ret = -EINVAL; | |
9876 | if (arg) | |
9877 | break; | |
9878 | ret = io_unregister_personality(ctx, nr_args); | |
9879 | break; | |
7e84e1c7 SG |
9880 | case IORING_REGISTER_ENABLE_RINGS: |
9881 | ret = -EINVAL; | |
9882 | if (arg || nr_args) | |
9883 | break; | |
9884 | ret = io_register_enable_rings(ctx); | |
9885 | break; | |
21b55dbc SG |
9886 | case IORING_REGISTER_RESTRICTIONS: |
9887 | ret = io_register_restrictions(ctx, arg, nr_args); | |
9888 | break; | |
edafccee JA |
9889 | default: |
9890 | ret = -EINVAL; | |
9891 | break; | |
9892 | } | |
9893 | ||
21b55dbc | 9894 | out: |
071698e1 | 9895 | if (io_register_op_must_quiesce(opcode)) { |
05f3fb3c | 9896 | /* bring the ctx back to life */ |
05f3fb3c | 9897 | percpu_ref_reinit(&ctx->refs); |
0f158b4c | 9898 | reinit_completion(&ctx->ref_comp); |
05f3fb3c | 9899 | } |
edafccee JA |
9900 | return ret; |
9901 | } | |
9902 | ||
9903 | SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode, | |
9904 | void __user *, arg, unsigned int, nr_args) | |
9905 | { | |
9906 | struct io_ring_ctx *ctx; | |
9907 | long ret = -EBADF; | |
9908 | struct fd f; | |
9909 | ||
9910 | f = fdget(fd); | |
9911 | if (!f.file) | |
9912 | return -EBADF; | |
9913 | ||
9914 | ret = -EOPNOTSUPP; | |
9915 | if (f.file->f_op != &io_uring_fops) | |
9916 | goto out_fput; | |
9917 | ||
9918 | ctx = f.file->private_data; | |
9919 | ||
b6c23dd5 PB |
9920 | io_run_task_work(); |
9921 | ||
edafccee JA |
9922 | mutex_lock(&ctx->uring_lock); |
9923 | ret = __io_uring_register(ctx, opcode, arg, nr_args); | |
9924 | mutex_unlock(&ctx->uring_lock); | |
c826bd7a DD |
9925 | trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, |
9926 | ctx->cq_ev_fd != NULL, ret); | |
edafccee JA |
9927 | out_fput: |
9928 | fdput(f); | |
9929 | return ret; | |
9930 | } | |
9931 | ||
2b188cc1 JA |
9932 | static int __init io_uring_init(void) |
9933 | { | |
d7f62e82 SM |
9934 | #define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \ |
9935 | BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \ | |
9936 | BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \ | |
9937 | } while (0) | |
9938 | ||
9939 | #define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \ | |
9940 | __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename) | |
9941 | BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64); | |
9942 | BUILD_BUG_SQE_ELEM(0, __u8, opcode); | |
9943 | BUILD_BUG_SQE_ELEM(1, __u8, flags); | |
9944 | BUILD_BUG_SQE_ELEM(2, __u16, ioprio); | |
9945 | BUILD_BUG_SQE_ELEM(4, __s32, fd); | |
9946 | BUILD_BUG_SQE_ELEM(8, __u64, off); | |
9947 | BUILD_BUG_SQE_ELEM(8, __u64, addr2); | |
9948 | BUILD_BUG_SQE_ELEM(16, __u64, addr); | |
7d67af2c | 9949 | BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in); |
d7f62e82 SM |
9950 | BUILD_BUG_SQE_ELEM(24, __u32, len); |
9951 | BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags); | |
9952 | BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags); | |
9953 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags); | |
9954 | BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags); | |
5769a351 JX |
9955 | BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events); |
9956 | BUILD_BUG_SQE_ELEM(28, __u32, poll32_events); | |
d7f62e82 SM |
9957 | BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags); |
9958 | BUILD_BUG_SQE_ELEM(28, __u32, msg_flags); | |
9959 | BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags); | |
9960 | BUILD_BUG_SQE_ELEM(28, __u32, accept_flags); | |
9961 | BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags); | |
9962 | BUILD_BUG_SQE_ELEM(28, __u32, open_flags); | |
9963 | BUILD_BUG_SQE_ELEM(28, __u32, statx_flags); | |
9964 | BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice); | |
7d67af2c | 9965 | BUILD_BUG_SQE_ELEM(28, __u32, splice_flags); |
d7f62e82 SM |
9966 | BUILD_BUG_SQE_ELEM(32, __u64, user_data); |
9967 | BUILD_BUG_SQE_ELEM(40, __u16, buf_index); | |
9968 | BUILD_BUG_SQE_ELEM(42, __u16, personality); | |
7d67af2c | 9969 | BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in); |
d7f62e82 | 9970 | |
d3656344 | 9971 | BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST); |
84557871 | 9972 | BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int)); |
91f245d5 JA |
9973 | req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC | |
9974 | SLAB_ACCOUNT); | |
2b188cc1 JA |
9975 | return 0; |
9976 | }; | |
9977 | __initcall(io_uring_init); |